Comprehensive Single-Cell RNA Atlas of Human Laryngeal Normal, Preneoplastic, and Tumorigenic States

Abstract

Purpose:

The mechanisms driving the progression of vocal cord leukoplakia (VCL) to laryngeal squamous cell carcinoma (LSCC) remain unclear, posing a significant barrier to the effective prevention, early diagnosis, and targeted treatment of LSCC. Therefore, it is essential to characterize the cellular microenvironmental differences between VCL and LSCC at single-cell resolution.

Experimental Design:

In the study, we conducted single-cell RNA sequencing and spatial transcriptomics on surgical tissue specimens obtained from 36 patients diagnosed with vocal cord polyps (VCP), VCL, and LSCC.

Results:

Our study generated the first single-cell atlas of VCP and VCL while expanding the cancer cell atlas of LSCC. This dataset comprises 318,907 cells and 12,679 spatial transcriptomic spots, allowing the identification of distinct cellular subclusters. We observed that VCL, as a transitional lesion between benign and malignant states, exhibits a hybrid microenvironment that mirrors VCP and LSCC, with early signs of immunosuppressive activity. Immunoregulatory cell populations demonstrate significant gene expression and functional pathway differences between VCL and LSCC.

Conclusions:

Our single-cell RNA sequencing and spatial transcriptomics analyses revealed the cellular heterogeneity underlying benign, precancerous, and malignant laryngeal lesions. We identified epithelial subclusters in VCL with malignant potential and observed shared immunosuppressive features with LSCC, suggesting their role in disease progression. These findings provide valuable insights into the molecular transition from VCL to LSCC and emphasize potential targets for early diagnosis and therapeutic intervention.

Translational Relevance.

Vocal cord leukoplakia (VCL) is among the most prevalent precancerous lesions associated with the development of laryngeal squamous cell carcinoma (LSCC); however, its underlying molecular mechanisms of progression remain unclear. We present the first single-cell RNA atlas of vocal cord polyps and VCL and expand the existing LSCC atlas by integrating single-cell RNA sequencing with spatial transcriptomics. At single-cell resolution, we delineated the cellular microenvironment of VCL and performed a comparative analysis of malignant epithelial subclusters in VCL and LSCC, focusing on differential gene expression and functional enrichment. Moreover, immune profiling revealed that VCL harbors an immunosuppressive microenvironment that, while sharing features with LSCC, exhibits distinct characteristics. These findings offer insights into the molecular transition from VCL to LSCC and provide a bioinformatics framework for early detection and targeted intervention. This comprehensive atlas advances our understanding of laryngeal lesion progression and supports precision medicine in head and neck carcinoma.

Introduction

Laryngeal squamous cell carcinoma (LSCC) is among the most prevalent malignancies of the head and neck, arising from the epithelial lining of the larynx. Most LSCC cases arise from precancerous lesions, including vocal cord leukoplakia (VCL), chronic hypertrophic laryngitis, and papilloma (1, 2). Of these, VCL is the most frequently encountered in clinical practice, manifesting as persistent grayish-white patches on the vocal cords with heterogeneous histopathologic features (3, 4). Malignant transformation rates vary according to the degree of dysplasia, with reported rates of 3.7% in nondysplastic, 10.1% in mild-to-moderate dysplastic, and 18.1% in severely dysplastic VCL cases (5, 6). LSCC is often characterized by insidious early symptoms, leading to late-stage diagnosis in more than 50% of patients, frequently accompanied by lymphatic or distant metastases (7). Furthermore, the loss of voice following total laryngectomy significantly diminishes the patient’s quality of life. Consequently, the early detection and effective prevention of precancerous laryngeal lesions are of paramount importance for improving clinical outcomes in LSCC.

The tumor microenvironment (TME) plays a crucial role in tumor initiation and progression. Comprising immune cells, stromal components, secreted factors, and the extracellular matrix, the TME interacts dynamically with epithelial cells to support tumor survival, invasion, and metastasis (8). It also promotes tumor growth by inducing angiogenesis and metabolic reprogramming, such as the Warburg effect, allowing tumors to adapt to hypoxic and acidic conditions (9). Elucidation of TME dynamics has deepened our understanding of tumor biology and accelerated the identification of diagnostic markers and therapeutic targets. TME-informed precision medicine has underpinned the success of immunotherapies in various malignancies, including lung, ovarian, and pancreatic cancers, leading to significant improvements in patient survival and quality of life (10–13). Despite these advances, the molecular mechanisms underlying the progression from VCL to LSCC remain uncertain, partly due to limited studies on the laryngeal TME. In particular, high-resolution studies using single-cell RNA sequencing (scRNA-seq) and spatial transcriptomics are lacking.

To investigate the role of the cellular microenvironment in laryngeal precancerous and malignant progression, we conducted this study with the following primary objectives: (i) to construct a scRNA-seq atlas, including benign [vocal cord polyps (VCP)], precancerous (VCL), and malignant lesions (LSCC), identifying key cellular populations and transcription factors (TF); (ii) to analyze copy-number variations (CNV), highly variable genes (HVG), biological functions, and signaling pathways within epithelial subclusters to elucidate the progression from VCL to LSCC; and (iii) to characterize immune cell heterogeneity and intercellular communication patterns that define the immunosuppressive microenvironment in VCL and LSCC. We applied scRNA-seq and spatial transcriptomics to VCP, VCL, and LSCC samples using the 10x Genomics platform, followed by integrative bioinformatics analysis. This comprehensive dataset offers a valuable resource for exploring the dynamic microenvironmental changes that underpin laryngeal tumorigenesis.

Materials and Methods

Declarations of ethical approval and consent to participate

This study was performed in accordance with the Declaration of Helsinki. The Ethics Committee of The First Affiliated Hospital, College of Medicine, Zhejiang University, approved the study protocols (reference number: IIT20240532B-R1). Written informed consent was obtained from all participants prior to sample collection.

