Evolving landscape of the gastric cancer immune microenvironment: From spatial architecture to precision biomarkers

Abstract

Gastric cancer remains a paradigm of therapeutic recalcitrance, driven by a complex ecosystem where therapeutic efficacy is dictated by the dynamic interplay between genomic instability and the tumor immune microenvironment. While biomarkers such as programmed death-ligand 1 expression and microsatellite instability currently guide therapeutic decisions, they offer only a static glimpse into a spatially and temporally evolving landscape. In this mini-review, we systematically delineate the co-evolution of spatial architecture, metabolic rewiring, and microbial interactions that orchestrate immune evasion in gastric cancer. We dissect how specific “cellular neighborhoods” – governed by the interplay between myofibroblastic cancer-associated fibroblasts and intratumoral microbiota like Fusobacterium nucleatum – construct physical and biological barriers to T-cell infiltration. Furthermore, we explore “invisible“ drivers of resistance, highlighting the synergistic potential of ferroptosis and pyroptosis in reshaping immunogenicity and the emerging role of the neuro-immune axis. Finally, we evaluate the clinical utility of next-generation biomarkers, ranging from tertiary lymphoid structure maturity and circulating tumor DNA molecular kinetics to artificial intelligence-driven “digital twins”. By integrating these multi-dimensional insights, we propose a strategic framework for precision immuno-oncology, transitioning from static profiling to holistic ecosystem engineering.

Key Words: Gastric cancer; Biomarkers; Tumor immune microenvironment; Tertiary lymphoid structures; Immune checkpoint inhibitors

Core Tip: Gastric cancer therapy must evolve beyond traditional approaches by focusing on the tumor immune microenvironment, which is shaped by spatial architecture, metabolic reprogramming, and microbiota interactions. The reliance on static biomarkers like programmed death-ligand 1 expression is shifting towards dynamic markers, such as tertiary lymphoid structure and circulating tumor DNA, to better predict treatment outcomes. A shift towards “ecosystem engineering” strategies, including vascular normalization, stromal reprogramming, and combination therapies like antibody-drug conjugates and chimeric antigen receptor T cells, is crucial for overcoming immune evasion and improving patient responses.

INTRODUCTION

The global burden and the transition from “therapeutic nihilism”

Gastric cancer (GC) remains a formidable global health challenge, ranking as the fifth most common malignancy and the fourth leading cause of cancer-related mortality worldwide[1]. The disease exhibits a stark geographical disparity, with an endemic burden in East Asia – particularly in China, Japan, and Korea – contrasted with a rising incidence of gastroesophageal junction tumors in Western populations[2]. This disparity is not merely statistical but reflects divergent immune-evolutionary trajectories. In East Asia, the high prevalence of chronic Helicobacter pylori infection acts as a persistent inflammatory stimulus, shaping a “pre-cancerous“ immune landscape characterized by the recruitment of myeloid-derived suppressor cells (MDSCs) and the induction of chronic T-cell exhaustion long before clinical malignancy emerges[3].

Despite refinements in surgical techniques and adjuvant regimens, the prognosis for advanced or metastatic GC has historically been dismal, with a median overall survival (OS) stagnating at 10-12 months for over a decade[4]. For years, the therapeutic landscape of advanced GC was characterized by a sense of “therapeutic nihilism“, a belief that the disease was too heterogeneous to be effectively targeted. Following the solitary success of the ToGA trial in 2010, which established Trastuzumab as the standard for human epidermal growth factor receptor 2 (HER2)-positive cases[5], the field endured a “decade of stagnation“ (2010-2020). During this period, numerous phase III trials targeting supposedly “pan-cancer“ oncogenic pathways, such as vascular endothelial growth factor (VEGF) (AVAGAST)[6], vascular endothelial growth factor receptor 2 (EXPAND), and mammalian target of rapamycin (mTOR) (GRANITE-1), failed to translate into survival benefits. This history of attrition underscored a fundamental biological reality: GC is not a monolithic entity driven by a single dominant oncogene, but a disease of profound inter-tumoral and intra-tumoral heterogeneity driven by complex evolutionary trajectories[3].

The “Renaissance“ of 2020-2025: Immunotherapy and beyond

The therapeutic landscape has undergone a seismic shift from 2020 to 2025, driven by the integration of immune checkpoint inhibitors (ICIs) and novel targeted agents[7]. However, a deep dive into the clinical data reveals a more nuanced reality than the headline results suggest.

The checkpoint era: Lessons from global vs regional divergence: The landmark CheckMate-649 trial fundamentally reshaped first-line standards by demonstrating that Nivolumab plus chemotherapy maintained a significant OS benefit compared to chemotherapy alone (hazard ratio = 0.79)[7]. Crucially, the 4-year follow-up data revealed a “long tail“ of survival, with approximately 17% of patients achieving long-term remission, suggesting that immunotherapy can induce a durable “reset“ of the immune rheostat in a specific subset of patients. Conversely, the failure of KEYNOTE-062 to show non-inferiority for Pembrolizumab monotherapy in the programmed death-ligand 1 (PD-L1) combined positive score (CPS) ≥ 1 population[8], and the regional differences observed in trials like ATTRACTION-4, highlight that baseline immune exhaustion states and subsequent-line therapies may vary across geographic populations[9].

Precision immunology and the “biomarker crisis“: The KEYNOTE-811 trial’s final analysis (2024) established a new paradigm for HER2-positive disease, showing that adding Pembrolizumab to Trastuzumab and chemotherapy significantly extended OS specifically in patients with PD-L1 CPS ≥ 1[10]. Simultaneously, the approval of Zolbetuximab (anti-CLDN18.2) in late 2024, based on the SPOTLIGHT and GLOW trials, opened a new therapeutic avenue for CLDN18.2-positive/HER2-negative patients, breaking the long-standing dependence on HER2 as the sole biomarker[11].

However, these triumphs have unveiled new clinical paradoxes. The objective response rate (ORR) for unselected populations plateaus at 20%-30%. A substantial subset of patients with “hot“ tumors (high PD-L1) exhibits primary resistance, while emerging resistance mechanisms often lead to rapid relapse. The current reliance on the CPS system is plagued by spatial heterogeneity – where a single biopsy may miss immune-active “hotspots“– and temporal evolution, as chemotherapy can alter the tumor immune microenvironment (TIME) phenotype after diagnosis[12].

Emerging modalities: Antibody-drug conjugates as microenvironment modifiers: The “Renaissance“ is further propelled by antibody-drug conjugates (ADCs) like Trastuzumab deruxtecan. Unlike traditional chemotherapy, the potent payload of ADCs induces immunogenic cell death (ICD), releasing danger-associated molecular patterns that activate dendritic cells and promote antigen cross-presentation[13]. This “bystander killing effect“ effectively turns “cold“ tumors “hot“, providing a rational blueprint for combining ADCs with programmed cell death protein 1 (PD-1) blockade to create a self-sustaining immunity cycle.

