Single-cell atlas of human gingiva unveils a NETs-related neutrophil subpopulation regulating periodontal immunity

Graphical abstract

Highlights

• •Single-cell RNA sequence analysis revealed the heterogeneity of neutrophils in human gingival tissues.
• •A NETs-related neutrophil subset was identified in gingival tissues for the first time.
• •The promotion of gingival inflammation and alveolar bone resorption in severe periodontitis by NETs was confirmed in vivo.
• •We built prediction model for periodontitis based on key NETs-related genes though six types of machine learning methods.
• •Gingival fibroblasts act on NETs-related neutrophil subset to aggravate periodontal immunopathology via MIF-CD74/CXCR4.

Abstract

Introduction

Exaggerated neutrophil recruitment and activation are the major features of pathological alterations in periodontitis, in which neutrophil extracellular traps (NETs) are considered to be responsible for inflammatory periodontal lesions. Despite the critical role of NETs in the development and progression of periodontitis, their specific functions and mechanisms remain unclear.

Objectives

To demonstrate the important functions and specific mechanisms of NETs involved in periodontal immunopathology.

Methods

We performed single-cell RNA sequencing on gingival tissues from both healthy individuals and patients diagnosed with periodontitis. High-dimensional weighted gene co-expression network analysis and pseudotime analysis were then applied to characterize the heterogeneity of neutrophils. Animal models of periodontitis were treated with NETs inhibitors to investigate the effects of NETs in severe periodontitis. Additionally, we established a periodontitis prediction model based on NETs-related genes using six types of machine learning methods. Cell-cell communication analysis was used to identify ligand-receptor pairs among the major cell groups within the immune microenvironment.

Results

We constructed a single-cell atlas of the periodontal microenvironment and obtained nine major cell populations. We further identified a NETs-related subgroup (NrNeu) in neutrophils. An in vivo inhibition experiment confirmed the involvement of NETs in gingival inflammatory infiltration and alveolar bone absorption in severe periodontitis. We further screened three key NETs-related genes (PTGS2, MME and SLC2A3) and verified that they have the potential to predict periodontitis. Moreover, our findings revealed that gingival fibroblasts had the most interactions with NrNeu and that they might facilitate the production of NETs through the MIF-CD74/CXCR4 axis in periodontitis.

Conclusion

This study highlights the pathogenic role of NETs in periodontal immunity and elucidates the specific regulatory relationship by which gingival fibroblasts activate NETs, which provides new insights into the clinical diagnosis and treatment of periodontitis.

Introduction

Neutrophils are the most abundant innate immune cells in oral-barrier tissues and are essential for the homeostatic regulation of periodontal immunity [1]. Excessive recruitment or activation of neutrophils leads to dysbiosis between periodontal microbes and the host immune response, which can ultimately elicit periodontitis [2]. In contrast, in patients with single gene mutations related to granulopoiesis or defective neutrophil recruitment/extravasation, it has been reported that neutrophil deficiency or functional defects can result in the early development of severe periodontitis [3]. Nevertheless, the mechanisms by which neutrophils contribute to the pathology of periodontitis are not fully understood.

Neutrophil extracellular traps (NETs) have been at the forefront of research on neutrophils in recent years. NETs are extracellular mesh-like structures composed of DNA and a variety of cytosolic and granule proteins that play a critical role in immune defense, inflammatory, autoimmune diseases, and cancer [4]. Numerous studies have revealed that patients with periodontitis have significantly increased levels of NETs in tissues such as the gingiva, crevicular exudate, and peripheral blood [5], [6], [7]. Notably, NETs represent a double-edged sword, as they can not only capture and eliminate pathogens, but their overaccumulation can exacerbate disease progression [8], [9], [10], [11]. NETs and extracellular histones (a vital component of NETs) can serve as early triggers for pathogenic inflammation and IL-17/Th17 immunopathology in periodontitis [11]. However, the crucial function and mechanisms of NETs in the development of periodontitis remain to be explored.

Single-cell RNA sequencing (scRNA-seq) is a mainstream single-cell sequencing technology that reveals the status and characteristics of different cells in tissues or organs through high-dimensional mapping of the transcriptome with single-cell resolution [12]. Several studies have mapped a single cell atlas of tissues such as the mucosa, as well as alveolar bone and peripheral blood mononuclear cells, which has greatly enriched the theory of periodontitis pathology and provided new perspectives on the role and mechanisms of neutrophils in periodontitis [13], [14], [15], [16], [17]. ScRNA-seq of gingival tissues from healthy individuals and those with periodontitis has suggested that neutrophils are crucial in oral mucosal immunity and can be specifically recruited by epithelial and stromal cells [13]. Additionally, a study of periodontitis mice with inflammatory bowel disease identified the function of the Prdx1 neutrophil subpopulation with proinflammatory phenotype infiltration [17]. Nevertheless, the heterogeneity of neutrophils in periodontitis awaits further investigation.

It is thus possible to analyze neutrophil heterogeneity and explore the role and mechanisms of NETs in periodontitis at the subpopulation level through single-cell RNA sequencing. Herein, we profiled a single-cell atlas of gingival tissues from both healthy individuals and patients diagnosed with periodontitis, analyzed neutrophil heterogeneity, and defined the NETs-related neutrophil subpopulation (NrNeu) for the first time. Subsequently, we determined the function of NETs in periodontitis by in vivo experiments and further constructed prediction models for periodontitis based on key NETs-related genes (NrGs). Finally, cellular communication analysis was used to reveal that gingival fibroblasts facilitate NETs-related neutrophil subsets to generate increased NETs production in periodontitis. Our study broadens the theories of periodontal immunity and offers new perspectives for the prevention and control of periodontitis.

