Spatially resolved single cell atlas deciphers SAA1 inflammatory epithelial cells

Abstract

Disruption of epithelial integrity is a pivotal event in inflammation, disease pathogenesis, and tissue homeostasis. To investigate these processes in chronic inflammatory disease, we performed spatial transcriptomics integrated with single-cell RNA sequencing (scRNA-seq) on human gingival tissue, coupled with metagenomic analysis of matched subgingival plaque. This approach allowed in situ characterization of epithelial heterogeneity and microbiome-epithelium-connective tissue crosstalk. We identified a distinct inflammatory epithelial subpopulation (SAA1+Epi), situated within the junctional epithelium, that becomes activated through the TLR2-PITX2 axis by Porphyromonas gingivalis lipopolysaccharide. These SAA1+Epi cells secrete TGFβ, which induces an inflammatory program in the connective tissue by driving the differentiation of inflammation-associated fibroblasts (C3+FB) via the PI3K/Akt pathway. Concurrently, SAA1+Epi cells express chemotactic factors such as CXCL6 to recruit NK cells, thereby sustaining the inflammatory niche. The transcription factor PITX2 emerged as a critical regulator of SAA1+Epi differentiation; targeting PITX2 suppressed  C3+FB induction and natural killer (NK) cells recruitment, ultimately attenuating periodontitis progression. Our findings position SAA1+Epi as a frontline responder to dysbiotic bacteria at the inflammatory interface and underscore its essential role in regulating epithelial-connective tissue homeostasis during inflammation.

Introduction

The epithelium serves as a sophisticated and dynamic interface that acts not only as a physical barrier but also as an active orchestrator of tissue homeostasis, immune surveillance, and host defense.^1,2 Following microbial challenge, the preservation of epithelial integrity and its communication with underlying connective tissue are critical to the pathogenesis of chronic inflammatory disorders.^3–5 In the intestinal epithelium, for instance, dysregulation of conserved pathways such as Wnt/β-catenin and Notch impairs crypt regeneration and Paneth cell function, compromising barrier integrity and promoting bacterial translocation in inflammatory bowel disease.^6,7 Similarly, in skin, defects in keratinocyte differentiation and filaggrin expression, together with altered transforming growth factor-β (TGF-β) signaling, sustain the chronic inflammation seen in atopic dermatitis.^8,9 Across tissues, the mechanisms driving epithelial dyshomeostasis during chronic inflammation are highly diverse, shaped by organ-specific microenvironments and functional demands.¹⁰ This contextual regulation underscores the need to study epithelial behavior within its native tissue framework to uncover both shared and unique principles of inflammatory dysregulation.

In the oral cavity, disruption of the gingival epithelial barrier can trigger periodontitis, a highly prevalent chronic inflammatory disease linked to systemic conditions including diabetes, cardiovascular disease, and adverse pregnancy outcomes.^11–15 Periodontal pathogens and their virulence factors disrupt intercellular junctions such as E-cadherin and occludin, facilitating bacterial invasion.^16–18 Activation of epithelial Toll-like receptors (e.g., TLR2, TLR4, TLR5) by microbial components engages mitogen-activated protein kinase (MAPK) and phosphoinositide 3-kinase (PI3K) signaling, leading to epithelial apoptosis and barrier breakdown.^19,20 Subsequent immune activation within the underlying connective tissue fuels a self-sustaining inflammatory cascade that promotes osteoclastogenesis and alveolar bone loss.^21–23 Fibroblasts further contribute to pathological remodeling via reactive oxygen species (ROS) and matrix metalloproteinases (MMPs) production, degrading the extracellular matrix and amplifying local inflammation.^24,25 Despite these insights, a spatially resolved understanding of epithelial-connective tissue crosstalk in periodontitis remains limited.

The application of single-cell RNA sequencing (scRNA-seq) and spatial transcriptomics (ST) in periodontium research has become a cutting-edge focus, revealing unprecedented cellular heterogeneity and intricate cell-cell communication networks. A comprehensive single-cell atlas of gingival tissues has been established, revealing the heterogeneity of immune cells across disease states, the differentiation potential of pericytes, and the transition of fibroblasts to a C3-secretory proinflammatory phenotype, as extensively documented in recent studies.^26–28 In addition, Quinn et al. integrated single-cell datasets to characterize keratinocyte heterogeneity, identifying a distinct differentiation state and immunomodulatory function of KRT19+ keratinocytes in periodontitis.²⁹ These cells notably upregulate the expression of genes encoding multiple effector cytokines, including CXCL1, CXCL8, IL1A, and IL1B. Furthermore, by coupling spatial proteomics, the authors revealed the spatial heterogeneity of the peri-epithelial immune microenvironment, demonstrating differently enriched clustering of innate and adaptive immune cells around keratinocyte subclusters. Through integrated scRNA-seq and ST analyzes, Zhang et al. revealed that cigarette smoke exposure compromises epithelial barrier integrity and disrupts fibroblast-epithelial cell crosstalk in gingival tissues.³⁰ Despite these insights, the mechanistic roles of specific epithelial subclusters, their differentiation dynamics, and the direct in situ impact of the local microbial community on epithelial remodeling remain incompletely characterized.

To address these gaps, in the present study, we integrated scRNA-seq, ST, and metagenomic sequencing of paired gingival tissues and subgingival plaque from healthy individuals and patients with periodontitis to investigate host-microbe interactions in situ. Briefly, we summarize here the conceptual framework that is further integrated in Fig. 8. We define inflammatory epithelial remodeling in periodontitis and identify Epithelium-3, located near the junctional epithelium, as a spatially restricted inflammatory epithelial region. Within this region, we further identified an SAA1+ inflammatory epithelial subcluster. The following therefore proceeds from spatial epithelial identification to microbial-triggered epithelial reprogramming, and finally to downstream connective tissue and immune microenvironment remodeling.

Fig. 8.

The mechanism diagram of SAA1+Epi in forming and maintaining the periodontal immune and inflammatory microenvironment. Epithelium-3 acts as the “invasive inflammatory area” in the gingival epithelium. The SAA1+Epi function in the Epithelium-3 is induced by P. gingivalis through the TLR2-ERK1/2-Smad2/3-PITX2 axis and interacts with the fibroblast and NK cells in the surrounding tissue

Results

Spatial transcriptomics integrated with scRNA-seq reveals the organizational landscape of the human gingival epithelium

To characterize periodontitis gingival epithelium features, we established ST (10× Genomics Visium) and scRNA-seq (10× Genomics Chromium) methods. Limited healthy gingival biopsy size allowed paired scRNA-seq for only a subset of controls. Spatial transcriptomics was performed on 4 healthy controls and 6 periodontitis patients, while scRNA-seq was conducted on a partially overlapping cohort of 2 healthy controls and 6 periodontitis patients, and the scRNA-seq reference was mapped onto the ST sections to reconstruct gingival tissue organization (Fig. 1a, b). A total of 3520 Visium spatial spots were obtained from ST sections. Damaged and unrecognizable spots were discarded. The epithelium region was first annotated based on the morphology of hematoxylin and eosin (H&E) staining of each tissue section (Fig. S1a, b) and further confirmed by the expression of known epithelial marker genes (KRT5, KRT76, and CSTB) (Fig. 1c, and Fig. S1c). The gingival epithelium could be clustered into 4 epithelium regions (Fig. 1d). In terms of histological anatomy, Epithelium-2 predominantly corresponded to the stratum spinosum of the sulcular epithelium, whereas Epithelium-1 aligned with both the stratum granulosum and stratum corneum layers. Of particular clinical significance, Epithelium-3 occupied the basal layer of the junctional epithelium. In contrast, Epithelium-4 maintained its characteristic location in the basal layer of the oral gingival epithelium, serving as an internal reference for epithelial homeostasis. Gene Set Variation Analysis (GSVA) revealed that the functions of all epithelium regions were consistent with the corresponding anatomical levels (Fig. S1d).

Fig. 1.

