A systematic review on the culture methods and applications of 3D tumoroids for cancer research and personalized medicine

Although a lot of research has been done on 3D models derived from stem cells in the past [15], the group of Hans Clevers pioneered the successful establishment of organoids in the modern era [14]. These organoids were generated in 2009 from murine intestinal stem cells, and since then many different research groups have adapted the establishment protocol to generate an extensive collection of organoid models from both animals and humans [16].

Typically, primary surgical resection material is cut into small fragments using scalpels, followed by an enzymatic digestion to dissociate the tissue into single cells or small cell clumps [17, 18]. However, as access to tissue samples can be anatomically and logistically limited, stem cells derived from fine needle aspirates of tissues [19‐21], fluid samples such as urine [22], ascites [23], broncho-alveolar lavage fluid [24], and bile [25], or enriched circulating tumor cells (CTCs) [26] might provide a minimally invasive alternative for organoid generation.

Organoids are mostly cultivated in 3D ECM hydrogel domes such as basement membrane extract, Matrigel® [27], or Geltrex®, in addition to a variety of xeno-free synthetic hydrogels [28]. Organoid culture medium provides a variety of niche-specific growth factors matched to the tissue of origin to ensure optimal growth conditions [17, 29, 30]. Common key components include activators of the canonical WNT pathway (R-Spondin), MAPK pathway (EGF), and inhibitors of TGF-β (A83-01) as well as BMP signaling (Noggin) to allow stem cell renewal and proliferation [11, 14, 17].

Stable organoid lines can be expanded long-term (> 10 passages) in vitro, and cryopreserved [31]. Of note, even though conventional submerged organoid cultures exclusively enrich for epithelial cells and fail to retain stromal components [32], organoids do recapitulate the tissue architecture on a microscale, and can mimic the function of the organ they resemble, while stably maintaining mutational and epigenetic states in line with cellular signaling [33‐35].

Organoids are used in many different research areas, including reconstructive medicine, evolutionary biology, and pathogen-host interactions, summarized in previously published reviews [36‐41]. In recent years, tumoroids derived from both animals and humans have been intensively used for preclinical cancer modeling, with a focus on cancers originating from epithelial cells (Fig. 1). However, efforts have been made to also establish tumoroids from non-epithelial cancers and rare cancer types, for example melanoma [42], glioblastoma [43, 44], rhabdomyosarcoma [45], epithelioid sarcoma [46], and rhabdoid cancer [47].

Disease modeling using tumoroids allows for the investigation of different hurdles of oncology such as tumor initiation and driver events of cancer, tumor heterogeneity or development of efficient therapies [10, 48‐50]. By establishing healthy organoids and altering cancer type specific genes in vitro, the transformation from healthy to cancerous cells can be studied in a 3D model focusing on different tissues and diseases. Additionally, by generating organoids and tumoroids directly from both healthy and tumor tissue, cancer driver mutations can be identified and used to discover patient-specific therapeutic targets [33, 51]. This way, organoids and tumoroids could be used for both answering basic questions of tumorigenesis and developing a personalized medicine approach to help advance the field of cancer research.

Given recent advancements in organoid technologies, we here summarize the latest developments in the field, outlining various methods and practical applications of these innovative model systems. We will focus on current uses for cancer research including both human- and animal-based methods. Furthermore, we will highlight recent work that has improved preclinical cancer models by using tumoroids in combination with other cell types to model the tumor microenvironment (TME). Moreover, studies focusing on advanced 3D culture systems based on fluidic devices and bioprinting, and grafting tumoroid models into animals for in vivo based studies, will be summarized. Finally, existing limitations are discussed, and future developments of the field are highlighted.

As tissue samples from animals often are easier accessible than most human biopsies, and as a big variety of murine cancer models exist, animal-derived tumoroids can be especially useful when human-derived tumoroids are not available. Therefore, many murine and other animal-derived 3D models have been established and used across all fields of cancer research (Table 1). Importantly, the availability of normal tissue samples and thus healthy organoid lines from animal models allows for the evaluation of adverse side effects of cancer therapies on healthy tissue [52].