Materials and methods

Biopsy collection

The specimen collection and analysis procedures were approved by the Ethics Committee of Southern Medical University (Approval No: NFEC-2022-296). Standardized gingival collar biopsies measuring 5 mm in length and 3 mm in width were obtained from donors under local anesthesia. Gingival biopsies from healthy individuals were harvested from areas without BOP that met the criteria for good oral health and had a probing depth (PD) < 3 mm. Biopsies from periodontitis patients were obtained from areas exhibiting severe inflammation and bone loss (BOP positive and PD > 5mm). The other inclusion criteria were: 1) no smoking, 2) no systemic diseases, 3) no intake of antibiotics or anti-inflammatory medications in the past 3 months, 4) no periodontal therapy within the last 6 months, 5) no pregnancy or breastfeeding, 6) no acute infections or allergies, and 7) no immunosuppressant treatment in the past 3 months. Healthy gingival tissue specimens were obtained during crown lengthening procedures, while biopsies from periodontitis patients were collected prior to the extraction of teeth with no retention value. Additional information (age, gender and other details) about donors is provided in Table S1.

Single-cell RNA sequencing

Single-cell suspension preparation

Three healthy and three periodontitis tissues were minced and digested with 2 mg/mL collagenase II (2275MG100, Biofroxx), 2 units/mL Dispase II, and 0.1 μg/mL DNase-I. Samples were incubated for 50 min at 37 °C, with shaking every 5 min on a shaker. After lysing erythrocytes and filtering, the single-cell suspensions were stained for viability assessment using Countess II. Subsequent experiments were continued using cells with activity of at least 80 %.

10X Genomics

Cellular suspensions were loaded on a 10X Chromium Controller (10X Genomics), and library preparation was performed according to the instructions of the 10X Chromium Next GEM Single Cell Library Kit v3 (10X Genomics). The sequencing strategy of Illumina Hiseq Novaseq PE150 was applied to detect a sequencing depth of ∼ 120G. The filtered feature barcode matrices produced by the Cell Ranger pipeline were employed as input, following conversion to a Seurat-compatible format using the ‘Read10X’ and ‘CreateSeuratObject’ functions. All subsequent processes were executed using the Seurat (v4.4.0) R package [18].

Data quality control

To ensure data quality, Cell Ranger was employed to distinguish reads from each cell using barcoding. Following filtering and processing, the number of cells, cell reads, and genes in the sample were statistically determined, eliminating heterogeneous cell populations with extremely distinct RNA contents. Subsequently, cells exhibiting an nFeature_RNA count between 200 and 2500 and a mitochondrial content of <15 % were selected for subsequent analysis.

Data integration, transformation, and standardization

For enhanced interpretability, the “NormalizeData” function from the Seurat package was employed for data normalization. Subsequently, the “FindVariableFeatures” function was used with a parameter setting of nfeatures = 2000 to identify highly variable genes. Following this, the “ScaleData” function and “RunPCA” (with the selection of PC equal to 15) were applied for dimensionality reduction using the FindNeighbors approach. Simultaneously, the “RunHarmony” function from the Harmony package was employed to mitigate batch effects.

Investigation of neutrophil heterogeneity

To further investigate the mechanistic role of neutrophils in the gingival tissues of patients with periodontitis, we employed dimensionality reduction clustering of neutrophils. Subsequently, differential gene analysis using FindAllMarkers was conducted to identify distinct genes within each neutrophil subgroup. Subsequently, the R package WGCNA was used for module gene analysis. We identified three modules in neutrophil genes, randomly distinguished by brown, blue and turquoise colours. Enrichment analysis was performed on the intersection of module genes associated with neutrophil subgroups and differentially expressed genes (DEGs) within each neutrophil subgroup using Metascape software (https://metascape.org/gp/index.html#/main/step1), leveraging the KEGG database. To explore the temporal relationships governing disease progression in individual neutrophil subgroups, we employed the monocle package for pseudotemporal analysis.

Identifying key genes associated with NETs using machine learning

The neutrophil profile (total 4019 cells) from our scRNA-seq data was used as a model development and internal validation dataset (IV). This dataset was randomly sampled 70 % (2814 cells) as internal training set (IV-T) and 30 % (1205 cells) as internal validation set (IV-V). For the external validation (EV), we used periodontitis dataset from the Gene Expression Omnibus database (access no: GSE16134, platform: GPL570), which includes 69 healthy samples and 241 periodontitis samples. This dataset was then randomly sampled 70 % (48 healthy samples and 169 periodontitis samples) as external training set (EV-T) and 30 % (21 healthy samples and 72 periodontitis samples) as external validation set (EV-V).

To explore potential key genes associated with NrNeu, we identified intersections among up-regulated genes and intersections among down-regulated genes in the following four sets of data: 1) the top 100 genes of the blue module, 2) the top 800 DEGs between healthy individuals and patients with periodontitis in NrNeu, 3) the top 800 DEGs specific to NrNeu, and 4) the DEGs between healthy individuals and patients with periodontitis in the GSE16134 dataset. Lasso regression and multifactor logistic regression were then conducted on the above intersections in order to screen for key genes. Six types of machine learning methods were employed to develop and validate the prediction models: linear discriminant analysis (lda), logistic regression, Ranger, Naive Bayes, Rpart, and support vector machines (svm). Key genes selection and the model development were conducted using IV-T, and the prediction models were evaluated using EV.

In order to validate the stability and overcome possible over-fitting or under-fitting of the models, k-fold cross-validation of the IV-T and EV-T was performed. The k-fold cross validation works by dividing the data set into k segments. In each run, one segment plays the role of the validation set whereas the other remaining segments (k-1) are the training set. Finally, the results from each of the folds are averaged and returned as the final results. In our study, 10 repetitions of a 5-fold cross validation were performed using mlr3verse package (Table S2 and S3). A schematic diagram describing the machine learning workflow is shown in Fig. S1A-D. Finally, model interpretation analysis was conducted using the permutation feature importance function implemented in the R package iml.

Cellular communication analysis

To investigate the communication relationships and intensity between neutrophils and other cells in gingival tissue, we employed the CellChat package (v1.6) for cell communication analysis. The CellChat package models the probability of cell communication by integrating the expression of ligands, receptors, and their auxiliary factor genes, guided by a priori knowledge of their interactions. We evaluated the total number of interactions and their strength across various cell types with respect to NrNeu.

