Human oral keratinocytes (HOKs) were purchased from Shanghai QuiCell Biotechnology Co., Ltd. The item number is QuiCell-Y1694. These were STR-certified and free of mycoplasma. Following resuscitation in cell culture flasks, the cells were incubated at 37 °C with 5% CO₂ for 3–4 hours to stabilize their condition. The complete culture medium consisted of DMEM (11965092, Thermo Fisher Scientific, Waltham, MA, USA) supplemented with 10% fetal bovine serum, 100 U/mL penicillin, and 0.1 mg/mL streptomycin. Upon reaching 70% confluency, the cells were detached using 0.25% trypsin and passaged at a 1:2 ratio [18].

Resuscitated HOKs were seeded into six-well culture plates, and 2 mL of complete medium (DMEM supplemented with 10% fetal bovine serum, penicillin, and streptomycin) was added to each well. The cells were cultured in an incubator at 37 °C and 5% CO₂ until they reached 70%–80% confluence. Treatment groups were exposed to 0, 1, and 10 µM concentrations of retinoic acid (RA), or 0, 100, and 500 nM concentrations of BMS493. The drug-containing medium was replaced every two days to ensure stable drug concentrations. On day 9 of culture, cell samples were collected for experiments, including Western blotting to detect changes in the expression of keratinization-related markers.

An siRNA that targets TNFRSF21 was designed and synthesized by TsingKe Biological (Beijing, China). The adenovirus vector GV315 containing TNFRSF21 (NM_014452.5) was purchased from TsingKe Biological (Beijing, China). Transfection was performed using Lipofectamine 2000 (TsingKe Biological) in accordance with the manufacturer’s instructions. All sequence details are provided in Supplementary Table 1.

All animal work was performed in accordance with the ARRIVE guidelines 2.0. The experimental procedures were approved by the Laboratory Animal Care and Use Committee at Anhui Medical University (Approval No. LLSC20232087). Eight C57BL/6 mice aged 8 months were randomly and equally divided into two groups via a simple randomization method using a random number generator. The first group (negative control) received an empty vector plasmid (Vector), while the second group was treated with a plasmid for the overexpression of TNFRSF21 (OE-TNFRSF21). Under surgical anesthesia induced by 2% sodium pentobarbital (50 mg/kg), all mice underwent full-thickness excisional wounds on their dorsal sides. Following hair removal on the back, two 6-mm full-thickness wounds were made along the midline of the back of each mouse using a sterile biopsy punch. Immediately after wounding, the mice were given multiple subcutaneous injections of either the empty vector or the OE-TNFRSF21 plasmid (four injections per wound site), with each wound receiving a single treatment. Additionally, by adding the opioid tramadol (0.1 mg/mL) to the drinking water to alleviate pain related to wounds [19]. On day 9 postoperatively, mice were euthanized via cervical dislocation following induction with 5% isoflurane anesthesia. The dorsal wound tissues were harvested for histological examination and fixed in a 4% paraformaldehyde solution, along with visceral organs, including the heart, liver, spleen, lungs, and kidneys. Subsequently, quantitative histomorphometric analysis was conducted on these tissues to evaluate the progress of wound healing and the overall systemic biocompatibility associated with plasmid therapy.

