1. Dataset selection and bulk dataset pre-processing

To investigate the transcriptional differences between diabetes patients with and without foot ulcers in foot (ulcer) skin or blood samples, we selected the bulk transcriptome dataset GSE134431 [8] and the single-cell transcriptome dataset GSE165816 [7] from Gene Expression Omnibus (GEO) repository. All data collection, processing, and sharing comply with the ethical requirements and privacy policies of the original data providers. Therefore, this study does not involve new ethical approval or participant consent. Due to ethical and moral considerations, the uploaders did not provide specific clinical information about the samples. After excluding irrelevant samples, GSE134431 included 8 foot skin samples from diabetes patients without foot ulcers, 7 foot skin (ulcer) samples from healing diabetic foot ulcers, and 6 foot skin (ulcer) samples from non-healing diabetic foot ulcers. GSE165816 included 8 foot skin samples from diabetes patients without foot ulcers, 2 peripheral blood samples from diabetes patients without foot ulcers, 9 foot skin (ulcer) samples from healing diabetic foot ulcers, 3 peripheral blood samples from healing diabetic foot ulcers, 5 foot skin (ulcer) samples from non-healing diabetic foot ulcers, and 2 peripheral blood samples from non-healing diabetic foot ulcers. To exclude genes with expression changes occurring during acute wound healing, we selected the acute wound healing-related gene chip dataset GSE28914 [9], which included 8 skin samples from day 0 of wound healing, 6 immediate post-biopsy skin samples (day 1), 6 skin samples from day 3 of wound healing, and 5 skin samples from day 7 of wound healing. Detailed information about the datasets can be found in S1 Table. The gene chip datasets or bulk sequencing datasets were normalized and subjected to PCA clustering analysis, as shown in S1A–S1D Fig.

2. Single-cell RNA sequencing data analysis

3. Analysis of cellular composition variation

The cell counts of each cell type in different groups (DS, DFUNH, DFUH) were determined, and the cell counts were divided by the total number of cells in the same group to calculate the cell type proportions. Based on these proportions, the percentages of specific cell types were calculated for each group. Then, the Log2FC (log fold change) between the DFUNH group and the DS group were calculated to identify cell types that changed in non-healing wounds of diabetes (|Log2FC| > 0.1). Similarly, the Log2FC between the DFUH group and the DFUNH group were calculated to identify cell types that changed in healing wounds (|Log2FC| > 0.1). Therefore, "rescue cell types" were defined as cell types that changed in non-healing wounds and were rescued in healing wounds. Additionally, the proportions of each cell type in each sample were calculated, and based on these proportions, the changes in cell type abundances between clinical groups were evaluated. A t-test was performed to assess the significance of cell type abundance changes between clinical groups (p-value < 0.05).

4. Analysis of cell-specific differentially expressed gene networks

The "FindMarkers" function in the Seurat package (version 4.3.0) was utilized to perform differential expression analysis of each cell type between different groups (DFUNH/DS and DFUH/DFUNH) using the Wilcoxon rank-sum test. Prior to conducting the differential expression analysis, cell types that were missing or had fewer than three cells in the comparison groups (DFUNH/DS and DFUH/DFUNH) were filtered out. Consequently, platelets, plasma-like dendritic cells, HSPCs (hematopoietic stem and progenitor cells), granulocytes, and granulocyte-monocyte progenitors were excluded from the peripheral blood. First, differentially expressed genes (DEGs) between the DFUNH and DS groups were identified, generating a dataset of DEGs associated with non-healing (unhealing DEGs) (|LogFC| > 0.25, adjusted p-value < 0.01). Subsequently, DEGs between the DFUH and DFUNH groups were identified, generating a dataset of DEGs associated with healing (healing DEGs) (|LogFC| > 0.25, adjusted p-value < 0.01). Based on the aforementioned results, "rescue DEGs" were defined as genes that were downregulated during the non-healing process and upregulated during the healing process (rescue upregulated DEGs), or genes that were upregulated during the non-healing process and downregulated during the healing process (rescue downregulated DEGs). Next, a cell-DEG network and a rose plot were constructed based on the aforementioned DEGs (unhealing DEGs, healing DEGs, and rescue DEGs) using single-cell RNA sequencing data, integrating the tissue and cell type sources of DEGs. The cell-DEG network was visualized using Cytoscape software (version 3.8.2) [14].

5. Short Time-series Expression Miner (STEM) analyses

Short Time-series Expression Miner (STEM) is a Java program used for clustering, comparing, and visualizing short time-series gene expression data from microarray experiments with 8 or fewer time points. In this study, the dynamic gene expression clusters associated with acute non-diabetic wound healing and diabetic wound healing were identified using the Short Time-series Expression Miner software (version 1.3.8) [15]. The dataset GSE28914 comprised four groups of samples at different time points during acute wound healing. For each group, the gene expression values were averaged to construct a matrix of mean expressions. Similarly, in the dataset GSE134431, three groups of skin wound samples in different states were used to construct a mean expression matrix using the same approach. The STEM clustering method was selected and other options were set as default. The gene expression profiles were clustered based on statistically significant values (P-value < 0.05). Genes that exhibit an increasing trend followed by a decreasing trend during the wound healing process are referred to as "rescue down" genes. On the other hand, genes that show a decreasing trend followed by an increasing trend are referred to as "rescue up" genes.

6. Gene Ontology (GO) and Protein-Protein Interaction (PPI) analysis

Gene Ontology (GO) and Protein-Protein Interaction (PPI) analyses were performed using the Metascape website. Default parameters were used for the analysis. For GO enrichment analysis, the minimum overlap was set to 3, the p-value cutoff was set to 0.01, and the minimum enrichment was set to 1.5. For PPI analysis, the minimum network size was set to 3, the maximum network size was set to 500, and physical core was used. The differential gene enrichment analysis among fibroblast subgroups was performed using the DAVID online website, and the enrichment results were visualized using R language.

7. Gene Set Enrichment Analysis (GSEA) analysis

Gene Set Enrichment Analysis (GSEA) is a common framework that integrates information from gene expression profiles into pathway or signature summaries. We used the GSEA method to estimate the activity of gene sets from bulk and single-cell RNA sequencing data. Specifically, for single-cell data, the "findmarkers" function was used to calculate differentially expressed genes between groups, with the filtering criteria set as adjusted p-value less than 0.05 and an average log2 fold change (avg_log2FC) greater than 0.5. The GSEA R package was applied to analyze the enrichment scores of the interested gene sets. Gene sets were considered significant when the adjusted p-value was less than 0.05 and the normalized enrichment score (NES) was greater than or equal to 1. For bulk data, the "limma" package in the online tool NetworkAnalyst was used to analyze the differentially expressed genes between the two groups (DFUNH vs DS, DFUH vs DFUNH). The GSEA R package was then used to analyze the enrichment scores of the interested gene sets, with the same filtering criteria as for the single-cell data. The gene sets were selected from the GSEA-msigDB database, specifically from c2.cp.v7.4.symbols.gmt (KEGG) and c5.go.bp.v7.4.symbols.gmt (GO-BP).

