Prognostic role of tumor-infiltrating lymphocytes in gastric cancer: a meta-analysis

Background In patients with gastric cancer, the prognostic value of tumor-infiltrating lymphocytes (TILs) is still controversial. A meta-analysis was performed to evaluate the prognostic value of TILs in gastric cancer. Materials and methods We identify studies from PubMed, Embase and the Cochrane Library to assess the prognostic effect of TILs in patients with gastric cancer. Fixed-effects models or random-effects models were used estimate the pooled hazard ratios (HRs) for overall survival (OS) and disease-free survival (DFS), which depend on the heterogeneity. Results A total of 31 observational studies including 4,185 patients were enrolled. For TILs subsets, the amount of CD8+, FOXP3+, CD3+, CD57+, CD20+, CD45RO+, Granzyme B+ and T-bet+ lymphocytes was significantly associated with improved survival (P < 0.05); moreover, the amount of CD3+ TILs in intra-tumoral compartment (IT) was the most significant prognostic marker (pooled HR = 0.52; 95% CI = 0.43–0.63; P < 0.001). However, CD4+ TILs was not statistically associated with patients’ survival. FOXP3+ TILs showed bidirectional prognostic roles which had positive effect in IT (pooled HR = 1.57; 95% CI = 1.04–2.37; P = 0.033) and negative effect in extra-tumoral compartment (ET) (pooled HR = 0.76; 95% CI = 0.60–0.96; P = 0.022). Conclusions This meta-analysis suggests that some TIL subsets could serve as prognostic biomarkers in gastric cancer. High-quality randomized controlled trials are needed to decide if these TILs could serve as targets for immunotherapy in gastric cancer.


INTRODUCTION
Gastric cancer has become a big health issue worldwide due to high morbidity and mortality [1].
Although the incidence and mortality of gastric cancer has been decreasing over the past decade, it is still the third cause of cancer death in men and the second in women in China [2]. Besides early diagnosis, appropriate treatment www.impactjournals.com/oncotarget/ Oncotarget, 2017, Vol. 8, (No. 34), pp: 57386-57398 Meta-Analysis plans identified from the prediction of patients' outcomes also contribute to successful treatment of gastric cancer. Therefore, further investigation is needed to identify newer tumor biomarkers with higher specificity and sensitivity in gastric cancer to determine the optimal therapeutic strategies and predict the prognosis of gastric cancers.
Tumor-infiltrating lymphocytes (TILs) are a heterogeneous group containing tumor-infiltrating T cells, B cells, and natural killer (NK) cells [3]. There are many specific antigens such as CD3, CD4, CD8, FOXP3, CD20, CD57 in the cell membrane of TILs. Usually, different cell surface antigens bind to specific type of lymphocytes. For example, CD3, CD4, CD8, FOXP3 bind to T cells; CD20 binds to B cells; and CD57 binds to NK cells. Thus, TILs play a bidirectional regulation roles in the tumor-associated immune responses: It can not only inhibit tumor growth by suppressing their outgrowth or destroying cancer cells, but also contribute to tumor progression either by creating a tumor microenvironment that stimulate tumor outgrowth or by selectively protecting tumor cell survival in an immunocompetent host [4].
TILs might be a prognostic biomarker in gastric cancer. Many studies have verified TILs for use in prognostic prediction besides their crucial role in tumorassociated immune responses in gastric cancer [5][6][7][8]. Lee et al. found that high infiltration of CD8 + T cells was associated with better outcome [5], whereas no correlation between the infiltration of CD8 + T cells and clinical outcome was found in Hass et al. study [6]. Zhou et al. revealed that patients with high density of FOXP3 + T cell had worse overall survival (OS) [7], but Kim et al. reported the opposite results [8]. Therefore, the prognostic role of TILs in gastric cancer remains controversial. It is also unclear whether the associations between the density of TILs and prognosis vary depending on survival-associated factors such as the endpoints and anatomical location. In view of the above-mentioned facts, to assess the prognostic effect of TILs in gastric cancer, we performed this meta-analysis.

Study characteristics
A total of 288 references were initially collected in our study. Among them, 43 articles reported the association between the density of TILs and patients' outcomes in gastric cancer. After further screening, 12 articles were excluded (one studying the same population, three lacking enough sample size, and eight without some useful data). Finally, we incorporated 31 articles including 4,185 patients from Italy, Japan, Germany, China, Korea, Turkey and France into this meta-analysis ( Figure 1) .
The main characteristics of studies were showed in Table 1. Among these, only five articles, including 875 patients with gastric cancer, reported the prognostic value of generalized TILs [9,10,13,15,29], and others focused on specific TIL subsets. The prognostic role of TILs in intra-tumoral compartment (IT) and extra-tumoral compartment (ET) were evaluated in 31 and 5 articles, respectively. 20 articles have large sample size (≥100). The details of tumor stage were provided in 21 articles, and the categorizations were various. Follow-up time was available in 27 articles, and the number of articles with median follow-up of more than 60 months was 11. The cut-off values contained mean value (n = 5), median value (n = 17), and others. Specific TIL subsets and generalized TILs were detected by means of immunohistochemistry (IHC) staining and Hematoxylin-eosin (HE) staining,

Subgroup analysis
Different TIL subsets have different functions in the process of anti-tumor immunoreaction response. Therefore, subgroup analyses based on TIL subsets were performed. Then, given that the function of TILs is associated with the distribution site, we further performed subgroup analyses based on the distribution site of TILs. Table 2 shows the major results of subgroup analyses.

