PFKM acts as a metabolic switch initiating donut-shaped mitochondrial remodeling to promote osteogenesis

Abstract

Osteogenic differentiation requires sophisticated mitochondrial adaptation to meet bioenergetic demands, yet regulatory checkpoints governing this organelle reorganization remain poorly defined. Through single-cell RNA sequencing reanalysis and metabolic intervention, this study unveils a non-canonical signaling role for PFKM, traditionally recognized solely as a glycolytic enzyme, in orchestrating mitochondrial remodeling during bone formation. Beyond its established metabolic function, Pfkm suppression triggers distinctive donut-shaped mitochondria through a novel signaling cascade. Mechanistically, Pfkm knockdown expands mitochondria-endoplasmic reticulum contacts (MERCs), facilitating mitochondrial calcium influx. Concomitantly, elevated CD38 suppresses protein kinase A (PKA) activity, inducing DRP1 dephosphorylation at Serine 656. This signaling integration promotes DRP1 mitochondrial translocation, driving the characteristic donut architecture. This structural transformation initiates comprehensive mitochondrial quality control (MQC) encompassing enhanced biogenesis, selective mitophagy, and mitochondrial-derived vesicles (MDVs) secretion, collectively optimizing the osteogenic microenvironment and cellular mineralization capacity. In vivo validation demonstrates that AAV-mediated Pfkm knockdown accelerates bone repair in rat calvarial and femoral defect models. This work establishes PFKM as a dual-function regulator bridging metabolism and mitochondrial signaling, offering a potent therapeutic avenue for bone regeneration.

Introduction

Bone marrow-derived mesenchymal stem cells (BMSCs) are central to regenerative medicine, owing to their multilineage differentiation potential governed by microenvironmental cues.^1,2 Among these fates, osteogenic differentiation is highly biosynthetically intensive, demanding a fundamental metabolic reprogramming to fulfill the energetic and structural requirements of matrix synthesis and mineralization.^3,4 While undifferentiated BMSCs rely predominantly on glycolysis to limit oxidative stress and protect genomic stability,^5–8 lineage commitment triggers a metabolic shift toward oxidative phosphorylation (OXPHOS) to fuel osteogenic transcription factors.^9–13 However, this metabolic transition inherently elevates mitochondrial reactive oxygen species (mtROS),¹⁴ which can impair differentiation if left unchecked.^15–17 To survive this oxidative surge, stem cells rely on the activation of the mitochondrial quality control (MQC) system.^18–24

As an all-around defense mechanism, MQC protects the cell from the inside out. Internally, this system maintains a tight balance between mitochondrial dynamics and biogenesis to reshape the network and replenish the functional pool, while simultaneously utilizing mitophagy to clear out damaged parts.^25–30 Moving beyond the cell, MQC extends its reach by releasing mitochondrial-derived vesicles (MDVs) into the surrounding space, which helps regulate its own mitochondrial pool and indirectly influences other related cellular functions.^31–34 While these different parts of MQC are known to keep cells healthy, how stem cells manage to coordinate this entire network alongside the initial metabolic shift remains a major unanswered question.

To achieve such a comprehensive adaptation, stem cells possess an intrinsic mechanism capable of sensing these metabolic shifts and transducing them into structural mitochondrial reorganization. Emerging evidence suggests that mitochondria-endoplasmic reticulum contact sites (MERCs) serve as this critical hub. By maintaining calcium homeostasis at these interfaces, MERCs can convert metabolic cues into localized calcium fluxes, which directly govern downstream signaling and dictate mitochondrial dynamics.³⁵

To identify metabolic checkpoints during stem cell osteogenesis and elucidate how they orchestrate the cellular network in response to metabolic shifts, we integrated single-cell RNA sequencing (scRNA-seq) reanalysis with functional metabolic screening. The result revealed that glycolytic suppression is a prerequisite for the BMSC-to-osteoblast transition, with pseudotime trajectory analysis specifically highlighting the progressive downregulation of the phosphofructokinase (Pfk) family, which catalyzes the rate-limiting step of glycolysis. Subsequent metabolic modulation pinpointed phosphofructokinase, muscle type (Pfkm) as exhibiting a distinctive response profile intimately linked to mitochondrial pathways. This prompted us to investigate whether PFKM serves as a regulatory node orchestrating both metabolic reprogramming and mitochondrial adaptation to drive osteogenic differentiation.

Here, we demonstrate that Pfkm suppression functions as a metabolic checkpoint that initiates osteogenic commitment through a highly coordinated mitochondrial reorganization cascade. Mechanistically, Pfkm knockdown drives a physical expansion of MERCs, enhancing localized mitochondrial calcium influx. Simultaneously, it elevates CD38 expression to suppress protein kinase A (PKA) signaling. These twin biophysical and biochemical pathways converge to promote DRP1 dephosphorylation at Ser656, directly catalyzing the structural emergence of specialized, donut-shaped mitochondria. Far from being a mere structural change, this architectural transformation serves as the engine driving the broader MQC cascade. By harmonizing mitochondrial biogenesis, mitophagy, and Mito/MDVs production, this process refines cellular fitness and establishes a permissive microenvironment for mineralization. In vivo validation using AAV-mediated Pfkm knockdown in rat calvarial and femoral defect models demonstrates robust acceleration of bone regeneration. Collectively, our findings position PFKM as a pivotal regulatory node linking metabolic reprogramming with mitochondrial remodeling to govern osteogenic fate determination and bone repair.

Results

Metabolic reprogramming of the osteogenic lineage progression during bone healing

To map the metabolic changes during bone regeneration, we first analyzed scRNA-seq data from murine fracture and control tissues using uniform manifold approximation and projection (UMAP)-based dimension reduction.³⁶ This analysis identified different clusters representing bone resident and infiltrating cell clusters (Figs. 1a and S1a), including: chondrocytes, osteoblast lineage cells (OBs), endothelial cells (ECs), bone marrow mesenchymal stem cells (BMSCs), mast cells (MCs), hematopoietic cells (HCs), smooth muscle cells (SCs), megakaryocytes, and neutrophils. Within the combined OBs and BMSCs subclusters, differentially expressed gene (DEG) and Gene Ontology (GO) enrichment analyses revealed that fracture injury significantly activates energy metabolism pathways, with a pronounced enrichment in mitochondrial ATP synthesis (Fig. S1b, c).

Fig. 1.

Single-cell RNA sequencing reveals metabolic reprogramming of the osteoblast lineage during bone fracture healing. a Uniform Manifold Approximation and Projection (UMAP) plot showing the distribution of single cells from mouse fracture and control bone samples. b Heatmap depicting glycolysis and oxidative phosphorylation (OXPHOS) pathway densities across different cell clusters, calculated using single-sample gene set enrichment analysis (ssGSEA). c t-distribution of random neighborhood embedding (t-SNE) of BMSCs and OBs cell subclusters in fracture and control samples. d Re-dimensionality reduction clustering and t-SNE distribution of osteoblast lineage clusters (including BMSCs and OBs clusters) in fracture and control samples. e Glycolytic pathway score of osteoblast lineage clusters along pseudotime trajectories in the fracture and control samples. f Expression patterns of glycolytic and osteogenic markers

Notably, UMAP density plotting showed that glycolysis was mainly limited to mature lineages like chondrocytes and OBs, whereas OXPHOS showed a much wider distribution that included the BMSCs cluster (Fig. 1b). Compared with the control group, the fracture injury group displayed higher OXPHOS scores across all cellular subclusters. Importantly, the upregulation of OXPHOS scores outpaced that of glycolysis, with this trend being most prominent in BMSCs (Fig. S1d). Concurrently, re-clustering of the osteogenesis-related cell subclusters (BMSCs and OBs) revealed a marked expansion in the proportions of Fkbp10+ and Bglap + OBs following injury (Figs. 1c, d and S1e). To dissect the temporal dynamics governing this metabolic shift during osteogenesis, we reconstructed the BMSC-to-OB developmental continuum using Monocle3 pseudotime analysis (Fig. S1f). This trajectory unmasked a divergence: while control samples maintained stable metabolic baselines, fractured samples exhibited a sharp, progressive decline in glycolytic activity along the differentiation timeline (Fig. 1e). Crucially, this attenuation of glycolytic flux coincided with a pronounced downregulation of the glycolysis-related gene Pfk family expression during the transition from BMSCs to mature osteoblasts, signaling that dampening glycolysis is essential to favor the OXPHOS-driven energetic demands required for lineage progression and bone healing (Fig. 1f).

PFKM functions as a metabolic checkpoint governing the glycolysis-to-OXPHOS shift and osteogenic differentiation

To validate whether the shift from glycolysis to OXPHOS is functionally required for BMSCs differentiation, we first treated rat BMSCs (rBMSCs) with metabolic inhibitors during osteogenic induction. Blocking mitochondrial ATP synthase with oligomycin A (1 and 5 µmol/L) significantly suppressed osteogenic differentiation, as evidenced by attenuated alkaline phosphatase (ALP) staining and activity at day 7 (Figs. 2a and S2a). Subsequent reverse transcription quantitative polymerase chain reaction (RT-qPCR) analysis revealed that oligomycin A markedly upregulated key glycolysis-related genes, notably hexokinase 2 (Hk2), Pfkm and phosphofructokinase, liver type (Pfkl) (Fig. S2b). These data indicate that while rBMSCs attempt to survive mitochondrial impairment via a compensatory shift toward glycolysis, this forced glycolytic state is incompatible with normal osteogenesis.

