Analysis of T-Cell Subsets Using Multiplex Immunohistochemistry and Clinical Outcomes of Immune Checkpoint Inhibitors in Advanced Gastric Cancer Patients

Article information

J Korean Cancer Assoc. 2025;.crt.2025.458

Publication date (electronic) : 2025 September 9

doi : https://doi.org/10.4143/crt.2025.458

Tae-Yong Kim^1,2, Jeesun Yoon¹, Dae-Won Lee^1,2, Yoonjin Kwak³, Hye Seung Lee³, Do-Youn Oh^,1,2,4

¹Department of Internal Medicine, Seoul National University Hospital, Seoul, Korea

²Cancer Research Institute, Seoul National University College of Medicine, Seoul, Korea

³Department of Pathology, Seoul National University Hospital, Seoul, Korea

⁴Integrated Major in Innovative Medical Science, Seoul National University Graduate School, Seoul, Korea

Correspondence: Do-Youn Oh, Department of Internal Medicine, Seoul National University College of Medicine, 101 Daehak-ro, Jongno-gu, Seoul 03080, Korea Tel: 82-2-2072-0701 E-mail: ohdoyoun@snu.ac.kr

*Tae-Yong Kim and Jeesun Yoon equally contributed to this work.

Received 2025 April 28; Accepted 2025 September 7.

Abstract

Introduction

Metastatic or recurrent gastric cancer (GC) remains incurable, with chemotherapy being the key treatment [1]. Immunotherapy has emerged as an essential component in treating advanced GC patients [2]. The CheckMate-649 study demonstrated that adding the programmed cell death-1 (PD-1) monoclonal antibody (mAb), nivolumab to platinum-doublet chemotherapy showed an improvement of survival outcome as first-line treatment for human epidermal growth factor receptor 2 (HER2)–negative GC [3]. Following the success of the KEYNOTE-859 study [4] and the Rationale-305 study [5], immunotherapy in combination with chemotherapy has become the standard of care as first-line treatment for HER2-negative GC. Additionally, in HER2-positive GC, the KEYNOTE-811 study demonstrated significant improvement of clinical outcomes by adding pembrolizumab to chemotherapy and trastuzumab [6], thereby establishing immune checkpoint inhibitors (ICIs) as an essential component of treatment strategy in advanced GC regardless of HER2.

Cytotoxic T-lymphocyte–associated antigen 4 (CTLA4) and PD-1 inhibit T cell signaling by interacting with CD80 or CD86 and programmed cell death-ligand 1 (PD-L1) or PD-L2, finally exhausting T-cells [7]. Several biomarkers including microsatellite instability (MSI), tumor mutational burden, PD-L1 expression, and tumor infiltrating lymphocytes (TILs) have been suggested to predict the response to ICIs across tumor types [8]. In case of GC, MSI, PD-L1 expression or Epstein-Barr virus (EBV) have been reported to be associated with the clinical efficacy of ICIs [9]. Additionally, with the emergence of single-cell RNA sequencing, the role of TILs in response to ICIs has been actively underway in GC. As a central axis of immune response, T cells have a key role in directly killing cancer cells and coordinating immune cells through cytokines [10]. Tumor-infiltrating CD8⁺ T cells have been associated with overall survival (OS) in diverse tumor types including GC [10-12]. Intratumoral Foxp3+RORgammat+ T cells have been associated with impaired function of CD8⁺ T cells and may predict poor prognosis in GC [13]. Analysis of peripheral blood T cells and T-cell gene signatures are also suggested to predict response to ICIs in GC [14,15]. Although these testing methods provide a deep understanding of T-cell subsets and their roles in GC, there are challenges to be applied to daily practice setting, which includes requirement of fresh and large tissue samples and financial toxicity, etc. Therefore, the current study was designed to assess intratumoral T-cell subsets using multiplex immunohistochemistry (mIHC), which could be more feasible method to predict clinical outcomes of ICIs in GC.

