Identification of Skin Multicellular Reprogramming Factors as Potential Treatment for Nonhealing Diabetic Foot Ulcers

Abstract

Objective:

In a complex environment such as that in a diabetic foot ulcer (DFU), multiple factors, including cross talk between distinct cell types of the affected tissue, play a significant role. We identified a transcription factor (TF) cocktail that induces a transition from nonhealing to healing states across multiple cell types.

Approach:

Thirty-three skin and wound single‐cell RNA‐sequencing samples (85,928 cells) from patients with diabetes with healing or nonhealing DFU were analyzed (GSE165816). The relative activity of cell type-specific TF in healing versus nonhealing DFU was compared, and the cumulative additive effect of different TF cocktails was assessed.

Results:

We used a cumulative additive-effect approach to identify five transcription factors, FOS Like 2, AP-1 Transcription Factor Subunit (FOSL2), CAMP Responsive Element Binding Protein 3 Like 1, RELB Proto-Oncogene, NF-KB Subunit, ETS Proto-Oncogene 1 (ETS1), and X-Box Binding Protein 1, whose targets include 66.5% of pro-healing genes and only 12.5% of anti-healing genes across all cell types. In vascular endothelial cells, this TF panel accounted for 95% of vasculature-development genes; in myeloid cells, it regulated 85% of antimicrobial-response genes. In silico knockout of ETS1 or FOSL2 shifted cells toward nonhealing states, whereas Nuclear Receptor Subfamily 3 Group C Member 1 knockout shifted endothelial cells, fibroblasts, and myeloid cells toward healing-associated state.

Innovation:

This work recognizes a TF panel that is likely to have therapeutic value in promoting healing in nonhealing DFU.

Conclusion:

In this work, we identified a set of candidate TFs with the potential to induce a cell state transition favoring a switch from nonhealing to healing outcomes in patients with nonhealing DFU. Overall, our gene regulatory network-driven TF cocktail provides a rational blueprint for reprogramming DFU cell states and paves the way toward targeted regenerative therapies.

Keywords

cell reprogramming cell state transition wound closure diabetic foot ulcer

INTRODUCTION

Diabetes mellitus affects approximately 550 million individuals around the world.¹ Each year, an estimated 18.6 million people with diabetes experience diabetic foot ulcers (DFU). DFU is known for its high recurrence rate and challenging treatment process.^1,2 Around 20% of patients with diabetes diagnosed with a foot ulcer undergo minor or major lower-extremity amputation.³ Infection is a major risk factor for amputation, occurring in almost 50% of DFUs.^4,5 The nonhealing outcome is caused by multiple factors, including poor vascularization, neuropathy, and defective immune cells.^6–8

Ahmed S. Abouhashem, PhD

Hossam Sharara, PhD

Coordination among multiple cell types, including endothelial cells, fibroblasts, macrophages, and keratinocytes, is required for proper healing.⁹ However, abnormalities across multiple cell types in the ulcer microenvironment impede proper healing.¹⁰ The application of single‐cell RNA‐sequencing (scRNA‐seq) enables analysis of gene expression at single‐cell resolution.¹⁰ Using scRNA-seq, distinct states for each cell type were identified in both healthy individuals and patients with diabetes, including those with healing and nonhealing diabetic foot ulcers (DFU).¹⁰ Phospholipase C gamma 2 was found to be downregulated across multiple cell types in diabetic skin, and its overexpression rescued hindlimb ischemia in diabetic mice.¹¹ Dermal microvascular endothelial cells derived from patients with DFU exhibit impaired tube‐forming capacity compared with healthy controls.¹² A healing-specific fibroblast subset with higher hypoxia-inducible factor 1 subunit alpha (HIF1A) expression was found in healing patients with DFU, suggesting its role in angiogenesis.¹⁰ The challenge arises from the complexity of the DFU microenvironment, where different cell types need to enter healing-specific states to achieve a proper healing outcome.

Modulating TF expression rewires downstream gene networks, enabling cell state transitions.^13,14 Previous literature has documented the alteration of cellular transcriptional programs by changing TF expression levels in both health and disease.¹⁵ These changes can significantly impact gene expression patterns, influencing various cellular functions and processes. Altered TF levels can contribute to disease progression or potentially offer therapeutic avenues.¹⁵ For example, downregulation of the TF GATA Binding Protein 4 rendered hepatoblastoma cells more sensitive to doxorubicin-induced apoptosis.¹⁶ Induction of the TF SRY-Box Transcription Factor 2 expression enhanced angiogenesis and keratinocyte migration in cutaneous wound healing.¹⁷ The TF Fli-1 Proto-Oncogene, ETS, was reported to contribute to enhanced tissue perfusion and wound closure.¹⁸ In addition, TFs are being used in cell fate reprogramming.^19,20 TF cocktails that drive coordinated state changes across multiple cell types represent a promising regenerative‐medicine strategy for chronic DFU. However, the main challenge lies in choosing which TFs to target in order to modulate their expression level to achieve the desired cell state. Although bulk RNA‐seq has contrasted healing versus nonhealing DFU, it lacks cell‐type resolution, and the master‐regulatory role of TFs in orchestrating multicellular healing remains unexplored.^21,22 However, these studies lacked cell-type resolution, resulting in the omission of TF expression differences across distinct cell types. Recent advancements in scRNA-seq have facilitated the identification of differentially expressed genes (DEGs), pathways, and cell–cell interactions among patients with DFU at the single-cell level.^10,23–25 These studies, however, primarily concentrated on identifying altered genes, pathways, and cellular interactions rather than examining TF activity at the cell-type level.¹⁰ In addition, their focus was on cellular heterogeneity within specific cell types, such as fibroblasts²³ or endothelial cells.²⁴ The current study specifically evaluates TF relative activity in healing and nonhealing DFU at the level of individual cell types.

The aim of this work was to identify a set of TFs that may have the potential to restore healing in nonhealing DFU by inducing a cell state transition from nonhealing to healing states in multiple cell types simultaneously. scRNA-seq datasets (GSE165816) from the wound edge of patients with DFU were analyzed to estimate the relative activity of TFs between healing and nonhealing DFU. Subsequently, the cumulative additive effect of TF cocktails that maximize the coverage of upregulated and minimize the coverage of downregulated genes in healing DFU was assessed, with the goal of restoring the skin’s ability to heal by shifting the gene expression profile from nonhealing to healing phenotypes in multiple cell types.

CLINICAL PROBLEM ADDRESSED

DFU with nonhealing outcomes pose a major burden on patients with diabetic and can lead to amputation. Current treatment options remain limited because of dysfunction across multiple cell types, including poor vascularization, neuropathy, and immune cell dysregulation. Addressing such a complex, multicellular pathology requires systemic strategies capable of reprogramming cells toward a pro-healing state. By analyzing scRNA-seq datasets from healing and nonhealing DFU, we uncovered a TF cocktail with the potential to induce healing at multicellular level, offering a new potential therapeutic for patients with nonhealing DFU.

MATERIALS AND METHODS

Data source

To identify alterations in transcriptional programs in diabetic healing and nonhealing DFU, we analyzed scRNA-seq data from the Gene Expression Omnibus database (GSE165816).¹⁰ The foot skin samples in the dataset include 94,325 cells from healthy skin samples (n = 11), diabetic skin samples without foot ulcers (n = 8), diabetic skin samples with healing foot ulcers (n = 9), and diabetic skin samples with nonhealing foot ulcers (n = 5). The authors of the original study collected scRNA-seq samples from patients with diabetic with plantar foot ulceration who underwent ulcer surgical resection, providing wound and peri-wound tissue for sampling. Patients with DFU were classified into healing and nonhealing groups based on a 12-week follow-up post-surgery.

