The impact of combined PD-L1 positive score on clinical response to nivolumab in patients with advanced esophageal squamous cell carcinoma

Nivolumab is recommended for patients with advanced esophageal squamous cell carcinoma (aESCC) refractory or intolerant to fluoropyrimidine- and platinum-based chemotherapy regardless of the tumor proportion score (TPS). However, the role of combined positive score (CPS) in predicting nivolumab efficacy remains unclear. We aimed to study whether TPS or CPS is a more suitable biomarker for predicting nivolumab efficacy in these patients. We retrospectively collected data from patients with aESCC treated with fluoropyrimidines and platinum and subsequently received nivolumab monotherapy between January 1, 2014 and September 15, 2020. Next, we evaluated the efficiencies of TPS and CPS in predicting the clinical response to nivolumab using PD-L1 IHC 22C3 pharmDx assay. This study included 50 patients (CPS groups: ≥ 10/1–10/ < 1, n = 24/18/8, respectively; TPS groups, ≥ 10%/1%–10%/ < 1%, n = 17/8/25, respectively). The median progression-free survival was 3.2, 2.5, and 1.5 months in the ≥ 10, 1–10 [hazard ratio (HR) vs. CPS of ≥ 10 group, 1.01; p = 0.98; adjusted HR, 1.33; p = 0.56], and < 1 CPS groups (HR vs. CPS of ≥ 10 group, 3.44; p = 0.006; adjusted HR, 1.67; p = 0.41), respectively. For the patients with CPS of ≥ 10/1–10/ < 1 and TPS of ≥ 10%/1%–10%/ < 1%, the objective response rate was 30%/25%/0% and 36%/0%/19% and the disease control rate was 60%/50%/12% (p = 0.06) and 65%/40%/38% (p = 0.30), respectively. This study suggests that a CPS of < 1 is not a strong predictor of efficacy but can predict the absence of response to nivolumab in patients with aESCC.


Introduction
Esophageal carcinoma is the seventh most common and sixth most life-threatening cancer type worldwide [1]. Esophageal squamous cell carcinoma (ESCC) accounts for the majority of esophageal carcinoma histological subtypes in Asia. Patients with advanced ESCC (aESCC) have a poor prognosis, with a median overall survival (OS) of 7.6-11.1 months [2][3][4][5]. However, to the best of our knowledge, no randomized clinical trial has compared the survival of patients with aESCC between treatment with first-line chemotherapy and best supportive care. Nevertheless, fluoropyrimidines plus platinum has been widely selected as first-line treatments until pembrolizumab plus chemotherapy was proposed to significantly improve OS in patients with untreated aESCC in a KEYNOTE-590 trial [6].
Recent studies have compared nivolumab and pembrolizumab, including human monoclonal anti-programmed cell death-1 (PD-1) antibodies, with chemotherapy in patients with aESCC with refraction or intolerance to fluoropyrimidines and platinum [7,8]. The ATT RAC TION-3 trial demonstrated that nivolumab significantly improves the OS of patients with aESCC regardless of the tumor proportion score (TPS) of programmed cell death ligand-1 (PD-L1) [7]. On the other hand, the KEYNOTE-181 trial revealed that pembrolizumab extends the OS of patients with advanced esophageal carcinoma (including adenocarcinoma) with a PD-L1 combined positive score (CPS) of ≥ 10 tumors [8]. Based on the results of the KEYNOTE-181 trial, CPS has been proposed as a factor useful in the indication of pembrolizumab for patients with aESCC. Nevertheless, the role of CPS in predicting the efficacy of nivolumab in these patients remains unclear. Furthermore, knowledge regarding whether CPS or TPS is more suitable as a biomarker for nivolumab efficacy in patients with aESCC may facilitate the development of new therapeutic agents in the future.
This retrospective study sought to evaluate the efficiencies of CPS and TPS in the prediction of clinical response to nivolumab in patients with aESCC who previously received fluoropyrimidine-and platinum-based chemotherapy.

Patients
We retrospectively evaluated the data of patients with aESCC who received nivolumab monotherapy at Aichi Cancer Center Hospital, Japan. Inclusion criteria were as follows: histological diagnosis of ESCC; initiation of nivolumab monotherapy between January 1, 2014 and September 15, 2020; resistance or intolerance to fluoropyrimidines and platinum; absence of comorbid malignancies; availability of tissue specimens for immunohistochemistry (IHC) and PD-L1 expression analysis before nivolumab initiation; and an Eastern Cooperative Oncology Group Performance Status (ECOG PS) of 0-2. Then, 240 mg nivolumab was intravenously administered for > 30 min every 2 weeks, 360 mg every 3 weeks, or 3 mg/kg every 2 weeks. Neoadjuvant chemotherapy was regarded as the first-line treatment if relapse occurred within 6 months after the last administration of the therapeutic agent. This study was approved by the institutional review board of the Aichi Cancer Center Hospital (R021076).

