PD-L1 expression as biomarker of efficacy of PD-1/PD-L1 checkpoint inhibitors in metastatic triple negative breast cancer: A systematic review and meta-analysis

Background Inhibitors of programmed cell death 1 (PD-1)/programmed cell death ligand 1(PD-L1) checkpoint have been approved for metastatic triple negative breast cancer (mTNBC) in patients positive for PD-L1 expression. Negative results from the recent phase III trials (IMPassion131 and IMPassion132) have raises questions on the efficacy of PD-1/PD-L1 checkpoint inhibitors and the predictive value of PD-L1 expression. Here we attempt to systematically analyze the biomarker value of PD-L1 expression for predicting the response of PD-1/PD-L1 checkpoint inhibitors in mTNBC. Materials and methods PubMed database was searched until Dec 2021 for studies evaluating PD-1/PD-L1 checkpoint inhibitors plus/minus chemotherapy in mTNBC. Outcome of interest included objective response rate (ORR), progression-free survival (PFS), and overall survival (OS). Review Manager (RevMan) version 5.4. was used for data-analysis. Results In total, 20 clinical trials comprising 3962 mTNBC patients (ICT: 2665 (67%); CT: 1297 (33%) were included in this study. Overall ORR was 22% (95%CI, 14-30%) and significant improvement was observed for PD-L1+ patients (ORR 1.78 [95%CI, 1.45-2.19], p<0.00001) as compared to PD-L1- cohort. Pooled outcome also indicated a significant 1-year PFS and 2-year OS advantage for patients with PD-L1 expression (1-year PFS: ORR 1.39 [95%CI, 1.04-1.85], p=0.02; I2 = 0%; 2-year OS: (ORR 2.47 [95%CI, 1.30-4.69], p=0.006; I2 = 63%). Subgroup analysis indicated that PD-L1 expression can successfully predict tumor response and 2-year OS benefit in mTNBC patients regardless of the type of investigating agent, line of treatment administration, and to some extent the type of treatment. Biomarker ability of PD-L1 expression to predict 1-year PFS was slightly better with pembrolizumab (p=0.09) than atezolizumab (p=0.18), and significantly better when treatment was administered in the first-line setting (OR 1.38 [95%CI, 1.02-1.87], p=0.04) and chemotherapy was added (OR 1.38 [95%CI, 1.02-1.86], p=0.03). Immune-related toxicity of any grade and grade≥3 was 39% (95%CI, 26%-52%) and 10% (95%CI, 8%-13%), respectively. Conclusions PD-L1 expression can predict objective response rate and 2-year OS in mTNBC patients receiving PD-1/PD-L1 checkpoint inhibitors. One-year PFS is also predicted in selected patients. PD-L1 expression can be a useful biomarker of efficacy of PD-1/PD-L1 checkpoint inhibitors in mTNBC.


Introduction
Triple negative breast cancer (TNBC), which constitutes about 15 to 20% of the breast cancer cases, is characterized by the lack of expression for hormone receptors (estrogen receptor [ER-] and progesterone receptor [PR-]) and human epidermal growth factor receptor 2 (HER2) (1)(2)(3). Multi-omics studies have further identified that majority of TNBC (about 55%-81%) express highly proliferative basal-like genes which can describe its aggressive nature (3). Prognosis is poor and 3-year recurrence is high despite a greater response to chemotherapy as compared to other breast cancer subgroups (3)(4)(5). Metastasis is common and is the major cause of death (2). 5-year survival rate is less than 30% for metastatic TNBC (2).
Immunotherapy in the form of immune checkpoint inhibition (ICI) was considered as an alternative and complementary cancer treatment option for TNBC due to its high genomic instability, infiltration of tumor-infiltrating lymphocytes (TILs) and elevated expression of programmed cell death protein ligand 1 (PD-L1) (6)(7)(8). Programmed cell death protein expressed on the T cells engages its ligand PD-L1 expressed on tumor cells thereby mediating tumor immune escape via suppression of antigen-specific T cell immune responses (8). Interruption of this PD-1/PD-L1 interaction with either anti-PD-1 monoclonal antibody (mAb) or anti-PD-L1 mAb results in activation of anti-tumor immune response (8) (Figure 1A). Additionally, PD-L1 is also expressed on other tumor-infiltrating immune cells mainly the antigen presenting cells (APCs) such as dendritic cells (DCs) and macrophages among others (9). Studies have identified the indispensable role of PD-L1 expression on such immune cells for the therapeutic efficacy of PD-1/PD-L1 blockade therapy (9,10). As such, PD-L1 expression on tumor cells as well as host immune cells is evaluated for their biomarker efficacy.
Durable responses were observed in the earlier TNBC trials for anti-PD-1 (pembrolizumab) and anti-PD-L1 (atezolizumab) monoclonal antibodies alone or in combination with chemotherapy (11)(12)(13)(14)(15). Results of subsequent phase III trials (IMPassion130 and KEYNOTE-355) had led to approval of these agents in combination with chemotherapy for advanced metastatic TNBC patients (16,17). The approval was based on the progression free survival benefit observed for TNBC patients receiving the combined treatment particularly in patients positive for PD-L1 expression. However, long term benefit is still unclear and OS results for KEYNOTE-355 is yet to come (17). Moreover, IMPassion131 phase III trial of atezolizumab showed no PFS advantage with combined treatment for PD-L1+ patients (18). These outcomes not only question the long-term benefit of PD-1 inhibitors but also cast doubt on the biomarker value of PD-L1 expression which had formed the basis for FDA approval.
In this study, we have gathered the efficacy data from various TNBC immunotherapy trials to evaluate the status of immunotherapy in advanced TNBC with a focus on PD-L1 expression as biomarker.

