T cell-inflamed gene expression profile and PD-L1 expression and pembrolizumab efficacy in advanced esophageal cancer

Aim: Investigate the relationship between response to pembrolizumab and expression of the 18-gene T cell-inflamed gene expression profile (Tcell inf GEP) or PD-L1 combined positive score (CPS) in esophageal cancer. Materials & methods: This analysis included heavily pretreated patients with advanced / metastatic esophageal / gastroesophageal junction adenocarcinoma or squamous cell carcinoma who received pembrolizumab in the single-arm, phase II study KEYNOTE-180. PD-L1 CPS was evaluated with PD-L1 IHC 22C3 pharmDx. Results: In patients with squamous cell carcinoma, trends toward enrichment for responders were observed for patients with PD-L1 CPS ≥ 10 tumors. In patients with adenocarcinoma, a trend was observed for Tcell inf GEP but not for PD-L1. Conclusion: Tcell inf GEP and PD-L1 CPS may enrich for responders to pembrolizumab in patients with esophageal cancer.

Biomarkers currently used or under investigation that predict immunotherapy response include expression of PD-L1 [1], tumor mutational burden [2,3] and gene expression profiles (GEPs) [2-4].The T cell-inflamed GEP (Tcell inf GEP) [2-4] was developed using data from clinical studies of pembrolizumab [5].The GEP has been examined in esophageal tumor samples, including those of 18 patients who were included in the population used to develop the Tcell inf GEP [4].In pembrolizumab-treated patients with advanced esophageal cancer in KEYNOTE- 028, the gene signature score (as a continuous variable) showed trends toward statistical significance in response (one-sided p = 0.107) and progression-free survival (PFS; one-sided p = 0.053) [6].
KEYNOTE-180 (NCT02559687) was a single-arm, phase II study of pembrolizumab in heavily pretreated patients with advanced/metastatic esophageal/gastroesophageal junction cancer, irrespective of PD-L1 status (n = 121); objective response rate (ORR) was 9.9%, median duration of response was not reached, and median overall survival (OS) was 5.8 months [7].
We explored the relationship between clinical outcomes of pembrolizumab and the Tcell inf GEP score and PD-L1 combined positive score (CPS) status by histology (squamous cell carcinoma [SCC] and adenocarcinoma) in patients with esophageal cancer from KEYNOTE-180.

Materials & methods
The design of KEYNOTE-180 has been described [7]; details are included in the supplement.In the current analysis, outcomes were assessed based on Tcell inf GEP score and PD-L1 CPS status; analysis by histologic subgroup was exploratory.Tumor expression levels of 18 genes were determined using the NanoString nCounter Analysis System from tumor samples, and the individual expression levels of the genes were combined as a weighted average to obtain a single Tcell inf GEP score.A prespecified, validated cutoff was used to divide tumors into 'low' and 'non-low' categories [4].PD-L1 expression was characterized using PD-L1 IHC 22C3 pharmDx (Agilent) and measured using CPS.
We report efficacy data in all patients who received ≥1 dose of pembrolizumab and had evaluable Tcell inf GEP or PD-L1.Only confirmed objective responses, defined as complete response plus partial response, were reported.
PD-L1 CPS and the Tcell inf GEP scores showed a modest positive correlation (Figure 1A).The prevalence of Tcell inf GEP and PD-L1 CPS by region and histology are reported in Supplementary Table 3; 46 tumors (38.0%) were discordant for PD-L1 CPS and Tcell inf GEP score (Table 1).A summary of responses is reported in Table 2. AUROC estimates were modest overall for Tcell inf GEP and PD-L1 when pooling histologies, but trends suggest an association with increased response to pembrolizumab (Figure 1).The clinical utility of the PD-L1 CPS cutoff is shown in Supplementary Table 4.
In patients with SCC, median PFS was 2.1 and 2.1 months by Tcell inf GEP non-low and Tcell inf GEP low score and 2.0 and 2.1 months by PD-L1 CPS ≥10 and CPS <10 status (Figure 2A-B).Median OS was 7.7 and 6.2 months by Tcell inf GEP non-low and Tcell inf GEP low score and 7.5 and 6.1 months by PD-L1 CPS ≥10 and CPS <10 status (Figure 2C-D).PFS and OS medians were similar across Tcell inf GEP and PD-L1 CPS subgroups in patients with adenocarcinoma.

