Multimodal profiling of oral squamous cell carcinoma identifies genomic alterations and expression programs associated with betel quid chewing

Highlights

• •Betel quid (BQ) chewing preferentially disrupts p53 and Hippo pathway in oral cancer.
• •BQ use modulates gene expression to promote hypoxia in oral cancer.
• •BQ chewing alters gene expression to promote cell-cell communication in oral cancer.
• •Oral cancer caused by BQ chewing represents a different molecular entity.

Abstract

Betel quid (BQ) chewing is a profound risk for oral squamous cell carcinoma (OSCC) in Southeast Asia. Yet, the detailed mechanisms by which BQ chewing damages the genome and creates a unique tumor niche that ultimately cause OSCC are still not fully understood. To address this, we conducted a multi-omics survey, including exome sequencing of tumor-normal pairs from 261 male patients with OSCC (129 habitual BQ chewers and 132 non-BQ users), alone with integrated single-cell and spatial transcriptomics of a set of tumors. Comparative analyses of the mutational catalog identified enrichment of significantly altered genes (e.g. mutations of TP53 and CHUK, copy gains of MAP3K13 and FADD, copy losses of CDKN2A) associated with BQ chewing. Assessment of oncogenic and co-occurring actionable alterations demonstrated frequently altered oncogenic pathways (Hippo and p53 signaling) and potential combination therapy opportunities linked to BQ use. In addition, evaluation of epithelial, immune, stromal expression programs in the corresponding tissue compartments revealed a shift of tumor microenvironment in BQ-related OSCC, characterized by induced hypoxia of tumor epithelium, altered immunosuppression of dendritic cells, and raised sprouting angiogenesis of tumor endothelium. Quantitative predictions of intercellular communications inferred a more heterogeneous cell-cell crosstalk among BQ-related OSCC, highlighted by extensive interactions of fibroblasts and dendritic cells with other non-epithelial cell types via mostly extracellular matrix-receptor signaling pathways. Collectively, these differences in genomic landscape and tumor niche suggest that OSCC caused by BQ chewing could be an etiological subtype different from their BQ-negative counterparts.

Introduction

Oral squamous cell carcinoma (OSCC) represents the major form of oral cancer and develops commonly among distinct ethnic groups, especially for those in Southeast Asia. The established causes of OSCC, involving human papillomavirus (HPV) infection and prolonged exposure to carcinogens [1], are mostly associated with unique lifestyles and social practices. In addition to tobacco use and alcohol consumption, betel quid (BQ) mastication remains a crucial etiological factor for oral tumorigenesis in Asian areas with a high incidence of OSCC, including Taiwan in which over 80 % of OSCC cases were long-term users of BQ [2].

Repeated exposure of carcinogenic substances is considered as a major contributor to OSCC related to chewing of areca nut or BQ products [3]. Among these carcinogens, arecoline, one predominant alkaloid in BQ, can undergo nitrosation to generate a myriad of nitrosamines, which elicit oncogenic effects on the oral cavity via induction of DNA damage and production of excessive reactive oxygen species (ROS) and pro-inflammatory cytokines [4]. Moreover, cumulative experimental evidence has indicated that BQ may orchestrate many molecular and cellular mechanisms, such as tissue hypoxia and epithelial-mesenchymal transition (EMT), to promote malignant transformation [4]. As a connection of BQ chewing with the pathogenesis of oral submucosal fibrosis, a precancerous condition, was demonstrated [5], chronic exposure to BQ constituents also facilitated a more aggressive phenotype in OSCC patients as evidenced by poor survival and increased recurrence [6].

Current genomic investigations have thoroughly profiled HPV-positive and HPV-negative head and neck cancers in the Western countries, forming the basis for improved molecular diagnostics and translational insights [7,8]. Despite a certain level of similarities in the mutational landscape, such as recurrent mutations in NOTCH1, CDKN2A, HRAS, and PIK3CA, HPV-positive OSCC tumors have been proposed as a distinct molecular entity as compared to their HPV-negative counterparts [8]. As the etiology and genomic aberrations of OSCC were not uniform across geographic areas, it is plausible that the outcome of mutagenic processes and expression programs operative in OSCC related to BQ chewing may be unique. Attempts to identify these mutational patterns associated with BQ chewing in oral oncogenesis have been made with a limited sample size or a dataset from different genomic foundations for comparisons [9,10]. However, the comprehensive genomic footprints and microenvironmental dynamics of BQ chewing-related OSCC are not fully understood. In this study, we examined 261 OSCC genomes (129 BQ users and 131 non-BQ users) through exome sequencing to dissect BQ-associated genomic abnormalities, in particular mutational signatures and oncogenic/targetable pathways. In addition, we combined single-cell and spatial transcriptomics (ST) upon a series of BQ-related and -unrelated OSCC samples, with the aims of characterizing the tumor microenvironment and exploring the expression programs and relationship among major cell types within.