Data preprocessing and cell type identification

The data analysis was performed using the Seurat library in R.^26–28 First, filtration of low-quality cells was performed. Cells with fewer than 200 detected genes, fewer than 1,000 reads, more than 50,000 reads, or with more than 30% of their transcripts coming from mitochondrial DNA were removed. The remaining 85,928 cells from 33 different samples were integrated using canonical correlation analysis to identify shared cell states across different samples. Next, the top 2,000 highly variable genes were used to perform principal component analysis. The top 25 principal components were then used for visualization using the t-distributed stochastic neighbor embedding and for clustering using the Louvain algorithm. To identify cell types within each cluster, manual annotation was performed using canonical markers (DCN, COL1A1 and COL1A2 for fibroblasts; TAGLN and ACTA2 for smooth muscle cells (SMC); KRT1, KRT5, and KRT14 for keratinocytes; VWF, PECAM1, and CDH5 for vascular endothelial cells; CD163 and IL1B for myeloid cells; CD3D, CCL5, and GZMB for NK and T cells [NKT] cells; TYRP1 for neurons; LYVE1 and CCL21 for lymphatic endothelial cells (L-endo); and TPSAB1 for mast cells) (Supplementary Figs. S1 and Figs. S2A).

Gene expression alteration between healing and nonhealing diabetic foot ulcer

To identify transcriptomic changes between healing and nonhealing DFU, the Wilcoxon rank-sum test was used to compare gene expression in each of the nine identified cell types in healing DFU with its counterpart in nonhealing DFU. Only genes detected in at least 20% of cells in either group were included in the analysis. To identify alterations in biological processes, gene set enrichment analysis (GSEA) was performed using the DEGs with adjusted p-value < 0.05 in each cell type, based on the Molecular Signatures Database.²⁹

Gene regulatory network inference

To assess the regulatory activity of TFs associated with healing DFU, we used the Python implementation of the Single-Cell rEgulatory Network Inference and Clustering (pySCENIC) to perform gene regulatory network (GRN) inference.³⁰ Regulatory modules (TF and its targets) were identified by inferring co-expression between TF and genes containing TF binding motifs in their promoters. Gene-motif rankings within ±10 kb of the transcription start site were obtained from RcisTarget databases (Genome Reference Consortium Human Build 38). GRNBoost was then used to infer the co-expression modules. Regulons were then defined by excluding indirect targets from the co-expressed modules based on the enrichment of DNA TF-binding motifs using the RcisTarget database.³¹ Network inference was performed separately for the following cell types: myeloid cells, keratinocytes, SMC, vascular endothelial cells, NKT cells, and fibroblasts. For each cell type, the inference was repeated 10 times without fixing a random seed to account for uncertainty in TF-target inference. Next, TF-target edge weights were normalized to a range between 0 and 1 based on the number of times each edge appeared across the runs.

TF relative activity estimation between healing and nonhealing DFU

The relative activity of each TF within distinct cell types in healing and nonhealing DFU was assessed based on the expression levels of its target genes in the respective cell type. The goal of the algorithm was to identify TFs with overrepresented targets among the DEG list, thereby highlighting their potential roles in healing. A positive score was assigned to a TF if its target genes were overrepresented among the top upregulated genes in healing DFU, indicating elevated activity. Conversely, a negative score was assigned if the target genes were predominantly found among the top downregulated genes in healing DFU, reflecting reduced activity.

Differentially expressed genes ranking

For each GRN, gene sets were constructed to include the inferred targets for each TF. DEGs with an adjusted p-value < 0.05 were identified for each cell type and sorted in descending order based on their log2 fold change values when comparing healing with nonhealing DFU. This ranking ensured that genes with higher differential expression contributed more substantially to the final score than those with lower differential expression, as shown in Equations (1) and (2).

GSEA-like enrichment

The relative activity of each TF within distinct cell types was assessed based on the expression levels of its target genes. TFs with targets overrepresented among the top upregulated genes in healing DFU were assigned positive scores, indicating elevated activity, while TFs with targets overrepresented among the top downregulated genes were assigned negative scores, reflecting reduced activity. To quantify this, a GSEA-like approach was employed.²⁹ Starting with a score of zero for each TF, the score was incremented when a gene was present in the TF’s target set and adjusted based on the TF-target edge weight and the log2 fold change of the gene as follows:

P_{hit} (T, i) = \sum_{g_{j} \in T} \frac{| w_{i j} r_{j} |}{N_{R}}, where N_{R} = \sum_{g_{j} \in T} | w_{i j} r_{j} |

(1)

where T represents a set of targets for the corresponding TF, i represents the index of the gene in the ordered DEG list, w_ij represents the TF-target edge weight, and r_j represents the log2 fold change of the target gene when comparing healing with nonhealing DFU. Conversely, the score was decremented if the gene was absent from the target set as follows:

P_{miss} (T, i) = \sum_{g_{j} \in T} \frac{1}{(N - N_{H})}

(2)

where T represents a set of targets for the corresponding TF, i represents the index of the gene in the ordered DEG list, N represents the total number of DEG, and N_H represents the number of targets for the corresponding TF. After iterating over all DEGs, the final TF score for each cell type was defined as the maximum deviation from zero, analogous to the standard GSEA enrichment score.

Significance testing and TF filtering

To assess the statistical significance of each TF score, we performed a permutation test by randomly shuffling the labels of the DEG 1,000 times. The resulting p-values were adjusted using the Benjamini–Hochberg procedure. We require TFs to have more than five targets to avoid enrichment from TF with small set of targets. TFs with an adjusted p-value < 0.05 and at least five targets among the DEGs in the corresponding cell type were considered for downstream analysis.

Indirect regulation of TF

To account for indirect targets, TF scores were further refined by incorporating contributions from TFs found among the direct targets of the original TF. The final TF score for each TF i was calculated as:

Adjusted_Scor e_{i} = α Scor e_{i} + (1 - α) {(\sum_{j \in T} w_{i j})}^{- 1} \times \sum_{j \in T} Scor e_{j} w_{i j}

(3)

where j represents a direct target of the TF i, which is also a TF, w_ij is the edge weight between the TF i and j, T is the set of TFs among the TF i’s direct targets excluding self-targeting, and α represents a scaling factor between 0 and 1. We evaluated three scenarios: α = 1 (considering direct targets only), α = 0 (indirect targets only), and α = 0.5 (average of direct and indirect contributions). This approach provided an estimate of TF activity, capturing both direct and GRN-mediated regulatory effects. The final TF activity scores are summarized in Supplementary Table S3.

Cumulative additive effect of TFs

To identify the optimal number of TFs that maximize coverage of upregulated genes and minimize coverage of downregulated genes in healing compared with nonhealing DFU, TFs with at least five targets in the DEG list and an adjusted p-value < 0.05 were considered, totaling 387 TFs. The cumulative score of each TF or TF cocktail represents the difference between the coverage of upregulated and downregulated genes. For each TF-target, the score is incremented by multiplying the edge weight by the log2 fold change between healing and nonhealing of the corresponding target, so higher scores are assigned to TFs targeting genes with higher log2 fold change values. Finally, the collective TF or TF cocktail scores represent the proportion of coverage of upregulated or downregulated targeted genes across all cell types. A greedy approach was employed, starting with the selection of the TF with the highest score, followed by the identification of the second-best TF to add, aiming to maximize the difference between the coverage of upregulated and downregulated genes. This approach was repeated to identify the best cocktail of 10 TFs. Finally, 100 permutations were generated by randomly shuffling the labels of the DEG used for calculating TF scores.

