Concordance of PD-L1 status in primary gastroesophageal adenocarcinoma and matched peritoneal metastases: a single institution study

Abstract

Background

Programmed death-ligand 1 (PD-L1) expression serves as a predictive biomarker for response to immune checkpoint inhibitors (ICIs) in metastatic gastroesophageal adenocarcinoma (mGEA). The peritoneum is one of the most common metastatic sites and is associated with a poor prognosis and apparently lower clinical efficacy of ICIs.

Patients and methods

We investigated the heterogeneity of PD-L1 expression in mGEA by examining its concordance between primary tumors and matched peritoneal metastases (PMs), and before and after systemic treatment. PD-L1 expression was assessed using combined positive score (CPS) by immunohistochemistry with VENTANA PD-L1 (SP263) assays on formalin-fixed paraffin-embedded tumor tissues. Results were reported using CPS cut-offs of 1 and 5.

Results

Twenty-two primary tumor and matched PM specimens from patients with mGEA were analyzed. The concordance of PD-L1 CPS was 54.5% with a CPS cut-off of 1 and 72.7% with a CPS cut-off of 5, highlighting spatial heterogeneity. Notably, none of the PD-L1-positive primary tumor samples tested positive in the matched PM specimens using the CPS ≥5 cut-off. Potential temporal heterogeneity of PD-L1 expression related to chemo(immuno)therapy administration was also observed, with a 55.6% concordance rate using the CPS ≥5 cut-off.

Conclusions

PD-L1 expression in PMs from mGEA is characterized by both spatial and potentially temporal heterogeneity, with the variability being more pronounced at lower CPS cut-off values. This variability complicates its role as a predictive biomarker for ICI outcomes. The high intrapatient discordance rate in PD-L1 CPS expression between positive primary tumor samples and matched PM specimens suggests that peritoneal specimens should not be used as the only source for PD-L1 assessment if representative tissue from other disease sites is available.

Highlights

• •PMs seem to derive less benefit from immunotherapy (ICI).
• •The predictive ability of PD-L1 for chemo-ICI suffers from spatial and temporal heterogeneity.
• •We described spatial–temporal heterogeneity of PD-L1 expression between primary tumors and PMs.
• •Further analyses will clarify the role of PD-L1 heterogeneity in determining outcomes for chemo-ICI.

Introduction

Gastric cancer is the fifth most diagnosed and the fourth most lethal cancer worldwide, with a global annual burden predicted to increase by 2040, mainly in younger adults.¹ In Europe, the 5-year survival rate remains poor (∼25%),² despite the recent positive results of immune checkpoint inhibitors (ICIs) in the programmed death-ligand 1 (PD-L1)-selected population of patients with metastatic gastroesophageal adenocarcinoma (mGEA).3, 4, 5

The peritoneum is one of the most common metastatic sites (present in 15%-30% of the cases at diagnosis) and the only site of recurrence in up to 60% of patients with resected gastric cancer.⁶ Peritoneal metastases (PMs) represent also one of the strongest poor prognostic determinants in this disease.^6,7 Pharmacokinetic studies showed that resistance of PMs to systemic chemotherapy has to be foreseen in the so-called peritoneal–plasma barrier, which reduces the uptake of the cytotoxic agents from plasma into peritoneal tumor nodules and ascites.⁸ This paradigm seems to apply also to pivotal ICIs plus chemotherapy combinations: recent biological data suggest that the immunity cell phenotype is extensively altered in patients with PMs with a shift toward B cells⁹ and immune-suppressive macrophage¹⁰ polarization which seems to contribute to local progression and aggressiveness. Indeed, data from CheckMate 649 indicate that adding nivolumab to chemotherapy significantly improves overall survival (OS) in the primary endpoint population with a PD-L1 combined positive score (CPS) ≥5, but the subset of patients with PMs appears to benefit less from this strategy.³

