TGF-β-induced fibrotic scar formation limits recovery of spinal cord injury

Abstract

Spinal cord injury (SCI) often causes long-term disability. But effective means to promote proper regeneration after SCI has so far failed to reach the clinic. Here, we report that fibrotic scar formation at injury sites prevents recovery after SCI and that the inhibition of fibrotic scar formation significantly improved SCI recovery in adult mice. We found that after SCI there is an elevation of macrophages, which are a primary source of activated transforming growth factor-β 1 (TGF-β1) that in turn recruits mesenchymal stromal/stem cells (MSCs) to induce their fibroblast differentiation, thus promoting scar formation. We also found that activated TGF-β1 acts on resident pericytes in the endothelial niche of the blood-spinal cord barrier to promote their differentiation into fibroblasts, which also contributes to scarring. Interrupting these pathways by selective genetic KOs or treatment with a TGF-β–neutralizing antibody inhibited scar formation and improved SCI functional recovery. Notably, we found that neonatal mice recover scarlessly after SCI and with no active TGF-β at the injury site. Together, these findings suggest that fibrotic scarring occurs due to elevated activation of TGF-β, and preventing such activation or neutralizing active TGF-β may be an approach to improve outcome after SCI.

Introduction

Spinal cord injuries (SCIs) can cause permanent paralysis at and distal to the injury site, with sensorimotor dysfunction due to poor recovery from axonal and neuronal damage. The World Health Organization estimates that every year 250 000 to 500 000 people globally will experience some form of SCI.^1,2 Most patients affected by SCI are relatively young, and of working age, and thus experience devastating physical, psychosocial, and financial effects of long-term disability.³ SCI typically leads to formation of scar tissue with glial and fibrotic components that limit the restoration of damaged spinal circuitry affecting sensory, motor, and autonomous functions below the injury site.^4–7 Whereas scar-forming astrocytes have been studied extensively, much less attention has been given to the fibrotic, non-glial component of the scar.^8,9 The fibrotic component of the scar consists of collagen, fibronectin, and fibroblasts that each present either a physical or molecular obstacle to the regeneration of severed axons.¹⁰ Scar formation is directly related to the degree of inflammation, as well as the number of macrophages at the site of injury.¹¹ However, the role of macrophages in fibrotic scar formation after SCI is poorly understood.

After SCI, macrophages are recruited to the lesion center. Once properly polarized they can act as a source of anti-inflammatory factors that promote tissue repair, but if improperly polarized they can also be a source of inflammatory cytokines that fuel sary injury.^12–14 Hematogenous macrophages can recruit fibroblasts to the injury site to form fibrotic scar.⁶ Interestingly, perivascular fibroblasts have been shown to be the primary source of collagen and fibronectin in kidney fibrosis.¹⁵ Emerging evidence has revealed that fibrotic scar formation involves a subset of perivascular cells that normally promote regeneration of tissue integrity, but if the injury persists or is too great this homeostatic process can go awry, leading to scar formation. This pathology involves the differentiation of this subset of perivascular cell into fibroblast-like cells that secrete extracellular matrix, promoting scarring.^16,17 The specific cytokine(s) involved in the recruitment of these pericytes is currently unknown.

Transforming growth factor-βs (TGF-βs), a family of 3 multifunctional peptide growth factors, are only present in mammals and are involved in tissue remodeling, repair, and homeostasis.^18–20 Importantly, many diseases are associated with aberrant activation of TGF-β, including various forms of organ fibrosis,^21–23 metastasis of tumors,^24,25 heterotopic ossification,²⁶ and osteoarthritis.^27,28 And notably, high levels of TGF-β1 have also been observed during the progression of SCI.²⁹

During tissue fibrosis, elevated active TGF-β induces fibroblast differentiation of mesenchymal stromal/stem cells (MSCs).^30–32 Thus, it is also possible that elevated active TGF-β at the SCI site recruits MSCs to induce fibroblast differentiation to promote fibrotic scar formation. Therefore, in this study, we investigated the role of active TGF-β in fibrotic scar formation in SCI. We found that active TGF-β was significantly elevated at the injury site after SCI, that it was primarily derived from recruited macrophages at the injury site, and that it induced fibroblast differentiation of both MSCs and perivascular cells. Inhibition of overactive TGF-β signaling by genetic or pharmacological means significantly reduced fibrotic scar formation and improved the recovery from SCI in adult mice. Importantly, as SCI in neonatal mice leads to scar-free healing,³³ we further found that neonatal mice did not show evidence of active TGF-β at the injury site, which likely explain how they can recover so effectively from SCI.

Results

Conditional knockout of TGF-β1 in the macrophage lineage reduces fibrotic scar formation in a mouse SCI model

To investigate the role of macrophages in SCI scar formation, we crossed LysM-cre mice with iDTR^flox/flox mice to generate iDTR_LysM-cre^−/−, in which macrophage/monocyte lineage cells are selectively eliminated with injection of diphtheria toxin (DT) daily for 3 days. The spinal cord was crushed with forceps at the T10 position of the spine to generate the mouse SCI model at day 0, and tissues were harvested for analysis at day 3, 5, 7, 14, 28 and 56 (Extended Data Fig. 1A, B and Extended Data Fig. 2A, B).³⁴ We prepared sections of SCI tissues and immunostained them for collagen III, a major component of fibrotic scar. We found that the collagen III–stained fibrotic scar size was significantly greater over time in SCI iDTR^flox/flox mice, peaking at 2 weeks after SCI, compared to sham controls at baseline, with the degree of collagen staining eventually diminishing for the remainder of the observation period. The SCI iDTR_LysM-cre^−/− mice showed a similar kinetics pattern of expression to the SCI iDTR^flox/flox mice, but at each time point the overall amount of collagen III staining was lower in the SCI iDTR_LysM-cre^−/− mice (Fig. 1a, b). Moreover, co-immunostaining of protein gene product 9.5 (PGP9.5) and fibronectin, a fibrotic scar marker (Extended Data Fig. 2C), further demonstrated that fibrotic scarring was significantly lower in SCI iDTR_LysM-cre^−/− mice compared to SCI iDTR^flox/flox mice (Extended Data Fig. 3D). The density of PGP9.5⁺ nerve fibers in the injured area was significantly greater in SCI iDTR_LysM-cre^−/− mice relative to their SCI iDTR^flox/flox littermates (Extended Data Fig. 3E). Importantly, the concentrations of active and total TGF-β1 in serum and in the spinal cord were significantly lower in SCI iDTR_LysM-cre^−/− mice relative to SCI iDTR^flox/flox mice (Extended Data Fig. 3F-I). Western blotting confirmed that the phosphorylation of Smad2 (pSmad2), a TGF-β downstream signaling transducer, was lower in SCI iDTR_LysM-cre^−/− mice compared to SCI iDTR^flox/flox mice (Extended Data Fig. 3J). Taken together, these results show that after SCI macrophages promote fibrotic scar formation, which is associated with activation of TGF-β signaling.

Fig. 1.

Conditional knockout of transforming growth factor-β 1 (Tgfβ1) in macrophage lineage cells reduces fibrotic scar formation in adult spinal cord injury (SCI) mice. a Representative images of immunofluorescent analysis of collagen III⁺ fibrotic scar at 1, 2, 4, and 8 weeks in iDTR_LysM-cre^−/− knockout mice after SCI, iDTR^flox/flox control mice after SCI, and iDTR^flox/flox control mice without surgery (sham). Scale bar, 200 µm. b Quantitative analysis of collagen III⁺ area at different time points (*P < 0.05, **P < 0.01, ***P < 0.001, n = 6). c Schematic diagram of TGF-β1 knockout in LysM-cre lineage cells. d Representative images of immunofluorescent analysis of protein gene product 9.5⁺ (PGP9.5⁺) (red) nerve fibers, fibronectin⁺ (green) fibrotic scar, and 4′,6-diamidino-2-phenylindole (DAPI) (blue) staining of nuclei at 4 weeks after SCI. Scale bar, 200 µm. e, f Quantitative analysis of the intensity value of PGP9.5 and fibronectin (*P < 0.05, **P < 0.01, n = 6). g Representative images of immunofluorescent analysis of nerve axon–specific β-III-tubulin⁺ (green), collagen III⁺ (red) fibrotic scar, and DAPI (blue) staining of nuclei at 4 weeks after SCI. Scale bar, 200 µm. h Quantitative analysis of the intensity value of collagen III (*P < 0.05, n = 6). i Representative images of immunofluorescent analysis of neurotransmitter marker 5-HT⁺ (red), β-III-tubulin⁺ (green), and DAPI (blue) staining of nuclei in spinal cord lesion site of T10 in Tgfb1^flox/flox sham group mice, Tgfb1^flox/flox control mice, and Tgfb1_LysM-cre^−/− mice at 4 weeks after SCI. Scale bar, 200 µm. Right images are high resolution versions of the boxed regions in the left images, scale bar, 50 µm. j, k Quantitative analysis of the intensity value of β-III-tubulin and 5-HT (*P < 0.05, **P < 0.01, n = 6)