Ligand–receptor–target links analysis

The NicheNet package (v2.0, https://github.com/saeyslab/nichenetr) was used to predict ligand-target links between fibroblasts and NrNeu in the RNA expression data of interacting cells, alongside current knowledge of signaling and gene regulatory networks. We defined NrNeu as receiver cells and fibroblasts as sender cells. The ligand activity analysis aimed to identify which fibroblast-produced ligands most significantly impacted gene expression in NrNeu. Receptors predictions were inferred based on the NicheNet prebuilt prior model, which utilizes the regulatory relationships among and between transcription factors, ligands, receptors, and downstream target genes.

Animal experiments

Mice

C57BL/6J mice (male) were purchased from Southern Medical University Animal Experimental Centre (China). All mice were maintained under specific pathogen-free (SPF) conditions. Mice were 6–8 weeks old at the beginning of the experiments. All animal experiments conformed to the relevant regulatory standards and were approved by the Animal Care and Use Committee of Nanfang Hospital (Approval No: IACUC-LAC-20230513-001).

Ligature-induced periodontitis

Experimental periodontitis was induced by ligature placement as described previously [19]. After anesthetizing the mice with sodium pentobarbital, a smooth 5–0 silk ligature was tied in a subgingival position around the left second maxillary molars for 2 weeks. Finally, the mice were euthanized, and the maxillae and maxillary gingiva were collected. In each group, the maxilla was isolated and prepared for histological analysis (n = 5). Maxillary gingivae were removed and prepared for quantitative real-time polymerase chain reaction analysis (n = 5), while alveolar bone and teeth were prepared for microcomputed tomography (micro-CT) analysis (n = 5).

DNase-I treatment

The mice were injected intraperitoneally with 400 units of DNase-I (D8071, Solarbio) in 0.9 % NaCl or 0.9 % NaCl alone, with an injection performed every 2 days.

Micro-CT analysis

Mandibular bone specimens were placed in 4 % paraformaldehyde for 24 h and transferred to 75 % ethanol for micro-CT scanning (SkyScan 1276, Bruker, Belgium). The alveolar bone supporting the second maxillary molars was chosen as the region of interest (ROI). The distance from the cementoenamel junction to the alveolar bone crest (CEJ-ABC), the bone volume fraction (BV/TV), the bone surface fraction (BS/BV), the trabecular number (Tb. N), and the trabecular thickness (Tb. Th) were analyzed using SkyScan CTAn software (v1.18.4.0).

Histology and immunofluorescence

Tissue processing

Human specimens and maxillae of mice were fixed in 4 % paraformaldehyde (BL539A, Biosharp) in phosphate buffered saline (PBS) at 4 °C overnight. The mouse maxillae were then decalcified in 10 % EDTA (G1105, Servicebio) for 2 weeks. Embedded in paraffin, all tissues were sectioned into 4-μm thick sections and then deparaffinized using xylenes and an ethanol gradient.

H&E staining

H&E staining was performed according to the manufacturer’s protocol (DH0020-2, Leagene), and the stained slides were dehydrated in ethanol and sealed in neutral resin. Images were acquired using a microscope (BX63, Olympus).

Immunofluorescence

The antigen was thermally repaired with sodium citrate–EDTA (IH0302, Leagene) for 30 min. After the inactivation of endogenous peroxidase by 3 % H2O2, the antigen was incubated with 5 % bovine serum albumin (BSA; R00911, Leagene) to block nonspecific antigens. All subsequent processes were conducted according to the manufacturer’s protocol (RS0035, Immunoway). The primary antibodies used were as follows: CD66b (human neutrophil marker, TD10151S, Abmart, 1:100), Ly6G (mouse neutrophil marker, MAB1037, 1:100), myeloperoxidase (marker of NETs, ab208670, Abcam, 1:200), citrullinated histone 3 (marker of NETs, ab5103, Abcam, 1:200), and FCGR3B (marker of NrNeu, CSB-PA008544LA01HU, CUSABIO, 1: 200). Microscopic images were taken using a fluorescence microscope (Panoramic SCAN II, 3DHISTECH Ltd.).

Cell isolation and culture

Human gingivae were cut into 1 × 1 × 1 mm pieces in collagenase II and then placed on the bottom of T25 culture bottles. After adhesion, tissues were maintained in DMEM (C11885500BT, Gibco) supplemented with 10 % fetal bovine serum (FBS; FBS-S500, NEWZERUM) and 1 % penicillin/streptomycin (15140122, Gibco). Over time, the fibroblasts crawled out of the tissue and formed a monolayer of cells. Fibroblasts at the first passage (P1) were used for the following experiments.

Neutrophils were collected from healthy donors using a peripheral blood neutrophil separation kit (P9040, Solarbio). According to the manufacturer’s instructions, the blood was separated into plasma, monocytes, neutrophils, and erythrocyte layers from top to bottom. The neutrophil layer was carefully collected into new tubes, and the erythrocytes mixed therein were lysed, before washing with PBS and centrifuging at 250 g for 10 min. Finally, the neutrophils were cultured in RPMI-1640 (C11875500BT, Gibco) medium with 10 % FBS. All of the relevant experiments were performed using human primary neutrophils.

Preparation of conditioned medium

Healthy and periodontitis gingival fibroblasts were cultured until they reached 80–90 % confluence and the medium was changed to RPMI-1640 medium with 10 % FBS. The conditioned medium (CMed) was harvested after 24  h and filtered through 0.2-μm cell strainers before storing at 20 °C.

Nets inducement and analysis

Neutrophils were counted and seeded on 60-mm dishes with 1 % BSA pretreated at a density of 3 × 10⁶ cells. To induce NETs formation, neutrophils were separately treated with 50 nM PMA (P849986, Macklin), 4 μM ionomycin (S1672, Beyotime), and CMed derived from HGFs or PGFs mixed at a 1:1 ratio with RPMI-1640 complete media. After 2 h of treatment, total cellular RNA was extracted.