Patients were classified based on clinical and radiographic (X-ray) criteria. Periodontal tissues were collected during tooth extraction or ridge repair, placed in sterile PBS tubes, and transported to the laboratory within 10 minutes for single-cell preparation and RNA sequencing. The periodontal therapy (PDT) patient group (n = 3) showed minimal gingival inflammation after treatment, but had severe clinical attachment and bone loss ($>$60%) [20]. Single-cell RNA sequencing (scRNA-seq) was performed by NovelBio Bio-Pharm Technology Co., Ltd, employing the NovelBrain Cloud Analysis Platform as described previously [21, 22]. scRNA-seq data were obtained from NCBI GEO (GSE171213). Raw sequencing data were processed with Cell Ranger (v7.1.0, 10x Genomics, Pleasanton, CA, USA) for alignment, barcode assignment, and unique molecular identifier (UMI) counting. Rigorous quality control (QC) steps were applied, with cells expressing 200 genes, cells with $>$20% mitochondrial reads, and potential doublets identified via the DoubletFinder algorithm (v2.0.3, Christopher S. McGinnis, University of California, San Francisco, CA, USA) excluded from downstream analyses. Cell normalization and regression analysis were performed using the Seurat package (v2.3.4, Satija Lab, New York Genome Center, New York, NY, USA) based on UMI counts and the proportion of mitochondrial genes, resulting in scaled expression data [23]. To mitigate potential batch effects that could compromise analytical accuracy, the Harmony package was employed using the top 3000 highly variable genes under default parameters [24]. Unsupervised clustering was conducted using a graph-based algorithm with a resolution of 0.8, based on the top 10 principal components obtained from principal component analysis (PCA). Marker genes were identified using the FindAllMarkers function with the Wilcoxon rank-sum test, applying the following criteria: log-fold change (lnFC) $>$0.25, p-value 0.05, and minimum expression percentage (min.pct) $>$0.1 [20]. After QC, a total of 622 single cells were retained. Clustering and dimensionality reduction were conducted using Seurat with t-Distributed Stochastic Neighbor Embedding (t-SNE) [25]. Ligand-receptor interaction analysis was performed using the CellPhoneDB toolkit to construct an interaction network of TNFRSF members and their ligands in oral epithelial cells. The String database (v12.0, STRING Consortium (SIB), Geneva, Switzerland) was used for visual analysis [26]. Bulk RNA-seq analysis was conducted on three samples from each group: negative control (NC), keratinization inhibition induced by RA treatment (10 µM), and keratinization promotion induced by BMS493 treatment (500 nM). Raw sequencing reads were aligned to the human genome (GRCh38) using STAR (v2.7.10a, Alexander Dobin, Cold Spring Harbor Laboratory, NY, USA). Differential expression analysis was performed using DESeq2 (v1.38.3, Simon Anders, European Molecular Biology Laboratory, Heidelberg, Germany) [27]. Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses were conducted using clusterProfiler (v4.8.1, Guangchuang Yu, Southern Medical University, Guangzhou, Guangdong, China) [28]. Differentially expressed genes (DEGs) were identified using a threshold of absolute log₂-fold change $>$1 and an adjusted p-value (FDR) 0.05.

Gingival tissues collected from clinical samples of healthy volunteers or periodontitis patients were homogenized in RIPA lysis buffer (1$\times$, P0013B, Beyotime Biotechnology, Shanghai, China) for protein extraction. The lysates were incubated on ice for 20 min with agitation, followed by centrifugation at 13,000 rpm and 4 °C for 10 min. The supernatant was transferred to a new 1.5 mL microcentrifuge tube, mixed 1:1 with 2$\times$ SDS-PAGE loading buffer (P0015B, Beyotime, Shanghai, China) containing protease and phosphatase inhibitors, boiled at 95 °C for 10 min, and stored at –20 °C or directly subjected to Western blot analysis. Protein samples were separated on a Bis-Tris Gel (12%, 20214ES08, Yeasen Biotechnology, Shanghai, China), and then transferred to a 0.45 µm pore size PVDF membrane (0.45 µm, IPVH00010, Merck, Darmstadt, Germany) in ice-cold transfer buffer for 2 h. After blocking with 5% non-fat milk for 1 h at room temperature, the membrane was incubated overnight at 4 °C with the primary antibody. After washing four times with TBST (Tris-buffered saline with Tween 20) for 7 min, the membrane was incubated with HRP-conjugated secondary antibodies (1:5000, SA00001-1/SA00001-2, Proteintech, Rosemont, IL, USA) for 1 h at room temperature, then washed four times with TBST (5 min each). The membrane was imaged using a CCD camera to visualize the immunoreactive bands following enhancement with chemiluminescence (Tanon-5200, Tanon Science & Technology Co., Ltd., Shanghai, China). GAPDH (Glyceraldehyde-3-Phosphate Dehydrogenase) was utilized as the housekeeping gene. Densitometric analysis was performed on all bands using GAPDH as the reference for normalization.