To determine whether activation of TGF-β by macrophages functionally contributes to fibrotic scar formation, we crossed LysM-cre mice with Tgfb1^flox/flox mice to generate Tgfb1_LysM-cre^−/− mice, in which TGF-β1 expression is selectively depleted in macrophage/monocyte lineage cells (Fig. 1c). We co-immunostained for PGP9.5 and fibronectin and found that there was significantly less fibrotic scar formation and a greater density of nerve fibers in the lesion area of SCI Tgfb1_LysM-cre^−/− mice relative to SCI Tgfb1^flox/flox mice (Fig. 1d-f). We confirmed these results by co-immunostaining β-III-tubulin with collagen III or with serotonin (5-HT) (Fig. 1g-k), as 5-HT availability in the spinal cord and the innervation it mediates are critical for the restoration of hindlimb motor function after SCI,^35–37 while β-III-tubulin is a neuron-specific cytoskeleton protein.^38,39 Importantly, the concentrations of active and total TGF-β1 in the serum and the spinal cord were significantly lower in SCI Tgfb1_LysM-cre^−/− mice compared to SCI Tgfb1^flox/flox mice (Fig. 2a-d). We confirmed that the number of pSmad2⁺ cells was significantly lower in Tgfb1_LysM-cre^−/− mice compared to SCI Tgfb1^flox/flox mice in western blot analysis (Fig. 2e). Moreover, by co-immunostaining of pSmad2 with platelet-derived growth factor receptor-β (PDGFR-β), a marker of pericytes, which are a subpopulation of MSCs, we found that pSmad2⁺ cells were primarily PDGFR-β⁺ pericytes in SCI Tgfb1^flox/flox mice, and that the number of pSmad2⁺PDGFR-β⁺ cells was significantly lower in Tgfb1_LysM-cre^−/− mice compared to SCI Tgfb1^flox/flox mice (Fig. 2f, g-i). These results suggest that TGF-β may diminishes pericyte accumulation by altering migration, proliferation, or differentiation of PDGFR-β⁺ pericytes into fibroblasts.

Fig. 2.

Conditional knockout of Tgfβ1 in macrophage lineage cells reduces pSmad2 levels and promotes functional recovery in SCI mice. a–d ELISA analysis showing the concentration of TGF-β in the spinal cord and the serum after SCI in each mouse for the groups indicated (*P < 0.05, **P < 0.01, n = 4). e Representative Western blots showing the activation of phosphorylated Smad (pSmad) signaling. f Representative images of immunofluorescent staining of pSmad2⁺ (red), platelet-derived growth factor receptor-β⁺ (PDGFR-β ⁺) (green) pericytes^, and DAPI (blue) staining of nuclei at 7 days after SCI. Scale bar, 10 µm. g Representative immunofluorescence images of PDGFR-β ⁺ (green) pericytes and DAPI (blue) staining of nuclei at 2 weeks after SCI. Scale bars, 200 µm. h Quantitative analysis of the number of pSmad2⁺ cells (****P < 0.000 1, n = 6). i Quantitative analysis of the number of PDGFR-β⁺ cells (** P < 0.01, n = 6). j, k Illustrations of the hot plate test and the von Frey test. l Quantitative analysis of hindpaw withdrawal time responding to temperature (hot plate test, **P < 0.01, n = 8). m, n Quantitative analysis of hindpaw withdrawal frequency responding to mechanical stimulation (von Frey test, 0.7 mN and 3.9 mN, *P < 0.05, **P < 0.01, n = 8). o Quantitative analysis of Basso Mouse Scale (BMS) score between Tgfb1^flox/flox control mice after SCI, Tgfb1_LysM-cre^−/− mice after SCI, and Tgfb1^flox/flox control mice without SCI (*P < 0.05, n = 8). PWF paw withdrawal frequency. LF left forepaw, RF right forepaw, RH right hindpaw. Statistical significance was determined by multifactorial ANOVA, and all data are shown as means ± standard deviations. Source data are provided as a Source Data file

Conditional knockout of Tgfb1 in macrophage lineage cells promotes functional recovery in SCI mice

We next investigated whether the reduced fibrotic scar formation in SCI iDTR_LysM-cre^−/− mice improved functional recovery after injury. As an initial study along these lines, we first performed co-immunostaining of 5-HT with β-III-tubulin and found that the extent of β-III-tubulin–stained neuron fibers was significantly greater in the lesion area of SCI iDTR_LysM-cre^−/− mice relative to SCI iDTR^flox/flox mice (Extended Data Fig. 3A, C). The concentrations of 5-HT were higher at the upper site of the crush lesion and more 5-HT positive neurotransmitters passed the lesion center in SCI iDTR_LysM-cre^−/− mice compared to SCI iDTR^flox/flox mice (Extended Data Fig. 3A, B). These results suggest that a reduced scar formation permits more neurites and neurotransmitters to cross the site of injury.

Next, we tested the actual degree of functional recovery in SCI iDTR_LysM-cre^−/− mice compared to SCI iDTR^flox/flox mice by measuring responses to hot plate tests (Extended Data Fig. 3D, F) and von Frey tests (Extended Data Fig. 3E, G, H), as well as determining Basso Mouse Scale (BMS) scores (Extended Data Fig. 3I). We found that by all these parameters the SCI iDTR_LysM-cre^−/− mice showed greater function after SCI than the SCI iDTR^flox/flox mice. Collectively, these findings show that elevated TGF-β signaling in pericytes induces fibrotic scar formation that impedes sensorimotor recovery after SCI.

Next, we evaluated the sensory and motor activity of the SCI mice after conditional knockout of Tgfb1 in macrophage lineage cells by a hot plate test (Fig. 2j). Sensory impairments in hindpaw withdrawal time on the hot plate test improved continuously from week 1 to week 8 in SCI Tgfb1_LysM-cre^−/− mice relative to SCI Tgfb1^flox/flox mice (Fig. 2l). Similar recovery results were observed in the manual von Frey test, BMS score and gait analysis of hindlimb function⁴⁰ (Fig. 2k, m, n). Specifically, hindpaw withdrawal frequency on the manual von Frey test increased significantly from 2 weeks after SCI (Fig. 2k, m, n) in SCI Tgfb1_LysM-cre^−/− mice, and motor impairment improved significantly from 28 days after SCI as assessed by BMS score (Fig. 2o). Taken together, these results suggest that fibrotic scar formation promoted by macrophage-mediated TGF-β activation inhibits the degree of functional recovery after SCI.

