Advances and challenges in human 3D solid tumor models

Abstract

The field of cancer biology and therapeutics has soared in the past several decades with new therapeutic modalities and options for patients, such as chemoradiotherapy, immunotherapy, and combination therapy. This dramatic success in expanding patient options is primarily attributed to the development of various model systems to elucidate drivers of oncogenesis, tumor maturation and evolution, and response to therapeutics. While mouse models have been a workhorse of cancer research, technological progress in ex vivo patient-derived tumor models has afforded more tunable and scrutable systems for patient-predictive platforms and mechanistic study. This review explores the technological innovations in 3D solid tumor models and their applicability to various aspects of cancer biology and identification of therapeutics. Features of the tumor and tumor microenvironment like spatial heterogeneity, multicellular populations and genomic variations are addressed and elaborated through the establishment of new in vitro models. We further address the integration of perfusable vasculature with 3D tumor models and the potentially wide-ranging applications of these more complex platforms in precision medicine and cancer immunotherapy. Finally, we provide an outlook on the future of experimental cancer models for both biological investigation and bench-to-bedside pipeline development.

Introduction

Cancer is the second leading cause of death, globally. The history of cancer therapy has ranged from surgical removal of the tumor tissues to chemotherapy and radiotherapy, and to the more recent immunotherapies which leverage the endogenous patient immune system or engineered immune cells to fight cancer. However, even with the significant clinical advancements and therapeutic options available, mortality and recurrence are extremely high, whereby patients either do not entirely respond to therapies or relapse after a period of time. Although early screening and improvements in therapeutic regimens have ameliorated prognosis and reduced mortality, enhancing patient response rates and sustaining remission remain significant challenges. Reliable experimental models that provide the key cellular and structural elements of cancer and features specific to the patient play a vital role in improving cancer treatments and therapeutic outcomes. Mouse models are one of the most widely utilized preclinical models of cancer therapies, including chemo/radiotherapy, immunotherapy, and combination therapy [1, 2]. Mouse models have been instrumental in elucidating the interactions between oncogenic and tumor suppressive genes commonly found in cancer, such as KRAS and P53 [1]. Besides transgenic mouse models used to recapitulate the etiology of various cancers such as lung or colorectal, patient-derived xenograft (PDX), orthotopic, and syngeneic mouse models are also employed to identify potential cells of origin, evaluate therapeutic interventions, and study the evolutions of cancer cells through their mutagenic trajectory [3, 4]. Though mouse models have significantly advanced our understanding of cancer and facilitated therapeutic development and testing, there are multiple shortcomings associated with the widespread applicability of these models. These include disparities between the mouse and human genome and regulatory elements, as well as the difficulty in generating certain mouse models such as the PDX [5–7]. In addition, the evolution of tumors in mice (i.e., organotropisms) differs from humans, as does the associated tumor immunology [8].

In the recent years there has been a surge in interest to design new culture platforms and devices for ex vivo models that recapitulate the human tumor etiology more closely. Advancements in stem cell biology, in vitro cell culture, and biomaterial fabrication have led to major breakthroughs in the realm of in vitro models of cancer. These emerging models treat tumors as analogous to organs with complex microenvironments including the stromal compartment containing extracellular matrix (ECM), cancer-associated fibroblasts (CAFs), the immune compartment, and vasculature. These contemporary multi-faceted models are conceptualized to bridge the gap between mouse and human, enabling the study of the etiology of human cancer disease progression in a defined and temporal manner.

Human cellular models were first developed using cell lines derived from human cervical cancer tissue (HeLa). Since then, these models have been extensively used to study cellular responses to interventions like radiation and various drug treatments, mechanisms of tumor development and chemotherapy efficacy, viral oncogenesis (e.g., HPV), and telomere biology [9–14]. Over time, multiple cancer cell lines have been established and widely utilized in cancer research [15, 16]. Indeed, high-throughput screening for drug discovery using cancer cell lines, enabled by the large numbers of otherwise isogenic cells, led to identification of numerous drugs [17]. Such in vitro screens also yielded insights into pathways involved in cancer therapy resistance as well as interactions between cancer cells of particular genetic backgrounds (i.e., synthetic lethality) [18–20]. While the 2D monolayer model is a valuable tool for studying drug resistance and sensitivity, it does not address the 3D gestalt that is an organ or tumor, effectively neglecting parameters like cell heterogeneity and cell-cell interactions, and contributions from the dynamic extracellular matrix (ECM). Figure 1 schematic depicts the tumor microenvironment composed of the heterogenous cell types and vasculature, as well as structural and secreted factors. Furthermore, the lack of the tumor microenvironmental (TME) components such as the immune compartment in the conventional 2D systems can significantly affect outcome measurements [21]. These avenues undoubtedly require the complex three-dimensional architecture as well as heterotypic cell-cell interactions. This has motivated 3D in vitro models that recapitulate various aspects of tumors, including heterogenous cell populations and the use of biomaterials with relevant physicochemical properties [22–24]. Besides better biological understanding, these models could serve as platforms for drug screening, discovery, and pre-clinical testing. Thus, the use of 3D cancer models, including spheroids, has blossomed in the past decade. Initially developed for drug and radiation testing, as well as for studying metastasis and its dependence on the TME, these models are now being leveraged to examine tumor-immune interactions and to predict patient response [25–27]. Recent advancements in organoid cultures have further opened new avenues for maintaining patient avatars to enable bench-to-bedside pipelines [28, 29]. These 3D models indeed enable the meticulous study of tumor evolution both spatially (i.e., primary to metastatic) and biologically (i.e., therapeutic resistance) [30–33].

Figure 1: Elements of the tumor microenvironment.