The paradigm shift: From tumor-centric to “ecosystem engineering“

The limitations of current strategies stem from a reductionist view that treats GC merely as a collection of somatic mutations. We are currently witnessing a paradigm shift towards an “ecosystem-centric“ perspective. The TIME in GC is not a passive scaffold but an active, structured co-conspirator[14,15].

The cellular sociology of GC: In this framework, the tumor is viewed through the lens of “cellular sociology“, where cancer cells engage in a dense network of inhibitory crosstalk with non-malignant components[14]. Recent breakthroughs in spatial transcriptomics and single-cell analysis have decoded this complexity, revealing that GC progression is governed by specific “cellular neighborhoods“[16]. For instance, the spatial exclusion of CD8+ T cells is orchestrated by distinct fibroblast subpopulations [myofibroblastic cancer-associated fibroblasts (myCAFs)] and reinforced by “invisible“ metabolic barriers such as ferroptosis-resistance and lactate accumulation[17].

The spatial architecture and biophysical barriers: Understanding the physical distribution of immune cells – specifically their proximity to the tumor nest – is as critical as quantifying their abundance. In “immune-excluded“ GC, T cells are abundant but restricted to the invasive margin, unable to penetrate tumor nests due to a physical barrier of dense collagen fibers and transforming growth factor beta (TGF-β) driven fibrosis[17]. Furthermore, the discovery of intratumoral microbiota (e.g., Fusobacterium nucleatum) and the neuro-immune axis adds new dimensions of regulation that act as “invisible“ drivers of resistance[18].

The philosophy of ecosystem engineering: This shift necessitates a transition from “killing cells“ to “engineering ecosystems“. Ecosystem engineering involves the holistic integration of genomic, spatial, and ecological insights to dismantle the TIME’s defenses[19]. By normalizing vessels to improve T-cell infiltration, neutralizing suppressive metabolites, and leveraging artificial intelligence (AI)-driven “digital twins“ to predict optimal combinations, we can finally breach the fortress of this recalcitrant disease.

Scope and roadmap of this review

To overcome current therapeutic bottlenecks, this Review provides a comprehensive synthesis of the evolving GC immune landscape. First, we delineate the spatial sociology of the TIME, moving beyond the “hot/cold“ dichotomy to dissect specific cell-cell interaction hubs and the trajectory of T-cell exhaustion. Second, we explore the molecular subtypes [Epstein-Barr virus (EBV), microsatellite instability (MSI), genomically stable (GS), chromosomal instability (CIN)] and how their unique genomic signatures dictate intrinsic sensitivity to immunotherapy. Third, we analyze the “invisible drivers“ of immune evasion, integrating metabolic reprogramming (ferroptosis/pyroptosis crosstalk), amino acid checkpoints, and epigenetic silencing. Fourth, we evaluate next-generation predictive biomarkers, ranging from tertiary lymphoid structure (TLS) maturity to dynamic circulating tumor DNA (ctDNA) molecular kinetics. Finally, we propose a strategic framework for precision immuno-oncology, utilizing AI-driven digital twins and mechanism-guided combination strategies to extend survival benefits to the majority of patients.

METHODOLOGY

We conducted a multi-stage literature search across PubMed, Web of Science, and EMBASE databases, covering the period from January 2010 to December 2025.

Search strategy

Our search utilized a combination of Medical Subject Headings and keywords, including “gastric cancer“, “tumor immune microenvironment“, “spatial transcriptomics“, “metabolic reprogramming“, and “immunotherapy resistance“.

Inclusion criteria

We prioritized peer-reviewed original research, meta-analyses, and high-impact clinical trial results (phase II/III) that provided mechanistic insights into immune evasion or validated predictive biomarkers.

Data integration

Multi-dimensional data from The Cancer Genome Atlas (TCGA) and single-cell RNA sequencing cohorts were cross-referenced to ensure the reproducibility of the “cellular neighborhood“ and “metabolic checkpoint“ theories discussed herein.

THE SPATIAL AND CELLULAR ARCHITECTURE OF THE GC TIME

The TIME of GC is increasingly recognized as a non-random, highly organized spatial system[20]. The clinical failure of many ICIs is not merely a consequence of “missing“ immune cells, but rather a failure of the spatial contexture – the precise arrangement and functional state of cells within specific histological niches. Recent advances in multiplex immunohistochemistry and spatial transcriptomics have enabled the classification of GC into three distinct spatial phenotypes, each governed by unique molecular and biophysical rules (Table 1)[21].

Table 1 Comparative spatial and molecular metrics of gastric cancer tumor immune microenvironment phenotypes.

Metric	Immune-inflamed (“hot“)	Immune-excluded (“cold“)	Immune-desert (“cold“)
Predominant subtype	Epstein-Barr virus+, microsatellite instability-high	Genomically stable (genomically stable)	Chromosomal Instability
Dominant chemokine axis	CXCL9/10/11-CXCR3	CXCL12-CXCR4	Low/absent
B-cell/TLS organization	Mature TLS (secondary follicles) with gastric cancer reaction; high B-cell diversity	Immature TLS or disorganized B-cell aggregates at the margin	Absent or rare B-cell clusters; lack of TLS formation
Stroma density (collagen)	Low/moderate	High (cross-linked lysyl oxidase-like 2 + fibers)	Variable
T-cell factor 1 + T-cell niche	Present in mature TLS	Restricted to invasive margin 15	Absent
Response to anti-programmed cell death protein 1	High (50%-70%)	Low (primary resistance)	Negligible

CXCL: C-X-C motif chemokine ligand; CXCR: C-X-C motif chemokine receptor; TLS: Tertiary lymphoid structure.

The three spatial phenotypes: A mechanistic taxonomy

The immune-inflamed phenotype (the “hot“ ecosystem): (1) TLS as immuno-sustaining hubs: In “hot“ GC, lymphocytes are often organized into TLS, which serve as sites for local antigen presentation and B-cell maturation[22,23]. We propose a three-stage maturation model in GC including lymphoid aggregates, characterized by unstructured T and B cell clusters; primary follicles, featuring an organized B-cell zone without a germinal center; and secondary (mature) follicles, containing CD21+ follicular dendritic cells and GZMB+ effector cells. Our analysis indicates that only mature TLS correlate with significant OS benefit, as they provide a protected “micro-niche“ that prevents the premature exhaustion of progenitor T cells[23]; and (2) The TCF1+ progenitor niche: Within these inflamed regions, a critical subset of T cells known as progenitor exhausted T cells (TPEX), marked by TCF1 expression, resides in protective niches[24]. These cells retain stem-like proliferative capacity and serve as the primary reservoir for the explosive T-cell expansion observed during successful anti-PD-1 treatment.

The immune-excluded phenotype (the “fortress“ model): This is the most prevalent phenotype in GC, where T cells are localized at the invasive margin but fail to penetrate the tumor nests[25].