Methods and materials
This systematic-review and meta-analysis was carried out following the updated version 2020 of PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) guidelines (19).

Eligibility criteria
The PICOT (Population, Index prognostic factors, Comparator prognostic factors, Outcomes, Timings, and Settings) system was utilized for identifying the key elements of this review.
◼ Target population and treatment: Advanced metastatic triple negative breast cancer (TNBC) patients receiving immunotherapy, mainly PD-1/PD-L1 checkpoint inhibitors, with or without chemotherapy.
◼ Index Prognostic factors: PD-L1 expression was the only main index prognostic factor.
◼ Comparator prognostic factors: not applicable for this review.

Study design and language restrictions
Single arm studies that provided treatment comparison based on PD-L1 status and comparative clinical trials (CTs) that compared PD-1/PD-L1 checkpoint inhibitors +/-chemotherapy to chemotherapy alone were pursued with English language restrictions.

Research strategy and study selection
PubMed was formally searched with several key terms until Dec, 2021. Further potential studies were identified through screening references of relevant articles. A step-wise procedure comprising retrieving, organizing, and screening was followed by two reviewers to select studies meeting the eligibility criteria. Disagreements were solved after consulting the third author.

Data extraction
Modified form "The Cochrane Collaboration Data Collection form for RCTs" was used for data extraction that is available from Cochrane website. Characteristics of the included studies and attributes of participants were extracted that included first author, publication year, trail designation, national clinical trial (NCT) registration number, trial design, number and type of participants, and treatment type. Participants' attributes included age, PD-L1 expression status, PD-L1 assessment assay, and the lines of previous therapy for metastatic disease. Relevant outcome data was also extracted for performing meta-analysis.

Quality assessment
The Cochrane Collaboration Tool were used to assess the quality of included randomized controlled trials (20). Assessments included sequence generation, allocation of sequence concealment, blinding of participants and personnel, blinding of outcomes and assessments, incomplete outcome data, selective outcome reporting, and other biases. Non-randomized clinical trials were assessed with the methodological index for non-randomized studies (MINORS), which contains eight items of assessment and is recognized as the most appropriate guideline to evaluate the methodological quality of non-randomized trials (21). As a meta-analysis of biomarker assessment, the risk of bias was also assessed by "The Quality In Prognosis Studies (QUIPS) tool" for studies reporting biomarker analysis (22, 23). The QUIPS tool used six important domains (1): study participation (2), study attrition (3), prognostic factor measurement (4), outcome measurement (5) study confounding, and (6) statistical analysis and reporting. Quality of evidence assessment was carried out with The Grading of Recommendations Assessment, Development and Evaluation (GRADE) (24). Major outcomes were graded as high, moderate, low, or very low depending upon the following elements: study design, risk of bias, inconsistencies, imprecision of the results, indirectness, and publication bias.

Measurement of treatment effect and data synthesis
Number of events for objective response rates (ORR: Complete Responses + Partial Responses) and adverse events were pooled using Mantel-Haenszel method (19, 25). Pooling of number of events for single-arm studies was carried out with the help of a stepby-step guide for meta-analysis of descriptive data analysis developed by Neyeloff, et al. (26). Hazard ratios for PFS and OS were recorded directly from the study or extracted from the K-M curves using the Digital Equalizer and methods for incorporating summary time-to-event data into the meta-analysis according to Tierney et al. (27,28). Pooling of HRs was done by applying inverse variance method (19, 25). Heterogeneity was assessed using Chi 2 test and I 2 value and graded as low (I 2 = 25%), moderate (I 2 = 50%), and high (I 2 = 75%) according to I 2 values (29). Random effects model was adopted when heterogeneity exceeded 50% (I 2 ≥ 50%). Significance level was set at p<0.05.

