PFKM governs metabolic shifts throughout skeletal muscle differentiation

Abstract

Metabolism is known to influence cell identity, but the underlying mechanisms remain unclear. Here we reveal spatiotemporal dynamics of phosphofructokinase 1 (PFK1), a key glycolytic enzyme, within the skeletal muscle lineage. The expression of PFKM (the muscle isoform of PFK1) is low in muscle stem cells and increases during differentiation. Mechanistically, Wnt signalling rapidly induces lysosomal degradation of PFKM through a methyl arginine degron motif, which gets selectively methylated by the protein arginine methyltransferase (PRMT1) and delivered to lysosomes through microautophagy. PFKM degradation shifts glucose metabolism from glycolysis to the pentose phosphate pathway. PFKM overexpression increases glycolysis and promotes differentiation into terminally differentiated myofibres. On the other hand, PFKM knockdown blunts differentiation, which can be rescued by supplementation with the downstream glycolytic intermediate 3-phosphoglycerate. In sum, our findings highlight the importance of compartmentalized metabolism in cell fate decisions.

Subcellular compartmentalization of the glycolytic enzyme PFKM regulates cell fate and metabolic switch during skeletal muscle differentiation.

Main

Cellular decision-making is linked with metabolite fluctuations that sustain bioenergetic homoeostasis. Recent studies confer a role for metabolic players as signalling molecules during tissue development, remodelling and homoeostasis. Metabolic enzymes, and even metabolites themselves, have been shown to actively control cell identity¹, embryogenesis², adaptive immunity³, mechanosensation⁴ and tumourigenesis^5,6. This emerging framework presents new questions about the underlying regulatory principles of biological systems. The molecular mechanisms underlying metabolite-guided biological regulation remain enigmatic, in part because of limitations in tools that can reliably probe across biological timescales^7,8. How the rapid turnover of metabolites (~seconds) and proteins (~minutes) directly causes downstream effects on transcription (~hours) and tissue behaviour (~days) is poorly defined.

Metabolic plasticity is fundamental for tissue metabolism and involves the precise control of nutrient flow through distinct pathways. For instance, glucose is a paragon metabolite and couples multiple branches of central metabolism⁹. Upon cellular import, glucose can be metabolized into cellular energy in the form of ATP through glycolysis or oxidative phosphorylation pathways. However, glucose and glycolytic intermediates are also precursors for alternative pathways of central metabolism^2,10. Branching from glycolysis into the hexosamine biosynthesis pathway or the pentose phosphate pathway (PPP) is essential for generating anabolic cofactors and reducing power, respectively. Moreover, recent studies have shown that alternating between branching pathways influences cell fate and tissue reprogramming¹¹. However, the subcellular processes that control the fate of nutrients through different pathways are limited.

Eukaryotic cells are highly compartmentalized. Here, we posit that subcellular compartmentation provides a generalized metabolic control mechanism for cell-level and tissue-level remodelling by physically segregating rate-limiting metabolic enzymes from their biochemical reactions. Metabolomics and live-cell imaging identified a rapid spatial mechanism that controls PFK1, which is essential for glycolysis and represents one of the most heavily studied enzymes. Importantly, temporally restricting PFKM switches glucose metabolism from glycolytic or oxidative phosphorylation to the PPP within the skeletal muscle lineage. Given that skeletal muscle is a key determinant of systemic health, these proof-of-concept studies in muscle cell models provide important mechanistic insight into tissue regenerative medicine by delineating an emerging set of molecular cues during myogenesis^12–15. Together, our findings support a model whereby the precise control of metabolic enzymes and compartmentalized metabolism synergize to enact specialized cellular behaviours.

Results

Characterizing rapid metabolite fluctuations by growth factors

Growth factor signalling is intricately linked with cellular metabolism and requires a precise integration of diverse biological timescales through molecular mechanisms that are incompletely understood¹⁶. Canonical Wnt is a universal growth pathway that is essential for embryogenesis, stem cell remodelling and tissue regeneration^17,18. Prior work has shown that Wnt signalling through β-catenin leads to metabolic reprogramming at the transcriptional level^17–19. However, whether Wnt stimulation and the rapid remodelling of intracellular landscapes influence metabolite levels before transcription activity is less defined. Given this uncertainty, we selected the canonical Wnt pathway as an initial paradigm. Cultured cells were treated with control buffer or Wnt3a, an endogenous ligand that binds to cognate Wnt receptors frizzled and low-density lipoprotein receptor-related protein 6 (LRP6) and is commonly used to stimulate Wnt signalling in cultured cell models^17,18. Cells were treated for 20–120 min, and metabolites were quantified by metabolomic liquid chromatography–mass spectrometry (LC–MS). Wnt-treated cells had increased levels of metabolites from upper glycolysis, including glucose, hexose-phosphate, glucose-6-phosphate (G6P) and fructose-6-phosphate (F6P) (Fig. 1a,b and Extended Data Fig. 1a)^19,20. Unexpectedly, the levels of fructose-1,6-bisphosphate (FBP) were reduced at these Wnt timepoints (Fig. 1b), which hinted at a possible upstream metabolic shunting process. In line with this reasoning, metabolites from the PPP branch of metabolism were increased, such as ribose-5-phosphate (R5P) and sedoheptulose-7-phosphate (S7P). In addition to macromolecule biosynthetic metabolites, PPP metabolism also generates critical metabolites for oxidation–reduction (redox) homoeostasis. Indeed, a notable increase was found in metabolite levels of glutathione disulfide (Fig. 1b). These data show that rapid cellular mechanisms control fluctuations between glycolysis and branching metabolism through the PPP.

Fig. 1. PFKM is degraded in lysosomes during Wnt signalling.

a, Schematic of metabolic pathways branching from glycolysis: G6P, F6P, FBP, pyruvate (Pyr), lactate (Lac), R5P, S7P, glutathione (GSH), glutathione disulfide (GSSG), glucosamine (GlcN), N-acetylglucosamine (GlcNAc) and N-acetylglucosamine-6-phosphate (GlcNAc-6P). b, Relative abundance of glucose-derived metabolites following treatment with control buffer or Wnt3a treatment (2 h) in HeLa cells by LC–MS (n = 3 biological replicates). Red, increase; blue, decrease. c, Representative immunofluorescence (IF) of GFP–PFKM (green), LAMP1 (pink) and DAPI (blue) in HeLa cells after a 20 min treatment with Wnt3a or control buffer. Scale bars, 5 μm; inset scale bar, 2.5 μm (n = 156 control cells and n = 137 treated cells from five fields of view). Right, colocalization analysis of PFKM and LAMP1 normalized by total cell count (×20 magnification). Statistical significance was determined by unpaired two-tailed Student’s t-test, and data are shown as mean ± s.e.m. d, Immunoblotting (IB) of PFKM, PFKL and PFKP in HeLa cells after control buffer or Wnt3a treatment for 15 min and 5 h. Right, protein levels normalized to loading control (Actin), assessed by densitometry of n = 3 biological replicates. Statistical significance was determined by one-way ANOVA with post hoc Dunnett’s analysis, and data are shown as mean ± s.e.m. (NS, P > 0.05). FC, fold change. e, IB of PFKM, PFKL and PFKP in HEK293 cells treated with a Wnt3a treatment time course with 10 μM cycloheximide (CHX). Right, protein levels normalized to loading control (H3), assessed by densitometry of n = 3 biological replicates. Statistical significance was determined by one-way ANOVA with post hoc Dunnett’s analysis, and data are shown as mean ± s.e.m. (NS, P > 0.05). f, IB of PFKM in HEK293 cells after control buffer or Wnt3a treatment for 20 min following a bafilomycin pretreatment (12 h). Right, protein levels normalized to loading control (Actin), assessed by densitometry of n = 3 biological replicates. Statistical significance was determined by two-way ANOVA with post hoc Dunnett’s analysis, and data are shown as mean ± s.e.m. (NS, P > 0.05). Exact P values are shown in graphs.

Source data

Extended Data Fig. 1. Wnt-triggered protein degradation is specific to the PFKM isoform.

a, Left, relative abundance of glucose and hexose-phosphate metabolites in HeLa cells after a 20 min treatment with Wnt3a or control buffer by liquid-chromatography mass spectrometry (LC-MS), (n = 2 biological replicates). Statistical significance was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. Right, MetaboAnalyst identification of impacted pathways with Wnt3a treatment, associated with Fig. 1b. b, Representative immunofluorescence (IF) of GFP-PFKM (green), LAMP1 (pink), and DAPI (blue) in HeLa cells treated with control buffer or Wnt3a (20 min) followed by digitonin permeabilization. Scale bars: 5 μm, inset: 2.5 μm (n = 51 control cells and n = 44 treated cells from 5 fields of view). Right, colocalization analysis of PFKM and LAMP1 normalized by total cell count (×20 magnification). Statistical significance was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. c, Immunoblotting (IB) of endogenous PFKM and GFP-PFKM in HEK293 cells treated with control buffer or Wnt3a (20 min). Right, protein levels normalized to loading control (Actin), assessed by densitometry of n = 3 biological replicates. Statistical significance was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. d, IB of PFKM, PFKL and PFKP in HEK293 cells following CA-LRP6 overexpression. Right, protein levels normalized to loading control (Actin), assessed by densitometry of n = 3 biological replicates. Statistical significance was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. (ns P>0.05). e, IB of PFKM, PFKL and PFKP in HEK293 cells and Hela cells after a cycloheximide time course. Right, protein levels normalized to loading control (Actin), assessed by densitometry of n = 3 biological replicates. Statistical significance was determined by one-way ANOVA with post-hoc Dunnett’s analysis and data are shown as mean ± s.e.m. (ns P>0.05). f, Representative IF of PFKM, PFKL, or PFKP (green), LAMP1, and DAPI in HeLa cells treated with control buffer or Wnt3a (20 min). Scale bars: 5 μm, inset: 2.5 μm (PFKM: n = 30 control cells and n = 57 treated cells from 10 fields of view. PFKL: n = 43 control cells and n = 35 treated cells from 5 fields of view. PFKP: n = 38 control cells and n = 37 treated cells from 5 fields of view). Right, colocalization analysis of PFKM, PFKL or PFKP and LAMP1 normalized by total cell count (×20 magnification). Statistical significance was determined by multiple unpaired t test analyses with Šidák-Bonferroni correction (α = 0.05) and data are shown as mean ± s.e.m. (ns P>0.05). g, Representative IF of PFKM, LAMP1, and DAPI in HeLa cells treated with control buffer or Wnt3a (20 min) treatment following a bafilomycin pretreatment (1 h). Scale bars: 5 μm, inset: 2.5 μm (n = 170 control cells and n = 161 treated cells from 5 fields of view). Right, colocalization analysis of PFKM and LAMP1 normalized by total cell count (×20 magnification). Statistical significance was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. h, IB of PFKM in HEK293 cells treated with control buffer or Wnt3a (20 min), with or without the presence of proteasome inhibitor MG132 (1 h). Right, protein levels normalized to loading control (Actin), assessed by densitometry of n = 3 biological replicates. Statistical significance was determined by two-way ANOVA with post-hoc Dunnett’s analysis and data are shown as mean ± s.e.m. (ns P>0.05). Exact P values are shown in graphs.

Source data

Wnt rapidly sequesters cytosolic PFK1 into lysosomes

Spatial and temporal control of protein distributions occur on the minute timescale, which aligns with the observed dynamics of metabolite levels in our analyses. PFK1 converts F6P into FBP²¹. We found an increase in glycolytic metabolites upstream of FBP. The role of Wnt signalling in PFK1 activity is unknown. Given the significant reduction of FBP after 2 h, we performed immunofluorescence microscopy of PFK1 localization after Wnt3a treatment for 20 min. In control cells, PFK1 was diffusely localized throughout the cytosol (Fig. 1c). By contrast, Wnt treatment rapidly shifted cytosolic PFK1 into punctate structures. Previous work demonstrated that PFK1 forms into phase condensates in response to nutrient levels^22–24. However, co-staining revealed that PFK1 was significantly localized into membrane-bounded vesicles marked by lysosomal-associated membrane protein 1 (LAMP1). Wnt-stimulated cells retained PFKM colocalization with LAMP1 in the presence of digitonin, which disrupts cholesterol patches in the plasma membrane while leaving intracellular organelles intact (Extended Data Fig. 1b). These data show that cellular PFK1 distributions are confined to lysosomes by Wnt pathway activation before the accumulation of upper glycolytic metabolites.

