Role of programmed cell death ligand-1 expression on prognostic and overall survival of breast cancer: A systematic review and meta-analysis.

BACKGROUND: Recently, the correlation of immunological checkpoint marker programmed cell death ligand-1 (PD-L1) and the prognosis of various cancers has been a research hotspot. The aim of this study is to examine the prognostic effect of PD-L1 in breast cancer.
METHODS: PubMed, EMBASE, Web of Science, the Cochrane Library database were searched for eligible studies and additional hand-searching were reviewed as an augmentation. Pooled hazard ratios (HR) and 95% confidence interval (CI) for overall survival (OS), cancer-specific survival (CSS), disease-free survival (DFS)/recurrence-free survival (RFS), and metastasis-free survival (MFS) were estimated using fixed- or random-effect models.
RESULTS: Data from 19 studies involving 12,505 patients were collected. Study quality was assessed according to guidelines for assessing quality in prognostic studies. PD-L1 expression was significantly associated with lymph node metastasis (P < .001), high tumor grade (P < .001), negative hormone receptor (P < .001), human epidermal growth factor receptor 2 (HER2) positivity (P < .001), high Ki67 (P < .001), and high tumor-infiltrating lymphocytes (TILs) (P < .001). PD-L1 expression had no significant impact on CSS (pooled HR 0.83, 95% CI = 0.64-1.09, P = .19) or MFS (pooled HR 1.11, 95% CI = 0.62-1.97, P = .72), but significantly correlated with shortened OS (pooled HR 1.52, 95% CI = 1.14-2.03, P = .004) and DFS (pooled HR 1.31, 95% CI = 1.14-1.51, P < .000). Subgroup analysis showed that not PD-L1 RNA expression, but protein expression was associated with shorter survival, in addition, the adverse prognostic effect of PD-L1 expression remained in luminal A, luminal B, and HER2 subtype, not in basal-like or triple-negative subtype.
CONCLUSIONS: An elevated PD-L1 expression significantly correlates with high-risk prognostic indicators and decreased survival in patients with breast cancer.

Introduction

Breast cancer is by far the most common malignant tumor in women worldwide.[ Advances in diagnosis, chemotherapy, endocrine therapy, and anti-human epidermal growth factor receptor 2 (HER2) therapy have significantly improved the survival of patients with breast cancer, but recurrence and metastasis remain the leading cause of breast cancer death.[ Cancer cells can also maintain an immunosuppressive microenvironment that favors tumor progression by expressing immune inhibitory signals.[ Interaction between programmed cell death ligand-1 (PD-L1 or CD274) and its receptor PD-1 is a major inhibitory pathway in maintaining an immunosuppressive tumor microenvironment. Interestingly, in recent years, inhibition of the immune checkpoint regulator PD-L1 or PD-1 is a new anticancer therapy.[

PD-L1 is one of the ligands of PD-1 and is expressed on hematopoietic cells, epithelial cells, and a number of tumor cells, including melanoma, lung, ovarian, and renal cell carcinomas. PD-1 is expressed on tumor-infiltrating CD8+ T cells, as well as CD4+ T cells, natural killer T cells, B cells, activated monocytes and dendritic cells. PD-L1 expressed on tumor cells bind themselves with PD-1 on the surface of T cells, thereby inhibiting T cells function, losing its killing effect on tumor cells.[ Moreover, upregulation of PD-L1 has been described closely in association with the clinicopathological status of cancer patients.[ Based on these results, targeting the PD-L1/PD-1 pathway to improve antitumor immune response is under investigation in multiple human cancers.[ PD-L1 has been reported not to be expressed in normal breast tissue but to be increased in nearly half of breast cancer. Some researchers have reported their paper with regards to PD-L1 expression in breast cancer and have raised concerns about the role of PD-L1 as a prognostic factor.[ However, its prognostic role in breast cancer is still under debate. Study by Qin et al[ evaluated the PD-L1 expression by immunohistochemical (IHC) staining, and revealed the association of high PD-L1 expression with poor prognosis in patients with breast cancer. This correlation was also validated in several other studies.[ On the contrary, Beckers et al[ demonstrated that PD-L1 expression improved outcome in triple-negative breast cancer.

