Identification of pro-fibrotic cellular subpopulations in fascia of gluteal muscle contracture using single-cell RNA sequencing

Abstract

Fibrosis is a common and integral pathological feature in various chronic diseases, capable of affecting any tissue or organ. Fibrosis within deep fascia is implicated in many myofascial disorders, including gluteal muscle contracture (GMC), Dupuytren’s disease, plantar fasciitis, iliotibial band syndrome, and chronic muscle pain. Despite its clinical significance, deep fascia fibrosis remains considerably under-researched compared to other fibrotic conditions. Single-cell RNA-sequencing (scRNA-seq) has been used to investigate cellular heterogeneity in fibrotic tissues. However, to our knowledge, only a few studies have applied scRNA-seq to explore cellular heterogeneity in deep fascia, and none have specifically examined fibrotic fascia. In this study, we performed scRNA-seq analysis on fibrotic fascia associated with GMC and compared them to nonfibrotic control fascial samples. Our findings show that fibroblast and macrophage cells play critical roles in pathological tissue remodeling within fibrotic deep fascia. We observed an upregulation of various collagens, proteoglycans, and extracellular matrix (ECM) glycoproteins in contracture deep fascia, attributed to the widespread activation of fibroblast subclusters. Additionally, two pro-fibrotic macrophage subpopulations, SPP1⁺ MP and ECM-like MP, appear to facilitate ECM deposition in fibrotic deep fascia by either regulating fibroblast activation or directly contributing to ECM production. The SPP1⁺ MP and ECM-like MP cells, as well as the signal interaction between SPP1⁺ MP and fibroblast cells, present potential therapeutic target for treating GMC and other related myofascial disorders.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12967-024-05889-y.

Introduction

Fibrosis is a prevalent and integral pathological condition in various chronic diseases, accounting for approximately 35% of global mortality and an even higher rate of 45% in the developed world [1]. It presents as the overgrowth and hardening of affected tissues. Pathologically, it is defined by an excessive accumulation of extracellular matrix (ECM) components, such as collagens and fibronectins [2]. Due to the lack of effective treatment for fibrotic diseases, extensive research efforts have been devoted to unravelling the cellular and molecular mechanisms underlying different fibrotic tissues. Such research aims to identify both common mechanisms shared across different tissue fibrosis and unique mechanisms specific to individual tissue fibrosis, thereby guiding therapeutic development through a deeper understanding of the pathological processes [3].

Gluteal muscle contracture (GMC), also known as gluteal fibrosis, is a clinical syndrome characterized by the contracture and distortion of the gluteal muscles, iliotibial band (ITB), tensor fasciae latae (TFL), and surrounding fasciae [4]. The fibrotic contracture of muscle and fascia tissues restricts hip external rotation and abduction, leading to walking difficulties, gait abnormalities, pain, and limb deformity [5]. While GMC is relatively rare globally, it is more prevalent in China, with a reported incidence of 1–2.5% in childhood [6]. This higher incidence is presumably related to both genetic factors and the historical use of diluent benzyl alcohol in intramuscular injection, which can cause persistent tissue damage and chronic inflammation, eventually leading to fibrosis in genetically predisposed individuals [5, 7]. In China, established GMC is often treated with invasive surgical interventions, such as division of contracture tissues or ablation of fibrotic fascia. However, these interventions often result in suboptimal outcomes, resulting in residual dysfunction or surgical complications like persistent hip pain, gluteal muscle wasting, postoperative hematoma, and muscle weakness [8, 9]. Currently, there is no specific treatment for mild GMC, nor are there any measures to prevent the post-fasciectomy recurrence of the disease [5].

GMC primarily manifests fibrotic changes in two tissues: muscle and fascia, with the latter specifically referring to deep fascia. Beyond GMC, deep fascia fibrosis is implicated in various other chronic conditions, such as Dupuytren’s disease, plantar fasciitis, iliotibial band syndrome, and chronic muscle pain [10, 11]. Unlike other tissues, deep fascia is predominantly composed of regularly arranged collagen fibers, with fibroblasts as the primary cellular component. Due to its crucial role in muscle force transmission, the deep fascia is particularly susceptible to injury, inflammation, and degenerative changes from overuse or misuse, which can lead to fibrotic lesions [12, 13]. Despite its clinical significance, deep fascia fibrosis remains significantly understudied compared to other fibrosis conditions. Even within myofascial disorders, research has primarily focused on understanding muscle fibrosis, with limited attention given to fibrotic changes in deep fascia [14, 15]. Therefore, elucidating the pathology and underlying mechanisms of deep fascia fibrosis represents an important and unmet medical need.

Single-cell RNA-sequencing (scRNA-seq) has emerged as a powerful tool for reconstructing cellular heterogeneity and cell-cell interactions across various fibrotic conditions, including those affecting muscle, skin, lung, and liver in the context of various chronic diseases [16–21]. Despite its widespread application, few studies have used scRNA-seq to investigate cellular heterogeneity in deep fascia, and no study has specifically examined the cellular mechanisms underlying fibrotic fascia [22].

In this study, we conducted scRNA-seq analysis on 10 fibrotic fascial bands associated with GMC and 4 nonfibrotic control fascial bands in this study. Our results suggest that fibroblast and macrophage cells play key roles in pathological tissue remodeling within the fibrotic deep fascia. We observed an upregulation of various collagens, proteoglycans, and ECM glycoproteins in contracture-affected deep fascia, driven by the widespread activation of fibroblast subclusters. Additionally, two pro-fibrotic macrophage subpopulations, SPP1⁺ MP and ECM-like MP, appear to contribute to ECM deposition in fibrotic deep fascia by either promoting fibroblast activation or directly producing ECM components.

Materials and methods

Subjects and samples

Fibrotic deep fasciae were obtained from the contracted and thickened upper ITB and TFL area of patients diagnosed with GMC during transverse fasciotomy, following a surgical procedure previously described [23]. Nonfibrotic control fasciae were obtained from a similar location in patients undergoing fracture reduction and internal fixation surgery for traumatic femoral fractures. All tissue samples were acquired at the Affiliated Changsha Central Hospital, Hengyang Medical School, University of South China, Changsha, China. Ethical approval for all research procedures was granted by the Ethics Committee at Changsha Central Hospital (Approval# 2022-S0064). Informed consent was obtained from all subjects prior to their participation.