Lysosomes govern isoform-specific degradation of PFKM (muscle)

Given that Wnt led to PFK1 relocalization into lysosomes, we next explored whether protein levels were impacted. PFK1 exists in three isoforms that are universally expressed across most cell types and historically named based on the tissue of their discovery: PFKL (liver), PFKP (platelet) and PFKM (muscle)²⁵. Protein levels of PFKL, PFKP and PFKM were evaluated during Wnt ligand treatments by immunoblotting. A striking reduction of PFKM protein levels was induced by Wnt treatment (15 min) relative to control cells (Fig. 1d). Wnt-induced proteolysis was further confirmed by using exogenous PFKM–GFP-expressing HEK293 cells (Extended Data Fig. 1c). PFKM protein levels were also reduced in cells expressing a constitutively active LRP6 (CA-LRP6) (Extended Data Fig. 1d), which is known to induce a potent Wnt signal²⁶. By contrast, PFKL and PFKP remained comparable amongst control or Wnt-treated cells (Fig. 1d). To evaluate whether the observed protein level changes reflected alternate baseline turnover, Wnt treatments were next performed with cycloheximide to inhibit protein synthesis. Control immunoblotting confirmed that baseline turnover rates of all three PFK1 isoforms were unchanged over 2 h (Extended Data Fig. 1e). Wnt treatments with cycloheximide similarly caused degradation of PFKM with no impact on PFKL or PFKP (Fig. 1e). Wnt had no effect on PFKP and PFKL distributions, further confirming PFKM selectivity (Extended Data Fig. 1f). Previous work has shown that PFKL and PFKP undergo ubiquitin-mediated degradation in the proteasome⁴, while the proteolytic pathway of PFKM is less defined. Co-treatment with a lysosomal inhibitor, bafilomycin, was sufficient to restore PFKM protein levels in the presence of Wnt ligands (20 min), which caused PFKM accumulation in lysosomes (Fig. 1f and Extended Data Fig. 1g). Co-treatment with the proteasomal inhibitor MG132 had no effect on PFKM protein levels (Extended Data Fig. 1h). These data indicate that lysosomal proteolysis exerts isoform-selective regulation of PFKM.

Arginine methylation drives PFKM proteolysis

Having shown that a unique proteolytic process is reserved for PFKM, we next addressed how this isoform specificity is conferred. Prior work has shown that PFK1 isoforms have distinct regulation and post-translational modifications, despite their high sequence homology (∼70%)^21,25,27. To assess PFKM-specific differences, we performed sequence analyses and identified a unique lysosomal degron motif within PFKM (Fig. 2a). Methyl arginine degron (MrDegron) is an emergent signal for lysosomal proteolysis^28–30. In the MrDegron pathway, a substrate is modified by asymmetric dimethylarginine (ADMA) modifications that initiate lysosomal delivery by microautophagy. As seen with other MrDegron substrates, the motif in PFKM was embedded within an intrinsically disordered region and contained a key arginine residue^28,29. Aligning with a selective lysosomal pathway for PFKM, MrDegron motifs were absent from PFKP and PFKL isoforms (Fig. 2a). Given this finding, we next examined whether PFKM is subjected to lysosomal proteolysis by MrDegron. Protein arginine methyltransferase 1 (PRMT1) catalyses type I arginine methylation modification of MrDegron substrates. Molecular docking with AlphaFold3 mapped the most favourable interactions to the catalytic domain of PRMT1 and the carboxyl terminus of PFKM, which contains the MrDegron motif (Fig. 2b and Extended Data Fig. 2a). Consistent with molecular docking, PRMT1 co-immunoprecipitated with PFKM purified from HEK293 cells (Fig. 2c). Immunofluorescence analyses showed that PFKM colocalized with PRMT1 upon Wnt stimulation (Extended Data Fig. 2b). Mass spectrometry was performed to assess methylation of PFKM purified from HEK293 cells. Further supporting an isoform-specific regulatory pathway for PFKM, arginine 774 was modified by ADMA within the MrDegron motif of PFKM that is absent in both other PFK isoforms (Fig. 2a,d and Extended Data Fig. 2c). Immunoblotting showed that an ADMA antibody recognized PFKM and that increasing Wnt activity decreased signal for PFKM and ADMA (Extended Data Fig. 2d). Immunofluorescence analyses showed that PFKM colocalization with the ADMA antibody was increased during Wnt signalling (Extended Data Fig. 2e), while lysosomal delivery was decreased with a PFKM arginine-to-lysine mutant (R774K (PFKM-RK)) relative to wild-type PFKM (PFKM-WT) (Extended Data Fig. 2f). We next examined the requirements of PRMT1 in this pathway. PFKM protein levels were assessed in the presence of Wnt3a co-treated with a type I PRMT inhibitor (MS023). Immunoblotting analyses show that PRMT inhibition was sufficient to restore PFKM levels during Wnt treatments (Extended Data Fig. 2g, lanes 2 and 4). To further establish this finding, cells were treated with short interfering RNA (siRNA) targeting PRMT1 or siControl RNA. PFKM protein levels were no longer reduced by Wnt signalling in PRMT1-depleted cells (Fig. 2e). These data show that PRMT1 catalyses arginine methylation of PFKM during lysosomal delivery.

Fig. 2. PFKM methylation triggers lysosomal delivery via microautophagy.

a, Schematic aligning PFKM, PFKL and PFKP C-terminal tails. The MrDegron (labelled in pink) lysosomal degradation motif is located within the intrinsically disordered region (IDR) of the PFKM C terminus. b, AlphaFold3 model of PRMT1 in complex with PFKM. PFKM is shown in green, PRMT1 is blue and the C-terminal region of PFKM (amino acids 770–780) is yellow. Inset highlights predicted associations of PFKM C terminus with the catalytically active domain of PRMT1. c, IB analysis of SNAP-PFKM and Halo-PRMT1 co-IP in stable WT-PRMT1-expressing HEK293 cells. OE, overexpression. Right, protein levels normalized to input (Actin), assessed by densitometry of n = 3 biological replicates. Statistical significance was determined by unpaired two-tailed Student’s t-test, and data are shown as mean ± s.e.m. d, MS spectrum showing identification of SNAP-PFKM peptide KRSGEAAV exhibiting a dimethyl modification at arginine residue R774 by LC–MS/MS. The SNAP-PFKM protein was immunoprecipitated from HEK293 cells after a bafilomycin pretreatment (4 h) and a 20 min Wnt3a treatment. Peptide was identified in a +2 charge state, shown by isotopic peak spacing of ~0.5 m/z units. e, IB of PFKM and PRMT1 in HEK293 cells with siControl or siPRMT1 and control buffer or 20 min Wnt3a treatment. Right, protein levels normalized to loading control (Actin), assessed by densitometry of n = 3 biological replicates. Statistical significance was determined by two-way ANOVA with post hoc Dunnett’s analysis, and data are shown as mean ± s.e.m. (NS, P > 0.5). f, Representative IF of GFP–PFKM (green), VPS4 (red), LAMP1 (pink) and DAPI (blue) in HeLa cells with control buffer or LLOMe treatment. Scale bar, 5 μm; inset scale bar, 2.5 μm (n = 358 control cells and n = 177 treated cells from five fields of view). Bottom right, colocalization analysis of PFKM, VPS4 and LAMP1 normalized by total cell count (×20 magnification). Statistical significance was determined by unpaired two-tailed Student’s t-test, and data are shown as mean ± s.e.m. Top right, schematic of methylated PFKM delivery into lysosomes by microautophagy. g, IB of PFKM and VPS4 in HEK293 cells under siControl or siVPS4A conditions with control buffer or 20 min Wnt3a treatment. Right, protein levels normalized to loading control (Actin), assessed by densitometry of n = 3 biological replicates. Statistical significance was determined by two-way ANOVA with post hoc Dunnett’s analysis, and data are shown as mean ± s.e.m. (NS, P > 0.5). Arrow denotes VPS4A isoform (bottom band). Exact P  values are shown on the graphs.

Source data

Extended Data Fig. 2. Arginine methylation tags PFKM for degradation.

a, PFKM (green) modeling of electrostatic interactions with protein arginine methyltransferase 1 (PRMT1, purple) with PFKM C terminus (yellow). Green circle indicates PRMT1 interaction site on PFKM. Color scale bar denotes −10 to 10 kT/e. b, Representative immunofluorescence (IF) of PFKM, PRMT1 (red), and DAPI (blue) in HeLa cells treated with control buffer or Wnt3a (20 min). Scale bars: 10 μm, inset: 2.5 μm (n = 106 control cells and n = 118 treated cells from 5 fields of view). Right, colocalization analysis of PFKM and PRMT1 normalized by total cell count (×20 magnification). Statistical significance was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. c, Left, chromatogram showing PFKM peptide KRSGEAAV abundance as area under the curve. Right, MS spectrum with identified b ions (blue) and y ions (red) for PFKM peptide KRSGEAAV. Gray represents peaks that were not assigned. MS spectrums associated with Fig. 2d. d, Immunoblotting (IB) of SNAP-PFKM and asymmetric dimethylarginine (ADMA) in HEK293 cells during increasing Wnt3a treatment. Right, protein levels normalized to loading control (Actin), assessed by densitometry of n = 3 biological replicates. Statistical significance was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. e, Representative IF of PFKM, ADMA (red), lysosomal-associated membrane protein 1 (LAMP1, pink), and DAPI in HeLa cells treated with control buffer or Wnt3a (20 min). Scale bars: 5 μm, inset: 2.5 μm (n = 74 control cells and n = 79 treated cells from 5 fields of view). Right, colocalization analysis of PFKM, ADMA and LAMP1 normalized by total cell count (×20 magnification). Statistical significance was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. f, Representative IF of wild-type PFKM (PFKM-WT) or methyl mutant R774K (PFKM-RK), LAMP1 and DAPI in HeLa cells treated with control buffer or Wnt3a (20 min). Scale bars: 5 μm (n = 207 PFKM-WT cells, n = 306 PFKM-WT and Wnt treated cells, n = 303 PFKM-RK cells, and n = 210 PFKM-RK and Wnt treated cells from 7 fields of view). Right, colocalization analysis of PFKM and LAMP1 normalized by total cell count (×20 magnification). Statistical significance was determined by two-way ANOVA with post-hoc Bonferroni’s analysis and data are shown as mean ± s.e.m. g, IB of PFKM and PRMT1 in HEK293 cells treated with control buffer or Wnt3a (20 min), with or without MS023 treatment. Right, protein levels normalized to loading control (Actin), assessed by densitometry of n = 3 biological replicates. Statistical significance was determined by two-way ANOVA with post-hoc Dunnett’s analysis and data are shown as mean ± s.e.m. (ns P>0.05). Exact P values are shown in graphs.

Source data

Microautophagy enables PFKM lysosomal delivery

Proteins are delivered into lysosomes by (macro)autophagy, chaperone-mediated autophagy and microautophagy^31–33. Upon methylation, MrDegron motifs are recognized by microautophagy machinery that enables lysosomal delivery through endosomal sorting complexes required for transport (ESCRT) proteins^28,29,31. As our data support a model that PFKM undergoes MrDegron-driven lysosomal degradation, we next examined the role of microautophagy. To gain initial insight, we performed an established microautophagy assay using L-leucyl-L-leucine methyl ester hydrobromide (LLOMe), a lysosome-damaging agent that leads to the recruitment of microautophagy regulatory proteins³¹. Vacuolar protein sorting-associated protein 4 (VPS4) is an ESCRT family member that is crucial for microautophagy, as it enables the final stage of membrane scission³¹. Compared to controls, LLOMe treatment led to the colocalization of PFKM with LAMP1 and VPS4 in triple-stained cells (Fig. 2f). Live-cell imaging tracking showed that the delivery of PFKM–GFP into lysosomes, marked by Lysotracker dye, occurs within minutes (Extended Data Fig. 3a and Supplementary Video 1), at a timescale consistent with microautophagy of methylated proteins^28–30. Biochemical analyses show that PFKM co-immunoprecipitates with VPS4 purified from HEK293 cells (Extended Data Fig. 3b). To examine whether microautophagy drives proteolysis, PFKM protein levels were evaluated in cells treated with siRNA targeting VPS4 or control RNA. Immunoblotting showed that PFKM protein levels were stabilized in VPS4-depleted cells during Wnt stimulation (Fig. 2g). To disrupt microautophagy activity, PFKM was next evaluated in cells expressing a dominant-negative VPS4–GFP mutant (VPS4-DN)²⁶. Immunofluorescence analyses showed that PFKM puncta were ablated in VPS4-DN-expressing cells compared to wild-type VPS4 (VPS4-WT) cells (Extended Data Fig. 3c). These data implicate microautophagy in PFKM lysosomal delivery, which further supports that PFKM is regulated through canonical MrDegron machinery.

Extended Data Fig. 3. PFKM is delivered into lysosomes via microautophagy.

a, Representative stills from live cell imaging of GFP-PFKM (green) and lysotracker (pink) in HeLa cells following Wnt3a stimulation for 40 min. Nucleus outlined by blue dashed line. Stills associated with Supplementary Video 1. Scale bars: 5 μm (n = 91 cells at 0 min, n = 94 cells at 3 min, n = 94 cells at 20 min, n = 95 cells at 25 min, and n = 95 cells at 40 min from 3 fields of view). Right, colocalization analysis of PFKM and lysotracker normalized by total cell count (×20 magnification). Statistical significance was determined by one-way ANOVA with post-hoc Dunnett’s analysis and data are shown as mean ± s.e.m. (ns P>0.05). b, Left, immunoblotting (IB) of co-immunoprecipitation of SNAP-PFKM and GFP-VPS4 in HEK293 cells. Middle, protein levels normalized to input (Actin), assessed by densitometry of n = 3 biological replicates. Statistical significance was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. Right, AlphaFold3 model of human VPS4 (pink) interacting with PFKM (green). c, Representative immunofluorescence (IF) of PFKM (green), vacuolar protein sorting-associated protein 4 (VPS4, red) of wild-type VPS4 (VPS4-WT) or dominant negative mutant VPS4-DN (VPS4-DN) and DAPI (blue) in HeLa cells. Scale bars: 5 μm, inset: 2.5 μm (n = 142 VPS4-WT cells and n = 150 VPS4-DN cells from 7 fields of view). Right, colocalization analysis of PFKM and VPS4 normalized by total cell count (×20 magnification). Statistical significance was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. Exact P values are shown in graphs.

