Abstract
Extensive studies have demonstrated the relationship between metabolic reprogramming and the tumor microenvironment. Here, we characterized the head and neck squamous cell carcinoma (HNSCC) evolutionary landscape using spatial metabolomics/transcriptomics, single-cell transcriptomics, and bulk multi-omics. Metabolic heterogeneity during HNSCC malignant transformation was identified, with significant enrichment in the purine metabolism. Integrating single-cell and bulk data, we developed a robust ligand-receptor-based signature (LRS) linked to NT5E, a key upstream regulator of purine metabolism, which served as an independent prognostic indicator. The low LRS subtype was associated with a high proportion of immune cell infiltration and improved response to immunotherapy. Notably, in vitro and in vivo experiments demonstrated that AMIGO2, a core molecule within the LRS, regulates tumor-associated purine metabolism, and that its downregulation suppresses tumor cell invasion and migration, inhibits myofibroblast differentiation, and promotes immune effector cell infiltration. Moreover, combining AMIGO2 targeting with anti-PD-1 therapy yielded superior efficacy. Consistent validation was also obtained in a clinical cohort of HNSCC and premalignancy patients.
Similar content being viewed by others
Introduction
Head and neck cancers are the sixth most common cancers worldwide, with squamous cell carcinoma accounting for approximately 90% of cases, making it the predominant pathological type1,2. Despite significant advances in multidisciplinary treatments-including surgery, radiotherapy, and chemotherapy—for head and neck squamous cell carcinoma (HNSCC) patients, the five-year survival rate remains relatively poor3,4. In recent years, immune checkpoint blockade therapy has become a first- or second-line treatment for recurrent and metastatic HNSCC, greatly improving clinical outcomes5. However, optimizing treatment response rates and improving overall survival (OS) remain key challenges.
Tumor cells accumulate metabolites to rapidly meet their energy and nutritional demands. HNSCC exhibits a range of metabolic reprogramming patterns6, including abnormal cholesterol metabolism7, glucose metabolism8, amino acid metabolism9, and polyamine metabolism10, which drive genomic mutations in tumor cells and remodel the tumor microenvironment (TME) to promote tumor growth. Guan et al. used bulk and single-cell RNA-seq (scRNA-seq) data to perform deconvolution and construct the metabolic feature-based prognostic model METArisk11. Similarly, Eldridge et al. used bulk metabolomics, observing that plasma metabolites may serve as prognostic biomarkers for HNSCC12. However, these methods lack detailed molecular location information and fail to adequately capture metabolic heterogeneity across various tissue regions. More recently, Ravi et al. employed spatial metabolomics to explore how regional metabolic changes drive the clonal evolution of glioblastoma13. As such, spatial omics approaches are critical for precisely identifying key metabolic pathways enriched in specific tumor regions in HNSCC.
Metabolic reprogramming is a key factor in tumor-immune interactions14. Metabolic patterns influence the innate or adaptive immune capabilities of the TME, as the abnormal consumption of metabolites leads to local hypoxia, decreased pH, and the accumulation of metabolic waste. This, in turn, suppresses immune cell proliferation and effector functions, promoting immune evasion and tumor progression15. Specifically, the kynurenine pathway from tryptophan in HNSCC can induce CD8 + T cell exhaustion16. Chen et al. discovered that combining radiotherapy with HSP60-targeted therapy weakens tumor cell glycolysis, thereby shaping an inflammatory microenvironment17. However, the mechanisms by which malignant cells communicate with TME cells through metabolic stress to modulate immune function remain unclear, and the clinical significance and therapeutic impact of the related metabolic signatures remain insufficiently understood.
In this study, we utilized multimodal data to elucidate the evolutionary characteristics of the TME during HNSCC progression (Fig. 1). Specifically, spatial metabolomics and transcriptomics analyses identified aberrant activation of the purine metabolism pathway within epithelial regions. Integration of bulk and scRNA-seq further pinpointed NT5E as a key metabolic regulatory gene embedded within the cellular communication network. Building on these findings, we developed a novel and robust TME-based prognostic model-the ligand-receptor-based signature (LRS)-using machine learning approaches, which highlighted AMIGO2 as a central model node and potential therapeutic target. Functional in vitro and in vivo experiments confirmed that AMIGO2 plays a critical role in promoting HNSCC malignancy and orchestrating TME remodeling. Collectively, our findings delineate the complex metabolic and transcriptional alterations underlying HNSCC evolution, providing a framework for both therapeutic intervention and prognostic prediction.
Illustration of the workflow and an overview of approaches used.
Results
Spatial metabolomics (SM) combined with weighted gene co-expression network analysis (WGCNA) identifies altered purine metabolic pathways in HNSCC malignant transformation
To investigate metabolic changes during HNSCC evolution, we collected samples from four patients with oral squamous cell carcinoma to conduct AFADESI-MSI. SM data (defined by SM1-SM4) were obtained to map the temporal and spatial distribution of metabolites (Fig. 2a), and UMAP clustering analysis was performed to categorize these metabolites (Fig. 2b). With the assistance of two pathologists, three epithelial pathological regions were annotated, including normal region (N-region), precancerous region (P-region), and tumor region (T-region) (Fig. 2c). WGCNA, which clusters metabolites into modules based on correlation patterns18, was applied to profile metabolic alterations during HNSCC malignant transformation. This analysis identified several co-expressed metabolite modules, designated by arbitrary colors (Supplementary Fig. 1a–e). Among them, the Blue module is the metabolite module-trait most strongly associated with HNSCC initiation (Fig. 2d). KEGG enrichment analysis indicated that metabolites in the Blue module were significantly enriched in the purine metabolism (Fig. 2e; Supplementary Data 1), including adenosine and adenosine monophosphate, whose levels progressively increased during tumor evolution (Fig. 2f, g). Given that PAICS and HPRT1 are key enzymes representing the de novo and salvage branches of purine metabolism, respectively, we next examined whether their tissue-level expression aligned with the pathway activation indicated by metabolomics. In an independent tissue cohort, immunohistochemistry (IHC) revealed a stepwise increase in PAICS and HPRT1 expression from normal epithelium through precancerous lesions to tumor regions (Fig. 2h, i), thereby corroborating the sustained activation of purine metabolism during the malignant transformation of HNSCC.
a Schematic of the experimental design. b UMAP plot of SM data generated from HNSCC samples. c H&E staining of the HNSCC sample. d Correlation between module metabolites and pathological progression features, with values represented as “Pearson r (P value)” in the cells. e KEGG enrichment analysis of metabolites in the Blue module. f, g Distribution of adenosine and adenosine monophosphate in HNSCC tissues. h, i IHC images and quantitative statistics of PAICS and HPRT1 expression in adjacent normal tissue, precancerous lesions, and tumor tissues. *P < 0.05, **P ≤ 0.01, ***P ≤ 0.001.
Transcriptomic profiling uncovers metabolic and microenvironmental alterations in HNSCC
To explore the alteration of purine metabolism at the cellular level, we performed quality control and removed batch effects from the scRNA-seq dataset. A total of 348,668 cells from 114 scRNA-seq samples were selected for subsequent analysis (Supplementary Fig. 2a, b). We then annotated 14 major cell types using specific marker genes (Fig. 3a, b). The changes in the proportion of cell types reflect the tissue heterogeneity. In tumor tissues, the proportion of epithelial and inflammatory cells increases (Fig. 3c). Fibroblasts, endothelial cells, and myeloid cells are often closely associated with tumor progression and contribute to the formation of an immunosuppressive microenvironment19,20. To explore this further, we performed subcellular clustering of fibroblasts, endothelial cells, and myeloid cells (Supplementary Fig. 2c–h). Our analysis revealed significant differences in the proportions of most cell types between normal and tumor tissues (Supplementary Fig. 2i–k), including ACKR1+ endothelial cells21, myofibroblasts22, and C1QA+ macrophages23. Notably, compared to normal epithelial cells, the pathways in cancer score and purine metabolism score were elevated in the tumor cells (Fig. 3d, e). Spatial transcriptomics (ST) analysis further confirmed the high level of purine metabolism in the malignant epithelial regions (Fig. 3f; Supplementary Fig. 3a). Kaplan-Meier survival analysis revealed that high expression of purine metabolism was associated with poor OS in patients from the TCGA and meta-GEO datasets (Fig. 3g, h). Interestingly, analysis of the TCGA-HNSCC dataset revealed that purine metabolism was positively correlated with tumor purity but negatively associated with immune score, stroma score, and cytolytic activity (CYT), suggesting its potential role in reshaping the TME (Supplementary Fig. 3b). In summary, these integrated analyses highlight dysregulated purine metabolism and TME remodeling during HNSCC malignant transformation.
a UMAP plot of 348,668 cells from scRNA-seq analysis. b Bubble plot of marker genes expressed in major cell types. c Cell type proportions in different tissue types. d, e Boxplot comparing the pathways in cancer score and purine metabolism score in epithelial cells of scRNA-seq datasets. f Spatial distribution of purine metabolism score in tissue slices from GSM6339632 and GSM6339637. g Kaplan–Meier survival analysis stratified by purine metabolism score in TCGA patients. h Kaplan-Meier survival analysis stratified by purine metabolism score in merge-GEO cohort patients.