PD-L1 immunostaining
PD-L1 was immunostained using PD-L1 IHC 22C3 pharmDx assay (22C3 pharmDx, Dako, Agilent Technologies, Carpinteria, CA, USA) according to the manufacturer's instructions. Staining was performed using Autostainer Link 48 (Dako, Agilent Technologies, Carpinteria, CA, USA). CPS was calculated as follows: the number of PD-L1-positive cells (tumor cells, macrophage, and lymphocytes) divided by the total number of tumor cells, multiplied by 100. TPS was calculated as follows: the percentage of PD-L1-positive tumor cells per the total number of total tumor cells. We used a three-tiered classification system for CPS. A CPS of < 1 was used based on previous important clinical trials on head and neck squamous cell carcinoma (HNSCC) and aESCC, in which CPSs of < 1 and ≥ 10, respectively, reliably differentiated responders from nonresponders for the indication of nivolumab [8,9]. Furthermore, we investigated whether a CPS of ≥ 20 or ≥ 50 can better identify patients with favorable outcomes. Based on a previous clinical trial on aESCC, patients were classified into the following three categories depending on their TPS: ≥ 10%, 1%-10%, and < 1% [7][8][9]. CPS and TPS were evaluated by two pathologists (KT and WH) who were blinded to patients' responses to nivolumab and outcomes.

Outcomes
Progression-free survival (PFS) was defined as the period between nivolumab initiation and the date of disease progression or death due to any reason, whichever occurred first. If a patient discontinued nivolumab for reasons other than disease progression or death, the patient was censored as of the date the physician decided to discontinue nivolumab. OS was defined as the period between nivolumab initiation and death due to any cause. Tumor responses were evaluated according to the Response Evaluation Criteria in Solid Tumors version 1.1. Furthermore, the deepness of response (DpR) was defined as the rate of tumor shrinkage from that observed on baseline CT.

Statistical analysis
To evaluate the characteristics of each group, Pearson's chisquare tests were performed using CPS or TPS. PFS and OS were determined using Kaplan-Meier analysis. Furthermore, hazard ratios (HRs) associated with PFS and OS were calculated using a univariate Cox proportional-hazards model. Afterward, adjusted HRs were calculated using a multivariate Cox proportional-hazards model with specific patient characteristics. These characteristics included age (≥ 65 vs. < 65 years), sex (male vs. female), ECOG PS (0 vs. ≥ 1), nivolumab treatment line (2nd vs. ≥ 3rd), stage (recurrent vs. nonrecurrent), previous esophagectomy (yes vs. no), previous radiotherapy (yes vs. no), previous taxane use (yes vs. no), number of metastatic sites (1 vs. ≥ 2), presence of lymph metastasis (yes vs. no), presence of liver metastasis (yes vs. no), and presence of lung metastasis (yes vs. no). Next, we assessed the differences in objective response rate (ORR) and disease control rate (DCR) of the patients using CPS or TPS with Pearson's chi-square tests. p values of < 0.05 were considered significant. Finally, the area under the curve (AUC) in the receiver-operating characteristic (ROC) curve was estimated to assess the discriminatory ability of CPS and TPS for predicting response, disease control, and disease progression in patients with aESCC treated with nivolumab. In the ROC curve, the y-axis shows sensitivity and the x-axis shows the false-positive rate. All statistical analyses were performed using JMP ® 16.1.0 (SAS Institute Inc., Cary, NC, USA).
In patients with measurable lesions who had both TPS of < 1% and CPS of ≥ 1 (n = 13), ORR and DCR were 31% and 54%, respectively. However, we observed no response and only 13% of DCR in patients with TPS of < 1% and CPS of < 1 (n = 8).
The AUC of CPS for response, disease control, and disease progression was 0.604, 0.622, and 0.622, whereas that of TPS was 0.468, 0.480, and 0.480, respectively. The AUC of CPS was numerically better than that of TPS (Supplementary Fig. 1).
Finally, we categorized the CPS group into five subgroups (CPS < 1, CPS 1-10, CPS 10-20, CPS 20-50, and CPS ≥ 50) and performed a validation analysis to identify the optimal three-tier stratification of CPS that is best associated with response to nivolumab or clinical outcomes (Supplementary Table 3a). The CPS ≥ 50 group comprised only two patients. Furthermore, ORR was 14% for the CPS 20-50 group, which exhibited outcomes poorer than the CPS 10-20 group. No gradual increase in ORR was observed even when the CPS ≥ 10 group was subgrouped further. Although the nonresponders were classified in the CPS < 1 group, the ORR and DCR of patients with a CPS of ≥ 1 were 28% and 56%, respectively (Supplementary Table 3b).

