Animal models have a crucial role in studying the biological behavior and molecular mechanisms of carcinogenesis and evaluating drug effectiveness. Cancer cell lines are often transplanted into immunodeficient mice to generate a model with easy manipulation and accessibility. However, this model loses the phenotypic and genetic heterogeneity, as well as the tumor microenvironment (TME) of the original tumors. A main advantage of PDXs in cancer research is that the tumor's histopathological architecture, cancer cells and surrounding stromal cells are largely preserved. Evidence suggests that the characteristics of PDX models are highly similar to those of parental tumors and their response to anti-cancer drugs is also similar to that of patients. PDXs were first reported in 1969 when the Danish scholar Rygaard transplanted human colon adenocarcinoma masses into nude mice (11). As with other tumor PDX models, immunodeficient mice have been used to establish PDX models of GC. The immunocompromised mouse strains widely used for PDX models are as follows: i) Nude mice, which lack a thymus and are unable to produce T cells, resulting in defective adaptive immune responses (12); ii) severe combined immune deficiency (SCID) mice, which lack both functional T and B lymphocytes (13). Human tumor engraftment efficiency is higher in SCID mice than in nude mice (14). Furthermore, SCID/beige mice, in addition to lacking T and B cells, have a severe deficiency of natural killer (NK) cell function, so the engraftment rate of human cancer cells is enhanced in SCID/beige mice compared to SCID mice (15,16); iii) nonobese diabetic (NOD)/SCID, interleukin 2 receptor (IL2R)-γ^null (NSG or NOG) mice and NOD/SCID Jak3^null (NOJ), in which T-, B- and NK-cell activity are completely absent, may markedly improve the efficiency of xenotransplantation (17-19); iv) BALB/c Rag-2^null/IL2R-γ^null and Rag-2^null/Jak3^null, in which macrophage-mediated phagocytosis of human cells may be reduced (20-22); and v) nude Rag-2^null/Jak3^null mice, established by crossing BALB/c Rag-2^null/Jak3^null mice and BALB/c nude mice, all serve as powerful tools for evaluating human tumor-host interactions (23). Choi et al (24) successfully established 15 GC PDX models with passaging to maintain tumors in nude or NOG mice (24.2%, 15/62); the genetic and histological characteristics of the primary tumors and PDX models were highly consistent. Karalis et al (25) established 23 PDX models from Western patients with GC with various ethnic backgrounds. In theory, highly immunosuppressed mouse strains may allow for higher tumor engraftment rates, but tumors implanted into NSG (16%) and nude (21%) mice had a similar engraftment rate, possibly because, as the immunodeficiency level increases in the recipient mice, the likelihood of developing B-cell lymphoma also increases, and the presence of B-cell lymphoma hinders the generation of solid tumor PDXs. Corso et al (10) generated a wide, multilevel platform of GC models, including 100 PDXs, organoids and primary cell lines. This PDX platform was the widest in an academic institution, and included all GC histologic and molecular types identified by The Cancer Genome Atlas. They also conducted a transcriptomic analysis of PDXs to identify a microsatellite instable (MSI) signature with the potential to assist in the development of precision medicine for GC (10).

