Research progress in the construction and application of immunotherapy resistance models

Abstract

The establishment and application of immunotherapy resistance models provide crucial tools and conceptual frameworks for better understanding resistance mechanisms, simulating and solving the immunotherapeutic drug resistance challenges faced by clinics, and validating drug efficacy and safety. This review comprehensively summarizes the current construction methods and applications of immunotherapeutic resistance models, while discussing the characteristic features and underlying resistance mechanisms of these models. With the aim of identifying strategies for constructing tumor immunotherapy resistance models that more closely recapitulate clinical scenarios, we provide mechanistic insights to facilitate the enhancement of immunotherapy sensitivity and advance the development of novel antitumor agents.

Supplementary Information

The online version contains supplementary material available at 10.1007/s00262-025-04221-x.

Numerous therapeutic regimens, including chemotherapy, molecularly targeted therapies, and immunotherapeutic agents, have been clinically implemented in oncology practice, yielding modest improvements in survival rates among cancer patients [1]. However, prolonged therapeutic administration frequently induces acquired drug resistance in malignant cells, ultimately resulting in ineffective treatment, tumor relapse, and metastatic progression [2]. Therefore, there is an urgent need to clarify the mechanisms of drug resistance, develop new drugs, and explore combinations of different drugs to provide more therapeutic options and improve therapeutic efficacy.

In vitro and in vivo models play a pivotal role in biomedical research, as they not only faithfully replicate the clinical manifestations of human diseases, but also establish essential experimental frameworks for elucidating disease mechanisms, identifying novel therapeutic targets, and advancing drug development [3, 4]. Current research efforts are increasingly employing cell-based models and animal-based models to elucidate the mechanisms of drug resistance, identify drug resistance-associated targets, and discover novel therapeutic strategies to overcome or mitigate treatment resistance. These models are indispensable tools in human research on drug resistance [5, 6].

In recent years, immune checkpoint inhibitors represented by PD-1/PD-L1 monoclonal antibodies have demonstrated remarkable progress in the treatment of various malignancies, conferring long-term survival benefits to a subset of patients. However, akin to other antitumor modalities such as chemotherapy and targeted therapies, immunotherapy is similarly confronted with the challenge of therapy resistance [7]. To address this pervasive and intricate issue, preclinical immunotherapy resistance models have become indispensable. In addition, the establishment of tumor resistance models at both in vitro and in vivo levels has become a critical pathway for preclinical drug target screening and pharmacodynamic evaluation, facilitating the identification of novel resistance-associated targets and advancing the development of next-generation antitumor therapeutics. This article delineates methodologies for constructing immunotherapeutic resistance models, including in vitro resistance models, human-derived resistant tissue xenografts, in vivo repeated dosing in tumor-bearing animals, modulation of immunotherapy resistance-associated gene expression, and fecal microbiota transplantation (FMT) from resistant patients. Additionally, a clinically applicable predictive model is constructed by deciphering the tumor microenvironment using bioinformatics and computational algorithms. The article also briefly exemplifies the application of each model and summarizes immunotherapy resistance mechanisms and model characteristics. These established models are proving essential for elucidating the dynamic interplay between resistant malignancies and the immune system, conducting preclinical antitumor drug evaluation, and facilitating the discovery of predictive biomarkers for immunotherapy, thereby providing critical experimental platforms for overcoming clinical resistance mechanisms.

In vitro construction of an immune drug resistance model

In vitro immunoresistant models can be established through acquisition of immune-resistant tissues, in vitro pharmacological induction, or by utilizing specific cell lines with inherent low immunogenicity to evade T cell-mediated cytotoxicity. This model serves as an important and indispensable tool for investigating immune resistance mechanisms, identifying immune resistance-associated therapeutic targets, and exploring combined therapeutic strategies.

Tumor samples from patients with acquired resistance to PD-1 blockade exhibit hallmarks of upregulated interferon-γ (IFN-γ)-related gene expression [8]. Sustained IFN-γ stimulation of cancer cells can confer resistance to immune checkpoint inhibitors (ICIs) therapy [9]. Compared to tumor models derived from IFN-γ-unstimulated cancer cells in mice, KP and LLC1 tumor models generated from cancer cells preconditioned with low-dose IFN-γ for 3–4 weeks in vitro exhibited attenuated responses to ICIs and unrestrained tumor growth. Treatment with IFN-γ in vitro can induce resistance to immune checkpoint inhibitors in vivo. This mechanism involves the upregulation of interferon-stimulated genes, resulting in a loss of MHC-I expression and enabling tumors to escape immune recognition [10]. Yan et al. [11] successfully established a therapy-resistant B16-F10 melanoma mouse model through IFN-γ-stimulated cells. Utilizing this model, they identified a novel prion-like chemical inducer, SAP, which effectively overcomes adaptive immune resistance in resistant melanoma-bearing mice, offering a new strategic approach to address immunotherapy resistance in melanoma. Acidosis is a hallmark feature of solid tumors [12]. Within the tumor microenvironment (TME), PD-L1 expression on cancer cells is predominantly regulated by IFN-γ [13]. In MC38 tumor-bearing mice preconditioned with IFN-γ, combined NaHCO3 and anti-PD-L1 treatment slowed tumor growth compared to anti-PD-L1 monotherapy, accompanied by increased CD3 + T cell infiltration. Research reveals that acidosis significantly enhances the expression of PD-L1 induced by IFN-γ on aggressive cancer cells. This PD-L1 engages with PD-1 on T cells, thereby suppressing their cytotoxic function and ultimately promoting tumor immune evasion. Neutralizing the acidic environment of TME and reducing tumor PD-L1 expression may pave the way for novel strategies to overcome resistance to immunotherapy [14].

