Introduction

Activities governing nutrient requirements and metabolic pathways in individual cells maintain tissue homeostasis and respond to stress through metabolite production. Adenosine triphosphate (ATP), produced via cytosolic glycolysis and mitochondrial oxidative phosphorylation (OXPHOS), is the universal energy currency of all organisms; it regulates all anabolic or catabolic cellular activities^1–3. Precise control of intracellular ATP concentrations is critical, as ATP is the rate determiner of many ATP-dependent biochemical reactions^4–9.

Hematopoietic stem cells (HSCs) are tissue stem cells at the apex of the hematopoietic hierarchy; their function is maintained throughout life by a rigorous metabolic program and a complex interplay of gene expression, epigenetic regulation, intracellular signaling, chromatin remodeling, autophagy, and environmental factors^10–14. Conventional analyses of the metabolic programs of hematopoietic stem and progenitor cells (HSPCs) have revealed diverse differentiation potentials and cell-cycling statuses and coordinated activities that maintain hematopoiesis^15–26. Among the HSPC fractions, HSCs possess unique cell cycle quiescence, high self-renewal and differentiation capacity in response to stimuli, and resistance to cellular stress, including reactive oxygen species and physiological aging^{10, 27–30}. These properties are regulated by a balance between glycolysis and mitochondrial OXPHOS, requiring biosynthesis of ATP and various macromolecules that confer resilience to stress³¹. Among the known regulators of ATP-producing pathways, glycolytic enzymes maintain HSCs and hematopoietic progenitor cells (HPCs) by regulating cellular survival and cell cycle quiescence^32–34. Loss of mitochondrial genes in HPSCs also induces HSC differentiation defects^35–37. Moreover, disrupting the mitochondrial complex III subunit depletes both differentiated hematopoietic cells and quiescent HSCs²¹. Although glycolysis and the tricarboxylic acid (TCA) cycle are metabolically linked, pyruvate dehydrogenase kinase activity, which can uncouple these pathways, is required to maintain HSC function^{32, 38}.

During HSC division, cell metabolism is reprogrammed to activate fatty acid β-oxidation and purine metabolism^39–41. Furthermore, Liang et al. reported that activated HSCs mainly rely on glycolysis as their energy source⁴². However, the mechanisms by which each ATP-producing pathway and their connections are differentially regulated between HSCs and differentiated cells at steady state, during cell cycling, or during stress remain unknown. Recently, it has been shown that deeply quiescent HSCs do not activate cell cycle under stress^43–45. Therefore, it remains unclear whether metabolic changes such as the individual ATP-producing pathways and their interconnections occur uniformly in all HSCs, including these deeply quiescent HSCs. Furthermore, the underlying hub metabolic enzyme responsible for changes in the metabolic system of HSCs under stress has not been identified. HSCs are essential for cell therapy, including HSC transplantation, and in order to comprehensively elucidate the metabolic systems that have attracted attention as their regulatory mechanisms, recent studies have included metabolomic analyses using rare cell types such as HSCs^{23, 46–49}, as well as isotope tracer analyses of undifferentiated hematopoietic cells purified after in vivo administration of isotopic glucose⁵⁰. Although these approaches are useful for obtaining comprehensive information on intracellular metabolites, they are not suited to track real-time changes in cellular metabolism at high resolution. Therefore, new approaches are necessary to analyze metabolites quantitatively and continuously without disturbing the physiological states of single cells while integrating the recently reported metabolome analysis techniques. In this study, we aimed to identify the key metabolic enzymes that define metabolic differences between steady-state and stress conditions in HSCs and elucidate their regulatory mechanisms using a quantitative and mathematical approach. Our findings provide a platform for quantitative metabolic analysis of rare cells such as HSCs, characterize the overall metabolic reprogramming of HSCs during stress loading, and highlight the key enzyme involved in this process.

Methods

Mice

C57BL/6 mice (7–16 weeks old, Ly5.2⁺) were purchased from Japan SLC (Shizuoka, Japan). C57BL/6 mice (Ly5.1⁺) were purchased from CLEA Japan (Shizuoka, Japan). Knock-in mice harboring GO-ATeam2^51–53 in the ROSA26 locus were generated in the Yamamoto laboratory. The GO-ATeam2 mice (8–16 weeks old) were used to analyze HSPCs. mVenus-p27K^- mice (17–20 weeks old) were provided by the Kitamura Laboratory at The Institute of Medical Science, The University of Tokyo and used for cell cycle analysis^{54, 55}. Mice were genotyped using PCR-based assays of tail DNA or transdermal enhanced green fluorescent protein (EGFP) fluorescence. All mice were maintained in the animal facility at the National Center for Global Health and Medicine Research Institute under specific pathogen-free conditions and fed ad-libitum. Mice were euthanized by cervical dislocation. All animal experiments were approved by the Institutional Animal Care and Use Committee (IACUC) at National Center for Global Health and Medicine Research Institute. Both male and female mice were used in experiments.

