Mitochondrial transfer from glia to neurons protects against peripheral neuropathy

Abstract

Primary sensory neurons in dorsal root ganglia (DRG) have long axons and a high demand for mitochondria, and mitochondrial dysfunction has been implicated in peripheral neuropathy after diabetes and chemotherapy^1,2. However, the mechanisms by which primary sensory neurons maintain their mitochondrial supply remain unclear. Satellite glial cells (SGCs) in DRG encircle sensory neurons and regulate neuronal activity and pain³. Here we show that SGCs are capable of transferring mitochondria to DRG sensory neurons in vitro, ex vivo and in vivo by the formation of tunnelling nanotubes with SGC-derived myosin 10 (MYO10). Scanning and transmission electron microscopy revealed tunnelling nanotube-like ultrastructures between SGCs and sensory neurons in mouse and human DRG. Blockade of mitochondrial transfer in naive mice leads to nerve degeneration and neuropathic pain. Single-nucleus RNA sequencing and in situ hybridization revealed that MYO10 is highly expressed in human SGCs. Furthermore, SGCs from DRG of people with diabetes exhibit reduced MYO10 expression and mitochondrial transfer to neurons. Adoptive transfer of human SGCs into the mouse DRG provides MYO10-dependent protection against peripheral neuropathy. This study uncovers a previously unrecognized role of peripheral glia and provides insights into small fibre neuropathy in diabetes, offering new therapeutic strategies for the management of neuropathic pain.

Mitochondria that are transported from satellite glial cells in dorsal root ganglia to peripheral sensory neurons through tunneling nanotube-like structures provide protection against peripheral neuropathy.

Main

Primary sensory neurons within the DRG are pseudounipolar, with long peripheral branches that innervate target organs such as the skin⁴. These neurons detect sensory inputs—including pain, touch, temperature and proprioception—and transmit them to the central nervous system via action potentials⁵. Neuronal activity, including rapid action potential firing, relies on tightly regulated mitochondrial function⁶. The peripheral axons of human DRG sensory neurons can extend up to 100 cm in the sciatic nerve. However, these long axons and their terminal branches in the skin face exceptional challenges in mitochondrial transport and delivery⁷. This raises the key question of how a single DRG neuron generates and maintains sufficient mitochondria to support its activity and axonal regeneration. SGCs exhibit a unique morphology, tightly enveloping the cell bodies of sensory neurons in the DRG³. However, the structural connections between SGCs and sensory neurons under both physiological and neuropathic conditions remain poorly understood. Mitochondria, often referred to as the powerhouse of the cell owing to their role in energy production⁸, were traditionally thought to be generated and maintained solely within individual cells. However, accumulating evidence suggests that mitochondria can be transferred between cells. For example, specialized structures called tunnelling nanotubes (TNTs) have been observed connecting different cell types in vitro^9,10, and have recently been implicated in mitochondrial transfer between microglia¹¹, from stem cells to endothelial cells¹², and from stem cells to T cells¹³. Although these findings highlight the dynamic interplay between different cell types and the roles of mitochondrial transfer in diverse physiological and pathological processes, the cell-specific mechanisms of mitochondrial transfer remain unclear, and in vivo evidence is still limited.

Mitochondrial transfer from mouse SGCs to neurons

To investigate mitochondrial transfer from SGCs to sensory neurons in vitro, we first prepared primary cultures of SGCs and neurons separately. Mouse primary SGC cultures were immunostained with established SGC markers, including FABP7, Kir4.1, AQP4, connexin 43 and glutamine synthetase. At 5 days in vitro (DIV5), 94.9–97.1% of cultured SGCs were positive for these markers (Extended Data Fig. 1a), validating the identity and purity of our SGC cultures. Next, we labelled SGCs with MitoTracker deep red dye¹⁴, co-cultured these cells with primary sensory neurons (without MitoTracker dye labelling) from DRG of Trpv1:Ai9 reporter mice, and captured images of the co-cultures after 24 h (Fig. 1a). Remarkably, we observed mitochondrial fluorescence not only in SGCs but also in Trpv1:Ai9⁺ neurons, indicating that mitochondria were transferred from SGCs to sensory neurons. Notably, we detected TNTs of more than 30 μm in length connecting SGCs and neurons, with mitochondria present inside these structures (Fig. 1b and Extended Data Fig. 1b). Quantitative analysis revealed that 83.3% of DRG neurons received mitochondria from SGCs, whereas only 31.3% of neurons exhibited visible TNTs (Fig. 1c). Live imaging of the co-cultures (Extended Data Fig. 1c) revealed that TNTs were formed transiently and could disassemble within tens of minutes (Extended Data Fig. 1d), which may explain the relatively low percentage of TNT-positive neurons. To determine whether mitochondrial transfer requires intrinsic neuronal activity, we treated co-cultures with tetrodotoxin (TTX) to block sodium channels. TTX reduced mitochondrial transfer (Extended Data Fig. 1e–g) without affecting the overall rate of TNT formation between SGCs and neurons (Extended Data Fig. 1h). Together, these results indicate that SGCs can transfer mitochondria to DRG neurons in an activity-dependent manner.

Extended Data Fig. 1. Characterization of mitochondrial transfer in mouse DRG cultures.

(a) Immunostaining FABP7, Kir4.1, AQP4, CX43, and GS in mouse SGC cultures, along with quantification of the percentage of marker-positive cells. (b) Triple staining showing mitochondrial (red arrow) transfer from a FABP7-labeled SGC to a Trpv1+ neuron. n = 3 biological repeats. (c-d) Schematic (c) and representative live-cell imaging (d) of TNT formation between SGCs and neurons. Top panels in d: MitoTracker labeling. Red arrows indicate mitochondria within a TNT, and white arrows indicate TNT breakage. Bottom panels in d: contrast images showing a broken TNT (white arrow). (e-f) Schematic (e) and representative images (f) showing mitochondrial transfer from SGCs to neurons in vehicle- and TTX-treated groups. White arrows indicate TNTs and red arrows indicate mitochondria (Mito). (g) Quantification of MitoTracker signal density in neurons interacting with SGCs. n = 31 neurons (TTX) and n = 32 neurons (vehicle) from 3 independent experiments. **** P < 0.0001. (h) Percentage of TNT-positive neurons. n = 60 neurons (TTX) and n = 66 neurons (vehicle) from 3 independent experiments. (i) Top, schematic of SGC and neuron co-cultures treated with different inhibitors. Bottom, representative MitoTracker images showing inhibitory effects of CytoB, Y-27632, Pitstop 2, and CBX on mitochondrial transfer from SGCs to neurons. Arrow indicates TNT. Scale bars are as indicated. (j) Quantification of MitoTracker signal density in neurons interacting with SGCs in 5 different groups. n = 41 (vehicle), 33 (CytoB), 32 (Y-27532), 30 (Pitstop2), and 37 (CBX) over 4 independent experiments each group. **** P < 0.0001, * P = 0.0187, * P = 0.0377. (k) Percentage of TNT-positive neurons showing inhibitory effects of CytoB, Y-27632, and CBX on TNT formation. Sample sizes are as indicated in j. Data are shown as means ± s.e.m. Unpaired t-test (two-sided, g), one-way ANOVA followed by Tukey’s multiple comparisons test (j).

Source Data

Fig. 1. Mitochondrial transfer from SGCs to neurons in co-culture and TNT-like structures in the mouse DRG.

a, Schematic of SGC–neuron co-cultures from mouse DRG. SGCs are labelled with MitoTracker dye and DRG neurons are from Trpv1:Ai9 mice. b, Left, image from an SGC–neuron co-culture showing a Trpv1⁺ neuron interacting with an SGC (scale bar, 20 μm). Right, enlarged view of the boxed area showing a TNT (white arrow) and a mitochondrion (Mito; red arrow) within the TNT (scale bar, 5 μm). c, Percentage of MitoTracker-positive (Mito⁺) and TNT-positive neurons. A total of 109 neurons from 6 independent experiments were included for quantification. d, Schematic for SEM of whole-mount DRG without sectioning. e, High-magnification SEM images of whole-mount mouse DRG showing a TNT-like structure (TNT-LS) with a bulge. n = 4 biological repeats. Scale bars: 5 μm (left); 1 μm (right). f, Schematic for SEM of sectioned mouse DRG. g, Representative SEM images of sectioned mouse DRG showing TNT-LS with bulges from SGCs to neuron. n = 4 biological repeats. Scale bars: 5 μm (left); 1 μm (right). h, Schematic for TEM of mouse DRG. i, Representative low-magnification TEM imaging showing the DRG neuron and SGCs. n = 4 biological repeats. The green asterisk indicates the nucleus of the neuron. Scale bar, 5 μm. j–l, Representative high-magnification TEM imaging showing a TNT-LS between SGC and neuron. ‘Mito’ in black indicates mitochondria in the cell (j–l); ‘Mito’ in red indicates mitochondria in the TNT-LS (k); ‘Vesicle’ indicates a vesicle in the TNT-LS (k,l); the black asterisk indicates the nucleus of an SGC. l, Enlarged view from k showing mitochondria (red arrows) and a vesicle inside the smooth TNT-LS. ER, endoplasmic reticulum. n = 3 biological repeats. Scale bars: 2 μm (j,k); 800 nm (l).

Source Data

Previous studies have proposed three primary mechanisms for mitochondrial transfer: (1) formation of TNTs¹¹, which can be disrupted by cytochalasin B (CytoB)¹⁵; (2) endocytosis and internalization of mitochondria by recipient cells¹⁶; and (3) transfer via gap junctions, which can be blocked by carbenoxolone¹⁷ (CBX). We investigated these mechanisms using pharmacological approaches. Treatment with CytoB effectively blocked mitochondrial transfer from SGCs to neurons in co-cultures. A similar blockade was observed with Y-27632, which is known to suppress uropod and TNT formation¹⁸ (Extended Data Fig. 1i–k). Moreover, mitochondrial transfer was suppressed by the endocytosis inhibitor Pitstop2¹⁹ and the gap junction blocker CBX (Extended Data Fig. 1j). CytoB, Y-27632 and CBX further suppressed TNT formation (Extended Data Fig. 1k). Together, our in vitro results support a model in which mitochondrial transfer from SGCs to sensory neurons is mediated by TNTs, endocytosis and gap junctions. To examine mitochondrial transfer ex vivo, whole-mount DRG tissue was co-cultured with SGCs pre-labelled with MitoTracker (Extended Data Fig. 2a). We observed extensive TNTs formed between MitoTracker-labelled SGCs and DRG cells, including neurons and SGCs (Extended Data Fig. 2b,c). Moreover, treatment of SGC-whole-mount DRG co-cultures with CytoB (Extended Data Fig. 2d) caused marked inhibition in SGC mitochondrial transfer to whole-mount DRG (Extended Data Fig. 2e,f). Macrophage-to-neuron mitochondrial transfer in mouse DRG contributes to the resolution of acute inflammatory pain²⁰. We conducted a macrophage–whole-mount DRG co-culture experiment and confirmed that macrophages can also transfer mitochondria to DRG neurons (Extended Data Fig. 2e–g). Together, these experiments provide strong evidence for mitochondria transfer from SGCs to sensory neurons both in vitro and ex vivo.

Extended Data Fig. 2. Characterization of mitochondrial transfer in mouse whole-mount DRG preparations.

(a) Schematic of the ex vivo co-culture setup: SGCs were labeled with MitoTracker and co-cultured with half-cut whole-mount DRG tissue. (b) Low-magnification image of MitoTracker fluorescence showing strong signals in DRG cells, TNTs, and SGCs. Notably, SGC-derived TNTs extend to the edge of the DRG (TNT-4/5, white arrow) and penetrate deep into the tissue (TNT-1 to −3, white arrows). The DRG boundary is outlined by a dotted line. n = 7 biological repeats. (c) High-magnification image (enlarged from boxed area in b) showing TNTs (TNT-1 to −3, white arrows) between SGCs and neurons (blue arrows) within the DRG. (d) Schematic of co-cultures using primary SGC or macrophage cultures and whole mount DRG from Trpv1:Ai9 reporter mice. (e) Top panels: representative MitoTracker (red) and Ai9 (green) fluorescence images of DRG co-cultures with SGCs without (left) and with CytoB treatment (middle), and with macrophages (right). DRG boundary is marked with a dotted line. Scale bars are as indicated. Bottom panels: high-magnification images (enlarged from boxed area in top panels) showing mitochondrial transfer to Trpv1+ neurons and the inhibitory effect of CytoB (middle). Notably, TNT-like structures were observed between SGCs and neurons (left and middle, indicated by arrows), but not between macrophages and neurons (right). In the left panels, strong red fluorescence obscures TNT visibility. Scale, 50 μm. (f) Quantification of MitoTracker signal density in neurons interacting with SGCs, demonstrating the effect of CytoB. n = 5/group. * P = 0.0379. (g) Comparison of mitochondrial transfer to DRG neurons from SGCs versus macrophages. n = 5/group. P = 0.6181. Data are shown as means ± s.e.m. Mice from both sexes were included. Unpaired t-test (two-sided).