The biophysical shield and mechanotransduction: The myCAFs orchestrate the deposition of a dense, cross-linked extracellular matrix[26]. In GC, high expression of lysyl oxidase-like 2 increases the stiffness of this matrix from 1.5 kPa to over 20 kPa[27]. This high-stiffness environment activates the YAP/TAZ mechanotransduction pathway, which inhibits the migratory velocity of T cells, reducing it from 10 μm/minute to less than 2 μm/minute[28].

The SPP1+ macrophage-cancer-associated fibroblast axis: High-resolution mapping identifies SPP1+ (Osteopontin) macrophages that co-localize with myCAFs[29]. These cells secrete SPP1, which binds to CD44 on fibroblasts to promote collagen cross-linking, further reinforcing the architectural resistance[30]. Although SPP1+ macrophage-cancer-associated fibroblast (CAF) interactions can be observed across multiple GC phenotypes in fibrotic contexts, current spatial transcriptomic evidence suggests that this axis is most prominently enriched and functionally dominant in immune-excluded tumors, particularly within the GS subtype.

The CCL2+ fibroblast–STAT3-activated macrophage axis: Recent high-resolution spatial transcriptomic mapping has further subdivided the immune-excluded GC microenvironment into discrete regional compartments. Notably, CCL2+ fibroblasts have been identified as key organizers of an immunosuppressive cellular neighborhood by actively recruiting and spatially retaining monocytes and macrophages through CCL2-CCR2 signaling. Within this niche, sustained STAT3 activation in macrophages reinforces an anti-inflammatory and tumor-supportive transcriptional program, thereby amplifying immune suppression and stromal remodeling. This CCL2+ fibroblast-STAT3-activated macrophage axis provides a representative mechanistic example of how specific stromal-myeloid interactions stabilize immune-excluded architectures in GC.

The immune-desert phenotype (immunological ignorance): (1) Antigenic silencing: Desert tumors often exhibit a “cold“ genomic profile with low tumor mutational burden (TMB) and frequent loss of heterozygosity at the human leukocyte antigen (HLA) locus, rendering them invisible to the adaptive immune system; and (2) The Wnt/βCatenin exclusion axis: Constitutive activation of the Wnt pathway in GC cells leads to the transcriptional repression of CCL4, a chemokine essential for the recruitment of conventional type 1 dendritic cells. Without resident DCs to sense danger signals, the adaptive immune system never initiates a T-cell response[31].

The B-cell landscape (from germinal center dynamics to plasma cell effector functions)

The intratumoral B-cell evolutionary trajectory: Within the mature TLS of GC, B cells undergo a highly regulated evolutionary process similar to that observed in secondary lymphoid organs[22].

Recruitment and aggregation: The process begins with the secretion of CXCL13 by T follicular helper cells and certain CXCL13+ CAFs, which recruits CXCR5+ B cells to the peritumoral stroma[23]. Germinal center reaction: In mature TLS (secondary follicles), B cells undergo somatic hypermutation and class-switch recombination. This “in situ“ maturation allows for the generation of high-affinity B-cell receptors specifically tailored to tumor-associated antigens[22].

Plasma cell differentiation: Mature B cells eventually differentiate into CD138+ plasma cells. In GC, these plasma cells are often found at the tumor-stroma interface, secreting large quantities of tumor-specific immunoglobulin G (IgG) and IgA.

Independent prognostic contribution and effector mechanisms: The presence of B cells and mature TLS has been identified as an independent predictor of favorable prognosis, often outperforming CD8+ T-cell density alone. This independent contribution is mediated through three primary axes: (1) Antibody-dependent cellular cytotoxicity: Tumor-specific antibodies produced by intratumoral plasma cells “coat“ cancer cells, facilitating their destruction by natural killer (NK) cells and macrophages via Fc receptor engagement. This is particularly critical in HER2+ and CLDN18.2+ GC subtypes[32]; (2) Professional antigen presentation: B cells within TLS function as highly efficient antigen-presenting cells. By processing tumor antigens and presenting them via major histocompatibility complex class II (MHC-II) to T follicular helper cells, B cells sustain the survival and expansion of the T-cell compartment within the hostile metabolic milieu[22]; and (3) Cytokine orchestration: Specialized B-cell subsets secrete pro-inflammatory cytokines such as interleukin (IL)-12 and interferon-γ (IFN-γ), which reinforce the type I immune response and prevent the premature exhaustion of the TPEX population[33].

B-cell induced spatial remodeling: Furthermore, B-cell rich TLS act as spatial “anchors“ for adaptive immunity. Tumors where T cells are spatially clustered around mature B-cell follicles exhibit a significantly higher ORR to ICIs compared to those with disorganized lymphocytic infiltration. This “proximity effect“ suggests that B cells provide essential survival signals that maintain T-cell fitness in the face of suppressive myCAFs and metabolic competition[34].

The myeloid landscape: Metabolic and functional suppressionmyeloid cells in GC are the primary source of the "metabolic brake"

Nutrient competition: MDSCs overexpress arginase-1, which hydrolyzes L-arginine – an amino acid required for T-cell receptor (TCR) signaling – thereby inducing T-cell anergy[35].

The acidic barrier: Accumulation of Lactate (up to 30 mmol/L) lowers the pH to 6.0-6.5, inhibiting the Ca²⁺-nuclear factor of activated T cells (NFAT) signaling pathway in T cells and leading to a state of reversible paralysis[36].

The neuro-immune axis: The "invisible" neural control of TIME

To achieve a holistic understanding of the GC ecosystem, we must move beyond immune and stromal cells to include the peripheral nervous system. GC is highly “neurotropic“, and the dense innervation within the stomach provides a direct regulatory highway for immune evasion.

Adrenergic signaling and T-cell paralysis: Gastric tumors frequently exhibit high nerve density (neo-neurogenesis), where sympathetic nerve fibers release norepinephrine (NE) into the TIME. This norepinephrine binds to β2-adrenergic receptors (β2-AR) on CD8+ T cells, triggering an intra-cellular cyclic adenosine monophosphate (cAMP)-protein kinase A signaling cascade that directly inhibits the production of granzyme B and IFN-γ. This “neural brake“ represents a spatial regulatory layer that functions independently of classical immune checkpoints[37].

Cholinergic regulation of macrophages: Conversely, parasympathetic (vagal) signaling via acetylcholine interacts with the α7 nicotinic acetylcholine receptor on tumor-associated macrophages (TAMs). This interaction promotes an M2-like polarization and suppresses the nuclear factor kappa B-mediated pro-inflammatory response, effectively creating a “cholinergic anti-inflammatory pathway“ that the tumor co-opts to maintain a suppressive environment[38].

The "vascular gatekeeper": Angiogenesis and endothelial anergy

The spatial exclusion of T cells often begins at the point of entry: The tumor vasculature. In GC, the vascular network is not just a nutrient supply line but a selective filter.