Tissue processing

After collection, fresh fascial samples were stored at 4 °C in the GEXSCOPE Tissue Preservation Solution (Cat# 1200050001, Singleron Biotechnologies, Nanjing, China) and processed within 24 h. Samples weighing over 200 mg were washed three times with Hanks Balanced Salt Solution and were, then minced into 1–2 mm³ pieces. Subsequently, they were digested in GEXSCOPE Tissue Dissociation Mix Buffer Solution (Cat# 1200000044, Singleron Biotechnologies, Nanjing, China) inside a thermostatic oscillator at 37 °C and 180 rpm for 15 min. Tissue lysates were filtered through a 40 μm sterile strainer (Cat# 352340, Corning, New York, USA), and any remaining red blood cells were eliminated by incubating in Red Blood Cell Lysis Buffer (Cat# 00-4333-57, ThermoFisher, Waltham, USA) at 4 °C for 10 min. The viability of final single-cell suspensions was determined using Trypan Blue (Cat# T8154, Sigma, St. Louis, USA) and acridine orange/propidium iodide staining (Cat# CS2-0106, Nexcelom, Lawrence, USA). Samples with a concentration exceeding 1.5 × 10⁵ cells/mL and 85% viability proceeded to the next step.

ScRNA-seq library preparation and next-generation sequencing

Single-cell suspensions, diluted in phosphate-buffered saline at a concentration of 1.5 × 10⁵ cells/mL, were loaded onto a microfluidic device. Subsequently, scRNA-seq libraries were constructed according to the manufacturer’s protocol (GEXSCOPE Single Cell RNA-seq Kit, Singleron Biotechnologies, Nanjing, China). The resulting scRNA-seq libraries were sequenced on the Illumina NovaSeq 6000 platform with 150 bp paired-end reads, aiming for a sequence depth of approximately 30,000 reads per cell.

Gene expression matrices were generated from raw reads using scopetools (https://anaconda.org/singleronbio/scopetools) by filtering out reads lacking polyT tails and extracting cell barcodes and unique molecular identifiers (UMI). Subsequently, adapters and polyA tails were trimmed, and the reads were aligned onto the GRCh38 reference genome using Ensemble (v92) gene annotation. The count of UMIs per gene per cell was calculated by grouping reads with the same cell barcode, UMI, and gene. Cell counts were determined using the Knee Plot method, and low-quality barcodes located to the right of the inflection point were excluded from future analysis.

Quality control and data analysis

Cells were excluded from analysis if they exhibited < 250 or > 4,000 expressed genes; contained > 20,000 UMIs; had > 25% of reads originating from mitochondrial genes; or were identified as doublets by DoubletFinder (v2.0.4). After these quality control measures, a total of 86,159 cells were retained for subsequent analysis. To remove batch effects, Harmony (v0.1.0) was applied, using a clustering penalty parameter of 2 [24]. All cells were recalibrated based on the calculated correction factor with ridge regression penalty parameter set to 1 and width of soft k-means clusters to 0.1. Clustering and correction steps were repeated, with a maximum of 20 iterations, until batch effects were eliminated.

ScRNA clustering and cell type annotation

The NormalizeData and ScaleData functions of Seurat (v4.1.0) were used for data normalization and scaling. Principal component analysis (PCA) was conducted for dimensionality reduction on the top 2,000 variable genes identified by the FindVariableFeatures function. Thirty principal components and a resolution of 0.1 were applied using the FindNeighbors and FindClusters functions, resulting in the generation of different cell clusters. The resulting clusters were further manually examined and visualized using the RunUMAP function. To assign cell types to each cluster, the average expression level and expression ratios of a set of well-established cell type-specific markers were evaluated [25]. Clusters assigned to the same cell lineage were grouped together for the future analyses.

Cell composition analysis

Changes in cell compositions, specifically the proportions of each cell type between the GMC and control groups, were analyzed and visualized using the R packages ggpubr (v0.4.0) and ggplot2 (v3.3.6). The unpaired Student’s t-test was applied for this analysis, with a p value < 0.05 considered statistically significant.

Differential gene expression analysis

Differentially expressed genes (DEGs) within a specific cell type, compared to all other cell types, were identified using the R package limma (v3.50.3) with linear model fitting and a Bayes test [26]. Additionally, the FindAllMarkers function from the Seurat package was employed using the Wilcoxon rank-sum test and Bonferroni corrected p values. Genes were considered differentially expressed if they were detected in at least 10% of cells in either group and had a False Discovery Rate (FDR) < 0.05 and an absolute log fold-change (|log(FC)|) > 0.25. This threshold allowed us to capture a substantial number of genes with significant variation while avoiding an excessive list of DEG2 that could obscure the focus.

Subset data analysis

To generate a Seurat object containing only the desired cell type from the integrated data, the SubsetData function from the Seurat package was used. The subset Seurat object underwent similar filtering, normalization, and integration procedures, as described above.

Pathway enrichment analysis

From the obtained DEGs, upregulated and downregulated genes were separately analyzed for pathway enrichments as previously described. Gene Ontology (GO) term enrichments, covering biological process, cellular component, and molecular function, as well as Kyoto Encyclopedia of Genes and Genomes pathway enrichments, were constructed using Metascape (http://metascape.org) [27].

Pathway RespOnsive GENes (PROGENy) analysis

The human version of the functional genomics tool PROGENy (v1.16.0) was used to infer signaling pathway activity from single-cell transcriptomic data [28]. PROGENy uses a core set of pathway-responsive genes derived from publicly available signal transduction data (https://saezlab.github.io/progeny/). The footprint genes, i.e. the top 500 most sensitive genes according to z-coefficients which was calculated by DEGs analysis following certain pathway perturbation, were first identified. These genes (Pathway RespOnsive GENes) were assigned weights to indicate the strength and direction of pathway modulation. Pathway activity scores were then scaled and matched to individual cell clusters to explore key signaling pathways and their significant changes across different cell types.

DoRothEA transcription factor analysis

To analyze transcriptional factor (TF) activity, we used the human version of DoRothEA (v1.6.0), a gene regulatory network containing nearly 2,000 transcription factors (TFs) and regulatory elements derived from various human and mouse tissues and cell types [29]. TF activity was inferred based on the expression levels of target genes. Each TF–target interaction was assigned a confidence level (A, B, or C) based on the amount of supporting evidence from the TF and its targets (https://bioconductor.org/packages/release/data/experiment/html/dorothea.html). TFs with high confidence levels (A) are considered reliable for mechanistic exploration, while those with moderate and lower confidence (B and C, respectively) are considered screenable and merely predictive, therefore requiring further validation.