Analysis of spatial transcriptome during wound healing

A spatial transcriptomics dataset generated using Visium technology was downloaded (GEO accession no. GSE241124).³² The dataset included skin samples from donors at various time points during wound healing, including uninjured skin and days 1, 7, and 30 post-wounding. Genes associated with vasculature development, upregulated in vascular endothelial cells, and genes related to antimicrobial response, upregulated in myeloid cells, in healing DFU samples were used to calculate scores for each Visium spot using the Ucell package in R.³³ In addition, TF activity scores were calculated for Visium spots using the top five TFs identified to maximize coverage of upregulated genes in healing patients with DFU. Targets shared by at least two cell types and having an edge weight of 1 were used for scoring TF activity in the Visium spots. Finally, the correlation between each TF activity score and the vasculature development or the antimicrobial response scores was calculated.

Cell subset analysis

To assess whether the TFs with elevated activity in healing DFU are expressed by specific cell subsets or broadly across entire cell populations, a second round of clustering was performed for fibroblasts, endothelial cells, and myeloid cells. Clustering was conducted independently for each cell type, followed by differential gene expression analysis comparing each identified cell subset with the remaining cells within the corresponding cell types. The proportions of each cell subset were then compared between healing and nonhealing DFU samples using the speckle R package.³⁴ Fibroblast subsets were further evaluated for expression of the vasculogenic gene signature,^18,35 which includes EPAS1, FOSL1, GLUL, HIF1A, Matrix Metallopeptidase 14 (MMP14), NRP2, VEGFC, TGFB1, PECAM1, VWF, and CDH5. Gene set scoring was performed using the AddModuleScore function in Seurat.

In silico knockout simulations of ETS1, FOSL2, and NR3C1

We utilized CellOracle to simulate cell identity shifts resulting from TF perturbations on the transcriptomic profiles of individual cell types along the developmental trajectory toward healing states.³⁶ Initially, Monocle3 was employed to construct baseline developmental flow fields for fibroblast, endothelial cell, and myeloid cell subsets independently.^37,38 The data were then preprocessed following the standard CellOracle pipeline using default parameters. Briefly, genes with zero expression were excluded, and the top 2,000 highly variable genes were retained, along with the top identified TFs. A GRN provided by CellOracle was then fit using ridge regression to perform in silico knockout simulations of ETS1, FOSL2, and NR3C1. For each simulation, the expression of the target TF was set to zero, and the resulting cell state transition probabilities were computed to infer potential shifts in cellular identity.

RESULTS

Microenvironment of the skin of patients with diabetic foot ulcers includes nine cell types

To characterize the cellular landscape of DFU microenvironment, we integrated scRNA-seq data comprising 85,928 cells from 33 samples (Fig. 1A; Supplementary Fig. S1). The dataset included intact skin from healthy individuals (n = 11), intact skin from diabetic individuals without foot ulcers (n = 8), and DFU tissue from patients with healing (n = 9) and nonhealing (n = 5) outcomes. Clusters were annotated using canonical marker genes, resulting in the identification of nine major cell types: fibroblasts, SMC, keratinocytes, vascular endothelial cells, myeloid cells, NKT cells, neurons, L-endo, and mast cells (Fig. 1B; Supplementary Fig. S2A, B).

Figure 1.

Multicellular gene expression alteration in nonhealing diabetic foot ulcers. (A) 33-foot skin scRNA-seq samples were retrieved from the GEO database (accession GSE165816), including 11 samples from healthy individuals, 8 samples from intact skin of diabetic individuals, 9 samples from diabetic individuals with healing foot ulcer, and 5 samples from diabetic individuals with nonhealing foot ulcer. Differential gene expression was performed at cell-type level, and the resulting DEGs were utilized to estimate transcription factor activity. Next, a combination of transcription factors was identified whose targets are more represented among upregulated and less represented among downregulated genes across multiple cell types simultaneously. (B) tSNE plot representing the identified nine cell types. (C) Stripe plot representing magnitude of DEGs in each cell type when comparing healing with nonhealing diabetic foot ulcers at cell-type level. Red dots represent upregulated genes in the healing group, green dots represent downregulated genes in the healing group, while gray dots represent genes with log2 fold change between −0.5 and 0.5. All dots represent genes with adjusted p-value < 0.05. (D) Number of upregulated or downregulated DEGs in healing compared with nonhealing diabetic foot ulcers (unique: in one cell type and shared: in more than one cell type). (E) The top 10 upregulated genes in the healing group in multiple cell types (upper), top 10 downregulated genes in the healing group in multiple cell types (middle), and genes with upregulation and downregulation expression levels in distinct cell types. DEGs, differentially expressed genes; GEO, gene expression omnibus; scRNA-seq, single-cell RNA sequencing; tSNE, t-distributed stochastic neighbor embedding.

Multicellular transcriptomic changes in nonhealing diabetic foot ulcers

DEG analysis between healing and nonhealing DFU conditions identified 7,057 genes with an adjusted p-value < 0.05 across all cell types (Fig. 1C). Among these, 902 genes showed substantial expression changes, with log2FC ≥ 0.5 or ≤ −0.5 in at least one cell type (Fig. 1D; Supplementary Table S1). Of these, 119 genes were upregulated in more than one cell type (shared upregulated), including collagen-related genes such as COL1A1, COL1A2, and COL3A1, suggesting extracellular matrix remodeling. In contrast, 67 genes were downregulated across multiple cell types (shared downregulated) (Fig. 1E). In addition, 353 genes were upregulated in a single cell type, 305 genes were downregulated in one cell type, and 58 genes exhibited both upregulation and downregulation in different cell types (bidirectional change; Fig. 1E). Notably, genes upregulated in healing DFU included the TF HIF1A, which has previously been implicated in promoting diabetic wound healing (Supplementary Table S1).^39,40

Multicellular disruption of key biological processes in nonhealing diabetic foot ulcers

GSEA of DEG revealed that several biological processes were altered in nonhealing DFU (Fig. 2A; Supplementary Figs. S3 and Figs. S4; Supplementary Table S2). Fibroblasts, SMC, and keratinocytes from healing DFU exhibited upregulation of biological processes related to endoderm development and endodermal cell differentiation (Fig. 2B). Notably, MMP14 and Collagen Type XII Alpha 1 Chain (COL12A1) were significantly upregulated in fibroblasts from healing DFU (Fig. 2C and D). MMP14 contributes to collagen remodeling in adult skin, and its knockout impairs fibroblast ability to process fibrillar collagen.⁴¹ COL12A1 supports myofibroblast survival and wound closure,⁴² and its knockout has been shown to result in delayed healing.⁴² Vascular endothelial cells in healing DFU also showed enrichment of vasculature development pathways, marked by upregulation of angiogenic transcription factors such as ETS Proto-Oncogene 1 (ETS1) and HIF1A (Fig. 2C and D).^40,43 In addition, myeloid cells in healing DFU exhibited increased expression of genes involved in the antimicrobial humoral response, including S100 Calcium Binding Protein A9 and A12 (S100A9 and S100A12), both of which play essential roles in suppressing biofilm formation and inhibiting bacterial growth (Fig. 2C and D).^44–46

Figure 2.