Source data

Dynamic fluctuations of PFKM levels align with muscle stem cell trajectories

After delineating an isoform-specific proteolytic pathway for PFKM, we next explored the functional impact on human physiology³⁴. Classical studies in rabbits show that PFKM is the sole isoform in skeletal muscle²¹. To assess whether PFKM expression profiles are conserved, we probed a human tissue single-cell transcriptomic database of 500,000 cells³⁵. Across 24 tissues and organs, PFKM levels were highest in muscle cell populations (Extended Data Fig. 4a). To gain further resolution, we next leveraged a human muscle database of single-cell and single-nucleus data from tissues collected from over 40 major human skeletal muscle cell populations¹³. PFKM expression levels were very low in the satellite muscle stem cell (MuSC) cluster, marked by high expression of paired box 7 (PAX7) (Fig. 3a–c and Extended Data Fig. 4b,c). By contrast, expression of PFKM increased significantly in clusters of differentiating myofibres, aligned with high expression of the core myogenic differentiation target genes myoblast determination protein (MYOD), myogenin (MYOG) and muscle myosin heavy chain 1 (MYH1) (Fig. 3c and Extended Data Fig. 4c). To further determine the dynamics of gene expression during differentiation of stem cells into myofibres, we generated pseudotime trajectories for the cumulative stem cell population within the atlas³⁶ using the R package monocle 2. We focused on differentially expressed genes across three groups of myogenic cells: MuSCs, slow-twitch (type I) myofibres and fast-twitch (type II) myofibres. We also performed pseudotime trajectory analysis on the above cells, which predicted a branched progression of cell states from MuSCs to maturing myofibres¹³. The beginning and ending branches of the pseudotime trajectory were composed of PAX7⁺ MuSCs and ACTA1⁺ terminally differentiating myofibres, respectively. Along this trajectory, the onset of PFKM expression was found in myogenic cells negative for PAX7 at the branching point, and at the highest levels in the mature cell population (Extended Data Fig. 4d). These data suggest that a dynamic range of PFKM expression levels accompanies progressively differentiating cell states within the skeletal muscle lineage. The differentiation-specific expression of PFKM prompted further investigation using a cultured quadricep-derived human muscle cell line³⁷. Muscle cells were evaluated over a 12 day time course of differentiation. Muscle cells displayed the classic myogenic cascade of increasing myogenin, titin (TTN), MYH8, MYH3 and the reduction of CDK1 and ID3 levels (Extended Data Fig. 4e)^37,38. PFKM transcript levels doubled within the first 24 h, while PFKM protein was increased after 72 h, as shown by immunoblotting analyses (Fig. 3d–f). Differentiation was further confirmed by rising levels of myosin heavy chain (MyHC) protein levels (Fig. 3e,f). PFKM protein levels increased during differentiation of a cultured mouse muscle cell line (C2C12), as assessed by immunoblotting (Extended Data Fig. 4f). Therefore, PFKM levels increase during differentiation of cultured muscle cell models.

Extended Data Fig. 4. PFKM levels increase in differentiated muscle populations.

a, PFKM single cell RNA-seq from tissue-wide human atlas across cell-types represented as a UMAP from 24 tissues (500,000 cells)³⁵. b, Single-cell and single-nucleus RNA-seq data from the human skeletal muscle atlas represented as a UMAP from 40 major muscle populations¹³. c, RNA-seq data from the human skeletal muscle atlas¹³ represented as a dot plot. Relative expression levels of PFKM and myogenic target genes across muscle stem cells (MuSC), slow-twitch myofibres (type I) and fast-twitch myofibres (type II). Paired box 7 (PAX7), myogenin (MYOG), myoblast determination protein (MYOD), myosin heavy chain 1 (MYH1). d, Pseudotime trajectory analysis of combined muscle stem cells (MuSC), slow-twitch myofibres and fast-twitch myofibres for PFKM, muscle stem cell marker PAX7 and myofibre marker actin α1 (ACTA1). Each dot represents a single cell, with the color gradient indicating gene expression level. Cells are colored by cell cluster, pseudotime, and pseudotime branch state. e, mRNA expression of myogenic target genes from bulk RNA-seq³⁷ in cultured human muscle cell line during a 12 day differentiation time course. Cyclin-dependent kinase 1 (CDK1), DNA binding protein inhibitor 3 (ID3), myogenin (MYOG), myosin heavy chain 3 (MYH3), myosin heavy chain 8 (MYH8), and titin (TTN), (n = 2 biological replicates). Statistical significance was determined by one-way ANOVA with post-hoc Dunnett’s analysis and data are shown as mean ± s.e.m. (CDK1: P=0.0009, P=0.0007, P=0.0002, P=0.0005, P=0.0010, P=0.0007. ID3: all values P<0.0001. MYOG: P<0.0001, P<0.0001, P<0.0001, P<0.0001, P=0.0004, P=0.0002. MYH3: P=0.0425. MYH8: P=0.0014, P=0.0013, P=0.0130, P=0.0074. TTN: P=0.0017, P<0.0001, P=0.0001, P<0.0001, P<0.0001. All ns P>0.05). f, Immunoblotting (IB) of PFKM in C2C12 muscle cells during a 3 day differentiation time course. Right, protein levels normalized to loading control (Actin), assessed by densitometry of n = 3 biological replicates. Statistical significance was determined by one-way ANOVA with post-hoc Dunnett’s analysis and data are shown as mean ± s.e.m. Exact P values are shown in graphs.

Source data

Fig. 3. PFKM is degraded in lysosomes in human muscle cells.

a, UMAP of single-cell (scRNA-seq) and single-nucleus (snRNA-seq) data from the human skeletal muscle atlas¹³, coloured by cell type. UMAP, uniform manifold approximation and projection. b, PFKM mRNA expression from the human skeletal muscle atlas¹³. PFKM expression is highest in differentiated myofibres, including type I slow-twitch myofibres and type II fast-twitch myofibres. PFKM expression is low in MuSCs. c, Track plots of differentiation and Wnt signalling genes in MuSC, myofibre type I and myofibre type II cell clusters across the human skeletal muscle atlas¹³. Within each track plot, individual cells are ranked from lowest expression (left) to highest expression (right), with bar height indicating relative expression per cell. Tracks are colour-coded as in a. Differentiation genes: paired box 7 (PAX7), myogenin (MYOG), myoblast determination protein (MYOD), myosin heavy chain 1 (MYH1). Wnt signalling genes: cMYC, glycogen synthase kinase-3β (GSK3B), beta catenin 1 (CTNNB1). d, PFKM mRNA expression from bulk RNA-seq³⁷ in cultured human muscle cells during a 12 day differentiation time course (n = 2 biological replicates). TPM, transcripts per million. Statistical significance was determined by one-way ANOVA with post hoc Dunnett’s analysis, and data are shown as mean ± s.e.m. Associated data in Extended Data Fig. 4 include additional target gene markers of differentiation. e,f, IB of PFKM and MyHC in cultured human muscle cells during a 12 day differentiation time course (f). Protein levels normalized to loading control (Actin), assessed by densitometry of n = 3 biological replicates (e). Statistical significance was determined by one-way ANOVA with post hoc Dunnett’s analysis, and data are shown as mean ± s.e.m. (NS, P > 0.5). g, IB of PFKM in cultured human muscle cells after differentiation (3 days), treated with control buffer or a 20 min Wnt3a treatment following a bafilomycin pretreatment (4 h). Right, protein levels normalized to loading control (vinculin), assessed by densitometry of n = 3 biological replicates. Statistical significance was determined by two-way ANOVA with post hoc Dunnett’s analysis, and data are shown as mean ± s.e.m. (NS, P > 0.05). h, Track plots of lysosomal genes in MuSC, myofibre type I and myofibre type II cell clusters across the human skeletal muscle atlas¹³. Within each track plot, gene expression data are presented as described in c. Lysosomal genes: LAMP1, cathepsin D (CTSD), cathepsin B (CTSB), ATPase H⁺-transporting V1 subunit B2 (ATP6V1B2), mucolipin 1 (MCOLN1), solute carrier family 11 member 1 (SLC11A1), solute carrier family 11 member 2 (SLC11A2), transcription factor EB (TFEB), transcription factor binding to IGHM enhancer 3 (TFE3), charged multivesicular body protein 2A (CHMP2A), charged multivesicular body protein 4B (CHMP4B), vacuolar protein sorting 4 isoform a (VPS4A) and vacuolar protein sorting 4 isoform b (VPS4B). i,j, Representative IF of PFKM (green), LAMP1 (pink) and DAPI (blue) in mononuclear (i) and multinuclear (j) cultured human muscle cells after differentiation (3 days). Cells were treated with control buffer or 20 min Wnt3a treatment. Insets are shown at ×20, ×40 and ×100 magnification. Scale bars, 10 μm at ×20 and ×40; 5 μm at ×100 (n = 109 control mononuclear cells, n = 107 treated mononuclear cells, n = 215 control multinuclear cells and n = 119 treated multinuclear cells from three fields of view). Right, colocalization analysis of PFKM and LAMP1 normalized by total nuclei count (×20 magnification). Statistical significance was determined by unpaired two-tailed Student’s t-test, and data are shown as mean ± s.e.m. (NS, P > 0.05). Exact P values are shown in graphs.

Source data

Compartmentalization precisely controls PFKM in myogenic cells

These analyses show that PFKM levels are dynamically controlled in muscle cells and tissues. Yet the route of PFKM proteolysis in muscle has yet to be examined. Given this knowledge gap, we conducted a series of assays to investigate the lysosomal proteolytic pathway in cultured human muscle cells. First, PFKM protein levels were evaluated in control or Wnt3a-treated conditions. Immunoblotting analyses showed that PFKM is rapidly reduced by Wnt3a treatment (20 min), as seen with non-muscle cells (Fig. 3g). In support of a lysosomal proteolytic pathway, co-treatment with bafilomycin restored PFKM to control cell levels during Wnt stimulation (Fig. 3g). To further establish this result, lysosomal delivery was assessed by live-cell imaging of PFKM–GFP-expressing muscle cells. As in HeLa cells, Wnt3a stimulation caused cytosolic PFKM particles to become colocalized with lysosomes marked by Lysotracker (Extended Data Fig. 5a and Supplementary Video 2) or LAMP1 in C2C12 cells (Extended Data Fig. 5b). PFKM was colocalized with PRMT1 and ADMA during Wnt3a treatment compared to controls (Extended Data Fig. 5c,d). These biochemical and image-based analyses show that the proteolytic regulation of PFKM in lysosomes occurs within muscle cells.

Extended Data Fig. 5. Mononuclear muscle populations have increased propensity for MrDegron-driven PFKM degradation.

a, Representative stills from live cell imaging of GFP-PFKM (green), lysotracker (pink) and Hoechst (blue) in undifferentiated cultured human muscle cells following Wnt3a stimulation for 20 min. Stills associated with Supplementary Video 2. Scale bars: 5 μm, inset: 2.5 μm (n = 499 control cells and n = 492 treated cells from 6 fields of view). Right, colocalization analysis of PFKM and lysotracker normalized by total cell count (×20 magnification). Statistical significance was determined by was determined by paired two-tailed Student’s t-test and data are shown as mean ± s.e.m. b, Representative immunofluorescence (IF) of PFKM, lysosomal-associated membrane protein 1 (LAMP1, pink), and DAPI in undifferentiated C2C12 muscle cells treated with control buffer or Wnt3a (20 min). Scale bars: 5 μm, inset: 2.5 μm (n = 285 control cells and n = 266 treated cells from 6 fields of view). Right, colocalization analysis of PFKM and LAMP1 normalized by total cell count (×20 magnification). Statistical significance was determined by was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. c, Representative IF of PFKM, protein arginine methyltransferase 1 (PRMT1, red), and DAPI in mononuclear cultured human muscle cells after differentiation (3 days). Cells were treated with control buffer or Wnt3a (20 min). Scale bars: 5 μm, inset: 2.5 μm (n = 152 control nuclei and n = 122 treated nuclei from 4 fields of view). Right, colocalization analysis of PFKM and PRMT1 normalized by total nuclei count (×20 magnification). Statistical significance was determined by was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. d, Representative IF of PFKM, asymmetric dimethylarginine (ADMA, red), and DAPI in mononuclear cultured human muscle cells after differentiation. Cells were treated with control buffer or Wnt3a (20 min). Scale bars: 5 μm, inset: 2.5 μm (n = 213 control nuclei and n = 245 treated nuclei from 4 fields of view). Right, colocalization analysis of PFKM and ADMA normalized by cell total nuclei (×20 magnification). Statistical significance was determined by was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. e, Pseudotime trajectory analysis of combined muscle stem cells (MuSC), slow-twitch myofibre and fast-twitch myofibre cells for Wnt signaling genes cMYC and glycogen synthase kinase-3β (GSK3B). Each dot represents a single cell, with the color gradient indicating gene expression level. f, Left, percentage of mononuclear and multinuclear cultured human muscle cells after differentiation in control and Wnt3a treated cells (n = 307 control mononuclear cells, n = 233 control multinuclear cells, n = 381 treated mononuclear cells, and n = 482 treated multinuclear cells from 3 fields of view). Right, quantification of PFKM mean fluorescence intensity (MFI) in mononuclear and multinuclear human muscle cells after differentiation normalized by nuclei count (n = 222 mononuclear cells and n = 275 multinuclear cells from 6 fields of view). Statistical significance was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. Graphs associated with Fig. 3i,j. Exact P values are shown in graphs.

