PD-L1 is overexpressed in 25-65% of GC. In general, the PD-L1/PD-1 axis is recognized as an important factor for the poor prognosis of patients with different GC types. Firstly, PD-L1 expressed on GC cell surfaces inhibit anti-tumor immunity via multiple pathways. For example, over-expression of PD-L1 prevents the body’s immune system from recognizing GC cells; effector T cells are unable to target GC cells, leading to suppressing the anti-tumor immune response (6–8). Secondly, PD-L1 also binds to the epidermal growth factor receptor (EGFR) and activates it to promote GC progression (9). In addition to GC cells, PD-L1 is highly expressed on immune cell surfaces in the tumor immune microenvironment, including lymphocytes, neutrophils, macrophages, and mast cells (10).

Microsatellite instability–high/deficient mismatch repair (MSI-H/dMMR) represents a specific subtype of tumor that accounts for 8–10% of GC characterized by high tumor immunogenicity and dense immune cell infiltration (58, 59). Given relatively low MSI-H GC incidences, no targeted Phase III clinical trials have been conducted. As ICIs have revolutionized the therapy of MSI-H/dMMR malignancies, their potential in neoadjuvant therapy is receiving more attention. MSI-H GC, characterized by defective DNA mismatch repair, exhibits a high frequency of somatic mutations, leading to the generation of abundant neoantigens (60). These neoantigens can activate robust anti-tumor T-cell responses and are associated with increased infiltration of anti-tumor immune cells within the tumor microenvironment, thereby rendering MSI-H tumors particularly sensitive to ICIs (61).

Epstein-Barr virus–associated GC (EBVaGC) accounts for 2-10% of all GC cases (74). Epstein-Barr virus (EBV) can cause a local immune response (75, 76). Analysis of tumor genomic profiles shows that EBV⁺ tumors are often microsatellite-stable (MSS) (77). Evidence exists of a low tumor mutational burden and stronger immune infiltration in EBV⁺ tumors compared to MSI-H tumors. In addition, EBV⁺ tumors showed higher expression of immune checkpoint pathway (PD-1, CTLA-4) genes and higher infiltration of lymphocytes (e.g., follicular helper T cells and CD8⁺ T cells (78)) in their RNA sequence data compared to MSS tumors. Therefore, the use of anti-PD-1 and anti–CTLA-4 monoclonal antibodies in a neoadjuvant setting is possible for patients with EBV⁺ GC.

The TMB was defined as the count of non-synonymous mutations per megabase of the tumor genome reflecting tumor immunogenicity (85). The somatic TMB may lead to the formation of new antigens, and the presence of additional antigens enhances the probability that tumor cells will be detected by immune cells that infiltrate the tumor (86), thereby activating T-cell-mediated anti-tumor responses (87). Thus, medications stimulating T-cell activation, such as monoclonal antibodies targeting PD-1 or PD-L1, might offer potent anticancer treatment, especially for individuals with an elevated TMB. Several studies have demonstrated a favorable association between TMB levels and response to ICIs in different types of tumors (88–90). As a result, in June 2020, pembrolizumab was approved by the FDA for all solid tumors with high tumor mutational load (TMB-H). However, the TMB threshold used to screen for immunotherapy efficacy has led to different results in different clinical trials.

A TMB ≥ 10 mutations/megabase is a clinically significant cut-off point (91). In the Keynote-158 study, investigators used pembrolizumab to treat people with solid tumors showing an advanced high-TMB (TMB ≥ 10 mut/Mb). Twenty-four (10.3%) patients with GC were included. The results of the study found an ORR of 29% in patients with TMB-H tumors, compared to only 6% in patients who were non-TMB-H. Exploratory outcome analysis showed a correlation between TMB, and the clinical results of first-line pembrolizumab treatment and pembrolizumab combination chemotherapy after adjusting for CPS (ORR, PFS, and OS; all P < 0.05); those cases that had a TMB ≥ 10 mut/Mb and were managed with pembrolizumab demonstrated better clinical benefits (ORR, PFS, and OS) (92). Two additional studies reached the same conclusion, but with different TMB thresholds of 12 mut/Mb and 20 mut/Mb, respectively. A phase Ib/II clinical trial conducted by F. Wang et al. revealed that for AGC patients who failed first-line chemotherapy receiving Toripalimab monotherapy, regardless of the expression level of PD-L1, the OS of patients with ≥ 12 mut/Mb was significantly better than that of patients with < 12 mut/Mb (32). The findings of Zhang et al. showed that in advanced, resectable MSI-H gastrointestinal tumors, patients with a high TMB (TMB ≥ 20 mut/Mb) responded well to neoadjuvant ICIs, regardless of PD-L1 levels. In the KEYNOTE-061 trial, pembrolizumab improved outcomes in patients with TMB ≥ 175 mutations/exome (whole exome sequencing assessment) compared to chemotherapy (19). The TMB exhibited significant associations with ORR, PFS, and OS (all, P < 0.05) in the pembrolizumab group, whereas no notable correlations were detected in the chemotherapy group.