Ion chromatography mass spectrometry (IC-MS) analysis

For metabolome analysis focused on glycolytic metabolites and nucleotides, anionic metabolites were measured using an orbitrap-type MS (Q-Exactive Focus; Thermo Fisher Scientific, Waltham, MA, USA) connected to a high-performance IC system (ICS-5000+, Thermo Fisher Scientific), enabling highly selective and sensitive metabolite quantification owing to the IC-separation and Fourier Transfer MS principle⁵⁶. The IC instrument was equipped with an anion electrolytic suppressor (Dionex AERS 500; Thermo Fisher Scientific) to convert the potassium hydroxide gradient into pure water before the sample entered the mass spectrometer. Separation was performed using a Dionex IonPac AS11-HC-4 μm IC column (Thermo Fisher Scientific). The IC flow rate was 0.25 mL/min supplemented post-column with 0.18 mL/min makeup flow of MeOH. The potassium hydroxide gradient conditions for IC separation were as follows: 1–100 mM (0–40 min), 100 mM (40–50 min), and 1 mM (50.1–60 min), with a column temperature of 30 °C. The Q-Exactive Focus mass spectrometer was operated under ESI negative mode for all detections. Full mass scan (m/z 70−900) was performed at a resolution of 70,000. The automatic gain control target was set at 3 × 10⁶ ions, and the maximum ion injection time was 100 ms. Source ionization parameters were optimized with spray voltage at 3 kV, and other parameters were as follows: transfer temperature, 320 °C; S-lens level, 50; heater temperature, 300 °C; sheath gas, 36; and aux gas, 10. Metabolite amounts were quantified from calibration curve data generated based on peak areas and respective metabolite amounts. Detailed methods for sample collection and IC-MS analysis are described in SUPPLEMENTARY METHODS.

Time-course analysis of ATP levels in single hematopoietic cells using GO-ATeam2 system

GO-ATeam2 is a ratiometric biosensor that monitors ATP concentration through Förster resonance energy transfer (FRET) from EGFP to the monomeric version of Kusabira Orange (mKO), regardless of the sensor expression levels⁵². Surface-marker-stained bone marrow (BM) mononuclear cells (MNCs) from GO-ATeam2 mice were dispensed into basal medium (Ba-M) containing minimal salts, vitamins, and buffers (HEPES and sodium bicarbonate), but no glucose or mitochondrial substrates (Table S1), or into medium containing mitochondrial substrates, pyruvate, lactate, fatty acids, and amino acids, but no glucose (PLFA medium). The FRET/EGFP ratio was analyzed in a real-time with a BD FACSAria IIIu Cell Sorter (BD Biosciences, Franklin Lakes, NJ, USA) under ambient pressure, and ATP levels were calculated. Detailed methods for sample collection and ATP concentration calculation are provided in SUPPLEMENTARY METHODS.

Statistical analysis

Data are presented as means ± SD, unless otherwise stated. For multiple comparisons, statistical significance was determined by Tukey’s multiple comparison test using the Tukey HSD function in R ×64 4.0.3 software (R Core Team, Vienna, Austria). A paired or unpaired two-tailed Student’s t-test and two-way ANOVA with Sidak’s test were used for experiments with two groups. A p-value <0.05 was considered statistically significant.

Data sharing

All relevant data are available from the corresponding author upon reasonable request.

Results

HSC cell cycling increases anaerobic glycolytic flux

To determine how cell cycle progression alters HSC metabolism in vivo, we intraperitoneally and intravenously treated mice with 5-fluorouracil (5-FU) to induce HSC cell cycling (Supplemental Figure 1A). For analysis after 5-FU administration, the Lineage (Lin)^- Sca-1⁺ c-Kit⁺ (LSK) gate was expanded to include as many HSCs as possible, for example high Sca-1-expressing cells and c-Kit-high to -dim Lin^- cells, based on the previous report⁵⁷ (Supplemental Figure 1B). This expanded LSK gate was consistent with the patterns of Sca-1 and c-Kit expression observed in EPCR⁺ Lin^- CD150⁺ CD48^- cells, and the presence of EPCR, as a surface marker, identified HSCs with high stem cell activity after 5-FU administration⁴¹. We observed a transient decrease in the number of quiescent HSCs (Ki67^-) and an increase in the number of cell-cycling HSCs (Ki67⁺) on day 6 after 5-FU treatment (Supplemental Figure 1C). Along with the loss of cell quiescence, ATP concentration in HSCs decreased transiently on day 6 (Supplemental Figure 1D). Because the route of administration of 5-FU (intraperitoneal or intravenous) made no difference in the Ki67 positivity rate of HSCs (Supplemental Figure 1E), we administered 5-FU intraperitoneally for remaining experiments. Two other methods were used to test whether cell cycle progression of HSCs after 5-FU treatment depends on the expression of EPCR. First, phosphorylation of Rb (pRb), a marker of cell cycle progression⁵⁸, was analyzed in HSCs after 5-FU treatment. Analysis of EPCR⁺ and EPCR^- HSCs showed increased pRb in HSCs from 5-FU-treated mice in both fractions compared to HSCs from phosphate-buffered saline (PBS)-treated mice, regardless of EPCR expression (Supplemental Figure 1F-G). Second, we used a G₀ marker mouse line⁵⁵. These mice expressed a fusion protein of the p27 inactivation mutant p27K^- and the fluorescent protein mVenus (G₀ marker), allowing prospective identification of G₀ cells. We tested whether the expression of G₀ marker in HSCs was altered after 5-FU administration to the G₀ marker mice (Supplemental Figure 1H) and found that 5-FU treatment reduced the frequency of G₀ marker-positive HSCs, regardless of the EPCR expression (Supplemental Figure 1I-J). This was not observed in the PBS group. These results indicated that 5-FU administration induced cell cycle progression of entire HSCs in mice.

