Human gingival specimens were procured from the Stomatological Hospital of Chongqing Medical University, comprising healthy controls and patients with periodontitis. None of these selected subjects showed any clinical evidence of recent infection or systemic disease, and none had a history of smoking or maxillofacial surgery, radiotherapy or chemotherapy.

C57BL/6 mice (age, 6 weeks old; weight, ≥20 g) were obtained from Chongqing Medical University and housed in a specific pathogen-free environment at constant temperature (22 °C) and humidity (50–60%). The ligature-induced periodontitis model was established by tying 5–0 silk ligatures around the left and right maxillary second molars as previously described (36). The ligatures were kept in position 12 d to promote microbial dental plaque accumulation and inflammation. During this period, the NOX2 inhibitor gp91 ds-tat was injected at 1.5mg/kg into the space between the first molar (M1) and the second molar (M2) using a micro-syringes from the maxillary to buccal side of the oral cavity every 2 d.

The human gingival cell line CA9–22 was purchased from Otwo Biotech. The experimental cells were initially maintained in DMEM medium (Gibco, Invitrogen, Carlsbad, CA, USA) supplemented with 10% fetal bovine serum (FBS, Biological Industries, Israel), 100 U/ml penicillin and 100 μg/ml streptomycin (Beyotime, Jiangsu, China) at 37°C and 5% CO₂.

Transient transfection experiments were performed using Lipofectamine 2000 (Invitrogen) and small interfering RNAs (siRNAs) were designed and synthesized by Gene Greate Bio (Wuhan, China) and were applied for 6 hours. The sequences were as follows: 5’-CUACCUAAGAUAGCGGUUGAUdTdT-3’ and 5’-AUCAACCGCUAUCUUAGGUAG dTdT-3’ for NOX2 siRNA, and 5’-UUCUCCGAACGUGUCACGUTT-3’ and 5’- ACGUGACACGUUCGGAGAATT-3’ for negative control siRNA. CA9–22 cells were then stimulated with 20 μg/mL Pg-LPS for 24h in the presence or absence of the following drugs: 400 μM H₂O₂, 100 μM PDTC (Pyrrolidinedithiocarbamate ammonium, NF-κB inhibitor), 1 μM TAK-242 (Resatorvid; Toll-like receptor 4 (TLR4) inhibitor), 100 nM Tofacitinib (JAK inhibitor) or 3 μM Fedratinib (JAK2 inhibitor) all obtained from MedChemExpress, Monmouth Junction, NJ, USA. All cell experiments were conducted at least three times independently.

Total RNA was extracted using TRIzol (Invitrogen) according to the manufacturer’s instructions and complementary DNA was synthesized using PrimeScriptRT reagent kit (Takara, Shiga, Japan). Quantitative (q)PCR analysis was performed using the CFX96 real-time PCR detection system (Bio-Rad, Hercules, CA, USA) and an ArtiCan^ATM SYBR qPCR Mix (Tsingke Biotech, Beijing, China). Primer sequences for qPCR are listed in Table 1.

The following thermocycling conditions were used for the qPCR: initial denaturation at 95 °C for 5 min; 40 cycles of denaturation at 95 °C for 30 s, annealing at 60 °C for 30 s, and extension at 72 °C for 30 s. Relative expression levels were determined using the 2^-ΔΔCq method and normalized to GAPDH.

Total protein was extracted using RIPA lysis buffer and protein concentrations were determined using a BCA Protein Assay Kit (Beyotime, Jiangsu, China). A total of 20 μg protein was separated by SDS-PAGE and electroblotted onto PVDF membranes. The membranes were incubated with the following primary antibodies overnight at 4 °C: NOX2 (1:2000, Huabio, China), GPX4 (1:10000, Huabio), SLC7A11 (1:1000, Huabio), JAK2 (1:1000, Huabio), STAT3 (1:1000, Huabio), TLR4 (1:1000, Zenbio, China), phospho-JAK2 (1:1000, Zenbio), phospho-STAT3 (1:1000, Zenbio), NF-κB p65 (1:1000, Cell Signaling Technology, Beverley, MA, USA), phospho-NF-κB p65 (1:1000, Cell Signaling), GAPDH (1:10000, Proteintech Biotechnology, China) and α-tubulin (1:10000, Proteintech). Following the primary antibody incubation, membranes were incubated with HRP-conjugated goat anti-rabbit IgG (H + L) (1:10000, Huabio) or HRP-conjugated goat anti-mouse IgG (H + L) (1:10000, Huabio) at room temperature for 1 h. The results were visualized with a chemiluminescence imaging system and further analyzed by ImageJ software Version 2.9.0 (NIH, Bethesda, MD, USA; https://imagej.net/ij/).