Sample acquisition

This study included 10 VCP, 10 VCL, 10 LSCC, and 10 LSCC precursor (LSCCP) tissues from 30 patients for scRNA-seq and paired tissues from six patients for spatial transcriptomics, all collected at the Department of Otolaryngology, The First Affiliated Hospital, College of Medicine, Zhejiang University (Supplementary Table S1). Both male and female patients were enrolled although sex was not treated as a biological variable due to the male predominance of laryngeal disease. All specimens were obtained from surgeries and met the inclusion criteria outlined in Supplementary Table S2. Due to ethical constraints, VCP tissues were used as substitutes for normal controls. LSCC samples were taken from the tumor core, and LSCCP samples were taken from the tumor margin, ensuring maximal distance from the lesion. Samples for scRNA-seq were preserved in Tissue Storage Solution, whereas those for spatial transcriptomics were fixed in 10% neutral-buffered formalin and embedded in paraffin (FFPE). All reagents, along with their catalog numbers and manufacturers, are listed in Supplementary Table S3.

Preparation of single-cell suspensions

Tissues were transported on ice in a sterile dish with 10 mL of 1× Dulbecco's Phosphate-Buffered Saline (DPBS) to remove residual storage solution and then mechanically minced. Enzymatic digestion was performed at 37°C with shaking at 50 rpm for approximately 40 minutes in 0.25% trypsin and 10 μg/mL DNase I in PBS with 5% FBS. To optimize yield and viability, cells were collected every 20 minutes. The suspensions were filtered through 40 μm strainers to remove debris, followed by red blood cell lysis. Cells were then washed with 1× DPBS containing 2% FBS. Cell viability was assessed using 0.4% Trypan Blue and quantified using a Countess II Automated Cell Counter (Thermo Fisher Scientific). Subsequently, samples were submitted to Majorbio Bio-Pharm Technology Co., Ltd. for downstream analysis (Fig. 1A).

Figure 1.

Single-cell atlas and transcriptional heterogeneity of VCP, VCL, LSCC, and LSCCP. A, Diagram showing the workflow of the present study. B, t-SNE showing the seven major cell types identified in the four types of tissue groups analyzed in this study. Marker genes for epithelial cells: KRT8, KRT15, KRT17, KRT18, KRT19, and EPCAM; marker genes for T cells: CD2D, CD3D, CD3E, and CD3G; marker genes for B cells: CD19, CD79A, CD79B, and MS4A1; marker genes for myeloid cells: CD33, CD68, CD1E, LYZ, and LAMP3; marker genes for NK cells: CD56, CD16, NKP46, and NKP30; marker genes for endothelial cells: PECAM1, CD34, CDH5, VWF, VEGFR, TEK, and CD54; marker genes for fibroblasts: COL1A1, COL1A2, COL3A1, ACTA2, FAP, and S100A4. All marker genes used for cell subtype annotation in this study are listed in Supplementary Table S5. C, Cellular composition and the numbers of cells of all cell types in the different types of samples examined in the present study. The sample-level proportions of cell types are shown in Supplementary Fig. S2H. D, Heatmap of intergroup variability based on HVGs. Red indicates high correlation between samples, whereas blue indicates low correlation. E, SCENIC results of all cell types in all samples. Red indicates that the TF is upregulated in the corresponding cell, and blue indicates downregulation. (A, Created in BioRender. Fu, Z. [2025] https://BioRender.com/s28k379.)

10x scRNA-seq library preparation and sequencing

scRNA-seq was conducted using the Chromium Single Cell 3′ Library & Gel Bead Kit version 3.1 on the 10x Genomics Chromium platform following the manufacturer’s protocols (10x Genomics). Barcoded gel beads with unique molecular identifiers (UMI) and cell-specific barcodes were loaded near saturation to encapsulate individual cells in gel beads in emulsion. Within gel beads in emulsion, cell lysis and hybridization of polyadenylated RNA to bead-bound oligonucleotides occurred. Reverse transcription labeled cDNA with a UMI and barcode at the 5′ end (matching the 3′ end of mRNA). After breaking the emulsion, barcoded cDNA underwent second-strand synthesis, adaptor ligation, and PCR amplification.

Sequencing libraries were prepared from amplified whole-transcriptome products enriched for 3′ ends tagged with UMIs and barcodes, per the Chromium version 3.1 protocol. Library quality and concentration were evaluated using a High Sensitivity DNA Chip (Agilent Bioanalyzer 2100) and Qubit High Sensitivity DNA Assay Kit (Thermo Fisher Scientific). Sequencing was performed on an Illumina NovaSeq 6000 system (Illumina) with 150 bp paired-end reads, generating about 100 GB of high-quality data.

scRNA-seq data processing

scRNA-seq data analysis was performed by Majorbio Co., Ltd. using the Majorbio Cloud analytics platform (www.cloud.majorbio.com). Software tools and databases are listed in Supplementary Table S4. Raw data were processed with Cell Ranger (version 7.1.0, 10x Genomics) to generate gene–barcode matrices. Quality control included exon mapping ratio, Q30 scores, barcode accuracy, and UMI counts. FASTQ files were aligned to the GRCh38.p13 human genome using STAR, and UMIs were quantified after filtering out barcodes not associated with viable cells. The resulting expression matrix was analyzed in Seurat (version 5.0) for quality control, normalization, dimensionality reduction, clustering, and cell type annotation.

All single-cell data were integrated using the Harmony algorithm to correct for batch effects and intersample variability, including differences in sequencing depth, UMI counts, gene expression levels, and mitochondrial gene content. Cells with <200 or >6,000 genes, a mitochondrial gene ratio >20%, or a hemoglobin gene ratio >1% were excluded. For dimensionality reduction and clustering, HVGs were normalized using Seurat’s “ScaleData” (RRID: SCR_007322), principal component analysis was used for dimensionality reduction, and clustering was performed with the Louvain algorithm. t-Distributed stochastic neighbor embedding (t-SNE) was used for visualization.

Identification of cell types and subtypes by nonlinear dimensional reduction (t-SNE)

Cell clustering was performed by constructing a shared nearest neighbor graph based on principal component analysis–reduced data, followed by Louvain community detection (14). Subclustering of specific populations used the same method. Batch effects were corrected using Harmony (15) prior to clustering, except for epithelial subclusters, in which uncorrected data were used to preserve biological variation. Differentially expressed genes (DEG) were identified using the Wilcoxon rank-sum test. Cell clusters in the resulting two-dimensional representation were annotated to known biological cell types using SingleR (16) and canonical marker genes (Supplementary Table S5). Clustering and annotation results were visualized using t-SNE plots.