In the past, cancer research mainly relied on genetically modified mouse models (GEMMs) targeting genes commonly mutated in human cancers [93]. Thus, a variety of organoids and tumoroids from these mouse models exist, which can further be easily genetically modified in vitro. Two recent studies focused on the development of preclinical 3D models for ovarian cancer based on murine cancer models [94, 95]. The knock-out of ovarian cancer-specific genes was induced using either a Cre recombinase-based system or CRISPR/Cas9 in organoids derived from healthy murine tissue in vitro, leading to their malignant transformation. The group of Zhang et al. hereby focused on the influence of the mutational heterogeneity on treatment response, and showed that both tumoroids and mice with the same mutational background can be treated successfully with identical combination therapies [94]. The group of Löhmussaar et al. established a tumor progression model for high-grade serous ovarian carcinoma from two different epithelial lineages of mouse ovaries. Their findings support the dual-origin hypothesis of these tumors and confirmed lineage-dependent differences in drug sensitivities [95]. Thus, these studies highlight how murine tumoroids can be used as a model for identifying novel genotype-informed treatments, and for investigating the differences and vulnerabilities of tumors arising from distinct cell types, leading to more patient-specific and efficient therapy options.

Another approach using tumoroids derived from a prostate cancer (PCa)-specific mouse model was reported by Chan et al., who investigated genetic driver mutations of treatment-induced transformation and plasticity [68]. The group induced malignant transformation of healthy organoids by knocking out the PCa-specific genes Pten, Rb1, and Tp53 in vitro and showed that over time the resulting tumoroids acquired an intermediate luminal-basal phenotype. An epithelial to mesenchymal (EMT) signature was further supporting tumoroid plasticity, a phenomenon that has recently also been investigated in pancreatic cancer (PaCa) organoids [96]. Of note, the authors reported that the PCa tumoroid plasticity was further induced by androgen ablation (the suppression or blockage of the androgen pathway using chemical compounds), the most common treatment for PCa, and identified JAK/STAT and FGFR signaling as the main drivers [68]. Since the inhibition of these two pathways converted the tumoroids back to an androgen ablation-sensitive more luminal phenotype, PCa patients may benefit from the same treatment. The researchers also tried to confirm their results in human PCa tumoroids. However, it is difficult to maintain human prostate organoids and tumoroids in long-term cultures, and the results did not overlap [68, 97, 98]. More studies are needed to verify these findings, but PCa is a good example for how animal-derived 3D models could be used as an alternative model for human cancer.

To investigate metastasis formation in addition to tumor heterogeneity, two of the main factors for poor overall survival of breast cancer (BrCa) patients, tumoroids from the C3(1)-TAg BrCa mouse model were used [54]. Individual tumoroid lines established from the same primary tumor showed heterogeneous invasive mechanisms that were classified into collective invasion or dissemination of single cells. Interestingly, KRAS expression was required for both mechanisms, and ERK inhibition blocked both invasive processes [54]. Additionally, it was shown that collective migration of BrCa cells is mediated by a keratin 14 and cadherin 3-positive subpopulation of leader cells [53]. These cells have an enhanced protrusive activity and interact with the TME to initiate invasion, and could become a novel target for preventing or treating BrCa progression. However, murine 3D cell models can not only be used to test the effectiveness of drugs, but also of other anti-cancer treatments like radiation. After irradiating murine intestinal organoid models, Du et al. demonstrated that the induction of inflammation with Zymosan-A promoted the regeneration of intestinal stem cells by up-regulation of ASCL2 [52]. Thus, Zymosan-A may be an effective radio-protective drug for the prevention of harmful side effects on surrounding healthy tissue and treatment of ionizing-radiation-induced intestinal injury.