Materials and Methods

Results

1. Patients

A total of 84 patients were analyzed (Table 1, S5 Table). The MSI-H, EBV, and non-MSI/EBV group included 7 (8.3%), 3 (3.6%), and 74 (88.1%) patients, respectively. Based on Lauren’s classification, all GCs harboring MSI-H were intestinal tumor type. GC harboring MSI-H and EBV showed a higher proportion of CPS ≥ 1 than non-MSI/non-EBV GC (85.7% and 100% vs. 40.5%, respectively). Majority of patients (63.1%) received ICI monotherapy. Among 72 patients who were available for tumor response assessment, 16 (22.2%) achieved complete response (CR) or partial response (PR), with two showing CR (2.8%) (Fig. 1A and B). Among the 16 responders, two were EBV-positive, six were MSI-H, and eight were in the CIN group, while there were no responders in the GS group (S5 Table, S6 Fig.). The mean OS and progression-free survival (PFS) of all study groups were 8.1 months (95% confidence interval [CI], 5.231 to 11.035) and 4.0 months (95% CI, 2.511 to 5.489), respectively. The median OS and PFS of MSI-H and EBV were longer than those of non-MSI/non-EBV group (Table 2). GC with CPS ≥ 1 showed longer OS than GC with CPS < 1 (17.7 vs. 5.9 months, p=0.010). Longer OS were also observed in early line of ICIs (17.7 vs. 4.1 months, p=0.010) and combination therapy (17.9 vs. 5.9 months, p=0.022).

Table 1.

Baseline characteristics of study populations

Fig. 1.

Tumor response of immune checkpoint inhibitors. (A) The waterfall plot for best tumor reduction from baseline. Each yellow (microsatellite instability–high [MSI-H]), green (Epstein-Barr virus [EBV]) and light blue (non–microsatellite instability [MSI]/non-EBV) bar means best change of sum of target lesions’ diameter from baseline, based on Response Evaluation Criteria in Solid Tumors (RECIST) 1.1 in each subject. Black star marks above or below each bar mean patients who received immune checkpoint inhibitors (ICs) as first- or second-line. (B) The spider plot for change of sum of target lesions’ diameter over time since the initiation of ICs, based on Response Evaluation Criteria in Solid Tumor 1.1. ICB, immune checkpoint blockades.

Table 2.

Summary of overall survival and progression-free survival

3. Intratumoral T-cell subsets and tumor response

GC with a high T_cyto/T_total ratio showed better tumor response than those with a low ratio (34.2% vs. 8.8%, p=0.034) (Table 3). A high ratio of memory T_cyto/T_mem was also associated with better tumor response (34.2% vs. 8.8%, p=0.026). However, the densities of T_cyto and PD-1– or TIM-3–expressing T_cyto were not statistically different between CR/PR and stable disease (SD) or progressive disease (PD) groups. While the average density of T_cyto was not different between tumor response, the average ratio of T_cyto/T_total was higher in CR/PR (0.53) than in PD (0.31) (p=0.021) (Fig. 2A). The average ratio of T_reg/T_help was also higher in CR/PR (0.46) than in SD (0.33) group (Fig. 2C). However, the average ratio of T_help/T_total and memory T_help/T_mem was lower in CR/PR group than PD group (Fig. 2B and D).

Table 3.

Summary of the response rate of ICIs

Fig. 2.

T-cell subsets and tumor response. (A) A T_cyto/T_total ratio according to tumor response. (B) A T_help/T_total ratio according to tumor response. (C) A T_reg/T_help ratio according to tumor response. (D) A memory T_help/T_mem ratio according to tumor response. CR, complete remission; NS, no significance; PD, progressive disease; PR, partial remission; SD, stable disease; T_cyto, cytotoxic T cells; T_help, helper T cells; T_mem, memory T cells; T_reg, regulatory T cells; T_total, total T cells.

4. Intratumoral T-cell subsets and survival

GC with a high density of T_cyto showed longer OS than those with a low density (18.2 vs. 6.5 months, p=0.001) (Table 4, Fig. 3A). A high ratio of T_cyto/T_total was associated with better OS (24.3 vs. 5.9 months, p < 0.001) (Fig. 3B). GC with a high density of PD-1 or TIM-3–expressing T_cyto showed better OS than those with a low density of such cells (17.9 vs. 7.0 months, p=0.024 and 18.2 vs. 7.0 months, p=0.003, respectively) (Fig. 3C and D). While no difference in OS was observed based on the ratio of T_reg/T_help (9.0 vs. 8.1 months, p=0.963), a low T_reg/T_total ratio was associated with prolonged OS (5.9 vs. 17.9 months, p=0.034) (Fig. 3E). Furthermore, a low density of memory T_reg were related with better OS (5.9 vs. 10.8 months, p=0.025), and a low ratio of T_mem/T_total was also related to prolonged OS (5.9 vs 10.4 months, p=0.027) (Fig. 3F). GC with a high ratio of memory T_cyto/T_mem and a low ratio of memory T_reg/T_mem showed longer OS, compared with a low or high ratio of those, respectively (18.1 vs. 5.9 months, p < 0.001 and 5.9 vs. 19.6 months, p=0.001, respectively) (Fig. 3G and H). PFS showed similar trends to OS, except for T_reg/T_total ratio, memory T_reg density, and memory T_reg/T_mem ratio (Table 4). Cox’s proportional hazard analysis indicated that the molecular subtypes (MSI/EBV vs. non-MSI/non-EBV), line of ICIs (first- or second-line vs. later), combination, and T_cyto density (high vs. low) were significantly associated with OS (S8 Table 3, S9 Fig.).