Differential expression analysis and functional enrichment

DEGs were identified using Seurat’s FindMarkers function, applying a likelihood ratio test. Genes with |log₂FC| > 0.25 and Q ≤ 0.05, where FC indicates fold change, were considered significant. Functional enrichment analyses were performed to assess the biological relevance of DEGs. Gene Ontology terms and metabolic pathway enrichment analyses were conducted using GOATOOLS (https://github.com/tanghaibao/Goatools), with statistical significance determined by Bonferroni-corrected P values ≤ 0.05. The Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis was based on human reference gene sets obtained from the gene set enrichment analysis database.

SCENIC and GSVA

Single-cell regulatory network inference and clustering (SCENIC) is a computational framework designed for the simultaneous reconstruction of gene regulatory networks and the identification of cellular states from scRNA-seq (17). Gene set variation analysis (GSVA) was performed to quantify the enrichment of predefined gene sets, offering insights into functional activity patterns across cell populations. The analysis was conducted using standard GSVA parameters, with the Hallmark gene sets from the Molecular Signatures Database (http://software.broadinstitute.org/gsea/downloads.jsp) serving as the reference (18).

Cell–cell communication analysis

Intercellular communication networks were analyzed using the CellChat package, which infers signaling interactions based on known ligand–receptor pairs (19). The input included scRNA-seq count matrices along with cell type annotations. Ligand–receptor pairs were filtered according to an expression threshold (default: 10%), followed by pairwise comparisons between different cell types. Statistical significance was assessed through enrichment analysis, and significant ligand–receptor interactions were identified based on the resulting P values.

CNV and benign/malignant classification of epithelial-derived cells

CNV analysis was performed using inferred CNV software, which integrates scRNA-seq data with reference datasets and genomic annotations, including gene locations and chromosomal coordinates. Gene expression profiles of epithelial-derived cells were compared with reference immune cells (T and B cells from VCP groups) to infer CNV patterns. The scCancer software was used to assess malignancy, evaluate cell-level CNV alterations, and assign malignancy scores based on the extent of gene copy-number deviations.

Integration of scRNA-seq data and Visium spatial transcriptomics data

The FFPE samples were submitted to Majorbio Bio-Pharm Technology Co., Ltd. for RNA extraction and quality assessment. Spatial transcriptomic profiling was conducted using the Visium platform, following the manufacturer’s protocol. FFPE sections were cut to a thickness of 5 μm, and all sequencing and data analyses were performed by Majorbio via the Majorbio Cloud analytics platform (www.cloud.majorbio.com). Combining scRNA-seq and Visium data improves spatial resolution. Cell types were deconvoluted using Robust Cell Type Decomposition (RCTD) (20), with scRNA-seq data as a reference. Gene expression at each spatial location was further examined using Loupe Browser 8 (https://www.10xgenomics.com/cn/support/software/loupe-browser/latest).

Data availability

The scRNA-seq data have been deposited into the National Center for Biotechnology Information Sequence Read Archive database (accession no. HRA009441; https://ngdc.cncb.ac.cn/gsa-human/browse/HRA009441) and the Gene Expression Omnibus database (accession no. GSE290927). All source data behind the graphs in the figures can be found in Supplementary Data. Any other data are available from the corresponding author upon reasonable request.

Statistics and reproducibility

Statistical analyses were performed using R (version 4.3.1) and RStudio (R Foundation for Statistical Computing; http://www.r-project.org). Analytic methods included Louvain clustering, SCENIC, CNV analysis, and CellChat. Statistical significance was evaluated using Z-scores and P values. A |Z-score| >2 was considered meaningful, with positive and negative values indicating activated and suppressed interactions, respectively. P values < 0.05 were considered statistically significant.

Results

Comprehensive characterization of scRNA-seq profiles in laryngeal lesions

Quality control of the scRNA-seq data included assessments of UMI counts, gene counts, and the proportion of mitochondrial gene expression per cell (Supplementary Fig. S1). In total, 318,907 single cells were captured and sequenced from 40 samples. Subsequent clustering analysis identified 27 distinct clusters, which were annotated into seven major cell types: epithelial cells, fibroblasts, T cells, B cells, NK cells, endothelial cells, and myeloid cells (Fig. 1B, Supplementary Fig. S2A and S2B). Cell identity was confirmed using t-SNE plots based on established marker genes (Supplementary Fig. S2C). LSCC samples exhibited a notably higher proportion of immune cells, suggesting an enriched tumor immune microenvironment. In the VCL group, a relatively higher proportion of fibroblasts was observed, along with an increased presence of endothelial cells (Fig. 1C). A heatmap of inter- and intragroup variability based on the top 2,000 HVGs is presented in Fig. 1D. This analysis revealed strong intragroup correlations, indicating consistent HVG expression patterns within each group. However, significant differences exist between groups, especially in LSCC, which demonstrated significant variability. Additionally, substantial intragroup heterogeneity was observed within the LSCC group, reflecting the intrinsic heterogeneity of the tumor. Expression patterns in the VCP and VCL groups were relatively similar, whereas those in the LSCC precursor (LSCCP) lesion group seemed intermediate between VCL/VCP and LSCC, indicating a possible transitional phenotype (Supplementary Fig. S2D). Intergroup variability was significantly greater than intragroup variability (P < 0.01; Supplementary Fig. S2E), reinforcing the robustness of the dataset for downstream bioinformatic analyses.

SCENIC analysis revealed distinct patterns of TF activity across the seven identified cell types (Fig. 1E). Among the immune cell populations—T cells, B cells, and myeloid cells—shared features were observed in TF expression profiles, indicating potential functional similarities. However, further studies are needed to elucidate the specific biological roles of individual subclusters within each cell type across the different laryngeal lesion types. In addition, gene and UMI count analyses indicated higher transcriptional activity in epithelial cells and fibroblasts, whereas B cells and NK cells exhibited lower activity levels (Supplementary Fig. S2F). Cell–cell communication analysis highlighted significant heterogeneity among laryngeal lesions. In particular, epithelial cells, T cells, and myeloid cells demonstrated marked differences in interaction strength, underscoring their central roles in modulating the TME dynamics (Supplementary Fig. S2G). All proportions of cell types and subclusters at the sample level were listed in Supplementary Fig. S2H and S2I. The distribution of subclusters and subtypes for each cell type in t-SNE plots was listed in Supplementary Fig. S3.