Given the discrepancy in PD-L1 assessment assay and relative small sample size of each individual study, we conducted a meta-analysis with newest and largest quantity of relevant publications[ to clearly investigate role of PD-L1 expression on prognostic and overall survival of breast cancer.

Materials and methods

Sine this study is a meta-analysis of previously published studies, the ethical approval and patient consent are not required.

Search strategy

A comprehensive search of PubMed, EMBASE, Web of Science, the Cochrane Library for relevant publications for the period up to June 10, 2017 was conducted. Databases were searched using the following terms, both as text words and Medical Subjects Heading [MeSH] terms: “Breast Neoplasm,” “Programmed Cell Death 1 Receptor,” and Keywords.

“Breast Cancer,” “PD-L1,” “B7-H1,” “CD274.” This search strategy was created by combining the above terms via the Boolean operator “OR” and “AND.” In addition, we augmented our computerized literature search by manually reviewing the reference lists of identified studies, relevant reviews, and meta-analyses. We also checked abstracts from the American Cancer Society of Clinical Oncology (ASCO) meetings available at http://meetinglibrary.asco.org for citations. The search criteria were limited to articles published in the English language. When the same population was included in different publications, the most recent study was used for analysis. The literature retrieval was conducted in duplication by 2 independent reviewers (SL and LC).

Eligibility criteria

To be included in this analysis, studies should meet the following inclusion criteria after the full text were reviewed: they focused on breast cancer; all selected cancer patients were pathologically confirmed; and correlation between PD-L1 and prognosis was described. Articles were excluded from the analyses based on the following criteria: non-English papers; non-human experiments; review articles, meeting abstracts, or case reports; duplicate publication; PD-L1 expressed on other cell (e.g., immune cell and stromal cell), not tumor cell; and insufficient data about hazard ratios (HR) and 95% confidence interval (CI), or the Kaplan–Meier curve could not be extracted.

Data extraction

Data were extracted from each study by 2 reviewers (SL and LC) independently according to the pre-specified selection criteria. Decisions were compared and disagreements about study selection were resolved by discussion or by involving a 3rd reviewer (JJ). The following information was extracted from the literatures: author; year of publication; country; number of patients; clinicopathological characteristics of patients; tumor stage; specimen; detection method; detection standard of positive/high PD-L1 expression. Survival data including HR and 95% CI for overall survival (OS), cancer-specific survival (CSS), disease-free survival (DFS)/recurrence-free survival (RFS) and metastasis-free survival (MFS) was directly extracted from tables or text of included studies for further pooled analysis.

Author contact

We sent e-mails to the corresponding authors (or any other author with a contact e-mail address listed on the main manuscript) if we could not get the full text or sufficient data.

Quality assessment

The quality of the selected articles was assessed according to guidelines for assessing quality in prognostic studies and 6 items relevant to this study were used.[

Data synthesis and statistical analysis

Clinicopathological data were presented as means and proportions, differences between groups were tested with Pearson chi-squared test. Statistical heterogeneity was assessed by means of the Cochran Q and I2 tests. A probability value of P < .1 or I2 ≥50% indicated the existence of significant heterogeneity.[ When substantial heterogeneity was observed, the pooled estimate calculated based on the random-effects model was reported using the DerSimonian and Laird method,[ which considers both within-study and between-study variations. If there was no significant heterogeneity, a fixed-effects model was adopted.

Sensitivity analysis was performed to assess the extent to which the combined estimates might be affected by individual studies, in which the meta-analysis estimates were computed after omission of each study in turn.[ The potential for publication bias was assessed using the Egger linear regression test and Begg rank correlation test for funnel plot asymmetry.[P value <.05 was considered statistically significant. All P values are 2-tailed. Meta-analysis was performed using Review Manager (RevMan, version 5.3, Copenhagen: The Nordic Cochrane Centre, The Cochrane Collaboration)[ and R software (version 3.2.3; R Core Team, Vienna, Austria).[

Results

Search results

We identified 1009 publications and the process of study selection is summarized in Fig. 1. After screening, 825 articles were eliminated because they were duplicates, non-human studies, reviews, case reports, meeting abstracts, or studies on other tumors. After reviewing the complete text of 184 records, 165 articles were excluded because 155 records were non-prognosis studies, while 4 articles did not have sufficient data for further analysis, 5 studies evaluated the prognostic role of PD-1 expressed on immune cells, and 1 study about prognostic role of PD-L1 positive immune cell. In total, 19 articles were available for meta-analysis because of their quality and availability of data.[

Figure 1

PRISMA flow chart of the literature search.

