Implantation models bring many benefits, including low cost and can be used to evaluate various HCC treatments. Implanting stable HCC cell lines into the recipient mice is widely used to establish xenograft models. Cell line xenografts can be sorted into ectopic and orthotopic models according to the implantation location.

Subcutaneous (ectopic) cell line xenograft models represent the most common and easiest model available in the studies of HCC. This model is achieved by implanting HCC cells subcutaneously into the flank of the immunodeficient mice. Immunodeficient mice commonly encompass nude mice, which lack B and T cells, and non-obese diabetic-severe combined immunodeficient mice (NOD-SCID), which have a diminished number of NK cells, NKT cells, and macrophages in addition to deficiencies in complement pathways. Although subcutaneous cell line xenograft models are relatively easy to create, facilitate the monitoring of tumor size alterations, and are effective for screening new compounds, they do not accurately replicate the tumor microenvironment, especially the immune response and immune infiltration. Furthermore, subcutaneous xenograft models infrequently exhibit the metastatic characteristics of HCC, potentially resulting in false-positive drug responses (25).

Albeit establishing an orthotopic xenograft model may be more labor-intensive than creating a subcutaneous xenograft model, it is better suited for preserving the tumor microenvironment (26, 27). Rao Q et al. evaluated four common approaches to establishing the orthotopic xenograft models, including intrahepatic implantation of syngenic tumor tissues derived from Hepa1-6 cells and intrahepatic, intrasplenic, intravenous inoculation of Hepa1-6 cells (28). C57BL6 mice were used for their ability to preserve the integrity of tumor tissues and cell-cell interactions (29, 30). Intrahepatic implantation was the optimal technique due to its 100% success rate, shortest time to tumor formation, highest metastatic rate, and ability to maintain the pathological characteristics in C57BL6 mice (28). Immunohistochemistry assays revealed localized CD3+ and Foxp3+ T lymphocyte infiltration around tumor sites compared with neighboring normal tissues in liver sections of the orthotopically implanted mice, similar to the immune environment in the HCC patients. In addition, they observed a reduction in the serum CD4+/CD8+ lymphocyte ratio, a decrease in serum IFN-γ levels, and an increase in serum IL-10 levels. These findings suggest tumor progression and the establishment of an immunosuppressive microenvironment. Yan Mingxia et al. established stable primary HCC patient-derived cell lines from established subcutaneous patient-derived xenografts with a success rate of 50% (2/4). For three years, the two cell lines stably expressed some of the HCC molecular markers detected by flow cytometry, revealing a high similarity to the original tumor with preservation of intertumoral heterogeneity (31). Therefore, the cell lines exhibited good potential for establishing cell-line xenografts that could maintain some intertumoral heterogeneity. However, a limitation of the orthotopic xenograft model was that tumor size could not be easily monitored without sacrificing the mouse. Yao et al. demonstrated a new method to orthotopically inject Hep3B cell lines into BALB/c athymic nude mice. They reported a concomitant increase in AFP levels with increased Hep3B tumor size, providing a minimally invasive method for monitoring tumor growth by measuring serum AFP levels (32).

It is well-established that patient-derived xenografts can better preserve the original tumor properties than 2D cell line culturing. PDX models are widely used to select new drugs or test new treatment strategies. PDX models are commonly established by injecting patient-derived liver tumor tissues heterotopically (subcutaneously) or orthotopically (into the liver) into the recipient mice (25). Both implanting methods have advantages and disadvantages. Subcutaneous implantation allows easy evaluation of tumor size, while orthotopic implantation provides the xenograft an environment more identical to the original tissue. To avoid rejection of the human cancer tissue, xenograft models often use immunocompromised mice as recipients.