The above experimental data suggest that the TMB status in GC may serve as a predictive biomarker for pembrolizumab efficacy. However, several factors limit its reliability. Heterogeneity: it should be noted that the TMB cutoff value varies across different tumor types. Therefore, it is necessary to determine the TMB cutoff in GC to establish its value as a predictive biomarker. Gene fusion: conventional TMB assays, such as whole exome sequencing (WES) or targeted panels, may not effectively capture gene fusion events, especially those in non-coding regions or involving complex rearrangements. Gene fusions may serve as oncogenic drivers in tumor progression (e.g., CLDN18-ARHGAP fusion in GC) (93) and can generate neoantigens. However, the impact of gene fusions on immunotherapy response is not captured by TMB (94), which may lead to an underestimation of immunogenic potential. Moreover, fusion-driven tumors (e.g., NTRK fusions) exhibit a distinct immune microenvironment, where high TMB may not reliably predict treatment efficacy but instead shows stronger correlation with targeted therapy response. Clonality: TMB only reflects the total mutational burden, encompassing both clonal (truncal) and subclonal mutations, yet clonal mutations alone may more reliably indicate immunogenicity. Subclonal mutations may contribute to immune escape due to spatial heterogeneity. The heterogeneity of mutational profiles between primary and metastatic sites may compromise the accuracy of TMB assessment based on a single biopsy. Quality of mutations: not all mutations are capable of generating effective immunogenicity. Both nonsynonymous mutations (which may generate neoantigens) and synonymous mutations (which do not alter the amino acid sequence and lack immunogenicity) are counted in TMB, yet only the former may influence immunotherapy response (95). Driver mutations (e.g., TP53, KRAS) may indirectly modulate immune response by altering the tumor microenvironment (96, 97). Although passenger mutations inflate TMB values, most fail to elicit effective immune responses due to issues such as protein stability or impaired MHC binding. TMB cannot distinguish their respective contributions. Mutation type: indels and frameshift mutations may generate highly immunogenic neoantigens, but certain detection methods (e.g., targeted panels) exhibit limited sensitivity for these mutation types. Logistic issue: the TMB is also affected by a variety of factors, such as tumor type, detection method, and analysis technique. For instance, WES covers the entire exonic region but is costly, while targeted panels only interrogate a limited set of genes (98), potentially leading to TMB underestimation. In bioinformatics analysis, both variant calling algorithms and germline mutation filtering criteria can significantly impact the final TMB calculation.

Currently, research is gradually emerging on combining TMB with other biomarkers to predict the effectiveness of immunotherapy. For instance, recent studies suggest that combining TMB with T-cell-inflamed gene expression profile (GEP) can achieve superior predictive performance. In the KEYNOTE clinical dataset, it has been demonstrated that patients with solid tumors, including GC, who exhibit both high TMB (TMB-H) and high GEP (GEP-H), achieve the highest ORR (99). The combination of TMB and TIDE has demonstrated robust predictive efficacy in breast cancer, lung adenocarcinoma, and hepatocellular carcinoma (100–103), with future studies planned to explore its predictive potential for immunotherapy response in GC. Moreover, genomic and transcriptomic factors, such as novel antigen presentation by MHC-I and II complexes, can predict ICI outcomes (87, 104). Currently, there is a lack of validation for such models in GC-specific cohorts. For GC patients who have completed immunotherapy clinical trials (e.g., ATTRACTION-4, KEYNOTE-062), retrospective extraction of transcriptomic data from archived samples can be performed to validate the predictive efficacy of the aforementioned models. Additionally, prospective randomized controlled clinical trials can be conducted to further evaluate the predictive capability of these models. It is important to note that the molecular heterogeneity of GC (e.g., EBV-positive type, genomically stable type) may affect the generalizability of the models. Subgroup analyses (e.g., stratified by PD-L1 CPS, MSI, and EBV status) can be conducted to validate the universality of the models. The results are promising, as they may provide new directions for the development of biomarkers for GC immunotherapy.

