A meta-validated immune infiltration-related gene model predicts prognosis and immunotherapy sensitivity in HNSCC

Head and neck squamous cell carcinoma (HNSCC), a common and aggressive malignancy with high morbidity and mortality, is one of the seven most common malignancies. Annually, there are about 800,000 new cases and more than 400,000 deaths worldwide [1]. Early-stage disease (stages I and II) is treated with single-modality surgery or radiotherapy contributing to high cure rates. However, due to the complex anatomy of head and neck, it is difficult to perform surgery. When patients are diagnosed with head and neck cancer, more than 50% of them are in clinical stage III or IV and lose their best chance of operation [2]. This is one of the reasons why the total global survival rate of HNSCC is only 50%. Besides, local recurrence or metastasis also leads to the poor prognosis of HNSCC.

Traditional treatments are not so effective for HNSCC. Even with aggressive therapy, loco-regional and distant recurrences after treatment are common and thus result in poor prognosis [3]. Despite the continuous innovation of treatment methods, there are still problems such as insufficient efficacy and excessive toxicity. With the advent of molecular targeted therapy, it is expected to replace cisplatin chemotherapy due to its less toxicity. The addition of the EGFR inhibitor cetuximab to radiotherapy has been shown to improve the prognosis of HNSCC patients compared with radiotherapy alone [4]. However, several recent studies indicated poor outcomes when cetuximab was given in HPV-associated HNSCC [5, 6]. Since traditional treatments and molecular targeted therapy cannot satisfy the treatment of HNSCC, immunotherapy has gradually attracted public attention. Taking the tumor heterogeneity and immune states of different individuals into account, it is necessary to identify the immune phenotypes of HNSCC to ensure that patients gain the maximum benefit from immunotherapy.

The tumor microenvironment (TME) is proved to be involved in tumor progression and treatment. Immune cells are most likely to be affected by TME [7]. Among them, tumor-infiltrating cells have attracted a lot of attention because of their duality and importance. They can target tumor cells and show anti-tumor activity. On the contrary, they can also exhibit pro-tumor activity and promote tumor development and metastases. In addition, regulatory T cells (Tregs) are considered to secrete suppressive cytokines such as TGF-β and IL-10, express cytotoxic T lymphocyte–associated protein 4 (CTLA-4), and significantly correlate with tumor progression in HNSCC [8]. Therefore, the investigation of TME in HNSCC to reveal the underlying mechanisms is important for the improvement of the diagnosis and treatment of HNSCC.

In the present study, we used weight gene co-expression network analysis (WGCNA) to identify immune infiltration-related gene modules in HNSCC and constructed a prognostic model based on LASSO Cox regression analysis. Nine genes in our risk model significantly influenced patients’ survival, and were effectively validated in the expression levels of mRNA and protein using GEPIA, HPA database and immunohistochemical method. We further investigated the landscape of immune infiltration, immunotherapy sensitivity and tumor mutation in two risk groups. Our results might help us deeply understand how TME affects patient’s clinical outcome and offer novel prognostic and therapeutic target of HNSCC.

In recent years, the tumor microenvironment has been regarded as a pivotal role in the progression of cancers including HNSCC [18]. The immune cells, stromal cells and extracellular components closely interact with tumor cells and form a complicated regulatory network to influence tumor growth and metastasis. The tumor often induces a suppressive microenvironment via impairing the function of both innate and adaptive immune cells to escape host’s immune surveillance [19]. Despite the application of traditional immune therapies in HNSCC, a large portion of patients show limited or no responses to current drugs. It is urgent and inevitable to find novel immune infiltration-related molecular targets in HNSCC tumor microenvironment.

Bioinformatic analysis has been widely used to investigate the tumor microenvironment profiles in various cancers. A recent study focusing on immune microenvironment of clear cell renal cell carcinoma identified critical immune subgroups via unsupervised consensus clustering and filtered hub genes from WGCNA modules [20]. Clustering procedures for modeling were applied in another study on the prognosis and immunotherapy response of lung squamous cell carcinoma [21]. LASSO regression analysis and multivariate Cox proportional model were selected as robust methods for the construction of prognostic gene signature [20, 22].