Conditional knockout of TGF-β type 2 receptor (Tgfbr2) in pericytes reduces fibrotic scar formation in SCI mice

To examine whether TGF-β-induced differentiation of pericytes contributes to fibrotic scarring after SCI, we crossed pericyte-specific Glast-creERT2 mice with Tgfbr2^flox/flox mice to knock out Tgfbr2 specifically in pericytes (Fig. 3A). We chose to focus on Tgfbr2 signaling as it is the only type 2 receptor in the TGF-β signaling pathway, and it has been shown to be involved in fibrosis in different organs.^41,42 By co-immunostaining for PGP9.5 and fibronectin, we found a greater extent neurons and nerve fibers, and less fibrotic scar formation in SCI Tgfbr2 _Glast-creER^−/− mice relative to SCI Tgfbr2^flox/flox mice (Fig. 3b-d). Similarly, by co-immunostaining for collagen III and β-III-tubulin, we confirmed that there was less fibrotic scar formation in SCI Tgfbr2 _Glast-creER^−/− mice compared to the SCI floxed controls (Fig. 3e, f). Moreover, by co-immunostaining for 5-HT and β-III-tubulin we found higher levels of 5-HT and more abundant nerve fibers across the site of SCI in SCI Tgfbr2 _Glast-creER^−/− mice compared with the SCI floxed controls (Fig. 3g-i), similar to the results with SCI Tgfb1_LysM-cre^−/− mice. In contrast, with less fibrotic scarring, glial scarring with abundant glial fibrillary acidic protein-positive (GFAP⁺) astrocytes were significantly greater in SCI Tgfbr2 _Glast-creER^−/− mice compared to SCI floxed controls (Extended Data Fig. 4A, B). Importantly, the number of pSmad2⁺PDGFR-β⁺ pericytes were significantly lower in SCI Tgfbr2 _Glast-creER^−/− mice compared to SCI floxed controls (Fig. 3j-l). By western blotting we found that pSmad2 levels were significantly greater in SCI Tgfbr2^flox/flox mice relative to sham-operated mice but also greater than in SCI Tgfbr2 _Glast-creER^−/− mice (Fig. 3m), indicating a large proportion of TGF-β signaling in pericytes.

Fig. 3.

Conditional knockout of TGF-β type 2 receptor (Tgfbr2) in pericytes reduces fibrotic scar formation in SCI mice. a Schematic diagram of Tgfbr2 knockout in Glast-Cre⁺ pericytes. b Representative images of immunofluorescent analysis of PGP9.5⁺ (red) nerve fibers, fibronectin⁺ (green) fibrotic scar, and DAPI (blue) staining of nuclei at 4 weeks after SCI. Scale bar, 200 µm. c, d Quantitative analysis of the intensity value of PGP9.5 and fibronectin (*P < 0.05, n = 6). e Representative images of immunofluorescent analysis of nerve axon–specific β-III-tubulin⁺ (green), collagen III⁺ (red) fibrotic scar, and DAPI (blue) staining of nuclei at 4 weeks after SCI. Scale bar, 200 µm. Right images are high resolution versions of the boxed regions in the left images. Scale bar, 50 µm. f Quantitative analysis of the intensity value of collagen III (*P < 0.05, n = 6). g Representative images of immunofluorescent analysis of neurotransmitter marker 5-HT⁺ (red), nerve axon–specific β-III-tubulin⁺ (green), and DAPI (blue) staining of nuclei in spinal cord lesion site of T10 in Tgfbr2^flox/flox sham group mice, Tgfbr2^flox/flox control mice, and Tgfbr2 _Glast-creER^−/− mice at 4 weeks after SCI. Scale bar, 200 µm. Right images are high-resolution versions of the boxed regions in the left images. Scale bar, 50 µm. h, i Quantitative analysis of the intensity value of 5-HT and β-III-tubulin (*P < 0.05, n = 6). j Representative images of immunofluorescent analysis of pSmad2⁺ (red), PDGFR-β ⁺ (green) pericytes and DAPI (blue) staining of nuclei at 7 days after SCI. Scale bar, 10 µm. k, l Quantitative analysis of the intensity mean value of PDGFR-β and the number of pSmad2⁺ cells (***P < 0.001, ****P < 0.000 1, n = 6). m Representative Western blots showing the activation of pSmad signaling. Statistical significance was determined by multifactorial ANOVA, and all data are shown as means ± standard deviations. Source data are provided as a Source Data file

To measure mouse behavior, we used the manual von Frey test, hot plate test, and BMS score, which showed significant SCI recovery in Tgfbr2 _Glast-creER^−/− mice relative to Tgfbr2^flox/flox control mice (Extended Data Fig. 5). Taken together, these results suggest that TGF-β signaling in pericytes is critical for fibrotic scar formation, which inhibits functional recovery in SCI mice.

Systemic injection of a TGF-β–neutralizing antibody attenuates fibrotic scar formation and improves functional recovery in SCI mice

We next examined whether inhibition of TGF-β activity attenuates fibrotic scar formation and improves neurological functional recovery. We injected a TGF-β–neutralizing antibody (1D11) or a control antibody of an identical immunoglobulin G (IgG) complex lacking any TGF-β-binding capabilities (13C4) intravenously 3 times a week over a 56 day observation period starting immediately after induction of SCI in wild-type, healthy mice (Fig. 4a). By co-immunostaining of PGP9.5 and fibronectin, as well as collagen III and β-III-tubulin, we found less fibrotic scar formation and a narrower nerve gap in the mice treated with 1D11 compared to those treated with the control antibody 13C4 (Fig. 4b-f), while glial scarring with abundant GFAP⁺ astrocytes was significantly greater in the 1D11-treatmed mice compared to the 13C4-treated controls (Extended Data Fig. 4C, D). Moreover, by co-staining for 5-HT and β-III-tubulin, we found that treatment with 1D11 was associated with significantly greater levels of 5-HT across the injury site and a greater extent of nerve fibers at the injury site in the 1D11-treated mice compared to the control-antibody-treated mice (Fig. 4g-i). As expected, the concentrations of active and total TGF-β1 in the serum and the spinal cord was significantly lower 1D11-treatment group compared to the control-treated group (Fig. 4j-m). Importantly, the number of pSmad2⁺ PDGFR-β⁺ pericytes was also lower in the 1D11-treated mice compared to the 13C4-treated mice (Extended Data Fig. 6A-C), which was confirmed by western blot analysis of pSmad2 expression (Extended Data Fig. 6D). As above, functional recovery was measured using different behavior tests, including manual von Frey tests, hot plate tests, and BMS scores of the hindlimb. We found that 1D11-treated mice showed significantly greater functional recovery compared to 13C4-treated control mice (Extended Data Fig. 6E-L).

Fig. 4.

Systemic injection of a TGF-β–neutralizing antibody promotes neurological recovery after SCI. a Schematic diagram illustrating the timeline and concentration of TGF-β–neutralizing antibody 1D11 injection used. b Representative images of immunofluorescent analysis of PGP9.5⁺ (red) nerve fibers, fibronectin⁺ (green) fibrotic scar, and DAPI (blue) staining of nuclei at 4 weeks after SCI. Scale bars, 500 µm. Right images are high-resolution versions of the boxed regions in the left images. Scale bar, 50 µm. c Representative images of immunofluorescent analysis of nerve axon–specific β-III-tubulin⁺ (green), collagen III⁺ (red) fibrotic scar, and DAPI (blue) staining of nuclei at 4 weeks after SCI. Scale bar, 200 µm. Right images are high-resolution versions of the boxed regions in the left images. Scale bar, 50 µm. d–f Quantitative analysis of the intensity value of fibronectin, PGP9.5 and Collagen III (*P < 0.05, **P < 0.01, ***P < 0.001, n = 6). g Representative images of immunofluorescent analysis of neurotransmitter marker 5-HT⁺ (red), nerve axon–specific β-III^-tubulin⁺ (green), and DAPI (blue) staining of nuclei in the spinal cord lesion site of T10 in sham group mice, 13C4 group control mice, and 1D11 group mice at 4 weeks after SCI. Scale bar, 200 µm. Right images are high-resolution versions of the boxed regions in the left images. Scale bar, 50 µm. h, i Quantitative analysis of the intensity value of 5-HT and β-III-tubulin (*P < 0.05, n = 6). j–m ELISA analysis showing the concentration of TGF-β in spinal cord and serum after SCI between 13C4 group control mice, 1D11 group mice, and 13C4 group control mice without surgery (sham) (*P < 0.05, **P < 0.01, n = 4). Statistical significance was determined by multifactorial ANOVA, and all data are shown as means ± standard deviations. Source data are provided as a Source Data file

MSC-to-Fibroblast differentiation results in secretion of axonal growth-inhibitory factors