Intracellular ROS levels were determined using 2’,7’-dichlorofluorescin diacetate (DCFH-DA, Beyotime). DCFH-DA was diluted with serum-free DMEM to a working solution (1:1000) and cells were incubated with this solution at 37°C for 20 min. After 3 washes with culture medium, cells were observed and photographed using a fluorescence microscope (Nikon, Melville, NY, USA). Finally, the collected cell suspensions were analyze using a fluorescence spectrophotometer (UV-1600, Shanghai Mapada Instruments Co., Ltd.).

A Lipid Peroxidation Assay Kit (Beyotime) was used to evaluate the relative concentration of malondialdehyde (MDA) in tissue and cell lysates following the manufacturer’s instructions. MDA was measured at 532 nm via the thiobarbituric acid (TBA) method. The concentrations were calculated based on the standard curve and normalized to corresponding protein concentration.

Cell viability was assessed using the Cell Counting Kit-8 (CCK-8, Beyotime, China) according to the manufacturer’s instructions. In brief, CA9–22 cells were seeded into 96-well plates for drug treatment experiments. Following incubation, 10 μL CCK8 solution was added and allowed to stand for 1 h at 37°C in the dark. The absorbance at 450 nm was measured using a microplate reader, and the optical density values of different groups were used to determine cell viability.

Paraffin-embedded sections were deparaffinized, subjected to antigen retrieval via microwave heating, and treated with 3% H₂O₂ to block endogenous peroxidase activity. After blocking with 3% bovine serum albumin (BSA), sections were incubated overnight at 4°C with anti-NOX2 primary antibody (1:200, Servicebio, China), followed by incubation with the secondary antibody. Sections were then developed with DAB and counterstained with hematoxylin.

Cellular immunofluorescence experiments utilized CA9–22 cells fixed with 4% paraformaldehyde (PFA) and washed with PBS for three times. The cells were then permeabilized and blocked using 0.5% Triton X-100 and 5% BSA, respectively and then incubated with anti-NOX2 primary antibody (1:500, Proteintech) at 4 °C overnight and with fluorescence dye-conjugated secondary antibodies at 37 °C 1 h. Nuclei were stained with DAPI for 15 min. Images were captured under a fluorescent microscope (Nikon).

Immuno-histo-fluorescence utilized mouse maxillae sections that were incubated with anti-GPX4 (1:200, Huabio) and anti-SLC7A11(1:200, Huabio) primary antibodies overnight at 4°C, followed by incubation with the secondary antibodies. Nuclei were counterstained with DAPI.

Hematoxylin and eosin (H&E) staining utilized paraffin-embedded sections sequentially stained with H&E to visualize nuclei and cytoplasm, respectively. For TRAP staining, the sections were stained using the Trap stain kit (Solarbio, Beijing, China) in accordance with the manufacturer’s instructions. Images were obtained using a microscope slide scanner. Osteoclasts with Trap positivity in the alveolar bone around the maxillary second molar were analyzed using ImageJ software.

The entire maxillary molars of C57BL/6 mice were removed and analyzed using a micro-CT system (Nemo, Pingseng Scientific, China). The rebuilt images of bone surfaces were used to perform three-dimensional histomorphometric analyses using the same density.

We initially consulted a microarray data set that encompassed 241 patients with periodontitis and 69 controls for this study that were obtained from the GEO database (GSE16134) (https://www.ncbi.nlm.nih.gov/geo/) (37, 38). R software was used for data processing, analysis, and figure generation. Using thresholds of |log₂FC| > 0.5 and adj. p < 0.05, differentially expressed genes (DEGs) were visualized as a volcano plot using “ggplot2”, while a heatmap of the DEGs was created with the “pheatmap” package (39, 40). “WGCNA” package was utilized to generate a heatmap depicting the correlations between gene modules and clinical traits (41). A proteomic dataset of gingival crevicular fluid from patients with periodontitis was obtained from the PRIDE database under the accession number PXD046328 (42, 43). Ferroptosis-related genes were retrieved from the FerrDb database (44). The STRING database (http://string-db.org, version 11.5) online tool was used to predict and visualize PPI network models based on the seven screened ferroptosis-related genes found to be active in patients with periodontitis (45). Functional enrichment analysis was performed and visualized using the “GOplot” package in R. The “ggpubr” R package was employed to produce violin plots illustrating gene expression levels in both control and periodontitis samples. Receiver operating characteristic (ROC) curves were generated using the “pROC” package in R.