Fig. 2.

Identification of PFKM as a metabolic checkpoint and its functional validation in promoting rBMSCs osteogenesis and metabolic reprogramming. a Alkaline phosphatase (ALP) staining of rBMSCs treated with oligomycin A (1 or 5 µmol/L) for 7 days under osteogenesis (n = 3). b ALP staining of rBMSCs treated with 2-DG (200 µmol/L) for 7 days under osteogenesis (n = 3). ALP staining (day 7) (c), and Alizarin Red S (ARS) staining (day 14) (d) in siPfkm and siNC groups under osteogenic induction (n = 3). qRT-PCR (e) and western blot (f) analysis of osteogenic-related markers in siPfkm and siNC groups under osteogenesis (n = 3). Oxygen consumption rate (OCR) analysis (g), glycolytic proton efflux rate (PER) analysis (h) in siPfkm and siNC groups (n = 3). OI osteogenic induction (0, 3, 7 days) ALP staining of Pfkm-knockdown and control rBMSCs subjected to metabolic inhibition with oligomycin A (i) or 2-DG (j) after 7 days of osteogenesis (n = 3). Data represent mean ± SD. *P < 0.05, **P < 0.01, and ***P < 0.001 by Student’s t test (e, g and h)

Given that an elevated glycolytic baseline correlated with impaired bone formation, we next tested whether direct pharmacological inhibition of glycolysis could conversely accelerate differentiation. Paradoxically, the glycolysis inhibitor 2-DG (200 µmol/L) also significantly blocked osteogenesis (Figs. 2b and S2c). While 2-DG treatment downregulated most glycolytic genes, Pfkm expression remained notably unchanged (Fig. S2d). This selective responsiveness suggested that PFK, rather than the typical coordinated glycolytic network, might act as a specialized metabolic checkpoint during osteogenesis.

Baseline expression analysis showed that Pfkm and platelet type (Pfkp) levels were comparable levels, both being lower than Pfkl (Fig. S2e). To systematically compare the functional roles of these PFK isoforms, we utilized small interfering RNAs (siRNAs) to individually target Pfkl, Pfkm, and Pfkp, selecting the sequences with the highest knockdown efficiencies for downstream validation (Fig. S2f). ALP assays revealed that the knockdown of Pfkm exerted a significantly superior pro-osteogenic effect compared to the NC group, Pfkl knockdown, Pfkp knockdown, or their combination (Figs. 2c and S2g, h). CCK-8 assays confirmed that this enhanced differentiation was not a confounding artifact of altered cell proliferation (Fig. S2i). Furthermore, Pfkm knockdown markedly accelerated extracellular matrix mineralization and advanced osteogenic differentiation, as demonstrated by Alizarin Red S (ARS) staining at day 14 (Figs. 2d and S2j), which correlated with a robust upregulation of key osteogenic markers at both the mRNA and protein levels (Fig. 2e, f). Ultimately, the distinctive regulatory behavior of PFKM prompted us to investigate its role in greater depth.

To characterize the regulatory role of PFKM in rBMSC metabolism, we assessed cellular respiratory and glycolytic flux during osteogenic induction using Seahorse XF technology. Pfkm knockdown triggered a robust enhancement in mitochondrial oxidative metabolism, evidenced by significantly elevated oxygen consumption rates (OCR). Compared with controls, siPfkm-treated cells exhibited higher basal respiration, maximal respiratory capacity, ATP production, and spare respiratory capacity, with this metabolic surge being most prominent at day 3 of osteogenic induction (Fig. 2g). Concurrently, we evaluated glycolytic function by monitoring the proton efflux rate (PER) and found that Pfkm knockdown markedly restricted glycolytic flux. Both basal and compensatory glycolysis were significantly suppressed in the knockdown group (Fig. 2h). Crucially, this PFKM-directed metabolic rewiring was functional, as Pfkm knockdown partially rescued the osteogenic suppression originally induced by oligomycin A and 2-DG treatments (Figs. 2i, j and S2k, l). Collectively, these metabolic profiles confirm that PFKM functions as a metabolic “brake” on osteogenesis. Since osteogenic differentiation is inherently coupled with elevated OXPHOS, the targeted suppression of Pfkm significantly accelerates this differentiation while concurrently boosting OXPHOS. This result indicates that PFKM enhances osteogenesis by acting as a critical metabolic switch that drives the bioenergetic shift in rBMSCs.

Pfkm knockdown initiates mitochondrial donut-shaped formation

To investigate the structural and functional consequences of the Pfkm-mediated metabolic switch, we next evaluated the functional state of the mitochondrial network during osteogenesis. Compared to the control group, the Pfkm-knockdown group exhibited significantly lower mtROS levels throughout differentiation (days 0, 3, 7, and 14), whereas their mitochondrial membrane potential (ΔΨm) remained stable and comparable between groups (Fig. S3a, b). This suppressed oxidative stress correlated with a concurrent upregulation of uncoupling protein 2 (Ucp2) expression, suggesting that enhanced mitochondrial uncoupling serves as an endogenous protective mechanism to mitigate mtROS production without compromising membrane potential during lineage progression (Fig. S3c).

Beyond these biochemical alterations, MitoTracker staining revealed that Pfkm knockdown triggered a striking, time-dependent structural remodeling of the mitochondrial network. Specifically, the mitochondrial reticulum transitioned from an elongated, interconnected network into fragmented, punctate forms by days 3 and 7 (Fig. 3a). Transmission electron microscopy (TEM) corroborated this morphological shift, capturing an increased frequency of mitochondrial fission events accompanied by a concomitant reduction in cristae density within Pfkm-knockdown group (Fig. 3b). To precisely characterize this fragmentation, we employed structured illumination microscopy (SIM), which confirmed that these punctate structures were specialized donut-shaped mitochondria. Notably, while cristae were clearly discernible in control cells, they were largely obscured in the donut-shaped mitochondria of knockdown cells, indicating a profound reorganization of the mitochondrial ultrastructure (Fig. 3c). This structural distinction indicates a targeted ultrastructural adaptation rather than generalized, pathological mitochondrial damage.

Fig. 3.

Pfkm knockdown triggers donut-shaped mitochondrial remodeling during rBMSCs osteogenesis. a Confocal images of rBMSCs stained with MitoTracker Deep Red (left) and quantitative analysis of mitochondrial aspect ratio (right) at days 0, 3, 7, and 14 of osteogenic induction in siPfkm and siNC groups (n = 3). OI osteogenic induction. Ultrastructural analysis. Transmission electron microscopy (TEM) images (red arrows indicate constricted mitochondria) (b) and structured illumination microscopy (SIM) images of siPfkm and siNC groups staining with PK Mito Red (red arrows indicate donut-shaped mitochondria) (c) following 3 days of osteogenesis (n = 3). Scale bars are as indicated. Data represent mean ± SD. *P < 0.05, **P < 0.01, ***P < 0.001 by Student’s t test (b)

Given that Pfkm knockdown simultaneously accelerated osteogenesis and altered mitochondrial architecture, we next sought to clarify whether conventional mitochondrial fragmentation alone is sufficient to drive bone formation. We therefore treated rBMSCs with either the mitochondrial fission promoter BC1618 or the fusion inhibitor MYLS22. Although both pharmacological agents successfully forced mitochondrial fragmentation (Fig. S4a), neither treatment could replicate the robust pro-osteogenic outcomes triggered by Pfkm knockdown, yielding only limited improvements in the osteogenic differentiation of rBMSCs (Fig. S4b). This modest enhancement suggested that conventional mitochondrial fragmentation alone could not fully account for the superior pro-osteogenic outcomes observed in Pfkm-knockdown cells. Consequently, we reasoned that Pfkm knockdown may trigger a distinct morphological and functional remodeling beyond classical fission, a hypothesis we explored further in the next section.

The donut-shaped morphology facilitates a comprehensive MQC program involving mitochondrial biogenesis, mitophagy and Mito/MDVs secretion

Given that this “donut-shaped” transformation typically serves as a structural precursor to advanced organelle clearance, we next characterized the broader landscape of MQC beyond mere morphological remodeling. Analysis of mitochondrial DNA (mtDNA) copy numbers indicated a substantial accumulation in the knockdown group relative to controls (Fig. 4a). This accumulation was accompanied by elevated expression of peroxisome proliferator-activated receptor γ coactivator 1-alpha (PGC-1α), the master regulator of mitochondrial biogenesis (Fig. 4b).

Fig. 4.