Spatial transcriptomics (ST) and scRNA-seq landscape in the gingival epithelium. a The gingival tissues were collected from 4 healthy controls and 6 periodontitis patients, and the gingival epithelium region. ST was performed on gingival tissues from 4 healthy controls and 6 periodontitis patients. scRNA-seq was performed on 2 healthy controls and 6 periodontitis patients. Because healthy gingival biopsies were limited in size and tissue allocation prioritized ST to preserve spatial information, paired ST and scRNA-seq data were available for only a subpopulation of healthy controls. b Schematic drawing of the combined scRNA-seq and ST. c The Epithelium in the slice of ST was recognized by marker genes (KRT76, KRT5, CSTB). d A dot plot showing the 4 major subclusters in different colors in the health group (n = 4) and periodontitis group (n = 6). e Uniform manifold approximation and projection (UMAP) of epithelial cell subclusters. The black circle indicates SAA1+Epi

A total of 37,122 cells were subsequently captured from the raw scRNA-seq data. Six cell clusters were identified: the epithelial cell cluster (KRT5), the endothelial cell cluster (PLVAP), the B and plasma cell cluster (CD79A), the natural killer (NK) cell (XCL1) and T-cell cluster (CD3D), the fibroblast cluster (COL3A1), and the myeloid cell cluster (LYZ) (Fig. S2a, b). The epithelial cells were then clustered into 6 distinct subclusters (Fig. 1e) according to their marker genes (Figs. S2c and S3d): COL17A1+Epi, KRT6C+Epi, IGLC2+Epi, VIM+Epi, SAA1+Epi, and SPRR1B+Epi. According to Gene Ontology (GO) functional analysis (Data S1), COL17A1+Epi and KRT6C+Epi have differentiation functions, while IGLC2+Epi plays a role in signal transduction. SAA1+Epi showed significant inflammatory features. VIM+Epi represents the epithelial-mesenchymal transformation (EMT) epithelial group, and SPRR1B+Epi is a group of mature keratinocytes.

SAA1+Epi epithelial subcluster is responsible for the pathogenesis of periodontitis

In terms of the landscape of the human gingival epithelium, Epithelium-3, which is the primary interface for subgingival plaque interactions, exhibited high expression of the genes MMP13, CXCL6, SAA2, and PTHLH, suggesting its potential role in gingival inflammation (Fig. S3a). Moreover, the proportion of Epithelium-3 in patients with periodontal disease significantly increased (Fig. S3b), indicating its function in periodontitis. Gene set enrichment analysis (GSEA) analysis revealed that Epithelium-3 was responsive to immune activity, especially bacteria and lipopolysaccharide (LPS) (Fig. 2a), which confirmed the special role of Epithelium-3 in bacterial defense. Epithelium-3 significantly upregulated pathways related to epithelial cell immune activation and proinflammatory immune responses (Fig. S1d), indicating that Epithelium-3 functions as both a microbial sentinel and an inflammatory orchestrator. Epithelium-3 directly converts plaque challenge into sustained tissue destruction, driving the transition from microbial colonization to chronic periodontitis.

Fig. 2.

SAA1+Epi performs the functions of the Epithelium-3. a Dot plot of functions enriched in the Epithelium-3 by using GSEA analysis with Bayesian Testing, P < 0.05. The X axis represents the enrichment score. The dot size represents the average score. The dot color represents the -log10^P Value. b Distribution of SAA1+Epi in different Epithelium regions was determined using the inferred proportion by robust cell type decomposition (RCTD), and P-values were determined by one-way ANOVA. ^*P < 0.05, ^**P < 0.01, ^***P < 0.001, ^****P < 0.000 1. c Visualized spatial distribution of SAA1+Epi in ST slices. d Dot plot of functions enriched in the 6 epithelial cell subclusters by using GSVA analysis with the Wilcoxon test, P < 0.05. The dot size represents the average score. The dot color represents the log₁₀P-value. e Representative images of mIF staining of DAPI (Blue), KRT5 (Red), and MMP13 (Green) in human gingiva tissues. n = 3 (Scale bars, 50 μm.) The white arrows represent SAA1+Epi

Furthermore, we identified SAA1+Epi, a functionally distinct epithelial subcluster within Epithelium-3, which exhibited pronounced enrichment in inflammatory and immune-regulatory pathways (Fig. S3c). Because each Visium spot may contain multiple cells, we used scRNA-seq as a reference dataset and applied robust cell type decomposition (RCTD) to infer the cellular composition of spatial spots. Thus, the spatial transcriptomic data were interpreted at spot-level resolution and further integrated with single-cell transcriptomic information to map epithelial subclusters within their tissue context. We found that SAA1+Epi was significantly greater in Epithelium-3 than in the other subclusters (Fig. 2b, and Fig. S2d) and that the location of SAA1+Epi was highly consistent with that of the Epithelium-3 region (Fig. 2c). Functional profiling confirmed that SAA1+Epi shared key markers (MMP13 and SAA2) and GO terms (e.g., response to bacterium and immune activation) with Epithelium-3 but displayed heightened inflammatory signatures (Fig. 2d, and Fig. S3d), positioning it as a pathogenic effector within this niche. Notably, SAA1+Epi was expanded in periodontitis patients (Fig. S3e), suggesting a potential role in disease progression. To validate its clinical relevance, we analyzed both mouse models and human gingival tissues. In the periodontitis mouse model, SAA1+Epi was markedly enriched in diseased gingiva (Fig. S3f, g), which correlated with bone loss (Fig. S3h). Similarly, in human periodontitis patients, SAA1+Epi accumulation aligned with inflammatory lesions (Fig. 2e), mirroring its spatial distribution in the ST data. Collectively, our multi-omics approach identified SAA1+Epi as a previously unrecognized pathogenic epithelial subcluster that orchestrates periodontitis progression through sustained inflammatory signaling and immune microenvironment remodeling in invasive front zone Epithelium-3. The consistent spatial enrichment of SAA1+Epi in both human periodontitis lesions and experimental models, coupled with its distinct proinflammatory signature, positions this epithelial subcluster as a key mediator of disease pathogenesis in periodontal inflammation.

P. gingivalis LPS enriched in subgingival dental plaque activates SAA1+Epi through the TLR2-PITX2 axis

The junctional epithelium serves as a critical barrier against periodontal pathogens, and the emergence of SAA1+Epi suggests a specialized host response to microbial challenge. To investigate the microbial triggers of SAA1+Epi, subgingival dental plaque and metagenomic sequencing analyzes from the same patient as ST slices (healthy controls, n = 2; periodontitis patients, n = 5) were performed. According to the results of the metagenomic sequencing, the microbial species compositions of periodontitis patients were quite different from those of healthy people (Fig. S4a, b), with the abundance of periodontal pathogens increasing significantly, especially those of the periodontal “red complex” (Porphyromonas gingivalis (P. gingivalis), Treponema denticola (T. denticola), Tannerella forsythia (T. forsythia) and Fusobacterium nucleatum (F. nucleatum)) compared with those of the “orange complex” (Fig. S4c), which is also consistent with previous research results.^31–33 Strikingly, single-cell host-microbiome interaction analysis (SAHMI) revealed P. gingivalis as the one of the most prominent microbial signals detected in association with epithelial cells in periodontitis, with significantly more microbial unique molecular identifiers (UMIs) than other pathogens (Fig. 3a). The functional annotation of P. gingivalis-invaded cells revealed profound upregulation of phagocytosis, immune activation, and chemotaxis pathways (Fig. 3b), mirroring the transcriptional signatures of SAA1+Epi and Epithelium-3 and suggesting that P. gingivalis plays a direct role in shaping this inflammatory epithelial subcluster. Moreover, P. gingivalis was enriched in the gingival epithelium layers collected from periodontitis patients and located in the gingival epithelium basal layer (Fig. 3c), indicating its role in breaching the junctional epithelium barrier and sustaining chronic inflammation. Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis of the metagenome of subgingival dental plaque revealed that the LPS-related pathway was significantly upregulated (Fig. S4d); in particular, the LPS-related pathway in P. gingivalis was the most upregulated pathway in periodontitis patients (Fig. S4e). Therefore, based on the metagenomic sequencing data, we identified all biosynthetic pathways related to LPS. The relative contribution of each taxonomic group to these pathways was quantified and visualized by Taxonomic Stratification of Function. Although F. nucleatum, P. gingivalis, and Prevotella intermedia were suggested to be major contributors to LPS-related pathways (Fig. S4f), the contribution of P. gingivalis to LPS was most significantly different between healthy controls and periodontitis patients (Fig. S4g), indicating that P. gingivalis is the dominant virulence factor that triggers SAA1+Epi and promotes periodontitis. To examine whether P. gingivalis LPS could activate an SAA1+Epi-associated inflammatory epithelial program, human gingival epithelial cells (HGE) were treated with P. gingivalis LPS. The expression of the most remarkable representative SAA1+Epi-associated markers, including SAA1, SAA2, MMP13, and CSF3, was significantly increased (Fig. S5a). Given that SAA1+Epi was spatially enriched in Epithelium-3 near the junctional epithelial interface, this in vitro model was used to mimic epithelial responses to bacterial LPS stimulation at the microbe-facing epithelial region. These results not only reveal that P. gingivalis LPS is the primary activator of SAA1+Epi differentiation but also highlight its ability to accelerate epithelial inflammation in periodontitis.

Fig. 3.