All results are presented as mean ± standard error of measurement (SEM). Data were analyzed with Prism (Version 8.0, GraphPad, Boston, MA, USA). The t-test and one-way ANOVA were used for statistical analysis of data between two or more groups. p < 0.05 was considered as being statistically significant (ns = not significant, * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001).

We initially examined a large GEO microarray dataset (GSE16134) that encompassed 241 patients with periodontitis and 69 healthy controls and identified differentially expressed genes (DEGs) between the two groups. We found 526 downregulated and 732 upregulated DEGs (Figure 1A). We then applied a WGCNA analysis to correlate DEGs with clinical traits and identifiedfour key modules (blue, light yellow, magenta, dark red) with correlation coefficients ≥0.75 (Supplementary Figures S1A–E). We additionally analyzed a proteomic dataset (PXD046328) that included 1536 gingival crevicular fluid proteins from periodontitis patients (42, 43) as well as ferroptosis-related genes that were retrieved from the FerrDb database. These data were intersected and this resulted in the identification of 7 genes that overlapped between the test categories (Figure 1B). Simultaneously, we constructed a protein-protein interaction network analysis on these genes using STRING database (Figure 1C). Subsequent GO enrichment analysis of the 7 genes revealed significant categories related to the following: biological processes (negative regulation of nucleocytoplasmic transport, reactive oxygen species metabolic process and respiratory burst); cellular components (NADPH oxidase complex) and molecular function (superoxide-generating NAD(P)H oxidase activity) (Figures 1D–F). These findings indicated a crucial role of oxidative stress and NADPH oxidase-mediated mechanisms in periodontitis. We therefore focused on cybb (encoding NOX2) that was upregulated in periodontitis (Figure 1G) and that demonstrated a potential for use as a diagnostic tool (AUC = 0.797; Figure 1H).

We further investigated whether NOX2 played a role in the pathology of periodontitis and examined its expression across clinical and experimental models. In human gingival tissues, NOX2 protein levels were significantly elevated in patients with periodontitis compared to healthy controls. (Figures 2A, B). In a mouse model of periodontitis, both NOX2 mRNA and protein expression were markedly increased (Figures 2C–E). In gingival epithelial cells stimulated with Pg-LPS, NOX2 expression was upregulated at both the transcriptional and translational levels in a concentration-dependent manner (Figures 2F–J). These results indicated that NOX2 expression was upregulated in periodontitis. Our resultsdemonstrated that Pg-LPS stimulation elevated TLR4 and p-p65 levels in a concentration- andtime-dependent manner similar to two other studies (46, 47) (Supplementary Figures S2A, B). To experimentally confirm a linkage between NOX2 expression and periodontitis, we treatedCA9–22 cells with Pg-LPS and examined NOX2 protein expression via Western blotting. Pg-LPSaddition induced NOX2 expression while pretreatment with the TLR4 inhibitor TAK-242 or the NF-κB inhibitor PDTC both significantly suppressed this upregulation (Supplementary Figures S2C, 2K). Together, these findings demonstrated that Pg-LPS upregulated NOX2 expression via the TLR4/NF-κB signaling pathway.

Human gingival tissue samples from periodontitis patients (GSE16134) were divided based on median NOX2 expression to enable gene expression profile analysis (Figures 3A, B). GSEA identified significant enrichment of “Ferroptosis” in tissues with high NOX2 expression indicating a correlation between NOX2 and ferroptosis in periodontitis (Figure 3C). To explore the biological role of NOX2, we used siRNA gene silencing to target the NOX2 gene in CA9–22 cells. We found significant suppression of NOX2 expression following NOX2 siRNA transfection (Figures 3D, E). To assess the intracellular labile iron levels, we found that intracellular Fe²⁺ increased significantly following Pg-LPS stimulation and knockdown of NOX2 reduced Fe²⁺ levels compared to the si-NC group (Figure 3F). Moreover, Pg-LPS induction significantly increased MDA levels compared with controls and the effect was reversed upon transfection with si-NOX2 (Figure 3G). Pg-LPS also significantly raised cellular ROS levels and these could be reduced by downregulating NOX2 with siRNA (Figures 3H, I). Additionally, data from Western blotting demonstrated that GPX4 and SLC7A11 expression levels in Pg-LPS induced CA9–22 cells were conspicuously reduced vs. controls and these effects were all reversed following transfection with si-NOX2 (Figures 3J–M). These findings demonstrated that NOX2 silencing suppressed Pg-LPS-induced ferroptosis in CA9–22 cells.

