Mouse Models for Immune Checkpoint Blockade Therapeutic Research in Oral Cancer

The most prevalent oral cancer globally is oral squamous cell carcinoma (OSCC). The invasion of adjacent bones and the metastasis to regional lymph nodes often lead to poor prognoses and shortened survival times in patients with OSCC. Encouraging immunotherapeutic responses have been seen with immune checkpoint inhibitors (ICIs); however, these positive responses to monotherapy have been limited to a small subset of patients. Therefore, it is urgent that further investigations into optimizing immunotherapies are conducted. Areas of research include identifying novel immune checkpoints and targets and tailoring treatment programs to meet the needs of individual patients. Furthermore, the advancement of combination therapies against OSCC is also critical. Thus, additional studies are needed to ensure clinical trials are successful. Mice models are advantageous in immunotherapy research with several advantages, such as relatively low costs and high tumor growth success rate. This review paper divided methods for establishing OSCC mouse models into four categories: syngeneic tumor models, chemical carcinogen induction, genetically engineered mouse, and humanized mouse. Each method has advantages and disadvantages that influence its application in OSCC research. This review comprehensively surveys the literature and summarizes the current mouse models used in immunotherapy, their advantages and disadvantages, and details relating to the cell lines for oral cancer growth. This review aims to present evidence and considerations for choosing a suitable model establishment method to investigate the early diagnosis, clinical treatment, and related pathogenesis of OSCC.


Introduction
Oral cancer is one of the most common malignant neoplasms in humans, endangering human health, of which more than 90% are oral squamous cell carcinomas (OSCC).
OSCC is a subtype of head and neck squamous cell carcinoma (HNSCC) with an estimated incidence of more than 370,000 new cases and 170,000 deaths annually [1][2][3][4]. Globally, the main risks of OSCC are tobacco smoking, alcohol drinking, and betel nut chewing, followed by infection with high-risk human papillomavirus (HPV) [5][6][7][8]. OSCC usually occurs in elderly patients; however, the incidence has increased in young people and is mainly due to HPV-associated oropharyngeal squamous cell carcinoma [9]. Over  orthotopically) into immunocompetent hosts. (B) Chemotoxic agent mouse models: Chemicals (carcinogens) are administered to induce OSCC growth. In mice treated with carcinogens, the tumor also forms spontaneously de novo. (C) genetically engineered mouse models use autochthonous tumor cell growth driven by the tissue-specific deletion of tumor suppressors or tissue-specific expression of oncogenes. (D) The humanized mouse can be further divided into two categories: (D1) humanized xenograft mice (peripheral blood mononuclear cells (PBMCs) or hematopoietic stem cells (HSCs) were implanted into immunodeficient mice. After 4-12 weeks, patient-derived tumors or cell line-derived tumors were implanted into mice orthotopically or subcutaneously to form humanized patient-derived xenograft (PDX) mice or humanized cell-derived xenograft (CDX) mice, respectively); (D2) using the expertise in the genetically engineered mouse model (GEMM) generation by CRISPR/Cas9, human immune checkpoint knock-in (KI) mice were generated that can be used to test specific molecular interactions between tumor cells and immune cells as well as within the immune system. The human immune checkpoint KI mice can further be used to establish human immune system (HIS) mice, resulting in humanized immune checkpoint KI mice. 4-NQO-4-nitroquinoline 1-oxide.

Humanized
Highly reflect promoter methylation in tumors and reproduced tumor heterogeneity.
Heterogeneity of the tumor is high.
High in cost. Difficult to set up. A partial TME is transplanted from the patient and is dependent on the transplantation site and donor immune cells.

Hu-PBMC
Simple to construct, the T cell transplantation efficiency is high and it is stable.
Limited study time due to short survival as well as the occurrence of GvHD.    K5 or K14-CrePR1-keratin 5 (K5) or keratin 14 (K14) tissue-specific promoter used in the overexpression of the oncogene K-ras G12D targeted to the oral epithelium of mice. CrePR1-Cre recombinase fused to a human progesterone receptor mutant.