Endothelial anergy and adhesion failure: In the “immune-desert“ and “excluded“ phenotypes, tumor-associated endothelial cells (TECs) exhibit a state of “anergy“. Driven by high levels of VEGF and basic fibroblast growth factor, TECs downregulate essential adhesion molecules such as intercellular adhesion molecule 1, vascular cell adhesion molecule 1, and E-selectin[39]. Without these “docking stations“, circulating T cells cannot undergo the rolling and adhesion required for extravasation, regardless of how many T cells are present in the systemic circulation.

Vascular remodeling by myCAFs: The myCAFs secrete VEGF-C and angiopoietin-2, which promote the formation of leaky, tortuous vessels[40]. The resulting high interstitial fluid pressure creates a physical outward flow that opposes the inward migration of immune cells, functionally sealing off the tumor nests from the host’s adaptive immune system.

The microbiota-immune crosstalk: Intratumoral "hidden residents"

The GC TIME is not sterile; it is an ecological niche for specific microbial colonizers that shape the local immune tone.

Fusobacterium nucleatum as an immune shield: Beyond its role in colorectal cancer, Fusobacterium nucleatum is enriched in advanced GC. It utilizes its surface protein fibroblast activation protein (FAP) 2 to bind to T-cell immunoreceptor with Ig and immunoreceptor tyrosine-based inhibitory motif domains on T cells and NK cells[41]. This binding delivers a potent inhibitory signal that mimics the PD-1/PD-L1 axis, providing a “microbial checkpoint“ that protects tumor cells from immune surveillance.

The Helicobacter pylori legacy: While Helicobacter pylori is often lost during the later stages of gastric atrophy and malignancy, the “epigenetic memory“ it leaves behind is profound. Chronic Helicobacter pylori infection induces a permanent state of Th17-mediated inflammation, which eventually recruits MDSCs and creates a “fertile soil“ for subsequent immune-excluded growth[42].

The complexity of these interactions suggests that the GC TIME is not merely a collection of isolated cells, but a sophisticated, multi-layered “interactome“ where neural, microbial, and vascular signals converge to dictate the fate of the anti-tumor immune response. To synthesize the cellular cross-talk mechanisms detailed throughout front part, we have categorized the primary ligand-receptor axes and their functional consequences in the following Table 2.

Table 2 High-resolution interactome of the gastric cancer microenvironment.

Interaction category	Signaling axis (ligand-receptor)	Key cellular sources	Spatial phenotype context	Functional consequence in gastric cancer
Microbial-immune	Fibroblast activation protein 2-T-cell immunoreceptor with immunoglobulin and immunoreceptor tyrosine-based inhibitory motif domains	Fusobacterium nucleatum to natural killer/T cells	Inflamed/excluded	Direct inhibition of lymphocyte cytotoxicity; “microbial checkpoint“
Neuro-immune	Norepinephrine-β2-AR	Sympathetic nerves to CD8+ T cells	Global (stress-induced)	Inhibition of granzyme B/interferon-γ production via cyclic adenosine monophosphate-protein kinase A signaling
Neuro-myeloid	Acetylcholine-alpha-7 nicotinic acetylcholine receptor	Vagus nerve to TAMs	Excluded/desert	Promotion of M2-like polarization; cholinergic anti-inflammatory shield
Fibro-immune	CXCL12-CXCR4	The myCAFs to T cells	Immune-excluded	Peritumoral T-cell sequestration; prevention of nest infiltration
Fibro-immune	SPP1-CD44	TAMs to myCAFs	Excluded	Strengthening of the extracellular matrix “fortress“ and collagen cross-linking
Vascular-immune	VEGF-A-VEGFR2	Tumor cells to Endothelium	Desert/excluded	Downregulation of intercellular adhesion molecule 1/vascular cell adhesion molecule 1; induction of endothelial anergy
Immune-immune	CXCL13-CXCR5	T follicular helper cells to B cells	Immune-inflamed	Recruitment of B cells to tertiary lymphoid structure; orchestration of germinal center reactions

CXCL: C-X-C motif chemokine ligand; CXCR: C-X-C motif chemokine receptor; myCAFs: Myofibroblastic cancer-associated fibroblasts; TAMs: Tumor-associated macrophages; VEGF-A: Vascular endothelial growth factor A; VEGFR2: Vascular endothelial growth factor receptor 2.

MOLECULAR SUBTYPES AND IMMUNE LANDSCAPES

The profound heterogeneity of the GC immune microenvironment is deeply rooted in tumor-intrinsic genomic and epigenetic alterations. The landmark classification by TCGA network established four distinct molecular subtypes: EBV-positive, microsatellite instability (MSI), genomically stable (GS), and CIN[43]. Each subtype possesses a unique “immune signature“, dictating its intrinsic sensitivity to immunotherapy and shifting the biomarker paradigm from a “one-size-fits-all“ approach to subtype-specific strategies[44].

EBV-positive GC: The "super-responder" phenotype

Accounting for approximately 9% of cases, EBV-positive GC represents the most immunologically active subset, consistently exhibiting the immune-inflamed (“hot“) phenotype.

Viral mimicry and innate recruitment: The expression of viral latent proteins (e.g., EBNA1) and non-coding RNAs (EBERs) serves as potent pathogen-associated molecular patterns. These signals trigger robust type I interferon pathways, facilitating the massive recruitment of NK cells and CD8+ cytotoxic T lymphocytes (CTLs) into the tumor parenchyma.

The 9p24.1 amplification strategy: A defining feature of this subtype is the focal amplification of the 9p24.1 locus, which harbors CD274 (PD-L1), PDCD1 LG2 (PD-L2), and JAK2. This genetic alteration leads to constitutive, high-level expression of PD-L1 on tumor cells, independent of inflammatory signaling. Consequently, these patients exhibit dramatic and durable responses to PD-1 blockade, often achieving exceptional ORR[45].

MSI GC: Neoantigen-driven adaptive immunity

The MSI subtype (approximately 22% of cases) results from mismatch repair (MMR) deficiency, leading to a hypermutated state characterized by high TMB.

Frameshift mutations and neoantigen load: The accumulation of insertion-deletion (indel) mutations generates aberrant frameshift peptides that are recognized as “non-self“ by the host immune system. This elicits robust T-cell infiltration and the frequent formation of mature TLS[22].

The “inflamed-but-inhibited“ paradox: To survive this intense immune pressure, MSI tumors upregulate multiple checkpoints, including PD-L1, LAG-3, and Indoleamine 2,3-dioxygenase 1 (IDO1), as adaptive resistance mechanisms. This state explains why MSI patients derive significant standard-of-care benefit from ICIs, such as the dual checkpoint blockade of Nivolumab plus Ipilimumab[46].

GS GC: The fortress of exclusion

The GS subtype (approximately 20% of cases) largely overlaps with the diffuse histological type and presents the most significant challenge for immunotherapy.