However, the mechanistic cause of the possibly reduced sensitivity of PMs to ICIs has not yet fully been clarified. Recent evidence from literature demonstrated that PD-L1 expression in tissue sections from primary tumor samples of patients with mGEA and PMs was widely present (e.g. 45.6% positive primary tumor samples with PD-L1 CPS ≥1 and 14.3% with PD-L1 CPS ≥10).¹¹ Nonetheless, significant spatial (between primary tumors and distant metastases) and temporal (before and after treatment) discordance of PD-L1 expression in patients with gastric cancer has also been described.^12,13 However, available studies were not focused on PMs, which were poorly represented in the reported series describing PD-L1 heterogeneity.

Although specific data regarding PD-L1 expression in peritoneal samples from mGEA are currently lacking, it is rational to investigate the putative mechanism of resistance in the biomarker expression intrinsic variability. Hence the aim of our work is to describe the spatial and temporal heterogeneity of PD-L1 expression in PMs from patients with mGEA, compared with its expression in matched primary tumors.

Materials and methods

Study population

We retrospectively selected formalin-fixed paraffin-embedded tumor samples from consecutive patients treated at our institution for mGEA from September 2015 to August 2023. Patients whose histological samples were available and adequate to assess PD-L1 expression on both primary tumors and PMs were enrolled. For each patient, pathological data [histology, grading, mismatch repair/microsatellite instability (MMR/MSI), and human epidermal growth factor receptor 2 (HER2) status], clinical characteristics (age, sex, performance status, primary tumor site, timing and sites of metastases, presence of ascites at diagnosis of advanced disease, peritoneal surgery, and peritoneal cancer index), and laboratory features (carcinoembryonic antigen and carbohydrate antigen 19.9 at diagnosis of advanced disease) were collected. In addition, data relative to administered systemic treatments (regimen types and activity and efficacy parameters) were recorded.

Tumor cells that exhibit intense, moderate, and faint complete or incomplete linear membrane staining were included in the PD-L1 CPS assessment. Tumor-associated lymphocytes, as well as macrophages with membrane and cytoplasmatic staining, were also included. Although neutrophils, eosinophils, plasma cells, stromal cells, necrotic cells, cellular debris, and platelets may show significant positivity, they were excluded when calculating the CPS.¹⁴

Both primary tumor and PMs were tested for PD-L1 CPS with immunohistochemistry (IHC) (clone SP263; Ventana Medical System) using the fully automated BenchMark ULTRA system (Roche-Ventana Medical Systems, Tucson, AZ) platform, according to the manufacturer’s instructions, and reported as exact score. All specimens were qualified by a histopathologist to confirm the requirements fulfillment and each tissue core was evaluated by two independent pathologists. For outcome correlations, both the PD-L1 CPS cut-offs of 5 and 1 were chosen, consistent with regulatory agency approval for nivolumab and pembrolizumab.^15,16

Details about techniques used for HER2 and MMR protein status assessment have been previously reported.¹⁷

The current analysis was conducted as part of protocol number 2020/140 (dated 16 November 2020), approved by the Ethics Committee of Veneto Institute of Oncology – IRCCS,¹⁷ and was conducted in accordance with the Helsinki Declaration of 1964 and later versions. Informed consent to be included in the study was obtained from all patients.

Objectives and statistics

The primary endpoint of the study was to report the concordance of PD-L1 CPS status between primary tumor and related PMs, according to the two chosen CPS cut-offs. The efficacy of first-line systemic therapies according to progression-free survival (PFS; defined as the time from the start of the first-line treatment to evidence of disease progression or death for any cause) and OS (defined as the time from mGEA diagnosis to death for any reason or last follow-up) was also described.

Descriptive statistics were provided as frequency (rates) for categorical data and as median values with ranges for continuous variables. The chi-square test (or Fisher’s exact test or McNemar’s test, when appropriate) was used to compare categorical data.

