Background
Lung cancer remains the leading cause of cancer incidence and death worldwide in 2018, when it was estimated to comprise 2.1 million new cases and cause 1.8 million deaths [
1]. Among all patients with lung cancer, lung adenocarcinoma accounts for the largest proportion, and its incidence continues to increase [
2]. The literature has shown that phenotypes of cancers are dependent on the tumor microenvironment, especially tumor-infiltrating immune cells [
3,
4]. The immune response is characterized by many types of cells, and differences in their potential prognostic value depend on the cancer type [
5]. Computational approaches were adopted to characterize immune-interactions in lung adenocarcinoma, mainly focused on B cell and CD8+ T cell [
6]. Meta-analysis also showed that CD8+, CD3+ and FOXP3+ T cells infiltration had significantly prognostic value in lung adenocarcinoma [
7]. Furthermore, the prognostic significance of integrated immune infiltration in some cancers has been validated, and this has been proposed to supplement the tumor, node, and metastasis (TNM) staging system for predicting patient survival [
8‐
13]. However, very few studies reported integrated immune infiltration in lung adenocarcinoma.
Most previous studies have adopted an immunohistochemistry method to evaluate the landscape of tumor-infiltrating immune cell subtypes. However, a bioinformatics tool, known as the CIBERSORT algorithm, has been previously developed for computational enumeration of immune cell subtypes. The landscape and proportions of tumor-infiltrating immune cells are evaluated on the basis of the changes in expression of immune-relative and other genes in one sample. This meta-gene approach has been well designed and validated in previous studies [
9,
14‐
16].
In the present study, CIBERSORT was used to estimate the fraction of immune cell types based on different lung adenocarcinoma gene expression profiles. Least absolute shrinkage and selection operator (LASSO) Cox regression analysis was utilized to establish an immune cell infiltrating score model in both training and validation cohorts. We believe that the immune infiltrating score could assist in predicting survival, along with clinical factors, in lung adenocarcinoma.
Discussion
In this study, we adopted the CIBERSORT method to calculate the fraction of 22 immune cell in lung adenocarcinoma. This newly developed algorithm can work accurately for bulk tumor samples profiled by microarray or RNA-Sequencing. And unlike flow or mass cytometry-based methods, it is applicable to archived RNA and cellular specimens [
26]. Previous studies have validated the efficacy of the CIBERSORT method [
14‐
16]. On the basis of the estimated fractions of signature immune cells, LASSO regression was utilized to construct an immune cell infiltrating score model, of which the predictive accuracy has been demonstrated previously [
14,
28,
30,
31]. In both cohorts, the prognostic value of the scoring model was confirmed. A lower score was associated with significantly better prognosis, which may be due to more immunity-activating lymphocyte infiltrations. Further analyses also implied that the score could significantly improve the accuracy of survival prediction when combined with the TNM staging system. Significant correlation between the score and the expression of common immune checkpoint regulators or inflammatory mediators was also observed, including
PD-
L1,
IDO1 and
LAG3, which further confirmed its potential values. Previous studies also indicated that the expression of
PD-
L1 and
LAG-
3 was related with early recurrence and poor prognosis [
32,
33]. In a word, we hope that our work will provide new insights into the construction of staging or prognostic models in lung adenocarcinoma.
The role of different tumor-infiltrating lymphocytes in lung adenocarcinoma has been explored separately [
7,
34]. Kinoshita et al. [
35] found that CD8+ T-cells accumulation was identified as a negative prognostic factor in lung adenocarcinoma, but not in lung squamous carcinoma. Comprehensive immune profiling of lung adenocarcinoma revealed plasma cell infiltration was related to worse prognosis [
36]. Mast cell exosomes were also reported to promote lung adenocarcinoma cell proliferation [
37]. The function of mast cells against or in favor of the cancers still requires investigations [
38,
39]. Previous studies also observed that the prognostic implication differs according to histological types and smoking status [
35,
40]. The potential role of tumor-infiltrating lymphocytes may require exploration comprehensively according to immune microenvironment. The functionality of CD8+ tumor-infiltrating lymphocytes was reported to be affected by competition between antitumor and exhaustion programs [
41]. Choi and Na investigated the relationship between the immune landscape and tumor glucose metabolism in lung adenocarcinoma [
42]. Varn et al. adopted computational approaches to characterize tumor-immune interactions, mainly focused on six immune cell subtypes [
6]. In our study, we adopted the CIBERSORT method with 22 cell phenotypes in lung adenocarcinoma like the study of Gentles et al. [
5]. The final formula of the immune cell infiltrating score was composed of 13 types of immune cells using LASSO regression, which may better characterize the potential internal interactions and microenvironment.
Several studies have focused on the relationships between tumor-infiltrating lymphocytes and the efficacy of adjuvant chemotherapy [
8,
43‐
45]. In this study, the potential associations with chemotherapy were not all statistically significant. The weak effect may be due to small sample size in this part of analyses and the difficulty in finding the optimal cut-off value for the score. Previous studies showed that the chemotherapy sensitivity may depend on the lymphocytes infiltrating the tumor [
8,
9]. One potential mechanism is that the interferon secreted by lymphocytes could sensitize cells to chemotherapy [
46]. The innate and adaptive immune responses may also be activated by immunogenic tumor cell death [
47]. More investigations may be required for the underlying mechanisms [
48]. Our study also indicated that the score was significantly associated with patient smoking status and histologic subtype, which may help us to better characterize patient subgroups and choose the optimal treatment.
The strength of this study is the large cohort derived from several institutions, which outnumbered than most similar researches of lung adenocarcinoma. We adopted a newly developed CIBERSORT method to estimate the level of tumor-infiltrating lymphocytes, which outperformed other methods (like immunohistochemistry) regarding noise, unknown mixture content and closely related cell types. There are also some limitations that should be noted. First, the public datasets in this study were based on different genechips, but we adopted the RMAExpress method which designed for Affymetrix
® microarrays and merged the raw data. All the datasets were collected with different study purposes, which may lead to heterogeneities in baseline features and therapies. Also, only retrospective analyses were made. Future prospective studies will be needed to confirm the findings. Second, some clinicopathologic factors were missing or incomplete, especially histological subtype, smoking status and adjuvant chemotherapy. Further investigations should collect more clinical factors and endpoints, along with possible alternative methods for testing and validation, which may provide more evidence for lung adenocarcinoma [
13]. In addition, optimal cut-off value and practical approaches are essential for future clinical application.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.