Structural barriers and E-cadherin loss: Defined by CDH1 loss and RHOA mutations, GS tumors exhibit a dispersive growth pattern. Immunologically, they are characterized by the highest enrichment of TGF-β signaling and pro-angiogenic signatures[47].

Mechanisms of stromal dominance: The activation of FAP+ myCAFs in GS tumors creates the physical “immune-excluded“ barrier discussed in front part, precluding T-cell entry[17]. Furthermore, a low neoantigen load renders these tumors “invisible“, necessitating strategies that modulate the stroma, such as targeting CLDN18.2 or TGF-β.

CIN GC: Aneuploidy and evasion

The CIN subtype (approximately 50% of cases) is characterized by marked aneuploidy and receptor tyrosine kinase amplifications.

The aneuploidy paradox: High levels of somatic copy number alterations in CIN tumors correlate with reduced immune infiltration[48]. This evasion likely involves the loss of heterozygosity of HLA machinery, preventing effective antigen presentation.

Targeting oncogenic remodeling: CIN tumors frequently harbor amplifications of HER2, VEGFA, or fibroblast growth factor receptor 2. Evidence from the KEYNOTE-811 trial suggests that oncogene inhibition (e.g., Trastuzumab) can favorably remodel the TIME by inducing ICD, potentially converting an immune-desert phenotype into an Inflamed one[7].

The divergent immunological trajectories of these four subtypes – ranging from the hyper-inflamed EBV+ niche to the fibrotic fortress of GS tumors – highlight the necessity of a genome-informed approach to immunotherapy. To provide a comprehensive overview of this relationship, the genomic drivers, spatial phenotypes, and dominant evasion mechanisms of each TCGA subtype are synthesized in Table 3.

Table 3 Integrative landscape of molecular subtypes and immune phenotypes.

Molecular subtype	Spatial phenotype	Key genomic drivers	Dominant immune mechanism	Recommended strategy
Epstein-Barr virus+	Inflamed (hot)	9p24.1 amp, PIK3CA mut	Viral pathogen-associated molecular patterns; constitutive programmed death-ligand 1	Programmed cell death protein 1 monotherapy
Microsatellite instability	Inflamed (hot)	Deficient mismatch repair, high tumor mutational burden	Neoantigen-driven recruitment	Dual checkpoint blockade
Genomically stable	Excluded (cold)	CDH1/RHOA mut, CLDN18.2	TGF-β fibrosis; myofibroblastic cancer-associated fibroblasts barriers	Anti-CLDN18.2/TGF-β inhibitors
Chromosomal instability	Desert/variable	TP53 mut, receptor tyrosine kinase amp, somatic copy number alterations	Human leukocyte antigen loss of heterozygosity; myeloid-driven suppression	Targeted + immune checkpoint inhibitors (e.g., human epidermal growth factor receptor 2-directed)

TGF-β: Transforming growth factor beta.

METABOLIC AND EPIGENETIC REWIRING: THE DEEP DRIVERS OF EVASION

Beyond physical barriers and receptor-ligand interactions, the GC TIME is characterized by a hostile metabolic milieu and profound epigenetic dysregulation. These “invisible“ mechanisms create a suppressive ecosystem that functionally paralyzes infiltrating immune cells, rendering them unresponsive even when physical contact with tumor cells is established[49].

The Warburg effect, lactate accumulation, and ionic checkpoints

GC cells exhibit a classical Warburg effect, prioritizing aerobic glycolysis over oxidative phosphorylation even under normoxic conditions[50]. This metabolic reprogramming is not merely a byproduct of rapid proliferation but a strategic weapon against host immunity.

The glucose tug-of-war and mTOR inhibition: Tumor cells typically overexpress the glucose transporter glucose transporter 1, outcompeting T cells for limited glucose uptake. Since effector T cells rely heavily on glycolysis for rapid expansion and the production of effector cytokines like IFN-γ, this glucose deprivation leads to metabolic insufficiency[51]. This state triggers the mTORC1-AMP-activated protein kinase axis in T cells, where the activation of AMP-activated protein kinase and subsequent inhibition of mTORC1 signaling force lymphocytes into a state of metabolic anergy or p53-mediated apoptosis[52].

Lactate-mediated epigenetic remodeling: The byproduct of hyper-glycolysis, lactate, is exported into the TIME via monocarboxylate transporter 4, resulting in local acidification (pH 6.0-6.5). Beyond inhibiting the nuclear translocation of NFAT in CTLs, high lactate concentrations promote histone lactylation in macrophages. This specific epigenetic modification drives the transcription of pro-tumorigenic genes, facilitating the polarization of TAMs toward an immunosuppressive M2-like phenotype through GPR132 activation[53,54].

Lactate as an epigenetic replicator (the histone lactylation mechanism): Beyond its role in physical acidification and the inhibition of ionic flux in T cells, lactate functions as a sophisticated precursor for a novel epigenetic modification: Histone lysine lactylation. This discovery has fundamentally shifted our understanding of how the “Warburg effect“ directly programs the immunosuppressive state of the GC ecosystem[54].

The metabolic-epigenetic bridge: In the hypoxic and glycolytic core of gastric tumors, intracellular lactate levels in TAMs rise significantly. This lactate is utilized by the p300 acetyltransferase (which also functions as a lactyltransferase) to add lactyl groups to lysine residues on histone H3 (e.g., histone H3 lysine 18 lactylation)[55].

Beyond p300, however, histone lactylation is increasingly recognized as a process regulated by a broader metabolic-epigenetic network rather than a single enzyme. Lactate availability itself is shaped by upstream metabolic enzymes such as lactate dehydrogenase and by monocarboxylate transporters that govern intracellular lactate flux. In addition, other acyltransferases with overlapping substrate permissiveness, as well as histone deacylases, may indirectly contribute to the dynamic deposition and removal of lactylation marks. Importantly, the precise enzyme hierarchy and regulatory specificity of histone lactylation in GC remain incompletely defined and warrant further mechanistic investigation.

Transcriptional reprogramming of TAMs: Unlike histone acetylation, which typically peaks early during inflammation, histone lactylation occurs as a “late-phase“ epigenetic mark[56]. In GC, this modification specifically drives the transcription of M2-like genes, such as Arg1, Mrc1 (encoding CD206), and various pro-fibrotic factors. This ensures that once a macrophage enters the high-lactate tumor neighborhood, it is epigenetically “locked“ into a suppressive phenotype that promotes tissue remodeling and T-cell exclusion[57].

The feedback loop of ecosystem evasion: The lactylation-driven expression of Arg1 further depletes L-arginine from the microenvironment[58], leading to the TCR zeta-chain defects discussed in front part. Thus, lactate-mediated epigenetic rewiring acts as a central node, linking the tumor’s glycolytic rate to the functional paralysis of both the innate and adaptive immune arms.

Ionic checkpoints: Potassium and calcium signaling in the acidic niche

The metabolic hostility of the GC TIME also manifests through altered ionic gradients that function as “checkpoints“ for T-cell activation.