Survival outcomes were calculated using the Kaplan–Meier method and compared through log-rank tests. Statistical analyses were carried out using MedCalc Statistical Software version 19.5.3 (MedCalc Software Ltd, Ostend, Belgium; https://www.medcalc.org; 2020).

Results

Patient and tissue sample characteristics

Twenty-two patients with primary tumors and related PMs available for analyses were included in the study. The selection process is illustrated in Figure 1, and the baseline characteristics of the population are reported in Table 1. The most common histological type was poorly differentiated adenocarcinoma with signet ring cell features (54.5%), while adenocarcinoma was present in 45.5% of patients. All patients had HER2-negative and proficient MMR disease. Metastatic disease was diagnosed synchronously in 17 (77%) patients. Matched primary tumor and PM samples were collected at baseline (before treatment) for 13 (59.1%) out of 22 patients. In addition, 20 (91%) patients had a defined peritoneal cancer index score for PMs at diagnostic staging laparoscopy, with a median score of 9.5 (range 0-21). Seven (31.9%) patients underwent peritoneal cytoreductive surgery, with six of these surgeries carried out after completing first-line chemotherapy that included oxaliplatin and a fluoropyrimidine [5-fluorouracil, FOLFOX, or capecitabine (XELOX)]. Of the entire population, only 4 (18.2%) patients received nivolumab plus chemotherapy as first-line treatment, whereas 16 (81.8%) received chemotherapy alone.

Figure 1.

Selection process of the study population. PD-L1, programmed death-ligand 1.

Table 1.

Summary of baseline demographic, clinical, and disease characteristics

Characteristics	Values
Age (years), median (range)	65 (37-81)
Gender, n (%)
Male	12 (54.5)
Female	10 (45.5)
ECOG performance status, n (%)
0	12 (54.5)
1	10 (45.5)
Primary tumor site, n (%)
Cardias	1 (4.5)
Body/antrum	21 (95.5)
Histology, n (%)
Adenocarcinoma	10 (45.5)
Signet-ring cell	12 (54.5)
Lauren type, n (%)
Diffuse	7 (31.9)
Intestinal	3 (13.6)
Mixed	2 (9.0)
Not defined	10 (45.5)
Grading, n (%)
2	1 (4.5)
3	15 (68.2)
Not defined	6 (27.3)
HER2 status, n (%)
Positive	0 (0)
Negative	22 (100)
MMR status, n (%)
MMR deficient	0 (0)
MMR proficient	22 (100)
Synchronous metastases, n (%)
Yes	17 (77.3)
No	5 (22.7)
Number of metastatic sites, n (%)
1	19 (86.4)
>1	3 (13.6)
Site of metastases (beyond peritoneum), n (%)
Distant nodes	2 (9.0)
Bone	1 (4.5)
Ascites, n (%)
Yes	6 (27.3)
No	16 (72.7)
CEA, n (%)
≤ULN	9 (40.8)
>ULN	4 (18.4)
Not available	9 (40.8)
CA19.9, n (%)
≤ULN	10 (45.6)
>ULN	4 (13.6)
Not available	9 (40.8)
Perioperative chemotherapy, n (%)
Yes	5 (22.7)
No	17 (77.3)
First-line chemotherapy + nivolumab, n (%)
Yes	4 (18.4)
No	18 (81.6)
Peritoneal surgery, n (%)
Yes	7 (31.9)
No	15 (68.1)
Peritoneal cancer index, median (range)	9.5 (1-21)

CA19.9, carbohydrate antigen 19.9; CEA, carcinoembryonic antigen; ECOG, Eastern Cooperative Oncology Group; HER2, human epidermal growth factor receptor 2; MMR, mismatch repair; ULN, upper limit of normal.