The tumor microenvironment is a collection of both heterotypic cells in multiple states and physicochemical factors resultant from both tumor cells and associated microenvironmental cells, driven by mutant cancerous cells seeking to expand and extravasate. The immune cells involved include CD8⁺ T cells, CD4⁺ T cells, regulatory T cells, B cells, neutrophiles, macrophages, and others. Their cell-cell interactions through physical or chemical means can affect CD8⁺ T cell infiltration and cytotoxicity. Cancer-associated fibroblasts can create stiffened stroma-tumor interfaces that hamper orthogonal T cell infiltration into the tumor islet, as well as secrete cytokines that render aberrant vasculature. In the dashed yellow box is an example of the effect of tumor vasculature-associated endothelial cell CLEVER-1 expression on T cell-tumor homing; T_reg = Regulatory T cell, CAF = cancer-associated fibroblast. Schematic made with Biorender.

Besides cells, the extracellular matrix properties also play important roles in tumor growth and immune cell infiltration, therefore motivating the incorporation of tumor/organ-specific ECM features into these engineered models. While there exist targeted efforts to use biomaterials approaches for replicating the ECM properties of the TME, we focus on the cellular aspects of recent 3D in vitro models and their applicability to cancer therapeutic research. We further guide readers to recent reviews regarding biomaterials tools to aid in 3D in vitro model development [34, 35]. This review particularly discusses various ex vivo tumor models ranging from native tumor tissue and/or cell culture to contemporary organoid and vascularized multicellular systems. These models culminate in a large set of tools to address various problems in both experimental tumor models and applications towards understanding cancer biology and identifying new therapeutic regimens including immunotherapy to advance precision medicine. Figure 2 depicts how ex vivo and in vitro cellular models of tumor could contribute to the pre-clinical and clinical studies to identify the optimal therapeutic regimens.

Figure 2: Three-dimensional models of the tumor and various applications in research and the clinic.

Ex vivo systems like tumor slices and explants are particularly well-poised for studying spatial localization of cells upon various perturbations (i.e., drug treatment), as well as clinical characterization with histology. Given the representation of tumor cells and tumor-associated cell types, the secretome of slices or explants could also potentially unveil key regulators of patient response to therapeutic regimens or immune cell infiltration into tumor epithelial islets. Spheroids are relatively easy to establish in in vitro culture, and thus could potentially be used for high-throughput screening or drug testing. Given their integrability with other cells, they could also be utilized to establish tumor-immune models. Organoids are particularly useful for biobanking and patient stratification, as the genomic and histological characteristics of the native tumor are conserved. Further, use of organoids for discovering new drug targets is an active avenue of research in oncology. The outer wheel illustrates the most direct applications of the various 3D tumor models, however there can be overlap between applications and models. Schematic made with Biorender.

3D cellular models of the tumor

Ex vivo models:

The great George Box coined the statistical mantra, “All models are wrong, but some are useful.” This sentiment, in addition to describing the utility of computational models, is entrenched in experimental modelling of the tumor. An ex vivo model of the tumor requires an adequate representation of the key elements relevant to the research question or development and validation of therapeutic intervention(s) (e.g., druggable targets, identification of treatment regimens). Thus, the most parsimonious of the tumor models is the ex vivo culture of intact tumor tissues (aka tumor explants) in which the cellular and structural milieu is retained. However, direct use of biopsied and excised tumor tissues is challenging due to the scarcity of material. This is often addressed by methods similar to early studies that selected for tumor-colony-forming cells [36–42]. These very initial efforts to directly profile patient tumor tissues relied on mechanically and enzymatically digesting tumor tissues to release cells. Though these cultures were considered “ex vivo,” these were mostly in vitro 2D cell cultures which selected for anchorage-dependent cellular growth. While these in vitro workarounds were necessitated by limitations of long-term explant culture (>1 week) and limited availability of tissue samples, they did not address the 3D structure and cellular crosstalk within the tumor tissue. In order to faithfully culture the tumor tissues ex vivo, the heterogenous cell populations should be retained and maintained long-term. However, during ex vivo culture, often the cells migrate out of the tumor bulk and form a monolayer which selectively enhances certain cell populations over others. The tumor tissue cultures often experience tissue disintegration. To address these challenges, many have utilized embeddings of tumor tissue in hydrogels (e.g., collagen gels) for enhanced stability of the cultures [43–46]. Hydrogels with tightly woven mesh networks are expected to confine the tumor bulk tissue and prevent cell efflux, thus offering structural confinement and long-term stability of the tumor tissue cultures in vitro. The choice of media is additionally challenging for these systems due to the potential of certain media formulations to preferentially select specific tumor-associated cell types over others, consequently influencing assay readouts.

The patient-derived explant (PDE), a hydrogel-embedded platform, has been explored for various applications [43–46]. PDEs are tumor resections which are cultured either in suspension or embedded within a hydrogel. PDE retention of cellular and acellular TME components makes it an attractive ex vivo pre-clinical model. These patient-derived explants have been used to study lymphocyte infiltration, drug response, and to retrospectively study gene and protein expression with immunohistochemistry (IHC) to identify correlations between patient response and specific drugs [47]. PDEs have also been demonstrated to capture patient-to-patient heterogeneity in chemotherapy response across different solid tumors like breast cancer and colorectal cancer [48, 49]. In concert with advances in single-cell resolution data acquisition such as flow cytometry, as well as high-sensitivity - and high-throughput - immunoassays (ELISA, multiplexed antibody arrays), PDEs are highly poised for comprehensive analyses of cell types and drug treatment-induced cell states.

In the recent days, PDEs are also being used for testing immunotherapies such as PD1 blockade [50]. Specifically, these studies have shown relevant cytokine release and the corresponding T cell activation from the PDEs that was further correlated with patient response to PD1 blockade treatment. Furthermore, PDEs from multiple patients were used to stratify patient cohorts into immune-inflamed, immune-desert, and immune-excluded. Among the immune-excluded group, the responders and non-responders were further differentiated. Together these findings enabled the characterization of T cell states between the patient populations and the identification of differential protein expression such as CTLA-4, PD-1, Tim-3 and TIGIT among the different patient cohorts and thereby correlating data matrices to patient outcomes. Importantly, the parallel use of excised tissue-based ex vivo models and clinical characterizations (e.g., histopathological, sequencing, biomarker) can further lead to identification of efficient therapeutic regimens and datasets to generate predictive models. For example, studies by Voabil et al. showed that the presence of tertiary lymphoid structures in the PDE was highly predictive of anti-PD1 therapy response [50]. These histopathological measurements in addition to cytokine profiling could culminate in large cohort-specific datasets that can be used to inform machine learning models to predict patient clinical response to various therapeutic regiments.