Results
The research strategy and study selection are illustrated in Figure 1B. A total of 1806 published studies were identified upon initial database search. Title and abstract screening excluded 1765 studies for various reasons including duplicates, irrelevant, lacking data of main outcomes, and not target agent/s, etcetera. Further scrutiny for eligibility and full text reading yielded 20 studies that were included in this meta-analysis (11)(12)(13)(14)(15)(16)(17)(18)(30)(31)(32)(33)(34)(35)(36)(37)(38)(39)(40)(41). General characteristics of the included studies and participants are presented in Table 1. The detailed treatment regimen is highlighted in Figure 1C, along with the national clinical trial number for each study. Overall, 3962 patients with mTNBC were available for analysis as depicted in Figure 1D. A total of 2665 (67%) participants had received PD-1 (nivolumab and pembrolizumab) or PD-L1 inhibitor (avelumab, atezolizumab, and durvalumab) as monotherapy or in combination with chemotherapy. This cohort was described as ICT cohort. While the remaining (33%) were administered with chemotherapy alone (CT cohort). Patients with PD-L1+ expression were 1453 (59%) and 1003 (41%) patients were negative for PD-L1 expression in the ICT cohort. Pembrolizumab was investigated in 12 studies; atezolizumab in 4; durvalumab in 2; and avelumab in only one study ( Figure 1C and Table 1). Pembrolizumab was also administered with induction chemotherapy in few trials (31,32) and was followed by radiotherapy in a single trial (33). Likewise, three studies had evaluated immune modulating agents along with pembrolizumab (34-36). However, these studies had only reported response rates in mTNBC and were not incorporated into the biomarker analysis due to lack of relevant data (33-36). PD-L1 expression was assessed with PD-L 1 I H C 22 C 3 p h a r m D x a s s a y i n s t u d i e s i n v o l v i n g pembrolizumab measuring the PD-L1 expression on tumor cells as well as immune cells which was termed as combined positive score (CPS) (11,12,17,30,41). Atezolizumab studies (Phase III RCTs) utilized VENTANA PD-L1 (SP142) assay assessing the PD-L1 expression on immune cells (IC) alone as the phase I trial had revealed a significant improvement based on the PD-L1 expression on immune cells and not tumor cells (TC) (15,16,18). Durvalumab also showed better predictive value of PD-L1 expression on immune cells in a phase II trial after an initial pilot study demonstrated no significant predictive value of PD-L1 expression assessed on tumor cells (38, 39). Details are highlighted in Table 1.

Quality assessment of the studies and evidence
Quality assessment was carried out using The Cochrane Collaboration Tool. Only three RCTs were double-blinded exhibiting a low risk of bias (16)(17)(18) (11-18, 30, 32, 33, 36-41) had low risk of bias (score 10-15) and 20% exhibited a moderate risk of bias (score <10) (31,34,35). QUIPS tool, which assess the risk of bias in the studies involving assessment of prognostic factors, showed 66% of the studies had low risk of bias. Nonetheless, 33% of the studies demonstrated high risk of bias in terms of participants (30), study attrition (30), prognostic factor (15,30), and confounding (40,41). Quality of evidence assessment based on the GRADE quality tool is

Subgroup analysis
As the included studies were heterogenous according to the type of therapeutic agents, line of treatment and the addition of chemotherapy, we further carried out subgroup analysis to investigate the predictive significance of PD-L1 expression ( Table 2). First of all, we considered the studies that either included pembrolizumab or atezolizumab as the investigating agent. The response was slightly higher in pembrolizumab cohort (OR 2.08 [95%CI, 1.49-2.90], p<0.0001) than atezolizumab (OR 1.55 [95%CI, 1.19-2.02], p=0.001) with no significant difference between the two agents (subgroup differences: Chi² = 1.86, df = 1 (P = 0.17), I² = 46.3%) ( Figure 5A). Pembrolizumab showed to  Figure 4B). A similar outlook was demonstrated for 2-year OS but the difference in IT cohort was close to significance (p=0.07) (Supplementary Figure 5B).
Lastly, we also evaluated if the significance of PD-L1 expression is maintained when therapy is administered in the second line setting? Predictive ability of PD-L1 expression was slightly better in first line setting ( Figure 4C). Although both subgroups showed significant 2-year OS advantage, the subgroups difference tended towards significance (p=0.08) (Supplementary Figure 5C).