Compared with cell-line xenografts, PDXs can better retain the tumor architecture, vasculature, stromal components and intertumoral heterogeneity ( Table 1 ). Julien et al. reported a very low gene expression difference between colorectal cancer patient samples and PDXs up to the 9^th passage (40). Moreover, PDXs can even preserve the human immune microenvironment for 4-5 passages in the recipient mice before being replaced by the mice cells (41, 42). Liver cancer PDXs were first established in 1996, and since then, many researchers have been dedicated to exploring ways to improve the resemblance of liver cancer PDXs to the original tumor tissues (30). Xu Wei et al. achieved a success rate of 28.1% (9/32) in HCC patient-derived subcutaneous xenografts and a success rate of 16.7% (1/6) in the subsequent passages of subcutaneous xenografts derived from the first xenograft tumor (33) while retaining the original tumor pathological characteristics. Armengol et al. successfully established 5 viable PDX models from 10 surgically resected HCC tumor tissues, as reported in their study (36). To verify the human origin of the passaged xenografts, they checked the microsatellite instability of two markers, D12S95 and BAT26. Yan Mingxia and colleagues collected resected tumor tissues from 24 HCC patients, implanted them subcutaneously into NOD/SCID mice with a success rate of 20.83% (5/24), and observed an increasing tumor take rate with the passages of mice (31). They also reported a tumorigenesis rate of 100% (8/8) by establishing orthotopic PDXs in the left hepatic lobe of BALB/c nude mice. Gu et al. established a stable cohort of 65 PDXs in BALB/c athymic mice out of 254 HCC patients (36). They demonstrated that the PDXs could accurately replicate the original tumor at both histological and gene expression levels while preserving intratumoral heterogeneity. Moreover, 81.3% (26/32) of the tested models showed detectable serum AFP, suggesting that PDXs have the potential to reflect the clinical characteristics of HCC on a large scale. Bissig-choisat et al., besides reporting the preservation of histologic, genetic and biological features of liver cancer PDX, also demonstrated the preservation of metastatic behavior (34). Gao and colleagues reported an intrahepatic metastasis rate of 75%(6/8), a bone metastasis rate of 37.5%(3/8) and a lung metastasis rate of 37.5%(3/8) in their HCC PDX models (43). Due to the increasing success rate of xenograft implantation and the faithful preservation of the original tumor features in liver cancer, PDX models have become reliable tools for validating current therapies and testing new therapies. Huynh and colleagues established HCC PDX lines and found that most of the current chemotherapeutic drugs for HCC yielded little or no antineoplastic effect in vivo ( 37). Hu et al. explored the antitumor effects of Hsp70-expressing oncolytic virus between cytokine-inducer killer (CIK) and non-CIK-infused mice using xenograft models derived from 10 HCC patients (38). A synergistic inhibitory effect was found with the co-expression of specific Hsp70 and CIK infusion.

Yet due to the differences between humans and mice, immune infiltration levels are also different in the TME. For instance, NK cells are the most abundant in the human liver, whereas NKT cells are the most abundant in the mouse liver (35, 44). Moreover, PDX models commonly use immunocompromised mice lacking some immune cell subsets, which can hardly mimic complete immune response, lymphangiogenesis or chemokine signaling. Additionally, PDX models can only preserve the human immune microenvironment for a short time, making long-term studies difficult. Moreover, different species bear different immune-cell-specific antigens, which leads to invalid results in immune-related target site selection (45). As a result, PDX models provide limited results in discovering immune-cell-related therapies.

However, recent scientific advances, such as the immunologically humanized mouse model, can reduce the gap between the two species. The model is modified to mimic the TME by expressing human immune cells and is commonly established through transferring human hematopoietic stem cells (HSCs), or peripheral blood mononuclear cells (PBMCs) derived from umbilical cord blood, fetal liver, bone marrow, or GM-CSF mobilized peripheral blood mononuclear cells into the bone marrow of the sublethally irradiated NSG mice. After developing a functional humanized immune system, PDX can be implanted subcutaneously or orthotopically into the recipient mice, recapitulating the immune microenvironment of the original tumor. Zhao et al. successfully established humanized immunity in NSG mice with HLA-I-matched HSCs (46). They found larger tumor sizes in the humanized PDX mice than in the NSG PDX mice, indicating a possibility of the tumor transforming the immune environment to facilitate tumor growth. Besides, they observed a significant decrease in infiltration levels in major human immune cells during xenograft tumor development in humanized mice. In addition, proinflammatory cytokine (IFN-γ, IL-2, IL-18, TNF-α) and cytolytic protein (granzyme A, granulysin) levels were analyzed and showed an initial increase during the early stage of the tumor and then decreased, corresponding to the clinical features of HCC patients. Lan and colleagues developed a novel humanized mouse model (BLT) that could establish sustained human hematopoiesis and functional human immune response by co-transplanting CD34+ hematopoietic stem cells (bone marrow, B), fetal liver (L) and thymus (T) into NOD/SCID mice (39). They reported that Thy/Liv/CD34+ mice could repopulate multilineage immune cells (B cells, T cells, dendritic cells) compared with a repopulation of mostly CD3+ T cells in Thy/Liv mice. Moreover, a strong in vivo immune response was observed in Thy/Liv/CD34+ mice as skin xenografts were rejected. Thus, humanized PDX model provides a promising tool for studying in vivo HCC behaviors and human immune responses.