The patient-derived tumor slice culture is another extensively used tumor tissue-based ex vivo model. Tumor slices are thin sections (< 500μm) of tumor tissue cultured in air-liquid interfaces with or without hydrogel support. Indeed, several groups are using tumor slices to pre-screen drugs for patient response, identifying robust sensitivity for response domains, combination therapy efficacy, and correlation to clinical response [51–54]. Tumor slice cultures of hepatocellular carcinoma have shown strong associations between patient drug responses and the tumor slice pathophysiological features [55]. For example, in hepatocellular carcinoma tumor tissue slices, CEA expression was found to be negatively correlated with clinical response to conventional chemotherapy. Such association provides a prognostic tool for therapy response [55]. Investigating baseline pathological signatures of tumor from the tumor slice culture in the same vein as tumor genome sequencing is thus very encouraging towards the promise of precision medicine. Like the parallel assays used in PDEs, tumor tissues can be simultaneously processed to perform IHC and TCR-seq in addition to employing the slice culture to test various therapeutic regimens and dosing studies [56]. Using this method, it was shown that T cell activation by PD-1/CXCR4 combination blockade could enhance tumor cell death. Together the findings can be used to develop predictive models towards immunotherapy.

In addition to preclinical drug testing, a deeper understanding of tumor architecture and its role on T cell localization/infiltration can be achieved using tumor slice cultures. By simply setting autologous T cells isolated from fresh human lung tumor samples onto the corresponding hydrogel-embedded tumor slices, researchers uncovered a role of stroma and changes in ECM physical properties on T cell infiltration into the tumor nests [57]. T cells preferentially localized within the stromal regions and migrated parallel to the tumor-stroma interface rather than infiltrating the tumor nests. Thereafter, leveraging CCL5-expressing tumor cells from a PDX was shown to stimulate improved T cell infiltration. Besides the tumor cells’ secretome and its effect on T cell infiltration, this study also demonstrated that ECM surrounding the tumor nests can act as a physical barrier to immune infiltration. Further, enzymatic disruption of the ECM with collagenase increased the number of contacts between T cells and tumor cells. Tumor slices are also well-suited to study heterotypic interactions given the maintenance of heterogenous tumor-associated cell populations. Towards this, it was shown that macrophages play a fundamental role in mediating T cell exclusion through long-lasting cellular contact with infiltrating T cells [58]. Thus, by chemically depleting macrophages, T cells exhibited both greater stromal motility and activation by anti-PD1 immunotherapy.

The scarcity of sample and the inability to expand PDEs or tumor slices preclude the longitudinal studies potentially relevant to understand cancer evolution and/or patient therapeutic response. Patient-derived xenografts (PDX) represent an alternative ex vivo model, and its establishment involves grafting patient native tumor tissue into immunodeficient host mice. These PDX could become vascularized by the host and undergo serial passaging of the established tumors across mice for long-term culture [7, 59]. However, PDX models remain difficult and require longer time periods (2–3 months) to successfully establish, and present difficult-to-deconvolute murine contributions to the outcomes of interest [5, 6, 60, 61]. Nonetheless, PDX tumor slice cultures have been used for predicting patient drug response, displaying a spectrum of sensitivity to common chemotherapies [51, 62–64].

Together, the afore-discussed studies demonstrate the potential application of patient-derived tumor tissues models for a multitude of drug treatment and cell profiling assays, of clear benefit to preclinical and precision medicine practices. However, these ex vivo systems can be highly variable, depending on the quantity and quality of the excised biological tissues, the media used to culture the tissues; they also suffer from potential areas of intratumoral hypoxia or necrosis, and a lack of vasculature [65, 66]. In addition, single biopsies of metastatic lesions could be potentially irrelevant to patient tumor biology and therapeutic efficacy. Nonetheless, efforts are being made to standardize tumor slice and PDE cultures, as well as to develop cryopreservation and biobanking methods for downstream analysis with varying levels of outcome [67–71].

Spheroids:

A more versatile method of culturing tumor materials ex vivo involves enzymatically and mechanically digesting the tumor tissue to single cells and creating cell clusters commonly known as spheroids. These emerging studies utilizing primary patient tumor cells to form spheroids represent a significant advancement in precision medicine and technological development for oncology, contrasting with conventional efforts that primarily utilize cell lines. The methodology for generating patient-derived primary tumor spheroids utilized traditional techniques, such as the hanging drop method and rotary cell culture, which are widely used for creating cell line-derived spheroids [72]. In addition to long-term culture, spheroids are well-suited for several experimental interrogations and modalities, such as understanding the mechanisms of tumor growth in various microenvironments, and co-culture with other tumor-associated cells like immune or stromal cells [73]. A number of studies have shown the potential of the spheroids derived from a patient’s tumor tissue, called patient tumor-derived spheroids, to predict clinical response to different therapeutic regimens. For example, patient-derived spheroids have been shown to differentially respond to chemo/immuno-therapy based on tumor genomic background and cellular distribution. This differential response has been observed in various tumor type including non-small cell lung cancer, ovarian cancer, breast cancer, and most recently in bladder cancer [74–77]. That the spheroids with larger diameters present hypoxic/necrotic cores after reaching 200–300μm diameter, due to the mass transport limitations, offers a unique opportunity to model hypoxia gradients. Indeed, numerous studies have leveraged this intricate limitation of oxygen transport across the tumor 3D models or hypoxia gradient to mirror the spatiotemporal oxygenation gradient observed in the native tumor, which can affect the effectiveness of both chemo- and radiotherapies [78–80]. Hence, spheroid models are uniquely positioned to study drug diffusion and resistance often recapitulating the patient tumor, but also might require vasculature to model drug infiltration [81]. In a recent study, the correlation between hypoxic gradients and drug resistance in colorectal cancer was investigated using cell line-derived tumor spheroids [82].

