Patient-derived cancer modeling for precision medicine in colorectal cancer: beyond the cancer cell line

ABSTRACT

Since effective immunotherapeutic agents such as immune checkpoint blockade to treat cancer have emerged, the need for reliable preclinical cancer models that can evaluate and discover such drugs became stronger than ever before. The traditional preclinical cancer model using a cancer cell line has several limitations to recapitulate intra-tumor heterogeneity and in-vivo tumor activity including interactions between tumor-microenvironment. In this review, we will go over various preclinical cancer models recently discovered including patient-derived xenografts, humanized mice, organoids, organotypic-tumor spheroids, and organ-on-a-chip models. Moreover, we will discuss the future directions of preclinical cancer research.

Introduction

Colorectal cancer is the third most common cancer diagnosed in men and the second most common cancer in women worldwide.¹ Surgery and chemotherapy are the main weapons against colorectal cancer. Major chemotherapy regimens are 5-fluorouracil (5-FU) combined with oxaliplatin (FOLFOX) or irinotecan (FOLFIRI). With the advent of targeted therapeutics such as cetuximab and bevacizumab, the oncologic outcome of colorectal cancer has dramatically improved. Immunotherapeutic agents such as pembrolizumab and nivolumab in combination with conventional regimens have been taking these achievements a step further, gaining exponential interest in oncology research and drug discovery.

The mainstay of the development of anti-cancer drugs is relying on preclinical cancer models. However, most chemical compounds that are effective in preclinical trials using cell-line based cancer models fail to show efficacy in clinical trials.² One of the main reasons for the low translatability is that preclinical models cannot accurately represent the biologic activity of tumor.³ Previous cancer models using cell line-based culture have some limitations to capture the intricate biologic phenomenon of tumors in the living body. They poorly retain intra-tumor heterogeneity (ITH). In addition, they can change genetically through passaging. It has been revealed that the cell line does not exhibit the full genetic spectrum of tumors originated, especially for rarer mutations in genetic analysis.⁴ Moreover, though tumor microenvironment (TME) consists of various cells other than cancer cells especially immune cells, fibroblasts, and extracellular matrix which play a crucial role in tumor growth and metastasis,⁵ cell line-based 2D culture models do not contain TME, and therefore, they have limitations for studying the dynamic interactions between the tumor and TME being regarded as a novel and effective therapeutic target.^6,7 To overcome these limitations, various patient-derived cancer model platforms have been developed.

Patient-derived xenograft (PDX) models are immunodeficient animals (commonly mice) engrafted with patients’ tumor cells or tissue. These in-vivo models are believed to be able to more reliably represent original tumors’ histopathology than traditional cell line-based cancer models. Recently, humanized mouse model generating well-differentiated multilineage human hematopoietic cells by knocking human cytokine genes into mice, and permitting the high rate of human cell engraftment has been found to be useful for studies involving the immune system of human and evaluating the response of immunotherapeutic drugs.⁸ However, animal models are inadequate for high throughput assay because they are expensive and labor-intensive to establish. In addition, experimental studies with humanized mice have some ethical issues. Firstly, fetal tissues such as fetal liver or fetal thymus are required as a source of cells.^9,10 Secondly, studies with humanized mice are kinds of chimera research and crossing species boundaries is regarded as ethically problematic in principle. Thirdly, in vivo studies involving animals basically raises issues of animal health and welfare.¹¹ Thus, a model that can simulate the pathophysiology of cancer as accurately as animal models while maintaining advantages of an in-vitro model with ease of manipulation, suitability for mass production, and relatively low cost is needed. Organoids were first introduced in 2009 by Hans Clevers and colleagues as a relatively new breakthrough model that can meet these needs.¹² They are a three-dimensional in-vitro culture platform founded from self-organizing stem cells. Since the first introduction of tumor organoids, some modifications including co-culture with peripheral blood lymphocytes¹³ and air-liquid interface (ALI) method¹⁴ have been applied to make them more suitable for studying immune-oncology and assessing novel therapeutics. Accordingly, organotypic tumor spheroids (OTS) containing human lymphoid and myeloid cells as well as cancer cells with key features of the TME can be used to effectively evaluate responses to immunotherapeutic agents such as immune checkpoint blockade.¹⁵ Furthermore, the organoids or organotypic tumor spheroids model can be used as a drug screening platform. For example, an organ-on-a-chip approach using special cell culture chips with multi-channel three-dimensional microfluidics has been newly employed, enabling researchers to adjust the local cellular, molecular, and chemical environments in a tailored manner for the purpose of the experiment.¹⁶ In this review, we will summarize each patient-derived cancer model currently used and discuss future directions of preclinical cancer research (Table 1).