Source Data

To obtain three-dimensional ultrastructures of TNTs in the DRG, we used scanning electron microscopy (SEM; Fig. 1d). We found abundant extracellular matrix-like structures (ECM-LS) on the surface of SGCs, which may provide additional structural support for neurons. Notably, we observed TNT-like structures (TNT-LS) connecting adjacent cells, with large vesicle-like bulges present within the TNT-LS in mouse DRG (Fig. 1e and Supplementary Fig. 1). To further visualize the detailed contacts between SGCs and neurons, we performed SEM imaging on horizontally half-sectioned DRG (Fig. 1f). This cryostat-based sectioning method provided improved visualization of both SGCs and neurons within the DRG architecture (Fig. 1g). TNT-LS between SGC and neuron were observed in mouse DRG, and often contained bulges consistent in size with mitochondria (Fig. 1g and Supplementary Fig. 2). As expected from a previous study²¹, the tube structures were largely disrupted by trypsin treatment (Extended Data Fig. 3a,b). Notably, the formation of TNT-LS between SGCs and neurons was impaired in mice with chemotherapy-induced peripheral neuropathy (CIPN), as evidenced by enlarged intercellular gaps and a reduced number of TNT-LS after paclitaxel (PTX) treatment (Extended Data Fig. 3c–f). These results indicate that mitochondria-transferring TNTs are present between SGCs and sensory neurons in the peripheral DRG and are susceptible to neurodegenerative conditions such as CIPN. We further utilized structured illumination microscopy (SIM) to resolve individual mitochondria in mouse SGC. Quantification of mitochondrial length revealed an increase in the number of mitochondria shorter than 2 μm following PTX treatment (Extended Data Fig. 3g–i). This increase in short mitochondria is associated with cell stress²², suggesting that the PTX treatment directly induced mitochondrial dysfunction in SGCs. To validate the presence of mitochondria within TNTs of the DRG, we performed transmission electron microscopy (TEM) on 70-nm sections of mouse DRG (Fig. 1h). Smooth TNT-LS were observed between SGCs and neurons (Fig. 1i–l and Extended Data Fig. 4a–c). Both mitochondria and vesicles were present within TNT-LS connecting SGCs and neurons (Fig. 1l, Extended Data Fig. 4a–c and Supplementary Fig. 3). Together with the SEM findings, these results provide ultrastructural evidence for the presence of TNT-LS containing mitochondria and vesicular cargo between SGCs and neurons in the DRG.

Extended Data Fig. 3. Scanning electron microscopy (SEM) imaging of mouse DRG and structured illumination microscopy (SIM) imaging of mitochondria in SGC cultures.

(a) Schematic of the trypsin-treated whole mount DRG preparation and SEM imaging. (b) Representative SEM images showing destruction of the tube structures in DRG after trypsin treatment. n = 2 biological repeats. (c) Schematic for the paclitaxel (PTX) model using intraperitoneal injections (2 mg/kg, every other day, four injections). (d) Representative SEM images showing gap between SGCs and neurons in the DRG at 1-, 2-, and 4-weeks post-paclitaxel treatment. “N” indicates sensory neuron; “S” indicates SGC; and arrow points to TNT-like structures; gap between the SGC and the neuron is also indicated. (e) Quantification of the percentage of DRG neurons showing increased gaps with SGCs one-week (n = 4), two-week (n = 4), and four-week (n = 3) post PTX treatment compared to naïve mice (n = 4). *** P = 0.0003, ** P = 0.0044, * P = 0.0495. (f) Quantification of TNT-like structures in the gaps between SGCs and neurons showing a significant increase at 4 weeks compared to 1-week post-PTX. n = 4 (PTX-1w), 4 (PTX-2w), and 3 (PTX-4w). * P = 0.021, P = 0.1513. (g-i) SIM imaging of mitochondria and the effects of PTX in SGC cultures. (g-h) Representative SIM images of mitochondria before (g) and 15 min after treatment with 1 μg/mL PTX (h). Right panels showing enlarged views of the boxed areas from the left panels. Arrows indicate mitochondria with short-length. Scale bars are as indicated. (i) PTX treatment increases the short-length mitochondria in SGCs. n = 6/group. *** P = 0.0007. Data are shown as means ± s.e.m and analyzed by one-way ANOVA followed by Tukey’s multiple comparisons test (e and f). Unpaired t-test (two-sided, i). Mice from both sexes were included.

Source Data

Extended Data Fig. 4. Transmission electron microscopy (TEM) imaging of mouse DRG.

(a) Low magnification TEM image showing DRG neurons and SGCs. n = 4 biological repeats. (b) Enlarged image (from the boxed area in a), showing a TNT-like structure (blue dotted line). (c) Enlarged image (from the boxed area in b), showing a tube containing mitochondria (red arrows) and vesicles (orange arrows), as well as ER from adjacent SGC and neuron. Black and green asterisks (*) indicate the nuclei of SGCs and neurons, respectively; ER: endoplasmic reticulum. Scale bars are as indicated.

In vivo mitochondrial transfer from SGCs to neurons

To investigate mitochondrial transfer from SGCs to sensory neurons in vivo, we crossed MitoTag mice, which express eGFP on the mitochondrial outer membrane²³, with Aldh1l1-cre^ERT2 mice to drive SGC-specific expression of MitoTag²⁴ (Extended Data Fig. 5a,b). Tamoxifen (Tam) was administered daily for five days to temporally induce Cre-recombinase activity (Fig. 2a). DRG were collected five and ten days after the Tam treatment. No MitoTag fluorescence signal was detected in mice without Tam treatment (Extended Data Fig. 5c), whereas MitoTag signal was present in SGCs, marked by the SGC marker FABP7, at both day 5 and day 10 post-Tam injection (Fig. 2b). Of note, we observed a substantial increase in MitoTag signal over time, with 23.0% of DRG neurons producing a signal on day 10 compared with only 2.9% on day 5 (Fig. 2b,c). Colocalization of MitoTag with the mitochondrial marker TOM20 further confirmed the mitochondrial identity (Extended Data Fig. 5d), supporting a time-dependent mitochondrial transfer from SGCs to sensory neurons. To explore the subcellular mechanism of this transfer, we administrated CytoB intrathecally and observed a decrease in MitoTag signal in DRG neurons (Fig. 2c), whereas MitoTag expression in SGCs remained detectable (Fig. 2b). We also performed intrathecal injection of Pitstop2 to block endocytosis in the DRG (Extended Data Fig. 5e,f), which partially decreased mitochondrial transfer from SGCs to neurons (Fig. 2c). Thus, both TNT and endocytosis mechanisms are involved in mitochondrial transfer in vivo. Next, we examined whether mitochondrial transfer is altered under pathological conditions. In a diabetic neuropathic pain model, induced by intraperitoneal injection of streptozotocin (STZ; Extended Data Fig. 5e,f), we found that STZ pretreatment but not post-treatment could block SGC-to-neuron mitochondrial transfer (Fig. 2c). Intraplantar injection of complete Freund’s adjuvant (CFA) is known to elicit inflammatory pain. However, CFA did not alter the percentage of neurons receiving mitochondrial transfer (Fig. 2c and Extended Data Fig. 5g,h).

Extended Data Fig. 5. Characterization of mitochondrial transfer in DRG of MitoTag mice.

(a) Schematic of the immunocapture of mitochondria in Aldh1l1:MitoTag mouse DRG. (b) Western blot showing the expression of GFP and mitochondrial marker COX4 in crude mitochondrial fraction (CMF), immunocapture (IP), and supernatant (Sup) groups. n = 2 biological repeats. (c) DRG fluorescence image of Aldh1l1:MitoTag mice without tamoxifen injection. n = 2 biological repeats. (d) Colocalization of MitoTag fluorescence signal with mitochondrial marker TOM20 in DRG section. (e) Schematic of experimental design of tamoxifen (Tam) treatment (5 daily injections) with DRG tissue collection on day 10, and additional treatment with Pitstop2 and streptozotocin (STZ). (f) Representative DRG images of MitoTag green fluorescence. Bottom panels: enlarge boxed areas from the top panels. Red and blue arrows indicate MitoTag positive and negative neurons, respectively. n = 4 biological repeats in control, Pitstop2, STZ-5d groups, and n = 3 biological repeats in STZ-15d group. (g) Schematic of experimental design of tamoxifen (Tam) treatment with DRG tissue collection on day 5 and additional treatment with complete Freund’s adjuvant (CFA). (h) Representative DRG images of MitoTag green fluorescence. Bottom panels: enlarge boxed areas from the top panels. Red and blue arrows indicate MitoTag positive and negative neurons, respectively. Scale bars are as indicated. Related to Fig. 2b, c. (i) Schematic of pain behavioral testing before (baseline, BL) and after spared nerve injury (SNI) surgery combined with peri-sciatic nerve implantation of bupivacaine-loaded film in Aldh1l1:MitoTag mice. (j) von Frey testing showing paw withdrawal threshold. Data are shown as means ± s.e.m. and analyzed by two-way ANOVA followed by Sidak’s multiple comparisons test. **** P < 0.0001.

Source Data

Fig. 2. In vivo mitochondrial transfer from SGCs to neurons in the DRG of MitoTag mice.

a, Diagram of Aldh1l1-cre^ERT2:MitoTag mice with timeline of Tam injections and DRG collection. IP, intraperitoneal injection; IT, intrathecal injection. b, Top, MitoTag signal in DRG sections. Bottom, enlarged views showing double staining with FABP7 and Nissl. White arrows, MitoTag⁺ SGCs; red arrows, MitoTag⁺ neurons; blue arrows, MitoTag^– neurons. c, The percentage of MitoTag⁺ neurons in mice treated with Tam for 5 days (n = 4), 10 days (n = 4), 10 days with Tam plus CytoB (n = 4), 10 days with Tam plus Pitstop2 (n = 4), Tam with STZ (5 days, n = 3), STZ (10 days) with Tam (n = 4), Tam with CFA (4 h, n = 4), and Tam with CFA (5 days, n = 4). **P = 0.0034. d, Diagram of SNI surgery in Aldh1l1-cre^ERT2:MitoTag mice. e, MitoTag signal in the contralateral and ipsilateral DRG. f, The percentage of MitoTag⁺ neurons in ipsilateral DRG (n = 4) and contralateral DRG (n = 4). **P = 0.002. g, Diagram of SNI surgery with ipsilateral peri-sciatic bupivacaine film implantation. h, MitoTag signal in DRG. Bupi, bupivacaine. i, The percentage of MitoTag⁺ neurons in ipsilateral (n = 4) and contralateral (n = 4) DRG. j, Schematic of injection of AAV-MaCPNS2-Syn-jRGECO1a into Aldh1l1-MitoTag mice, SNI surgery and ex vivo calcium imaging in DRG. k, DRG calcium imaging of jRGECO1a in Aldh1l1-MitoTag mice. l, Heat map of MitoTag⁺ (green) and jRGECO1a⁺ neurons (red) from k. m, Quantification of jRGECO1a fluorescence. n = 45 cells from 9 DRG of 3 different mice. n, Colocalization of MitoTag⁺ neurons and jRGECO1a⁺ neurons. Data are mean ± s.e.m. One-way ANOVA followed by Tukey’s multiple comparisons test (c); two-sided unpaired t-test (f,i,m). ****P < 0.0001, ***P < 0.001, **P < 0.01, *P < 0.05; NS, not significant (P ≥ 0.05).