Given the above results that TGF-β may act on more than just pericytes, we next investigated which other cell types may be recruited by TGF-β for fibrotic scar formation. As TGF-β induces pericytes in the stromal cell lineage to undergo fibroblast differentiation, we first performed a pericyte lineage-tracing experiment using Solute Carrier Family 1 Member 3 (Glast)-CreER^T2::ROSA26-tdTomato mice by crossing specific inducible Glast-CreER^T2 transgenic mice with ROSA26-tdTomato reporter mice. Injection of tamoxifen in Glast-CreER^T2::ROSA26-tdTomato mice induced genetically labeling of a subpopulation of pericytes and their progeny that potentially contribute to fibrotic scarring after SCI (Fig. 5a, b). Immunostaining of PDGFR-β showed that tdTomato⁺ cells comprised more than 80% of the PDGFR-β⁺ cells after SCI, indicating successfully labeling of pericytes with tdTomato (Fig. 5c, d). Importantly, injection of 1D11resulted in significantly less tdTomato-labeled pericytes in the fibrotic scar relative to injection of 13C4 control antibody, suggesting that pericytes are recruited by TGF-β for fibrotic scar formation (Fig. 5E, F). Moreover, immunostaining of leptin receptor (LepR), a marker for MSCs, showed that 12% of tdTomato⁺ pericytes were LepR⁺ after SCI (Fig. 5g, h). Expression of type III collagen, which was largely co-localized with LepR at the injury area, was significantly lower in the 1D11-treatment group relative to the 13C4-treatment group (Fig. 5i, j), indicating fibrotic differentiation of pericytes. Interestingly, immunostaining of fibroblast-specific protein 1 (FSP1) demonstrated that nearly 40% of FSP1⁺ fibroblasts are LepR⁺ MSCs, indicating that tdTomato-labeled pericytes represent a subpopulation of MSC lineage cells that have undergone fibroblast differentiation (Fig. 5k, l).

Fig. 5.

LepR⁺ MSCs are the primary cells for fibrotic scar formation. a, b Diagram illustrating the genetic strategy to trace type A pericytes and the timeline of tamoxifen injection. c Representative images of immunofluorescent analysis of tdT⁺ (red) type A pericytes, PDGFR-β ⁺ (green) pericytes, and DAPI (blue) staining of nuclei in the spinal cord lesion site of T10 wild-type (WT) mice at 2 weeks after SCI. Scale bar, 50 µm. Left images are high-resolution versions of the boxed regions in the right images. Scale bar, 10 µm. d Quantitative analysis of percentage of PDGFR-β cells/tdTomato⁺ type A pericytes cells (***P < 0.001, n = 4). e Representative images of immunofluorescent analysis of collagen III⁺ (Col-III green) fibrotic scar, tdT ⁺ (red) type A pericytes, and DAPI (blue) staining of nuclei in spinal cord lesion site of T10 in 13C4 group control mice and 1D11 group mice at 4 weeks after SCI. Scale bar, 50 µm. f Quantitative analysis of percentage of collagen III⁺ area /tdTomato⁺ type A pericytes field (***P < 0.001, n = 4). g Representative images of immunofluorescent analysis of LepR⁺ (green) MSCs, tdT ⁺(red) type A pericytes, and DAPI (blue) staining of nuclei in sham group mice, 13C4 group control mice, and 1D11 group mice at 4 weeks after SCI. Scale bar, 50 µm. h Quantitative analysis of the percentage of LepR⁺ cells/tdTomato⁺ cells (*P < 0.05, n = 4). i Representative images of immunofluorescent analysis of LepR⁺ (green) MSCs, collagen III⁺ (Col-III, green) fibrotic scar^, and DAPI (blue) staining of nuclei. Scale bar, 50 µm. j Quantitative analysis of the percentage of collagen III⁺ area/ LepR⁺ MSCs field (**P < 0.01, n = 4). k Representative images of immunofluorescent analysis of LepR⁺ (green) MSCs, FSP1 ⁺ (red) type A pericytes^, and DAPI (blue) staining of nuclei. Scale bar, 50 µm. l Quantitative analysis of the percentage of LepR⁺ MSCs/FSP1 ⁺ cells (**P < 0.01, n = 4). Statistical significance was determined by multifactorial ANOVA, and all data are shown as means ± standard deviations. Source data are provided as a Source Data file

Single-cell RNA-sequencing shows elevated active TGF-β–induced fibroblast differentiation of MSCs in SCI mice

We then used single-cell RNA-sequencing (scRNA-seq) technology to investigate the type of cells present at the fibrotic scar in SCI. 1D11 or 13C4 was injected intravenously in SCI mice immediately after injury with sham-operated mice as control. Tissues from the SCI mice were harvested at day 3 post injection and single cells were prepared for sequencing. After quality control, we obtained high-quality RNA-seq results from these three SCI single-cell samples. Using a combination of automated cell typing and differential expression, we identified endothelial cells, fibroblasts, MSCs, pericytes, microglia and neurons in all samples (Fig. 6a, b and Extended Data Fig. 7A-C).

Fig. 6.

Single-cell RNA-seq reveals elevated active TGF-β-induced fibroblast differentiation of MSCs after SCI. a Clustered and annotated data from all sample groups (1D11, 13C4, and Sham) plotted on UMAP coordinates and identifying the indicated cell types. b UMAP plot of cell type annotations split by sample group. c Heat map of enriched expression of TGF-β pathway genes in pericytes from the 1D11 versus 13C4 sample groups. d–g Violin plots of TGF-β ligand and receptor expression across cell types, split by sample group. h Dot plot of the mean expression and fraction of cells in each sample group expressing three key axon growth regulators: two growth inhibitors (Sema3a and Ephb2) and one growth factor (Cd248). i Dot plot of significant (P < 0.05) ligand-receptor interactions between cell types of interest, as determined by CellPhoneDB

Enrichment analysis showed that the expression of downstream signaling genes was significantly greater in SCI mice treated with 13C4 relative to the sham group, and almost all elevated expression of TGF-β downstream genes in the 13C4 group was significantly inhibited in the mice injected with 1D11 (Fig. 6c). Moreover, expression of the TGF-β ligands and receptors, Tgfb1, Tgfb2, Tgfbr1 and Tgfbr2, was elevated in pericytes, MSCs, endothelial cells, fibroblasts and microglia of the 13C4 injection group relative to the sham group, and significantly inhibited in the 1D11-treated group (Fig. 6d-g), indicating that these cell types are specifically responsive to inhibition of TGF-β signaling as a potential target for fibrosis. These results collectively support the conclusion that TGF-β pathway activity is significantly enhanced in the 13C4 treatment group and identify putative cell types with key roles in TGF-β-induced fibrotic scar formation following SCI.

Next, we interrogated the expression of known axon growth inhibitors and growth factors to test the hypothesis that inhibition of TGF-β signaling could encourage axon growth by modulating the secretion of these mediators. The mean expression of several known axon growth inhibitors such as Sema3a and Ephb2 was found to be lower in the 1D11 group as compared to the 13C4 group, while mean expression of axon growth factor CD248 was found to be higher in the 1D11 group as compared to the 13C4 and Sham groups, indicating that suppression of the TGF-β signaling may contribute to axonal regeneration (Fig. 6h, i).

To determine the relationship between TGF-β-responsive cell types of interest, we performed trajectory analysis to reconstruct putative lineages connecting developmentally-related cell types. We also performed a ligand-receptor analysis using CellPhoneDB to identify significant ligand-receptor interactions between cell types of interest. Because fibrotic scar occurs in 7 days post SCI, as shown above, and as single cell preparation was impossible once a scar is formed, samples for scRNA-seq analyses were harvested 3 days post SCI when fibroblast differentiation is still in the intermediate stage. Thus, we performed lineage trajectory analysis on fibroblasts as well as MSCs, pericytes, and endothelial cells by computing pseudotime and found that MSCs give rise to fibroblasts at day 3 post SCI (Extended Data Fig. 8A-E). Trajectory analysis returned two putative lineages: one representing a possible endothelial-to-mesenchymal transition of endothelial cells to fibroblasts through an MSC intermediate, and the other representing the differentiation of pericytes to fibroblasts through an MSC intermediate (Extended Data Fig. 8B-E). Analysis of dynamically expressed genes along pseudotime of the endothelial cell lineage revealed gene expression changes indicative of endothelial-to-mesenchymal transition, such as loss of tyrosine kinase with immunoglobulin like and EGF like domains 1 (Tie1)⁴³ and von Willebrand factor (vWF)⁴⁴ expression in endothelial cells and expression of collagen synthesis genes in fibroblasts (Extended Data Fig. 8A). CellPhoneDB analysis further identified significant interactions between relevant ligand-receptor pairs with known roles in fibrosis, such as collagen-integrin binding and NOTCH-JAG⁴⁵ interactions amongst fibroblasts, MSCs, pericytes and endothelial cells (Fig. 6i). Taken together, our scRNA-seq analysis confirms that enriched TGF-β downstream signaling in the control antibody-treated sample, reveals enhanced cell-cell signaling in MSCs, fibroblasts, and pericytes, and it may indicate a possible role for endothelial-to-mesenchymal transition in fibrotic scar formation.