A PD-1 antibody-resistant BT-549 breast cancer cell line was established through coculture with primary T cells under adherent conditions, supplemented with repeated cycles (5 passages) of murine PD-1 antibody RMP1-14. Compared to parental cells, these resistant cells exhibited upregulated expression of the receptor tyrosine kinase Tyro3 and cyclin-dependent kinase 9 (CDK9). In xenograft models, Tyro3 or CDK9 knockdown BT-549 tumors reversed PD-1 antibody resistance, whereas Tyro3/CDK9 overexpression or activation correlated with anti-PD-1/PD-L1 therapy resistance in triple-negative breast cancer patients. These findings suggest that combining CDK9 or Tyro3 inhibitors with immune checkpoint inhibitors may enhance immunotherapy response rates in patients [15].

Based on TME characteristics and response to immunotherapy, tumors are classified as ‘hot’ or ‘cold’ ‘Hot’ tumors exhibit high tumor mutational burden (TMB), enhanced immunogenicity, and dense infiltration of immune effector cells, which correlates with improved clinical responses to ICIs. In contrast, ‘cold’ tumors are characterized by low TMB, sparse immune effector cell infiltration, and an immunosuppressive TME dominated by regulatory T cells (Tregs) or myeloid-derived suppressor cells (MDSCs), resulting in primary resistance or acquired refractoriness to ICIs therapy [16]. The glioblastoma cell line CT2A [17], triple-negative breast cancer cell line 4T1 [18], melanoma cell line B16-F10 [19], and colorectal cancer cell line CT26 [20] are recognized as ‘cold’ tumor models, commonly utilized to investigate mechanisms of tumor immune evasion and therapy resistance. Li et al. [21] identified that elevated expression of Membrane Spanning 4-Domains A4A (MS4A4A) may serve as a biomarker for immunotherapy nonresponsiveness in cancer patients. In PD-1 antibody-resistant CT26 and B16-F10 tumor-bearing mice, combination therapy with MS4A4A-targeting monoclonal antibodies and PD-1 blockade restored sensitivity to anti-PD-1 treatment. These findings suggest that targeting MS4A4A represents a promising strategy to enhance the efficacy of immune checkpoint inhibitors and a novel immunotherapeutic approach against cancer. Patients frequently use dexamethasone to alleviate tumor-related cerebral edema. Iorgulescu JB et al. [22] evaluated the effect of dexamethasone on the PD-1 blockade response in an immunotherapy-resistant CT-2A glioblastoma mouse model, which realistically reflects the response to PD-1 antibodies in glioblastoma patients. The study results revealed that dexamethasone limits the effectiveness of immune checkpoint blockade (ICB) therapy. It is imperative to recognize that the administration of dexamethasone for tumor-associated edema may compromise the clinical benefits of immunotherapy.

Antigen loss and defective antigen presentation are among the underlying mechanisms driving resistance to ICIs. Given that NK cell cytotoxicity is independent of antigen recognition, activating NK cells via STING agonist-loaded nanoparticles emerges as a promising strategy to overcome ICI resistance in cancer immunotherapy [23]. The B16-F10 melanoma lung metastasis model is a model with PD-1 treatment resistance. Mice were constructed with the B16-F10 melanoma lung metastasis model by tail vein injection of B16-F10-luc2 cells. Anti-PD-1 monotherapies did not show antitumor effects in this model, whereas the combination of STING agonist-loaded nanoparticles with PD-1 blockade demonstrated synergistic tumor suppression and mitigated resistance to anti-PD-1 therapy [24]. Lipid nanoparticles loaded with STING agonists could be promising candidates for combining with anti-PD-1 in the treatment of drug-resistant tumors.

Immuno-resistant tissues are one of the pathways by which drug-resistant cells are acquired. The drug-resistant cell line underwent five rounds of selection in vivo with continuous anti-PD-1 mAb treatment. Following these rounds, anti-PD-1 immunotherapy was ineffective in controlling MC38 tumor growth. The phenotypic resistance characteristics were retained by the MC38 cells after several subculture expansions in vitro. This indicates that the acquired resistance resulted from inherited alterations in the tumor cells [25]. Compared to the parental MC38 cell line, therapy-resistant MC38 tumor cells exhibited significantly elevated expression of glycoprotein nonmetastatic melanoma protein B (GPNMB). CRISPR-Cas9-mediated knockout of GPNMB in resistant MC38 cells demonstrated that loss of GPNMB expression abrogated tumor cell resistance to PD-1 blockade therapy in vivo. GPNMB may represent an attractive novel therapeutic target for tracking adaptive resistance to therapeutic PD-1 blockade.

Organoids are three-dimensional in vitro culture models derived from tumor patient tissues or stem cells. Tumor organoids can better preserve the molecular architecture, pathological characteristics, genetic profiles of parental tumor tissues, and play an important role in tumor research, especially as highly reliable experimental models in the fields of drug screening, disease modeling, and drug resistance research [26]. XUE et al. [27] established a patient-derived organoid (PDO) model based on an air–liquid interface system using surgically resected tissues from patients with advanced clear cell renal cell carcinoma (ccRCC). This model successfully recapitulated the immune exhaustion status characteristic of ccRCC and was employed to evaluate the efficacy of the immunotherapeutic agent toripalimab in enhancing treatment response. Toripalimab treatment increased the CD8 + T cell/CD4 + T cell ratio and promoted tumor cell apoptosis in the PDO model, suggesting that toripalimab mitigates CD8 + T cell exhaustion and reverses the immune-depleted TME. This study validates tumor organoids as a promising and reliable preclinical model for predicting immunotherapy responses in renal carcinoma. A novel model of spontaneous peritoneal metastases in microsatellite instability-high colorectal cancer resistant to ICB was established by engrafting PDO into the cecum of humanized mice. Peritoneal metastases in this model are insensitive to ICB and are clinically highly relevant. Peritoneal metastases in this model are insensitive to ICB and are clinically highly relevant. It is plausible that an immunosuppressive ascitic milieu, characterized by high levels of IL-10 and TGF-β isoforms, mediates the observed resistance to ICB in peritoneal metastases [28].