Source Data

Neuronal activation and action potential firing critically depend on mitochondrial function⁶, and abnormal neuronal activity after nerve injury drives neuropathic pain²⁵. To investigate the relationship between neuronal activity and mitochondrial transfer, we performed spared nerve injury (SNI) surgery at the time of the first Tam injection and analysed DRG tissue 5 days later (Fig. 2d). Remarkably, the percentage of MitoTag⁺ neurons was increased in the ipsilateral L4–L5 DRG compared with the contralateral side 5 days post-SNI (Fig. 2e,f). Additionally, eGFP⁺ mitochondrial signals were observed in axons, suggesting further mitochondrial transport from neuronal cell bodies to axonal processes (Fig. 2e). To further determine the dependence of mitochondrial transfer on neuronal hyperactivity, we implemented a sustained nerve blockade using bupivacaine-containing PEU (poly(ester urea)) films²⁶ wrapped around the sciatic nerve during SNI surgery (Fig. 2g). This blockade could increase the mechanical pain threshold above pre-injury baseline levels in SNI mice (Extended Data Fig. 5i,j). Notably, the percentage of MitoTag⁺ neurons was not different between the ipsilateral and contralateral DRG (Fig. 2g–i), suggesting that neuronal activity is required for mitochondrial transfer. Mitochondria remained present in SGCs after the nerve blockade (Fig. 2h), suggesting that the blockade impaired transfer rather than biogenesis of mitochondria. Together, these results suggest that SGC-to-neuron mitochondrial transfer is regulated by neuronal activity and specific pathological conditions.

Given the role of spinal cord astrocytes in regulating physiological and pathological pain, we next investigated whether mitochondrial transfer occurs between astrocytes and neurons in the spinal cord using the same Aldh1l1-cre^ERT2 system. No MitoTag signal was detected in Tam-untreated animals (Extended Data Fig. 6a). We did not observe any mitochondrial transfer from astrocytes to spinal neurons at either five or ten days post-Tam induction, but MitoTag labelling clearly delineated individual astrocyte territories (Extended Data Fig. 6b,c). SEM imaging of the spinal cord revealed cilia-like structures, but no tube-like connections between astrocytes and neurons or between astrocytes themselves (Extended Data Fig. 6d,e). To evaluate the directionality of mitochondrial transfer, we crossed MitoTag mice with Advillin-cre mice to drive sensory neuron-specific expression. This resulted in strong MitoTag labelling in DRG neurons but only minimal signal in SGCs (labelled with FABP7), indicating predominantly one-way transfer from SGCs to neurons. Similarly, mitochondrial transfer from neurons to macrophages (labelled with IBA1 (also known as AIF1)) in the DRG was minimal (Extended Data Fig. 7a–c). Finally, enabling MitoTag expression in macrophages using Cx3cr1-cre^ERT2 mice did not result in detectable mitochondrial transfer from macrophages to sensory neurons in the DRG (Extended Data Fig. 7d–f). Together, these in vivo findings strongly support specific and unidirectional mitochondrial transfer from SGCs to sensory neurons under the experimental conditions tested.

Extended Data Fig. 6. Characterization of mitochondrial transfer in the spinal cord dorsal horn of Aldh1l1: MitoTag mice and SEM imaging in the dorsal horn.

(a) Left: Timeline of tamoxifen treatment (5 daily injections) followed by spinal cord tissue collection at day 0, day 5, and day 10 from Aldh1l1:MitoTag mice. Right: MitoTag fluorescence imaging showing no signal at day 0 without tamoxifen treatment. (b-c) In the spinal cord dorsal horn, MitoTag signal is only present in astrocytes (GFAP+) but not in neurons (NeuN+) 5 days (b) and 10 days (c) after tamoxifen treatment. Boxed areas are enlarged in the bottom panels showing Aldh1l1:MitoTag fluorescence (bottom left) and triple staining of Mitotag/GFAP/NeuN (bottom right). Notably, no mitochondrial transfer was observed in neurons (NeuN+). n = 3 biological repeats. (d-e) SEM imaging on the L4-L5 segment of mouse spinal cord dorsal horn. (d) Schematic of tissue preparation for SEM imaging. (e) Left: low-magnification SEM image showing dorsal root, dorsal horn (grey matter), and white matter. Dotted curve line indicates the dorsal horn boundary. Middle: enlarged boxed area from the left panel. Right: enlarged boxed area from the middle panel. Arrows indicate the cilia-like structures. n = 4 biological repeats. Scale bars are as indicated.

Extended Data Fig. 7. Characterization of mitochondrial transfer in the DRG of Advillin: MitoTag and Cx3cr1: MitoTag mice.

(a) Schematic of crossing Advillin-cre and MitoTag mice. (b) Images of DRG sections showing Advillin:MitoTag (green) and FABP7 (red) fluorescence signal. Red arrows indicate colocalization of MitoTag and FABP7; blue arrows indicate FABP7 without MitoTag. (c) DRG section images showing Advillin:MitoTag fluorescence signal (green) and Iba1 staining (red). Red arrows indicate colocalization of MitoTag and Iba1, and blue arrows indicate macrophages (Iba1+) without MitoTag. (d) Diagram of crossing Cx3cr1-cre/ERT2 and MitoTag mice, along with the timeline of tamoxifen injection and DRG collection at day 5 and day 10. (e-f) MitoTag fluorescence signal (green), Iba1 signal (red), and Nissl signal (gray) at 5 days (e) and 10 days (f) after tamoxifen injection. Red arrows indicate colocalization of MitoTag and Iba1 in macrophages, and blue arrows indicate neurons without MitoTag. n = 3 biological repeats (b, c, e, and f). Scale bars are as indicated.

To investigate SGC-to-neuron mitochondrial transfer in relation to neuronal hyperactivity after SNI, we used ex vivo calcium imaging in DRG using jRGECO1a, a sensitive calcium indicator with red fluorescence²⁷. Adeno-associated viruses (AAVs) encoding jRGECO1a (AAV-MaCPNS2-Syn-jRGECO1a) were injected intraperitoneally into Aldh1l1:MitoTag mice at postnatal day 0 (P0; Fig. 2j). Four weeks later, mice underwent SNI surgery followed by five Tam injections, and L4–L5 DRG were collected for ex vivo calcium imaging (Fig. 2j–n). jRGECO1a fluorescence was detected in many small-sized DRG neurons (Fig. 2k), predominantly in MitoTag-negative neurons (Fig. 2k–m). Among 334 neurons in the SNI group, only 3 (less than 1%) were positive for both MitoTag and jRGECO1a (Fig. 2n). In non-SNI controls, both MitoTag and jRGECO1a signals were greatly reduced, with no colocalization (Extended Data Fig. 8a–d). Cell size analysis showed MitoTag⁺ signals in medium-to-large neurons, whereas jRGECO1a⁺ signals appeared mainly in small-to-medium neurons (Extended Data Fig. 8e). As a positive control, KCl depolarization evoked jRGECO1a responses in neurons of all sizes, confirming that there was no size-selective labelling (Extended Data Fig. 8f,g).

Extended Data Fig. 8. Additional characterization of ex vivo calcium imaging in mouse DRG.

(a) Representative images of calcium signal (jRGECO1a⁺, red) and mitochondrial signal (MitoTag⁺, green) in DRG neurons of Aldh1l1-Mito mice with tamoxifen (Tam) treatment for 5 days, followed by tissue collection on day 5. Left, low-magnification image showing both red and green channels. The boxed area was enlarged in the right panels for single red or green channel. G1 and G2 indicate two GECO1a-positive neurons. M1 indicates a mitoTag-labeled neuron. n = 3 biological repeats. (b-c) Heatmap of MitoTag- and jRGECO1a-positive neurons from (a) with MitoTag (green) and jRGECO1a (red) fluorescence signal. (d) Lack of MitoTag and jRGECO1a co-expression in 20 MitoTag⁺ neurons and 81 jGECO1a+ neurons in naïve mice without nerve injury. (e) Cell size frequency distribution of total DRG neurons (n = 12 DRG from 4 mice), MitoTag⁺ neurons (n = 9 DRG from 3 mice), and jRGECO1a⁺ neurons (n = 9 DRG from 3 mice) 5 days after SNI surgery. (f) Representative images of calcium indicator jRGECO1a in DRG of Aldh1l1:MitoTag mice following KCl treatment (60 mM, 1 min). Scale bars are as indicated. (g) Size frequency distribution of the jRGECO1a⁺ neurons in DRG of naïve mice with KCl treatment, demonstrating extensive activation in neurons with various diameters. n = 3. Data are shown as means ± s.e.m. Sample sizes are indicated by individual dots inside columns.

Source Data

Together, these findings indicate that: (1) SNI induces neuronal hyperactivity primarily in small-sized neurons; and (2) mitochondrial transfer from SGCs may preferentially protect medium-to-large neurons from becoming hyperactive after nerve injury. This protective mechanism may have implications for the pathogenesis of small fibre neuropathy in neuropathic pain²⁸.

MYO10 regulates TNT formation and pain

TNTs contain F-actin but lack microtubules⁹, contributing to their high elasticity. Ultrastructural analysis by SEM and cryo-electron microscopy has revealed the presence of MYO10 in TNTs²⁹. Analysis of a published mouse DRG single-cell RNA-sequencing (scRNA-seq) dataset³⁰ revealed Myo10 enrichment in SGCs (Fig. 3a). Immunostaining confirmed MYO10 expression in mouse SGCs, co-localizing with the SGC marker FABP7. (Fig. 3b and Extended Data Fig. 9a). To evaluate the role of MYO10 in TNT formation, we knocked down Myo10 in cultured SGCs with specific small interfering RNA (siRNA) and co-cultured them with DRG neurons (Fig. 3c). Immunocytochemistry revealed MYO10 expression in SGCs and TNTs connecting to neurons (Fig. 3d). MYO10 knockdown reduced the MYO10 signal (Fig. 3e), markedly decreased TNTs (Fig. 3f), and impaired mitochondrial transfer to neurons (Fig. 3g). We validated these results in vivo by microinjecting siRNA targeting Myo10 (siMyo10) or control siRNA into DRG of Aldh1l1:MitoTag mice (Fig. 3h). siMyo10 decreased SGC-to-neuron mitochondrial transfer (Fig. 3i,j), lowered paw withdrawal threshold at day 3 (Extended Data Fig. 9b,c), and disrupted TNT-LS in the DRG (Extended Data Fig. 9d–f and Supplementary Fig. 4). To further assess MYO10 in TNT formation and pain, we analysed Myo10-knockout (tm1d) mice, which we generated in an earlier study³¹. Since homozygous deletion is semi-lethal with frequent exencephaly³¹, we used heterozygous (Myo10^+/−) mice for behavioural and anatomical analyses. Compared with wild-type controls, Myo10^+/− mice exhibited heightened mechanical and thermal sensitivity (Fig. 3k,l and Extended Data Fig. 9g–i). SEM revealed wider gaps between SGCs and neurons and reduced TNT-LS following Myo10 knockdown (Fig. 3m–p). Together, these findings demonstrate that MYO10 is essential for TNT formation, mitochondrial transfer and SGC–neuron contact, and even partial loss disrupts TNT formation and promotes pain hypersensitivity.

Fig. 3. MYO10 in DRG SGCs regulates TNT formation, mitochondrial transfer, axonal degeneration and pain in mice.

a, Analysis of scRNA-seq data³⁰ (reproduced with license number 6122760569440) reveals the expression of Myo10 in mouse DRG. b, Immunohistochemistry of MYO10 in mouse DRG. N denotes neurons. Scale bar, 10 μm. c, Schematic of mitochondrial transfer from SGCs to neurons and its blockade by knockdown of Myo10. d, Signals of MYO10 and MitoTracker in SGCs and neuron co-cultures treated with control siRNA (siCtrl) or siMyo10. Scale bars, 10 μm. e, Quantification of integrated MYO10 density. n = 4 per group. *P = 0.0285. f,g, Percentage of TNT⁺ neurons (f) and MitoTracker⁺ neurons (g) in control and Myo10 siRNA groups. h, Diagram of experimental setup for Aldh1l1-cre^ERT2:MitoTag mice treated with Tam and siMyo10. i, MitoTag signal in DRG of siCtrl and siMyo10 groups. Red and blue arrows indicate MitoTag⁺ and MitoTag⁻ neurons, respectively. j, Quantification of the percentage of MitoTag⁺ neurons in control (n = 3) and siMyo10 (n = 4) groups. k, Schematic of wild-type (WT) and Myo10^+/− mice for behavioural study. l, von Frey test in wild-type (n = 6) and Myo10^+/− mice (n = 6). PWT, paw withdrawal threshold. m,n, Representative SEM images of DRG from wild-type (m) and Myo10^+/− (n) mice. Asterisks indicate neurons with visible gaps between surrounding SGCs. N1 and N2 indicate neurons; S1 to S7 indicate SGCs; the white arrow (m) indicates a TNT with bulge; the blue arrow (n) points to an irregular TNT spanning a gap between SGC and neuron. o, The percentage of DRG neurons having gaps. n = 4 (wild type), n = 6 (Myo10^+/−). p, Quantification of TNTs within SGCs. n = 4 (wild type), n = 6 (Myo10^+/−). Data are shown as means ± s.e.m. Two-sided unpaired t-test. cLTMR, C-fibre low-threshold mechanoreceptor; Firbo, fibroblast cells; Immune, immune cells; NF, neurofilament⁺ neurons; NP, non-peptidergic neurons; PEP, peptidergic neurons; Schwann, Schwann cells; SST, somatostatin⁺ neurons; TRPM8, TRPM8⁺ neurons; Vascular, endothelial cells.