NT5E is aberrantly expressed in tumors, modulates purine metabolism, and drives HNSCC progression
To further investigate the molecular mechanisms of metabolism in HNSCC, we selected upstream regulatory genes of metabolic enzymes based on purine metabolites in the Blue module and KEGG database. Differential expression analysis using TCGA data revealed 23 genes that were aberrantly expressed in tumor samples (Fig. 4a). We then compare the prognostic value of these 23 genes in the TCGA and meta-GEO datasets. Univariate Cox regression analysis revealed that only changes in NT5E expression significantly affected patient prognosis (Fig. 4b). Additionally, compared to normal epithelial cells, NT5E was highly expressed in tumor epithelial cells (Supplementary Fig. 4a). AB680 is a potent and selective NT5E inhibitor that has demonstrated efficacy in suppressing tumor progression in pancreatic cancer24 and melanoma25. This efficacy is attributed not only to its role in restraining adenosine accumulation but also to its broader modulation of the purine metabolite pool, thereby deregulating immune suppression. In addition, it has a long half-life and is well tolerated, which facilitates clinical application26,27. In vitro, AB680 treatment downregulated the expression of purine metabolism-related proteins (PAICS and HPRT1) and inhibited the epithelial-mesenchymal transition (EMT) process in HNSCC cells (Supplementary Fig. 4b, c). In bulk RNA-seq (Fig. 4c), HNSCC cell lines treated with AB680 showed downregulation of 281 genes following treatment (Fig. 4d; Supplementary Data 2). GSEA revealed that AB680 exposure significantly affected purine metabolism pathways (Supplementary Fig. 4d). Enrichment analysis of the 281 differentially expressed genes (DEGs) indicated significant associations with biological processes such as extracellular matrix organization, cytokine-cytokine receptor interaction, regulation of chemokine production, and regulation of cell-cell adhesion (Fig. 4e). In immunocompetent mice bearing subcutaneous MOC-1 tumors (Supplementary Fig. 4e), AB680-treated allografts exhibited significantly reduced tumor volumes and weights (Supplementary Fig. 4f–j). Furthermore, AB680 treatment recruited more activated NK cells and effector CD8 + T cells, while reducing the infiltration of exhausted CD8 + T cells, thereby enhancing anti-tumor immune responses compared to controls (Supplementary Fig. 4k).
a Differential expression profiles of 31 purine metabolism regulators in normal (n = 44) and HNSCC samples (n = 509) from TCGA dataset. b Univariate Cox regression analysis of 23 purine metabolism regulators in TCGA and merge-GEO datasets. c Strategy for sequencing Fadu cells treated with AB680. d Heatmap showing the gene downregulation profile following AB680 treatment. e Biological pathway analysis of Fadu cells after AB680 treatment. f Schematic diagram of the screening strategy for potential receptor-ligand genes in cell-cell communication. g Bubble plot of receptor-ligand interactions between different epithelial cells and myofibroblast, INSR+ endothelial cells, and C1QA+ macrophages. Ns: no significant difference, *P < 0.05, **P ≤ 0.01, ***P ≤ 0.001.
Given that TME reprogramming is often initiated by the activation of key ligand-receptor signaling pathways, we intersected the above 281 DEGs with ligand-receptor genes from the CellPhoneDB database and performed cell-cell communication analyses using scRNA-seq datasets (Fig. 4f). The results indicated that NT5E+ epithelial cells exhibited strong interactions with myofibroblasts, INSR+ endothelial cells, and C1QA+ macrophages. Ligand-receptor pairs such as FN1-NT5E, TNC-ITGA5, and SERPINE1-PLAUR were among the 52 enriched interactions identified, suggesting that NT5E may contribute to the recruitment or activation of immunosuppressive stromal and immune cells within the TME (Fig. 4g). In conclusion, NT5E is a potential key molecule linking purine metabolism in HNSCC to the formation of an immune-suppressive TME.
Unsupervised clustering analysis identifies two ligand-receptor related clusters consistent with patient classification
Ligand-receptor pairs play a crucial role in cell communications28. They often alter tumor dynamics and evolutionary trajectories by influencing cell interactions, regulating biological behaviors, and driving tumor initiation and progression29,30. To capture metabolism-driven cross-cell effects, we performed a ligand-receptor network analysis focused on purine metabolism to derive patient stratifications that determine prognosis and microenvironmental states. From this analysis, 52 enriched ligand-receptor genes were selected for further evaluation. To assess their prognostic impact, we conducted univariate Cox regression in the merged HNSCC cohort and identified 17 prognosis-related genes, which exhibited strong intercorrelation (Supplementary Fig. 5a, b). Based on the Consensus Clustering method, we divided the patients in the merge-HNSCC cohort into two clusters. Cluster2 (C2) demonstrated a better prognosis than C1 (Supplementary Fig. 5c, d). To explore the pathway changes that might explain the survival differences, we conducted differential analysis of the clusters and identified 524 DEGs. Enrichment analysis revealed that C1 was associated with EMT, positive regulation of angiogenesis, cancer pathways, and the PI3K-Akt signaling pathway, whereas C2 was closely associated with the regulation of T cell activation, inflammatory response, adaptive immune response, and antigen processing and presentation (Supplementary Fig. 5e). GSVA indicated that C1 was enriched in malignant tumor progression and metabolic regulation, including hypoxia, Wnt β-catenin signaling, glycolysis, purine metabolism, and purine biosynthesis (Supplementary Fig. 5f, g). C1 showed greater regulation of immune suppression and tumor metabolic reprogramming, while C2 showed greater enrichment of immune cells such as B cells, NK cells, and CD8 + T cells, as well as the activation of immune-related pathways, such as tertiary lymphoid structures (TLS), T-cell receptor (TCR) signaling, B-cell receptor (BCR) signaling pathways (Supplementary Fig. 5h). Collectively, these findings suggest that C2 is characterized by immune activation and inflammatory response, while C1 exhibits more pronounced immune suppression.
Construction and validation of the LRS model based on machine learning integration
To develop a more robust clinical prediction model, we applied XGboost and LASSO algorithms to the 524 DEGs identified between C1 and C2 subtypes, narrowing the candidate genes down to 65. These genes were further tested using Cox regression with survival data, yielding 21 genes. Subsequently, a machine learning-based integration program was constructed to establish a consensus model (Fig. 5a; Supplementary Data 3). Notably, the Random Survival Forests (RSF) algorithm achieved the highest average concordance index (C-index = 0.764), and 8 genes were ultimately selected to construct the LRS model (Supplementary Fig. 6a). Kaplan-Meier survival analysis showed that patients in the low LRS group had significantly better OS compared to those in the high LRS group across the TCGA-HNSCC, merge-GEO, and ICGC-OSCC cohorts (Fig. 5b–d). At the clinical-subgroup level, we conducted stage-stratified validation of the LRS across the TCGA-HNSCC, merged-GEO, and ICGC-OSCC cohorts. In both stage I-II and stage III-IV disease, patients with a low LRS consistently showed superior OS compared with those with a high LRS (Supplementary Fig. 6b–f).
a Flowchart for constructing the LRS model. b–d Kaplan-Meier analysis for LRS subtypes in TCGA-HNSCC, merge-GEO, and ICGC-OSCC cohorts. e Pie chart showing the differences between LRS high and low groups determined by Chi-square test. f Univariate and multivariate Cox regression analysis of LRS score in relation to HPV status, age, gender, tumor stage, and radiation in the TCGA cohort. g Kaplan–Meier analysis of LRS subtypes in the in-house cohort.
Additionally, we observed that in the TCGA dataset, the LRS model was significantly associated with survival status, HPV status, smoking status, tumor stage, and radiation treatment (Fig. 5e). Compared to traditional classification studies of HNSCC, the LRS signature was significantly upregulated in the Wound Healing, Non-Immune, Mesenchymal, and SCC-3 subtypes, suggesting that a higher LRS score involved in tumor progression and invasion, and is enriched in immune desert subtypes (Supplementary Fig. 7a–d). Moreover, in the multivariate Cox regression risk model, the LRS was confirmed as an independent prognostic biomarker (for OS, HR = 13.25, 95% CI = 8.87 to 19.77, P < 0.001) (Fig. 5f). Surprisingly, the LRS model outperformed other signatures in nearly all datasets when comparing their C-index values (Supplementary Fig. 7e; Supplementary Data 4). Importantly, we collected samples from in-house cohort, performed qRT-PCR, and constructed the model using the RSF approach. We found that patients with a high LRS score had a poor prognosis (Fig. 5g; Supplementary Fig. 7f).