All PM samples were obtained from surgically resected peritoneal nodules. Using a PD-L1 CPS cut-off of 5, 1 patient (4.5%) out of 22 was classified as positive, while using a cut-off of 1, 12 patients (54.5%) were classified as positive. Primary tumor samples were collected as endoscopic biopsies for 9 patients (40.9%) and as surgical specimens for 13 patients (59.1%). The type of sample analyzed did not influence the results of PD-L1 expression assessment in primary tumors. With a PD-L1 CPS cut-off of 5, 2 out of 9 (22.2%) biopsy samples and 3 out of 13 (23.1%) surgical specimens were classified as positive (P = 0.963). Using a PD-L1 CPS cut-off of 1, 3 out of 9 (33.3%) biopsy samples and 3 out of 13 (30.8%) surgical specimens were classified as positive (P = 0.899).

At the time of the analysis, the median follow-up was 34.6 months. The overall median OS from the initial diagnosis of GEA was 21.5 months (95% CI 12.6-46.3 months). For those with metastatic disease, the median OS was 13.2 months (95% CI 11.5-24.9 months). No significant difference in OS was observed according to PD-L1 CPS status determined from the primary tumor, using either of the two cut-offs (PD-L1 CPS ≥5 versus <5: hazard ratio 1.02, 95% CI 0.22-4.65, P = 0.98; PD-L1 CPS ≥1 versus <1: hazard ratio 0.75, 95% CI 0.19-2.97, P = 0.70). Of the four patients treated with nivolumab plus chemotherapy, all had synchronous peritoneal-only metastases. In this subgroup, the median PFS and OS were 5.6 and 9.7 months, respectively. Notably, all these four patients had discordant results: primary tumors showed a CPS of ≥5 in all cases, while PMs had a CPS of <5 in all cases. This subgroup exhibited a heterogeneous pattern of objective response to treatment, both overall and when considering each disease site separately (Supplementary Table S1, available at https://doi.org/10.1016/j.esmogo.2024.100089).

PD-L1 spatial heterogeneity

For primary tumor specimens, the PD-L1 CPS ranged from 0 to 18, with a median CPS of 1.25. By contrast, for PM specimens, the PD-L1 CPS ranged from 0 to 6.5, with a median CPS of 1. Five (22.7%) primary tumors had a PD-L1 CPS ≥5, compared with only one (4.5%) PM (P = 0.079). Twelve (54.5%) patients exhibited PD-L1 CPS positivity in both the primary tumor and PM samples.

A matched comparison of PD-L1 expression between primary gastric tumors and PMs was carried out in all 22 patients. The results of PD-L1 spatial heterogeneity are summarized in Tables 2 and 3. Representative matched samples demonstrating IHC PD-L1 spatial heterogeneity are shown in Figure 2. Using a CPS cut-off of 5, the PD-L1 status of paired primary tumors and PMs was concordant in 72.7% of cases (16/22). Of the 17 PD-L1-negative primary tumors, 16 (94%) also had a PD-L1 CPS of <5 in the PMs. By contrast, none of the five PD-L1-positive primary tumors showed a PD-L1 CPS of ≥5 in the PMs. With a threshold CPS of 1, the concordance rate was 54.5% (12/22). Of the 10 PD-L1-negative primary tumors, 5 (50%) were also PD-L1 negative in the PMs. Of the 12 PD-L1-positive primary tumors, 7 (58%) were PD-L1 positive in the PMs. An analysis of spatial heterogeneity based on the type of primary tumor sample (biopsy or surgical specimens) yielded similar results (Supplementary Table S2, available at https://doi.org/10.1016/j.esmogo.2024.100089).

Table 2.

Comparison of PD-L1 expression in primary tumors versus peritoneal metastases according to a PD-L1 CPS cut-off of 5

		Peritoneal metastases PD-L1
		CPS ≥5	CPS <5	Total	Concordance rate, %	P valuea
Primary tumor PD-L1	CPS ≥5	0	5	5	0
	CPS <5	1	16	17	94.1
	Total	1	21	22	72.7	0.22

CPS, combined positive score; PD-L1, programmed death-ligand 1.