Core matrisome and ECM regulator scores

Gene correlation analysis was used to calculate ECM regulator scores. Macrophages and fibroblasts were subject to clustering at a high-resolution and matching with the Matrisome-master list from the human version of MatrisomeDB (https://matrisomedb.org/) [30]. Scores for the core Matrisome and ECM regulators were determined based on the average expression levels of the relevant genes in these cells obtained, as described above.

Cell trajectory analysis

To project the differentiation trajectories of macrophages and fibroblasts, we analyzed the differentiation status of these cells using the R package CytoTRACE (v.0.3.3) as previously described [31]. CytoTRACE values were calculated by measuring the expression levels of the top 200 genes that serve as markers of immaturity and differentiation. Subsequently, single-cell pseudotime trajectories were inferred using the R package Monocle2 (v2.2.0) for macrophages and Slingshot (v2.2.1) for fibroblasts, incorporating the previously obtained CytoTRACE values [32, 33].

Cell–cell communication analysis

Intercellular communications between cell types were analyzed using the R package CellChat (v1.6.0) and the ligand–receptor interaction reference database CellChatDB.huma (http://www.cellchat.org/) as previously described [34]. The function identifyCommunicationPatterns and network analysis were used to predict major signaling inputs and outputs for a specific cell type. Interaction strength, a metric reflecting the communication probability between a given ligand–receptor interaction, was calculated using the computeCommunProbPathway function.

The R package Nichenet (v2.0.5) was employed to predict the effects of ligands on their target genes [35], using ligand–receptor pairing information (lr_network), a target gene set (ligand_target_matrix), and gene regulatory networks (weighted_networks) from a public database (https://zenodo.org/record/7074291). The function nichenet_seuratobj_aggregate was used to screen the DEGs (|log(FC)| > 0.25, expression ratio > 0.1) of the sender cells and the receiver cells, and to predict ligand–target genes using the ligand–target gene regulation model. The top 60% of ligand–target gene pairs from the top35-ranked ligands were considered active and hence visualized.

Masson’s trichrome staining and microscopy

The tissues were fixed in 4% paraformaldehyde at 4 °C overnight, followed by sequential dehydration in xylene and ethanol. Next, the tissues were embedded in paraffin. Paraffin-embedded tissue blocks were sectioned at a thickness of 3 μm onto microscope slides. Masson’s trichrome staining was used to detect collagen fibers following previous described protocol. The staining procedure was done at the Affiliated Hospital of Guilin Medical University, Guilin, China, in their pathology core lab. Images were captured at 40 x magnification using bright-light microscopy (Leica, Wetzlar, Germany).

Real-time quantitative PCR

Total RNA was extracted from fascial tissues using TRIzol Reagent™ Solution (Cat# AM9738, ThermoFisher, Waltham, USA) and reverse transcribed into cDNA using RevertAid First Strand cDNA Synthesis Kit (Cat# K1622, ThermoFisher, Waltham, USA). The cDNA template was subjected to quantitative real-time PCR (qPCR) following the manufacturer’s instructions for Maxima SYBR Green qPCR Master Mix (Cat# K0253, ThermoFisher, Waltham, USA) using QuantStudio™ 5 (Thermo Fisher Scientific) and with gene-specific primers (Table S1). The 2^−ΔΔCT method was used to analyze the relative changes in gene expression, normalized against an internal control 18S ribosomal RNA.

Statistical analysis

All data are presented as mean ± SD. Statistical analyses were performed using GraphPad Prism (v9). Differences between two-set samples were assessed using Welch’s test or an unpaired two-tailed t-test. A p value < 0.05 was considered statistically significant.

Code Availability

All original codes will be deposited at GitHub and be available online before publication.

Data Availability

The datasets generated during this study are available at the NCBI GEO accession code GEO: GSE273293.

Results

A single-cell atlas of the normal and GMC-associated deep fasciae

To unravel the cellular heterogeneity associated with deep fascia fibrosis, we performed scRNA-seq on fascial tissues from the upper ITB and TFL regions obtained from patients with GMC and nonfibrotic condition (Fig. 1A and Table S2). In GMC cases, contracture fascial tissues were identified under arthroscopy and extracted via arthroscopy-guided transverse fasciotomy (Fig. S1A). For the control group, fascial tissues from a similar location were obtained during fixation surgery for acute femoral injuries. As expected, compared with control fascia, the contracture fascial tissues exhibited increased expression of collagen genes (COL1A1, COL3A1, COL6A1, and COL14A1) and enhanced collagen fiber staining, confirming their fibrotic status (Fig. S1B and S1C).

Fig. 1.

Single-cell landscape of control and contracture GMC fascia bands. (A) Schematic of workflow of single-cell RNA-sequencing for cells from contracture GMC fascia bands and control samples. (B) UMAP visualization of 86,159 cells from jointly analyzed contracture GMC fascia bands (n = 10) and control samples (n = 4). (C) Heatmap of differentially expressed genes and their average expression levels (coded by colors) in each cell type. Columns denote cell type; rows denote genes; and genes associated with ECM pathway are marked with an asterisk. (D) UMAP visualization of cells from separately analyzed contracture GMC fascia bands (n = 10) and control samples (n = 4). (E) Violin plots showing the expression of cell type-specific markers in each cell type. (F) Bar plots showing the average proportion of each cell type in GMC and control fascia bands. (G) Boxplot comparing the relative proportion of each cell type between GMC fascia bands and control bands. P values were calculated by the Welch-test. (H) Scatter plot showing the number of differentially expressed genes and their fold changes between GMC and control bands across cell types. GMC: gluteal muscle contracture; Con: control. *: p < 0.05

ScRNA-seq and unbiased clustering of 86,159 cells from 14 samples (10 GMC-associated and 4 control fasciae) revealed a total of 9 cell clusters (Fig. 1B). Quality control metrics demonstrated comparable consistency across these clusters (Fig. S2A and S2B). Based on the differential expression of known cell type-specific marker genes, these clusters were manually categorized into seven known cell types (Fig. 1B and C). Cluster 7, identified as muscle stem cells, was predominantly observed in sample Con_04 (87.2%), likely due to residual muscle tissue contamination (Fig. S3A and Table S3). All other clusters were consistently present between groups and across all samples (Fig. 1D, Fig. S3B, and Table S3). Consequently, cluster 7 was excluded from subsequent analyses.