Heterogeneity and microenvironment dynamics of epithelial cells in different laryngeal lesions

Epithelial cells play a crucial role in laryngeal lesions. Without batch effect correction, all epithelial cells were reclustered into 35 subclusters. The subcluster distribution in t-SNE plots and epithelial cell proportions demonstrated heterogeneity across groups (Fig. 2A). Using B- and T-cell genomes from the VCP group as references, CNV analysis was performed on all epithelial subclusters in the VCL and LSCC groups. Elevated CNV levels were observed in VCL and LSCC, with higher levels in LSCC (Fig. 2B; Supplementary Fig. S4A). High-resolution CNV heatmaps for all groups are shown in Supplementary Fig. S4B–S4E. Based on CNV results, group comparisons, and prior studies, subclusters with CNV scores above 200—specifically subclusters 4, 10, 20, 21, 24, 27, 29, 31, 33, and 34—were preliminarily classified as malignant (21). KEGG analysis indicated that these malignant subclusters were primarily enriched in pathways related to cell damage and repair, tumor progression, and virus infection. Notably, virus- or bacteria-associated pathways, such as Epstein–Barr virus, human papillomavirus, and Helicobacter pylori, were identified (Fig. 2C). SCENIC analysis revealed heterogeneity in TF expression across epithelial subclusters (Supplementary Fig. S4F). Unlike malignant subclusters predominantly found in LSCC, subclusters SC24 and SC27 in VCL showed distinct TF enrichment, including SREBF2 and BACH1.

Figure 2.

Epithelial cell transcriptional heterogeneity of VCL and LSCC. A, Cellular composition of epithelial cell subclusters in the different types of samples examined. The sample-level proportions of cell types are shown in Supplementary Fig. S2H. B, Chromosomal landscape of inferred CNVs distinguishing epithelial cell subclusters from different types of samples based on scRNA-seq data. The reference is B and T cells in VCP. Chromosomal amplifications are shown in red, and deletions are shown in blue. C, KEGG results showing enrichment pathway activation differences among malignant-like epithelial cell subclusters. The y-axis represents KEGG pathway names, and the x-axis indicates the rich factor. Dot size reflects the number of genes associated with each pathway, whereas dot color represents the range of adjusted P values. D, GSVA results showing the enrichment levels of epithelial cell malignancy–associated gene sets in VCL and LSCC. Colors represent the normalized enrichment scores of each gene set across cell types, with red indicating higher enrichment scores and blue indicating lower values. E, Comparison of relative information flow between two groups. The x-axis represents the relative information flow (ranging from 0% to 100%), indicating the importance of each signaling pathway. The y-axis lists the corresponding signaling pathways. Left: Dark blue bars denote values in VCL, whereas orange bars represent VCP. Right: Dark blue bars denote values in LSCC, whereas orange bars represent VCL.

To investigate malignancy-associated transcriptional changes further, we performed GSVA using the upregulated genes in these malignant epithelial subclusters as a reference, comparing the VCL and LSCC groups (Fig. 2D). Gene sets used for GSVA were derived from the HALLMARK collection in the Molecular Signatures Database to ensure unbiased pathway selection. Our findings revealed that LSCC samples generally exhibited higher enrichment scores in malignancy-related pathways, whereas VCL samples displayed lower levels of enrichment. Notably, 43 gene sets exhibited statistically significant differences between LSCC and VCL groups, emphasizing a distinct divergence in their transcriptional landscapes (raw data are provided in Supplementary Data). GSVA revealed that certain VCL samples—such as VCL2, which presented histologic features of high-grade dysplasia—demonstrated pathway enrichment patterns closely resembling those of LSCC tumor tissues. These findings suggest that although the majority of VCL samples maintain transcriptional profiles distinct from LSCC, a subset may already exhibit molecular signatures indicative of elevated malignant potential. This observation is consistent with clinical evidence linking severe dysplasia to an increased risk of malignant transformation.

Analysis of cell–cell communication among epithelial cell subclusters revealed that signaling activity was more pronounced in the LSCC group, particularly within the malignant epithelial subclusters (Supplementary Fig. S4G). This suggests that these subclusters play a pivotal role in mediating intercellular communication within the LSCC TME. A comparative analysis of communication pathways between epithelial subclusters demonstrated that certain pathways, such as FN1, PERIOSTIN, DESMOSOME, and SLURP, exhibited higher inferred activity or expression in the VCL group compared with LSCC. Furthermore, the VCL group exhibited elevated expression of cell–cell communication pathways—including COMPLEMENT, SLURP, ITGB2, EGF, CD99, JAM, and APP—when compared with the VCP group (Fig. 2E).

T-cell subcluster reprogramming shapes the immunosuppressive landscape in VCL and LSCC

T cells play a pivotal role in antitumor immunity and serve as key modulators of the immune microenvironment. To elucidate the functional heterogeneity of T cells, we reclustered them into 15 subclusters and identified six distinct T-cell types: CD4⁺ naïve T cells, CD8⁺ effector T cells, proliferating T cells, Th17 cells, regulatory T cells (Treg), and other T-cell subsets (Fig. 3A). Comparative analysis revealed that intercellular communication patterns and intensities among these T-cell subtypes were largely conserved between VCL and LSCC. However, in VCL, signaling via the MHC-I, CLEC, and CD99 pathways was significantly more active. Notably, CD8⁺ effector T cells exhibited strong incoming signal intensities in VCL and LSCC, emphasizing their robust engagement in immune responses (Fig. 3B).

Figure 3.