Discussion
In the primary analysis of KEYNOTE-180, pembrolizumab provided durable antitumor activity and an acceptable safety profile in some patients with heavily pretreated advanced/metastatic esophageal cancer [7].Patients with SCC demonstrated higher ORR than patients with adenocarcinoma (14.3 vs 5.2%) [7].Response was also evaluated by PD-L1; ORR was 13.8 versus 6.3% in patients with CPS ≥10 versus <10 tumors [7].These data suggest the possibility that biomarkers and disease characteristics can be used to enrich for higher efficacy, which was further investigated in the current analysis.
Higher levels of PD-L1 expression have generally been shown to correlate with PD-1/PD-L1 inhibitor response across tumor types, with some variability, possibly because of differing definitions of PD-L1 positivity and assays [8].PD-L1 CPS has been incorporated into regulatory approvals for multiple indications of pembrolizumab, including CPS ≥10 for esophageal cancer [5].The phase III KEYNOTE-181 trial of pembrolizumab established PD-L1 CPS ≥10 as a cutoff for efficacy in SCC [9].Other biomarkers may also be useful in predicting response to pembrolizumab in esophageal cancer.
In this analysis, we present clinical outcomes based on Tcell inf GEP score and PD-L1 CPS status by histology.Although both Tcell inf GEP and PD-L1 are viewed as indicative of IFN-γ-driven inflammation, there is considerable discordancy between these two biomarkers, with lower correlation than has been observed in other tumor types [3,10], indicating different aspects of the tumor microenvironment are captured by each marker.Consistent with the findings in the primary analysis, response rates were numerically higher among patients with SCC than among those with adenocarcinoma regardless of Tcell inf GEP score or PD-L1 CPS status.In the analysis of ORR, Tcell inf GEP non-low appeared to enrich for response among patients with adenocarcinoma.PD-L1 CPS ≥10 enriched for response among with SCC but not for patients with adenocarcinoma.Estimates of median PFS were similar across all biomarker subgroups, whereas median OS was marginally longer among patients with SCC whose tumors were classified as Tcell inf GEP non-low versus Tcell inf GEP low and whose tumors expressed PD-L1 CPS ≥10 versus CPS <10.Patients with adenocarcinoma had similar median OS across the Tcell inf GEP and PD-L1 CPS subgroups.
Limitations of the current study include the single-arm nonrandomized study design, the modest sample sizes and the small number of responders, leading to wide CIs; all results should be interpreted with caution.Larger randomized studies will facilitate better interpretation of the relationship between these inflammatory biomarkers and time-to-event end points.

Conclusion
Our findings suggest that these measures of inflammation -PD-L1 and Tcell inf GEP -may enrich for positive clinical outcomes from treatment with pembrolizumab.In SCC, a trend toward enrichment was observed for patients with PD-L1 CPS ≥10 tumors.In adenocarcinoma, a trend was observed for Tcell inf GEP but not for PD-L1 CPS.Additional studies are needed to facilitate understanding of the molecular correlates in adenocarcinoma.The cytokine IFN-γ has an important role in immune regulation that can be exploited by cancer cells [4]; work in a pan-tumor setting led to the development of an 18-gene Tcell inf GEP as a biomarker for pembrolizumab efficacy [2,4].Findings in KEYNOTE-180 are consistent with analyses using esophageal tumor samples from KEYNOTE-028 showing an improved propensity for response to pembrolizumab with higher levels of IFN-γrelated gene expression [6].

Supplementary data
To view the supplementary data that accompany this paper please visit the journal website at: www.futuremedicine.com/doi/suppl/10.2217/fon-2021-1134

Figure 1 .
Figure 1.Correlation of T cell-inflamed gene expression profile and PD-L1 combined positive score and associations with response to pembrolizumab.(A) The Spearman correlation was 0.17.AUROC for (B) Tcell inf GEP score and (C) PD-L1 CPS status.AUROC: Area under the receiver operating curve; CPS: Combined positive score; CR: Complete response; NR: Nonresponder; PR: Partial response; R: Responder; Tcell inf GEP: T cell-inflamed gene expression profile.

SCCSCCFigure 2 .
Figure 2. Kaplan-Meier estimates of survival outcomes and the T cell-inflamed gene expression profile score and PD-L1 combined positive score status by histology.Kaplan-Meier estimates of PFS by histology, (A) Tcell inf GEP score, and (B) PD-L1 CPS status and Kaplan-Meier estimates of OS by histology, (C) Tcell inf GEP score and (D) PD-L1 CPS status.AC: Adenocarcinoma; CPS: Combined positive score; OS: Overall survival; PFS: Progression-free survival; SCC: Squamous cell carcinoma; Tcell inf GEP: T cell-inflamed gene expression profile.

Table 1 .
T cell-inflamed gene expression profile score by PD-L1 combined positive score status.

55 Tcell inf GEP non-low n = 26 Tcell inf GEP low n = 37 Tcell inf GEP non-low n = 25 Tcell inf GEP low
Patients who had a baseline assessment but no postbaseline assessment on the data cutoff date because of missing data, discontinuation from the study or death before the first postbaseline imaging.AC: Adenocarcinoma; CPS: Combined positive score; RECIST v1.1: Response Evaluation Criteria in Solid Tumors: version 1.1; SCC: Squamous cell carcinoma; Tcell inf GEP: T cell-inflamed gene expression profile.
†Objective response defined as complete response plus partial response.‡ Disease control defined as complete response plus partial response plus stable disease.§