While spheroids present a more versatile 3D tumor model, the issue of dominant cell types emerging through exposed culture in an ex vivo tissue culture persists. The media formulations utilized for spheroid cultures significantly affect the representation of cell populations within them. For example, a recent study has shown that non-small cell lung cancer (NSCLC) specimens from different excising procedures (i.e., biopsy, pleural effusion, surgery) can spontaneously form patient-derived tumor-like cell clusters (PTCs) when cultured in media supplemented with a cocktail of growth factors like hepatocyte growth factor, epidermal growth factor, and fibroblast growth factor [83]. These PTCs containing tumor epithelial cells, fibroblasts and immune cells maintained the features for at least three weeks. PTCs were not only able to recapitulate the structural and genomic features of the native tumors, but they were also used to stratify patients both retrospectively and prospectively to different therapeutic interventions. Interestingly, the immune cells within the PTCs could be reinvigorated with PD1 blockade and thus also functioned sufficiently to predict patient response to checkpoint blockade. Although, while the PTCs could predict different patient outcomes, they failed to predict the response of patients presenting stable disease upon treatment. This patient population with stable disease manifests from a mildly effective drug against the specific tumor, and the low predictive capacity of PTC for this population indicates the lower sensitivity to nuanced patient response. This study suggest PTCs are exceptional at predicting the extremities of patient response, but less so for more spectral and dampened therapy resistance. This may be due to the stochastic nature of cell aggregation in PTCs, neglecting biological/deterministic associations of tumor epithelial, stromal, and immune cells. If the representation of the true patient tumor is highly variable among the PTCs tested, the corresponding sensitivity of PTC drug treatment assays for therapeutic edge cases will be low. A potential solution to this disparity could be addressed by actively incorporating tumor-associated cell types with patient-derived tumor spheroids as in multicellular tumor spheroids (MCTS) described below.

Extensive research has pointed to the contribution of tumor-associated cells like stromal and immune cells on therapeutic response to chemo/immuno-therapies [84–86]. Disparities in multicellular populations between the spheroid and native tumor can lead to inaccurate outcome measurements. Thus, improvements on patient-derived spheroids to be used for assays like drug screening primarily utilized incorporation of other cell types (co-culture) to create spheroids with heterotypic cell populations commonly termed as multi-cellular tumor spheroids (MCTS). For example, liver MCTS comprised of primary hepatocellular carcinoma cells, fibroblasts, endothelial cells, and hepatic stellate cells were developed [87]. These MCTS demonstrated strikingly differential sensitivities to 5-fluorouracil (5-FU), sorafenib, and cisplatin, at times even displaying opposite response to 2D cultures and corresponding monocellular or homotypic spheroids. Complementing this study, studies have incorporated fibroblast cells into cancer spheroids to identify drug toxicities against stromal cells [88]. These studies have identified a therapeutic index parameter to differentiate drugs specific to cancer cells vs. stromal cells.

MCTS have also been deployed for tumor-immune modeling. The first use of this approach for probing immune infiltration was developed in 2017, using colon adenocarcinoma cell lines, fibroblasts, and peripheral blood mononuclear cells (PBMCs) [89]. This study showed selective targeting of T cell bispecific antibodies to tumor cells or fibroblasts within spheroids, and subsequent activation of PBMCs to achieve cell-specific apoptosis. A later effort used colorectal cancer spheroids in co-cultures with allogeneic T or NK cells to assess NK cell-enhancing interventions on tumor spheroid infiltration [90]. Further using autologous tumor-infiltrating lymphocytes (TILs) with matched patient-derived spheroids, this study identified NKG2A blockade as a potential combination therapy with MICA/B neutralization to repress tumor-immune evasion. These studies together demonstrate the potential of advanced cellular models on fine-tuning therapeutic regimens and identifying new druggable targets to advance precision medicine.

While MCTS can be used to probe tumor-T cell cytotoxicity and potentially cellular dependencies on the stromal cells, the spatiotemporal resolution of these features can be lost in bulk co-culture models. Therefore, patient-derived spheroids in conjunction with microfabrication such as micro-patterning and microfluidics have been used to spatially specify and confine cell populations and facilitate high-resolution longitudinal imaging [75, 91–93]. In addition, supernatant from the co-culture systems can be profiled for cytokine release. Thus, using high-resolution imaging and multiplexed cytokine profiling of patient-derived organotypic tumor spheroids (PDOTS), Jenkins et al. demonstrated the use of PDOTS for identifying responders and non-responders to PD-1 blockade [93]. In addition, multiplex cytokine profiling identified CCL19 and CXCL13 upregulation in concordance with increased immune infiltration in patients treated with PD-1 blockade. Finally, a novel immunotherapy TBK1/IKKε inhibition was tested for its ability to both stimulate tumor cytotoxicity and activate T cells. Therefore, integrating bioengineering tools such as microfluidics and fabrication represents a significant improvement in cell-based tumor models such as MCTS, pairing the ability to spatially designate different cell types with enhanced cell culturing techniques and assay development.

In short, patient-derived spheroids offer 3D tumor models encompassing monocellular or multicellular compositions with cells in different proliferative and metabolic states, thus enabling preclinical drug discovery and testing for stage-specific therapeutic resistance. However, the preservation of histopathological and genomic features to the patient tumor is often lost over time during in vitro culture, and this is particularly poignant given that spheroids are difficult to expand, cryopreserve and biobank [69]. The lack of fidelity for continuous ex vivo expansion, and inability to subject them to extensive long-term culture for analysis after treatment regimens of months to years make their application in interrogating tumor evolution and its role in therapeutic resistance challenging.