HSC cell cycling is preceded by the activation of intracellular ATP-related pathways that metabolize extracellular nutrients, including glucose^{39, 40}, which are utilized in both ATP-producing and -consuming pathways, determining cellular ATP levels. Therefore, we examined the metabolic flux of glucose by performing in vitro IC-MS tracer analysis with uniformly carbon-labeled (U-¹³C₆) glucose to determine the pathways driving changes in ATP in 5-FU-treated HSCs (Figure 1A; Table S2). To avoid metabolite changes, samples were continuously chilled on ice during cell preparation, and the process from euthanasia to cell preparation was performed in the shortest possible time (see “Preparation and storage of in vitro U-¹³C₆-glucose tracer samples” section under “Supplementary Methods” for more information). We found that changes in metabolite levels before and after sorting were present but limited (Supplemental Figure 2A). This result is consistent with the finding that the cell purification process does not significantly affect metabolite levels when sufficient care is taken in cell preparation⁵⁰. In 5-FU-treated HSCs, the levels of glycolytic metabolites derived from U-¹³C₆-glucose were double those observed in PBS-treated HSCs (Figure 1B-C; Supplemental Figure 2B). The total levels of TCA cycle intermediates derived from U-¹³C₆-glucose were similar between PBS- and 5-FU-treated cells (Figure 1D; Supplemental Figure 2B). Levels of U-¹³C₆-glucose-derived intermediates involved in the pentose phosphate pathway (PPP) and nucleic acid synthesis (NAS) were two-fold higher in 5-FU-treated than in PBS-treated HSCs, whereas no significant differences in the levels of metabolites were observed between both groups (Figure 1E-F; Supplemental Figure 2B). Notably, the labeling rate of metabolites during the first half of glycolysis was almost 100% in both groups, allowing us to easily track the labeled metabolites (Supplemental Figure 2C-E). This was thought to be due to the rapid replacement of unlabeled metabolites with labeled metabolites during exposure to U-¹³C₆-glucose because of the generally rapid glycolytic reaction. Next, we evaluated whether glucose uptake in HSCs after 5-FU administration was differentially affected by the expression of EPCR. The fluorescent analog of glucose, 2-(N-(7-nitrobenz-2-oxa-1,3-diazol-4-yl)amino)-2-deoxyglucose (2-NBDG), was administered intravenously to mice⁵⁰ and its uptake in EPCR⁺ and EPCR^- HSCs was assayed (Figure 1G). Regardless of the EPCR expression, the 2-NBDG uptake was greater in HSCs treated with 5-FU than in those treated with PBS (Figure 1H-J). Furthermore, compared with HSCs cultured under quiescence-maintaining conditions, those cultured under proliferative conditions were more resistant to OXPHOS inhibition by oligomycin (Supplemental Figure 1H; Table S1). Overall, the results showed that 5-FU-treated HSCs exhibited activated glycolytic flux, increasing the turnover of ATP. Moreover, glycolytic flux into mitochondria was equally unchanged in PBS- and 5-FU-treated-HSCs, supporting that 5-FU activated anaerobic glycolysis in HSCs.

Figure 1.

OXPHOS-inhibited HSCs exhibit compensatory glycolytic flux

Previous studies using mouse models of mitochondrial disease or defects in genes involved in electron transport chain and OXPHOS suggest that mitochondrial energy production is essential for maintaining HSC function^{21, 35–37}, as is the glycolytic system. However, there have been no quantitative reports on how OXPHOS-inhibited HSCs can adapt their metabolism. To understand HSC metabolism under OXPHOS inhibition, we performed in vitro U-¹³C₆-glucose tracer analysis of oligomycin-treated HSCs (Figure 2A; Table S3). Similar to 5-FU-treated HSCs (Figure 1), oligomycin-treated HSCs exhibited glycolytic system activation (Figure 2B-C; Supplemental Figure 2B). Metabolite flux to the TCA cycle and PPP was unchanged, but flux to the NAS was significantly reduced in oligomycin-treated HSCs compared to that in steady-state HSCs (Figure 2D-F; Supplemental Figure 2B). The results suggested that OXPHOS-inhibited HSCs activated compensatory glycolytic flux and suppressed NAS flux. As with 5-FU-treated HSCs, analysis of oligomycin-treated HSCs also showed almost 100% labeling of metabolites in the first half of glycolysis (Supplemental Figure 2F-H), allowing us to easily track the labeled metabolites. To further validate the compensatory glycolytic activation of HSCs under OXPHOS inhibition, a Mito Stress test was performed on HSCs and other differentiated myeloid progenitors (MyPs, Lin^-Sca-1^-c-Kit⁺ (LKS^-) cells). The results showed that extracellular acidification rates (ECAR) were elevated in HSCs after oligomycin treatment compared to before oligomycin treatment (Figure 2G-H). No increase in ECAR was observed in MyPs (Figure 2G-H), supporting that inhibition of OXPHOS activated anaerobic glycolysis specifically in HSCs.

Figure 2.

Phosphofructokinase (PFK) metabolism in HSCs is activated during proliferation and OXPHOS inhibition