LRS model is closely associated with TME remodeling and clinical treatment response
We further evaluated the role of the LRS model in HNSCC development and progression. The LRS score is increased in the TP53 mutant status (Fig. 6a). Chromosomal alterations status in the two LRS feature subgroups showed distinct frequent changes, with deletions and amplifications being particularly prominent in the high LRS group (Fig. 6b, c). We performed correlation analysis on two datasets (Fig. 6d) and observed a significant negative correlation between the LRS score and immune-related pathways, including CYT, TLS, TCR, BCR, inflammatory signature, and antigen processing and presentation. In contrast, various oncogenic signaling pathways, such as hypoxia, angiogenesis, WNT signaling, and pro-tumor cytokines, were significantly positively correlated with LRS scores. Additionally, the LRS score is primarily associated with immune checkpoint molecules, and factors involved in extracellular matrix remodeling (Fig. 6e). We also assessed the immune treatment response in patients from the TCGA, GEO-meta, and ICGC cohorts, finding that patients with high LRS scores had higher TIDE scores, indicating a poor response to immune checkpoint inhibitor therapy (Fig. 6f–h). Subsequently, we identified clinically-relevant targeted drugs and analyzed their association with different groups. The patients with high LRS exhibited higher IC50 values for drugs, suggesting that low LRS score patients are more sensitive to Sorafenib, Cisplatin, Erlotinib, and 5-Fluorouracil (Fig. 6i–l).
a Boxplot comparing LRS scores in HNSCC samples from TCGA with different TP53 mutation statuses. b, c Boxplots showing higher frequency of amplification/deletion mutations in the high LRS score group. d Correlation analysis between LRS score and cancer-related pathways in TCGA-HNSCC and merge-GEO cohorts. e Heatmap depicting the correlation between LRS scores and representative immune activities. f–h Violin plots showing significant differences in TIDE scores associated with LRS subtypes across TCGA-HNSCC, merge-GEO, and ICGC-OSCC cohorts. i–l Comparison of drug sensitivity between LRS subtypes for four commonly used clinical drugs: Sorafenib, Cisplatin, Erlotinib, and 5-Fluorouracil. *P < 0.05, **P ≤ 0.01, ***P ≤ 0.001.
Overview of AMIGO2 in human cancers
We visualized the expression of eight genes in the LRS model within scRNA-seq data and found that their expression was heterogeneous across cell clusters (Fig. 7a). After defining patient categories based on the LRS score, it was notable that the proportion of immune cells such as T cells, B cells, and NK cells showed higher proportion levels in the low LRS group, while the proportion of fibroblasts, endothelial cells, and epithelial cells showed lower proportion levels in the same group (Fig. 7b). Interestingly, scRNA-seq analysis revealed that HPV-negative patients exhibited higher LRS scores (Supplementary Fig. 8a).
a Expression of AMIGO2, STC2, INHBA, DKK1, VEGFC, SPOCK1, MT1E, and FOXD1 in scRNA-seq UMAP plots. b Bar plot showing cell type proportions under LRS subtypes from scRNA-seq data. c AMIGO2, ranked highly for prognostic importance, was selected for further analysis. d AMIGO2 expression is upregulated in most tumor tissues across various cancers in TCGA pan-cancer dataset. e Evaluation of AMIGO2’s prognostic value (OS) in TCGA pan-cancer dataset. f Correlation analysis between AMIGO2 expression and representative signatures in TCGA pan-cancer dataset. Ns: no significant difference, *P < 0.05, **P ≤ 0.01, ***P ≤ 0.001.
Further analysis with the Ranger RFS and SWSFS algorithms ranked AMIGO2 as the most important factor affecting patient survival (Fig. 7c; Supplementary Fig. 8b). Given the limited research on AMIGO2 in tumors, we evaluated its expression across various cancers. Overall, AMIGO2 was overexpressed in multiple cancer types (Fig. 7d). Additionally, its expression was validated for prognostic value in OS in an independent TCGA pan-cancer cohort (Fig. 7e). High AMIGO2 levels were significantly associated with pathways such as EMT, TGF-β signaling, hypoxia, angiogenesis, WNT signaling, matrix remodeling, and cancer-associated fibroblasts (CAFs). Downregulation of AMIGO2 was associated with TCR, BCR, T cell co-stimulation, tumor-infiltrating lymphocytes (TILs), activated CD8 + T cells, and activated CD4 + T cells (Fig. 7f). More importantly, in HNSCC and other cancers, AMIGO2 upregulation was positively correlated with purine metabolism scores and purine biosynthesis scores (Supplementary Fig. 8c, d). Interestingly, as a key enzyme in purine metabolism, NT5E shows a significant positive correlation with AMIGO2 across the TCGA-HNSCC, merged-GEO, and ICGC-OSCC datasets (Supplementary Fig. 8e). In vitro, RNA-seq following pharmacologic inhibition of NT5E with the selective inhibitor AB680 revealed reduced AMIGO2 transcript abundance (Supplementary Fig. 8f). Consistently, AB680 treatment further decreased AMIGO2 mRNA and protein levels (Supplementary Fig. 8g, h). Together, these population-based correlations and perturbational experiments support a direct biological link between NT5E and AMIGO2. To further explore the potential relationship between AMIGO2 and immune suppression, we performed an integrated analysis using 75 immune-related genes. In the pan-cancer dataset, we detected significant negative correlations between AMIGO2 expression and the majority of immune gene expressions (Supplementary Fig. 8i). Based on these findings, we confirm that AMIGO2 plays a crucial role in remodeling the TME and driving HNSCC progression.
AMIGO2 involvement in HNSCC evolution
We constructed AMIGO2 knockdown models in Fadu and SCC-25 cells using shRNA (Supplementary Fig. 9a, b). Knockdown of AMIGO2 significantly reduced the expression of key proteins (PAICS and HPRT1) involved in purine metabolism (Fig. 8a). Furthermore, E-cadherin expression was upregulated in the shAMIGO2 group, while the expression of MMP9, N-cadherin, and Vimentin was downregulated (Fig. 8b), indicating that high expression of AMIGO2 promotes HNSCC EMT. Expectedly, shAMIGO2-transfected HNSCC cells exhibited significantly reduced migration and invasion (Fig. 8c; Supplementary Fig. 9c). Notably, co-culture experiments with MOC-1 and 3T3 cells showed that AMIGO2 deficiency led to a significant reduction in the expression of classic myofibroblast markers, including FAP, α-SMA, and Vimentin (Fig. 8d, Supplementary Fig. 9d). Furthermore, stratified differential expression in the TCGA-HNSCC cohort based on AMIGO2 levels, followed by pathway enrichment, revealed significant over-representation of the PI3K-AKT signaling pathway and Pathways in cancer (Supplementary Fig. 9e). Consistently, GSEA on TCGA-HNSCC data showed significant enrichment of the PI3K-AKT program in high-AMIGO2 samples (Supplementary Fig. 9f). Given the central role of PI3K-AKT in oncogenesis, we next examined nodal proteins after AMIGO2 perturbation. Mechanistic experiments indicated that AMIGO2 knockdown did not change PI3K phosphorylation, but selectively attenuated p-PDK1, p-AKT, and p-mTOR, implying that AMIGO2 regulates AKT-mTOR activation via PDK1 and acts downstream of-or independently from-PI3K (Supplementary Fig. 9g). To define the molecular basis, reciprocal co-immunoprecipitation in FaDu cells detected AMIGO2 in PDK1 immunoprecipitates and PDK1 in AMIGO2 immunoprecipitates (Supplementary Fig. 9h, i), demonstrating a direct AMIGO2-PDK1 interaction. Collectively, these findings indicate that AMIGO2 drives tumor cell behavior and TME remodeling primarily by tuning the PDK1-AKT-mTOR axis rather than broadly increasing PI3K activity.
a Western blot analysis showing downregulation of PAICS and HPRT1 proteins in shAMIGO2-treated HNSCC cells. b Western blot analysis showing alteration of EMT pathway proteins in shAMIGO2-treated HNSCC cells. c shAMIGO2 significantly inhibits cell migration and invasion in HNSCC cells. d Schematic diagram of the fibroblast co-culture model in vitro (left), Western blot analysis showing decreased expression of FAP, α-SMA, and Vimentin proteins in the shAmigo2 group (right). e Schematic illustration of establishing orthotopic tongue tumors by injecting MOC-1 cells. f Representative H&E-stained sections of orthotopic tongue tumors established by injecting MOC-1 cells into C57BL/6 mice. g Representative images of orthotopic tumors formed in the tongues of C57BL/6 mice following MOC-1 cell injection. Treatment groups included Vehicle, Anti-PD-1, shAmigo2, and shAmigo2+Anti-PD-1 combination. h Tumor growth curves of orthotopic tongue tumors with treatments initiated around day 3 post-implantation. i Flow cytometry analysis of NK cell infiltration, total CD8 + T cell proportion, PD-1 + CD8 + T cells, Tim3 + CD8 + T cells, GZMB + CD8 + T cells, and IFNγ + CD8 + T cells in tumors from the different treatment groups with representative images. *P < 0.05, **P ≤ 0.01, ***P ≤ 0.001.
To evaluate the impact of targeting AMIGO2 on the HNSCC TME in an anatomy-relevant setting, we established an orthotopic tongue implantation model (Fig. 8e, f). ShAMIGO2 significantly suppressed tumor growth, and its combination with anti-PD-1 further augmented the antitumor effect (Fig. 8g, h). Immune profiling showed that shAMIGO2 increased NK-cell and CD8 + T-cell infiltration and enhanced CD8 + T-cell effector function, while reducing the proportions of PD-1+ and Tim-3 + CD8 + T cells (Fig. 8i). As expected, in a subcutaneous tumor model shAMIGO2 alone inhibited tumor growth, whereas the addition of anti-PD-1 achieved superior tumor control and significantly extended OS to 80 days (Supplementary Fig. 10b–f). TME reprogramming mirrored that in the orthotopic setting (Supplementary Fig. 10g).
