We collected retrospective data from 100 patients with advanced GC (n = 59) or UC (n = 41) that were treated with palliative chemotherapy (n = 100) and anti-programmed death 1 (PD-1)/PD-ligand (L)-1 immunotherapy (n = 49) at Samsung Medical Center between December 2016 and January 2020. The median age was 61.0 (33–81) years and 30 (61.2%) patients were male. All the patients present with GC were stage IIB–IV disease at diagnosis and have experienced local recurrence or metastasis at treatment for ICI. For UC patients, they were all locally advanced stage II–IIIb disease stages (Supplementary Table S1). Responses of the 49 patients treated with immunotherapy were assessed every 6–12 weeks according to the Immune Response Evaluation Criteria in Solid Tumors (iRECIST) (17). Data from patients with at least 6 weeks of follow up were included. The primary clinical endpoint was the objective response rate (ORR), defined as a complete (CR) or partial (PR) response. Patients with progressive (PD) or stable (SD) disease were classified as non-responders. Clinicopathological data were retrospectively extracted from electronic medical records. This study proceeded in accordance with the Institutional Review Board guidelines (IRB No. 2018-09-041-001) for data analysis and investigational treatment, and written informed consent from the patients was also obtained to analyze their innominate data.

Total RNA was isolated from FFPE tumor tissues using the ReliaPrep™ FFPE Total RNA Miniprep System (Promega Corp., Madison, WI, USA), and a amplified using a high-capacity cDNA reverse transcription kit (Thermo Fisher Scientific Inc., Waltham, MA, USA) as described by the manufacturer. Target genes were analyzed using a gene expression assay with forward and reverse primers and an Applied Biosystems FAM-labeled MGB TaqMan^TM probe (Thermo Fisher Scientific Inc.) as we previously described (18). We found that the PD-L1 IHC results correlated with those of NanoString nCounter assays (19), we used CD274 TaqMan probes spanning exon 1–2 (assay ID; Hs01125296_m1), 3–4 (assay ID; Hs00204257_m1), and 5–6 (assay ID; Hs01125301_m1) boundaries for qRT-PCR (Supplementary Figure S1). These sequences were amplified by PCR in triplicate under the following conditions using QuantStudio 6 (Thermo Fisher Scientific Inc.): 2 min at 50°C and 10 min at 94°C, followed by 40 cycles of 95°C for 15 s and 60°C for 60 s. Threshold cycle (Ct) values for each sequence were calculated for each and averaged, and normalized to the mean of the reference gene GUSB2 (assay ID: Hs99999908_m1), which was stably expressed (18). The mRNA expression of each gene was measured using the 2^^-ΔCt (ΔCt = ΔCt_{target gene−}ΔCt_GUSB2) method.

Gastric FFPE tissue blocks were cut into 4-μm sections and stained using an Autostainer Link 48 system and Dako PD-L1 IHC 22C3 pharmDx kits (both from Agilent Technologies Inc., Santa Clara, CA, USA) (2). A rabbit anti-human PD-L1 monoclonal antibody (clone SP142; Ventana Medical Systems, Tucson, AZ, USA) was used as described for UC samples (20). The CPS of PD-L1 expression was calculated as the number of PD-L1-stained GC tumors and ICs divided by the total number of viable tumor cells, multiplied by 100. The concordance rate between qRT-PCR and IHC was evaluated using CPS cut-offs of 1 and 10 for GC. Infiltrative ICs covering ≥ 5 of a UC tumor area were defined as PD-L1-positive. For positive control, we used positive cell lines provided by PD-L1 IHC 22C3 pharmDx and tonsil tissues. For negative control, we used MCF-7 cell lines provided by PD-L1 IHC 22C3 pharmDx. Benign human tonsil is tissue control as it contains both positive and negative staining epithelial and immune cells and can serve as both a positive and negative tissue control for VENTANA PD-L1 (SP142) Assay staining (21).