Syngeneic Mouse Models
Syngeneic OSCC mouse models are produced using allografts of immortalized mouse tumor cell lines. The models can efficiently prevent tissue rejection or graft-versus-host disease (GvHD), which results from transplanting tumor cells into mice. In addition, these models have the inherent advantage of fast establishment, high stability, and high consistency of transplanted tumors. Therefore, they are widely used in immuno-oncology studies and show great potential in developing novel OSCC treatments, especially immunotherapy [70]. Tumor cells can be injected orthotopically or ectopically. The orthotopic model supplies a more exact tumor microenvironment, while the subcutaneous model facilitates tumor monitoring and handling. For oral cancer induction, orthotopic models are performed by injecting tumor cells into the oral cavity, while ectopic models most commonly receive subcutaneous injections in the flanks [24].
In addition, several researchers use 4-nitroquinoline-1-oxide (4NQO) to induce OSCC in a mouse and then inject the OSCC obtained from the mouse into another mouse of the same species to establish a syngeneic model. This is similar to the construction of patient-derived xenograft (PDX) models. For example, Chen et al. used 4NQO (100 µg/mL in drinking water) to induce C57BL/6 mice over 16 weeks; the mice were sacrificed at week 28 to generate the mouse tongue OSCC cell lines MTCQ1 and MTCQ2. Afterward, the MTCQ1 or MTCQ2 cells were injected into the flank or tongue of new mice to establish ectopic or orthotopic mouse models, respectively [43]. Compared with human SAS tongue SCC cell lines, MTCQ cells have lower proliferation ability but far higher abilities of migration/invasion. Such capabilities are demonstrated through the identification of extensive cervical lymph node metastasis and lung metastasis resulting from an MTCQ1 cell subclone. Several therapeutic approaches have been tested using this model, including anti-PD-L1 immunotherapy, cisplatin therapy, and miRNAs (especially miR-134). In addition, Chen et al. [44] established a syngeneic model using 4NQO-induced OSCC transgenic mice. In this model, K14-EGFP-miR-211 transgenic mice were induced using 4NQO (100 µg/mL in drinking water) over 16 weeks and then sacrificed. Tissue isolated from OSCC lesions on the dorsal surface of the tongue was then used to produce cell lines MOC-L1 to -L4. These were subsequently used to create orthotopic xenografts and real-time in vivo tumor imaging by injecting cells (5 × 10 6 ) into the central tongue of C57BL/6 mice. These cells were also used to measure the efficacy of cisplatin therapy and study distant metastasis. Chen et al. [45] established the NHRI-HN1 and NHRI-HN2 cell lines from 4-NQO/arecoline-induced murine tongue tumors and further selected for cell stemness by in vitro sphere culture to evaluate potential immunotherapy for OSCC in East and Southeast Asia. NHRI-HN1 or NHRI-HN2 cells (5 × 10 5 ) in 50 µL in sterile phosphate-buffered saline (PBS) were injected into the buccal mucosa of mice that were sacrificed 40 days after injection.
Syngeneic tumor models have also been applied to the investigation of the anti-tumor activity of ICIs, including anti-programmed death (PD)-1/anti-PD-ligand 1 (L1) antibodies [71,72] and anti-cytotoxic T lymphocyte-associated antigen 4 (CTLA-4) [46]. The time taken to produce syngeneic tumor models is short, as tumor growth happens within a few weeks [25,73]. However, such rapid tumor growth can prevent the assessment of immunotherapeutics, as the treatment effect is often progressive and estimated by improving survival [74]. This makes syngeneic models unsuitable for assessing immunotherapy drugs at the early stages of tumor development [75]. The syngeneic OSCC mouse model is a viable tool for immuno-oncology. Still, the main problem is that the model only represents mouse oral cancer and forms mouse tumors with mouse targets. Mice and humans differ in compositions and mechanisms, and some targets in humans are absent or unresponsive in mice.