To investigate whether glycolytic activation in HSCs after 5-FU treatment and OXPHOS inhibition could be demonstrated through unbiased mathematical simulations, we performed quantitative ¹³C metabolic flux analysis (¹³C-MFA). After generating a metabolic model for isotope labeling enrichment and setting appropriate lactate efflux values, a simulation was conducted using the labeled metabolite abundance data obtained from isotope tracer analysis. The appropriate lactate efflux for quiescent HSC (PBS-treated HSC) was determined to 65 after experimenting with values from 0–100. The lactate efflux of 5-FU- or oligomycin-treated HSCs was higher than that of quiescent HSCs based on the observation that labeled glycolytic metabolite levels were particularly elevated in in vitro tracer analysis (see “Quantitative ¹³C-MFA with OpenMebius” under “Supplementary Methods” for more information). As a result, the variation in the flux values of all enzymatic reactions calculated in HSCs after 5-FU or oligomycin treatment became smaller compared to quiescent HSCs, suggesting that HSCs strictly regulated their metabolism in response to stress (Supplemental Figure 3A-C). Unlike PBS-treated HSCs, those treated with 5-FU or oligomycin exhibited preferential glycolytic activation rather than TCA- or PPP-based metabolic strategies; the first half of the glycolytic system appeared to be the site of metabolic activation (Figure 3A-D; Supplemental Figure 3D-U, Table S4). This increase in metabolic flux upstream of the glycolytic pathway was also supported by our in vitro tracer analysis (Figure 1B and Figure 2B), suggesting that ¹³C-MFA was a valid metabolic simulation. Among the reactions in the first half of glycolysis, phosphorylation of fructose 6-phosphate by PFK is the irreversible and rate-limiting reaction⁵⁹. A detailed review of in vitro isotope tracer analysis results showed that the ratio of fructose 1,6-bisphosphate (F1,6BP; the product of PFK) to F6P (the substrate of PFK) was greatly elevated in HSCs during proliferation and OXPHOS inhibition (Figure 3J-K). Together with the results of quantitative ¹³C-MFA, these findings suggested that HSCs exhibit elevated glycolytic flux relative to mitochondrial activity by increasing PFK enzyme activity under various stress conditions.

Figure 3.

HSCs under stress exhibit activation of glycolysis-initiated TCA cycle and NAS

To investigate the long-term glucose utilization of HSCs, we performed an in vivo tracer analysis with U-¹³C₆ glucose based on recent reports^{47, 50} (Supplemental Figure 4A; see “Preparation and storage of in vivo U-¹³C₆-glucose tracer samples” under “Supplementary Methods” for more information). In HSCs from 5-FU-treated mice, we observed increased labeling of glycolytic metabolites such as dihydroxyacetone phosphate, glycerol-3-phosphate, and phosphoenolpyruvate, as well as NAS metabolites such as inosine monophosphate and ATP, and those derived from TCA cycle such as aspartic acid and glutamate, compared to HSCs from PBS-treated mice (Supplemental Figure 4B-I, Table S5). When the amount of U-¹³C₆-glucose-derived labeled metabolites in each pathway was calculated, more glucose-derived metabolites entered TCA cycle in the 5-FU-treated group than PBS-treated group (Supplemental Figure 4J). Thus, although short-term (10–30 min) in vitro tracer analysis showed that HSCs exhibited more potent activation of anaerobic glycolysis than of other pathways in response to 5-FU administration, long-term (approx. 3 h) labeling by in vivo tracer analysis revealed that glycolysis-initiated TCA cycle and NAS flux were activated in addition to enhanced anaerobic glycolysis.

PFKFB3 accelerates glycolytic ATP production during HSC cell cycling

In vitro and in vivo tracer analysis results collectively suggested that the activation of glycolysis catalyzed by PFK may have been the starting point for the activation of the entire HSC metabolism. To analyze the contribution of PFK to ATP metabolism in steady-state or stressed HSCs, we needed to develop an experimental system that could measure the dynamics of ATP concentrations in HSCs in a non-destructive, real-time manner. To this end, we used knock-in GO-ATeam2 mice as a FRET-based biosensor of ATP concentration (see “Time-course analysis of ATP levels in single hematopoietic cells using GO-ATeam2 system” under “Methods” for more information.). The number of BMMNCs, as well as the frequency of HSCs (CD150⁺CD48-LSK) and other progenitor cells, in the BM of GO-ATeam2⁺ mice were almost unchanged compared to of C57B6/J mice, except for a mild decrease in the frequency of the Lin^- fraction (Supplementary Figure 5A-C). Using BMMNCs derived from GO-ATeam2⁺ mice, we developed a method to detect changes in ATP concentration with high temporal resolution when the activity of PFK was modulated (METHODS, Supplemental Figure 5D-F). To validate our methods, we measured ATP concentrations in HSCs and MyPs with or without various nutrients (see “Time-course analysis of FRET values” under “Supplementary Methods” for more information.). MyPs showed more rapid decreases in ATP concentration than HSCs, suggesting higher ATP consumption by progenitors (Supplemental Figure 5G-H). Adding glucose to the medium suppressed this decrease in MyPs; however, other metabolites (e.g., pyruvate, lactate, and fatty acids) had minimal effects, suggesting that ATP levels are glycolysis-dependent in MyPs (Supplemental Figure 5G-H), consistent with previous reports that the aerobic glycolytic enzyme M2 pyruvate kinase isoform (PKM2) is required for progenitor cell function³³.

Further, we analyzed ATP consumption and metabolic dependency of cell-cycling HSCs after 5-FU administration (Figure 4A). After inhibiting glycolysis using 2-deoxy-D-glucose (2-DG) with other mitochondrial substrates, 5-FU-treated HSCs showed more rapid decreases in ATP concentration than PBS-treated HSCs (Figure 4B-C). In contrast, OXPHOS inhibition by oligomycin without glucose or mitochondrial substrates decreased the ATP concentration to a similar extent in both 5-FU- and PBS-treated HSCs, although 5-FU-treated HSCs showed earlier ATP exhaustion (Figure 4D-E). These data suggest that 5-FU-treated-HSCs upregulated ATP production via glycolysis, rather than relying on mitochondria. Apoptosis assay revealed a slight increase in early apoptotic cells (annexin V⁺ propidium iodide [PI]^-) after 2-DG treatment and a slight decrease in the number of viable cells (Annexin V^- PI^-) after oligomycin treatment, both to a very limited extent (approx. 5%) compared to the degree of ATP decrease, suggesting that the decrease in ATP after 2-DG or oligomycin treatment did not simply reflect cell death (Supplementary Figure 5I).