Next, 1 × 10⁵ neutrophils were seeded on poly-L-lysine-covered 24-well chamber slides and allowed to settle for 30 min. Cells were incubated in CMed derived from HGFs or PGFs mixed at a 1:1 ratio with RPMI-1640 complete media for 3 h. Neutrophils were treated for 3  h at 37 °C under 5 % CO2 and then stained with 20 μM SYTOX Green (KGA261, KEYGEN BIOTECH) for 10  min. Images were taken using a microscope (BX63, Olympus). Referring to existing research, the area of NETs coverage and dead cells (positive for SYTOX Green) were quantified using ImageJ (Fiji) software [20].

Quantitative real-time polymerase chain reaction

Total RNA was isolated using an EZ-press RNA Purification Kit (B0004D, EZBioscience) for cells and an Animal Total RNA Isolation Kit (RE-03011, Foregene) for tissues. Subsequently, RNA was reverse-transcribed using a Colour Reverse Transcription Kit (A0010CGQ, EZBioscience). The resulting cDNA was amplified using a MagicSYBR Mixture (CW3008, CWBIO) on an ABI Quant-Studio 5 system. The primer sequences are listed in Table 1. The 2^–ΔΔCt values were used to determine the fold changes in target gene expression.

Table 1.

The primer sequences.

Target genes	Sequence
GAPDH	Forward: CTGGGCTACACTGAGCACC Reverse: AAGTGGTCGTTGAGGGCAATG
GAPDH (mouse)	Forward: GACTTCAACAGCAACTCCCAC Reverse: TCCACCACCCTGTTGCTGTA
β-actin	Forward: ATTGCCGACAGGATGCAGAA Reverse: GCTGATCCACATCTGCTGGAA
FCGR3B	Forward: ACTGCTCTGCTACTTCTAGTTTCA Reverse: ATGGACTTCTAGCTGCACCG
MNDA	Forward: AGCGTACACAAGAAGAACACAA Reverse: GTTTCAGCTTGCGGTCAACTG
SMCHD1	Forward: TGCTACAGTCGGTCAATCAGT Reverse: AACCAGTGTGTCATAGTGAGGTA
SLC2A3	Forward: GCTGGGCATCGTTGTTGGA Reverse: GCACTTTGTAGGATAGCAGGAAG
S100P	Forward: AAGGATGCCGTGGATAAATTGC Reverse: ACACGATGAACTCACTGAAGTC
PTGS2	Forward: CTGGCGCTCAGCCATACAG Reverse: CGCACTTATACTGGTCAAATCCC
MME	Forward: AGAAGAAACAGCGATGGACTCC Reverse: CATAGAGTGCGATCATTGTCACA
MIF	Forward: CTGCACAGCATCGGCAAGAT Reverse: ATAGTTGATGTAGACCCTGTCCG
CD74	Forward: CCGGCTGGACAAACTGACA Reverse: GGTGCATCACATGGTCCTCTG
CXCR4	Forward: ACGCCACCAACAGTCAGAG Reverse: AGTCGGGAATAGTCAGCAGGA
CXCL15 (mouse)	Forward: TCGAGACCATTTACTGCAACAG Reverse: CATTGCCGGTGGAAATTCCTT
IL-1β (mouse)	Forward: GAAAGACGGCACACCCACCCT Reverse: GCTCTGCTTGTGAGGTGCTGATGTA
MMP9 (mouse)	Forward: CTGGACAGCCAGACACTAAAG Reverse: CTCGCGGCAAGTCTTCAGAG
TNFα (mouse)	Forward: ACAGAAAGCATGATCCGCG Reverse: GCCCCCCATCTTTTGGG

Statistical analysis

The data were analyzed using GraphPad Prism (v9, GraphPad Software). For all experiments, the mean ± SEM was calculated for at least three independent experiments. Unpaired two-tailed Student’s t test was used to compare the means between two independent groups, and one-way ANOVA was used to compare the means between two or more groups. The specific P values are detailed in the figure legends and were considered statistically significant at P < 0.05.

Results

Single-cell landscape of human gingival tissues

ScRNA-seq analysis was conducted on gingival tissues obtained from a cohort consisting of three healthy individuals and three individuals diagnosed with periodontitis (Fig. 1A). After implementing quality control exclusion, 54,801 single cells were retained, comprising 29,454 cells from the healthy control group and 25,347 cells from the periodontitis group (Fig. S2 and B). Using uniform manifold approximation and projection (UMAP), we identified nine major cell populations, including fibroblasts, endothelial cells, epithelial cells, T/NK cells, macrophages, neutrophils, mast cells, B cells, and plasma cells (Fig. 1B and Table S4) [13], [15]. The heatmap showed the marker gene expression levels associated with each cell population (Fig. 1C and S2C).

Fig. 1.

Single-cell landscape of human gingival tissues (A)Study flow diagram. (B)Uniform manifold approximation and projection (UMAP) plot representation of major cell types in gingiva tissues (All: n = 6, 54,801 cells; Health: n = 3, 29,454 cells; Periodontitis: n = 3, 25,347 cells). (C)Heatmap of the expression levels of marker genes in different cell types. (D)Relative fold change of total cell numbers between periodontitis and health for all clusters depicted in (B). (E)Immunofluorescence staining of neutrophil marker (CD66b) in human gingivae. White arrows indicate neutrophils. White scale bar = 100  μm; yellow scale bar = 8 μm. (F)Immunofluorescence staining of neutrophil marker (Ly6g) in mice gingivae. White arrows indicate neutrophils. White scale bar = 200  μm; yellow scale bar = 10 μm. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

We then compared the fold-change differences in each cell population in the periodontitis group relative to the healthy group (Fig. 1D and S2D). The results showed that stromal, epithelial, and mast cells were significantly reduced in periodontitis, whereas plasma cells, B cells, and neutrophils were significantly increased. Notably, neutrophils exhibited the most dramatic increase. Consequently, increased neutrophil infiltration in periodontitis gingival tissues from humans and mice were confirmed by immunofluorescence staining of CD66b or Ly6G (Fig. 1E). These results suggest that neutrophils are recruited in periodontitis and play an essential role in periodontal immunity [2].