In general, the results obtained using murine organoid systems that mimic human disease adequately reflect human physiology and mechanisms [52‐54, 68, 94, 95]. However, rodents and mammals are inherently different, which can lead to unexpected discrepancies during translation to the clinic [99, 100]. Recently, the use of organoids and tumoroids derived from farm or companion animals has become more popular in the field of cancer research, as animals like cattle, horses, pigs, monkeys, dogs, and cats have a more similar anatomy to humans [101]. One big difference to rodents is that these animals can develop spontaneous cancer lesions, which makes them more suitable to study the mechanisms of cancer initiation and metastasis [83]. For example, in dogs bladder cancer (BlCa) arises spontaneously with similar pathology and genetics to human disease [89], while murine tumoroids do not fully reflect the characteristics of human BlCa [102]. Therefore, Elbadawy et al. focused on canine tumoroids derived from CTCs from urine samples to generate a more representative model to study BlCa [89]. Later the same group also established canine healthy bladder organoids to study the transformation of healthy cells to cancer cells [102]. The canine BlCa 3D models were then used to show that trametinib, a drug mainly used for melanoma treatment [103], and extracts from the Chaga mushroom [104], could be a potential therapy option for patients with BlCa. Recently, the same researchers also established canine healthy lung and lung cancer 3D models that resemble human disease as a new model for molecular analysis and drug testing of lung cancer [85]. It can be expected that additional organoid systems will be established from non-conventional animals in the near future, some of which might have high translational impact for human cancer [105, 106].

Patient-derived tumoroids (PDTs) maintain the characteristics of the primary tumor and can thus be used as a model to identify the most efficient cancer treatment for individual patients, laying the foundation for personalized medicine [107]. Therefore, tremendous efforts have been made to generate organoids and tumoroids from diverse tissues and tumor types. Until now, PDTs from more than 20 different carcinoma types, including BrCa, lung cancer, and colorectal cancer (CRC) have been established (Table 2).

Due to their advantages over 2D cell culture, PDTs are widely used for drug screening and therapy response prediction of patients [17]. Additionally, as it was shown that both animal- and human-derived tumoroids do not acquire additional mutations during long-term culture in vitro [148], numerous studies have already proven the usefulness of such 3D systems for precision medicine including CRC [113, 149‐151], PaCa [152‐154], as well as rare cancer types such as Merkel cell carcinoma [147]. In CRC PDTs, the response of patients to oxaliplatin was predicted with a sensitivity of 70% and a specificity of 71% [155]. Additionally, transcriptome analysis revealed that differences in oxaliplatin response were mediated by distinct genetic features, and 18 specific genetic alterations were identified as a potential biomarker panel for oxaliplatin resistance in CRC patients to aid clinical decision making. Another study focusing on the resistance of PaCa patients to neoadjuvant chemotherapy confirmed a high heterogeneity in patient responses, supporting the notion that personalized medicine approaches might be more efficient [156]. Hennig et al. hypothesized that the observed resistance is mediated by resistant clones residing in the tumor that get enriched under systemic treatment. In addition to intrinsic resistance of tumors, acquired resistance can also hinder efficient treatment. Most research on the underlying mechanisms of this phenomenon have been generated using 2D cell lines, but recently resistant tumoroid models have been established as a more representative system [157‐159].

