High PD-L1 expression as a negative prognostic factor in stage III but not in stage II gastric cancer

Abstract

BACKGROUND

The prognostic value of programmed death ligand-1 (PD-L1) expression in patients with gastric or gastroesophageal junction cancer (G/GEJC) at different stages remains unclear. We hypothesized that high PD-L1 expression is associated with poor survival in patients with pathological stage III (pIII) G/GEJC.

AIM

To investigate the relation between PD-L1 expression and survival outcomes in patients with pathological stage II (pII)-pIII G/GEJC.

METHODS

In this retrospective cohort study, we collected the clinicopathological data of 388 consecutive patients with pII/pIII G/GEJC who underwent gastrectomy without neoadjuvant therapy at Nanfang Hospital, Southern Medical University. Postoperative tissue samples were collected to evaluate PD-L1 expression. Combined positive score (CPS) ≥ 1 was defined as PD-L1-positive, and CPS ≥ 5 as high PD-L1 expression. Survival outcomes were analyzed using Kaplan-Meier and Cox proportional hazards models.

RESULTS

In pII G/GEJC, no significant differences were found in 3-year disease-free survival (DFS; 88.5% vs 87.9%, P = 0.939) or 3-year overall survival (OS; 92.3% vs 89.9%, P = 0.621) between PD-L1-positive and PD-L1-negative groups or between high and low PD-L1 expression groups (3-year DFS: 91.8% vs 84.2%, P = 0.138; 3-year OS: 94.2% vs 88.4%, P = 0.233). In pIII G/GEJC, PD-L1-positive patients had poorer 3-year DFS (52.2% vs 67.8%, P = 0.030) and 3-year OS (65.1% vs 78.2%, P = 0.007) than PD-L1-negative patients. High PD-L1 expression was associated with significantly inferior 3-year DFS (50.2% vs 64.8%, P = 0.023) and 3-year OS (64.0% vs 75.1%, P = 0.036) compared to low PD-L1 expression. Multivariate analysis revealed that high PD-L1 expression was independently associated with shorter DFS (HR = 1.624, P = 0.027) and OS (HR = 1.653, P = 0.043) in patients with pIII G/GEJC. Sensitivity analyses confirmed the robustness of the OS findings.

CONCLUSION

High PD-L1 expression serves as an independent adverse prognostic biomarker in stage III but not stage II G/GEJC. These findings indicate a stage-dependent prognostic value of PD-L1 expression in G/GEJC.

Key Words: Survival; Prognosis; Locally advanced gastric cancer; Combined positive score; Programmed death ligand-1

Core Tip: This retrospective cohort study revealed that high programmed death ligand-1 (PD-L1) expression was an independent adverse prognostic factor for disease-free and overall survival in pathological stage III gastric or gastroesophageal junction cancer (G/GEJC). However, in pathological stage II G/GEJC, patients with high PD-L1 expression tended to have better survival despite the lack of statistical significance. These findings align with the subgroup analysis of the KEYNOTE-585 trial, which indicated the clinical benefit of anti-programmed death 1 therapy in stage III G/GEJC but not in stage II. This study provides theoretical support for designing future perioperative precision immunotherapy trials in G/GEJC.

INTRODUCTION

As recent estimates indicate, the global burden of gastric or gastroesophageal junction cancer (G/GEJC) remains substantial[1,2]. In China, 72.32% of patients are diagnosed with locally advanced G/GEJC (including stages II and III), and 41.92% are diagnosed with stage III[3]. The recurrence risk within 3 years for patients with pathological stage III (pIII) G/GEJC ranges from 31.6% to 48.9%[4-6].

The current standard treatment for locally advanced G/GEJC involves D2 gastrectomy with perioperative chemotherapy. Although this therapy has moderately improved survival, the prognosis of these patients remains poor. Between 32.09% and 47.90% of patients with locally advanced G/GEJC die within 5 years of surgery[7-9]. It was previously believed that the factors driving recurrence included advanced TNM stage, presence of signet ring cell carcinoma, and the overexpression of HER-2, etc.[7,10].

Programmed death ligand-1 (PD-L1) is expressed on cancer cells and interacts with programmed death 1 (PD-1) receptors on T cells. This binding inhibits T-cell activation and prevents T cells from attacking cancer cells, thereby facilitating immune evasion and promoting tumor progression[11,12]. Several studies, including CheckMate-649[13], ORIENT-16[14], and KEYNOTE-811[15], have demonstrated that anti-PD-1 treatment can substantially prolong the survival of patients with stage IV G/GEJC. Accumulating evidence suggests that higher PD-L1 expression tends to be associated with better therapeutic efficacy. In a subgroup analysis of the KEYNOTE 585 study[16], which explored combining immunotherapy and chemotherapy as neoadjuvant treatment, anti-PD-1 therapy indicated a more favorable benefit trend in clinical stage III (cIII) than in clinical stage II (cII) G/GEJC.