Potassium efflux and stemness: High rates of necrosis in GC release high concentrations of extracellular potassium (K+). This ionic imbalance inhibits the uptake of nutrients and prevents the effector differentiation of T cells, potentially contributing to the maintenance of the “quiescent“ TPEX state in TLS, albeit at the cost of immediate cytolytic function[59].

Acid-sensing ion channels: The acidic pH (6.0-6.5) triggered by lactate accumulation activates acid-sensing ion channels on infiltrating CTLs. This activation interferes with the calcium (Ca²⁺) oscillations required for NFAT translocation, effectively creating an “ionic shield“ that renders T cells unresponsive even when they achieve physical TCR engagement with the tumor cell[60].

Amino acid checkpoints and the integrated stress response

Amino acids are critical rheostats for immune cell fate, and GC cells create “nutritional checkpoints“ by depleting essential amino acids in the local environment.

The indoleamine 2,3-dioxygenase 1-kynurenine-aryl hydrocarbon receptor axis: IDO1 is frequently upregulated in GC, particularly in the MSI subtype. IDO1 catalyzes the rate-limiting degradation of tryptophan into kynurenine (Kyn)[61]. This process exerts a “double-hit“ on immunity: Tryptophan depletion triggers the stress-response kinase general control nonderepressible 2 in T cells, leading to growth arrest. Concurrently, Kyn acts as an endogenous ligand for the aryl hydrocarbon receptor, which promotes the differentiation of regulatory T cells while suppressing the cytolytic potential of CD8+ T cells[62,63].

Arginine depletion and the TCR zeta-chain defect: MDSCs in the GC stroma secrete high levels of arginase-1, hydrolyzing L-arginine. Since T cells lack the enzymes to synthesize arginine de novo, its depletion leads to the systemic downregulation of the CD3 zeta chain within the TCR complex. This structural defect impairs signal transduction upon antigen recognition, effectively “decoupling“ the T cell from its target[64,65].

Hypoxia, adenosine, and the ectonucleotidase brake

As tumors outgrow their blood supply, hypoxia becomes a hallmark of the GC microenvironment, primarily driven by the stabilization of hypoxia-inducible factor-1 alpha.

The CD39-CD73 synergy: Hypoxia induces the expression of ectonucleotidases CD39 and CD73 on tumor cells and endothelial cells. These enzymes sequentially hydrolyze extracellular adenosine triphosphate – normally a pro-inflammatory “danger signal“– into adenosine, a potent immunosuppressor[66].

A2a receptor signaling and cAMP accumulation: Adenosine binds to the high-affinity A2a receptor on T cells and NK cells. This binding triggers an intracellular accumulation of cAMP, which activates protein kinase A and blocks the secretion of pro-inflammatory cytokines such as IL-2 and tumor necrosis factor-alpha. This pathway represents a major resistance mechanism to PD-1 blockade in refractory GC[67].

Ferroptosis and pyroptosis: The emerging role of ICD

We must explore “invisible“ drivers of resistance, highlighting the synergistic potential of ferroptosis and pyroptosis in reshaping immunogenicity

Ferroptosis-resistance as an immune barrier: GC cells often develop resistance to ferroptosis – a form of iron-dependent programmed cell death – by upregulating glutathione peroxidase 4 or solute carrier family 7 member 11. This resistance prevents the release of danger-associated molecular patterns that would otherwise activate dendritic cells[68].

Pyroptosis and the “cold-to-hot“ transition: In contrast, inducing pyroptosis – a highly pro-inflammatory form of cell death – via the activation of Gasdermin D/E can remodel the TIME. Dying tumor cells release high concentrations of IL-1β and high-mobility group box-1, which recruit effector lymphocytes and overcome the “immune-desert“ phenotype[69].

Epigenetic silencing and viral mimicry

Epigenetic alterations, including DNA hypermethylation and histone modification, contribute to “hard“ immune evasion in GC.

MHC-I downregulation via methylation: GC cells transcriptionally silence the antigen processing and presentation machinery, including HLA-A/B/C and B2M, via promoter hypermethylation. This renders tumor cells functionally invisible to CTLs, facilitating immune evasion regardless of the tumor’s mutational load[70].

Reactivating endogenous retroviruses: A novel therapeutic vulnerability involves the reactivation of silenced endogenous retroviruses. Treatment with DNA methyltransferase inhibitors (e.g., decitabine) can induce the accumulation of cytoplasmic double-stranded RNA from these endogenous retroviruses. This double-stranded RNA is sensed by pattern recognition receptors like MDA5 and RIG-I, mimicking a viral infection and triggering a robust type I interferon response. This “viral mimicry“ can turn a “cold“ tumor “hot“ by recruiting immune cells and upregulating MHC molecules[71].

Integrative metabolic signatures of immune evasion

The transition from a “tumor-centric“ metabolic view to an “ecosystem-centric“ one reveals that metabolic waste products are, in fact, sophisticated signaling molecules. As we have delineated, the competition for glucose and amino acids creates a “zero-sum game“ where the tumor cells’ survival is directly coupled to the immunological paralysis of the host. To provide a holistic view of this metabolic-epigenetic axis, the key regulatory enzymes, their downstream signaling consequences, and the resulting immune checkpoints are synthesized in Table 4.

Table 4 Integrative metabolic checkpoints and their epigenetic consequences in gastric cancer.

Metabolic axis	Key mediators/enzymes	Effect on the tumor immune microenvironment	Epigenetic and signaling impact	Potential intervention
Glucose/Lactate	Glucose transporter 1, MCT4, LDHA	Local acidification (pH 6.0-6.5); glucose starvation	Histone lactylation (histone H3 lysine 18 lactylation); inhibition of nuclear factor of activated T cells translocation	MCT4 inhibitors; LDH inhibitors
Tryptophan	IDO1, tryptophan 2,3-dioxygenase	Tryptophan depletion; kynurenine accumulation	Activation of Aryl hydrocarbon receptor; Induction of suppressive regulatory T cells via general control nonderepressible 2	IDO1 inhibitors (e.g., Epacadostat)
Arginine	Arg-1	L-arginine starvation	Downregulation of the CD3 zeta-chain; T-cell anergy	Arg-1 inhibitors
Adenosine	CD39, CD73	Adenosine triphosphate to adenosine conversion	Activation of A2aR; intracellular cyclic adenosine monophosphate/protein kinase A signaling	A2aR antagonists (e.g., Ciforadenant)
Lipid/iron	Glutathione peroxidase 4, solute carrier family 7 member 11	Ferroptosis-resistance	Prevention of danger-associated molecular patterns release (suppression of immunogenic cell death)	Ferroptosis inducers (e.g., RAS-selective lethal 3)
Epigenetic/viral	DNA methyltransferase 1, EZH2	Silencing of endogenous retroviruses and human leukocyte antigen genes	Major histocompatibility complex class I transcriptional silencing; viral mimicry failure	DNA methyltransferase inhibitors; EZH2 inhibitors

Arg-1: Arginase-1; A2aR: A2a receptor; EZH2: Enhancer of zeste homolog 2; IDO1: Indoleamine 2,3-dioxygenase 1; LDH: Lactate dehydrogenase; MCT4 Monocarboxylate transporter 4.