Safety concerns
Toxicity data were reported in the form of treatment-related adverse events (TRAEs) and immunotherapy-related adverse events (IRAEs).

Category
Subgroups Significant improvement with PD-L1 expression ORR, objective response rate; PFS, progression-free survival; OS, overall survival; IT, immunotherapy; ICT, immunotherapy plus chemotherapy; L, line of therapy. diff * indicates the statistical difference between two subgroups. Y indicates significant association with PD-L1 expression. X indicates no significant association with PD-L1 expression.

Discussion
Immune checkpoint inhibition (ICI) therapy has been extensively evaluated in the management of triple negative breast cancer patients. ICI as monotherapy or in combination with chemotherapy have demonstrated anti-cancer activity in earlystage as well as metastatic TNBC (11)(12)(13)(14)(15)(16)(17)(18)(42)(43)(44)(45)(46). Our study result indicates a 22% response rate for PD-1/PD-L1 checkpoint inhibitors as monotherapy or in combination with chemotherapy in mTNBC. The later approach was also shown superior to chemotherapy alone in terms of ORR and PFS. However, no overall survival advantage was demonstrated for the combined approach.
PD-L1 expression has been regarded as prognostic factor in various cancers (47)(48)(49)(50). Its biomarker value to predict response to PD-1/PD-L1 checkpoint inhibition has also been acknowledged across cancers (50)(51)(52)(53)(54)(55). Among breast cancer subtypes, TNBC exhibit higher tumor mutational burden (TMB), tumor infiltrates of lymphocytes (TILs) and PD-L1 expression, indicating its candidacy for immune checkpoint inhibition therapy (6)(7)(8). In TNBC, PD-L1 expression as biomarker have failed to predict pathological response rate in the neoadjuvant setting and longterm OS benefit in the metastatic setting (56). Our study systematically accumulated data to analyze the biomarker significance of PD-L1 expression in the metastatic TNBC patients. The outcome revealed that PD-L1 expression can significantly predict objective response, 1-year PFS, 2-year OS and overall survival. However, overall PFS was not significantly different between these two groups based on the direct comparison analysis. Nonetheless, the studies that provided the direct PFS comparison between PD-L1+ and PD-L1-patients were comprised of smaller number of participants (12,13,15,30). While major phase III trials reported treatment difference in each subgroup (PD-L1+/-cohort) separately with no direct comparison between the PD-L1+ and PD-L1-patients (16)(17)(18). Besides, a much larger population cohort would be required to detect a PFS difference of any significant value in a population with such a heterogenous and extensive disease burden as mTNBC (57). On the other hand, OS is affected by several other patient factors other than the treatment as compared to PFS which is a direct measure of clinical performance (58). In the treatment comparative analysis, the response rate, PFS and OS was significantly higher with ICT in PD-L1 positive population. While PD-L1 negative patients failed to derive such benefit. Overall, these outcomes strongly underline biomarker value of PD-L1 expression in mTNBC.
PD-L1 expression was assessed on the tumor cells and/or the tumor infiltrating lymphocytes and macrophages (11-18, 30, 32, 38, 39, 41). Studies involving pembrolizumab had reported the combined positive score (CPS), defined as the ratio of PD-L1positive cells (tumor cells, lymphocytes, and macrophages) out of the total number of tumor cells × 100 (11,12,17,30,41). The prognostic and predictive efficacy of atezolizumab was initially assessed with PD-L1 expression on the tumor cells (TC) and immune cells (IC) (13,15). However, the phase Ib trial conducted by Emens, et al. indicated a significant predictive value for PD-L1 expression on immune cells only (15). Therefore, subsequent phase III trials had only evaluated the predictive efficacy of atezolizumab based on the IC-derived PD-L1 which was defined as a percentage of tumor area (consisting of TC and associated intra-tumoral and continuous peri-tumoral stroma) occupied by IC with discernible PD-L1 staining of any intensity (16,18). Subgroup analysis revealed that predictive value of PD-L1 expression was maintained regardless of the scoring algorithm used. PD-L1 expression was able to predict ORR and 2-year OS to both of these agents. Although a near to significant improvement in 1-year PFS was observed for pembrolizumab in PD-L1+ patients, PD-L1 expression failed to predict 1-year PFS for these two subgroups. This subgroup analysis also established that the type of detection assay (SP142 used for atezolizumab & 222C3 for pembrolizumab) also had no impact on the clinical activity. This outcome is in concordance with the observations made in comparative analysis of these immunohistochemistry assays in IMPassion130 