Data collection

The scRNA-seq profiles, including normal tissues (n = 9), tumor tissues (n = 20), and lymph nodes (n = 4), were downloaded from GSE181919. Gene expression data and related clinical information for TCGA-HNSC (normal = 44, tumor = 499) were obtained from UCSC Xena, with one patient (TCGA-CQ-A4CA-01) excluded due to missing survival data. For verification, datasets GSE41613, GSE65858, GSE42743, GSE30784, GSE25099, GSE37991 and GSE31056 were employed. Additionally, GBM-PRJNA482620 and Melanoma-PHS000452 from the TIGER database were acquired for immunotherapy studies. All expression data were log2 transformed.

Quality controlling, processing and clustering of scRNA-seq data

Quality control and cell type annotation for GSE181919 were conducted following methods described in the corresponding article.43 In brief, the “Seurat 4.0” R package was used to filter the raw data of the gene expression matrix, resulting in the inclusion of a total of 47,711 cells for subsequent analysis. Batch effects of samples were corrected using “RunHarmony” in the “Harmony” R package. Dimension reduction and clustering were performed using UMAP with the “RunUMAP” and “FindClusters” functions. Cell types were annotated manually based on marker genes calculated by “FindAllMarkers” function. The “inferCNV” R package was exploited to distinguish malignant cells from epithelial cells of tumor tissues by calculate copy number variations (CNVs).

Clarification of SGOC metabolic heterogeneity

For tumor metabolic heterogeneity analysis, single-sample gene set enrichment analysis (ssGSEA) was carried out using the “GSVA” package. The differential expressions of 114 metabolism pathways between non-malignant and malignant epithelial cells, as well as normal and tumor tissues, were calculated using the “Limma” package. Next, 425 genes involved in serine, glycine, one-carbon, folate, methionine, purine, and pyrimidine were downloaded from the Molecular Signature Database (MSigDB) and used to construct the SGOC metabolic network with five branched pathways44,45,46 (Table S2).

Malignant cells were classified into five groups using UMAP with the “RunUMAP” and “FindClusters” functions. The DEGs of the clusters were filtered using a threshold of adjusted P-values of < 0.05 and |log2FC | > 0.20. Kyoto encyclopaedia of genes and genomes (KEGG) pathway analysis was performed using the ‘clusterProfiler’ R package to evaluate the enrichment pathways of the sub-clusters. Then, deconvolution of the bulk RNA-Seq datasets were was performed using “MuSiC” R package.47 Univariate Cox regression analysis and Kaplan–Meier (K-M) survival curve analyses were conducted to evaluate prognostic value using the ‘survminer’ and ‘survival’ R packages.

Construction of a prognostic SGOC-related gene signature

The differentially expressed SGOC metabolic genes between non-malignant and malignant cells in scRNA-seq data were calculated by the “Findmarkers” algorithm with the same threshold as mentioned above. The prognostic values of the dysregulated SGOC genes were determined using univariate Cox regression analysis. Subsequently, 13 dysregulated prognostic SGOC genes (P < 0.05) were selected to construct a prognostic signature through LASSO-Cox regression analysis using the ‘glmnet’ R package. Penalty parameter lambda (λ) of the model was determined by 10-fold cross-validation. The risk score of each patient was calculated according to the normalized expression of the candidate genes (Expi) and their corresponding regression coefficients (Coei). The formula for the risk score was constructed as follows:

$$}\;\, }=\mathop\limits_^(\times )$$

The final formula is as follows: Risk Score= 0.256 × PLOD2 + 0.251 × HPRT1 + 0.407 × TBPL1-0.112 × SLC44A4.

The risk score was calculated for each HNSCC patient, and based on this score, patients were stratified into high-risk and low-risk groups using an optimal cutoff value. Kaplan−Meier survival curve analysis was then conducted to evaluate the prognostic value of the SGOC gene signature using the ‘survminer’ and ‘survival’ R packages.

