Integration RNA bulk and single cell RNA sequencing to explore the change of glycolysis-related immune microenvironment and construct prognostic signature in head and neck squamous cell carcinoma

Highlights

• •High glycolysis level indicates poor prognosis of patients with head and neck squamous cell carcinoma.
• •RNA bulk and single cell RNA sequencing database of head and neck squamous cell carcinoma analysis showed that patients with high glycolysis levels had less infiltration of macrophages, T cells and monocytes.
• •The risk signature was verified by the external verification dataset, and the results show that the prediction effect is good.
• •In vitro functional and Western blot assays confirmed that the above five risk genes(MUCL1,TRIML2,RAB3B,SPINK6,IGSF11) affect tumor function and related to the process of glycolysis.

Abstract

Background

Glycolysis is an indispensable process for tumor cell,but the effect of glycolysis on the prognosis and immune cell infiltration of head and neck squamous cell carcinoma is not clear.

Methods

Based on RNA bulk and single cell RNA sequencing data of head and neck squamous cell carcinoma from The Cancer Genome Atlas(TCGA) and GSE195832, the effect of glycolysis level on immune cell infiltration was analyzed. Then, we obtained the prognostic genes related to glycolysis through survival analysis to construct prognostic risk signature. Our sample and GSE65858 datasets are used as external verification datasets to verify the validity of the signature. Finally, we used Western blot and cell function assays to determine the relationship between risk genes and glycolysis and the function of prognostic genes.

Result

The level of glycolysis was related to the prognosis of head and neck tumors (P = 0.0044). The results of immune infiltration analysis of TCGA database showed that high level glycolysis subgroup had less infiltration of macrophages, T cells and monocytes. Results of single cell sequencing analysis validates the above results. Additionally, Five risk genes(MUCL1,TRIML2,RAB3B,SPINK6,IGSF11) were selected to construct signature.Risk score was an independent prognostic factor(P < 0.01). The external validation set also shows the same result. In vitro functional and Western blot assays confirmed that the above five genes affect tumor function and related to the process of glycolysis.

Conclusion

Glycolysis-related risk signatures can be used to predict the prognosis and immune infiltration of head and neck squamous cell carcinoma.

Introduction

In the national cancer statistics in 2020, the incidence of head and neck tumors ranks seventh among all cancers. New cases and mortality accounted for 4.9 % and 4.7 % of all cancers, respectively [1]. And because the location of the disease is partly in the pharynx and larynx, it leads to difficulty in breathing and swallowing. Surgical treatment may result in loss of voice due to removal of the location of the disease. The application of vocal prosthesis has achieved certain results, but the effect of surgery on the quality of life of patients is beyond doubt [2].Although the diagnosis and treatment of head and neck squamous cell carcinoma have achieved good results, the 5-year survival rate of patients with head and neck squamous cell carcinoma is still less than 50 % and has a high recurrence and metastasis rate [3,4]. It is necessary to find a better treatment.

In the past, single tumor immunotherapy did not get the desired results, mostly because the inhibitory effect of metabolic reprogramming on immune cells in microenvironment was not taken into account [5].Abnormal energy metabolism is the basic feature of tumor, and abnormal glucose metabolism is the most important feature of tumor [6]. Glycolysis was obviously activated in head and neck squamous cell carcinoma, and the increase of key enzymes related to glycolysis occurred in oral squamous cell carcinoma, laryngeal squamous cell carcinoma and pharyngeal carcinoma [[7], [8], [9]]. The process of glycolysis not only increases energy supply by itself, but also promotes immune escape of tumor cells. The high concentration of lactic acid produced during glycolysis can promote the M2 polarization of macrophages, promote the expression of HIF-1 α in macrophages and lead to T cell apoptosis, and inhibit the immune lethality of mature CD8+T cells and NK cells [[10], [11], [12], [13]]. Glycolysis can also affect the indirect killing effect of cytotoxic T cells on tumor cells by acting on TNF-α signal pathway [14]. Clinical studies have shown that the level of glycolysis is negatively correlated with the overall survival time of tumor patients [15]. The clinical studies of 311 patients have shown that the serum LDH level higher than 1000 IU /L indicates distant metastasis of the tumor [16]. To sum up, glycolysis can provide energy for tumor cells and promote immunosuppression, resulting in a poor prognosis of patients.Recently, researchers inhibit the expression of glucose transporter 1(GLUT1) to inhibit the secretion of lactic acid in glioblastoma so as to reduce the number of macrophages M2 and regulatory cells, and increase the sensitivity of immunotherapy [17]. In the same year, researchers shunt lactic acid to mitochondria by acting lithium carbonate (LC) to reduce the inhibition of lactic acid on CD8+T cells and achieve the effect of cancer immunotherapy [18]. Thus it can be seen that targeted glycolysis cannot only inhibit the development of tumor, but also relieve the immunosuppression of tumor microenvironment, which can improve the sensitivity of immunotherapy.

In the past, There are few studies on glucose metabolism of head and neck squamous cell carcinoma. We obtained the RNA expression data and clinical data of head and neck squamous cell carcinoma through TCGA database to analyze the effect of high and low glycolysis levels on survival and the difference of immune infiltration between the two groups. Finally, in order to better serve the clinic, we establish glycolysis-related risk prognostic signature, using the Gene Expression Omnibus(GEO) and our sample database as external verification sets to verify the reliability of the risk prognostics signature. At the same time, the function of the above risk genes were verified by in vitro assays.