In our study, immune scores were calculated to estimate the infiltrating level of immune cells by ESTIMATE algorithm which was widely used to infer tumor purity. Compared with other immune infiltration-related HNSCC risk prediction models, our model not only has good predictive performance on prognosis, but also can predict patient response to immunotherapy. Unlike other validation methods using single or multiple datasets, we used a prognostic meta-analysis to test the applicability and stability of our model.

Our results suggested that patients with high immune scores had significantly ameliorated prognosis. We propose the hypothesis that enhanced immune infiltration levels in HNSCC promote anti-tumor responses and thus contain the tumor progression. On the contrary, low immune score indicating suppressed immune response can be a risk factor for the prognosis of HNSCC patients. Consistent with our findings, high immune score was significantly correlated with favorable survivals in gastric cancer and osteosarcoma [23, 24]. But elevated immune score can also indicate poor overall survivals as described in clear cell renal cell carcinoma [25]. It is speculated that the practical effect of immune infiltration on tumor microenvironment is attributed to not only the quantity of infiltrated immune cells, but also the functional activity and interactive patterns with the tumor.

The risk model we constructed consists of nine genes: MORF4L2, CTSL1, TBC1D2, WIPF1, CXCL13, TMEM173, C5orf15, LIPA, and ISG20. MORF4L2 is a component of the NuA4 histone acetyltransferase complex involved in the activation of oncogene and proto-oncogene-mediated growth induction, and replicative senescence, apoptosis, and DNA repair. Cathepsin L (CTSL1), a lysosomal cysteine protease member, is mainly involved in the terminal degradation of intracellular phosphorylated proteins [26]. Increasing evidences indicate that CTSL1 is highly and specifically expressed in various cancers [27, 28]. TBC1D2 is a GTPase-activating protein of Rab7 GTPase. In breast cancer cells, persistent Rac1 activity enhanced escape of β4 integrin from lysosomal degradation depending on actin-related protein 2/3 and TBC1D2 [29]. WIPF1, also known as the WASP-interacting protein (WIP), drives the oncogenic activity of mutant p53. Knockdown of WIPF1 in glioblastoma and breast cancer cells expressing mutant p53 reduced the proliferation and growth ability of cancer stem-like cells and decreased the expression of cancer stem-like markers such as CD44, CD133, and TAZ/YAP. WIPF1 knockdown inhibits the growth of glioblastoma tumor cells and breast cancer cells in vivo [30]. CXCL13 is a chemokine capable of promoting B cell migration [31]. Previous studies have shown that CXCL13 is associated with the prognosis of various cancers including oral squamous cell carcinoma and breast cancer [32, 33]. Over the past few decades, TMEM173 (also known as STING or STING1) was found to play an important role in the production of type I interferons and proinflammatory cytokines. STING1-dependent signaling networks regulate autophagic degradation and different patterns of cell death. Insufficient or overactivation of the STING1 pathway is associated with various pathological conditions, such as tumorigenesis, infection, disseminated intravascular coagulation, autoimmune disease and tissue damage [34]. Recently, TMEM173 was reported to correlate with the clinical status and immune response of HNSCC patients and can be used as a biomarker for improving prognosis [35]. C5orf15 (chromosome 5 open reading frame 15) is predicted to be an integral component of the membrane and haven’t been investigated yet. LIPA (lipase A) functions to catalyze the degradation of low-density lipoproteins to generate free fatty acids and cholesterol. Since hypoxia and hypermetabolism are characteristics of the tumor microenvironment, fatty acid turnover is usually high to meet the requirement of energy and biosynthesis [36]. Lipophagy may play a dual pro- and anti-tumor role. The expression of lysosomal acid lipase (LAL) was suggested to improve lipid metabolism and reduce metastasis in lung and liver cancer [37]. ISG20 is a kind of interferon-induced antiviral exoribonuclease mainly acting on single-strand RNA, and exerts antiviral activity against multiple RNA viruses in an exonuclease-dependent manner [38]. Whereas, high ISG20 expression was found to significantly associated with poor prognosis in liver cancer and clear cell renal cell carcinoma, which was proved to enhance angiogenesis, tumor cell proliferation and metastasis [39, 40].