Material and methods

Subject enrollment and sample collection

The cohort of this study comprising 261 male OSCC patients who had been neither previously treated nor proven metastatic disease at the time of diagnosis was accrued from 2008 to 2023, with the approval by the institutional review board of Chung Shan Medical University Hospital, Taichung, Taiwan. All participants provided informed written consent at enrollment. Specimen collection, tumor staging, and definition of habitual risks can be found in supplemental methods.

Exome sequencing

Genomic DNA was isolated by using the DNeasy Blood & Tissue Kit (Qiagen) according to the manufacturer’s instructions. In-solution enrichment of coding exons and flanking intronic sequences for 261 pairs of qualified DNA samples was conducted by the SureSelect Human All Exon V5 or V6 kit (Agilent Technologies). DNA library was prepared and sequenced on an Illumina HiSeq instrument.

Detection of significantly altered somatic variants

Calling of somatic single-nucleotide variations (SNVs) and small insertions/deletions (INDELs) was performed by MuTect2 through the Genome Analysis Toolkit (GATK) 4.0, and significantly mutated genes were detected by MutSigCV. Identification of somatic copy-number variants (CNVs) was conducted by Control-FREEC, and significantly recurrent CNVs were inferred with GISTIC2.0. For calling somatic structural variations (SVs), BreakDancer was used, and variant calling in details can be found in supplemental methods.

Annotation of oncogenic and clinically actionable variants

Annotation of functionally or clinically relevant variants was performed by using OncoKB. In addition, putatively targetable variants were also identified with PHIAL based on the database of tumor alterations relevant to genomics-driven therapy (TARGET). Detailed methods can be found in supplemental materials.

Single-cell RNA-sequencing (scRNA-seq) and data processing

scRNA-seq libraries were prepared using Chromium Single cell 3′ GEM, Library, & Gel beads kit (10X Genomics) and sequenced on an Illumina NovaSeq 6000 instrument. Gene-cell matrices were generated by using Cell Ranger Pipeline (10X Genomics) and analyzed with the Seurat R package. Tissue processing, preparation of cell suspension, library construction and sequencing, data quality control and normalization, dimension reduction, visualization, and scoring of expression programs in detail are available in supplemental methods.

Cell type annotation

For determining major cell types and subtypes, we employed unsupervised clustering and differential expression to extract top differentially expressed genes (DEGs) with cell type-specific expression known from literature. Cellular subtypes were further delineated by re-clustering subsets of broadly defined cell types and re-analyzing with the same approach. Cell type annotation in detail can be found in supplemental methods.

Spatial transcriptomics (ST) and data analysis

ST experiment was performed by using Visium Spatial for FFPE Gene Expression Kit, Human Transcriptome, 6.5 mm (10x Genomics). Sequencing reads were analyzed with Space Ranger, and the filtered expression matrix was imported into Seurat package for downstream analyses. Normalization across spots was performed with the SCTransform function of Seurat, and cell-type deconvolution was conducted by SPOTlight with our scRNA-seq dataset (10 major cell types) being used as reference. Processing, staining, and imaging of tissues, library construction and sequencing, data processing, normalization, and deconvolution in detail can be found in supplemental methods.

Analyzing the status of expression programs

To explore the program expression status for each cell type of individual ST tissues, we applied a uniform strategy by which spots enriched with a given cell type from each sample were selected for assessing the expression score of a given program. The expression intensity of a set of cell-type specific DEGs (n = 30) derived from scRNA-seq was used to select spots enriched with a given cell types from each sample. According to the distribution of maker expression among all spots across different samples, the cut-off for epithelial cells, fibroblasts, and monocytes/macrophages was set at top 50 % of spots, as only the upper quartile was selected for T, NK, dendritic, and endothelial cells. The program expression of each selected spot was then scored and normalized to its score of cell type-specific markers (n = 30) for inter-sample comparisons.

Cell–Cell Communication Analysis

The ST gene expression data were imputed by MIST, and normalized expression matrix was imported with information regarding spot-spot distance calculated by converting the spatial coordinates in pixels to micrometers. Spatially-proximal intercellular communication was inferred by an updated version (v2.1.0) of CellChat R package. Detailed methods are available in supplemental materials.

Statistical analysis

Demographic and clinical data were compared by Mann-Whitney U test. Difference in mutational signature exposures between groups was tested by Student’s t-test. Comparisons of oncogenic pathway and actionable alterations were performed using the χ² test, unless otherwise indicated. Enrichment of expression programs and cell types from deconvolution of ST data was determined by Wilcoxon rank sum test. All reported p values were two-tailed, and a p value of <0.05 was considered significant.