P. gingivalis LPS induced TLR2 and PITX2 expression increased in SAA1+Epi. a The heatmap of the proportion of microbe UMIs of P. gingivalis, T. denticola, T. forsythia, and F. nucleatum reads in total healthy control and periodontitis groups’ reads using the scRNA-seq dataset. The thermodynamic value represents the relative UMIs in every group. b The dotplot shows GO functional analysis of differentially expressed genes (DEGs) in cells associated with the Porphyromonas genus and without those associated with the Porphyromonas genus. P-values were determined by the Wilcoxon test, and the thermodynamic value represents the P-value. The X axis represents the enrichment score. c Representative images of Hematoxylin and Eosin (H&E) staining, P. gingivalis (Green) and DAPI nuclei (Blue) stained with FISH in health and periodontitis groups (n = 3). The black box represents the enlarged portion of the fluorescent image. The black dots represent the boundary between the epithelium area and the connective tissue. d A Heatmap of PRRs gene expression level in 6 epithelium cell subclusters, the thermodynamic value represents the relative expression level. e, g SAA1 mRNA relative expression in inhibitors (C29) and PITX2 siRNA treated with P. gingivalis LPS in HGE cells. ^*P < 0.05, ^**P < 0.01, ^***P < 0.001, ^****P < 0.000 1. f SCENIC analysis of transcription factors of all epithelial cell subclusters

To determine how P. gingivalis LPS induces SAA1+Epi, the expression patterns of pattern recognition receptors (PRRs) in all types of epithelial cell subclusters were first analyzed. Toll-like receptors (TLRs) were highly expressed in epithelial cells, and the expression of TLR2 was highest in the SAA1+Epi group (Fig. 3d). By combining and comparing the results of scRNA-seq and metagenomic sequencing from the same patient, the relative expression of both TLR2 and P. gingivalis LPS was significantly increased in periodontitis patients (Fig. S5b), and P. gingivalis LPS also significantly upregulated the relative level of TLR2 in HGE cells (Fig. S5c). Moreover, inhibiting TLR2 expression significantly decreased the ability of P. gingivalis LPS to induce SAA1+Epi (Fig. 3e, and Fig. S5d). To elucidate the mechanism by which P. gingivalis LPS activated SAA1+Epi through TLR2, the transcription factors (TFs) FOXC2, PITX2, and CEBPB were identified as possible core TFs according to the Single-Cell Regulatory Network Inference and Clustering (SCENIC) analysis (Fig. 3f), in which PITX2 could regulate most marker genes of SAA1+Epi (Data S2) and P. gingivalis LPS significantly increased the expression of PITX2 in HGE cells (Fig. S5e). The inhibition of TLR2 expression significantly decreased the expression of PITX2 (Fig. S5d), and the knockdown of PITX2 in HGEs significantly reduced the ability of P. gingivalis LPS to induce SAA1+Epi (Fig. 3g, and Fig. S5f–i), but the overexpression of PITX2 in HGEs significantly increased the ability of P. gingivalis LPS to induce SAA1+Epi (Fig. 4a, and Fig. S5j, k), indicating that PITX2 is a functionally important downstream TF in the TLR2-associated induction of SAA1+Epi. The phosphorylation of ERK1/2 and Smad2/3 has been reported to be important for TLR2 signaling,^34–36 and PITX2 can be regulated by Smad2/3;^37,38 therefore, ERK1/2 and Smad2/3 phosphorylation were inhibited in HGEs. This inhibition significantly reduced the ability of P. gingivalis LPS to induce SAA1+Epi (Fig. 4b–e, and Figs. S5l and S6a–g). Importantly, inhibition of TLR2 repressed the phosphorylation of both ERK1/2 and Smad2/3 (Fig. 4b, and Fig. S5l), and inhibition of ERK1/2 phosphorylation repressed the phosphorylation of Smad2/3 (Fig. 4e, and Fig. S6f), whereas inhibition of Smad2/3 phosphorylation did not significantly influence the phosphorylation of ERK1/2 (Fig. 4e, and Fig. S6g). Neither knockdown nor overexpression of PITX2 affected the phosphorylation of either ERK1/2 or Smad2/3 (Fig. 4b, c, e, and Fig. S6b, c), indicating that P. gingivalis LPS activated SAA1+Epi through the TLR2-ERK1/2-Smad2/3-PITX2 axis (Fig. S6h). The hierarchical organization of this signaling cascade, from microbial ligand recognition to specific transcriptional regulation, provides a framework for understanding epithelial responses to periodontal pathogens and suggests novel strategies for modulating these responses in therapeutic contexts.

Fig. 4.

TLR2-PITX2 axis activates SAA1+Epi differentiation through P. gingivalis LPS. a SAA1 mRNA relative expression in wildtype and overexpressing (OE) PITX2 HGE cells treated with P. gingivalis LPS. b Western blot analysis of the levels of total and phosphorylated ERK1/2 and Smad2/3 proteins in HGE cells with inhibitors (C29) and PITX2 siRNA treated with P. gingivalis LPS. c Western blot analysis of the levels of total and phosphorylated ERK1/2 and Smad2/3 proteins in OE PITX2 HGE cells treated with P. gingivalis LPS. d SAA1 mRNA relative expression in different inhibitors (SCH772984 and LY364947) treated with P. gingivalis LPS in HGE cells. e Western blot analysis of the levels of total and phosphorylated ERK1/2 and Smad2/3 proteins in HGE cells with inhibitors (SCH772984 and LY364947) treated with P. gingivalis LPS.^*P < 0.05, ^**P < 0.01, ^***P < 0.001, ^****P < 0.000 1

PITX2 suppresses SAA1+Epi-induced pathogenesis

To verify that the phosphorylation of both ERK1/2 and Smad2/3 activated the TLR2-PITX2 axis to induce SAA1+Epi and promote the development of periodontitis, multiplex immunofluorescence (mIF) staining was performed in periodontitis model mice infected with P. gingivalis. The p-Erk1/2 and p-Smad2/3 expression levels were significantly higher in the mice with periodontitis than in the control mice, and p-ERK1/2 and p-Smad2/3 colocalized with Saa1+Epi (Fig. 5a, b), indicating that P. gingivalis LPS potentiates ERK1/2 and Smad2/3 phosphorylation, which in turn drives Saa1+Epi induction and exacerbates periodontitis progression.

Fig. 5.

PITX2 as a core transcription factor (TF) in activating SAA1+Epi. a, b Represent images of mIF staining of DAPI (Blue), KRT5 (Red), MMP13 (Green), p-ERK1/2 (Magenta), and p-Smad2/3 (Magenta) in mice. Yellow arrows show the Saa1+Epi in the mouse epithelium. Magenta arrows show the p-ERK1/2 cells in a, and the p-Smad2/3 cells in b. (Scale bars, 50 μm.) c Quantification of bone density parameters and alveolar bone region. Results are shown as mean ± SD, n = 7. d The represent images of H&E staining of the mice periodontitis models (n = 3). e Representative images show the distance between the alveolar bone crest and the cementoenamel junction in the maxillary (n = 7) through 3D reconstructed analysis. (Scale bars, 1 mm) f Represent images of mIF staining (n = 3) of DAPI (Blue), KRT5 (Magenta), and MMP13 (Green). (Scale bars, 50 μm.) The White arrows show the Saa1+Epi in the mouse epithelium

Given the pivotal role of SAA1+Epi in periodontal pathogenesis, we next assessed whether pharmacological inhibition of PITX2 could mitigate periodontitis. Localized delivery of Pitx2 siRNA in a periodontitis mouse model was performed (Fig. S6i). The inhibition of Pitx2 expression significantly inhibited the induction of Saa1+Epi and reversed the progression of periodontitis, as the injection of Pitx2 siRNA significantly reversed alveolar bone loss, bone density loss, and local inflammatory cell infiltration but decreased osteoclast activity (Fig. 5c–f). Collectively, these data establish PITX2 as a master regulator of SAA1+Epi-driven pathogenesis and suggest its targeted inhibition as a precision strategy for periodontitis intervention.