THE EVOLUTION OF PREDICTIVE BIOMARKERS: FROM STATIC HISTOLOGY TO DYNAMIC PRECISION

The rapid expansion of the GC therapeutic armamentarium has outpaced the development of robust predictive biomarkers. While PD-L1 CPS and MSI status remain the regulatory gold standards, they are increasingly viewed as static snapshots that fail to capture the spatial and temporal plasticity of the TIME[72].

The PD-L1 paradox: Quantity vs spatial context

PD-L1 expression is the most widely used biomarker, yet its predictive value is plagued by significant biological and technical noise.

Spatial and temporal heterogeneity: In GC, PD-L1 expression can vary between the primary tumor and metastatic lymph nodes in up to 40% of cases[73]. Furthermore, “interferon-driven“ (adaptive) PD-L1 expression is transient. A biopsy taken at diagnosis may not reflect the immune landscape after several cycles of chemotherapy, which can either induce “immunogenic priming“ or select for PD-L1-negative resistant clones.

The “bystander“ problem: The CPS system aggregates PD-L1 expression on both tumor cells and infiltrating immune cells. However, high-resolution mapping shows that PD-L1 on macrophages may have a more potent suppressive effect on T-cell activation than PD-L1 on tumor cells, yet current scoring systems do not distinguish between these cellular sources[74].

TLS: The "immune school" maturity

As discussed in front part, the mere presence of T cells is insufficient. The presence and maturity of TLS are emerging as superior predictors of ICI response.

The maturation grading system: TLS should be evaluated as functional units rather than simple cell clusters. Mature TLS (containing germinal centers and CD21+ follicular dendritic cells) act as local hubs for B-cell clonal expansion and affinity maturation. Patients with a high density of mature TLS exhibit a significantly higher “immune fitness“, leading to durable responses to PD-1 blockade even in tumors with moderate PD-L1 expression[75].

Spatial geometry: The distance between T cells and TLS – the “immune-spatial distance“– is a novel metric. Tumors where T cells are clustered within 50 μm of a mature TLS are more likely to exhibit an “inflamed“ phenotype than those with diffuse, disorganized infiltration[76].

Limitations and standardization challenges: Despite the growing evidence supporting TLS maturity as a promising biomarker in GC, several important limitations should be acknowledged. First, there is currently no universally accepted definition or grading system for TLS maturity, resulting in variability across studies and potential inter-observer inconsistency. Second, sampling bias represents a significant challenge, as TLS are spatially heterogeneous and may be underrepresented in small biopsy specimens. Moreover, it remains unresolved whether mature TLS act as direct causal drivers of improved immunotherapy response or instead reflect an underlying state of baseline immune fitness. TLS formation may therefore represent a permissive immune ecosystem rather than an independent determinant of treatment efficacy. Finally, technical standardization – including reproducible scoring criteria, spatial quantification methods, and clinically relevant cut-off values – will be essential before TLS maturity can be reliably implemented in routine clinical decision-making.

Liquid biopsies and ctDNA: Capturing the evolution of resistance

Molecular residual disease: The ctDNA provides a real-time window into tumor burden. In the adjuvant setting, patients who remain ctDNA-positive after surgery have an extraordinarily high risk of recurrence. Integrating ctDNA monitoring with immunotherapy (as seen in the PANDA and GERCOR trials) allows for the early identification of “molecular relapse“ months before radiological evidence appears[77].

Clonal evolution monitoring: Longitudinal ctDNA sequencing can detect the emergence of resistance mutations (e.g., JAK1/2 mutations or B2M loss) during treatment, allowing clinicians to switch strategies before clinical progression occurs[78].

AI-driven “digital twins“ and spatial intelligence

The future lies in the integration of multi-modal data through AI.

Deep learning for spatial features: Recent advances in AI have enabled the extraction of thousands of spatial features from routine hematoxylin and eosin or multiplex-stained slides, uncovering patterns of immune exclusion and stromal organization that are not readily discernible by conventional pathology. In this context, so-called “digital twins“ refer to computational representations of a patient’s TIME that integrate spatial, cellular, and molecular information.

At present, however, AI-driven digital twins should be regarded as exploratory and hypothesis-generating research tools rather than clinically validated decision-making systems. Most existing applications in GC and other solid tumors are based on retrospective analyses or pilot-level studies, aiming to stratify immune phenotypes, infer mechanisms of resistance, or simulate potential therapeutic perturbations in silico. While these models may suggest whether certain combination strategies (e.g., PD-1 blockade combined with anti-angiogenic therapy) could theoretically alleviate stromal or vascular barriers, prospective validation and regulatory standardization are still lacking.

Nevertheless, as spatial transcriptomics, high-throughput imaging, and longitudinal clinical datasets continue to mature, AI-driven spatial intelligence platforms may evolve into powerful translational frameworks. Their future value will likely lie in treatment hypothesis prioritization, trial design optimization, and patient stratification, rather than direct treatment selection at the current stage.

OVERCOMING THERAPEUTIC BOTTLENECKS: TOWARDS ECOSYSTEM ENGINEERING

The elucidated mechanisms of immune evasion – spatial exclusion, metabolic suppression, and genomic invisibility – provide a rational blueprint for designing next-generation combination therapies. The goal is no longer just to “release the brake“ on T cells (using ICIs alone) but to fundamentally remodel the hostile ecosystem that restricts them[15].

Vascular normalization: Re-opening the "highway" of infiltration

The chaotic and leaky vasculature of GC, driven by VEGF overexpression, creates a hypoxic, high-pressure barrier that excludes T cells and accumulates immunosuppressive metabolites like adenosine and lactate[79].

The low-dose anti-angiogenic paradox: High doses of anti-VEGF agents can lead to excessive pruning of vessels, exacerbating hypoxia[80]. However, “vascular normalization“ using low-dose TKIs (e.g., Lenvatinib, Regorafenib, or Apatinib) induces a structural and functional maturation of the endothelial barrier[81].

Mechanistic synergy: This normalization improves oxygenation, thereby downregulating the hypoxia-inducible factor-1 alpha/adenosine axis, and provides a stable “highway“ for CD8+ T cells to infiltrate the tumor core[82]. Trials like REGONIVO have exemplified this synergy, showing remarkable response rates in refractory GC by converting “cold“ tumors “hot“ through this vascular-immune remodeling[83].

Stroma normalization: Dismantling the fibrotic fortress

For the majority of GC patients with the “immune-excluded“ phenotype (particularly the GS subtype), the priority is to dismantle the physical barrier constructed by FAP+ myCAFs.