Given the unexpected effects of AMIGO2 in altering the HNSCC phenotype, we evaluated the expression of AMIGO2 in tissue samples from precancerous and HNSCC patients, along with its correlation to changes in the TME. IHC analysis showed that AMIGO2 expression increased as HNSCC progressed malignantly (Fig. 9a). Based on IHC results, we divided our tumor samples into High-AMIGO2 and Low-AMIGO2 groups. Quantitative analysis of multiplex immunofluorescence showed that AMIGO2 expression was positively correlated with the intra-tumoral myofibroblast marker FAP and negatively correlated with the expression levels of CD8 and CD56 (Fig. 9b). These results indicate that AMIGO2 levels in tumors are associated with the formation of an immune-suppressive microenvironment.
a IHC images and quantitative statistics of AMIGO2 expression in adjacent normal tissue, precancerous lesions, and tumor tissues. b Representative images and quantification of CD8, CD56, FAP, and PanCK staining in tumor tissues from high-AMIGO2 and low-AMIGO2 patient groups. c Graphical abstract for this study. Ns: no significant difference. *P < 0.05, **P ≤ 0.01, ***P ≤ 0.001.
Discussion
Tumors hijack nucleotide metabolism reprogramming to support their growth and evade immune surveillance. Targeting purine metabolism could be an effective strategy to restrain tumor progression31. Here, we utilized SM combined with WGCNA to analyze the spatial distribution of metabolic differences in HNSCC, leading to the identification of altered purine metabolism pathways. Notably, multimodal data analyses further revealed elevated purine metabolism in malignant cells and its potential association with tumor immune regulation. Previous research has shown that purine metabolites, particularly adenosine and inosine, dominate the metabolic profile in HNSCC32. By integrating bulk RNA-seq with scRNA-seq, we identified NT5E as a critical regulatory enzyme in purine metabolism in HNSCC. NT5E has been shown to promote HNSCC cell proliferation, migration, and the formation of invasive pseudopodia33. NT5E is also highly expressed in CAFs, where it enhances immune checkpoint activity through an A2B receptor-mediated positive feedback loop34. Furthermore, NT5E expression correlates with poor responses to cetuximab in HPV-negative, late-stage metastatic HNSCC patients35. To our knowledge, this study represents a pioneering effort in integrating multi-omics and purine metabolism to stratify prognosis and characterize the TME in HNSCC.
We constructed the LRS model using machine learning algorithms, enhancing the risk stratification of HNSCC patients. This model demonstrated robust predictive performance across different datasets and was further validated from in-house cohort using qRT-PCR. Notably, the predictive value of LRS surpassed that of other known HNSCC-related features. Compared to the low LRS subtype, the high LRS subtype exhibited a substantially lower copy number burden. Additionally, we found that the high LRS subtype was more likely to harbor TP53 mutations. Subsequent studies indicated that this subtype was significantly positively correlated with pathways such as hypoxia, angiogenesis, WNT signaling, and pro-tumor cytokines, which may explain its higher rates of recurrence and mortality. As mentioned earlier, under hypoxic conditions, HNSCC exosomes promote the activation of CAFs, which in turn facilitates HNSCC metastasis36. Similarly, VEGF37 and WNT signaling38 are known mediators of HNSCC progression and metastasis. In contrast, the low LRS subgroup was primarily enriched in immune-related pathways, including co-stimulation of T cells, TLS, TCR signaling, BCR signaling, and antigen processing and presentation, all of which are mature pathways that promote antitumor immune responses39,40,41,42. Furthermore, the low LRS patients displayed increased immune cell infiltration, upregulated effector cytokines, and improved immune recognition, suggesting these patients may exhibit enhanced responses to immune checkpoint blockade therapy. ScRNA-seq analysis and TIDE algorithm results further validated these findings. Moreover, our evidence indicates that patients with a low LRS score are more likely to benefit from treatments such as sorafenib, cisplatin, erlotinib, and 5-fluorouracil. The integrated analysis demonstrates that LRS is a superior biomarker for predicting clinical outcomes in HNSCC.
HPV infection is associated with HNSCC in the oropharynx, larynx, and oral cavity, with approximately 25%-30% of patients being HPV-positive, typically exhibiting better therapeutic responses and survival rates43. HPV encodes circE7, which suppresses Galectin-9 and impairs cytotoxic T cell function44, yet HPV-positive tumors often show stronger immune infiltration and improved responses to PD-1-directed immunotherapy, with E2 and E5 proteins serving as dominant CD8 + T cell targets45. Additionally, E6 protein suppresses SERPINB3, promoting cisplatin sensitivity46. Our study revealed distinct metabolic heterogeneity, with HPV-positive tumors exhibiting significantly lower LRS scores, which correlated with increased immune cell infiltration, suggesting that virus-driven tumors may foster a more permissive immune environment through reduced purine biosynthesis. These findings highlight that purine metabolism may represent a therapeutic target, where HPV typing could guide personalized strategies: enhancing immune-metabolic synergy in HPV-positive patients and targeting purine biosynthesis in HPV-negative tumors.
Compared with adjacent normal tissues, AMIGO2 is upregulated in tumor cells. Further analysis identified AMIGO2 as the most influential factor affecting patient survival. The human AMIGO2 gene encodes a type I transmembrane protein originally described in neuronal processes, comprising six leucine-rich repeats and an extracellular immunoglobulin-like domain47. AMIGO2 promotes endothelial adhesion to diverse cancer cells, thereby facilitating liver metastasis48, and its expression has been proposed as a predictor of cervical cancer recurrence49. Prior studies report that AMIGO2 binds the PDK1 PH domain to promote AKT Thr308 phosphorylation and angiogenesis50, and that in lung cancer AMIGO2 is associated with enhancement of the PDK1/AKT axis, increased invasion, and chemoresistance51. Consistent with these reports, our results demonstrate a regulatory relationship between AMIGO2 and the PDK1-AKT-mTOR signaling. Moreover, the selective NT5E inhibitor AB680 modulates AMIGO2 expression, supporting a model in which the NT5E-metabolic axis establishes an immunosuppressive TME that is reinforced by AMIGO2-driven amplification of intracellular signaling.
The function of AMIGO2 is complex and multifaceted. In our study, we found that AMIGO2 regulates EMT and purine metabolism in HNSCC, and that its knockdown significantly reduced tumor cell migration and invasion. Co-culture experiments demonstrated that inhibiting AMIGO2 in tumor cells reduced the expression of classic myofibroblast markers, including FAP, α-SMA, and Vimentin. A substantial body of evidence suggests that CAFs are critical components of the TME and can promote cancer progression or impair immune therapy sensitivity across various malignancies. Our in vivo experiments confirmed that AMIGO2 promotes tumor growth in xenograft models, leading to NK and CD8 + T cell rejection and exhaustion. In clinical HNSCC samples, AMIGO2 expression correlated with malignant transformation, accompanied by low levels of NK and CD8 + T cell. A previous study reported that exosomes derived from HNSCC exhibit high purine content, which can drive tumor immune evasion52. Therefore, AMIGO2 may represent an alternative pathway through which purine metabolism mediates immune microenvironment remodeling.
Our study has limitations. First, the omics cohorts were of modest size, which may introduce inter-individual variability; multicenter validation with larger cohorts is warranted. Second, incomplete clinical and molecular annotations in public datasets may obscure associations between the LRS and specific clinical traits. Third, although we place AMIGO2 within a membrane-proximal PDK1-AKT-mTOR axis, the upstream-downstream links by which AMIGO2 drives remodeling of the metabolic microenvironment and influences immunotherapy response require deeper mechanistic dissection. Finally, our spatial omics measurements are spot-level, so integrating single-cell-resolution spatial technologies will be necessary to resolve cell-intrinsic metabolic programs and in situ cell-cell interactions within the TME.
In conclusion, we established the metabolomics-based LRS model for HNSCC patients through comprehensive data mining and investigated the therapeutic potential of AMIGO2 (Fig. 9c). These findings validate the hypothesis that tumor metabolic profiling, based on expression signatures, can be applied in clinical practice to determine tumor microenvironment phenotypes and guide treatment strategies.
Methods
Collection of human clinical samples
Samples from treatment-naïve HNSCC patients, premalignant patients (including cases of oral leukoplakia and lichen planus with or without dysplasia), and tumor-matched adjacent normal tissues were collected with approval from the Research Ethics Committee of the First Affiliated Hospital of the University of Science and Technology of China (Registration number: 2023KY222). This study explicitly affirms its full compliance with the ethical principles outlined in the Declaration of Helsinki, thereby ensuring the standards of ethical conduct in research involving human participants. All samples were double-confirmed by the Department of Pathology. Written informed consent was obtained from all patients prior to tissue collection. Detailed patient information is provided in Supplementary Data 5. For public datasets, the same stringent manual review criteria were applied to head and neck samples, and only those annotated as squamous cell carcinoma or healthy oral mucosa were retained (Supplementary Data 6).