Organoids:

The organoid is another 3D multi-cellular model. Organoid and spheroid cultures have distinct but overlapping features in cellular sources and experimental protocol for their generation. In contrast to spheroids, however, organoids are stem/progenitor cell-originated, self-assembled 3D organotypic structures generated following the encapsulation of a small number of cells within an ECM (e.g., Matrigel, synthetic biomaterials) and cultured in a cocktail medium of defined growth factor(s)-/pathway inhibitor(s), and cytokines. The organoid formation from adult stem cells was demonstrated first with Lgr5⁺ intestinal cryptvillus organoids [94]. While organoids provide very interesting models to study organ development and disease progression, the development of cancer organoids from patient tumor tissue has opened up new avenues and applications extending into drug discovery, screening, immunotherapy, and precision medicine; to date, tumor organoids have been generated from various cancer tissues in vitro [95–99]. In contrast to the short-lived cancer spheroids, patient-derived organoids (PDOs) can be expanded and grown extensively in vitro without compromising the genetic states, thus providing an “unlimited” potential biological material. Not only are organoids useful for therapeutic testing and patient response determination and stratification, but they are also capable of reflecting tumor evolution, acquired resistance, and metastasis over longer culture timespans [30–33]. Moreover, several studies have established the predictive capacity of PDOs for patient response to chemotherapy such as in colorectal and gastrointestinal cancers, interestingly demonstrating even greater preservation of subclonal tumor populations in cultured organoids compared to bulk tumors [100–103].

The ability of tumor organoids to be expanded extensively enables biobanking of PDOs, including colorectal, breast, lung, gastric, and bladder cancer, thus enabling testing therapeutic regimens for a wide range of heterogenous patient populations [31, 102, 104–107]. These biobanked PDOs have already started to contribute significantly towards the potential clinical guidance of cancer therapies for patient subsets. For example, studies with biobanked PDOs have unveiled a key ERK inhibitor treatment for a specific subset of liver cancer patients, improved adjuvant gemcitabine treatment of predicted pancreatic cancer patient responders, and a molecular subtype of gastric cancer vulnerable to specific chemotherapies [108–110]. In addition to guiding treatment, PDOs have also been instrumental in discovering therapeutically viable combinations, exemplified by the recent discovery of the microtubule-targeting vinorelbine inducing a cytostatic-to-cytotoxic switch in KRAS-mutant colon cancer [111]. Finally, a useful application of PDOs is the generation of both patient-matched tumor and normal organoids from tumor-adjacent tissue, thus providing a gauge on on-target off-tumor cytotoxicity [112].

Similar to spheroid cultures, PDOs are amenable to co-culture with other tumor-associated cell types including immune cells. Tumor-immune organoid models can be generated by growing organoids encapsulated with a biomaterial such as Matrigel followed by incorporating various immune components such as tumor-associated macrophages (TAMs), dendritic cells (DCs), TILs, or PBMCs. These in vitro longevity and expansion potential of the PDOs allow sufficient time to prepare other cell types for co-culture. A handful of studies have explored the applicability of organoids for testing cytotoxicity of T cells or model immune checkpoint inhibition (ICI) using T cell-tumor-organoid co-cultures [113–116]. Integrating cell-specific cytokine secretion along with real-time imaging of the tumor organoid provides unique insights into the role of gradients of secreted factors and cell-cell interactions on infiltration of immune cells to the tumor and its environment.

One such study used a colorectal cancer (CRC) organoid to characterize antigen-specific killing with anti-EpCAM chimeric antigen receptor natural killer cells (CAR NK) [117]. Using a CAR targeting a mutant EGFR overexpressed in CRC, the study showed differences in cytotoxicity against normal and tumor CRC organoids, thus characterizing on-target off-tissue effects which are pervasive in CAR-T therapies [118]. This model would thus be useful for direct testing of engineered T cells for target specificity. In addition, a highly patient-representative tumor-immune model was developed using air-liquid interface (ALI) PDOs [115]. With minimal perturbations to the patient biological material, it was shown that the major patient stromal and immune components, such as CAFs and TILs, were preserved in air-liquid supported PDO cultures and that the PDOs were amenable to serial passaging and cryopreservation. Finally, CD8+ T cells within these PDOs were activated by ICI with nivolumab, and the expression of effector protein genes like IFNG and GMZB was increased upon ICI, thus motivating this platform for wide-ranging applications like ICI response prediction and mechanistic studies on the effect of ICI on T cells.

Despite the significant advancements made in generating patient tumor-derived organoids, their applicability towards precision medicine such as drug response and determination of therapeutic regimens is still limited due to difficulty in organoid establishment from patients and sample availability [119]. While the patient-specific organoid formation is difficult to mitigate, engineering approaches such as droplet microfluidics to generate organoids can be adapted to circumvent the shortage of biological materials [120, 121]. The matrix-supported droplets containing single cells from homogenized patient tumor digest expand to form organoids and can be used to generate homogenous organoids for drug testing or T-cell-tumor cytotoxicity studies. One such study utilized microfluidic droplet to generate organoids from various cancers such as colorectal, lung, ovarian, kidney, and breast [120]. That the organoids can be formed from single cells makes the organoid model an attractive ex vivo platform for patient tumor avatars. The miniaturized organoid platform can capture the spatial heterogeneity within the tumor by producing differential organoids from the excised tissue, thus motivating the use of miniaturized platforms for organoid establishment and culture.

While the potential of these PDOs is grand, there exist significant challenges especially when used towards stratification of patients based on drug efficacy. There are instances where ideal stratification is difficult due to limited understanding of mechanism of action for chemotherapies. For example, patient-derived models of colorectal cancer from 29 patients were utilized for stratification based on response to chemotherapies in monotherapy or combination [100]. The study found that colorectal cancer PDOs could distinguish responders to irinotecan both in mono- and combination-therapy with 5-FU, but not for combination therapy using 5-FU and oxaliplatin [100]. This likely resulted from an inadequate representation of relevant tumor-associated cellular components such as the stroma or immune system and/or adaptive changes encountered during the ex vivo culture. Additional variables could include the treatments status and response of the patients from which the PDOs have been generated[122]. Furthermore, the inability to predict response to 5-FU/oxaliplatin combination therapy underscores the poor mechanistic understanding of the multilevel effects of a particular drug combination on the tumor. This could likely be remedied by improved organoid culturing systems with patient-specific fidelity.