As confirmed by the recent improvements of transcriptome and proteome analysis on single cell level, one of the remaining problems for efficient cancer treatment is the intra-tumoral and inter-patient heterogeneity [160, 161]. Studies showed that between different regions of the same tumor, drugs can have varying inhibitory effects [162, 163]. This highlights the fact that biopsies might not be representative for the drug response of the whole tumor, and that specific combinatorial therapies might be necessary to target all tumor subclones [164, 165]. To analyze whether PDTs can reflect the intra-tumoral heterogeneity of a primary tumor, De Witte et al. compared the drug responses of ovarian cancer PDTs derived from different sites of the same primary tumor for a small patient cohort [162]. The group reported that individual tumoroids showed a differential response of 31%, defined as a more than 10-fold change in IC50 value, to the same treatment, proving inter-patient heterogeneity. Additionally, individual tumoroids from the same primary tumor showed high differences in drug response in six out of seven cases, suggesting that PDTs genetically maintain the heterogeneity and drug sensitivity of the original tumor [162]. Interestingly, it was recently shown that a more physiological culture medium changes the drug response of ovarian tumoroids compared to commonly used organoid medium, highlighting the necessity for careful selection of experimental conditions [166]. Using a similar approach, the intra-tumoral drug response differences of liver cancer were investigated [163]. In this study, 27 PDTs were generated from different areas of five primary lesions and a total of 129 cancer drugs were tested. Even though a high inter-patient and intra-tumoral heterogeneity was confirmed, the researchers identified several pan-effective drugs that showed an inhibitory effect for most of the tested tumoroids [163].

The high heterogeneity of cancer is a significant problem for efficient treatment; however, metastasis formation is still the main reason for cancer related death. To elucidate the differential drug responses between primary tumors and matched liver metastases of CRC, Mo et al. generated tumoroids from both samples of 25 patients, and showed that the intra- and inter-patient heterogeneity is reflected by the in vitro 3D models [167]. Interestingly, although transcriptomic analysis revealed differences between primary and metastatic tumoroids, drug sensitivities were highly consistent. Thus, the authors suggested that the response of metastatic lesions to specific drugs could be predicted using tumoroids derived from primary tissue for a personalized medicine approach [167].

The predictive potential of PDTs and their translation to the clinics has also been investigated for various other cancer types, such as PaCa [168], and brain cancer [169], two highly aggressive cancer types with limited or inefficient treatment options. In a translational approach, tumoroids of different stages of PaCa and brain tumors were used to predict the optimal treatment for each patient in a clinically relevant time frame [168, 169]. Even though both studies confirmed that tumoroids reflect inter-patient heterogeneity and patient-specific drug responses, limitations of varying efficiencies of PDT generation, time management, but also cost, have to be overcome for direct translation of tumoroid research to the clinics. Especially for rare cancers, where a major problem for efficient treatment is the lack of prognostic and diagnostic biomarkers due to small patient numbers [170, 171], tumoroids might be suitable to improve treatment response and survival rates of patients [172]. For example, PDTs of PD-L1 negative mucinous adenocarcinoma of the appendix were used to predict response to chemotherapeutic drugs and targeted therapy for one individual patient [173]. PDT drug response correlated well with the patient response, and the tyrosine kinase inhibitor dasatinib was identified as a possible treatment option for this patient.

In addition to evaluating the efficiency of commonly used treatments for patient-specific precision medicine, PDTs can also be used to test the efficacy of novel treatment regimes. As an example, lung cancer PDTs are highly sensitive to the tyrosine kinase inhibitors dabrafenib and trametinib, which are commonly used for melanoma treatment, suggesting these drugs as a novel therapy for lung cancer [174].

In summary, several studies have provided clinically relevant evidence that PDTs can be a suitable model for a precision medicine approach to overcome the problem of high heterogeneity in drug responses between patients. In line with this, PDTs have also been used to identify pre-existing and acquired resistance to commonly used drugs, to avoid treatment of patients with inefficient therapies.

