Background
Gastric cancer is a common cancer and a leading cause of cancer-related deaths worldwide [
1]. The clinicopathologic characteristics are routinely revealed though Lauren/WHO classification and tumor-node-metastasis (TNM) staging system for prognostication which is also critical for the selection of appropriate treatment [
2]. However, gastric cancer is a heterogeneous disease, and its outcomes can vary significantly even for patients with similar clinical features and treatment regimens, suggesting that clinicopathologic characteristics and current classifications are insufficient for prognostication and risk stratification [
3,
4]. Hence, identification of novel markers providing more predictive value is highly demanded for improving the prognostication for gastric cancer.
Gastric cancer tissue is highly heterogeneous, where malignant cells are in an intricate relationship with tumor microenvironment, including immune cells, vessels and fibroblasts [
5‐
7]. Either through structural and functional abnormality of tumor vasculature or deterioration of the diffusion geometry of blood vessels, the vessels and fibroblasts cells of tumor microenvironment influence O
2 perfusion and diffusion, and therefore, leading to the development of hypoxia in that tissue area [
8]. Hypoxia has been reported as one of characteristic hallmarks of solid tumors that directly contribute to the malignant properties of cancers, including tumor progression, invasion and metastasis [
9‐
12]. Meanwhile, immune cell is also a potentially powerful force that can prevent or slow tumor growth, which is associated with tumor invasion and metastasis [
13‐
15]. Interestingly, increasing evidences have found that the direct or indirect interaction between hypoxia and immune status in gastric cancer microenvironment [
16,
17], although their underlying mechanisms remains unclear.
In this study, we speculated that immune and hypoxia interaction could provide prognostic value for gastric cancer patients. Through a series of systematic analyses, we developed a novel gene signature by incorporating immune and hypoxia status into the current clinicopathologic characteristics and staging system, aimed to improve the prognostication of gastric cancer.
Discussion
Considering the widely varying prognostic outcomes of gastric cancer, it is of great importance to establish a robust classifier to stratify patients with different risks and prognoses, which is critical to maximize the benefits brought by the personalized treatment and timely follow-up. In this study, the comprehensive mining of the transcriptional profiles and microenvironment characteristics was aimed to construct a tool to help address this important clinical issue. We found that both the hypoxia and immune status of the tumor microenvironment was associated with gastric cancer patient survival. Moreover, the inverse effects of hypoxia and immune status were significantly associated with prognosis, even after stratifying patients by clinicopathologic risk factors. Finally, a hypoxia-immune-based three-gene signature was developed as a prognosis classifier, with promising performance in risk stratification among the discovery cohort and three independent cohorts. These findings represent a new insight to improve discussions on patient prognostication and stratification through considering the microenvironment characteristics and transcriptomics.
The immune and hypoxia microenvironment plays a critical role in the tumorigenesis and progression of gastric cancer [
10,
11,
13‐
15,
25]. On one hand, the immune incapability in the tumor microenvironment has been reported as an essential mechanism for solid cancers to evade from host immunity [
13,
15,
25‐
27]. It was found that higher expression of immune-related gene predicts better prognosis in both EBV-positive and EBV-negative gastric cancer patients [
25]. Also, the estimation of immune cells in tumor tissues could improve the accuracy of TNM staging system for prognostication in gastric cancer [
28]. On the other hand, the hypoxic microenvironment promotes tumor malignancy by activating angiogenesis and increases cell migration and expansion toward cancer stem cell phenotype by altering cell skeleton and extracellular matrix [
10‐
12]. These findings were similar with the results in current study. We found that the protective DEGs, mainly containing the immune-related DEGs, could take part in the activation of immune cells, migration of immune cells and release of inflammatory factors and those risk DEGs, of which the majority was hypoxia-related DEGs, were associated with cytoskeleton, cell junction and pathway involved in epithelial-mesenchymal transformation such as Wnt pathway. More importantly, it has been reported that hypoxia incapacitated immune effector cells [
17] and enhanced the activity of immunosuppressive cells [
16], and immune escape [
29] and tumor cell adaptations to hypoxia [
30] could promote and perpetuate immunosuppression. In current analysis, we found that the hypoxia and immune status made inverse effects on patient prognosis in gastric cancer; higher hypoxia status was associated with poor prognosis while higher immune status could indicate better outcomes; and the impact of the inverse interaction on survival also observed after combining hypoxia and immune status. Thus, the hypoxia and immune status accompanied with their interaction in tumor microenvironment and its linking to gastric cancer progression could provide improved discussion with gastric cancer regarding prognosis.