Chemotoxic Agent Mouse Models
Chemical carcinogen-induced (CI) tumor models are easier to establish experimentally and the cancers induced by carcinogen treatment are more like sporadic human cancers [73,76]. The mechanism of action can be divided into genotoxic (GTX) and nongenotoxic (NGTX). In the part of GTX, carcinogen directly acts on DNA or chromosomes and injures them [77], while the mechanism of NGTX action is first to trigger many other cellular effects, thus leading to changes in intracellular signaling pathways [78,79].
GTX carcinogens have electrophilic properties and can covalently bind to DNA to form DNA adducts and thus cause mutations [80,81]. GTX carcinogens can be said to be the initiator of cancer. For example, benzo(a)pyrene (BaP), a tobacco-related carcinogen, can form bulky BaP-DNA adducts and basic sites or, by reactive oxygen species (ROS) and metabolites generated during metabolism, causes DNA damage [82,83]. In contrast, NGTX carcinogens do not directly bind to DNA but affect the cell cycle, cause chronic inflammation [84], activate steroid hormone receptors (SHRs) to activate silent genetic pathways and generate reactive oxygen species (ROS), or cause immunosuppression [85]. In general, NGTX carcinogens have different actions, including mitogenic promotion (phenobarbital) [86], affecting the receptor-interacting protein-mediated pathway (2,3,7,8tetrachlorodibenzo-p-dioxin) [84,87,88], cytotoxicity (asbestos) [89], or interfering with gap junction intercellular communication (chloroform) [90]. The epigenetic mechanisms by these NGTX carcinogens cause genetic instability, which in the corresponding microenvironment leads to tumorigenesis [91].
CI models are usually established for a long time, but they have great genomic complexity, which accurately reflects the real situation of human tumorigenesis. Furthermore, with the larger mutational burden, neoantigens with different degrees of immunogenicity may be generated, thus affecting the immunogenicity of the tumor [92]. The generation of unknown neoantigens makes it challenging to determine and affect immune responses using the CI model. As such, these models are rarely utilized to assess the effectiveness of ICIs. However, the mouse model of oral cancer induced by 4NQO has been established and used to evaluate the efficacy of the anti-PD-1 antibody [93,94].

4NQO-Induced OSCC Mouse Model
Immunocompetent C57BL/6, BALB/c, CF-1, and CBA mice can be used to generate 4NQO-induced OSCC mouse models. Most studies applying this approach have used 6-8-week-old adult mice, but Vincent-Chong et al. noticed that 92% of aged mice (65-70 weeks of age) developed severe dysplasia/invasive squamous cell carcinoma, compared with 69% of young mice (7-12 weeks of age) [100]. The exposure of C57BL/6 and CBA mice to 4NQO in drinking water produced multiple precancerous lesions and carcinogenesis on the esophagus and tongue. CBA mice were more sensitive to 4NQO induction than the C57BL/6 strain [101,102].
The administration of 4NQO can be either topical or in drinking water. Both methods successfully establish OSCC mouse models [100,[103][104][105][106][107]. Initially, researchers used a topical daub of 4NQO to induce OSCC, which was the same method as using DMBA to induce SCC in hamster pouches [103,106]. Schoop et al. applied 4NQO (5 mg/mL) dissolved in propylene glycol to the tongues of male CBA mice three times a week over 16 weeks [103]. Such treatment produced continued development of hyperplasia; mild, moderate, and severe dysplasia; and squamous cell carcinoma between weeks 24 to 40. This method ensured that 4NQO was localized to the oral cavity and reduced digestive tract exposure. In contrast, using 4NQO-containing drinking water to model OSCC is considered a more natural mode of administration, faster to model, and less painful to mice than topical application [100,104,105,107].
The amount and timing of 4NQO used in the drinking water method varied between studies. The concentrations of 4NQO most commonly used to induce OSCC in drinking water are 50 and 100 µg/mL [47,48]. A higher 4NQO concentration results in faster OSCC establishment. Mice are typically exposed to 4NQO-containing drinking water for 16 weeks (treatment period), followed by normal drinking water for 6-16 weeks (developmental period) [100,104,105,107]. At present, no standardized method exists for producing 4NQOinduced OSCC mouse models, and researchers can adjust these parameters according to the needs of the study. Tang et al. gave CBA mice different concentrations of 4NQO in drinking water: 100 µg/mL (16 weeks), 100 µg/mL (8 weeks), 50 µg/mL (8 weeks), and 20 µg/mL (8 weeks). All mice in the 100 µg/mL, and 50 µg/mL groups formed oral lesions, and the malignancy of oral lesions were 21.0 ± 0.9%, 15.1 ± 0.8%, 10.9 ± 0.8%, and 11.1 ± 0.8%, respectively [101].
The 4NQO-induced OSCC mouse model is widely used in research, given its similarities to human disease at the molecular level and regarding host immune activity, pathological changes, and pathogenesis. This model is particularly beneficial in the development of biomarkers for early diagnosis and translation of the epithelium of the material. Furthermore, the pathogenesis of OSCC can be studied using 4NQO as it can induce primary OSCC and mimics tobacco-related gene mutations [108]. The advantages of primary OSCC is that it closely mimics the tumor microenvironment in humans [100,109] and interacts with cytokines [47], mesenchymal stem cells [110], natural killer (NK) cells [111,112], microbiomes [113], and angiogenesis [114]. As a result, such lesions can be used to investigate the early identification and treatment of precancerous OSCC lesions [94,115,116]. In addition, given the similarities in host immune activity, this model allows for tumor immunology during tumorigenesis and development and related topics such as immunosuppression [49,[117][118][119] and immunotherapy to be investigated [99,120,121]. However, a limitation of this model is the low probability of metastasis and bone invasion, limiting its use for investigating malignant tumor invasion and metastasis. In conclusion, given the model's similarities to human OSCC it is best suited to investigating diagnostic and prognostic markers of OSCC.