Characteristics of the studies and study quality

The main characteristics of 19 eligible studies are presented in Table 1. The publication years ranged from 2014 to 2017, and a total of 12,505 breast cancer patients from Switzerland, USA, Austria, France, Australia, Korea, China, Brazil, Italy, UK, Spain, Canada, and Japan were included. The number of patients in each study ranged from 97 to 5454. Study by Tymoszuk et al[ contained 2 cohorts and reported separately (cohort A and cohort B). The PD-L1 expression levels were measured in protein by IHC in 15 studies and detected in RNA level in 4 studies by microarrays, Fluorescent RNAscope assay, or quantitative Real-Time PCR Experiments. To evaluate PD-L1 positivity in selected studies, different cut-off were used according to various scoring systems including Histo-score system (H-score), quantitative fluorescence (QIF) detection score, PD-L1 expression in tumor/normal breast samples (T/NB ratio), 4-point scale, Allred score, or Immunoscore (staining intensity and percentage of PD-L1 positive tumor cells). PD-L1 positive rate in all studies ranged from 6.4% to 76.4%.

Table 1

Main characteristics of the included studies.

Figure 2 summarized the methodological quality of these 19 included studies assessed by guidelines for assessing quality in prognostic studies with 6 items in assessing potential opportunity for bias. All of these included studies had a representative study participation and scored “low risk”; no loss to follow-up was associated with expression of PD-L1 in any study and all scored “low risk”; they all showed cut-off or scoring systems in evaluating PD-L1 positivity and scored “low risk”; all of 19 studies described information about primary antibody used in IHC or probe in RNA detection and scored “low risk”; blind interpretation were reported in 8 studies[ and scored “low risk,” others did not described that and scored “unclear”; all studies conducted survival analyses based on univariate and/or multivariate survival analyses. There is no selective reporting of results and scored “low risk.” Results of quality assessment were summarized in Fig. 3.

Figure 2

Quality assessment according to guidelines for assessing quality in prognostic studies.

Figure 3

Summary of methodological quality of each included studies on basis of review authors’ judgments on 6 items relevant to this review from guidelines for assessing quality in prognostic studies.

PD-L1 expression and clinicopathological features

Correlations between PD-L1 expression and clinicopathological features were analyzed and showed in Table 2. High PD-L1 expression associated with lymph node metastasis[ (P < .001), high tumor grade[ (P < .001), negative hormone receptor[ (P < .001), positive HER2[ (P < .001), high Ki67[ (P < .001), and high tumor-infiltrating lymphocytes (TILs)[ (P < .001). However, neither T stage[ (P = .501) nor patients’ age[ (P = .500) was significantly correlated with PD-L1 expression.

Table 2

Associations between PD-L1 expression and clinicopathological features.

PD-L1 expression and patient survival

We assessed the prognostic value of PD-L1 expression in terms of OS, CSS, DFS/RFS, and MFS. For OS, altogether 16 studies[ (n = 4719) reported OS data. Significant heterogeneity existed among included studies (I = 67%, Cochrane Q P < .000). Pooled result by random model revealed PD-L1 expression associated with poor prognosis in term of shortened OS (pooled HR 1.52, 95%CI = 1.14–2.03, P = .004) (Fig. 4A). Four studies[ (n = 3724) focused on CSS and no heterogeneity was existed among these studies (I2 = 0%, Cochrane Q, P = .79). Pooled result by fixed model revealed PD-L1 expression had no impact on CSS (pooled HR 0.83, 95% CI = 0.64–1.09, P = .19) (Fig. 4B).

Figure 4

Forest plots of hazard ratios for survival based on PD-L1 expression. A, OS. B, CSS. C, DFS. D, MFS. CI = confidence interval, CSS = cancer-specific survival, DFS = disease-free survival, IV = inverse variance, MFS = metastasis-free survival, OS = overall survival.