Functional enrichment analysis of high- and low- risk HNSCC patients

DEGs between high- and low-risk groups in the TCGA-HNSC cohort were calculated using the R packages “limma”. For scRNA-seq data, malignant cells were extracted, and the risk score was calculated. The top 30% of risk scores were defined as the high risk group, while the bottom 30% were defined as the low risk group. DEGs between high- and low-risk groups in the scRNA-seq cohort were calculated using the R packages “FindMarkers”. The ssGSEA was carried out to analyze the enrichment of metabolism pathways. GO and KEGG pathway analysis were performed using the ‘clusterProfiler’ R package. The P-values of ssGSEA, GO terms and KEGG pathways were corrected. The correlation between SGOC-risk score and cell cycle score was calculated using Spearman correlation analysis.

The stromal score, immune score, ESTIMATE score and tumor purity score of each sample were computed using the “ESTIMATE” R package. The xCell score calculated by “xcell” R package provides a comprehensive tumor microenvironment landscape.48 The levels of immune cells in the TME were estimated by the CIBERSORT, xCell, and MCPcounter algorithms. Additionally, the correlations between the risk score and immune score, or immune cells infiltration, were calculated using Spearman correlation analysis, and the results were plotted by the “corrplot” R package.

Cell-cell communication

T cells were re-clustered to perform the analysis of cell-cell communication through “RunHarmony”, “RunUMAP” and “FindClusters” functions. CD8 + T cells, high, low and median risk malignant cells were selected and “CellCall” algorithm49 was used to analyze the intercellular communication.

Cell culture

Human HNSCC cell lines (HSC3, HSC4, HSC6, Cal33, CAL27, SCC1, HN6) were maintained in our laboratory in Guangzhou, China.12,50,51 CAL27, CAL33, HSC3, HSC6, and SCC1 cells were cultured in Dulbecco’s modified Eagle’s medium (DMEM; Gibco) supplemented with 10% fetal bovine serum (FBS; Gibco, USA). HN6 cells were grown in DMEM/F-12 (Gibco, USA) supplemented with 10% FBS. HSC4 cells were cultured in Minimum Essential Medium (MEM; Gibco, USA) supplemented with 10% FBS. The following inhibitors were used in this study: mycophenolic acid (MPA, Selleck, USA), mycophenolate mofetil (MMF, Selleck, USA).

Plasmid construction and transfection

HNSCC cells with stable knockdown of IMPDH1 were generated as previously reported.12,50 The plko.1-shNC-GFP-puro, plko.1-IMPDH1-sh1-GFP-puro, and plko.1-IMPDH1-sh2-GFP-puro plasmids were obtained from Long Bioscience (Guangzhou, China). Briefly, the IMPDH1-shs or shNC plasmids were co-transfected with lentivirus packaging plasmids psPAX2 and pMD2.G into 293FT cells using polyethylenimine. Subsequently, lentiviral particles were collected and used to infect HSC6 cells. The stably transfected cells with shNC or IMPDH1-shs were selected using puromycin (10 mg/ml) (Sigma). The efficiency of knockdown was evaluated by quantitative real-time PCR and western blotting assays.

Quantitative real-time PCR assay (qPCR)

The mRNA level of IMPDH1 was measured using qPCR. The SYBR Green-based qPCR analysis was conducted with the Light-Cycler 96 system (Roche). GAPDH served as an endogenous control for IMPDH1. The relative expression levels were calculated using the comparative threshold cycle equation (2 -ΔΔCT). The primer sequences were used as follows: IMPDH1, Forward Primer: 5’-CAGCAGGTGTGACGTTGAAAG-3’; Reverse Primer: 5’-AGCTCATCGCAATCATTGACG-3’. GAPDH, Forward Primer: 5’-CTCCTCCTGTTCGACAGTCAGC-3’, Reverse Primer: 5’-CCCAATACGACCAAATCCGTT-3’.

CCK-8 assay

For cell viability assessment, the CCK-8 assay was applied. Approximately 1.5 × 103 cells per well were seeded in 96-well plates. Following incubation with either DMSO or MPA treatment for the indicated times (0, 1, 2, 3 and 4 days), the cell medium was discarded, and cells were exposed to a mixture of 10 μL CCK8 (Sigma-Aldrich, USA) and 100 μL serum-free medium for an additional 2 h before detection. Cell viability was recorded based on absorbance readings at 450 nm using a spectrophotometric plate reader (Biotek, USA).