Tobacco-Related Chemical Carcinogens-Induced OSCC
Chemical carcinogens associated with tobacco smoke include B(a)P, DB(a,l)P, and NNN. B(a)P is a member of the family of PAHs found in tobacco smoke, charcoal-grilled foods, contaminated water, engine exhaust gas, and soil. DB(a,l)P is a potent carcinogen produced in cigarette smoke. NNN is a nitrosamine generated during the curing and subsequent processing of tobacco. The researchers used these carcinogens to induce tumorous formation in mice and found that cancer incidence in other parts of the mouse body, including the liver, lungs, forestomach, esophagus, and larynx, was higher than in the tongue, thus limiting their application on OSCC models. (Table 3) [58][59][60][122][123][124][125].

4NQO Combined with Other Chemical Carcinogens
Researchers have used 4NQO combined with other carcinogens, such as arecoline or alcohol to induce oral cancer. Arecoline, a betel nut-derived alkaloid, plays a crucial role in oral cancer progression. Alcohol is another risk factor for oral cancer. 4NQO, arecoline, or alcohol alone have little effect on inducing oral cancer in mice, but when arecoline or alcohol is combined with 4NQO, the incidence of tongue tumors in mice can even reach 100% (Table 3) [56,57,[126][127][128][129].
In general, carcinogens cannot induce specific genes during oral carcinogenesis nor can they induce tumors at specific sites, so the CI models can be assisted by xenograft models or transgenic mouse models to meet individual research needs.

Genetically Engineered Mouse Models
Genetically engineered mouse models (GEMMs) include "loss-of-function" and "gainof-function" categories. "Loss of function" uses gene knockout or knockdown technology to deplete or silence target genes such as oncogenes, tumor-suppressor genes, or metabolic genes. "Gain-of-function" uses the knock-in technique to overexpress oncogenes for the function study of oncogenes in vivo. According to the particularity, GEMM is divided into normal GEMM or conditional GEMM. Normal GEMM alters target genes throughout the entire organism. However, this method is inconsistent with tumor formation, given that tumorigenesis is triggered by a single cell in which multiple mutated gene loci inhibit apoptosis and promote proliferation [129,130]. Conditional GEMM is spatiotemporally specific, and the use of this technique can specifically alter target genes in different tissues (space) or at different periods (time). The resulting pathological changes were similar to those of primary human OSCC and mouse OSCC induced by 4NQO. Taking L2D1 + /p53 +/− and L2D1 + /p53 −/− mice as examples, at 5-6 months old, the histological examination demonstrated hyperplasia, hyperkeratosis, severe epithelial dysplasia, and cancer [61]. GEMM features an immunocompetent environment allowing tumors to grow, making this model suitable for immuno-oncology research [70]. C57BL/6 was the first mouse strain to undergo whole-genome sequencing and is considered the "standard" inbred strain that provides the genetic background for many targeted genes. C57BL/6 is often used in genetic experiments as a transgenic mouse model to recreate human genetic defects. In addition to C57BL/6 mice, Balb/c mice can also be used as the genetic background for GEMMs [41,70].

LSL-Kras G12D Mice
Two oral cancer GEMMs are known to be associated with K-ras mutations. Researchers use the keratin 14 (K14) or keratin 5 (K5) promoters, which tend to express in the oral cavity, to drive overexpression of the K-ras G12D oncogene in mouse oral epithelial cells [62,131]. Caulin et al. demonstrated that the K-ras G12D oncogene driven by the K14 or K5 promoter is distributed in a modified recombinase, Cre, fused to a deletion mutant human progesterone receptor, which fails to bind its ligand progesterone but can be activated by progesterone antagonists, such as RU486. The treatment of RU486 caused K-ras G12D overexpression in mouse oral epithelial cells and developed oral squamous papilloma [62]. In an additional model, the K5 promoter-driven expression of the K-ras G12D oncogene is controlled by tet-responsive elements. The administration of doxycycline to mice can also be used to induce overexpression of K-ras G12D [131]. Such overexpression results in dysplasia and SCC affecting the tongue, oral mucosa, esophagus, forestomach, or skin.