Nets-related subpopulations identified by neutrophil heterogeneity analysis

We next focused on the neutrophil population. To this end, neutrophils were further subdivided into four subsets (Fig. 2A A and S3A), including N0 (FCGR3B, MNDA, SMCHD1, S100A8, and SLC2A3), N1 (RPS3, RPL13, RPL10, RPL12, and RPS8), N2 (CCRL2, CCL4, C15of48, CCL3L1, and CCL3), and N3 (FDCSP, SFN, FABP5, S100A2, and KRT14) (Fig. 2C, S3B and S3C). Fold change analysis of each subpopulation between the periodontitis and healthy groups showed that N3 was almost exclusively present in the healthy group, while N0, N1, and N2 were increased in periodontitis, with N0 showing the most significant increase (Fig. 2B and Table S5).

Fig. 2.

Identified NETs-related subpopulations by neutrophil heterogeneity analysis (A)The UMAP plot distribution of the subclusters of neutrophil (4,032 cells) in healthy and periodontitis gingiva and the pie charts showing their proportion. (B)Relative fold change of total numbers between periodontitis and health for all neutrophil subclusters. (C)The UMAP plots of the marker genes for N0. (D)WGCNA analysis of neutrophils identified 3 modules of co-expressed genes. (E)Bubble graph showing the expression of gene modules in neutrophil subpopulations. (F)Venn diagram representation of the intersection of the top 100 of blue modules genes and the top 800 DEGs of NrNeu. Bar graph demonstrating the KEGG enrichment results of the above intersection. (G)Bar graphs of the relative expression of NrNeu specific marker genes in human neutrophils treated or not with NETs inducers (PMA or ionomycin). Data shown as means ± SEM with *P < 0.05, **P < 0.01 and ****P < 0.0001 by one-way ANOVA. (H)The left pseudotime trajectory is shown coloured in a gradient from dark to light blue and the start of pseudotime is indicated. The middle pseudotime trajectory is shown coloured by different cell states. The right pseudotime trajectory is shown coloured by healthy and periodontitis groups. (I)The pseudotime trajectory of four neutrophil subclusters. (J)Representative images of multicolour Immunofluorescence staining showing the expression of FCGR3B, MPO and CitH3 in human gingival sections. White scale bar = 100 μm; yellow scale bar = 10 μm. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

To explore the specific biological functions of each subpopulation, we next performed weighted gene co-expression network analysis (WGCNA) of neutrophil genes, which identified three modules: brown, blue, and turquoise (Fig. 2D). By analysing the average expression of the module genes in each subpopulation, we found that the blue module genes were highly expressed in N0, the brown module in N2, and the turquoise module mainly in N3 (Fig. 2E). KEGG pathway enrichment analyzes were conducted on the intersections of the top 100 module genes and the top 800 DEGs of the corresponding subpopulations. The results indicated that the intersection genes of the blue module and N0 were markedly enriched in the Neutrophil extracellular trap formation pathway (Fig. 2F). To demonstrate the association between N0 and NETs, we isolated neutrophils from peripheral blood and analyzed the NOX-dependent and NOX-independent NETs formation induced by PMA and ionomycin in vitro. We found that the expression of FCGR3B, SLC2A3, and S100P was significantly increased in both induction methods, implying that these genes are involved in both NOX-dependent and NOX-independent pathways [21], [22]. Intriguingly, SMCHD1 expression was elevated when NETs were induced with PMA while MNDA expression was significantly elevated when ionomycin was used (Fig. 2G). These findings indicate that SMCHD1 is mostly present in the NOX-dependent NETs generation pathway, while MNDA is primarily present in the NOX-independent NETs generation pathway. In addition, colocalization of FCGR3B with NET markers (MPO and CitH3) further confirmed the function of N0 in generating NETs (Fig. 2J) [23], [24], [25], [26]. Moreover, high activities of NETs-related pathways were detected in N0, suggesting that NETs formation occurs mainly in N0 compared to other subpopulations (Fig. S3D). In according with the above results, we defined N0 as “NETs-related neutrophil” (NrNeu) and the genes that were both in the blue module and NrNeu as “NETs-related genes” (NrGs).

Analogously, the intersections of the top 100 brown module genes and the top 800 DEGs from N2 were mostly enriched in the autophagy–animal pathway, leading to the designation of N2 as “Autophagic Neutrophil” (ANeu) (Fig. S3E). In parallel, the intersection genes of the turquoise module and N3 were mainly involved in pathways related to physiological metabolic processes such as ribosome and oxidative phosphorylation; thus, we named this module “Normal Neutrophil” (NNeu) (Fig. S3F). Pathway enrichment showed that N1, which did not exhibit a high expression of modular genes, mainly participated in pathways related to pathogen infection, such as Coronavirus disease, Phagosome, Salmonella infection, and Prion disease. Neutrophils are the first line of host defence against infection, and one of their key pathogen-eliminating mechanisms is phagocytosis. Therefore, we designated N1 as “Phagocytic Neutrophil” (PNeu) (Fig. S3G).

Next, to determine the lineage relationships, we performed pseudotime analysis on the neutrophil subpopulations to display the variation in cellular composition from healthy to periodontitis (Fig. 2H). NNeu was distributed at the beginning of the trajectory, which was consistent with the result showing that this subpopulation was predominantly derived from gingival samples from the healthy group. Subsequently, cells in the periodontitis group differentiated into two distinct fates, with stage 2 slightly preceding stage 3. RNeu was primarily distributed in stage 2, whereas NrNeu and ANeu were located in stage 3. Notably, the positions of NrNeu and ANeu prompted us to speculate that a close relationship exists between these two subpopulations in periodontitis progression (Fig. 2I).