We used CPS ≥1 and ≥ 10 for GC, and IC ≥5 for UC to compare IHC with qRT-PCR. To calculate the sensitivity, specificity, positive (PPV) and negative (NPV) predictive values, and accuracy, a positive IHC result was considered as CPS ≥1 or ≥ 10 for GC, and IC ≥ 5 for urothelial carcinoma. Predicted responses based on tumor type, IHC results, and qRT-PCR results were evaluated using logistic regression.

The ORR (CR/PR) and disease control rate (DCR; CR/PR/SD) were compared with the CD274 mRNA qRT-PCR results using two-tailed unpaired Student t-tests. The diagnostic values of panels were assessed by calculating the area under the receiver operating characteristics (ROC) curve (AUC). Kaplan–Meier estimates of progression-free (PFS) and disease-specific survival (DSS) were compared using log-rank tests. All graphs were generated using GraphPad Prism v. 9.0 (GraphPad Software Inc., San Diego, CA, USA). Statistical significance was set at P < 0.05. All data were statistically analyzed using SPSS software version 27.0 (IBM Corp., Armonk, NY, USA).

The 22C3 pharmDx assay identified PD-L1 positivity with CPS ≥ 1 and ≥ 10 in 32 (54.2%) and 13 (22 %) of 59 GC samples, respectively. The mean PD-L1 CPS in GC was 9.24 (0–95). The Ventana SP142 assay identified PD-L1 positivity with IC ≥ 5 in 12 (29.3%) of 41 UCs. The mean PD-L1 IC in urothelial carcinomas was 10.46 (0–95) (Figure 1).

The mean RQ (range) of relative CD274 mRNA expression spanning exons 1–2, 3–4, and 5–6 were 0.1004 (0–2.4897), 0.2371 (0–7.5214), and 0.0928 (0–3.7064), respectively. These values closely correlated (Spearman correlations: r² = 0.92 for exons 1–2 and 3–4; r² = 0.89 for exons 1–2 and 5–6, and r² = 0.99 for exons 3–4 and 5–6; Figure 2A). The PD-L1 scores in 100 evaluated samples closely correlated with CD274 mRNA expression spanning exons 1–2 (r² = 0.81), 3–4 (r²=0.65), and 5–6 (r² = 0.59; Figure 2A). In GC, The PD-L1 CPS score with 22C3 pharmDx significantly correlated with the exon 1–2 (r² = 0.81), 3–4 (r² = 0.67), and 5–6 (r² = 0.62) junctions of CD274 (Figure 2B). The Ventana SP142 PD-L1 IC score was significantly associated in UC with exon 1–2 (r² = 0.93), exon 3–4 (r² = 0.82), and exon 5–6 (r² = 0.76) junctions of CD274 (Figure 2C).

The RQ cutoffs of CD274 mRNA expression in exon 1–2, 3–4, and 5–6 junctions were evaluated as the AUC based on PD-L1 CPS cut-offs of 1 and 10 for GC and PD-L1 IC cut-offs of 5 for UC (Supplementary Table S2 and Supplementary Figure S2). At a CPS cutoff of 10, the highest AUC in GC was 0.783, obtained from CD274 mRNA expression at the exon 1–2 junction with a cut-off of 0.0722 (P < 0.0001). The highest AUC of UC based on PD-L1 IC cut-offs of IC 5 was 0.781, obtained from CD274 mRNA expression in the exon 1–2 junction with a cut-off of 0.0480 (P < 0.0001).

Between May 2018 and October 2020, 49 patients were treated with anti PD-1/PD-L1 agents, and treatment responses to treatment with pembrolizumab (n = 16), nivolumab (n = 16), atezolizumab (n = 13), and durvalumab (n = 4) were evaluated during > 6 weeks of followup (Supplementary Table S1). The median number PD-1/PD-L1 cycles was 8.9 (range, 1–37) as of May 20, 2021, and the patients were followed up for a median of 11.3 months. Table 1 summarizes the clinicopathological characteristics of the patients treated with anti-PD-1/PD-L1.