GSEA also identified significant enrichment of the “JAK-STAT signaling pathway” in tissues with high NOX2 expression (Figure 4A) and indicated a correlation between NOX2 and JAK-STAT signaling in periodontitis.Considering the active involvement of the JAK2-STAT3 signaling pathway in ferroptosis andperiodontitis pathogenesis (48, 49), we further investigated whether NOX2 silencing provided a protective mechanism. We therefore examined JAK2 and STAT3 and phospho-JAK1 levels in Pg-LPS stimulated CA9–22 cells. JAK1 phosphorylation was not significantly activated confirming that this JAK isoform was not involved and further supporting the specificity of Pg-LPS in activating JAK2 (Supplementary Figure S3A). In contrast, NOX2 silencing significantly reduced Pg-LPS-induced increases in the levels of phosphorylated JAK2 and STAT3 (Figures 4B–D). Additionally, since NOX isoforms play central roles in catalytic ROS generation, we soughtto determine whether NOX2-mediated ROS was responsible for these alterations in JAK2-STAT3activation. We first examined the expression levels of other NOX subtypes (excluding NOX2) in response to Pg-LPS stimulation and we found no significant changes in any of these genes indicating that the ROS induced by Pg-LPS primarily originates from NOX2 (Supplementary Figure S4). Furthermore, we treated CA9–22 cells with exogenous H₂O₂ to mimic ROS effects in vitro. As H₂O₂ levels increased, the levels of the ferroptosis-related proteins GPX4 and SLC7A11 declined, while phosphorylation levels of JAK2 and STAT3 were elevated (Figure 4E). This implicated ROS-induced activation of the JAK2–STAT3 pathway. To verify therole of JAK2-STAT3 in ferroptosis, the JAK2-specific inhibitor Fedratinib and JAK inhibitorTofacitinib were employed (Supplementary Figure S5A). Fedratinib effectively inhibited p-JAK2 and p-STAT3 expression (Supplementary Figure S5B). In H₂O₂-treated CA9–22 cells, GPX4 and SLC7A11 expression levels were markedly decreased. These effects were reversed by pretreatment with either Fedratinib or Tofacitinib (Figures 4F, G). These findings demonstrated that NOX2 mediates ferroptosis in vitro by producing ROS and subsequently modulating the JAK2–STAT3 pathway.

To elucidate a role for NOX2 in periodontitis in vivo, we administered the NOX2 inhibitor gp91 ds-tat in the experimentally-induced mouse model of periodontitis (Figure 5A). Immunohistochemistry staining revealed a significant increase in NOX2 in the ligature group vs. controls while gp91 ds-tat treatment reduced NOX2 expression (Figure 5B). Lipid peroxidation assays also indicated that MDA levels were elevated in periodontitis mice and this could be attenuated by NOX2 inhibition (Figure 5C). The ligature-induced mice also exhibited reduced GPX4 and SLC7A11 protein levels vs. controls and gp91 ds-tat treatment reversed these effects (Figures 5D–F). These findings were further corroborated using immunofluorescence staining of the cells(Supplementary Figures S6A–D).

Together, these findings indicated that NOX2 inhibition reduced ferroptosis in mice with periodontitis. We next investigated whether a down-regulation of NOX2 could mitigate periodontitis. The gp91 ds-tat treatment group had a thicker periodontal epithelial fibrous layer, better alveolar bone height and morphology and fewer infiltrating inflammatory cells than did the ligation group (Figure 5G). Since osteoclasts are the primary participant in bone resorption, we examined whether there was an imbalance in the osteoclast population. We found fewer osteoclasts in periodontal tissue in the gp91ds-tat group than in the ligation group using TRAP staining (Figures 5H, I). A micro-CT analysis also indicated obvious alveolar bone resorption in the ligation group compared with controls and interestingly, gp91 ds-tat treatment reversed this condition (Figure 5J). Furthermore, ligation induced significant increases in both cementoenamel junction-alveolar bone crest (CEJ-ABC) distance and trabecular spacing (Tb.Sp) while the bone volume fraction (BV/TV) was decreased. These pathological changes were effectively normalized by gp91 ds-tat administration where CEJ-ABC measurements and Tb.Sp were decreased while BV/TV levels were increased (Figures 5K–M). Taken together, these findings indicated that NOX2 inhibition may confer a protective effect by suppressing ferroptosis in the ligature-induced periodontitis model.