Fourteen studies[ (n = 4241) provided DFS/RFS data and no significant heterogeneity existed among included studies (I2 = 48%, Cochrane Q P = .02). Pooled result by fixed model showed that PD-L1 overexpression was associated with shorter DFS/RFS in patients with breast cancer than PD-L1 negative expression (pooled HR 1.31, 95% CI = 1.14–1.51, P < .000) (Fig. 4C). For MFS, 3 studies[ presented MFS data (n = 2035). Significant heterogeneity existed among included studies (I2 = 61%, Cochrane Q P = .08). Pooled result by random model revealed PD-L1 expression had no significant effect on MFS (pooled HR 1.11, 95% CI = 0.62–1.97, P = .72) (Fig. 4D).

Sensitivity and subgroup analysis

Sensitivity analysis was conducted to determine whether the exclusion of each study resulted in a significant difference. We performed the same pooled calculus after omitting each study in turn, and no change was calculated for OS, CSS, DFS, and MFS, indicating that our results were statistically robust. Since different PD-L1 expression level and breast cancer subtypes have been assessed in these included studies and could be potential confounding factors. Subgroup analyses were conducted to evaluate variations in PD-L1 protein/RNA level and intrinsic subtype for breast cancer. In protein-level subgroup, statistical difference was determined for OS, DFS, and MFS, not for CSS, while in RNA-level subgroup, PD-L1 status was not significantly associated with any end point (Table S1). In subgroup analysis by intrinsic subtype, PD-L1 expression was associated with shorten OS and/or DFS in luminal A subtype, luminal B (HER2– and HER2+) subtype and HER2 subtype. Of note, neither OS nor DFS was associated with PD-L1 expression in basal-like subtype or triple-negative subtype. For CSS and MFS, only one study showed PD-L1 expression was associated with longer CSS and MFS in basal-like subtype (Table S2).

Publication bias

Begg and Egger tests did not reveal publication bias affecting the hazard ratios for OS, CSS, DFS, and MFS. The P values for these tests were present in Table S3.

Discussion

A growing body of evidence suggests that the PD-L1/PD-1 pathway plays a key role in tumor immune escape. The correlations between PD-L1 expressions and different tumors have been studied by many researches.[ PD-L1 expression was also investigated as an indicator of survival for breast cancer in numerous studies,[ however, the results were inconsistent and conflicting. A study of 870 patients reported patients with high PD-L1 expression had decreased DFS, MFS, and OS compared with those with no PD-L1 expression, indicating that PD-L1 expression is an indicator of poor prognosis in breast cancer patients.[ Conversely, a study of 636 stage I–III breast carcinomas showed that PD-L1 mRNA expression is related to improved RFS.[ Several other studies reported no significant difference between the locoregional recurrence or survival of patients with high PD-L1 expression and patients with no PD-L1 expression.[ These conflicting results warrant further exploration. To arrive at a reasonable conclusion, we searched and performed this meta-analysis including 19 studies with a total of 12,505 patients. The present meta-analysis provided strong evidence that PD-L1 expression on tumors is significantly associated with worse OS (HR 1.52, 95% CI = 1.14–2.03, P = .004) and DFS (HR 1.31, 95% CI = 1.14–1.51, P = .0002) in breast cancer, while no effect on CSS (HR 0.83, 95% CI = 0.64–1.09, P = .19) or MFS (HR 1.11, 95% CI = 0.62–1.97, P = .72). Maybe less studies focusing on CSS and MFS is the reason and the effect of PD-L1 on CSS or MFS is a subject of ongoing investigation. Further subgroup analyses by intrinsic subtype confirmed that the adverse prognostic effect of PD-L1 expression remained in luminal A, luminal B, and HER2 subtype, not in basal-like or triple-negative subtype.

In addition, when the clinicopathological features were considered, high PD-L1 expression was associated with lymph node involved, high tumor grade, negative hormone receptor, positive HER2, high Ki67, and the presence of TILs. The finding that PD-L1 expression is associated with the above high-risk prognostic factors in breast cancer could indicate that activation of the PD-L1/PD-1 pathway may help these tumors evade antitumor immune response, these tumor cells even consequently proliferate and spread more rapidly. These results might strengthen the sensitivity and specificity of PD-L1 in predicting the clinical survival of breast cancer.