Source Data

Extended Data Fig. 9. Behavioral and SEM analyses following Myo10 si-RNA knockdown and Myo10 knockout in mice.

(a) Triple staining (left) of MYO10 (red), FABP7 (green), and DAPI (blue) and double staining (right) showing co-localization of MYO10 and FABP7 in SGCs of mouse DRG. n = 3 biological repeats. (b-c) Altered mechanical pain sensitivity following siRNA knockdown of Myo10 in naïve mice. (b) Schematic of intra-DRG injection of si-RNA (Myo10, 4 μg in 1 μl) and associated pain behavior testing. (c) von Frey test showing decreased paw withdrawal threshold 3 days after Myo10-targeting siRNA treatment. n = 12 mice per group. ** P = 0.004. (d) Schematic of the intra-DRG injection of si-RNA (Myo10) and SEM imaging. (e-f) SEM images of DRG from control mice (e) and mice with si-RNA knockdown of Myo10 (f), showing a loss of TNT-like structures after Myo10 knockdown. Arrow indicates a TNT-like tube. Scale bars are as indicated. (g-i) Behavioral test in WT and Myo10^+/− mice. (g) Schematic of mechanical and thermal pain tests in wild-type (WT) and Myo10^+/− mice. (h) Paw withdrawal frequency to 0.16 g von Frey filament, showing heightened response in Myo10^+/− mice (n = 6) compared to WT mice (n = 6). *** P = 0.0003. (i) Hargreaves test showing the decreased paw withdrawal latency in Myo10^+/− mice (n = 6) compared to WT mice (n = 6). * P = 0.0351. Data are shown as means ± s.e.m. and analyzed by two-way ANOVA followed by Sidak’s multiple comparisons test (c). Unpaired t-test (two-sided, h and i).

Source Data

To test the role of TNTs in vivo, we microinjected MitoTracker into the DRG. MitoTracker signal was observed in both neurons and SGCs, but application of CytoB decreased signal intensity and the percentage of labelled neurons (Extended Data Fig. 10a–c), underscoring the crucial role of TNTs in mitochondrial transfer in vivo. To assess mitochondrial health, we treated SGCs with PTX and co-cultured these cells with healthy neurons (Extended Data Fig. 10d). JC-1 aggregate signals were reduced in both SGCs and neurons, suggesting transfer of dysfunctional mitochondria from SGCs to neurons. CytoB blocked this PTX-induced loss (Extended Data Fig. 10d,e). Seahorse analysis further showed that CytoB reduced oxygen consumption rate (OCR) in SGC–neuron co-cultures (Extended Data Fig. 10f). Sensory neurons damaged by nerve injury or chemotherapy exhibit abnormal hyperactivity that drives pain sensitivity³². PTX pretreatment enhanced calcium responses to capsaicin in primary cultured DRG neurons (Extended Data Fig. 11a–c). Notably, co-culture with SGCs suppressed this effect, but CytoB abolished this protection, implicating TNTs in SGCs-mediated neuroprotection (Extended Data Fig. 11a–c). Similarly, SGCs prevented PTX-induced increase of reactive oxygen species (ROS), an effect that is lost with CytoB (Extended Data Fig. 11d). Capsazepine (CPZ) blocked capsaicin responses under all conditions, confirming TRPV1 specificity (Extended Data Fig. 11e,f). Together, these findings suggest that PTX increases neuronal ROS, driving hyperactivity, whereas SGCs counteract these effects by transferring mitochondria to neurons via TNTs to reduce ROS and suppress hyperexcitability.

Extended Data Fig. 10. CytoB treatment blocks mitochondrial transfer to mouse neurons in vivo and in vitro.

(a-c) CytoB treatment blocks mitochondrial transfer to mouse neurons in vivo. (a-b) Representative images of vehicle-treated (a) and CytoB-treated (b) mice 1 day following DRG microinjection of MitoTracker dye. (a) DRG images from vehicle-treated mice and the box is enlarged in right panels showing MitoTag, GFAP, Nissl, and merged images. Arrows indicate MitoTracker-positive neurons. (b) DRG images from CytoB-treated mice and the box is enlarged in right panels showing MitoTag, GFAP, Nissl, and merged images. Notably, MitoTag labeling in neurons is blocked by CytoB. (c) Quantification of MitoTracker density (left) and the percentage of Mito+ neurons (right). n = 6/group. *** P = 0.0004; ** P = 0.0024. (d-f) CytoB treatment blocks mitochondrial transfer to healthy mouse neurons in vitro. (d) Left panel: schematic of DRG neuron-SGC co-cultures under 3 different conditions: 1) no treatment, 2) SGCs treated with PTX (1 μg/ml, 1 h), 3) SGCs treated with both PTX and CytoB (3.5 μM, 24 h). Right panels: SGC-neuron co-culture images of bright-field (grey color) and JC-1 aggregate signal (red color) under different conditions as indicated. Scale bar is indicated. (e) Quantification of JC-1 aggregate signal density in co-cultures. n = 12/group. *** P = 0.0002, ** P = 0.0039, * P = 0.0445, P > 0.9999, n.s., no significance. (f) Oxygen consumption rate (OCR) showing an inhibitory effect of CytoB treatment (3.5 μM) in DRG SGC-neuron co-cultures. n = 6/group. ** P = 0.0086. Data are shown as means ± s.e.m. and analyzed by Unpaired t-test (two-sided, c and f) and Two-way ANOVA followed by Sidak’s multiple comparisons test (e). Mice from both sexes were included.

Source Data

Extended Data Fig. 11. Mitochondrial transfer from SGCs to neurons prevents chemotherapy-induced neuronal hyperactivation and oxidative stress.

(a) Calcium imaging of DRG neurons co-cultured with SGCs from Advillin:GCamp6f mice before (baseline) and after capsaicin perfusion (100 nM, 2 min), demonstrating the effects of PTX (1 μg/mL) and CytoB (3.5 μM). Scale bar: 100 μm. (b-c) Traces of calcium responses to capsaicin and KCl (b), and Amplitudes of capsaicin-induced calcium responses, showing the effects of paclitaxel and CytoB (c) in dissociated DRG neurons under four indicated conditions. n = 288 neurons (vehicle), 296 neurons (PTX), 110 neurons (PTX + SGCs + vehicle), and 101 neurons (PTX + SGCs + CytoB). **** P < 0.0001, P = 0.1938, * P = 0.022. (d) Reactive oxygen species (ROS) assay in DRG SGC-neuron co-cultures showing the effects of paclitaxel and CytoB. n = 12 cultures/group. ** P = 0.0046, * P = 0.0209, * P = 0.024. (e-f) TRPV1 antagonist capsazepine (CPZ) blocks PTX-induced calcium response in DRG neurons. (e) Traces of calcium responses to capsaicin (100 nM, 2 min) in dissociated DRG neurons with PTX and CPZ (100 μM) treatment in 4 different groups. (f) Amplitudes of calcium responses. n = 162 (vehicle), 157 (PTX), 198 (PTX + SGCs + vehicle), and 200 (PTX + SGCs + CytoB). P = 0.9704, P = 0.5901. Data are shown as means ± s.e.m. and analyzed by One-way ANOVA followed by Tukey’s multiple comparisons test (c, d, and f). n.s. no significance. Mice from both sexes were included.

Source Data

Given the long axons of DRG neurons, impaired mitochondrial trafficking has been strongly implicated in neuropathic pain conditions including CIPN and diabetic peripheral neuropathy (DPN). To test the role of SGC-derived mitochondria in neuropathic pain, we adoptively transferred MitoTracker-labelled SGCs (treated with vehicle or PTX) into mouse DRG. Mitochondrial signals appeared in both neurons and GFAP⁺ SGCs and extended into spinal nerve axons, indicating anterograde transport (Extended Data Fig. 12a–e). PTX-treated SGCs exhibited impaired mitochondrial bioenergetics (Extended Data Fig. 12c), and their transfer did not support axonal mitochondrial transport (Extended Data Fig. 12d,e), indicating that dysfunctional SGC mitochondria cannot sustain long-range neuronal delivery. To determine whether dysfunctional SGC mitochondria affect axonal growth, we performed SGC–neuron co-cultures. PTX-treated SGCs impaired axonal growth, an effect that was reversed by CytoB, implicating TNT-mediated transfer (Extended Data Fig. 12f,g). In vivo, intra-DRG CytoB induced dose-dependent mechanical hypersensitivity (days 1–5), similar to the mitochondrial toxin antimycin A (Extended Data Fig. 12h,i). Both treatments also reduced intraepidermal nerve fibre (IENF) density (Extended Data Fig. 12j). Similarly, low-dose PTX reduced IENFs two weeks after treatment (Extended Data Fig. 12k,l). Collectively, these findings demonstrate that chemotherapy damages SGCs, and impaired SGC-to-neuron mitochondrial transfer contributes to IENF loss and CIPN.

Extended Data Fig. 12. Chemotherapy impairs mitochondrial transfer from SGCs to nerve axons and causes nerve degeneration.

(a) Schematic of adoptive transfer of SGCs to DRG. (b) Mitochondrial fluorescence in neurons and SGCs in DRG. White and red arrows indicate MitoTracker⁺ SGCs and neurons, respectively. (c) PTX treatment decreased OCR in cultured SGCs. n = 6 per group. *** P = 0.0004. (d) Mitochondrial fluorescence signal in the spinal nerve after adoptive transfer of SGCs. (e) Quantification of integrated density of MitoTracker fluorescence in vehicle (n = 6) and PTX-treated (n = 5) group. **** P < 0.0001. (f) Left, schematic of neuron-SGC co-cultures. Right, βIII-tubulin immunostaining showing axonal outgrowth of neurons. (g) Quantification of axonal outgrowth in neuron-SGC co-cultures. n = 15 neurons per group from three independent experiments. **** P < 0.0001, *** P = 0.0002, * P = 0.0472. (h) Schematic of microinjection of CytoB and Antimycin A into DRG. (i) Paw withdrawal threshold decreased after CytoB and Antimycin A treatment. n = 6/group. *** P = 0.0008, P = 0.3112, ** P = 0.0012; *** P = 0.0008, P = 0.9875, *** P = 0.0005; ** P = 0.001, P = 0.07, *** P = 0.0005; *** P = 0.0008, ** P = 0.0029, *** P = 0.0006; *** P = 0.0008, ** P = 0.0065, *** P = 0.0005. (j) Representative images of PGP9.5 staining and quantification of IENF density. Asterisks (*) indicate IENFs. n = 6/group. ** P = 0.0015, *** P = 0.0004. (k) Representative images of PGP9.5 staining (k) and quantification of IENF density (l) in naïve and PTX-treated group. n = 5/group. **** P < 0.0001. Data are shown as means ± s.e.m. and were statistically analyzed by unpaired t-test (two-sided, c, e, and l), One-way ANOVA followed by Tukey’s multiple comparisons test (g and j), and Two-way ANOVA followed by Tukey’s multiple comparisons test (i). n.s., not significant.

Source Data

Role of MYO10 and TNTs in human SGCs

Mouse and human SGCs exhibit distinct transcriptomes, which may hinder translation of pain therapeutics³³. To examine whether TNT-LS exist in human DRG, we performed SEM on DRG tissues provided by National Disease Research Interchange (NDRI) from healthy donors and those with diabetes (Fig. 4a, Extended Data Fig. 13a–h and Supplementary Table 1). Immunostaining confirmed SGC-specific expression of FABP7 (Fig. 4b), with inner SGCs closely enwrapping neurons and others positioned 10–20 μm away as outer SGCs. High-resolution SEM imaging revealed TNT-LS connecting SGCs to neurons and to each other, often containing vesicle-like bulges (Fig. 4c and Extended Data Fig. 13d,e). To profile human SGCs, we performed single-nucleus RNA sequencing (snRNA-seq) on DRG from two donors. Clustering of 11,576 human DRG nuclei identified 17 clusters and 6 major cell types (Fig. 4d,e, Extended Data Fig. 14a–f and Supplementary Table 2). SGCs represent the largest cell population in the human DRG, comprising 31.1% of total cells (Fig. 4f) and separated into two clusters, cluster 1 and cluster 2, on the basis of gene expression profiles (enriched genes listed in Supplementary Table 3). Consistent with a prior report³⁴, human SGCs were identified by the expression of FABP7 and EDNRB (Extended Data Fig. 14a). MYO10 was highly enriched in SGCs (Fig. 4g), with a violin plot showing much higher expression in SGCs than neurons (P < 0.0001; Fig. 4h). RNAscope in situ hybridization confirmed MYO10 colocalization with FABP7 in SGCs (Fig. 4i). Notably, MYO10 expression in SGCs was downregulated in DRG from donors with diabetes compared with those without (Fig. 4i, j).

