PD-L1 expression in lung cancer and its correlation with driver mutations: a meta-analysis

Although many studies have addressed the prognostic value of programmed cell death-ligand 1 (PD-L1) expression in lung cancer, the results remain controversial. A systematic search of the PubMed, EMBASE, and Cochrane Library databases was performed to identify the correlation between PD-L1 expression and driver mutations and overall survival (OS). This meta-analysis enrolled a total of 11,444 patients for 47 studies, and the pooled results showed that increased PD-L1 expression was associated with poor prognosis (HR = 1.40, 95% CI: 1.19–1.65, P < 0.001). In subgroup analysis stratified according to histology types, the pooled results demonstrated that increased PD-L1 expression was an unfavorable prognostic factor for non-small cell lung cancer (NSCLC) (HR = 1.26, 95% CI: 1.05–1.52, P = 0.01) and pulmonary lymphoepithelioma-like carcinoma (LELC) (HR = 3.04, 95% CI: 1.19–7.77, P = 0.02), rather than small cell lung cancer (SCLC) (HR = 0.62, 95% CI: 0.27–1.39, P = 0.24). The pooled ORs indicated that PD-L1 expression was associated with gender, smoking status, histology, differentiation, tumour size, lymph nodal metastasis, TNM stage and EGFR mutation. However, PD-L1 expression was not correlated with ALK rearrangement and KRAS mutations.

on the surface of cells in NSCLC, and exposure to gefitinib also lead to PD-L1 up-regulation 17 . Another study showed the upregulation expression of PD-L1 in NSCLC as a result of an EGFR mutation and ALK rearrangement via common downstream signalling pathways mediated by PI3K-AKT and by MEK-ERK 18 , implicating driver mutations in the regulation of the expression of immunosuppressive molecules.
We therefore conducted a comprehensive meta-analysis to investigate the significance of PD-L1 expression as a prognostic marker and to determine the relation of PD-L1 expression to clinicopathological features and driver mutations in lung cancer patients.

Results
Search results and characteristics of studies. The literature review process is shown in Fig. 1. The initial search strategies retrieved a total of 2,402 potentially relevant articles. After screening the titles or abstracts, 1977 studies were excluded as irreverent, non-English, in vivo/in vitro studies, case reports, reviews, meta-analyses and comments. After reading the full texts of the remaining articles, 45 studies lacking sufficient data for further analysis were discarded. Forty-seven studies with 11,444 patients were finally included for meta-analysis.
Correlation of PD-L1 expression with prognosis. The correlation between the expression of PD-L1 and overall survival (OS) in lung cancer is shown in Fig. 2. The meta-analysis indicated that PD-L1 expression is a correlative factor of OS, with the pooled hazard ratio (HR) values of 1.40 (95% CI: 1.19-1.65, P < 0.001) for OS using a random model with significant heterogeneity (I 2 = 79%, P < 0.001).

Correlation of PD-L1 with clinicopathological features. The correlation between PD-L1 expression
and clinicopathological parameters of lung cancer is shown in Supplementary Figs 1   Publication bias and sensitivity analysis. Begg's and Egger's test were performed to evaluate the publication bias in the literature. And no indicator of publication bias among these studies was present. The P values for these tests were 0.237 and 0.120, respectively (Fig. 5). (Statistical significance was set at P < 0.05). Meanwhile, the sensitivity analysis was performed to assess the stability of the present meta-analysis by omitting one study.
The results demonstrated that none of the studies influenced the overall HRs, suggesting that the results of the study are credible.