Figure 4.

PFK is allosterically activated by 6-phosphofructo-2-kinase/fructose-2,6-bisphosphatase (PFKFB). Among the four isozymes of mammalian PFKFB, PFKFB3 is the most favorable for PFK activation⁶⁰. We investigated whether PFKFB3 contributes to glycolytic plasticity in HSCs during proliferation. When treated with the PFKFB3 inhibitor AZ PFKFB3 26, compared with HSCs from PBS-treated mice, HSCs from 5-FU-treated mice showed decreased ATP levels (Figure 4F, H, Supplementary Figure 5J). Although AMPK activates PFKFB3 in other contexts⁶¹, AMPK inhibition by dorsomorphin did not alter ATP concentration in 5-FU-treated-HSCs (Figure 4G, I).

OXPHOS inhibition accelerates glycolysis to sustain ATP levels in HSCs, but not in progenitors

To assess differences in metabolic dependence between steady-state or stressed and naturally proliferating HPCs, we altered ATP metabolism in HSCs and progenitors using 2-DG or oligomycin (Figure 5A). Oligomycin treatment rapidly depleted ATP in HSCs and all HPC fractions (green lines in Figure 5B-C; Supplemental Figure 6A-D). Treatment with 2-DG decreased ATP concentrations for a short amount of time (∼12 min) in HSCs and HPCs, but ATP reduction was less evident than that induced by oligomycin (blue lines in Figure 5B-C; Supplemental Figure 6A-D). The ATP reduction induced by 2-DG treatment was particularly low (∼15%) in HSCs, multipotent progenitor cells (MPPs), and common lymphoid progenitors (CLPs) relative to that in common myeloid progenitors (CMPs), granulocytes-macrophage progenitors (GMPs), and megakaryocyte-erythrocyte progenitors (MEPs) (Figure 5D).

Figure 5.

Next, we investigated the role of glycolysis in ATP production during OXPHOS inhibition by combining oligomycin administration and glucose supplementation. ATP concentration remained more stable in HSCs treated with oligomycin and glucose than in those treated only with oligomycin. Similar results were not seen in HPCs, indicating that HSCs have the plasticity to upregulate glycolytic ATP production to meet demands (orange lines in Figure 5B-C; Supplemental Figure 6A-D, summarized in Figure 5E). Similar to oligomycin treatment, rotenone (Complex I inhibitor) and carbonyl cyanide 4-(trifluoromethoxy)phenylhydrazone (FCCP, mitochondrial uncoupler) treatments, which inhibit OXPHOS-derived ATP production, also decreased ATP concentrations in HSCs, but not when administered simultaneously with glucose (Supplementary Figure 6E-F). Furthermore, with oligomycin, HSCs, but not HPCs, maintained ATP concentrations at low glucose levels (50 mg/dL) (Supplemental Figure 6G). These analyses suggest that ATP was produced by mitochondrial OXPHOS in steady-state HSCs, and that only HSCs, but not HPCs, maintained ATP production by glycolysis when OXPHOS was compromised.

PFKFB3 accelerates glycolytic ATP production during OXPHOS inhibition

Next, to understand whether PFKFB3 contributes to ATP production in HSCs under OXPHOS inhibition, we evaluated PFKFB3 function under OXPHOS inhibition using the GO-ATeam2⁺ BMMNCs. In oligomycin-treated HSCs, PFKFB3 inhibition led to rapidly decreased ATP concentration that was not observed in HSCs not treated with oligomycin (Figure 5F-I). We examined the effects of HSPC metabolic regulators on ATP levels in oligomycin-treated HSCs. Inhibiting PKM2, which accelerates glycolysis in steady-state progenitors³³, significantly reduced ATP levels in oligomycin-treated HSCs (Supplemental Figure 6H, J). Inhibiting LKB1, a kinase upstream of AMPK^{62, 63}, did not affect the ATP concentration in oligomycin-treated HSCs (Supplemental Figure 6I, K), whereas levels of adenosine monophosphate (AMP), which also activates AMPK, increased in oligomycin-treated but not in 5-FU-treated HSCs (Supplemental Figure 6L). This may explain differences in AMPK-dependent ATP production between proliferative HSCs and HSCs under OXPHOS inhibition.

Next, we tested the effects of PFKFB3 on ATP concentration in HPCs. Unlike HSCs, HPCs exhibited PFKFB3-dependent ATP production, even without oligomycin (Supplemental Figure 6M-Q). Therefore, ATP production in steady-state HSCs was PFKFB3-independent, and proliferative stimulation or OXPHOS inhibition plastically activated glycolytic ATP production in a PFKFB3-dependent manner to meet ATP demand.

PFKFB3 activity renders HSCs dependent on glycolysis

Next, we investigated whether PFKFB3 activity itself confers glycolytic dependence on HSCs. We retrovirally overexpressed Pfkfb3 in HSCs and performed cell cycle analysis (Figure 5J). Pfkfb3-overexpressed HSCs increased the proportion of cells in the S/G2/M phase and decreased the number of G₀ cells compared to mock-overexpressed HSCs (Figure 5K). Next, we retrovirally overexpressed Pfkfb3 in GO-ATeam2⁺ HSCs and performed real-time ATP measurement (Figure 5J). Pfkfb3-overexpressing GO-ATeam2⁺ HSCs did not show changes in ATP concentrations relative to those in mock-transduced cells (Figure 5L; Supplemental Figure 6R). Upon 2-DG treatment, Pfkfb3-overexpressing HSCs showed a greater decrease in ATP concentration than mock-transduced HSCs did (Figure 5M; Supplemental Figure 6S). However, oligomycin treatment of both mock-transduced and Pfkfb3-overexpressing HSCs decreased ATP concentration to comparable levels (Figure 5N; Supplemental Figure 6T). Notably, Pfkfb3-overexpressing HSCs recovered ATP levels more effectively under low glucose conditions (12.5 mg/dL) than did mock-transduced HSCs (Figure 5O; Supplemental Figure 6U). These data suggest that PFKFB3 directly conferred glycolytic dependence onto HSCs by modulating the cell cycle and increasing their ATP-generating capacity via glycolysis under metabolic stress.