Immunotherapy-resistant cells, generated through in vitro induction or isolated from immunotherapy-resistant tissues, are easily scalable and suitable for large-scale expansion and high-throughput drug screening. Their short experimental cycle facilitates rapid validation of molecular mechanisms underlying immune resistance. Two-dimensional (2D) monolayer cultures of immunotherapy-resistant cells lack critical TME components such as fibroblasts, endothelial cells, and macrophages, and fail to recapitulate the in vivo tumor three-dimensional (3D) architecture, extracellular matrix, hypoxia, or metabolic heterogeneity, thereby limiting comprehensive investigation into immune resistance mechanisms. Long-term passaging of tumor cell lines may lead to the loss of heterogeneity inherent to the original tumor, resulting in discrepancies between cellular immune resistance phenotypes and clinical patient response profiles [29, 30]. These disadvantages prompted the development of models that more closely mimic in vivo conditions. PDO models exhibit high establishment efficiency, typically requiring days to weeks for development. PDO models from immunotherapy-resistant patients retain tumor spatial architecture, cellular heterogeneity, and TME components, closely recapitulating the authentic therapy-resistant microenvironment. Current advancements include the establishment of PDO coculture models with cancer-associated fibroblasts or immune cells. These refined systems not only facilitate the study of dynamic cross talk between malignant cells and key tumor microenvironment components but also demonstrate high accuracy in predicting immunotherapy sensitivity in patients [31]. PDO models require precise and high-quality culture conditions and lack vascularization structures and a fully functional immune system [32]. While cell-based models are well suited for preliminary exploration of tumor resistance mechanisms and high-throughput screening of candidate drugs, organoids exhibit superior advantages in mimicking complex immune-resistant TME and enabling personalized studies. A complementary integration of both systems can provide a more comprehensive toolkit for investigating immunotherapy resistance mechanisms and advancing clinical drug development [33] (Fig. 1).

Fig. 1.

From in vitro immunoresistance to in vivo immunoresistance modeling

Establishment of immunotherapy-resistant tumor models in tumor-bearing animals via repeated in vivo administration

Immunotherapy-sensitive tumor cell lines can be selected and inoculated subcutaneously into mice. The tumor-bearing mice are then subjected to long-term drug induction, and tumor tissues from mice unresponsive to PD-1/PD-L1 antibody therapy are continuously passaged into the next generation of mice. Tumors treated with the drug initially exhibit slow growth, gradually accelerate in growth rate, and eventually show reduced drug sensitivity, thereby establishing an immunotherapy-resistant model [34].

The 344SQ murine lung carcinoma cell line was subcutaneously inoculated into female 129 Sv/ev mice. Beginning on day 4 post-engraftment, anti-PD-1 antibodies or isotype control IgG antibodies were administered intraperitoneally twice weekly. Tumors refractory to anti-PD-1 antibody therapy were surgically resected, enzymatically dissociated into single-cell suspensions, expanded in vitro, and serially reinoculated into syngeneic murine hosts through four iterative in vivo passages. This selection protocol generated phenotypically stable anti-PD-1 antibody-resistant lung adenocarcinoma models and corresponding resistant cell lines. Anti-PD-1 antibody-resistant tumors exhibited downregulation of major histocompatibility complex (MHC) class I and class II molecules as well as β2-microglobulin, concomitant with reduced infiltration and activation of CD4 + and CD8 + T cells. Localized radiotherapy can resensitize anti-PD-1-resistant lung adenocarcinoma tumors to PD-1 blockade therapy through radiation-induced upregulation of MHC class I antigen presentation [35]. Leveraging this resistant model, Barsoumian et al.[36] discovered that radiotherapy combined with NLRP3 agonism stimulates the immune system and augments systemic antitumor responses, resulting in delayed tumor progression and even complete clearance of tumors at unirradiated sites. Mice treated with either X-ray therapy (XRT) + NLRP3 agonist or the triple combination (XRT + NLRP3 agonist + anti-PD-1 antibody) exhibited elevated serum levels of proinflammatory cytokines—including IL-1β, IL-4, IL-12, IL-17, IFN-γ, and GM-CSF—triggering coordinated local tumor microenvironment remodeling and systemic immune activation.

Tumor fragments demonstrating refractoriness to anti-PD-1 antibody therapy were selected and orthotopically reimplanted into mammary fat pads of subsequent-generation BALB/c nude mice recipients, while maintaining concurrent anti-PD-1 therapeutic administration. Through 3–4 iterative cycles of in vivo selection, mice bearing the final-passage tumors exhibited diminished therapeutic response, thereby generating a validated PD-1 blockade-resistant breast cancer model [15]. High expression of tumor-associated macrophage receptor family members Tyro3, Axl, and MerTK was detected in PD-1 antibody-resistant breast cancer tumor tissues. Mechanistically, activation of these receptor tyrosine kinases drives macrophage polarization from an immunostimulatory M1 phenotype to an immunosuppressive M2 phenotype, facilitating tumor progression and suppressing antitumor immunity. Therapeutic targeting of the TAM receptor family (Tyro3, Axl, MerTK) to reverse M2 macrophage polarization represents a promising strategy for restoring antitumor immunity and overcoming resistance to PD-1 blockade [37, 38].

MC38 tumors subjected to three serial in vivo passages under anti-PD-1 antibody treatment developed acquired resistance to PD-1 blockade. Resistant tumors exhibited hyperactivation of the TGFβ and Notch signaling pathways, as well as diminished infiltration of cytotoxic immune populations, including CD8 + T cells and NK cells [39]. Mechanistically, hyperactivation of TGFβ and Notch signaling pathways drives Tregs expansion while suppressing the proliferation and cytotoxic activity of effector T cells and NK cells. This immunosuppressive reprogramming correlated with poor clinical response to checkpoint inhibitors in patients. Dual inhibition of TGFβ and Notch signaling synergistically restored antitumor immunity when combined with PD-1 blockade, significantly delaying the growth of resistant tumors [40, 41].