Donut-shaped mitochondria accompany mitochondrial biogenesis and mitophagy during osteogenic differentiation. a Relative mtDNA copy number in siPfkm and siNC groups at days 0, 3, 7, and 14 of osteogenesis (n = 3). b Western blot analysis of PGC-1α protein expression in siPfkm and siNC groups at days 0, 3, 7, and 14 of osteogenesis (n = 3). c Western blot analysis of autophagy and mitochondrial markers in siPfkm and siNC groups at days 0, 3, 7, and 14 of osteogenesis (n = 3). d TEM images showing mitophagosomes in the Pfkm-knockdown group after 3 days of osteogenesis (n = 3). e Confocal images (left) and fluorescence line profile analysis (right) of LC3B (Red)-mitochondria (Green) co-localization in siPfkm and siNC groups after 3 days of osteogenesis (n = 3). Scale bars are as indicated. Data represent mean ± SD. *P < 0.05, **P < 0.01, and ***P < 0.001 by Student’s t test (a)

To confirm whether accelerated degradation occurs in tandem with this enhanced biogenesis, we evaluated key mitophagy markers during osteogenic induction. Pfkm-knockdown cells exhibited elevated LC3B-II levels alongside a significant clearance of TOMM20 and P62, indicating increased formation of mitochondrial autophagosomes and accelerated lysosomal degradation (Fig. 4c). The induction of mitophagy upon Pfkm knockdown was further confirmed by TEM imaging, which revealed distinct double-membrane structures engulfing mitochondria (Fig. 4d). Furthermore, immunofluorescence revealed a significant increase in LC3B-mitochondria colocalization, providing robust evidence that Pfkm knockdown triggers heightened mitophagy (Fig. 4e). To investigate whether this accelerated mitochondrial turnover is directly required for the pro-osteogenic phenotype, we treated Pfkm-knockdown cells with Mdivi-1, a dual inhibitor of mitochondrial fission and mitophagy, during osteogenic induction. Mdivi-1 treatment successfully attenuated the heightened mitophagy flux triggered by Pfkm knockdown, as indicated by reduced mitochondrial-LC3B fluorescence colocalization, decreased LC3B-II levels, and a rescue of TOMM20 expression (Fig. S5a, b). Morphological analysis via mitochondrial staining further revealed that Mdivi-1 largely reverted the fragmented, donut-shaped mitochondria back to an elongated, reticular network (Fig. S5c). Crucially, this pharmacological inhibition of fission and mitophagy significantly reversed the pro-osteogenic phenotype of Pfkm-knockdown rBMSCs, resulting in a substantial reduction in ALP activity and ARS-stained mineralized matrix (Fig. S5d). Additionally, the autophagy inhibitor 3-methyladenine (3-MA) was applied to further validate these findings. Consistent with the Mdivi-1 results, 3-MA treatment successfully restored mitochondrial morphology to an interconnected network structure, which was accompanied by a rescue of TOMM20 expression and a concomitant decrease in LC3B-II levels. Ultimately, this pharmacological disruption of the mitophagy program also suppressed the pro-osteogenic effects originally exerted by Pfkm knockdown (Fig. S5e, f, g).

Given the observed mitochondrial fragmentation and turnover, we next investigated whether these cells released mitochondrial components into the extracellular space. Following the isolation protocol schematically illustrated (Fig. 5a), we extracted particulate components from the osteogenic induction medium. Subsequent MitoTracker Green staining of the precipitate revealed a greater abundance of extracellular mitochondrial structures—encompassing both larger fragments and smaller Mito/MDVs—in the experimental group, with total particle diameters ranging from 0.5 to 5 μm (predominantly 0.5–1 μm) (Figs. 5b and S6a). Consistently, western blot analysis showed enriched expression of the mitochondrial markers COX IV and TOMM20 in the experimental group (Fig. 5c). TEM revealed MDVs diameters predominantly ranging from 70 to 100 nm, consistent with previous reports,³⁷ and confirmed their increased abundance in the Pfkm-knockdown group (Fig. 5d). To test whether these secreted mitochondrial structures possess functional bioactivity, we performed a co-culture assay. Stably expressing mitochondria-targeted GFP rBMSCs were knocked down for Pfkm, and their secreted Mito/MDVs were collected and co-cultured with naive mCherry-expressing rBMSCs for 24 h. Confocal imaging confirmed that the recipient cells successfully internalized these green-fluorescent mitochondrial particles (Figs. 5e and S6b). Remarkably, the internalization of these Pfkm-knockdown Mito/MDVs significantly enhanced the baseline osteogenic differentiation potential of the naive recipient cells (Fig. 5f). Collectively, our findings illustrated that Pfkm knockdown promotes osteogenic differentiation by triggering a comprehensive remodeling of MQC, impacting mitochondrial dynamics (fission/fusion), biogenesis, mitophagy, and the secretion of functional Mito/MDVs.

Fig. 5.

Donut-shaped mitochondria mediate the secretion of functional Mito/MDVs, facilitating osteogenic differentiation in recipient cells. a Schematic illustration of the procedure for isolating secreted Mito/MDVs from rBMSC conditioned medium. Confocal images (b) and western blot analysis (c) of the isolated Mito/MDVs from siPfkm and siNC groups (n = 3). d TEM images of isolated Mito/MDVs (green arrows: double-membrane MDVs; yellow arrows: single-membrane MDVs; red arrows: mitochondria) (n = 3). e 3D confocal images showing recipient rBMSCs internalizing GFP-tagged Mito/MDVs from virus-transduced donor cells after 24 h co-culture (n = 3). f ALP (day 7) and ARS (day 14) staining with respective quantitative analyses in rBMSCs treated with isolated Mito/MDVs (n = 3). Scale bars are as indicated. Data represent mean ± SD. *P < 0.05, **P < 0.01, and ***P < 0.001 by Student’s t test (f)

DRP1 dephosphorylation at Ser656 drives donut-shaped mitochondrial remodeling

To elucidate the precise molecular machinery driving the Pfkm-dependent mitochondrial morphological switch, we focused on the genes or proteins that orchestrate organelle dynamics. While mitochondrial fusion requires mitofusin 1/2 (MFN1/2) and optic atrophy 1 (OPA1), fission primarily depends on dynamin-related protein 1 (DRP1) and its corresponding outer membrane receptors. We first examined both total DRP1 expression and its site-specific phosphorylation status. Although total DRP1 levels remained unchanged, Pfkm knockdown specifically triggered a substantial, site-specific decrease in DRP1 phosphorylation at the Ser656 residue, while leaving the Ser635 site entirely unaffected (Figs. 6a and S7a).

Fig. 6.

PFKM regulates mitochondrial fission through DRP1 dephosphorylation at Ser656. a Western blot analysis of p-DRP1 (Ser656) in siPfkm and siNC groups at 0, 1, 3, and 5 h of osteogenesis (n = 3). b Confocal images (left) and fluorescence line profile (right) of DRP1 (Green)-mitochondria (Red) co-localization in siPfkm and siNC groups (n = 3). c Mitochondrial morphology in siPfkm-rBMSCs expressing the DRP1-S656D phosphomimetic mutant (n = 3). d Western blot analysis of the secreted Mito/MDVs from each group (n = 3). e ALP staining (day 7) and ARS staining (day 14) in siPfkm or siNC-rBMSCs expressing DRP1-WT or the DRP1-S656D phosphomimetic mutant (n = 3).Protein levels of p-PKA substrate and p-PKA C (Thr197) (0, 1, 3 and 5 h) in siPfkm-rBMSCs (f), and in Forskolin-treated siPfkm-rBMSCs at 3 h (g) of osteogenesis. h Western blot analysis of p-DRP1 (Ser656) in siPfkm cells with 10 μmol/L forskolin after 3 h of osteogenesis (n = 3). Scale bars are as indicated

To determine how this specific dephosphorylation event influences DRP1 kinetics, we established cells stably expressing fluorescently tagged DRP1. Pfkm knockdown dramatically accelerated DRP1 translocation from the cytosol to mitochondria, as reflected by significantly enhanced mitochondrial-DRP1 colocalization (Fig. 6b). To definitively test whether the phosphorylation state of Ser656 directly dictates these structural shifts, we generated a phosphomimetic DRP1 variant (S656D-EGFP), substituting serine with aspartate to mimic constitutive phosphorylation (Fig. S7b, c). Remarkably, introducing the S656D mutation effectively blocked DRP1 recruitment to the organelle surface, suppressed the formation of donut-shaped mitochondria, and completely restored the interconnected tubular network that had been disrupted by Pfkm knockdown (Figs. 6c and S7d).

We next evaluated whether this DRP1-mediated remodeling program is structurally tied to extracellular vesicle release. Western blot analysis of mitochondrial pellets isolated from culture supernatants revealed that expressing the S656D mutant in Pfkm-knockdown cells substantially diminished extracellular levels of COX IV and TOMM20, proving that Ser656 dephosphorylation-induced fission is critical for the secretion of functional Mito/MDVs (Fig. 6d). Furthermore, introducing the S656D mutation into Pfkm-knockdown cells successfully reversed the Pfkm knockdown-induced alterations by decreasing both the elevated mtDNA copy numbers and the heightened mitophagy flux (Fig. S7e, f). Crucially, functional assays, including ALP and ARS staining, subsequently showed that the S656D mutation partially suppressed the pro-osteogenic effects of Pfkm knockdown (Figs. 6e and S7g).

