Effectiveness of anti-PD-1/PD-L1 antibodies in urothelial carcinoma patients with different PD-L1 expression levels: a meta-analysis

Background Urothelial carcinoma ranks the ninth among malignant cancers. We conducted this study to identify which patients could benefit more from the treatment of programmed death-1 (PD–1)/programmed death-ligand1 (PD–L1) inhibitors. Materials and Methods We performed literature searches, combined data from qualified literature and performed comparative analyses on the effectiveness of anti-PD–1/PD–L1 antibodies in patients with different PD–L1 expression levels. Results We divided patients into three groups according to the percentages of PD–L1-positive cells, namely the low- PD-L1 (PD-L1 < 1%), the medium-PD-L1 (PD-L1 ≥ 1 and < 5%) and the high–PD–L1 (PD-L1 ≥ 5%) groups. We found that the high-PD-L1 group responded significantly better than other groups (P = 0.0003, ORs = 0.45, 95%CI: 0.29-071; P = 0.0009, ORs = 0.43, 95%CI: 0.25-0.73, for low-PD-L1 and medium-PD-L1 groups, respectively), while the latter two groups responded similarly (P = 0.90, ORs = 1.06, 95%CI: 0.62-1.83) to both PD–1 and PD–L1 inhibitors. Furthermore, we found that the medium-PD–L1 and high-PD–L1 groups responded similarly to PD-1/ PD-L1 inhibitors (P = 0.65, ORs = 1.11, 95%CI: 0.69–1.77), while the low-PD–L1 group responded better to PD-1 inhibitors than PD-L1 inhibitors (P = 0.046, ORs = 1.92, 95%CI: 0.98–3.89). Conclusions Our results suggest that PD–L1 positive patients should be defined as those with ≥ 5% or greaterPD-L1-positive cells. PD-1 antibodies performed better only in the low-group patients, likely because they could block the interactions of PD–1 with both PD–L1 and PD–L2.


INTRODUCTION
The incidence of urothelial carcinoma has become the ninth of all malignant tumors and the first in urinary tract cancers [1]. It leads to the death of more than 165,000 people every year all over the world [2]. Urothelial cancer can be divided into two groups: upper tract urothelial carcinoma and urinary bladder carcinoma. Both of the them occur more often in males than in females [3]. Platinum-based chemotherapy has been the first-line choice in advanced and metastatic urothelial cancer, but often has very poor prognosis [4]. It is thus urgent that new therapeutic options should be introduced for patients with urothelial cancer.
Tumor microenvironment plays an important role in tumor immunity [5]. Programmed death-ligand1 (PD-L1) is found to be expressed in human tumor-associated antigen presenting cells, and can suppress the immune system through its receptor programmed death-1 (PD-1) and lead to immune escape of tumor cells [6,7]. So far a few PD-1 and PD-L1 antibodies are commercially available, including PD-1 antibodies nivolumab and
It has been shown in many clinical trials that cancer patients with higher expression of PD-L1 received better antitumor effect to PD-1/PD-L1 inhibitors [8,11]. Similar results have been observed in urothelial cancer [11,16]. However, in recent studies there have been some seemingly different observations. For example, some researchers found that patients with ≥ 1% PD-L1-positive cells responded better to PD-1/PD-L1 antibodies; however, some other studies found similar effects only in patients with ≥ 5% PD-L1-positive cells [17]. Although the limited numbers of patients in individual studies could be a reasonable explanation to these different results, they did raise the critical question regarding the threshold for distinguishing PD-1/PD-L1 positive patients from negative ones [18]. In addition, it is still unclear for oncologists that which antibodies (i.e. PD-1 or PD-L1 antibodies) work better for patients with urothelial cancer. We therefore performed a meta-analysis to address the following two questions. First, what threshold should be used for distinguishing PD-1/PD-L1 positive patients from negative ones? Second, which antibodies (i.e. PD-1 or PD-L1 antibodies) work better for patients with urothelial cancer?

Six eligible studies were included for analysis
We retrieved in total 95 articles from public databases using the above-mentioned keywords. After reading abstracts and full texts, we kept six articles for further analysis; in total 828 participants were tested in the six studies. Please consult Figure 1 for the overall workflow of the literature selection, the criteria of the selection and the numbers of articles remained/excluded at each step.
Three out of the six studies evaluated PD-1 antibodies, of which two evaluated nivolumab [19,20] and the other evaluated pembrolizumab [21]. Three studies evaluated PD-L1 antibodies, of which two evaluated atezolizumab [11,16] and the other evaluated durvalumab [22]. All of the studies were single-arm trials (see Table 1 for details).

Patients with higher-ratio of PD-L1-positive cells responded better to PD-1/PD-L1 antibodies
We first tested whether patients with higher-ratio of PD-L1-positive cells responded better to PD-1/PD-L1 antibodies. Either of 1% and 5% was often used in the literature to classify patients into "positive" and "negative" groups, according to whether there are more than 1% or 5% PD-L1-positive cells in a patient.