Source data

The differentiation-dependent increase of PFKM prompted us to re-analyse expression patterns of signalling pathways previously implicated in myogenic lineage regulation. Prior work showed that precise temporal control of canonical Wnt signalling occurs during skeletal muscle lineage differentiation. MuSCs displayed the highest levels of the core Wnt target genes cMYC and β-catenin (CTNNB1), consistent with previous studies^39,40. The Wnt pathway destruction complex comprises Axin, adenomatous polyposis coli (APC), casein kinase I (CKI) and glycogen synthase kinase-3β (GSK3B) and continuously degrades β-catenin to reduce Wnt signalling. The levels of GSK3B were the highest within differentiating cell populations (Fig. 3c). Several hallmark lysosomal genes further showed enrichment within MuSCs, which displayed the highest levels of microautophagic machinery genes in addition to lysosomal membrane proteins (Fig. 3h). These data indicate that PFKM inversely correlates with key constituents in the MrDegron lysosomal-proteolytic axis across different muscle cell populations. To gain additional molecular insight, we next evaluated the lysosomal delivery of PFKM in early-stage mononucleated cells and in multinucleated cells. Wnt3a treatment led to the rapid delivery of PFKM into lysosomes of mononuclear cells (Fig. 3i). Conversely, no significant changes were observed with PFKM localization within multinucleated cells, as PFKM remained diffuse throughout the cytosol in control and Wnt3a conditions (Fig. 3j). Consistent with an increase during differentiation, multinucleated cells also displayed higher levels of PFKM relative to mononuclear cells at an earlier stage of differentiation (Extended Data Fig. 5f). These data support a role for the PFKM–lysosomal axis in muscle and suggest that proteolysis is dominant within early stages of differentiation.

PFKM regulates processes during muscle cell differentiation

The extensive remodelling of PFKM levels in muscle cells and tissues hinted at a potential role for PFKM during differentiation. Whether PFKM regulates muscle differentiation is unknown. To explore this possibility, we assessed cultured muscle cells upon PFKM depletion using siRNA targeting the 3′ untranslated region of PFKM. Immunoblotting confirmed the efficiency of PFKM knockdown (PFKM-KD) in muscle cells over differentiation (Fig. 4a and Extended Data Fig. 6a,b). Given the onset of PFKM and myogenic target genes at 24 h (Fig. 3d and Extended Data Fig. 4e), we evaluated the effects of PFKM depletion at the downstream timepoint of 72 h differentiation. Transcript levels of myogenic regulatory factor 5 (MYF5) and MYH1 were decreased in PFKM-KD cells compared to siControl conditions (Fig. 4b). Immunoblotting showed that MyHC protein levels were reduced in PFKM-KD cells (Fig. 4c). To further characterize differences in muscle differentiation, image-based analyses were performed to evaluate myogenin⁺ nuclei and myofibre maturation. In control cells, myogenin⁺ nuclei were observed for total cell populations (Fig. 4d,e). By contrast, PFKM-KD cells displayed a significantly reduced number of myogenin⁺ nuclei (Fig. 4d,e). In support of altered differentiation, PFKM-KD cells also displayed lower myosin staining levels relative to control cells (Fig. 4f,g). Similarly, PFKM-KD also reduced myogenin⁺ nuclei and myosin levels in C2C12 cells (Extended Data Fig. 6b–d). Additional quantification evaluated the fraction of multinuclear cells across the entire population and the average number of nuclei per MyHC⁺ multinucleated (myotube) cells⁴¹. PFKM-KD conditions had a reduced number of multinucleated cells in human (Fig. 4f and Extended Data Fig. 6e) and C2C12 cells (Extended Data Fig. 6d,e). Furthermore, PFKM-KD conditions also displayed fewer nuclei per myosin-positive multinuclear cells (myotubes) (Fig. 4f and Extended Data Fig. 6d,f). Total nuclei numbers were consistent between control and PFKM-KD cells (Extended Data Fig. 6g). These data indicate that the loss of PFKM decreases differentiation markers and fusion.

Fig. 4. Depletion of PFKM hinders muscle cell differentiation.

a, IB of PFKM, PFKP and PFKL in cultured human muscle cells after differentiation (3 days) with siControl or siPFKM treatment. Right, protein levels normalized to loading control (Actin), assessed by densitometry of n = 3 biological replicates. Statistical significance was determined by unpaired two-tailed Student’s t-test, and data are shown as mean ± s.e.m. (NS, P > 0.05). b, Quantitative PCR with reverse transcription (RT–qPCR) analysis of the relative transcript levels of myogenic factor 5 (MYF5) and myosin heavy chain 1 (MYH1) in cultured human muscle cells after differentiation (3 days) with siControl or siPFKM treatment (n = 6 biological replicates). Statistical significance was determined by multiple unpaired two-tailed Student’s t-test analyses with Šidák–Bonferroni correction (α = 0.05), and data are shown as mean ± s.e.m. c, IB of PFKM and MyHC in cultured human muscle cells after differentiation (3 days) with siControl or siPFKM treatment. Right, protein levels normalized to loading control (Actin), assessed by densitometry of n = 3 biological replicates. Statistical significance was determined by unpaired two-tailed Student’s t-test, and data are shown as mean ± s.e.m. d, Representative IF of myogenin (MyoG, pink) and DAPI (blue) in cultured human muscle cells after differentiation (3 days) with siControl or siPFKM treatment, with or without PFKM overexpression of PFKM-WT or PFKM-RK. Scale bars, 10 μm, n = 790 control nuclei, n = 1,050 siPFKM nuclei, n = 533 PFKM-WT nuclei, n = 595 siPFKM and PFKM-WT nuclei, n = 506 PFKM-RK nuclei and n = 562 siPFKM and PFKM-RK nuclei from three ROIs. e, Colocalization analysis of MyoG and DAPI normalized by total nuclei count (×10 magnification). Statistical significance was determined by two-way ANOVA with post hoc Dunnett’s analysis, and data are shown as mean ± s.e.m. (NS, P > 0.05). f, Representative IF of MyHC (red) and DAPI (blue) in cultured human muscle cells after differentiation (3 days) with siControl or siPFKM treatment, with or without PFKM overexpression of PFKM-WT or PFKM-RK. Scale bars, 20 μm, n = 1,665 control nuclei, n = 1,462 siPFKM nuclei, n = 1,239 PFKM-WT nuclei, n = 1,014 siPFKM and PFKM-WT nuclei, n = 886 PFKM-RK nuclei and n = 1,137 siPFKM and PFKM-RK nuclei from four fields of view. g, Quantification of mean fluorescence intensity (MFI) of MyHC per field of view (×10 magnification). Statistical significance was determined by two-way ANOVA with post hoc Dunnett’s analysis, and data are shown as mean ± s.e.m. (NS, P > 0.05). h, Representative IF of MyoG (pink) and DAPI (blue) in undifferentiated (day 0) cultured human muscle cells in control or PFKM-WT overexpression. Scale bars, 10 μm (n = 379 control cells and n = 298 PFKM-WT cells from five fields of view). Right, colocalization analysis of MyoG and DAPI normalized by total cell count (×20 magnification). Statistical significance was determined by unpaired two-tailed Student’s t-test, and data are shown as mean ± s.e.m. i, Representative IF of Myh1 (red) and DAPI (blue) in undifferentiated (day 0) cultured human muscle cells in control or PFKM-WT overexpression. Scale bars, 10 μm (n = 290 control cells and n = 233 PFKM-WT cells from four fields of view). Right, quantification of MFI of Myh1 per field of view (×20 magnification). Statistical significance was determined by unpaired two-tailed Student’s t-test, and data are shown as mean ± s.e.m. j, Schematic showing low PFKM leads to altered differentiation, and overexpression of both PFKM-WT and PFKM-RK rescues differentiation. Exact P values are shown in graphs.

Source data

Extended Data Fig. 6. PFKM depletion interferes with both muscle differentiation and fusion.

a, Immunoblotting (IB) of PFKM in undifferentiated cultured human muscle cells with siControl or siPFKM treatment. Right, protein levels normalized to loading control (Actin), assessed by densitometry of n = 3 biological replicates. Statistical significance was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. b, IB of PFKM in C2C12 muscle cells after differentiation (3 days) with siControl or siPFKM treatment. Right, protein levels normalized to loading control (Actin), assessed by densitometry of n = 3 biological replicates. Statistical significance was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. c, Representative immunofluorescence (IF) of myogenin (MyoG, pink) and DAPI (blue) in C2C12 muscle cells after differentiation with siControl or siPFKM treatment. Scale bar: 30 μm (n = 811 control nuclei and n = 1148 siPFKM nuclei from 4 regions of interest (ROIs)). Right, colocalization analysis of MyoG and DAPI normalized by total nuclei count (10x×10 magnification). Statistical significance was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. d, Representative IF of myosin heavy chain (MyHC, red) and DAPI (blue) in C2C12 muscle cells after differentiation with siControl or siPFKM treatment. Scale bar: 30 μm (n = 3783 control nuclei and n = 3126 siPFKM nuclei from 4 fields of view). Right, quantification of mean fluorescence intensity (MFI) of MyHC per field of view (×10 magnification). Statistical significance was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. e, Fusion index calculations were performed using the percentage of contacting nuclei over total nuclei in cultured muscle cells after differentiation with siControl or siPFKM treatment. Left, quantification of human muscle cells (n = 2586 control nuclei and n = 1868 siPFKM nuclei from 6 ROIs). Right, quantification of C2C12 muscle cells (n = 2203 control nuclei and n = 1458 siPFKM nuclei from 6 ROIs). Statistical significance was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. f, Fusion index calculations were performed using the percentage of nuclei in MyHC+ multinucleated cells (myotubes) over total nuclei in cultured muscle cells after differentiation with siControl or siPFKM treatment. Left, quantification of human muscle cells (n = 2159 control nuclei and n = 1598 siPFKM nuclei from 5 fields of view). Right, quantification of C2C12 muscle cells (n = 1849 control nuclei and n = 1199 siPFKM nuclei from 5 ROIs). Statistical significance was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. g, Total nuclei per ROI in cultured muscle cells after differentiation in siControl or siPFKM treatment. Left, quantification of human muscle cells (n = 1724 control nuclei and n = 1829 siPFKM nuclei from 5 ROIs). Right, quantification of C2C12 muscle cells (n = 1640 control nuclei and n = 1459 siPFKM nuclei from 5 ROIs). Statistical significance was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. (ns P>0.05). h-i, IB of PFKM in cultured human muscle cells after differentiation with siControl or siPFKM treatment, with or without PFKM overexpression of wild-type (PFKM-WT) or methyl-mutant R774K (PFKM-RK). Right, protein levels normalized to loading control (Actin), assessed by densitometry of n = 3 biological replicates. Statistical significance was determined by two-way ANOVA with post-hoc Dunnett’s analysis and data are shown as mean ± s.e.m. (ns P>0.05). j, IB of PFKM in undifferentiated cultured human muscle cells in control or PFKM-WT conditions. Right, protein levels normalized to loading control (Actin), assessed by densitometry of n = 3 biological replicates. Statistical significance was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. k, Cell proliferation assay in control or PFKM-WT conditions after 24 h in undifferentiated human muscle cells, (n = 3 biological replicates). Statistical significance was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. Exact P values are shown in graphs.

Source data

We next aimed to confirm that the observed effects on differentiation were caused by PFKM depletion. Differentiation assays were repeated in muscle cells following the exogenous expression of siRNA-resistant PFKM-WT and PFKM-RK. Immunoblotting confirmed PFKM expression in control and PFKM-KD cells (Extended Data Fig. 6h,i). Expression of either PFKM-WT or PFKM-RK was sufficient to increase the levels of myogenin⁺ nuclei (Fig. 4d,e) and myosin staining in PFKM-KD cells to the levels of control cells (Fig. 4f,g). As a final metric, we examined the effect of increasing PFKM levels in muscle cells before differentiation. Compared to control cells, PFKM-overexpressing cells displayed higher levels of myogenin and MyHC, even in the absence of differentiation media (Fig. 4h,i and Extended Data Fig. 6j). Accordingly, exogenous PFKM also decreases proliferation in undifferentiated cells (Extended Data Fig. 6k), in line with muscle cells exiting the cell cycle to begin expressing differentiation markers¹⁵. These data further support a specific role for PFKM during muscle differentiation (Fig. 4j).