High PD-L1 expression may be associated with survival outcomes in patients with pIII G/GEJC. However, the role of PD-L1 expression in the survival outcomes of patients with pIII G/GEJC remains unclear. Therefore, we evaluated its impact on the prognosis of patients with locally advanced G/GEJC and explored differences in the tumor microenvironment (TME) between patients with high PD-L1 expression in pathological stage II (pII) and pIII.

MATERIALS AND METHODS

Patients and clinical data collection

From December 2018 to November 2021, 388 consecutive patients with G/GEJC who did not receive neoadjuvant therapy and were histologically confirmed as pII or pIII after radical gastrectomy were identified from the clinical gastrointestinal tumor database of the Department of General Surgery at Nanfang Hospital, Southern Medical University[17]. A flowchart of the study population is shown in Figure 1. A total of 1057 patients with G/GEJC were recorded in our database. Of these, 669 patients were excluded because of the absence of a PD-L1 combined positive score (CPS) evaluation (n = 132), receipt of preoperative neoadjuvant therapy (n = 129), stage I or IV disease (n = 386), multiple primary malignancies (n = 1), or incomplete medical records (n = 21). Ultimately, 388 patients were eligible for inclusion. Clinicopathological and follow-up data were collected.

Figure 1 Study flow diagram. A total of 1057 patients with gastric or gastroesophageal junction cancer (G/GEJC) were recorded in our database. Of these, 669 patients were excluded owing to the absence of a programmed death ligand-1 (PD-L1) combined positive score (CPS) assessment (n = 132), receipt of preoperative neoadjuvant therapy (n = 129), stage I or IV disease (n = 386), multiple primary malignancies (n = 1), or incomplete medical records (n = 21). The remaining 388 consecutive patients did not receive neoadjuvant therapy and were histologically confirmed as pathological stage II (pII) or pathological stage III (pIII) after gastrectomy. Among them, 159 patients were pII and 229 were pIII G/GEJC. In pII G/GEJC, 107 (67.3%) patients were PD-L1-positive (CPS ≥ 1) and 85 (53.5%) had high PD-L1 expression (CPS ≥ 5). In pIII G/GEJC, 155 (67.7%) patients were PD-L1-positive and 116 (50.7%) had high PD-L1 expression. G/GEJC: Gastric or gastroesophageal junction cancer; AJCC: American Joint Committee on Cancer; PD-L1: Programmed death ligand-1; CPS: Combined positive score; pII: Pathological stage II; pIII: Pathological stage III.

Clinical characteristics, including sex, age, tumor location and size, histology, grade, TNM stage (AJCC 8^th edition)[18], Ki-67 index, Epstein-Barr virus (EBV), mismatch repair (MMR), and HER-2 status, were obtained from medical records, pathology reports, and discharge summaries.

The enrolled patients were divided into PD-L1-positive (CPS ≥ 1) and PD-L1-negative (CPS < 1) groups according to their PD-L1 expression. Additionally, they were categorized into high PD-L1 expression (CPS ≥ 5) and low PD-L1 expression (CPS < 5) groups. Clinical characteristics and prognostic differences were compared between the PD-L1-positive and PD-L1-negative groups and between the high and low PD-L1 expression groups.

Patients’ treatment

None of the enrolled patients had received preoperative neoadjuvant therapy. Postoperative adjuvant chemotherapy was recommended for all patients following gastrectomy. Treatment regimens included fluoropyrimidine alone or with oxaliplatin. In this study, a total of 229 patients (59.0%) received postoperative adjuvant chemotherapy, representing 50.3% (80/159) of patients with pII disease and 65.1% (149/229) of those with pIII disease. None of the patients received immunotherapy before recurrence.

Recurrence occurred in 110 patients, accounting for 11.3% (18/159) of patients with pII disease and 40.2% (92/229) of those with pIII disease. Among these patients with recurrence, 33.6% (37/110) received PD-1 inhibitors plus chemotherapy.

Immunohistochemistry and multiplex immunofluorescence

PD-L1 expression has been routinely assessed at Nanfang Hospital since 2018[19], and all PD-L1 testing results in this study were obtained from pathology reports issued by pathologists at our center. Immunohistochemical staining was performed using an automated immunohistochemistry (IHC) stainer (Ventana, Tucson, AZ, United States). Anti-PD-L1 antibodies (Dako, 22C3) were used to evaluate PD-L1 protein levels in gastric adenocarcinoma tissues. PD-L1 expression was quantified using the CPS, defined as the number of PD-L1-positive cells (including tumor cells, lymphocytes, and macrophages) divided by the total number of viable tumor cells, multiplied by 100[20].