L2D1 + /p53 +/− and L2D1 + /p53 −/− Mice
Opitz et al. designed the first GEMM specifically for the development of HNSCC. They used the ED-L2 promoter (L2) from the Epstein-Barr virus to specifically modify genes in squamous epithelial cells in the mouth and esophagus. These modifications produced L2-cyclin D1 (L2D1 + ) mice, which contained L2 fused to human cyclin D1 cDNA. These mice were further crossed with p53 +/− and p53 -/− mice to produce composite mice predisposed to aggressive oral-esophageal SCC. In addition, cell lines generated from the oral epithelial cells of these composite mice could be used to induce tumors in athymic nu/nu mice [61,132] Tgfbr1/Pten double conditional knockout (2cKO) mice can be induced in 4 weeks with tamoxifen (tam) to produce hyperplastic changes affecting the oral epithelium. Furthermore, after ten weeks, visible tumors develop that are well-differentiated SCC, mimicking HNSCC. PTEN is a negative PI3K/Akt pathway regulator and tumor suppressor gene. The interaction between PTEN and the PI3K/Akt pathway is crucial in HNSCC carcinogenesis. In contrast, TGF-β signaling is understood to be both tumor-promoting and tumor-suppressing and affects the carcinogenesis of various cancers. In Tgfbr1/Pten 2cKO mice, the loss of PTEN produces epithelium hyperproliferation and the loss of Tgfbr1 causes immune suppression and increases inflammation-enhanced angiogenesis. Thus, this model suggests that HNSCC carcinogenesis is affected by oncogenic epithelial changes and the tumor microenvironment [63]. In the p53 R172H ; K5.CrePR1 and p53 flox/flox ; K5 CrePR1 models, cutaneous SCC occurred more frequently than oral SCC and there were metastases. In contrast, the incidence of OSCC in p53 flox/flox ; K5 CrePR1 mice was 25% and only 16% in p53 R172H ; K5.CrePR1 mice developed OSCC. In both models, deletion of the Cdkn2a gene enhanced skin tumor formation and promoted metastasis. However, the role of this gene in oral tumors remains to be explored. The gain and the loss of the p53 function were observed to be associated with carcinogenesis in HNSCC in both models [64].
In conclusion, the tumor microenvironment of GEMMs in mice differs significantly from humans due to the presence of transgenes in the stromal cells. Transgene expression can be localized through oral mucosa-specific promoters, such as the K14 or K5, to increase the specificity of the model. However, this does not entirely prevent leaky expression in other tissues. In addition, the natural carcinogenesis of oral cancer is never predominated by a single gene. Therefore, using specific genes (such as Akt or K-ras) to produce tumors in transgenic mice may differ to human oncogenic processes. This observation is aptly demonstrated by the low frequency of K-ras mutations found in human head and neck cancers [133,134].