8. Cell-cell communication analysis

The cell-cell communication analysis was conducted using the CellChat [16] software package (version 1.1.0) in R language. Skin cells were annotated based on their cell types, including differentiated keratinocyte (DiffKera), basal keratinocyte (BasalKera), smooth muscle cells (SMCs), fibroblast (Fibro), vascular endothelial cells (VasEndo), T lymphocyte (T_lympho), natural killer cell (NK), M1 macrophage (M1), M2 macrophage (M2), melanocytes/schwann cell (Melano_Schwann), lymphatic endothelial cells (LymphEndo), B lymphocyte (B_lympho), plasma cell (Plasma_cell), mast cell (Mast_cell), and merkel cell (Merkel_cell). The "createCellChat" function was used to create a CellChat object. After annotating the objects with relevant labels and identifying the differentially expressed genes, the "computeCommunProb" function was used to infer the communication probabilities between cells. The "computeCommunProbPathway" function was used to generate intercellular communication for each cell signaling pathway. The "netVisual_chord_gene" function was used to generate the visualization. The "subsetCommunication" function was employed to determine the upregulated or downregulated ligand-receptor (LR) pairs in the cell communication network. The filtering criteria were as follows: upregulated LR: ligand.logFC > = 0.2, receptor.logFC > = 0.2; downregulated LR: ligand.logFC < = -0.2, receptor.logFC < = -0.2.

9. Single-cell trajectory analysis and CytoTRACE analysis

In this study, we employed the Monocle 3 [17] algorithm to perform single-cell trajectory analysis and detected gene expression changes in fibroblast subtypes during cellular transitions. Monocle 3 utilizes a single-cell trajectory analysis strategy that leverages algorithms to learn gene expression changes occurring in each cell during dynamic biological processes. By constructing a global "trajectory" of gene expression changes, each cell can be placed in the appropriate position within a reduced-dimensional space for trajectory analysis. Additionally, CytoTRACE [18] analysis was performed to predict the differentiation status of cells in the single-cell RNA sequencing data, enabling the identification of cell subtypes located at the beginning of the cell trajectory.

10. Selection of differentially expressed genes in fibroblast subpopulations

The FindMarker function was used to identify differentially expressed genes between the "good" and "bad" cells. The selection criteria were an absolute log fold change (logFC) greater than or equal to 0.25, an adjusted p-value less than 0.01, and the Wilcoxon test was employed.

11. Curve analysis of Receiver Operating Characteristics (ROC)

We used the pROC function in the R package to create Receiver Operating Characteristic (ROC) curves to determine the area under the curve (AUC) for screening signature genes and evaluating their diagnostic value. The potential of its role as a molecular biomarker was evaluated based on the value of the AUC [19].

1. Construction of a single-cell atlas of diabetic foot ulcers

In order to identify local and systemic factors associated with DFUs healing, we performed scRNA-Seq analysis on foot skin samples (n = 8) and peripheral blood samples (n = 2) from patients with diabetes without foot ulcers, foot ulcer samples (n = 9) and peripheral blood samples (n = 3) from patients with healed DFUs, and foot ulcer samples (n = 5) and peripheral blood samples (n = 2) from patients with non-healed DFUs. After quality control (S2A and S2B Fig), we analyzed 52335 skin cells and 23567 peripheral blood cells, respectively, and created a gene expression matrix for each cell. We used these matrixs to perform dimensionality reduction via UMAP and graph-based clustering, identifying 29 and 17 orthogonal cell clusters in the skin and peripheral blood, respectively (S3A, S3B, S3E and S3F Fig). The expression of established cell-specific marker genes aided in annotating these cell clusters into 15 and 13 distinct cell types, respectively (Fig 1A and 1F and S3C, S3D, S3G and S3H Fig). Typical cell types in the skin included smooth muscle cells (SMCs) (TAGLN⁺, ACTA2⁺), fibroblasts (Fibro) (DCN⁺, CFD⁺), vascular endothelial cells (VasEndo) (ACKR1⁺), T lymphocytes (T_lympho) (CD3D⁺), differentiated keratinocytes (Diffkera) (KRT1⁺, KRT10⁺), basal keratinocytes (BasalKera) (KRT5⁺, KRT14⁺), natural killer cells (NK) (CCL5⁺, GZMB⁺), M1 macrophages (M1) (IL1B⁺), M2 macrophages (M2) (CD163⁺), melanocytes and Schwann cells (Melano_Schwann) (MLANA⁺, CDH19⁺), lymphatic endothelial cells (LymphEndo) (CCL21⁺), B lymphocytes (B_lympho) (CD79A⁺, MS4A1⁺), plasma cells (plasma_cell) (MZB1⁺), mast cells (Mast_cell) (TPSAB1⁺) and Merkel cells (Merkel_cell) (KRT18⁺) (Fig 1F and S3C, S3D Fig). Typical cell types in peripheral blood included natural killer cells (NK) (CCL5⁺, GZMB⁺), natural killer T cells (NKT) (CD3D⁺, CCL5⁺), Erythrocytes (Erythro) (HBB⁺), dendritic cells (DCs) (GZMB⁺, IRF8⁺), T lymphocytes (T_lympho) (CD3D⁺), CD14⁺ monocytes (CD14_Mono) (CD14⁺, S100A9⁺), CD16⁺ monocytes (CD16_Mono) (FCGR3A⁺, MS4A7⁺), B lymphocytes (B_lympho) (CD79A⁺, MS4A1⁺), plasmacytoid dendritic cells (pDCs) (LILRA4⁺), platelets (ITGA2B⁺, ITGB3⁺, SELP⁺, PPBP⁺), granulocytes (CD63⁺, ENPP3⁺, CEACAM8⁺), granulocyte-monocyte progenitors (ADK⁺, ALDH4A1⁺, ANXA1⁺), and hematopoietic stem and progenitor cells (HSPCs) (CD34⁺) (Fig 1A and S3G, S3H Fig). Comparison and statistical analysis of cell type abundance demonstrated significant differences in the concentrations between different clinical groups and each sample (Fig 1B–1E and 1G–1J). Statistical analysis of cell abundance revealed significant differences in Fibro in skin tissue and plasmacytoid dendritic cells and T cells in blood between clinical groups (S4A and S4B Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Single-cell RNA sequencing revealed identification and characterization of cell types in diabetic foot ulcers (DFUs) skin and peripheral blood.