PFKFB3 methylation by PRMT1 supports ATP production by cell-cycling HSCs

Next, we investigated how 5-FU-treated-HSCs regulate PFKFB3 independently of AMPK (Figure 4G, I). PFKFB3 activity is regulated at multiple levels⁶⁴, and PFKFB3 transcript and protein levels in HSCs remained unchanged during 5-FU-induced cell cycling (Figure 6A-B). Phosphorylation can also regulate PFKFB3 activity^{61, 65, 66}; however, we observed no change in PFKFB3 phosphorylation in 5-FU-treated-HSCs (Figure 6C). Upon oligomycin exposure, PFKFB3 was phosphorylated by AMPK in the HSCs (Figure 6D). PFKFB3 is also methylated, and its activity is upregulated by protein arginine methyltransferase 1 (PRMT1)⁶⁷. We observed that Prmt1 expression increased in 5-FU-treated-HSCs relative to that in PBS-treated-HSCs (Figure 6E). Furthermore, PFKFB3 methylation was significantly induced in 5-FU-treated-HSCs than in PBS-treated-HSCs (Figure 6F). Treatment of HSCs with a PRMT1 inhibitor decreased PFKFB3 methylation (Figure 6G), suggesting that PRMT1 catalyzed PFKFB3 methylation. To investigate whether glycolytic activity in HSCs was regulated by m-PFKFB3, mice treated with PBS or 5-FU were injected with 2-NBDG, and m-PFKFB3 levels in HSCs with high and low 2-NBDG uptake were quantified. Regardless of PBS or 5-FU treatment, HSCs with high 2-NBDG uptake exhibited higher m-PFKFB3 levels than those with low uptake (Figure 6H), suggesting that m-PFKFB3 regulated the activity of the glycolytic system in HSCs.

Figure 6.

Further, we analyzed the potential effects of PRMT1 inhibition on ATP concentration in GO-ATeam2⁺ HSCs. Treatment with the PRMT1 inhibitor significantly decreased ATP levels in 5-FU-treated-HSCs than in PBS-treated-HSCs (Figure 6I-J). These findings indicated that ATP levels in 5-FU-treated-HSCs were supported by PRMT1 methylation–mediated PFKFB3 activation.

PFKFB3 contributes to HSPC pool expansion and stress hematopoiesis maintenance

Finally, we analyzed PFKFB3 function in HSCs during hematopoiesis. We cultured HSCs with a PFKFB3 inhibitor in vitro under quiescence-maintaining or proliferative conditions (Supplementary Figure 7A)⁶⁸. Cell count in HSC-derived colonies decreased following treatment with a PFKFB3 inhibitor under proliferative, but not quiescence-maintaining, conditions (Supplementary Figure 7B).

We also knocked out Pfkfb3 in HSCs using the less toxic, vector-free CRISPR-Cas9 system and cultured the cells under quiescence-maintaining or proliferative conditions (Supplementary Figure 7A) based on recent reports by Shiroshita et al.⁶⁹. Again, cell numbers in Pfkfb3-knockout (KO) HSC–derived colonies decreased only in proliferative cultures when compared to control cultures (Rosa26-KO HSCs) (Supplementary Figure 7C, E, F). We retrovirally overexpressed Pfkfb3 in HSCs and cultured them under quiescence maintenance or proliferative conditions (Supplementary Figure 7A). Pfkfb3-overexpressing HSC colonies showed increased cell count compared to that of mock-transduced cells, but only under proliferative conditions (Supplementary Figure 7D).

To assess PFKFB3 function in HSCs in vivo, we transplanted Pfkfb3-KO HSCs (Ly5.2⁺) or wild type (WT) control HSCs into lethally irradiated recipients (Ly5.1⁺) as well as Ly5.1⁺ competitor cells (Figure 7A), and the behavior of Pfkfb3-KO cells was evaluated by Sanger sequencing of peripheral blood (PB) cells⁶⁹. In the KO group, donor-derived chimerism in PB cells decreased relative to that in the WT control group during the early phase (1 month post-transplant) but recovered thereafter (Figure 7B). Next, we retrovirally transduced Ly5.2⁺ HSCs with Pfkfb3 S461E (Pfkfb3CA), a constitutively active PFKFB3 mutant, and transplanted them into lethally irradiated recipients (Ly5.2⁺), along with Ly5.1⁺ competitor cells (Figure 7A, Supplementary Figure 7G). Donor chimerism during the early post-transplant period in the Pfkfb3CA-overexpressing group was significantly higher than that in the mock-transduced group (Figure 7C). These findings suggested that PFKFB3 was required for differentiation and proliferation of HSCs to expand the HSPC pool. Therefore, we compared the contribution of PFKFB3 to HSPC function at steady state and after myeloproliferative stimulation. Pfkfb3- or Rosa-KO HSPCs were transplanted into recipients (Ly5.1⁺). After 2 months, recipients received 5-FU intraperitoneally, and the dynamics of Pfkfb3- or Rosa-KO cell abundance in PB was assessed (Figure 7D). In PB cells prior to 5-FU administration, Pfkfb3- or Rosa-KO HSPC-derived blood cells were almost equally present, suggesting that the loss of PFKFB3 did not affect steady-state blood cell production (Figure 7E). However, after 5-FU administration, Pfkfb3-KO HSPC-derived blood cell abundance was reduced compared to that in the Rosa group (Figure 7E). This change occurred on day 6 after 5-FU administration (day 1), when the cell cycle of HSCs was activated (Supplementary Figure 1B), suggesting that PFKFB3 served a function in the proliferation and differentiation of HSPCs from HSCs.