POLE/POLD1 mutations are thought to be linked to a high TMB, an enhanced tumor immune reaction and a better response to ICIs (105). It has been shown that POLE and POLD1 mutations are able to be used as stand-alone biomarkers for anticipating the efficacy of pan-cancer immunotherapy, and that they are also markers of a poor prognosis (106–108). Zhu et al. found that patients with GC and POLE/POLD1 mutations typically exhibit an acquired immune resistance to the TME, with an increase in PD-L1 expression and an elevation of TMB (109). The above suggests that mutations of the POLE/POLD1 gene can be used as a biomarker to improve the clinical efficacy of neoadjuvant immunotherapy in people with GC; identical results were obtained in NCT03012581. In this trial, Rousseau et al. prospectively evaluated the efficacy of nivolumab monotherapy in patients with advanced POLE/POLD1 mutated solid tumors. Two patients (9%) with GC were included in this trial. It was found that only tumors with selective pathogenic mutations in the catalytic site of the DNA-binding or nucleic acid exonuclease structural domains exhibited high mutational loads, high T-cell infiltration, and high response rates to anti-PD1 monotherapy (110).

Given the variation in time and space of GC, blood-based predictive biomarkers from liquid biopsies have surfaced as a hopeful strategy for anticipating responses to immunotherapy (111). Circulating tumor DNA (ctDNA) is one widely studied biomarker in liquid biopsies (112). Jin et al. performed next-generation sequencing testing on 46 patients with metastatic GC treated with neoadjuvant PD-1 inhibitor immunotherapy. They showed that patients with a > 25% decrease in the frequency of the largest variant allele in the ctDNA assay had a longer median PFS (7.3 m vs. 3.6 m; P = 0.0011) and higher ORR (53.3% vs. 13.3%). The study also found that patients with TGFBR2, RHOA, and PREX2 mutations in the GC had significantly shorter PFS than those without mutations (113). Results from another study that included 61 patients with metastatic GC receiving neoadjuvant therapy with pembrolizumab showed that a decrease in the concentration of ctDNA in the sixth week of continuous ctDNA monitoring can be used to predict a benefit from neoadjuvant immunotherapy (73). Qiao et al. treated patients with unresectable advanced GC using neoadjuvant dendritic cells mixed with cytokine-induced killer cells (DC-CIK; an immune cell therapy) in combination with chemotherapy. They found that the frequency and number of ctDNA mutations were reduced in 19 patients (63.3%) after DC-CIK infusion. Reduced ctDNA mutation frequency was associated with improved PFS and OS (P = 0.001) (114). The above suggests that ctDNA testing is useful in anticipating the performance of individualized neoadjuvant immunotherapy in patients with GC. A prospective phase II clinical trial (NCT05594381) is currently underway to investigate the feasibility of ctDNA for evaluating the performance of PD-1 inhibitors in combination with a SOX regimen as a neoadjuvant therapy for locally progressive GC. Another aim of this study is to construct a model for evaluating the efficacy of treatment so as to clarify the applicability of the neoadjuvant immunotherapy for locally progressive GC. The results are eagerly anticipated.

Liquid biopsy offers advantages. Compared to other emerging biomarkers (such as TMB, PD-L1, or immune gene signatures), ctDNA testing exhibits higher sensitivity, enabling the detection of minimal residual disease (MRD) tumor burden and early recurrence. Moreover, from a practical standpoint, ctDNA’s blood-based detection offers dual advantages: enhanced clinical accessibility and serial monitoring capacity for tracking disease progression. However, it also has limitations as a screening approach for predictive biomarkers in GC immunotherapy. First of all, technical methods and sample quality limit the sensitivity of liquid biopsies, which may fail to detect tumor biomarkers that are at low concentration levels. Second, compared to other biomarkers such as TMB, PD-L1, or immune gene signature in tumor specimen, biomarkers in the blood may not be tumor-specific, leading to false-positive results. Therefore, given the significant clinical potential of ctDNA, more studies are needed to further improve the reliability of liquid biopsy, thus contributing to the screening of a GC population sensitive to immunotherapy.