(A, F) Uniform Manifold Approximation and Projection (UMAP) embeddings generated from datasets containing 52335 cells (A) and 23567 cells (F), respectively. Cells are colored based on orthogonal-generated clusters and annotated manually according to their cell types (Skin cell types: smooth muscle cells, SMCs; fibroblasts, Fibro; vascular endothelial cells, VasEndo; T lymphocytes, T_lympho; differentiated keratinocytes, Diffkera; basal keratinocytes, BasalKera; natural killer cells, NK; M1 macrophages, M1; M2 macrophages, M2; melanocytes and Schwann cells, Melano_Schwann; lymphatic endothelial cells, LymphEndo; B lymphocytes, B_lympho; plasma cells, Plasma_cell; mast cells, Mast_cell; and Merkel cells, merkel_cell. Typical cell types in peripheral blood: natural killer cells, NK; natural killer T cells, NKT; Erythrocytes, Erythro; dendritic cells, DC; T lymphocytes, T_lympho; CD14⁺ monocytes, CD14_Mono; CD16⁺ monocytes, CD16_Mono; B lymphocytes, B_lympho; plasmacytoid dendritic cells, pDCs; platelets; granulocytes; granulocyte-monocyte progenitors, GMPs; and hematopoietic stem and progenitor cells, HSPCs). (B, G) UMAP embeddings of peripheral blood cells (B) and skin cells (G), colored based on clinical group, from all analyzed samples. (C, H) UMAP embeddings of peripheral blood cells (C) and skin cells (H), colored based on sample source, from all analyzed samples. (D, I) The stacked bar charts display the proportions of different cell types in the three clinical groups of peripheral blood (D) and skin (I). Light blue: DFU-Healers, dark blue: DFU-non-Healers, purple-red: patients with diabetes without foot ulcers. Cell types with significant differences between clinical groups are marked with an asterisk. The differences in proportions of each cell type in each group are shown in S4 Fig. (E, J) The stacked bar charts display the proportions of different cell types in each analyzed sample of peripheral blood (E) and skin (J). Different colors represent different cell types.

https://doi.org/10.1371/journal.pone.0306248.g001

2. Reconstruction of the cellular ecosystem for diabetic wound healing

The cells may exhibit continuous variations or minor differences between different clinical groups, which may be overlooked at a statistically significant level. To further delineate the dynamic and subtle changes in cell composition in diabetic skin, diabetic non-healing wounds (DFUNH), and diabetic healing wounds (DFUH), we compared the proportional changes of each cell type in skin wound tissues and peripheral blood between the DS, DFUNH, and DFUH groups, respectively. Fig 2 demonstrates the cell types that exhibit changes in proportion in non-healing wounds and undergo reversal in healing wounds (see Methods). Overall, we observed changes in the proportions of various cell types in the DFUNH group compared to the DS group, many of which were rescued in the DFUH group (Fig 2A–2D). For example, the proportions of T lymphocytes, plasma cells, melanocytes, mast cells, M2 cells in skin and T lymphocytes in peripheral blood increased in the DFUNH group, while they decreased in the DFUH group. Similarly, the proportions of fibroblasts in skin and CD14 monocytes, CD16 monocytes, hematopoietic stem and progenitor cells (HSPCs), plasmacytoid dendritic cells, and platelets in periheral blood decreased in the DFUNH group, while they increased in DFUH group (Fig 2A–2D). These changes in cell proportions may have negative or positive effects on wound healing. Fibroblasts are an important cell type in wound healing, involved in different stages of the healing process. They produce collagen and other matrix proteins during the healing process, providing structural support and the necessary matrix for wound repair. Fibroblasts are also involved in angiogenesis, promoting the formation of new blood vessels by secreting angiogenic factors, thus improving the blood supply to the wound site [20]. However, diabetes affects the function and activity of fibroblasts. Under high glucose conditions, the proliferation and migration abilities of fibroblasts decrease, and the quality of collagen production is reduced [4]. The specific role of plasmacytoid dendritic cells (pDCs) in the wound healing process in diabetes is not yet clear. pDCs play an important role in immune responses, primarily involved in the immune response to viral infections and regulating the immune system by producing interferon-alpha (IFN-alpha) in large quantities [21]. We speculate that pDCs may affect the healing of diabetic wounds through the aforementioned mechanisms. In a thymectomized rat model, depleted CD8 T cells have a positive impact on wound healing, while CD4 T cell depletion has a negative effect [22]. To explore the differences in CD4+ and CD8+ T cells between the DS, DFUNH, and DFUH groups, we further divided peripheral blood T cells into CD4+ and CD8+ subgroups and analyzed the changes in the composition of CD4+ and CD8+ cells among the three groups. Our study results showed a downregulation of CD4+ T cells and an upregulation of CD8+ T cells in DFUNH, although not statistically significant (S5A–S5D Fig). This suggests that the depletion of CD8+ T cells may be beneficial for the healing of diabetic wounds. Plasma cells are abundant in DFUs non-healing wounds, indicating a possible association with B lymphocyte differentiation and non-healing wounds. Generally, M1 macrophages promote inflammatory responses, while M2 macrophages primarily inhibit inflammation and promote angiogenesis. Our results showed an increase in the proportion of M2 cells in DFUNH, while a decrease in proportion was observed in DFUH, consistent with the study by Theocharidis et al. (Fig 2C) [7]. Compared to the DS group, the M1/M2 ratio decreased in the DFUNH group, and compared to the DFUNH group, this ratio significantly increased in the DFUH group (Fig 2F). Regarding monocytes in the blood, the classical phenotype (CD14+ monocytes) is similar to the pro-inflammatory phenotype described earlier, while the non-classical phenotype (CD16+ monocytes) is similar to anti-inflammatory monocytes [23]. Similar to macrophages, the CD14/CD16 ratio decreased in the DFUNH group compared to the DS group, and this ratio significantly increased in the DFUH group compared to the DFUNH group (Fig 2E). Furthermore, a lower negative regulation score for local inflammation response was observed in DFUH wound sites, while a higher negative regulation score for systemic inflammation response was observed in peripheral blood, indicating local inflammatory activation and systemic inflammation suppression in healing wounds (Fig 2G and 2H). These results suggest that a moderate local inflammatory response is necessary for the healing of diabetic wounds.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Dynamic composition changes of cell types during the diabetic wound healing process.

(A-B) UMAP plots showing the composition changes of various cell types in the skin (A) and peripheral blood (B) of the DS, DFUNH, and DFUH groups. (C-D) Relative changes in cell proportions among the three groups in skin (C) and periheral blood (D). The numbers on the right indicate the logarithmic fold change in cell proportions (DFUNH/DS and DFUH/DFUNH). The cell types marked in red are downregulated in non-healing wounds and upregulated in healing wounds, while the cell types marked in blue are upregulated in non-healing wounds and downregulated in healing wounds. Gray indicates cell types with no significant proportional changes. (E) UMAP plot displaying the changes in CD14 /CD16 monocyte cell proportions. (F) UMAP plot showing the changes in M1/M2 macrophage proportions. (G-H) Differences in the scores of negative regulation of inflammatory response in local wound and peripheral blood among the three clinical groups. * indicates p<0.05. **** indicates p<0.0001.

https://doi.org/10.1371/journal.pone.0306248.g002

3. Cross-cell expression changes of reversed non-healing associated genes in healing wounds

In order to elucidate the molecular events associated with healing and non-healing, we identified thousands of DEGs between DFUNH and DS, as well as between DFUH and DFUNH, which we referred to as "non-healing-related DEGs" and "healing-related DEGs," respectively. Through further comprehensive comparative analysis of these differentially expressed genes (DEGs), key DEGs that exhibited opposite expression trends in healing wounds and non-healing wounds were identified, and therefore referred to as "rescue DEGs" (Figs 3A–3D and 4A–4D; see Methods section for details).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Screening of unhealing genes, healing genes, and rescue genes in skin.