Subcutaneous transplantation of CT26 cells in BALB/c-hPD1 mice treated with PD-1 antibody was performed to screen nonresponsive tumors, which were subsequently reimplanted into untreated mice and subjected to repeated PD-1 antibody treatment. After four consecutive cycles of screening, a PD-1 resistance-induced model was successfully established. The induced PD-1-resistant CT26 cell line maintained a stable resistant phenotype following in vitro passaging and cryopreservation. In this model, Serine/Threonine Kinase 11 (STK11) was identified to be critically associated with PD-1 resistance. STK11 depletion led to reduced CD8 + T cell infiltration and significant accumulation of MDSCs. Furthermore, STK11 deficiency correlated with decreased PD-L1 expression in tumor cells, potentially impairing immune system recognition and attack mechanisms, consequently diminishing the efficacy of immunotherapy [42]. These findings suggest that STK11 may serve as a crucial therapeutic target associated with resistance to immune checkpoint inhibitors [43].

PD-1 resistance in vivo models were established through treatment/reimplantation cycles using MC38 colorectal cancer and MBT-2 bladder cancer models. In these models, elevated Serpinf1 expression was identified to drive PD-1 antibody resistance in MC38 tumors, while Serpinf1 knockdown restored therapeutic sensitivity in both MC38 and MBT-2 resistant models. Mechanistically, Serpinf1 overexpression correlated with increased free fatty acid (FFA) production and reduced CD8 + T cell activation. Notably, the reversal of PD-1 treatment resistance by orlistat, an FFA biosynthesis suppressor, implicates Serpinf1 as a key mediator of immunoevasion via metabolic reprogramming, thereby identifying Serpinf1 targeting as a critical therapeutic strategy for enhancing immunotherapy [34].

Establishing immunotherapy-resistant tumor models through repeated in vivo administration in tumor-bearing animals serves as a critical approach for investigating tumor immune resistance-associated targets and screening strategies to reverse therapeutic resistance. The repeated administration of immunotherapeutic agents to induce resistance mimics the gradual development of acquired resistance observed in patients undergoing prolonged treatment rather than transient, in vitro-induced acute resistance. This process faithfully models the clinical emergence of acquired resistance during long-term therapeutic regimens. This model not only retains the intact TME, immune cell infiltration, and systemic immune regulatory networks, but also faithfully recapitulates the dynamic interplay between tumors and the immune system during the development of immunotherapy resistance [44]. However, this model has limitations. The successful establishment of resistance models is highly dependent on the design of dosing regimens and is susceptible to experimental variability, such as inter-individual variability among animals or microbial colonization, which may lead to inconsistent results. Moreover, significant interspecies variations between murine and human immune systems—such as differential distribution patterns of immune checkpoint molecules and divergent sensitivity profiles of cytokine receptors—may limit the translatability of resistance mechanisms to human clinical settings [45, 46] (Fig. 2).

Fig. 2.

Establishment of immunotherapy-resistant tumor models in tumor-bearing animals via repeated in vivo administration

Development of immunotherapy-resistant models via patient-derived resistant xenografts

In vitro immunoresistant models can be established either by pharmacological induction or isolation from drug-resistant tissues, and are then subcutaneously or orthotopically implanted into mice to generate in vivo immunoresistant models. However, cell lines frequently acquire unintended phenotypic alterations during in vitro culture, leading to diminished similarity to parental resistant tumors [30]. To address this limitation and develop models that better recapitulate human therapeutic resistance, immunotherapy-resistant patient-derived xenograft (PDX) models have been established. These involve engrafting tumor fragments from cancer patients into humanized immunodeficient mice, serially subjecting them to ICI treatment and transplanting nonresponsive tumors into secondary recipient mice over multiple cycles. Immunotherapy-resistant PDX models typically retain the genomic characteristics of patients at different stages, with various subtypes, and diverse treatment histories, even after extended passaging. Owing to these advantages, immunotherapy-resistant PDX models serve as an ideal choice for anticancer drug resistance research, including preclinical evaluation of new therapeutics, validation of novel combination regimens, identification of drug-sensitive patient populations, and investigation of resistance mechanisms [47].

Fresh tumor tissue from immunotherapy-resistant patients was dissected into fragments and engrafted into female immunodeficient mice previously humanized with CD34 + hematopoietic stem cells. T cell bispecific antibody therapy was administered twice weekly, with drug concentration escalation permitted when tumor volume reached 300 mm³. Treatment was discontinued upon tumor regression and reinitiated upon recurrence. Mice were humanely euthanized before tumors exceeded 1,000 mm³, followed by tumor resection for processing into single-cell suspensions or reimplantation as tissue fragments into secondary recipient mice. Following three serial passages, an acquired immunotherapy-resistant PDX model was successfully established [48].

An immune-resistant melanoma PDX model constructed in human PBMC-reconstructed NSG mice transplanted with surgical tumor fragments from patients with PD-1-blocked refractory melanoma demonstrated superior efficacy of treatment with a prion-like chemical inducer known as SAP. Effective tumor growth inhibition was achieved with SAP, which was superior to that of anti-PD-1 treatment. Studies have evaluated anti-PD-1 immunotherapy in humanized PDX (huPDX) murine models of high-grade serous ovarian cancer. Mice treated with nivolumab exhibited no significant survival benefit and limited therapeutic efficacy. However, combining ICIs with either histone deacetylase inhibitors or antiangiogenic agents enhanced tumor immunogenicity and potentiated the antitumor immune response [11, 49]. An microsatellite stable (MSS) colorectal cancer (CRC) huPDX model was treated with nivolumab. Tumor growth inhibition was noted in the initial 10-day period, after which rapid progression occurred, consistent with clinical reports in anti-PD-1-treated MSS-CRC patients. The absence of human T cell infiltration in MSS-CRC huPDX models may account for their reduced response to nivolumab. Clinically, immunotherapeutic approaches aimed at increasing T cell levels and enhancing T cell function could be combined [50].