study, which indicated significant prevalence differences (prevalence rates for PD-L1 IC ≥1% cutoff for SP142, SP263, and 22C3 were 46.4%, 74.9%, and 73.1%, respectively. And for PD-L1 22C3 CPS ≥1 was 80.9%) but similar clinical activity with A+nP vs P+nP for all four subgroups (SP142 IC ≥1%, SP263 IC ≥1%, 22C3 IC ≥1%, and 22C3 CPS ≥1; PFS HR = 0.60 to 0.68; OS HR = 0.74 to 0.79) (59). These observations indicate that the lack of analytical equivalency among PD-L1 expression detection assays exert no impact on the biomarker ability of PD-L1 expression in predicting clinical activity. PD-L1 expression also varies between primary tumor and metastatic sites as previously been reported for mTNBC (60,61). The specific site of PD-L1 assessment was not clearly reported in these trials but a post hoc analysis of IMPassion130 revealed a difference in PD-L1 expression between primary tumor and metastatic sites (44% vs 36%, p=0.014) of mTNBC patients with no apparent influence on the predictive value of PD-L1 expression (59). A negative impact has been reported for number of previous lines of therapy on the response to PD-1 checkpoint inhibitors in lung cancer and melanoma (62)(63)(64). Likewise, overall response was higher in mTNBC patients receiving PD-1/PD-L1 checkpoint inhibitors as first line treatment compared to the patients previously treated for metastatic disease (12)(13)(14)(15)(16). However, previous therapy had no effect on the biomarker efficacy of PD-L1 expression according to the outcomes of our study. Nonetheless, 1-year PFS could only be predicted by PD-L1 expression when the treatment was administered in the first line setting.
Moreover, use of PD-1/PD-L1 checkpoint inhibitors in mTNBC have shown durable responses as monotherapy but with high heterogeneity from study to study and its supremacy over chemotherapy alone is controversial (11)(12)(13)(14)(15)(16)(17)(18). In fact, combined approach has also yield contradictory results in phase III trials in terms of PFS and OS (16,18). Although PD-L1 expression was able to predict clinical response when analysis was restricted to singleagent and combined approach separately, 1-year PFS and 2-year OS was only predicted when immunotherapy was combined with chemotherapy. It may indicate that the two treatments are applied in combination may induce strong and long-term responses which also result in survival benefits. In addition, preclinical evidence also indicates a synergistic interplay between these two treatments. Chemotherapeutic agents have shown augment mutational load and neoantigen presentation, suppress the immune-suppressive cells, sensitize tumor cells to effector cytokines produced by T cells, induce the expression of major histocompatibility complex class I (MHC-I) and PD-L1 on tumor cells (65)(66)(67)(68)(69)(70)(71)(72)(73). These outcomes firmly support the use of PD-L1 expression as prognostic and predictive of response to immune checkpoint inhibition in mTNBC. As expected, the combination of two treatments has led to added toxicity mainly observed in the form of immune-related adverse events. Toxicity associated with immune checkpoint inhibitors has been recognized as a separate entity of special consideration and management needs (74). Nonetheless, no late-onset or long-term safety concerns were reported with addition of PD-1 checkpoint inhibitors.
Our study is limited by several factors. First, a number of nonrandomized clinical trials were included which may impart certain degree of inevitable heterogeneity. Second, certain studies involved small number of patients which contributed high heterogeneity to the pooled outcomes of ORR and adverse events of single-arm studies. Certain dissimilarities were noticed in these studies in reference to the type of agent (PD-1 or PD-L1 inhibitor), chemotherapy regimen, PD-L1 expression assessment assays, treatment combination (IT versus ICT) and line of therapy. Subgroup analysis were undertaken to address these issues; nonetheless, these differences may still impact pooled analysis.

Conclusion
In summary, the results indicate that PD-L1 expression can successfully predict objective response rate and 2-year OS in mTNBC patients receiving PD-1/PD-L1 checkpoint inhibitors plus/minus chemotherapy. Short-term progression-free survival (1-year) is only predicted by PD-L1 expression when treatment is administered in the first-line setting and with chemotherapy. It can also predict the ORR, PFS and OS for PD-L1+ mTNBC patients receiving PD-1 checkpoint inhibitors plus chemotherapy as compared to chemotherapy alone. Overall, PD-L1 expression can be a useful biomarker of efficacy of PD-1/PD-L1 checkpoint inhibitors in mTNBC.

Data availability statement
The original contributions presented in the study are included in the article/Supplementary Material. Further inquiries can be directed to the corresponding authors.