Finally, we traced the upstream mechanism driving DRP1 dephosphorylation. Given that protein kinase A (PKA) directly phosphorylates DRP1 at Ser656,^23,38 we measured PKA activity in Pfkm-knockdown cells. As expected, Pfkm knockdown markedly reduced PKA activity, but this was restored by the cyclic adenosine monophosphate (cAMP) activator forskolin (Fig. 6f, g). Concurrently, forskolin also reversed the DRP1 dephosphorylation at Ser656 caused by Pfkm knockdown, establishing that Pfkm regulates DRP1 Ser656 phosphorylation through the PKA activity (Fig. 6h).

Enhanced MERCs formation drives calcium-dependent mitochondrial fission to promote osteogenic differentiation

To unravel the molecular mechanisms underlying the enhanced osteogenic phenotype induced by Pfkm knockdown, we performed RNA-seq analysis. Phenotypic tracking via ALP staining and activity assays identified day 3 of induction as the onset of significant divergence between the experimental and control groups (Fig. S8a). RNA-seq analysis at this critical time point revealed 795 differentially expressed genes between Pfkm-knockdown and control groups, among which Cd38 (ADPRC 1) emerged as one of the most strongly upregulated genes (Fig. S8b). Consistently, Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway and Gene Set Enrichment Analysis (GSEA) indicated a significant enrichment in the calcium signaling pathway and calcium channel regulator activities, both of which are closely linked to CD38 function (Fig. S8c, d). This robust upregulation was further validated at both the mRNA and protein levels via RT-qPCR and western blotting (Fig. S9a, b).

To determine whether CD38 acts as the primary functional bridge between PFKM and downstream phenotypes, we pharmacologically inhibited CD38. Notably, administration of a CD38 inhibitor significantly attenuated the pro-osteogenic effects originally enhanced by Pfkm knockdown, as demonstrated by diminished ALP staining and activity (Fig. S9c). Corroborating these functional shifts, mitochondrial morphology tracking revealed that the donut-shaped morphology induced by Pfkm knockdown was largely restored to an interconnected reticular network upon CD38 inhibitor 1 treatment (Fig. S9d). This rescue was further accompanied by restored PKA activity and increased Drp1 Ser656 phosphorylation (Fig. S9e, f). Collectively, these findings establish CD38 as a central downstream mediator coupling PFKM-directed metabolic signaling to mitochondrial architecture and subsequent osteogenic commitment.

To elucidate how CD38-mediated remodeling of mitochondrial architecture drives osteogenesis, we analyzed downstream Gene Ontology (GO) terms. GO enrichment analysis highlighted prominent alterations in terms related to calcium ion binding, the endoplasmic reticulum (ER), and mitochondria (Fig. S10a). Driven by these structural and organelle clues, we systematically examined key regulators of ER-mitochondria communication. Western blot analysis demonstrated sustained elevation of CD38 and phosphodiesterase 1A (PDE1A) in the Pfkm-knockdown group throughout osteogenesis (days 0, 3, 7, and 14) (Fig. S10b). Consistent with the above result, RT-qPCR showed upregulation of mRNA expression for genes associated with ER and mitochondrial calcium channels, including Ip3r1 (inositol 1,4,5-trisphosphate receptor type 1), Pde1a, Cd38, Ryr1 (ryanodine receptor 1), Mcu (mitochondrial calcium uniporter), Micu1 (mitochondrial calcium uptake 1), and Vdac1 (voltage-dependent anion-selective channel 1), as well as genes governing MERCs, such as Esyt1 (extended synaptotagmin-like protein 1), Grp75 (glucose-regulated protein 75), and Pacs2 (phosphofurin acidic cluster sorting protein 2) (Fig. S10c).

We next sought to verify whether this transcriptional program translated into physical alterations at the organelle interface. Dual fluorescence probing of the ER and mitochondria demonstrated a significantly higher co-localization ratio in the knockdown group, indicating enhanced MERCs formation (Fig. 7a). To surpass the resolution and diffraction limits of fluorescence microscopy, we employed TEM to precisely visualize the nanoscale gap of these contact sites (typically 10–50 nm). TEM imaging confirmed a significantly narrowed inter-organelle distance (Fig. 7b) and an increased frequency of MERC structures in Pfkm-knockdown cells (Fig. 7c).

Fig. 7.

MERCs-mediated mitochondrial calcium influx promotes osteogenesis in Pfkm-knockdown rBMSCs. a SIM images showing the co-localization of mitochondria (Red) and the ER(Green) in siPfkm and siNC groups at day 3 of osteogenic induction (n = 3). b TEM images (left) and quantitative analysis (right) of the distance between the ER and mitochondria in siPfkm and siNC groups after 3 days of osteogenic induction (n = 3). c TEM images (left) and quantification of MERCs (right) in siPfkm and siNC groups after 3 days of osteogenic induction (n = 3). Mitochondrial Ca²⁺ dynamics assessed by Rhod-2AM (d) and quantification of peak fluorescence intensity (e) during osteogenic induction (D0, D1, D3) (n = 6). f Western blot analysis of phosphorylated DRP1 levels in Pfkm-knockdown treated with MCU-i4 after 3 h of osteogenic induction (n = 3). g ALP staining (day 7) and ARS staining (day 14) in siPfkm-rBMSCs treated with MCU-i4 (n = 3). Scale bars are as indicated. Data represent mean ± SD. *P < 0.05, **P < 0.01, and ***P < 0.001 by Student’s t test (b, c, e)

Given that MERCs serve as structural conduits for inter-organelle communication, we investigated whether this physical expansion facilitated enhanced ER-to-mitochondria Ca²⁺ transfer. Using Rhod-2 AM in a Ca²⁺-free medium, we monitored mitochondrial Ca²⁺ uptake and observed that the knockdown group exhibited significantly amplified mitochondrial Ca²⁺ flux at days 0, 1, and 3, peaking sharply at day 1 (Fig. 7d, e). To establish a direct causal link between this calcium influx and downstream mitochondrial remodeling, we sequentially blocked the pathway at two distinct regulatory nodes. Pharmacological inhibition of the upstream driver using CD38 inhibitor 1, or the downstream channel using the specific mitochondrial calcium uniporter inhibitor-MCU-i4, successfully blunted the elevated mitochondrial Ca²⁺ flux induced by Pfkm knockdown (Fig. S11a, b).

Crucially, MCU-i4 effectively reversed the Pfkm knockdown-induced dephosphorylation of DRP1 at Ser656 (Fig. 7f). Consistent with this biochemical rescue, MCU-i4 treatment reduced DRP1 recruitment to the mitochondria, successfully restoring mitochondrial morphology from fragmented, donut-shaped structures back to a tubular reticular network (Fig. S11c, d). Furthermore, western blot analysis revealed that MCU-i4 treatment also suppressed downstream mitophagy (Figure S11e). Finally, functional assays via ALP and ARS staining demonstrated that MCU-i4 successfully attenuated the pro-osteogenic phenotype triggered by Pfkm knockdown (Figs. 7g and S11f). Taken together, these data outline a complete mechanistic cascade whereby Pfkm knockdown drives MERCs expansion and Ca²⁺ influx to trigger DRP1-mediated fission and mitophagy, ultimately dictating osteogenic differentiation.

The ERK/c-Jun axis transcriptionally upregulates CD38 to drive mitochondrial remodeling and osteogenic differentiation

To uncover the precise upstream molecular cascade by which Pfkm knockdown initiates Cd38 expression, we turned our attention to transcriptional regulatory pathways. Previous studies have established that Cd38 transcription is regulated via functional Activator Protein 1 (AP-1) response elements located within its promoter region.^39–42 Utilizing the JASPAR database, we identified putative AP-1 binding sites within the rat Cd38 promoter, the core sequences of which are evolutionarily conserved across humans, mice, and rats (Fig. 8a). Given that c-Jun is a major functional subunit of the AP-1 complex, we examined its activation status to validate these predictions. Western blot analysis revealed increased phosphorylation of c-Jun, alongside elevated ERK1/2 phosphorylation, in Pfkm-knockdown cells (Fig. 8b, c). To establish a direct causal link between this activated MAPK/ERK signaling cascade and Cd38 transcription, cells were treated with the selective ERK1/2 inhibitor SCH772984. This pharmacological intervention significantly suppressed the Pfkm knockdown-induced upregulation of key downstream targets, including p-c-Jun, CD38, PDE1A, and p-DRP1 (Fig. 8d). Consequently, SCH772984 successfully rescued mitochondrial dynamics by restoring the reticular mitochondrial network and robustly reducing the recruitment of DRP1 to the mitochondria (Figs. 8e and S12). Western blot analysis further revealed that ERK1/2 inhibition following Pfkm knockdown also suppressed downstream mitophagy (Fig. 8f). Consistently, functional assays confirmed that SCH772984 treatment attenuated the pro-osteogenic phenotype caused by Pfkm knockdown, as evidenced by a substantial decrease in ALP staining at day 7 and ARS-stained mineralized matrix at day 14 (Fig. 8g). Collectively, these findings demonstrate that Pfkm knockdown upregulates CD38 expression through a dual mechanism involving c-Jun-mediated promoter activation and ERK1/2-driven transcript stabilization, thereby driving downstream mitochondrial fission, mitophagy, and the pro-osteogenic phenotype.⁴²

Fig. 8.