The majority of cells (66.8 ± 17.7%) were identified as fibroblasts, comprising two clusters prominently marked by the expression of ECM genes such as LUM, DCN, and FBLN1 [36]. Endothelial cells, characterized by genes like PECAM1, CLDN5, and TSPAN7 [18], and smooth muscle cells (SMCs), expressing ACTA2, MYL9, and MYH11, were also readily seen in the fascial tissues. Three clusters were annotated as immune cells, including NK/T cells, mast cells, and macrophages. Among these, macrophages accounted for a significant proportion (8.5 ± 7.0% of all cells and 70.8 ± 0.4% of total immune population) and expressed canonical markers such as CD68, RNASE1, and C1QA (Fig. 1E and Table S4).

Next, we analyzed the proportions of different cell types in GMC and control fasciae (Fig. 1F and S4). Both groups exhibited similar proportions of fibroblasts, endothelial cells, and SMCs (Fig. 1F and Table S5). Consistent with previous reports highlighting the critical role of macrophage recruitment and activation in fibrosis, macrophage proportions were significantly increased in GMC-associated fasciae compared to controls (11.2 ± 6.6% vs. 1.8 ± 1.0%, Fig. 1G). We also examined DEGs between GMC and control fasciae across cell lineages (Fig. 1H and Table S6). Functional enrichment analysis of upregulated genes revealed activation of collagen-containing ECM, closely linked to fibrosis pathology. In contrast, downregulated genes were mainly associated with ribosomal and mitochondrial functions (Fig. S5 and S6). Macrophages displayed the highest number of upregulated genes and were the significantly increased cell proportions in GMC fasciae (Fig. 1G and H), indicating substantial signaling alterations in this cell type.

Identifying the pathogenic macrophage subclusters in GMC-associated deep fasciae

To characterize the macrophage compositions and identify the pro-fibrotic subpopulations driving GMC, we extracted and re-clustered all 7,143 macrophages from both GMC and control groups into 7 annotated subclusters (Fig. 2A–D, Fig. S7–9, Table S7, and Table S8). Among them, the Monocyte MP subcluster represented a naive (M0) macrophage population with high expression of S100 genes [37]. The Proliferative MP subcluster exhibited high levels of cell-cycle regulation genes (MKI67, TOP2A, and CDKN3), indicating a proliferative characteristic [38]. The MoDCs MP subcluster expressed high level of antigen-presenting genes (CD1C, FCER1A, and MHC II) as well as classical pro-inflammatory (M1) macrophage marker genes (TNF, CD86, and HLA-DRB1), representing a tissue-infiltrated antigen-presenting macrophage. The Proinflamm MP subcluster displayed both M1 and anti-inflammatory (M2) profiles, with high levels of pro-inflammatory chemokines (CXCL2, CCL3, and CXCL3) and cytokine-activating AP-1 genes (JUN and FOSB), suggesting its involvement in inflammatory responses.

Fig. 2.

Characterization of macrophage cell types in control and contracture fascia bands. (A) UMAP depicting 7,143 macrophage cells from jointly analyzed contracture GMC (n = 10) and control fascia bands (n = 4). (B) UMAP embedding of macrophage cells from separately analyzed contracture GMC (n = 10) and control fascia bands (n = 4). (C) Bubble plot showing expression of marker genes across macrophage subclusters. Dot size represents the percentage of expressed; dot color reflects the relative expression level. (D) Bar plots showing the average proportion of macrophage subclusters in GMC and control fascia bands. (E) Violin plots showing ECM regulator scores stratified by macrophage subclusters. (F) Violin plots showing core matrisome scores stratified by macrophage subclusters. (G) Heatmap showing PROGENy pathway activities of macrophage subclusters. (H) UMAP plot of CytoTRACE score (coded by colors) in the macrophage cells. (I-J) Pseudotime trajectory analysis using Monocle 2. Cells are colored by pseudotime (I) or overlaying with cluster assignment (J). (K) DoRothEA transcription factor analysis of macrophage subclusters. MP: macrophage; GMC: gluteal muscle contracture; Con: control

Three macrophage subclusters appeared to be M2-like or M2-polarized: LYVE1⁺ MP, SPP1⁺ MP, and ECM-like MP. The LYVE1⁺ MP subcluster expressed high levels of tissue-resident macrophage marker gene LYVE1, complement system-related genes (C1QA, C1QB, and MARCO), and the M2 macrophage marker CD163. Although not as typical as LYVE1⁺ MP, the SPP1⁺ MP subcluster presented relatively high level of M2 marker genes CCL18 and CCL22, suggesting its M2-likely nature (Fig. 2C and S7). In SPP1⁺ MP, increased expression of SPP1, along with lipid metabolism-associated genes (APOE, LPL, and FABP5) and matrisome-related genes (MMP9 and CTSK), was also noted, indicating enriched functions in lipid metabolism and ECM remodeling. In addition, we also identified an ECM-like MP subcluster, which expressed high level of ECM and ECM regulation-related genes (CLU, DCN, LUM, PRELP, and FMOD), suggesting its direct involvement in producing ECM components, especially proteoglycans. The ECM-like MP subcluster didn’t display a typical M1 or M2 paradigm but exhibited more M2-like characteristics. Notably, these three M2-associated clusters, were more expanded in GMC-associated fasciae (Fig. 2D and S10), consistent with previous observations that M2 macrophages, rather than M1 macrophages, are the main population contributing to matrix deposition and tissue remodeling during various fibrotic processes.

To unbiasedly identify pro-fibrotic macrophage subpopulations, we scored each macrophage subcluster based on the expression of genes defined as ECM regulators [30]. Our data showed that higher ECM regulator scores in M2-associated clusters, with the highest in the SPP1⁺ MP subcluster (Fig. 2E). Treating phorbol 12-myristate 13-acetage (PMA)-differentiated THP-1 macrophages with IL4, a classical M2 inducer, recapitulated the increased expression of M2 marker gene ALOX15 and SPP1⁺ MP markers, including SPP1, CSTB, FABP5, and TREM2 (Fig. S11). A recent study reported the existence and importance of SPP1⁺MAM⁺, a universal M2-polarized, SPP1⁺, matrisome-associated, fibrogenic macrophage subpopulation, in a wide range of human fibrotic diseases [39]. Therefore, we examined whether our SPP1⁺ MP corresponds to the reported SPP1⁺MAM⁺. Notably, most differentially upregulated genes in SPP1⁺MAM⁺ (such as SPP1, MMP9, LPL, CHI3L1, MATK, and GPC4) were also highly expressed in our SPP1⁺ MP, suggesting a consistent profile between our SPP1⁺ MP and the reported fibrogenic SPP1⁺MAM⁺ (Fig. S12).