T-cell heterogeneity in VCP, VCL, LSCC, and LSCCP. A, Cellular composition of T-cell types in the different types of samples examined. Marker genes for CD4⁺ T naïve cells: CD4, CCR7, LTB, SELL, CD27, IL7R, and LEF1; marker genes for CD8⁺ T effector cells: CD8, CXCR4, GZMA, NKG7, GNLY, GZMK, GZMB, and GZMH; marker genes for proliferating T cells: HMGN2, RRM2, TOP2A, and MKI67; marker genes for Th17 cells: RORA, IL6ST, and IL17RA; marker genes for Tregs: FOXP3, IKZF2, and IL2RA (Supplementary Table S5). T-cell subclusters lacking definitive marker genes for known T-cell subtypes were collectively classified as “other T cells.” The sample-level proportions of cell types are shown in Supplementary Fig. S2I. B, Heatmap of intercellular communication intensities between T-cell types in VCL and LSCC. The x-axis shows T-cell types, and the y-axis lists ligand genes. Heatmap colors indicate relative signaling intensity. The top and right bar plots represent total intensity for each cell type and signaling pathway. C, Differential gene expression of Tregs between the VCL and LSCC groups. VCL served as the control group. Red indicates genes upregulated in Tregs from the LSCC group, whereas blue indicates genes downregulated in LSCC (also upregulated in VCL). Significantly DEGs were defined by P_adj. < 0.05 and |log₂FC| ≥ 0.25, where FC indicates fold change. D, KEGG results showing enrichment pathway activation differences in upregulated genes of Tregs between VCL and LSCC. The y-axis represents KEGG pathway names, and the x-axis indicates the rich factor. Dot size reflects the number of genes associated with each pathway, whereas dot color represents the range of P_adj. AGE-RAGE, advanced glycation end-product–receptor for advanced glycation end-product; GVHD, graft-versus-host disease; MD, mitochondrial dysfunction; NOD, nucleotide-binding oligomerization domain; ROS, reactive oxygen species; TNF, tumor necrosis factor.

A differential gene expression analysis of Tregs revealed distinct immunoregulatory profiles between the two conditions. Tregs in LSCC demonstrated significantly elevated expression of STAT family members and key immunosuppressive genes, including IL-10, CD274, and TGFB1. Conversely, STAT1 expression was higher in VCL-derived Tregs (Fig. 3C). Furthermore, LSCC Tregs were more enriched in immunosuppressive signaling pathways, such as the TNF signaling cascade and oxidative phosphorylation, consistent with enhanced metabolic activity and immunosuppressive function. These findings suggest that Tregs in LSCC reside in a heightened immunosuppressive state. In contrast, Tregs from VCL were more transcriptionally associated with immune activation and antiviral responses, reflective of a precancerous immunologic milieu. Nonetheless, the presence of immunosuppression-associated genes in VCL-derived Tregs implies an early functional shift toward immunoregulatory activity (Fig. 3D).

Heterogeneity and immunosuppressive roles of Bregs in VCL and LSCC

As the sole immune cells capable of producing antibodies, B cells play an essential role in orchestrating adaptive immune responses. In our study, B cells were reclustered into 13 distinct subclusters, including memory B cells, naïve B cells, germinal center B cells, regulatory B cells (Breg), plasma cells, and other B-cell subsets. The distribution of these subclusters varied across the four tissue types, emphasizing the heterogeneity of B cells in benign and malignant laryngeal lesions (Fig. 4A). Beyond the conventional B-cell subsets, we identified Bregs within laryngeal lesions, which were present in all four tissue types but were more prevalent in VCL and LSCC. KEGG pathway analysis revealed significant enrichment of ribosomal pathways in subclusters SC0 and SC6, indicating heightened antibody secretion activity in these Bregs. In addition, SC8 Bregs were enriched in PD-L1 signaling pathways and were predominantly found in LSCC and VCL, suggesting a contribution to the immunosuppressive TME in these lesions (Fig. 4B).

Figure 4.

B-cell heterogeneity in VCP, VCL, LSCC, and LSCCP. A, Cellular composition of B-cell types in the different types of samples examined. Marker genes for memory B cells: MS4A1, CD27, AIM2, TNFRSF13B, CRIP2, and ITGB; marker genes for naïve B cells: MS4A1, IGHD, FCER2, TCL1A, and IL4R; marker genes for germinal center (GC) B cells: AICDA, RGS13, and GCSAM; marker genes for Bregs: IL10, CD1D, CD5, and TGFB1; marker genes for plasma cells: CD38, SDC1, and MZB1 (Supplementary Table S5). B-cell subclusters lacking definitive marker genes for known B-cell subtypes were collectively classified as “other B cells.” The sample-level proportions of cell types are shown in Supplementary Fig. S2I. B, KEGG results showing enrichment pathway activation differences in Bregs SC0, SC6, and SC8. The y-axis represents KEGG pathway names, and the x-axis indicates the rich factor. Dot size reflects the number of genes associated with each pathway, whereas dot color represents the range of P_adj. C, Differential gene expression of Bregs between the VCL and LSCC groups. VCL served as the control group. Red indicates genes upregulated in Bregs from the LSCC group, whereas blue indicates genes downregulated in LSCC (also upregulated in VCL). Significantly DEGs were defined by P_adj. < 0.05 and |log₂FC| ≥ 0.25. D, KEGG results showing enrichment pathway activation differences in Bregs between VCL and LSCC. The y-axis represents KEGG pathway names, and the x-axis indicates the rich factor. Dot size reflects the number of genes associated with each pathway, whereas dot color represents the range of P_adj. E, Heatmap of intercellular communication intensities between B cell types in VCL and LSCC. The x-axis shows B-cell types, and the y-axis lists ligand genes. Heatmap colors indicate relative signaling intensity. The top and right bar plots represent total intensity for each cell type and signaling pathway. HIV-I, HIV type I; HTLV-I, human T-cell lymphotrophic virus, type I. COVID-19, coronavirus disease 2019; GVHD, graft-versus-host disease; KSHV, Kaposi’s sarcoma–associated herpesvirus.