The ERK/c-Jun pathway mediates Pfkm knockdown-induced CD38 expression and osteogenic differentiation. a Motif enrichment analysis identifying transcription factor binding sites in CD38 promoter region. Western blot analysis of signaling pathway in siPfkm and siNC groups during osteogenic induction (0, 1, 3, and 5 h). Panels show the phosphorylation kinetics of c-Jun (b) and ERK (c), as well as pathway-related protein expression in the presence of the inhibitor SCH772984 (0.1 and 0.2 μmol/L) (d) (n = 3). e Confocal images of mitochondrial morphology and quantification of mitochondrial aspect ratio in siPfkm-rBMSCs treated with SCH772984 (0.1 and 0.2 μmol/L) during osteogenic induction (n = 3). f Western blot analysis of TOMM20 and LC3B protein expression in siPfkm-rBMSCs treated with SCH772984 (0.1 and 0.2 μmol/L) after 3 days of osteogenic induction (n = 3). g ALP staining (day 7) and ARS staining (day 14), with corresponding quantitative analyses in siPfkm-rBMSCs following SCH772984 treatment (0.1 and 0.2 μmol/L) and osteogenic induction (n = 3). Scale bars are as indicated. Data represent mean ± SD. *P < 0.05, **P < 0.01, and ***P < 0.001 by one-way ANOVA with Dunnett’s test vs. control (g)

AAV-mediated Pfkm suppression accelerates calvarial and femoral bone regeneration in vivo

To assess whether Pfkm knockdown could accelerate bone repair under physiological conditions, we utilized rat models featuring calvarial and femoral defects. Prior to evaluating bone healing, we validated the in vivo efficiency of adeno-associated virus (AAV)-mediated gene suppression. Four weeks post-injection, rBMSCs harvested from the treated animals showed a marked reduction in PFKM protein levels, establishing an optimal viral dosage of 2.0 × 10¹¹ viral genomes (vg) (Fig. S13a). Crucially, these ex vivo harvested cells fully recapitulated our prior in vitro mechanistic findings by exhibiting a fragmented, donut-shaped mitochondrial morphology (Fig. S13b), and displaying enhanced osteogenic capacity, as evidenced by ALP staining and activity (Fig. S13c).

Having confirmed successful gene knockdown and the subsequent activation of this mitochondrial-osteogenic axis ex vivo, we next evaluated the therapeutic efficacy of the intervention in vivo. In the calvarial defect model, rats received a local subcutaneous administration of AAV-shPfkm two weeks prior to surgery, followed by the implantation of a bilayer collagen membrane during the operation to cover the defects (Fig. S13d). Micro-CT analysis at four weeks post-surgery revealed markedly enhanced bone regeneration in the AAV-shPfkm group, characterized by significant increases in bone volume fraction (BV/TV), trabecular thickness (Tb.Th), and bone mineral density (BMD) (Figs. 9a and S13e). This radiological evidence of accelerated bone formation velocity was further supported by dynamic histomorphometry via sequential calcein labeling, which showed substantial elevations in the mineral apposition rate (MAR), bone formation rate per bone surface (BFR/BS), and mineralizing surface per bone surface (MS/BS) (Fig. S13f, g). These findings were further supported by histological evaluations: hematoxylin and eosin (H&E) and Masson staining showed highly organized collagen deposition, and osteopontin (OPN) immunofluorescence confirmed robust osteogenic marker expression in the newly formed tissue (Fig. 9b).

Fig. 9.

AAV-shPfkm treatment enhances bone regeneration in rat critical-sized bone defects. a Micro-CT images of calvarial defects 4 weeks post-surgery (n = 6). b H&E staining, Masson staining, and immunofluorescence staining in the calvarial defects region (n = 3). c Micro-CT images of femoral defects 2 weeks post-surgery (n = 4). d H&E staining, Masson staining, and immunofluorescence staining in the femoral defect region (n = 3). Scale bars are as indicated

Given the distinct embryonic origins and mechanical profiles of flat and long bones, we next utilized the femoral defect model to evaluate the broad-spectrum efficacy and clinical relevance of targeting Pfkm. In this long-bone model, which was established three weeks post-intramedullary AAV injection (Fig. S13h), Micro-CT analysis at two weeks post-surgery demonstrated markedly enhanced bone regeneration in the AAV-shPfkm group, with significantly higher BV/TV, BMD, and trabecular number (Tb.N) (Figs. 9c and S13i). Consistent with these micro-CT data, subsequent histological assessments (H&E and Masson) and OPN immunofluorescence confirmed accelerated structural repair of the femoral defects (Fig. 9d). These integrated in vivo results demonstrate that targeting the Pfkm is a promising strategy for enhancing bone regeneration.

Discussion

While undifferentiated MSCs rely predominantly on glycolysis for self-renewal, lineage commitment necessitates a metabolic shift toward OXPHOS to meet heightened bioenergetic demands. This transition is coupled with robust MQC, which safeguards highly active mitochondria against oxidative stress-induced damage, thereby ensuring a sustained energy supply and a homeostatic intracellular environment essential for successful differentiation.^9,10,43–48 However, the upstream molecular sensors that orchestrate metabolic reprogramming with mitochondrial remodeling during osteogenesis have remained elusive.

Our study addresses this gap by establishing PFKM as a master regulator coupling metabolic reprogramming directly to mitochondrial renovation. Its suppression activates an ERK/c-Jun/CD38 signaling axis, which subsequently orchestrates mitochondrial dynamics through MERC-mediated Ca²⁺ influx and PKA activity inhibition. This sequence of events facilitates DRP1-mediated fission, yielding specialized, donut-shaped mitochondria that serve as functional hubs for accelerated mitochondrial biogenesis, selective mitophagy, and Mito/MDVs release. By accelerating the turnover of compromised components and replenishing the mitochondrial pool, this PFKM-driven renovation program optimizes bioenergetic fitness to ultimately drive osteogenic differentiation.

Canonical metabolic enzymes, such as hexokinase, PFKM, pyruvate kinase, succinate dehydrogenase, and nicotinamide adenine dinucleotide (NADH) dehydrogenase, are well-established for their enzymatic roles in catalyzing glycolysis and OXPHOS. However, an emerging body of evidence suggests that these enzymes also execute non-canonical “moonlighting” functions, playing indispensable roles in gene expression, DNA repair, cell cycle regulation and signal transduction.⁴⁹ As an isoform of PFK, PFKM serves as a primary rate-limiting enzyme in glycolysis by catalyzing the phosphorylation of fructose-6-phosphate. Although previous studies have indicated that Pfkm silencing can inhibit both glycolysis and OXPHOS, triggering ATP depletion and apoptosis in H9c2 cells,⁵⁰ our Seahorse OCR and PER analyses reveal that Pfkm knockdown in rBMSCs triggers metabolic reprogramming, shifting the metabolic flux from glycolysis toward OXPHOS. This transition initially reflects the modulation of its canonical function.

Intriguingly, this metabolic shift toward OXPHOS does not compromise cellular viability; instead, it triggers a robust compensatory mechanism characterized by an expansion of total mitochondrial mass to match the heightened bioenergetic demands of osteogenesis. This bioenergetic surge is accompanied by a profound structural and functional remodeling of the mitochondrial network. Notably, the resulting donut-shaped mitochondria observed in Pfkm-knockdown cells represent an adaptive configuration rather than a precursor to cell death. Although extensive mitochondrial fragmentation typically correlates with apoptosis, these donut-shaped structures retain their membrane potential and mirror the hyper-fused, protective organelle responses seen during hypoxia-reoxygenation stress.⁵¹ Benefiting from this specific morphology, donut-shaped mitochondria exhibit a high surface-area-to-volume ratio and are highly refractory to self-degradation via mitophagy.^52–54 Instead of undergoing autonomous clearance, they survive to serve as specialized structural scaffolds, facilitating the targeted mitophagic clearance of remaining, severely damaged fragments. This structural preservation is accompanied by a distinct asymmetry in protein and nucleic acid dynamics: localized mtDNA replication inside the matrix occurs much faster than the de novo synthesis of nuclear-encoded proteins like TOMM20, which requires time-consuming cytoplasmic translation and subsequent mitochondrial import. This coexistence of rapid autonomous mtDNA replication and targeted structural protein degradation explains the seemingly paradoxical finding of elevated mtDNA alongside reduced TOMM20 expression.