The TME includes the extracellular matrix and stromal cells, which include cancer-associated fibroblasts (CAFs), immune cells, pro-inflammatory cells and other components. The interaction between the TME and tumor cells has a prominent role in tumor progression, metastasis and therapeutic response. However, during xenograft growth, human stromal cells originally present in patient-derived tumors are gradually replaced by murine counterparts, which may hinder the analysis of tumor-stroma interaction in humans, as certain cytokines from mouse stroma may not have an impact on human carcinoma cells in PDX models (26). To overcome this limitation in PDX models, humanized mouse models have been generated. Researchers engrafted the human immune system and human tumor tissues in animal models, allowing the human immune system to reconstitute in the immunodeficient mice with patient tumor engraftment. Improved humanized mouse models have also been developed, such as i) the human peripheral blood lymphocytes (Hu-PBL) model, ii) the Hu-CD34⁺ model and iii) the bone marrow-liver-thymus (BLT) mice model. In 1988, Mosier et al (27) established the first of these, the Hu-PBL model, by injecting peripheral blood mononuclear cells (PBMCs) intraperitoneally (i.p.) or intravenously (i.v.) into SCID mice. After the transplantation of PBMC, human CD3⁺ T cells could be detected within one week, and ~50% of human CD45⁺ cells could be detected in the peripheral blood of mice after approximately four weeks. The advantage of this approach was that PBMCs were readily available and easy to manipulate, but the transplanted mice developed lethal graft vs. host disease (GVHD) within 2-3 weeks caused by the human T cells attacking mouse tissue, limiting the model's utility. More importantly, these mice are incapable of mounting adaptive immune responses with their engrafted immune systems. With the Hu-CD34⁺ model, immunodeficient mice were first given sublethal irradiation to deplete mouse hematopoietic stem cells (HSCs). Then, human CD34⁺ HSCs from human umbilical cord blood, adult bone morrow, granulocyte colony-stimulating factor-mobilized PBMCs or fetal liver was injected i.v. or i.p. into newborn or adult immunodeficient mice. In this model, the CD34⁺ HSCs can differentiate into various mature blood cells, such as T cells, B cells, NK cells or myeloid cells. However, human-derived T-cell development was low due to the lack of a human thymus. To address this problem, Lan et al (28) established the Hu-BLT model in 2006, in which immunodeficient mice (NOD/SCID) were also treated with sublethal (2-3 Gy) whole-body irradiation, after which human fetal liver and thymus tissue were transplanted into the subrenal capsule of adult immunodeficient mice, and autologous CD34⁺ human HSCs from the same fetal liver or bone marrow were injected i.v. into the mice, resulting in a stable model 3-4 months after transplantation. This method achieved a significant reduction in GVHD symptoms, but the limited donor source and the complexity of establishment have restricted the use of this model to study the human immune microenvironment and infectious diseases. To our knowledge, GC PDX models using humanized mice have not been reported thus far. However, a cell-derived xenograft (CDX) model of GC using humanized mice has been used to evaluate the biological roles of Zinc Finger Protein 64 (ZFP64) in GC for nab-paclitaxel resistance (29). In this study, 3-week-old NSG mice were injected with cord blood-derived CD34⁺HSCs; subsequently, human GC HGC-27 cells were subcutaneously implanted in the humanized mice. The integration of tumor progression analysis and humanized mouse models offered a novel approach for evaluating tumor cell drug resistance, as well as the role of the immune system in response to chemotherapy.

PDX models can be established by orthotopic or heterotopic (e.g. subcutaneous, intravenous or intraperitoneal injection) implantation. Heterotopic engraftments with subcutaneous injection of patient-derived cancer tissues have been widely used, as it is easier to manipulate and to monitor tumor growth. However, compared with heterotopic engraftments, orthotopic models are more clinically relevant and more suitable for the interpretation of the mechanisms of cancer metastasis, development and progression (30). In 1993, Furukawa et al (31) established the first patient-derived orthotopic nude mouse models of GC. In total, tissues from 36 patients with advanced GC were transplanted orthotopically into nude mice, yielding 20 tumors (56%, 20/36). GC commonly metastasizes to the liver, lymph nodes, peritoneum, lung and bone, either through direct invasion or via distant metastases by lymphatic, hematogenous or intraperitoneal spread (32). Hepatic metastases, observed in ~50% of patients with GC, are the most common distant metastases, with a survival rate of 4% at five years (33). In one study, the tumor tissues of five patients with clinical liver metastases also developed liver metastases in nude mice (31). Existing orthotopic implantation methods of GC are used to establish orthotopic stomach tumor models for studying cancer biology or organ metastasis. However, only certain types of malignant material have been successfully transplanted, such as single-cell suspensions or a firm fragment of tumor. In 2021, the Jackson Laboratory developed a novel, completely closed, orthotopic GC animal model in NSG mice using diverse tumor materials, such as soft tissues, semi-liquids or culture derivatives (34). This novel method overcame the weaknesses of the existing methodologies that supported using only single-cell suspensions or a firm tumor fragment. Although their approach required advanced surgical techniques, this procedure can generate an appropriate animal model for numerous research purposes, including exploration of biomarker functions, testing the efficacy of anti-tumor drugs and utilizing GC organoids.