^aMcNemar’s test.

Table 3.

Comparison of PD-L1 expression in primary tumors versus peritoneal metastases according to a PD-L1 CPS cut-off of 1

		Peritoneal metastases PD-L1
		CPS ≥1	CPS <1	Total	Concordance rate, %	P valuea
Primary tumor PD-L1	CPS ≥1	7	5	12	58.3
	CPS <1	5	5	10	50
	Total	12	10	22	54.5	0.99

CPS, combined positive score; PD-L1, programmed death-ligand 1.

^aMcNemar’s test.

Figure 2.

Formalin-fixed paraffin-embedded tumor tissue representative of immunohistochemistry programmed death-ligand 1 (PD-L1) spatial heterogeneity. (A) Primary tumor sample, PD-L1 expression staining by CPS (PD-L1 CPS = 15-20) (×10); (B) peritoneal metastasis sample, PD-L1 expression staining by CPS (PD-L1 CPS = 3.5) (×10).

Notably, in four cases, two or more concomitantly resected metastases from different peritoneal sites were available for analysis. We observed concordant PD-L1 CPS results in three patients: two had a PD-L1 CPS of 1 and one had a PD-L1 CPS of 4. However, we found discordant results between two different peritoneal implants in one patient, with PD-L1 CPS scores of 0 and 3, respectively. Interestingly, the primary tumor in this patient was positive for PD-L1, with a CPS of 11.

To account for the potential impact of collection timing on spatial heterogeneity, we also explored the concordance status between primary tumor and PM samples collected within the same time frame (i.e. synchronous lesions submitted to concomitant resection and/or biopsy). Overall, 13 patients were included in this subanalysis. The results were consistent with those reported in the general population. Using a cut-off CPS of 5, we observed a 76.9% concordance rate. In particular, there was 100% concordance among samples with PD-L1 CPS <5 in both the primary tumor and PMs. However, all three primary tumor cases with PD-L1 CPS ≥5 were negative in the PM. Using a cut-off CPS of 1, we observed a 69.2% concordance rate. In detail, there was a 62.5% concordance rate for the PD-L1 CPS ≥1 group and an 80% concordance rate for the PD-L1 CPS <1 group (Supplementary Tables S3 and S4, available at https://doi.org/10.1016/j.esmogo.2024.100089).

PD-L1 temporal heterogeneity

In nine patients, PD-L1 determination was carried out on primary tumor and PM specimens collected before and after receiving systemic chemotherapy, respectively. In two of these cases, patients received a combination of chemotherapy and ICI.

PD-L1 temporal heterogeneity findings are reported in Tables 4 and 5. Using a cut-off CPS of 5, the concordance rate for pre- and post-treatment PD-L1 positivity was 55.6% (5/9). Four pretreatment tumors with PD-L1 CPS ≥5 became PD-L1 CPS <5 after treatment, whereas no tumors with PD-L1 CPS <5 before treatment converted to PD-L1 positive after treatment. When using a threshold CPS of 1, the concordance rate dropped to 33.3% (3/9). Two PD-L1-negative pretreatment tumors became PD-L1 positive after treatment, whereas four PD-L1-positive pretreatment tumors became PD-L1 negative after treatment.

Table 4.

Comparison of PD-L1 expression before versus after therapy according to a PD-L1 CPS cut-off of 5

		Post-therapy PD-L1
		CPS ≥5	CPS <5	Total	Concordance rate, %	P valuea
Pretherapy PD-L1	CPS ≥5	0	4	4	0
	CPS <5	0	5	5	100
	Total	0	9	9	55.6	0.12

CPS, combined positive score; PD-L1, programmed death-ligand 1.

^aMcNemar’s test.

Table 5.