Neonatal mice show complete recovery from SCI with no detectable active TGF-β levels and without fibrotic scar formation

All 3 members of the TGF-β family, comprising TGF-β1, 2, and 3, are expressed in mammals, but they are not expressed in aquatic vertebrates nor amphibians, suggesting they arose later in evolution. Importantly, non-mammalian vertebrates do not face the same obstacles as mammals when it comes to recovery from SCI.^46–48 Interestingly, neonatal mice do not exhibit fibrotic scar formation within 7 days after SCI. We therefore tested whether fibrotic scar formation occurs at all in neonatal mice after SCI. And if it does not, we posited that recovery from SCI in neonatal mice could be expected to occur as it does in amphibians. In brief, the spinal cord was crushed with forceps to generate SCI in neonatal mice on days 2, 7 and 12 after birth. SCI tissues were harvested, and sections were prepared (Fig. 7a). Immunostaining of pSmad2 showed that pSmad2⁺ cells were not detectable in the SCI site in postnatal day 2 and 7 (P2 and P7) mice as they were in sham-operated neonatal mice on day 7 (Fig. 7b, c), whereas the number of pSmad2⁺ cells was significantly increased in postnatal day 12 (P12) mice.

Fig. 7.

Neonatal mice completely recover from SCI without scarring. a Schematic diagram illustrating the timeline of the experimental procedures. b Representative images of immunofluorescent analysis of pSmad2 (red), PDGFR-β (green), and DAPI (blue) in the spinal cord lesion site in neonatal mice on days 2, 7, and 12 with or without SCI. Scale bar, 50 µm. Right images are high-resolution versions of the boxed regions in the left images. Scale bar, 50 µm. c, d Quantitative analysis of the number of pSmad2⁺ cells and PDGFR-β⁺ cells (**P < 0.01, ***P < 0.001, ****P < 0.000 1, n = 6). e Representative images of immunofluorescent analysis of β-III-tubulin (green), collagen III (red), and DAPI (blue) in the spinal cord lesion site in neonatal mice on days 2 and 12, and adult mice with or without SCI. Scale bar, 200 µm. f, g Quantitative analysis of the intensity mean value of collagen III⁺ fibrotic scar and β-III-tubulin⁺ nerves (*P < 0.05, ****P < 0.000 1, n = 6). Statistical significance was determined by multifactorial ANOVA, and all data are shown as means ± standard deviations. Source data are provided as a Source Data file

Furthermore, by co-immunostaining for PDGFR-β, a marker of pericytes, and pSmad2, we found that PDGFR-β was expressed similarly in SCI P2 and P7 mice, and sham P7 mice, but significantly increased in P12 mice after SCI (Fig. 7b, d), suggesting fibrotic scar formation on day 12. Interestingly, immunostaining of collagen III demonstrated no fibrotic scar formation in SCI P2 mice, similar to sham mice, while there was notable fibrotic scar formation on day 12, though with a smaller size relative to adult SCI fibrotic scar (Fig. 7e, f). Consistent with these findings, immunostaining of β-III-tubulin showed nerve growth through the SCI area with no scar in P2 mice, similar to sham mice (Fig. 7e, g), whereas the fibrotic scar blocked nerve growth through the SCI area in P12 mice.

Because the behavior tests can not be conducted in neonatal mice, we examined the recovery of P2 and P12 mice at 7 days and 1 month after SCI by video recording. Importantly, the P2 mice were almost completely recovered at 7 days and 1 month after SCI, similar to the sham littermates (Video 1 and 3). However, the hindlimbs of P12 mice were still paralyzed at 7 days and 1 month after SCI (Video 2 and 4). These results show that overactive TGF-β–induced fibrotic scar of SCI inhibits recovery in mice after the first week of the neonatal period.

Discussion

Scar formation, including glial and fibrotic scars at SCI sites, is believed to be the primary obstacle for axon regeneration and functional recovery.^49–53 Although glial scarring has been well studied on the cellular and structural levels, particularly for the potential benefits the formation of glial scarring has for functional recovery from SCI,^54–58 little is known about fibrotic scarring. Previous studies of fibrotic scarring have been primarily in vitro. Fibrotic scar formation is believed, through fibrosis, to inhibit axonal regeneration after SCI.^59–61 We found that overactive TGF-β at SCI sites recruits and induces fibroblast differentiation of MSCs, as well as resident pericytes, altering their secretome, and thus promoting the formation of fibrotic scar. Importantly, inhibition of TGF-β activity, either genetically or pharmacologically, reduced fibrotic scar formation and significantly improved recovery from SCI in mice. Strikingly, neonatal mice did not show TGF-β activity at the SCI site and showed full recovery after injury. Collectively, our findings show that fibrotic scarring results from abnormal TGF-β activity, and that such scarring is the primary obstacle to functional recovery after SCI.

In evolution, TGF-βs exist only in mammals and are known to promote tissue repair, remodeling, and homeostasis in adulthood.^62,63 Unlike most cytokines, TGF-βs are expressed in an inactive latent form and deposited in the matrix upon secretion. After injury or during tissue remodeling, TGF-βs are activated to recruit stem cells or progenitors to the injury site where these cells undergo differentiation for tissue repairing and restoring homeostasis.^64–67 In addition, TGF-β can be either a tumor suppressor or a promoter depending on the temporal stage of the cancer.^63,68 In the early stage of tumors initiation, TGF-β limits the growth of tumor cells through its antiproliferative and proapoptotic actions.⁶⁹ Conversely, during tumor progression, TGF-β acts as a tumor-promoter by inducing proliferation, angiogenesis, and immunosuppression, and thus promotes invasion and metastasis of cancer.⁷⁰ In spinal cord injury, we believe TGF-β plays dual functions as well. On one hand, TGF-β can recruit immune cells and vascular cells to stabilize the vascular environment and seal the wound, on the other hand, TGF-β could recruit scar-forming cells flooded into the lesion site to form the fibrotic scar in the later stage. Our findings were supported by Wang et al.,⁷¹ who demonstrated that miR-21-5p functions in an amplifying circuit to enhance TGF-β signaling events in the activation of spinal fibroblasts and suggest that miR-21-5p is a potential therapeutic target in the treatment of fibrotic scar formation after SCI.⁷¹ However, fibrosis in various tissues is often associated with overactive TGF-β.^26,72,73 Inhibition of TGF-β activity has been reported in preventing glial scarring of SCI.^74,75 Regarding the absence of TGF-β pathway activation in neonatal mice, we believe that TGF-β serves as a critical upstream factor for fibrotic scar formation and is primarily secreted by activated macrophages. In neonatal SCI models, however, typical fibrotic scars fail to form, suggesting that macrophages are not sufficiently recruited or activated following SCI in neonates. Consequently, the TGF-β signaling pathway remains largely inactive. Similarly, TGF-β becomes overactive to recruits MSCs for fibroblast differentiation and fibrotic scar formation after SCI.^76,77 The nature of the fibrotic scar formation is to repair the tissue injury quickly for survival. However, scar tissues also become a physical and chemical barrier for axon regeneration. This partially explains why the recovery after SCI is particularly difficult in mammals. Apparently, the mechanism of wound repair involving TGF-β has not been developed yet in early neonatal mice, which provides an ideal model to study the role of fibrotic scar formation in SCI recovery. In our study, TGF-β neutralization antibody was injected intraperitoneally with no observed immune dysregulation or impaired wound healing at distant sites. However, prolonged systemic inhibition of TGF-β could potentially lead to adverse effects such as increased susceptibility to infection or delayed wound healing, as reported in previous studies.^78–80 Also, it has been reported that macrophages play an important role in scar contraction, hematoma and myelin fragments clearing. The fibrotic scar area was greater in clodronate treated group (macrophages deletion group) relative to PBS treated group.⁵ These discrepancies may be attributable to two principal factors. First, macrophages exert temporally distinct and functionally divergent roles during SCI progression. In the acute phase, they are essential for hematoma resolution, phagocytosis of cellular and myelin debris, and modulation of the inflammatory milieu—functions that are indispensable for initiating repair processes. In contrast, at later stages, a subset of alternatively activated (M2-like) macrophages promotes fibrotic scar formation through secretion of pro-fibrotic mediators such as TGF-β. Accordingly, the timing and specificity of macrophage depletion critically influence the resultant tissue outcomes, particularly with respect to fibrotic matrix deposition. s, the present study employed a LysM-Cre::iDTR transgenic model with administration of DT to selectively ablate peripheral myeloid-derived macrophages and, to a lesser extent, neutrophils, while largely sparing microglia. Microglia are the earliest innate immune responders following SCI, rapidly mobilizing to the lesion site and forming a containment barrier around the hematoma.^81,82 Within the first 72 h post-injury, microglia play a dominant role in debris clearance relative to infiltrating monocyte-derived macrophages. This selective preservation of microglia may account for the lack of observable differences in hematoma resolution and myelin clearance between DT-treated and control animals in our study, despite effective depletion of peripheral macrophages. As TGF-β1 expression was not specifically quantified in microglia at the injury site, microglia could not completely be excluded for local TGF-β1 production.