Protein A/G beads (50 µL per reaction, HY-K0202, MedChemExpress, Monmouth Junction, NJ, USA) were washed with an appropriate amount of IP buffer, then centrifuged, and the supernatant discarded. TNFRSF21 or IgG antibody (2 µg) was added before incubation for 4–6 h at 4 °C to couple the antibody with the beads. The supernatant from lysed HOKs was then collected, and 100 µL was taken as the Input group. The remaining cell lysate was added to the pre-treated beads and incubated at 4 °C overnight to allow the TNFRSF21 protein to bind with the specific antibody. Next, the beads were washed three times with lysis buffer to remove unbound proteins. Loading buffer (5$\times$) was subsequently added and the sample heated at 95 °C for 15 minutes for analysis by Western blotting, or storage at –20 °C. Western blotting was performed using standard protocols, with the antibodies listed in Supplementary Table 1.

Tissue sections (4 µm) were subjected to H&E staining using standardized protocols. Briefly, deparaffinized sections were stained with Mayer’s hematoxylin (4 min) and then differentiated (1% acid ethanol, 3 sec), blued (0.2% ammonia water), and counterstained with eosin Y (3 min). After dehydration and mounting, the slides were imaged using a Nikon Eclipse E100 microscope (Nikon Eclipse E100, Nikon Instruments Inc., Tokyo, Japan).

Tissue sections were deparaffinized in xylene (2 $\times$ 15 min) and then rehydrated through a graded ethanol series (100%, 100%, 85%, and 75% for 5 min each) followed by rinsing in distilled water. Antigen retrieval was performed in citrate buffer (10 mM, pH 6.0) using microwave heating (95 °C, 20 min). After cooling, slides were washed in PBS (3 $\times$ 5 min).

Tissue sections were encircled with a hydrophobic barrier, treated with an autofluorescence quencher (G1221, Servicebio, Wuhan, Hubei, China) for 5 min, and rinsed under running water (10 min). Non-specific binding was blocked with 3% BSA/PBS (30 min, RT). Primary antibodies (diluted in PBS) were applied and incubated overnight at 4 °C in a humidified chamber. After PBS washes (3 $\times$ 5 min), fluorophore-conjugated and species-matched secondary antibodies (Servicebio, Wuhan, Hubei, China) were applied (50 min, RT, dark). Nuclei were counterstained with DAPI (1 µg/mL, 10 min, dark). Sections were then mounted with anti-fade medium (Servicebio, G1401) and imaged using a fluorescence microscope (Nikon Eclipse Ni-E, Tokyo, Japan) with standard filter sets (DAPI: Ex 330–380 nm/Em 420 nm; FITC: Ex 465–495 nm/Em 515–555 nm; Cy3: Ex 510–560 nm/Em 590 nm).

To investigate whether TNFRSF regulates the healing process after treatment for severe periodontitis, we analyzed single-cell RNAseq of gingival epithelial tissues collected after initial PDT [20]. t-SNE cluster analysis performed on the scRNA-seq data of oral epithelial cells revealed 7 distinct cell clusters (Fig. 1A). The differentiation trajectory was analyzed and visualized with partition-based graph abstraction (PAGA) (Fig. 1B). KEGG enrichment analysis revealed significant enrichment of the TNF signaling pathway within the Basal_V and Pro_KC cell clusters. In addition, signaling pathways related to tight junctions, inflammatory regulation, and interaction with extracellular mechanisms were also significantly enriched (Fig. 1C). Subsequently, we aggregated the expression data of TNFRSF across all cell clusters and observed that TNFRSF21 was prominently expressed in each cluster. This consistent expression pattern indicates that TNFRSF21 functions as a pivotal marker in modulating the healing processes of periodontitis (Fig. 1D).