As the hypoxia and immune activity in tumor microenvironment is complicated, there is no public biomarker using mRNA expression pattern to estimate their status [
31,
32]. Indeed, as tumors develop regions of hypoxia, tumor can also react favorably to hypoxic conditions as well as recovery of tumor blood and nutrient supply to some extent [
32,
33]. Thus, it is not powerful (and likely to omit important biology process information) to determine the hypoxia status by a single biomarker [
10,
11,
34]. The machine learning algorithm
t-SNE provides an elegant and robust dimensionality reduction approach, which has been applied to explore potential subtypes in prostate cancer [
35] and breast cancer [
36]. In the present study, the nonlinear cluster method of
t-SNE identified distinct patterns of hypoxic tumor microenvironment based on a set of two hundred hypoxia hallmark genes; further, expression changes of HIF-1 targeting genes were analyzed to explore their association with the hypoxia process. When it came to immune status, the ESTIMATE algorithm was used to impute immune scores to predict the level of infiltrating immune cells based on 141 specific gene signatures of immune cells. It was a newly developed algorithm that takes advantage of the unique properties of the transcriptional profiles of cancer tissues to infer tumor cellularity as well as the different infiltrating normal cells [
21]. Subsequent works have applied the ESTIMATE algorithm to prostate cancer [
37], breast cancer [
38], and colon cancer [
39], showing the effectiveness of such big-data based algorithms, although combination of immune characteristics with hypoxia status has not been investigated in detail.
Important roles of the signature genes identified in this study have been previously reported in multiple types of cancers. TAGLN encodes a shape change and transformation sensitive actin-binding protein. Overexpression of TAGLN was associated with cell invasion, which in turn contributed to promoting cancer metastasis [
40]. Notably, the expression of TAGLN is significantly induced by hypoxia in lung adenocarcinoma [
41]. Another risk gene PPP1R14A, has been reported to drive Ras pathway and tumorigenesis via inactivation of the tumor suppressor merlin [
42]. These results were consistent with the results in this study that overexpression of TAGLN and PPP1R14A could be unfavorable factors for patient’s outcomes. The protective gene CXCR6 is known as a chemokine receptor, which is selectively expressed in NK cells, T cells, and plasma cells. It is responsible for the chemotactic migration of immune cells to cancer tissues, which has the potential to kill cancer cells [
43,
44]. In our study, CXCR6 was identified as an immune-related, favorable gene for prognosis in gastric cancer. However, the abovementioned three signature genes were seldom studied in the context of combination of immune and hypoxia. Thus, signature genes identified in this study could provide underlying targets for experimental design in the laboratory to elucidate molecular mechanisms in gastric cancer.
Many potential targets, even with their detailed mechanism of action, have been revealed to play critical roles in tumor development and progression. However, it remains challenging for clinicians and researchers to translate these efforts and findings from laboratory into clinical settings. Integration of molecular and genetic characteristics and clinicopathologic factors provides a new insight for this issue regarding precision prognostication and individualized treatment. In current study, it was worth mentioning that patients in the low-risk group seemed not benefit from adjuvant chemotherapy. This provides a hint that hypoxia and immune status could serve as underlying markers for selecting sensitive patients to chemotherapy. Ongoing efforts on characterizing properties of tumor cell and its microenvironment are making intrinsic and extrinsic variations more and more clear; meanwhile, emerging techniques of sequencing make it possible for individualized risk stratification and treatment at molecular level in clinical application. Thus, the findings in current study, which links the genetic profiles and microenvironment characteristics to patient prognosis, could potentially provide translational value for clinical management of patients with gastric cancer.
Several limitations exist in this study. First, although several independent external validations were performed in this study, it was difficult to cover all variations among patients from different geographical regions when tissues and information were retrospectively collected in publicly available databases. Second, considering that the microenvironment characteristics might be distinct in different tumor regions, such as tumor core and invasive margin. Samples used for analyses were all collected from the core of tumor, and it is impossible to evaluate the immune and hypoxia status in different tumor regions. Thus, findings in this study are waiting for further validation by well-designed, prospective, multicenter studies.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.