SAA1+Epi induced inflammatory fibroblasts through TGF-β to aggravate periodontitis

After clarifying the characteristics of SAA1+Epi from the Epithelium-3 inflammatory region and its therapeutic targeting benefit in periodontitis treatment, we further investigated how inflammatory SAA1+Epi transmitted inflammatory signals to reshape the local inflammatory microenvironment and promote the development of periodontitis. First, the inflammatory-adjacent region stromal spots beneath Epithelium-3 and the noninflammatory-adjacent region beneath Epithelium-4 on ST slices were located (Fig. S7a). Compared with the noninflammatory-adjacent region, the inflammatory-adjacent region had significantly upregulated “inflammatory response” and “immune response” (Fig. S7b), indicating that the inflammatory microenvironment beneath the junctional epithelium and SAA1+Epi could remodel this microenvironment. The communication between SAA1+Epi and other cells from stromal tissue was subsequently analyzed, and fibroblasts were determined to be the primary cells that received the signal modules from the SAA1+Epi (Fig. S7c). Four subclusters of myofibroblasts (CCL2+myoFB, IGLC+myoFB, USTN+myoFB and COL4A1+myoFB), three subclusters of fibroblasts (C3+FB, IGHG+FB and ACTA2+FB), and two subclusters of neurofunctional fibroblasts (neo1+FB and neo2+FB) were identified from the scRNA-seq data analysis (Fig. 6a). C3+FB is a type of inflammatory fibroblast that not only regulates the inflammatory response but also mediates epithelial proliferation (Fig. S7d). GO enrichment analysis suggested that C3+FB was associated with NF-κB and PI3K-Akt related inflammatory signatures, together with complement and caspase-related programs (Fig. S7e). Notably, C3+FB demonstrated the most robust intercellular crosstalk with SAA1+Epi (Fig. S7f), forming a pathogenic axis that perpetuates gingival inflammation. mIF staining of the periodontitis mice also revealed that more C3+FB were enriched near Saa1+Epi, but inhibition of Saa1+Epi by the injection of siPitx2 significantly decreased both Saa1+Epi and C3+FB (Fig. 6b, c). Critically, this fibroblast-epithelial alliance was conserved in human periodontitis, where C3+FB accumulated prominently beneath SAA1+Epi-enriched zones (Fig. 6d), reinforcing its clinical relevance. Together, these findings illuminate a previously unrecognized SAA1+Epi/C3+FB signaling circuit that acts as a keystone driver of inflammatory environment maintenance, offering a novel framework for multitargeted therapeutic strategies aimed at disrupting this synergistic interplay in periodontitis pathogenesis. TGFβ and PI3K-AKT were the pathways most markedly upregulated by the interactions between SAA1+Epi and C3+FB; therefore, we hypothesized that SAA1+Epi induced C3+FB through TGFβ or PI3K-Akt-mediated regulation (Fig. 6e). To confirm this direct communication, SAA1+Epi were induced by P. gingivalis LPS and cocultured with fibroblasts in Transwell plates (Fig. S8a). The most remarkable marker genes in C3+FB (C3, SULF2, and FGF7) were significantly upregulated (Fig. 6f, and Fig. S8b). The relative level of TGFβ mRNA and phosphorylation of both PI3K and Akt were significantly increased in fibroblasts (Fig. 6g, and Fig. S8c, d). The inhibition of PITX2 reduced the induction of C3+FB, but the overexpression of PITX2 increased their activation (Fig. 6h, i, and Fig. S8e–g), indicating that SAA1+Epi could directly activate the inflammatory C3+FB. Moreover, inhibition of TGFβ I/II receptors in the Transwell co-culture system significantly reduced the induction of C3+FB-associated markers in fibroblasts, indicating that this epithelial cell-fibroblast communication depends, at least in part, on TGFβ receptor-related signaling in fibroblasts (Fig. S8h).

Fig. 6.

SAA1+Epi induced C3+FB enrichment in the region close to the Epithelium-3. a The UMAP of Fibroblast subclusters. b The representative images of mIF staining of DAPI (Blue), KRT5 (Yellow), MMP13 (Green), COL3A1 (Red), and C3 (Cyan) in mice. Green arrows: Saa1+Epi. White arrows: C3+FB. c Quantification of the percentage of C3+/COL3A1+ cells determined by immunofluorescence staining (n = 3). d The representative images of mIF staining of DAPI (Blue), KRT5 (Red), MMP13 (Yellow), COL3A1 (Cyan), and C3 (Green) in humans. Cyan arrows: C3+FB. Orange arrows: SAA1+Epi. KEGG functional analysis of DEGs of SAA1+Epi ligand and C3+FB receptors. e KEGG functional analysis of DEGs of SAA1+Epi ligand and C3+FB receptors. The X axis represents the enrichment score. f, g C3, TGFβ mRNA relative expression level in fibroblasts in transwell assays. h, i C3 mRNA relative expression level in fibroblasts in transwell assays. ^*P < 0.05, ^**P < 0.01, ^***P < 0.001, ^****P < 0.000 1. Scale bars, 50 μm

SAA1+Epi may contribute to NK cell recruitment through chemokine networks

To further determine the impact of inflammatory SAA1+Epi on the microenvironment beyond its ability to induce inflammatory C3+FB, the communication of ligand-receptors between SAA1+Epi or all fibroblasts and immune cells was analyzed, and SAA1+Epi and fibroblasts were shown to engage mainly in cross-talk with myeloid cells and NK cells (Fig. S9a), suggesting that NK cells and myeloid cells may serve as underappreciated effectors in periodontitis pathogenesis. The inflammatory adjacent region beneath Epithelium-3 also showed enrichment of “positive regulation of natural killer cell chemotaxis” (Fig. S7b), suggesting the active recruitment and functional engagement of NK cells in disease progression.

Through comprehensive cytokine profiling of epithelial subclusters, we identified SAA1+Epi as a potent inflammatory hub that markedly upregulated the expression of proinflammatory mediators, including CSF3, CSF2, and IL10 (Fig. S9b). Based on the annotation and clustering of all immune cells (Fig. 7a, b), SAA1+Epi had the closest interaction with myeloid cells and NK cells (Fig. 7c, and Fig. S9c). In contrast, NK cells were enriched in the region adjacent to the inflammation in periodontitis (Fig. 7d, e, and Fig. S9d, e), indicating that SAA1+Epi possibly recruited more NK cells to the surrounding gingival tissue. This selective NK cell recruitment pattern, coupled with the cytokine signature of SAA1+Epi, supports an association between SAA1+Epi and NK cell recruitment.

Fig. 7.

SAA1+Epi recruits NK cells to shape the immune environment. a, b UMAP of immune cell subclusters of scRNA-seq data. c Number of ligand-receptor interactions among SAA1+Epi and immune cells. d NK cells relative ratio in the stroma region, unpaired Student’s t-test, ^*P < 0.05. e Bar chart of NK cells relative ratio in the inflammatory adjacent region and the non-inflammatory adjacent region in the slices of ST. f Representative images of mIF staining of DAPI (Blue), KRT5 (Red), MMP13 (Green), and CD56 (Magenta) in human gingiva tissues. Green arrows: SAA1+Epi. Magenta arrows: NK cells. (Scale bars, 50 μm) g Representative images of mIF staining of DAPI (Blue), KRT5 (Red), MMP13 (Green), and CD56 (Magenta) in the mouse gingival tissues. Green arrows: Saa1+Epi. Magenta arrows: NK cells. (Scale bars, 50 μm). Quantification of CD56 immunofluorescence intensity per mm² (n = 3). ^*P < 0.05, ^**P < 0.01

By analyzing the communication functions between SAA1+Epi and NK cells (Data S3), we found that the terms “NK cells mediated cytotoxicity” and “antigen processing and presentation” were significantly enriched (Fig. S9f). Moreover, differentially expressed ligands and receptors between SAA1+Epi and NK cells (Fig. S9g), “cytokine-cytokine receptor interaction” and “chemokine signaling pathway”, were enriched in periodontitis patients (Fig. S9h), indicating that SAA1+Epi activated the immune activity of NK cells through chemokines. The expression of different types of chemokines in epithelial subclusters was analyzed, and SAA1+Epi highly expressed CXCL6, CCL25, and CXCL5, which could activate the expression of the chemokine receptors CXCR1 and CXCR2 in NK cells (Fig. S9i). Compared with the inflamed gingival tissues from volunteers diagnosed with periodontitis, fewer SAA1+Epi and recruited NK cells were observed in healthy controls (Fig. 7f). In the periodontitis mouse model, Saa1+Epi was significantly increased in mice with periodontitis, and more NK cells were recruited near the Saa1+Epi, while the inhibition of Pitx2 decreased Saa1+Epi and NK cell recruitment (Fig. 7g), indicating that inflammatory Saa1+Epi recruited NK cells to promote the development of periodontitis and that targeting Pitx2 to inhibit Saa1+Epi could reshape the periodontal inflammatory microenvironment. These findings extend existing knowledge of epithelial-immune communication in the junctional epithelium region by implicating an SAA1+Epi-associated chemokine program.

In order to comprehensively summarize the role of gingival epithelial heterogeneity in the pathogenesis of periodontitis, we have constructed a comprehensive conceptual framework (Fig. 8). Our study elucidates a previously unrecognized pathogenic circuit involved in periodontitis pathogenesis in situ, wherein P. gingivalis LPS-mediated activation of the TLR2-PITX2 axis orchestrates the reprogramming of gingival epithelial cells into a specific inflammatory SAA1+Epi subcluster. This discovery fundamentally advances our understanding of periodontal disease mechanisms by identifying the gingival epithelium not only as a passive barrier but also as an active orchestrator of disease progression through its acquisition of a proinflammatory phenotype. Transformed SAA1+Epi cells establish a self-amplifying inflammatory niche through the following dual mechanisms: TGFβ-mediated recruitment of pathogenic C3+FB subclusters and paracrine CXCL6-dependent activation of NK cells, thereby creating a robust feedforward loop that perpetuates tissue destruction and alveolar bone loss. Importantly, our identification of the TLR2-PITX2 axis as the central regulatory hub coordinating epithelial dysfunction, stromal remodeling, and immune hyperactivation represents a paradigm shift in periodontitis research, bridging the critical knowledge gap regarding epithelial contribution to disease pathogenesis (Fig. 8). These findings provide a novel conceptual framework that moves beyond conventional microbial-centric models, offering transformative therapeutic potential through targeted disruption of this host-derived pathogenic axis, which may enable true disease modification rather than symptomatic microbial suppression in the management of periodontitis.