EdU click chemistry assay and fluorescence imaging

Briefly, 1 × 105 cells per well were seeded in confocal dishes. Following incubation with either DMSO or MPA treatment for 24 h, cells were exposed to 1 μL EdU (Beyotime, China). Then, cells were washed with phosphate-buffered saline (PBS, Sigma-Aldrich, USA), fixed in a 4% paraformaldehyde solution, permeabilized with 0.3% Triton-X 100 in PBS, and subjected to incubation with a reaction cocktail containing the necessary compounds for bonding of Alexa Fluor® 488 azide with EdU. Finally, cell nuclei were stained with DAPI (Sigma-Aldrich, USA). Cell imaging was performed using a laser scanning microscope (LSM 980, ZEISS).

Flow cytometry analysis

For cell cycle detection, CAL27 and HSC6 cells treated with either DMSO or MPA were seeded into 60 mm dishes. Once cell confluence reached 80%, the cells were fixed in 70% cold ethanol and then subjected to testing with the Cell Cycle Detection Kit (KeyGEN, China). The fluorescence signals were recorded using flow cytometry (Cytoflex, Beckman Coulter, USA).

For the apoptosis assay, CAL27 and HSC6 cells treated with DMSO, MPA or guanosine were seeded in 6-well plates. After incubating with serum-free medium for 24 h, the cells were harvested. An Annexin VFITC/ PI Apoptosis Detection Kit (KeyGEN, China) was used to detect apoptotic cells. HSC6 IMPDH1-shs or shNC cells were dyed by Annexin V-APC/DAPI Apoptosis Kit (Procell, China). The flow cytometer was performed to count these apoptotic cells.

For detecting the expressions cytokines of CD8 + T cells, tumor or spleen tissues derived from C3H mice were grinded into single cells. Then, cells were stimulated in vitro with cell stimulation cocktail (1:500, TNB-4975-UL100, Tonbo Biosciences) for 5 h at 37 °C with 5% CO2, followed by PI staining for 15 min and surface markers staining for 30 min in the dark. Next, cells were fixed and permeabilized with intracellular fixation and permeabilization buffer, and stained with intracellular cytokine antibodies according to the manufacturer’s instructions. The following antibodies were used: CD8α (FITC conjugated, 53-6.7, Cat#100706, Biolegend), CD3e (APC conjugated, 145-2C11, Cat#100312, Biolegend), PD-1(Percp-Cy5.5 conjugated, 29 F.1A12, Cat#135208, Biolegend), LAG3 (Bv421-Conjugated, C9B7W, Cat#125221, Biolegend), Tim-3 (PE-Cy7 conjugated, RMT3-23, Cat#25-5870-82, eBioscience), IFN-γ (APC-eFlour 780 conjugated, XMG1.2, Cat#47-7311-82, eBioscience), TOX (PE-conjugated, TXRX10, Cat#12-6502-80, eBioscience).

Western blotting

Cells were washed with ice-cold PBS and lysed with RIPA strong lysis buffer (Sigma-Aldrich, USA) supplemented with 1% protease and 1% phosphatase inhibitors (Beyotime, China). Then, 5× loading buffer (Beyotime, China) was added to the protein samples and cooked at 99°C for denaturation. The lysates were loaded onto 10% SDS-PAGE gel for separation and transferred to a 0.22 μm PVDF membrane (Millipore, USA). After blocking in 1× Protein Free Rapid Blocking Buffer (EpiZyme), the membranes were incubated with primary antibodies at 4 °C overnight, followed by incubation with species-matched secondary antibodies. Finally, the antigen-antibody reaction was tested by enhanced chemiluminescence (ThermoFisher, USA). The following antibodies were used: GAPDH (60004-1, 1:3 000, Proteintech), p53 (10442-1-AP, 1:1 000, Proteintech), B23/NPM1 (60096-1, 1:1 000, Proteintech), GNL3 (67169-1, 1:1 000, Proteintech), Cytochrome-C (66264-1, 1:2 000, Proteintech), rabbit IgG HRP-linked (7074, 1:3 000, CST), mouse IgG HRP-linked (7076, 1:3 000, CST), Cleaved PARP (5625, 1:1 000, CST), PARP (9532, 1:1 000, CST), Cleaved caspase-9 (20750, 1:1 000, CST), IMPDH1 (861791, 1:500, Zen BioScience).