Targeting the TGF-β signaling loop: TGF-β is the master regulator of stromal stiffness and T-cell exclusion. Combining ICIs with TGF-β inhibitors (e.g., Bintrafusp alfa) or DKK1 inhibitors (e.g., DKN-01) aims to reprogram pro-tumorigenic CAFs into a quiescent state, effectively “opening the door“ for lymphocytic entry[84].

ADCs and the bystander effect: ADCs like Trastuzumab deruxtecan revolutionize this space by inducing ICD. The potent topoisomerase I inhibitor payload released from dying cells can penetrate the dense stroma (the “bystander effect“), killing neighboring cells and releasing a cascade of neoantigens that prime new T-cell responses[85].

Next-generation immune engagement: Chimeric antigen receptor T cells and bispecifics

To overcome the limitations of passive ICI therapy, active engagement of both innate and adaptive immunity is required.

The claudin-18.2 frontier: Zolbetuximab works primarily through antibody-dependent cellular cytotoxicity and complement-dependent cytotoxicity, recruiting NK cells and macrophages to the tumor site. Combining this with chemotherapy leverages the innate immune system to breach the tumor defense, paving the way for adaptive immunity[86].

Bispecific antibodies: Emerging agents like Cadonilimab (PD-1/cytotoxic T-lymphocyte-associated protein 4 bispecific) allow for simultaneous blockade of the priming phase (cytotoxic T-lymphocyte-associated protein 4) and the effector phase (PD-1) within the same tissue site[87]. This dual-targeting strategy has shown superior survival compared to chemotherapy alone in 1 L GC settings (COMPASSION-15 trial)[88].

The clinical success of such agents underscores a broader shift: We are moving from targeting single proteins to engineering entire immune ecosystems. To provide a comprehensive roadmap for this transition, Table 5 synthesizes the mechanism-guided strategies designed to dismantle the biophysical and metabolic defenses of the GC TIME, transforming them into self-sustaining cycles of anti-tumor immunity.

Table 5 Mechanism-guided ecosystem engineering strategies in gastric cancer.

Engineering objective	Target mechanism/axis	Therapeutic strategy	Biological rationale	Representative agents/trials
Breaching physical barriers	Myofibroblastic cancer-associated fibroblasts-lysyl oxidase-like 2/TGF-β	TGF-β traps; Fibroblast activation protein-targeted chimeric antigen receptor T cells	Dismantling the collagen “fortress“ to enable T-cell infiltration	Vactosertib; AB122
Normalizing the niche	VEGF-A/VEGFR2	Anti-angiogenesis + immune checkpoint inhibitor	Reversing endothelial anergy; reducing high interstitial fluid pressure	Regorafenib + Nivolumab (LEAP-005)
Metabolic rescue	Lactate/histone H3 lysine 18 lactylation	Monocarboxylate transporter 4 or lactate dehydrogenase inhibitors	Reversing histone lactylation to “unlock“ suppressive TAMs	Preclinical/early phase
Metabolic rescue	Adenosine/A2aR	CD73 inhibitors; A2aR antagonists	Abrogating cyclic adenosine monophosphate -mediated paralysis in the hypoxic core	Oleclumab; Ciforadenant
Priming immunity	ADC-induced immunogenic cell death	Trastuzumab deruxtecan; CLDN18.2-ADC	Triggering danger-associated molecular patterns release (calreticulin/high-mobility group box-1) to activate conventional type 1 dendritic cells	DESTINY-gastric03
Dual checkpoint synergy	Programmed cell death protein 1 and cytotoxic T-lymphocyte-associated protein 4	Bispecific antibodies	Simultaneous modulation of priming (lymph node) and effector (tumor) phases	Cadonilimab (AK104)
Neural modulation	Β2-AR signaling	Repurposed β-blockers	Preventing stress-induced catecholamine inhibition of cytotoxic T lymphocytes fitness	Propranolol (phase II)

ADC: Antibody-drug conjugate; A2aR: A2a receptor; TGF-β: Transforming growth factor beta; VEGF-A: Vascular endothelial growth factor A; VEGFR2: Vascular endothelial growth factor receptor 2.

FUTURE PERSPECTIVES: TOWARDS DIGITAL AND DYNAMIC ONCOLOGY

As we stand on the precipice of a new era, the integration of high-dimensional data and AI will define the next decade of GC research. The transition from “biomarker discovery“ to “ecosystem engineering“ necessitates a radical shift in how we process biological information.

AI and the rise of "spatial intelligence"

The immense information contained in a standard hematoxylin and eosin slide is largely untapped by human eyes. Future diagnostic workflows will likely transcend simple cell counts to embrace “spatial intelligence“.

Computational staining: Deep learning algorithms can now segment individual cells and quantify the “spatial mixing“ of lymphocytes and tumor cells. These “computational stains“ can predict MSI status or PD-L1 expression directly from morphological patterns with high accuracy, often outperforming traditional IHC[89].

Automated immune scoring: We envision a clinical pipeline where an AI pre-screen of all biopsy slides provides an instant “immune score“. This score integrates spatial architecture and cellular density to guide treatment decisions – such as whether to prioritize vascular normalization or dual checkpoint blockade – before molecular sequencing results are even available.

The "digital twin" of the immune microenvironment

The ultimate goal of precision immuno-oncology is the development of patient-specific “digital twins“.

Multi-modal integration: These computational models will integrate genomic (TMB), transcriptomic (Interferon score), and metabolic (Hypoxia score) data into a virtual representation of the patient’s specific TIME.

In silico simulation: By simulating how a patient’s specific ecosystem would react to various perturbations – for instance, “What if we add a VEGF inhibitor to this excluded phenotype? “– these models can predict the optimal combination strategy in silico. This approach will minimize trial-and-error in the clinic and accelerate the arrival of truly personalized medicine.

From static snapshots to dynamic ecosystem monitoring

The failure of current strategies often stems from treating GC as a static entity. We must transition to a model of “dynamic oncology“.

The liquid biopsy revolution: Longitudinal analysis of ctDNA and circulating immune cells will allow for the real-time tracking of the TIME’s evolution. Detecting the early emergence of resistance clones (e.g., JAK1/2 or B2M mutations) will enable adaptive therapy switches before clinical deterioration.

Ecological management: We have moved from simple histology to molecular subtypes, from cell counts to spatial architecture, and finally, from static biopsies to dynamic tracking.

CONCLUSION

The journey to conquer GC has evolved from a “search for the magic bullet“ to the “management of a complex ecosystem“. We now recognize that the failure of immunotherapy in the majority of patients is not a failure of the drugs themselves, but a failure to address the specific spatial, metabolic, and genomic barriers that protect the tumor. By dismantling these defenses layer by layer – normalizing the vasculature, neutralizing suppressive metabolites, and recruiting effectors through multi-modal integration – we can finally breach the fortress of this recalcitrant disease. The future of GC therapy lies in holistic ecosystem engineering, ensuring that the survival benefit of the “long tail“ is extended to the majority of patients.