Fibroblast differentiation of MSCs initiates fibrotic scar formation after SCI. Here we show that both TGF-β1 and pSmad2 levels were significantly increased after SCI. Conditional knockout of Tgfbr2 in Glast-positive pericyte lineage cells significantly decreased fibronectin expression and improved the recovery of SCI in mice. Given that pericytes are a subpopulation of MSCs, our results suggest that MSCs were recruited by overactive TGF-β1 for fibrotic scar formation. Single-cell sequencing analysis also revealed that TGF-β receptors, ligands, and their downstream signaling were significantly increased after SCI within 3 days. Moreover, TGF-β signaling pathway profiling of pericytes showed a significant increase, and our pseudotime trajectory analysis based on harmonized principal components illustrates a lineage progression from pericytes → MSCs → fibroblasts following SCI. Since there has been no unique marker for MSCs and Glast expression is relatively specific for pericytes, Glast-Cre was used to drive conditional knockout of Tgfbr2 and used in lineage-tracing experiment. Interestingly, discrepant results regarding the role of pericytes in SCI have been reported.^60,83–85 Our results provide an explanation for these discrepancies. Moreover, multiple cell types may contribute to fibrotic scar formation,⁸⁶ especially given the lack of highly specific markers for fibroblasts. In our dataset, we also observed that endothelial cells—shown in Extended Data Fig. 8B–E—may undergo a transition along the endothelial → MSC → fibroblast lineage.

Pericytes have potential stem cell capacity and reside within the vascular basement membrane to stabilize the vessels and tight junctions between endothelial cells and to support the structural integrity of the blood–spinal cord barrier. After SCI, TGF-β1 is overactivated to induce migration of pericytes for fibrotic scar formation, which damages the structural integrity of the blood–spinal cord barrier. Exogenous pericyte transplantation has been shown to enhance the structural integrity and function of the new blood vessels, particularly for the regulation of neurovascular function, as a potential therapy for SCI.^87–89 The transplanted pericytes likely repair blood vessels, compensating for the deficit of pericytes resulting from their depletion due to overactive TGF-β during fibrotic scar formation after SCI. Notably, an increase in GFAP⁺ astrocytes were observed following the deletion of Tgfbr2 in pericytes (Extended Data Fig. 4A, B). Previous studies have reported a “compensatory response,” in which the change of fibrotic scar formation triggers modulated astrocytic scarring.⁹⁰ For instance, in Glast-Rasless transgenic mice—where fibrotic scar formation is mitigated by suppressing fibroblast proliferation—astrocyte proliferation is reduced, and the border between astrocytic and fibrotic scars becomes disrupted.⁸⁴ These findings suggest a reciprocal interaction between astrocytic and fibrotic scars, which together contribute to maintaining the structural stability of each scar type following SCI. SCI could also result in vascular impairment and blood spinal cord barrier deterioration. And Vascular endothelial growth factor (VEGF), a strong pro-angiogenic factor, involves angiogenic pathways in pathological disease.⁹¹ In aortic endothelial cells, TGF-β stimulation of an ALK5/TβRII-Smad2 complex enhances expression of VEGF. In addition, pericytes is one the most important cells to form the vessel, increase blood flow, preserve blood-brain barrier function and regulate immune cell entry to the central nervous system.⁹² Thus, we believe TGF-β can not only induce pericytes differentiation into fibroblasts, but also has effect on angiogenesis by targeting pericytes after spinal cord injury. Moreover, the research on glial scar are polarized. Some studies suggest that glial scar can inhibit axon growth,^17,93 whereas others claimed glia scar aids axon regeneration.^7,94 Our data showed conditional knockout TGF-β1 receptor in pericytes increased number of GFAP⁺ astrocytes along with improved functional recovery, which may support the protective function of gliosis after spinal cord injury. While our study primarily focused on TGF-β1, our transcriptomic data (Fig. 6c) suggest that other signaling pathways may also contribute to fibrotic scar formation. For instance, PDGF-related downstream effectors such as Crk and Pik3cd were upregulated in the control group (13C4) compared to the TGF-β1–neutralized group (1D11). In addition, transcriptional factors Ccar1 and Arid1a for CTGF expression were differentially expressed, indicating potential involvement of CTGF pathway. These findings are in consistency with previous reports that PDGF signaling^95,96 and CTGF^97,98 can act in parallel or downstream of TGF-β to promote fibroblast activation and collagen deposition.

In summary, we took multiple approaches to demonstrate that overactive TGF-β recruit MSCs for fibrotic scar formation after SCI. In LysM-cre::iDTR^flox/flox mice, macrophage lineage cells were eliminated and TGF-β activity was significantly decreased, suggesting macrophages may be responsible for increased TGF-β1 expression and over activation. TGF-β1 activity was also significantly decreased in LysM-cre::Tgfb1^flox/flox mice, which further validated the notion that macrophages are key source of TGF-β1 expression. Furthermore, findings in Glast-creERT2::Tgfbr2 ^flox/flox mice demonstrate that TGF-β signaling in pericytes drives their fibroblast differentiation. Although the effects on scarring and recovery was more potent using the TGF-β neutralizing antibody 1D11, all three conditional knockout mice showed significantly improved functional recovery after SCI. Furthermore, neonatal mice with no active TGF-β showed no fibrotic scar formation and near-complete recovery after SCI, further supporting the notion that overactive TGF-β is responsible for fibrotic scar formation and inhibition of recovery from SCI. Together, our findings suggest that inhibition of abnormal overactivation of TGF-β may present a potential new therapy for SCI.

Materials and methods

Genetic labeling of transgenic mice

Recombination in Glast-Rasless-YFP mice was induced by a daily intraperitoneal injection of 2 mg of tamoxifen (20 mg/mL in 1:9 ethanol:corn oil, Sigma-Aldrich, St. Louis, MO, USA) for 5 consecutive days (Extended Data Fig. 1B). Vehicle mice (matched mice with the same genotype) received the same number of injections of the solvent (1:9 ethanol:corn oil) without tamoxifen. Mice were randomly assigned to the vehicle or tamoxifen group. Glast-Tdtomato mice were recombined with tamoxifen using the above protocol. Injuries were performed after a 7-day clearing period starting after the last tamoxifen injection. Tamoxifen and its active metabolite 4-hydroxytamoxifen have a half-life of 6–12 h in the mouse.^84,99 A previous study analyzing CreERT2 distribution in the adult mouse spinal cord 6 days after the last tamoxifen administration showed that there is no CreERT2 in the nucleus of cells at this time, directly demonstrating that tamoxifen has been cleared at this time point.¹⁰⁰ Therefore, the chosen 7-day washout period ensures that no tamoxifen is left at the time of injury or after, which could affect the response to injury. Moreover, the 7-day washout period guarantees that all recombination occurs before the injury; therefore, if cells other than type A pericytes start expressing the Glast-CreERT2 transgene in response to the injury, recombination will not occur.⁸⁴

Mice

All young adult C57BL/6 J male and female mice were 8 weeks of age at the time of SCI. All mice were housed in a 12 h light/dark cycle in a specific pathogen-free facility with controlled temperature ( ~ 18–23 °C) and humidity (40%–60%) and were allowed free access to food and water. Animal care, including manual bladder voiding, was performed at least twice daily or as needed after SCI for the duration of the experiment. All mice were maintained in the animal facility of The Johns Hopkins University School of Medicine (Baltimore, MD, USA). The experimental protocols were reviewed and approved by the Institutional Animal Care and Use Committee of The Johns Hopkins University.