PDXs have emerged as valuable models for predicting drug responses in GC treatment. However, their limitations, including being time-consuming and having a lower engraftment rate, hinder their clinical application in patients with advanced GC due to rapid disease progression. Thus, there is an urgent need for a rapid and dependable alternative approach to evaluating drug sensitivity. The hollow fiber assay has been proposed as a preliminary screening tool for anticancer agents to identify sensitive tumor cell lines (35), but this approach did not have high similarity with clinical results. In 2018, Shanghai LIDE Biotech Co., Ltd. developed a rapid drug screening model named OncoVee^® MiniPDX (36). In this model, hollow fiber capsules were filled with patient-derived GC tumors and then implanted subcutaneously into mice, and they are permitted to grow for 7 days. The system has shown high similarity between compound responses of miniPDX and corresponding PDX. Several study groups have also reported the use of miniPDX models for the treatment of GC and found that drug screening through this system can provide significant benefits for patients with GC (37-41). MiniPDX in combination with next-generation sequencing (NGS) can be used to rapidly evaluate drug sensitivity in patients with GC and identify key genetic mutations (39). A single-arm, open-label phase I clinical study utilizing miniPDX models to evaluate HER2-negative medium-advanced GC/gastroesophageal junction cancer chemotherapy regimens and targeted agents resulted in favorable antitumor activity and safety outcomes (41). In the future, it will presumably be possible to co-transfer fresh cancer tissues with autoimmune cells (PBMCs or tumor-infiltrating lymphocytes) from the same patient with GC into minicapsules and engraft them into immunodeficient mice, which can capture the human TME to a maximum extent, allowing for the evaluation of the efficacy of immunotherapy drugs.

The presence of stromal cells in the TME, such as endothelial cells, immune cells and CAFs, contributes significantly to tumorigenesis, metastasis and treatment resistance (57). Tumors expressing programmed cell death-ligand 1 (PD-L1) interact with CD8⁺ cytotoxic T lymphocytes (CTLs) expressing programmed cell death protein 1 (PD-1) to inhibit CTL proliferation and survival, leading to tumor evasion of immune surveillance, which in turn leads to increased proliferation of cancer cells (58,59). More than 40% of patients with GC have tumors that express PD-L1 (60). However, only 22% (8/36) of patients with GC have had an overall response to anti-PD-1 antibody pembrolizumab (61). Therefore, improved preclinical models are needed that can predict the efficacy of immune therapies to enhance the survival of patients with GC. In most organoid models, these crucial components of the TME are absent. Given the lack of immune cells, a co-culture model system was established to overcome this drawback. Co-culturing cancer organoids with immune cells or fibroblasts provided a valuable tool for investigating the TME and molecular interactions in cancer treatment. Current checkpoint blockade immunotherapy has shown remarkable efficacy in unblocking T cells that are negatively controlled, leading to T cell-mediated anticancer responses. Several studies of cancer precision medicine have utilized co-culturing of GC PDOs with immune cells in combination with checkpoint blockade inhibitors. Chakrabarti et al (62) established a GC patient-derived organoids/immune cell co-culturing system. Before co-culturing organoids with CTLs, researchers pulsed antigen-presenting dendritic cells (DCs) with tumor antigens and then cultured autologous CTLs with the DCs to increase cytolytic activity and proliferation of tumor-specific T lymphocytes. Using this autologous organoid/immune cell co-culture system, they found that HER2 expression may promote immune evasion in GC that was mediated by PD-L1. This co-culturing strategy provided a suitable preclinical model for studying the effect of anti-HER2-targeted therapy in combination with anti-PD-L1 immunotherapy for patients with GC (63). In addition, this system was used to investigate the differentiation and immunosuppressive function of myeloid-derived suppressor cells (MDSCs) (64). In another study, PDO/immune cell co-cultures demonstrated that gastric organoids expressing PD-L1 were not responsive to nivolumab in vitro when PMN-MDSCs were present. However, when PMN-MDSCs were depleted in these co-cultures, the organoids became sensitive to anti-PD-1/PD-L1-induced cancer cell death (64), suggesting that MDSCs with immunosuppressive function had an important role in the TME of GC. These studies have provided valuable insight into predicting alternative drug regimens and studying the GC microenvironment using GC PDOs. Thus, advances in co-culture organoid techniques may yield additional clinical treatment strategies using targeted therapy and immunotherapy. This platform can also benefit patients with GC by generating individualized therapy data more rapidly than animal models.