Multicellular biological process alteration in nonhealing diabetic foot ulcers. (A) Number of biological processes upregulated or downregulated in healing compared with nonhealing diabetic foot ulcers (unique: in one cell type and shared: in more than one cell type). (B) Common upregulated biological processes in three cell types in healing compared with nonhealing diabetic foot ulcers. (C) Selected three biological processes upregulated in healing compared with nonhealing diabetic foot ulcers in fibroblasts (left), vascular endothelial cells (middle), and myeloid cells (right). Colored horizontal bars (red to blue) represent the ordered list of the DEGs in descending order based on log2 fold change between healing and nonhealing ulcers in the corresponding cell types. Vertical black lines represent the intersection between the DEGs and the corresponding biological process. (D) Expression level of two example upregulated genes from the corresponding biological process (MMP14 and COL12A1: involved in endoderm development, ETS1 and HIF1A: involved in vasculature development, and S100A9 and S100A12: involved in antimicrobial humoral response). COL12A1, collagen type XII alpha 1 chain; ETS1, ETS proto-oncogene 1; HIF1A, hypoxia inducible factor 1 subunit alpha; MMP14, matrix metallopeptidase 14.

Identification of crucial TFs associated with DFU healing via relative activity estimation

To identify TFs regulating the observed transcriptomic changes, we inferred cell-type-specific GRNs and estimated TF relative activities in healing versus nonhealing DFU (Fig. 3A; Supplementary Table S3). Among the top candidates, Nuclear Factor Kappa B Subunit 1 (NFKB1) showed consistently elevated activity across all cell types in healing DFU, reflected by the upregulation of both its direct (α = 1) and indirect (α = 0) targets (Fig. 3B). NFKB1-mediated innate immune signaling has been implicated in the transdifferentiation of fibroblasts into endothelial-like cells, facilitating vasculature development.^47,48 Similarly, ETS1 exhibited increased activity in nonimmune cells, specifically fibroblasts, keratinocytes, SMC, and vascular endothelial cells, where both its direct and indirect targets were upregulated (Fig. 3B). Conversely, Nuclear Receptor Subfamily 3 Group C Member 1 (NR3C1; glucocorticoid receptor) was more active in nonhealing DFU, with upregulation of its target genes suggesting a potential role in impairing wound healing (Supplementary Fig. S5).⁴⁹

Figure 3.

Transcription factor relative activity estimation between healing and nonhealing diabetic foot ulcer. (A) Schematic diagram of the approach. Using pySCENIC, cell type-specific gene regulatory network was inferred 10 times. For each TF regulon (TF + its targets), the TF relative activity between healing and nonhealing diabetic foot ulcer was estimated, resulting in identification of 395 TFs with adjusted p-value < 0.05 in at least one cell type (among which 387 TFs have at least 5 targets). The resulting TF estimates were adjusted to account for direct and indirect targets (see methods). (B) The top 15 TFs with higher relative activity in healing compared with nonhealing diabetic foot ulcers based on the expression level of direct targets only (alpha = 1), indirect targets only (alpha = 0), or the average score (alpha = 0.5). TFs were ranked based on the number of times they have been identified with higher relative activity in distinct cell types. pySCENIC, Python implementation of the Single-Cell rEgulatory Network Inference and Clustering; TF, transcription factor.

The identified TF cocktail upregulates pro-healing genes in silico

Next, we applied a greedy selection algorithm across multiple cell types to identify a TF cocktail capable of simultaneously regulating gene expression programs across diverse cell types (Fig. 4A; Supplementary Table S4). This approach yielded a five-TF cocktail consisting of FOS Like 2, AP-1 Transcription Factor Subunit (FOSL2), CAMP Responsive Element Binding Protein 3 Like 1 (CREB3L1), RELB Proto-Oncogene, NF-κB Subunit (RELB), ETS1, and X-Box Binding Protein 1 (XBP1). Collectively, these TFs covered 66.5% of the genes upregulated in healing DFU while targeting only 12.5% of downregulated genes across all cell types (Fig. 4B; Supplementary Fig. S6A), indicating that the TF cocktail preferentially induces pro-healing genes while minimally mistargeting genes downregulated in the healing state. Among cell types, myeloid cells exhibited the highest responsiveness to this TF cocktail, with 85% of upregulated genes captured, followed by SMC at 78%, vascular endothelial cells at 71%, and fibroblasts at 67% (Fig. 4C). Notably, XBP1 has previously been shown to promote in vitro tube formation in dermal microvascular endothelial cells, underscoring its potential role in vascular regeneration.⁵⁰

Figure 4.

Cumulative additive effect of transcription factors covers upregulated genes in multiple biological processes simultaneously. (A) Schematic diagram of the approach. The TFs with at least five targets and with adjusted p-value < 0.05 (n = 387) in at least one cell type when estimating the TFs relative activity between healing and nonhealing diabetic foot ulcer were used. The difference between coverage of upregulated and downregulated genes was used to score TF or TF cocktail. A greedy approach was used to identify best 10 TF to maximize difference between coverage of upregulated and downregulated genes in healing DFU compared with nonhealing DFU. (B) The greedy approach results in selecting the best 10 TFs that maximize coverage of upregulated genes and minimize coverage of downregulated genes across multiple cell types in healing compared with nonhealing DFU. Blue dashed lines represent the cumulative proportion of upregulated gene coverage (red) and downregulated gene coverage (green) using five TFs. (C) Pie charts representing percentage of genes covered using the top five TFs in B in each cell type among upregulated and downregulated genes in healing compared with nonhealing DFU. Percentage of coverage of upregulated genes among all upregulated genes in the corresponding cell type is labeled. (D) The intersection between the targets of the TF cocktail FOSL2, CREB3L1, RELB, ETS1, and XBP1 and the DEGs involved in three biological processes: vasculature development in vascular endothelial cells (left), endoderm development in fibroblasts (middle), and antimicrobial humoral response in myeloid cells (right). DFU, diabetic foot ulcer; CREB3L1, CAMP Responsive Element Binding Protein 3 Like 1; FOSL2, FOS like 2; RELB, RELB proto-oncogene; XBP1, X-Box binding protein 1.

On the other hand, the TF cocktail comprising Fos Proto-Oncogene, AP-1 Transcription Factor Subunit (FOS), D-Box Binding PAR BZIP Transcription Factor (DBP), Nuclear Receptor Subfamily 3 Group C Member 1 (NR3C1), Kruppel Like Factor 4 (KLF4), and Basic Leucine Zipper ATF-Like Transcription Factor (BATF) collectively targeted 59.2% of the downregulated genes and 14% of the upregulated genes across all cell types (Supplementary Fig. S6B–D). These TFs were considered inhibitory to the healing process, consistent with the known role of the glucocorticoid receptor NR3C1 in delaying wound healing.⁴⁹ The TF difference scores, calculated as the difference between coverage of upregulated and downregulated genes under random permutation of gene labels, ranged from 0.7% to 9% for maximizing upregulated gene coverage and 0.6–8% for maximizing downregulated gene coverage (Supplementary Table S4).