Patients with PD-L1 < 1% could respond better to PD-1 antibodies than to PD-L1 antibodies
We next sought to compare the effectiveness of PD-1 and PD-L1 antibodies in patients with different PD-L1-positive cell ratios. In the literature, patients were often grouped into the following subgroups: PD-L1 < 1%, PD-L1 ≥ 1% but < 5%, and PD-L1 ≥ 5%; we referred them as to Low-PD-L1, Medium-PD-L1 and High-PD-L1 groups respectively in the following analysis. As shown in Figure  3A, we found no significant differences between the two types of antibodies in the Medium-and High-groups (P > 0.05, Fisher's Exact Test). In some studies, patients of the two groups were often combined; again, we found no significant differences in the combined datasets in the responses to PD-1 and PD-L1 antibodies.
Surprisingly, we found in the Low-group, patients had significantly better objective responses to PD-1 antibodies than to PD-L1 antibodies (P = 0.046, OR = 1.92, 95%CI: 0.98, 3.89; Fisher's Exact Test) ( Figure 3A). It is known that PD-1 antibodies block the interaction between PD-1 with PD-L1 and PD-L2, while PD-L1 antibodies only block the interaction between PD-1 with PD-L1 [7,23]; therefore it is very likely that PD-1 antibodies are more sensitive to lower ratio of PD-L1possitive cells than to PD-L1 antibodies.

5% level of the PD-L1 positive cells should be used as the cutoff to distinguish PD-L1 positive patients from the negative ones
We also compared the responses of patients with different PD-L1-positive cell ratios to PD-1/PD-L1 antibodies. As shown in Figure 3, we found that patients in the High-group responded better than the other two groups www.impactjournals.com/oncotarget  Figure 3B). These results suggested that a PD-L1-positive ratio of 5% or greater instead of 1% should be used as a threshold to define PD-L1 "positive" patients.

DISCUSSION
In this study, we performed a meta-analysis on the effectiveness of anti-PD-1/PD-L1 antibodies in urothelial carcinoma patients with different ratios of PD-L1-positive cells. We first confirmed previous findings that patients with higher ratios of PD-L1-positive cells responded significantly better to both PD-1 and PD-L1 antibodies than those with lower ratios of PD-L1-positive cells.  [22] I/II NCT01693562 Durvalumab [2] All 31% (95%CI 17. [1] PD-1inhibitors: nivolumab and pembrolizumab. [2] PD-L1 inhibitors: atezolizumab and durvalumab. www.impactjournals.com/oncotarget We also checked whether patients responded better to one type of antibodies than the other. We found that in most cases, patients had similar ORRs between the two types of antibodies in the Medium-PD-L1 and High-PD-L1 groups. However, in patients with less than 1% PD-L1-positive cells (the Low-group), the objective response ratio was statistically better to PD-1 antibodies than to PD-L1 antibodies. This is likely that PD-1 antibodies are more sensitive to lower ratio of PD-L1-possitive cells, becausePD-1 antibodies block the interaction between PD-1 with PD-L1 and PD-L2, while PD-L1 antibodies only block the interaction between PD-1 with PD-L1 [7,23]. However, due to limited numbers of patients involved in the studies, further clinical trials should be conducted in order to fully compare the effectiveness of all PD-1/PD-L1 antibodies in patients in the Low-group.
The definition of PD-L1 "positive" patients is not very clear so far [18]. Cutoffs of 1%, 5%, 10%have been used in many studies of urothelial cancer [21,24], as in other cancers, such as NSCLC and melanoma [17,25,26]. For example, in the six studies analyzed, four chose 1% as the cutoff to define "positive" and "negative" patients. Our analysis indicates that patients in the Low-(PD-L1 < 1%) and Medium-(PD-L1 ≥ 1% but < 5%) groups had similar ORRs to PD-1/PD-L1 antibodies; medians OS and median PFS are also similar in the two groups. Conversely, the ORRs to the antibodies in the High-group (PD-L1 ≥ 5%) were significantly better than the other two groups. These results suggested a ratio of PD-L1positive cells of 5% and greater should be used as the cutoff to define PD-L1-"positive" patients in urothelial cancer. The 5% cutoff is also used in other cancers [27]. However, recently Skala and Farajre commended a much higher cutoff (i.e. 50%) for urothelial cancer [28,29]. Due to limited data so far, we could not rule out the possibility that a higher ratio (such as 10% or even 50%) should be used.
In conclusion, PD-L1 positive patients of urothelial cancer response more than negative ones, and we suggest 5% or greater as the cutoff to distinguish them. PD-1 antibodies is better than PD-L1 antibodies in PD-L1 < 1% subgroup with urothelial carcinomas patients. Further clinical trials need to be performed to establish the position of immunotherapy in urothelial cancer.