Figure 7.

To investigate the mechanisms underlying the short-term effects of PFKFB3 on hematopoiesis after bone marrow transplantation (BMT), we evaluated cell cycle and apoptosis of Pfkfb3-KO or -overexpressing HSPCs on day 2 after BMT (Figure 7F). Cell cycle was analyzed by Ki67/Hoechst33432 staining and in vivo BrdU labeling ⁵⁰, which showed that cell cycle progression was suppressed in Pfkfb3-KO HSPCs (Figure 7G-J). In contrast, Pfkfb3 KO cells did not exhibit aberrant apoptosis of HSPCs after BMT (Figure 7K). Furthermore, we examined the cell cycle of HSPCs overexpressing Pfkfb3CA on day 2 after BMT (Figure 7L) and found that Pfkfb3CA-overexpressing HSPCs showed accelerated cell cycle compared to mock-overexpressing HSPCs (Figure 7M-N). These data indicate that PFKFB3 governs cell cycle progression and promotes the production of differentiated cells from HSCs in proliferative environments, both in vitro in the presence of cytokine stimulation and immediately after transplantation in vivo.

Discussion

In this study, by combining metabolomic tracing of U-¹³C₆-labeled glucose and ¹³C-MFA, we quantitatively identified the metabolic programs used by HSCs during steady-state, cell-cycling, and OXPHOS inhibition. Under proliferative stress, HSCs uniformly shift from mitochondrial respiration to glycolytic ATP production and PPP activation, which represent hallmarks of cell-cycling mammalian cells⁷⁰. Previous reports have emphasized the importance of glycolysis in maintaining HSC quiescence, but have primarily analyzed HSCs in transplant assays, wherein HSCs must enter the cell cycle^{32, 71}. Prior analysis of repopulation capacity, which is positively correlated with enhanced glycolysis, may have overestimated glycolytic ATP production and overlooked mitochondrial ATP production during native hematopoiesis. In fact, some studies have suggested that OXPHOS activity is important for HSC maintenance and function²¹.

Our method was based on recently reported quantitative metabolic analysis techniques for very small numbers of cells^{23, 46–50}, such as HSCs, and expands our knowledge of HSC metabolism during stress hematopoiesis. In our study, 5-FU administration in mice transiently decreased ATP concentration in HSCs in parallel with cell cycle progression, suggesting that HSC differentiation and cell cycle progression are closely related to intracellular metabolism and can be monitored by measuring ATP concentration. We mainly analyzed a mixture of EPCR⁺ and EPCR^- HSCs, and we believe that the observed cell cycle progression and promotion of glycolysis in both EPCR⁺ and EPCR^- HSCs support the validity of our claims (Figure 1J, Supplementary Figure 1F-J). According to ¹³C-MFA enzymatic reaction flux of PFK in 5-FU-treated HSCs indicated a relative increase of approximately 10%. However, the flux value obtained by ¹³C-MFA was calculated with glucose uptake as 100. Thus, when combined with the overall increase in the glycolytic pool demonstrated by in vitro isotopic glucose tracer analysis and in vivo NBDG analysis, rapid acceleration of glycolysis becomes evident throughout the HSCs, including subpopulations that were less responsive to stress^43–45. These findings are consistent with reports suggesting that HSCs have relatively low biosynthetic activity^{72, 73} that is rapidly activated in response to cell proliferation stimuli^{40, 74}. Notably, we found that HSCs could accelerate glycolytic ATP production to fully compensate for mitochondrial ATP production under OXPHOS inhibition, a phenomenon that is difficult to identify without real-time ATP analysis. On the other hand, we observed that ATP production in steady-state or cell-cycling HSCs and in naturally proliferating HPCs depended more on mitochondrial OXPHOS than on glycolysis; inhibiting glycolysis in steady-state HSCs resulted in only mild ATP decreases. This is in contrast with the rapid decrease in ATP levels caused by OXPHOS inhibition. Because fatty acid β-oxidation is necessary for HSC self-renewal and quiescence^{39, 68}, fatty acids may support mitochondrial ATP production independently of fluxes from glycolysis. Furthermore, although glycolysis and TCA cycle are decoupled in steady-state HSCs, in response to cell cycle progression, anaerobic glycolytic metabolism in HSCs is enhanced (Figure 1) and fluxes to TCA cycle and PPP from the glycolytic system are also promoted (Supplementary Figure 4). The difference in the degree of coupling between glycolysis and the TCA cycle between in vitro and in vivo tracer analysis may be explained by the difference in labeling time (10–30 min vs. 3 h). Altogether, these data provide direct evidence that glycolysis and TCA cycle become functionally uncoupled in quiescent HSCs^{32, 38}. Our findings are also consistent with previous reports of OXPHOS activation associated with HSC proliferation^{32, 36, 75, 76}. In other words, HSCs exhibit an increased proportion of anaerobic glycolysis–derived ATP by PFKFB3 upon proliferation and OXPHOS inhibition; furthermore, the glycolytic system is the starting point of metabolic activation and is indispensable for the overall enhancement of HSC metabolism (Figure 7H).