Spatial metabolomics (SM) detection
The embedded samples were stored at −80 °C before being sectioned. The samples were cut into consecutive sagittal slices 10 μm about 10 slices by a cryostat microtome (Leica CM 1950, Leica Microsystem, Germany) and were thaw-mounted on positive charge desorption plate (Thermo Scientific, U.S.A). Sections were stored at −80 °C before further analysis. They were desiccated at −20 °C for 1 hour and then at room temperature for 2 hours before mass spectrometry imaging (MSI) analysis. Meanwhile, an adjacent slice was left for hematoxylin-eosin (H&E) staining.
In brief, this experiment was carried out with an AFADESI-MSI platform (Beijing Victor Technology Co., LTD, Beijing, China) in tandem with a Q-Orbitrap mass spectrometer (MS, Q Exactive, Thermo Scientific, U.S.A.). Here, the solvent formula was acetonitrile (ACN) /H2O (8:2) at negative mode and ACN/H2O (8:2) at positive mode and the solvent flow rate was 1.5 μL/min, the transporting gas flow rate was 45 L/min, the spray voltage was set at 7 kV, and the distance between the sample surface and the sprayer was 3 mm as was the distance from the sprayer to the ion transporting tube. The MS resolution was set at 20,000, the mass range was 70-1200 Da, the automated gain control target was 2E6, the maximum injection time was set to 200 ms, the S-lens voltage was 55 V, and the capillary temperature was 350 °C . The MSI experiment was carried out with a constant rate of 0.2 mm/s continuously scanning the surface of the sample section in the x direction and a 40 μm vertical step in the y direction.
SM data analysis
The raw data (.raw files) were first converted to imzML format using imzMLConverter, then imported into MSiReader (an open-access MSI software based on the Matlab platform) for ion image reconstruction. Background subtraction was performed using the Cardinal software package. All MS images were normalized by total ion count in each pixel. Region-specific MS profiles were extracted by aligning high-spatial resolution H&E images. The ions detected by AFADESI were annotated by the pySM pipeline and an in-house SmetDB database (Lumingbio, Shanghai, China). Raw metabolomic data detected by AFADESI-MSI have been uploaded to the metaspace platform (https://metaspace2020.eu/project/HNSCC-AMIGO2).
RNA sequencing (RNA-seq) data acquisition and processing
Total RNA was isolated from cultured cells using TRIzol. After quality control, ribosomal RNA (rRNA) was removed from total RNA using the Ribo Zero Gold Kit. Libraries were then constructed and sequenced on the Illumina HiSeq XTen platform. The sequencing data were aligned to the human reference genome (GRCh38), and differential expression analysis between groups was performed using DESeq2 when the criteria fold change > 1.5 and P < 0.05 were met. Other detailed processing and analysis methods are referenced from our previous study53. The processed RNA-seq data supporting findings of this study is displayed in Supplementary Data 7.
Public bulk data collection and analysis
Patient genomic sequencing, RNA sequencing, and clinical data were downloaded from public repositories, including the GEO, The International Cancer Genome Consortium (ICGC) (https://dcc.icgc.org/), and The Cancer Genome Atlas (TCGA) (https://www.cancer.gov/tcga/). Based on previous studies, patient subtypes and tumor-related scores were organized and retrieved53,54. This study involves five published HNSCC datasets, including TCGA-HNSCC, ICGC, GSE65858, GSE41613, and GSE42743 cohorts. For transcriptomic data, we processed and merged the datasets as previously reported53, consolidating GSE65858, GSE41613, and GSE42743 into a merged GEO cohort (merge-GEO), and integrating the TCGA-HNSCC dataset into a merged HNSCC cohort (merge-HNSCC). Additionally, 32 pan-cancer datasets were downloaded from the UCSC Xena database (https://xenabrowser.net/datapages/) (Supplementary Data 6 and 8).
Public single-cell RNA-seq (scRNA-seq) datasets collection and analysis
In order to explore cellular heterogeneity in the HNSCC microenvironment, we performed an integrative analysis of four publicly available HNSCC scRNA-seq datasets55,56,57,58. A detailed report on the individual donors and specimens including clinical characteristics is provided in Supplementary Data 6 and 9. The raw count matrices from each dataset were merged using Scanpy (V1.9.1) for subsequent quality control analysis. For gene filtering, we removed genes that were expressed in fewer than 50 cells or associated with noise and dissociation artifacts. Regarding cell filtering, we applied the following criteria: (1) expressed genes per cell ranged from 500 to 6000, (2) mitochondrial RNA content was below 20%, and (3) the total counts per cell did not exceed 40,000. Potential doublets were identified and eliminated using DoubletDetection (https://doi.org/10.5281/zenodo.2678041) with default settings. Following these steps, 348,668 high-quality cells were retained for downstream analysis.
We then normalized the gene expression matrix by scaling the total UMI count of each cell to 10,000, followed by logarithmic transformation. Principal component analysis (PCA) was conducted using the top 3000 variable genes, and batch effects across different samples and datasets were corrected using the harmony algorithm (sigma = 0.07, lambda = 1.0, theta = 0.5). The first 50 principal components (PCs) were used to construct a neighborhood graph and generate Uniform Manifold Approximation and Projection (UMAP) visualizations. In the first round of clustering, we identified major cell types-including T cells, B cells, NK cells, plasma cells, monocytes, macrophages, dendritic cells (DC), mast cells, fibroblasts, smooth muscle cells (SMC), epithelial cells, salivary gland epithelium (SG Epithelium), endothelial cells, and lymphatic endothelial cells (LEC)-using Louvain clustering at a resolution of 0.3 based on canonical markers. In the second round of clustering, we further characterized subpopulations within each cell type according to differentially expressed genes.
Public spatial transcriptomics (ST) datasets collection and analysis
To investigate the spatial distribution of purine metabolism activity, we collected 21 10x Visium ST slices from three public datasetss59,60,61 (Supplementary Data 6). The routine analysis pipeline of Scanpy with default parameters was used to preprocess the ST data. The purine metabolism pathway score was calculated using “score_genes” function with default parameters.
Consensus clustering and machine learning
Based on the expression of ligand-receptor genes associated with the prognostic value of NT5E, consensus clustering using a resampling method was performed on the merged HNSCC cohort. This process was carried out using the ConsensusClusterPlus package. Subsequently, the optimal number of clusters was determined by integrating consensus score matrices, CDF curves, PCA scores, and Nbclust, in a process that involved 1000 iterations to ensure the stability of the classification.
As previously described62, we integrated 10 machine learning algorithms and 101 algorithm combinations to develop a ligand-receptor-based signature (LRS) with high accuracy and stability for HNSCC. Validation was performed in other datasets, and the Harrell concordance index (C-index) was calculated for all validation datasets. The model with the highest average C-index was considered the best for subsequent analysis.
Ranger, a weighted version of the random forest algorithm, was used to evaluate the impact of selected gene expression on patient OS, and the variable importance score (VIS) for each gene was provided using the R package “ranger”. Sliding window sequential forward feature selection (SWSFS) was used to identify the most important genes, with the methods described earlier54, selecting a gene set with the lowest out-of-bag (OOB) error as candidate prognostic factors for further analysis.
Weighted gene co-expression network analysis (WGCNA) construction, pathway annotation, ssGSEA, and gene set variation analysis (GSVA)
For the SM data of HNSCC samples, a weighted co-expression network was constructed using the WGCNA R package to identify metabolic modules and assess the relative importance of individual metabolites. We selected hub metabolites in the module as candidate metabolites for further analysis. Detailed procedures are as previously described63. Pathway analysis was performed on metabolites clustered in the same module using the MetaboAnalyst platform (https://www.metaboanalyst.ca/), and gene set enrichment analysis was conducted using the Metascape platform (https://metascape.org/gp/index.html), followed by visualization.
Gene set enrichment analysis (GSEA) was performed using the “ClusterProfiler” R package. To quantify the activity of gene sets, we referred to previous studies using GSVA and ssGSEA methods54 to calculate the differences in signaling pathway activity between groups. Gene sets were sourced from the MSigDB database (https://www.gsea-msigdb.org/gsea/msigdb/index.jsp) and previous research54,59.
LRS subtype analysis
We performed Kaplan-Meier survival analysis using the “survminer” R package to determine the survival outcomes of high and low LRS score subtypes. To assess the significance of the LRS score in relation to clinical features of patients, univariate and multivariate analyses were conducted to establish a Cox proportional hazards regression model. Chi-square tests were performed to evaluate the differences in the number of mutated genes between LRS subtypes, and GISTIC 2.0 (https://cloud.genepattern.org) was used to identify copy number alterations (CNAs), categorizing genes with amplifications or deletions. The total number of genes with arms or focal regions deleted or amplified was calculated as the burden of copy number loss or gain.
As previously described, the Tumor Immune Dysfunction and Exclusion (TIDE) algorithm was used to predict response to immune checkpoint inhibitor treatment64. To predict chemotherapy responses in different LRS subtypes, we evaluated commonly used HNSCC chemotherapeutic agents using the Genomics of Drug Sensitivity in Cancer (GDSC) database65. The half-maximal inhibitory concentration (IC50) values were estimated using the pRRophetic algorithm. In addition, candidate compounds targeting LRS subtypes were identified, and potential therapeutic strategies for HNSCC patients were proposed based on these findings.