Discussion
To the best of our knowledge, this is the first study to evaluate the association between PD-L1 expression scores (CPS and TPS) and tumor response to nivolumab in patients with aESCC. Patients with a CPS of < 1 showed poor response to nivolumab and shorter survival outcomes than those in other groups. In addition, a CPS of < 1 indicated the absence of response to nivolumab in previously treated patients with aESCC.
Although combination therapy with anti-PD-1 antibodies is the predominant first-line treatment for patients with aESCC, some patients, such as those showing early relapse after the last administration of neoadjuvant chemotherapy or definitive chemoradiation therapy, still require nivolumab monotherapy as the standard of care. A biomarker to predict nivolumab efficacy may help plan suitable treatment regimens. In addition, identifying a nivolumab biomarker may contribute to the development of nivolumab-based treatment strategies for aESCC in the future-for example, as a stratified factor, a factor for subgroup analysis, or an eligibility criterion for future clinical trials. TPS and CPS are used to score the extent of PD-L1 expression in cancer cells as well as the surrounding immune cells in various malignancies. TPS functioned as an efficient biomarker in phase 3 trials on the efficacy of pembrolizumab for non-small cell lung cancer (NSCLC); thus, it has been accepted as a useful biomarker for indicating the first-line pembrolizumab treatment in patients with NSCLC [10][11][12]. However, based on the results of phase 3 trials, second-line nivolumab showed survival benefits in patients with NSCLC or HNSCC regardless of their TPS [13,14]. In addition, TPS did not differentiate between responders and nonresponders in other malignancies such as gastric cancer or renal cell carcinoma [15,16]. Therefore, TPS is not a universal biomarker for immune checkpoint inhibitors; this fact indicates an urgent need for another scoring system of PD-L1 expression for therapeutic indications in patients with various cancers. The KEYNOTE-048 trial showed that although pembrolizumab plus chemotherapy or pembrolizumab monotherapy extended the OS of patients with HNSCC, patients in the CPS < 1 group exhibited no benefits [9]. Thus, CPS may help select the first-line regimen for patients with HNSCC. Furthermore, pembrolizumab extends the OS in patients with gastric or esophageal cancer who have high CPS [8,17]. In this study, although the AUCs of CPS for response, disease control, and disease progression were not high (0.604, 0.622, and 0.622), they were higher than those of TPS for all efficacy outcomes. Our findings suggest that CPS is a better factor than TPS for predicting In this study, the proportion of CPS and TPS groups was similar to that of the previous phase 3 trials [7,8]. PD-L1 staining was performed using PD-L1 IHC 22C3 pharmDx assay, similar to the method used in the KEYNOTE-181 trial. However, TPS was evaluated using the PD-L1 IHC 28-8 pharmDx assay in the phase 3 ATT RAC TION-3 trial. In previous clinical trials, although the 22C3 pharmDx assay was used to evaluate the efficacy of pembrolizumab, the 28-8 pharmDx assay was selected in the present study to assess nivolumab efficacy. Krigsfeld et al. reported that concordance was high between the 28-8 and 22C3 assays across several tumor types and biopsy locations [18]. They studied various types of malignancies (e.g., lung cancer, melanoma, HNSCC, and urothelial carcinoma). Thus, this result may be substantially applicable to ESCC.
Our study has some limitations. First, the sample size was limited in this retrospective observational study. Second, PD-L1 expression immediately before nivolumab initiation could not be evaluated. Nevertheless, the reliability of CPS to predict the efficacy of nivolumab should be validated in future studies. Patients with aESCC with a CPS of < 1 showed no response to nivolumab. Therefore, chemotherapy may be an appropriate option for these patients, although the efficacy of chemotherapy has not been investigated in patients with aESCC with a CPS of < 1. Hence, a CPS-based strategy that helps plan individualized treatment regimens for patients with ESCC should be explored in the future.
In conclusion, our study suggests that a CPS of < 1 is not a strong predictor of efficacy but indicates the absence of response to nivolumab in patients with aESCC.