Discussion
High PD-L1 expression has been observed in various solid tumours, and a previous study demonstrated that the expression of PD-L1 contributes to poor prognosis 65 . Although heavily investigated; it remains controversial for the prognostic value of PD-L1 expression in lung cancer, reflecting the inconsistent results of previous studies. This meta-analysis included 47 studies with 11,444 patients to evaluate the significance of increased PD-L1 to the prognosis of lung cancer. The results of the present analysis showed increased PD-L1 expression was associated with poor prognosis in lung cancer patients. According to subgroup analysis, high PD-L1 expression was an indicator of poor prognosis in Asian populations, but not in non-Asian populations, suggesting that the association between PD-L1 expression and prognosis is dependent on ethnicity. Different histological types of lung cancer process different biological characteristics. To reduce the heterogeneity of study, we performed a subgroup analysis on the basis of different histological types. The pooled results demonstrated that increased PD-L1 expression was an adverse prognostic factor for NSCLC and LELC, but not for SCLC. Our study analyzed the relationship between PD-L1 expression and prognoses of LELC and SCLC for the first time. This study provides important evidence on the prognostic value of the PD-L1 expression in LELC and SCLC patients. A potential correlation between PD-L1 expression and OS of patients with NSCLC was evaluated in previous meta-analyses [66][67][68][69] . The results of three meta-analyses revealed that NSCLC patients with increased PD-L1 expression had a poor OS [66][67][68] . Another meta-analysis did not indicate PD-L1 as a prognostic predictor for NSCLC 69 . However, the combined sample size of the four meta-analyses was relatively small. In addition, the four meta-analyses did not include SCLC and LELC, nor the investigation of the association between increased PD-L1 expression and driver mutations. Compared with those meta-analyses, more studies have been included in our research. Different thresholds to define positivity expression and particularly different baseline characteristics hinder the comparison of different studies reporting correlation of PD-L1 expression with OS in NSCLC. Standardized methods and definitions of PD-L1 positivity are clearly needed to facilitate studies of PD-L1 as a prognostic biomarker. Thus, a large multicenter study using the same antibody and cutoff of PD-L1 expression may be helpful to obtain more accurate results.
Several clinical trials using anti-PD-1 and anti-PD-L1 monoclonal antibodies, including nivolumab (BMS-936558) 70,71 , pembrolizumab (MK-3475) 72 , and atezolizumab (MPDL3280A) 73 have shown promising clinical activity in advanced NSCLC. In the era of precision medicine, it is particularly important to screen patients who are most likely to benefit from PD-1/PD-L1 antibody immunotherapy. Preliminary results suggested that high PD-L1 expression was associated with higher clinical activity of anti PD-1/PD-L1 monoclonal antibodies 74 . Therefore, the identification of patients with high PD-L1 expression is a vital question for anti-PD-1/PD-L1 therapy. In the present study, we investigated the relation of PD-L1 expression to clinicopathological factors. According to the pooled analysis, the expression of PD-L1 was increased in male, smoker, patients with SCC, a higher histological grade, larger tumour size, positive lymph nodal metastasis, and later clinical stage. These patients might benefit more from treatment targeting the PD-1/PD-L1 pathway. These data suggest that increased PD-L1 expression might promote lung cancer invasion and metastasis, leading to the poor prognosis of patients with lung cancer. It has been reported in several studies regarding the association of smoking status with PD-L1 expression in patients with lung cancer. Some studies have shown that the expression of PD-L1 was significantly higher in smokers 28,46 , whereas other studies could not confirm this finding 23,24,45,54 . In the present study, patients with high PD-L1 expression were associated with smoking status in lung cancer patients.
Accumulating evidence revealed the relationship between PD-L1 expression and driver mutations. EGFR mutations represent one of the most frequent driver mutations in NSCLC, particularly in ADC. Previous studies revealed that activating EGFR mutations induced PD-L1 expression in EGFR-driven NSCLC in cell lines and an animal model 21,75 . Moreover, as observed in NSCLC cell lines, there was a high level of PD-L1 expression in NSCLC patients harboring EGFR mutations 21,24,45 . However, some studies have shown that PD-L1 positivity was more frequent in EGFR wild-type 28,44,46 , and other studies have shown no association between PD-L1 expression and EGFR mutations 23,26,27 .The present meta-analysis investigated the correlation of PD-L1 expression with EGFR mutations in lung cancers. The results of the present study showed that high PD-L1 expression was associated with EGFR mutations. The discrepancies among different studies might reflect the heterogeneous study population and variable definitions of PD-L1 expression. Additional studies are needed to further analyze this issue. In addition, we showed that increased PD-L1 is not associated with ALK rearrangements and KRAS mutations.
There are several limitations of the present study that should be acknowledged. First, the sample size of SCLC and LELC studies included in the present meta-analysis was relatively small; therefore, the pooled data might be less than the statistical power. Hence, additional well-designed studies with larger sample sizes are needed to provide a more comprehensive evaluation of the prognostic value of PD-L1 expression in patients with SCLC and LELC. Second, the HR values of some studies were extracted from survival curves, which is less reliable than direct data provided in the original literature. Third, the distinct antibodies and different cut-off levels of PD-L1 expression among diverse studies might also impact the accuracy of prognostic estimation for lung cancer. Moreover, some inevitable publication bias might exist in the present meta-analysis, as many negative studies could not be published. Furthermore, significant heterogeneity existed in the results, although we calculated the pooled subgroup data using random-effects models. The observed heterogeneity might reflect differences based on different baseline characteristics, study designs, or treatment protocols. Finally, all of included studies were retrospectively collected, which might have introduced heterogeneity from variable treatments.
In conclusion, despite the limitations described above, this study presents the first meta-analysis to systematically assess the association of PD-L1 expression with lung cancer survival and driver mutations. The results demonstrated that high PD-L1 expression represents an unfavorable biomarker in LELC and NSCLC, but not in SCLC. In addition, increased PD-L1 expression is correlated with EGFR wild-type status. To strengthen these findings, the validation of the prognostic value of PD-L1 expression in patients with lung cancer requires further studies.