OS
Eleven articles focused on the OS of patients with high density of CD8 + TILs in IT [5, 6, 8, 11, 17-21, 34, 35]. The pooled analysis revealed that high density of infiltration of CD8 + T lymphocytes was statistically significant associated with better OS (pooled HR = 0.73;    (Table 3). We further performed a sensitivity analysis which demonstrated the result pattern was not impacted by any single study. The results of metaregression showed that ethnicity (P = 0.838), sample size (P = 0.996) and cut-off value (P = 0.959) all had no statistically significant influence on the combined effect size for OS. Four articles evaluated the OS of patients with high density of CD8 + TILs in ET [6,21,35], and the pooled analysis demonstrated that high density of infiltration of CD8 + T lymphocytes was statistically significant associated with better OS (pooled HR = 0.71; 95% CI = 0.53-0.97; P = 0.029) ( Figure 2B).

CD4 + T lymphocyte subset
Four articles researched clinical outcome of patients with high CD4 + T cells infiltration [17,20,21,35]. However, subgroup analysis demonstrated that CD4 + T cells infiltration in IT/ET was not significantly associated with survival outcomes (Table 2).

Publication bias
We only evaluated the publication bias of subgroups that contained over 10 studies. First, we use the funnel plot to detect the publication bias. The funnel plots of the CD8 + and FOXP3 + TILs in IT were substantially symmetric ( Figure 3). Then, we used Begg's and Egger's tests to evaluated the publication bias. The P values were all greater than 0.05. Therefore, our meta-analysis has no publication bias.

DISCUSSION
The prognostic value of TILs in gastric cancer was quantitatively evaluated in this meta-analysis. Considering the bidirectional role of TILs in tumor-associated immune responses, it is necessary to assess the prognostic role of different TIL subsets. Therefore, we conduct subgroup analyses based on subset of TILs. Each subset of TILs plays its own roles in the development of gastric cancer, which contribute to their prognostic roles in gastric cancer. CD8, FOXP3, CD3, CD4, CD45RO, Granzyme B and T-bet are mainly located in the surface of T lymphocytes. Some of them are associated with favorable OS and/or DFS. CD8 and Granzyme B are surface antigens of cytotoxic T lymphocytes (CTLs) that are the main effective cells in the anti-tumor immune response [21]. This is in agreement with the results of our study in which these two subsets were associated with better OS and/or DFS. CD3 is a common surface antigen of T cells, and Arigami et al.

Figure 3: Funnel plots of CD8 + T cells (A) and FOXP3 + Tregs (B) infiltration in tumor nest.
reported the density of CD3 + TILs decreased during tumor progression [36]. It is noteworthy that the prognostic effect of CD3 + TILs in IT appeared most significant in our study. CD45RO is the most specific surface antigen of memory T cells, and this subset could prevent peritoneal dissemination of gastric cancer by inhibiting tumor growth and invasion [37]. T-bet is the most specific marker for T helper type 1 (Th1) cells, but besides Th1 cells, CD8 + CD4 + and CD56 + lymphocytes also express T-bet [24]. And for CD8 + CD4 + and CD56 + lymphocytes, T-bet is a T-box transcription factor which is crucial for the their differentiation and functions [38,39]. Furthermore, other surface antigens located in T lymphocytes showed different prognostic roles. CD4 + lymphocytes include a group of heterogeneous T lymphocytes, which can secret diverse cytokines, and IL-17 is one of the most representative cytokines. Studies showed that IL-17 could promote tumor angiogenesis, proliferation, migration and invasion [40,41]. On the contrary, many studies also confirmed IL-17 has anti-tumor effects. Bencherit et al. reported IL-17 could increase the infiltration and cytotoxicity of CTLs [42]. Antonysamy et al. found IL-17 could also increase the rate of mature and function of dendritic cells (DCs) [43]. Therefore, the roles of CD4 + TILs are very complicated [44], and in our study, the prognostic value of CD4 + TILs shows no significance. FOXP3 is the most specific marker on regulatory T cells (Tregs) that are generally considered to be immunosuppressive and block of effective antitumor immunity, therefore are associated with poor outcome in several kinds of tumors [45,46]. But in our study, FOXP3 + TILs played bidirectional prognostic roles including positive effect in IT and negative effect in ET. That may be because FOXP3 + TILs in different anatomical locations may have opposite functions and influences between each other [21].
B cells and NK cells also play important roles in tumor-associated immune responses. CD20 generally exists in the surface of B cells. CD20 + B cells not only contribute to cellular immunity by providing costimulatory signals to T cells and/or serving as antigen-presenting cells, but also to humoral immunity by producing antibody [47,48]. CD57 is a marker of NK cells which could attack tumor cells directly, and the recruitment of NK cells could exhibit strong antitumor activity and generate a better prognosis in gastric cancer [49,50]. The functions of these two subsets were consistent with our results.
In addition to prognostic biomarker for gastric cancer, we observed that TILs are potential targets for immunotherapy. Immunotherapy has long been regarded as a powerful anti-tumor treatment which is less toxic and tumor-specific [51]. This treatment contains adoptive cell therapy, monoclonal antibody-based treatment and cancer vaccines. CD8 + TILs have crucial targets in these courses. By blocking the interaction of CD8 + T cell-related receptors (CTLA-4 and PD-1) and ligands (PD-L2 and PD-L1), anti-tumor immunity is enhanced in patients with advanced solid tumor, including gastric cancer [52]. Champiat et al. also confirmed that augmentation of the local immune response might be a promising target for new immunotherapy [53]. According to our study, the accumulation of CD8 + , CD3 + , CD57 + TILs and depletion of FOXP3 + TILs were all favorable prognostic factors in gastric cancer. Based on these studies, we propounded CD3 + , CD57 + and FOXP3 + TILs could also serve as targets in immunotherapy like CD8 + TILs. However, we still need more prospective studies to verify the feasibility of other subsets in future gastric cancer immunotherapy.
To our knowledge, this is the first meta-analysis of evaluating the prognostic role of TILs in gastric cancer.
The results from our study should be interpreted with caution for following reasons. First, some studies used different cut-off values that could reduce the practicability of TILs in the process of estimating prognosis. Second, when several HRs were not directly provided in the original studies, we obtained them by calculating the data extracted from the survival curves. Thus, small statistical errors were inevitable. Third, there are still some studies that only used univariate analysis, although univariate analysis may overestimate effect sizes compared to multivariate analysis under common settings. Finally, there are few investigations of some subsets such as CD45RO + , Granzyme B + and T-bet + TILs. This could result in relatively insufficient data in the subgroup analysis.
In conclusion, our meta-analysis provides strong evidence that TILs might serve as a robust marker for predicting the outcome of patients with gastric cancer. Especially, high density of FOXP3 + TILs in the ET was a negative prognostic factor, which was opposite to previous results [6,54]. We still need more randomized controlled trials to further confirm the results of this meta-analysis.