Discussion

The integrity of the epithelial barrier is fundamental for maintaining tissue homeostasis across organ systems.³⁹ The mechanisms through which microbial-induced epithelial injury drives connective-immune remodeling during chronic inflammatory diseases require systematic investigation.^40–42 In this study, we employed periodontitis in the oral cavity as a model to investigate the molecular mechanisms of the disruption of epithelial homeostasis mediated by imbalanced microorganisms and the interaction between epithelial cells and connective tissues. Our results delineate a defined epithelial reprogramming pathway driven by P. gingivalis and mediated through the TLR2-PITX2-SAA1 axis, revealing a spatially restricted inflammatory epithelial subcluster (SAA1+Epi) that orchestrates multicellular crosstalk with fibroblasts and immune cells to perpetuate local inflammation and mediate tissue destruction.

ST analysis revealed four distinct epithelial zones with unique spatial distributions. Notably, the term “response to bacterium” indicated that Epithelium-3 was strongly associated with periodontal pathogens, and we identified this zone as an “invasive inflammatory zone”. Epithelium-3 exhibited significant expansion in inflammatory gingiva and markedly upregulated the expression of proinflammatory mediators, including SAA2 and MMP13. Intriguingly, Quinn et al. reported a KRT19+ epithelial zone in which genes with differential expression support the potential immune roles of various structural cell types and may be related to periodontal pathogen invasion in periodontitis (n = 30).²⁹ ST analysis revealed that the expression of some of the same marker genes in Epithelium-3 was consistent with that in the KRT19+ epithelium and was located in a similar area adjacent to the basal cell layer. Furthermore, Shen et al. reported the expression of Epi-1 (KRT5, KRT14, and KRT15) (n = 8), which may be related to inflammation in the epithelial basal cell layer.⁴³ In inflamed gingival tissue, the proportion of Epi-1 increased. Given its strategic positioning at the microbial-epithelial interface and its pronounced inflammatory signature, we propose that Epithelium-3 serves as a critical gateway for initial bacterial invasion into the epithelium.

A key conceptual advance of this work is the recognition that the gingival epithelial cells are not merely passive targets of microbial challenge, but active contributors to inflammatory niche formation. In a previous study, Williams et al. reported that H.Epi 1.2 is a special subcluster of epithelial cells found only in gingival tissues (n = 8), and its gene expression characteristics are consistent with those of the inflammatory response, such as IL1B, SLPI, SAA1, and S100A8.²⁶ However, H.Epi 1.2 only showed characteristics of neutrophil activation, indicating that the epithelium participates in the immune response in the periodontium. We determined that the SAA1+Epi subcluster is the principal inflammatory effector, exhibiting marked upregulation of proinflammatory markers (SAA1, SAA2, MMP13, and CSF3) along with abundant expression of inflammatory chemokines and cytokines. Importantly, SAA1+Epi was distributed mainly in Epithelium-3. Functional analyzes confirmed its pivotal role in the bacterial response and immunoinflammatory processes. More importantly, for the analysis of the in situ paired samples, we more precisely reconstructed that SAA1+Epi at the epithelial spatial location is the main inflammatory functional characteristic cell in Epithelium-3. Within this practical context, our study design integrated ST, scRNA-seq, and orthogonal experimental validation to strengthen biological inference beyond any single dataset alone. Taken together, these observations place gingival epithelial heterogeneity into a more explicit host-microbe-tissue framework and advance current understanding of how epithelial dysregulation may actively contribute to periodontal disease progression.

Although previous studies have proposed that specific epithelial regions or subcluster cells respond to inflammation, the mechanism by which microbes directly modulate epithelial cell heterogeneity remains unexplored.^44–46 Our study revealed pronounced infiltration of the keystone pathogen P. gingivalis within distinct gingival epithelial niches, with its LPS identified as a critical virulence effector. Mechanistically, P. gingivalis LPS engages TLR2 on epithelial cell surfaces, initiating a signaling cascade characterized by ERK1/2-Smad2/3 phosphorylation and culminating in PITX2-mediated transcriptional activation. This axis drives the differentiation of SAA1+Epi, directly linking microbial challenge to inflammatory epithelial remodeling. Therefore, we verified that a specific pathway, the P. gingivalis-TLR2-PITX2 axis, could activate the differentiation of inflammatory epithelium SAA1+Epi. One limitation of the proposed ERK1/2-Smad2/3-PITX2 signaling model is that the hierarchy was mainly inferred from pharmacological inhibition and Western blot analysis. Although these results are consistent with ERK1/2 acting upstream of Smad2/3 in the present system, inhibitor-related off-target effects and parallel pathways cannot be fully excluded. Moreover, whether Smad2/3 directly regulates PITX2 transcription remains to be determined, and future studies using Smad-binding motif analysis, chromatin immunoprecipitation, and promoter-reporter assays will be needed to validate this regulatory relationship. At the same time, increasing evidence indicates that periodontitis is driven by polymicrobial interactions rather than by a single pathogen alone. In addition to P. gingivalis, other key periodontal pathogens, such as F. nucleatum, have also been reported to disrupt epithelial barrier integrity, activate inflammatory signal pathways, and modulate host immune responses, suggesting that multiple microbial species may converge on epithelial remodeling through distinct but potentially overlapping mechanisms. Likewise, although our data support a TLR2-dependent response in the present system, previous studies have also highlighted an important role for TLR4 signaling in periodontal inflammation, particularly in mediating host responses to Gram-negative bacterial products and amplifying downstream inflammatory cascades. These observations suggest that the regulation of inflammatory epithelial states in vivo is likely more complex and may involve coordinated inputs from different microbial ligands and pattern-recognition receptors. From this perspective, our results support P. gingivalis as an important driver of epithelial inflammatory remodeling, while also pointing to a broader host-microbe regulatory network that warrants further investigation. Collectively, these findings highlight the importance of locally enriched microbial virulence signals in shaping epithelial inflammatory states and further support a spatially organized host-microbe interaction framework in human periodontitis.

In our study, PITX2 is an important transcription factor that induces the differentiation of SAA1+Epi. Through integrated multi-omics analysis and in vivo/in vitro validation, our data support PITX2 as a major regulator of the SAA1+Epi program under the present experimental conditions. Genetic or pharmacological inhibition of PITX2 significantly suppressed SAA1+Epi generation and effectively ameliorated experimental periodontitis. Although local siPitx2 delivery is not strictly cell-type-specific, its administration near the junctional epithelium and the reduction of MMP13/KRT5 double-positive epithelial cells after treatment support suppression of the epithelial inflammatory program. Future studies using epithelial-specific Pitx2 deletion or cell-type-targeted delivery will be needed to further define the cell-type-specific role of PITX2 in vivo. Previous studies have established the pleiotropic roles of transcription factor PITX2 in embryonic development and disease pathogenesis, particularly in regulating organogenesis of the heart, eyes, and pituitary gland.^47–49 However, its functional involvement in inflammatory regulation, especially its potential in governing inflammatory epithelial cell differentiation, remains largely unexplored.^50,51 Despite these limitations, our mechanistic dissection of bacteria-induced SAA1+Epi differentiation provides crucial insights for modulating periodontal immunoregulation. The identification of the PITX2-mediated regulatory axis offers therapeutic targets for the precision treatment of periodontal disease, potentially informing strategies to restore tissue homeostasis while controlling pathological inflammation.