Humanized Mouse Models
The above syngeneic OSCC mouse models enable mouse tumors with a mouse cell microenvironment to be established, making it possible to investigate the processes of the tumor formation and its interaction with the TME cells. However, to assess the effectiveness of immunotherapeutic treatments, models using both human tumors and immune cells are required as the mouse immune system does not always correspond to that found in humans [135]. The demand for such models is seen in the high failure rate of therapeutics between preclinical mice trials and subsequent human trials [136].
Human xenograft models represent an alternative and are the transplantation of human cells into an immunocompromised host. Athymic nude mice or severe-combined-immunodeficiency (SCID) mice are often used for human xenografts and can be used to test the effectiveness of cytotoxic drugs [137,138]. However, the lack of immunity makes this model unsuitable for the direct investigation of immunotherapy.
Athymic nude mice (classic) have severe T cell dysfunction but still have B cells, as well as neutrophils, dendritic cells (DCs), NK cells, and other components of the innate immune system [139]. SCID mice are deficient in DNA-dependent protein kinases and thus affect T-and B-cell development [140]. SCID mice have a higher human tumor engraftment efficiency than nude mice [141]. In addition, SCID mice are also the first strain to be injected with PBMCs or to be transplanted with human hematopoietic stem cells (HSCs) for the development of humanized mouse models [142,143].
Regarding the host of the humanized model, the premise of establishing a humanized mouse with a human immune system is the construction of immunodeficient mice, followed by continuous optimization and development, because only by destroying the recipient mouse's immune system first can the human tissue or cells transplanted into its body rebuild a functional human immune system. From the earliest nude mice (nude) to the later SCID mice and Rag −/− mice to NOD-SCID mice, these early immunodeficient mice, either due to the presence of innate immunity resulting in a low success rate of human cell engraftment or due to a high sensitivity to irradiation or due to a shorter life span, all have limited the application of humanized mice in research in varying degrees [143][144][145][146][147].
The development of immunodeficient mice ushered in a milestone breakthrough after 2000 AD. It has been found that interfering with the deletion of the interleukin 2 (IL-2) receptor gamma chain (IL-2Rγc) not only causes severe defects in mouse T and B cells but also interferes with the development of NK cells, which can further improve the transplantation effect of human cells into mice. On this basis, a variety of more advanced severe immunodeficiency mice have been bred by researchers around the world and become the most widely used humanized mouse model hosts for immunotherapies, including BALB/c-Rag2 −/− γc −/− (BRG), NOD/shi-SCID γc −/− (NOG), NOD/SCID-γc −/− (NSG), etc. [148][149][150][151].
Regarding the establishment of a humanized model, according to the reconstruction method of the human immune system, two types of humanized mouse models are commonly used: (1) The Hu-PBMC mouse model Peripheral blood mononuclear cells (PBMC) are cells in the peripheral blood with a single nucleus composed of lymphocytes (T or B cells), monocytes, phagocytes, dendritic cells, and a small number of other cell types. It is an important cellular component of the body's immune response function.
The Hu-PBMC model, also known as the Hu-PBL (peripheral blood lymphocyte, PBL) model, is a simple and economic humanized mouse model of the immune system. Constructed by an intraperitoneal (IP) or an intravenous (IV) injection of mature lymphocytes (derived from PBMC) into immunodeficient host mice, this model is often used to study the activation of human effector T cells and to evaluate immunosuppressive drugs.
The preparation period for the Hu-PBMC model is short. After PBMC transplantation, human CD3 + T cells can be detected as soon as one week; in about 2 weeks, the immune cells will be rapidly reconstituted; in about 4 weeks, about 50% of human CD45 + cells can be detected in the peripheral blood of mice, about 90% of them are CD3 + T cells, and the ratio of CD4 + :CD8 + T cells is about 1:1. However, the Hu-PBMC model develops lethal graft-versus-host disease (GvHD), the extent of which is directly related to the engraftment of human T cells, as assessed by weight loss in mice. GvHD generally occurs 2-3 weeks after transplantation, so the experimental observation window is short [152].
(2) The Hu-HSC mouse model Another approach involves the injection of human CD34 + HSCs into immunodeficient host mice, which requires first sublethal irradiation of the host mice to deplete mouse HSCs and to facilitate engraftment of human HSCs. Such models, also called hu-CD34 + , or hu-SRC (SCID-repopulating cell) models, have been widely used to study human hematopoietic development, cell-mediated immune responses, and viral infectious diseases such as HIV and EBV.
Hematopoietic stem cells (HSCs) are a type of cell in human hematopoietic tissues that can self-renew and differentiate into various blood cells. CD34 antigen is widely recognized as a representative surface marker of hematopoietic stem/progenitor cells. It is a highly glycosylated type I transmembrane glycoprotein that regulates cell adhesion and promotes cell adhesion to the bone marrow matrix.
A variety of hematopoietic stem cells can be generated by intravenous (IV) or intrafemoral (IF) injection of human CD34 + HSCs from human umbilical cord blood, bone marrow, G-CSF-activated peripheral blood, or fetal liver into adult immunodeficient mice, but T cells are produced in small quantities and are not functional. Alternatively, by intravenous injection (intracardiac or intrahepatic), the transplantation of human CD34 + HSCs into neonatal recipient mice (less than 4 weeks old) results in good human cell engraftment and production of NK cells, macrophages, B cells, T cells, and DCs. Embryonic liver and umbilical cord blood are the most used sources of human CD34 + HSCs, which are more likely to colonize immunodeficient mice than adult HSCs [152].
For tumor research, the establishment of a humanized xenograft mouse model that allows interaction between the human tumor-immune system and the use of novel mouse models to explore the treatment of immune checkpoint inhibitors are currently emerging development directions.