Results

Subject characteristics, study design, and overview of multi-omics assays

To assemble a comprehensive map for the genomic alterations and expression programs of OSCC associated with BQ chewing, we conducted multi-omics measurements from a cross-sectional cohort of 261 male subjects (Figure 1A), comprising 129 habitual BQ chewers and 132 non-BQ users (Table S1). In addition to dissecting BQ-associated genomic abnormalities via exome sequencing, we combined single-cell and spatial transcriptomics (ST) to characterize the expression programs and cell-cell interactions within the tumor microenvironment. All patients enrolled were male, and several anatomical sites were represented, including buccal mucosa (42.1 %), tongue (28.3 %), gingiva (10.3 %), lip (8.4 %), and others (10.7 %). Of the patients recruited, 74.3 % and 36.8 % reported a history of tobacco and alcohol use, respectively. Histological examination confirmed advanced stage III/IV in 51.7 % of patients. Lymph node metastasis occurred in 34.5 % of cases.

Fig 1.

Overview of study design and significantly altered genes in BQ-related and -unrelated OSCC. (A) Schematic diagram illustrating a single-source experimental design that comprises whole-exome sequencing (WES), single-cell RNA sequencing (scRNA-seq), and spatial transcriptomics (ST) in order to systematically address genomic alterations and expression programs associated with betel quid (BQ) chewing. A catalog of somatic alterations was obtained through WES of tumor-normal pairs from 261 male patients with OSCC to characterize cancer diver genes, mutational signatures, and oncogenic/actionable pathways between BQ-related (BQ+) and -unrelated OSCC (BQ-). In addition, scRNA-seq of 14,590 individual cells (after QC) was integrated into ST profiling of 24,742 spots from 10 tumors of the buccal mucosa, to dissect BQ-associated gene expression programs and cellular interactions within the tumor microenvironment. (B) Mutation rate, demographic and clinical features, habitual exposure to behavioral risks, and landscape of genomic alterations of 247 OSCC patients. Patients exposed to habitual risks are indicated as a filled square. *, subjected to single-cell RNA sequencing; #, subjected to spatial transcriptomics. (C) Frequency of recurrent CNVs (gains in red; losses in blue) along the genome for BQ-related (BQ+, n = 124) and -unrelated OSCC (BQ-, n = 123). The vertical axis indicates the frequency of CNVs in the cohort determined by summing up WES samples with ploidy N > 2.5 (gain, red, positive values) or N < 1.5 (loss, blue, negative values), divided by the total number. x-axis, chromosomal coordinates Chr 1–22. Dark segments denote genomic segments that are significantly enriched in BQ+ or in BQ- samples, based on an adjusted p value of < 0.05 using the proportion test to compare each segment in two cohorts. Selected genes in recurrent copy gains and losses are labeled, and the total number of genes within these dark peaks is given in brackets. Genes listed in the cancer hallmark pages of the COSMIC cancer gene census [50] are marked in bold.

Mutational landscape

For deciphering BQ-related genomic aberrations in oral cancer, we initially assessed somatic variants and mutational signatures through exome sequencing of tumor-normal pairs from 261 cases. A mutational catalog, comprising 49,492 single nucleotide variants (SNVs) and 3,970 insertion or deletion (INDEL) variants, was profiled, through which 13 significantly mutated genes were detected (Figure 1B and Table S2), highlighting a certain degree of consistency with previous findings [7,8,11], such as recurrent mutations in HRAS and PIK3CA and inactivating mutations in NOTCH1. Among these potential OSCC driver genes, TP53 and CHUK were mutated more frequently in BQ-related OSCC than in BQ-unrelated samples (Figure S1A). Further evaluation for regions of recurrent copy-number gains or losses demonstrated that the frequency (Figure 1C) and significance (Figure S1B) of genomic segments with copy-number changes were distinct between BQ-related and -unrelated tumors, although patterns were similar (e.g. significant and frequent gains in Chr3q, 8q, 11q, and 19p; losses in Chr3p, 8p, and 9p). In general, BQ chewing appeared favorable to the increased frequencies of recurrent copy gains but less so to that of copy losses (Figure 1C). Specifically, amplification of several significantly gained regions (e.g. Chr3q29, containing MAP3K13; 11q13, containing FADD and VEGFB; 19p13, containing JUNB) occurred more frequently in BQ-related OSCC than in BQ-unrelated OSCC. Of note, focal deletion of another significantly depleted region in Chr9p21 comprising CDKN2A and CDKN2B (with a q value of 10⁻²¹ and 10⁻⁷ for BQ-related and -unrelated samples, respectively) was enriched among BQ-related OSCC as compared to those in BQ-unrelated tumors. These differences in genomic landscape reflect a long-standing conjecture that OSCC caused by BQ chewing could be an etiological subtype different from those unexposed to this environmental risk.