Cell lines and culture conditions
The human SCC-25, Fadu, and mouse embryonic fibroblast 3T3-L1 cell lines were purchased from Procell Life Science & Technology Co., Ltd., and the MOC-1 cell line was obtained from Meisen Chinese Tissue Culture Collections. All cell lines were authenticated by STR profiling and tested negative for mycoplasma contamination. All cell lines used in this study were cultured in DMEM (Gibco) supplemented with 10% FBS (PAN) and 5% (v/v) CO2 in a 37°C incubator.
For AMIGO2 (human) shRNA and Amigo2 (mouse) shRNA, lentiviral vectors (HanBio, Shanghai, China) were transfected into SCC-25, Fadu, and MOC-1 cells (Supplementary Data 10). As a control, lentiviral vectors containing short hairpin RNA sequences targeting non-mammalian genes were used. After 48 hours of transduction, puromycin selection was applied, and cells were cultured.
In the co-culture experiment, 3T3-L1 cells were seeded at a density of 2 × 105 cells per well in 6-well plates with DMEM (containing 10% FBS), and then cultured with MOC-1 cells in Transwell inserts with 0.4 μm pores for 72 hours. After co-cultivation, the cells were collected for subsequent experiments.
Western blotting and IHC
Western blotting and IHC staining were performed using methods similar to those described in previous studies66. The following primary antibodies were used for Western blotting: AMIGO2 (1:500, HUAAN, catalog No: ER1903-66), GAPDH (1:1000, CST, catalog No: 5174S, RRID: AB_10622025), E-cadherin (1:1000, CST, catalog No: 3195S, RRID: AB_2291471), N-cadherin (1:1000, CST, catalog No: 13116S, RRID: AB_2687616), HPRT1 (1:1000, Proteintech Cat# 15059-1-AP, RRID: AB_10638622), PAICS (1:1000, Proteintech Cat# 12967-1-AP, RRID: AB_10638449), Vimentin (1:1000, CST, catalog No: 5741S, RRID: AB_10695459), MMP9 (1:1000, CST, catalog No: 13667S, RRID: AB_2798289), FAP (Thermo Fisher Scientific Cat# PA5-99313, RRID: AB_2818246), α-SMA (Cell Signaling Technology Cat# 19245, RRID: AB_2734735), PI3K (Cell Signaling Technology Cat# 4292, RRID:AB_329869), Phospho-PI3K (Cell Signaling Technology Cat# 17366, RRID:AB_2895293), AKT (Cell Signaling Technology Cat# 9272, RRID: AB_329827), Phospho-Akt (Thr308) (Cell Signaling Technology Cat# 9275, RRID: AB_329828), Phospho-Akt (Ser473) (Cell Signaling Technology Cat# 9271, RRID: AB_329825), PDK1 (Thermo Fisher Scientific Cat# A302-130A, RRID:AB_1720395), Phospho-PDK1 (Ser241) (Cell Signaling Technology Cat# 3061, RRID: AB_2161919), mTOR (Abcam Cat# ab134903, RRID: AB_2800465) and mTOR (phospho S2481) (Abcam Cat# ab137133, RRID:AB_2800466).
For IHC staining, AMIGO2 (1:200, HUAAN, catalog No: ER1903-66), HPRT1 (1:500, Proteintech Cat# 15059-1-AP, RRID: AB_10638622), and PAICS (1:50, Proteintech Cat# 12967-1-AP, RRID: AB_10638449) were used as primary antibodies. IHC scoring was performed by two independent pathologists who evaluated each sample according to a standardized scoring system and scale. In cases of discrepancy, the average score was taken. The semi-quantitative evaluation was based on staining intensity (weak: 1; moderate: 2; strong: 3) and the abundance of positive cells (≤5%: 0; 6%–25%: 1; 26%–50%: 2; 51%–75%: 3; ≥76%: 4). The final staining score was obtained by multiplying the staining intensity with the abundance of positive cells, yielding the final data67.
RNA extraction, RT-PCR, and qRT-PCR
Total RNA was extracted from cultured cells and tissues using TRIzol reagent. cDNA was synthesized using a reverse transcription kit, and the cDNA was used as a template for detecting gene expression levels by SYBR Green. β-actin was used as an internal control. The primers used for qRT-PCR are listed in Supplementary Data 10.
Cell invasion and migration assays
For the invasion assay, Matrigel was diluted with serum-free medium and used to coat the Transwell chambers. Cells (1×105 cells per well) were seeded into the upper chambers, which were then transferred to 24-well plates containing the lower chambers filled with complete medium (10% FBS) and cultured for 48 hours. The number of cells that invaded into the lower chamber was counted. For the migration assay, no Matrigel was used for coating, and cells were cultured for 24 hours. Other steps were the same as for the invasion assay.
In vivo tumor implantation and experimentation
For in vivo tumor growth studies, approximately 1×106 MOC-1 cells (sh-NC and sh-AMIGO2) were injected into the axillary region of 5–6-week-old male C57BL/6 mice. Treatment was then performed with anti-PD-1 (BioXCell, Cat#BE0273, clone29F.1A12) or isotype control anti-IgG (BioXCell, Cat#BE0089, clone2A3) antibodies. The mice were randomly assigned to 4 groups (8 mice per group): Vehicle group, Anti-PD-1 treatment group (10 mg/kg), shAMIGO2 group, and shAMIGO2+Anti-PD-1 combination treatment group. Tumor volume (TV) was measured at regular intervals, and calculated using the formula: TV was calculated using the formula = 0.5 × L × W2, where L is the length and W is the width of the tumor. Three mice per group were kept for survival analysis. In addition, to inhibit NT5E enzymatic activity, AB680 (MedChemExpress) was administered intraperitoneally at a dose of 20 mg/kg every four days, starting four days after tumor induction and continuing until the end of the experiment. After 27 days, the mice were euthanized, and subcutaneous tumors were collected for photography, weighing, and fresh tissue analysis for flow cytometry.
Orthotopic tongue implantation was performed as previously described68,69. Briefly, an orthotopic syngeneic oral carcinoma model was established by submucosal injection of 7.5×105 MOC-1 cells into the dorsal tongue of C57BL/6 mice. A 1 mL syringe was used to slowly deliver the cell suspension into the right lateral tongue margin, approximately 1.5 mm from the tip, resulting in the formation of a visible bulge at the injection site. Tumor growth was subsequently monitored and quantified at regular intervals using caliper measurements. All animal experiments were approved by the Animal Care and Use Committee of the First Affiliated Hospital of the University of Science and Technology of China, and were conducted in accordance with institutional guidelines and ethical standards (Registration number: 2024-N(A)-0244).
Tumor-infiltrating immune cells flow cytometry analysis
Tumor tissue was dissociated using a tumor dissociation kit, grinding and digesting mouse subcutaneous tumor-bearing tissue. The resulting cell suspension was filtered through a 100 µm cell strainer, and tumor cell counts were performed. The suspension was washed once with PBS containing 10% FBS and further stained with surface antibodies: Anti-CD3e-Brilliant Violet 421™ (BioLegend, 145-2C11, Cat. 100336), Anti-CD8a-APC (BioLegend, 53-6.7, Cat. 100712), Anti-CD4-Brilliant Violet 510™ (BD Biosciences, RM4-5, Cat. 563106), Anti-TIM-3-PE/Cyanine7 (BioLegend, RMT3-23, Cat. 119716), Anti-CD45-APC/Fire™ 750 (BioLegend, 30-F11, Cat. 103154), Anti-PD-1-PE/Dazzle™ 594 (BioLegend, RMP1-30, Cat. 135228), and Anti-NK1.1-PE (BD Biosciences, PK136 (RUO), Cat. 557391). After staining, the cells were fixed and permeabilized using cell permeabilization reagent (BioLegend, Cat. 420801) and fixation buffer (BioLegend, Cat. 421002). Subsequently, intracellular staining was performed with Anti-IFN-γ-PerCP/Cyanine5.5 (BioLegend, XMG1.2, Cat. 505822) and Anti-Granzyme B Recombinant-FITC (BioLegend, QA16A02, Cat. 372206). In addition, we used the related Isotype control antibodies: PE/Cyanine7 Rat lgG2a, k lsotype Ctrl (BioLegend, RTK2758, Cat. 400521), PE/Dazzlem 594 Rat IgG2a, k Isotype Ctrl (BioLegend, RTK2758, Cat. 400557), PerCP/Cyanine5.5 Rat IgG1, k Isotype Ctrl (BioLegend, RTK2071, Cat. 400425) and FITC Mouse IgGl, k Isotype Ctrl (ICFC) (BioLegend, MOPC-21, Cat. 400137). Flow cytometry was performed on a Beckman Navios flow cytometer, and data were analyzed using FlowJo 10.8.1.
Multiplex immunofluorescence
Paraffin-embedded and formalin-fixed tissue samples were sectioned into 4 µm slices. Deparaffinization and antigen retrieval were performed to expose antigens in the tissue slices. Specific antibodies against CD8 (1:800, BIOSS, catalog No: bsm-60735R), FAP (1:800, HUABio, catalog No: ET1704-23), CD56 (1:400, HUABio, catalog No: HA601126), and PANCK (1:2000, HUABio, catalog No: HA601138) were used to bind to the corresponding proteins in the tissue. After secondary antibody binding, fluorochrome-conjugated secondary antibodies were applied. The process was repeated multiple times, followed by DAPI nuclear counterstaining to complete the staining procedure. Stained slides were mounted with anti-fade mounting medium and analyzed using a fluorescence scanning imaging system and HALO analysis software to observe the expression of CD8, FAP, CD56, and PANCK proteins in the samples. Positive cell density was calculated based on protein markers in single cells.