Multiplex immunofluorescence (mIF) staining was carried out using an Akoya OPAL Polaris 7-Color Automation IHC Kit (NEL871001KT). The antibody panel included antibodies against CD163 (Abcam, ab182422, 1:500), CD68 (Abcam, ab213363, 1:1000), PD-1 (CST, D4W2J, 86163S, 1:200), PD-L1 (CST, E1 L3N, 13684S, 1:400)[21-23], CD3 (Dako, A0452 IR503, 1:1), CD4 (Abcam, ab133616, 1:100), CD8 (Abcam, ab178089, 1:200), CD56 (Abcam, ab75813, 1:1000), CD20 (Dako, L26 IR604, 1:1), and FOXP3 (Abcam, ab20034, 1:100). Based on these markers, multiple immune cell subsets were quantified, including total T cells (CD3+), cytotoxic T cells (CD8+), helper T cells (CD3+ CD4+), regulatory T cells (CD3+ CD4+ FOXP3+), PD-1-positive effector T cells (PD-1+ CD8+), PD-1-positive cells (PD-1+), PD-L1-positive macrophages (CD68+ PD-L1+), M1 macrophages (CD68+ CD163-), M2 macrophages (CD68+ CD163+), natural killer cells (CD56+), B cells (CD20+) and tertiary lymphoid structures (TLS; CD3+ CD20+). Tumor parenchyma and stroma were distinguished by Pan-CK staining (Abcam, ab7753, 1:100; Akoya Biosciences). Stained slides were scanned using the Vectra Polaris Quantitative Pathology Imaging System (Akoya Biosciences), and all scans for each slide were merged into a single image. Multilayered images were then imported and analyzed using AP-TIME software (3D Medicines). The quantities of various cell populations were expressed as the number of stained cells per square millimeter and as the percentage of positively stained cells among all nucleated cells.

Follow-up

Postoperatively, the patients were routinely followed up every 3 months for the first 2 years, every 6 months for the subsequent 3 years, and annually thereafter. Follow-up was conducted via outpatient visits, telephone calls, or mail. The final follow-up was conducted on June 30, 2023. Disease-free survival (DFS) and overall survival (OS) were recorded. DFS was defined as the time from surgery to tumor recurrence (local recurrence or distant metastasis) or death, while OS was defined as the time from surgery to death.

Statistical analysis

Statistical analyses were performed using IBM SPSS version 27.0 and GraphPad Prism. Continuous variables with a normal distribution were reported as mean ± SD and analyzed using the Student's t-test, whereas non-normally distributed variables were presented as median [interquartile range (IQR)] and analyzed using the Mann-Whitney U test. Categorical variables were described as counts and percentages and analyzed using the χ² or Fisher's exact tests, as appropriate. Survival probability was estimated using the Kaplan-Meier method and compared using the log-rank test. Sensitivity analyses were performed by censoring patients at the time of PD-1 inhibitor initiation to exclude the potential confounding effect of PD-1 inhibitor treatment on OS. Variables with a P value < 0.10 in the univariate analyses, or considered clinically significant, were further evaluated in the multivariate analysis using Cox proportional hazards regression models to identify independent predictors of DFS and OS. Statistical significance was defined as a two-tailed P value < 0.05.

RESULTS

Baseline characteristics

A total of 388 patients with locally advanced G/GEJC who underwent D2 gastrectomy were enrolled, including 159 (41.0%) with pII disease and 229 (59.0%) with pIII disease. Among patients with pIII disease, 99 (43.2%), 84 (36.7%), and 46 (20.1%) had pIIIA, pIIIB, and pIIIC G/GEJC, respectively. A total of 262 (67.5%) patients were PD-L1-positive, and 201 (51.8%) had high PD-L1 expression in the entire cohort. Other clinicopathological features are summarized in Table 1. Additionally, the median follow-up period was 38.00 months (IQR: 28.00-48.00 months), calculated using the reverse Kaplan-Meier method.

Table 1 Baseline patient characteristics.

Variables	Cases	Total
Sex		388
Male	249 (64.2)
Female	139 (35.8)
Age (year)	58.05 ± 12.61
Tumor size (cm)		387
< 5	266 (68.7)
≥ 5	121 (31.3)
Lesion sites		386
Stomach	294 (76.2)
GEJ	92 (23.8)
Histology		385
Sig	141 (36.6)
Others	244 (63.4)
Grade		381
G1-G2	80 (21.0)
G3	301 (79.0)
Pathological stage		388
pII	159 (41.0)
pIIIA	99 (25.5)
pIIIB	84 (21.6)
pIIIC	46 (11.9)
Postoperative adjuvant chemotherapy		388
Yes	229 (59.0)
No	159 (41.0)
HER-2 expression		369
Negative	337 (91.3)
Positive	32 (8.7)
MMR status		368
dMMR	20 (5.4)
pMMR	348 (94.6)
EBV status		376
Negative	360 (95.7)
Positive	16 (4.3)
PD-L1 CPS		388
Negative (< 1)	126 (32.5)
Positive (≥ 1)	262 (67.5)
pII		159
Negative (< 1)	52 (32.7)
Positive (≥ 1)	107 (67.3)
pIII		229
Negative (< 1)	74 (32.3)
Positive (≥ 1)	155 (67.7)
PD-L1 CPS		388
Low expression (< 5)	187 (48.2)
High expression (≥ 5)	201 (51.8)
pII		159
Low expression (< 5)	74 (46.5)
High expression (≥ 5)	85 (53.5)
pIII		229
Low expression (< 5)	113 (49.3)
High expression (≥ 5)	116 (50.7)
Ki-67 index	59.31 ± 23.04

Data were shown as n (%) or mean ± SD. GEJ: Gastroesophageal junction; Sig: Signet ring cell carcinoma; pII: Pathological stage II; pIII: Pathological stage III; MMR: Mismatch repair; dMMR: Deficient mismatch repair; pMMR: Proficient mismatch repair; EBV: Epstein-Barr virus; PD-L1: Programmed death ligand-1; CPS: Combined positive score.