Oncogenic signaling pathways

Genomic variants in signaling pathways orchestrate common hallmarks of cancer, but the extent of genomic alterations in these pathways between BQ-related and -unrelated OSCC remains unexplored. To address this, we assessed 10 canonical signaling pathways frequently altered in cancer genomes, including cell cycle, Hippo signaling, Myc signaling, Notch signaling, oxidative stress response/Nrf2, phosphatidylinositol 3-kinase (PI3K) signaling, receptor-tyrosine kinase (RTK)-Ras signaling, transforming growth factor β (TGFβ) signaling, p53 and Wnt signaling [12]. Through filtering the putative passenger variants and calculating the fraction of samples with at least one oncogenic variant in the gene members of these pathways, we found that the RTK-Ras and TGFβ signaling are the one with the highest and lowest frequency of oncogenic alterations, respectively, in our cohort (Table S3 and Figure 2 and S2). The ranking of frequently altered pathways differed between BQ-related and -unrelated OSCC. The pathway with the highest frequency of oncogenic alterations in BQ-unrelated OSCC is the RTK-Ras signaling (57 % of BQ-unrelated OSCC), followed by cell cycle (50 %), PI3K (48 %), Hippo (45 %), and p53 signaling (44 %). However, in BQ-related OSCC, top altered pathways are the Hippo (63 % of BQ-related OSCC), RTK-Ras (62 %), p53 (61 %), PI3K (60 %), and cell cycle pathway (47 %). Of these pathways, a significant increase in the frequency of oncogenic alterations related to the use of BQ was observed in the Hippo and p53 signaling (Figure 2), where FAT1 and TP53 are the most frequently altered gene, respectively. In addition, the mutagenic impact of BQ on shaping the cell cycle signaling of OSCC was anatomical site-specific, as 59 % and 35 % of samples were altered in BQ users with the cancers of the tongue and bacco mucosa, respectively (Table S3 and Figure S2). By assessing the proportion of oncogenic alteration types, we found that the use of BQ was associated with a significant increase in the frequency of oncogenic gene amplifications in the PI3K pathways across distinct anatomical sites of oral cavity (p = 0.0015, 0.0097, and 1.87E-6, for tongue carcinoma, cancers of the buccal mucosa, and overall OSCC, respectively) (Figure S2). Nevertheless, such BQ-associated gene amplifications were observed in the Wnt signaling pathway of tongue cancers (p = 0.0023) but not in that of buccal mucosa cancers. These results reveal an impact of habitual BQ consumption on carving the detailed landscape of oncogenic pathway alterations in oral cancer.

Fig 2.

Hippo, p53, and RTK-Ras pathway alterations across BQ-related and -unrelated tumors from distinct anatomic sites of the oral cavity. Alteration frequencies in the members of the Hippo, p53, and RTK-Ras pathway within cancers of the tongue (Tongue, n = 32 and 38 for BQ+ and BQ- tumors, respectively), buccal mucosa (BM, n = 48 and 56 for BQ+ and BQ- tumors, respectively), and overall oral cavity (OSCC, n = 124 and 123 for BQ+ and BQ- tumors, respectively). Only known or likely oncogenic variants in each gene are considered. Color side bars denote the fraction of samples affected by each type of somatic alterations or a combination of them for each pathway gene. Grey side bars show the proportion of samples bearing oncogenic alterations for the given pathway. SNV, single-nucleotide variation; INDEL, small insertion/deletion; AMP, amplification; DEL, deletion; SV, structural variation; *adjusted p < 0.05, by the chi-square test.

Putative actionability

To explore the clinical utility of our exome sequencing results, we used two databases to evaluate the druggable alterations in our samples: OncoKB and PHIAL (see Supplemental Results). Assessment by the OncoKB revealed that a slightly higher proportion of BQ-related OSCC tumors carried at least one (80.65 %) or multiple (41.12 %) putatively targetable alterations with different levels of evidence, in comparison with BQ-unrelated OSCC (72.36 % with at least one, and 36.59 % with two or more) (Figure 3A-B). Overall, the most common putatively druggable gene bearing alterations with actionability for standard care (Levels 1 or 2) and investigational therapies (Levels 3 and 4) in our dataset was PIK3CA and CDKN2A, respectively (Figure 3C). Of note, nearly all CDK4 and FGFR1 alterations that are potential biomarkers for investigational therapies (Levels 3 and 4) were seen exclusively in BQ-related samples across different anatomical sites of oral cavity (Figure 3C). Furthermore, based on the occurrence of targetable alterations, we sought for candidate drug combinations that could be effective for BQ-related or -unrelated OSCC. Overall, a combination of PI3K and EGFR inhibitors was the most commonly indicated combination (Figure 3D). Additional combination therapies that might be particularly beneficial for BQ-driven OSCC (difference in the frequency with co-alteration of actionable variants>15 %) include CDK4 and FGFR inhibitors in tongue carcinoma (18 % vs 0 %), PI3K and CDK4 inhibitors in cancers of buccal mucosa (31 % vs 9 %), PI3K and FGFR inhibitors in cancers of buccal mucosa (31 % vs 9 %), and PI3K and PARP inhibitors in cancers of buccal mucosa (34 % vs 18 %). This assessment demonstrates a broad spectrum of possible tumor site-specific and BQ-associated therapeutic combinations that can be examined in pre-clinical and clinical contexts.

Fig 3.