The PDX models are directly derived from immunotherapy-resistant tumor tissues of patients, preserving the genomic, transcriptomic, and epigenomic features of the original tumors, which can more realistically reflect the key mechanisms of immunotherapy resistance in the clinic, and are suitable for the study of TME interactions [51]. By matching patient’s clinical data, it can be used to predict immunoresistance markers and screen individualized treatment regimens. Traditional PDX models utilize immunodeficient mice, which lack a functional human immune system and thus fail to fully replicate interactions between human immune cells and tumors. This limitation can be partially addressed by huPDX models, where immunodeficient mice are pre-engrafted with human immune cells [52]. However, the PDX resistance model is characterized by prolonged experimental cycles, low success rates, and high maintenance costs. NSG mice engrafted with human peripheral blood mononuclear cells are susceptible to the onset of severe xenogeneic graft-versus-host disease as early as 3–4 weeks post-injection [53]. The balance of hematopoietic and immune cells in mice remains distinct from that in humans. Moreover, matching HLA presents a significant challenge due to the allogeneic origin of hematopoietic stem cells (HSCs) versus patient-derived tumors [54] (Supplementary Fig. S1).

Development of immunotherapy-resistant models through targeted genetic modification

Constructing immune resistance models based on specific genetic targets is indispensable for investigating immunotherapy resistance mechanisms. Mutations, overexpression, and silencing of specific genes represent key mechanisms driving immune resistance development. To modulate gene expression in tumor cells, techniques including gene editing, lentiviral vector-mediated gene transduction, and other approaches are employed to construct models with overexpression, knockdown, or knockout of target genes. These models are likely closely associated with tumor immune evasion [55–57]. Elucidating the mechanisms by which these genes contribute to resistance enables the exploration of targeted intervention strategies. Combining such approaches with immunotherapies holds promise for overcoming immunotherapy resistance and improving therapeutic outcomes in patients [58].

Interleukin-4-Induced-1 (IL-4I1) is an amino acid oxidase that depletes essential amino acids and generates metabolites toxic to cytotoxic T cells, thereby impacting the antitumor efficacy of ICIs. Elevated expression of IL-4I1 has been observed in melanoma patients resistant to PD-1 inhibitors [59]. Hirose et al. [60] engineered IL-4I1-overexpressing B16-F10 melanoma cells via retroviral-mediated gene transduction. These cells were subcutaneously injected into 6-week-old female C57BL/6 J mice, resulting in tumors refractory to PD-L1 antibody therapy and exhibiting significantly reduced CD8 + T cell infiltration. The overexpression of IL-4I1 not only upregulates immunosuppressive genes in tumors but also enhances the expression and secretion of chemokines CCL2 and CCL10, thereby recruiting tumor-associated macrophages and contributing to the formation of an immunosuppressive tumor microenvironment. In melanoma patients, early detection of IL-4I1 expression may be helpful in predicting the response to PD-1 inhibitor therapy, as well as the potential risk of drug resistance.

Upregulated expression of miR-20a-5p in exosomes released by tumor cells has been observed in various malignancies, including triple-negative breast cancer (TNBC) [61]. TNBC cells stably expressing miR-20a-5p were implanted into the 7–8 week-old NSG immunodeficient mice. These tumor-bearing mice exhibited resistance to PD-1 antibody therapy. The nuclear protein ataxia–telangiectasia gene (NPAT) was highly expressed in immature CD8 + T cells and was critical for their rapid proliferation. Exosomes with upregulated miR-20a-5p expression are taken up by CD8 + T cells, leading to reduced NPAT expression and impaired effector functions in these cells [62]. The immunotherapy resistance model demonstrated that miR-20a-5p upregulation may serve as a potential biomarker for malignancies, particularly TNBC. This discovery holds clinical applicability for early diagnosis, prognostic stratification, and prediction of immunotherapy response in patients.

Researches have demonstrated that the expression of ganglioside-2 (GD2) on the surface of tumor cells promotes their proliferation, migration and invasion [63, 64]. These biological properties are associated with the amplification and expression of the neuroblastoma-derived MYC oncogene (N-MYC). The neuroblastoma 9464D cell line was sequentially transduced with lentiviral vectors encoding GD2 synthase and GD3 synthase to generate a GD2-high-expressing variant (designated "9464D-GD2"). Xenograft models established using 9464D-GD2 cells exhibited insensitivity to anti-CTLA-4 monotherapy, which was associated with N-MYC mutations and a low TMB [65]. N-MYC induces the upregulation of PD-L1 in tumor cells. The interaction of PD-L1 with its receptor PD-1 on T cells triggers immunosuppression [66]. Evidence shows that N-MYC-targeting inhibitors can overcome acquired resistance in neuroblastoma [67].

Studies demonstrate that Forkhead Box Protein A1(FOXA1) is downregulated in nasopharyngeal carcinoma (NPC) tissues and functions as a tumor suppressor in NPC development and progression [68]. Co-culturing primary tumor-specific T cells with FOXA1-silenced NPC cells induced apoptosis in in vitro-activated tumor-specific CD8⁺ T cells and reduced the expression of cytotoxic effector molecules. Furthermore, in nude mice receiving adoptive T cell therapy, FOXA1 overexpression enhanced the therapeutic efficacy of atezolizumab against NPC. Mechanistic studies revealed that FOXA1 inhibits IFN-γ-induced nuclear translocation of STAT1, restricting downstream transcription by IRF1 and consequently suppressing PD-L1 expression [69]. FOXA1 expression may function as an important predictor of treatment resistance and a druggable target to enhance sensitivity of prostate cancer and bladder cancer to immunotherapeutic and chemotherapeutic agents [70].

Genetic engineering techniques enable targeted editing, insertion, or deletion of specific genes, allowing precise construction of customized immune resistance models with enhanced research specificity. Compared to conventional approaches for establishing immunotherapy resistance models, this method demonstrates significantly higher efficiency in model development while exhibiting lower time and cost requirements, enabling rapid screening of resistance-associated targets with enhanced mechanistic validation [71]. However, its limitations primarily stem from an overreliance on fully elucidated mechanisms of immunotherapy resistance and refined genetic modification techniques. Gene editing techniques may affect nontarget genes, presenting off-target risks [72]. Clinical resistance is frequently driven by coexisting multiple genes, which cannot be recapitulated by single-gene modification models. Certain resistance mechanisms remain non-replicable through genetic editing [73] (Supplementary Fig. S2).