PDTs are not only suitable to predict the response of tumor cells to chemotherapy or targeted drugs, but also to other treatments like radiotherapy [141, 175‐178]. Radiation directly leads to the induction of single-strand and double-strand breaks in DNA, as well as the generation of reactive oxygen species and the upregulation of oxidative stress signaling pathways, and is thus commonly used as a combinatorial treatment with chemotherapy [179]. For patients with locally advanced rectal cancer, the standard treatment is neoadjuvant chemoradiation followed by total mesorectal excision of the tumor [175]. Drug response of rectal PDTs matched chemoradiation responses in patients with an accuracy of up to 84% [175, 176], and a sensitivity of 78.01% with 91.97% specificity [176]. The predictive potential of PDTs for radiotherapy has also been confirmed for head and neck squamous cell carcinoma (HNSCC), where the combination of the EGFR inhibitor cetuximab and radiotherapy resulted in increased cell death in vitro, and showed higher efficacy compared to radiotherapy alone [141]. A recent study from the same researchers confirmed the predictive potential of HNSCC tumoroids by correlating tumoroid therapy response with patient clinical response to model the efficacy of chemoradiation for these patients, and the further use for biomarker selection and validation [177]. In addition, PDTs offer an advanced model to study the dynamics and mechanisms of radioresistance. Glioblastoma PDTs reflected in vivo resistance mechanisms mediated by treatment-induced senescence to combination treatment with temozolomide and radiation, suggesting the use of PDTs for studying the underlying mechanisms of drug resistance [178].

To investigate the interplay between cancer cells and their surrounding environment, advanced co-culture models, including the culturing of tumoroids with cancer-associated fibroblasts (CAFs) [187, 188], lymphocytes [189], and myeloid cells such as macrophages [190‐192], or even bacteria [193] have been used. Indirect cell interactions can be mimicked by the addition of cell-conditioned media, which contains secreted molecules, to another cell type. However, this co-culture system only allows for single-directional response, which can be overcome by the use of transwell cultures. Individual cell types are separated via a mesh leading to diffusion of secreted molecules and therefore to a multi-directional interaction [13, 188]. However, as direct cell-cell contact is essential for various physiological processes, indirect co-culture systems can only be used to investigate the effect of metabolites, cytokines, and other secreted molecules. In direct co-culture systems different cell types are embedded in the same ECM and then either submerged in culture medium or maintained as an ALI. Because of the direct cell-cell contact, different aspects of the interaction between the tumor and its TME can be studied. However, as it is more complex to isolate specific cell types for downstream analysis, single cell read outs or optical techniques are required to efficiently analyze cell-cell dependencies or their crosstalk [187, 190, 194, 195]. As important factors normally found in the TME are lost upon dissociation of the tumor tissue and the subsequent artificial assembly of the isolated cells for direct co-culture systems, Ootani et al. established a method to culture cells in their original environment by preserving the TME using ALI models [196]. Dissociated tissue is hereby directly embedded into an ECM such as collagen type I and/or a hydrogel, seeded onto a transwell mesh, and then cultured in a way that only the basal part of the ECM or tissue is in contact with the culturing medium while the other part is exposed to the atmosphere [183, 188] (Fig. 2A). ALI models have already been established for several cancer types, such as salivary gland cancer and kidney cancer (Fig. 2B). This method leads to the differentiation and maturation of the cell types present in the tissue, resulting in organoids or tumoroids containing epithelial cells, as well as stromal and immune cells [13, 183, 197]. This way, PDTs can be cultured for months while mostly preserving the original TME, thus reflecting the physiological tumor site. However, the limited availability of primary tumor tissue, the slow growth of a tissue ALI culture, and the partial loss of stromal cells during long-term culture, makes this method unsuitable for high-throughput analyses such as drug screens [198]. Tissue ALI culture can nevertheless be a powerful direct co-culture model for studying processes like tumor progression, therapy response to selected drugs, or the presence and infiltration of immune cells in the original tumor tissue.

Bioprinting allows for the exact positioning of different cell types and biomaterials in a 3D space with the help of a mechanical and computer-assisted system to mimic the in vivo spatial architecture of a tissue or tumor and its microenvironment [216]. The biggest advantage of using a bioprinting-based approach to generate preclinical cancer models is the standardization of cell dispensing [217], and the possibility to construct an artificial 3D tumor including different cell types, structures, and ECMs for more precise personalized medicine approaches [218].