MATERIALS AND METHODS
We conducted this meta-analysis according to the Observational Studies in Epidemiology (MOOSE) guidelines [56] and the Systematic Reviews and Meta-Analyses (PRISMA) guidelines [55].

Search strategy
We searched the literatures published prior to November 2016 through PubMed, Embase and the Cochrane. Keywords used in the search process were "TILs OR tumor-infiltrating lymphocyte OR intratumoral lymphocyte OR tumor infiltrating lymphocyte OR intratumoral lymphocyte OR intratumoral lymphocyte" (all fields) AND "stomach OR gastric" (all fields) AND "tumour OR tumor OR cancer OR neoplasm OR carcinoma" (all fields) AND "prognostic OR prognosis OR outcome OR survival" (all fields). We browsed not only the title and abstract but also the full text of identified articles. The availability evaluation and database search were conducted independently by two reviewers. The third reviewer resolved the disagreement from two reviewers.

Inclusion and exclusion criteria
Only the literatures meeting the following criteria were eligible for inclusion: (1) studies researching the prognostic effect of TILs or associated TIL subsets in gastric cancer; (2) the expression of TILs or associated TIL subsets in IT or ET of patients with gastric cancer; (3) HR and 95% confidence intervals (CI) could be extracted; (4) the sample size of studies was greater than 50. Several literatures were excluded according to the following criteria: (1) studies with small sample size (< 50) were excluded to avoid the publication bias from too small sample size; (2) reviews, letters, case reports, animal trials and conference abstracts were excluded; (3) only the complete or most recent study was enrolled if one patient cohort were researched by multiple studies.

Data extraction
The author information, publication year, TIL subsets and distribution site, origin of population, tumor stage, cut-off values, sample size, follow-up, detected methods, HRs and 95% CIs were collected by two researchers. OS and DFS (where possible) were chosen as the indexes of this meta-analysis. Recurrence-free survival (RFS) was used as a substitute if DFS was lacking because RFS is similar to DFS. Compared with univariate analysis, the result from multivariate analysis was preferred because it had improved precision after accounting for confounding factors.

Statistical analysis
The prognostic effect of different levels of TILs in gastric cancer was evaluated by HRs and 95% CIs. If the HRs and 95% CIs were reported in the study, we extracted them directly. Otherwise, we obtained them by calculating the data extracted from Kaplan-Meier survival curves using Engauge Digitizer version 4.1 [57,58]. The I 2 statistic and Chi-square test (P value) were used to assess the statistical heterogeneity [59,60]. Heterogeneity was suggested when I 2 was below 50% and/or P value was greater than 0.05. The source of heterogeneity was explored primarily through meta-regression and subgroup analysis. Funnel plot, Egger's and Begg's tests were performed to evaluate the publication bias [61]. All analyses were performed using STATA version 12.0 (Stata Corporation, College Station, TX, USA).