Materials and methods

Acquisition of datasets

We downloaded RNA sequencing data and clinical information of head and neck squamous cell carcinoma from GDC web(https://portal.gdc.cancer.gov/), including 520 tumor samples and 44 normal samples. Then we obtained the single-cell RNA sequencing data (GSE195832(Platforms:GPL24676, read lengths: 2 × 150 bp (PE150)) of 4 cases of untreated head and neck squamous cell carcinoma from GEO database.. In this study, two datasets are used as external validation sets.One of which was GSE65858(Platforms:GPL10558 read lengths:about 50 bp). The RNA sequencing data and clinical information of 270 patients with head and neck squamous cell carcinoma were included. Another verification dataset included 59 patients with head and neck squamous cell carcinoma from Hebei Medical University Sample Library, which was approved by the Research Ethics Committee of the second Hospital of Hebei Medical University (2021-R012-01). All patients obtained informed consent by signing an ethical consent form. Table 1 shows the details of the patients in the sample bank.

Table 1.

Clinical sample information.

	Clinical dataset(n = 59)
Age	62.85+7.51
Male	56(94.92 %)
Stage
Ⅰ	11(18.64 %)
Ⅱ	12(20.34 %)
Ⅲ	15(25.42 %)
Ⅳ	21(35.59 %)
T
T1	13(22.03 %)
T2	18(30.51 %)
T3	14(23.73 %)
T4	14(23.73 %)
N
N0	40(67.80 %)
N1	9(15.25 %)
N2	10(16.95 %)
N3	0
M
M0	59(100 %)
M1	0

Data processing

GSE195832: We downloaded Single cell transcriptome data of 4 untreated patients from the GEO website. Each sample generates barcodes.tsv, genes.tsv, and matrix.mts files containing the transcriptional count of each sample.We uploaded the above data to R studio software, and then used "Seurat" package to filter the cells by using percentage of mitochondrial genes (percent.mt) < 5,& percentage of red blood cell gene (Percent.HB) < 20 & 200<the number of genes detected in each cell (nFeature_RNA) < 5500 as the screening condition. we standardize the data to find the first 2000 hypervariable genes of each sample. Then merge the four sample data into “Seurat objects”, which are linear dimensionality reduction by PCA, and the cells are clustered by“Run TSNE”. Use "FindAllMarkers" to find the top 10 markers of cell groups, and annotate the cell groups through the CellMarker website(http://bio-bigdata.hrbmu.edu.cn/CellMarker).

GSE65858: We downloaded the sequencing data of 270 patients with head and neck squamous cell carcinoma from the GEO database. After processing the outliers of the expressed data, we removed the genes with zero expression in all the samples, then normalized the samples, and finally reduced dimension by “PCA” (The expression level of the sample has been standardized by log2).

Principal components and Kaplan-Meier survival analysis

Tricarboxylic acid cycle, pentose phosphate pathway and glycolysis related genes were obtained through KEGG and MSigDB website (https://www.kegg.jp; http://www.gsea-msigdb.org/gsea/msigdb). According to the gene expression, the samples were divided into high- and low- expression groups of glucose metabolism by single sample gene set enrichment analysis (ssGSEA) algorithm, and principal component analysis (PCA) was performed. Then, according to the death and survival time of the patients with head and neck squamous cell carcinoma, Kaplan-Meier survival analysis was performed on the high and low risk groups, and the survival differences of the three high and low expression groups were obtained.

Difference analysis

Patients with head and neck squamous cell carcinoma were divided into high- and low- risk groups, and the differences between the two subgroups were analyzed by "LIMMA" package. The differentially expressed genes between high- and low- expression groups were obtained by using | logFC | > 1 and P Value < 0.01 as the screening condition.

Immune microenvironment score

The immune infiltration of 22 kinds of immune cells, immune score,stromal score, the scores of MHC molecules, effector cells, suppressor cells and immune checkpoints were obtained by "CiberSort", "ESTIMATE" and "IPS" algorithms, respectively, and the differences of immune infiltration between high- and low- glycolysis expression subgroups were analyzed.

Establishment of glycolysis-related signature

We used "LIMMA" package to analyze the difference between the high- and the low- expression subgroups of glycolysis to obtain the differentially expressed mRNAs. Then "survival" and "survminer" packages were used to analyze the survival of the differentially expressed mRNAs by log-rank test and univariate Cox. Using P Value < 0.05 as the screening criteria, 5 mRNAs related to the survival of head and neck squamous cell carcinoma were screened. We obtained the risk coefficient of mRNAs by multivariate Cox survival analysis, and calculated the risk score of patients. Then univariate and multivariate Cox were used to analyze the impact of risk score on prognosis. Finally, a Nomogram prognostic model was constructed by combining sex, age, stage and other clinical factors. The effectiveness of Nomogram construction was verified by the correction curve.

Cell culture

Head and neck squamous cell carcinoma cell line (TU177,AMC-HN-8) was purchased from Beina Chuanglian Institute of Biotechnology in Beijing. The cells were cultured in RPMI 1640 and DMEM containing 10 % fetal bovine serum, and the incubator was set at 37 °C and 5 % CO₂. The above culture medium was purchased from GIBCO 108 (Thermo Fisher Science,Inc.)

Quantitative real-time polymerase chain reaction

We used EaStep ®Super Total RNA extraction Kit (USA,Promega) to extract tissue RNA from 48 head and neck squamous cell carcinoma samples and their matched non-cancer samples. After RNA extraction, the mass and concentration of RNA were measured by spectrophotometer, and then cDNA was obtained by reverse transcription with Transcriptor First Strand cDNA Synthesis Kit (Roche, Germany). Finally, quantitative real-time polymerase chain reaction (qRT-PCR) was performed with GoTaq ®qPCR Master Mix (Promega, USA). The relative expression of the gene was calculated by 2-∆∆Ct (with 18s as the internal reference).

The sequence of primers used in the experiment is as follows:

• MUCL1(forwards): 5 ʹ- TCTGCCCAGAATCCGACAAC-3 ʹ

• MUCL1(reverse): 5ʹ-AATGTCTTTACGAGCAGTGGTAG-3 ʹ

• RAB3B (forwards): 5 ʹ- CCGCTATGCTGATGACACGTT-3 ʹ

• RAB3B (reverse): 5 ʹ-ACGGTAGACTGTCTTCACCTTG-3 ʹ

• TRIML2(forwards): 5 ʹ-GCCACCGAGCTAGAGGAGAT-3 ʹ

• TRIML2(reverse): 5 ʹ-CTTGAGCAATGCCAAGGTGC-3 ʹ

• SPINK6(forwards): 5 ʹ-TGACTGTGGTGAGTTCCAGGA-3 ʹ

• SPINK6(reverse): 5 ʹ-CCACTTTTCACTATGGCCTTACA-3 ʹ

• IGSF11(forwards): 5 ʹ-GTCATTTGGATGGTCACTCCTC-3 ʹ

• IGSF11(reverse): 5 ʹ-AATCCTACCCTACCGTGGAAC-3 ʹ

• 18S RNA(forward): 5ʹ- GTAACCCGTTGAACCCCATT-3ʹ

• 18S RNA(reverse): 5ʹ- CCATCCAATCGGTAGTAGCG-3ʹ

Construction of shRNA plasmid of target mRNA

The pGenesil-1 plasmid was digested with BamHI and HindIII, and the cut plasmid was obtained by gel extraction. The single strand of the target gene was obtained by designing primers, and the double-stranded DNA of the target gene was obtained by heating 3 min at 100 °C. the primer sequence can be seen in supplementary file 1. Finally, the cut plasmid and the double-stranded DNA of the target gene were ligated by T4 ligase. The success of the connection was judged by the sequencing results. The knock-down efficiency of shRNA can be verified by the above qRT-PCR.

Cell transfection

We inoculated TU177 and AMC—HN-8 in six-well plate. When the cell fusion rate was 90 %−95 %, 4 μg plasmid (shMUCL1,shTRIML2,shRAB3B,shSPINK6,shIGSF11) and control group (pGenesil-1) were combined with 7.5 μg Hieff Trans Liposomal Transfection Reagent (Yeasen Biotechnology, Shanghai) were transferred into cells. 48 h later, the transfection efficiency was observed, and then the follow-up assays were carried out.

Cell proliferation assay

The transfected cells were digested with trypsin and implanted in a 96-well plate with a number of 2 × 10³ cells per well, repeated 3 times in each experimental subgroup. After 0 h, 24 h, 48 h and 72 h, 20 μL MTS (CellTiter96 ®AQueousOne Solution Cell Proliferation Analysis Kit) (Promega, USA) was added to the 96-well plate and cultured for 3 h. The optical density (OD) at 490 nm was measured by Spark ®143 multimode microplate meter (Spark 10m Tecan, Switzerland).

Invasion and migration assay

Remove the Matrigel from the -80 °C refrigerator and put it on the freezer. Then diluted with serum-free medium at 1:14 ratio. Use a precooled 200 μl gun head to absorb 50 μl diluted matrix glue and add it to the upper chamber. Then incubated in a 37 °C incubator for more than 30 min, the Matrigel was fully solidified, and then the remaining liquid in the upper chamber was sucked out. 650 μl serum-free medium was added to the lower chamber of the migration and invasion experiment. The serum medium containing 1 × 10⁵ transfected cells was added to the upper chamber. Then put it into the incubator for 24 h, carefully erase the non-invasive or non-migratory cells in the upper chamber after 24 h, fix 20 min in 4 % paraformaldehyde, wash and dry with PBS, and then dye 20 min with crystal violet. After washing and drying, the cells were imaged and counted in at least three random areas.

Clone formation assay

The transfected TU177 or AMC-HN-8 cells were inoculated in a six-well plate with 2 × 10³ cells/well and cultured in an incubator at 37 °C and 5 % CO₂ for 14 days. After 14 days, the medium in the six-well plate was sucked out, PBS was cleaned for 3 times, 20 min was fixed with 4 % paraformaldehyde, and then 0.5 % crystal violet stained 20 min. Finally, after washing and drying, the number of clone formation formed in the six-hole plate was counted.

Western blot

When the fusion rate of TU177 and AMC—HN-8 cells in the six-well plate reached 90 %−95 %, it was transferred into the target gene knock-down plasmid and its control plasmid. After 48 h, the transfection efficiency was observed, and the lysate (RIPA (SolarBio, China): PMSF (SolarBio, China): protease inhibitor (Promega, USA) = 100:1:1) was used. The protein products were mixed with SDS-PAGE loading buffer (Solabio, China) at 4:1 after the protein concentration was measured by BCA method, and 5 min was boiled in a metal bath at 95 °C. Then we added the same amount of (20 μg) protein mixture to the 10 % sodium dodecyl sulfate-polyacrylamide gel electrophoresis plate for electrophoresis. The separated protein was transferred to the polyvinylidene fluoride membrane, then the membrane was incubated in 5 % skimmed milk for 2 h, and then incubated overnight with the corresponding first antibody. The TBST washed the membrane three times the next day, and then incubated the membrane with the corresponding second antibody for 2 h. The chemiluminescence of the film was displayed by fluorescence XRS+ (Bio-Rad, USA). The main antibodies include: PKM2 (1:1000; R1603–5,Huabio,China); PFKP(1:1000; HA500472;Huabio,China);β actin (1:10,000; 81,115–1-RR, Proteintech, Wuhan, China). The second antibody was Anti-Rabbit IgG (Haul) (1:10,000, SA00001–2, Proteintech, Wuhan, China).

Results

Glycolysis level is related to the prognosis of patients with head and neck squamous cell carcinoma

The important pathway of glycolysis tumor energy supply affects the occurrence and development of tumors. In order to study the effect of glycolysis on the prognosis of patients with head and neck squamous cell carcinoma, we download glycolysis-related genes through KEGG and MSigDB website and divide patients into two groups with high and low glycolysis score by ssGSEA algorithm. Principal component analysis showed that there were significant differences in expression between the above two groups (Fig. 1A). Then the Kaplan-Meier survival analysis of the two groups showed that the survival time of the group with high glycolysis score was significantly shorter than that of the group with low glycolysis score (P = 0.0044, Hazard ratio=1.560) (Fig. 1B).Multivariate Cox survival analysis confirmed that glycolysis score was still an independent prognostic factor after excluding the effects of age, sex and stage(P = 0.003, Hazard ratio=1.614) (Supplementary Fig. 1).Other glucose metabolism-related pathways such as pentose phosphate pathway and tricarboxylic acid cycle did not show the same extremely significant results (Fig. 1A,B). Based on the above results, we can conclude that there is a significant correlation between glycolysis and the prognosis of head and neck squamous cell carcinoma.

Fig. 1.

Effects of different levels of glycolysis on the prognosis of patients with head and neck squamous cell carcinoma. (A): PCA clustering of tricarboxylic acid cycle, oxidative phosphorylation and glycolysis at different levels in patients. (B): Kaplan-Meier analysis of the survival results of high and low levels of tricarboxylic acid cycle, oxidative phosphorylation and glycolysis in patients.(HR= Hazard Ratio).

Glycolysis is associated with immune cell infiltration in head and neck squamous cell carcinoma

Previous studies have proved that glycolysis is related to tumor immunity and can promote tumor cell immune escape [10]. We also believe that the expression level of glycolysis in head and neck squamous cell carcinoma is related to the degree of tumor immune cell infiltration. In order to verify this conjecture and analyze the specific types of immune cells, we performed "ESTIMATE" immune score on 520 patients with head and neck squamous cell carcinoma in TCGA database.It was found that the high expression group had higher level of stromal score and lower immune score (Fig. 2A). Then the IPS scores of the patients were analyzed, and the immunophenotypic scores (MHC molecules, effector cells, inhibitory cells, checkpoints) were obtained. The score of MHC molecules and effector cells decreased, and score of the checkpoint increased (Fig. 2B). Finally, we performed "CiberSort" immune score on 520 patients, and the infiltration of 22 kinds of immune cells in each sample was obtained. it was found that the infiltration of monocytes, macrophages M1, CD4+T cells,CD8+T cells and natural killer cells activated decreased and macrophage M0 infiltration increased in the high glycolysis group (Fig 2C). Combined with the analysis results of the above TCGA database, we can know that the infiltration of macrophages, natural killer cells and T cells decreased and immune escape occurred in the high expression group of glycolysis-related genes.

Fig. 2.

Effect of high and low levels of glycolysis on immune cell infiltration in patients with head and neck squamous cell carcinoma.(A): Differences in stromal scores, immune scores and estimate scores between high and low levels of glycolysis. (B): Differences in immunophenotypic scores (MHC molecules, effector cells, suppressor cells, checkpoints) between high and low levels of glycolysis. (C): The difference of immune cell infiltration between high and low levels of glycolysis.

In the single cell sequencing dataset, the number of macrophages and T cells in the high score group of glycolysis decreased

In order to make up for the deficiency of TCGA data analysis, we obtained the single cell sequencing data (GSE195832) of head and neck squamous cell carcinoma from GEO database to verify the changes of immune cells. According to the above filtration conditions, the total number of cells was obtained and divided into 14 cell subgroups. 8 cell subsets were identified as "Tumor cells", "Endothelial cells", "Macrophage", "Fibroblast", "T cells", "Natural killer cells", "Monocyte" and "Plasma cells" by the cell markers "LAMB3", "VWF", "C1QB", "DCN", "CD69", "GZMB", "S100A9″ and "MZB1". Then the glycolysis scores of 8 cell subsets were obtained by "AUcell" package (Fig. 3A), which showed that the glycolysis scores of tumor cells, monocytes and macrophages were higher (Fig. 3B). In order to further analyze the effect of glycolysis level on immune cells, we divided the cells into high and low score subgroup according to the median of glycolysis score. The proportion of tumor cells with high score was higher, but the proportion of macrophages, T cells and monocytes was lower (Fig. 3C). The above results validate the results of immunoassay in TCGA database. We also analyzed the expression of glycolysis markers (PKM,PFKP,GAPDH,HK1,LDHA,LDHB) in eight cell groups, and we found that the expression of glycolysis markers in tumor cells was significantly higher than that in other cell subsets (Fig. 3D). Combined with the above results, it can be inferred that the process of glycolysis mainly occurs in tumor cells, and in cells active in the process of glycolysis, there is less infiltration of macrophages, monocytes and T cells, and immune escape occurs.

Fig. 3.

Processing and analysis of single cell sequencing data of head and neck squamous cell carcinoma (A) Grouping, annotation, and cell glycolysis AUC score of single cell sequencing data (B) Differences in glycolytic AUC scores among different cell groups (C) Differences in the proportion of cell groups in the high and low score groups of glycolysis (D) The differential expression of glycolytic characteristic marker genes (PKM, PFKP, GAPDH, HK1, LDHA, LDHB) among different cell groups.

Establishment of prognostic signature related to glycolysis

In order to make our research results better serve the clinic, we established prognostic signatures based on glycolysis in patients with head and neck squamous cell carcinoma. First of all, we analyzed the differential expression of the above high- and low- glycolysis expression groups by "LIMMA" package. Taking | logFC | > 1 and P Value < 0.01 as screening conditions, a total of 151 differentially expressed mRNAs(DEmRNAs) were obtained, including 94 up-regulated DEmRNAs and 57 down-regulated DEmRNAs (Fig. 4A). DEmRNAs can be found in supplementary file 2. Enrichment analysis of the above DEmRNAs showed that Glycolysis pathway, Fatty acid degradation, IL-17 signal pathway, Glutathione metabolism and Pathways in cancer, and Cell adhesion molecules (CAMs) were enriched. The results of enrichment analysis are shown in Table 2. After Kaplan-Meier and univariate Cox survival analysis of the DEmRNAs, 5 genes(MUCL1,TRIML2,RAB3B,SPINK6,IGSF11)were found to be related to the prognosis of patients with head and neck squamous cell carcinoma (Fig. 4B–F,Table 3).The differences in the expression of the above five genes were also verified in our clinical samples (Fig. 4G–K).In order to evaluate the effect of the above five risk genes on the prognosis of head and neck squamous cell carcinoma, the risk coefficients of the above mRNAs were obtained by multivariate Cox analysis, and the risk scores of patients were calculated:

Fig. 4.

Screening of risk genes for glycolysis prognostic signature. (a): Analysis of differences in gene expression between patients with high and low levels of glycolysis. (B–F): Kaplan-Meier analysis of survival results between high and low expression groups of five risk genes (MUCL1, TRIML2, RAB3B, SPINK6, IGSF11). (G–K): The differential expression of the five genes in 48 samples of head and neck squamous cell carcinoma between cancer and corresponding non-cancerous tissues.

Table 2.

KEGG enrichment pathway.

Pathways	ID	P-Value	Gene
Retinol metabolism	hsa00830	5.25E-10	ADH7|ALDH1A1|UGT1A8|CYP26A1| ADH1C|UGT1A6|UGT1A7|UGT1A1
Drug metabolism - cytochrome P450	hsa00982	8.87E-10	ADH7|GSTA1|MGST1|UGT1A8|ADH1C| UGT1A6|UGT1A7|UGT1A1
Metabolism of xenobiotics by cytochrome P450	hsa00980	1.32E-09	ADH7|GSTA1|MGST1|UGT1A8|ADH1C| UGT1A6|UGT1A7|UGT1A1
Chemical carcinogenesis	hsa05204	2.30E-09	ADH7|GSTA1|MGST1|UGT1A8|ADH1C| UGT1A6|UGT1A7|UGT1A1
Metabolic pathways	hsa01100	8.05E-07	RIMKLA|PNLIPRP3|PCYT1B|ADH7| UGT8|GALNT5|CEL|NOS2|GSTA1| CHST9|ALOX15|MGST1|UGT1A8| CYP26A1|PLA2G2F|ALDH1A1|ADH1C| UGT1A6|UGT1A7|UGT1A1
Drug metabolism - other enzymes	hsa00983	9.78E-07	GSTA1|MGST1|UGT1A8|UGT1A6| UGT1A7|UGT1A1
Porphyrin and chlorophyll metabolism	hsa00860	1.04E-06	UGT1A6|UGT1A7|UGT1A8|UGT1A1| CP
Ascorbate and aldarate metabolism	hsa00053	6.02E-06	UGT1A6|UGT1A7|UGT1A8|UGT1A1
Pentose and glucuronate interconversions	hsa00040	1.38E-05	UGT1A6|UGT1A7|UGT1A8|UGT1A1
Steroid hormone biosynthesis	hsa00140	0.0001102	UGT1A6|UGT1A7|UGT1A8|UGT1A1
Pathways in cancer	hsa05200	0.00116175	IGF2|NOS2|CLGN|GSTA1|FGF19|MGST1| SHH|HES5
Cell adhesion molecules (CAMs)	hsa04514	0.00272502	VTCN1|NRXN3|CLDN3|IGSF11
IL-17 signaling pathway	hsa04657	0.00607951	MUC5AC|CLGN|S100A7
Linoleic acid metabolism	hsa00591	0.00630935	ALOX15|PLA2G2F
Pancreatic secretion	hsa04972	0.00699322	CEL|CHRM3|PLA2G2F
Proteoglycans in cancer	hsa05205	0.00845599	SHH|IGF2|GPC3|HSPG2
Tyrosine metabolism	hsa00350	0.00937314	ADH1C|ADH7
Ferroptosis	hsa04216	0.01136571	ALOX15|CP
Fat digestion and absorption	hsa04975	0.01189044	CEL|PLA2G2F
Ras signaling pathway	hsa04014	0.0131756	IGF2|FGF19|NTRK2|PLA2G2F
Fatty acid degradation	hsa00071	0.01352703	ADH1C|ADH7
ABC transporters	hsa02010	0.01409308	DEFB1|ABCA13
Ether lipid metabolism	hsa00565	0.01525544	UGT8|PLA2G2F
Glutathione metabolism	hsa00480	0.0209688	MGST1|GSTA1
Glycerolipid metabolism	hsa00561	0.02446903	PNLIPRP3|CEL
Arachidonic acid metabolism	hsa00590	0.02593139	ALOX15|PLA2G2F
Phosphonate and phosphinate metabolism	hsa00440	0.02652625	PCYT1B
Hepatocellular carcinoma	hsa05225	0.02833257	IGF2|MGST1|GSTA1
Staphylococcus aureus infection	hsa05150	0.02973755	DSG1|DEFB1
Glycolysis / Gluconeogenesis	hsa00010	0.02973755	ADH1C|ADH7
Platinum drug resistance	hsa01524	0.03375087	MGST1|GSTA1
ECM-receptor interaction	hsa04512	0.04509023	FREM2|HSPG2
Insulin secretion	hsa04911	0.04509023	CHRM3|PCLO
Salivary secretion	hsa04970	0.04882485	CHRM3|STATH

Table 3.

Univariate Cox survival analysis.

	coef	P value	HR	95 %CI
MUCL1	-0.415	0.01	0.66	0.482–0.904
TRIML2	-0.414	0.011	0.661	0.48–0.91
RAB3B	-0.459	0.004	0.632	0.461–0.866
SPINK6	0.442	0.007	1.555	1.13–2.14
IGSF11	-0.39	0.016	0.677	0.493–0.93

HR: Hazard Ratio;CI: Confidence Interval.

Riskscore=0.5608538*MUCL1 expression + 0.6005111* TRIML2 expression +0.3225742* RAB3B expression-0.6619177* SPINK6 expression+0.3264303* IGSF11 expression

With the increase of risk score, the survival time of patients shortened and the number of deaths increased (Fig. 5A-B). At the same time, the survival time of the high-risk group was significantly shorter than that of the low-risk group(P < 0.0001, Hazard ratio=2.039) (Fig. 5C).There were also significant differences in the expression of five risk genes between the two groups (Fig. 5D). We also verified this conclusion in the GEO external verification set and our sample bank clinical samples, the survival time of patients shortened and the number of deaths increased with the increase of risk score (Fig 5E,F, I,J); the survival time of patients in high-risk group was shorter than that in the low risk group(P = 0.0076, Hazard ratio=1.728 & P = 0.0038, Hazard ratio=1.804) (Fig 5G, K);and there were differences in the distribution of five risk genes between the high-risk and the low-risk group (Fig 5H, L). According to the above results, it is known that the risk score has good applicability.

Fig. 5.

Glycolysis-related risk score. (A): Risk score distribution of patients with head and neck squamous cell carcinoma in the TCGA cohort. (B): Survival status of patients with high and low risk of glycolysis in the TCGA cohort. (C): Kaplan-Meier survival analysis of survival outcomes between high and low risk groups in the TCGA cohort. (D):The expression of five risk genes in the TCGA cohort with high and low glycolysis scores. (E): Risk score distribution of patients with head and neck squamous cell carcinoma in the GEO cohort. (F): Survival status of patients with high and low risk of glycolysis in the GEO cohort. (G): Kaplan-Meier survival analysis of survival results between high and low risk groups in the TCGA cohort. (H): The expression of five risk genes in the high and low glycolysis score group in the GEO cohort. (I): Risk score distribution of patients with head and neck squamous cell carcinoma in the clinical validation cohort. (J): Survival status of patients in the high and low risk group of glycolysis in the clinical validation cohort. (K): Kaplan-Meier survival analysis of survival outcomes between high and low risk groups in the clinical validation cohort. (L): The expression of 5 risk genes in the high and low glycolysis score group in the clinical verification cohort.

Glycolysis-related risk score is an independent prognostic factor

We performed univariate Cox analysis of risk scores and clinical factors(stage, sex, age) on the TCGA training set and two validation sets. Risk score and age were correlated with prognosis (Fig. 6A). In order to determine whether glycolysis related riskscore is an independent prognostic factor, We performed multivariate Cox analysis in three datasets. The result showed that risk score excluded the effect of age,sex and stage were still independent prognostic factors (P < 0.01) (Fig. 6B). In order to further visualize the impact of risk score on patient survival, we constructed a Nomogram model to predict patient survival. The higher the risk score and the older the age, the worse the prognosis of the patients (Fig. 6C). Both TCGA database and external verification set show that Nomogram's prediction of 3-year and 5-year survival conditions is in good agreement with the actual survival conditions (Fig. 6D).

Fig. 6.

Construction of Nomogram model. (A): Univariate Cox survival analysis of risk scores and clinical factors in TCGA datasets and validation sets (GEO, clinical sample database). (B): Multivariate Cox survival analysis of risk scores and clinical factors in TCGA datasets and validation sets (GEO, clinical sample database). (C): Nomogram constructed by the TCGA training cohort. (D): Correction curves based on data from TCGA and validation set (GEO, clinical sample database).

Functional experiment of five risk genes related to glycolysis

In order to determine the effect of five risk genes on the function of tumor cells, we constructed the knockdown plasmid (shMUCL1,shTRIML2,shRAB3B,shSPINK6,shIGSF11) of the above five risk genes, and the knockdown efficiency has been verified (Supplementary Fig. 2). The knockdown plasmid and the control plasmid (pGenesil-1) were transfected into TU177 and AMC—HM-8 cells. 48 h later, the transfection efficiency was determined by fluorescence, and then tumor cell viability, migration, invasion and colony formation were tested. The results of functional assays of TU177 and AMC—HN-8 cells showed that the viability, migration, invasion and clone formation ability of tumor cells decreased after MUCL1,TRIML2,RAB3B and IGSF11 knockdown, while the effect of SPINK6 knockdown was the opposite (Fig. 7A–F). To sum up, it is known that the above five risk genes affect the function of tumor cells.

Fig. 7.

Functional experiment of five risk genes. (A-B): MTS assay was used to verify the changes in cell viability after knocking down five risk genes in TU177 and AMC—HN-8 cell lines. (C-D): Transwells assay was used to verify the changes in invasion and migration ability after knocking down five risk genes in TU177 and AMC—HN-8 cell lines. (E-F): Colony formation assay was used to verify the changes in cell proliferation after knocking down five risk genes in TU177 and AMC—HN-8 cell lines. (G-H): Western blots assay was used to verify the changes in glycolysis marker protein (PFKP,PKM2) after knocking down five risk genes in TU177 and AMC—HN-8 cell lines.

Risk genes are associated with glycolysis

In order to verify whether the risk gene is related to the glycolysis process, we transferred the knock-down plasmid and control group (pGenesil-1) into TU177 and AMC—HN-8 cells. 48 h later, the protein was collected and detected by Western blot. According to the results, the protein content of PKFP and PKM2 decreased after MUCL1,TRIML2,RAB3B and IGSF11 knockdown, while PKFP and PKM2 increased after SPINK6 knockdown. Based on the above results, it can be concluded that the above risk genes are related to glycolysis (Fig. 7G-H).

Discussion

In summary, the glycolysis level affected the prognosis of head and neck squamous cell carcinoma, and the results of RNA-bulk sequencing and single cell sequencing data analysis showed that the infiltration of macrophages, T cells and monocytes in the high glycolysis group was less than that in the low glycolysis group. We analyzed the difference between the high and low score groups of glycolysis, and a total of 151 DEGs were obtained.Glycolysis pathway, Fatty acid degradation, IL-17 signal pathway, Glutathione metabolism and Pathways in cancer, CAMs were enriched by enrichment analysis of the above genes. Five prognostic risk genes (MUCL1, TRIML2, RAB3B, SPINK6, IGSF11) were screened to construct prognostic risk signature, and the verification results of the external verification set show that the signature has good applicability.

Our study found that the survival time of the high-risk group of glycolysis was significantly shorter than that of the low-risk group (P = 0.0044, Hazard ratio=1.560). Even when there is sufficient oxygen, tumor cells still choose glycolysis as their own way of energy supply [19]. The amount of ATP produced by a single metabolic process of glycolysis is less than that of oxidative phosphorylation, but the metabolic rate of glycolysis is more than 100 times that of oxidative phosphorylation, so it can obtain the necessary energy more quickly for its own growth, migration and invasion [20]. Secondly, glycolysis can promote the occurrence of pentose phosphate pathway, resulting in an increase in the production of NADPH and 5-ribose phosphate, which are an important precursor of lipid and nucleic acid biosynthesis in tumor cells [21]. Glycolysis also produces a large amount of lactic acid by competing for the energy supply of immune cells and inhibits the infiltration of immune cells [22,23]. Previous studies have also shown that lactic acid levels are inversely proportional to the survival time of patients [15]. To sum up, tumor cells achieve their own energy and material metabolism through glycolysis pathway, while avoiding the attack of immune cells, affecting the survival time and prognosis of patients.

In our study, we found that the higher glycolysis risk score, the less the number of macrophage M1, CD4+T cells and CD8+T cells. The massive proliferation of tumor cells utilizes the energy needed for the survival of immune cells and produces a large number of immunosuppressive substances [24,25]. It is known that the energy acquisition pathway of macrophages M1 is glycolysis. The capacitation pathway of macrophages M2 is mainly fatty acid oxidation and oxidative phosphorylation [26,27]. Glycolysis can "educate" macrophages into M2 type, which promotes their own growth, and inhibit M1 polarization with the characteristics of attacking tumor cells [28]. The macrophages M2 can also promote angiogenesis and tumor growth through HIF-1α, vascular endothelial growth factor (VEGF) and other factors [29]. This phenomenon leads to vascular system disorder and extremely hypoxic environment, which further limits the infiltration of immune cells. The differentiation, proliferation and infiltration of T cells require a large amount of energy supply [30]. Energy deficiency in tumor microenvironment can also limit the capacity of immature CD8+T cells and inhibit maturation [31]. High lactic acid environment also reduces proliferation,cytokine production and NAD+ levels, thus limiting the maturation of CD4+T and CD8+T cells and killing mature effector T cells [32]. To sum up, high glycolysis activities compete for the energy of tumor microenvironment and produce immunosuppressive factors, which limit the growth, polarization and maturation of immune cells, resulting in immunosuppression of tumor microenvironment.

The results of enrichment analysis showed that Glycolysis pathway, Fatty acid degradation, IL-17 signal pathway, Glutathione metabolism and Pathways in cancer, CAMs were enriched. In the past, most of the differential genes related to head and neck tumors were obtained from the differential analysis of malignant tumors and their matched normal tissues, so they were mostly enriched in the pathways related to tumor proliferation, migration and invasion, such as cell cycle, stress, hypoxia, epithelial differentiation, and partial epithelial-tomesenchymal transition (p-EMT) [33]. The DEGs used in our enrichment analysis come from the analysis of high and low score groups of glycolysis, so the enrichment pathway is not limited to those related to tumor proliferation, migration and invasion.According to the above discussion, it can be known that tumor cells give priority to glycolysis as the main way of energy acquisition, so it will lead to a large amount of glucose consumption, and the tumor microenvironment will enter a state of low glucose and high lactic acid. The main energy supply mode of the microenvironment changes from oxidative phosphorylation to fatty acid oxidation, so the fatty acid metabolic pathway is activated [34]. Glycolysis produces a large amount of NADPH, which can maintain a sufficient amount of reduced glutathione in tumor cells, which is also an important reason for tumor cells to develop chemotherapy resistance [35]. IL-17 cannot only promote the M2 interaction between tumor cells and macrophages through the CCL2-CCR2 axis, but also promote the expression of metalloproteinases (MMP9) and promote tumor metastasis [36]. The above enrichment results can indirectly explain the effects of glycolysis on tumor metabolism and microenvironment.

We have obtained five gene construction risk signature, of which IGSF11 has been shown to be associated with glycolysis of breast cancer [37], and the other four genes are associated with tumor migration and invasion [[38], [39], [40], [41]]. However, the relationship between these four genes and glycolysis has not been reported in the studies. Two external verification sets are used to verify the risk signature, and the prediction performance is good. As everyone knows,effective criteria can help clinicians to judge the effectiveness of treatment and the risk of recurrence can effectively avoid the occurrence of excessive diagnosis and treatment. For example, risk prognostic signatures can be used to predict the risk of distant recurrence, overall survival and sensitivity to adjuvant chemotherapy in early breast cancer [42,43]. In 2007, FDA approved a risk signature which was developed based on analysis of the expression data of 78 young patients with ER+ and LN- breast cancer. In the external verification set of 295 patients with invasive breast cancer, the risk signature performed well and the predictive value was significantly better than the traditional clinicopathological factors [44,45]. Our study can be used to judge the prognosis, immune cell infiltration and clinical immunotherapy effect through further assays.

Limitations: This study requires further experimental verification of functional immune system and glycolysis modification in vitro before clinical usage; at the same time, bigger, independent cohorts are needed to validate the signature; we also need to expand the cell lines of head and neck squamous cell carcinoma to verify the above conclusions in the future.

Conclusions

Combined with the results of RNA sequencing and single cell sequencing dataset, we can determine that the prognosis of patients with active glycolysis is poor. The microenvironment of tumor cells with active glycolysis reduced the infiltration of macrophages M1, monocytes and T cells, resulting in tumor cells from immune attack. Based on the above findings, targeted glycolysis is a more effective and reasonable treatment. In order to better serve the clinic, we build a prognostic signature related to glycolysis.The verification of the external verification set to determine the prognostic markers has better predictive efficiency and applicability, and can better serve the clinic.

Data availability

The datasets analysed during the current study are available in the GDC repository [TCGA-HNSC] [https://portal.gdc.cancer.gov/], GSE195832 [https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE195832] and GSE65858 [https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE65858].

Funding

This work was funded by improvement of Innovation ability of Hebei Otorhinolaryngology Clinical Medical Research Center (Grant No. 20577716D).

Ethics approval and consent

The data of this study were downloaded from the open database TCGA and GEO. Cancer tissues and paracancerous tissues are obtained from patients with Head and neck squamous cell carcinoma in the Second Hospital of Hebei Medical University. All experimental protocols were approved by the Research Ethics Committee of the second hospital of Hebei Medical University. Informed consent obtained from all patients.This study meets the requirements of the Helsinki Declaration.

CRediT authorship contribution statement

Qian Nie: Methodology, Software, Writing – original draft. Huan Cao: Data curation, Supervision, Writing – review & editing. Jianwang Yang: Software, Writing – review & editing. Tao Liu: Writing – review & editing. Baoshan Wang: Conceptualization, Methodology, Writing – review & editing.

Declaration of competing interest

The authors declare no competing interests.