Advances in stem cell technology have contributed significantly to cancer research and in vitro preclinical models. In 2007, Takahashi et al. first demonstrated the generation of induced pluripotent stem cells (iPSCs) from human skin fibroblasts with 4 defined factors: Oct3/4, Sox2, Klf4, and c-Myc (47). IPSCs have a great self-renewal capacity and differentiation potential, providing a platform for studying pathogenesis, drug screening and regenerative medicine. Moreover, patient-specific iPSCs hold the potential for treating liver cancer since these cells maintain the genetic background of their donor.

According to Kim and colleagues, pluripotent liver cancer cells were produced from four HCC cell lines using the retroviral introduction of genes associated with reprogramming, which displayed distinctive colony morphology and tumor marker expression compared to the original tumors, and demonstrated pluripotency by expressing multiple marker genes for pluripotency (48). Afify and colleagues established liver cancer stem cells (CSCs) from iPSCs by culturing in a conditioned medium (CM) for the HCC cell line Huh7. This CM could replicate a microenvironment like chronic inflammation, generating CSC without genetic manipulation (49). These studies provide methods to produce liver CSCs, defined as a tumorigenic subpopulation in liver cancer, contributing to tumor metastasis and recurrence (50). In this way, researchers can depict basic and oncological features of liver CSCs and elucidate molecular mechanisms underlying CSCs development and cancer progression. The efficacy of novel therapies has also been tested with iPSC-derived liver cancer models. In a recent study, researchers knocked down p21 of human iPSCs and observed tumorigenicity during induction differentiation of iPSCs to hepatocyte-like cells. Upon combination therapy of acyclic retinoid with AKR1B10 inhibitor, hepatoma-like cells were induced into normal hepatocytes (51).

In summary, some early attempts have been made to apply iPSCs in liver cancer modeling and therapy exploration. However, as iPSCs fail to preserve the 3D architecture and local TME and liver cancers are complex diseases evolving based on versatile genetic and environmental alterations, their potential applications are limited.

Initially developed by Smith and colleagues in 1985 for toxicity testing, precision-cut liver slices have since expanded their applications beyond toxicology and are now commonly used to model chronic liver conditions (52, 53). PCLS model preparation starts by slicing tissue cylinders with a 5-8 mm diameter into a reproducible thickness, usually 250-300 μm, which allows the diffusion of nutrients and oxygen to the inner cell layer. Subsequently, the liver slices are put in continuously submerged or dynamic culture systems and can keep viable for several days in optimal conditions (54). In PCLS models, various cellular components of liver tissue are preserved, including hepatocytes, hepatic stellate, Kupffer and endothelial cells, and functionality of all liver cell types enables research on liver function from a multicellular perspective. Moreover, unlike traditional in vitro models, PCLSs maintain the original extracellular matrix, including proteoglycans, glycoproteins and collagens of different types with the initial alignment of cells and cell-cell and cell-matrix communications. Furthermore, PCLS models offer a promising approach for studying the process of liver carcinogenesis and may provide early predictive value for HCC diagnosis, as they preserve in vivo cell-cell and cell-ECM interactions on a multicellular basis, thereby bridging the gap between in vivo and in vitro models. Currently, PCLS models have been widely used to establish versatile ex-vivo models of liver disease, including ALD models, NAFLD models, viral hepatitis models and liver cancer models (55–58). Overall, PCLSs are robust and reliable tools for studying mechanisms underlying liver injury and discovering novel therapeutic strategies.

Several studies have employed precision-cut liver slices to evaluate patient-specific responses to anti-cancer therapies in primary liver cancer ( Table 2 ). In 2006, Kern MA and colleagues added selective cyclooxygenase-2 inhibitor Meloxicam to a surgical resected PCLS model of HCC and confirmed the antitumor effect of COX-2 inhibition with significantly increased tumor cell apoptosis and reduced tumor proliferation. Their research demonstrated the potential of the PCLS model in analyzing apoptosis at the tissue level and assessing drug effect through direct comparison with the ‘unaffected’ non-tumorous tissue of the same patient (63). Similarly, Zhang et al. reported the application of PCLS in systematic drug screening, using a method combined with cryopreservation and improved cell viability (59). The effect and safety of oncolytic viruses have also been tested with PCLS models. Zimmermann et al. established a series of PCLS with tissue from primary and secondary liver tumors and infected them with the oncolytic measles vaccine virus (MeV). With PCLS, a multicellular model with original ECM, the penetration and spreading capabilities of MeV were measured, and comparative testing of genetically variant MeV vaccine strains was enabled, bringing hope to the preselection of oncolytic viruses for virotherapy in a patient-specific manner (61). Besides drug response prediction with PCLS, other studies emphasized the investigation of tumor immunology, inter- and intra-tumoral heterogeneity and growth properties of liver cancer (60).