Fig. 1.

Tumor necrosis factor receptor superfamily 21 (TNFRSF21) is a key molecule during periodontal regeneration and repair. (A) The main cell types present in the gingival epithelium after periodontal treatment were identified by the Seurat R package and visualized by t-Distributed Stochastic Neighbor Embedding (t-SNE). (B) Partition-based graph abstraction (PAGA) was employed to analyze and visualize the differentiation trajectories. (C) Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis of Basal_V and Pro_KCs showed significant activation of the TNF signaling pathway. (D) Violin plot illustrating the expression levels of selected members of the TNFRSF family during the periodontitis healing process.

To further examine specific gene expression in periodontal epithelial tissues, we compared the expression of TNFRSF21 and KRTs between individuals with periodontitis and healthy controls. The expression levels of both TNFRSF21 and KRTs were found to be significantly reduced in the periodontitis group compared to the control group (Fig. 2A). Subsequently, we established an in vitro concentration gradient model for inhibiting and promoting keratinization in HOK cells using RA, a small molecule drug, and its inhibitor BMS493, respectively [29, 30, 31]. This model was designed to simulate the processes of incomplete keratinization in epithelial tissues during the onset of periodontitis, and of re-keratinization following treatment and recovery. Western blot analysis confirmed that expression levels for TNFRSF21 and KRTs (including KRT8 and KRT18) were significantly reduced in the keratinization inhibition group (RA), whereas they were significantly increased in the keratinization promotion group (BMS493; Fig. 2B). Differential gene analysis of bulk RNA sequencing showed that RA treatment reduced the expression of keratin proteins such as KRT9, KRT16, and KRT17, while increasing the expression of chemokines like C-C Motif Chemokine Ligand 2 (CCL2). Furthermore, bulk RNA sequencing data revealed excellent intra-group homogeneity and clear inter-group separation across the NC, RA, and BMS493 groups (Fig. 2C). Gene Ontology (GO) functional enrichment analysis revealed the DEGs were significantly correlated with cell differentiation, immune regulation, epithelial migration, and keratinization (Fig. 2D). These findings imply that epithelial regeneration, reduction of inflammation, and the critical regulatory role of keratinization are integral to both the pathogenesis of periodontitis and its therapeutic recovery.

Fig. 2.

Changes in TNFRSF21 expression during the keratinization process, as revealed by the in vitro keratinization induction model. (A) Western blot analysis showed a significant reduction in the expression levels of keratin and TNFRSF21 in the gingival epithelial tissues of periodontitis patients (PD) compared to healthy controls (HP). (B) In vitro models showed decreased TNFRSF21 and keratin expression in the keratinization inhibition group (treated with retinoic acid (RA)), and increased expression in the keratinization promotion group (treated with BMS493). (C) Bulk RNA sequencing revealed high intra-group consistency and clear segregation between the negative control (NC), RA, and BMS493 groups. (D) GO enrichment analysis of differentially expressed genes identified by bulk-RNA sequencing.

To further investigate the potential regulatory role of TNFRSF21, we employed gene knockdown and overexpression techniques to suppress and enhance its expression. Validation of the knockdown efficiency revealed markedly decreased TNFRSF21 protein expression (Fig. 3A). We also examined alterations in the expression levels of target KRTs, including KRT8, KRT18, and Claudin-1 (Fig. 3B). Subsequently, we constructed a plasmid for TNFRSF21 overexpression and transfected it into HOK cells. This resulted in increased expression of TNFRSF21, as well as increased expression levels of KRT8, KRT18, and Claudin-1 (Fig. 3C). Together, these findings support a regulatory role of TNFRSF21 in keratin expression by HOK cells.

Fig. 3.