Putative actionability across BQ-related and -unrelated tumors from distinct anatomic sites of the oral cavity. (A-B) Frequencies of therapeutic actionability in cancers of the tongue (Tongue), buccal mucosa (BM), and overall oral cavity (OSCC), broken down by the level of evidence (A) and number of actionable variants detected (B). Samples are classified by variants considered actionable (the highest level of evidence) or not actionable but oncogenic. (C) Frequencies of actionable alterations per gene. For genes with multiple levels for different variants, more than one row is shown. (D) Frequencies of potential drug combinations indicated by the co-alteration of targetable variants.

Cellular components and their expression programs

In addition to assessing BQ-associated genomic abnormalities, we combined single-cell and spatial transcriptomics (ST) to comparatively characterize the transcriptional programs of diverse cell populations within the tumor microenvironment. After quality control filtering, transcriptomes of 14,590 cells from two tumors of the buccal mucosa and 24,743 spots from 10 tumors of the same anatomical site were profiled by scRNA-seq and ST (Table S4 and Figure S3), respectively. Unsupervised clustering identified 10 major cell types, including epithelial cells, immune cells (T, NK, B/plasma cells, monocytes/macrophages, dendritic and mast cells), and stromal cells (fibroblasts, endothelial cells, and myocytes) (Figure 4A-B), all bearing resemblance to cell compositions from prior scRNA-seq studies of head and neck cancer [8,13,14]. To clarify whether BQ use caused a shift in cell compositions of the OSCC microenvironment, we performed deconvolution analyses to determine cell-type proportions in ST spots (Figure 4C-D). Despite a considerable inter-sample heterogeneity in cell-type deconvolution data across different tumors, proportions of spots enriched with monocytes and macrophages (MoMΦ) in BQ-related samples were higher than that of their BQ-unrelated counterparts (Figure 4E and S4), suggesting a promotive effect of BQ on the recruitment of MoMΦ to the tumor niches.

Fig 4.

Integration of single-cell and spatial transcriptomics reveals the promotion of monocyte recruitment and hypoxic epithelium in BQ-related OSCC. (A) UMAP representation of 14,590 single cells profiled in the presenting work colored by major cell types. (B) Expression levels of diverse marker genes by annotated cell clusters. (C) Spot deconvolution into proportion of major cell types identified from scRNA-seq experiments and colored as (A). Pie charts displaying the spatial distribution of the estimated cell type proportion per spot. (D) Estimated proportion of monocyte/macrophage (MoMΦ) and expression intensity of MoMΦ signature genes. (E) Proportions of spots enriched with T, NK, dendritic cell, or MoMΦ across samples. Stacked bars exhibiting proportions of spots with different abundance of a given cell type. Difference was determined between BQ+ and BQ- samples at the proportion of ≥ 0.3. * p < 0.05, by Wilcoxon rank sum test. (F) Expression of top 10 shared basal, cycling, differentiating, and mesenchymal marker genes across individual cells of three tumor subpopulations. (G) Violin plots of relative signature score for three tumor subpopulations in epithelium-enriched spots across samples. (H) Violin plots of relative hypoxia score in epithelium-enriched spots across samples. Association was determined between BQ+ and BQ- samples by Wilcoxon rank sum test. (I) Spatial feature plots showing expression of select epithelial (SERPINB5 and KRT6B) and hypoxia marker genes (NDRG1, BHLHE40, PGK1, and ADM) from Patient VT5 and VT9.

We next compared the epithelial, stromal, and immune (see Supplemental Results) expression programs identified from our scRNA-seq data as well as other studies (Table S5). For tumor cells, scRNA-seq profiling of 756 epithelial cells, whose malignant transformation was verified by inferring large-scale chromosomal CNVs (Figure S5A-B), revealed three tumor subpopulations (P1-3), each recapitulating one or a blend of pre-defined expression programs (Figure 4F). Of note, one (P3), exhibiting shared basal, cycling, and mesenchymal marker genes, was corresponding to a previously described program, which was unique to squamous cell carcinoma and correlated with invasive behavior and EMT [15] (Figure S5C-E). However, none of these programs was associated with BQ through evaluation of their epithelial expression intensity in ST samples (Figure 4G). Moreover, we further analyzed other tumor expression patterns extracted from relevant studies (Table S5) to improve the resolution of gene signature characterization. Among these programs, the epithelial expression of a hypoxia signature but not the EMT programs in BQ-related tumors was higher than that among BQ-unrelated samples (Figure S6), and this trend was mostly reproduced by using different gene lists from multiple related studies (Figure 4H-I and Table S5). These implicate a role of BQ in contributing to tissue hypoxia during the development of OSCC.