Table 1.

Comparisons of patient-derived cancer models.

	Advantages	Disadvantages
PDX	Retain ITH Includes tumor microenvironment Preserve genetic mutations	Require high cost and time-consuming work Genetic diversion toward the mouse during passaging Generated in immunodeficient environment
Humanized PDX	Mimics human immune system Recapitulate tumor and immune interactions Retain ITH Includes tumor microenvironment Preserve genetic mutations	Unavailable of large numbers of autologous HSCs Require high cost and time-consuming work
PDO	Retain in-vivo like 3-D architecture and complexity High success rates Appropriate for constructing living biobanks	Drift toward clonality Lack of evidence to maintain ITH Artificial niche (much more growth factors and cytokines than physiologic environment)
OTS	Reflect the human immune microenvironment Same as PDO	Incomplete tumor microenvironment (only containing immune compartment) Lack of capturing systemic immunity
Organ-on-a-chip	Enable fine control over microenvironment Ideal platform for high-throughput drug screening Potential to generate multiple organs in a single integrated system	Difficult to optimize and standardize Difficult to scale up Complex devices to fabricate

PDX; patient-derived xenografts, ITH; intra-tumor heterogeneity, HSC; hematopoietic stem cells, PDO; patient-derived organoids, OTS; organotypic tumor spheroids.

Modeling cancer phenotype: intra-tumor heterogeneity

Within a single tumor, diversity has been revealed in several aspects. Regarding structured geographical diversity, various types of cells including cancer cells, mesenchymal cells, endovascular cells, and immune cells are comprising the tumor.¹⁷ More importantly, at the molecular level, there exists clonal heterogeneity derived from genomic instability and non-clonal heterogeneity caused by epigenetic gene regulation, both of which are involved in drug resistance as key mechanisms.^18-20 ITH has been observed in many types of malignancy including colorectal cancer.²¹ Understanding and characterizing the ITH is necessary for developing targeted anti-cancer drugs.^22,23 While traditional cell line-based two-dimensional culture models have been the backbone of cancer research to date, they not only comprise cancer cells without TME but also lack ITH. To fill the gap between preclinical and clinical trials for improving translatability, more exquisite and reliable patient-derived cancer models are needed.

Patient-derived xenograft models

PDX as a model for drug response

Mouse xenografts are constructed on the base of the immunodeficient mouse, usually nonobese diabetic/severe combined immunodeficient (NOD/SCID) mouse or Balb/c nude mouse.^24-26 In contrast with traditional cell line xenografts, PDXs are engrafted with mechanically chopped or chemically-digested fresh tumor tissue-derived directly from patients with cancer. Tumor tissues could be engrafted heterotopically or orthotopically. Heterotopic PDXs engrafted to the subcutaneous flank of a mouse are easier for cell transfer and tumor growth monitoring than orthotopic PDXs. Orthotopic PDXs are constructed with engraftment of tumor tissue to the ideal location of the mouse. They are technically difficult and time consuming to be established. However, it has been shown that orthotopic PDXs can represent the patient’s tumors better and predict response to drugs more accurately than heterotopic PDXs.²⁷ Especially, colon cancer cells show diverse responses to drugs according to the anatomical location of the graft.²⁸