Knockdown and overexpression of TNFRSF21 demonstrate its regulatory role during keratinization. (A) Verification of TNFRSF21 knockdown efficiency with siRNA. (B) Knockdown of TNFRSF21 reduced the expression of KRTs. (C) Overexpression of TNFRSF21 significantly increased the expression of KRTs. KRTs, keratinization-related genes.

TNFRSF is a class of cell surface receptor that typically mediates physiological functions upon binding to its ligands [12]. We next investigated interactions between the TNFRSF family and their ligands during treatment for periodontitis and subsequent healing. Through protein–protein interaction (PPI) analysis of scRNA-seq data (Fig. 4A), we identified a significant interaction between TNFRSF21 and amyloid precursor protein (APP). Further validation via STRING database analysis of DEGs confirmed this interaction (Fig. 4B). Additionally, an immunoprecipitation (IP) assay provided direct evidence for the interaction between TNFRSF21 and APP in HOK cells (Fig. 4C). These observations confirm the strong interaction between TNFRSF21 and its ligand APP throughout the periodontitis recovery phase. Moreover, we speculate that the complex formation between TNFRSF21 and APP may play a synergistic role in modulating the functions of oral epithelial keratinocytes.

Fig. 4.

Interaction of TNFRSF21 and amyloid precursor protein (APP) during keratinization. (A) Protein–protein interaction (PPI) analysis revealed that TNFRSF21 and APP undergo significant interaction. The dashed box highlights that APP and TNFRSF21 have the strongest PPI within the TNFRSF family. (B) This interaction was further validated using the String database. (C) Western blot analysis confirmed the interaction between TNFRSF21 and APP in human oral keratinocytes (HOKs) cells.

We next conducted animal experiments to further investigate the role of TNFRSF21 in promoting keratinization and wound healing. Using a puncher, we created a 6-mm diameter wound on the back of C57BL/6 mice and subsequently injected plasmids into the wound site that contained either an empty vector, or a vector overexpressing TNFRSF21. Wound conditions were monitored and documented on days 1, 3, 5, 7, and 9 post-injection (Fig. 5A). On day 9, a biopsy of the wound tissue was performed, followed by hematoxylin and eosin (H&E) staining and immunofluorescence analysis. Our findings revealed that the plasmid overexpression group had approached full wound closure by day 9. In contrast, the control group still presented with a substantial wound area at day 9. Additionally, mice treated with the TNFRSF21-overexpressing plasmid showed enhanced wound healing, characterized by a more rapid decrease in wound size and elevated keratin expression levels compared to the control group (Fig. 5B,C,E). Systemic effects of the overexpressing plasmid on the heart, liver, spleen, lung, and kidney tissues were also investigated by H&E staining, but no evidence of toxicity was observed (Fig. 5D).

Fig. 5.

Overexpression of TNFRSF21 promotes keratinized wound healing in mouse dorsal skin. (A) Diagram of the experimental design used to evaluate skin wound healing in C57BL/6 mice. On day 0, a 6-mm full-thickness skin incision was created using a biopsy punch as the wound model, followed by the injection of either empty vector or overexpression plasmids into the wound site. (B) Representative hematoxylin and eosin (H&E)-stained skin sections on day 9 after the injection of empty vector or overexpression plasmid. Scale bar = 200 µm. (C) Photographs of the wound area in C57BL/6 mice treated with either an empty vector or an overexpression plasmid on days 0, 3, 5, and 9 during the wound healing process. Scale bar = 2 mm. (D) Representative H&E-stained sections of the heart, liver, spleen, lung, and kidney on day 9 after injection with empty vector or overexpression plasmid. Scale bar = 200 µm. (E) Skin sections stained by immunofluorescence on day 9 after injection of empty vector or overexpression plasmid. The field of view is 5$\times$, scale bar = 500 µm. The yellow boxes are enlarged views. The field of view is 10$\times$, scale bar = 250 µm.