Survival analysis of PD-L1 expression on DFS and OS in locally advanced G/GEJC

In patients with pII-pIII G/GEJC, the 3-year DFS rate was 66.8% in the PD-L1-positive group and 76.4% in the PD-L1-negative group (log-rank P = 0.054; Figure 2A). The 3-year OS rate was 75.6% in the PD-L1-positive group vs 83.5% in the PD-L1-negative group (log-rank P = 0.022; Figure 2B).

Figure 2 Survival outcomes according to programmed death ligand-1 expression status in the overall cohort of patients with locally advanced gastric or gastroesophageal junction cancer. A and B: Kaplan-Meier curves showing disease-free survival (DFS) and overall survival (OS) between programmed death ligand-1 (PD-L1)-positive and PD-L1-negative groups; C and D: Kaplan-Meier curves showing DFS and OS between high and low PD-L1 expression groups. PD-L1: Programmed death ligand-1.

In the high vs low PD-L1 expression groups, the 3-year DFS rates were 67.5% vs 72.7% (log-rank P = 0.242), and the 3-year OS rates were 76.1% vs 80.6% (log-rank P = 0.208; Figure 2C and D).

Survival analysis of PD-L1 expression on DFS and OS in pII G/GEJC: At the last follow-up, 18 (11.3%) patients experienced recurrence, and 15 (9.4%) patients died. No significant differences were observed in 3-year DFS and 3-year OS between the PD-L1-positive and PD-L1-negative groups (3-year DFS: 88.5% vs 87.9%, log-rank P = 0.939; 3-year OS: 92.3% vs 89.9%, log-rank P = 0.621; Figure 3A and B). Similar findings were observed in the sensitivity analysis for OS (3-year OS: 93.7% vs 89.9%, log-rank P = 0.485) (Supplementary Figure 1A). Forest plot analysis confirmed that DFS and OS were similar between the two groups across all subgroups (DFS: Supplementary Figure 2A; OS: Supplementary Figure 2B).

Figure 3 Survival outcomes according to programmed death ligand-1 expression in patients with pathological stage II gastric or gastroesophageal junction cancer. A and B: Kaplan-Meier curves showing disease-free survival (DFS) and overall survival (OS) between programmed death ligand-1 (PD-L1)-positive and PD-L1-negative groups; C and D: Kaplan-Meier curves showing DFS and OS between high and low PD-L1 expression groups. No significant differences in DFS or OS were observed between the groups. PD-L1: Programmed death ligand-1.

Likewise, no substantial differences were detected between the high and low PD-L1 expression groups (3-year DFS: 91.8% vs 84.2%, log-rank P = 0.138; 3-year OS: 94.2% vs 88.4%, log-rank P = 0.233; Figure 3C and D). Sensitivity analysis also showed no significant differences in OS between the two groups (3-year OS: 96.0% vs 88.2%, log-rank P = 0.131; Supplementary Figure 1B). Forest plot analysis of DFS between the high and low PD-L1 expression groups is shown in Supplementary Figure 2C. Subgroup analyses for OS yielded similar results for both groups (Supplementary Figure 2D).

Survival analysis of PD-L1 expression on DFS and OS in pIII G/GEJC: During follow-up, recurrence occurred in 92 (40.2%) patients, and 73 (31.9%) patients died. The 3-year DFS rates were 52.2% in the PD-L1-positive group and 67.8% in the PD-L1-negative group (HR = 1.653, 95%CI: 1.083-2.526, log-rank P = 0.030; Figure 4A), and the 3-year OS rates were 65.1% vs 78.2% (HR = 2.048, 95%CI: 1.278-3.282, log-rank P = 0.007; Figure 4B) in the PD-L1-positive and -negative groups, respectively. In sensitivity analysis, the PD-L1-positive group was still associated with worse OS than the PD-L1-negative group (3-year OS: 67.6% vs 80.7%, log-rank P = 0.036; Supplementary Figure 3A). DFS and OS were significantly shorter in the PD-L1-positive group among men and among those with tumor size < 5 cm, poor differentiation, and proficient MMR (pMMR). Additional subgroup analyses are presented in Supplementary Figure 4A and B.

Figure 4 Survival outcomes according to programmed death ligand-1 expression in patients with pathological stage III gastric or gastroesophageal junction cancer. A and B: Kaplan-Meier curves showing disease-free survival (DFS) and overall survival (OS) between programmed death ligand-1 (PD-L1)-positive and PD-L1-negative groups; C and D: Kaplan-Meier curves showing DFS and OS between high and low PD-L1 expression groups. Patients with PD-L1-positive status or high expression had significantly shorter DFS and OS than those with PD-L1-negative or low expression. PD-L1: Programmed death ligand-1.