(A, C) Heatmaps display the distribution of DEGs in the skin across different cell types. Each column represents a cell type, and each row represents a gene. Red indicates genes that are upregulated in DFUNH compared to DS, as well as genes that are upregulated in DFUH compared to DFUNH (LogFC > 0.5, adjusted p-value < 0.05). Blue indicates genes that are downregulated in DFUNH compared to DS, as well as genes that are downregulated in DFUH compared to DFUNH (LogFC < -0.5, adjusted p-value < 0.05). Gray indicates no significant change (|LogFC| < 0.5). Unhealing DEGs refer to genes that show changes in the DFUNH group compared to the DS group, while healing DEGs refer to genes that show changes in the DFUH group compared to the DFUNH group. Rescue DEGs are genes that exhibit opposite changes in expression compared to unhealing DEGs within healing DEGs. (B, D) Venn diagrams display the number of unhealing, healing, and rescue DEGs. The overlapping regions represent the number of downregulated rescue DEGs (B) and upregulated rescue DEGs (D). (E, F) The rose plots display the number and cell type distribution of unhealing DEGs, healing DEGs (E), and rescue DEGs (F) in the skin.

https://doi.org/10.1371/journal.pone.0306248.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Screening of unhealing genes, healing genes, and rescue genes in blood.

(A, C) Heatmaps display the distribution of DEGs in the peripheral blood across different cell types. Each column represents a cell type, and each row represents a gene. Red indicates genes that are upregulated in DFUNH compared to DS, as well as genes that are upregulated in DFUH compared to DFUNH (LogFC > 0.5, adjusted p-value < 0.05). Blue indicates genes that are downregulated in DFUNH compared to DS, as well as genes that are downregulated in DFUH compared to DFUNH (LogFC < -0.5, adjusted p-value < 0.05). Gray indicates no significant change (|LogFC| < 0.5). Unhealing DEGs refer to genes that show changes in the DFUNH group compared to the DS group, while healing DEGs refer to genes that show changes in the DFUH group compared to the DFUNH group. Rescue DEGs are genes that exhibit opposite changes in expression compared to unhealing DEGs within healing DEGs. (B, D) Venn diagrams display the number of unhealing, healing, and rescue DEGs. The overlapping regions represent the number of downregulated rescue DEGs (B) and upregulated rescue DEGs (D). (E, F) The rose plots display the number and cell type distribution of unhealing DEGs, healing DEGs (E), and rescue DEGs (F) in the peripheral blood.

https://doi.org/10.1371/journal.pone.0306248.g004

Next, to distinguish the distribution of non-healing DEGs and healing DEGs in each cell type, we attributed the non-healing, healing, and rescue DEGs to each cell type. Clearly, as shown in the rose plot and cell-gene network plot, non-healing-related and healing-related genes are predominantly distributed in keratinocytes (both differentiated and undifferentiated), fibroblasts, M1, and M2 macrophages in the skin tissue (Fig 3E and S6A and S6B Fig). Similarly, rescue DEGs are also primarily distributed in these cell types (Fig 3F and S6C Fig). This suggests that keratinocytes, fibroblasts, M1, and M2 are the main responsive cell types in the wound healing process and play a significant role. In peripheral blood, non-healing DEGs were mainly distributed in erythrocytes, B cells, NKT cells, CD14, and CD16 cells, while healing-related DEGs were mainly distributed in DCs (Fig 4E and S6D and S6E Fig). Rescue DEGs were primarily distributed in B cells, CD14 monocytes, and erythrocytes (Fig 4F and S6F Fig). This indicates that B cells, CD14 monocytes, and erythrocytes are the main blood cell types involved in the response during wound healing. These analyses highlight the cell specificity of unhealing and healing DEG distributions and suggest that a multi-cell type comparison may provide a more comprehensive portrait for a better understanding of the mechanisms underlying unhealing and healing.

To further narrow down the scope of DEGs and exclude genes with similar trends of changes during acute wound healing, we performed a combined analysis of the selected DEGs with the bulk transcriptome datasets GSE134431 (diabetes wound-related) and GSE28914 (acute wound healing-related). Firstly, using STEM software, we obtained the rescue up and rescue down genes in GSE134431 and GSE28914, respectively (see Methods). Specifically, genes that followed an increasing and then decreasing trend in expression throughout the entire wound healing process were referred to as rescue down genes, while genes that exhibited a decreasing and then increasing trend were referred to as rescue up genes. As shown in Fig 5A, genes contained in clusters 30, 31, 32, and 49 of the GSE28914 dataset were rescue down genes, while clusters 21, 22, 7, and 17 contained rescue up genes. Similarly, in the GSE134431 dataset, clusters 1, 6, and 5 contained rescue up genes, while clusters 14, 10, and 9 contained rescue down genes (Fig 5B). First, we took the intersection of rescue up and rescue down genes in the skin tissue from GSE165816 and rescue up and rescue down genes in GSE134431, resulting in 54 common rescue down genes and 149 common rescue up genes (Fig 5C and 5D). Then, we took the intersection of the aforementioned common genes with the rescue down and rescue up genes during acute wound healing to exclude genes that showed changes in the acute wound healing process, resulting in 54 unique rescue down genes and 146 unique rescue up genes in diabetes wounds (Fig 5E and 5F). The changes in these genes were considered specific to the diabetes wound healing process.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Identification of genes exhibiting specific changes during the wound healing process.

(A) Time kinetics of gene expression during acute non-diabetic wound healing. The black line represents the overall trend in each profile, and color-coded profiles indicate statistically significant differences (P < 0.05). Profiles of the same color represent clusters with similar expression trends. (B) Time kinetics of gene expression during diabetic wound healing. The black line represents the overall trend in each profile, and color-coded profiles indicate statistically significant differences (P < 0.05). Profiles of the same color represent clusters with similar expression trends. (C-D) Common rescue down (C) or rescue up (D) genes in bulk transcriptome (GSE134431) and single-cell transcriptome (GSE165816) of diabetic wounds. (E-F) Intersection of rescue down or rescue up genes common to diabetic wounds and rescue down/up genes during acute wound healing, to exclude intersecting genes and obtain uniquely expressed rescue down (E) or rescue up (F) genes during diabetic wound healing. (G-H) Venn diagram showing the number of shared upregulated (H) or downregulated (G) rescue DEGs in the skin and peripheral blood.