We purchased 8-week-old LysM-cre (Stock number: 004781), iDTR^flox/flox (Stock number: 007900), Tgfb1^flox/flox (Stock number: 010721) and Glast-creERT2 (Stock number: 012586) mouse strains from the Jackson Laboratory (Bar Harbor, ME, USA). Tgfbr2^flox/flox were obtained from the laboratory of H.L. Moses (Vanderbilt University, Nashville, Tennessee, USA).¹⁰¹ We generated LysM-cre::iDTR^flox/flox mice (iDTR_LysM-cre^−/−) by crossing LysM-cre mice with iDTR^flox/flox mice, LysM-cre::Tgfb1^flox/flox mice (Tgfb1_LysM-cre^−/−) by crossing LysM-cre mice with Tgfb1^flox/flox mice and Glast-creERT2:: Tgfbr2^flox/flox mice (Tgfbr2_Glast-creER^−/−) by crossing Glast-creERT2 mice with Tgfbr2^flox/flox mice. We induced SCI in 8-week-old WT, iDTR_LysM-cre^−/−, iDTR^flox/flox, Tgfb1_LysM-cre^−/−, Tgfb1^flox/flox, Tgfbr2_Glast-creER^−/−, and Tgfbr2^flox/flox female mice and humanely killed them at different time points.

For the antibody treatment experiments, we purchased 8-week-old C57BL/6 J (WT, Stock number: 000664) mice from the Jackson Laboratory, and they were intraperitoneally injected with 13C4 (R&D Systems, Minneapolis, MN) or 1D11 (R&D Systems), 5 mg/kg body weight, 3 times a week for 1, 2, or 4 weeks immediately after SCI (Extended Data Fig. 1B). 1D11 is a monoclonal antibody that neutralizes 3 major active TGF-β isoforms (TGF-β1, -2, and-3), the known ligands for the TGF-β receptor kinase.

We purchased 8-week-old R26R-TdTomato mice (Stock number: 007909) from the Jackson Laboratory (Bar Harbor, ME, USA). We generated Glast-creERT2::R26R- TdTomato mice by crossing Glast-creERT2 mice with R26R- TdTomato mice. We performed SCI operations on 8-week-old Nestin-creERT2::R26R- TdTomato male mice. Three days after surgery, we treated the mice with 80 mg/kg body weight of tamoxifen 3 times a week for 2 or 4 weeks and humanely killed the mice at 2 or 4 weeks after surgery.

Surgical procedures

We anesthetized the mice at 8 weeks of age with ketamine (Vetalar, Ketaset, Ketalar; 100 mg/kg, intraperitoneally) and xylazine (Rompun, Sedazine, Anased; 10 mg/kg, intraperitoneally). We then shaved the hair ranging from 2–3 cm above and below the T10 position of the back (the highest raised point on the back) and disinfected the operating table and surgical instruments in advance, laying towels to prepare the instruments. After a 1-cm skin incision was made at the T10 position on the back, the muscles in the local area were bluntly separated, using gauze to stop bleeding, and then the T10 lamina was exposed. Subsequently, severe crush SCIs were made at the level of T10 after laminectomy of a single vertebra by using Dumont #5 Forceps (Fine Science Tools, Foster City, CA, USA) without spacers and with a tip width of 0.5 mm to completely compress the entire spinal cord laterally from both sides for 15 s.^7,53,102,103 Spastic tail swings, retraction, and fluttering of the lower extremities and body, paralysis of both lower extremities, and dura mater congestion and were observed, indicating successful modeling. Finally, the wound was washed with normal saline. After hemostasis with gelatin sponge, the muscle and skin were sutured layer by layer. Warm blankets were used to rewarm mice postoperatively to prevent hypothermia. Mice were housed in cages, and manual assistance was provided to mice at least twice a day to empty the bladder to avoid urinary system infections. The sham group did not involve spinal cord compression but the other aspects of the surgical procedure were the same as the above.

Neonatal (P2/P7/P12) mice were anesthetized using isoflurane. A laminectomy was performed at the thoracic level (T10) until the spinal cord was exposed completely from side-to-side. The spinal cord was then fully crushed for 15 s. After SCI, the pups were returned to the mother. Feeding was monitored closely in the first week after surgery. Nutra-Gel diets (Bio-Serv, Flemington, NJ, USA) or breeder chow diets were provided to avoid cannibalism.

ELISA (enzyme-linked immunosorbent assay) and Western blot analyses

We determined the concentration of active and total TGF-β1 in the conditioned media using the ELISA Development Kit (R&D Systems, MB100B) according to the manufacturer’s instructions. Western blot analyses were conducted on the protein of lysates from the in vivo spinal cord. The spinal cord lysates were centrifuged, and the supernatants were separated by SDS–PAGE (sodium dodecyl sulfate–polyacrylamide gel electrophoresis) and blotted on polyvinylidene fluoride membrane (Bio-Rad Laboratories, Hercules, CA, USA). After incubation in specific antibodies, we detected proteins using an enhanced chemiluminescence kit (Amersham Biosciences, Little Chalfont, UK). We used antibodies recognizing mouse pSmad2 (1:1 000, Cell Signaling Technology, Inc., Danvers, MA, USA) and Smad2 (1:1 000, Cell Signaling Technology, Inc.) to examine the protein concentrations in the lysates.

Histochemistry and immunohistochemistry

At the time of killing, the spinal cords were collected and fixed in 4% paraformaldehyde overnight and then embedded in paraffin or optimal cutting temperature compound after being dehydrated with 30% sucrose for 48 h (Sakura Finetek, Torrance, CA, USA). Four-μm-thick, coronal-oriented sections of the spinal cord were processed for immunohistochemistry staining using a standard protocol. Thirty-μm-thick, coronal-oriented sections were prepared for nerve-related immunofluorescent staining, and 10 μm-thick, coronal-oriented sections were used for scar-related and other immunofluorescent staining using a standard protocol. The sections were incubated with primary antibodies to rabbit 5-HT (1:50, sc-65495, Santa Cruz Biotechnology, Dallas, TX, USA), mouse β-III-tubulin (1:100, MA1-118, Invitrogen, Carlsbad, CA, USA), rabbit PGP9.5 (1:250, ab108986, Abcam, Cambridge, UK), mouse PGP9.5 (1:50, ab8189, Abcam), rabbit Fibronectin (1:100, ab2413, Abcam), mouse Fibronectin (1:100, ab6328, Abcam), rabbit Collagen III (1:100, ab7778, Abcam), chicken GFAP (1:500, ab4674, Abcam), mouse Collagen1α1 (1:50, sc-293182, Santa Cruz Biotechnology), rabbit Phospho SAMD2 (1:100, 44-244 G, Invitrogen), rabbit PDGFR-β (1:100, ab32570, Abcam), goat PDGFR-β (1:100, AF1042, R&D Systems), rabbit FSP1 (1:300, 07-2274, MilliporeSigma, Burlington, MA, USA), and chicken green fluorescent protein (1:250, ab13970, Abcam) overnight at 4 °C. Then, the corresponding sary antibodies were added onto the sections for 1 h while avoiding light. For immunohistochemistry, a horseradish peroxidase–streptavidin detection system (Dako, Carpinteria, CA, USA) was subsequently used to detect the immunoactivity, followed by counterstaining with hematoxylin (Sigma-Aldrich). For immunofluorescent staining, the sections were counterstained with 4′,6-diamidino-2-phenylindole (DAPI) (H-1200, Vector Laboratories, Burlingame, CA, USA). The sample images were observed and captured by a fluorescence microscope (BX51, DP71, Olympus Scientific Solutions Americas Inc., Waltham, MA, USA) or confocal microscope (LSM 780, Zeiss, Oberkochen, Germany). ImageJ software (National Institutes of Health, Bethesda, MD, USA) was used for quantitative analysis. We calculated nerve and scar area as described previously.¹⁰⁴

Quantification analysis

Five alternate sections per mouse, spanning the lesion center and spaced 40 μm apart, were immunostained for 5-HT, β-III-tubulin, PGP9.5, fibronectin, collagen III, pSmad2, PDGFR-β, GFAP, and DAPI. The lesion site was photographed using a Zeiss LSM 780 confocal microscope or Olympus BX51 microscope, and the lesion center was manually outlined. Measurements were performed using ImageJ/Fiji software. The nerve gap, which was determined using ImageJ/Fiji, was the distance between the rostral and caudal ends of PGP9.5⁺ nerve in the injury area. The fibrotic scar area occupied by collagen III signal was thresholded and determined by ImageJ/Fiji software.