We have constructed heterotopic PDX models to evaluate the drug response of original human tumors (Figure 1). Tumor samples were collected immediately after surgery from colorectal patients and cut into 1- to 2-mm3-pieces in 100 mL of Matrigel. Tumor fragments were transplanted into the flanks of 6-week-old female Balb/c nu/nu mice (Orientbio, Korea). When the mean tumor volume reached approximately 80–100 mm3, mice were randomly assigned to the Irinotecan treatment group and the control group. The dose of Irinotecan was 20 mg/kg, and the drug was injected intraperitoneally once a week. We concluded that the size of the tumor and the weight of the mouse are reliable surrogate markers for drug response. They can be easily measured to draw growth curves. In our other previously published study, we have validated the high degree of pathologic similarity of colorectal cancer PDXs to original tumors with histologic and immunohistochemical analyses.²⁹ Regarding molecular analysis including genetic mutations, genomic alterations, and gene expression patterns, characteristics of original tumors are well preserved in the corresponding xenograft tumors. Meanwhile, because PDXs could be the surrogates of a patient who provided the tumor tissue, they have the potential to provide the clinical information and predict the oncologic outcomes such as recurrence of cancer.^30,31 In our previous study,³⁰ we included 242 tumor specimens from colorectal cancer patients to establish PDX models and evaluated the relationship between the ability of the tumor to engraft (tumorigenicity) and the clinical characteristics and oncologic outcomes. We concluded that tumorigenicity was correlated with advanced tumor and moderate or poor differentiation. In addition, tumorigenicity was confirmed as an independent predictor of poor disease-free survival in patients with stage III colorectal cancer. Therefore, the aggressiveness of the tumor could be inferred from tumorigenicity in the PDX model, which may be an effective marker for poor oncologic outcomes.

Figure 1.

Preclinical study in the PDX model.

(a) Tumor growth curve of colorectal cancer PDX treated with Irinotecan. (b) Weights of individual tumors from each group. Error bars, ± s.d. (n = 5 per group). (c) Mouse weight during treatment. Statistical analysis was performed using a one-way analysis of variance followed by the Newman–Keuls posttest. (d) Representative photographs of individual tumors from each group. Tumor volume and body weight were recorded at regular intervals until tumors reached approximately 1,000mm3. To compare the groups, Student’s t-test was performed, and the p-value was indicated as asterisk mark. (** P < .01, *** P < .001).

Hui Gao and colleagues have conducted a large-scale in-vivo screen to model inter-patient response heterogeneity using colorectal PDXs.³² Their experimental paradigm substantially improved the translatability of in-vivo experiments. However, immense cost and time are needed to establish such a large cohort, PDX model is unsuitable as a high-throughput drug screening platform. Moreover, particular genetic diversion such as accumulation of copy number alterations specific for mouse tissue during passaging has been observed in PDXs.³³

Humanized PDX models: next-generation PDX models to evaluate cancer immunotherapies

As current immunotherapeutic compounds are only effective in a small subset of patients with colorectal cancer,³⁴ a deeper understanding of the interaction between the immune system and tumor cells with microenvironment is required to discover more widely effective therapeutics. A humanized mouse that mimics the immune system of the human is a valid platform to study tumor-immune interactions and evaluate the response to immunotherapeutic drugs. To construct a humanized mouse, the reconstitution of the human immune system to background immunodeficient mouse with impaired T, B, and NK cell populations to avoid rejection of the reconstituted human immune system is needed. NOG (NOD/Shi-scid/IL-2Rγnull) mouse has been found to be more advantageous than the traditional nude mouse or severe combined immunodeficiency (SCID) mouse as a background mouse for humanization.^35-37 CD34+ hematopoietic stem cells (HSCs) or peripheral blood mononuclear cells (PBMCs) are injected via the tail vein into irradiated 3- to 4- week-old mice, although CD34+ HSCs are preferred due to the possibility of graft-versus-host disease when using PBMCs. Major sources of CD34+ HSCs are fetal-liver and umbilical cord blood. After the injection of CD34+ cells, the level of humanization is monitored for 10 weeks using flow cytometry method and compared to levels of human CD45 (huCD45), human CD3 (huCD3), and human CD19 (huCD19) in the peripheral blood. After constructing humanized mouse, tumor response and immune response in the periphery and within the tumor to immunotherapeutic agents can then be evaluated.