HPCs and leukemic cells accelerate glycolytic ATP production using PKM2 for differentiation and transformation, respectively³³; however, we demonstrated that glycolytic acceleration does not fully compensate for mitochondrial ATP production in HPCs. Mechanistically, PFKFB3 increased glycolytic activity in HSCs to maintain ATP concentrations during proliferation and OXPHOS inhibition. Under steady-state conditions, naturally proliferating HPCs rely on PFKFB3 for ATP production, whereas HSCs do not. Therefore, we can infer that quiescent HSCs at steady state can produce ATP via PFKFB3 activation in response to stress, enabling ATP generation. Furthermore, overexpression of Pfkfb3 in HSCs increased glycolytic dependency, suggesting that PFKFB3 itself can modulate metabolic dependency in HSCs. Changes in glycolytic dependency in HSCs overexpressing Pfkfb3 may seem small (0.06–0.13 mM) (Figure 5M, O). However, it is noteworthy that the rate of the reaction catalyzed by PFK varies greatly within a very narrow range of ATP concentrations, less than 1 mM. Webb et al. analyzed the factors controlling PFK activity and reported that the reaction rate of PFK varies by approximately 40% in the 0.3–1 mM ATP concentration range⁷⁷. The reason that differences in glycolytic dependence could be detected in cells overexpressing Pfkfb3 may be that the ATP concentration at the time of analysis was approximately 0.5–0.6 mM, which is within the range where a small change in ATP concentration can dynamically alter PFK activity.

PFKFB3 supports hematopoiesis in contexts that require robust HSPC proliferation in vitro and in vivo. We showed that the positive or negative effect of Pfkfb3 overexpression or KO on differentiated blood cell production is gradually lost after BMT. This is because HSPCs require PFKFB3 for cell cycle progression during stress hematopoiesis in the early phase after BMT (Figure 7F-J, and L-N). However, HSCs no longer require PFKFB3 for a certain period of time after BMT, probably because they regain a quiescent state. This is consistent with the fact that inhibition of PFKFB3 in quiescent HSCs does not reduce the ATP concentration (Figure 5F, H), suggesting that the activity of PFKFB3 is plastically modified. HSC metabolic plasticity is also illustrated by the mode of PFKFB3 activation, differing depending on stress type. During proliferative stress, PRMT1 methylates PFKFB3 in the HSCs to promote glycolytic ATP production, a modification that increases its activity⁶⁷. PRMT1 is required for stress hematopoiesis⁷⁸, but its downstream targets in HSCs remain unclear. Our results strongly suggest that PRMT1 targets PFKFB3 to stimulate glycolysis in HSCs. In contrast, under OXPHOS inhibition, PFKFB3 phosphorylation by AMPK is induced—another modification that also upregulates its activity. These two PFKFB3 protein modifications allow for flexible regulation of ATP production by glycolysis, even under simultaneous and different stresses. In fact, the constitutively active S461E PFKFB3 mutant, designed to mimic phosphorylation in response to OXPHOS inhibition, enhanced HSC reconstitution capacity after transplantation, suggesting that even if PFKFB3 is activated by one stress (in this case, proliferative), it has the activation capacity to respond to a different stress (i.e., mitochondrial). Therefore, the functions of phosphorylated and methylated forms of PFKFB3 are to some extent interchangeable, and either modification can be used to handle diverse stresses.

In summary, we found that HSCs exhibit a highly dynamic range of glycolytic flux. Our study highlights glycolysis as a pivotal source of energy production in stressed HSCs, and indicates that OXPHOS, although an important source of ATP, can be uncoupled from glycolysis in steady-state HSCs without compromising ATP levels. Because multiple PFKFB3 modifications safeguard HSCs against different stresses by accelerating glycolysis, interventions targeting these might effectively induce or manage stress hematopoiesis. This study provides a platform for comprehensive and quantitative real-time analysis of ATP concentration and its dynamics in HSPCs. Our approach allows for analysis of metabolic programs in rare cells and detection of various metabolic activities within a diverse cell population, making it applicable to the analysis of various tissue systems in normal and diseased states.

Acknowledgements

We thank E. Lamar for preparation of the manuscript, T. Kitamura for providing mVenus-p27K-mice, and N. Toyama-Sorimachi and H. Shindou for their critical reading of the manuscript. This work was supported in part by KAKENHI grants from MEXT/JSPS (JP19K17847 to H.K.; JP19K17877, JP21J01690 to D.K.; JP18H02845, JP20K21621, JP21H02957 to K.T.), AMED grants (JP22zf0127007 to M.S.; JP18ck0106444, JP18ae0201014, JP20bm0704042, JP20gm1210011 to K.T.), grants from the National Center for Global Health and Medicine (29-1015, 20A1010 to H.K.; 26-001, 21A2001 to K.T.), the Takeda Science Foundation (to D.K. and K.T.), a JB Research Grant (to D.K.), the Human Biology Microbiome Quantum Research Center (WPI-Bio2Q) supported by MEXT (to M.S.), and the MEXT Joint Usage/Research Center Program at the Advanced Medical Research Center, Yokohama City University (to K.T.).

Author Contributions

S.W., H.K., Y. Sugiura, M.Y., K.S., Y. Sorimachi, D.K., S.K., M.O., A.N., K.M., M.H., and S.T. performed the study and analyzed data; T. Yamamoto, T. Yabushita, Y.T, H.H., S.O., N.G., T.T., A.N-I., M.S., A.I., T.S., and K.T. provided scientific advice and materials; S.W., H. K., A.N-I. and K.T. wrote the manuscript; and K.T. conceived the project and supervised the research.

Conflict of interest

The authors declare no competing financial interests.