Early studies focused on generating such models by using bioprinted cancer cell lines or single cell suspensions [184]. Recently, the dispersion of organoids and tumoroids together with stromal cells allowed for the development of co-culture models including the TME (Fig. 2C) [219].

CRC microtissues were produced according to patient-specific colonoscopy images by printing PDTs surrounded by healthy organoids to model the interaction of a tumor with normal adjacent tissue [220]. As the in vitro treatment response of these microtissues to the standard 5-FU therapy reflected the patient response, the model could be used as a more physiological drug screening platform. Additionally, the patient-specific risk of tumor invasion into the surrounding tissues was calculated in this model by correlating the number and distance of invading tumor cells. This could provide a real-time quantitative readout for analyzing cancer progression and metastasis [220].

In another recent study, microtissues consisting of patient-derived lung tumoroids in co-culture with matched CAFs and endothelial cells were generated [221]. After printing the vessel structures and seeding the CAFs, the tumoroids suspended in a hydrogel derived from porcine lung tissue were printed into the same compartment. An active fusion of the stromal cells and tumoroids was observed, and microvessels formed to directly interact with the other cell types. After administering the drug poziotinib through the vessel structures it was seen that both the endothelial cells and CAFs, but also the CAF-secreted matrix, protected the lung tumoroids from the treatment. Thus, this model could be used to further test the influence of cell-cell or cell-matrix interactions on the efficacy of drug delivery to the tumor tissue [221].

Apart from printing models mimicking the TME, bioprinting can be used for standardized cell dispensing for high-throughput applications. However, one drawback is the difficulty of printing bioink into small wells as the utilized ECMs can spread in the well and thus do not maintain the shape necessary for 3D cell growth. To circumvent the spreading, patient-derived glioblastoma or sarcoma cells were mixed with a bioink consisting of hyaluronic acid as well as collagen, and then printed into wells coated with gelatin, which was later removed and substituted with medium [222]. In addition, acoustic bioprinting of small droplets onto a hydrophobic substrate was used to generate BlCa-derived tumoroids consisting of both cancer cells and CAFs [223]. This method allows for the generation of large numbers of uniform tumoroids that mimic the TME, which can easily be dispensed into small wells and be used in high-throughput drug screens for personalized therapy.

Thus, the combination of 3D cell models and bioprinting could help to facilitate reproducible drug testing, and the investigation of cellular processes in a 3D space with multiple cell types and structures. However, these printed models are still static and do not reflect the effect of mechanical forces like dynamic flow, or chemical gradients, which influence the tumor cell phenotype in vivo [224].

The important physical and chemical characteristics of the TME can be modeled with microfluidic systems, where cells are cultured in chambers connected to microchannels supplying oxygen, growth factors, or drugs via dynamic perfusion [225‐227] (Fig. 2D). In a proof of concept study, CRC PDTs cultured on fluidic devices showed a higher viability, proliferation rate, and tumoroid formation efficiency, compared to tumoroids cultured in static ECM drops on a plate [228]. Additionally, chip-grown tumoroids were bigger in size and showed a multilayered morphology with crypt-like projections similar to the human colon, while the conventional tumoroids formed more cystic monocellular structures. Interestingly, when testing the response of PDTs to 5-FU no significant differences were seen between the culturing methods, suggesting that chip-grown 3D cells maintain their phenotype, while having a growth advantage, and thus validating the use of fluidic devices for tumoroid culture and disease modeling [228].

However, the tumor niche consists of multiple cell types, and an extensive vasculature, through which nutrients, oxygen or drugs get delivered to the cancer cells. In order to mimic this vascular transport of factors and metabolites to the tumor, endothelial cells together with fibroblasts were grown in microfluidic chambers to produce a 3D microvascular network, before adding primary BrCa tumoroid-like structures into a separate adjacent compartment [186]. Using this setup, vessel outgrowth towards the tumor cell chamber was observed in addition to tumor cells invading the vasculature chamber after inducing EMT via TGF-β treatment. Aside from its suitability for physiological drug testing, this chip design could be a useful tool to study and easily visualize differences in angiogenesis, proliferation, and migration [186]. Importantly, it has already been shown before that fluidic devices provide a more physiological model to study cancer cell migration [229].