Correspondingly, the 3-year DFS rates were 50.2% in the high PD-L1 expression group vs 64.8% in the low PD-L1 expression group (HR = 1.602, 95%CI: 1.064-2.413, log-rank P = 0.023; Figure 4C) and the 3-year OS rates were 64.0% vs 75.1% (HR = 1.628, 95%CI: 1.028-2.578, log-rank P = 0.036; Figure 4D) in the high and low PD-L1 expression groups, respectively. In the sensitivity analysis, the associations between high PD-L1 expression and poor OS remained statistically significant (3-year OS: 65.6% vs 79.0%, log-rank P = 0.033; Supplementary Figure 3B). Forest plot analysis revealed that in patients with pIII G/GEJC, poor DFS and OS were associated with high PD-L1 expression in men and in those with tumor size < 5 cm, pMMR or EBV-negative tumors. Detailed subgroup results are provided in Supplementary Figure 4C and D.

TME analysis of high PD-L1 expression in locally advanced G/GEJC: Based on the aforementioned survival outcomes, eight patients with high PD-L1 expression—four each with pII and pIII G/GEJC—were randomly selected for preliminary exploratory mIF analysis, as detailed in Supplementary Table 1. In this small subset, pIII tumors showed a trend toward a lower abundance of PD-1+ cells than pII tumors. Among the six patients with evaluable TLS, TLS was detected in two of two pIII cases and one of four pII cases. The abundance and proportion of TLS appeared to be higher in pIII than in pII G/GEJC. CD8+ T cells in patients with pIII tumors tended to be more enriched within the tumor parenchyma than within the stroma (three of four patients), whereas this pattern was not observed in patients with pII tumors. Among the macrophage subsets, M1 macrophages tended to localize within the stroma in pIII tumors (three of four patients), whereas no comparable trend was observed in pII tumors. No clear stage-related trends were identified in the remaining immune cell populations in this limited exploratory cohort.

Relation between PD-L1 expression and clinicopathological characteristics in pIII G/GEJC

As summarized in Supplementary Table 2, compared with the PD-L1-negative group, the PD-L1-positive group had a significantly higher Ki-67 index (62.85 ± 21.90 vs 52.85 ± 23.97, P = 0.002). By contrast, there were no significant differences in sex (P = 0.106), age (P = 0.064), tumor size (P = 0.767), lesion sites (P = 0.205), histologic subtype (P = 0.133), differentiation grade (P = 0.268), pathological TNM stage (P = 0.434), HER-2 expression (P = 0.203), MMR status (P = 0.667), EBV status (P = 0.761), or postoperative adjuvant chemotherapy (P = 0.801).

Similarly, the high expression group also had a significantly higher Ki-67 index than that of the low expression group (63.78 ± 22.09 vs 55.36 ± 23.28, P = 0.006). Additionally, patients in the high expression group were older (59.51 ± 12.05 vs 55.41 ± 13.31, P = 0.015) and had a lower incidence of signet ring cell carcinoma (35.1% vs 48.2%, P = 0.045) than those in the low expression group. No significant differences were observed in the remaining clinicopathological features (all P > 0.05; Table 2).

Table 2 Clinicopathological features associated with high programmed death ligand-1 expression group in pathological stage III gastric or gastroesophageal junction cancer.

Variables	High PD-L1 expression (n = 116)	Low PD-L1 expression (n = 113)	Statistic	P value
Sex			1.914	0.167
Male	78 (67.2)	66 (58.4)
Female	38 (32.8)	47 (41.6)
Age (year)	59.51 ± 12.05	55.41 ± 13.31	-2.445	0.015
Tumor size (cm)			0.002	0.964
< 5	76 (66.1)	75 (66.4)
≥ 5	39 (33.9)	38 (33.6)
Lesion sites			3.194	0.074
Stomach	79 (69.3)	90 (79.6)
GEJ	35 (30.7)	23 (20.4)
Histology			4.007	0.045
Sig	40 (35.1)	54 (48.2)
Others	74 (64.9)	58 (51.8)
Grade			0.888	0.346
G1-G2	19 (17.0)	14 (12.5)
G3	93 (83.0)	98 (87.5)
Pathological stage			1.263	0.532
pIIIA	51 (44.0)	48 (42.5)
pIIIB	39 (33.6)	45 (39.8)
pIIIC	26 (22.4)	20 (17.7)
HER-2 expression			0.728	0.394
Negative	100 (89.3)	100 (92.6)
Positive	12 (10.7)	8 (7.4)
MMR status			0.000	1.000
dMMR	5 (4.3)	4 (3.5)
pMMR	111 (95.7)	109 (96.5)
EBV status			1.779	0.182
Negative	105 (93.8)	108 (98.2)
Positive	7 (6.3)	2 (1.8)
Ki-67 index	63.78 ± 22.09	55.36 ± 23.28	-2.795	0.006
Postoperative adjuvant chemotherapy			0.929	0.335
Yes	72 (62.1)	77 (68.1)
No	44 (37.9)	36 (31.9)

Data were shown as n (%) or mean ± SD. GEJ: Gastroesophageal junction; Sig: Signet ring cell carcinoma; pIII: Pathological stage III; MMR: Mismatch repair; dMMR: Deficient mismatch repair; pMMR: Proficient mismatch repair; EBV: Epstein-Barr virus; PD-L1: Programmed death ligand-1.

