Prognostic value of PD –L1 expression in patients with primary solid tumors

Programmed death-ligand 1 (PD-L1) is thought to play a critical role in immune escape by cancer, but whether PD-L1 expression can influence prognosis of patients with solid tumors is controversial. Therefore, we meta-analyzed available data on whether PD-L1 expression correlates with overall survival (OS) in such patients. PubMed, EMBASE and other databases were systematically searched for cohort or case-control studies examining the possible correlation between PD-L1 expression and OS of patients with solid tumors. OS was compared between patients positive or negative for PD-L1 expression using scatter plots, and subgroup analyses were performed based on tumor type and patient characteristics. Data from 59 studies involving 20,004 patients with solid tumors were meta-analyzed. The median percentage of tumors positive for PD-L1 was 30.1%. OS was significantly lower in PD-L1-positive patients than in PD-L1-negative patients at 1 year (P = 0.039), 3 years (P < 0.001) and 5 years (P < 0.001). The risk ratios of OS (and associated 95% confidence intervals) were 2.02 (1.56-2.60) at 1 year, 1.57 (1.34-1.83) at 3 years and 1.43 (1.24-1.64) at 5 years. Similar results were obtained in subgroup analyses based on patient ethnicity or tumor type. The available evidence suggests that PD-L1 expression negatively affects the prognosis of patients with solid tumors. PD-L1 might serve as an efficient prognostic indicator in solid tumor and may represent the important new therapeutic target.


INTRODUCTION
Immune co-stimulatory and co-inhibitory receptors determined the functional outcome of T cell receptor (TCR) signaling and immune surveillance [1]. Tumors can modulate the interactions between inhibitory receptors and ligands to scape immune responses [2,3]. For example, the co-inhibitory receptor programmed cell death 1 (PD-1) plays a key role in cancer immune, especially in the immune escape phase [4]. PD-1 can be expressed in activated CD4 + and CD8 + T cells, but also in some natural killer cells and B cells [5]. When PD-1 binds to the ligand PD-L1 (B7-H1) expressed on the surface of tumors, it strongly inhibits the production of T cells and cytokines [6,7], promoting tumor cell growth and immune escape [8,9].
In addition to serving as a therapeutic target, PD-L1 may also be useful as a prognostic biomarker [22]. However, whether PD-L1 expression is associated with worse prognosis for patients with primary solid tumors remains controversial [20][21][22]. Therefore we metaanalyzed all available evidence to address this question comprehensively.

RESULTS
A total of 1,258 records were retrieved from PUBMED, EMBASE, Web of Science and EBSCO ( Figure 1). After excluding 825 duplicate publications, we reviewed the abstracts and titles of the remaining 433 articles. This led to the exclusion of another 288 records that were not original research articles published in English. The remaining articles were read in full, leading to the exclusion of 86 records because they did not deal with human patients or solid tumors, or because they failed to report adequate outcomes data. In the end, 59 articles were included in the meta-analysis.

PD-L1 expression and OS across all studies
Meta-analysis of data from all 59 studies showed that the median OS rate was significantly lower in PD-L1-positive patients than in PD-L1-negative patients at 1 year (P = 0.039), 3 years (P < 0.001) and 5 years (P < 0.001; Figure 2 Figure 2).

Subgroup analysis by tumor type
Given the significant heterogeneity in the metaanalysis involving all 59 studies, we performed a series of subgroup analyses to eamine the possible correlation between PD-L1 expression and OS. PD-L1 expression was associated with worse 1-year OS for the following types of solid tumor (   Table 2).