Comparison of PD-L1 expression before versus after therapy according to a PD-L1 CPS cut-off of 1

		Post-therapy PD-L1
		CPS ≥1	CPS <1	Total	Concordance rate, %	P valuea
Pretherapy PD-L1	CPS ≥1	2	4	6	33.3
	CPS <1	2	1	3	33.3
	Total	4	5	9	33.3	0.69

CPS, combined positive score; PD-L1, programmed death-ligand 1.

^aMcNemar’s test.

Discussion

Our study investigated the heterogeneity of PD-L1 expression in mGEA, comparing primary tumors and PMs. We found that spatial heterogeneity could be a critical issue even for PMs, as demonstrated previously also for other disease sites: in fact, using a CPS cut-off of 5, we reported a 72.7% concordance between primary and peritoneal specimens, indicating that a discordant PD-L1 result could be reported in about one out of four cases. Notably, with a CPS cut-off of 5, all PD-L1-positive primary tumors had PD-L1-negative matched PMs, whereas there was a 94.1% concordance between primary lesions and PMs among PD-L1-negative primary tumors. Using a CPS cut-off of 1, the concordance rate was lower (54.5%) and greater variability of the PM positivity status was observed, with 50% of PD-L1-negative primary tumor samples being positive on PMs, and 42% vice versa. Even when restricting the analyses to samples collected concomitantly from both tumor sites (thus limiting the potential confounding effect of systemic treatments received in the meantime), the results did not change. Besides, when different PMs from the same patient were assessed, results did not change significantly, with only one out of four cases showing different PD-L1 expression between peritoneal nodules. With regard to putative temporal heterogeneity, we reported a 56% concordance of PD-L1 status between pre- and post-treatment tumors, with a threshold CPS of 5. In particular, all PD-L1-positive pretreatment samples became negative in the matched post-treatment PM analysis.

Immunotherapy has revolutionized the systemic treatment landscape of mGEA, particularly in the first-line setting. The CheckMate 649 study, which evaluated the addition of the anti-PD-1 monoclonal antibody nivolumab to first-line chemotherapy (XELOX or FOLFOX), led to EMA approval for HER2-negative disease in this setting. This approval was eventually reinforced by the results of other trials with different agents,^4,5 even in HER2-positive disease.¹⁸ In all the aforementioned studies, the magnitude of survival benefit increased with PD-L1 expression and the clinical relevance of the additive benefit from nivolumab or pembrolizumab addition was very limited (if any) in the PD-L1-negative subgroup. Therefore PD-L1 expression emerged as the most valuable predictive biomarker for chemoimmunotherapy in mGEA, with MSI/MMR status being the only factor with stronger predictive ability.¹⁹ Moreover, subgroup analyses as well as different clinical series suggest that PM and ascites are associated with lower benefits from ICIs in GEA, even in the MSI-high/deficient-MMR subset.^3,20, 21, 22

However, PD-L1 CPS is still considered a suboptimal biomarker as a result of the use of different cut-off values and scoring systems across different trials with different agents, interobserver and interlaboratory variability, and primarily spatial and temporal heterogeneity.12, 13, 14^,23, 24, 25, 26

Spatial heterogeneity of PD-L1 expression has been observed in multiple cancer types and is attributed to heterogeneous PD-L1 staining in different sections of tumor samples, as well as to the different PD-L1 expression between primary tumors and metastases.²⁷ In this regard, Ye et al.²⁸ suggested that for patients with inoperable gastric cancer, more than five biopsy specimens are needed to assess the status of PD-L1 expression and identify patients who would benefit from immunotherapy. In GEA, significant spatial discordance of PD-L1 expression was observed in both resectable and metastatic diseases (Supplementary Table S5, available at https://doi.org/10.1016/j.esmogo.2024.100089). In resected tumors, Gao et al.²⁹ found that regional lymph node metastases had higher PD-L1-positive rates than matched primary tumors. By contrast, Zhou et al.¹² showed that in advanced stages, distant metastases had lower PD-L1-positive rates than matched primary tumors, with concordance rates of 61% and 84% using CPS cut-offs of 1 and 10, respectively. Interestingly, there was no clear dependence of PD-L1 concordance on the metastatic site. The concordance rate between the primary tumor and PMs was 67% against a cut-off of 1.¹²