Humanized Xenograft Mice
Humanized mouse models with co-transplantation of the human immune system and human tumors (xenografts) are effective tools for developing new strategies for tumor immunotherapy. If classified by the source of the graft, xenograft models include CDX, which is derived from tumor cells cultured in vitro, and PDX, which is derived from fresh tumor tissue from patients. If classified by the location of inoculation, xenograft models include both ectopic and orthotopic. In ectopic models, tumor cells are injected subcutaneously into the flank or back of the mouse. In contrast, in orthotopic models, tumor cells are often grafted onto the tongue of the mouse [70,153]. Implanting human tumor cells, CDX, or tumor tissue, PDX, into humanized mice based on the human immune system can better simulate the human tumor microenvironment. It is used to study the growth of tumors in the human immune environment and it can evaluate the efficacy and related mechanisms of anti-tumor therapy, especially immunotherapy [154]. Tumor cell or tissue inoculation is performed on the Hu-PBMC model or Hu-HSC model to establish tumor-bearing humanized mouse models, which can be used to evaluate the therapeutic effects of immune checkpoint inhibitors or dual checkpoint inhibitors: (a) Assessing the effects of immune checkpoint inhibitor therapy A growing number of studies have shown that humanized mouse models of the immune system have unique advantages in the study of immune checkpoint inhibitors; for example, inoculating KHOS cells in Hu-PBMC mice to establish a humanized osteosarcoma mouse model of the immune system. Moreover, using this model to prove that PD-1 inhibitors can effectively inhibit the lung metastasis of osteosarcoma. The anti-PD-1 antibody reduced tumor size by more than 50% in Hu-PBMC mice inoculated with human lymphoma SCC-3 cells or glioblastoma U87 cell models. The humanized mouse tumorbearing model of the immune system is a good platform for evaluating immune checkpoint inhibitors [155].
(b) Assess the effect of combination therapy, e.g., dual checkpoint inhibitors Hu-PBMC mice were inoculated with human colorectal HT-29 cells and gastric cancer tissue and treated with urelumab (anti-hCD137) or/and nivolumab (anti-PD-1). It was found that the combination therapy or monoclonal antibody alone can slow down tumor growth, the combination therapy did not significantly improve the efficacy, and the tumor grafts increased IFNγ-secreting human T cells and decreased human Treg cells. In the Hu-HSC model of EBV-associated lymphoma, the combination of PD-1 and CTLA-4 inhibitors can effectively inhibit the growth of EBV-induced diffuse large B-cell lymphoma and the anti-tumor effect is better than that of monotherapy [156,157].
Morton et al. successfully generated the TME in a humanized PDX mouse model of HNSCC. In xenochimeric mice (XactMice), tumors contained human cells derived from humanized bone marrow and a dynamic microenvironment containing human B and T cell populations, cytokine expression, and lymphangiogenesis. These mice accurately represented the in vivo growth of the original tumor [67]. This HNSCC humanized mouse model was further optimized by Morton et al. [68]. Through the infusion of mesenchymal stem cells (MSCs) and human hematopoietic stem and progenitor cells (HSPCs), the percentage of human immune cells present in the bone marrow of established mice could be doubled compared with mice only implanted with HSPCs. In addition, a 9-38-fold increase in mature peripheral human immune cells was seen for the dual infusion, and these mice had more MSCs, cytotoxic T cells, and regulatory T cells. Therefore, the dual infusion of MSCs and HSPCs produced a higher degree of humanization, which improved the accuracy of the model. The humanized mouse model is an emerging technology and thus still has many limitations. For example, tissue incompatibility may cause immune responses in humanized mice [158]. This suggestion is supported by findings that transplanted human immune cells, predominantly T cells, produce GvHD and attack recipient mice, resulting in death. As a result, only a short period is available to perform experiments [159]. Nonetheless, the humanized mouse is the closest to replicating human disease, thus ongoing development of this technology is needed.