The use of cancer organoids for testing TIL and TCR-T/CAR-T cells in adoptive cell therapy (ACT) is another important area of research. This approach leverages highly characterized organoids that preserve genetic mutations which could impart antigen-specificity of the T cell product. Such systems offering longitudinal imaging could quantify cytotoxicity for thresholding responses. However, the lack of TME representation in cancer organoids could result in missed outcomes and skewed conclusions of therapeutic regimens for patients, especially considering the multitude of patients who do not respond to immunotherapy despite tumor PD-L1 expression and sufficient tumor mutational burden, or due to potentially immune-excluded tumors [122–124]. Therefore, although the field of 3D models of the tumor provides several options (explants and 3D cellular models), there is a great need to create systems affording more mutability, heterogeneity and potentially greater physiological representation of the tumor.

Furthermore, a fundamentally limiting factor of all discussed models is the lack of a vascular component. In addition to mass transport, the vascular system serves as a conduit for other cell types such as immune cells and influences their behavior, while additionally representing a key component of solid tumors [125, 126]. Tumor vasculature, a highly aberrant, tortuous, and leaky network generated through hyperangiogenic signaling from both tumor cells and associated stromal cells results in increased interstitial fluid pressure and limited drug and immune penetration [127–129]. In addition, endothelial cells of the tumor vasculature and surrounding blood or lymphatics can both influence the recruitment of immune cell types like regulatory T cells and express checkpoint blockade proteins such as PD-L1 to inhibit cytotoxic T cell endothelial transmigration and tumor infiltration [130–132]. This inhibition is illustrated in Figure 1, where shear stress activates endothelial cell CLEVER-1 and PD-L1 expression, recruiting Tregs and activating immune checkpoints. Therefore, tissue engineered platforms which incorporate vasculature and 3D tumor models can potentially enable studies for drug testing, adoptive cell therapy, and precision medicine and will comprise the remainder of this review.

Vascularized 3D tumor models

Multiple methods have been developed to generate vasculature in vitro, such as micromolding [133–138], sacrificial hydrogels for generating cytocompatible tubes [139–145], and bioprinting [146–153]. In addition to these architecturally driven vascularization methods, microvascular networks formed by endothelial cells have been extensively developed and utilized [154–158]. Microfluidic-supported organs-on-chip platforms that incorporate microvascular networks are useful to not only extend the viability of 3D organ models, but also to understand the pathophysiology of various diseases, fluid dynamics within these environments, and the effects of heterogenous cell types and reciprocal interactions between endothelial cells. Given the importance of vascularization in cancer pathogenesis and progression, vascular structures have been incorporated with various ex vivo models of tumor including explants, spheroids and organoids.

Vascularized explants:

Initial studies involved the incorporation of vasculature into PDEs in efforts to increase their in vitro longevity and recapitulate an important aspect of the TME. Lance Munn and colleagues have pioneered tissue engineering methods to achieve the integration of vascular networks with tumor explants [159–161]. One such approach involves placing tumor explants of 0.2mm-0.5mm diameter onto microvascular beds created from human umbilical vein-derived endothelial cells (HUVECs) and pulmonary smooth muscle cells in a 96-well plate. Using this approach, the authors have identified that the tumor microenvironment of the PDE was largely composed of cell-derived ECM and supported by optimal vascularization of the explant. A follow-up study using pancreatic tumor explants demonstrated heightened secretion of several angiogenesis-related cytokines and potentially serves as a more patient-representative tumor model for guiding personalized treatment regimens [160]. This approach was more defined than previously described bioreactor cultures of tumor explants, which though enabled long-term culture of tumor explants but yielded low resolution on contributing cell types to phenotypes of interest (i.e., response to Bortezomib) [162]. To date, no other studies have been reported which vascularize tumor explants but results thus far nonetheless represent a fundamental improvement in ex vivo platforms and warrant further technological developments. Since the early studies that demonstrated the ability to vascularize tumor explants, the integration of microfluidics has taken a large effort, where perfusable vasculature is integrated with 3D cellular models of tumor like spheroids and organoids.

Vascularized spheroids:

By far the largest effort in vascularizing 3D tumor models - partly due to ease of generation and culture – uses spheroids [125–129, 163]. Using spheroids, fibroblasts, endothelial cells, and other tumor-associated cells, several groups have developed vascularized cancer spheroids and exploited them for drug testing [164–172]. In general, spheroids can be pre-vascularized with endothelial cells through direct co-culture or integrated with existing microvascular networks (mVNs). For long-term maintenance of mVNs, the addition of an ancillary cell, fibroblasts, is used to induce angiogenesis through secreted factors and reciprocal vessel-stabilizing interactions between endothelium and fibroblasts [177]. While there is a growing appreciation of the relevance of specific fibroblast populations – whether tissue-resident, induced, or recruited – there is limited research into the effect of specific fibroblast populations on microvascular network formation, function, or phenotypes of interest in drug or immune cytotoxicity studies with cancer spheroids. Given the importance of cancer associated fibroblasts (CAFs) on cancer growth and the tumor physicochemical environment, as well as in promoting neo-angiogenesis in the tumor, the incorporation of CAFs would render 3D tumor models more physiologically relevant and potentially improve therapeutic discovery [178]. However, CAF incorporation is much more complex because of the different CAF subpopulations. Specifically, cancer-associated fibroblasts can be myofibroblastic (myCAF), inflammatory (iCAF), antigen-presenting (apCAF) or others, further underscoring the importance of considering differential CAF processes in tumor control and vascular remodeling, especially as certain CAF subsets have been implicated in clinical response to chemo/immunotherapy [179–183]. In addition, CAFs can be highly secretory, and both their secretome and cell-cell contact with endothelial cells were recently shown to attenuate mVN formation [184]. Moreover, vascularized 3D cellular models formed the basis of studying drug transport into tumor spheroids as well as resistance to drugs such as in the case of temozolomide in glioblastoma [164–168, 170]. Vascularized cancer spheroids have also been recently employed to test TAM polarization and its influence on monocyte recruitment to MCTS (Figure 3A) [185]. The study identified monocytes as generally pro-tumoral, whereby blocking the monocyte chemoattractant secretion through a multispecific antibody against CCR2, CSF-1R and TGF-β could enhance TAM polarization to an anti-tumoral M1 cell state.