Modeling immunotherapy resistance via fecal microbiota transplantation from resistant patients

In recent years, accumulating evidence has demonstrated that the composition and homeostasis of the gut microbiota play a pivotal role in shaping the immune system and influencing resistance to immunotherapy [74, 75]. Targeted modulation strategies of the gut microbiota to ameliorate resistance to cancer immunotherapy in patients have been validated. As evidenced by several clinical trials, the combined use of FMT with ICIs can enhance clinical responses and mitigate resistance to ICB therapy in cancer patients [76, 77]. MSS-CRC are typically resistant to anti-PD-1 therapy. A CRC-associated pathogen, Fusobacterium nucleatum (Fn), significantly enhances the sensitivity of MSS-CRC to anti-PD-1 treatment. FMT from Fn-high MSS-CRC patients into germ-free MSS-CRC mice resulted in significantly enhanced anti-PD-1 responses compared to FMT from Fn-low donors [78].

While transplantation of gut microbiota from immunotherapy-responsive patients has shown potential to overcome treatment resistance, emerging evidence confirms that transferring microbiota from nonresponders may conversely induce therapeutic resistance [75]. FMT provides a novel approach for constructing immunotherapy-resistant models. Fecal pellets were collected from patients nonresponsive to anti-PD-1 antibody therapy and prepared into a suspension. The suspension was gavaged three times a week to 6–8 week-old germ-free mice and an appropriate amount of the suspension was applied to the mice’s fur. Two weeks later, mice were inoculated with tumor cells and treated with anti-PD-1 antibody therapy, and no significant suppression of tumor growth was observed [79]. In a murine model of anti-PD-1 therapy-resistant non-small cell lung cancer established via FMT. Yu et al. demonstrated that baicalin enhanced gut microbiota diversity, notably enriching short-chain fatty acid-producing genera Akkermansia and Clostridia_UCG-014. These microbiota shifts correlated with increased CD8 + T cell infiltration in the tumor immune microenvironment, elevated production of TNF-α and IFN-γ by CD8 + T cells, and attenuated immunosuppressive effects mediated by Tregs, collectively ameliorating resistance to PD-1 blockade therapy [80]. Compared to healthy controls, the efficacy of the PD-1 antibody therapy was significantly impaired in tumor-bearing mice that received FMT from colorectal cancer patients with treatment resistance. Pectin substantially enhanced the antitumor efficacy of PD-1 antibody in this model. Treatment with PD-1 antibody plus pectin increased T cell infiltration and activation within the tumor microenvironment. This beneficial effect is likely mediated by the immunomodulatory properties of butyrate, a gut microbial metabolite [81].

Transplanting the entire gut microbiome from immunotherapy-resistant patients into germ-free mice preserves the microbial community architecture associated with resistance, overcoming the limitations of single-strain validation and enabling an integrated analysis of the gut microbiota–tumor immune microenvironment axis [82]. This modeling approach remains in its preliminary research phase. Critical parameters, such as fecal sample dosage, transplantation timing, donor/recipient mouse selection, processing protocols, criteria for successful model establishment, and model reproducibility, require systematic experimental validation to be refined and optimized, ultimately enabling the generation of more reliable immune-resistant models [83] (Supplementary Fig. S3).

Developing an immunotherapy resistance prediction model through bioinformatics and computational algorithms for precision immunotherapy

Developing preclinical models that mimic human immune checkpoint inhibitor resistance is a crucial approach for exploring the biological mechanisms of resistance, identifying potential therapeutic targets, and validating combination strategies. While these traditional in vitro and in vivo models have demonstrated significant value in elucidating core resistance pathways and discovering key targets, they remain inherently limited in guiding the selection of clinical combination therapies. This limitation primarily stems from inherent species differences between preclinical models and human patients, encompassing immune system composition, tumor progression, and gut microbiota. Consequently, combination therapies demonstrating significant synergistic efficacy in mouse models of immunotherapy resistance often exhibit diminished or nonsignificant effects when advanced to human clinical trials [84, 85]. This phenomenon highlights the challenge of predicting clinical translation outcomes through preclinical research. To overcome this bottleneck, a more translationally promising strategy lies in leveraging real-world clinical data. By integrating and analyzing multi-omics data alongside clinical follow-up information from large cohorts of patients treated with immune checkpoint inhibitors, machine learning-based predictive models for immunotherapy efficacy can be developed [85]. Such models employ computational algorithms to compare individual patient’s tumor biological characteristics against vast databases of known treatment outcomes from large patient samples [86]. This approach aims to transcend species barriers, relying directly on human evidence to identify optimal combination therapies for immune-resistant patients, ultimately advancing tumor immunotherapy toward a higher level of precision medicine.

This study systematically analyzed the heterogeneity of the tumor immune microenvironment by integrating bulk and single-cell transcriptomic data from hepatocellular carcinoma patients, thereby generating an immune gene signature score. Based on this immune gene signature score, seven algorithms were employed to identify gene sets associated with immune checkpoint inhibitor response, and patients were classified into distinct immune subtypes using unsupervised consensus clustering. Furthermore, cross-scale integration of bulk and single-cell data elucidated immune subtype-specific cellular architectures and molecular profiles, enabling identification of key cellular populations associated with therapeutic response. Based on these findings, potential drug candidates were screened for patients with distinct immune subtypes, providing both theoretical foundation and strategic support for precision immunotherapy in hepatocellular carcinoma [87]. By integrating single-cell and bulk RNA sequencing data from the GEO database, this study employed the Scissor algorithm to identify cell subpopulations positively and negatively correlated with immunotherapy response, and utilized the BayesPrism algorithm to quantify TME cellular composition. The analysis revealed enrichment of T cells and B cells in responders versus increased macrophages in nonresponders, with B cell proportion positively correlated with prognosis. Based on these findings, a nine-gene predictive model for immunotherapy efficacy was successfully constructed. This model demonstrated superior predictive performance across multiple independent cohorts, providing novel biomarkers and theoretical foundations for precision immunotherapy in melanoma [88]. In adrenocortical carcinoma, the integrated application of xCell, weighted gene co-expression network analysis, and LASSO-Cox regression delineated tumor microenvironment subtypes characterized by distinct immune checkpoint expression patterns, macrophage infiltration, and immunotherapy sensitivity. Core genes were selected to construct a gene signature model for prognostic prediction. These findings not only deepen our understanding of adrenal cortical carcinoma TME heterogeneity but also provide novel biomarkers and clinical strategies for precision immunotherapy application and patient prognosis assessment [89].