Table 4.

Summary of overall survival and progression-free survival

Fig. 3.

Kaplan-Meier survival curves, depending on the density and ratio of T-cell subsets. Overall survival (OS), according to T_cyto density (A), T_cyto/T_total ratio (B), programmed cell death-1 (PD-1)–expressing T_cyto (C), TIM-3–expressing T_cyto (D), T_reg/T_total ratio (E), T_mem/T_total ratio (F), memory T_cyto/T_mem ratio (G), and memory T_reg/T_mem ratio (H). T_cyto, cytotoxic T cells; TIM-3, T-cell immunoglobulin and mucin domain-3; T_mem, memory T cells; T_reg, regulatory T cells; T_total, total T cells.

Discussion

Immunotherapy has become a key anti-cancer agent in the treatment of advanced GC. In particular, the role of T cells is fundamental, and various approaches are conducted to gain a deeper understanding of T cell role. Our study showed that intratumoral T-cell subsets were associated with response to ICIs (Table 3). T_cyto positively affected survival on ICIs treatment (S8 Table). Interestingly, the ratio of memory T-cell subsets, for example, memory T_cyto or memory T_reg over total T_mem were also associated with clinical efficacy of ICIs. These results are meaningful because this study evaluated the responsiveness to immunotherapy in patients with GC who received immunotherapy alone, thereby demonstrating the role of T-cell subsets as a predictive marker for immunotherapy. Additionally, these findings are consistent with recent analyses of T-cell subsets through single-cell RNA sequencing in patients with GC who received cytotoxic chemotherapy and immunotherapy combination [14,17,18].

CD8⁺ cytotoxic T lymphocytes (CTLs) have a major role in killing and eliminating cancer cells via release of granules or induction of FasL-mediated apoptosis [19]. Intratumoral CD8⁺ CTLs were associated with anti-tumor activity of ICIs in a preclinical study [20], which is in line with our findings. Tumor antigen (TA)–specific CTLs mainly have an exhausted phenotype, suggesting CTLs in tumor are easily reinvigorated by ICIs [21]. Although we did not directly evaluate TA-specific CTLs, intratumoral T_cyto were likely to be TA CTLs, contributing to better anti-tumor activity of ICIs.

Interestingly, a high density of PD-1– or TIM-3–expressing T_cyto was significantly associated with better clinical efficacy of ICIs in our study. CTL with PD-1^high were an exhausted phenotype and associated with worst prognosis [22]. However, some preclinical study suggested that PD-1^high CTL were more easily reinvigorated by anti–PD-1 monoclonal antibodies than PD-1^low CTL ex vivo [23], and patients with high levels of circulating PD-1^high CTLs showed better responses to immunotherapy [17]. These findings are consistent with our observation that a high density of PD-1– or TIM-3–expressing T_cyto was associated with better clinical efficacy of ICIs in our study. Therefore, in-depth functional studies are needed in the future to comprehensively understand the conflicting research results between the expression of T cell exhaustion markers such as PD-1 or TIM-3 and the response to immunotherapy.

Memory T-cell subsets, for example, memory T_cyto or memory T_reg, were associated with clinical outcomes of ICIs in our study. While a high density of CTLs linked to tissue-resident memory T cells (Trm) and CTL from Trm cells showed enhanced cytotoxicity, memory T_reg decreased anti-tumor immunity, leading to tumor progression [24,25]. Taken together, our findings that a high ratio of memory T_cyto/T_mem and a low T_reg/T_mem was associated with longer OS are reasonable.