We also scored these macrophage subclusters for their expression of core ECM components, including collagens, glycoproteins, and proteoglycan. ECM-like MP was the macrophage subpopulation presenting the highest core matrisome score (Fig. 2F). ECM-related genes like DCN, CLU, LUM, COMP, COL1A2, COL6A2, and MMP3 were all highly expressed in this subpopulation (Fig. S13). Concomitantly, ECM-like MP also showed a high ECM regulator score, suggesting a potential self-regulation mechanism (Fig. 2E).

We next analyzed macrophage subclusters for signal pathway activity using PROGENy [40]. As expected, MoDCs MP displayed strong activity in classical, pro-inflammatory M1 pathways, such as TNFa, JAK-STAT, and NFkB (Fig. 2G). Consistent with VEGF’s role in M2-polarization of macrophages, with LYVE1⁺ MP displayed high VEGF activity (Fig. 2G) [41]. In ECM-like MP, the activity of canonical fibrogenic TGFβ pathway and the expression level of TGFβR II (type II TGFβ receptor) were notably robust (Fig. 2G and S14), further suggesting a potential self-regulation mechanism. Since TGFβ binding to TGFβR II is crucial for forming the heterotetrameric TGFβR complex and triggering downstream signaling, we examined the main sources of TGFβ in this case. Naive Monocyte MP and the ECM-like MP subclusters expressed the highest level of TGFβ1 and TGFβ2, respectively, suggesting that ECM-like MP subcluster can be activated by both pre-secreted TGFβ1 or self-secreted TGFβ2 (Fig. S14) [42]. In SPP1⁺  MP, significant TRAIL and TNFa pathway activities were observed, consistent with findings that TRAIL supports the anti-inflammatory function of macrophages, and a pro-fibrotic SPP1⁺ macrophage population in chronic kidney disease and heart failure displays increased TNFa signaling [30, 43, 44].

Altogether, our data suggest that SPP1⁺ MP is a fibrogenic macrophage subcluster with an important regulatory role in deep fascia fibrosis, while ECM-like MP may directly contribute to ECM deposition through TGFβ pathway activation.

Origin of SPP1⁺ MP and ECM-like MP in deep fasciae

We next conducted single-cell trajectory analysis to investigate the origin of the SPP1⁺ MP and ECM-like MP subclusters. CytoTRACE and Monocle differentiation trajectory analyses identified two distinct destinies: one from the Monocyte MP towards the LYVE1⁺ MP and another from the Monocyte MP towards the SPP1⁺ MP and ECM-like MP, with the diverging point being somewhere around the Proinflamm MP. Importantly, SPP1⁺ MP and ECM-like MP were at the end of the trajectory and preceded by LYVE1⁺ MP (Fig. 2H–J). By clustering enriched genes along the differentiation trajectory, we uncovered four stages of gene expression during the development process from Monocyte MP to SPP1⁺ MP and ECM-like MP. Stage 1, 2, and 3 displayed enriched GO terms including responding to stimuli and T-cell activation, antigen presentation, and macrophage activation, respectively. Stage 4 was mainly associated with SPP1⁺ MP and ECM-like MP cells, with enriched GO terms of ECM formation and remodeling (Fig. S15).

To further understand the transcription factors controlling the gene expression of the SPP1⁺ MP and ECM-like MP, we analyzed DoRothEA transcription factor activity in these subclusters. Interestingly, SPP1⁺ MP was characterized by high activity of transcription factors ATF1 and PPARG, both have been previously reported to be associated with M2 polarization (Fig. 2K) [45, 46]. ECM-like MP showed high activity of Hippo pathway transcription factor TEAD1, along with high expression levels of Hippo downstream effectors such as YAP1, TAZ, and TEAD1, suggesting that Hippo pathway might contribute to the ECM production of ECM-like MP (Fig. 2K and S16) [47].

Revealing fibroblast heterogeneity in GMC fibrotic fasciae

Fibroblasts are the predominant cell type in connective tissue, also the population playing the central role in the pathological accumulation of ECM during fibrosis. Therefore, we conducted unsupervised clustering on a total of 59,948 fibroblasts from both control and GMC groups, and we revealed further heterogeneity with five fibroblast subclusters (Fig. 3A and B). These subclusters were annotated based on their dominant subcluster-specific genes as PI16⁺ FB, JUNB⁺ FB, COMP⁺ FB, CXCL14⁺ FB, and MFAP5⁺ FB (Fig. 3C, Fig. S17, and Table S9). Upon examining the expression patterns of marker genes and analyzing the Pearson correlation of overall gene expression across fibroblast subpopulations, we observed similar profiles between MFAP5⁺ FB and CXCL14⁺ FB, a similarity between COMP⁺ FB and JUNB⁺ FB, but a unique profile for PI16⁺ FB (Fig. 3C and S18).

Fig. 3.

Characterization of fibroblast cell types in contracture GMC and control fascia bands. (A) UMAP depicting 59,948 fibroblast cells from jointly analyzed contracture GMC (n = 10) and control fascia bands (n = 4). (B) UMAP embedding of fibroblast cells from separately analyzed contracture GMC (n = 10) and control fascia bands (n = 4). (C) Bubble plot showing expression of marker genes across fibroblast subclusters. Dot size represents the percentage of expressed; dot color reflects the relative expression level. (D) UMAP plot of CytoTRACE score (coded by colors) in the fibroblast cells. (E) Trajectory analysis using the Slingshot algorithm, showing the trajectories of fibroblast subsets connected by differentiation state. (F) Violin plots showing core matrisome scores stratified by fibroblast subclusters. (G) Violin plots showing ECM regulator scores stratified by fibroblast subclusters. (H) Bar plots showing the average proportion of fibroblast subclusters in GMC and control fascia bands. (I) Feature plots showing the expression distribution of ACTA2 (the up panel) or COL1A1 (the down panel) in fibroblast cells between control and GMC groups. (J) Column plot showing the percentage of α-SMA⁺/COL1A1⁺ myofibroblasts in fibroblast subclusters between control and GMC groups. (K-M) Heatmap showing the differential expression of collagens (K), proteoglycans (L), and glycoproteins (M) between contracture GMC fascia bands and control samples across fibroblast subclusters. FB: fibroblast; GMC: gluteal muscle contracture; Con: control