A recent study suggests that high OXPHOS demand drives mitochondria to segregate into distinct subpopulations: one enriched with pyrroline-5-carboxylate synthase (P5CS) that lacks ATP synthase and defined cristae, and another enriched with ATP synthase that possesses highly ordered cristae and superior OXPHOS capacity.⁵⁵ Strikingly, our donut-shaped mitochondria closely mirror this phenotype by lacking typical cristae. We therefore hypothesize that Pfkm knockdown induces a similar functional specialization, warranting further studies to characterize these subsets and define their roles in osteogenesis.

Beyond these metabolic and buffering roles, the donut morphology specifically promotes MDVs formation, thereby integrating structural remodeling with targeted quality control. The cargo of these MDVs—potentially including ATP synthase subunits along with Complex III and IV components, specific lipids, or mtDNA fragments—might provide “pre-assembled” metabolic machinery that primes recipient cells for differentiation.⁵⁶ MDVs and mitophagy often function as complementary, sequentially-activated quality control mechanisms. Under mild stress, MDVs serve as a first-line defense by selectively removing damaged components to preserve overall organelle function; however, when remodeling demands exceed MDVs capacity, mitophagy is triggered to eliminate compromised organelles.^57–59

This sophisticated, multi-layered quality control network ultimately protects Pfkm-knockdown cells from the damage typically associated with enhanced oxidative metabolism. While enhanced OXPHOS conventionally increases mtROS production via ETC hyperpolarization—a prime example of typical mitochondrial dynamics—Pfkm-knockdown cells paradoxically suppress mtROS accumulation. This metabolic balance comes from a multi-layered protective mechanism. Crucially, the robust enhancement in ATP production and OXPHOS is highly likely a direct consequence of increased mitochondrial biogenesis and elevated PGC-1α expression, rather than the hyperactivation or overloading of individual mitochondrion. Furthermore, the stable mitochondrial membrane potential shows that proton pump translocation remains balanced, which avoids the excessive voltage that drives electron leakage. This process is further supported by the significant upregulation of Ucp2, which inhibits the production of mtROS through mild uncoupling without disrupting ATP synthesis. When combined with the clearance of damaged mitochondria through enhanced mitophagy, these cooperative systems allow rBMSCs to safely increase ATP production while decreasing mtROS levels to support osteogenesis.

Our data demonstrate that Pfkm knockdown actively facilitates the formation of MERCs—dynamic membrane domains that tether the ER to mitochondria to facilitate lipid and ion exchange.^60–62 Consequently, this structural optimization and the concurrent expansion of the mitochondrial surface area further support the osteogenic secretory phenotype by reshaping the functional interfaces between these two organelles. The formation and structural stability of MERCs depend on “tethering proteins” such as (vesicle-associated membrane protein)-associated protein B and C (VAPB)-protein tyrosine phosphatase-interacting protein 51 (PTPIP51) and the IP3R-GRP75-VDAC complex.⁶³ While recent research has shown that protein kinase RNA-like ER kinase (PERK) can structurally support MERCs assembly independently of kinase activity,⁶⁴ artificial elevation of mitochondrial Ca²⁺ via MCU overexpression can also feedback-activate IP3R expression to promote contact site assembly.⁶⁵ This mechanism aligns closely with our model, wherein Pfkm knockdown significantly elevates Ip3r1 expression, providing a clear molecular basis for enhanced structural MERCs expansion.

Beyond this catalytic shift, we further identified a non-canonical moonlighting pathway triggered by Pfkm knockdown: the activation of the ERK/c-Jun signaling axis, which transcriptionally upregulates CD38 expression to stimulate mitochondrial biogenesis and fission. Within these expanded structural interfaces, this upregulated CD38 establishes a highly localized calcium microdomain, thereby safeguarding mitochondrial calcium homeostasis and organelle integrity. While physiological Ca²⁺ uptake typically enhances respiration,^66–70 excessive calcium oscillations can trigger the opening of the mitochondrial permeability transition pore (mPTP) to cause bioenergetic collapse.^71,72 Pfkm-knockdown cells escape this crisis by harnessing the PFKM/CD38 axis as a homeostatic checkpoint to constrain calcium signaling within safe boundaries. This protective gating role is strongly supported by external genetic data; for instance, Cd38-knockout stem cells experience severe calcium depletion and metabolic collapse.⁷³

Following this molecular bridge, the resulting calcium surge activates Ca²⁺-sensitive PDE1A, which hydrolyzes cAMP and suppresses PKA activity.⁷⁴ We show that these CD38-driven calcium surges suppress PKA activity, thereby inducing DRP1 dephosphorylation at its core inhibitory site, Ser656 (rat Ser656; human Ser637). By lifting this PKA-mediated molecular brake, Pfkm suppression drives extensive DRP1 recruitment and initiates mitochondrial fission.²³ Importantly, this extensive DRP1 recruitment does not merely drive organelle fission; it also actively coordinates targeted quality control through vesicle biogenesis. Most MDVs biogenesis events are known to be DRP1-independent,^57,75 although a specialized subset of TOMM20+ MDVs does require DRP1.⁷⁶ Crucially, our finding that the DRP1-S656D mutation attenuates MDVs release not only corroborates this DRP1-dependent mechanism but also demonstrates its functional importance in osteogenic differentiation.

Notably, the Ser10 residue of histone H3 is a shared phosphorylation target for both ERK and stress-activated P38, and given that other PFK isoforms (like PFKP) directly interact with ERK2,^77–79 we hypothesize that PFKM may competitively modulate shared chromatin sites or physically sequester ERK1/2. Accordingly, the exact molecular interfaces and regulatory sites governing this PFKM-ERK interaction need further investigation.

Ultimately, these findings demonstrate that PFKM acts as a metabolic coordinator, harmonizing its canonical catalytic roles and non-canonical signaling functions to dynamically steer stem cell fate. While this study establishes PFKM as a critical metabolic regulator of osteogenesis, several dimensions warrant further exploration. First, although we elucidated the structural and signaling consequences of Pfkm knockdown, the precise rewiring of metabolic fluxes remains to be fully mapped. Future isotopic tracing experiments utilizing glucose or glutamine will be essential to visualize dynamic changes in glycolytic flux and reactions, providing direct evidence at the metabolite level. Second, the hypothesized existence of a functional mitochondrial subpopulation requires further evidence to confirm whether P5CS is specifically enriched within donut-shaped structures. Finally, while we identified MDVs as bioactive osteogenic messengers, their specific cargo has not been systematically characterized. Multi-omics profiling to identify the proteins, lipids, or nucleic acids ferried by these vesicles represents a crucial next step to pinpoint the active components driving intercellular communication. Addressing these questions will complete the theoretical framework of the metabolic-mitochondrial-osteogenic axis.

Conclusion

In summary, our work identifies PFKM as a critical metabolic checkpoint whose inhibition unlocks the osteogenic potential of rBMSCs. This axis drives the transition of the mitochondrial network into a donut-shaped topology that enables synchronized mitochondrial biogenesis, mitophagy and MDVs formation. Specifically, these coordinated MQC processes ensure the bioenergetic fitness required for bone regeneration. Ultimately, by revealing how PFKM integrates its canonical metabolic function with non-canonical signaling roles to orchestrate this “metabolic-MQC” coupling, our work provides a new point for understanding stem cell fate determination and offers novel, mechanism-based strategies for regenerative medicine (Fig. 10).

Fig. 10.

Schematic diagram. This study unveils a non-canonical signaling role for PFKM in osteogenic differentiation beyond its metabolic function. Pfkm knockdown triggers characteristic donut-shaped mitochondria by expanding mitochondria-ER contacts and elevating CD38, which inhibits PKA activity and induces DRP1 dephosphorylation. This architectural transformation initiates comprehensive mitochondrial quality control—biogenesis, mitophagy, and MDVs secretion—optimizing cellular fitness for mineralization. In vivo validation demonstrates accelerated bone repair, establishing PFKM as a novel therapeutic target for bone regeneration

Materials and methods

Cell culture

The isolation of Sprague-Dawley rat BMSCs was conducted following established protocols.⁸⁰ The cells were maintained in α-MEM (Gibco, C12571500BT) supplemented with 10% (v/v) fetal bovine serum (FBS) (Pan, P30-3302) and 1% (v/v) penicillin-streptomycin solution (Gibco, 15140122). Culturing was performed in a 37 °C, 5% CO₂ incubator with 90%–95% humidity, and the medium was replenished every 2–3 days.

Animal ethics

This study was conducted in compliance with protocols approved by the Ethics Committee of Fujian Medical University (Approval No. IACUC FJMU 2025-0048). Eight-week-old male Sprague-Dawley rats were procured from hfkbio (China) and randomly assigned to experimental groups. All animals were maintained under standard conditions (22–24 °C, 12-h light/dark cycle) with 3 rats per cage at the institutional Laboratory Animal Research Center.