PFKM links metabolic landscapes of muscle cells during differentiation

A muscle stem cell adopts a unique profile of metabolites at each stage of the differentiation process^15,42. How such metabolite fluctuations directly influence muscle cell reprogramming is incompletely understood. The unexpected finding that PFKM impacts muscle cell differentiation led us to question the mechanism of action. Classically known as the gatekeeper of glycolysis, PFKM lies at a critical node for determining the cellular fate of glucose. In Tarui disease, the deficiency of PFKM disrupts the metabolism of glucose through glycolysis, and cells accumulate upstream metabolites like glucose-6-phosphate, the precursor of the PPP. Our analyses identified that PFKM levels are lowest in MuSCs. Intriguingly, previous studies hint at a preference for PPP in MuSCs as they are known to be exposed to high levels of oxidative stress^32,43. To explore the different branches of glucose metabolism, we probed the human skeletal muscle transcriptomics¹³. PPP gene signatures were enriched specifically within MuSC populations, which tapered off within both slow-twitch and fast-twitch differentiated myofibres (Fig. 5a and Extended Data Fig. 7a). Accordingly, the relative levels of PFKM expression inversely correlated with PPP target gene levels within MuSCs and both type I and type II differentiated myofibres (Extended Data Fig. 7a). Gene signatures of glycolysis and oxidative phosphorylation were highly enriched within slow-twitch and fast-twitch myofibres (Fig. 5a and Extended Data Fig. 7a), in line with prior studies⁴². By mapping the related gene expression onto pseudotime trajectories, we found that cells with higher expression of PPP pathway genes were mainly in the MuSC states, while cells with elevated expression of glycolytic and oxidative phosphorylation pathway-related genes became progressively enriched along the differentiation trajectories to myofibre populations (Extended Data Fig. 7b).

Fig. 5. PFKM diverts glucose metabolism to enable muscle differentiation.

a, Heatmap of the PPP and glycolytic gene signatures from RNA-seq of single-cell and single-nucleus data of human skeletal muscle atlas in slow-twitch myofibre (type I) and fast-twitch myofibre (type II) or MuSCs¹³. PPP genes: phosphogluconate dehydrogenase (PGD), phosphoribosyl pyrophosphate synthetase 1 (PRPS1), aldolase A (ALDOA), aldolase C (ALDOC), ribulose-5-phosphate-3-epimerase (RPE), 6-phosphogluconolactonase (PGLS), glucose-6-phosphate dehydrogenase (G6PD), transketolase (TKT) and phosphoglucomutase 2 (PGM2). Glycolysis genes: hexokinase 2 (HK2), acetyl-CoA synthetase 2 (ACSS2), phosphoglucomutase 1 (PGM1), phosphoglycerate mutase 2 (PGAM2), enolase 3 (ENO3), galactose mutase (GALM), hexokinase 1 (HK1), phosphoglycerate kinase 1 (PGK1), aldehyde dehydrogenase 1B1 (ALDH1B1) and glucose-6-phosphate isomerase (GPI). Red, increase; blue, decrease. b, PFKM and PPP gene correlations with regression of LAMP1 or PRMT1 by mediation analysis using GTEx, shown as boxplots (n = 7 genes). PFKM–LAMP1 correlations: natural variation (min., 2.1; max., 10.9; median, 6.7) and adjusted (min., 0.3; max., 3.1; median, 2.5). PFKM–PRMT1 correlations: natural variation (min., 2.3; max., 11.0; median, 6.3) and adjusted (min., 0.1; max., 4.8; median, 0.8). Whiskers indicate minima and maxima, and boxes represent the interquartile range. Statistical significance was determined by two-tailed Wilcoxon tests. c, RT–qPCR relative transcript levels of genes (PPP, blue) and (glycolysis, pink) in cultured human muscle cells after differentiation (3 days) with siControl or siPFKM treatment (n = 6 biological replicates). Statistical significance was determined by multiple unpaired two-tailed Student’s t-test analyses with Šidák–Bonferroni correction (α = 0.05), and data are shown as mean ± s.e.m. PHGDH, phosphoglycerate dehydrogenase; PKM2, pyruvate kinase isoform M2. d, Relative abundance of 3-PG, pyruvate and citrate in siControl or siPFKM treatment (fold change) in differentiated (day 3) cultured human muscle cells by gas chromatography (GC)–MS (n = 6 biological replicates). Statistical significance was determined by unpaired two-tailed Student’s t-test, and data are shown as mean ± s.e.m. e, Schematic of labelled glucose and fractional labelling of 3-PG using 1,2-¹³C labelled glucose in siControl or siPFKM treatment (fold change) in differentiated (day 3) cultured human muscle cells by GC–MS (n = 3 biological replicates). Statistical significance was determined by one-way ANOVA with post hoc Bonferroni’s analysis, and data are shown as mean ± s.e.m. (NS, P > 0.05). f, Measurement of OCR in cultured human muscle cells after differentiation (3 days) with siControl or siPFKM treatment (fold change) at basal levels or in response to indicated compounds: oligomycin (1.5 µM, ATP-linked respiration), BAM15 (2.5 μM, maximal respiration) and rotenone and antimycin A (5 μM, reserve capacity) (n = 3 biological replicates). Statistical significance was determined by one-way ANOVA with post hoc Bonferroni analysis, and data are shown as mean ± s.e.m. g, RT–qPCR relative transcript levels of gene targets myosin heavy chain 1 (MYH1) in cultured human muscle cells after differentiation (3 days) with siControl or siPFKM treatment, with or without exogenous 3-PG supplementation (2 mM) (n = 6 biological replicates). Statistical significance was determined by two-way ANOVA with post hoc Dunnett’s analysis, and data are shown as mean ± s.e.m. (NS, P > 0.5). h, IB of MyHC in cultured human muscle cells after differentiation (3 days) with siControl or siPFKM treatment, with or without exogenous 3-PG supplementation (2 mM). Right, protein levels normalized to loading control (Actin), assessed by densitometry of n = 3 biological replicates. Statistical significance was determined by two-way ANOVA with post hoc Dunnett’s analysis, and data are shown as mean ± s.e.m. (NS, P > 0.5). i, Representative IF of MyoG (pink) and DAPI (blue) in cultured human muscle cells after differentiation (3 days) with siControl or siPFKM treatment, with or without exogenous 3-PG supplementation (2 mM). Scale bar, 20 μm; inset scale bar, 10 μm (n = 661 control nuclei, n = 488 siPFKM nuclei, n = 587 3-PG-treated nuclei and n = 460 siPFKM and 3-PG-treated nuclei from five fields of view). Asterisks denote MyoG⁺ nuclei in representative images. Right, colocalization analysis of MyoG and DAPI normalized by total nuclei count (×20 magnification). Statistical significance was determined by two-way ANOVA with post hoc Dunnett’s analysis, and data are shown as mean ± s.e.m. (NS, P > 0.5). j, Schematic showing that low PFKM leads to altered differentiation, and exogenous supplementation of 3-PG metabolite rescues differentiation. Exact P  values are shown in graphs.

Source data

Extended Data Fig. 7. PFKM depletion impacts glucose metabolism through glycolysis and PPP.

a, RNA-seq data from the human skeletal muscle atlas of 40 major human skeletal muscle populations collected from single-cell and single-nucleus data¹³. Relative expression levels of metabolic genes across slow-twitch myofibres, fast-twitch myofibres, and muscle stem cells (MuSC) by dot plot. Pentose phosphate pathway (PPP) genes: ribulose-5-phosphate-3-epimerase (RPE), aldolase A (ALDOA), aldolase C (ALDOC), phosphoribosyl pyrophosphate synthetase 2 (PRPS2), 6-phosphogluconolactonase (PGLS), 6-phosphogluconate dehydrogenase (PGD), phosphoribosyl pyrophosphate synthetase 1 (PRPS1), glucose-6-phosphate dehydrogenase (G6PD), transketolase (TKT). Glycolysis/TCA cycle genes: phosphoglycerate kinase 1 (PGK1), glucose-6-phosphate isomerase (GPI), hexokinase 1 (HK1), hexokinase 2 (HK2), phosphoglucomutase 1 (PGM1), phosphoglycerate mutase 2 (PGAM2), enolase 3 (ENO3), citrate synthase (CS), aconitase 2 (ACO2), isocitrate dehydrogenase 2 (IDH2), isocitrate dehydrogenase 3 (IDH3A), oxoglutarate dehydrogenase (OGDH), succinate-CoA ligase (SUCLA2), acetyl-CoA synthetase 2 (ACSS2), aldehyde dehydrogenase 1B1 (ALDH1B1). b, Pseudotime trajectory analysis of combined muscle stem cells (MuSC), slow-twitch myofibre and fast-twitch myofibre myogenic cells for PPP and glycolysis/TCA cycle genes. Each dot represents a single cell, with the color gradient indicating gene expression level. c, Mediation analysis of PFKM and LAMP1 or PRMT1 in relation to PPP genes using GTEx (n = 310 donors). Statistical significance was determined by two-tailed Student’s t statistics testing for correlation. Heatmaps are associated with Fig. 5b. Exact P values are shown in graphs.

Source data

To explore a potential causal relationship between PFKM expression and metabolic gene expression, we performed a regression-based mediation analysis using population variation. We aimed to understand whether the variance of specific genes marking lysosomes could explain strong correlation structures, such as those between PPP and PFKM. Using skeletal muscle gene expression from 310 individuals available in GTEx, mediation analysis was applied by regressing out the expression of lysosomal marker LAMP1 to assess the relationship between PFKM and PPP genes mediated by the presence of LAMP1. As expected, PFKM expression was significantly correlated with various PPP genes in the presence of LAMP1 (Fig. 5b and Extended Data Fig. 7c). After adding LAMP1 as a covariate to this model, many of the PFKM–PPP gene correlations were lost. Similar results were also found using PRMT1 and explain the significance between PFKM and PPP pathways (Fig. 5b and Extended Data Fig. 7c). These genetic analyses show that PRMT1 and lysosomal proteins are potential mediators of PFKM relationships with alternative glucose metabolism pathways in muscle.

Metabolite shunting enables muscle cell differentiation

Muscle cells have unique metabolic demands at different stages of differentiation^15,42. Our analyses of skeletal muscle tissue show that the PPP is highly expressed within MuSC, while PFKM is acquired during differentiation. This led us to explore whether controlling PFKM levels reinforces the use of glucose between alternative branches of metabolism. We evaluated whether the loss of PFKM in differentiated muscle cells would impact metabolic gene levels. Compared to control cells, PFKM-KD cells caused a significant decrease in the levels of downstream glycolytic genes and pathways that branch off glycolysis downstream of PFKM, phosphoglycerate dehydrogenase (PHGDH) and pyruvate kinase (PKM2) (Fig. 5c). By contrast, PFKM-KD cells had a marked increase in central PPP enzymes, glucose-6-phosphate dehydrogenase (G6PD), phosphogluconate dehydrogenase (PGD) and transketolase (TKT) (Fig. 5c), consistent with the inverse expression of PFKM and PPP in skeletal muscle tissue. PFKM depletion in HEK293 cells recapitulated the gene expression changes of glycolytic and PPP enzymes observed in muscle cells (Extended Data Fig. 8a). Congruently, PFKM-KD cells also had reduced levels of intermediate metabolites of glycolysis (FBP and pyruvate), as expected (Extended Data Fig. 8b). No significant differences in cell counts of control and PFKM-KD were found over the course of multiple days (Extended Data Fig. 8c). PFKM-KD increased the cellular capacity to resist oxidative stress induced by exposure to hydrogen peroxide over multiple days (Extended Data Fig. 8c), aligning with an expected phenotype of NADPH production through PPP activity⁴⁴. These data align with the inverse relationship of PFKM and PPP in skeletal muscle tissues to show that PFKM depletion increases PPP gene signatures in differentiated muscle cells.