Fig. 4. MYO10 expression, TNT formation and mitochondrial transfer in human DRG SGCs and the impact of diabetes.

a, Schematic of sectioned human DRG processed with SEM and immunostaining using adjacent sections. b, Left, representative three-dimensional SEM images. Middle and right, staining of FABP7 (SGCs), Nissl (neurons) and DAPI (nuclei). Neurons are labelled N1 to N8, and N1 is enlarged in the bottom panels. SGCs are labelled S1 to S18. Asterisks indicate outer SGCs. c, High-magnification SEM images showing TNT-LS between an SGC and a neuron (left) and between two SGCs (right). Arrows point to TNT-LS with bulges. Scale bars, 1 μm. d, Schematic for snRNA-seq. e, Uniform manifold approximation and projection (UMAP) of 11,576 DRG nuclei identifies 17 clusters corresponding to 6 major cell types, including SGCs, neurons, connective tissue cells, endothelial cells, immune cells and Schwann cells. f, Proportions of six major DRG cell types. g, UMAP plot showing MYO10 expression across different clusters. h, Violin plot illustrating distinct MYO10 expression levels in single SGCs and neurons. Two-sided Mann–Whitney test. i, In situ hybridization of MYO10 in human DRG tissue from non-diabetic and diabetic donors. Arrows indicate MYO10 mRNA puncta that co-localize with FABP7. Scale bars, 20 μm. j, Quantification of MYO10 puncta surrounding individual DRG neurons per 3,000 μm² area. Two-sided unpaired t-test. n = 4 donors without diabetes; n = 5 donors with diabetes; 4 images (each represented by a small dot) from each donor. k, Schematic of primary SGC–neuron co-culture from non-diabetic and diabetic DRG. l, Representative images of MitoTracker in SGC–neuron co-cultures. Red arrows point to mitochondria within TNTs. m, Quantification of MitoTracker density as the integrated fluorescence intensity in neurons, normalized to the corresponding intensity in interacting SGCs. Two-sided unpaired t-test. n = 6 neurons from 2 independent experiments (no diabetes); n = 9 neurons from 3 independent experiments (diabetes). Data are mean ± s.e.m.

Source Data

Extended Data Fig. 13. SEM images of human DRG from non-diabetic and diabetic donors.

(a) Schematic of non-diabetic human donors for whole mount DRG experiment and SEM imaging. (b) Left, low-magnification SEM image. Middle, enlarged box area from the left. Right, enlarged boxed area from the middle. Arrow points a tube-like structure with a bulge. n = 3 biological repeats. (c) Schematic of non-diabetic human donors for sectioned DRG experiment and SEM imaging. (d) low magnification image of N1 and N2 neurons. (e) Left, enlarged boxed area of d, showing multiple SGCs (S1 to S9). Right, enlarged boxed area of the left, showing TNT-like structure with a bulge (arrows) near S1. n = 3 biological repeats. (f) Schematic of diabetic human donors for sectioned DRG experiment and SEM imaging. (g) Left, SEM image of a neuron (N) and surrounding SGCs (S1 to S7). Right: enlarged boxed area from the left, showing micrometer gap between SGCs and neurons and an unsmooth TNT-LS (indicated by a blue arrow). n = 3 biological repeats. (h) A neuron is surrounded by multi-layers of non-neuronal cells (presumably SGCs) in diabetic DRG (indicated by black circles). n = 3 biological repeats. Scale bars are as indicated.

Extended Data Fig. 14. snRNA-seq transcriptomic analysis reveals six major cell populations in healthy human DRG.

(a-b) UMAP plots showing SGC markers FABP7 and EDNRB (a) and immune cell markers CSF1R and LY86 (b). (c) UMAP plots reveal seven neuronal populations including: 1) neurofilament population (FXYD7+), 2) Aβ-LTMR (HS3ST4 + ), 3) proprioceptors (NXPH1+), 4) peptidergic nociceptors (GAL + ), 5) C-fiber low threshold mechanoreceptor (GFRA2+/ and POU4F2+), 6) somatostatin-positive pruriceptors (SST+), and 7) cold-sensing TRPM8+). (d-f) UMAP plots showing Schwann cell markers MPZ and GLDN (d), connective tissue marker COL1A1 (e), and endothelial cell marker PECAM1 (f).

We next examined morphological changes of SGCs and TNTs in human DRG from donors with or without diabetes. DRG from individuals with diabetes showed enlarged neuron–SGC gaps (0.1 μm in control versus 1–4 μm in diabetes), irregular and unsmooth TNTs, and multilayered SGCs (Extended Data Fig. 13g,h). Consistently, SEM of CIPN mice two weeks after PTX revealed enlarged gaps compared with naive and recovered mice, with TNTs appearing more visible owing to gap expansion (Extended Data Fig. 3c–f). Whereas less than 8% of neurons displayed clear gaps under normal conditions, around 50% did so after PTX (Extended Data Fig. 3d–f). These results imply that neuron–SGC separation, enlarged gaps and disrupted TNTs after chemotherapy impair neuro–glial interactions and contribute to peripheral neuropathy. We tested mitochondrial transfer between SGCs and neurons from human DRG donors with and without diabetes. MitoTracker-labelled SGCs transferred mitochondria to neurons via TNTs in non-diabetic cultures, but this transfer was reduced in diabetic DRG (Fig. 4k–m). Directionality assays showed that transfer was more efficient from SGCs to neurons than the reverse (Extended Data Fig. 15a–d). Notably, 83.3% of neurons received mitochondria, with 35.3% being associated with TNT structures (Extended Data Fig. 15e). These findings suggest that neurons preferentially acquire mitochondria from SGCs, probably reflecting their high energy demands.

Extended Data Fig. 15. Comparison of mitochondrial transfer from SGCs to neurons versus neurons to SGCs in human DRG cell cultures.

(a) Schematic of two different mitochondrial transfer approaches between SGCs and neurons: 1) MitoTracker labeled SGCs and co-culture with neurons and 2) MitoTracker labeled neurons and co-culture with SGCs. (b-c) Representative images of MitoTracker labeled mitochondrial transfer from SGC to neuron (b) and from neuron to SGC (c). White arrows point to TNTs, and red arrows point to mitochondria within TNTs. Scale bar: 20 μm. (d) Quantification of mitochondrial transfer efficiency as the receiver/donor ratio of MitoTracker signal intensity in two groups. Data are shown as means ± s.e.m. Unpaired t-test (two-sided). **** P < 0.0001. n = 21 cells (left column) and n = 25 cells (right column) from 2 to 4 independent experiments. (e) Percentage of MitoTracker-positive and TNT-positive neurons of the SGC-neuron co-cultures from human DRG. A total of 54 neurons from 6 independent experiments were included for quantification.

Source Data

Adoptive transfer to mitigate neuropathic pain

To investigate whether adoptive transfer of healthy SGCs alleviates neuropathic pain via mitochondrial transfer, we injected human SGCs from healthy donors into the lumbar DRG of diabetic db/db mice. Pretreatment of SGCs with siRNA targeting MYO10 (siMYO10) abolished the analgesic effect observed with control siRNA-treated SGCs, which reduced mechanical hypersensitivity and enhanced mitochondrial function (Fig. 5a–c). To further evaluate the role of MYO10, we transferred SGCs from wild-type or Myo10^+/− mice into PTX-treated mice, followed by behavioural testing (Fig. 5d). Wild-type SGCs reduced mechanical hypersensitivity within 1–2 days, whereas Myo10^+/− SGCs did not do so (Fig. 5e). These results indicate that adoptively transferred SGCs can alleviate neuropathic pain in the CIPN model and that MYO10 is critical for SGC-to-neuron mitochondrial transfer via TNTs. We next explored direct mitochondrial transfer as a strategy for managing neuropathic pain. MitoTracker-labelled mitochondria isolated from cultured SGCs were injected into the DRG (Extended Data Fig. 16a), where they localized to both neurons and SGCs without TNT formation (Extended Data Fig. 16b,c). CytoB had no effect on mitochondrial uptake; however, Pitstop2 reduced MitoTracker signals (Extended Data Fig. 16d–f), indicating that endocytosis mediates internalization of adoptively transferred mitochondria.

Fig. 5. Adoptive transfer of SGCs or SGC-derived mitochondria protects against axonal degeneration and alleviates neuropathic pain.

a, Schematic illustrating human SGC culture treated with siCtrl or siMYO10 and intra-DRG microinjection into db/db diabetic mice. b, Results of von Frey test showing the analgesic effect of transferred SGCs, which is reversed by siMYO10 treatment. 1 day vehicle versus siCtrl SGCs: **P = 0.0099; 2 days vehicle versus siCtrl SGCs: **P = 0.006; 3 days vehicle versus siCtrl SGCs: *P = 0.0241. c, OCR of DRG from mice treated with vehicle, or with SGCs pretreated with siCtrl or siMYO10. n = 6 per group. **P = 0.0048. d, Schematic of SGCs from wild-type and Myo10^+/− mice and intra-DRG microinjection. e, Analgesic effect following intra-DRG injection of wild-type SGCs. 1 day vehicle versus wild-type SGCs: *P = 0.0225; 2 days vehicle versus wild-type SGCs: *P = 0.0251. f, Schematic of human SGC culture, mitochondria isolation and intra-DRG injection. g, Analgesic effect of mitochondrial transfer from non-diabetic human SGCs. 6 h: ***P = 0.0009 (vehicle versus non-diabetic SGCs), *P = 0.0157 (vehicle versus diabetic SGCs), *P = 0.0295 (diabetic versus non-diabetic SGCs). 1 day: ****P < 0.0001 (vehicle versus non-diabetic SGCs). 2 days: **P < 0.0014 (vehicle versus non-diabetic SGCs), *P < 0.0123 (vehicle versus diabetic SGCs), ** P < 0.0054 (diabetic versus non-diabetic SGCs). h, Schematic of mitochondrial isolation and intra-DRG injection. i, Analgesic effect of mitochondrial transfer from a non-diabetic donor. Mitochondria from non-diabetic SGCs, baseline versus 1 day: *P = 0.0417; 2 days diabetic versus non-diabetic SGCs: *P = 0.0442. j, Left, images of PGP9.5 staining. Asterisks indicate IENFs. Scale bar, 50 μm. Right, quantification of IENFs. n = 7 per group. *P = 0.0475. Data are mean ± s.e.m. Two-way ANOVA with Tukey’s multiple comparisons tests (b,e,g,i); one-way ANOVA followed by Tukey’s multiple comparisons (c); two-sided unpaired t-test (j). k, Schematic showing potential therapeutic strategies using mitochondrial transfer for neuropathic pain: (1) adoptive transfer of SGCs; and (2) transfer of isolated SGC mitochondria into the DRG.

Source Data

Extended Data Fig. 16. Effects of adoptive transfer of mouse mitochondria on evoked and spontaneous pain in paclitaxel-injected mice and OCR analysis in human DRG SGCs.

(a-f) Mitochondrial uptake by DRG cells following intra-DRG injection of MitoTracker labeled mitochondria. (a) Schematic of mitochondrial isolation from MitoTracker labeled SGCs and intra-DRG injection. (b-e) Low magnification DRG images showing triple staining of MitoTracker labeled mitochondria (red), FABP7-labeled SGCs (green), and NeuN-labeled neurons (blue) from untreated mice (b-c), CytoB (3.5 μM)-treated mice (d), and Pitstop2 (25 μM)-treated mice (e). (c) Enlarged image from the box in b showing adoptively transferred mitochondria in both NeuN+ neuron (blue arrow) and FABP7+ SGCs (white arrows). Bottom panels in (d-e) are enlarged images from the boxed areas in top panels. Scale bars are as indicated. (f) Mito density quantification shows that MitoTracker+ mitochondrial uptake in DRG neurons and SGCs is blocked by Pitstop2, not by CytoB. n = 4 (Mitochondrial + vehicle), n = 3 (Mitochondrial + CytoB), and n = 3 (Mitochondrial + Pitstop2). ** P = 0.0066, P = 0.3434; n.s., no significance. (g-j) Paclitaxel-induced evoked pain and spontaneous pain and the effects of mitochondrial transfer. (g) Schematic of mitochondrial isolation from mouse SGCs, treatment with the mitochondrial complex III inhibitor myxothiazol (1 mM, 10 min), and intra-DRG injection in the PTX model. (h) Analgesic effects of adoptively transferred mitochondria and its blockade by myxothiazol. n = 10/group. **** P < 0.0001, *** P = 0.0003; **** P < 0.0001, *** P = 0.0003. (i-j) Assessment of spontaneous pain with conditioned place preference test (CPP) scores. * P = 0.0224. n = 10 mice/group. (k) OCR analysis of SGC cultures of human DRG from diabetic and non-diabetic donors, showing mitochondrial dysfunction in diabetes. n = 10 cultures/group. * P = 0.0477. Data are shown as means ± s.e.m. and were statistically analyzed by One-way ANOVA followed by Tukey’s multiple comparisons test (f), Two-way ANOVA followed by Sidak’s multiple comparisons test (h), and unpaired t-tests (two-sided, j and k). Mice from both sexes were included for analysis.