Recent studies have begun to focus on the relationship between the gut microbiota, and tumor progression and treatment outcomes. This has been proposed as a potential biomarker for participating in the outcome of immunotherapy in solid tumors (115, 116). Although specific gut microbes play a role in GC progression and immunomodulation, insufficient evidence exists to support their potential as efficacy biomarkers in neoadjuvant immunotherapy for GC.

Recent research indicates that H. pylori infection may hinder the growth and anti-tumor functions of CD8⁺ T cells, foster the transformation of naive T cells into Tregs, and modulate the production of inflammatory mediators. These actions influence the TIME, dampen host immune reactions, and diminish the effectiveness of immunotherapy for GC (117–119). Therefore, patients with locally advanced GC should be aware of their H. pylori infection status before receiving neoadjuvant immunotherapy. However, forward-looking research exploring the relationship between H. pylori and prognosis in GC immunotherapy is lacking. In addition, the use of H. pylori as a predictive marker of immunotherapy efficacy needs to be thoroughly demonstrated (120).

LAG-3 is an inhibitory receptor on cell surfaces, which negatively regulates both CD8⁺ and CD4⁺T cell activity. It maintains immune system homeostasis under normal physiological conditions (121, 122), while LAG-3 and Treg cell interactions stimulate Treg activity, strengthen immune tolerance, and indirectly suppress dendritic cell (DC) function, thereby facilitating tumor cell immune escape. Consequently, LAG-3 is a promising therapeutic target in cancer immunotherapy, complementing the PD-1/PD-L1 pathway (123). Currently, the joint application of LAG-3 and PD-1 inhibitors is a topical research field. Kelly et al., in their innovative study, combined PD-1 and LAG-3 inhibitors with chemoradiotherapy for the neoadjuvant treatment of patients with gastroesophageal cancer (124). The authors reported that higher baseline PD-L1 (CPS ≥ 5) and LAG-3 expression was associated with superior pathological responses, with a Phase Ib clinical study also providing safety insights on combining PD-1 and LAG-3 for gastroesophageal cancer. But, in a Phase II study (RELATIVITY-060), combined nivolumab and the LAG-3 inhibitor relatlimab with chemotherapy for advanced G/GEJC failed to reach its primary endpoint (125). Therefore, more research is required to test the safety and effectiveness of PD-1 inhibitors when combined with LAG-3 inhibitors as an immunotherapy for patients with GC.

CTLA-4 is a co-inhibitory molecule on activated T and regulatory T cell surfaces (Tregs). It interacts with B7-1/B7–2 ligands on antigen-presenting cells, thereby inhibiting CD28-mediated signaling, which is responsible for T cell activation. Monoclonal antibodies targeting CTLA-4 have been shown to block its competition with CD28 for binding to B7, which in turn activates the CD28 signaling cascade and reduces immunosuppressive Treg cell populations in the TME (126). As the world’s first PD-1/CTLA-4 bispecific antibody, in the COMPASSION-15 (AK104-302) Phase III trial, cadonilimab showed significant efficacy in patients with different PD-L1 expression levels. In the trial, the proportion of individuals with PD-L1 CPS < 5 and CPS < 1 reached 49.8% and 23%, respectively, which was better reflected real-world patients. Therefore, these data confirmed the unique advantages of cadonilimab in individuals with low PD-L1 expression, and reflecting this, on September 30^th, 2024, the National Drug Administration officially approved its combination with XELOX for the first-line treatment of unresectable locally advanced recurrent or metastatic G/GEJ adenocarcinoma.

TIM-3 is encoded by HARVCR2 (127) and is an emerging target for cancer immunotherapy (128). Studies have shown that TIM-3 expression levels in GC tissue are significantly higher than in normal gastric mucosa tissue (129, 130), and positively correlated with PD-1/PD-L1 expression levels in GC tissue (131). Chen et al., reported that TIM-3 potentially promoted CD8⁺ T cell dysfunction in GC, manifested by decreased IFN-γ, perforin, and Granzyme B (GzmB) levels, but increased PD-1 and CTLA-4 levels (132). Although HARVCR2 mRNA is elevated in most GC subtypes, it is more highly expressed in EBV-positive and MSI subtypes, as characterized by increased immune characteristics and higher immunotherapy responsiveness (73). Further research is required to examine the potential benefits of anti-TIM-3 inhibitors in patients with GC, and investigate their potential in combination with anti-PD-1/PD-L1 inhibitors.