Humanized mouse models may be unique ex-vivo platforms that simulate the dynamic interactions between tumor, TME, and human immune system. They overcome the drawbacks of the traditional PDX models constructed by the wildtype or immunodeficient mouse which do not contain the human immune components. Recent studies reported that humanized mouse models could provide a useful and informative tool in pre-clinical cancer immunotherapy research.^38,39 They could also be utilized as the screening tool for efficacy and the safety of cancer immunotherapy drugs such as anti-PD-1 or anti-CTLA4 antibodies.³⁹ Therefore, patient-derived humanized mouse models could help physicians to make a choice for immunotherapeutic drugs for each patient in the concept of personalized precision medicine.^40,41

However, creating a humanized mouse requires a very involved technique and has difficulties in carefully fine-tune many experiment variables with low success rate.^8,42 Moreover, the unavailability of a large number of autologous CD34+ HSCs to generate cohorts for study purposes limits the applicability of humanized mouse as a standard preclinical cancer model.

Patient-derived in-vitro cancer models

Organoids

Tumor organoids are three-dimensional primary tumor cell cultures retaining histopathological and genetic features of the original tumor. Early models of tumor organoids only comprised malignant epithelial cells without stomal component.^15,43 Recently, more reliable models consisting of both tumor cells and tumor-infiltrating lymphocytes (TILs) than simple organoids have been acutely required.^14,44,45

To obtain tumor-reactive T cells, peripheral blood lymphocytes can be a source.¹³ Co-culture of peripheral blood mononuclear cells with matched tumor organoids can generate a patient-specific model system for capturing tumor-specific T cell responses to cancers in a personalized manner. This approach evades the requirement for samples derived from the resection of specimens to isolate TILs. Also, the ability to expand circulating tumor-reactive T cells by co-culture with tumor organoids offers a clinically feasible strategy to generate patient-specific T cell products for adoptive T cell transfer.

Calvin Kuo and colleagues have previously presented a special culture method called ALI to generate organoid models containing tightly integrated epithelial and stromal components of normal stomach, pancreas, and colon.^46,47 They have advanced this method to culture clinical tumor samples as patient-derived organoids (PDOs) preserving complex histological TME architecture with tumor parenchyma and stroma, including functional, tumor-specific TILs.¹⁴ In their recent study, it has been shown that diverse endogenous immune cell types including B, NK cells, and macrophages besides T cells are preserved in their ALI PDO models, distinct from the reconstitution of clonally expanded or T cell receptor-engineered TIL populations added to tumor organoids.

PDOs have many advantages. Firstly, they have great potential for constructing living biobanks for various types of cancer.^48-53 Secondly, they provide almost the ideal platform for high-throughput drug screening assay because the cost for establishing PDOs is much less than that for establishing PDXs. In addition, they have high success rates.^14,54-57 PDOs can also be used as an adequate platform for studying tumorigenesis.⁵⁸ By combining with CRISPR/Cas9 technologies, drug-genotype correlations could also be investigated.⁵⁹ Drost et al.⁵⁹ have shown that genetically engineered organoids with loss of APC, TP53, and activating mutations in KRAS and PIK3CA in normal healthy cells could be the informative ingredients for studying tumorigenesis to exhibit gaining of malignant potential.

However, whether organoids can maintain tumor heterogeneity reliably has not been fully validated yet. Although tumor organoids show a diverse spectrum of cells phenotypically and genetically at the beginning of the culture, a drift toward clonality has been observed in the study of fluorescent labeling of tumor organoids, resulting in a single color after 30–40 days.⁵⁸ Cellular barcoding studies for clonal analysis of tumor organoids should be performed to evaluate the heterogeneity of tumor organoids through passaging.⁶⁰ Lastly, another disadvantage of organoids is that the optimal culture medium for growing organoids contains much more growth factors and cytokines than the physiologic environment.^61,62 As the composition of culture media could affect the growth and maintenance of the cells, the results for drug responses obtained using unphysiological culture media might not be reliable.