Kim 2016
South  N, number of studies; OS, overall survival; RR, risk ratio; 95% CI, 95% confidence interval * These meta-analyses were performed using a fixed-effects model. All other meta-analyses were performed using a random-effects model. www.impactjournals.com/oncotarget DISCUSSION While studies published more than a decade ago established that PD-L1 promotes cancer immune escape [82,83] and that blocking PD-L1 can improve the antitumor efficacy of anti-tumor responses [84][85][86], whether PD-L1 expression by solid tumors negatively affects patient prognosis remains unclear. Here we reviewed 59 studies involving 20,004 patients with 11 types of solid tumors and found strong evidence that PD-L1 expression is associated with significantly lower OS at 1, 3 and 5 years. This effect was observed in meta-analyses involving   PD-L1 positive expression is associated with viral infection and chronic inflammation [87]. Expression of PD-L1 and/or PD-1 has been described for numerous types of cancers associated with viral infection [88], including polycyclic virus-associated Merkel cell carcinoma [89], hepatitis B virus-associated hepatocellular carcinoma [33], human papillomavirus-associated head and neck cancer, and Epstein-Barr virus-related nasopharyngeal carcinoma [90]. In patients with hepatocellular carcinoma, PD-L1 expression was significantly higher in tumor macrophages than in matched normal tissues, and expression correlated with tumor grade [25].
Our results are consistent with previous reports that PD-L1 expression is associated with worse 5-year outcome in patients with gastrointestinal carcinomas such as esophageal cancer and gastric cancer [70,79] as well as colorectal cancer [25]. The precise mechanisms whereby PD-L1 expression may worsen prognosis are unknown; When PD-1 binds to the ligand PD-L1 (B7-H1) expressed on the surface of tumors, PD-1 has been shown to promote tumor cell-specific T cell inactivation or apoptosis [12].
The results of this meta-analysis should be interpreted cautiously because of several limitations. One is the lack of a standardized assay and cut-off value for classifying patients as PD-L1-positive. This may help explain the high heterogeneity observed across the included studies. Another limitation is our exclusion of gray literature, which may have increased the risk of publication bias and selection bias.
Despite these limitations, this large meta-analysis provides strong evidence that expression of PD-L1 may be a meaningful index for predicting prognosis in a wide variety of patients with solid tumors. These findings justify more focused prognostic studies in welldefined patient populations in which a panel of clinically relevant outcomes beyond only OS are considered.

Literature search
PubMed, EMBASE, Web of Science and EBSCO were searched through 15 January 2017 to identify cohort and case-control studies examining the relationship between PD-L1 expression and prognosis of patients with solid tumors. The following search terms were used: programmed death-ligand 1, PD-L1, B7-H1, CD274 and solid tumor.

Inclusion and exclusion criteria
To be included in our meta-analysis, studies had to involve (1) primary solid tumors in human patients; (2) The main content of the articles is to analyze the relationship between the expression of PD-L1 and the prognosis of solid tumors in patients; (3) a hospitalbased or population-based case-control or cohort design, regardless of sample size; (4) immunohistochemical assay of PD-L1 expression as high and low PD-L1 expression; (5) all patients underwent surgery; and (6) adequate reporting of overall survival (OS) data. When eligible studies involved overlapping patient populations, only the most recent or complete report was included. Studies were excluded if they were letter, summary of meeting and review; if they were published in a language other than English; or if they failed to report adequate data; or they investigated metastatic tumors. Gray literature (Reports and papers that were not published in PubMed, EMBASE, Web of Science and EBSCO) was not included into this study. Reference lists within identified articles were also searched manually to identify additional articles.

Meta-analysis outcomes
The primary outcome in the meta-analysis was OS. This outcome was compared between patients showing high or positive PD-L1 expression and patients showing low or no expression, as defined within the individual studies.

Data collection
Two researchers (P.-C.Y, X.X) independently screened studies for inclusion. Disagreements were resolved by discussion and, when necessary, consultation with a third author (S.Z). The first author's name, year of publication, country, number of patients, and tumor type were extracted from each study, and OS results for 1, 3 and 5 years were extracted from tables or Kaplan-Meier curves.

Statistical analysis
Forest plots of OS were generated using RevMan 5.3 (Cochrane Collaboration, Copenhagen, Denmark). Weighted risk ratio (RR) estimates were generated from pooled data using Mantel-Haenszel random-effects meta-analysis, unless no statistically heterogeneity, in which case fixed-effects meta-analysis was performed. Statistical heterogeneity in meta-analyses was assessed using Cochrane's Q and I 2 statistics. Survival results were analyzed using scatter plots generated in Prism 5 (Graphpad Software, San Diego, USA). The results for different patient groups were compared using the log-rank test. The threshold of statistical significance was defined as P < 0.05.

Author contributions
X.X, J.-H.Z and L.L conceived the study; P.-C.Y collected and analyzed the data; X.X drafted the manuscript; all authors have read and approved the final version to be published.