AFP has a wide range of biological functions, with previous research showing that it directly stimulated cancer cell proliferation and growth while also inhibiting apoptosis (133). AFP also inhibited monocyte differentiation to fully functional DCs and prevented them presenting foreign antigens to CD8⁺ lymphocytes via MHC signaling (134, 135). Furthermore, AFP down-regulated Toll-like receptor 4 expression on DCs and inhibited pro-inflammatory cytokine secretion, including interleukin-12 and tumor necrosis factor-α. These cytokines were shown to stimulate CD4⁺ and CD8⁺ lymphocyte production in immunotherapy (136). Additionally, AFP also induced ThCD4⁺ lymphocytes to differentiate into Tregs, thus negatively regulating immunotherapy. Zhang et al., reported that AFP levels predicted ICI efficacy in treating advanced GC, i.e., high baseline AFP levels were associated with reduced disease control rate (DCR) during ICI treatment and also shortened PFS and OS (137). However, the exact mechanisms whereby AFP levels affect ICI efficacy in patients with GC remain unclear and further research is needed.

In 2004, LSD-1 was characterized as the first histone demethylase (138); its expression was significantly elevated in GC and it promoted GC proliferation and metastasis (139–142). LSD-1 also mediates epithelial-mesenchymal transition (143) in GC via H3K4me2 demethylation, thereby promoting drug resistance, disease recurrence, and disease invasion and metastasis (144). Shen et al., reported that LSD-1 deletion offset its immunosuppressive functions by reducing PD-L1 levels in exosomes and inhibiting its transport to other cancer cells, thereby restoring T cell killing functions in the GC microenvironment (145). Therefore, LSD-1 may function as a new immunotherapy target against GC, with the new LSD-1 inhibitor 5ac inhibiting mouse GC cell growth (146).

An increasing body of evidence now indicates that reprogrammed energy metabolism has critical roles in GC progression (147). Therefore, more in-depth research on the metabolic changes in the GC TME may provide new markers/therapeutic targets for neoadjuvant GC immunotherapy. Yang et al., used database resources to identify eight genes associated with fatty acid metabolism, which correlated with GC prognosis outcomes. The authors developed ‘FRAS’, a model whereby FRAS scores effectively identified patients with GC who were likely to benefit from anti-CTLA-4 antibody immunotherapy. Among the eight genes, RGS2 was significantly correlated with the TMB and CD8⁺ T cell infiltration in GC, suggesting that RGS2 may be a potential target for future immunotherapy strategies toward GC (148).

On the one hand, the results obtained in previous clinical trials of PD-1/PD-L1 and TMB as screening indicators of whether GC patients can undergo immunotherapy have not been satisfactory. On the other hand, MSI-H and EBV GC patient population only accounts for a small proportion of all GC patients. Therefore, finding new positive predictive biomarkers and exploring negative biomarkers (that identify subjects with clear lack of benefit from specific therapy) are necessary. Promoters are cis-regulatory elements found upstream of transcription start sites, with > 50% of human genes having multiple promoters, i.e., AP (149). AP allow transcription to start at different transcription start sites, which then produces different 5’ untranslated regions and first exons, thereby enhancing mRNA and protein subtype diversity (150). Sundar et al., reported that patients with metastatic GC with high AP use expressed lower CD8A, GZMA, and PFR1 levels (151) (cytolytic T-cell activity marker (152, 153)), indicating that AP use in metastatic GC was inversely correlated with anti-tumor immunity. Subsequent studies reported that patients with advanced GC with higher AP activity showed higher CD8A and PRF1 levels (CTL surface markers) and lower LAG-3 and TIM-3 levels, thus creating an inhibitory immune microenvironment and a mechanism for GC immune escape (154).

Homologous recombination (HR) is a highly accurate DNA repair mechanism (155). On one hand, HRD induces DNA repair defects, leading to the accumulation of more mutations and the generation of neoantigens, which enhances tumor response to ICI (156). On the other hand, HRD may promote the formation of an anti-tumor immune microenvironment by increasing tumor-infiltrating lymphocytes (TILs) (157), suggesting its potential as a biomarker for predicting immunotherapy response. The study by Fan et al. revealed that HRD-positive GC patients exhibited significantly longer OS following ICI treatment compared to other GC patients. More importantly, the study also demonstrated positive correlations between HRD status and both TMB-H and MSI-H. Additionally, HRD-positive GC patients showed increased CD8⁺ T cell infiltration after ICIs treatment (158). These findings suggest that HRD has the potential to serve as a predictive biomarker for immunotherapy efficacy in GC.