To investigate the functional differences in Bregs among laryngeal lesion types, we compared the gene expression profiles of Bregs between VCL and LSCC (Fig. 4C). In LSCC-derived Bregs, there was significant upregulation of immunosuppressive genes, including STAT3, STAT4, TGFB1, PRDM1, and IRF4, implying a more pronounced immunosuppressive phenotype and a potential role in TME remodeling. Furthermore, mitochondrial metabolism-related genes MT-CO1 and MT-CO2 were upregulated in LSCC Bregs. Conversely, Bregs from VCL showed elevated expression of genes associated with antigen presentation and immune activation, indicating a role in promoting immune responsiveness and regulatory balance. Moreover, LSCC Bregs exhibited enhanced ribosomal activity, whereas VCL Bregs were primarily involved in antiviral immune processes (Fig. 4D). Notably, LSCC-derived Bregs demonstrated increased incoming signaling activity. Overall, B cells in LSCC displayed a more dynamic and complex intercellular communication profile, involving, in particular, MIF, FN1, TENASCIN, and ANNEXIN signaling pathways (Fig. 4E).

Myeloid cell heterogeneity and signaling dynamics in VCL and LSCC

All myeloid cells were clustered into 22 subclusters and annotated based on signature gene expression profiles as neutrophils, dendritic cells, basophils, macrophages, and monocytes (Fig. 5A). An increased abundance of neutrophils was observed in LSCC tissues, suggesting a heightened inflammatory TME and implicating their involvement in tumor infiltration. In contrast, monocytes and macrophages were the predominant myeloid populations in VCP and VCL, indicating a different immune landscape in benign and precancerous lesions. SCENIC analysis revealed robust immune activity across all myeloid cell types in both benign and malignant lesions, as illustrated by the heatmap (Fig. 5B). Neutrophils exhibited elevated expression of STAT family members, nuclear factor NF-κB1, and RELB, emphasizing their proinflammatory potential mediated via the NF-κB and STAT signaling pathways. Dendritic cells were enriched for antiviral TFs, such as IRF7, IRF8, and REL, suggesting that virus-induced immune modulation may contribute to disease progression. Macrophages demonstrated increased expression of CEBPD and PPARG, indicative of an anti-inflammatory M2 phenotype associated with tissue repair and microenvironmental regulation.

Figure 5.

Myeloid cell heterogeneity in VCP, VCL, LSCC, and LSCCP. A, Cellular composition of myeloid cell types in the different types of samples examined. Marker genes for neutrophils: CEACAM1, and CXCR2; marker genes for dendritic cells: CCR7, FSCN1, and LAMP3; marker genes for basophils: CD63, ENPP3, and IL3RA; marker genes for macrophages: CD14,CD163, and APOE; marker genes for monocytes: VCAN, FCN1, and S100A12 (Supplementary Table S5). The sample-level proportions of cell types are shown in Supplementary Fig. S2I. B, SCENIC analysis showing TFs in different myeloid cell types. Red indicates that the TF is upregulated in the corresponding cell, and blue indicates downregulation. C, Heatmap of intercellular communication intensities between myeloid cell types in VCL and LSCC. The x-axis shows myeloid cell types, and the y-axis lists ligand genes. Heatmap colors indicate relative signaling intensity. The top and right bar plots represent total intensity for each cell type and signaling pathway.

Marked differences in myeloid cell signaling dynamics were observed between VCL and LSCC (Fig. 5C). In LSCC, both incoming and outgoing signaling activities were more intense and intricate, particularly within neutrophils and macrophages. Key inflammatory and chemotactic signaling pathways—including CXCL, CCL, TNF, and IL-6—were significantly enriched in LSCC, reflecting a proinflammatory TME with enhanced intercellular communication. Conversely, VCL exhibited relatively lower signaling activity, with greater enrichment of pathways such as FN1 and ITGB2, pointing to a microenvironment more focused on extracellular matrix remodeling and tissue repair.

Integration of scRNA-seq and spatial transcriptomics reveals cellular heterogeneity and an immunosuppressive microenvironment in VCL and LSCC

The scRNA-seq results provide single-cell resolution transcriptional analysis but lack spatial context. To overcome this limitation, we integrated scRNA-seq data with spatial transcriptomics to reconstruct the spatial architecture of tissues and validate cell–cell communication and Treg distribution. Using previously generated scRNA-seq datasets, we inferred the cellular composition of each spatial transcriptomics spot. Hematoxylin and eosin staining for all samples is presented in Supplementary Fig. S5. By mapping the expression profiles of cell type–specific marker genes derived from identified subclusters, we annotated the dominant cell types within each spot’s microenvironment (Supplementary Fig. S6), determining the predominant cellular identity per spot (Fig. 6A; Supplementary Fig. S7). Tumor tissues exhibited greater complexity and cellular heterogeneity, indicative of an active TME. Notably, LSCC contained a significantly higher proportion of plasma cell-enriched spots compared with other tissues, with plasma cells accounting for approximately 10% of all spots, whereas their representation in other tissue types was generally below 1%. This observation corroborates our previous scRNA-seq findings. As central effectors of humoral immunity, plasma cells may contribute substantially to immune microenvironment remodeling in LSCC. Furthermore, in regions of immune cell aggregation within LSCC, we observed spots dominated by Bregs and Tregs, reinforcing the conclusion that the LSCC immune microenvironment is predominantly immunosuppressive. Regulatory immune cells were also present in other tissue types, albeit at varying proportions. However, due to limitations in spatial resolution, these cells were not distinctly visualized in Fig. 6A.

Figure 6.

Integration of scRNA-seq results with spatial transcriptomics to assign spatial characteristics. A, The spatial distribution of epithelial cells, immune cells, and stromal cells was analyzed within the samples for spatial transcriptomics. scRNA-seq results from this study were used as a reference. The proportion of specific cells within each tissue is shown in the Supplementary Data. B, Cell–cell communication in VCL and LSCC samples for spatial transcriptomics. Higher-resolution images are presented in Supplementary Fig. S8A. C, Co-expression results of marker genes in Tregs and immunosuppression. The color of each spot represents the gene co-expression pattern within that spot. Blue indicates the expression of immunosuppression-related genes, yellow represents the expression of Treg-specific marker genes, and green denotes co-expression of both gene sets at the same spatial location. The results were calculated using Loupe Browser 8.0. Higher-resolution images are presented in Supplementary Fig. S8B.