PI16⁺ fibroblasts have been previously reported as a universal, steady-state fibroblast subtype present in multiple mouse and human tissues [48]. Consistently, in our trajectory analysis, PI16⁺ FB represented the least differentiated subtype, serving as the origin point for the development of other fibroblast subpopulations (Fig. 3D and E). One differentiation path, from JUNB⁺ FB to COMP⁺ FB, involved sequential activation of AP-1 genes (JUNB and FOSB) and cartilage/tendon linage-associated genes (CRTAC1, COMP, and TNC), suggesting a progression from quiescent PI16⁺ fibroblast to COMP⁺ tenocytes (Fig. 3C–E and S19). COMP, a key ECM component, maintains tendon and cartilage structure, and COMP⁺ FB has recently been reported with fibrosis in conditions like scleroderma [49, 50]. In another trajectory, CXCL14⁺ FB, following PI16⁺ FB, expressed high level of complement activation genes (CXCL14, CXCL12, C3, and CFD), suggesting an inflammatory role. Subsequently, MFAP5⁺ FB expressed microfibril-associated glycoprotein MFAP5 and matrix metalloproteinase MMP2, indicating its involvement in ECM remodeling (Fig. 3C–E and S19). Importantly, MFAP5⁺ FB has been found to interact with macrophages, promoting ECM remodeling in tumor microenvironment, and recombinant MFAP5 itself enhances fibroblast migration and pro-fibrotic gene expression in skin fibrosis mouse model [51–53]. As expected, the more differentiated MFAP5⁺ FB and COMP⁺ FB subpopulations exhibited higher matrisome and ECM regulator scores compared to less differentiated subpopulations CXCL14⁺ FB and JUNB⁺ FB (Fig. 3F and G).

Surprisingly, when comparing the proportions of different fibroblast subpopulation between the control and GMC groups, we found no significant differences in their composition (Fig. 3H, Fig. S20, and Table S10). Myofibroblasts, a subset of activated, fibrogenic cells marked by the expression of α-smooth muscle actin (α-SMA; ACTA2) and COL1A1 are normally considered as the central mediator of tissue fibrosis. To our surprise, although α-SMA⁺/COL1A1⁺ myofibroblasts were readily seen in all fibroblast subpopulations, they were not significant upregulated in GMC tissues or in GMC-related fibroblast subpopulation compared to the control group (Fig. 3I and J).

To understand how fibroblast subpopulations contribute to the ECM accumulation in GMC fascia bands, we examined collagen expression across subpopulations. At bulk level, compared to control fasciae, GMC fascia showed significantly higher expression of COL1A1/2, COL3A1, COL6A1/2, and COL14A1 (Fig. S1B and S21). These upregulations were not confined to a specific fibroblast subpopulation; instead, most were upregulated across multiple subclusters in both differentiation paths, with a trend of more upregulation towards the end of both trajectories (i.e., MFAP5⁺ FB and COMP⁺ FB, Fig. 3K). Notably, COL14A1 was uniquely upregulated in the CXCL14⁺ FB to MFAP5⁺ FB trajectory, with higher expression in the less differentiated CXCL14⁺ FB (Fig. 3K).

In addition to collagens, non-collagen ECM components also play important roles in collagen dynamics, including deposition, remodeling, and breakdown of collagen. These non-collagen components contribute to fibrotic diseases by influencing various cellular functions like proliferation, migration, differentiation, and immune regulation [54, 55]. Therefore, we also analyzed the expression of non-collagen ECM components, including glycoproteins, proteoglycans, ECM regulators, ECM-affiliated proteins, and ECM-binding secreted factors in our samples. A wide range of non-collagen ECM components was found significantly changed in the GMC group. Although multiple cell types were involved in the regulation of ECM-binding secreted factors and ECM-affiliated proteins, fibroblasts were the major source of the upregulated ECM glycoproteins and proteoglycans (Fig. S22). Similarly, the upregulations in proteoglycans and ECM glycoproteins occurred across multiple fibroblast subpopulations (Fig. 3L and M). For example, proteoglycans such as LUM, ASPN, PRG4, DCN, VCAN, and BGN and ECM glycoproteins like CTGF and PCOLCE2 were upregulated in fibroblasts from both cartilage/tendon and inflammation linages.

In summary, among the five fibroblast subclusters, PI16⁺ FB was the least activated, serving as a root fibroblast, while other fibroblast subclusters collectively contributed to the increased ECM composition in GMC-associated deep fasciae (Fig. 3F and G). This suggests that fibrotic changes in deep fascia likely result from the collective activation of diverse fibroblast subpopulations, rather than from the activation of myofibroblasts, as seen in other tissues [56].

Cellular communication between macrophages and fibroblasts in GMC fibrotic fasciae

To identify potential mediators of pathological macrophage–fibroblast communication, we analyzed ligand–receptor interactions between different cell types. Overall, the GMC group showed reduced interaction number and strength between most cell types, compared to controls. However, an enhanced communication strength was observed between macrophages and fibroblasts in the GMC condition (Fig. 4A). Transmission electron microscopy (TEM) further revealed fibroblasts and macrophages in relatively close proximity between 2-50 μm range (average distance: 17.5 ± 12.7 μm, Fig. S23), suggesting possible signaling interaction between these two cell populations.

Fig. 4.

The crosstalk of macrophage–fibroblast was increased in GMC. (A) Network plots showing differential inter-population ligand–receptor interactions between GMC fascia bands and control bands. Red line represents increased interaction in GMC fascia bands; blue line represents decreased interaction in GMC fascia bands. (B) Heatmaps showing differential ligand–receptor interactions from macrophage subclusters to fibroblast subclusters comparing GMC fascia bands and control bands. Red represents increased interactions in GMC fascia bands, while blue represents decreased interactions in GMC fascia bands. (C) Predicted ligands from SPP1⁺ MP that regulate respective target genes in fibroblasts. Edges are scaled by the inferred regulatory potential of the interaction. MP: macrophage; FB: fibroblast; GMC: gluteal muscle contracture; Con: control

Among the seven macrophage subtypes, only SPP1⁺ MP displayed a ubiquitously increased communication with multiple fibroblast subtypes, while leaving PI16⁺ FB largely unaffected, in the GMC condition (Fig. 4B). Therefore, we explored the molecular signals driving the widespread activation of various fibroblasts by SPP1⁺ MP. One of the highest-ranked SPP1⁺ MP-derived ligands was the well-established ECM regulator gene TGFβ1, promoting the upregulation of proteoglycan FMOD and VCAN, ECM glycoprotein SPARC, and ECM regulator TIMP3 in fibroblasts (Fig. 4C). Other SPP1⁺ MP-derived ligands, such as SPARC, TIMP1, HBEGF, DCN, MMP9, CXCL2, SERPINF1, ICAM1, and SPP1, were associated with the regulation of core matrisome genes, including COLA3A1, CHAD, KERA, VCAN, BGN, SPARC, FBN1, and TIMP3 (Fig. 4C, Fig. S24, and Table S11). Notably, SPARC, a SPP1⁺ MP expressed ligand that drives the increased expression of proteoglycan CHAD and KERA, was also found to be upregulated by fibroblasts (Fig. 4C). These SPP1⁺ MP-derived signals likely contribute to the increased expression of collagen, proteoglycans, ECM glycoproteins, and ECM regulator in GMC fibroblasts.