Source Data

We isolated mitochondria from primary mouse cultures of SGCs, treated them with vehicle or the mitochondrial complex III inhibitor myxothiazol²⁰, and injected them into the DRG of PTX-injected mice (Extended Data Fig. 16g). Transfer of healthy mitochondria mitigated mechanical pain, whereas transfer of myxothiazol-treated mitochondria abolished this effect (Extended Data Fig. 16h). We also performed conditioned place preference (CPP) to measure ongoing pain³⁵ and found increased CPP scores in the mitochondria-treated group (Extended Data Fig. 16i,j). Furthermore, we conducted a cross-species mitochondrial transfer by administration of human mitochondria, isolated from human SGCs of non-diabetic and diabetic donors, into the lumbar DRG of PTX-treated mice (Fig. 5f). Seahorse measurement showed reduced basal OCR in diabetic SGCs compared with non-diabetic controls (Extended Data Fig. 16k). Mitochondria from healthy SGCs reduced mechanical pain for two days, whereas those from diabetic SGCs produced only transient relief (6 h; Fig. 5g). In db/db mice, transfer of mitochondrial from non-diabetic but not diabetic SGCs reduced neuropathic pain (Fig. 5h,i) and increased IENF density in hindpaw skin (Fig. 5j). Taken together, these in vivo results support our in vitro findings and highlight the therapeutic potential of SGC-derived mitochondria in treating nerve degeneration and neuropathic pain.

Discussion

Our study provides comprehensive in vitro, ex vivo and in vivo evidence that SGCs transfer mitochondria to adjacent sensory neurons in the DRG, thus revealing a protective role against peripheral neuropathy. Although TNT-mediated mitochondrial transfer was first identified two decades ago⁹, in vivo evidence has been elusive. Our scanning and TEM imaging revealed TNT-like tubes containing mitochondria in mouse and human DRG. We also identified MYO10 as a key regulator of TNT formation between SGCs and neurons in the DRG (Supplementary Fig. 5). MYO10 is highly enriched in SGCs, as demonstrated by in situ hybridization and snRNA-seq. Our knockdown experiments demonstrate that MYO10-mediated TNT formation is essential for SGC-to-neuron mitochondrial transfer in vitro and in vivo. In addition, endocytosis and connexin 43-mediated gap junction communication are also critically involved. Moreover, connexin 43-containing gap junctions not only regulate pain³ but also stabilize TNTs and facilitate mitochondrial transfer¹⁰.

Mitochondrial transfer has been implicated in diverse diseases, including obesity³⁶, stroke¹⁴, inflammatory pain²⁰ and cancer^37,38. Our study demonstrates that SGC-to-neuron mitochondrial transfer occurs under physiological conditions, and its dysregulation drives neuropathic pain in animal models of nerve injury, CIPN and DPN. Mitochondrial dysfunction, a hallmark of CIPN² and DPN¹, is associated with pain, tingling and paraesthesia³⁹, along with IENF loss⁴⁰. Our findings provide direct evidence that impaired mitochondrial transfer from SGCs to DRG neuronal cell bodies is sufficient to trigger IENF degeneration and neuropathic pain behaviours. Small fibre neuropathy is common in chronic pain conditions, including CIPN, DPN and fibromyalgia⁴¹, We found that SGC-derived mitochondrial transfer preferentially targets medium- and large-sized neurons, but not small nociceptors, providing mechanistic insights into small fibre neuropathy.

Mitochondria represent important drug targets in neurodegenerative disorders and metabolic diseases⁴². Our gain-of-function approaches propose multiple strategies for mitochondrial therapy in peripheral neuropathy, including adoptive transfer of SGCs to the DRG and direct delivery of SGC-derived mitochondria (Fig. 5k and Extended Data Fig. 17). Cross-species transfer of human mitochondria transfer to mouse DRG also confers protection, consistent with reports of rapid and efficient interspecies mitochondrial fusion⁴³. A limitation of this study is that Aldh1l1 is expressed in other glia, such as astrocytes; future work will use lines with greater SGC specificity. Further studies are needed to validate TNT-like structures using higher-resolution and more specific electron microscopy approaches. In summary, SGCs regulate pain via multiple neuro–glial pathways involving ATP, cytokines, potassium channels and gap junctions^3,33. We identified a previously unrecognized mechanism in which SGCs transfer mitochondria to sensory neurons, protecting against neuropathic pain. We demonstrated TNT-like ultrastructures in mouse and human DRG and established that MYO10-mediated mitochondrial transfer is a key protective pathway. These findings offer mechanistic insight into small fibre neuropathy and highlight mitochondrial transfer as a potential therapeutic strategy.

Extended Data Fig. 17. Schematic illustrations of mitochondrial transfer from SGCs to neurons in DRG in health, disease (CIPN/DPN), and treatment conditions.

Top, SGC-neuron mitochondrial transfer in health conditions in physiological status. Middle, disruption of SGC-neuron mitochondrial transfer in peripheral neuropathy-associated diseases, such as chemotherapy-induced peripheral neuropathy (CIPN) and diabetic peripheral neuropathy (DPN), induces nerve degeneration and neuropathic pain. Bottom, therapeutic approaches targeting mitochondrial transfer via SGC transfer or mitochondrial transfer can enhance nerve regeneration and alleviate neuropathic pain.

Methods

Animals

Transgenic male and female mice (8 to 12 weeks of age) were used for behavioural and biochemical experiments. Trpv1-Cre mice (JAX: 017769), Aldh1l1-cre^ERT2 mice (JAX: 031008), Advillin-Cre mice (JAX: 032536), Cx3cr1-cre^ERT2 mice (JAX: 020940), Ai9 mice (JAX: 007909), GCaMP6f mice (JAX: 024105), MitoTag mice (JAX: 032675), and BKS.Cg-Dock7^{m +/+} Lepr^db/J mice (JAX: 000642) were purchased from Jackson Laboratory. Male and female CD1 mice were purchased from Charles River Laboratories. Generation of Myo10 knockout mice (tm1d) and mouse genotyping were described previously³¹. Mice were maintained at the Duke animal facility. db/db mice aged 12–16 weeks were used. All mouse experiments were approved by the Duke University Institutional Animal Care and Use Committee (IACUC). Mice were housed in an AAALAC-accredited animal facility under a 12 h:12 h light:dark cycle, with food and water provided ad libitum. All animals were housed at 22 ± 1 °C and 30–70 % humidity.

Human DRG tissues

A total of 16 human DRG samples were obtained from donors with and without diabetes (Supplementary Table 1) through the NDRI, under exemption approval from the Duke Institutional Review Board (IRB, Pro00051508). Diabetes was identified on the basis of medical history provided in the NDRI reports.

Reagents

PTX (T7402), Tam (T5648), oligomycin (O4876), carbonyl cyanide 4-(trifluoromethoxy) phenylhydrazone (C2920), rotenone (R8875), antimycin A (A8674), myxothiazol (T5580), CBX (C4790), CytoB (C2743) and Y-27632 (688001) were obtained from Sigma. Pitstop2 (ab120687) was obtained from Abcam. CPZ was obtained from MCE (HY-15640). TTX (1078) was obtained from Tocris. siMyo10 (siRNA ID: 157002), siMYO10 (siRNA ID: 118391) and negative control siRNA (4459405) were obtained from Ambion Life Technology.

Primary cultures from mouse DRG

Mouse DRG neuron culture

Mouse primary DRG neuron cultures were prepared as described⁴⁴. In brief, DRG were digested with collagenase (0.2 mg ml⁻¹, Sigma, 10103578001) and Dispase-II (3 mg ml⁻¹, Sigma, D4693) for 60 min at 37 °C. The cells were then mechanically dissociated with a pipette, filtered through a 70-µm nylon mesh, and centrifuged at 300g for 5 min. The cells were plated on glass coverslips coated with 0.1 mg ml⁻¹ poly-d-lysine (Thermo, A3890401) and cultured in Neurobasal medium supplemented with 10% FBS, 2% B27 supplement, and 1% penicillin-streptomycin (Neurobasal-supplemented medium) at 37 °C with 5% CO₂.

Mouse DRG SGC culture

Mouse DRG SGCs were prepared as previously described¹⁵. DRG were enzymatically digested with collagenase and Dispase-II for 90 min at 37 °C. The tissue was then mechanically dissociated using a pipette, filtered through a 70-μm nylon mesh, and centrifuged at 300g for 5 min. Cells were seeded onto 35-mm cell culture dishes (VWR, 10861-656) without poly-d-lysine coating and maintained in Neurobasal-supplemented medium at 37 °C with 5% CO₂. On day 3 in vitro (DIV3), non-SGCs—including neurons—were removed by vigorously shaking the culture dishes by hand for 15–20 s, leaving behind a purified adherent layer of SGCs.

Mouse DRG SGC–neuron co-cultures and mitochondrial transfer analysis

For in vitro mitochondrial transfer experiments, DIV5 SGCs were labelled with 20 nM MitoTracker Deep Red (Thermo Fisher, M22426) in culture medium at 37 °C with 5% CO₂ for 30 min. Cells were then washed three times with PBS. MitoTracker-labelled SGCs were collected using 0.25% Trypsin-EDTA (Gibco, 25200-056) and seeded onto coverslips pre-seeded with cultured DRG neurons. Mitochondrial transfer inhibitors, including the TNT formation inhibitor CytoB (3.5 μM) and Y-27632 (10 μM), the endocytosis inhibitor Pitstop 2 (20 μM) or the gap junction blocker CBX (20 μM), were added to the SGC–neuron co-cultures. After 24 h of incubation, cells were imaged using a Zeiss LSM 880 inverted confocal microscope.

Immunocytochemistry in mouse DRG SGC cultures

Primary cultured SGCs were fixed in 4% paraformaldehyde (PFA), then incubated overnight at 4 °C with the primary antibodies: GFAP (mouse, 1:400, Millipore Sigma, G3893), FABP7 (mouse, 1:1,000, Neuromics, MO22188), Kir4.1 (rabbit, 1:500, Alomone, APC-035), AQP4 (rabbit, 1:500, Proteintech, 16473-1-AP), connexin 43 (CX43; rabbit, 1:500, Zymed, 71-0700), glutamine synthetase (GS; rabbit, 1:500, Novus, NB110-41404), and MYO10 (rabbit, 1:500, Sigma, HPA024223). After PBS washes, cells were incubated with appropriate fluorescent secondary antibodies for 1 h at room temperature. Fluorescent images were acquired using a Zeiss LSM 880 confocal microscope.

DRG neuron cultures and assessment of neurite outgrowth

Primary mouse DRG neurons were co-cultured with SGCs for 72 h and subsequently fixed with 4% PFA. Fixed cells were subjected to immunocytochemistry using a βIII-tubulin antibody (1:1,000, Abcam, ab18207) overnight at 4 °C. Immunofluorescence images were acquired with a Nikon fluorescence microscope, and neurite length was quantified using the SNT plugin in ImageJ.

Human DRG SGC–neuron co-cultures and mitochondrial transfer analysis

Postmortem L2–L5 DRG were collected from diabetic and healthy donors and delivered in an ice-cold cell culture medium to the laboratory at Duke University within 24–48 h postmortem. Human DRG cultures were prepared as described⁴⁵. DRG were digested with collagenase and Dispase-II for 120 to 150 min at 37 °C. Cells were then mechanically dissociated using pipettes and centrifuged (300g for 5 min). Cells were seeded onto poly-d-lysine-coated glass coverslips for neuron culture and into uncoated dishes for SGC culture, and grown in Neurobasal-supplemented medium. On DIV2, non-SGCs were removed by vigorously shaking the culture dishes by hand for 15–20 s. Human SGCs were incubated with 20 nM MitoTracker Deep Red in culture medium at 37 °C with 5% CO₂ for 30 min, then co-cultured with human DRG neurons for 24 h.