Additionally, we tested a potential effect of BQ consumption on reshaping OSCC stromal expression signatures, with a focus on cancer-associated fibroblasts (CAFs) and tumor endothelial cells (TECs). A total of 1,001 CAFs from our scRNA-seq assay were stratified into five clusters, of which three were assigned into an immunomodulating phenotype (iCAF1-3, expressing IL6) and two into a contractile phenotype (myoCAF1-2, expressing ACTA2 and MYLK) based on a consensus document [16] (Figure S7A-B). Among these CAF subsets, iCAF1 was found to exhibit potent expression of many hallmarks that drive cancer development, such as angiogenesis, EMT, and inflammation (Figure S7C) but failed to be associated with BQ use in our ST dataset (Figure S7D). Since the EMT program has been suggested to localize in the tumor niches in proximity to CAFs [14] and associated with etiology of BQ-related OSCC, the expression of EMT programs from relevant studies (Table S5) was assessed in CAF-enriched compartments of ST tumors. However, no association of BQ chewing was detected with the EMT program of CAFs (Figure S7E). Moreover, 850 TECs from our scRNA-seq procedure were grouped into three major subtypes (TEC1-3), of which TEC2 resembled a tip cell phenotype of endothelial sprouts characterized by invasive features and regulated predominantly by the Notch signaling [17] (Figure 5A-D). Strikingly, evaluation of endothelial program expression in the TEC-enriched compartments of ST samples showed that, rather than TEC2 or the tip-like cell, the expression intensity of a stalk cell signature and an angiogenic TEC program (Table S5) in BQ-related tumors was higher than that of BQ-unrelated samples (Figure 5E-F), indicating a role of BQ in promoting angiogenesis of TECs. Collectively, our survey links the use of BQ to the shift of OSCC microenvironment, involving induced hypoxia of tumor epithelium, altered phenotypes of DCs, and raised sprouting angiogenesis of TECs.

Fig 5.

Promotion of sprouting angiogenesis in the endothelium of BQ-related OSCC. (A) UMAP plot of 850 endothelial cells profiled in the presenting work colored by cell subsets. (B) Violin plots of the previously-defined tip-cell and stalk-cell signature scores in three endothelial subpopulations. (C) Heatmap depicts differences in pathway activities for each hallmark MSigDB program among three endothelial subsets. Each column is normalized by z-score to indicate the relative pathway activities. (D) Heatmap depicts expression of genes in Notch signaling. (E) Violin plots of relative program scores in endothelium-enriched spots across samples. Association was determined between BQ+ and BQ- samples by Wilcoxon rank sum test. AngioEC, angiogenic endothelial cell. (F) Spatial feature plots showing expression of select endothelial (VWF, ENG, and CD34) and stalk-cell marker genes (JAM2, ACKR1, and TSPAN7) from Patient VT2 and VT7.

Cell-cell crosstalk

Given the considerable importance of communications among cells in the development of head and neck cancers [[18], [19], [20]], we conducted quantitative inferences of intercellular communication networks to explore the impact of BQ chewing on reconstructing OSCC microenvironment. As epithelial-epithelial communications contributed to a significant portion of ligand-receptor interactions in both BQ-related and -unrelated tumors, extensive communications of fibroblasts and DCs with other cell types were inferred particularly in BQ-related OSCC (Figure 6A), indicating a unique cell communication landscape within this etiological subtype. Specifically, key ECM-receptor pathways, such as collagen and laminin, exhibited abundant signaling interactions between fibroblasts and diverse non-epithelial cell types in BQ-related tumors (Figure 6B). However, this communication pattern was not seen in BQ-unrelated OSCC, where considerable collagen and laminin signaling events were mostly restricted to fibroblast-epithelial interactions only. Moreover, such BQ-associated effect on fibroblast-mediated cell signaling does not involve key secreted signaling pathways, such as SPP1 and WNT. Similarly, key ECM-receptor pathways largely accounted for the interactions of DCs with other non-epithelial cell types in BQ-related OSCC but less so than in BQ-unrelated tumors (Figure 6C). In addition, to a less extent, an impact of BQ use on DC-lymphocyte communications driven by chemotaxis pathways, CCL and CXCL, was noted. These predicted networks indicate a more heterogeneous cell-cell crosstalk in BQ-related OSCC, involving substantial interactions of fibroblasts and DCs with other non-epithelial cell types largely via ECM-receptor signaling pathways.

Fig 6.

A unique cell communication landscape within the niche of BQ-related OSCC. (A) Heatmaps depict the number of ligand-receptor interactions between pairs of cell groups in aggregated BQ+ and BQ- samples. (B-C) The inferred communication networks of individual ligand-receptor pairs associated with a given signaling pathway between fibroblasts and other cell groups (B) and between dendritic cells (DC) and other cell groups (C). The edge width is proportional to the indicated number of ligand-receptor pairs. MoMΦ, monocyte/macrophage; DC, dendritic cell; Epi. cell, epithelial cell; Endo. cell, endothelial cell.