Cranial defect model

Two weeks prior to surgery, eight-week-old SD rats (approximately 250 g) were shaved and received a subcutaneous injection of AAV virus (2.0 × 10¹¹ vg) (Obio, China). Under isoflurane anesthesia, a midline incision was made to expose the skull, and the underlying periosteum was gently reflected. Using a drill, two 5-mm diameter circular craniotomies were created bilaterally adjacent to the sagittal suture, with continuous irrigation of pre-cooled saline to minimize thermal injury. Based on experimental groups, collagen membranes were implanted into the defects. Closure was achieved by suturing the fascial and skin layers sequentially. Postoperative gentamycin (2–4 mg/kg) was administered intramuscularly to prevent infection. For fluorescent bone labeling, calcein (Solarbio, C7600) (40 mg/kg) was injected intraperitoneally 10 days and 3 days prior to euthanasia. Following a 4-week healing period, rats were sacrificed for tissue harvest and subsequent analysis.

Femoral defect model

The femoral defect model followed a two-stage surgical protocol. Rats first received a unilateral injection of AAV virus (2.0 × 10¹¹ vg) into the femoral marrow cavity. Three weeks later, a 3-mm diameter circular cortical defect was created in the same femur, and tissue samples were harvested two weeks post-defect induction. Postoperative gentamycin (2–4 mg/kg) was administered intramuscularly to prevent infection.

Single-cell data acquisition, quality control, and cell annotation

For the single-cell RNA sequencing (scRNA-seq) analysis of bone fractures, we utilised the GSE154247 dataset, specifically targeting non-hematopoietic bone stromal cells (BSCs) from Mus musculus femurs at 14 days post-fracture and corresponding control bone tissues. To identify distinct cellular clusters, we performed unsupervised clustering using the Leiden algorithm (resolution = 0.3) applied to a shared nearest neighbor (SNN) graph. Dimensionality reduction and spatial visualization were achieved via Uniform Manifold Approximation and Projection (UMAP). Differentially expressed genes (DEGs) were defined using the Wilcoxon rank-sum test, applying thresholds of |log₂ fold change (FC)| > 0.5 and P < 0.05. Cell-type annotation was conducted through a dual-step process: initial automated classification via SingleR, followed by rigorous manual refinement based on established canonical marker genes.⁸¹

Pathway enrichment, activity scoring, and pseudotime trajectory analysis

We identified cluster-specific DEGs (log₂ FC > 0.5, P < 0.05) using the FindAllMarkers function and subsequently performed GO enrichment analysis to characterize the biological themes of the cell subclusters. Pathway score at the single-cell level was quantified using the single-sample gene set enrichment analysis (ssGSEA) method.⁸² To reconstruct developmental lineages, we performed pseudotime trajectory analysis using the monocle3 software package.⁸³ Furthermore, we retrieved the “HALLMARK OXIDATIVE PHOSPHORYLATION” and “HALLMARK GLYCOLYSIS” gene sets from the MSigDB Mouse Collections (https://www.gsea-msigdb.org/gsea/msigdb/mouse/). The AddModuleScore function was then employed to calculate the scores of these metabolic gene sets, allowing us to track their dynamic shifts along the identified pseudotime trajectory.

Osteogenic differentiation protocol

Upon reaching 80%–90% confluence, cells at the third passage were induced towards osteogenic differentiation. The induction medium comprised α-MEM supplemented with 10% (v/v) FBS, 1% (v/v) penicillin-streptomycin, 10 mmol/L β-glycerophosphate (Sigma, G9422), 50 μg/mL L-ascorbate (Sigma, A4544), and 10 nmol/L dexamethasone (Sigma, D4902). This medium was refreshed every 2–3 days during the induction period.

Alkaline phosphatase (ALP) staining and activity assay

After 7 days of induction, cells were rinsed twice with phosphate-buffered saline (PBS) and fixed in 4% (w/v) paraformaldehyde for 30 min. ALP staining was conducted using a commercial kit (Beyotime, C3206) following the manufacturer’s instructions. ALP activity was quantitatively assessed with an ALP assay kit (Beyotime, P0321), with three replicates (n = 3) analyzed for statistical significance.

Alizarin Red S staining

Matrix mineralization was evaluated by Alizarin Red S staining after 14 days of induction. Cells were fixed as previously described and exposed to a freshly prepared Alizarin Red S solution (OriCell, ALIR-1000) in the absence of light. The stained nodules were dissolved in 10% (v/v) cetylpyridinium chloride (Macklin, C830716). Subsequent absorbance measurement at 562 nm was carried out in three replicates (n = 3) for statistical quantification.

RNA knockdown

The sequences of the small interfering RNAs (siRNAs) targeting Pfkm, purchased from GenePharma, are listed in Supplementary Table 1. Cell transfection was performed using Lipofectamine RNAiMAX (Invitrogen, 13778150) according to the manufacturer’s protocol.

Fluorescent staining for labeling mitochondria and endoplasmic reticulum

Mitochondria were fluorescently labeled with one of the following probes according to experimental needs and channel compatibility: MitoTracker Green (Beyotime, C1048), MitoTracker Deep Red (Beyotime, C1032), or PK Mito Red (Genvivotech, PKMR-2). Staining in each case was conducted following the manufacturer’s recommended protocol. The endoplasmic reticulum was labeled in live cells by transfecting with EZ Cap™ ER-EGFP Probe mRNA (m1Ψ) (APExBIO, R1112) using Lipofectamine RNAiMAX in accordance with the manufacturer’s instructions. Fluorescence images were acquired using either a laser scanning confocal microscope (Olympus FV3000, Japan) or a Multi-SIM system (Beijing Naxi Technical Corporation, China), equipped with a 100×/1.49 NA oil-immersion objective (Nikon, Japan).

Cell viability

A Cell Counting Kit-8 (Dojindo, CK04) was used to evaluate cell proliferation according to the manufacturer’s protocol. Cells were seeded in 96-well plates at a density of 3 × 10³ cells per well. After 1, 3, and 5 days of culture, 10% CCK-8 reagent was added to each well and incubated for 1 h. The absorbance was subsequently measured at 450 nm using a microplate reader (SpectraMax, USA).

Quantitative real-time reverse transcription polymerase chain reaction (qRT-PCR)

Total RNA was isolated with TRIzol Reagent (Invitrogen, 15596026CN) following the manufacturer’s instructions. Subsequently, the extracted RNA was reverse-transcribed into complementary DNA (cDNA) using a commercial Reverse Transcription kit (Yeasen, 11141ES60). Reverse transcription quantitative polymerase chain reaction (RT-qPCR) was conducted using the SYBR Green PCR reagent kit (Yeasen, 11202ES50) on an ABI QuantStudio 5 Real-Time PCR System (Applied Biosystems, USA). The sequences of the primers used are provided in Supplementary Table 2. Expression levels of all target mRNAs were normalized to that of α-tubulin.

mtDNA copy number quantification assay

Genomic DNA was extracted from cultured cells using the TIANamp Genomic DNA Kit (Tiangen Biotech, DP304-03). Mitochondrial DNA copy number (mtDNA-CN) was determined by quantitative PCR (qPCR) targeting the mitochondrial genes ND1 and ND5, with the single-copy nuclear gene β-actin used as the target for nuclear DNA. The sequences of the primers used are provided in Supplementary Table 2. The data were analyzed using the 2^−ΔΔCt method. The obtained values were normalized to nuclear DNA and are reported as the number of mitochondrial DNA copies per nuclear genome (mtDNA/nDNA).

Western blotting analysis

Cells were lysed in RIPA buffer (Beyotime, P0013B) containing a broad-spectrum protease and phosphatase inhibitor cocktail (Boster, AR1141). Subsequently, the samples were lysed via ultrasonication with three repeated pulses (3 s each, 6 s apart). The resulting lysates were centrifuged, and the supernatants were harvested for Western blot. Protein samples were electrophoresed on 10% SDS-PAGE gels (Bio-Rad, USA) and transferred to 0.22 μm PVDF membranes (Millipore, IPVH00010). After blocking, the membranes were incubated with designated primary antibodies at 4 °C overnight. The blots were then treated with HRP-conjugated secondary antibodies, and signal detection was carried out using an ECL substrate (Bio-Rad, USA). Details of all antibodies used can be found in Supplementary Table 3. To assess site-specific DRP1 phosphorylation, antibodies targeting human epitopes were utilized based on their cross-reactivity with rat proteins. Accordingly, the human DRP1 Ser637 and Ser616 epitopes were used to monitor the corresponding phosphorylation sites at Ser656 and Ser635 in rat DRP1, respectively.

Isolation of secreted mitochondria and mitochondrial-derived vesicles (MDVs)

To recover secreted mitochondria and MDVs, the osteogenic conditioned medium was subjected to a tiered centrifugation series. Initial centrifugation at 300 × g removed cellular debris, followed by dual rounds at 3 000 × g (15 min each) to eliminate larger organelles and apoptotic bodies at 4 °C. The remaining supernatant then underwent two high-speed runs at 18 000 × g for 30 min at 4 °C. The final pellets, containing the enriched mitochondrial fractions and MDVs, were collected and resuspended for downstream experiments.