To evade from the immune system's monitoring, tumor cells in microenvironment can modulate PD-L1 expression via 2 major pathways, the extracellular pathway and the intracellular signaling pathway. The former is induced by IFNγ production from TILs and subsequent IFNGRs/JAK/STAT signaling in tumor cells,[ this pathway depends on the presence of TILs.[ The latter does not depend on the presence of the TILs and multiple mechanisms can lead to PD-L1 expression, including chromosomal amplification,[ activating mutation in epidermal growth factor receptor,[ or activation of the phosphoinositide 3-kinases/protein kinase B/mammalian target of rapamycin pathway.[ PD-L1 expressed on tumor cells bind themselves with PD-1 on the surface of T cells, thereby inhibiting T cells function, losing its killing effect on tumor cells.[ This reveals that antitumor immunity is elicited against many solid tumors, it is also affected by immunosuppressive factors. PD-L1 not only induces tumorigenesis and invasiveness, but also makes tumor cells less susceptible to specific CD8+ T cells.[ Results from these preclinical in vivo models make breast cancer an attractive candidate for immunotherapies targeted against this molecule. In addition, results from subgroup analysis by PD-L1 expression level showed that high PD-L1 protein expression was associated with shorter survival, but high PD-L1 RNA expression did not have any impact on survival.

Tumoral PD-L1 expression is of considerable clinical interest due to the recent development of PD-L1/ PD-1 blocking antibodies. A number of antibodies directed against PD-L1 (atezolizumab, avelumab, durvalumab) or PD-1 (nivolumab, pembrolizumab) are currently under clinical investigation.[ The early phase I clinical studies targeting PD-L1/PD-1 pathway with monoclonal antibodies have received substantial attention. Emens presented effect of atezolizumab in a phase I trial in patients with metastatic triple-negative breast cancer (TNBC). Among 21 patients, 3 patients had partial remission and 2 patients had complete remission. Overall, the 24-week PFS rate was 33%.[ In another phase I b trial with avelumab for 168 patients with metastatic or locally advanced breast cancer, Heery et al[ presented that 9 patients responded to treatment (1 complete response and 8 partial responses). In the phase I trial KEYNOTE-12, pembrolizumab has been used to determine whether it is effective in the treatment of breast cancer. 58.6% of patients screened positive for PD-L1 and the overall response rate was 18.5%.[ Currently, there are ongoing phase II (KEYNOTE-86, NCT02447003) and phase III clinical trials (KEYNOTE-119, NCT02555657) that will evaluate pembrolizumab as a monotreatment for TNBC while other phase I–III studies investigate the combination of pembrolizumab with chemotherapy. Furthermore, the anti-PD-L1 durvalumab (MEDI4736) and the anti-PD-1 nivolumab (BMS-936558/MDX-1106) are under investigation in breast cancer.[ However, these findings are also consistent with the viewpoint that over-expression of PD-L1 indicates a poor prognosis and therapeutic blockade of the PD-L1/ PD-1 pathway might be a valid treatment approach in breast cancer.

Despite our efforts in performing a comprehensive and accurate analysis, yet several limitations should be taken into account when interpreting results. Firstly, the analysis was limited to articles published in English. Secondly, a majority of the selected studies measured PD-L1 expression by IHC, the variable detection antibodies, tissue preparation, processing variability might account for the high variability in PD-L1 positive rate reported by different authors. Finally, the techniques and cut-off values for evaluating PD-L1 expression were different among the included studies, which might have caused some of the heterogeneity. A standardized methodology should be set up to improve consistency and reproducibility in the measurement of PD-L1 for future studies.

Conclusions

In summary, the current evidence shows that an elevated PD-L1 expression is a negative prognostic factor in breast cancer. More multicenter studies with larger sample size are warranted to present more reliable results of the clinical relevance and precise molecular explanation for the abnormal expression of PD-L1 in the future.

Acknowledgments

Authors thank all the authors whose publications were included in our meta-analysis.

Author contributions

Conceptualization: Shichao Li.

Data curation: Shichao Li, Li Chen.

Formal analysis: Shichao Li, Li Chen.

Investigation: Li Chen.

Methodology: Shichao Li.

Project administration: Li Chen.

Resources: Li Chen.

Software: Shichao Li.

Validation: Jun Jiang.

Visualization: Jun Jiang.

Writing – original draft: Shichao Li.

Writing – review & editing: Jun Jiang.