In the context of spatial heterogeneity, our results align with those reported by Zhou and colleagues.¹² However, they did not observe a clear pattern of PD-L1 status modification after chemotherapy. Recently, however, temporal heterogeneity of PD-L1 expression has been described in relation to chemotherapy administration, primarily indicating a reduction in PD-L1 expression following treatment. Here, we observed a marked temporal heterogeneity, with a 56% concordance of PD-L1 status between pre- and post-treatment tumors at a CPS threshold of 5. Notably, all PD-L1-positive pretreatment samples became negative in the matched post-treatment PM analysis.

Spatial and temporal heterogeneity of PD-L1 expression might ultimately affect the outcomes of regimens containing ICIs. As a result of the small number of patients treated with ICIs in our case series, we were unable to provide evidence of this correlation. However, the four patients in our study who were treated with nivolumab plus chemotherapy exhibited discrepancies in PD-L1 between the primary tumor and PMs. They achieved conflicting objective responses at different tumor sites and reported lower PFS and OS outcomes compared with CheckMate 649,³ highlighting the poor prognosis of peritoneal carcinomatosis even with immunotherapy.

To the best of our knowledge, this is the first study to report both spatial and temporal heterogeneity in mGEA with a specific focus on PMs. However, some limitations have to be pointed out. First, it is a retrospective study with a small sample size from a single institution, so confirmation in larger, independent cohorts is needed. Second, matched primary tumor and PM samples were available at baseline for only 13 out of 22 patients. Therefore the discrepancy in PD-L1 status observed between primary tumors and PMs in patients exposed to systemic therapy should not be interpreted solely as temporal heterogeneity (induced by the selective pressure of treatment) but also as spatial heterogeneity (due to different microenvironments in primary tumors and PMs). In addition, our study cannot definitively determine whether PD-L1 heterogeneity contributes to resistance to ICIs, as only four patients were treated with nivolumab-containing regimens, yielding conflicting results. Thus, larger studies with ICI-treated patients and post hoc analyses from randomized trials are needed. Moreover, in our study, 40.9% of primary tumor specimens were obtained through endoscopic biopsies, which prevented us from investigating intratumoral heterogeneity in the primary lesions. Lastly, our cohort included patients treated since 2015. It is well-known that, like many other biomarkers evaluated by IHC, PD-L1 expression may deteriorate over time, largely due to tissue storage conditions.³⁰ Therefore we focused on samples collected within this time frame from a single institution, ensuring consistent fixation protocols and sample storage practices that remained unchanged throughout the years.

In conclusion, our analysis confirmed both spatial and possibly temporal discrepancies in PD-L1 CPS expression between PMs and matched primary gastroesophageal tumors. This discrepancy occurred regardless of the biomarker cut-off value, with at least one-quarter of patients showing discordant results. Specifically, we found that all patients with PD-L1-positive primary tumors had PD-L1-negative PMs. These findings suggest that relying solely on peritoneal specimens for PD-L1 assessment may be problematic. Therefore efforts should be made to include evaluable primary tumor samples (either biopsies or surgical specimens) in patients with PMs. As adequate material from both the primary tumor and distant metastases is essential for biomarker assessment in GEA, we recommend that if PMs are the only available samples for PD-L1 assessment, a rebiopsy of the primary tumor (if archival tissue is unavailable or inadequate for IHC) or other metastatic sites should be considered. In this regard, guidelines and recommendations about the number and location of endoscopic biopsies should be followed.¹⁴ Further studies with larger cohorts are needed to confirm our findings and to determine the optimal anatomical site and timing for evaluating PD-L1 expression.