The pathogenic relevance of SAA1+Epi extends beyond its intrinsic inflammatory signature. By integrating scRNA-seq, multiplex immunofluorescence (mIF), and in vitro cell culture experiments, we validated that blocking SAA1+Epi significantly inhibits C3+FB enrichment through TGFβ signaling, while simultaneously activating fibroblast inflammatory phenotypes via the PI3K/AKT pathway. On the other hand, our study elucidated the regulatory role of SAA1+Epi on NK cells. We found that SAA1+Epi significantly upregulates the expression of chemokines (CXCL6, CXCL5, CXCL1) in NK cells, facilitating their recruitment. Ren et al. also identified that C3+FB could promote M1 polarization of macrophages and differentiation of osteoclasts, thereby increasing the levels of inflammation and destruction in periodontal soft and hard tissues.²⁸ Our results demonstrated the effect of SAA1+Epi on the differentiation of C3+FB, further confirming the significant role of SAA1+Epi in maintaining the local inflammatory environment and exacerbating periodontitis. As key innate immune effectors, neutrophils rapidly initiate antimicrobial responses through phagocytosis and ROS production.^52,53 Periodontal tissues coordinate the expression of local defense mediators to guide neutrophil extravasation from the vasculature and migration through gingival tissues, culminating in their accumulation within the gingival sulcus.^54,55 Our research emphasized the significant role of NK cells in periodontitis. Crosstalk between SAA1+Epi and connective tissues establishes an immunomodulatory network that sustains the inflammatory microenvironment and exacerbates periodontitis, highlighting SAA1+Epi inhibition as a promising therapeutic target for periodontal disease. Although we focused on NK cells because they showed consistent spatial enrichment adjacent to inflamed Epithelium-3 and were reduced after PITX2 inhibition, our interaction analyzes also suggested potential communication with additional lymphocyte populations, including cytotoxic T cells and MAIT cells. These populations were not investigated in depth here, and future work will be required to determine whether they represent parallel or cooperative immune axes downstream of inflammatory epithelial remodeling.

Interestingly, in the analysis of epithelial cell subclusters, except for SAA1+Epi, we found that VIM+Epi showed elevated expression of EMT markers,⁵⁶ including VIM, TWIST, CXCL13, and THY, and VIM+Epi brought out more communication with SAA1+Epi than other epithelial cell clusters, indicating that it may aggravate periodontitis with SAA1+Epi collaboration. Some studies revealed that EMT involves the destruction of interepithelial connexin,^57–59 which will increase the permeability of the epithelium, make it easier for local pathogens to invade internal tissues, and aggravate tissue inflammation.^60,61 In the future, we will deeply explore the relations between VIM+Epi and SAA1+Epi to provide more comprehensive insights into the pathogenesis of periodontal inflammation.^62,63

An important consideration is the relatively small number of healthy control samples in this study. Although the inclusion of clinically healthy gingival tissues is essential for defining disease-associated epithelial remodeling, acquisition of such tissues is inherently constrained by ethical considerations, because only limited amounts of clinically healthy gingiva can be obtained without causing unnecessary injury. Therefore, we prioritized biological signals that showed not only statistical differences but also consistent effect direction, spatial localization, cross-modality concordance among ST, scRNA-seq, and metagenomic datasets, and orthogonal validation in human tissues, mouse models, and in vitro systems. Nevertheless, the limited control sample size may reduce the ability to detect subtle transcriptional or microbial differences and may limit the generalizability of the findings to broader patient populations. Future studies with larger, independently recruited, and clinically balanced cohorts will be necessary to obtain more precise effect size estimates, improve statistical power, control for demographic and clinical confounders, and further validate the SAA1+Epi-centered epithelial remodeling model in periodontitis.

In summary, our study employs in situ multi-omics approaches to delineate a spatially restricted “invasive inflammatory zone” and systematically characterize the functional role and regulatory mechanisms of the inflammatory epithelial subcluster SAA1+Epi within this critical region. Despite the inherent challenges associated with studying bacterial-epithelial interfaces, we have identified several actionable signaling pathways and transcription factors with translational potential. Our findings reveal how a spatially organized host-microbe interface drives inflammatory tissue destruction, nominating a druggable transcriptional node for targeted intervention in chronic inflammatory conditions. Collectively, this work advances our understanding of epithelial homeostasis by establishing that microbial disruption of epithelial-connective tissue crosstalk represents a pivotal mechanism in the progression from tissue health to inflammatory disease across the organ epithelial systems.

Materials and methods

Sample collection

Gingival specimens were harvested from 4 healthy people and 6 periodontitis patients who signed informed consent and were following endorsement by the Institutional Review Board of West China Hospital of Stomatology (WCHSIRB-D-2023-293). Fresh samples were immediately used in scRNA-seq and ST. Written informed consent was obtained from all participants, with explicit communication of the research aims prior to enrollment. Because healthy gingival tissue collection was constrained by ethical considerations and the need to avoid excessive removal of clinically healthy tissue, sample allocation prioritized spatial transcriptomics (n = 10). Consequently, paired scRNA-seq data (n = 8) and metagenomic sequencing (n = 7) were available for only a subpopulation of healthy control samples after tissue allocation and quality control.

Healthy sites showed no obvious gingival inflammation, probing depth ≤ 3 mm, no clinical attachment loss, and no bleeding on probing at the sampling site. Periodontitis samples were obtained from sites with clinical signs of periodontal inflammation, increased probing depth, clinical attachment loss, and bleeding on probing. The exclusion criteria included antibiotic use within the previous 3 months, periodontal treatment within the previous 3 months, use of systemic corticosteroids or immunosuppressive drugs, pregnancy or lactation, uncontrolled systemic diseases, autoimmune diseases, malignancy, severe hepatic or renal disorders, and acute oral infection unrelated to periodontitis. Demographic and clinical information, including age, sex, smoking status, systemic comorbidities, recent medication history, periodontal treatment history, sampling sites, and periodontal clinical parameters, was collected and summarized in Data S4.

Animals

Animal samples were collected following endorsement by the Institutional Review Board of West China Hospital of Stomatology (WCHSIRB-D-2023-528). The animal study was conducted in accordance with the ARRIVE guidelines to ensure ethical and transparent reporting of in vivo experiments. Eight-week-old male C57BL/6J mice were fed in this project (Dashuo Experimental Animal Laboratories, Chengdu). All animals were treated at the State Key Laboratory of Oral Diseases, Sichuan University, in standard housing and feeding conditions with pathogen-free facilities for a 12 h cycle of day and night alternation. The construction of the model was carried out under isoflurane gas anesthesia. A 6-0 silk suture was passed through the interdental space between the maxillary first and second molars and finally secured around the maxillary second molar bilaterally to induce experimental periodontitis. The ligatures were positioned in the maxillary molar region throughout the modeling period.⁶⁴ P. gingivalis W83 was smeared on the silk sutures every 2 days for 2 weeks after ligation. Finally, the mice were sacrificed via CO₂ asphyxiation.

Micro-computed tomography (Micro-CT)

After fixation, images of maxillary alveolar bone were acquired using micro-CT (SCANCO Medical, Bruettisellen, Switzerland) with the parameters: 10 μm resolution at 80 kV and 500 μA. The mice maxillary alveolar bones were selected for the region of interest (ROI) of 10 μm spatial resolution. On the basis of the ROI, all 3D reconstruction images we captured of the mouse maxillary samples were analyzed for periodontal alveolar bone loss using SCANCO Visualizer software (Version 1.1.18.0). The severity of periodontal alveolar bone loss was shown by scaling the vertical distance of the cementoenamel junction (CEJ) to the alveolar bone crest (ABC). The loss scores were averaged in every group. Bone volume fraction (BV) and trabecular volume fraction (TV) were assessed in mm³.

Cell lines and cell culture

Human gingival epithelial cells (HGE) and human gingival fibroblasts (HGF) were incubated in DMEM (Gibco, C11995500BT) with 10% fetal bovine serum (FBS, ZETA, Z7185FBS-500), 1% penicillin and streptomycin (Gibco, 15140122). P. gingivalis LPS (100 ng/mL, InvivoGen) was used to stimulate HGE cells for 24 h. All cells were incubated with 5% CO₂ at 37 °C.

RNA isolation and quantitative PCR (qPCR)

Total RNA samples were isolated using the RNAiso Easy kit (Takara, TCH021). Then, the RNA was reverse transcribed into cDNA using the HiScript III RT SuperMix kit (Vazyme, R323-01). qPCR assay was implemented using the Taq Pro Universal SYBR qPCR Master Mix kit (Vazyme, Q712-02) according to the specifications. Gene expression was calculated using the previously described method. All primers used in this experiment were listed in Table S1. All primers used in this paper.

Fluorescence in situ hybridization (FISH)

FISH staining was implemented as described.⁶⁵ Tissue sections were permeabilized at room temperature with 0.1% Triton X-100. Then, lysozyme (30 mg/mL, Solarbio, 12650-88-3) was used to maintain for 20 min at 37 °C in an incubator. Then, the sections were dehydrated separately in ethanol (50%, 75%, 80%, 90%, 95%, and 100%, each for 3 min). The P. gingivalis probe (5′-CAATACTCGTATCGCCCGTTATTC-3′) was contained in a hybridization buffer conjugated with Alexa Fluor 488 (200 nmol/L). Next, the sample was incubated again in the hybridization buffer at 46 °C for 1.5 h. After that, the sample was rinsed with a washing solution for 15 min at 48 °C and pre-cooled deionized water at 4 °C. DAPI was used to show the nuclei. The formula of the hybridization buffer is 0.9 mol/L NaCl, 20 mmol/L Tris-HCl, 0.01% SDS, and 20% formamide. The formula of the washing solution is 20 mmol/L Tris-HCl, 5 mmol/LEDTA, 0.01% SDS, and 215 mmol/L NaCl.