Organotypic tumor spheroids

To expand the capabilities of organoids, a novel microfluidic ex-vivo culture system consisting of epithelial cells and stromal cells grown in collagen hydrogels has been presented by Jenkins and colleagues.¹⁵ They validated that spheroids (40–100 μm) derived from tumors contained relatively similar fractions of T lymphocytes and myeloid populations in a recent study and named them as patient-derived organotypic tumor spheroids (PDOTS) and mouse-derived organotypic tumor spheroids (MDOTS) derived from human and mouse, respectively.¹⁵ These models could be employed to evaluate short-term response to PD-1/PD-L1-targeted therapies and cytokine expression profiling can be used as a functional readout of activity. The authors also revealed some features of resistance to immunotherapy.

We built PDOTS of colorectal cancer patients (Figure 2). Immunofluorescence staining using cell surface antibodies to delineate tumor cells and immune cells is of great utility. Tumor cells are readily detected using EpCAM antibodies. Staining for CD3 readily identifies effector CD3+T cells present in PDOTS, which are necessary effectors following treatment with immunotherapies. Therefore, we could visually confirm that tumor cells and immune cells were successfully co-cultured in this model and their spatial arrangement effectively resembled the TME. This model has great potential to be widely used in the study of TME and screening tool for immunotherapeutic drugs.

Figure 2.

Patient-derived organotypic tumor spheroids (OTS).

(a) Phase-contrast imaging (4X, 10X) of OTS derived from the colorectal cancer patients. (b) Immunofluorescence staining of CD3 immune cells in patient-derived OTS. Fresh tumor samples were obtained from the Department of Surgery at Samsung Medical Center in accordance with protocols approved by the Institutional Review Board. For the establishment of PDOTS, fresh tumor specimens were received in media (DMEM or RPMI) on ice and minced in a 10 cm dish using sterile forceps and scalpel. The minced tumor was resuspended in high-glucose DMEM with 100 U mL−1 type IV collagenase (Life Technologies, Carlsbad, CA), and 125 mg/ml Dispase II (Life Technologies, Carlsbad, CA) for 30 min at 37°C. Following digestion, samples were pelleted and resuspended in fresh media and passed over 100 μm filters. Cell fractions were pelleted and embedded in Matrigel (BD Bioscience) on ice and plated into 24‑well plates (1x104 cells with 50 μl of Matrigel per well). Matrigel was polymerized at 37°C for 15 min. PDOTS were fixed in 4% paraformaldehyde for 10 min at room temperature, permeabilized with 0.2% Triton X‑100 in PBS for 5 min at room temperature and incubated in blocking buffer (2% BSA and 0.2% Triton X‑100 in PBS) for 1 h at room temperature. PDOTS were labeled with antibodies against EpCAM, clone 9C4, (BioLegend, cat no. 324210, 1:100 dilution) and CD3 (BD Pharmingen, cat no.555340, 1:100 dilution) overnight at 4°C. Cells were analyzed with a laser scanning confocal microscope (magnification, x20). DAPI and protein signals were detected at excitation wavelengths of 633 and 488 nm, respectively.

The OTS model has strength for recapitulating the in-vivo use of immune checkpoint inhibitors or combination with molecularly targeted agents because this model reflects the immune microenvironment with relatively similar fractions of T lymphocytes (CD3+, CD4+, and CD8+) and myeloid (CD11b+ and/or CD11c+) populations.⁶³ Deng and colleagues have used the OTS model in support of beneficial effects of CDK4/6 inhibition on tumor immunity.⁶⁴ The authors were able to reveal that CDK4/6 inhibition in combination with an anti-PD-1 antibody could increase tumor cell deaths and reduce chemokines.