Prognostic factors in pIII G/GEJC

The univariate and multivariate Cox proportional hazards models for patients with pIII G/GEJC are presented in Table 3. In the univariate analysis, a more advanced TNM stage and high PD-L1 expression were significantly associated with shorter DFS and OS. Further multivariate analysis identified high PD-L1 expression as an independent adverse prognostic factor for DFS (HR = 1.624, 95%CI: 1.056-2.500, P = 0.027) and OS (HR = 1.653, 95%CI: 1.016-2.688, P = 0.043). TNM stage was also independently associated with both DFS and OS (P < 0.001).

Table 3 Univariate and multivariate analyses of clinicopathological features associated with disease-free survival and overall survival for pathological stage III gastric or gastroesophageal junction cancer.

Variables	DFS-univariate		DFS-multivariate		OS-univariate		OS-multivariate
	HR (95%CI)	P value	HR (95%CI)	P value	HR (95%CI)	P value	HR (95%CI)	P value
Sex
Male	1.000				1.000
Female	1.101 (0.725, 1.670)	0.652			1.067 (0.670, 1.702)	0.784
Age (year)
< 65	1.000				1.000
≥ 65	1.249 (0.810, 1.926)	0.314			1.782 (0.772, 4.112)	0.176
Tumor size (cm)
< 5	1.000				1.000
≥ 5	1.143 (0.743, 1.759)	0.542			1.394 (0.869, 2.238)	0.169
Lesion sites
Stomach	1.000				1.000
GEJ	1.228 (0.778, 1.938)	0.377			1.058 (0.631, 1.772)	0.831
Histology
Other	1.000				1.000
Sig	0.900 (0.590, 1.373)	0.626			0.962 (0.602, 1.537)	0.870
Grade
G1-G2	1.000				1.000
G3	0.873 (0.501, 1.520)	0.631			0.755 (0.414, 1.380)	0.362
Pathological stage
pIIIA	1.000	< 0.001	1.000	< 0.001	1.000	< 0.001	1.000	< 0.001
pIIIB	1.716 (1.030, 2.861)	0.038	1.719 (1.018, 2.903)	0.043	1.953 (1.080, 3.533)	0.027	1.897 (1.026, 3.509)	0.041
pIIIC	3.584 (2.130, 6.029)	< 0.001	3.423 (2.012, 5.822)	< 0.001	4.269 (2.348, 7.761)	< 0.001	4.237 (2.291, 7.837)	< 0.001
HER-2 expression
Negative	1.000		1.000		1.000		1.000
Positive	1.798 (0.977, 3.308)	0.059	1.680 (0.910, 3.100)	0.097	1.332 (0.660, 2.689)	0.423	1.148 (0.566, 2.328)	0.702
MMR status
dMMR	1.000				1.000
pMMR	0.505 (0.221, 1.156)	0.106			0.475 (0.191, 1.178)	0.108
EBV status
Negative	1.000				1.000
Positive	0.671 (0.212, 2.123)	0.497			0.555 (0.136, 2.266)	0.412
PD-L1 expression
Low	1.000		1.000		1.000		1.000
High	1.608 (1.060, 2.438)	0.026	1.624 (1.056, 2.500)	0.027	1.639 (1.026, 2.619)	0.039	1.653 (1.016, 2.688)	0.043

DFS: Disease-free survival; OS: Overall survival; GEJ: Gastroesophageal junction; Sig: Signet ring cell carcinoma; pIII: Pathological stage III; MMR: Mismatch repair; dMMR: Deficient mismatch repair; pMMR: Proficient mismatch repair; EBV: Epstein-Barr virus; PD-L1: Programmed death ligand-1.

DISCUSSION

The precise population that would benefit from perioperative immunotherapy has not yet been clearly identified. Although PD-L1 CPS has been correlated with the efficacy of immunotherapy in stage IV G/GEJC[13-15], no clear relation has been observed between PD-L1 expression and the efficacy of immunotherapy for locally advanced G/GEJC. Subgroup analysis of the KEYNOTE-585 study indicated that patients with cIII G/GEJC were more likely to benefit from PD-1 antibodies than those with cII G/GEJC[16]. Previous studies exploring the prognostic relevance of PD-L1 in G/GEJC have yielded inconsistent results[24-28]. The role of baseline PD-L1 expression in patients with locally advanced G/GEJC at different stages remains unclear. In this study, we evaluated the prognostic significance of PD-L1 expression in 388 patients with pII-pIII G/GEJC who underwent curative gastrectomy without neoadjuvant therapy. The results showed that in pIII G/GEJC, high PD-L1 expression was associated with worse DFS and OS, whereas in pII G/GEJC, it exhibited a trend toward better survival. Sensitivity analyses confirmed the robustness of the OS findings. Further multivariate analysis revealed that high PD-L1 expression was an independent risk factor for postoperative recurrence and mortality in patients with pIII G/GEJC.