https://doi.org/10.1371/journal.pone.0306248.g005

Next, we explored the biological significance of these unique rescue DEGs in diabetes wounds using Gene Ontology (GO) and pathway analysis (http://metascape.org/). For rescue down DEGs, enrichment was observed in categories such as skin development, IL-18 signaling pathway, establishment of skin barrier, PIDCMYB PATHWAY, cellular response to cytokine stimulus, cellular response to hydrogen peroxide, Glucocorticoid receptor pathway, regulation of epithelial cell proliferation, muscle tissue development, regulation of angiogenesis, and response to growth factor (Fig 6A). For rescue up DEGs, enrichment was observed in categories such as cellular response to cytokine stimulus, antigen processing and presentation of peptide antigen, innate immune response, response to interleukin-7, antigen processing and presentation of endogenous antigen, neutrophil degranulation, regulation of hematopoiesis, negative regulation of ubiquitin protein ligase activity, negative regulation of cell population proliferation, degradation of the extracellular matrix, positive regulation of translation, VEGFA-VEGFR2 signaling pathway, positive regulation of apoptotic process, response to interferon-beta, and burn (Fig 6B). Further PPI analysis revealed that the core network functions constructed by these rescue down and rescue up genes were mainly associated with antigen processing and presentation, interferon signaling, and intermediate filament organization (Fig 6C). The gene expression scores of the GO terms "antigen processing and presentation" and "response to interferon beta" in the three clinical groups were further determined using the AddModuleScore function, and the results were consistent with the above findings (S7A and S7B Fig). Antigen-presenting cells, such as macrophages and dendritic cells, present and process antigens from the wound environment and migrate to lymph nodes, where they present the processed antigens to T cells. This leads to the activation of specific immune responses against the antigens and helps clear pathogens or abnormal cells present in the wound [24]. Impairment of antigen processing and presentation in DFUNH may result in clearance defects of pathogens and other antigens, leading to the sustained non-healing state of the wound. Interferon signaling plays an important regulatory role in antigen presentation and immune response by regulating the activation, maturation, and MHC molecule expression of APC cells, as well as directly affecting the functions of other immune cells [25, 26]. Studies have shown that the absence of IFN-γ and IFN-κ impairs wound healing in diabetes wounds [27, 28]. To identify the shared rescue DEGs between skin tissue and peripheral blood, we created Venn diagrams of rescue DEGs in different tissues. The results showed that six downregulated unhealing-related genes (HLA-C, CD74, HLA-DRA, HLA-B, HLA-DQA, TUBA1B) were reactivated in healing wounds in both skin and peripheral blood (Fig 5G and 5H). These DEGs were downregulated in non-healing wounds and restored in healing wounds. HLA-B and HLA-C belong to the MHC class I molecules, while HLA-DRA and HLA-DQA belong to the MHC class II molecules. HLA-II molecules can enhance the binding and presentation capacity of antigen-presenting lymphocytes, mediating the actions of other immune components [29]. These molecules also act as co-stimulatory molecules, enhancing inflammation and mediating the production of cytokines such as interferons [30], which are associated with increased collagen deposition [31]. The systemic (in blood) downregulation of the above-mentioned HLA molecules may be one of the reasons for impaired antigen processing and presentation in DFUNH. In addition, we evaluated the potential diagnostic value of these 6 genes in the bulk dataset GSE134431. Among them, CD74, HLA-B, and HLA-DRA showed AUC values greater than 0.7, indicating that these three genes have the potential to be biomarkers for non-healing DFUs (S8 Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. The enrichment analysis of diabetes-specific rescue genes.

(A-B) GO enrichment analysis of the aforementioned uniquely expressed rescue down (A) and rescue up (B) genes. (C) PPI network constructed by the common rescue down and rescue up genes and the core module within the network. Different colors represent different core modules.

https://doi.org/10.1371/journal.pone.0306248.g006

4. Disrupted ECM signaling in non-healing diabetic foot ulcers

The extracellular matrix (ECM) is a complex network of macromolecules secreted by cells into the extracellular space, accounting for more than one-third of the body’s mass. ECM components not only provide dynamic tissue integrity but also serve as signaling molecules involved in driving numerous biological reactions, making their dysregulation a direct or indirect cause of most chronic diseases [32]. From an immunological perspective, the ECM also includes various secreted proteins, including cytokines, chemokines, and growth factors, which potentially participate in immune cell regulation [33]. The results presented herein demonstrate a decrease in the proportion of fibroblasts, the primary cell type responsible for synthesizing and secreting ECM major components, in non-healing diabetic foot ulcers (DFUNH). To further investigate the differential ECM-related signaling activities in DS, DFUNH, and DFUH group, we utilized the AddModuleScore function to evaluate the gene expression scores of ECM-receptor signaling pathways among different clinical groups. The results showed elevated ECM degradation and assembly activities in both the DFUH and DS groups compared to the DFUNH group, with higher ECM-receptor signaling pathway scores observed in the DFUH and DS groups compared to the DFUNH group (Fig 7A–7C). These findings suggest an inactivation of the ECM-receptor signaling pathway in DFUNH. Moreover, the enrichment analysis of GSE165816 and bulk transcriptome data confirmed the downregulation of ECM-receptor signaling levels in DFUNH (Fig 7D, 7E, 7G and 7H). Additionally, fibroblasts were identified as the cell type with the highest ECM receptor signaling score (Fig 7F), indicating their dominant role in ECM-receptor signaling.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Activity of ECM assembly and degradation, as well as ECM-receptor signaling, in the skin tissues of three clinical groups.

(A) Gene expression scores of ECM degradation signals in the single-cell datasets DS, DFUNH, and DFUH groups. ****, P<0.05. (B) Gene expression scores of ECM assembly signals in the single-cell datasets DS, DFUNH, and DFUH groups. ****, P<0.05. (C) Gene expression scores of ECM-receptor interaction signals in the single-cell datasets DS, DFUNH, and DFUH groups. ****, P<0.05. (D) Enrichment results of ECM-receptor interaction signals in DFUNH compared to DS. Adjusted p <0.05, |NES|> = 1. Data source: GSE165816. (E) Enrichment results of ECM-receptor interaction signals in DFUH compared to DFUNH. Adjusted p <0.05, |NES|> = 1. Data source: GSE165816. (F) Gene expression scores of ECM-receptor interaction signals in various skin cell types. Data source: GSE165816. The dashed line represents the baseline level of base mean. (G) Enrichment results of ECM-receptor interaction signals in DFUNH compared to DS. Adjusted p <0.05, |NES|> = 1. Data source: GSE134431. (H) Enrichment results of ECM-receptor interaction signals in DFUH compared to DFUNH. Adjusted p <0.05, |NES|> = 1. Data source: GSE134431.