Humanized Immune Checkpoint Knock-In Mice
Humanized xenografted mice can be prohibitively expensive and laborious to develop. However, an alternative method is the introduction of knock-in (KI) human immune checkpoint genes into mice, especially when it comes to testing drugs that target the immune checkpoints of human immune cells or human tumor cells [160].
An antigenic immune response is regulated by a balance between inhibitory and stimulatory signals through signaling pathways of receptors and ligands called immune checkpoints. Several checkpoint molecules, for example CTLA-4 and PD-1, can affect the functionality of both adaptive and innate immune cells [161]. Tumor cells manipulate these pathways by either suppressing stimulatory immune regulators or overexpressing inhibitory immune checkpoints. Immune checkpoints are an area of interest to immunooncology, which seeks to modulate the immune system to identify and destroy cancerous cells [162]. Immune checkpoint blockers (ICBs) are a novel field in drug research and are developed to block ligand-receptor binding between T cells and cancer cells [163].
The differences in the immune systems between humans and mice (60% homology) can cause discrepancies between preclinical and clinical studies. For example, the inhibitory receptors on NK cells that recognize MHC class I molecules differ and do not exhibit crossreactivity (KIRs in humans and Ly49s in mice). Therefore, studying these receptors in immunocompetent mouse models is challenging; murine Ly49s do not bind to human MHC and vice versa [163,164]. Therefore, early research into NK/ILC checkpoint inhibitors depended on immunodeficient mice transplanted with human tumor cells and only one type of human immune cell. Thus, better preclinical models that mimic human cancer development and our immune system will provide a more realistic TME to measure the success of ICBs. To connect the laboratory with clinical translation, several ready-made mouse models in which human immune checkpoint orthologs and immune regulatory genes are knocked in to replace all or part of endogenous mouse genes.
Murine cells can be edited to express a human form of a murine protein while still being regulated by murine elements. Therefore, these edited mice have a single gene alternation to their immune system, retaining the overall intracellular signaling [165,166]. These mice are called human immune checkpoint knock-in (KI) mice. These mice do not possess human immune cells, so they are not viewed as human immune system (HIS) mice. Nevertheless, these mice can be used to assess the anti-tumor and binding and efficacy of antibodies against discreet human checkpoint molecules. The human immune checkpoint KI mice can further be used to establish HIS mice, resulting in humanized immune checkpoint KI mice. These humanized knock-in immune checkpoint mouse models are ideal for producing tumor mouse models for drug testing and basic research. Commercially available mice can be purchased that express human CTLA-4, human PD-1 and CTLA-4, or both human PD-L1 and PD-1 without murine PD-L1 or PD-1 [162]. These models can help assess the direct and secondary effects of targeted antibody therapies on tumor growth in vivo, reduce off-target toxicities, and optimize antibody binding kinetics to reduce the incidence of severe immunotherapy-related adverse events.
The humanized immune-checkpoint mice model has some other advantages as follows: Precise humanized gene insertion: mitigates against any possible interference with the humanized gene expression. The target protein confirmation, in particular the binding region, remains unaffected.
Highly efficient models: such models undergo comprehensive QC and validation using clinical-grade checkpoint blockers to ensure adequate expression and correct signal transduction of human immune checkpoint genes.
Fully functional and potent immune system: the murine immune system is not compromised; instead, specific immune cells will express the human version of checkpoints of interest.

Conclusions and Future Perspectives
Immunotherapies are changing the treatment landscape for oral cancer. Compared with current treatments, immunotherapies improve survival and reduce toxicity. ICIs have now been advanced from second-line to first-line therapy for recurrent or metastatic (R/M) oral cancer. The mouse models used in related studies have various advantages and disadvantages. Syngeneic, GEM, and humanized PDX models all have several limitations. When it comes to recapitulating the heterogeneous anti-tumor immune responses observed in immuno-oncology clinical trials, the limitations mentioned above pose challenges for any single model type. However, further development of more specialized preclinical models will enable us to define their selection to accurately replicate the human malignancies and to better assess responsiveness to immunotherapy [74]. Accurately determining the initial experimental question is crucial when choosing an appropriate model.
Although current research primarily relies on syngeneic tumor models, increasingly complex GEMMs models can accurately replicate the autochthonous tumor microenvironment, enhancing the predictive value of tumor immunotherapy in the preclinical setting. Additionally, advancements in humanized PDX models can potentially develop a "coclinical" approach to engraft patient tumors into preclinical models of humanized immune reconstitution to guide treatment decisions [74]. While generating humanized PDX models has focused on recreating the patient's immune system, a crucial advancement will be incorporating microbiome analysis into preclinical models. Given the sensitivity of the microbiome, which can be affected by, among others, the manufacturer and the study housing environment, it is important to include these variables when measuring the immunotherapeutic response [167].
Another critical area of preclinical tumor models that needs further development is the accurate modeling of immune-related adverse events (irAEs), as available preclinical models do not replicate the severity, kinetics, and nature of the toxicities observed in ICBtreated patients. Because immuno-oncology drugs can present unpredictable patterns of irAEs in patients, identifying the early biomarkers of irAEs and approaches for reversing lethal toxicities are crucial for the safety of immuno-oncology combination treatments in the clinic [168]. Such improvements include using models that have reduced kinetics of tumor regression (to facilitate the development of irAEs), incorporating novel ways to determine toxicity, as well as assessing these treatments in models more sensitive to autoimmunity (Treg-depleted model systems [169]). Because irAEs can have a multitude of manifestations in various tumor histologies, incorporating models that correctly present the irAEs seen in the clinic is of the utmost importance [170].