Figure 3: Immune cell perfusion into 3D tumor models using different strategies.

A: A pre-established microvascular bed can incorporate the MDA-MB-231 cancer spheroid for perfusion of monocytes to assess recruitment and extravasation (231 TFM = heterotypic spheroid comprised of MDA-MB-231 breast cancer cells, fibroblasts, macrophages); Adapted from [185]. B: A multi-layer microfluidic device can accommodate an endothelial cell monolayer above a 3D tumor model chamber; The endothelium is migrated across by CAR-T cells and CAR-T cells bind to cancer spheroids; The use of dasatinib to modulate cytokine release (CR) syndrome is explored; Adapted from [191].

Vascularized organoids:

The goal of vascularizing cancer organoids has only recently been elaborated. The first demonstration of a vascularized cancer organoid was reported in 2023 by Choi et al. [186]. Using microextrusion-based 3D bioprinting, lung fibroblasts and lung cancer organoids (LCO) were seeded in a decellularized lung ECM surrounding a gelatin-templated HUVEC-lined tube. This resulted in HUVEC-LCO contact and subsequent differential response to the chemotherapeutic, poziotinib. These studies highlight how the vascular compartment affected the LCO response to drug. Following up on this, the same group recently reported a similar vascularized patient-derived gastric cancer organoid platform that accurately reflected patient-specific responses to anti-VEGFR2 treatment [187]. The differential sprouting of vasculature upon anti-VEGFR2 treatment further supports the need to incorporate vasculature into cellular models of cancer. Specifically, the extent of vascular sprouting itself could potentially predict patient response to anti-angiogenic or other chemo/immuno-therapies. Clinical trials of anti-angiogenic drugs in single-agent or in combination with immunotherapy have witnessed mixed results, thus motivating better predictive in vitro models to stratify patient populations for appropriate therapeutic options [188]. Noo Li Jeon and colleagues very recently reported the first vascularized cancer organoid within a microfluidic platform, leveraging a pre-established microvasculature [189]. The vascularized tissue on mesh-assisted platform (VT-MAP) is a high-throughput 3D vascularized organoid system for testing drug delivery accompanied by a computer vision-enabled imaging analysis for quantifying drug cytotoxicity. Using patient-derived colorectal cancer organoids, the authors demonstrated patient-specific responses to single-agent or combination chemotherapy contrasting the inconsistent responses seen in PDX or isolated organoids of the same patients. Thus, the incorporation of organoids with perfusable microvasculature recapitulated the more complex in vivo environment of the patient tumor, resulting in similar drug responses seen in the clinic. An orthogonal technique, trapping organoids within restriction chambers in a microfluidic device, was recently demonstrated and enabled vascularization of the blood vessel or pancreatic islet organoids by apposed HUVEC monolayer sprouting [190]. This technique was further enhanced in a recent report of vascularized breast cancer-on-chip system (Figure 3B) [191]. The authors designed a multi-layer microfluidic device harboring organoid traps beneath an endothelial monolayer, in which perfused CAR-T cells underwent transendothelial migration and adhered to spheroids or PDOs. Using dasatinib to regulate cytokine release syndrome (CRS) from CAR-T cell hyperactivation yielded differentiable CRS based on the time of treatment, thus motivating the use of such vascularized 3D cellular models for patient-specific prediction of CRS prior to CAR-T therapy. This study and the previously described study on TAM polarization in a vascularized spheroid device exemplify two encouraging methods for leveraging perfusable vascularized tumor models for immune cell perfusion and testing, as illustrated in the schematic in Figure 4. Importantly, these two methods (tumor on mVN vs. tumor apposed to endothelial monolayer) are agnostic of tumor model, potentially applicable to explants, and cellular models of cancer such as spheroids and organoids.

Figure 4: Methods to perfuse immune cells into 3D tumor models.

The pre-establishment of a microvascular bed for integration with the cancer spheroid is one method to create a perfusable system. In this case, the self-assembled microvasculature presents perfusable lumens through which immune cells can flow and eventually interact with the incorporated tumor model. Alternatively, a multi-layer device can enable the formation of an endothelial monolayer. The microfluidic channels can then be perfused with immune cells for transendothelial migration into the tumor compartment. Schematic made with Biorender.

Conclusions and future directions

Cancer can be considered as a complex functional tissue that works in concert with its microenvironment to evolve and metastasize to target organs. Our understanding of the cancer relies on our ability to generate experimental models that accurately replicate the intricate features of tumor and tumor microenvironment. The field of employing explant and cellular models to study cancer has come a long way since its inception with the HeLa cells. Advances in tissue engineering have significantly transformed oncology in the recent times and the fundamental breakthroughs in 3D tissue models can not only reduce the use of animal models but also unveil advanced predictive in vitro tissue and cellular models to identify effective therapeutic regimens. Current progress in multi-cellular 3D models— spheroids and organoids—further advance the field by providing patient avatars for both precision medicine and biological interrogations. These models have realized the patient-specific high-throughput drug screening to identify the optimal therapeutic regimen and new treatment combination. The patient-derived tumor organoids can additionally offer a unique opportunity to investigate cellular heterogeneity of tumors owing to their ability to emulate the genetic and phenotypic heterogeneity as in the native tumors. Despite all these advancements, numerous hurdles must be overcome to enable wide-spread adoption of 3D tumor models in clinical practice and biological investigation.