https://doi.org/10.1371/journal.pone.0306248.g007

To further explore the molecular events associated with ECM-related signaling during the wound healing process in diabetic ulcers, the distribution of ECM-receptor signaling molecules in the skin tissues of DS, DFUNH, and DFUH groups was analyzed using Cellchat. The cell-cell communication network depicted by the computational analysis revealed enhanced or diminished ECM-receptor signaling between DS and DFUNH, as well as between DFUNH and DFUH (Fig 8A and 8B, S9A and S9B Fig). Similarly, rescue up and rescue down ligand-receptor pairs were obtained using a similar approach for ECM-receptor signaling. Rescue up ligand-receptor pairs refer to those that are downregulated in DFUNH (compared to DS) and upregulated in DFUH (compared to DFUNH). Rescue down ligand-receptor pairs refer to those that are upregulated in DFUNH (compared to DS) and downregulated in DFUH (compared to DFUNH) (see methods for details). Cell communication analysis revealed a significant reduction in COLLAGEN signaling in DFUNH tissues compared to DS, primarily involving COL1 and COL6 ligands and their corresponding receptors, along with diminished FN1, THBS, and TENASCIN signaling. These signals were largely restored in DFUH (S3 Table). Collagen is a major structural component of the extracellular matrix, providing support and structure to various tissues in the body [34]. Studies have shown that the effective domains of recombinant human COL1A1-collagen can induce cell proliferation and collagen synthesis in human dermal fibroblasts, as well as enhance cell migration and elastin production. Peptides derived from human COL1A2 have also been shown to promote wound healing and elastin production. Human collagen I alpha-2-derived peptide can improve the synthesis of type I collagen, cell proliferation, cell migration, and elastin synthesis [35]. Additionally, fragments of type I collagen can serve as potent chemotactic agents for neutrophils, enhance phagocytosis and immune response, and regulate gene expression, making them inflammatory mediators [36, 37]. They can also effectively stimulate angiogenesis in vitro and in vivo through the involvement of specific integrin receptors [38]. Type VI collagen, a non-fibrillar collagen, is expressed in many connective tissues and participates in the formation of the ECM. It interacts with various key ECM components, including type I and type II fibrillar collagens, basement membrane type IV collagens, fibronectin, and others. In connective tissue ECM, type VI collagen forms highly branched fibrillar networks surrounding the main fibrous collagens I, II, and III. Additionally, through its interaction with basement membrane type IV collagen, it anchors blood vessels, nerves, and mesenchymal cells. It also serves as a reservoir for platelet-derived growth factors, keratinocyte growth factor, matrix metalloproteinases-1, -2, -3, -8, and -9, IL-2, and cytokine thymic stromal lymphopoietin, regulating their activity and availability [39]. Type VI collagen acts as an early sensor for injury/repair responses and can regulate fibrogenesis by modulating cell-cell interactions, stimulating mesenchymal cell proliferation, and preventing apoptosis [40].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 8. Changes in ligand-receptor interactions in ECM-receptor signaling between different cell types between the DS and DFUNH groups, as well as between the DFUNH and DFUH groups.

(A-B) Network diagrams depict the changes in ligand-receptor interactions between different cell types in the skin DFUNH/DS (A) and DFUH/DFUNH (B) comparison groups. Communication between cells is represented by connecting lines. The thickness of the lines is positively correlated with the number or intensity of ligand-receptor interaction events. Red or blue color indicates an increase or decrease in signaling in the second dataset (DFUNH or DFUH) compared to the first dataset (DS or DFUNH). (C-D) Extracted ligand-receptor pairs included in Fibro, BasalKera, DiffKera, M1, M2, and VasEndo cells for "rescue up" (C) and "rescue down" (D). The starting point of an arrow represents the cell containing the ligand, while the endpoint of the arrow points to the cell type containing the receptor targeted by that ligand. "Rescue up" and "rescue down" are indicated by red and blue arrows, respectively.

https://doi.org/10.1371/journal.pone.0306248.g008

It is worth noting that the restored COL1A1, COL1A2, COL6A1, and COL6A2 in DFUH mainly originate from M1 and M2 cells, while their corresponding receptors ITGA2, ITGB1, ITGA3, ITGA9, SDC1, and SDC4 are predominantly present in BasalKera and DiffKera cells (Fig 8C and S3 Table). This suggests that weakened cell communication between the COL1 and COL6 ligands derived from M1 and M2 cells and their corresponding receptors from BasalKera and DiffKera cells may be an important factor contributing to the non-healing of wounds. However, there is also a small subset of COLLAGEN signaling (mainly COL1 and COL6) between specific cell types that increases in DFUNH and decreases in DFUH (rescue down) (Fig 8D and S4 Table). Since the majority of COLLAGEN signaling is diminished in DFUNH, these increased COLLAGEN signals in DFUNH can be disregarded.

Furthermore, we observed an imbalance in THBS signaling in DFUNH. Specifically, THBS1 is mainly present in the rescue up signal, while THBS2 is mainly present in the rescue down signal. In the rescue up signal, the THBS1 ligand, derived primarily from Fibro, corresponds to receptors from various cell types such as BasalKera, LymphoEndo, M2, DiffKera, and Merkel cells (S3 Table). In the rescue down signal, the THBS2 ligand is predominantly derived from BasalKera, and its corresponding receptors are distributed in M1, M2, Fibro, and DiffKera. THBS1 and THBS2 seem to play distinct and non-overlapping roles in skin wound healing. Studies have shown that THBS1 knockout mice exhibit delayed wound healing, prolonged inflammation, and reduced TGF-β in the wound bed [41, 42]. This indicates the positive role of THBS1 in wound healing. In contrast, THBS2 knockout mice show accelerated full-thickness wound healing, enhanced angiogenesis, and changes in the remodeling phase of healing [43, 44]. This suggests the negative role of THBS2 in wound healing. Additionally, LAMININ signaling is only present in the rescue down signal and absent in the rescue up signal. The LAMA4 ligand mainly originates from Fibro, while its corresponding receptors ITGA6, ITGB1, ITGB4 are located in LymphoEndo and VasEndo. Similarly, FN1 signaling is exclusively found in the rescue up signal. The FN1 ligand is primarily derived from M2 and VasEndo, and its corresponding receptors ITGA3, ITGA5, ITGB1, SDC1, and SDC4 are distributed in BasalKera, VasEndo, DiffKera, and Merkel cells. FN1 plays a crucial role in wound healing by promoting cell adhesion, migration, angiogenesis, and providing a scaffold for tissue regeneration [45–47]. Its interaction with various cell types and growth factors helps coordinate effective repair of damaged tissue. The loss of FN1 signaling in DFUNH may be an important factor contributing to the non-healing of wounds [48–50].

The ligand-receptor pairs TNXB-SDC4 are upregulated in DFUNH and downregulated in DFUH, while the pairs TNC-SDC1 and TNC-SDC4 are downregulated in DFUNH and upregulated in DFUH. This suggests that TNC and TNXB may play different biological roles in DFUNH. During the inflammatory phase, TNC can induce the synthesis of pro-inflammatory cytokines through TLR4. In the proliferative phase, the significant increase in TNC promotes angiogenesis [51]. Similarly, we evaluated the potential diagnostic value of COL1A1, COL1A2, COL6A1, COL6A2, FN1, THBS1, THBS2, TNC, and TNXB genes in the bulk dataset GSE134431. Among them, COL6A1, COL6A2, and FN1 showed AUC values greater than 0.7, indicating that these three genes also have the potential to be biomarkers for non-healing DFUs (S8 Fig).