Using microfluidic devices with tumoroids alone, recent papers already described the advantages of chip-based models for a personalized medicine approach and more representative drug testing compared to static models. Both Mazzocchi et al. [143] and Jung et al. [230] used PDTs to perform medium-throughput drug screens for mesothelioma or small cell lung cancer (SCLC) patients, respectively. Drug response to specific chemotherapeutic therapies was different between the tumoroid lines with different mutational backgrounds but correlated well with the patient response. Importantly, SCLC tumoroids on the chip maintained a chemotherapy resistant core, which might be due to insufficient drug delivery also observed in in vivo tumors [230].

To facilitate the translation of this drug screening method to the clinics, Schuster et al. developed an automated high-throughput screening platform on a chip [231]. The authors constructed a microfluidic system with which 20 independent experimental conditions could be tested on up to 10 different patient-derived 3D cell lines. A customizable software allows for the stimulation, assaying, and imaging of the tumoroids without human intervention. Interestingly, combinatorial treatments and time-dependent administration with multiple rounds of treatment showed greater therapy efficacy on PaCa tumoroids compared to single treatments [231].

To investigate the influence of the tumor stroma and the method of drug delivery on treatment efficiency, a fluidic system mimicking the TME for drug testing on PaCa PDTs has recently been established [232]. The vasculature was modeled by growing endothelial cells in a perfusable microfluidic scaffold. Tumoroids co-cultured with fibroblasts, which resulted in increased proliferation of cancer cells and enhanced tissue stiffness, were then added into adjacent compartments. When administering the chemotherapeutic drug gemcitabine to the co-culture via diffusion through the vessel-like structures, the drug had a reduced efficiency compared to the administration to tumoroid single cultures or to static co-cultures. Thus, the authors were able to recapitulate the microenvironment of a vascularized pancreatic tumor and to demonstrate the inhibitory effect of the vasculature cell barrier and the stroma on drug efficiency [232].

Interestingly, another PaCa-on-a-Chip model was used to show that stroma-targeting agents do not influence the cell viability of tumoroids in monoculture, but lead to a significant increase of chemotherapeutic anti-cancer effects in a fluidic device co-culture [233]. Using a similar model, it was reported that different drugs show different efficacies in targeting PaCa cells in normoxic versus hypoxic conditions [234]. Thus, both publications highlight the importance of including cells of the TME and relevant oxygen levels for microfluidic drug response studies.

In conclusion, tumoroid-on-a-chip models provide a powerful tool to investigate biological processes and to assess drug responses in an efficient and more physiological manner compared to static tumoroid cultures.

To study the tumorigenic potential of cancer cells based on common mutational patterns specific for endometrial cancer, murine tumoroids that overexpress Kras^G12D and carry genetic deletions of various tumor suppressor genes (Pten, Tp53) were allografted subcutaneously into nude mice [237]. Different mutational backgrounds led to the development of specific tumor subtypes with corresponding tumoroids having an either more cystic or spindle-like morphology because of an irreversible EMT during transformation. This also correlated with a less or more aggressive phenotype, respectively [237] (Fig. 3B).

As the specific cancer subtype also plays an important role in the aggressiveness and treatment of PaCa, Miyabayashi et al. established an in vivo model to investigate the role of patient-specific genetic and epigenetic aberrations for PaCa heterogeneity [240]. The authors derived tumoroids from patients and xenografted them orthotopically into the pancreatic ducts of immunodeficient mice. Tumors were either indolent with a luminal morphology or progressed to a more aggressive basal-like subtype based on epigenetic aberrations and deregulation of KRAS. In contrast to Maru et al. [237], this group argued that cellular plasticity is mainly influenced by the stroma and microenvironment rather than cellular clonality, thus suggesting that allo- and xenografts should be performed orthotopically to recapitulate the corresponding tumor niche for translatable results [240].