Small interfering RNA (siRNA) transfection

HGE cell transfection was performed using TSnanofect V2 (Tsingke, TSV405) according to the instructions. Cells were plated and grown to 50% confluency for transfection. The transfection complex was prepared with 10 μL TSnanofect V2 and 5 μL of siRNA (100 nmol/L) in 250 μL Opti-MEM. After 8 h of transfection, the old medium was discarded, and the new fresh culture medium was added. The cells were then cultured for another 48 h. siRNAs (Table S1. All primers used in this paper) were used to knockdown PITX2. siRNA Pitx2 (20 nmol/L/time) and a negative control were provided by Tsingke Biotechnology Company and injected locally into periodontal tissue. The periodontitis model was established for 3 weeks, during which Chol-siRNA was injected once every 3 days.

Overexpression plasmid transfection

The human PITX2 coding sequences were cloned into the vector pCDH-CMV-MCS-EF1-RFP-T2A-puro to generate PITX2 expression plasmids. The lentiviral vector and packaging plasmids were co-transfected into 293T cells for lentivirus production. Cells were incubated with the lentivirus for 48 h in 8 μg/mL polybrene (Yeasen, 40804ES76). Finally, used to screen for gene-edited cells.

Western blot

Total protein was extracted using radioimmunoprecipitation assay (RIPA) buffer (Beyotime, P0013B). 10% and 12% sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE) gels (Yeasen, 36246ES10, 36253ES10) were used for performing electrophoresis and then transformed to polyvinylidene difluoride (PVDF) membranes. Next, the PVDF membranes were rinsed in the TBST buffer containing 5% BSA (Solarbio, SW3015). Then, the PVDF membranes were incubated in the primary antibodies under refrigeration (4 °C) overnight. Following primary antibody treatment, samples were exposed to secondary antibody for 1 h at room temperature. In terms of the manufacturer’s instructions (Bio-Rad), signals were captured using an enhanced chemiluminescence system. All the antibodies used in the experiments listed in Table S2. All antibodies used in this paper, and the dilution ratio should be used in accordance with the instructions.

mIF assay

mIF was performed on human gingival tissue and mouse maxillary alveolar bone. The tissues were collected and quickly fixed in 4% paraformaldehyde (Solarbio, P1110) at 4 °C. Next, samples were prepared with paraffin sections. Then, the samples experience deparaffinization using xylene, followed by rehydration through a graded ethanol series. 0.5% Triton X-100 (Solarbio, 9002-93-1) was used to permeabilize. Heat-based antigen retrieval was conducted at 95 °C. After samples were cooled naturally, they were maintained in 3% H₂O₂ for 30 min to block endogenous peroxidase activity. Then, to block, 5% goat serum (Solarbio, SL038) in PBST for 1 h. Primary antibodies were incubated at 4 °C overnight. Rinse the samples by using PBS on the second day for 3 times. Place the samples in PBS and shake them on a pendulum shaker for washing three times, each time for 5 min. After the slices were slightly spun dry, the corresponding TSA was dropped into the circle and incubated at room temperature in the dark for 10 min. Finally, the samples were incubated with DAPI (Thermo, D1306) for 5 min and mounted with coverslips. The pictures were captured with a laser scanning confocal microscope (LSCM) (Olympus). All the antibodies used in the articles are listed in Table S2. All antibodies used in this paper, and the dilution ratio should be used in accordance with the instructions.

Transwell assay

Transwell assays were employed to evaluate the influence of SAA1+HGE on Fibroblasts. Fibroblasts were seeded in 6-well Transwell insert chambers (Nest) after HGE cells were stimulated by P. gingivalis LPS with or without inhibitors or PITX2 siRNA. Cells were harvested after 24 h for further tests.

Spatial transcriptomic sequencing

Fresh gingival tissue samples were oriented with optimal cutting temperature compound (OCT) embedding medium using peel-away molds prior to cryopreservation. Spatial transcriptomic sequencing was performed using a 10× Genomics Visium spatial transcriptome platform according to the manufacturer’s specifications. Serial 10 μm cryosections were mounted on Visium Spatial Gene Expression Slides, followed by H&E staining and brightfield imaging. Subsequent processing included tissue permeabilization, cDNA synthesis via reverse transcription, amplification of cDNA, and DNA library preparation using the Visium Spatial Gene Expression Slide & Reagent Kit (PN-1000184) and Visium Spatial Gene Expression Slide Kit (PN-1000185), based on the standardized 10× Genomics protocol.

Single-cell RNA library preparation and sequencing

Every sample was minced into tiny pieces (< 1 mm³) and then digested in digestion media containing collagenase (Roche, Basel, Switzerland) and DNase I for 30 min at 37 °C. Next, the suspension was filtered using a 40 μm strainer (ThermoFisher Scientific, USA) and then washed with PBS twice. After centrifugation at 300 × g at 4 °C for 6 min, the cell pellet was resuspended and was ready for trypan blue staining and scRNA-seq.

Data processing

Spatial transcriptomics data processing was performed using the 10× Genomics Space Ranger pipeline (Version 2.0.1). FASTQ format files of raw sequencing data were aligned to the GRCh38 human reference genome, with gene expression quantification and spatial barcode assignment subsequently following the manufacturer’s computational workflow. The number of tissue-covered spots in each section ranged from 600 to 1 292, while the gene counts were 3 728–5 631 per spot. The clustering resolution was selected based on preservation of spatial continuity, histological interpretability, and marker gene coherence across epithelial regions. To distinguish tissue overlaying spots from the background, tissue overlaying spots were detected according to the images. The filtered UMI count matrix was then analyzed using the Seurat (Version 4.3.0) R package. SCTransform was used to normalize data and identify the top 3 000 highly variable genes (HVGs). Principal component analysis (PCA) was conducted to reduce dimensionality on the log-transformed gene-barcode matrices of top variable genes. Graph-based clustering was performed to cluster cells according to their gene expression profile with the FindClusters function. Cells were visualized using a 2-dimensional Uniform Manifold Approximation and Projection (UMAP) algorithm with the RunUMAP function. The FindAllMarkers function (test.use = presto) was used to identify marker genes of each cluster. To infer the cell-type composition of each spot, RCTD (Version 1.1.0) was applied. Specifically, the default parameters were used in the create. RCTD function, except for a minimum number of cells > 1 for each cell type and a minimum of UMI counts > 1 per pixel, and the doublet_mode parameter was set to FALSE in the run.RCTD function. Differentially expressed genes (DEGs) were selected using the Seurat function FindMarkers (test.use = presto). P value < 0.05 and |log₂ fold change| > 0.58 were set as the threshold for significantly differential expression. GO enrichment and KEGG pathway enrichment analysis of DEGs were respectively performed using R (Version 4.0.3) based on the hypergeometric distribution.

Processing scRNA-seq data using the 10× Genomics Cell Ranger pipeline (Version 7.0.1). Similarly, the single-cell RNA sequencing raw reads (FASTQ files) were aligned to the GRCh38 human reference genome, and UMI counts were aggregated for each cellular barcode. The resulting UMI count matrix was imported into Seurat (Version 4.0.0) for downstream analysis. To ensure data quality, cells were filtered based on the following criteria: (1) Genes detected per cell (<200); (2) Total UMI counts per cell (<1 000); (3) Gene-to-UMI ratio (log₁₀(GenesPerUMI) < 0.7); (4) Mitochondrial transcript content (>10% of UMIs mapped to mitochondrial genes); (5) Hemoglobin transcript content (>5% of UMIs mapped to hemoglobin genes). The DoubletFinder package (Version 2.0.3) was applied to recognize possible doublet cells. Furthermore, in the subsequent analysis, we discarded these cells. Library size normalization was processed by implementing the NormalizeData function to attain the normalized gene expression data. Subsequently, the global-scaling normalization method “LogNormalize” normalized the gene expression level to estimate every cell by the total expression, multiplied by a scaling factor (10 000 by default), and log-transformed the results.