Nets foster periodontal immunopathologies

To gain further insight into the effect of NETs in periodontitis, we treated control and LIP mice with DNase I, a proven NETs inhibitor (Fig. 3A). Immunofluorescence analysis of mouse maxillary tissues showed that the expression of NETs markers (MPO and CitH3) was significantly increased in LIP mice compared to healthy control mice, while in the LIP group, the expression was visibly reduced in the DNase I-treated group compared to the vehicle group (Fig. 3B), indicating suppression of NETs. We further explored the effect of NETs inhibition on periodontitis in LIP mice. HE staining showed that after inhibiting NETs, the gingival tissues of LIP mice exhibited reduced inflammatory cell infiltration and attachment loss (Fig. 3C). Additionally, DNase I treatment clearly decreased the expression of gingival inflammatory factors (IL-1β, TNF-α, CXCL15, and MMP9) in LIP mice (Fig. 3D), indicating that NETs are capable of regulating local inflammation. The micro-CT analysis demonstrated that suppressing NETs mitigated alveolar bone resorption in LIP mice (Fig. 3E and F). Our findings suggest that NETs are critical for inflammatory infiltration and alveolar bone loss in periodontitis, which makes them a potential therapeutic target.

Fig. 3.

NETs foster periodontal immunopathologies (A)Schematic illustration of the LIP mouse model and intraperitoneal injection of DNase-I. (B)Immunofluorescence for CitH3 and MPO in gingival tissues from control (CTRL) and LIP mice, with or without DNase-I. Scale bar = 200 μm. (C)H&E staining images of the maxillae from mice. The black arrows show the location of cementoenamel junction (CEJ). The black lines represent attaching loss (AL) showing the distance from the CEJ to the base of periodontal pocket. Scale bar = 200 μm. (D)Bar graphs of the relative expression of CXCL15, IL-1β, MMP9 and TNFα in mice gingivae. Data shown as means ± SEM with *P < 0.05, **P < 0.01 and ***P < 0.001 by one-way ANOVA. (E)Micro-CT images of the maxillae in CTRL and LIP mice, with or without DNase-I. The red areas indicate the level of bone loss in the various experimental groups, whose the upper red line indicates the cemento-enamel junction (CEJ), and the lower line indicates the alveolar bone crest (ABC). (F)Quantification of CEJ-ABC length, bone mass and microarchitecture in the maxillae. Quantification results shown as means ± SEM with *P < 0.05, **P < 0.01, ***P < 0.001 and ****P < 0.0001 by one-way ANOVA. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Prediction models for periodontitis based on key NrGs

Considering that NETs effectively boost periodontal pathology, we next constructed NrG-derived prediction models for periodontitis (Fig. 4A and Fig. S1A) [27]. In brief, intersections were computed separately with genes up- or down-regulated in periodontitis within NrGs and those identified in periodontitis within GSE16134 (see “Materials and methods”). The fourteen genes obtained from the above intersections were subjected to lasso regression and multifactorial logistic regression. Three key genes were finally screened, including PTGS2, MME, and SLC2A3, which were up-regulated in periodontitis in NrNeu and GSE16134 (Fig. 4B and C).

Fig. 4.

Prediction models for periodontitis based on NrGs (A)Flow chart of the prediction models construction and its validation. (B)Bubble plot of the expression of SLC2A3, PTGS2 and MME in varies neutrophil subculsters in healthy and periodontitis states. (C)Volcano plot for DEGs of periodontitis vs. health in GSE16134. (D)Box plots showing the AUC values of all machine learning algorithms in the internal training set (IV-T). (E)Box plots showing the AUC values of all machine learning algorithms in the external training set (EV-T). (F)ROC curve of all machine learning algorithms on the internal training set (IV-V). (G)ROC curve of all machine learning algorithms on the external training set (EV-V). (H)Bubble plot of the expression of SLC2A3, PTGS2 and MME in healthy and periodontitis states in scRNA-seq data. (I)Bar graphs for the relative expression of PTGS2, MME and SLC2A3 in human healthy and periodontitis gingivae. Results shown as means ± SEM with **P < 0.01, ***P < 0.001 and ****P < 0.0001 by Unpaired two-tailed Student's t-test.

We then built periodontitis prediction models of the three key genes using six machine learning algorithms. During modeling, k-fold cross-validation (k = 5) was used to enhance the stability of the model. Moreover, the cross-validation results proved that over-fitting and under-fitting were avoided as much as possible during the modeling process (Table S2 and S3). For IV, the median AUC values of IV-T were distributed in the range of 0.65 to 0.75 (Fig. 4D), while those of IV-V were predominantly greater than 0.70 (Fig. 4F). To verify the generalization capability of our prediction models, the independent dataset GSE16134 was used for EV. The median AUC values of EV-T for all six models were greater than 0.85, and the results of EV-V were highly consistent with those of EV-T (Fig. 4E and G). We applied the feature importance to identify the contribution of each feature to the model prediction. The contribution of all three genes to the modeling was relatively high, though the importance of each gene varied in different models. In IV, the importance of the three genes was almost equal in lda, logistic regression and ranger models. PTGS2 was the most important feature in navie bayes and rpart models, while MME and SLC2A3 were the most important feature in svm model (Fig. S4). In EV, PTGS2 was the most important feature across all models (Fig. S5). Moreover, the up-regulation of these three key genes in periodontitis was validated in our scRNA-seq data and in clinical gingival samples (Fig. 4H and 4I). These results suggest that the three key genes based on NrGs have good potential for periodontitis prediction at both the neutrophil and gingival tissue levels.