Although these in-vitro systems are among the most adequate models to discover and test effective immunotherapy combinations, they have some limitations. In the TME, various types of cells that can interact with tumor cells and influence experimental results such as fibroblasts, endothelial cells, and immune cells. However, the focus of OTS is only on the immune compartment. Because the OTS model mimics immunologic features of the local environment surrounding the tumor, the dynamics of systemic immunity such as priming of naïve T cells in lymph nodes as one major target for anti-CTLA4 is not captured in this model.⁶⁵

Applications of patient-derived in-vitro cancer model: organ-on-a-chip

As an alternative to mouse models and cell culture models, ‘Organ-on-a-chip’ or ‘Tumor-on-a-chip’ approach that integrates microfluidics, microfabrication, tissue engineering, and biomaterials research has emerged recently as a novel platform for cancer research.^66-71 This microfluidic system offers improved spatial organization through controlled compartmentalization and well-designed control over the diffusion of soluble factors. Tumor-stromal interactions, tumor-vasculature interactions, and interactions with non-cellular components including extracellular matrix and chemokines can be studied on chips in real-time. Furthermore, these culture-systems can be applied as drug delivery and screening tools. Thus, they have great potential to enable high-throughput drug screening in a well-controlled scalable manner.⁶⁸ Efficacies of drugs that target tumor cells directly and the effects of anti-angiogenic agents can be tested by assessing responses of VEGF-induced and lung fibroblast-induced chemotaxis of human umbilical vein endothelial cells (HUVECs).⁷² Despite recent rapid progress in the organ-on-a-chip model, a lot of devices are still complex to fabricate and integrated genetic quantification is quite difficult in these systems.^73,74 Through continued development and advancement, this approach will lead to powerful in-vitro pre-clinical cancer models that can facilitate the discovery of anti-cancer drugs.

Future directions of patient-derived cancer models

Although novel immunotherapeutic agents show promising clinical responses, the vast majority of patients do not respond to current therapeutics.⁷⁵ In particular, patients with colorectal cancer who respond well to immunotherapy are defective in mismatch repair genes only account for 15% of all colorectal cancer patients.⁷⁶ Possible causes of failure of treatments are the paucity of immunogenic antigens, insufficient antigen presentation, and alternative expression of immune checkpoint molecules.⁷⁷ Moreover, reliable biologic markers that can accurately predict which patients will show a good response to immunotherapy have not been identified yet. Therefore, extending benefits of immunotherapy to a large fraction of patients who are unresponsive to the current regimen are unmet needs. To improve the response rate of immunotherapy, several combinations of drugs such as anti-PD-1/PD-L1 combined with CTLA-4 and/or immune checkpoint blocker parallel with targeted agents or traditional chemotherapeutic agents have been tested in clinical trials.^78-80 Furthermore, there is also an active study of various understandings that determine successful immunotherapy, including the TME, tumor-infiltrative lymphocytes, stroma and vasculature around the tumor, and whole-body elements (peripheral lymphocytes, intestinal microbes, and/or hormonal changes). A recent preclinical study has suggested that the gut microbiome can modulate tumor response to checkpoint blockade immunotherapy, although this result has not been well-established in human cancer patients.^81,82 Therefore, the gut microbiome of humans should be considered when constructing more precise preclinical models.^66,81,83-86 The most ideal preclinical cancer model may recapitulate all these in-vivo features. Organ chips have substantial advantages in that we can integrate each chip representing one organ or tumors to construct the whole body systematically, namely human body-on-a-chips.⁸⁷ These theoretical models mimicking whole-body physiology, controlling fluid flow and cell viability while permitting real-time observation of the culture tissues can be used to predict human pharmacokinetic and pharmacodynamic responses of drugs outside the human body. However, the ambiguity of whether such an artificial system would really be physiological still deserves criticism.

Conclusions

In summary, many of the widely used preclinical cancer models today have their own strengths and weaknesses. These models are being rapidly improved, and translatability of preclinical trials is being enhanced. Researchers and clinicians should be able to utilize appropriate models according to the contents of their study and the purpose of their experiments.

Funding Statement

This review received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors.

Author contributions

D.H.P wrote the manuscript. H.K.H carried out the experiment and edited figures. Y.B.C and W.Y.L devised the main conceptual ideas and supervised the entire process of this review.

Disclosure of potential conflicts of interest

No potential conflicts of interest were disclosed.