Cell type annotation

The top 2 000 HVGs were identified using Seurat’s FindVariableFeatures function (selection method = ‘vst’), with mean expression calculated via FastExpMean and dispersion modeled by FastLogVMR. To mitigate batch effects, we used the mutual nearest neighbors correction algorithm⁶⁶ performed in the batchelor R package (Version 1.6.3). Subsequent analyzes included: (1) Graph-based clustering using the FindClusters function to group cells by transcriptional similarity; (2) Nonlinear dimensionality reduction via UMAP implemented in the RunUMAP function, enabling 2D visualization of cellular heterogeneity. For spatial transcriptomic analysis, epithelial regions were first delineated according to H&E morphology, then confirmed using a panel of epithelial marker genes, and further interpreted in conjunction with RCTD-based cell-type decomposition. For scRNA-seq analysis, cell identities were assigned based on the combined expression of multiple canonical marker genes (generally at least three markers per cell type), together with cluster-specific DEGs and consistency with previously reported gingival cell atlases. Specifically, epithelial cells were identified by epithelial markers such as KRT5, KRT14, KRT76, EPCAM, and CSTB; endothelial cells by PLVAP, PECAM1, EMCN, and KDR; B/plasma cells by CD79A, CD74, MS4A1, MZB1, JCHAIN, and SDC1; T/NK cells by CD3D, CD3E, TRBC1, NKG7, XCL1, and KLRD1; fibroblasts by COL1A1, COL3A1, DCN, LUM, and PDGFRA; and myeloid cells by LYZ, FCN1, S100A8, S100A9, CTSS, and LST1.

Cell communication analysis

To predict cell-cell communication analysis systematically, CellPhoneDB (Version2.0) was used to compute ligand-receptor interactions from scRNA-seq data. If 10% of the cells of that type had non-zero read counts for the ligand/receptor encoding gene in a particular cell type, a ligand or a receptor will be defined as “expressed”. After running 1 000 permutations, P-values were calculated with the normal distribution curve generated from the permuted LR pair interaction scores. Igraph and Circlize R packages were displayed to construct cell-cell communication networks and heatmaps.

SCENIC analysis

Transcription factor (TF) activity was analyzed across individual cell subclusters using SCENIC (Single-Cell Regulatory Network Inference and Clustering; Version 1.1.2.1).

SCENIC analysis was conducted employing the RcisTarget (Version 1.2.1) motif database and GRNBoost algorithm, with default parameters applied throughout the analysis. Regulon activity in individual cells was quantified using AUCell (Version 1.4.1). To assess cell type specificity, we computed regulon specificity scores based on Jensen–Shannon divergence, measuring the similarity between binary regulon activity patterns and cell type assignments. Additionally, connection specificity indices (CSI) were derived for every regulon using the scFunctions package (https://github.com/FloWuenne/scFunctions/).

Differential expression and pathway analysis

Differential gene expression analysis between different cell clusters or groups was applied using the Seurat package. DEGs were filtered using the following criteria: (1) P < 0.05 and (2) logFC > 0.5. The expression of genes across different groups was visualized using dot plots.

Gene set variation analysis (GSVA)

GSEABase R package (Version 1.44.0) was used as usual to assess GSVA. All of the gene sets were acquired from the KEGG database (https://www.kegg.jp/) and processed for subsequent analysis. Furthermore, the GSVA R package (Version 1.30.0) was used to estimate pathway activity was allocated to individual cells in the GSVA analysis. Finally, according to the LIMMA package (Version 3.38.3), we screened per cell’s differences in pathway activities. The path diagram is presented through forms such as bubble diagrams and heat maps.

Gene set enrichment analysis (GSEA)

Gene Ontology (GO) and KEGG pathway enrichment analysis were performed using GSEA with the Molecular Signatures Database (MSigDB, Version 7.2), specifically utilizing the C5 (GO gene sets) and C2 (curated KEGG gene sets) collections. The GO and KEGG pathway enrichment results were filtered by P < 0.05, and the top 200 enriched pathways are displayed in bar plots.

Single-cell analysis of host-microbiome interactions (SAHMI)

Raw sequencing data underwent dual-alignment processing, where host transcripts were mapped to the GRCh38 reference genome using Cell Ranger (Version 7.0.1) while microbial reads were aligned against a curated pan-microbial database via Kraken 2 (Version 2.1.2), followed by stringent quality control to exclude low-quality cells and ambient RNA contamination. Genomes from the Human Oral Microbiome Database (HOMD) (https://homd.org), which are not in the standard database, were also downloaded and built using kraken 2-build with default parameters. SAHMI analysis suggested host cell-associated microbial signals enriched for microbial reads, although we cannot fully exclude contributions from ambient or residual microbial RNA.

DNA extraction and Metagenomic sequencing

Total genomic DNA was extracted from dental plaque samples using a modified phenol-chloroform protocol incorporating mechanical lysis (0.1 mm glass beads) and enzymatic digestion (lysozyme/proteinase K), followed by RNase A treatment and column-based purification (ZymoBIOMICS DNA Miniprep Kit). DNA integrity was verified through 1%–1.5% agarose gel electrophoresis, while purity (A₂₆₀/A₂₈₀ ≥ 1.8; A₂₆₀/A₂₃₀ ≥ 2.0) and concentration (>1 ng/μL) were quantified using Qubit 4.0 Fluorometer and NanoDrop spectrophotometry. For library preparation, 50–100 ng of high-quality DNA was fragmented (Covaris M220, 350 bp insert size), end-repaired, and ligated with unique dual-index adapters (Illumina TruSeq DNA PCR-Free). Shear the gDNA sample by transferring each DNA sample to each Covaris tube. Transfer fragmented DNA from each Covaris tube to the corresponding well of the new 0.3 ml PCR plate. For each sample, according to the standard protocol provided by Illumina, Inc., the purified metagenomic DNA was cleaved into segments and libraries. Matched subgingival plaque samples for metagenomic sequencing were available from 7 donors (2 healthy controls and 5 periodontitis patients), reflecting the practical constraints of paired clinical sample collection in this cohort.

Metagenomic analysis

Metagenomic sequencing and bioinformatic processing were performed by OE Biotech Co., Ltd (Shanghai, China). Raw sequencing data (FASTQ format) underwent quality control through Trimmomatic (Version 0.36) for adapter removal and read filtering. For samples derived from host-associated environments, host DNA contamination was removed by alignment to the reference genome using Bowtie 2 (Version 2.2.9), with subsequent exclusion of mapped reads. Cleaned reads were assembled into contigs using MEGAHIT (Version 1.1.2), followed by scaffold fragmentation at gap regions (Scaftig) with a minimum length threshold of 200 bp (or 500 bp). Finally, open reading frames (ORFs) were predicted from assembled scaffolds using Prodigal (Version 2.6.3) and translated into corresponding amino acid sequences. A non-redundant gene catalog was constructed from all predicted genes using CD-HIT (Version 4.6.7) with stringent clustering parameters (95% sequence identity and 90% coverage), where the longest sequence within each cluster was designated as the representative. Quality-filtered reads from each sample were then mapped to this reference gene set (95% identity threshold) using Bowtie2 (Version 2.2.9) for gene abundance quantification. Functional annotation of representative amino acid sequences was performed against NR, KEGG, COG, Swiss-Prot, and GO databases (E-value cutoff: 1 × 10⁻⁵). Taxonomic classification and species abundance profiling were derived from the NR database lineage information, with species-level abundances calculated by aggregating corresponding gene counts. Taxonomic abundance profiles were generated across all standard hierarchical levels (Domain to Species). For functional characterization, gene sets were annotated against the CAZy database using hmmscan (Version 3.1b2), with carbohydrate-active enzyme (CAZyme) abundances derived from summed gene counts. Multivariate analyzes, including PCA, were performed in R (Version 3.2.0) to visualize species and functional abundance patterns. Beta-diversity was assessed through both principal coordinate analysis (PCoA) and non-metric multidimensional scaling of distance matrices.

Statistical analysis

For cell experiments, data were shown as mean ± SD. Two-group analysis used an unpaired two-tailed t-test. Multiple comparisons analysis used a one-way analysis of variance (ANOVA) with Tukey’s adjustment or two-way ANOVA. Correlations between variables were analyzed by the Pearson test. ST and scRNA-seq data were determined by Wilcoxon test and Bayesian Test. GraphPad Prism software (Version 10.0) was used in drawing figures of all analyzes, with independent experimental replicates conducted at least twice to ensure reproducibility. Statistical significance was defined as P < 0.05.

Author contributions

Y.J. Wu, Z.F. Su performed ScRNA-seq and ST, and finished the bioinformatics analyses; B.W. Zhang collected the clinical samples; Y.J. Wu performed the cell and animals assistant with F.J. Zhou; Y.J. Wu and B. Ren wrote the manuscript; R.H. Xi, B. Ren, X. Peng, Y. Lu and Y.Q. Li assistant designed the project; J.Y. Li, X.Y. Wang, F. Chen, L. Yue and X.D. Zhou supervised the project; B. Ren and J.Y. Li revised the manuscript; B.W. Zhang, Z.F. Su, B. Ren and J.Y. Li acquired the funding.

Data availability

Metagenomic sequencing raw data have been deposited to National Center for Biotechnology Information (NCBI) under the BioProject number PRJNA1263682, GSE297924, GSE299838. (Reviewer access).

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41368-026-00464-1.