In the field of tumor immunotherapy, constructing predictive models for immune therapy resistance based on bioinformatics data and machine learning algorithms holds significant potential for clinical translation. Such models are crucial for precisely identifying patient subgroups responsive to immune checkpoint inhibitors and guiding personalized combination therapy strategies. Developing computational models that elucidate immune tolerance mechanisms based on tumor microenvironment characteristics and immune cell subset transcriptomic expression profiles in specific patient populations will provide a theoretical foundation for mechanism-driven combination therapy selection. This is crucial for reversing immune resistance states and improving clinical outcomes (Supplementary Fig. S4 and Table 1).

Table 1.

Mechanisms of resistance, therapeutic strategies, model applications, and model characteristics for each immunotherapy resistance model

Model category	Mechanisms of tumor immune resistance	Intervention strategies	Model applications	Model characteristics
In vitro-induced immunotherapy resistance models	Sustained IFN-γ signaling activation → upregulation of ISGs → T cell exhaustion[10]	SAP chemical inducer → reversal of sustained IFN-γ signaling → overcomes melanoma therapy resistance	Preliminary screening of immunotherapy resistance targets and high-throughput screening of immunotherapy candidate drugs	Advantages: Short cycle time, easy expansion, rapid validation of immune resistance mechanisms Limitations: Lack of TME components, and potential loss of inherent tumor heterogeneity during prolonged passaging
	Acidosis potentiates PD-L1 expression → IFN-γ + low pH → PD-L1 upregulation → immune evasion[14]	NaHCO₃ + Anti-PD-L1 → neutralizes tumor acidity → enhanced T cell Infiltration
	Cold tumor phenotype → low TMB + MDSCs/tregs infiltration → primary immunotherapy resistance[16]	Targeting MS4A4A + Anti-PD-1 → restoring therapeutic sensitivity in CT26/B16-F10 tumors
	Antigen loss and defective antigen presentation[24]	STING agonist-loaded nanoparticles + Anti-PD-1 → NK cell activation
	Tyro3/CDK9 overexpression mediates anti-PD-1 resistance in breast cancer[15]	Tyro3/CDK9 Inhibitors + ICIs → therapeutic sensitization
	Glucocorticoids impair immunotherapy efficacy[22]	Restrict dexamethasone use during ICB therapy in glioma patients with peritumoral edema
	Elevated levels of immunosuppressive molecules → immune exhaustion[28]	Toripalimab → Elevated CD8 + /CD4 + T cell ratio with increased tumor cell apoptosis → reversal of immunotherapy resistance in PDO models
In vivo immunotherapy resistance induction models	MHC Class I/II downregulation → reduced T cell infiltration → immunotherapy resistance in lung cancer[35]	Radiotherapy + NLRP3 Agonist + anti-PD-1 → upregulation of MHC Class I/II → reversal of therapy resistance	Simulation of clinical progressive immunological resistance and evaluation of combination therapies	Advantages: Preservation of intact tumor microenvironment Limitations: Model construction affected by multifactorial variation with inherent human mouse immune system inconsistency
	TAM Kinase (Tyro3/Axl/MerTK) activation → M2-like macrophage polarization → immunotherapy resistance in breast cancer[15]	TAM receptor inhibitors → reversal of M2 phenotypic polarization
	TGFβ / Notch pathway activation → tregs expansion + NK cell reduction[39]	Dual pathway inhibitor + anti-PD-1 → synergistic inhibition of colorectal cancer progression
	STK11 deficiency → MDSCs accumulation + PD-L1 downregulation → resistance to therapy in CT26 models[43]	Targeting the STK11 pathway → restoration of CD8⁺ T cell function
	Serpinf1 overexpression → elevated FFA levels → CD8⁺ T cell dysfunction[34]	Orlistat → inhibition of FFA synthesis → reversal of anti-PD-1 resistance
PDX immunotherapy resistance models	Effector immune cells with functional defects[50]	HDAC inhibitors/antiangiogenic agents + ICI → enhanced response to immunotherapy for ovarian cancer SAP chemical inducer → tumor growth inhibition	Personalized immunotherapy response prediction and elucidation of clinical resistance mechanisms	Advantages: Preservation of patient-specific genomic characteristics Limitations: High risk of graft-versus-host disease, prolonged experimental duration, elevated maintenance costs, and uncertainties in human–mouse HLA matching
Targeted genetic engineering models of immunotherapy resistance	IL-4I1 overexpression → depletion of essential amino acids and production of toxic metabolites → CD8⁺ T cell dysfunction[60]	Early IL-4I1 detection → prediction of anti-PD-1 response	Functional validation of specific genes and discovery of immunotherapy resistance biomarkers	Advantages: Precise construction of customized immune resistance models with study specificity Limitations: Off-target risks and inability to recapitulate polygenic co-occurrence-mediated therapy resistance
	N-MYC overexpression → upregulates PD-L1 → immunotherapy resistance in neuroblastoma[65]	N-MYC inhibitor + Anti-PD-1 therapy → overcoming immunotherapy resistance
	miR-20a-5p upregulation → inhibition of NPAT in CD8⁺ T cells → TNBC immunotherapy resistance[62]	miR-20a-5p as a predictive biomarker for TNBC
	FOXA1 silencing → promoting PD-L1 expression → nasopharyngeal cancer resistance[69]	FOXA1 overexpression + Atezolizumab → sensitization to immunotherapy
Host–microbiota interaction-mediated resistance models	Colonization of immunotherapy-resistant gut microbiota: gut microbiota of transplantation-resistant patients → inhibition of anti-PD-1 efficacy by an unfavorable gut microbiota for immunotherapy[75]	Baicalin enriches Akkermansia spp. and Clostridia_UCG-014 spp. → Increased production of TNFɑ and IFNγ by CD8⁺ T cells and diminished immunosuppression by Tregs[80]	Microbiota–immune crosstalk investigation	Advantages: Recapitulation of the microbiota–immune axis Limitations: Lack of standardization (unresolved dosage protocols and inconsistent transplantation times)
		Pectin → increased butyrate production → enhanced T cell function → increased sensitivity to anti-PD-1 therapy[81]