In our results, the expression of multiple immune checkpoints differed between high-risk and low-risk groups based on our risk model. Blockade of PD1 with nivolumab or pembrolizumab produces durable antitumor efficacy in patients with recurrent or metastatic HNSCC, although only 15–20% of patients respond to treatment [41]. As a PD-1 inhibitor, pembrolizumab can be used in combination with cytotoxic chemotherapy for recurrent or metastatic HNSCC, and a recent clinical trial demonstrated promising clinical activity of pembrolizumab in combination with cetuximab in the treatment of recurrent or metastatic HNSCC [42]. Combined immunotherapy targeting PD-L1 and CTLA-4 has shown enhanced activity in several tumor types. However, further study found no statistically significant difference in OS between durvalumab plus tremelimumab treatment and standard treatment [43]. One study showed that increases in PD-1 and TIM-3 TILs during cetuximab treatment were inversely associated with response in HNSCC patients. Blocking these immune checkpoint receptors may enhance cetuximab-based cancer immunotherapy, potentially improving clinical outcomes in patients with HNSCC [44]. Using the TIDE algorithm, we found that the score of the risk model was significantly positively correlated with the TIDE score. In conclusion, immune checkpoint blockade (ICB) therapy has important value in the treatment of HNSCC, and our risk model has potential value in predicting patient response to it.

In results of immune infiltration analysis, memory B cells, CD8+ T cells, follicular helper T cells, and regulatory T cells were enriched in the low-risk group, while activated dendritic cells and activated mast cells elevated in the high-risk group. CD8+ cytotoxic T cells are capable of releasing granzymes and perforin to directly target tumor cells. Activated CD4+ or CD8+ T cells can also produce anti-tumor cytokines such as IFN-γ to inhibit tumor growth and recruit other immune cells.

It was reported that higher CD8+ tumor infiltrating T-lymphocytes were correlated with improved survival and predicted to be a favorable prognostic factor in HNSCC [45, 46]. Although Tregs are typically immunosuppressive and contribute to the immune escape of tumor, studies found that a high infiltration level of Foxp3+ Tregs was significantly associated with longer survival time of HNSCC patients, which were in accordance with our results [47]. The increased Foxp3+ Tregs in the low-risk group may indicate persistently enhancing immune responses and thereby inhibit tumor progression. The roles of tumor-infiltrating B cells in HNSCC haven’t been clearly elucidated yet since they are so few and excluded in most immune infiltration analysis. A study found that activated, antigen-presenting and memory B cells were enriched in the TME of HNSCC, and further suggested the dual effect of B cells due to their plasticity and heterogeneity [48]. Dendritic cells have been described as a strong antigen-presenting cells (APCs) and to mediate the activation of T cells [49]. However, few studies have explored their roles in HNSCC. The high level of activated dendritic cells in the high-risk group can be related to the attenuated inhibitory effect of Tregs to some extent. Mast cells are widely considered to produce regulatory cytokines targeting various immune cells to participate in anti-infective response, allergy and autoimmunity diseases. Low mast cell density was considered to associated with reduced survival in HNSCC [50].

We comprehensively analyzed the microenvironment and immune cell infiltration in HNSCC, and further built a nine-gene risk model to explore the prognostic value of immune infiltration-related biomarkers. These findings reveal the pivotal role of tumor microenvironment in HNSCC and can provide new molecular targets for the immunotherapy of HNSCC patients.