The identified TF cocktail upregulates pro-healing biological processes in silico

The identified TF cocktail was found to concurrently regulate multiple biological processes across distinct cell types. The targets of the top five TFs encompassed 95% of the vasculature development genes upregulated in vascular endothelial cells (including HIF1A, NRP2, FLT1, and KDR), 90% of the endodermal development genes upregulated in fibroblasts (such as MMP14), and 85% of the antimicrobial response genes upregulated in myeloid cells of healing DFU (including S100A9 and S100A12) (Fig. 4D; Supplementary Table S5). Among these, CREB3L1 and ETS1 exhibited the strongest positive correlation with vasculature development scores in Visium spatial transcriptomics spots from skin samples at different wound healing time points (Fig. 5A–F). Notably, ETS1 also showed high activity in spots adjacent to the wound edge that displayed low vasculature development scores (Fig. 5B), suggesting that ETS1 may have additional roles in wound healing beyond angiogenesis. Consistently, overexpression of ETS1 has been reported to inhibit keratinocyte terminal differentiation, thereby maintaining them in a migratory state.⁵¹ Furthermore, antimicrobial response scores correlated most strongly with the TF XBP1 (Fig. 5G–J).

Figure 5.

CREB3L1 and ETS1 correlate positively with vasculature development during wound healing. (A) Activity of CREB3L1 during normal wound healing. (B) Activity of ETS1 during normal wound healing. (C) Vasculature development score during normal wound healing. (D) Scores from A, B, and C over time. (E) and (F) Correlation between CREB3L1 or ETS1 activity and vasculature development. (G) Activity of XBP1 during normal wound healing. (H) Antimicrobial response score during normal wound healing. (I) Scores from G and H over time. (J) Correlation between XBP1 activity and antimicrobial response.

Intercellular activation of VEGF signaling

To identify specific cell subsets expressing the TFs of interest, a second round of clustering analysis was performed on endothelial cells, fibroblasts, and myeloid cells (Fig. 6A, F, and J). Endothelial cell subset 0 exhibited a higher proportion in the healing DFU samples (proportion = 0.44 in healing versus 0.26 in nonhealing) and showed elevated expression of ETS1, VEGFR1, and VEGFR2 (Fig. 6B–E). Similarly, fibroblast subset 1 and myeloid subset 1 were more abundant in healing DFU samples (fibroblast subset 1 proportion = 0.49 in healing versus 0.18 in nonhealing; myeloid subset 1 proportion = 0.29 in healing versus 0.14 in nonhealing) (Fig. 6G and K). Within these subsets, ETS1 expression was higher in fibroblast subset 1, while FOSL2 expression was elevated in myeloid subset 1 (Fig. 6I and M). Notably, both healing-associated subsets (fibroblast subset 1 and myeloid subset 1) exhibited significantly increased expression of Vascular Endothelial Growth Factor A (VEGFA) (Fig. 6H and L). Graphical abstract highlighted the summary of the roles played by the identified TFs across distinct cell types.

Figure 6.

Intercellular activation of Vascular Endothelial Growth Factor (VEGF) signaling. (A) Uniform Manifold Approximation and Projection (UMAP) plot representing vascular endothelial cell subsets in healing and nonhealing DFU samples. (B) Endothelial cell subset 0 proportion in healing and nonhealing DFU samples among endothelial cell population. Ratio of proportions is 1.71 in healing with p-value <0.05. (C–E) Expression level of selected genes in endothelial cell subsets. (F) UMAP plot representing fibroblast cell subsets in healing and nonhealing DFU samples. (G) Fibroblast subset 1 proportion in healing and nonhealing DFU samples among fibroblast population. Ratio of proportions is 2.68 in healing with p-value < 0.05. (H) and (I) Expression level of selected genes in fibroblast subsets. (J) UMAP plot representing fibroblast cell subsets in healing and nonhealing DFU samples. (K) Myeloid cell subset 1 proportion in healing and nonhealing DFU samples among myeloid cell population. Ratio of proportions is 2.07 in healing with p-value < 0.05. (L) and (M) Expression level of selected genes in myeloid cell subsets.

Figure 7.

CellOracle knockout simulation of core transcription factors induces cell state transition toward distinct states. (A) UMAP plots of fibroblast (left), endothelial cell (middle), and myeloid cell subsets from healing and nonhealing scRNA-seq samples. Red circles represent healing-specific cell subsets. (B) Developmental flow of cells. Arrows represent the progression direction. (C) Knockout simulation of ETS1, FOSL2, and NR3C1 in fibroblast (left), endothelial cell (middle), and myeloid cell (right) subset trajectory perturbation. NR3C1, nuclear receptor subfamily 3 group C member 1.

Beyond ETS1 and VEGFA, fibroblast subset 1 also showed higher expression of MMP1 and MMP3, markers linked to a healing-specific fibroblast population in DFU (Supplementary Fig. S7).¹⁰ In addition, this fibroblast subset demonstrated increased vasculogenic scores and elevated expression of PECAM1, consistent with a previously described fibroblast state termed vasculogenic fibroblasts (Supplementary Fig. S7).¹⁸

The identified TF cocktail induces cell state transition toward healing states in silico

We leveraged Monocle3 pseudotime trajectory analysis to infer the baseline flow of cells between states (Fig. 7A and B). Using CellOracle, we simulated knockout effects of the TFs ETS1 and FOSL2 (healing-specific) and NR3C1 (nonhealing-specific). Knockout simulation of ETS1 in fibroblasts and endothelial cells caused cells to shift away from healing-associated states, specifically, fibroblast subset 1 and endothelial subset 0 (Fig. 7C). Similarly, knockout of FOSL2 in myeloid cells shifted cells away from the healing-specific subset 1 (Fig. 7C). In contrast, knockout of NR3C1, a TF associated with nonhealing states, shifted cells toward healing-associated subsets across fibroblasts (subset 1), endothelial cells (subset 0), and myeloid cells (subset 1) (Fig. 7C). Figure 8 represents a graphical abstract summarizing the roles of the identified transcription factors across distinct cell types.

Figure 8.

Graphical summary. The TF combination ETS1, CREB3L1, RELB, FOSL2 and XBP1 works at multicellular level. CREB3L1 and ETS1 correlates positively with vasculature development. ETS1 has higher expression in healing-specific non-immune cells (endothelial cells, fibroblasts, smooth muscle cells and keratinocytes). Its expression in endothelial cells correlates positively with VEGFR1, VEGFR2 and HIF1A expression. In fibroblasts, ETS1 is expressed in the healing-specific subset which has higher expression of VEGFA to induce vasculature development. On the other hand, FOSL2 is expressed in the healing-specific myeloid cell subset and correlates positively with VEGFA and Heparin-Binding EGF-Like Growth Factor (HBEGF) with roles in blood vessel development and keratinocyte migration, respectively.

TAKE-HOME MESSAGES

A subset of DEG showed consistent regulation across multiple cell types, highlighting shared transcriptional programs in DFU healing.

The TF cocktail (FOSL2, CREB3L1, RELB, ETS1, and XBP1) collectively regulates 66.5% of the genes upregulated in healing DFU across diverse cell types.

The cocktail encompasses genes involved in vasculature development, endodermal development, and antimicrobial response.

The cocktail may contribute to intercellular activation of VEGF signaling.

In silico knockout of the cocktail members shifted cells away from healing-specific states, whereas knockout of NR3C1 redirected cells toward healing-specific subsets.