Anti-PD-1/PD-L1 responders (CR/PR, n = 16) and non-responders (PD/SD, n = 33) were identified using the iRECIST category of ORR. The expression of PD-L1 (P = 0.010) and high CD274 mRNA expression (P < 0.001) were significantly associated with the response to immunotherapy. The ROC curve for the predictive performance of PD-L1 IHC and mRNA expression of CD274 at exon 1–2 was discriminatory. The AUC and 95% confidence intervals (CIs) were 0.76 (0.61–0.91) for PD-L1 and 0.75 (0.59–0.91) for mRNA expression of CD274 exon 1–2. These findings were similar using the iRECIST category of DCR (CR/PR/SD, n = 30 and PD, n = 19). Furthermore, PD-L1 expression (P = 0.015) and high CD274 mRNA expression (P = 0.038) predicted responses to immunotherapy with AUCs of 0.70 (0.55–0.86) and 0.68 (0.53–0.83), respectively. In GC, the expression of PD-L1 (P = 0.002) and high CD274 mRNA expression (P=0.041) were significantly associated with the response to immunotherapy. In UC, the expression of PD-L1 (P = 0.147) and high CD274 mRNA expression (P=0.008) did not reach statistical significance in predicting response to immunotherapy (Figure 3A). The ROC curve for the predictive performance of PD-L1 IHC and mRNA expression of CD274 at exon 1–2 was discriminatory. In GC, the AUC and 95% confidence intervals (CIs) were 0.80 (0.63–0.97) for PD-L1 and 0.69 (0.47–0.92) for mRNA expression of CD274 exon 1–2. In UC, the AUC and 95% Cis were 0.68 (0.36–0.99) for PD-L1 and 0.87 (0.67–1.00) for mRNA expression of CD274 exon 1–2 (Figure 3B). These findings were similar using the iRECIST category of DCR (CR/PR/SD, n = 15 and PD, n = 18) in GC. PD-L1 expression (P = 0.008) and high CD274 mRNA expression (P = 0.017) predicted responses to immunotherapy with AUCs of 0.73 (0.56–0.90) and 0.71 (0.53–0.90), respectively, in GC. In UC, anti-PD-1/PD-L1 responders (n = 15) and non-responders (n = 1) were identified using the iRECIST category of DCR. PD-L1 expression (P = 0.375) and high CD274 mRNA expression (P = 0.250) predicted responses to immunotherapy with AUCs of 0.67 (0.43–0.91) and 0.90 (0.71–1.00), respectively (Figures 3C, D).

The PFS was closely associated with PD-L1 expression (P = 0.018) and high CD274mRNA expression spanning the exon 1–2 junction (P = 0.010) in GC (Figure 4A). The association was also similar between DSS and PD-L1 expression (P = 0.047). However, DSS was not significantly associated with mRNA expression (P = 0.134); Figure 4B). The expression of PD-L1 was significantly associated with PFS (P = 0.016) and DSS (P = 0.009) in UC, whereas the CD274mRNA expression at exon 1–2 junction did not significantly correlate with PFS and DSS (Supplementary Figure S3).

The clinical value of PD-L1 assessment with IHC and qRT-PCR was compared using the standard parameters of sensitivity, specificity, PPV, NPV, and accuracy (Table 2). We used two cut-offs for GC samples (CPS ≥ 1% and 10%; RQ ≥ 0.0276 and ≥ 0.0772) to ensure the optimal performance to predict responses for immunotherapy. The CPS ≥ 1% for PD-L1 was the most sensitive (90%), and qRT-PCR with a RQ cutoff of 0.0722 was the most specific (100%) in GC. The sensitivity was highest in GC samples with CPS ≥ 1 (90%) although the PPV was very low (50%). The sensitivity (66.7%) and specificity (90%) of detecting UC were higher with qRT-PCR and the AUC values higher than those in PD-L1 IHC.