Discussion

In this comparative survey, we extended the unique features of BQ-related OSCC to include recurrent mutations, focal somatic copy-number gains and losses, mutational signatures, oncogenic pathway alterations, potential druggability, distinct gene expression programs, and cell-cell communications. Our findings suggest that habitual use of BQ disrupts tumor suppressor pathways, p53 and Hippo, to ruin cell growth and genomic stability, coupled with dysregulated expression patterns to promote tissue hypoxia, tumor angiogenesis, and dendritic immunosuppression. These data reiterate the notion that OSCCs caused by BQ chewing represent a different molecular entity, as compared to those free of this risk.

As previous OSCC studies pertaining to the mutation status of partial TP53 gene (DNA binding region only) associated with BQ chewing have been widely contradictory [[21], [22], [23], [24], [25]], we demonstrated that TP53 was mutated more commonly in BQ-related OSCC, and p53 signaling was a frequently altered oncogenic pathway linked to BQ use. These observations reinforce the etiological role of BQ in oral tumorigenesis. Intriguingly, we also found that CDKN2A deletion but not mutation was enriched among BQ-related OSCC. Genomic alterations in CDKN2A and TP53 have been reported as the core of the earliest genetic events in a direct analysis of oral premalignant lesions, related to their genomic positions in loci of loss of heterozygosity [26]. As a link of BQ chewing to the pathogenesis of oral submucosal fibrosis, a precancerous condition, was demonstrated [5], our data together with others’ findings suggest that BQ-mediated mutagenesis in oral epithelium represents an early clonal event of potential pathophysiological significance. Moreover, we identified another BQ-associated OSCC driver gene, CHUK, which encodes the kinase IKKα that inhibits NF-κB activation [27]. Constitutive activation of NF-κB has been observed in numerous cancers, yet oncogenic mutations in NF-κB genes are scarce and merely occur in upstream components of the signaling cascade [28]. Our data for the first time connect NF-κB to the impact of BQ chewing on OSCC development. Overall, this oncogenic gene/pathway landscape in BQ-related OSCC is meant to offer an additional resource for fostering the understanding of pathogenic mechanisms and advancing early diagnosis of this malignancy.

Combination of surgery, radiotherapy or chemotherapy remains the standard treatment of choice for OSCC but fails to provide significant improvement in survival and quality of life in patients. Due to a highly heterogeneous nature of genetic complexity inherent in OSCC, targeted oral cancer therapies are still challenging [29]. In our cohorts, the most common putatively druggable gene bearing alterations with actionability for standard care was PIK3CA, and a joint inhibition of PI3K and EGFR was the most commonly indicated combination. These data from our sequencing experiments are in concordance with the results of recent clinical studies where treatment with buparlisib (a PI3K inhibitor) plus cetuximab (an EGFR inhibitor) significantly improved the disease control rate and prolonged survival outcomes of recurrent and metastatic head and neck cancer patients [30,31]. However, resistance of OSCC to chemotherapy, radiotherapy and some targeted therapies is common, particularly for tumors driven by BQ [32,33]. Of note, dysregulation of Hippo signaling, the top altered oncogenic pathway detected in BQ-related OSCC (63 %) in our survey, was shown to mediate intrinsic and acquired resistance to cancer therapies [34], in addition to its prevailing effects on tumor development. Emerging evidence shows that targeting aberrant activation of Hippo-YAP/TAZ signaling could overcome the resistance of many solid tumors to multiple subtypes of chemotherapy, including anti-microtubule, anti-metabolites, and DNA damage agents [35,36]. Furthermore, dysregulation of Hippo signaling and subsequent activation of YAP/TAZ were functionally involved in the resistance mechanism to numerous targeted therapies, particularly against EGFR, MAPK, and CDK inhibitors [37]. Thus, combination therapies with Hippo-YAP/TAZ inhibitors [38] may improve the efficacy of targeted and chemotherapies by reducing therapy resistance in cancers with aberrant regulation of Hippo signaling, including BQ-related OSCC.

In addition, by deconvoluting BQ-related OSCC genomes, we identified mutational signatures (SBS and ID), whose contribution to the total number of mutations was higher in tumors from BQ chewers. Among these, one such mutational pattern, corresponding to SBS5 in COSMIC, was previously identified as a direct consequence of exposure to a carcinogenic nitrosamine compound [39]. Since BQ-derived nitrosamines can generate bulky DNA adducts that are substrates of nucleotide excision repair [40], habitual BQ use may saturate the DNA repair capability of OSCC, implicating an involvement of BQ-derived nitrosamines in proposed etiology of SBS5. Notably, as BQ chewing was shown to cause more shorter INDELs (<3 bp) found at mono/polynucleotide repeats in OSCC [41], enrichment of a de novo extracted ID signature featured as a blend of COSMIC ID1, ID8, and ID10 was found in our BQ-related samples, implying the presence of BQ-specific ID signatures.