DISCUSSION

Nonhealing DFUs represent a major clinical challenge, leading to lower-limb amputation in approximately 20% of affected patients.³ Poor healing outcomes are primarily driven by impaired vascularization, peripheral neuropathy, and their susceptibility to infection.^1,2 These risk factors, compromising multiple cell types, highlight the need for a systemic, multicellular approach to promote effective healing in nonhealing DFUs. Many of the identified altered biological processes in the nonhealing compared with healing DFU relate to known risk factors for impaired diabetic ulcer healing and revascularization, which offers a further sort of validation from earlier studies. Vascular endothelial cells in healing DFU samples overexpress genes involved in vasculature development, such as ETS1 and HIF1A.^43,52 Besides, the observed downregulation of endoderm formation and development across multiple cell types in nonhealing DFU suggests that coordinated interactions among different cell types are essential for effective DFU healing. Furthermore, the upregulated NFKB1 activity in healing DFU, alongside the reduced antimicrobial response in myeloid cells from nonhealing DFU, underscores the critical role of a properly regulated immune response for wound repair. Collectively, these findings align with the current understanding that nonhealing DFU results from impaired multicellular functionality, which contributes to the pathophysiology complexity of diabetic ulcers.

In this work, we constructed cell type-specific GRNs from DFU samples with healing and nonhealing outcomes. We identified a set of TFs that may promote healing by driving cell state transitions from nonhealing to healing states. TF activity estimation revealed that certain TFs, such as ETS1, exhibit elevated activity in healing compared with nonhealing DFU across multiple cell types simultaneously: endothelial cells, fibroblasts, keratinocytes, and SMC. This supports the potential for identifying a cocktail of TFs that can broadly target upregulated genes while minimizing targeting of downregulated genes in healing DFU at multicellular level.

The TF cocktail comprising FOSL2, CREB3L1, RELB, ETS1, and XBP1 collectively regulates 66.5% of upregulated genes and 12.5% of downregulated genes across multiple cell types. The direct downstream targets of this TF cocktail include key genes involved in vasculature development, such as HIF1A and NRP2, as well as VEGF receptors VEGFR1 and VEGFR2 (FLT1 and KDR) in vascular endothelial cells; genes with role in endoderm development like MMP14 in fibroblasts; and antimicrobial genes such as S100A12 in myeloid cells, which help inhibit bacterial growth and biofilm formation.^44,45

At the cell subset level, knockout simulations further support the role of the identified TF cocktail in maintaining healing-specific states. ETS1 in silico knockout shifted endothelial cells and fibroblasts away from healing-specific subsets, while FOSL2 in silico knockout shifted myeloid cells away from the healing-specific subsets. This further highlights that a mixture of TFs is needed for inducing cell state transition toward healing states at multiple cell types simultaneously. Conversely, in silico knockout of NR3C1, which has higher activities in nonhealing DFU, shifted cells toward healing-specific subsets across fibroblasts, endothelial cells, and myeloid cells. Interestingly, subsets expressing the TF ETS1 and FOSL2 were found to interact to activate VEGF signaling in vascular endothelial cells to promote angiogenesis. Specifically, healing-specific subsets of myeloid cells and fibroblasts express the ligand VEGFA, alongside the TFs FOSL2 and ETS1, respectively, while the healing-specific subset of vascular endothelial cells expresses VEGFR1 and VEGFR2 receptors alongside the TF ETS1. Conversely, the downregulation of VEGF ligands, receptors, and these TFs in the nonhealing group may underlie the impaired VEGF signaling and limited angiogenesis observed. This highlights the role of the identified TFs in enhancing intercellular communication for inducing VEGF signaling and stimulating angiogenesis during DFU healing.

The identified TF cocktail presents promising candidates for further evaluation of their potential to promote healing in nonhealing DFU. In contrast, assessing antagonists targeting TFs with elevated activity in nonhealing DFU may be of equal importance. Cells from nonhealing DFU exhibit higher activity of the glucocorticoid receptor NR3C1, which can activate glucocorticoid signaling pathways that impair the healing process.⁴⁹ However, delivering the TF cocktail remains a significant challenge. Current in vivo TF delivery methods include viral vectors, nanocarriers, cell-penetrating peptides, and various transfection techniques, which have demonstrated success in mouse models for skin reprogramming and enhanced wound healing.^53–56 Experimental validation is essential to confirm that overexpression of these TFs effectively upregulates their target genes in vitro and in vivo. In addition, the analysis was based on transcriptomics data to predict TF targets without proteomics or chromatin data validation. Furthermore, incorporating the impact of target genes on lipid signaling pathways and how known drugs regulate their expression level could provide additional layers of biological insight and enhance the interpretability of the findings.^57,58 Although TF-based therapies face obstacles such as delivery efficiency and potential off-target effects, encouraging results from animal studies over the past few years and ongoing advances in delivery technologies provide a promising path toward clinical translation. In addition, this analysis is based on a single cohort of samples, and the results may differ when applied to other cohorts. However, the findings are supported by consistent results observed in Visium wound-healing samples, where the same transcription factors demonstrated high activity during healing, as well as in unpublished Xenium datasets.

INNOVATION

At present, multiple factors involving various cell types have been implicated in the nonhealing outcomes observed in some patients with diabetes with DFU. Therefore, it is critical to identify master regulators capable of orchestrating healing across these diverse cell populations. Our integrative GRN approach is the first to nominate a multi-TF cocktail (FOSL2, CREB3L1, RELB, ETS1, XBP1) controlling >65% of healing-associated genes.

AUTHORS’ CONTRIBUTIONS

A.S.A. and H.S. conceived and designed the work. A.S.A. wrote the analysis code with contribution from S.K.S., M.E., A.A.E., and H.S. A.S.A., K.S., and C.K.S. wrote the article. Project administration by H.M.E.A. All authors read, interpreted or provided comments on the article. A.S.A. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Footnotes

FUNDING INFORMATION

This work was supported in part by National Institute of Diabetes and Digestive and Kidney Diseases (NIDDK) grants DK136814 to K.S. and DK141513 and DK135447 to C.K.S. This work was also supported in part by U.S. Department of Defense grants W81XWH-22-1-0146 and HT9425-24-1-0131 to K.S. Research programs led by K.S. and C.K.S. were also supported by Commonwealth of Pennsylvania (BT McGowan) research grants. Additional support was provided by a research grant to H.M.E.A. from the American University in Cairo.

AUTHOR DISCLOSURE AND GHOSTWRITING

The authors declared no potential conflicts of interest with respect to publication of this article, authorship, and research.

DATA AVAILABILITY

The dataset was downloaded and reanalyzed from the National Center for Biotechnology Information Gene Expression Omnibus (GEO) repository (accession number: GSE165816). The code for calculating the indirect activity of a TF, based on its direct score and the GRN, is implemented in the R package RTFactivity, available under the ASA-SysBio project on GitHub:

CONSENT FOR PUBLICATION

All the authors listed have approved the article that is enclosed.

ADDITIONAL INFORMATION

Supplementary information is available for this paper.

Correspondence and requests should be addressed to H.S. or A.S.A.

ABOUT THE AUTHORS

Ahmed S. Abouhashem, PhD, is an Assistant Professor of Surgery at the McGowan Institute for Regenerative Medicine (MIRM), University of Pittsburgh School of Medicine. Sherine K. Saber, BS, is a Master student in Biotechnology, the American University in Cairo (AUC). Sara Abouzekry, MS, is a PhD student in Biotechnology, AUC. Mohammed Elkholy, BS, was an undergraduate student in Computer Science, AUC. Ahmed Moustafa, PhD, is a Professor of Bioinformatics and Director of Systems Genomics Lab, AUC. Ahmed Abdellatif, PhD, is an Associate Professor of Biomedical and Neurosciences, AUC. Hassan M. E. Azzazy, PhD, is a Distinguished University Professor of Chemistry, AUC. Ahmed A Elbaz, MS, was a master student in Biotechnology, AUC. Kanhaiya Singh, PhD, is an Associate Professor of Surgery at the MIRM, University of Pittsburgh School of Medicine. Chandan K. Sen, PhD, is a Professor of Surgery and Director at the McGowan Institute for Regenerative Medicine, University of Pittsburgh School of Medicine. Hossam Sharara, PhD, is an Assistant Professor of Computer Science, AUC.