Periodontitis gingival fibroblasts enhance NrNeu to form NETs via MIF-CD74/CXCR4

Dissecting the interactions between NrNeu and other cell populations in the periodontal immune microenvironment will facilitate further elucidation of the mechanisms underlying NETs formation in periodontitis. Cell-cell communication analysis identified that fibroblasts exerted the strongest effect on NrNeu under both healthy and periodontitis conditions, suggesting that fibroblasts contribute to the formation of NETs (Fig. 5A-C). We next stimulated neutrophils with a conditioned medium obtained from healthy and periodontitis gingival fibroblasts (HGFs CMed and PGFs CMed), which showed that the latter caused more cell death and NETs production (Fig. 5D and E). Subsequently, we applied ligand activity prediction to find the most probable ligands in fibroblasts and their downstream targeted genes in NrNeu, with the result that MIF was the top candidate involved in NrNeu regulation (Fig. 5F). Indeed, MIF expression was upgraded in periodontitis fibroblasts, which was also observed in PGFs (Fig. 5H). Further analysis revealed that the receptors for MIF were CD74, CXCR4, and CXCR2 (Fig. 5G). Among them, upregulation of CD74 and CXCR4 expression was observed in NrNeu in periodontitis, as well as in neutrophils stimulated by PGFs CMed (Fig. 5H).

Fig. 5.

Periodontitis gingival fibroblasts enhance NrNeu forming NETs via MIF-CD74/CXCR4 (A)The circo plots illustrating the total numbers and strength of intercellular communication between NrNeu and other cell populations. (B)The circo plots showing the numbers and strength of intercellular communication between NrNeu and other cell populations in healthy state. (C)The circo plots demonstrating the numbers and strength of intercellular communication between NrNeu and other cell populations in periodontitis state. (D)Representative images of neutrophils treated by HGFs CMed or PGFs CMed for 3 h. DNA released by rupture of the cell membrane was stained with Sytox Green (green). White arrows represent to NETs. Scale bar = 200 μm. (E)(Upper) Quantification of the area of the field covered by SYTOX green positive neutrophil-derived extracellular DNA relative to the number of neutrophils in each field after treatment with HGFs CMed or PGFs CMed for 3 h. (Lower) The percentage of dead neutrophils after treatment based on the number of SYTOX green positive neutrophils. Data shown as means ± SEM with **P < 0.01 by Unpaired two-tailed Student's t-test. (F)Heatmap for the relationship between ligands expressed by Fibroblast and potential target genes in NrNeu. (G)Heatmap for the relationship between ligands expressed by Fibroblast and potential receptors in NrNeu. The expression of the corresponding ligands for Fibroblast and receptors for NrNeu in healthy (H) and periodontitis (P) states are displayed on the right and below in the form of heatmaps. (H)Box plots indicating MIF expression in Fibroblast and CD74 and CXCR4 expression in NrNeu. Bar graphs representation for the relative expression of MIF in healthy and periodontitis fibroblasts and the relative expression of CD74 and CXCR4 in neutrophils treated with HGFs CMed or PGFs CMed. Data shown as means ± SEM with *P < 0.05 and ****P < 0.0001 by Unpaired two-tailed Student's t-test. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Discussion

Neutrophil infiltration is a hallmark of periodontitis, and NETs are a crucial trigger of histopathological damage in periodontitis. In this study, we deciphered the phenotypic and functional diversity of neutrophils in human gingiva tissues and first defined NETs-related neutrophil subpopulations in gingival tissue using single-cell techniques. We also constructed a prediction model for periodontitis based on the key genes of NETs and unveiled the molecular mechanism of elevated production of NETs in periodontitis, which may contribute to disease diagnostics and pathology.

Similar to previously published studies, our analysis revealed four major cell compartments: epithelial cells, stromal cells (including fibroblasts and endothelial cells), and immune cells [13], [14], [15]. Among immune cells, T cells are the most abundant, indicating their important role in periodontal homeostasis and disease progression. In periodontitis, both epithelial cells and stromal cells are notably reduced, whereas immune cells are significantly increased. Specifically, our results revealed that neutrophils show the greatest increase in periodontitis. Correspondingly, many studies have confirmed the substantial increase in neutrophils in periodontitis and their role in disease progression [11], [3], [28].

Previous single-cell sequencing for periodontitis lacked subpopulation analysis of human gingival neutrophils, whereas our study delved into the heterogeneity of neutrophils. We employed WGCNA and identified three gene modules, each highly expressed in a different subpopulation. WGCNA is a method used to cluster highly associated genes into gene modules that may be relevant to cell typing because highly associated genes are likely to be detected in the same subsets [29]. A previous study on somatosensory neurons applied single-cell sequencing combined with WGCNA to analyze the functional heterogeneity of different neuronal clusters [30]. Accordingly, we obtained phenotype-related modular genes to analyze the functional characteristics of neutrophil subpopulations. In contrast to focusing only on DEGs, WGCNA uses information from the most changed genes or all genes to characterize highly synergistically changed gene sets and correlate them with phenotypes; this not only makes full use of the information but also contributes to the study of gene interactions at a comprehensive level.

We set the five most significant DEGs as markers for NrNeu, including FCGR3B, MNDA, SLC2A3, S100P, and SMCHD1. FCGR3B encodes a protein that belongs to the Fc receptor family for IgG molecules, and it has been reported to be involved in NETs formation induced by immune complexes [26]. FCGR3B expression is also involved in a specific class of neutrophils, low-density neutrophils (LDNs) [31]. LDNs release large numbers of NETs, which have been documented to progress inflammation in diseases such as lupus and Aureus-infected diabetes [32]. MNDA has been identified in different NETs-related proteomics studies and may undergo citrullinated modifications [33], [34], [35]. In our study, the expression of MNDA was significantly up-regulated in NOX-independent NETs formation. Upon stimulation by inflammation mediators, MNDA relocates to the cytoplasm and undergoes caspase-dependent cleavage, which is required for formation and activation of the inflammasome [36]. It has been observed that inflammasomes are involved in NETs formation induced by nigericin or ionomycin [37]. NETs activation with nigericin or ionomycin does not require PKC, ROS, MPO, NE or transcriptional activation, which are part of the NOX-independent pathway [38]. Thus, we hypothesize that MNDA may participate in NOX-independent NETs formation through inflammasome-dependent signaling. S100P (S100 Calcium Binding Protein P) is a member of the S100 family and acts as a calcium sensor that contributes to cellular calcium signaling [39]. Remarkably, some studies have highlighted the critical role of calcium signaling in the generation of NETs [40], [41], [22]. Meanwhile, S100 proteins are required for activation of the receptor for advanced glycation end products (RAGE) in responsive cells [39]. RAGE is required for platelet stimulation of NETs via HMGB1, and the autocrine/paracrine role of endogenous RAGE ligands is essential in the late stages of NETs formation [42], [43]. Thus, we speculate that S100P may be involved in NETs formation by regulating calcium signaling or activating RAGE. SMCHD1 is a non-canonical member of the structural maintenance of chromosomes protein family that plays a key role in epigenetic silencing by regulating chromatin architecture [44]. There are no relevant studies that have reported on its role in NOX-dependent NETs generation at present, underscoring the need for further exploration in future research.