Cellular plasticity and clonality can not only be influenced by the microenvironment, but also by specific therapies [241]. To investigate resistance mechanisms of androgen pathway directed therapy, Lee et al. generated tumoroids from patient-derived xenograft models of bone metastatic PCa. Androgen ablation led to the development of reversibly dormant basal-luminal-like hybrid cells, highlighting how some cancer therapies might even support the progression of the disease, thus suggesting that treatment regimens might have to be reviewed [241]. Apart from tumor heterogeneity and plasticity hindering efficient treatment, metastasis formation is one of the leading causes of cancer-related deaths [242]. CTCs can spread to distant secondary sites to induce metastasis [243]. As CTCs are a rare and heterogeneous cell population, De Angelis et al. established a model to study the biological features and possible drug sensitivities of these cells [244]. The group derived tumoroids from both orthotopic CRC xenografts and CTCs isolated from the blood of the same mice. As expected, CTC-derived tumoroids showed a more aggressive and migratory phenotype in addition to higher stem cell and EMT marker expression compared to the xenograft-derived tumoroids. The authors also showed that CTC-derived tumoroids are highly sensitive to drugs targeting the Survivin pathway including YM155 and quercetin, providing a new possible treatment to inhibit metastasis formation. As this model reflected the properties of CTCs isolated from CRC patients, De Angelis et al. hypothesized that in the future patient-specific drug sensitivities could be studied in a metastasis xenograft model established from the patient’s primary tumor even before the onset of metastatic disease, and thus prevent tumor progression [244].

The same group also developed another preclinical metastasis model for CRC [245]. After orthotopically grafting patient-derived cells into immunodeficient mice, the authors generated tumoroids from the primary tumor site and from spontaneous liver metastases and analyzed the EMT state together with the drug response. As expected, metastatic tumoroids displayed a more mesenchymal phenotype and increased chemoresistance. Interestingly, high numbers of CTCs in patients correlated with an increased engraftment efficiency of the corresponding tumor-derived cells, again highlighting the important role of CTCs for cancer progression [245].

EMT is not only an important characteristic of CTCs, but also of the cancer cells in tumor lesions. Recently it was shown that hybrid cells, presenting both epithelial and mesenchymal characteristics, can be found in human tumor tissues, GEMMs, and tumoroids of BrCa [246]. The number of these cells correlated with worse overall survival in patients, while driving invasion in vitro. Using tail vein injection or orthotopic engraftment of tumor cells or tumoroids, respectively, heterogeneous EMT states were identified within the same tumor. Sequential molecular EMT programs were essential for tumor cell invasion and colonization during the metastatic process. Thus, both models could be used to further investigate the influence of the EMT status on patient survival [246].

To further understand the mechanism of metastasis formation, the role of the immune system in the metastatic niche was also recently investigated in a PCa model of liver metastasis [239]. Murine tumoroids generated from GEMMs with PCa specific mutations were orthotopically allografted into immunocompetent mice, which resulted in metastatic spread to the liver after only 30 days. This model is not only less time consuming compared to original GEMMs and xenografts, and more reliable than patient-derived cell lines, but importantly also allows for the investigation of the immune cells in the metastatic niche. The authors found that an immune suppressive microenvironment mediated mainly by neutrophils, anti-inflammatory macrophages, and CD4 + T cells was essential for the survival of tumor cells and progression of metastases (Fig. 3D). This model might be relevant for testing drugs targeting immune cells, and for elucidating cell-cell crosstalk in the metastatic niche [239].

In summary, organoids and tumoroids can be used in an in vivo setting to investigate many aspects of cancer development and progression, highlighting the advantages of orthotopic grafts and allografts. Thus, these models are highly suitable for drug testing and the development of novel therapies.