Combining the results of various studies on the prognostic correlation between PD-L1 expression and tumors, no definitive conclusion has been drawn[29-32]. Studies examining the association between PD-L1 expression and G/GEJC have also reported contradictory findings. Sughayer et al[24] and Sotnikova et al[25] found that CPS positivity in patients with G/GEJC correlated with improved OS. By contrast, Lian et al[26] and Zhou et al[27] concluded that CPS positivity was not significantly associated with survival in patients with G/GEJC. Additionally, other studies considered CPS positivity a poor prognostic factor for G/GEJC[28]. These discrepancies may be attributed to variations in patient staging, criteria used to define PD-L1 positivity, or the PD-L1 detection kits employed.

A meta-analysis examining the prognostic significance of PD-L1 expression in lung cancer found that high PD-L1 expression was associated with a favorable outcome in stage I lung adenocarcinoma [progression-free survival (PFS), P = 0.0154; OS, P = 0.0144] but with an adverse outcome in stage II/III/IV squamous cell lung cancer (PFS, P = 0.0016; OS, P = 0.0014)[33]. A study conducted by Pereira et al[34], which included 284 patients with G/GEJC (pathological stage I-IV), found that patients with CPS negativity (< 1, clone SP142) had worse DFS and OS than those with CPS positivity in resectable G/GEJC, although the differences did not reach predefined statistical significance (DFS: P = 0.052; OS: 52.8% vs 68.2%, P = 0.081). However, Zhang et al[35] reported that among 132 patients with pII-pIII G/GEJC, those with PD-L1 positivity, defined as moderate or intense staining using the antibody ab58810 (rabbit polyclonal, Abcam), had significantly poorer survival than those with PD-L1-negativity (5-year OS: 50.7% vs 83.1%, P < 0.001). Multivariate analysis confirmed that PD-L1 positivity (P = 0.001) was an independent prognostic factor for locally advanced G/GEJC. The aforementioned studies did not differentiate prognostic value by stage. In the study by Pereira et al[34], patients with pIII-pIV G/GEJC accounted for a slightly smaller proportion (45.8%, 130/284). In Zhang et al's study[35], patients with pIII G/GEJC accounted for a larger proportion (84.1%, 111/132). To further assess the role of PD-L1 in different stages of G/GEJC, we utilized CPS (Dako, 22C3), which is recommended by current guidelines for PD-L1 evaluation, and stratified locally advanced G/GEJC into pII and pIII. Both survival and multivariate analyses revealed that high PD-L1 expression was a negative prognostic factor for pIII G/GEJC. Conversely, although no statistical significance was reached, our study observed a trend suggesting a better prognosis in patients with high PD-L1 expression in pII G/GEJC.

The KEYNOTE-585 trial was the first phase III study to report long-term survival outcomes of perioperative immunotherapy plus chemotherapy in patients with locally advanced G/GEJC[16]. Among the CPS subgroups, no significant survival benefit was observed with the combination of pembrolizumab and chemotherapy in patients with high CPS [event-free survival (EFS): CPS ≥ 10, HR = 0.70, 95%CI: 0.46-1.04; CPS < 10, HR = 0.85, 95%CI: 0.67-1.09; OS: CPS ≥ 10, HR = 0.82, 95%CI: 0.53-1.28; CPS < 10, HR = 0.91, 95%CI: 0.70-1.17]. However, tumor-stage subgroup analyses revealed that in patients with cIII G/GEJC, pembrolizumab plus chemotherapy significantly improved both median EFS (HR = 0.73, 95%CI: 0.58-0.91) and median OS (HR = 0.75, 95%CI: 0.59-0.96). By contrast, in patients with cII G/GEJC, the addition of a PD-1 inhibitor to chemotherapy did not significantly improve median EFS (HR = 1.02, 95%CI: 0.64-1.61) or median OS (HR = 1.35, 95%CI: 0.84-2.19). Unfortunately, the efficacy of pembrolizumab plus chemotherapy in patients with high CPS remains unclear across different stages (cII and cIII). For patients with pII G/GEJC, high PD-L1 expression (CPS ≥ 5) was not identified as a poor prognostic factor in our study and even showed a trend toward improved absolute survival. Meanwhile, subgroup analysis of the cII cohort in KEYNOTE-585 suggested that combination immunotherapy did not significantly improve survival. Therefore, it may be more precise for future investigations to focus on patients with stage III G/GEJC to evaluate the efficacy of neoadjuvant immunotherapy plus chemotherapy. Additionally, more attention should be paid to the efficacy of PD-1 inhibitors in combination with other therapies in patients with high CPS in stage III G/GEJC.