Discussion

Deep fascia fibrosis refers to the abnormal thickening and hardening of the fascial tissue surrounding bones, muscles, and nerves [10]. Fibrotic fascia loses its flexibility and elasticity, leading to pain, stiffness, and movement difficulty [12]. Deep fascia fibrosis is involved in various chronical conditions, such as GMC, Dupuytren’s disease, and chronic muscle pain. Notably, about 80% of people experiencing myofascial pain in their lifetime [10, 57]. However, research has long primarily focused on the features and mechanisms of the associated muscle tissue, often neglecting the significance of fascial pathophysiology [58]. Currently, the dominant hypothesis suggests that resident fibroblasts within fascia tissue are stimulated by certain factors to undergo abnormal differentiation into myofibroblasts, which in turn develop into a contracture cord characterized by increased myofibroblast activity and ECM deposition [19]. Much of this hypothesis stemmed from the observations that, first, myofibroblasts play a vital role in fibrosis in the majority of examined organs and tissues, and second, increased ACTA2 expression and more myofibroblasts were reported in myofascial tissue associated with Dupuytren’s contracture affecting the palmar fascia [19, 59]. Therapeutic efforts have thus focused on reducing myofibroblast activation and collagen deposition to treat fibrotic diseases. To date, the cellular compositions and molecular mechanisms of deep fascia fibrosis remains poorly understood. Using surgically resected contracture fascia tissues from GMC patients and nonfibrotic control fascia samples from similar anatomic locations, we aimed to study the cellular landscape of fibrotic fascia tissue using scRNA-seq technology.

In this study, we noticed that the predominant cells of the fascial tissue and the main contributor to ECM deposition in fibrotic deep fascia were, in fact, fibroblasts. However, contrary to the conventional theory, the number of ACTA2⁺ myofibroblasts was not increased in the fibrotic fascial bands. Interestingly, Zhang et al. similarly observed no significant increase of myofibroblasts in a mouse model of muscle repair and fibrosis [60]. In addition, the activation of myofibroblast may vary among tissues, as the density of myofibroblasts in human fascia differs across different parts of the body—for example, significantly higher in lumbar fascia compared to plantar fascia and ITB [61]. Supporting our findings, a few studies also suggested that ACTA2⁺ myofibroblast are not solely responsible for abnormal tissue fibrogenesis. In mouse models of bleomycin-induced lung fibrosis and unilateral ureteral obstruction-induced kidney fibrosis, selectively blocking TGFβ activating in ACTA2⁺ fibroblasts did not protect mice from fibrogenesis, while ACTA2⁻ fibroblasts isolated from the fibrotic tissues were similarly capable of activating TGFβ as ACTA2⁺ fibroblasts [62]. In mdx^5cv mice, a model for Duchenne muscular dystrophy, the level of ACTA2 expression in fibrogenic cells was not correlated with the severity of muscle fibrosis or the level of collagen expression [63]. To the best of our knowledge, this study is the first to observe that ACTA2 might not be a definitive marker of collagen-producing fibroblasts in human fascia fibrosis.

In our samples, four fibroblast subtypes differentiated from the primary PI16⁺ FB through two trajectory lineages: the cartilage/tendon linage (JUNB⁺ FB and COMP⁺ FB) and the inflammation linage (CXCL14⁺ FB and MFAP5⁺ FB). All four differentiated fibroblasts showed upregulated expression of various ECM components. Previous studies have reported that two fibril-forming collagens, type I and type III collagens, were upregulated in contracture ITB associated with GMC [7]. Beyond these (COL1A1/2 and COL3A1), we observed increased expression of non-fibril-forming collagens and fibril-associated collagen (COL6A1/2 and COL14A1) as well. COL1A1/2, COL3A1, and COL6A1/2 were upregulated in fibroblasts from both the cartilage/tendon and inflammation linages. Interestingly, COL14A1 was primarily expressed in fibroblasts from inflammation-linage, with higher expression in the less differentiated CXCL14⁺ FB than in the more differentiated MFAP5⁺ FB. This observation aligns with a recent report suggesting that type XIV collagen plays a regulatory role in the early stages of collagen fibrillogenesis [64].

In addition to collagens, we noticed that over twenty non-collagen ECM components were significantly overexpressed by fibroblasts in GMC-associated fascia band. Cartilage/tendon linage and inflammation-linage fibroblasts may differentially contribute the ECM composition in fibrotic fascia bands, as more proteoglycans were upregulated in cartilage/tendon linage fibroblast, while more ECM glycoproteins were upregulated in inflammation linage fibroblasts (Fig. 3L and M). Notably, non-collagen ECM components including PRG4, VCAN, and MFAP5 were mostly associated with late-differentiated MFAP5⁺ FB, indicating their involvement in the late stage of inflammation-related collagen deposition.

Our study identified two pro-fibrotic macrophage populations associated with the contracture GMC fascia band. The first population is SPP1⁺ MP, marked by the expression of SPP1, M2 marker genes, lipid metabolism-associated genes, and matrisome-related genes. SPP1⁺ macrophage was initially identified in liver cirrhosis and have since been studied in human lung, liver, and heart fibrosis [16, 65, 66]. For instance, Hoeft et al. described a pro-fibrotic SPP1⁺ macrophage population, characterized by the markers SPP1 and FN1, which exhibits latent pro-inflammatory activation with increased TNF signaling in human chronic kidney disease and heart failure [30]. Another recent multi-organ fibrosis study discovered a universal M2 polarized, SPP1⁺, matrisome-associated macrophage population (SPP1⁺MAM⁺) in the fibrotic tissues of the heart, lung, liver, kidney, skin, and uterus, suggesting a broader implication of SPP1⁺ macrophages in human fibrotic diseases. Similarly, in the tumor microenvironment, SPP1⁺ macrophage with enhanced glycolysis contribute to tumor metastasis by promoting matrix remodeling [67]. While not tested in GMC context yet, SPP1⁺ MP have been shown to activate fibroblasts and promote ECM deposition in other fibrotic models. In IPF mouse models, SPP1 binds to CD44 to induce fibrin synthesis, with SPP1 knockout significantly inhibiting pulmonary fibrosis [68]. Additionally, blocking SPP1⁺ macrophages by CXCL4 has been shown to inhibit fibrosis in heart and kidney in mouse models [30]. Through ligand–receptor interaction analysis, we identified SPP1⁺ MP as the key macrophage subset responsible for the broad fibroblast activation in GMC fascia band. Ligands such as TGF-β1, SPARC, TIMP1, HBEGF, and SPP1, derived from SPP1⁺ MP, have potential roles of upregulating various ECM components in fibroblasts. Therefore, our data not only corroborated the role of SPP1⁺ macrophage in human fibrotic disease but also suggested a distinct mechanism by which SPP1⁺ MP promoted ECM restructuring through broad regulation of fibroblast activation [69].