Single-cell RNA sequencing (scRNA-seq) is a valuable approach that enables analysis of cancer expression profiles at the single-cell level, allowing identification and characterization of unique subpopulations with specific biological behaviors (65). Numerous studies examining the heterogeneity of tumor cells and comprehensive dynamics in the TME have been performed using scRNA-seq in GC (66-68). Jiang et al (66) were the first to evaluate the heterogeneity of GC primary tumors and metastases in different organs at the single-cell level, demonstrating the characteristics of different organ-tropism metastases of GC and identifying effective therapeutic targets. Li et al (67) utilized scRNA-seq to study the role of CAFs in the GC TME, including their classification, function, origin, interaction with other cell subsets and spatial distribution in different pathological types. They found distinct roles of CAFs in regulating various aspects of TME biology, including immune modulation, invasion, migration and angiogenesis. Of note, their study demonstrated that a specific type of CAFs, known as extracellular matrix CAFs, exhibited an enhanced chemotaxis ability for attracting M2 macrophages and their presence was associated with poor prognosis of patients with GC (67). GC commonly metastasizes to lymph nodes. Qian et al (68) conducted a comprehensive analysis of the transcriptome profiles of GC tissues of primary tumors and metastatic lymph nodes (MLNs) at the single-cell level. They discovered that dysfunctional neutrophil polarization and maturation had a vital role in lymph node metastasis of GC. In addition, secreted phosphoprotein 1 (SPP1) signaling, an immune checkpoint, can be activated in MLNs. Hence, targeting the disordered neutrophils and SPP1 signaling may be novel strategies to treat and prevent lymph node metastasis of GC.

Among the platforms to study GC, patient-derived organoids have emerged as a promising system for investigating tumor behavior and the influence of TME components. To elucidate the stepwise progression of the disease from dysplasia to different stages of adenocarcinoma, including well-differentiated, poorly-differentiated and metastatic, Lu et al (69) generated a series of genetically-edited gastric organoids in mice. Through scRNA-seq analyses and functional studies, they identified an interaction between tumor cells and macrophages, facilitated by integrin α6/β4 and fibronectin 1, which had an important role in promoting GC progression and metastasis (69). To study the extent to which GC organoid in vitro culture affects transcriptional lineage states or cellular proportions compared with primary cancer cells in vivo, Kumar et al (56) performed an overall analysis of cell states between primary GC cells and organoids by scRNA-seq and found similarities and differences between primary GC tissues and organoids. Similar to primary tumors, tumor PDO epithelial cells showed upregulation of cancer-associated modules and GC-related genes compared with normal PDO epithelial cells (56). In addition, they found differences between primary tumors and PDOs; for instance, stromal and epithelial cell clusters were significantly enriched in PDOs, while lymphoid and plasma cell clusters were depleted (56). A gene-expression comparison between PDO and primary samples found that plasma cells exhibited the most significant differences in gene expression profile in PDO models, whereas epithelial signatures were relatively more conserved. Altogether, gastric organoids are soon expected to use combinatorial single-cell methods, including epigenetic, genetic and transcriptional analyses, and spatial context, to further enhance our understanding of the mechanisms underlying GC development.