Migration and invasion assays

Cells were scratched using a 10 μL pipette after the cells in the six-well plate. Photos were taken at 0 h and 24 h after scratching. To evaluate the migration capability, 6 × 104 CAL27 or HSC6 cells in 100 μL serum-free medium were added to the upper layer of transwell chamber (Corning, USA). The lower chamber of transwell was supplemented with DMEM medium with complete serum-medium. After 18 h for migration and 24 h for invasion, cells on the lower surface were fixed with 4% paraformaldehyde solution, stained with 0.4% crystalviolet (Beyotime, China), and counted under a microscope.

Immunofluorescence assay

Cells were harvested and fixed with 4% paraformaldehyde. After permeabilization using 0.3% Triton-X 100 in PBS, cells were incubated with the primary antibodies at 4 °C overnight. Then, cells were stained with species-matched fluorescent secondary antibodies. Nuclei were stained with DAPI. The slides were viewed using a laser scanning microscope. The following antibodies were used: B23/NPM1 (60096-1, 1:200, Proteintech), GNL3 (67169-1, 1:200, Proteintech), Donkey anti-Mouse IgG (H + L) Alexa Fluor Plus 488 (A32766, 1:1 000, ThermoFisher).

GTP and ATP content measurement

CAL27 or HSC6 cells treated with DMSO, MPA or guanosine were seeded in 60 mm dishes and subsequently harvested. Then cells were incubated with 80% cold carbinol (V/V) at 4 °C and collected. After centrifugation, the supernatant was collected and dried by a vacuum centrifugal concentrator at −50 °C to collect the precipitate. Then, the samples were dissolved in 60% acetonitrile before detecting the ATP and GTP content using a Triple quadrupole LC/MS.

In vivo tumor models

All animal research procedures were carried out in strict accordance with the detailed rules of the Institutional Animal Care and Use Committee of Sun Yat-Sen University, with approval numbers 2023000092 and 2023003617. A total of 1 × 106 HSC6 cells were subcutaneously injected into twelve female BALB/c nude mice (4–6 weeks old), while 5 × 106 SCC7 cells were subcutaneously injected into twelve female C3H mice (4 weeks old). Upon reaching tumor volumes of approximately 40 mm3, the mice were randomly divided into 2 groups and intraperitoneally injected with either DMSO or MMF every 2 days. Accordingly, the tumor volumes were recorded. After 2 weeks of injection and tumor growth, the mice were sacrificed, and primary tumors and spleens were collected.

Immunochemistry (IHC) analysis

IHC was performed on xenograft mice tissues.50 Briefly, dewaxing was carried out using xylene, followed by rehydration using alcohol with a gradient concentration. Endogenous peroxidase activity was blocked by 3% H2O2. Then, the slices were subjected to citrate-mediated high-temperature antigen retrieval. Afterward, slides were blocked with goat serum and incubated with primary antibodies overnight. Following washing with TBST solution, slides were incubated with secondary antibodies at room temperature. The Apreio AT2 digital whole slide scanner (Leica, Wetzlar, Germany) was applied to scan the slices. The following antibodies were used: B23/NPM1 (60096-1, 1:200, Proteintech), Cytochrome-C (66264-1, 1:2 000, Proteintech), p53 (10442-1-AP, 1:1 000, Proteintech).

Statistical analysis

The bioinformatics and statistical analyses were conducted using R 4.3.1, SPSS 27 and GraphPad Prism 8.0 softwares. The Student’s t test or ANOVAs (one- or two-way) was used to preform statistical analyses. Data presented as the mean ± SD were extracted from at least three independent experiments. P < 0.05 was considered as significant.