For example, the outcome from the retrospective clinical studies that used organoids or spheroids to predict the patient response to therapeutic regimens suggests the need of continued validation using larger cohorts and development of protocols to realize the widespread application. Finally, benchmarking multiple aspects of organoid/spheroid platforms such as their establishment and culture, characterization of tumor features (i.e., tumor mutational burden, biomarkers), and incorporation with tumor-relevant cell populations will be necessary to advance their use in precision medicine.

Another avenue that needs improvement is developing strategies to achieve long-term culture of the patient-derived tumor models; this includes explants and 3D cellular models. This would be particularly important for probing the evolution of cancer over time and throughout treatment courses, as the tumor models could acquire mutations adapting to a changing TME or specific drug treatment. This is exemplified by therapy resistance through acquired mutations in rare cell populations during the course of treatments.

A useful faculty of in vitro 3D models is the integration of experimental assays with baseline characterizations. Such multi-modal approach could contribute to large-scale datasets containing transcriptomic and genomic information about the patient tumor, treatment regimens used for stratified patient populations, and data acquired through various assays from the patient-derived tumor models. Such integration of data could accelerate the generation of machine learning-enabled predictive models whose parameters are learned from a diverse set of sources including patient characteristics and tumor model assay readouts like drug-induced cytotoxicity or drug resistance. This will surely be a key focus of many groups in both industry and academia to leverage artificial intelligence for future innovation in oncology, especially as the predictive models could inform new drug targets based on patient genotype or changes in the TME.

Patient-derived cellular models such as organoids can also advance our understanding of the roles of immune cells and tumor-immunity on patient therapeutic response and tumor maturation/evolution. Tumor organoid models that recapitulate the tumor tissue with immune microenvironment by preserving endogenous stromal components including various immune cells, or by adding exogenous immune cells, cancer-associated fibroblasts (CAFs), vasculature, and other relevant cell populations will enable better drug evaluation for both response and mechanism. Such patient-specific in vitro tumor models with immune TME compartment could be invaluable tools to understanding the ever-evolving novel immunotherapies such as cancer vaccines, bispecific antibodies, and drug-antibody conjugates. The response of patient subpopulations to immunotherapies could be assessed using these patient-derived cellular models, especially by incorporating heterogenous, multicellular/polyclonal immune populations like TILs.

Perfusable vascularized tumor models that incorporate PBMCs or other immune cells can model the dynamic interactions between the cancer and immune cells, including effector T cell priming/activation, T cell trafficking/infiltration into tumor tissues, and recognition/killing of cancer cells by T cells. Given the antigen-specificity of current FDA-approved adoptive T cell products like TILs or CAR-T cells, it could be highly beneficial to leverage perfusable vascularized tumor models for pre-treatment testing of manufactured products (batch and lot release) with potency assays. While there are endless possible applications of vascularized tumor models including towards potency assay development, the scalability of these platforms represents a significant hurdle, as most of these systems are inherently lower throughput.

Finally, the generation of more complex in vitro tumor models that capture intertumoral and intratumoral heterogeneity could be a potentially groundbreaking avenue in 3D solid tumor modeling. Especially given the role of tumor-immune landscapes on immune therapy efficacy (i.e., immune-hot vs. immune-excluded), it would be beneficial to create models that capture the spatial multicellular heterogeneity often seen within the tumor. Identifying key factors involved in driving these landscapes could reveal therapeutics targeting them, as well as evolve our ability to convert immune-excluded tumors into immune-hot tumors or vice versa. Immune switches like these could theoretically serve our mechanistic understanding of tumor immunity as well as improve existing adoptive T cell therapies through novel pre-treatment regimens.

Together, the engineering of new systems to recapitulate the complex tumor using technologies like bioprinting and microfluidics has spawned a new era of tumor modeling. Seeking to retain the fundamentally high-throughput nature of 2D cultures thus represents the future of 3D tumor modeling, as it is relatively clear how representation of TME elements and structures will be achieved. Indeed, organoid and microfluidic-enabled cellular models are not inherently high-throughput, requiring inventive techniques and new strategies to be able to deploy them for preclinical drug testing and precision medicine using patient-derived models. Nonetheless, the current state of tumor modeling affords researchers and clinicians an abundance of options depending on the biological question. Therefore, there will likely not emerge a one-size-fits-all platform that can achieve the desired outcome or resolution for all applications, and hence users will need to weigh the pros, cons, technical expertise and resources required to generate the appropriate model. Such is the case where cytokine release is a primary measurement and would benefit from an explant model with millions of cells, or an imaging assay using microfluidic devices that utilize a transparent substrate for easier imaging. In conclusion, the myriad tools available to both biological and clinical research are constantly expanding, and future innovation in the space of experimental modeling of the tumor could potentially yield great benefit to precision medicine and fundamental oncology research by both enabling therapeutic development/testing and motivating highly specific and efficient cancer patient therapy.

Biographies

Naveen Natesh is a recent Ph.D. graduate of the lab of Professor Shyni Varghese in the Department of Biomedical Engineering at Duke University. His research focuses on developing tunable and physiologically relevant in vitro models of cancer by employing 3D tumor models like organoids and microfluidic-enabled vascular systems. Specifically, he seeks to understand the differential effects of tumor microenvironmental components on T cell-tumor infiltration and cytotoxicity. He is currently pursuing a career in industry.

Dr. Shyni Varghese is the Laszlo Ormandy Distinguished Professor of Orthopaedic Surgery at Duke University, with a unique true triple appointment in Biomedical Engineering, Mechanical Engineering & Materials Science, and Orthopaedic Surgery. She is Duke’s first MEDx Investigator, an interdisciplinary role bridging medicine and engineering. Dr. Varghese's research covers a wide range of topics, including aging, stem cells, biomaterials, microphysiological systems, each aiming to drive advances in regenerative medicine and therapeutic development.