Recently, blockade of immune checkpoint molecules with monoclonal antibodies has emerged as a promising strategy in several malignancies [1–6]. Programmed death 1 (PD-1), which belongs to the CD28 family of proteins, is a negative costimulatory receptor expressed on the surfaced of activated T cells [7]. The binding of PD-1 and its ligands, PD-L1 and PD-L2 in tumor or immune cells, can inhibit a cytotoxic T-cell response, which leads tumor cells to escape from immune surveillance [7]. Accordingly, blockade of this interaction restores the antitumor activity of T cells [7]. Clinical trials of anti-PD-1/PD-L1 monoclonal antibodies have shown durable anti-tumor response and improved overall survival in several malignancies [1–6].

A phase III ATTRACTION-2 trial of nivolumab, a fully human IgG4 monoclonal antibody (mAb) against PD-1, for patients (pts) with advanced gastric cancer (AGC) after two or more previous line chemotherapies showed a survival benefit, leading to the approval of nivolumab for AGC in Japan [8]. Exploratory analysis of ATTRACTION-2 suggested a survival benefit of nivolumab regardless of PD-L1 expression on tumor cells, thus nivolumab have been used without any restriction by biomarkers [8].

Pembrolizumab, another PD-1 mAb, also demonstrated encouraging anti-tumor activity with acceptable safety for PD-L1 positive AGC in phase II and III trials [9, 10], where PD-L1 expression has been evaluated as combined positive score (CPS) counting both tumor cells and immune cells. A relationship between greater PD-L1 CPS and a greater treatment effect was suggested in phase II and III trials of pembrolizumab [9, 10]. ORR in pts with CPS ≥ 10, CPS ≥ 1, and CPS < 1 (PD-L1-) were 25, 16, and 2%, respectively [10]. Recently, the US Food and Drug Administration approved pembrolizumab for pts with microsatellite instability-high or mismatch repair (MMR) deficient solid tumors including AGC based on the durable response in several trials [11–13]. In addition to PD-L1 expression and MMR deficiency, a small study suggested that high tumor mutation burden (TMB) and EBV positive status were associated with response to pembrolizumab [14]. However, predictive factors of nivolumab for AGC have not yet been evaluated. Also, overlapping of several clinicopathological and molecular features have not yet been discussed in detail.

In order to establish the better selection of pts who may derive greater benefit from PD-1 blockade, we investigated clinicopathological and molecular features of responders to nivolumab for AGC.

In this study, we investigated the characteristics of responders to nivolumab for pts with AGC. To our knowledge, this is the first report to provide detailed information on clinicopathological and molecular features associated with response to nivolumab for AGC.

The results of subgroup analysis of phase II and III trials of pembrolizumab showed that better PS was associated with a higher response rate and longer overall survival [10, 11]. Consistent with these results, pts with PS of 0 had better ORR and PFS compared to those with PS of 1 or 2 in our study. Furthermore, after excluding pts with MMR-D from the analysis, PS of 0 was an only clinical factor significantly associated with responders in pts with MMR-P, suggesting that it is important to assess general condition before the initiation of PD-1 blockade for the prediction of efficacy. Although the exact explanations for the correlation between PS and clinical outcomes of PD-1 blockade were not established, pts with poor PS may not stay on treatment long enough to achieve a response.

In our analysis, PD-L1 expression in TC was significantly associated with responders to nivolumab for AGC, which was contrary to that of subgroup analysis from ATTRACTION-2 [8]. Furthermore, after excluding pts with MMR-D, impact of PD-L1 in TC was still statistically significant. Different anti-PD-L1 antibodies (SP142 or SP263) in our study from those (28–8 or 22C3) in these previous studies of nivolumab or pembrolizumab [9, 10, 14] might affect the predictive value of PD-L1 expression. Also, ORR and PFS tended to be better in pts with CPS ≥ 10 overlapping substantially with PD-L1+ in TC in our analysis; 5 of 14 responders had both CPS ≥ 10 and PD-L1+ in TC. Impact of CPS on the efficacy of PD-1 blockade will also be evaluated in the ongoing phase III KEYNOTE­062 trial (NCT02494583), which compared the efficacy of cytotoxic agents combined to pembrolizumab with that of cytotoxic agents and that of pembrolizumab monotherapy in pts with untreated AGC.

ORR was significantly higher in AGC pts with PIK3CA mutation in our study, which was consistent with a recent study analyzing genomic correlates of response to immune checkpoint blockade in microsatellite-stable solid tumors [16]. It is also suggested that PIK3CA mutation have been linked with APOBEC signatures which is highly proficient at generating DNA breaks whose repair can trigger the formation of single-strand hypermutation substrates [17]. Moreover, in gastric cancer, it has been well known that APOBEC-mutation signature and PIK3CA mutation were frequently observed in EBV+ pts [18]. Meanwhile, it is reported that PIK3CA mutation is strongly associated with the MSI molecular subgroup [19]. Among 4 responders with PIK3CA mutation in our study, 3 were MMR-D, and only additional one patient with MMR-P, no EBV+, and PD-L1 in TC with CPS ≥ 10 had mutation in PIK3CA lie in E542K, which has been reported to be associated with APOBEC signature. Thus, the predictive value of PIK3CA mutation alone in AGC needs further investigations. Most recently, extremely high ORR (100%) of pembrolizumab was reported in 6 pts with EBV+ AGC [14], which was inconsistent with our result showing that 1 of 4 pts with EBV+ (25%) achieved an objective response. Notably, no EBV pts in our study showed CPS ≥ 10, which was different from previous study [14]. Our pervious study showed not all EBV+ tumors showed high PD-L1 expression [15], thus both EBV status and PD-L1 expression should be evaluated in a larger cohort.

High TMB was not associated with response to nivolumab in our study, though it was notable that 4 of 8 responder pts with MMR-P had high TMB. It has been reported that the estimated TMB based on panel sequencing showed relatively high discordance compared with TMB calculated from whole exome sequencing in tumors with relatively low number of mutations [20], which might lead to the results in this study which did not show good correlation between anti-tumor response and TMB. Recently, Kim ST et al. reported that high TMB was a potential biomarker of pembrolizumab for AGC [14]. However, most pts with high TMB had MMR-D status, and not all pts with high TMB achieved an objective response [14]. Thus, precise mechanism regarding the influence of TMB to the efficacy of PD-1/PD-L1 blockade should be investigated in the near future.

Interestingly, ORR was 31% in pts with at least one of the following factors; MMR-D, high-TMB, EBV+, and PD-L1+ in TC vs. 0% in those without these factors, suggesting that pre-screening of these biomarkers might be useful to predict clinical benefit of anti-PD-1/PD-L1 blockade in AGC.

It is important to note the limitations of the present study. First, this was a single-institution study with a limited sample size. Second, we did not analyze PD-L1 expression, MMR, EBV status, and cancer genome alterations in all the pts enrolled in this study, which warrants further evaluations in a larger cohort.

In conclusion, we identified some clinicopathological and molecular characteristics associated with responders to nivolumab for pts with AGC. Combining these features might be useful for the better selection of pts who may derive greater benefit from PD-1 blockade. However, further investigations in larger cohorts are needed to confirm precise biomarkers of PD-1/PD-L1 blockade for AGC.