Whole-mount DRG co-cultured with mouse SGCs and macrophages

Bone marrow-derived macrophages were prepared as described²⁰. In brief, L-929 cells (CCL-1) were obtained from ATCC and authenticated by ATCC. Bone marrow was collected from mice and cultured in DMEM supplemented with 10% heat-inactivated fetal bovine serum, 1% penicillin-streptomycin, and 20% L-929 cell-conditioned medium at 37 °C with 5% CO₂. Macrophages or SGCs were seeded on coverslips and labelled with 20 nM MitoTracker Deep Red then co-culture with the whole-mount DRG. CytoB, (3.5 μM) or vehicle was added to medium. Images were captured using a Zeiss LSM 880 inverted confocal microscope.

siRNA transfection in vitro

Transfection was performed using Lipofectamine 2000 (Thermo Fisher, 11668027). In brief, siRNA targeting Myo10, MYO10 and control siRNA was diluted in 100 μl Opti-MEM, and 0.4 μl of Lipofectamine 2000 was diluted in a separate 100 μl Opti-MEM solution. Two solutions were combined and incubated at room temperature for 5 min before being added to the culture dish to make a 15 nM siRNA concentration. The culture medium was replaced 6 h after transfection. siRNA sequences are included as follows: siMyo10: sense, CCUACAAGCAGAGUACAAUtt; antisense, AUUGUACUCUGCUUGUAGGtg; siMYO10: sense: GGUAUUCACUUACAAGCAGtt; antisense: CUGCUUGUAAGUGAAUACCtg.

Mitochondrial health measurement

Mitochondrial membrane potential was assessed using the JC-1 dye (Invitrogen, T3168), a commonly used indicator of mitochondrial health⁴⁶. Mouse primary SGCs were incubated with JC-1 (2 μg ml⁻¹) for 30 min at 37 °C. JC-1 accumulates in mitochondria in a membrane potential-dependent manner: monomeric forms emit green fluorescence (excitation 485 nm/emission 516 nm), while aggregated forms emit red fluorescence (excitation 579 nm/emission 599 nm), with higher red signal indicating healthier mitochondria. JC-1 fluorescence was monitored in real time using a VivaView FL Incubator Fluorescence Microscope (Olympus) over 24 h. Differential interference contrast images were automatically captured every 30 min at multiple positions.

SIM imaging of mitochondria

Mitochondria of SGCs were labelled with 20 nM MitoTracker Green (Thermo, M7514). SIM imaging was performed using an inverted Zeiss Elyra 7 microscope equipped with a 63× oil-immersion objective. SIM² image processing was conducted using Zeiss Zen Black software. Mitochondrial length was quantified using the Mitochondria Analyzer plugin in ImageJ (v.1.53q).

OCR measurement

OCR was measured using an XFe96 Seahorse Extracellular Flux Analyzer (Agilent) as described¹¹. In brief, primary DRG neurons and SGCs were seeded in XFe96 cell culture microplates (Agilent, 101085-004) and incubated at 37 °C with 5% CO₂ before measurement. On the day of the experiment, cells were washed and placed in Seahorse XF base medium (Agilent, 103575-100) containing 10 mM glucose (Agilent, 103577-100), 1 mM pyruvate (Agilent, 103578-100) and 2 mM glutamine (Agilent, 103579-100). Each assay cycle consisted of 3 min of mixing, 1 min of waiting, and 3 min of OCR measurements. Following three baseline OCR measurements, oligomycin (0.5 μM), FCCP (2 μM), and rotenone/antimycin A (0.5 μM each) were sequentially added. OCR values were normalized to the cell number. Fresh tissue mitochondrial bioenergetics were also measured with the XFe96 Seahorse Extracellular Flux Analyzer as described⁴⁷. In brief, sciatic nerves or whole-mount DRG were freshly isolated and placed into XFe96 spheroid microplates (Agilent, 102978-100) containing Seahorse XF base medium supplemented with 5.5 mM glucose, 0.5 mM sodium pyruvate and 1 mM glutamine. OCR was normalized to protein content. Oligomycin (12 μM), FCCP (20 μM), and rotenone/antimycin A (20 μM each) were sequentially injected during the assay. An assay cycle of 3 min mixing, 3 min waiting, and 4 min measurement was repeated 3 times for baseline rates and after each port injection.

Calcium imaging in mouse DRG neuron–SGC co-cultures

DRG primary neurons were cultured from Advillin:GCaMP6f mice. Cells were plated on coverslips precoated with poly-D-lysine (Corning, 354087) and grown in a Neurobasal- supplemented medium. Neurons were co-cultured with SGCs following the previously described procedures. The calcium imaging buffer composition was as follows: 140 mM NaCl, 5 mM KCl, 2 mM CaCl₂, 1 mM MgCl₂, 10 mM HEPES, and 10 mM glucose. The buffer solution was perfused through the culture during baseline acquisition, followed by the addition of 100 nM capsaicin, and 60 mM KCl. The calcium indicator Fura2-AM was used for calcium imaging for the TRPV1 antagonist experiment. Dissociated mouse DRG neurons were loaded with 5 μM Fura2-AM (Invitrogen, Thermo Fisher Scientific, F1221) for 45 min and then replaced with calcium imaging buffer. Cells were pretreated with CPZ (100 μM)⁴⁸ for 3 min prior to capsaicin perfusion. This perfusion procedure was applied uniformly across all four experimental groups. Calcium signals were captured and expressed as F values, representing fluorescence intensity⁴⁹.

ROS measurement in mouse DRG neuron–SGC co-cultures

ROS levels were measured using the ROS Assay Kit (Thermo Fisher, 88-5930-74) following the manufacturer’s instructions. Primary DRG neuron cultures were assigned to four experimental groups with overnight treatment: (1) vehicle control; (2) PTX (1 μg ml⁻¹) (3) PTX with SGC co-culture; and (4) PTX with SGC co-culture plus CytoB (3.5 μM).

SEM of mouse DRG and spinal cord

For whole-mount mouse DRG and spinal cord imaging, fresh L4–L5 DRG and lumber spinal cord were collected and cut with Vannas spring scissors (FST, 15019-10) to create a flat surface. DRG and spinal cord were fixed in 3% glutaraldehyde in PBS buffer for 1 h at room temperature after tissue collection. After dehydration with 100% ethanol overnight at 4 °C, the samples were ready for the next preparation step. To examine whether trypsin affects the structure of DRG (Extended Data Fig. 3a,b), DRG tissues were treated with 0.25% trypsin for 20 min at 37 °C, followed by hydrolysis with 8 N HCl at room temperature for 20 min, and then fixed in 3% glutaraldehyde, according to a previously described SEM protocol for DRG²¹. To improve the visualization of TNTs, mice were transcardially perfused with PBS and followed by 4% PFA, and then L4–L5 DRG were collected and immersed in 30% sucrose for over 3 nights at 4 °C. Subsequently, DRG were sectioned by a cryostat (Leica CM 1950). Alternatively, DRG were directly placed in 3% glutaraldehyde for 1 h at room temperature for fixation, followed by dehydration with 100% ethanol overnight at 4 °C. Sample preparation by the critical point drying (Ladd CPD3), then the DRG and spinal cord samples were sputter-coated with gold for 300 s at 12 mA (Denton Desk V). Imaging was performed using an Apreo 2 scanning electron microscope (ThermoFisher Scientific) with a 2.00 kV accelerating voltage and 25 pA emission current at Duke University Shared Materials Instrumentation Facility. Multiple detectors (ETD, T1, T2 and T3) were utilized, and immersion mode with T1, T2 and T3 detectors was used to capture high-magnification images.

SEM and immunostaining of human DRG

Fresh human DRG (L2–L5) were fixed overnight in 4% paraformaldehyde upon delivery, and then immersed in 30% sucrose for at least 3 nights at 4 °C. Free-floating sections (30 μm) were cut using a cryostat. Some sections were processed for SEM, while adjacent sections were used for immunohistochemistry (IHC). For SEM, imaging was performed using an Apreo 2 scanning electron microscope. Multiple detectors (ETD, T1, T2, and T3) were used. For immunohistochemistry, adjacent sections were blocked with 5% bovine serum albumin (BSA) for 1 h at room temperature, followed by overnight incubation at 4 °C with anti-FABP7 antibody (rabbit, 1:100). The next day, sections were incubated with Alexa Fluor 488-conjugated anti-rabbit secondary antibody (1:400) and Nissl/NeuroTrace-640 for 1 h at room temperature. Sections were mounted using DAPI Fluoromount-G and imaged with a Zeiss LSM 780 confocal microscope.

TEM of mouse DRG

Dissected DRG tissue was fixed in 2.5% glutaraldehyde in 0.1 M sodium cacodylate buffer (pH 7.4) for at least 1 h at room temperature, then stored at 4 °C. Samples were washed with 0.1 M sodium cacodylate buffer, followed by post-fixation in 1% osmium tetroxide in 0.1 M sodium cacodylate buffer (pH 7.2) for 2 h at room temperature. After three additional rinses in buffer, tissues were dehydrated through a graded acetone series and embedded in epoxy resin. Semi-thin sections (500 nm) and ultrathin sections (70–80 nm) were cut using a LEICA EM UC7 ultramicrotome (Leica). Ultrathin sections were collected on copper grids, post-stained with uranyl acetate and lead citrate, and imaged using a JEOL 2100 Transmission Electron Microscope (Duke Center for Electron Microscopy and Nanoscale Technology) at an accelerating voltage of 120 kV.

Animal models of neuropathic pain and inflammatory pain

CIPN was induced by intraperitoneal injections of PTX at 2 mg kg⁻¹ every other day for 4 doses. Type 2 DPN was modelled using db/db mice (JAX: 000642). Type 1 diabetes was induced by intraperitoneal administration of STZ (150 mg kg⁻¹; Sigma-Aldrich, S0130)⁵⁰. SNI surgery was performed as previously described⁵¹. In brief, the common peroneal and tibial nerves were ligated and transected, while the sural nerve was left intact. Inflammatory pain was induced by intraplantar injection of 20 μl CFA into the hindpaw⁵².

Nerve blockade by bio-resorbable PEU films for controlled bupivacaine release

We previously utilized bio-resorbable PEU films for the controlled release of local anaesthetics in pain management^26,53. Bupivacaine-loaded films (40% w/w, ~0.3 mg bupivacaine per film) and blank control films were prepared in the laboratory of M.L.B. Bupivacaine-containing films were wrapped around the ipsilateral sciatic nerve during SNI surgery.

Mitochondrial imaging in the DRG and spinal cord of MitoTag mice

MitoTag mice were crossed with Aldh1l1-cre^ERT2, Advillin-cre and Cx3cr1-cre^ERT2 lines. In inducible Cre lines, recombination was induced by Tam administration⁵⁴. Tam was prepared in a solution of 10% ethanol and 90% corn oil (Sigma, C8267). Mice received intraperitoneal injections of Tam at 100 mg kg⁻¹ once daily for 5 consecutive days. DRG and spinal cord tissues were collected for mitochondrial imaging analysis.

Mitochondrial isolation and immunocapture in Aldh1l1-MitoTag mice DRG

Mitochondria were isolated from fresh dissected DRG tissue as described²³. In brief, fresh DRG were dissected from Aldh1l1-MitoTag mice. The samples were then subjected to nitrogen cavitation, and after depressurization, nuclei and debris were removed by centrifugation. The supernatant was filtered through a pre-separation filter (130-041-407; Miltenyi Biotec) to obtain the crude mitochondrial fraction. For immunocapture of mitochondria against eGFP, the crude mitochondrial fraction was incubated with GFP beads for 1 h. LS columns (Miltenyi Biotec) were then used to separate GFP microbead-coated mitochondria (immunoprecipitate) and mitochondria without GFP (supernatant) from the solution. For western blot analysis, GFP antibodies (goat, 1:1,000, Abcam, ab5450) were used to examine the eGFP, and anti-COX4 (rabbit, 1:1,000, Abcam, ab16056) was used as a mitochondrial loading control.