Discussion

The development of resistance to immunotherapy represents a leading cause of failure for anticancer agents [90]. A comprehensive understanding of the mechanisms of resistance is crucial. In-depth elucidation of these resistance mechanisms will facilitate the discovery of novel drug targets, enable the design of more effective therapeutics, and provide a theoretical foundation for developing combination therapies. Understanding resistance mechanisms is intrinsically linked to models of immunotherapy resistance [91]. Creating models of immune resistance is key to systematically understanding different resistance pathways. This will provide a scientific foundation for developing new treatment methods and improving clinical therapeutic strategies.

Developing effective preclinical models to reliably elucidate tumor resistance mechanisms and predict antitumor efficacy in human clinical trials of immunotherapy remains a critical unmet need. In vitro models of immune resistance play a crucial role in elucidating resistance mechanisms and the initial screening of novel antitumor agents. However, their primary limitation lies in the absence of key components of TME, preventing the simulation of complete "tumor-immune" microenvironmental interactions. Resistance models constructed by repeatedly administering immunotherapeutic agents to tumor-bearing mice effectively mimic the clinical development of acquired resistance during long-term treatment. These models preserve a relatively intact resistant TME and systemic immune regulatory network. Nevertheless, significant differences between murine and human immune systems exist, potentially causing resistance mechanisms observed in animal models to fail to fully recapitulate in clinical patients. Models established by transplanting human-derived resistant tumor tissue into mice with a humanized immune system offer a more clinically relevant approach. They better simulate the interactions between human immune cells and tumor cells, providing a powerful tool for uncovering key mechanisms underlying clinical immunotherapy resistance. Furthermore, constructing more refined immunotherapy resistance models based on specific genetic modifications or manipulation of the gut microbiota focuses on identifying genetic or microbial features associated with specific resistance phenotypes. This enables the exploration of more targeted intervention strategies. By integrating transcriptomic data from immunotherapy patients with traditional in vitro and in vivo experimental models, we systematically analyze the tumor immune microenvironment to identify key gene signatures associated with immunotherapy response. This enables the construction and validation of an integrated model capable of both predicting treatment efficacy and assessing prognosis, ultimately driving clinical decision-making to select personalized treatment strategies for drug-resistant patients. In conclusion, these distinct tumor immunotherapy resistance models play a vital role in simulating clinical resistance phenotypes, discovering potential novel resistance targets, developing targeted therapeutics, and exploring effective combination treatment strategies. Based on the unique characteristics of each model, and using strategies that capitalize on their respective strengths and promote complementary strengths, these different models together form a comprehensive research continuum. This continuum ranges from basic mechanistic exploration to clinical translational prediction, ultimately serving the overall goal of overcoming immunotherapy resistance and improving patient prognosis.

Supplementary Information

Below is the link to the electronic supplementary material.

List of symbols

2D

Two-dimensional

3D

Three-dimensional

ccRCC

Clear cell renal cell carcinoma

CDK9

Cyclin-dependent kinase 9

CRC

Colorectal cancer

FFA

Free fatty acid

FMT

Fecal microbiota transplantation

Fn

Fusobacterium nucleatum

FOXA1

Forkhead Box Protein A1

GD2

Ganglioside-2

GPNMB

Glycoprotein nonmetastatic melanoma protein B

HSCs

Hematopoietic stem cells

HuPDX

Humanized PDX

ICB

Immune checkpoint blockade

ICIs

Immune checkpoint inhibitors

IFN-γ

Interferon-γ

IL-4I1

Interleukin-4-Induced-1

MDSCs

Myeloid-derived suppressor cells

MHC

Major histocompatibility complex

MS4A4A

Membrane spanning 4-domains A4A

MSS

Microsatellite stable

N-MYC

Neuroblastoma-derived MYC

NPAT

Nuclear protein ataxia–telangiectasia

NPC

Nasopharyngeal carcinoma

PDO

Patient-derived organoid

PDX

Patient-derived xenograft

STK11

Serine/threonine kinase 11

TMB

Tumor mutational burden

TME

Tumor microenvironment

TNBC

Triple-negative breast cancer

Tregs

Regulatory T cells

XRT

X-ray therapy

Author’s contribution

Huiling Li helped in writing—original draft. Shenzhan Wang contributed to writing—review & editing. Xiaoxi Li helped in writing—review & editing. Chao Wu contributed to writing—review & editing. Yingnan Feng helped in writing—review & editing. Xin Hu contributed to writing—review & editing. Lan Zhang helped in writing—review & editing, supervision, resources, project administration. Xianzhe Dong contributed to writing—review & editing, supervision, resources, project administration, funding acquisition.

Funding

This work was supported by the National Natural Science Foundation of China (No.82271765, 81773778); Talent Project established by Chinese Pharmaceutical Association Hospital Pharmacy department (No.CPA-Z05-ZC-2023–003); Beijing Hospitals Authority Youth Programme (Code: QML20230811).

Data availability

No datasets were generated or analysed during the current study.

Declarations

Conflict of interest

The authors declare no competing interests.