Seahorse assay

Metabolic phenotyping, specifically Oxygen Consumption Rate (OCR) and Proton Efflux Rate (PER), was conducted using the Seahorse XF24 Extracellular Flux Analyzer (Agilent). After 0, 3, and 7 days of induction, the cells were trypsinized and seeded at 2 × 10⁴ cells per well into XF24-well microplates and cultured until they reached ~80%–90% confluence. Prior to the assay, the growth medium was replaced with 500 μL of Seahorse XF DMEM, and the cells were equilibrated for 60 min at 37 °C in a non-CO₂ incubator. OCR was evaluated via the Mito stress test kit, utilizing sequential injections of oligomycin (1.5 μmol/L), FCCP (1.5 μmol/L), and a mixture of rotenone/antimycin A (0.5 μmol/L). PER was assessed using the Glycolysis rate test kit. All XF data were normalized to final cell counts and analyzed using Wave 2.6.4 software.

Plasmid construction, viral vector preparation, and viral transduction

The plasmids (pLV3-CMV-Drp1(rat)-EGFP-EF1a-Puro, pLV3-CMV-mCherry-Map1lc3b(rat)-Puro, pLV3-CMV-Drp1(rat) S656D-EGFP-EF1a-Puro, pLV3-CMV-Mito-cox8(rat)-EGFP-Puro) were commercially constructed by Yanming Bio (Shenzhen, China). For lentiviral production, the respective transfer plasmids were co-transfected along with the packaging plasmids psPAX2 and pMD2.G into HEK293T cells using polyethyleneimine (Yeasen, 40815ES03). Viral supernatants were harvested at 48 and 72 h post-transfection, concentrated via ultracentrifugation, and titrated by RT-qPCR following standard procedures.

For transduction, rBMSCs were seeded in 12-well plates and infected with the lentiviruses at a multiplicity of infection (MOI) of 20 in the presence of polybrene. The culture medium was replaced with fresh complete medium 24 h after transduction. Fluorescence microscopy performed at 72 h post-transduction confirmed both transduction efficiency and fluorescence expression for all constructs.

Measurement of mitochondrial membrane potential and mitochondrial reactive oxygen species (mtROS)

Cells were seeded in confocal dishes at a density of 3 × 10⁴ cells per dish. For measurement of the mitochondrial membrane potential, and the mitochondrial ROS, the cells were incubated with TMRE (MCE, HY-D0985A) or MitoSOX (HY-D1055, MCE) for 30 min at 37 °C, respectively. Image acquisition was performed using a confocal laser scanning microscope (Olympus FV3000, Japan).

Measurement of mitochondrial Ca²⁺ uptake

rBMSCs were seeded in 96-well plates, cultured to 70%–80% confluence, and treated as indicated. Cells were washed with HBSS (without Ca²⁺/Mg²⁺), loaded with Rhod-2 AM (Beyotime, S1062M) in serum-free HBSS for 30–60 min at 37 °C in darkness, then washed thrice and incubated in fresh HBSS for 20–30 min for complete de-esterification. Fluorescence intensity was measured using a microplate reader (excitation 552 nm, emission 581 nm, bottom-read mode).

RNA sequencing and bioinformatics analysis

The RNA extraction, library preparation, and sequencing procedures were performed by Shanghai Biotree Technology Co., Ltd. Total RNA was extracted using Trizol reagent and assessed for quality using an Agilent Bioanalyzer 2100. Libraries were prepared from high-quality RNA (RIN > 7.0) following poly(A) selection and fragmentation. cDNA synthesis, end repair, A-tailing, and adapter ligation were performed using standard protocols. Final libraries were sequenced on an Illumina NovaSeq 6000 platform in PE150 mode. Differential expression analysis between comparison groups was conducted using DESeq2, with genes showing |log₂(FC) | ≥1 and FDR < 0.05 considered statistically significant. Functional enrichment analysis of differentially expressed genes was carried out for GO terms and Kyoto Encyclopedia of Genes and Genomes (KEGG) pathways. Gene Set Enrichment Analysis (GSEA) was also performed to evaluate the enrichment of predefined gene sets under |NES | >1, NOM P-val < 0.05, and FDR Q-val < 0.25. All enrichment results were visualized using ggplot2 in R.

Flow cytometry

For secreted mitochondria and MDVs staining, the suspensions were incubated with the MitoTracker Green for 30 min at 37 °C. Samples were analyzed using an LSRFortessa X-20 flow cytometer (BD Biosciences, USA), and data were processed with FlowJo software.

Immunofluorescence staining

For immunofluorescence analysis, tissue sections were first permeabilized with 0.5% Triton X-100 (BioFroxx, 1139ML100) for 15 min at room temperature and then blocked with 5% BSA (BioFroxx, 4240GR500) for 30 min. Sections were subsequently incubated with primary antibody overnight at 4 °C, followed by incubation with an Alexa Fluor 488-conjugated secondary antibody for 1 h at room temperature in darkness. Nuclei were counterstained with DAPI (MCE, HY-D0814). Image acquisition was performed using a confocal laser scanning microscope (Olympus FV3000, Japan).

Micro-computed tomography (Micro-CT) and bone morphometric analysis

Following euthanasia, rat skull and femur specimens were harvested and fixed in 4% paraformaldehyde for 24 h. Bone architecture was then assessed using a high-resolution micro-CT system (NEMO Micro CT, NMC-200, PINGSENG Healthcare, China). Bone reconstruction at the defect sites was performed with the Cruiser image acquisition software, followed by quantitative bone morphometric analysis using the Avatar software package.

Histological staining and analysis

Bone specimens were decalcified in 0.5 mol/L EDTA for 6 weeks. Subsequently, for paraffin sections, samples were embedded in paraffin wax and cut into 5 μm-thick slices. For frozen sections, samples were embedded in OCT compound and cryosectioned at 10 μm thickness. Finally, staining with Hematoxylin and Eosin (H&E) (Servicebio, G1005) and Masson’s Trichrome (Maxim, MST-8003) was performed following the manufacturer’s instructions.

Transmission electron microscopy

For transmission electron microscopy analysis, cell pellets were collected by centrifugation until a visible pellet formed. The samples were primarily fixed with an aldehyde-based electron microscopy fixative (Aspen Bio, Cat# AS1063, China) at 4 °C for 2–4 h, followed by three 15-min washes with 0.1 mol/L PBS. Post-fixation was then performed using 1% osmium tetroxide in 0.1 mol/L PBS at room temperature for 2 h, after which the samples were washed again with PBS three times. Dehydration was carried out in a graded series of ethanol (50%, 70%, 80%, 90%, 95%, and two changes of 100% anhydrous ethanol), 15 min per step. The dehydrated samples were subsequently infiltrated with a 1:1 mixture of acetone and SPI-PON 812 epoxy resin (SPI, Cat# 90529-77-4, USA) overnight, followed by pure resin for another overnight period. The samples were then embedded in fresh resin and polymerized at 60 °C for 48 h. Ultrathin sections (60–80 nm) were cut using a Leica UC7 ultramicrotome (Leica, Germany) with a diamond knife (Diatome, Ultra 45°, Swiss). The sections were double-stained with 2% uranyl acetate and lead citrate, each for 15 min, and dried overnight at room temperature. Finally, the prepared grids were observed, and images were captured under an FEI Tecnai G2 20 TWIN transmission electron microscope.

Mitochondrial morphology analysis

The aspect ratio of mitochondria was measured and statistically analyzed using ImageJ (version 1.54) and Mitochondria Analyzer (version 2.3).

Statistical analysis

All statistical analyses were performed using GraphPad Prism (version 9.3.0). Data are presented as mean ± SD. Statistical comparisons among more than two groups were performed using one-way ANOVA followed by either Tukey’s post hoc test or Dunnett’s test against control unless otherwise specified. Comparisons between two groups were analyzed using Student’s t test. P value of less than 0.05 was considered statistically significant, with asterisks denoting significance levels as *P < 0.05, **P < 0.01, and ***P < 0.001.

Supplementary information

Supplementary Table ^{(23.3KB, docx)}

Supplementary figure ^{(17.3MB, pdf)}

Supplementary information ^{(20KB, docx)}

Author contributions

N.W. Huang, J. Chen and Y.J. Ou conceived the study and designed the experiments. N.W. Huang designed and performed the majority of the in vitro and in vivo experiments. Y. Li, J. Lu, Y.F. Xing, W.P. Chen, and W. Li assisted with the in vivo experiments and bioinformatics analyses. K.D. Chen provided assistance with the Seahorse metabolic assays. G.Y. Zheng, P.Y. Hu, K.X. He, H.Y. Lin and Z.Z. Xue contributed to the in vivo experiments, as well as histological sectioning and staining. W.X. Yuan, Y.W. Zhou and J.J. Li performed flow cytometry and histological staining. N.W. Huang and Y.J. Ou analyzed the data and discussed the findings. N.W. Huang drafted the manuscript. Y.J. Ou, S.H. Zhang and J. Chen reviewed and edited the manuscript. All authors have read and approved the final version of the manuscript.

Data availability

No original code was generated in this study. Any additional information required to reanalyze the data reported in this paper is available from the corresponding author upon request.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41368-026-00460-5.