5. MMPs and TIMPs are widely downregulated in DFUNH

Matrix metalloproteinases (MMPs) are a family of zinc-dependent extracellular matrix (ECM) remodeling enzymes that can degrade almost all components of the ECM. Changes in the expression of MMPs lead to abnormal degradation of the ECM, which is the initial cause of chronic degenerative diseases and the development of vascular complications in diabetes. Tissue inhibitors of matrix metalloproteinases (TIMPs) within the ECM inhibit the proteolytic activity of MMPs. TIMPs are important regulatory factors in ECM metabolism, tissue remodeling, and cellular behavior. Therefore, similar to MMPs, TIMPs regulate angiogenesis, cell proliferation, and cell apoptosis. The regulatory roles of MMPs and TIMPs can affect the interaction and signaling between cells and the ECM, thereby regulating the activity and effects of ECM receptor signaling pathways [52, 53]. If the strict balance between MMP activity and inhibition is disrupted, wounds can progress to a state of increased ECM degradation, altered cytokine profiles, and degradation of growth factors, ultimately leading to delayed or impaired wound closure [54, 55]. We analyzed the expression of 21 human MMPs and 4 TIMPs in DFUs skin and blood. As shown in Fig 9, the expression levels of MMP2, MMP14, TIMP1, and TIMP2 were decreased in the skin and blood of the DFUNH group compared to the DS group, while their expression was restored in the DFUH group compared to the DFUNH group (Fig 9A–9F). MMP2 and MMP14 are mainly expressed in Fibro (S10C and S10D Fig), TIMP1 is mainly expressed in Fibro, LymphoEndo, M1, M2, Mast cell, and SMCs, and TIMP2 is mainly expressed in Fibro, LymphoEndo, M1, M2, Mast cell, Melano_schwann, and SMCs (S10A and S10B Fig). In fibroblasts, TIMP1 and MMP14 were downregulated in DFUNH compared to DS, while their expression was restored in DFUH compared to DFUNH (S10A and S10D Fig). TIMP2 and MMP2 were both highly expressed in DFUNH and DFUH compared to the DS group, but their expression was higher in DFUH (S10B and S10C Fig).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 9. Expression patterns of MMPs and TIMPs in three clinical groups.

(A-B) Dot plots depicting the expression of MMP and TIMP gene families in the skin (A) and peripheral blood (B) of the three clinical groups. The size of the dots represents the percentage of gene expression in different clinical groups, while the color represents the average expression level of the genes; dark blue: low, yellow: high. (C-F) Overall expression patterns of MMP2, MMP14, TIMP1, and TIMP2 in the skin tissue of the three clinical groups. ***, P < 0.001; ****, P < 0.0001.

https://doi.org/10.1371/journal.pone.0306248.g009

Studies have shown that MMP2 accelerates cell migration. MMP2 regulates angiogenesis in the wound healing process by activating pro-angiogenic cytokines, including TNF-α and VEGF, and by producing anti-angiogenic peptides, such as endostatin derived from type XVII collagen, expressed in the basement membrane [56, 57]. However, there are also studies suggesting that the activity of MMP2 is several-fold higher in non-healing wounds compared to healing wounds, and inhibiting MMP2 contributes to promoting the healing process [58]. MMP14 activates MMP-2 by degrading pro-MMP-2 [59]. The absence of MMP14 results in defective turnover of type I collagen and impaired activation of MMP-2 [60, 61]. In addition, MMP14 regulates epithelial cell proliferation in the wound healing process by altering the expression of KGF receptors [62]. Besides controlling proliferation, MMP14 also promotes epithelial cell migration in vitro by cleaving syndecan-1, CD44, and laminin-332 [63–65]. MMP14 knockout mice exhibit early onset and increased mortality rates [66]. TIMP-1 is present in the epithelial cells of healing excision and burn wounds, and it is also expressed in wound fibroblasts, particularly in fibroblasts near human blood vessels [67, 68]. Evidence suggests that skin injuries in diabetes lead to increased production of TIMP-1 and type I and III procollagen in diabetic skin animal models. Proper levels of active TIMP-1 protein can effectively protect cells treated with AGEs from apoptosis. Local administration of active TIMP-1 protein or TIMP-1 gene therapy at the wound site can be used as a strategy to accelerate diabetic wound healing [69]. TIMP-1 can inhibit the activity of MMP-9 [70]. TIMP-1 may participate in epidermal regeneration by stabilizing the basement membrane zone and regulating matrix remodeling and wound bed angiogenesis. A biopsy of chronic diabetic skin ulcer wound tissue showed decreased expression of TIMP-2 [71]. The lack of TIMP-2 near the migrating epithelial wound edge may lead to uncontrolled activity of MMP-2 in chronic ulcers. TIMP-2 has been shown to impair [65] or accelerate [72] cell migration in vitro. In fact, recombinant human TIMP-2 (rh-TIMP-2) accelerates wound closure in diabetic mice, and GPI anchor modification of TIMP-1 (TIMP-1-GPI) increases the rate of healing in a excision wound model in humans [72]. In hypoxic human keratinocytes, secretion of MMP-2, MMP-9, and TIMP-2 is reduced, while in human monocytes, MMP-9 reduction and TIMP-1 increase are observed [73, 74]. Our results imply a protective role of MMP2, MMP14, TIMP1, and TIMP2 in wound healing.

6. Healing-related fibroblast subtypes drive DFUs healing through promoting matrix remodeling

To further characterize the role of fibroblasts in wound healing, we conducted a detailed analysis of fibroblasts, resulting in the identification of 10 subclusters (Fig 10A), representing distinct molecular states or fibroblast subtypes. Most subclusters exhibited distinct expression profiles, indicating the heterogeneity of the fibroblast population (Fig 11). To further explore the heterogeneity of each subcluster, we performed gene ontology (GO) enrichment analysis for the top 6 genes in each subcluster. The results showed that Cluster 0 represented classical fibroblast types, with enrichment in processes such as extracellular matrix organization, extracellular structure organization, external encapsulating structure organization, collagen fibril organization, and collagen metabolic process (Fig 11). Cluster 1 was associated with complement activation and immune response. Cluster 2 was related to response to UV and ECM degradation. Cluster 3 was associated with leukocyte migration and chemotaxis. Cluster 4 was related to defense response to viruses and symbionts. Cluster 5 was associated with leukocyte aggregation. Cluster 6 was related to regulation of immune response and insulin-like growth factor receptor signaling pathway. Cluster 7 was related to fructose 1,6-bisphosphate metabolic process, lipoxygenase pathway, response to folic acid, UV protection, and fructose metabolic process, indicating its involvement in metabolic regulation. Cluster 8 was associated with kidney development, and Cluster 9 was related to granulocyte/neutrophil migration and chemotaxis (Fig 11).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 10. Identification and characterization of fibroblast subpopulations with specific gene features associated with healing DFUs.