In contrast, spatial mapping of cell–cell interactions in VCL was less evident, potentially due to limitations associated with the size of FFPE samples or sectioning artifacts. In VCL, interactions were primarily confined to epithelial cells and fibroblasts. LSCC demonstrated a significantly higher degree of intercellular communication, particularly between epithelial cells, plasma cells, Tregs, and Bregs (Fig. 6B). Higher-resolution images were presented in Supplementary Fig. S8A. Therefore, we analyzed the spatial co-expression of canonical Treg markers (FOXP3, IL2RA, and CTLA4) alongside immunosuppression-associated genes (CD274, IL10, TGFB1, STAT2, and STAT4; Fig. 6C). Higher-resolution images were presented in Supplementary Fig. S8B. The observation suggested that although Tregs contribute to immune regulation in both VCL and LSCC, those in LSCC tend to exhibit higher expression of immunosuppressive genes. The presence of immunosuppressive Tregs in VCL, although at lower abundance and with weaker spatial signals, may reflect the early establishment of immune suppression during lesion progression.

Discussion

With advances in scRNA-seq technologies, numerous human atlases—especially for tumors—have been developed to elucidate lesion development. The transition from precancerous to malignant states remains a key focus (22, 23). scRNA-seq has proven instrumental in characterizing tissue heterogeneity, immune landscapes, radiation sensitivity, and therapeutic resistance in various cancers, including breast and kidney malignancies (24, 25). In LSCC, prior studies have explored in situ tumors and lymphatic metastases at single-cell resolution (26, 27). However, a comprehensive single-cell atlas encompassing benign, precancerous, and malignant laryngeal lesions has yet to be established. Chronic exposure to low-grade inflammatory stimuli—mediated by TFs and proinflammatory cytokines—plays a pivotal role in tumor initiation and progression (28). In our research, although these lesions share core cell types, their TMEs show substantial heterogeneity. VCL and LSCC notably exhibit immune cell enrichment and hallmarks of an immunosuppressive microenvironment.

Malignant epithelial cells are detectable in VCL, exhibiting CNV levels that are intermediate between those observed in LSCC and VCP, supporting the classification of VCL as a transitional stage in laryngeal tumorigenesis. Human papillomavirus and Epstein–Barr virus have been implicated in LSCC pathogenesis, whereas laryngopharyngeal reflux may promote VCL and LSCC by damaging the laryngeal mucosa (29–31). In epithelial subclusters associated with malignancy, enrichment analysis identified the activation of pathways related to viral and bacterial infections, consistent with clinical findings and emphasizing the potential contribution of infectious agents to VCL progression.

Additionally, SCENIC analysis revealed distinct TF expression profiles in malignant epithelial subclusters. SREBF2, a key regulator of lipid and cholesterol metabolism, is overexpressed in esophageal squamous cell carcinoma and has recognized oncogenic potential (32). BACH1, implicated in breast and liver cancers, plays a critical role in oxidative stress response and energy metabolism (33, 34). As the principal cell type defining the malignant potential of laryngeal lesions, epithelial cells in VCL may be modulated by infection- or inflammation-related stimuli within the local microenvironment. Progressive CNV accumulation and altered TF activity contribute to stepwise malignant transformation. Moreover, differences in TF expression and intercellular communication between LSCC and VCL may provide prognostic insights and aid in the identification of novel therapeutic targets for early intervention.

Our scRNA-seq analysis reveals that LSCC-associated immune subclusters display gene expression profiles indicative of immune chemotaxis and epithelial-to-mesenchymal transition, including activation of the STAT family. STAT proteins regulate key genes, such as nitric oxide synthase, BCL-2 and p21, which are involved in apoptosis and cell-cycle regulation and function as tumor suppressors in various malignancies (35). In breast cancer, co-expression of PD-L1 and phosphorylated STAT1 has been correlated with poor prognosis and advanced disease stages, indicating a role for phosphorylated STAT1 in immune evasion (36). In addition, STAT2 upregulates SLC27A3, a gene involved in lipid metabolism and mitophagy, contributing to drug resistance in clear-cell renal cell carcinoma (37). In our study, elevated STAT expression and activity in LSCC-associated immune cells support their protumorigenic role and involvement in TME remodeling.

Tumorigenesis is a multifaceted process involving the malignant transformation of tumor cells and dynamic alterations within the surrounding microenvironment. The interplay between immunostimulatory and immunosuppressive signals largely determines tumor development or elimination. In laryngeal lesions, the microenvironment is primarily composed of immune and stromal cell populations. An immune-activating milieu enables effective recognition and clearance of tumor cells, whereas an immunosuppressive environment promotes immune evasion and tumor persistence (38, 39). Our study identified regulatory immune cell populations, such as Tregs and Bregs, in VCL and LSCC, with a higher degree of enrichment observed in LSCC. These findings suggest that immunosuppressive mechanisms may play a pivotal role in the progression from VCL to LSCC, emphasizing the potential of immunotherapeutic interventions to prevent malignant transformation.

Tregs and Bregs act as immunosuppressive modulators within the TME. Although essential for maintaining immune homeostasis, their suppressive activity can facilitate immune evasion in malignancies (40–42). Treg-mediated tolerance is implicated in metastatic progression and is recognized as a promising target for immunotherapy (43). We observed functional remodeling of regulatory immune cells during lesion progression. In LSCC, Tregs and Bregs displayed features of enhanced immunosuppression, whereas their counterparts in VCL were more associated with antiviral responses. These findings suggest that Tregs and Bregs undergo functional remodeling during the transition from precancerous to malignant states, contributing to the establishment of an immunosuppressive TME. Spatial transcriptomic analysis confirmed increased infiltration of immunosuppressive Tregs in LSCC. Notably, this functional shift may play a critical role in tumor progression and presents a potential avenue for early diagnosis and therapeutic intervention.