Extended Data Fig. 8. PFKM levels regulate glucose utilization between glycolysis and PPP.

a, RT–qPCR of gene targets glucose-6-phosphate dehydrogenase (G6PD), phosphogluconate dehydrogenase (PGD), and transketolase (TKT) (PPP, blue) and hexokinase 1 (HK1) phosphoglycerate dehydrogenase (PHGDH) and glyceraldehyde-3-phosophate dehydrogenase (GAPDH) (glycolysis, pink) in HEK293 cells with siControl or siPFKM treatment (n = 6 biological replicates). Statistical significance was determined by multiple unpaired two-tailed Student’s t-test analyses with Šidák-Bonferroni correction (α = 0.05) and data are shown as mean ± s.e.m. b, Relative abundance of fructose-1,6-bisphosphate (FBP) and pyruvate (Pyr) with siControl or siPFKM treatment (fold change) in HEK293 cells by gas-chromatography mass spectrometry (GC-MS), (n = 3 biological replicates). Statistical significance was determined by multiple unpaired two-tailed Student’s t-test analyses with Šidák-Bonferroni correction (α = 0.05) and data are shown as mean ± s.e.m. c, Cell proliferation assay in HeLa cells with siControl or siPFKM treatment and H₂O₂ for 3 days. (n = 3 biological replicates). Statistical significance was determined by one-way ANOVA with post-hoc Dunnett’s analysis and data are shown as mean ± s.e.m. (ns P>0.05). d, Abundance of 3-phosphoglycerate (3-PG), pyruvate, citrate and ribose-5-phosphate (R5P) in siControl or siPFKM treatment (fold change) in undifferentiated (day 0) cultured human muscle cells by GC-MS (3-PG, pyruvate, and citrate n = 6 biological replicates. R5P n = 3 biological replicates). Statistical significance was determined by unpaired two-tailed Student’s t-test and data are shown as mean ± s.e.m. e, Fractional labeling of 3-PG using 1,2-¹³C labelled glucose in siControl or siPFKM treatment (fold change) in undifferentiated cultured human muscle cells by GC-MS (n = 3 biological replicates). Statistical significance was determined by one-way ANOVA with post-hoc Bonferroni’s analysis and data are shown as mean ± s.e.m. (ns P>0.05). f, ATP production calculated from oxygen consumption rates (OCR) in undifferentiated cultured human muscle cells (orange, left) and 3 day differentiated human muscle cells (purple, right) with siControl or siPFKM treatment, (n = 3 biological replicates). Statistical significance was determined by one-way ANOVA with post-hoc Bonferroni’s analysis and data are shown as mean ± s.e.m. (ns P>0.5). g, Measurement of extracellular acidification rate (ECAR) in undifferentiated cultured human muscle cells (orange, left) and 3 day differentiated human muscle cells (purple, right) with siControl or siPFKM treatment (fold change) after Oligomycin, BAM15, and Rotenone/Antimycin A (n = 3 biological replicates). Statistical significance was determined by one-way ANOVA with post-hoc Bonferroni’s analysis and data are shown as mean ± s.e.m. (ns P>0.05). h, Measurement of OCR in undifferentiated cultured human muscle cells with siControl or siPFKM treatment (fold change) after indicated compounds: Oligomycin (ATP-linked Respiration), BAM15 (Maximal Respiration), and Rotenone/Antimycin A (Reserve Capacity) (n = 3 biological replicates). Significance was determined by one-way ANOVA with post-hoc Bonferroni’s analysis and data are shown as mean ± s.e.m. (ns P>0.05). i, OCR to ECAR ratios in undifferentiated cultured human muscle cells (orange, left) and 3 day differentiated human muscle cells (purple, right) with siControl or siPFKM treatment after Oligomycin, BAM15, and Rotenone/Antimycin A (n = 3 biological replicates). Significance was determined by two-way ANOVA with post-hoc Bonferroni’s analysis and data are shown as mean ± s.e.m. (ns P>0.05). j, Quantification of total nuclei count per field of view, associated with Fig. 5i. (n = 351 control nuclei, n = 314 siPFKM nuclei, n = 346 3-PG treated nuclei, and n = 285 siPFKM and 3-PG treated nuclei from 5 fields of view). Statistical significance was determined by two-way ANOVA with post-hoc Dunnett’s analysis and data are shown as mean ± s.e.m. (ns P>0.5). k, Immunoblotting quantification of PFKM in cultured human muscle cells after differentiation (3 days) with siControl or siPFKM treatment, with or without exogenous 3-PG supplementation. Quantifications are associated with Fig. 5h. Protein levels normalized to loading control (Actin), assessed by densitometry of n = 3 biological replicates. Statistical significance was determined by two-way ANOVA with post-hoc Dunnett’s analysis and data are shown as mean ± s.e.m. (ns P>0.5). Exact P values are shown in graphs.

Source data

PFKM regulates glucose metabolism in muscle cells

The shift in metabolic gene signatures hinted at the possibility that PFKM depletion alters cellular metabolism. To assess this prospect, we evaluated the metabolite changes of differentiated muscle cells after PFKM depletion. Compared to control cells, PFKM-KD cells displayed lower levels of the downstream glycolytic metabolites 3-phosphoglycerate (3-PG) and pyruvate as well as the TCA cycle intermediate citrate (Fig. 5d). Complementary analyses were performed with PFKM-KD in undifferentiated muscle, which showed similar trends in the reduced metabolite levels (Extended Data Fig. 8d). Additionally, PFKM-KD cells displayed higher levels of the PPP metabolite ribose-5-phosphate relative to controls (Extended Data Fig. 8d). Previous studies showed that the depletion of the alternative PFK1 isoforms redirects glucose into the PPP in neutrophils and cancer cells^3,5,45. To gain additional insights into our model, we performed 1,2-¹³C-labelled glucose tracing and monitored the M + 1 fraction of 3-PG^10,45,46. Previous studies have linked higher M + 1-labelling of downstream glycolytic intermediates to PPP metabolism^46,47. Relative to controls, PFKM-KD cells displayed a higher level of M + 1-labelled 3-PG within differentiated muscle cells (Fig. 5e). No significant changes were observed in the fractional labelling of undifferentiated muscle cells (Extended Data Fig. 8e).

Human patients with loss-of-function PFKM mutations have defective glucose use through glycolysis and oxidative phosphorylation³⁴. The reduced levels of TCA metabolites in our system led us to evaluate the impact of PFKM depletion on oxidative phosphorylation using Seahorse analyses to measure the oxygen consumption rate (OCR) and the extracellular acidification rate (ECAR). PFKM-KD cells displayed lower basal OCR levels than control cells (Fig. 5f). Additionally, PFKM-KD cells had reduced OCR rates after BAM15 injection, indicating a decrease in maximal respiratory capacity, and had lower ATP production as calculated from OCR measurements (Fig. 5f and Extended Data Fig. 8f). No significant changes were observed in ECAR or OCR rates between control and PFKM-KD cells before differentiation (Extended Data Fig. 8g,h). No significant changes were observed in the OCR:ECAR ratios at basal states or following treatments, which suggests that a proportional shift in oxidative phosphorylation and glycolysis occurs in PFKM-KD cells (Extended Data Fig. 8i)⁴⁵. These data are consistent with metabolite level changes and support a role for PFKM in the glycolysis and oxidative phosphorylation pathways of differentiated muscle cells.

Downstream metabolites restored muscle cells lacking PFKM

Metabolic gene signatures support a model in which PFKM protein levels tune glucose use between branching pathways. Given that PFKM-KD reduced differentiation efficiency and shifted glucose into PPP, we reasoned that PFKM could function as a provider of key metabolites for muscle differentiation. To test this idea, we evaluated whether directly supplementing downstream glycolytic metabolites would be sufficient to restore defective differentiation of PFKM-KD cells. PFKM generates several intermediates, such as 3-PG, which was decreased in our metabolic tracing above. The 3-PG metabolite is non-toxic and displays high cell permeability. Leveraging these characteristics, differentiation assays were performed in control and PFKM-KD cells in the presence of 3-PG supplementation. Differentiation efficiency, marked by myogenin⁺ nuclei and myosin levels, was significantly increased by 3-PG supplementation of PFKM-KD cells. Total cell numbers were consistent across all conditions (Extended Data Fig. 8j). In addition, 3-PG was sufficient to increase the levels of MYH1 transcripts and MyHC protein expression of PFKM-KD to the level of control cells (Fig. 5g,h and Extended Data Fig. 8k). Moreover, immunofluorescence analyses showed that the myogenin⁺ nuclei of PFKM-KD cells were restored to the level of controls in the presence of 3-PG (Fig. 5i,j). These analyses align with a role for PFKM to generate key metabolite intermediates during the muscle differentiation processes.

In sum, the collective findings of this work demonstrate a precisely tuned compartmentalization and associated metabolic shunting that modulate cellular transition processes.

Discussion

An emerging conceptualization of metabolism involves metabolic enzymes, and even metabolites themselves, to extend past bioenergetics to actively instruct cell fate and embryogenic programmes^2,8,9. Our findings reveal that the spatiotemporal remodelling of a metabolic enzyme, PFKM, and the associated metabolic shunting, enables critical sugar processing during muscle cell differentiation. A MrDegron modification sparks PFKM microautophagy, and the absence of cytosolic PFKM reduces downstream glycolytic metabolites and biosynthetic building blocks. Our phenotypic and gene-based analyses support a role for PFKM in muscle health, as reducing PFKM levels caused alterations of muscle differentiation and maturation in cultured cell models. Given that PPP is the largest NADPH producer across most cell types, our data suggest that a metabolite branch point, at the level of PFKM¹⁰, may have evolved as a mechanistic determinant for glucose use within muscle populations.

Cellular compartmentalization is essential for precise spatiotemporal regulation of biological processes^23,48. There is an increasing appreciation that protein distribution and subcellular compartmentation provide temporal control of metabolite fluctuations on minute timescales^27,48, before expression levels of glucose transporters and glycolytic genes that collectively enhance glycolytic metabolism⁹. Lysosomal compartmentation rapidly dictates the abundance of vitamin B6 metabolism during cell growth²⁹. The present study contributes to emerging functions of lysosomes and protein turnover in metabolic control. PFKM is post-translationally modulated (MrDegron) to provide spatiotemporal control of glucose consumption through glycolytic and oxidative phosphorylation or PPP. Interestingly, proteolysis of PFKP and PFKL through the proteasome is essential for cellular sensing of mechanical stress and substrate stiffness⁴. Our findings align with the role of protein turnover in PFK regulation and introduce a degradative organelle—the lysosome—as an isoform-specific pathway for PFKM. PRMT1 activity during delivery of methylated proteins to lysosomes is an established characteristic of Wnt-activated cells. Building on this knowledge, we showed that Wnt ligands caused a rapid accumulation of metabolites in upper glycolysis, in agreement with prior literature on Wnt and glucose metabolism¹⁹. Wnt signalling and metabolic shunting are central processes during embryogenesis² and cancer. Breast cancers first use glucose metabolism in the hexosamine biosynthesis pathway to promote metastasis by generating hexosamine-derived cofactors for integrin α_vβ₃ modifications during invasion, while cells at newly colonized sites switch to glycolysis for rapid proliferation⁵. During embryogenesis, two waves of glucose are reported that first prioritize the hexosamine biosynthesis pathway for glycosylation and then enter glycolysis for energy production². Stem cell fates can also be tuned by downstream use of glucose-sourced carbons, as highlighted by a non-canonical TCA cycle¹. Future investigations are warranted that explore whether Wnt can also drive glucose flow into hexosamine metabolism.

Despite the high level of sequence homology, PFK1 isoforms can exert distinct regulation within a single cell type^6,24,25. Indeed, PFKP and PFKL display distinct metabolite interaction profiles in the MIDAS platform²⁷. Structural elucidation by cryo-electron microscopy identified several unique features of each isoform, which may explain the differences in F6P affinity and ATP inhibition that have been reported for many years²⁵. By delineating an isoform-selective proteolytic pathway, our findings offer insight into how each PFK isoform functions across different cell and tissue types and may potentially perform specialized moonlighting functions⁴⁹. Notably, F6P and FBP, the metabolites impacted by differential PFKM compartmentation, have moonlighting functions outside of metabolism and could also contribute to the downstream cellular responses^27,50. PPP serves to generate NADPH for redox homoeostasis and macromolecule biosynthesis. Prior studies have shown that the levels of PFKL and PFKP isoforms are specifically controlled to enable glucose shunting between PPP and glycolysis. PFKL shunts metabolites between branching pathways to enable an oxidative burst that is critical for immune responses upon exposure to pathogens³. Genetic mutations in PFKM are found in patients with Tarui disease who display an overactive production of several PPP metabolites³⁴. Thus, the pathological roles of aberrant PFKM metabolic shunting have been established but unappreciated until now. Metabolic shunting provides versatility for diverse mechanisms that enable cellular adaptation on rapid timescales. Under low glycolysis conditions, PFKL tetramers disperse into monomers that tether lipid droplets to the mitochondria for fatty acid transfer and lipolytic β-oxidation⁴⁹. Our data show that PFKM moonlights as a fateful determinant in muscle, in alignment with the expanding role of metabolic branching in physiology⁵. Muscle cells exhibit distinct metabolic landscapes during the ordered, multi-step process of differentiation, which involves key myogenic factors, withdrawal from the cell cycle and fusion into multinucleated myotubes that eventually form complex skeletal muscle architectures⁵¹. Our work aligns with prior studies of muscle metabolism that suggest that the acquisition of PFKM during differentiation may have evolved to help cells metabolically adapt to differentiation processes^11,42,52. We show that PFKM is selectively expressed in mature fibre types, whereas PRMT1 and lysosomal genes are attenuated during differentiation¹³, suggesting that cells regulate PFKM at the gene and protein level. Fittingly, recent work supports a dynamic fluctuation of the universal methyl donor S-adenosyl-methionine that accompanies muscle differentiation⁵³.