Generation of AAV-MaCPNS2-Syn-jRGECO1a for DRG neuron targeting

AAV production was performed using the following plasmids: pAdDeltaF6 (Addgene #112867), pUCmini-iCAP-AAV.MaCPNS2 (Addgene #185137), and pAAV.Syn.NES-jRGECO1a.WPRE.SV40 (Addgene #100854). HEK293T cells (CRL-3216) obtained from ATCC, and authentication was performed by ATCC through morphological and STR profiling. HEK293T cells were not tested for mycoplasma contamination. HEK293T cells were cultured in DMEM medium for transfection. Cells were transfected with 30 μg pAdDeltaF6, 15 μg pUCmini-iCAP-AAV.MaCPNS2, and 15 μg jRGECO1a plasmids using PEI MAX (Polysciences, 24765). After 72 h, cells were collected and lysed in 4 ml lysis buffer (15 mM NaCl, 5 mM Tris-HCl, pH 8.5) through three freeze-thaw cycles. Lysates were incubated with benzonase (50 U ml⁻¹; Millipore, 70664) for 30 min at 37 °C and then centrifuged at 4,500 rpm for 30 min at 4 °C. The supernatant was layered onto a stepwise iodixanol gradient (15%, 25%, 40% and 60%) and centrifuged at 67,000 rpm for 1.5 h at 18 °C using a Beckman Ti-70 rotor. The viral fraction was collected from the 40%–60% interface, and concentrated using a 100 kDa molecular weight cutoff filter (Millipore, UFC910008). Purified viral aliquots were stored at –80 °C until use.

Simultaneous ex vivo calcium and mitochondrial imaging in DRG neurons from MitoTag mice

AAV-MaCPNS2-Syn-jRGECO1a (3 × 10¹¹ viral genomes per mouse) was administered intraperitoneally to MitoTag mice at postnatal day 0. Four weeks post-injection, mice underwent SNI surgery and received Tam (100 mg kg⁻¹, intraperitoneal injection) once daily for 5 consecutive days. Ex vivo calcium imaging was performed 3–7 days after surgery. L4–L6 DRG were dissected and incubated in artificial cerebrospinal fluid. Live imaging was conducted using a Zeiss LSM 780 upright confocal microscope equipped with a 20× water-immersion objective. Images were acquired with 50 cycles over a 15 min time-lapse session for each DRG. Data were analysed using FIJI (ImageJ) software.

Intra-ganglionic (DRG) injection of siRNA

siMyo10 and siCtrl were mixed with RVG-9R peptide in 5% glucose at a peptide/siRNA molar ratio of 2:1 before use, following a previously described protocol⁵⁵. Intra-DRG injection was performed as described previously⁵⁶. In brief, a partial laminectomy was performed to expose the left L4 and L5 DRG. A total volume of 1 μl containing 4 μg of siRNA was delivered to each DRG using a gelatin sponge as the delivery matrix.

Intrathecal injection

For intrathecal drug delivery, antimycin A (110 ng), CytoB (70 ng and 350 ng), or Pitstop2 (120 ng) was diluted in 10 μl of PBS and administered via lumbar puncture using a 30-gauge needle between the L5–L6 vertebral levels. Successful intrathecal delivery was confirmed by observing a characteristic tail-flick response⁵⁷.

In vivo labelling of mitochondria in the DRG

Mitochondrial labelling in the DRG was performed based on a previously described protocol with minor modifications⁵⁸. A total volume of 0.6 μl Mitotracker Deep Red FM (2 μM; Thermo, M22426), with or without CytoB, was injected into the L4–L5 DRG using a Hamilton syringe connected to a glass micropipette. Injected DRG were collected 24 h post-injection, embedded in OCT medium, and sectioned at 20 μm thickness using a cryostat. Immunohistochemistry was then performed on these sections using antibodies against GFAP, along with Nissl staining.

Adoptive transfer of SGCs to mice

SGCs were collected from primary cell cultures, and 8,000 cells suspended in 1 μl PBS were injected intra-ganglionically into the L4 and L5 DRG using a Hamilton syringe connected to a glass micropipette. In select experiments, SGCs were pre-labelled with 20 nM MitoTracker Deep Red FM in culture medium for 30 min at 37 °C, followed by three PBS washes prior to injection.

Mitochondria isolation and adoptive transfer of mitochondria via intra-ganglionic injection

Mitochondria were isolated from SGCs following a previously described protocol⁵⁹ with minor modifications. In brief, SGCs were detached from culture dishes, pelleted, and resuspended in MIB buffer (210 mM D-mannitol, 70 mM sucrose, 5 mM HEPES, 1 mM EGTA, and 0.5% (w/v) fatty acid-free BSA, pH 7.2) at 4 °C. Cells were homogenized with 25 strokes and centrifuged at 600g for 5 min at 4 °C. The supernatant was collected and centrifuged at 7,000g for 10 min at 4 °C. Mitochondria isolated from 10,000 SGCs in 1 μl of PBS were injected intra-ganglionically into the L4 and L5 DRG using a Hamilton syringe and glass micropipette. In select experiments, to inhibit mitochondrial oxidative phosphorylation, SGCs were pretreated with 1 mM myxothiazol (Sigma, T5580) for 10 min prior to mitochondrial isolation.

Behavioural tests for pain in mice

For von Frey testing, mechanical sensitivity was assessed with a set of von Frey filaments (Stoelting) with logarithmically increasing stiffness ranging from 0.02 g to 2.56 g, which were applied to the hindpaw plantar surface, and a quick withdrawal or licking response to the stimulus was considered a positive response. PWT was calculated using the up-down method. For assessing mechanical allodynia, paw withdrawal frequency (PWF) to a 0.16 g von Frey filament was measured by observing reflexive withdrawal responses within 10 stimulations, and a quick withdrawal or licking response to the stimulus was considered a positive response. Hargreaves test was used to assess heat sensitivity. Paw withdrawal latency was measured with a Hargreaves radiant heat apparatus (IITC Life Science). CPP assay was used to measure ongoing pain in mice⁴⁴. Mice were habituated for 3 days with 30 min of preconditioning in a two-compartment CPP chamber consisting of white and dark chambers. The baseline behaviour of the mice was recorded on the fourth day using a camera and automatically tracked for 15 min with ANY-Maze software (Stoelting). On the fifth day (conditioning day), mice underwent a 30 min pairing session without intra-DRG injection in the preferred CPP chamber in the morning session, followed by a 30 min pairing session with intra-DRG injection in the non-preferred CPP chamber in the afternoon session. Twenty-four hours later, mice were placed in the CPP test box with access to both chambers, and their behaviour was recorded for 15 min. The time spent in both chambers was analysed using ANY-Maze software. The CPP score, measured in seconds, was calculated as the inverse of the time spent in the preferred chamber, using the formula: post-preference time minus pre-preference time.

Immunohistochemistry in mouse DRG, spinal cord, nerve, and skin tissues

Following terminal anaesthesia with isoflurane, mice were transcardially perfused with PBS, followed by 20 ml of 4% PFA in PBS. The DRG, spinal nerves and hindpaw skins were collected, fixed in 4% PFA and immersed in 30% sucrose at 4 °C for at least 3 nights. Tissues were then embedded in OCT medium (Tissue-Tek), and sections were prepared using a cryostat at the following thicknesses: DRG (20 μm), spinal cord (30 μm), spinal nerve (20 μm), and skin tissue (25 μm). Tissue sections were then blocked in a solution containing 5% BSA and 0.3% Triton X-100 for 1 h at room temperature. Next, sections were incubated overnight at 4 °C with the following primary antibodies: anti-FABP7 (mouse, 1:1,000, Neuromics, MO22188), anti-Iba1 (rabbit, 1:800, Wako, 019-19741), anti-GFAP (mouse, 1:400, Sigma, G3893), anti-MYO10 (rabbit, 1:500, Sigma, HPA024223), anti-TOM20 (rabbit, 1:600, Proteintech, 11802-1-AP) and anti-PGP9.5 (rabbit, 1:200, Thermo, PA5-29012). Following several washes with PBS, the sections were incubated with species-specific secondary antibodies and Nissl/NeuroTracer-640 (1:100, Thermo, N21483). Finally, the sections were mounted with coverslips using DAPI Fluoromount-G mounting medium (Southern Biotech, 0100-20). Images were acquired using a Zeiss LSM 780 confocal microscope. For quantification of immunofluorescence staining, the same acquisition settings were applied to images requiring comparison under different conditions. Three sections were randomly selected from each mouse, and the integrated density of the fluorescence signal per section was measured using ImageJ software (v.1.53q). For intraepidermal nerve IENF analysis in mouse hindpaw skins, all ascending nerve fibre branches crossing into the epidermis were counted. Three randomly selected skin sections were analysed for each mouse.

In situ hybridization and immunohistochemistry in human DRG

Fresh human DRG were immediately fixed in 4% paraformaldehyde overnight at 4 °C upon delivery. Free-floating sections (30 µm) were then cut using a cryostat. For MYO10 in situ hybridization, a human MYO10 RNAscope probe (RNAscope Probe Hs-MYO10, 440691), designed by Advanced Cell Diagnostics was used. The RNAscope protocol was followed according to the manufacturer’s instructions. Following the RNAscope steps, immunohistochemistry was performed as described below. The sections were blocked with 5% BSA for 1 h at room temperature and then incubated with anti-FABP7 antibody (rabbit, 1:100, Thermo, PA5-24949) overnight at 4 °C. Images were captured using a Zeiss LSM 780 confocal microscope with consistent acquisition settings across different samples. The integrated density of the fluorescence signal per section was measured using Image J (v.1.53q).

snRNA-seq and data analysis in human DRG

Human DRG tissues were snap-frozen and stored at –80 °C. Nuclei isolation was performed as previously described⁶⁰. Single-nucleus capture was performed using the 10x Genomics Chromium Single Cell 3′ system (v.3.1). Libraries from individual nuclei samples were pooled and sequenced on an MGISEQ-2000 platform. snRNA-seq reads were processed using Cell Ranger v.4.0.0⁶¹, and alignment to the Human GRCh38 (GENCODE v.32/Ensembl98) reference genome performed using default parameters. Downstream analysis was conducted with Seurat v.4.3.0⁶² using R v.4.3.0. Cells were filtered out if they expressed fewer than 200 genes or had more than 10% mitochondrial reads. Gene expression was then normalized and scaled using the NormalizeData and ScaleData functions from Seurat, respectively, with default settings, as well as the FindVariableFeatures function to identify the highly variable genes. Principal component analysis was carried out, and downstream analyses were based on the top 20 principal components. To address batch effects, datasets were integrated, re-normalized, scaled, and batch-corrected using Harmony v.0.1.1⁶³. Unsupervised clustering was done using the FindClusters function from Seurat with a resolution of 0.8. UMAP was applied to reduce the data to two dimensions. FindMarkers with default settings to get gene markers, and cell types within each cluster were annotated based on known marker genes and genes that were differentially expressed within each cluster.

Statistical analysis

Data are expressed as mean ± s.e.m. Statistical analyses were completed with Prism GraphPad 8.4. The sample sizes were based on our previous studies^44,45. Each data point corresponds to an individual animal. All data were included in the analyses (no outliers removed). Data were analysed using unpaired t-test (two-sided) or Mann–Whitney test (two-sided) for comparison between two groups, one-way ANOVA followed by Tukey post hoc test, and two-way ANOVA followed by Sidak’s multiple comparisons test for two groups and Tukey’s multiple comparisons test for more than two groups. A significance level of P < 0.05 was considered statistically significant. Additional statistical details are included in Supplementary Table 4.

Reporting summary

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

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at 10.1038/s41586-025-09896-x.

Author contributions

J.X. and R.-R.J. developed the project. J.X., Y.L., S.B., A.M., S.C., V.Z. and W.H. conducted experiments and data analyses. C.N. analysed snRNA-seq data of human DRG. M.L and T.L. contributed to the experiment of AAV design and packaging. Z.Y. and D.V. provided expertise in snRNA-seq. M.P.R.S. and C.E. contributed to mitochondrial isolation and immunocapture assay experiments in MitoTag mice. R.E.C. provided Myo10^tm1d mice and expertise on MYO10 experiments. M.L.B. contributed to the nerve blockade experiment with bupivacaine-loaded films. J.X. and R.-R.J. wrote the manuscript. All authors reviewed the manuscript and provided input.

Peer review

Peer review information

Nature thanks the anonymous reviewer(s) for their contribution to the peer review of this work. Peer review reports are available.

Data availability

All data reported in this paper will be shared by the corresponding author upon request. Any additional information required to reanalyse the data reported in this paper is available from the corresponding author upon request. Source data are provided with this paper. Human DRG single-nucleus RNA sequencing (snRNA-seq) data have been deposited in Gene Expression Omnibus (GEO) under accession number GEO: GSE 310000.

Competing interests

The authors declare no competing interests

Extended data

is available for this paper at 10.1038/s41586-025-09896-x.

Supplementary information

The online version contains supplementary material available at 10.1038/s41586-025-09896-x.