Myeloid cells, originating from common myeloid progenitors, are critical components of the innate immune system and serve as the first line of defense against pathogens (44). In LSCC, we observed a significant increase in neutrophil populations, characterized by active intercellular communication and transcriptional profiles enriched for STAT family TFs. Myeloid cells in LSCC showed increased abundance and more complex communication networks, with enrichment of inflammation-related signaling pathways. Activation of CXCL and CCL pathways suggests enhanced myeloid recruitment to tumor sites, promoting tumor growth and metastasis (45, 46). M2-like macrophages in LSCC secrete IL-6 and TNF, activating inflammatory cascades and reinforcing immunosuppression by supporting regulatory immune cells and facilitating immune evasion (47). Elevated expression of CXCL1 and CXCL2 has been associated with glioblastoma invasion and poor prognosis, whereas CCL2-driven macrophage recruitment promotes fibrosis and therapy resistance in breast cancer (46, 48). In our study, macrophages displayed an M2-like phenotype, suggesting a potential role in VCL-to-LSCC progression. These findings emphasize the need for further investigation into the contributions of myeloid cells to laryngeal tumorigenesis.

However, several limitations exist in our study. Although each group included 10 samples, tumor heterogeneity emphasizes the need for larger cohorts to improve the generalizability and robustness of the findings. Future studies should incorporate larger sample sizes, multiomics approaches, and include in vivo and in vitro functional validation to elucidate the molecular mechanisms underlying VCL progression and identify potential biomarkers for early diagnosis and treatment.

In conclusion, we used scRNA-seq and spatial transcriptomics to profile benign and precancerous laryngeal lesions, expanding the LSCC single-cell atlas. At single-cell resolution, we delineated epithelial subclusters exhibiting malignant potential and associated signaling pathways. Comparative analysis of the immune microenvironments in VCL and LSCC revealed notable similarities, implicating immunosuppression as a critical driver of malignant progression. These findings advance our understanding of the molecular transition from VCL to LSCC and provide a valuable single-cell resource for investigating the early stages of laryngeal tumorigenesis.

Supplementary Material

Supplementary Table S1

Patients demographics of the 40 samples included in scRNA-seq

Supplementary Table S2

The inclusion and exclusion criteria of tissue samples in the study

Supplementary Table S3

The reagents, catalog number and their manufacturers in this study

Supplementary Table S4

The databases and software referenced and utilized in this study

Supplementary Table S5

Marker genes for cell type classification in this study

Supplementary Data S1

All raw data for the analyses and visualizations in this study.

Supplementary Figure S1

Supplementary Figure S1. (A) Quality control and cell filtering of sc-RNA seq data.

Supplementary Figure S2

Supplementary Figure S2. (A) The tSNE plots results of individual samples within the VCP and VCL groups. (B) The tSNE plots results of individual samples within the LSCC and LSCCP groups. (C) t-SNE plots showing the main distribution in each of the seven cell types identified in the four types of tissue samples analyzed in this study. (D) Principal Component Analysis (PCA) of Sample Groups. Illustrates the clustering of LSCC, LSCCP, VCL, and VCP samples based on principal components. (E) Comparison of Euclidean Distances Within and Between Groups. Violin plot showing the Euclidean distance distribution between-group and within-group, with statistical significance (Wilcoxon test, p = 7.8e-07). (F) Gene statistics and unique molecular identifiers (UMIs) of all cell types in different types of samples in this study. (G) Incoming and outgoing cell-cell communication intensities among the seven major cell types across the four tissue groups. (H) The stacked bar plots of cell type and subcluster proportions in each sample. a. All cell type; b. epithelial cell subclusters; c. fibroblast subclusters; d. endothelial cell subclusters. (I) The stacked bar plots of cell type and subcluster proportions in each sample. a. T cell subclusters; b. T cell type; c. B cell subclusters; d. B cell type; e. myeloid cell subclusters; f. myeloid cell type.

Supplementary Figure S3

Supplementary Figure S3. t-SNE plots showing subclusters and celltypes identified in six main celltypes of the four different tissue samples. (A) Epithelial cells; (B) T cells; (C) B cells; (D) Myeloid cells; (E) Endothelial cells; (F) Fibroblasts.

Supplementary Figure S4

Supplementary Figure S4. (A) CNV scores of 21 epithelial cell subclusters in four different types of tissue samples. B cells and T cells from the VCP group were used as the reference. (B–E) Inferred chromosomal CNV landscapes distinguishing epithelial cell subclusters from the VCP (B), VCL (C), LSCC (D), and LSCCP (E) groups based on scRNA-seq data. B cells and T cells from the VCP group served as the reference. Chromosomal amplifications are shown in red and deletions in blue. (F) SCENIC analysis identifying transcription factors in epithelial cell subclusters; malignant subclusters are annotated in yellow. (G) Comparison plot result of epithelial cell subclusters outgoing and incoming signal strength among four different tissues.

Supplementary Figure S5

Supplementary Figure S5. HE staining images of all samples for spatial transcriptomics. (A) ST-VCP1; (B) ST-VCP2; (C) ST-VCL1; (D) ST-VCL2; (E) ST-LSCC1; (F) ST-LSCC2; (G) ST-LSCCP1; (H) ST-LSCCP2.

Supplementary Figure S6

Supplementary Figure S6. (A) Based on scRNA-seq data, all spots were annotated into seven major cell types.

Supplementary Figure S7

Supplementary Figure S7 All celltypes spatialplot in different samples. (A) ST-VCP1; (B) ST-VCP2; (C) ST-VCL1; (D) ST-VCL2; (E) ST-LSCC1; (F) ST-LSCC2; (G) ST-LSCCP1; (H) ST-LSCCP2.

Supplementary Figure S8

Supplementary Figure S8 (A) Higher-resolution image of Fig. 6B. Cell-cell communication in VCL and LSCC sample for spatial transcriptomics. (B) Higher-resolution image of Fig. 6C. Co-expression results of marker genes in Tregs and immunosuppression. The color of each spot represents the gene co-expression pattern within that spot. Green means genes of Tregs and immunosuppression co-expressed in the same spot.

Authors’ Disclosures

No disclosures were reported.

Authors’ Contributions

Z.-M. Fu: Data curation, formal analysis, supervision, visualization, methodology, writing–original draft, writing–review and editing. Y.-Y. Bao: Conceptualization, resources, supervision. L.-B. Dai: Data curation, supervision, investigation. J.-T. Zhong: Data curation, software, validation. H.-C. Chen: Data curation, formal analysis, methodology. Z. Chen: Data curation, software, investigation. S.-H. Zhou: Conceptualization, resources, funding acquisition, project administration, writing–review and editing.