Oral cancer is a malignant tumor with suppressed immune surveillance. PD-L1 was highly expressed in oral cancer cells and involved in immune escape [11]. In 2017, nivolumab targeting PD-1 was approved to treat advanced oral cancer with metastasis or recurrence [12]. Two years later, pembrolizumab, another PD-1 antibody come into use for oral cancer [13, 14]. However, therapeutic effects of these PD-1 inhibitors would be mitigated in patients with certain characteristics, such as high expression of inhibitory T-cell receptors [4]. Thus, it is imperative to identify immune-related markers for more individualized immunotherapy of oral cancer.
To enhance the robustness of prediction, IRGPs were applied to be integrated to the prognostic signature. Herein, gene expression data were not required in this model, instead the relative expression of two genes were needed. The application of IRGPs brought about two advantages for prognostic prediction. Firstly, we did not have to perform standardization for gene expression from different platforms. Secondly, the effect of technical bias of different platforms was minimized on gene expression.
At first, 17 IRGPs associated with prognosis of oral cancer were screened out with DEG and LASSO analyses and integrated into a risk score model. The oral cancer patients were separated into high- and low-risk groups based on the critical risk score point. And a higher overall survival rate was detected in low-risk patients. Besides, the mortality risk increased with higher riskScore. Subsequently, univariate and multivariate Cox analysis was performed to assess the correlation between IRGP-model and clinical parameters including age, gender, stage, tumor and node status. The results revealed that IRGP-model can be considered as an independent prediction factor for oral cancer prognosis.IRG-based model was widely applied in stratification for various cancers, such as low-grade glioma, ovarian cancer and melanoma [15-17]. Of note, multiple studies have developed prognostic models for oral cancer frequently based on IRGs. For example, Yao et al. integrated four IRGs and node status to develop a prognostic model for head and neck squamous cell carcinoma (HNSCC)[18]. Huang et al. stratified patients with oral squamous cell carcinoma into high- and low-risk groups based on 9 IRGs [19]. However, there is on study based on IRGPs for oral cancer. Zhang et al. constructed a 14-IRGPs signature for HNSCC with a relatively dismal predictive performance (AUC=0.7<0.75). In our study, the 17-IRGPs signature had a good predictive performance for oral cancer (AUC=0.774>0.75).
In the signature, the top 3 IRGPs strongest correlated with riskScore were TNFRSF12A|TNC (positive), PLAU|DEFB1 (positive) and OASL|SPP1 (negative). Given the negative correlation between riskScore and prognosis, TNFRSF12A, PLAU and SPP1 were correlated with prognosis negatively. Whereas, TNC, DEFB1 and OASLwere correlated with prognosis positively. Meanwhile, mutation analysis revealed that TNC was identified as the most frequently mutated gene in high-risk group. Accumulating studies have reported the association of these genes with prognosis in various cancers. Several studies have reported that the high expression of TNFRSF12A andPLAU were involved in worse outcome for HNSCC patients [18, 20]. Xu et al. found that SPP1 was increased and TNC was decreased in OSCC tissues[21]. The up-regulation of SPP1 was involved in worse outcome for HNSCC patients. And the level of TNC expression was decreased with higher stage [21]. However, Chi et al. identified the up-regulation of TNC and OASL in saliva samples from OSCC patients [22]. This is rational that the expression of genes are different in different tissues. Human beta-defensin-1 (DEFB1) has been reported to be a potential tumor suppressor in prostate and renal cancer [23, 24]. And recently, DEFB1 was confirmed to be positively and independently associated with OSCC survival [25]. In addition, the abnormal expression of most other genes have been also reported in various cancers, including oral cancer [26-28]. Collectively, most genes in the model have been investigated in various cancers. In our study, the integration of these 17 gene pairs suggested the important role of immune system in oral cancer.
Considering the critical role of TME in cancers, we investigated the abundance of infiltrative immune cells in oral cancer. Among the 22 types of immune cells, 10 types were significantly different between high- and low-risk groups. Many studies have explored the prognostic values of immune cells. Song et al. has demonstrated that naïve B cells and regulatory T cells (Tregs) indicated a favorable survival in head and neck cancer [29].Whereas, eosinophils and activated mast cells indicated a poorer outcome [29]. Besides, Eosinophils have been reported to be involved in metastasis and negatively associated with cancer prognosis [29, 30]. Consistently, in our current study, after grouping based on our model, the abundance of naïve B cells and regulatory T cells (Tregs) was significantly enhanced in low-risk patients. And eosinophils and activated mast cells were highly expressed in high-risk group. Memory T cells are reported to play roles in eliminating tumor cells and activated memory CD4 T cells indicates an improved survival in several cancers [31, 32]. We also identified the high enrichment of activated memory CD4 T cells and CD8 T cellsin low-risk group. In addition, we detected the significant difference of mast cells, plasma cells and naïve CD4 T cells between high- and low-risk groups. Mast cells could influence tumor progression by regulating inflammation[33]. Activated mast cells was found to be associated with poor prognosis of several cancer [33, 34]. Similarly, in our study, activated mast cells were enriched in high-risk group, whereas resting mast cells were highly expressed in low-risk group. The results were consistent with subsequent GSEA analysis that various immune-related GO terms and signaling pathways were enriched in low-risk group, including T cell activation involved in immune response, positive regulation of T cell proliferation, adaptive immune response, natural killer cell mediated cytotoxicity, cell adhesion molecules cams, T cell receptor signaling pathway, etc. These results demonstrated that the immune cells significantly enriched in low-risk group may play promising roles in improving the prognosis of oral cancer, which awaits further investigation.
Tumor mutation burden (TMB) has been investigated as a promising biomarker in various cancers [35, 36]. SNPs is a common type of gene variation, which are caused by point mutations. SNPs has been reported to be associated with tumorigensis and prognosis of cancers, including oral cancer [37, 38]. And TNC was identified to be the most frequent mutated gene among the 17 IRGPs in oral cancer. There is a possibility that the SNPs affected the immune cells infiltration in oral cancer based on a previous study [39]. To our best knowledge, this is the first study that constructed a IRGPs-based prognostic model for oral cancer and comprehensively analyzed infiltration of immune cells and TMB. However, there are several limitation in our study. Although the 17-IRGP signature were constructed and validated based on TCGA database, an individual database should be introduced to validate our model. Our study is retrospective, and needs to be corrected for clinical application.