PD-1 is primarily expressed on the surface of T, B, and myeloid cells. PD-1 antibodies restore the antitumor function of immune cells and prevent immune evasion by tumor cells by binding to the PD-1 receptor on immune cells and blocking its interactions with ligands[36]. PD-L1 plays a crucial role in tumor immunosuppression on both tumor and immune cells. The effect of PD-L1 on immune cells is primarily manifested by negatively regulating T-cell activation and restricting T-cell migration[37,38]. In our preliminary exploratory TME analysis, we observed a trend toward greater intratumoral enrichment of CD8+ T cells in pIII tumors. Although this observation is hypothesis-generating, it is biologically plausible that interactions between PD-L1 on tumor cells and immune cells might cooperatively dampen antitumor immune responses and facilitate tumor immune evasion[39,40]. This conceptual framework may partly help contextualize why patients with high PD-L1 expression experienced poor survival in pIII G/GEJC in our cohort. Tang et al[41] reported that PD-L1 in host cells is essential for PD-L1 blockade-mediated tumor regression. They found that blocking PD-L1 within tumors alone was not sufficient for tumor regression, as PD-L1 on host immune cells restricts T-cell trafficking, thereby reducing the efficacy of PD-1/PD-L1 blockade therapy. Based on these insights, we cautiously speculate that the distribution of PD-L1 on tumor cells vs immune cells may differ between pII and pIII G/GEJC. Such differences may potentially be related to the stage-dependent prognostic impact of high PD-L1 expression observed in our study. And it might partly explain why patients with stage III G/GEJC benefited from PD-1 antibodies in the KEYNOTE-585 trial, while patients with stage II G/GEJC did not. However, these interpretations remain exploratory and require validation in larger datasets.

We also noted that patients with high PD-L1 expression showed a trend toward a higher abundance of PD-1+ cells in pII tumors compared with that in pIII tumors in our preliminary TME subset. A previous study demonstrated that patients who co-expressed PD-1 and PD-L1 had a better prognosis (log-rank P = 0.019)[42]. Further multivariate analysis confirmed that co-expression was independently associated with improved outcomes (HR = 0.48, 95%CI: 0.27-0.84, P = 0.010). Conversely, patients with pIII tumors tended to have fewer PD-1+ cells in this exploratory subset, and patients with pIII G/GEJC experienced poorer survival outcomes in our cohort. A mechanistic study demonstrated that PD-1 deficiency inhibited the adaptive immune response and exacerbated the tumor immunosuppressive microenvironment by enhancing CD30+ Treg cell function[43].

Collectively, these preliminary observations raise the possibility that high PD-L1 expression may reflect distinct immunobiological states at different disease stages. In pIII G/GEJC, the combination of great tumor burden, deep invasion, and a tendency toward the presence of TLS and intratumoral CD8+ T cells may reflect a chronically activated yet adaptively immunosuppressed microenvironment[44-47]. By contrast, in pII G/GEJC, high PD-L1 expression tended to have a relatively greater abundance of PD-1+ cells and more favorable survival outcomes. This may suggest that tumors may be in the early phase of immune recognition[48]. The tumor burden was relatively lower, and the local immune microenvironment might be characterized as a predominantly immune-activated state with concurrent PD-1/PD-L1 axis engagement[49]. Within this context, high PD-L1 expression may reflect active immune surveillance rather than a fully established immunosuppressive state[46,47]. This hypothesis might provide context for why high PD-L1 expression was associated with worse prognosis in patients with pIII G/GEJC, but tended to be associated with better survival in patients with pII G/GEJC. It may also potentially offer a biological rationale for the limited efficacy of PD-1 inhibitors in stage II disease reported in the KEYNOTE-585 trial. These interpretations remain exploratory because of the limited TME sample size, and further studies are required to confirm these potential biological differences. We have initiated additional TME investigations to elucidate these mechanisms better and support the development of optimized treatment strategies.

This study had some limitations. First, 37 (33.6%) patients with recurrence received PD-1 inhibitors, which may have partially altered the original difference in OS between the PD-L1-positive/high-expression group and the PD-L1-negative/low-expression group. However, our sensitivity analyses reaffirmed the validity of the OS results and suggested that PD-1 inhibitor treatment did not confound our study findings. Second, the median follow-up duration was 38.00 months, which is relatively short for patients with pII G/GEJC. Finally, the study was retrospective in design, and the TME analysis was based on a small exploratory subset.

CONCLUSION

High PD-L1 expression was associated with worse DFS and OS in patients with stage III G/GEJC and served as an independent prognostic factor for poor survival. These findings indicate the stage-dependent prognostic value of PD-L1 expression and could inform future investigations on perioperative immunotherapy strategies for G/GEJC.

ACKNOWLEDGEMENTS

Abstract was selected for online publication at the 2025 ASCO Annual Meeting (May 30, 2025 - June 3, 2025).