Earlier studies employing in vitro and animal models treated with active components or crude extract of BQ have delineated cellular and molecular effects of BQ consumption on oral tumorigenesis, including facilitation of EMT, autophagy initiation, tissue hypoxia, and cancer stemness conversion [4]. Here, unlike a great variation in the expression of EMT programs across samples, an increase in epithelial expression intensity of hypoxia programs was observed among BQ-related OSCC specimens via an integration of scRNA-seq and ST data. Such BQ-triggered hypoxic condition has been linked to increased chemoresistance [42], providing additional explanation for a treatment-refractory phenotype of BQ-related OSCC. In addition, tissue hypoxia gives rise to augmentation of tumor angiogenesis by promoting the mitogenic and migratory activities of TECs [43]. In cultured endothelial cells, ROS induction by a BQ carcinogen, arecoline, stimulated the expression of a proangiogenic enzyme, hemeoxygenase-1 [44]. These are in concordance with our observation that expression of a stalk cell (an endothelial cell type that forms the body of endothelial sprouts) signature and an angiogenic TEC program was higher in BQ-related specimens, highlights a promotive role of BQ in sprouting angiogenesis. Intriguingly, a dendritic expression signature of T cell suppression and an activated phenotype that closely resembled mregDCs were highly expressed in BQ-related OSCC. DCs are key components of the tumor niche and can be classified into distinct phenotypic subsets that either promote or suppress anti-tumor T-cell responses through different mechanisms involving secreted signaling and cell-cell contact [45]. However, how chronic exposure to BQ constituents affects dendritic functions in the immune cell ecosystem of OSCC remains mostly unknown. With the identification of a similar mregDC signature, a gene program enriched in immunoregulatory and maturation genes expressed by conventional DCs after uptake of tumor antigens [46], it is conceivable that massive amounts of neoantigens derived from BQ-mediated genotoxicity may reshape OSCC-associated DC biology. Our data indicate that BQ use drives oral tumorigenesis by facilitating an immunosuppressive dendritic program that restrains anti-tumor immunity in BQ-related OSCC.

Another intriguing finding from our integrative analyses of scRNA-seq and ST data is the presence of a more heterogeneous cell-cell crosstalk in BQ-related OSCC, featured by extensive interactions of fibroblasts with other non-epithelial cell types mainly via ECM-receptor signaling pathways. BQ chewing was known to cause oral submucous fibrosis, a precancerous condition with accumulation of collagen being a notable pathological manifestation [5]. In fibroblast cells, arecoline stimulated collagen synthesis [47], and other BQ constituents, such as tannins, flavonoids, and catechins facilitated the crosslinking of collagen fibers, resulting an attenuated susceptibility to collagenases [48]. Our data together with others’ dictate a shift in OSCC microenvironment where BQ chewing promotes the communications between stromal fibroblasts and other non-epithelial cell types largely through modulating the turnover and remodeling of ECM components.

To assemble a comprehensive map for the genomic alterations and gene expression signatures of OSCC associated with BQ chewing, additional efforts are needed to address several study limitations. One concern is that the composition of BQ varies geographically. There are three forms of commonly-used BQ products in Taiwan, all of which are linked to unique molecular pathology of oral cancer [49]. Thus, some BQ-associated genomic abnormalities detected in this study may be restricted to certain ethnic or geographic cohorts. Another issue is that the influences from covariate behaviors of BQ chewers, prominently alcohol drinking and cigarette smoking, cannot be ruled out. In addition, a modest sample size of OSCC tissues was profiled by ST, which might present challenges concerning generalizability of findings. Ideally, these expression programs and relationship among major cell types within OSCC niche can be replicated in an independent cohort or verified via mechanistic animal assays. Such experiments are typically cost- and time-consuming; therefore, our quantitative inferences could serve as a useful map of BQ-associated OSCC microenvironment.

Conclusions

Collectively, our results extend the unique characteristics of BQ-related OSCC to comprise recurrent mutations, focal somatic copy-number gains and losses, mutational signatures, oncogenic pathway alterations, potential druggability, distinct gene expression programs, and cell-cell communications. We anticipate these data to be helpful for improving the understanding of pathogenic mechanisms and for advancing early diagnosis and treatment efficacy of this malignancy.

Ethics approval and consent to participate

This study was approved by the institutional review board of Chung Shan Medical University Hospital, Taichung, Taiwan (CS1-20091). All participants provided informed written consent at enrollment.

Consent for publication

Not applicable.

CRediT authorship contribution statement

Shih-Chi Su: Data curation, Writing – original draft, Conceptualization. Chiao-Wen Lin: Formal analysis, Resources. Mu-Kuan Chen: Resources. Yi-Chan Lee: Resources, Investigation. Chun-Wen Su: Investigation, Formal analysis. Shi Bai: Formal analysis, Methodology. Hansraj Jangir: Formal analysis. Chun-Yi Chuang: Formal analysis. Wen-Hung Chung: Data curation. Lun-Ching Chang: Visualization, Formal analysis, Writing – original draft. Shun-Fa Yang: Conceptualization, Writing – review & editing, Data curation.

Declaration of competing interest

The authors declare that they have no competing interests.