TCGA studies have classified GC into four molecular subtypes: GC with MSI-H, EBV, GS, and CIN [16]. The NGS results in this study dataset were insufficient to accurately classify the samples according to the TCGA classification. To improve this limitation, we conducted an additional analysis to divide the samples into two groups, estimated to be CIN and GS groups, based on the IHC results (EGFR, c-MYC, c-MET) and histological features obtained from the clinical data of individual patients. GC harboring EBV was associated with PD-L1/L2 overexpression and immune cell signaling. As expected, GC with MSI-H and EBV showed a better tumor response and survival, compared with non-MSI/non-EBV GC in our study (Table 1, S10 Table), which was consistent with a previous study [9]. PD-L1, a well-known biomarker that can predict responses to ICIs in GC. Similarly, GC with CPS ≥ 1 also showed longer survival than GC with CPS < 1 in our study (17.7 and 5.9 months, p=0.010). Besides molecular subtypes and CPS, the line of treatment and single or combination treatment affected clinical outcomes of ICIs in our study. To exclude the influence of aforementioned factors, we re-analyzed OS according to tumor biology, CPS, the line of treatment, and with or without combination. As can be seen in S11-S13 Tables, T_cyto, PD-1, TIM-3, T_help, and T_mem cell subsets were still meaningful to OS, and especially T_cyto density was a meaningful marker regardless of aforementioned clinicopathologic factors (S8 Table).

For analysis of T-cell subsets in a tissue or blood, fluorescence-activated cell sorting (FACS) is a commonly used method. However, FACS needs a fresh and adequate amount of tissue, which is a major limitation on translational research using a human tissue. Recent advances in technology, from bulk transcriptome analysis to single-cell transcriptome analysis, have provided deeper insights into the role of intratumoral T-cell subsets. However, it is not realistic to apply these analyses to all patients in clinical practice setting and immediately utilize the results in treatment. IHC can simply assess T cell surface markers; however, it cannot read two or more surface markers in the same section, due to the cross-reactivity of secondary antibodies [26]. Multiplex IHC overcomes the limitation of conventional IHC and enables analysis of T-cell subsets even in a small biopsy FFPE tumor tissue [27]. In this context, our study suggested the use of cost-effective multiplex IHC to analyze T-cell subsets and predict responses to ICIs.

This study has some limitations. First, the number of samples was small. In the study planning stage, we excluded patients who had received immunotherapy in combination with chemotherapy in order to observe a homogeneous response to immunotherapy. Recently, most patients with advanced GC receive immunotherapy as a first-line setting in combination with chemotherapy. Therefore, we did not additionally include patients in the front-line setting after immunotherapy was widely applied. Due to these limitations, a large-scale external validation study using patient samples from actual clinical settings where immunotherapy monotherapy was administered is necessary to validate our findings in the future. Second, the target patient population is heterogeneous, consisting of patients with diverse treatment settings. Currently, ICI monotherapy for advanced GC is being conducted in the third- or later-line setting, with an objective response rate of approximately 10%. Considering that treatment response rates decrease as the later-line setting in advanced cancer, this study included patients who participated in clinical trials using immunotherapy as first- or second-line treatment. Because our study population was heterogeneous, careful interpretation was necessary to account for the influence of confounding variables. Further external validation in a homogeneous patient population is required. Third, during the analysis of T-cell subsets, the cutoff was standardized to the median, which may lack biological relevance. We attempted to apply receiver operating characteristic curves based on response status and maximize Log-rank statistics based on survival for each parameter; however, due to the small sample size mentioned earlier, it was difficult to present a statistically significant, unified cutoff determination method for all parameters. Fourth, line of treatment was not unified. Among 84 patients in this study, 38 (45.2%) were in the second-line setting, and 28 (33.3%) were in the third-line setting. To overcome the small sample size, we analyzed all patients without distinguishing treatment lines and classified each T cell parameter into high/low groups based on the median value. When classified in this manner, there were no significant differences in patient distribution between the high/low groups according to treatment line. Further studies are needed with a more homogeneous patient population within the same treatment line setting.

In conclusion, our study shows that intratumoral T_cyto play an important role in responses to ICIs, and PD-1 or TIM-3–expressing T_cyto are associated with clinical efficacy of ICIs. Memory T-cell subsets, including memory T_cyto or T_reg, are also essential for response to ICIs. These findings suggest that interaction between tumor and intratumoral T cells is crucial to determine the activity of ICIs in GC. To assess T cell subsets, multiplex IHC is a feasible method using archival gastric tumor tissue.

Electronic Supplementary Material

Article information Continued