Overall, PCLS provides a versatile tool for liver cancer study with advantages in preserving complex 3D architectures and cell-cell interplay compared with traditional cell cultures. Additionally, it is easier to build and capable of presenting the immune landscape in human malignancies efficiently compared with organoids and xenografts. However, there are a few drawbacks to using the PCLS platform. Though the presence and metabolism activity of various cellular components in PCLS is confirmed when established, unexpected proliferation and functional deterioration may occur during incubation. Furthermore, the slicing process during the preparation of PCLS may bring out unavoidable damage and trigger a repair and regenerative response, resulting in fibrosis during culture. Additionally, the lifespan of PCLS is rather short, and the model is non-renewable, which hinders long-term study with the platform and limits the reproducibility of experiments, calling for other patient-derived models that are easier to maintain and available for biobanking.

In recent days, the development of 3D cultures has enabled the establishment of novel in vitro cancer models that resemble the primary tumor epithelium that they derived from genetically and phenotypically. Organoids are self-organizing 3D structures that mimic the original in vivo architecture of tumors and have shown promising applications in precision medicine (62). PDOs are organoids derived from patient samples collected through surgical resections or biopsies. To culture PDOs, the tumor tissues undergo physical or enzymatical disassociation first and are subsequently embedded in an extracellular matrix (ECM), growing in the specific culture media containing growth factors and/or inhibitors required by that tissue (64). The choice of biological or synthetic scaffold mimicking the ECM depends on the tissue exhibiting different porosity, permeability, surface chemistry, and mechanical characteristics (65). Through a combination of various ECM and growth factors, researchers can now mimic the native TME in PDO, including cell-cell and cell-matrix interactions, which are lost in traditional 2D cell culture. Specifically, emerging PDOs that recapitulate TME provide more accurate and versatile tools to testify existing therapies, discover new drugs, and guide personalized treatment plans in PLC.

In recent years, considerable efforts have been devoted to developing alternative 3D culture methods that can overcome the limitations of organoids in onco-immune research, such as the absence of native stromal and immune cells. Moreover, the capability of PDOs in guiding clinical drug decisions is still restricted as it takes time to generate enough organoids for rapid drug screening.

Multicellular tumor spheroid (MCTS) is a 3D model whereby tumor cells grow as spherical colonies in suspension culture, with supplemented cell types to mimic the complex tumor tissue microenvironment. Compared to other preclinical models, spheroid models offer several advantages, including ease of maintenance, preservation of in vivo tumor growth kinetics and chemoresistance, and suitability for high-throughput drug testing (99). Yeonhwa Song and colleagues demonstrated an MCTS culture system in liver cancer, using patient-derived HCC cell lines and stromal cells from human hepatic stellate cells, human fibroblasts, and human umbilical vein endothelial cells. Drug testing revealed a clear selective response to sorafenib, 5-FU and cisplatin among MCTSs, indicating its potential role in optimizing individualized treatment (100).

In 2018, Jenkins et al. described a new 3D microfluidic patient-derived organotypic tumor spheroid (PDOTS), which preserved autologous, tumor-infiltrating immune cells. Fresh tumor specimens from patients underwent physical and enzymatic dissociation first, and spheroids ranging from 40–100 μm were filtered for subsequent PDOTS culture and profiling. The 3D microfluidic culture was enabled through a single “3D cell culture chip” presented with three self-contained microfluidic chambers. In microfluidic PDOTS, the response to PD-1 blockade and novel combination therapies was examined, suggesting that CDK4/6 inhibition augmented the response to PD-1 blockade (101, 102).

However, to guide drug selection and immune profiling in real-world clinical settings, the patient-derived model should be easy to grow from minimum tissue in a short time in addition to preservation of native TME. Shen et al. designed a novel clinical-biopsy-derived pipeline-patient-derived micro-organospheres (MOSs), which could be applied to evaluate antitumor drug response and to identify new treatment options in less than 14 days, a satisfying timeframe for guiding clinical treatment. Micro-organospheres are based on microfluidic technology and are characterized by their small size and large surface-to-volume ratio, which make T-cell infiltration into MOS less problematic. Functional analysis revealed that MOS could be used for modeling the response to personalized chemo, targeted, and radiation therapies. Moreover, the correlation between MOS assay readout and clinical outcomes has been validated in a pilot clinical trial involving eight metastatic colorectal cancer (CRC) patients (103). An automated MOS imaging pipeline combined with machine learning has been recently established, delivering on the promise of rapid tissue prototyping and high-throughput therapeutic profiling (104).

To provide a clear comparison, we summarized the advantages, disadvantages, and applications of each patient-derived model ( Table 4 ).