Another pro-fibrotic macrophage population identified in the contracture GMC fascia band was ECM-like MP, marked by high expression of ECM and ECM regulation-related genes. While macrophages are commonly known for their role in inducing ECM production by activating fibroblasts, emerging evidence suggests that macrophages can also directly produce ECM components during tissue repair [70, 71]. In our study, the ECM-like MP expressed high level of ECM-related genes, including DCN, CLU, LUM, COMP, COL1A2, COL6A2, and MMP3. Additionally, this subpopulation also expressed high level of TGFβR II, indicating its potential involvement in ECM deposition through the canonical fibrogenic TGFβ pathway.

Currently treatments for idiopathic pulmonary fibrosis (IPF) include nintedanib, a tyrosine kinase inhibitor, and pirfenidone, a likely TGF-β regulator, with no approved therapies for fibrotic conditions. Several promising investigational drugs are in phase II or III clinical trials, including the PDE4 inhibitor (BI-1015550) for IPF; IL-4/IL-13 inhibitors (SAR156597) for IPF and systemic sclerosis; anti-TNF antibody (adalimumab) for Dupuytren’s disease; and recombinant human pentraxin-2 (PRM-151) for pulmonary fibrosis and myelofibrosis [3]. Our analysis suggests that SPP1⁺ MP-derived signals may drive increased collagen and non-collagen ECM expression in GMC fibroblasts, with TGF-β1 being a key regulator of fibroblast activation. Additionally, ECM-like MP, another collagen-production population, is also regulated by TGF-β signaling. Increased TNF signaling was observed in SPP1⁺ MP. These findings indicate that targeting TGF-β and TNF pathways may be worth investigating for slowing or preventing GMC progression. Moreover, SPP1⁺ MP influenced ECM production in fibroblasts through signaling pathways such as SPARC and HBEGF, both of which have shown anti-fibrotic effects when blocked in vivo [72–74]. Effective anti-fibrosis strategies may require targeting multiple pathways through combined interventions, and our findings provide potential targets for developing combination therapies for GMC treatment.

This study faced several limitations. First, sample collection, particularly for control samples, was challenging. Second, while widely used, scRNA-seq technology has known limitations like technical noise, cell stress during dissociation, transcript coverage bias, and dropout events. Although increasing sample numbers helped meet analysis requirement, these issues may not be fully resolved [75]. Additionally, scRNA-seq lacks spatial context, which is crucial for understanding cell interactions in their native microenvironment. Future study using combining scRNA-seq with spatial transcriptomics RNA sequencing (stRNA-seq) could provide a more comprehensive understanding of the cellular and molecular components in GMC [76]. Another significant challenge is the lack of validated in vitro and in vivo models for deep fascia fibrosis. While we attempted to isolate primary fibroblasts from GMC fascia tissue using type I collagenase dissociation [77], cell adhesion was unstable within 24 h of culturing, and viability declined sharply within 48 h. To address this, we are currently developing a rat GMC model to support future research. Finally, the dense ECM of fascia tissue resulted in high background signals and weak specificity in IHC and IF staining, as have been reported by other groups [78]. We anticipate similar challenges with stRNA-seq until more expertise is gained. Although we visualized fibroblast-macrophage proximity using TEM, overcoming staining difficulties or employing experimental models will be crucial for future studies to better explore molecular mechanisms [79].

In conclusion, this study makes several key contributions. It is the first to dissect the cellular composition and molecular signals of GMC-associated contracture deep fascia at single-cell resolution. We demonstrate a widespread upregulation of collagens, proteoglycans, and ECM glycoproteins in contracture fascia, driven by pan-activation of various fibroblasts. Additionally, we identified two pro-fibrotic macrophage subpopulations: ECM-like MP, which directly contributes to ECM deposition via the canonical TGFβ pathway, and SPP1⁺ MP, which regulates fibroblast activation and ECM restructuring. By highlighting the critical roles of specific fibroblast and macrophage subclusters in fibrotic remodeling, this study suggests potential therapeutic strategies for GMC and opens new avenues for research into the mechanisms of fascial fibrosis and related fibrotic conditions. Future exploration could focus on how these cells contribute to fascial fibrosis, whether similar mechanisms are involved in related fibrotic muscle tissues, and whether these cellular processes offer valuable therapeutic targets for treating GMC and other myofascial disorders.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

DW: designed the study, supervised experiments, wrote the manuscript, and revised the manuscript. LL: designed the study, performed sample collection, and obtained informed consent from all subjects. WZ: conducted experiments, performed data analysis, and wrote the manuscript. ZL: processed samples, conducted experiments, and performed data analysis. WC: conducted TEM experiments and data analysis. ZW, SM, and LZ: performed data analysis.

Funding

This study was supported by the Science and Technology Innovation Program of Hunan Province, China (Grant Number: 2021SK1014) and the Postgraduate Scientific Research Innovation Project of Hunan Province, China (Grant Number: CX20210916). The funders had no role in the design of the study; in the collection, analyses, or interpretation of data; in the writing of the manuscript, or in the decision to publish the results.

Data availability

All data sets analyzed during this study are included in this published article. Human single-cell RNA-seq data was uploaded to the NCBI GEO with the code GSE273293. All original codes will be deposited at GitHub and be available online before publication.

Declarations

Ethics approval and consent to participate

The study was conducted according to the guidelines of the Declaration of Helsinki, and ethical approval for all research procedures was granted by the Ethics Committee at Changsha Central Hospital (Approval# 2022-S0064). Informed consent was obtained from all subjects prior to their participation.

Conflict of interest

The authors declare that they have no known competing interests.