(A) UMAP analysis describes 10 fibroblast subpopulations at a resolution of 0.1. (B) Relative changes in cell proportions across the three groups of subpopulations. The numbers on the right represent Log2FC values of cell proportions (DFUNH/DS and DFUH/DFUNH). The cell subpopulation types highlighted in red are downregulated during the unhealing process and upregulated during the healing process, while the cell subpopulation types highlighted in blue are upregulated during the unhealing process and downregulated during the healing process. Gray indicates cell subpopulation types with no significant proportional changes. (C) UMAP dimensionality reduction plot of cell subpopulation distribution in the DS, DFUNH, and DFUH groups. (D) Cluster 0 classified as "normal" cells with typical skin fibroblast functions, clusters 4 and 6 classified as "good" cells, and clusters 1, 5, and 9 classified as "bad" cells.

https://doi.org/10.1371/journal.pone.0306248.g010

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 11. Heatmap showing the highest expression genes in each subgroup and their GO enrichment analysis results.

https://doi.org/10.1371/journal.pone.0306248.g011

We compared the proportions of fibroblast subclusters in DS, DFUNH, and DFUH. Subclusters 1, 5, and 9 showed higher proportions in DFUNH, while their proportions decreased in DFUH (Fig 10B and 10C). On the other hand, subclusters 4 and 6 had lower proportions in DFUNH but showed a rebound in proportions in DFUH (Fig 10B and 10C). Since subclusters 4 and 6 mainly originated from the DFUH group (Fig 10C), we defined them as healing-related fibroblast subtypes. Similarly, we defined subclusters 1, 5, and 9 as rescue-down subtypes and subclusters 4 and 6 as rescue-up subtypes.

To investigate the differentiation sequence of fibroblast subtypes, we used Cytotrace to predict the starting subcluster and the order of differentiation. The results showed that Cluster 2 was located at the starting point of differentiation, followed by clusters 9, 0, 4, 6, 1, 3, 7, 5, and 8 (Fig 12A). Assuming that clusters 1, 5, and 9 have a negative impact on wound healing, while clusters 4 and 6 have a positive impact, we referred to clusters 1, 5, and 9 as "bad" fibroblasts and clusters 4 and 6 as "good" fibroblasts (Fig 10D).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 12. Pseudotemporal analysis of fibroblast subpopulations and analysis of biological differences among subpopulations.

(A) Cytotrace determined the differentiation order or sequential appearance of subpopulations. Cytotrace scores from high to low correspond to the order of cell differentiation or sequential appearance. (B) Monocle3 inferred trajectories of fibroblast clusters. From deep purple to light yellow represents a putative temporal trajectory of fibroblast differentiation. (C) Genes significantly changed during the transition from a normal phenotype to a bad phenotype and then to a good phenotype. (D-E) KEGG and GO enrichment analysis of differentially expressed genes between good and bad cells. Red represents enrichment entries of upregulated genes in good cells, and blue represents enrichment entries of downregulated genes in good cells. The x-axis represents the log of the p-value corresponding to each enrichment entry. Positive values indicate upregulation, and negative values indicate downregulation.

https://doi.org/10.1371/journal.pone.0306248.g012

To explore the differences in the differentiation fate from classical fibroblasts to "bad" or "good" subtypes, we performed pseudo-time analysis. Cluster 0 represented classical fibroblast types, and its proportion did not significantly change among the groups. Therefore, we considered Cluster 0 as "normal" cells and used it as the starting point for pseudo-time analysis, while "bad" and "good" cells represented the two different differentiation fates (Fig 12B). Pseudo-time analysis results revealed significant gene expression changes during the transition from "normal" to "bad" or "good" cell phenotypes. Genes such as APCDD1, APOD, APOE, CFD, GSN, MT-ATP6, MT-CO3, PLA2G2A, and PRG4 were upregulated in the transition from "normal" to "bad" phenotype but showed a decreasing trend in expression during the transition from "bad" to "good" phenotype (Fig 12C). These genes with altered expression may be involved in the phenotypic transition of different fibroblast subtypes, influencing wound healing. Similarly, ASPN, COL1A1, COL1A2, COL3A1, SFRP4, SPARC, and TNC showed decreased expression in the transition from "normal" to "bad" phenotype but increased expression in the transition from "bad" to "good" phenotype (Fig 12C). The results of the ROC curve analysis indicated that APOD, APOE, COL3A1, PRG4, and TUBA1B have the potential to serve as biomarkers for non-healing DFUs (S8 Fig).

ASPN promotes ECM signaling, adhesion, and migration. ASPN, an ECM protein, has been found to inhibit NF-κB activity induced by TLR2 and TLR4 and the expression of pro-inflammatory cytokines in macrophages [75]. TLR4-mediated inflammation drives the synergistic effect of hypoxia and hyperglycemia on impaired diabetic wound healing [76]. Therefore, excessive expression of ASPN may be an important determinant of DFUs healing. In a recent study, a distinct cluster of fibroblasts enriched in ASPN and POSTN was described as interstitial cells and demonstrated a more dermal localization [77]. The roles of COL1A1, COL1A2, and TNC in promoting wound healing have been described previously and are not further elaborated here. Studies have shown that wound healing is significantly delayed in SPARC-deficient mice, indicating the positive role of SPARC in wound healing [78].

We further investigated the biological functional differences between "good" and "bad" cells. GO and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were performed for differentially expressed genes between these two cell types. GO enrichment analysis revealed that upregulated genes in "good" cells, compared to "bad" cells, were involved in cell adhesion, cytoplasmic translation, collagen fibril organization, extracellular matrix organization, response to wounding, negative regulation of BMP signaling pathway, and negative regulation of monocyte chemotaxis. Downregulated genes were associated with cellular respiration, aging, complement activation, negative regulation of amyloid fibril formation, and killing of cells of other organisms (Fig 12E). Aging prolongs the inflammatory phase and increases the production of reactive oxygen species (ROS), leading to further impairment of chronic wound healing [79]. Increased levels of C3, C3a, and C3d have been observed in chronic wounds, burns, or traumatic wounds [80–84], indicating uncontrolled complement activation in these types of wounds. Additionally, studies on animals lacking complement components and patients receiving complement inhibitors have confirmed the importance of controlling the complement system in wound healing [85–90]. Activation of the complement system in chronic wounds leads to cell death and enhanced inflammation, resulting in further damage and impaired wound healing [91]. Attenuation of complement activation through specific inhibitors is considered an innovative wound care strategy. Furthermore, KEGG enrichment analysis revealed that upregulated genes were involved in focal adhesion, TGF-beta signaling pathway, regulation of actin cytoskeleton, and phagosome, while downregulated genes were associated with oxidative phosphorylation, complement and coagulation cascades, and the PI3K-AKT signaling pathway (Fig 12D). Transforming growth factor-beta (TGFβ) plays a crucial role in maintaining skin homeostasis and is important for reepithelialization, inflammation, angiogenesis, and granulation tissue formation during wound healing. Numerous studies have shown that exogenous application of recombinant TGFβ1 accelerates healing in animal models [92, 93], while exogenous inhibition of TGFβ signaling using antibodies targeting all TGFβ subtypes impairs epithelialization and granulation tissue formation [94].