Co-immunoprecipitation (Co-IP)
To assess the endogenous interaction between AMIGO2 and PDK1, FaDu cells were cultured to 80-90% confluence and rinsed with ice-cold PBS. Cells were lysed in NP-40 lysis buffer (50 mM Tris-HCl, pH 7.5; 150 mM NaCl; 1% NP-40; 1 mM EDTA; supplemented with protease and phosphatase inhibitors) for 30 min at 4°C. Lysates were cleared by centrifugation at 12,000 × g for 15 min, and protein concentration was determined using a BCA assay. For each immunoprecipitation, 1 mg of total protein was precleared with Protein A/G magnetic beads for 1 h at 4°C. The precleared lysates were then incubated overnight at 4°C with either anti-AMIGO2 antibody (Santa Cruz Biotechnology, Cat# sc-373699, RRID: AB_10920216) or anti-PDK1 antibody (Thermo Fisher Scientific Cat# A302-130A, RRID:AB_1720395). Equal amounts of isotype-matched IgG served as negative controls, and 10% of the lysates were retained as input controls. Immune complexes were captured by adding 20–40 μL of Protein A/G magnetic beads and incubating for an additional 1–2 h at 4°C. The beads were washed four to five times with lysis buffer, and bound proteins were eluted by boiling in 2×Laemmli sample buffer at 95°C for 5 min. The eluates were resolved by SDS-PAGE, followed by Western blotting and detection using enhanced chemiluminescence (ECL).
Statistical analysis
For comparisons between two groups, normally distributed variables and non-normally distributed variables were assessed using unpaired Student’s t-test and Mann–Whitney U test, respectively. For comparisons involving more than two groups, one-way analysis of variance (ANOVA) or the Kruskal-Wallis test was used. Pearson or Spearman correlation analysis was used to assess the linear relationship between two groups. All statistical analyses were performed using R software (V 4.4.1). Each experiment was repeated three or more times, and data are presented as mean ± standard deviation (SD). Statistical significance is indicated as follows: ns, not significant; *P < 0.05; **P ≤ 0.01; ***P ≤ 0.001.
Data availability
All data are available in the main text or in the Supplementary Data. The generated spatial metabolomics dataset is available in the metaspace platform query (https://metaspace2020.eu/project/HNSCC-AMIGO2).
Code availability
No new algorithms were developed for this manuscript. All software tools and computational methods used in this study are described in detail in the Methods section. The custom analysis codes generated during this study are available from the corresponding author upon reasonable request.
References
Shi, Y. et al. Finotonlimab with chemotherapy in recurrent or metastatic head and neck cancer: a randomized phase 3 trial. Nat. Med. 30, 2568–2575 (2024).
Van den Bossche, V. et al. PPARalpha-mediated lipid metabolism reprogramming supports anti-EGFR therapy resistance in head and neck squamous cell carcinoma. Nat. Commun. 16, 1237 (2025).
Cramer, J. D., Burtness, B., Le, Q. T. & Ferris, R. L. The changing therapeutic landscape of head and neck cancer. Nat. Rev. Clin. Oncol. 16, 669–683 (2019).
Li, K. et al. Single cell analysis unveils B cell-dominated immune subtypes in HNSCC for enhanced prognostic and therapeutic stratification. Int J. Oral. Sci. 16, 29 (2024).
Li, J., Luo, Z., Jiang, S. & Li, J. Advancements in neoadjuvant immune checkpoint inhibitor therapy for locally advanced head and neck squamous carcinoma: a narrative review. Int Immunopharmacol. 134, 112200 (2024).
De Martino, M., Rathmell, J. C., Galluzzi, L. & Vanpouille-Box, C. Author Correction: cancer cell metabolism and antitumour immunity. Nat. Rev. Immunol. 24, 537 (2024).
Zhao, X., Guo, B., Sun, W., Yu, J. & Cui, L. Targeting squalene epoxidase confers metabolic vulnerability and overcomes chemoresistance in HNSCC. Adv. Sci. 10, e2206878 (2023).
Gao, L. et al. Mediation of PKM2-dependent glycolytic and non-glycolytic pathways by ENO2 in head and neck cancer development. J. Exp. Clin. Cancer Res. 42, 1 (2023).
Wang, L. et al. Integrative single-cell and bulk transcriptomes analyses reveals heterogeneity of serine-glycine-one-carbon metabolism with distinct prognoses and therapeutic vulnerabilities in HNSCC. Int. J. Oral. Sci. 16, 44 (2024).
Zhi, Y. et al. Spatial Transcriptomic and metabolomic landscapes of oral submucous fibrosis-derived oral squamous cell carcinoma and its tumor microenvironment. Adv. Sci. 11, e2306515 (2024).
Guan, J. et al. Cellular hierarchy framework based on single-cell/multi-patient sample sequencing reveals metabolic biomarker PYGL as a therapeutic target for HNSCC. J. Exp. Clin. Cancer Res. 42, 162 (2023).
Eldridge, R. C. et al. Plasma metabolic phenotypes of HPV-associated versus smoking-associated head and neck cancer and patient survival. Cancer Epidemiol. Biomark. Prev. 30, 1858–1866 (2021).
Ravi, V. M. et al. Spatially resolved multi-omics deciphers bidirectional tumor-host interdependence in glioblastoma. Cancer Cell 40, 639–655.e13 (2022).
Dang, Q. et al. Cancer immunometabolism: advent, challenges, and perspective. Mol. Cancer 23, 72 (2024).
Chuang, Y. M., Tzeng, S. F., Ho, P. C. & Tsai, C. H. Immunosurveillance encounters cancer metabolism. EMBO Rep. 25, 471–488 (2024).
Zhang, X. Y. et al. Metabolic landscape of head and neck squamous cell carcinoma informs a novel kynurenine/Siglec-15 axis in immune escape. Cancer Commun. 44, 670–694 (2024).
Chen, F. et al. HSP90 inhibition suppresses tumor glycolytic flux to potentiate the therapeutic efficacy of radiotherapy for head and neck cancer. Sci. Adv. 10, eadk3663 (2024).
Langfelder, P. & Horvath, S. WGCNA: an R package for weighted correlation network analysis. BMC Bioinformatics 9, 559 (2008).
Liu, Z. L. et al. Single cell deciphering of progression trajectories of the tumor ecosystem in head and neck cancer. Nat. Commun. 15, 2595 (2024).
Wang, Q., Zhao, Y., Tan, G. & Ai, J. Single cell analysis revealed SFRP2 cancer associated fibroblasts drive tumorigenesis in head and neck squamous cell carcinoma. NPJ Precis. Oncol. 8, 228 (2024).
Lee, C. M. et al. BRCA1 mutation promotes sprouting angiogenesis in inflammatory cancer-associated fibroblast of triple-negative breast cancer. Cell Death Discov. 10, 5 (2024).
You, E. et al. Disruption of cellular plasticity by repeat RNAs in human pancreatic cancer. Cell 187, 7232–7247.e23 (2024).
Bai, Q. et al. Single-cell landscape of immune cells in human livers affected by HBV-related cirrhosis. JHEP Rep. 5, 100883 (2023).
Chen, Q. et al. Tumor microenvironment responsive CD8(+) T cells and myeloid-derived suppressor cells to trigger CD73 inhibitor AB680-based synergistic therapy for pancreatic cancer. Adv. Sci. 10, e2302498 (2023).
Piovesan, D. et al. Targeting CD73 with AB680 (Quemliclustat), a novel and potent small-molecule CD73 inhibitor, restores immune functionality and facilitates antitumor immunity. Mol. Cancer Ther. 21, 948–959 (2022).
Bi, C., Patel, J. S. & Liang, S. H. Development of CD73 inhibitors in tumor immunotherapy and opportunities in imaging and combination therapy. J. Med. Chem. 68, 6860–6869 (2025).
Lawson, K. V. et al. Discovery of AB680: a potent and selective inhibitor of CD73. J. Med. Chem. 63, 11448–11468 (2020).
Su, J. et al. Cell-cell communication: new insights and clinical implications. Sig. Transduct. Target Ther. 9, 196 (2024).
Yan, P. et al. Midkine as a driver of age-related changes and increase in mammary tumorigenesis. Cancer Cell 42, 1936–1954.e9 (2024).
Xu, Q. et al. Multiple cancer cell types release LIF and Gal3 to hijack neural signals. Cell Res. 34, 345–354 (2024).
Tran, D. H. et al. De novo and salvage purine synthesis pathways across tissues and tumors. Cell 187, 3602–3618.e20 (2024).
Li, W. et al. Analysis of metabolomics and transcriptomics data in head and neck squamous cell carcinoma. Oral. Dis. 29, 1464–1479 (2023).
Xue, F. et al. CD73 facilitates invadopodia formation and boosts malignancy of head and neck squamous cell carcinoma via the MAPK signaling pathway. Cancer Sci. 113, 2704–2715 (2022).