Methods
This meta-analysis was performed according to the preferred reporting items for systematic reviews and meta-analysis (PRISMA) statement 76 . The present study was based on data from previously published studies, and therefore, ethical approval was not required. Literature search. We conducted a systematic literature search for published articles in the PubMed, EMBASE, and Cochrane databases from January 1999 to July 2017. The search terms included the following keywords: (PD-L1 OR B7-H1 OR CD274 OR programmed cell death 1 ligand 1 protein) AND (lung cancer OR lung neoplasms OR pulmonary cancers). Furthermore, we manually searched the abstracts of the annual meetings of American Society of Clinical Oncology (ASCO), European Society of Medical Oncology c (ESMO) and the World Conference of Lung Cancer (WCLC) from 1999 to 2017. To explore additional studies, we also reviewed the reference lists of relevant articles. Eligibility criteria. The following inclusion criteria were used: (1) All patients were histologically confirmed as having lung cancer; (2) PD-L1 expression was detected by immunohistochemistry (IHC) or quantitative immunofluorescence (QIF) in primary NSCLC tissue; (3) Studies showed a correlation between PD-L1 expression and overall survival; (4) Studies showed a correlation between PD-L1 expression and clinicopathological features; (5) Studies provided sufficient information to extract the HR and 95% CI date for OS; and (6) articles were published in English. Studies that did not meet the inclusion criteria were excluded. When several studies were conducted using the same cohort of patients, only the most recent study was included.
Data extraction. Two authors (ZMH and LGL) independently conducted the data extraction, and a third reviewer (WY) resolved any discrepancies. The following information was extracted: name of the author, year of publication, country, tumour type, number of patients, stage, detection method, PD-L1-positive expression, outcome, clinicopathological parameters and HRs and 95% CIs for OS. When the HR values were not directly reported, we obtained the additional data from the original authors. And the data was extracted from survival curves using the methods of Parmar under the circumstances of no response 77 . Two reviewers (ZS and WY) independently conducted the quality assessment for each study using the NOS, and any discrepancies were resolved after revisiting the original study and discussion until consensus was reached. The NOS maximum possible score was 9 points, and studies that receiving a score of 6 or higher were considered high quality 78 .
Statistical methods. The HR and its 95% CI values were used to evaluate the association between PD-L1 expression and survival, and the pooled OR and 95% CI values were used to determine the relationship between PD-L1 expression and clinicopathological features. Statistical heterogeneity between studies was assessed using the chi-squared test and I 2 . A P value <0.1 orI 2 values of >50% were indicative of significant heterogeneity, with a random effects model being used; a fixed effects model otherwise. A subgroup analysis was conducted to explore the potential heterogeneity among studies. Potential publication bias was assessed using Egger's and Begg's tests.