How stem cells sense tissue damage and participate in orchestrating the response for tissue repair are essential questions in cell and tissue reprogramming fields. In contrast to the stem cells of continually renewing tissues like the blood, intestine and skin, the signals that activate quiescent stem cells of the muscle and brain are less understood. Muscle stem cells are integral for tissue repair throughout life. Re-engaging muscle stem cells therapeutically with pharmacologic or nutrient-based solutions offers a critical resource to accelerate physiologic muscle recovery in human diseases like muscular dystrophy, cancer cachexia and diabetes^54–56. Our findings align with prior established literature of canonical Wnt signalling in skeletal muscle repair and extend regulation with a Wnt–lysosomal–PFK axis. Although PFKM expression in adult muscle has been previously defined, we propose that a metabolic demand for PPP activity in undifferentiated muscle cells⁵⁷ necessitates a need to maintain low PFKM levels. During differentiation, lysosomal gene signatures decrease and PFKM levels are restored from nascent synthesis, which in turn increases glucose use for energetics. This model synergizes with currently accepted paradigms in muscle biology. First, arginine methylation in muscle stem cells has been established as a key regulator of muscle stem cell functions, whereby the genetic ablation of PRMT1 or PRMT6 causes a loss of regenerative capacity. Second, the downregulation of lysosomal-autophagy pathways is critical for differentiation into myotubes, as the constitutive activity of lysosomes blocks multinucleation and fusion. In sum, this line of research presents a distinct vantage point for muscle metabolism and stem cell remodelling and could provide insight into strategies to reconstitute stem cell activity for tissue regeneration.

Methods

Cell lines and tissue culture

Cell lines were obtained from the American Type Culture Collection. HeLa cells (CCL-2), HEK293 cells (CRL-1573), L cells (CRL-2647) and C2C12 cells (CRL-1772) were cultured in Dulbecco’s modified Eagle medium (DMEM) (10313021, Gibco) with 10% heat-inactivated FBS (A5670801, Gibco), 1% penicillin–streptomycin (15140122, Gibco), 1% L-glutamine (25030081, Gibco) and 0.25 mg ml⁻¹ Plasmocin (ant-mpp, Invivogen). A stable HEK293 cell line expressing WT PRMT1–HaloTag-Myc was previously generated through plasmid transfection and G418 selection³⁰. C2C12 differentiation was initiated by switching to high-glucose DMEM containing 2% horse serum (SH30074.02, Cytiva) and 1% penicillin–streptomycin once cells reached 80–90% confluency. The human quadricep-derived muscle cell line (B0-41) was previously immortalized from healthy myoblasts³⁷. Cultured human muscle cells were grown in high-glucose DMEM supplemented with 20% FBS, 1% penicillin–streptomycin and 2% Ultroser G (Crescent Chemical Co.). Differentiation was initiated by switching to high-glucose DMEM containing 2% FBS and 1% ITS-A (51300044, Thermo Fisher) once cells reached 80–90% confluency. All cells were tested biweekly for contamination with the Mycoplasma Detection Kit (Invivogen, rep-mys).

Cell treatments

For Wnt treatments, endogenous Wnt3a secretion was inhibited using 5 μM IWP-2 for 1 h (I0536, Millipore Sigma) across all conditions. Wnt conditions were treated with Wnt3a ligands (315-20, PeproTech) diluted 1:1,000 in Wnt3a-conditioned L cell media. For the digitonin detergent assays, cells were treated with 6.5 μg ml⁻¹ digitonin (D141, Sigma-Aldrich) for 10 min following Wnt treatment. Cycloheximide (C1988, Sigma-Aldrich) was used at 10 μM. Bafilomycin A1 (11038, Cayman Chemical) was used at 500 nM for 1 h for immunofluorescence microscopy experiments. For non-muscle cell immunoblotting, bafilomycin was used at 500 nM working concentration overnight, followed by a 1 mM treatment for 1 h. For use in muscle cell immunoblotting, bafilomycin was used at 400 nM for 4 h. LLOMe (L7393, Sigma-Aldrich) was used at 1 mM for 1 h, followed by removal and a 3 h incubation. MS023 (SML1555, Sigma-Aldrich) was used overnight at 10 μM, followed by a 4 h incubation at 20 μM the following day. MG132 (A11043, AdooQ Bioscience) was used for 1 h at 5 μM concentration. For human muscle cell treatments with 3-PG (P8877, Sigma-Aldrich), 2 mM was maintained in culture throughout the 3 day differentiation period.

Immunofluorescence

Cells were seeded on coverslips and left to adhere overnight. For cultured muscle cell lines, coverslips were coated with 1× poly-L-lysine (1339A, Newcomer Supply). After cell treatments, cells were washed with PBS and fixed with 4% paraformaldehyde (sc-281692; Santa Cruz Technology). Coverslips were permeabilized with 0.2% Triton X-100 and blocked with 0.5% BSA blocking buffer. Coverslips were incubated with primary antibodies overnight at 4 °C before being washed and incubated with secondary antibodies for 40 min at room temperature (20–24 °C) (Supplementary Table 1). Finally, coverslips were mounted onto glass slides with ProLong Gold antifade reagent with 4′,6-diamidino-2-phenylindole (DAPI) (P36931, Invitrogen). Images were acquired on a Zeiss LSM-900 microscope with Airyscan using ×10, ×20 and ×40 water immersion and ×63 oil immersion. C2C12 murine muscle cell images were taken on an ECHO Revolution microscope using ×10 magnification. The excitation lasers used to capture the images were 488, 568, 630 and 405 nm. The same brightness and contrast profile was applied to all images within the same experiment. Three to ten images were captured per condition, where each image was considered a field of view. Zeiss (v.3.4) (blue edition) and Fiji (v.1.54p) (ImageJ, National Institutes of Health (NIH)) imaging softwares were used for image analyses. For live-cell imaging, cells were plated in glass-bottom eight-well chamber slides. Transfected cells were incubated with Lysotracker Red DND-99 (L7528, Thermo Fisher) and Hoechst 33258 (H3569, Invitrogen) following the manufacturer’s instructions.

Transient transfections and siRNA knockdown

Cells were transfected with DNA (0.75–1 μg ml⁻¹) or siRNA (50–100 nM) for 24–72 h using Lipofectamine 2000 (11668019, Invitrogen) according to the manufacturer’s instructions. For cultured muscle cell experiments combining both PFKM knockdown and overexpression, knockdown was initiated first, followed by overexpression after 48 h. ON-TARGETplus SMART-pool siRNA targeting human PFKM (L-006765-00-0005), human VPS4A (L-013092-00-0005), human PRMT1 (L-010102-00-0005) and mouse PFKM (L-065512-00-0005) were all purchased from Dharmacon. Predesigned Dicer-substrate siRNA (DsiRNA) for human PFKM that targets the 3′ untranslated region (hs.Ri.PFKM.13.1) and Negative Control DsiRNA (51-01-14-03) were purchased from Integrated DNA Technologies.

Immunoblotting

Cell lysates were prepared by the addition of 2× Laemmli sample buffer and dithiothreitol (DTT). Samples were then boiled at 95 °C for 10 min before being resolved by 4–20% SDS–PAGE gel electrophoresis using standard protocols. After transfer, nitrocellulose membranes were placed on a shaker in blocking buffer (5% nonfat dry milk in PBS) for 1 h at room temperature. Blots were incubated with primary antibodies overnight at 4 °C before being washed and incubated with secondary antibodies for 1 h at room temperature (Supplementary Table 1). Images were acquired using the iBright FL1500 imaging system. C2C12 lysates were prepared with 4× LDS Sample Buffer (NP0007, Thermo Fisher) and 10× Reducing Agent (NP0009, Thermo Fisher), heated at 70 °C for 10 min, resolved on 4–20% TGX Stain-Free gels (5678095, Bio-Rad) and transferred to PVDF membranes following the manufacturer’s instructions. C2C12 blots were imaged using SuperSignal West Dura ECL Substrate (34075, Thermo Fisher) on a Bio-Rad ChemiDoc imaging system. Fiji (v.1.54p) (ImageJ, NIH) was used for blot analyses.

Cell proliferation

HeLa cells transfected with siPFKM were seeded at 1.0 × 10⁵ cells per ml and treated in triplicate with 50 μM H₂O₂ (H325-500, Fisher Scientific) for 3 days. For cultured human muscle, cells were transfected in triplicate with PFKM overexpression and counted after 24 h. All cells were counted using the Countess II Automated Cell Counter.

LC–MS/MS for protein methylation

Stable HEK293 cell lines expressing WT PRMT1–HaloTag-Myc were transiently transfected with SNAP-tagged PFKM and pretreated with 400 nM bafilomycin A1 for 4 h before undergoing 20 min Wnt treatment in the presence of 400 nM bafilomycin and 10 μM MG132. Cells were collected in lysis buffer (50 mM Tris (pH 7.9), 150 mM NaCl, 1 mM EGTA, 1 mM EDTA, 1 mM DTT, 10 µM leupeptin, 100 µM AEBSF and 2 mM PMSF) and clarified by centrifugation at 16,000g for 5 min at 4 °C. Clarified lysates were incubated with anti-SNAP magnetic beads equilibrated with Tris-buffered saline (TBS) (S9145S, New England Biolabs) for 3 h on a rotator at 4 °C. Following incubation, beads were washed with TBS and eluted in 1% formic acid. Immunoprecipitated samples were reduced with DTT (10 mM, 95 °C, 10 min) and alkylated (30 mM iodoacetamide, room temperature, dark, 1 h) before digestion with sequencing grade trypsin (Promega, 1/50 (w/w), 37 °C, 18 h). Samples were analysed by LC–MS using an ACQUITY UPLC H-class LC coupled to a Xevo QTOF (Waters Corp). Observed m/z were mapped to an in silico digestion of PFKM using Biopharmalynx (Waters v.1.3.5; 30 ppm; fixed modification, carbamidomethylation of cysteine; variable modifications, oxidation of methionine and methylation of arginine). Peptides were injected onto a reverse-phase C4 column (ACQUITY UPLC Protein BEH C4 column, 300 Å, 1.7 μm, 2.1 mm × 50 mm; Waters Corp) in a buffer stream of 3% acetonitrile (ACN) in 0.1% formic acid (0.3 ml min⁻¹) and desalted for 0.5 min before gradient elution from 3% to 60% ACN over 1.5 min. The Xevo Z-spray source was operated with a capillary voltage of 3,000 V and a cone voltage of 40 V (NaCsi calibration, Leu749 enkephalin lock mass). Nitrogen was used as the desolvation gas at 350 °C and a total flow of 800 l h⁻¹.

LC–MS for metabolomics

HeLa cells were treated with Wnt3a ligand for 20 or 120 min. Cells were washed with 2 ml of 150 mM ammonium acetate solution (pH 7.4, 4 °C) and collected after a 15 min incubation in 500 ml of dH₂O:MeOH solution (1:4) at −80 °C. Samples were spun at 4 °C for 10 min at 17,000g, and supernatants were transferred to fresh vials and dried with an EZ2-Elite lyophilizer (Genevac). Dried metabolites were resuspended in 100 ml of dH₂O:ACN solution (1:1) and spun at 17,000g for 10 min. Samples were transferred to high-performance liquid chromatography (HPLC) autosampler vials and ran on a Vanquish (Thermo Fisher Scientific) UHPLC system with mobile phase A (5 mM NH₄AcO, pH 9.9) and mobile phase B (ACN) at a flow rate of 200 ml min⁻¹ on a Luna 3 µm NH₂ column (100 Å; 150 × 2.0 mm) at 40 °C with a gradient from 15 to 95% A in 18 min followed by an 11 min isocratic step. The UHPLC was coupled to a Q-Exactive (Thermo Fisher Scientific) mass analyser running in polarity switching mode at 3.5 kV with an MS1 resolution of 70,000. Metabolites were identified by exact mass (MS1), retention time and by fragmentation patterns (MS2) at normalized collision energy based on internal standards. Quantification was performed by area under the curve integration of MS1 ion chromatograms with the MZmine 2 software package. Area values were then normalized to cell count averages.

MetaboAnalyst analysis

Metabolite pathway analysis was performed on LC–MS Wnt metabolomics data using MetaboAnalyst 6.0. Metabolite compound names were used as identifiers, and the Homo sapiens KEGG Pathway Database was used as the set library. A scatter plot was used as the visualization method, a hypergeometric test was chosen as the enrichment method and relative-betweenness centrality was used as the topology measure. All compounds in the selected pathway library were used as the reference metabolome. At the time of access, KEGG pathway information had been updated as of December 2024.

Gas chromatography–MS for metabolomics

For HEK293 whole-cell metabolomics, cells were washed with cold 0.9% NaCl saline. Then, 1 ml of 80% methanol (diluted in HPLC-grade water) with norvaline as an internal standard was added to cells, followed by transfer of the six-well plates to −80 °C for 15 min. Whole-cell extracts were collected, transferred to a tube and centrifuged at 18,000g for 10 min at 4 °C. The supernatant was then dried by a speed vacuum and incubated with 50 μl of methoxyamine hydrochloride (10 mg ml⁻¹ in pyridine) at 42 °C for 1 h. After cooling, the samples were incubated with 100 μl of tert-butyldimethylsilyl ether at 70 °C for 1 h.