Yu, M. et al. CD73 on cancer-associated fibroblasts enhanced by the A(2B)-mediated feedforward circuit enforces an immune checkpoint. Nat. Commun. 11, 515 (2020).
Shi, E. et al. 5’-Ectonucleotidase CD73/NT5E supports EGFR-mediated invasion of HPV-negative head and neck carcinoma cells. J. Biomed. Sci. 30, 72 (2023).
Duan, Y. et al. Hypoxia-induced miR-5100 promotes exosome-mediated activation of cancer-associated fibroblasts and metastasis of head and neck squamous cell carcinoma. Cell Death Dis. 15, 215 (2024).
Hefni, E. et al. Angiopoietin-like 4 induces head and neck squamous cell carcinoma cell migration through the NRP1/ABL1/PXN pathway. Cell Signal. 108, 110697 (2023).
Huang, Q. et al. WNT7A promotes tumorigenesis of head and neck squamous cell carcinoma via activating FZD7/JAK1/STAT3 signaling. Int. J. Oral. Sci. 16, 7 (2024).
Lagos, J. et al. B cell adapter for PI 3-kinase (BCAP) coordinates antigen internalization and trafficking through the B cell receptor. Sci. Adv. 10, eadp1747 (2024).
Neve-Oz, Y., Sherman, E. & Raveh, B. Bayesian metamodeling of early T-cell antigen receptor signaling accounts for its nanoscale activation patterns. Front. Immunol. 15, 1412221 (2024).
Tokita, S., Kanaseki, T. & Torigoe, T. Neoantigen prioritization based on antigen processing and presentation. Front. Immunol. 15, 1487378 (2024).
Tooley, K. et al. Pan-cancer mapping of single CD8(+) T cell profiles reveals a TCF1:CXCR6 axis regulating CD28 co-stimulation and anti-tumor immunity. Cell Rep. Med. 5, 101640 (2024).
Feng, B. et al. Ligand-receptor interactions combined with histopathology for improved prognostic modeling in HPV-negative head and neck squamous cell carcinoma. NPJ Precis. Oncol. 9, 57 (2025).
Ge, J. et al. Human papillomavirus-encoded circular RNA circE7 promotes immune evasion in head and neck squamous cell carcinoma. Nat. Commun. 15, 8609 (2024).
Eberhardt, C. S. et al. Functional HPV-specific PD-1(+) stem-like CD8 T cells in head and neck cancer. Nature 597, 279–284 (2021).
Huang, Z. et al. HPV enhances HNSCC chemosensitization by inhibiting SERPINB3 expression to disrupt the fanconi anemia pathway. Adv. Sci. 10, e2202437 (2022).
Kuja-Panula, J., Kiiltomaki, M., Yamashiro, T., Rouhiainen, A. & Rauvala, H. AMIGO, a transmembrane protein implicated in axon tract development, defines a novel protein family with leucine-rich repeats. J. Cell Biol. 160, 963–973 (2003).
Izutsu, R. et al. AMIGO2 contained in cancer cell-derived extracellular vesicles enhances the adhesion of liver endothelial cells to cancer cells. Sci. Rep. 12, 792 (2022).
Iida, Y. et al. AMIGO2 expression as a predictor of recurrence in cervical cancer with intermediate risk. Mol. Clin. Oncol. 19, 56 (2023).
Park, H. et al. AMIGO2, a novel membrane anchor of PDK1, controls cell survival and angiogenesis via Akt activation. J. Cell Biol. 211, 619–637 (2015).
Chen, L. K. et al. AMIGO2 attenuates innate cisplatin sensitivity by suppression of GSDME-conferred pyroptosis in non-small cell lung cancer. J. Cell Mol. Med. 27, 2412–2423 (2023).
Ludwig, N., Gillespie, D. G., Reichert, T. E., Jackson, E. K. & Whiteside, T. L. Purine metabolites in tumor-derived exosomes may facilitate immune escape of head and neck squamous cell carcinoma. Cancers 12, 1602 (2020).
Yang, Z. et al. Histone lysine methyltransferase SMYD3 promotes oral squamous cell carcinoma tumorigenesis via H3K4me3-mediated HMGA2 transcription. Clin. Epigenet. 15, 92 (2023).
Yang, Z. et al. YKT6, as a potential predictor of prognosis and immunotherapy response for oral squamous cell carcinoma, is related to cell invasion, metastasis, and CD8+ T cell infiltration. Oncoimmunology 10, 1938890 (2021).
Bill, R. et al. CXCL9:SPP1 macrophage polarity identifies a network of cellular programs that control human cancers. Science 381, 515–524 (2023).
Kurten, C. H. L. et al. Investigating immune and non-immune cell interactions in head and neck tumors by single-cell RNA sequencing. Nat. Commun. 12, 7338 (2021).
Puram, S. V. et al. Cellular states are coupled to genomic and viral heterogeneity in HPV-related oropharyngeal carcinoma. Nat. Genet. 55, 640–650 (2023).
Williams, D. W. et al. Human oral mucosa cell atlas reveals a stromal-neutrophil axis regulating tissue immunity. Cell 184, 4090–4104.e15 (2021).
Arora, R. et al. Spatial transcriptomics reveals distinct and conserved tumor core and edge architectures that predict survival and targeted therapy response. Nat. Commun. 14, 5029 (2023).
Galeano Nino, J. L. et al. Effect of the intratumoral microbiota on spatial and cellular heterogeneity in cancer. Nature 611, 810–817 (2022).
Cheng, H. Y. et al. Snail-regulated exosomal microRNA-21 suppresses NLRP3 inflammasome activity to enhance cisplatin resistance. J. Immunother. Cancer 10, e004832 (2022).
Liu, Z. et al. Machine learning-based integration develops an immune-derived lncRNA signature for improving outcomes in colorectal cancer. Nat. Commun. 13, 816 (2022).
Yang, Z. et al. Identification of AUNIP as a candidate diagnostic and prognostic biomarker for oral squamous cell carcinoma. EBioMedicine 47, 44–57 (2019).
Charoentong, P. et al. Pan-cancer immunogenomic analyses reveal genotype-immunophenotype relationships and predictors of response to checkpoint blockade. Cell Rep. 18, 248–262 (2017).
Yang, W. et al. Genomics of Drug Sensitivity in Cancer (GDSC): a resource for therapeutic biomarker discovery in cancer cells. Nucleic Acids Res. 41, D955–D961 (2013).
Zhang, S. et al. mTORC1 promotes ARID1A degradation and oncogenic chromatin remodeling in hepatocellular carcinoma. Cancer Res. 81, 5652–5665 (2021).
Galon, J. & Lanzi, A. Immunoscore and its introduction in clinical practice. Q. J. Nucl. Med. Mol. Imaging 64, 152–161 (2020).
Liu, N. et al. Immunotherapy and senolytics in head and neck squamous cell carcinoma: phase 2 trial results. Nat. Med. 31, 3047–3061 (2025).
Tao, Z. Y., Wang, L., Zhu, W. Y., Zhang, G. & Su, Y. X. Lingual denervation improves the efficacy of anti-PD-1 immunotherapy in oral squamous cell carcinomas by downregulating TGFbeta signaling. Cancer Res. Commun. 4, 418–430 (2024).
Acknowledgements
We thank EditSprings for providing professional language services for this manuscript (https://www.editsprings.cn). We thank the Laboratory Medicine Center of the First Hospital of the University of Science and Technology of China for fluorescence-activated cell sorting support, and the Laboratory Animal Research Center of the First Hospital of the University of Science and Technology of China for animal research support. This work was supported by the National Natural Science Foundation of China (Nos. 82203277), China Postdoctoral Science Foundation (Nos. 2023T160624 and 2022M713043), Fundamental Research Funds for the Central Universities (Nos. WK9100000052), and Anhui Province Postdoctoral Science Foundation (No. 2023B697). We are grateful to Shanghai Lu Ming Biotech Co., Ltd. (Shanghai, China) for performing AFADESI-based spatially resolved metabolomics. We also acknowledge Hefei GeneIntel Medical Technology Co., Ltd. (Hefei, China) for their insightful suggestions and technical support.
Author information
Authors and Affiliations
Contributions
G.L. and Y.H. performed the in vitro and in vivo experiments. XY. and WD. analyzed the data. LZ., H.H., Y.W., and J.C. collected the clinical samples. Q.Z. and L.S. provided help with mouse subcutaneous tumor modeling and drug handling. P.H. and W.C. provided experimental help with molecular biology. S.L. and W.G. helped to revise the figure and manuscript. X.M., Y.Z., and H.L. helped to guide the study and revise the manuscript. S.Z. and Z.Y. developed the concept, designed the experiments, and guided the whole study. G.L. and Z.Y. wrote and revised the manuscript.
Corresponding authors
Ethics declarations
Competing interests
The authors declare no competing interests.
Additional information
Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.
About this article
Cite this article
Liu, G., Yao, X., Hou, Y. et al. Metabolomic and transcriptomic profiling of HNSCC identifies AMIGO2 as a therapeutic target modulating tumor microenvironment. npj Precis. Onc. 9, 358 (2025). https://doi.org/10.1038/s41698-025-01132-z
Received:
Accepted:
Published:
Version of record:
DOI: https://doi.org/10.1038/s41698-025-01132-z