Supplemental Material

Abbreviations and Acronyms

References

Armstrong

, Tan

, Boulton

AJM

, et al. Diabetic foot ulcers: A review. JAMA, 2023; 330(1):62–75; doi: 10.1001/jama.2023.10578

Xiang

, Cai

, Xiao

, et al. Single-cell sequencing technology in diabetic wound healing: New insights into the progenitors-based repair strategies. World J Stem Cells, 2024; 16(5):462–466; doi: 10.4252/wjsc.v16.i5.462

McDermott

, Fang

, Boulton

AJM

, et al. Etiology, epidemiology, and disparities in the burden of diabetic foot ulcers. Diabetes Care, 2023; 46(1):209–221; doi: 10.2337/dci22-0043

Lipsky

, Senneville

, Abbas

, et al.; International Working Group on the Diabetic Foot (IWGDF). Guidelines on the diagnosis and treatment of foot infection in persons with diabetes (IWGDF 2019 update). Diabetes Metab Res Rev, 2020; 36 (Suppl 1):e3280; doi: 10.1002/dmrr.3280

Cortes-Penfield

, Armstrong

, Brennan

, et al. Evaluation and management of diabetes-related foot infections. Clin Infect Dis, 2023; 77(3):e1–e13; doi: 10.1093/cid/ciad255

Yang

, Zhu

, Han

, et al. Pathophysiology of peripheral arterial disease in diabetes mellitus. J Diabetes, 2017; 9(2):133–140; doi: 10.1111/1753-0407.12474

Feldman

, Callaghan

, Pop-Busui

, et al. Diabetic neuropathy. Nat Rev Dis Primers, 2019; 5(1):41; doi: 10.1038/s41572-019-0092-1

Sawaya

, Stone

, Brooks

, et al. Deregulated immune cell recruitment orchestrated by FOXM1 impairs human diabetic wound healing. Nat Commun, 2020; 11(1):4678; doi: 10.1038/s41467-020-18276-0

Zubair

, Ahmad

. Role of growth factors and cytokines in diabetic foot ulcer healing: A detailed review. Rev Endocr Metab Disord, 2019; 20(2):207–217; doi: 10.1007/s11154-019-09492-1

10.

Theocharidis

, Thomas

, Sarkar

, et al. Single cell transcriptomic landscape of diabetic foot ulcers. Nat Commun, 2022; 13(1):181; doi: 10.1038/s41467-021-27801-8

11.

Rustagi

, Abouhashem

, Verma

, et al. Endothelial phospholipase Cγ2 improves outcomes of diabetic ischemic limb rescue following VEGF therapy. Diabetes, 2022; 71(5):1149–1165; doi: 10.2337/db21-0830

12.

, Li

, Lu

, et al. Single-cell RNA-seq and bulk-seq identify RAB17 as a potential regulator of angiogenesis by human dermal microvascular endothelial cells in diabetic foot ulcers. Burns Trauma, 2023; 11:tkad020; doi: 10.1093/burnst/tkad020

13.

Wang

, Wang

, Huang

, et al. Understanding transcription factor regulation by integrating gene expression and DNase I Hypersensitive sites. Biomed Res Int, 2015; 2015:757530; doi: 10.1155/2015/757530

14.

Grath

, Dai

. Direct cell reprogramming for tissue engineering and regenerative medicine. J Biol Eng, 2019; 13:14; doi: 10.1186/s13036-019-0144-9

15.

Lee

, Young

. Transcriptional regulation and its misregulation in disease. Cell, 2013; 152(6):1237–1251; doi: 10.1016/j.cell.2013.02.014

16.

Soini

, Pihlajoki

, Kyrönlahti

, et al. Downregulation of transcription factor GATA4 sensitizes human hepatoblastoma cells to doxorubicin-induced apoptosis. Tumour Biol, 2017; 39(3):1010428317695016; doi: 10.1177/1010428317695016

17.

Uchiyama

, Nayak

, Graf

, et al. SOX2 epidermal overexpression promotes cutaneous wound healing via activation of EGFR/MEK/ERK signaling mediated by EGFR ligands. J Invest Dermatol, 2019; 139(8):1809–1820.e8; doi: 10.1016/j.jid.2019.02.004

18.

Pal

, Ghatak

, Singh

, et al. Identification of a physiologic vasculogenic fibroblast state to achieve tissue repair. Nat Commun, 2023; 14(1):1129; doi: 10.1038/s41467-023-36665-z

19.

Takahashi

, Yamanaka

. Induction of pluripotent stem cells from mouse embryonic and adult fibroblast cultures by defined factors. Cell, 2006; 126(4):663–676; doi: 10.1016/j.cell.2006.07.024

20.

Pang

, Yang

, Vierbuchen

, et al. Induction of human neuronal cells by defined transcription factors. Nature, 2011; 476(7359):220–223; doi: 10.1038/nature10202

21.

Rai

. Transcriptomics revealed differentially expressed transcription factors and MicroRNAs in human diabetic foot ulcers. Proteomes, 2024; 12(4):32; doi: 10.3390/proteomes12040032

22.

Fan

, Zeng

, Yu

. Analysis of mRNA expression profile in the treatment of diabetic foot ulcer healing by tibial cortex transverse distraction. Sci Rep, 2024; 14(1):24865; doi: 10.1038/s41598-024-76291-3

23.

Wang

, Tang

, Zhou

, et al. Identification of CGNL1 as a diagnostic marker in fibroblasts of diabetic foot ulcers: Insights from single cell RNA sequencing and bulk sequencing data. Int J Immunopathol Pharmacol, 2024; 38:3946320241265945; doi: 10.1177/03946320241265945

24.

, Liu

, Zhao

, et al. Single-cell profiling reveals transcriptomic signatures of vascular endothelial cells in non-healing diabetic foot ulcers. Front Endocrinol (Lausanne), 2023; 14:1275612; doi: 10.3389/fendo.2023.1275612

25.

Chen

, Zou

. Combined analysis of single-cell sequencing and bulk transcriptome sequencing reveals new mechanisms for non-healing diabetic foot ulcers. PLoS One, 2024; 19(7):e0306248; doi: 10.1371/journal.pone.0306248

26.

Hao

, Hao

, Andersen-Nissen

, et al. Integrated analysis of multimodal single-cell data. Cell, 2021; 184(13):3573–3587.e29; doi: 10.1016/j.cell.2021.04.048

27.

Stuart

, Butler

, Hoffman

, et al. Comprehensive integration of single-cell data. Cell, 2019; 177(7):1888–1902.e21; doi: 10.1016/j.cell.2019.05.031

28.

Butler

, Hoffman

, Smibert

, et al. Integrating single-cell transcriptomic data across different conditions, technologies, and species. Nat Biotechnol, 2018; 36(5):411–420; doi: 10.1038/nbt.4096

29.