For labelled glucose experiments, cells were incubated with 20 mM of 1,2-¹³C-labelled glucose (453188, Sigma-Aldrich) in glucose-free DMEM (11966025, Gibco) for 2 h. Metabolites were then extracted as described above for HEK293 whole-cell metabolomics. Derivatized samples were analysed using an Agilent 7890 gas chromatograph coupled to a 5977B mass selective detector. Separation was achieved on an Agilent HP-5MS UI column (30 m × 0.25 mm × 0.25 μm film thickness). The gas chromatograph oven programme began at 60 °C (1 min hold), ramped to 300 °C at 10 °C min⁻¹ and held for 5 min. Helium was used as the carrier gas at a constant flow of 1.0 ml min⁻¹. Injections were made in splitless mode with an inlet temperature of 280 °C. The MS operated in electron ionization mode at 70 eV with a scan range of m/z = 50–650. Metabolite levels were normalized to total protein content for each sample. All samples were analysed using an Agilent 7820A gas chromatograph system and an Agilent 5977 mass spectrometer. Metabolite peaks were analysed for quality and quantity using standards as reference. All isotope tracing experiment data were corrected for natural isotope abundance. MassHunter Qualitative Analysis software was used to extract ion chromatograms, and MassHunter Quantitative Analysis was used to quantify peak areas.

DNA constructs

The full-length WT-GFP–PFKM construct was purchased from SinoBiological (HG14133-ANG). WT-VPS4–GFP, DN-VPS4–GFP and CA-LRP6 constructs were previously generated and used to study lysosomal delivery and Wnt sigaling^26,28. A SNAP-PFKM construct (Supplementary Data 1) was generated from a SNAP-GSK3β plasmid³⁰ by using BamHI/NotI restriction enzymes (R3136S/R3189S, New England BioLabs) and the following PFKM primers (Integrated DNA Technologies): forward, 5′-GATCGGATCCACCCATGAAGAGCACCATGCAG-3′; reverse, 5′-GATCGCGGCCGCGACGGCCGCTTCCCCG-3′. A mutant R774K GFP–PFKM construct (Supplementary Data 1) was generated from WT-GFP–PFKM by using pfuUltra DNA polymerase (600385, Agilent) and Dpn1 enzyme (R0176S, New England Biolabs) and using the following PFKM primers (Integrated DNA Technologies): forward, 5′-ACATCACCCGGAAGAAGTCCGGGGAAGCGG-3′; reverse, 5′-CCGCTTCCCCGGACTTCTTCCGGGTGATGT-3′. Newly generated constructs are available upon reasonable request from the corresponding author.

Co-immunoprecipitation

For PRMT1–PFKM co-immunoprecipitation (co-IP), stable HEK293 cells expressing WT PRMT1–HaloTag-Myc were transiently transfected with SNAP-tagged PFKM as described above. For VPS4–PFKM co-IP, HEK293 cells were transfected with SNAP-tagged PFKM and WT-VPS4–GFP as described above. The following day, cells were treated with 400 nM bafilomycin A1 for 4 h. Cells were then lysed by scraping in lysis buffer (50 mM Tris (pH 7.9), 150 mM NaCl, 1 mM EGTA, 1 mM EDTA, 1 mM DTT, 10 µM leupeptin, 100 µM AEBSF and 2 mM PMSF) and clarified by centrifugation at 16,000g for 5 min at 4 °C. Clarified lysates were incubated with anti-Myc magnetic beads for the PRMT1–PFKM co-IP (88842, Thermo Fisher Scientific, equilibrated with 3 × 10 vol/vol TBS washes) for 3 h on a rotator at 4 °C. For the VPS4–PFKM co-IP, clarified lysates were incubated with anti-SNAP magnetic beads (S9145S, New England Biolabs, equilibrated with 3 × 10 vol/vol TBS washes) for 3 h on a rotator at 4 °C. Following incubation, beads were washed three times with TBS and eluted in 4× Laemmli buffer with DTT, followed by incubation at 95 °C for 15 min. Lysates and immunoprecipitated samples were analysed by immunoblotting as described above.

Quantitative PCR

For analysing gene expression by qPCR, RNA was isolated from samples using the Direct-zol RNA Miniprep Kit (R2050, Zymo Research). cDNA was generated using 100–200 ng RNA per reaction and reverse transcriptase (18090050), dNTP mix (FERR0192), random primers (48190011) and RT reaction buffer (18090050B), all purchased from Thermo Fisher. qPCR was performed using Power Track SYBR Green Master Mix (A46109, Applied Biosciences). Genes of interest were normalized to RPL13A. Primers are listed in Supplementary Table 2.

Molecular docking

Molecular 3D models of human PRMT1 (UniProt ID Q99873) interacting with human PFKM (UniProt ID P08237) were constructed using the FASTA amino acid sequence for each as input for AlphaFold3. The PFK–PRMT1 predicted complex with the closest interaction with the PFK C-terminal tail was used for all models. For electrostatic potential calculations, proteins were protonated and pKas were calculated with PDB2PQR. Subsequently, the electrostatic potential was produced using ABPS. The output ‘.pqr’ and ‘.dx’ files were used to show the surface electrostatic potential in UCSF ChimeraX. A colour scale bar of the surface potential ranged from −10 to +10 kT e⁻¹. The FASTA sequence for each PDB entry was used to generate AlphaFold3 models of PFKM in complex with human VPS4 (PDB 1XWI). All models were visualized and coloured in ChimeraX.

Single-cell and single-nucleus RNA sequencing

To analyse the single-cell and single-nucleus muscle gene expression profiles, datasets were imported into Seurat (v.5). Cell types were visualized using uniform manifold approximation and projection, overlaying key gene expression data to examine spatial distribution within skeletal muscle. Pathway-specific genes were curated into metabolic categories, including the PPP, glycolysis and the TCA cycle, for downstream analysis. Track plots of the specific genes were produced in Python (Scanpy v.1.11.4, AnnData v.0.12.1, NumPy v.2.2.6, Pandas v.2.3.1, Matplotlib v.3.10.5). For each cell type, bicorrelation coefficients were calculated between selected metabolic genes, including PFKM and LAMP1, as well as pathway-specific genes. Correlation values were visualized in a pathway heatmap using pheatmap, with statistically significant relationships annotated. Pseudotime analysis was performed on a subset of data containing MuSC, slow-twitch myofibre and fast-twitch myofibre cell types, with 10% down-sampling. Pseudotime trajectories were reconstructed with the DDRTree algorithm in Monocle2 (ref. ³⁶), and cells were ordered using the orderCells function. To orient pseudotime from MuSC, the MuSC-enriched state was set as the root. Gene expression dynamics along the trajectories were visualized with plot_cell_trajectory.

Bulk-cell RNA sequencing

Sequencing done in cultured human skeletal muscle cells throughout differentiation was previously published and deposited in the database of Genotypes and Phenotypes (dbGaP, phs002554.v2.p1)³⁷.

Seahorse flux analysis (Mito Stress Test)

Cultured human muscle cells were washed and incubated in Seahorse assay media (XF DMEM with 10 mM glucose, 2 mM glutamine and 2 mM sodium pyruvate). Cells were equilibrated in a non-CO₂ incubator at 37 °C for 1 h before basal measurements were taken. OCR and ECAR were measured at baseline, followed by sequential addition of 1.5 μM oligomycin, 2.5 μM BAM15 and 5 μM rotenone–antimycin A. Data were collected in triplicate with the Seahorse XF96 Extracellular Flux Analyzer (Agilent). Data were normalized to protein concentration and analysed using Seahorse Wave software (v.2.6.4; Agilent). Protein concentration was obtained through a standard BCA assay, in which cells were lysed in 20 μl of 0.1% Triton X-100 Tris pH 7.4, followed by the addition of 200 μl of BCA reagent. Equations for ATP production calculations were derived from Agilent protocols and are as follows: mitoATP production rate (pmol ATP min⁻¹) = OCR_ATP × 2 × P/O ratio, where the P/O ratio is the number of molecules of ATP synthesized per atom of oxygen reduced by an electron pair, and OCR_ATP (pmol O₂ min⁻¹) = OCR_basal – OCR_oligo.

Imaging quantification of colocalization

Lysosomal analyses using immunofluorescence microscopy were performed using standard protocols³¹. For colocalization analyses, bona fide puncta were defined based on intensity and size using a consistent thresholding method with background signal subtraction. Colocalization was defined as the presence of bright puncta in multiple channels that spatially overlapped and existed within the same subcellular region in the same cell. Puncta overlap was manually validated by counting the number of puncta present across multiple channels (all channels of interest bright and punctate).

Imaging quantification of muscle cell differentiation

Muscle differentiation parameters were quantified by using previously published strategies^41,54. Each image at ×20 and ×40 was considered a field of view, and images at ×10 were subdivided into four regions of interest (ROIs) for quantification. For analyses quantifying myogenin expression, myogenin⁺ nuclei were defined based on intensity and size using a consistent thresholding method with background signal subtraction. Myogenin⁺ nuclei were quantified by counting nuclei exhibiting overlap with the myogenin fluorescence signal of nuclear size and morphology; DAPI staining was used to identify nuclei. All colocalization counts were normalized to cell or nuclei count per image or ROI. For mean fluorescence intensity quantifications of Myh1 and MyHC, the mean fluorescence intensity was measured for each field of view or ROI and then normalized to the average of the control values.

Imaging quantification of fusion index

Muscle fusion index values were quantified by following established methodologies^54,58. Parameter 1 assessed differentiation through the proportion of contacting nuclei relative to the total nuclei in a field of view or ROI; more than two nuclei directly touching within the same cell were considered to be in contact. Nuclei were visualized through DAPI staining with normalized threshold across all conditions. Parameter 2 assessed differentiation through the number of nuclei in MyHC⁺ myotubes per total nuclei count in a field of view or ROI; myotubes were defined as any cell with two or more nuclei and MyHC immunofluorescence intensity above the normalized threshold.

Statistical analyses

GraphPad Prism (v.10) was used for statistical analyses. A two-sided, unpaired Student’s t-test was used to compare variables between two groups (d.f. = n₁ + n₂ − 2). Multiple unpaired t-tests were done with the Šidák–Bonferroni correction (α = 0.05) to control for false discovery rates. Ordinary one-way ANOVA with Bartlett’s correction or the Geisser–Greenhouse correction and Bonferroni’s or Dunnett’s post hoc tests were performed as necessary to compare three or more groups against each other (d.f. = n − 1). Two-way ANOVA (α = 0.05) with Bonferroni’s or Dunnett’s post hoc tests were performed for experiments testing multiple variables. Data distribution was assumed to be normal, but this was not formally tested. Outliers were removed using the ROUT method on GraphPad Prism, where Q = 1%. Sample sizes were not predetermined and were based on sample sizes commonly used in previous studies^{26,28–30,54}.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

L.V.A. supervised the study. M.C. and L.V.A. conceptualized the study and wrote the paper. L.V.A., M.C., S.T.N., X.K., R.L.W., A.G., C.N.F., L.J.S., A.R.L.S. and K.Y. designed and performed cell biology experiments. L.V.A., M.C., Y.Y., E.A.H., M.K. and D.N. designed and performed metabolic analyses. J.E.C. and R.M. conducted molecular docking. L.V.A., M.C., R.L.W., C.R.L., S.D., Y.L., M.Z., K.S., M.M.S. and M.V.P. analysed data. L.V.A. and K.Y. secured funding.

Peer review

Peer review information

Nature Metabolism thanks the anonymous reviewers for their contribution to the peer review of this work. Primary Handling Editors: Alfredo Gimenez-Cassina and Jean Nakhle, in collaboration with the Nature Metabolism team.

Data availability

All source data reporting the findings of this paper, such as uncropped and unprocessed blots, sequences of generated plasmids and numerical source data, are included as Supplementary Information. Public databases used for sequencing analyses are cited as used in the article and include the Tabula Sapiens Consortium (https://tabula-sapiens.sf.czbiohub.org), the Human Skeletal Muscle Aging Atlas (https://www.muscleageingcellatlas.org), the Genotype-Tissue Expression (GTEx) Portal (https://gtexportal.org/home) and the database of Genotypes and Phenotypes (dbGaP) (https://dbgap.ncbi.nlm.nih.gov/home). Source data are provided with this paper.

Code availability

To promote transparency and facilitate reproducibility, we have made the code for our single-cell muscle gene expression profile analyses publicly available on GitHub at https://github.com/mingqizh/muscle_metabolic_switching. This repository includes detailed scripts and all the raw data for analysing Tabula Sapiens skeletal muscle gene expression.

Competing interests

The authors declare no competing interests.

Extended data

is available for this paper at 10.1038/s42255-026-01457-4.

Supplementary information

The online version contains supplementary material available at 10.1038/s42255-026-01457-4.