Behavioral testing

Behavioral testing was performed before surgery and weekly after surgery. All behavioral tests were performed by the same investigator, who was blinded to the study groups.

The hindpaw withdrawal frequency in response to a mechanical stimulus was determined using von Frey filaments of 0.7 mN and 3.9 mN (Stoelting Co., Wood Dale, IL, USA). Mice were placed on a metal wire mesh grid covered with a clear plastic cage. Mice were allowed to acclimatize to the environment for 30 min before testing. Von Frey filaments were applied to the mid-plantar surface of the hindpaw through the mesh floor with enough pressure to buckle the filaments. Probing was performed only when the mouse’s paw was in contact with the floor. A trial consisted of application of a von Frey filament to the hindpaw 10 times at 1 s intervals. If withdrawal occurred after application, it was recorded, and the next application was performed similarly when the mouse’s paw was again in contact with the floor. Mechanical withdrawal frequency was calculated as the percentage of withdrawals in response to 10 applications.

We performed automated gait analysis preoperatively and 1, 2, 4, and 8 weeks postoperatively using a “CatWalk” system (Noldus, Leesburg, VA, USA). All experiments were performed during the same period of the day (1:00 PM to 4:00 PM) and analyzed as previously reported.^105,106 Briefly, we trained mice to cross the CatWalk walkway daily for 7 days before SCI or control operation. During the test, each mouse was placed individually in the CatWalk walkway, which consists of a glass plate (100 × 15 × 0.6 cm) plus two Plexiglas walls, spaced 8 cm apart. The mice were allowed to walk freely and traverse from one side of the walkway glass plate to the other. Two infrared light beams spaced 90 cm apart were used to detect the arrival of the mouse and to control the start and end of data acquisition. We recorded these data when the room was completely dark, with the exception of the light from the computer screen. LED light from an encased fluorescent lamp was emitted inside the glass plate and completely internally reflected. When the mouse paws contacted the glass plate, light was reflected down and the illuminated contact area was recorded with a high-speed color video camera positioned under the glass plate and connected to a computer running CatWalk software, version 9.1 (Noldus). We compared stride length, print length, and print area in each run of each mouse at each time point. Paired Student’s t-tests were used for statistical analysis.

The hot plate test was used to calculate analgesic activity using the method described by Eddy and Leimbach¹⁰⁷ with minor modifications. Mice were retained on a hot plate having a stable temperature of 42 °C. Each mouse was placed individually on the hot plate to observe their reaction to heat-evoked nociceptive responses (licking of the forepaws and eventually jumping). The time taken for either paw licking or jumping was recorded. The latency until mice showed the first signs of discomfort (hindpaw lifting, hindpaw licking, or jumping) was recorded, and responses were determined at 1, 2, and 4 weeks after SCI.

Mice were also tested for hindlimb functional deficits at 1, 3, and 5 days and 1, 2, 4, 6, and 8 weeks (n = 6 per group) after SCI. Hindlimb locomotor recovery was assessed in an open field using the BMS score.⁴⁰ This scale ranges from 0 (indicating complete paralysis) to 9 (indicating normal movement of the hindlimbs). Performance of the left and right hindlimbs was averaged to obtain the BMS score.

Single-cell RNA-sequencing (scRNA-seq) analysis

Single-cell suspension was prepared using the 10X Genomics Chromium Single Cell 3’ Reagent Kit v3 (10X Genomics) according to the manufacturer’s protocol. The quantity and quality of cDNA were assessed using an Agilent 2100 Expert High Sensitivity DNA Assay. cDNA samples were sequenced on one lane of a NovaSeq 6000 S2 flowcell at Johns Hopkins School of Medicine for Genomics and Bioinformatics. Sequence alignment to the GRCm38 (mm10) reference genome was performed using Cell Ranger v.5.0.1.¹⁰⁸ More than 430 million reads were obtained for each sample. The average number of genes detected per cell was 848 ± 103 (mean ± s.e.m). Subsequent quality control filtering, normalization, clustering, and differential gene expression analysis was performed using Seurat (v.4.0.0, https://github.com/satijalab/seurat).¹⁰⁹ Quality control, normalization, log transformation, and highly variable gene identification were performed separately for each condition (sham, 13C4, 1D11). For each dataset, genes expressed in fewer than five cells were removed and ribosomal, mitochondrial, and Metastasis Associated Lung Adenocarcinoma Transcript 1 (MALAT1) genes were removed. Doublets and debris were removed by selecting for cells with mitochondrial gene content less than 25% of their total reads, a minimum of 200 unique features, and a maximum number of unique features set as the 95th percentile of the number of features detected per cell. The datasets were each normalized, log transformed, and highly variable genes were identified. After quality control filtering, the datasets contained 233 cells (1D11 condition), 281 cells (13C4 condition), and 238 cells (Sham condition), respectively.

Batch correction, clustering, and cell type annotation

The datasets for the three conditions (1D11, 13C4, and Sham) were merged in Seurat, and highly variable genes for the merged data were found to be the union of highly variable gene (HVG) for each individual sample and for the merged sample (total 4122 genes). The merged dataset was scaled, principal component analysis (PCA) was performed, and scores for expression of G2M and S phase cell cycle markers were computed and visualized on principal components (PC) coordinates. Batch correction was performed using Harmony (version 0.1.0, https://github.com/immunogenomics/harmony)¹¹⁰ to re-compute corrected embeddings. The Leiden algorithm was used to perform clustering on the corrected embeddings and clusters were visualized on a uniform manifold approximation and projection (UMAP) embedding. Cluster annotation was performed by visualizing key marker gene expression on a UMAP embedding, performing differential expression analysis using DESeq2 (version 1.30.1),¹¹¹ and performing classification using the Python version of SingleCellNet (https://github.com/pcahan1/PySingleCellNet)¹¹² using a Random Forest classifier trained on Tabula Muris Senis 10X data from skeletal muscle.¹¹³

Downstream analyses: pseudotime and ligand-receptor interactions

Pseudotime was computed on Harmony embeddings using Slingshot (version 1.8.0, https://github.com/kstreet13/slingshot).¹¹⁴ The change in gene expression over pseudotime of key TGF-β pathway genes was visualized in a heat map. Ligand-receptor interactions were interrogated using CellPhoneDB (version 2.0, https://github.com/Teichlab/cellphonedb),¹¹⁵ a repository of receptor-ligand interactions which accounts for subunit architecture and method to infer cell-cell communication networks from scRNA-seq data. CellPhoneDB’s dot plot function was used to visualize significant ligand-receptor interactions (P < 0.05) between cell types of interest. TGF-β target enrichment was performed using enrichR (https://github.com/wjawaid/enrichR)^116,117 to measure the extent to which genes more highly expressed in 13C4 overlapped with TGF-β effector target genes, as determined by Chromatin immunoprecipitation (ChIP)-Seq data of Smad1, Smad2, Smad3, and Smad4, as compiled in Epoch (https://github.com/pcahan1/epoch).

Statistics

Data are presented as means ± standard deviations. The comparisons among different groups were performed using multifactorial analysis of variance (ANOVA). When ANOVA testing indicated overall significance of main effects without interaction between them, the difference between individual time points and sites was assessed by post hoc tests. The level of significance was set at P < 0.05. All data analyses were performed using SPSS Statistics analysis software, version 15.0 (SPSS Inc, IBM Corp., Armonk, NY, USA).

Author contributions

D.P. designed and conducted the majority of the experiments and prepared the manuscript. P.W. helped with LysM-cre mice preparation and performed some data analysis. K.N. helped with single cell RNA-sequencing. P.C. provided suggestions for the project. X.C. developed the concept, supervised the project and wrote most of the manuscript.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41413-026-00507-7.