To clarify the effects of NETs in severe periodontitis, we constructed ligation-induced periodontitis mice with NETs inhibition. Our findings support the idea that NETs can enhance gingival inflammation and alveolar bone resorption in severe periodontitis, and these pathological processes can be rescued by NETs inhibitors. Similar results were obtained in the early stages of periodontitis and ligneous periodontitis in mice [10], [11], suggesting that targeting NETs represents a promising approach for treating periodontitis. Convincing evidence points to a link between NETs and the pathogenesis of many diseases, including sepsis, systemic lupus erythematosus and rheumatoid arthritis [45], [46]. Hence, an increasing number of researchers are turning their attention to NETs-targeted therapy [47], [48], [49]. Among these, recombinant human DNase-I (rhDNase), a safe, United States Food and Drug Administration-approved drug, can effectively degrade NETs. Although rhDNase has been widely used clinically for NETs-associated diseases, its limitations include a short half-life and high clearance rate [47]. In recent years, studies have been conducted to solve these problems by transforming rhDNase into hydrogels or coating them with nanoparticles [50], [51]. Thus, modification of rhDNase may broaden the path for periodontal therapy.

Given the influential role of NETs in periodontitis, coupled with the fact that current clinical diagnostic methods for periodontitis cannot detect the disease until it has progressed to a specific stage, developing an early prediction model for periodontitis based on key NrGs could help us understand and manage periodontal conditions. We successfully filtered out three key genes, including SLC2A3, PTGS2, and MME. As a marker gene of NrNeu, SLC2A3 facilitates glucose transport and can also mediate the uptake of various other monosaccharides across the cell membrane. The formation of NETs is strictly dependent on glucose, and glucose uptake is increased during PMA stimulation of neutrophils, suggesting that SLC2A3 is involved in the process of glucose transport during NETs formation [52]. In addition, SLC2A3 was primarily up-regulated in neutrophils in the periodontium of patients with periodontitis and positively correlated with neutrophil infiltration [53]. Activated by NETs through the Toll-like receptor (TLR2), prostaglandin-endoperoxide synthase 2 (PTGS2), more widely known as cyclooxygenase-2 (COX-2), acts as an important factor in periodontitis by mediating the inflammatory response in periodontal tissues [54], [55], [56]. The protein encoded by MME is a neutral endopeptidase that is predominantly expressed on neutrophils and fibroblasts in periodontitis, and its expression correlates with the severity of periodontal disease [57], [58]. Validated high prediction potential with internal and external validation, these three genes require further in-depth exploration to determine their potential utility in clinical decision-making processes.

We then extended our view from neutrophils to the periodontal tissue microenvironment and found that fibroblasts in gingival tissues had the most significant cellular communication relationship with NrNeu in both healthy and periodontitis groups. Previous studies have shown that stromal cells, predominantly fibroblasts, are important for the regulation of neutrophil chemotaxis and immune infiltration [13]. Likewise, cellular experiments confirmed that supernatants from inflammatory gingival fibroblasts significantly activated NETs production from peripheral blood neutrophils. The performed analysis also illustrated that gingival fibroblasts most likely prompt the production of NETs by NrNeu via MIF-CD74/CXCR4 (Fig. 6). MIF regulates different cellular functions, including proliferation, differentiation, and cell survival/apoptosis, but is also involved in cell-mediated immunity, immunoregulation, and inflammation [59], [60]. In periodontitis and periradicular periodontitis, MIF can enhance the production and activity of osteoclasts, facilitating alveolar bone loss, in a process that is linked to neutrophil accumulation [61], [62], [63]. In addition, evidence suggests that MIF fosters neutrophils to produce NETs [64], [65]. Furthermore, MIF requires CD74 to activate ERK1/2 signaling and CD74 and CXCR4 to activate JNK signaling, both of which have been reported to be involved in the generation of NETs [59], [66], [67], [68]. In this case, we speculate that gingival fibroblasts boost NETs production through MIF-CD74/CXCR4 activation of ERK1/2 or JUK signaling. Overall, by understanding the upstream regulation of NETs, we can identify improved clinical therapeutic targets and design better clinical medications to treat periodontitis.

Fig. 6.

gingival fibroblasts most likely prompt the production of NETs by NrNeu via MIF-CD74/CXCR4 In inflamed gingival tissues, fibroblasts are reduced while neutrophils are increased, and fibroblasts may contribute to increased production of NETs via the MIF-CD74/CXCR4 axis, which in turn promotes the progression of periodontitis.

Conclusion

Our findings identified a NETs-related neutrophil subgroup in gingival tissue that may pave the way for subsequent in-depth exploration of the mechanisms of NETs in the context of periodontitis. We proposed a potential prediction model for periodontitis and uncovered a role for MIF-CD74/CXCR4 as a pathway that may be relevant to fibroblasts promoting the formation of NETs, provoking inflammatory pathological injury.

Compliance with ethics requirements

All procedures followed were in accordance with the ethical standards of the responsible committee on human experimentation (institutional and national) and with the Helsinki Declaration of 1975, as revised in 2008 (5). Informed consent was obtained from all patients for being included in the study.

All Institutional and National Guidelines for the care and use of animals (fisheries) were followed.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary material

The following are the Supplementary data to this article: