Multi-omics characterization of a scoring system to quantify hypoxia patterns in patients with head and neck squamous cell carcinoma

The data sources and analysis methods used in this study are shown in Fig. 1. Bulk tumor tissue RNAseq and somatic mutation information, copy number variation (CNV), clinical characteristics, and survival data were all obtained from The Cancer Genome Atlas (TCGA)(https://​tcga-data.​nci.​nih.​gov/​tcga/​, accessed 12 June 2022) and Gene Expression Omnibus(GEO) (http://​www.​ncbi.​nlm.​nih.​gov/​geo/​, accessed 12 June 2022). A total of 547 samples were collected from TCGA (503 HNSCC patients and 44 normal tissues), and 270 and 97 HNSCC patient samples were obtained from the GSE65858 and GSE41613 datasets, respectively (Table 1 and Table 2). The FPKM value from TCGA was converted to transcripts per kilobase million (TPM) and combined with GSE65858 as a data matrix. After excluding some cases with missing survival time, a total of 770 samples were obtained and used for follow-up and analysis.

Single-cell RNAseq analysis was conducted on five healthy donors and twenty-six patients with HNSCC from the GEO database (GSE139324) (Table 3). During data integration, ‘harmony’ package(version: 0.1.0) was able to mitigate batch effects [17]. The Expression data were normalized with the ‘NormalizeData’ function in the ‘Seurat’ R package (version: 4.1.1) [18]. The first 2,000 highly variable genes were identified using the ‘FingVariableGenes’ function of the ‘Seurat’ R package. Principal component analysis (PCA) was used to explore the sources of these highly variable genes via the ‘RunPCA’ function of the ‘Seurat’ R package, and cell subpopulations were annotated using the ‘singleR’ R package (version:1.10.0) and the marker genes from the previous study [19]. The distribution of the cellular components was plotted using the ‘TSNE’ R package (version: 0.16) [20]. The crosstalk of different cell subtypes in HNSCC was explored using ‘CellChat’ R package (version:1.5.0) [21].

A total of 49 HRGs obtained from the buffa hypoxia metagene dataset in the Molecular Signature Database (MSigDB) were used to identify different hypoxia-related patterns (version 7.4, https://​www.​gsea-msigdb.​org/​gsea/​msigdb/​, accessed 12 June 2022). The sample was clustered using the ‘ConsensusClusterPlus’ R package (version:1.60.0) and the hierarchical clustering method was repeated 1,000 times to ensure cluster stability [22].

Single-sample gene set enrichment analysis (ssGSEA) was used to quantify immune cell infiltration and function in each sample using the ‘GSVA’ R package (version:1.44.0) [23]. In addition, 48 immune checkpoints were obtained from prior studies [24, 25].Stromal/immune/Estimate scores were calculated with the ‘estimate’ R package (version:1.0.13) [26]. The correlation between the HLA-DPA1/CD4 expression and marker genes of various immune cell types was analyzed using the Tumor Immune Estimation Resource (TIMER) database (https://​cistrome.​shinyapps.​io/​timer/​, accessed 12 September 2022) [27]. In addition, the correlation between hub genes and immune cell infiltration was assessed using Gene Set Cancer Analysis (GSCA) (http://​bioinfo.​life.​hust.​edu.​cn/​GSCA/​#/​expression/​, accessed 12 September 2022) [28].

The ‘GSVA’ R package (version:1.44.0) was used to clarify the biological processes of different hypoxia clusters or cell subpopulations [23]. Hypoxia-related pathways were presented as heatmaps in the hypoxia clusters and bar plots of different cell subpopulations (all adjusted P < 0.05).

First, three hypoxia clusters were identified based on the expression of 49 HRGs using consensus clustering analysis. Second, differentially expressed genes (DEGs) in the three clusters were retrieved using the ‘limma’ R package (version: 3.52.1). A total of 688 DEGs were obtained after the intersection among the 3 hypoxia clusters (adjusted P < 0.001) and evaluated using univariate Cox regression analysis. Of these, 207 DEGs associated with disease prognosis were obtained and used to further analysis (P < 0.05). The hypoxia score (HS) of each sample was calculated based on the PCA analysis as follows:

‘i’ is the expression of the 207 prognostic DEGs. The HS of each patient in the GSE41613 and each cell in the GSE139324 datasets was designed using a similar formula, respectively.

The response of the HHSG and LHSG groups to drug therapy and immunotherapy was compared. The ‘pRRophetic’ R package (version: 0.5) was used to determine the concentration of various chemotherapy, targeted inhibitor, and immunotherapy drugs that caused a 50% reduction (IC50) in growth [29]. Immunophenoscores (IPS) for HNSCC were then downloaded from the Cancer Immunome Atlas (TCIA) (https://​tcia.​at/​, accessed 20 June 2022) [30]. The IPS of the HHSG and LHSG were compared to predict sensitivity to immunotherapy. In addition, the Tumor Immune Dysfunction and Exclusion (TIDE) algorithm (http://​tide.​dfci.​harvard.​edu, accessed 20 June 2022) was used to predict HNSCC patient response to immune checkpoint inhibitors [31]. A total of 14 cisplatin-sensitive and 4 cisplatin-resistant HNSCC cell lines from the GSE102787 dataset were used to assess the ability of the HS to predict chemotherapy responses.

The weighted gene co-expression network analysis (WGCNA) R package (version:1.71)was used to obtain the key prognostic HRGs in the HNSCC [32]. Based on the TCGA-HNSCC gene expression matrix, the suitable power exponent was selected and the adjacency matrix was converted to the topological overlap matrix. Then, the correlation analysis between the gene consensus modules with HS was performed, and the top 2 correlation coefficient modules positively correlated with HS were selected for further analyses. After intersection of the genes obtained from the 2 module and the 49 HRGs and a careful literature search, the key prognostic HRG was identified [30]. The Human Protein Atlas (HPA) database was used to compare key prognostic HRG protein expression between normal and tumor tissue in HNSCC.

SCC-4 cells were purchased from BeNa Culture Collection (Xinyang, China) and cultured in 90%Dulbecco’s Modified Eagle Medium/Nutrient Mixture F-12(DMEM-H/F12; Procell Life, China) supplemented with 10% fetal bovine serum (FBS; Procell Life, China) and 400 ng/ml hydrocortisone. The SCC-4 cells were incubated with 5% carbon dioxide at 37 °C, plasmid transfection was carried out using Lipofectamine 3000 reagent (Thermo Fisher Scientific, China).

As described previously, western blot analysis was performed [33]. In brief, after the total protein was extracted, the BCA protein concentration reagent is quantified and then subjected to SDS-PAGE. A 20 μl protein sample was then transferred to a PVDF membrane, which was repeatedly washed before adding chemiluminescent solution (ECL, Thermo Fisher Scientific, China) to image the protein.

Cell growth and viability were measured using a Cell Counting Kit-8 (CCK-8) kit (Solarbio, Beijing, China), cells in a 96-well plate, about 2 × 10³ cells in each well. After 48 h incubation, 10 μl of CCK-8 was added to each well and incubated at 37 °C for 1 h. The optical density at 450 nm of each well was measured once at 0,24, 48, 72 and 96 h.

Cell migration analysis was performed according to the descriptions in a previous study [34]. Briefly, all the groups of cells were first resuspended in medium and the transwell chambers (Corning, USA) were removed after 24 h of incubation, followed by fixation of the invading cells with 4% paraformaldehyde and staining with 0.1% crystalline violet. Finally, transwell chambers were inverted and photographed under a microscope.

All statistical analyses in this study were conducted using R language software (version 4.2.1) (Bell Laboratories, formerly AT&T, now Lucent Technologies). The ‘maftool’ R package (version 2.12.0) and ‘rcircos’ R package (version 1.2.2) were used to draw a waterfall plot of the gene mutation status of different clusters and illustrate the position of the HRG CNV on the chromosomes, respectively [35, 36]. Variables in data analysis including continuous(e.g., HRGs expressions, HS, OS time) and categorical variables(e.g., survival status). Kaplan-Merier curves are presented and the differences in the prognosis of different types of patients between the curves are tested using log-rank test. Wilcoxon tests were used to compare measurements between two groups(e.g., the 49 HRGs expressions between normal and cancer, TMB, stromal/immune/ESTIMATE scores between LHSG and HHSG). In comparison, Kruskal–Wallis tests were used for the comparison among three or more than three groups (e.g., the immune infiltration, immune-related function scores and the immune checkpoints expressions among three hypoxia clusters, HS among the three hypoxia clusters, three gene clusters, ten cell subpopulations). Correlation analysis was performed using the Spearman test. A P < 0.05 was considered statistically significant.

Analysis showed that CNV was common among HRGs associated with HNSCC. While CNV is most often characterized by a reduction in copy number, more than half of the HRGs showed an increase in copy number (Fig. 2A). Ninety-nine (22.86%) of the 433 samples were found to contain HRG mutations, of which ESRP1 was the most frequently mutated gene (Fig. 2B). CNV was also detected on these HRGs (Fig. 2C). Gene expression in normal and cancerous tissues was assessed and CNV was shown to affect HRG expression in HNSCC. While ANKRD376 expression was significantly lower in cancer than in normal tissues, all other HRGs, except TUBA1A and ESRRP1, were significantly higher in cancer tissues (Fig. 2D). In addition, most HRGs were highly expressed in tumor tissues, suggesting that they may contribute to the occurrence and development of HNSCC.

Hypoxia-related gene expression profiling of 770 HNSCC samples was obtained from TCGA and GSE65858 datasets. Univariate COX and correlation analysis were used to explore the interactions, connections, and effects of these genes on disease prognosis. Of these, 40 HRGs were risk factors (HR > 1), and two were protective factors (HR < 1) for HNSCC prognosis (all P < 0.05). Except for a significant negative correlation between HK2 and MAD2L2 expression, other prognosis-related HRGs were positively correlated (Fig. 3A). This finding suggests that there is possible crosstalk between these HRGs, which may correlate with disease prognosis and tumor heterogeneity among HNSCC patients.

A consensus cluster analysis of 770 samples was conducted to further explore the relationship between patient prognosis and the hypoxic signature and three HNSCC subclasses were identified, defined as hypoxia clusters A, B, and C (Fig. 3B). Of the clusters, hypoxic cluster C was associated with the best prognosis, while hypoxia cluster B was linked to the worst (P < 0.05) (Fig. 3C). Heatmap results showed that hypoxia clusters C and B had the lowest and highest HRGs expression, respectively (Fig. 3D). In contrast, hypoxia cluster A had the second-best prognosis with intermediate HRGs expression. Collectively, these results suggested that higher HRG expression was associated with poor HNSCC patient prognosis and that HRGs may play a key role in the malignancy of this disease.

To better understand differences in the tumor microenvironment (TME) associated with the three hypoxia clusters, ssGSEA was used to assess immune cell infiltration and function in each cluster. Significant differences in the infiltration of 23 immune cell types, except for gamma delta T cells, were observed among the three hypoxia populations(all P < 0.05) (Fig. 3E). The TME of hypoxia cluster C was highly infiltrated by immune cells, including immunosuppressive cells such as myeloid-derived suppressor cells (MDSCs), macrophages, mast cells, and regulatory T cells (Tregs). A previous study found that hypoxia was linked to tumor immunosuppression by affecting immune cell infiltration [37]. MDSCs and Tregs are shown to be very active in highly hypoxia TME [11]. An additional study found that hypoxia can promote the development of malignant macrophages that aid tumor development [38]. High infiltration of immunosuppressive cells was associated with the poor prognosis of patients with HNSCC, pancreatic adenocarcinoma, esophageal cancer, or colon cancer [39‐41]. The ranking of several cells with anti-tumor effects, including CD4 + T cells, CD8 + T cells, and natural killer (NK) cells, from higher to lower infiltration, were hypoxia cluster C, hypoxia cluster B and hypoxia cluster A (P < 0.05), possibly explaining why patients with three hypoxia clusters had the different prognosis.

APC inhibition, APC stimulation, CCR, immune checkpoints, cytolytic activity, HLA, inflammation promotion, T cell inhibition, T cell stimulation, and type II IFN responses also differed among the three clusters (Fig. 3F), and each of these processes was highly enriched in hypoxia cluster C patients with HNSCC, while the parainflammation was highly enriched in hypoxia cluster A. Given the increased use of immune checkpoint inhibitor (ICI)-based therapies, differences in immune checkpoint (IC) gene expression were also compared among the three clusters. The expression of CD274, TNFRSF25, VTCB1, TNFSF18 and other ICs differed considerably among the clusters (Fig. 3G). KEGG-related GSVA was conducted to explore differences in the biological behaviors of the three hypoxia clusters (Fig. 3H). Diverse signaling pathways associated with inflammation, ebb signaling pathway, pathogenic escherichia coli infection were significantly enriched in cluster A. Elevated RNA degradation, cell cycle activity, spliceosome function, nucleotide excision repair, and mismatch repair reflected the high degree of hypoxia in cluster B while signaling pathways related to immunity, such as primary immunodeficiency, and the immune network for IgA production was significantly enriched in cluster C.

Taken together, these results showed that immune activation is strong in hypoxia cluster C, the inflammatory response is strong in hypoxia cluster A, and the degree of hypoxia is especially high in hypoxia cluster B. Hypoxia cluster A, Band C is the ‘immune-inflamed’, ‘immune-cold’ and ‘immune-hot’ phenotype, respectively. The consistency of the immune and prognostic profiles among the three groups supported this classification method as scientific and reasonable.

To identify the potential biological behavior of each hypoxia phenotype, 5,258 DEGs were identified in hypoxia cluster A vs. B, 4,113 DEGs were identified in hypoxia cluster A vs. C, and 6,915 DEGs were identified in hypoxia cluster B vs. C, respectively, using the ‘limma’ R package. An additional 688 overlapping DEGs were obtained through intersection analysis.

To further identify other phenotypic differences caused by hypoxic patterns, the overlapping DEGs were used to stratify patients into three DEG-related hypoxia clusters (gene clusters A–C) using an unsupervised cluster (Fig. 4A). Survival analyses showed that patients in gene cluster B had the worst prognosis (P < 0.001) (Fig. 4B). Most patients in gene clusters A, B, and C corresponded to hypoxic group clusters A, B, and C, respectively (Fig. 4C). Except for ANKRD37, the expression of all 48 HRGs differed significantly among the different gene clusters (Fig. 4F). Thus, the gene clusters corresponded well with the hypoxia-related clusters.

Considering the heterogeneity and complexity of individual hypoxia patterns, hypoxia patterns were quantified in all HNSCC patients using PCA and the results were expressed as an HS. After obtaining the HS optimal cutoff value(−7.012175) using the ‘maxstat’ R package (version:0.7–25) based on the TCGA and GSE65858 database, patients with HS values ≥ the optimal cutoff value were defined as the high HS group (HHSG) while those with HS values < the optimal cutoff value were defined as the low HS group (LHSG) [42]. Patients in the HHSG had a better prognosis than those in the LHSG using TCGA, GSE65858 (P < 0.05) (Fig. 4D) and GSE41463 (P < 0.05) data (Fig. 4E). Thus, HS has strong predictive power for the prognosis of patients with HNSCC.

To determine the relationship between HPV16 infection status, hypoxia clusters, gene clusters, and HS in each patient, HPV16 positive, negative, and unknown patients were classified into three hypoxia clusters and then divided into three gene clusters. Most patients in gene clusters A and C belonged to HHSG, while most patients in gene cluster B belonged to LHSG (Fig. 4G).

To further explore the correlation between immune cell infiltration in the TME and HS, immune cell infiltration and immune cell function scores were calculated in each sample using ssGSEA and a correlation analysis was conducted using the HS (Fig. 4H). There was a significant positive correlation between the HS and both immune cell infiltration and function. In addition, significant differences in the HS were observed among the three hypoxia clustering modes. Hypoxia cluster C had the highest median HS values, while hypoxic cluster B had the lowest median HS values (all P < 0.001) (Fig. 4I). Similar results were found in gene clusters C and B (all P < 0.001) (Fig. 4J). In summary, patients with hypoxia cluster C, gene cluster C and HHSG had the best prognosis, whereas those with hypoxia cluster B, gene cluster B, and LHSG had the worst prognosis. These results not only demonstrate the consistency of the predictive effect but also suggest that quantifying hypoxia status can help to predict patient prognosis and assess their immune-related characteristics.

Tumor mutation burden (TMB) is a key criterion for successful immunotherapy [43].Thus, a series of analyses were conducted on somatic mutations associated with HNSCC. No differences in TMB were observed between the HHSG and LHSG groups (P > 0.05) and there was no significant association between the HS and TMB (P > 0.05) (Fig. 5A, B). The low HRG mutation frequency may account for these findings. Given that TMB is a prognostic factor in many types of cancer, a role for TMB and HS in HNSCC prognostic status was explored [44]. Prognosis was shown to be significantly higher among patients in the low TMB group than in the high TMB group (P < 0.05) (Fig. 5F). A German multi-center retrospective study found similar results [45]. Further results of the stratified prognostic analysis showed that patients with HHSG and low TMB had a large survival advantage (Fig. 5G). Since the frequency of HRG mutations in HNSCC is not high, a particular gene mutation in HNSCC was assessed and the top 20 driver genes with the highest mutation frequency were visualized. TP53 had the highest mutation frequency in both groups, followed by TTN (Fig. 5D, E).

The TME is critical to understanding the tumorigenesis and progression of many cancer types. Thus, stromal/immune/Estimate scores were calculated based on the ESTIMATE algorithm and the relationship between these scores and the HS was explored in patient prognosis. Findings showed that the immune/Estimate scores were higher in the HHSG than in the LHSG (all P < 0.05) (Fig. 5C). Next, patients were divided into high and low groups based on the median values of their stromal/immune/Estimate scores. The stratified prognostic analysis showed that those with a combination of HHSG and high stromal/immune/Estimate scores had a significant survival advantage (Fig. 5H–J). These data suggest that HS is a good prognostic indicator and that its combination with TMB and stromal/immune/Estimate scores can help to further refine patient prognosis.

To assess the value of HS in predicting the clinical efficacy of HNSCC, the sensitivity of different drugs among patients in the HHSG and LHSG were analyzed using TCGA, GSE41613, and GSE65858 data. Using intersection analysis, 11 drugs were obtained with median IC50 values that differed between the two groups (Fig. 6). While the median IC50 was lower for the HHSG than the LHSG for rapamycin, the median IC50 was significantly higher for the HHSG than the LHSG for the other 10 drugs (all P < 0.001).IDO1, LAG3, GTLA4, ICOS, TIGIT, PDCD1, TNFRSF8, TNFRSF14, IDO2, CD40, TNFRSF4, TNFSF14, CD244, HAVCR2, CD27, BTLA, LGALS9, CD48, CD40LG, ADORA2A, and CD200R1 had higher expression in the HHSG than in the LHSG (all P < 0.05), while TNFSF9, CD270, CD274, CD44, TNFSF18, CD200, NPR1 had lower expression (Fig. 6M). Studies have demonstrated the ability of IPS to predict the immunotherapy response of patients with colon cancer and uterine corpus endometrial carcinoma [30, 46]. Thus, differences in IPS were assessed between HHSG and LHSG. The IPS (Fig. 7A), IPS-PD1/PD-L1/PD-L2 (Fig. 7B), IPS-CTLA4 (Fig. 7C), IPS-PD1/PD-L1/PD-L2 + CTLA4 (Fig. 7D), and TIDE scores (P < 0.05) (Fig. 7E) were significantly higher in the HHSG than in the LHSG (all P < 0.05). Meanwhile, the HS was significantly higher for the cisplatin-resistant group than for the cisplatin-sensitive group (P < 0.05) (Fig. 7F). Thus, HS has a strong ability to predict the effects of targeted drugs, immunotherapies, and chemotherapeutics.

As WGCNA is a powerful tool to find key genes that are highly correlated to corresponding modules, we adopted it to identify the key genes with high correlations with HS. Firstly, number 8 was chosen as the soft threshold (Additional file 1:Figure S1A). Subsequently, we obtained 9 modules (Additional file 1: Figure S1B). Thetwo modules, including black and brown,had significant positive correlations with HS, while the other four modules(green, blue, turquoise, grey) had negative correlations with HS(all P < 0.001, Additional file 1: Figure S1C). We selected the black and brown modules for further screening. After the intersection between module genes and HRGs was performed, CHCHD2 was finally identified as a key gene. A previous study found that over-expression of CHCHD2 could promote the expression of HIF-1α to adapt the hypoxia microenvironment in non-small cell lung cancer [47]. However, the exact mechanisms between CHCHD2 and hypoxia in HNSCC is still not clear. Then we further explored the function of this gene in HNSCC. Based on the HPA database, CHCHD2 was highly expressed in HNSCC tissue and lowly expressed in normal tissue (Fig. 8A). Meanwhile, patients with higher expression of CHCHD2 had a poorer prognosis than the patients with lower expression (all P < 0.05, Fig. 8B). Besides, in vitro experiments showed that overexpression of the CHCHD2 promoted the migration and invasion of HNSCC cells (Fig. 8C–E). Taken together, it seems worthwhile to further explore the specific roles of CHCHD2 in hypoxia-driven HNSCC progression.

A single-cell sequencing-based analysis was performed to further reveal the hypoxia mechanism in different immune cells. Immune cell samples from five tonsil tissues, thirty-one peripheral blood mononuclear cells (PBMC), and 26 tumor-infiltrating lymphocytes (TIL) were obtained from the single-cell sequencing dataset, GSE139324. A total of 121,347 cells from 62 samples were assessed, of which 99,261 were from patients with HNSCC and 22,086 were from the normal population (Fig. 9A–B, Additional file 2: Figure S2A–B). Subsequently, 19 clusters were presented by graph-based t-distributed stochastic neighbor embedding (TSNE) (Additional file 2: Figure S2C), and the top five markers of the 19 clusters were presented in dotplot(Additional file 2: Figure S2D). Based on the expression levels of marker genes those obtained from previous literature (Additional file 2: Figure S2E), ten cell subpopulations were identified and shown in TSNE (Fig. 9C). Besides, the fraction of different cell subsets in normal and tumor, PBMC, tonsil, and TILs,19 clusters were shown in Fig. 9E–G, respectively. Each cell was assigned an HS using the same PCA method (Fig. 9D). The HS of the CD16 + cells, Dendritic cells,pDC cells, B cells, and CD4 + Tconv cells in the tumor tissue were lower than those in normal tissue (P < 0.05), while NK cells, CD8 + cells, and CD14 + cells vice versa (P < 0.05) (Fig. 9H). The HS of CD4 + Tconv cells was significantly lowest, while pDC cells was highest in the ten cell types (Additional file 3: Figure S3A, B). The GSVA analysis was used to further clarify the biological functions of different cell subpopulations (Additional file 3: Figure S3D). GSVA-based biological analysis showed that hypoxia pathways were active in the CD16 + cells, and Dendritic cells, and inhibited in the CD4 + Tconv cells, B cells, and CD8 + cells. Collectively, these results may suggest potential targets and mechanisms of the hypoxia in the tumor immune microenvironment (TIME) of HNSCC.

To further investigate the interaction network in the TIME in HNSCC, we used ‘Cellchat’ R package to uncover alterations in cell crosstalk between normal tissues populations and HNSCC. Compared with the normal tissues, the overall number and strength of interactions were all increased (Fig. 10A), and the number and strength of most of the interactions among all the immune cells were increased and stronger, respectively (Fig. 10B). The above results indicate TIME is a complex milieu. Then we further compared the signals patterns between tissue from normal populations and HNSCC. The overall signal patterns of normal and HNSCC were cleanly presented in Fig. 10C. For example, the SELPLG signal strength was ranked as the first in the CD4 + T conv cells derived from HNSCCs. Next, the focus was set on the CD4 + T conv cells as its lowest HS among all the immune cells and hypoxia-related function. Compared with normal tissues, we found the most significant changes number of ligand-receptor between the CD4 + T conv cells were pDC cells in the HNSCC, and the number of ligand-receptor among these immune cells was increased (Fig. 10D). Furthermore, we analyzed the receptor ligands that may regulate communication between CD4 + T conv cells and other immune cells. For example, CD4 + T conv communicated with pDC cells using the CD74 + CXCR4, CD74 + CCD44, which was not found in normal tissues (Fig. 10E), suggesting the role of CD4 + T conv cell-derived CD74 in the progress of HNSCC. A previous study found that CD74 and migration inhibitory factor (MIF) plays a vital role in HNSCC progression [48]. Besides, Zhu et al. [49] revealed that MIF bind to CD74/CXCR2, CD74/CXCR4 could play a vital role in the hypoxia-induced tumor growth in HNSCC. Thus, the association of hypoxia and the communication between CD4 + T conv cells and pDC cells deserves further investigation.

As CD4 + T Conv cells communicated with pDC cells mainly using the MHC-II pathway (Fig. 10E), we further investigated the MHC class II (MHC II) pathway gene expression in different immune cells between normal tissues and HNSCC. Compared with the normal tissues, the expression of HLA-DPA1 was increased significantly in the CD4 + T Conv cells, while receptor CD4 was also increased significantly in the pDC cells (Fig. 11A), explaining the reason for the construction of the MHC-II pathway between CD4 + T Conv cells and pDC cells in the tissues from HNSCC. HLA-DPA1, an MHC II molecule, is primarily expressed on professional antigen-presenting cells and is responsible for presenting exogenous peptide antigens to CD4 + T cells [50]. CD4 is the marker gene of CD4 + T cells. Given the close relations between MHC II, CD4 and immune, we adopted the TIMER database to analyze the relationship between HLA-DPA1 and CD4 expression and representative markers of different immune infiltrating cells in patients with HNSCC, respectively. The adjusted results based on tumor purity showed a significant correlation between HLA-DPA1, CD4, and all immune cell types (Table 4) (all P < 0.05). To validate this result, different algorithms were adopted using the GSCA website and ssGSEA algorithms. All immune cells except mucosal-associated invariant T were significantly associated with HLA-DPA1 and CD4 expression in HNSCC patients (Table 5) (all P < 0.05), and the ssGSEA algorithms also revealed that all 23 immune cells infiltration abundance had significant associations with both the expression of HLA-DPA1 and CD4 (Additional file 3: Figure S3C) (all P < 0.05). These results demonstrate the tight link between HLA-DPA, CD4, and HNSCC-specific immunity. Additionally, recent studies suggest that tumor-specific MHC-II may also function as a tumor suppressor, and expression is associated with favorable outcomes in lung cancer and pan-cancer patients [51‐53]. Besides, CD4 + T cells play an anti-tumor role in human bladder cancer [54], and the higher levels of CD4 + T cells infiltration had a better prognosis than the lower levels in sarcoma patients [55]. Therefore, it is reasonable to believe that HLA-DPA1 and CD4 may play antitumor effects in HNSCC. We further analyzed the association between HLA-DPA1 and CD4 and prognosis. Survival analysis defined patients with HLA-DPA1 and CD4 mRNA expression levels  ≥ optimal cutoff value (of 7.7662 and 6.175818, respectively) as the high group and those with expression < optimal cutoff value as the low group. Patients in the high group of HLA-DPA1 had higher OS than those in the low group in TCGA, GSE65858 (P < 0.05), and GSE41613 datasets (P < 0.05) (Fig. 11D–E), while the same results were also validated in CD4(Fig. 11G–H). Besides, we divided the patients into HLA-DPA1 high/CD4 high and the other three groups. The survival result showed patients with low expression of HLA-DPA1 and high expression of CD4, high expression of HLA-DRA1 and high expression of CD4 had better prognoses than the other two groups (Fig. 11F, I). Altogether, the above results support the hypothesis that HLA-DPA1/CD4 axis plays antitumor effects in HNSCC. The HPA database was used to compare HLA-DRA protein expression in normal and HNSCC tissues. Interestingly, HLA-DRA was low expressed, while CD4 was weakly expressed in HNSCC tissues, both HLA-DPA1/CD4 were not detected in the normal tissues (Fig. 11B–C). The positive correlation of the HLA-DPA1/CD4 expression with some immune cells with anti-tumor effects such as NK cells may explain this phenomenon.