Subramanian

, Tamayo

, Mootha

, et al. Gene set enrichment analysis: A knowledge-based approach for interpreting genome-wide expression profiles. Proceedings of the National Academy of Sciences of the United States of America, 2005; 102(43):15545–15550; doi: 10.1073/pnas.0506580102

30.

Kumar

, Mishra

, Athar

, et al. Inference of gene regulatory network from single-cell transcriptomic data using pySCENIC. Methods Mol Biol, 2021; 2328:171–182; doi: 10.1007/978-1-0716-1534-8_10

31.

Aibar

, González-Blas

, Moerman

, et al. SCENIC: Single-cell regulatory network inference and clustering. Nat Methods, 2017; 14(11):1083–1086; doi: 10.1038/nmeth.4463

32.

, Liu

, Zhang

, et al. The lncRNA SNHG26 drives the inflammatory-to-proliferative state transition of keratinocyte progenitor cells during wound healing. Nat Commun, 2024; 15(1):8637; doi: 10.1038/s41467-024-52783-8

33.

Andreatta

, Carmona

. UCell: Robust and scalable single-cell gene signature scoring. Comput Struct Biotechnol J, 2021; 19:3796–3798; doi: 10.1016/j.csbj.2021.06.043

34.

Phipson

, Sim

, Porrello

, et al. Propeller: Testing for differences in cell type proportions in single cell data. Bioinformatics, 2022; 38(20):4720–4726; doi: 10.1093/bioinformatics/btac582

35.

Mohanty

, Singh

, Kumar

, et al. Vasculogenic skin reprogramming requires TET-mediated gene demethylation in fibroblasts for rescuing impaired perfusion in diabetes. Nat Commun, 2024; 15(1):10277; doi: 10.1038/s41467-024-54385-w

36.

Kamimoto

, Stringa

, Hoffmann

, et al. Dissecting cell identity via network inference and in silico gene perturbation. Nature, 2023; 614(7949):742–751; doi: 10.1038/s41586-022-05688-9

37.

Trapnell

, Cacchiarelli

, Grimsby

, et al. The dynamics and regulators of cell fate decisions are revealed by pseudotemporal ordering of single cells. Nat Biotechnol, 2014; 32(4):381–386; doi: 10.1038/nbt.2859

38.

Qiu

, Mao

, Tang

, et al. Reversed graph embedding resolves complex single-cell trajectories. Nat Methods, 2017; 14(10):979–982; doi: 10.1038/nmeth.4402

39.

Zhu

, Wei

, Zhang

, et al. Mechanism of damage of HIF-1 signaling in chronic diabetic foot ulcers and its related therapeutic perspectives. Heliyon, 2024; 10(3):e24656; doi: 10.1016/j.heliyon.2024.e24656

40.

, Ko

, Li

, et al. A small molecule HIF-1α stabilizer that accelerates diabetic wound healing. Nat Commun, 2021; 12(1):3363; doi: 10.1038/s41467-021-23448-7

41.

Zigrino

, Brinckmann

, Niehoff

, et al. Fibroblast-derived MMP-14 regulates collagen homeostasis in adult skin. J Invest Dermatol, 2016; 136(8):1575–1583; doi: 10.1016/j.jid.2016.03.036

42.

Donovan

, Cogswell

, Sun

, et al. Collagen XII regulates stromal wound closure. Exp Eye Res, 2023; 230:109456; doi: 10.1016/j.exer.2023.109456

43.

Chen

, Fu

, Day

, et al. VEGF amplifies transcription through ETS1 acetylation to enable angiogenesis. Nat Commun, 2017; 8(1):383; doi: 10.1038/s41467-017-00405-x

44.

Mishra

, Ch

, Hong

, et al. Antimicrobial peptide S100A12 (calgranulin C) inhibits growth, biofilm formation, pyoverdine secretion and suppresses type VI secretion system in Pseudomonas aeruginosa. Microb Pathog, 2022; 169:105654; doi: 10.1016/j.micpath.2022.105654

45.

Realegeno

, Kelly-Scumpia

, Dang

, et al. S100A12 is part of the antimicrobial network against mycobacterium leprae in human macrophages. PLoS Pathog, 2016; 12(6):e1005705; doi: 10.1371/journal.ppat.1005705

46.

Fyhrquist

, Muirhead

, Prast-Nielsen

, et al. Microbe-host interplay in atopic dermatitis and psoriasis. Nat Commun, 2019; 10(1):4703; doi: 10.1038/s41467-019-12253-y

47.

Sayed

, Wong

, Ospino

, et al. Transdifferentiation of human fibroblasts to endothelial cells: Role of innate immunity. Circulation, 2015; 131(3):300–309; doi: 10.1161/CIRCULATIONAHA.113.007394

48.

Meng

, Lv

, Chanda

, et al. Reservoir of fibroblasts promotes recovery from limb ischemia. Circulation, 2020; 142(17):1647–1662; doi: 10.1161/CIRCULATIONAHA.120.046872

49.

Slominski

, Zmijewski

. Glucocorticoids inhibit wound healing: Novel mechanism of action. J Invest Dermatol, 2017; 137(5):1012–1014; doi: 10.1016/j.jid.2017.01.024

50.

Wang

, Li

, Xu

, et al. Boosting UPR transcriptional activator XBP1 accelerates acute wound healing. PNAS Nexus, 2023; 2(3):pgad050; doi: 10.1093/pnasnexus/pgad050

51.

Nagarajan

, Chin

, Wang

, et al. Ets1 blocks terminal differentiation of keratinocytes and induces expression of matrix metalloproteases and innate immune mediators. J Cell Sci, 2010; 123(Pt 20):3566–3575; doi: 10.1242/jcs.062240

52.

Alique

, Sánchez-López

, Bodega

, et al. Hypoxia-inducible factor-1α: The master regulator of endothelial cell senescence in vascular aging. Cells, 2020; 9(1):195; doi: 10.3390/cells9010195

53.

Ulasov

, Rosenkranz

, Sobolev

. Transcription factors: Time to deliver. J Control Release, 2018; 269:24–35; doi: 10.1016/j.jconrel.2017.11.004

54.

Rilo-Alvarez

, Ledo

, Vidal

, et al. Delivery of transcription factors as modulators of cell differentiation. Drug Deliv Transl Res, 2021; 11(2):426–444; doi: 10.1007/s13346-021-00931-8

55.

Gallego-Perez

, Pal

, Ghatak

, et al. Topical tissue nano-transfection mediates non-viral stroma reprogramming and rescue. Nat Nanotechnol, 2017; 12(10):974–979; doi: 10.1038/nnano.2017.134

56.

Verma

, Sen

, Srivastava

, et al. Tissue nanotransfection-based endothelial PLCγ2-targeted epigenetic gene editing rescues perfusion and diabetic ischemic wound healing. Mol Ther, 2025; 33(3):950–969; doi: 10.1016/j.ymthe.2025.01.034S1525-0016(25)00038-3. Advance online publication

57.

Bhamidipati

, Hajj

, Ghoneim

, et al. Diabetic wound vasculopathy and neuropathy: Spotlight on wound lipid signaling. Adv Wound Care, 2025; doi: 10.1177/21621918251366681

58.

Gharbia

, Abouhashem

, Moqidem

, et al. Adult skin fibroblast state change in murine wound healing. Sci Rep, 2023; 13(1):886; doi: 10.1038/s41598-022-27152-4

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

1.05 MB

0.04 MB

0.03 MB

0.11 MB

0.06 MB