From bench to bedside: Advancements in patient-derived xenografts for predicting therapy outcomes in colorectal cancer

Abstract

Colorectal cancer is the third most diagnosed malignancy and the second-leading cause of cancer-related deaths worldwide. Management includes a combination of surgery, radiotherapy, and systemic therapy that is tailored to the stage of the disease. However, each tumor has a unique genetic profile that influences the treatment response and the overall prognosis. Biomarkers guide treatment decisions, but many chemotherapeutics lack reliable predictors. To bridge this gap patient-derived xenograft models were developed and are valuable preclinical tools. These systems utilize patient-derived tumor tissue grafted into an animal host that provides a platform for personalized drug profiling. This article surveyed recent advances in mouse and zebrafish colorectal cancer patient-derived xenografts, emphasizing their clinical utility for functional precision oncology. We explored the impact of these models on translational research, discussed current limitations, and outlined key priorities for future development.

Key Words: Zebrafish; Mouse; Patient-derived xenografts; Colorectal cancer; Translational medicine

Core Tip: This minireview discussed the use of patient-derived xenograft (PDX) models in mice and zebrafish and their impact on predicting treatment outcomes in patients with colorectal cancer. Each model has its own unique advantages and disadvantages that are useful in certain situations. Mouse PDX models are better suited for longer-term and in-depth studies, and zebrafish PDX models are better for chemoprofiling and short-term treatment guidance. We also discussed the applicability of these models for personalized medicine and the synergy with current biomarkers to better select treatment.

INTRODUCTION

Colorectal cancer (CRC) is one of the leading causes of morbidity and mortality in the world[1]. In 2020 it was the third most diagnosed cancer, with 1900000 new cases. It ranked second in cancer-related deaths, with over 900000 fatalities. Importantly, geographical disparities exist in CRC. The highest incidence rates have been in developed regions (such as North America, Europe, and Australia/New Zealand). Developing countries have had lower incidence rates, reflecting differences in diet, lifestyle, and access to screening[2]. However, this pattern is changing with an increasing incidence in low-income and middle-income countries as societal habits evolve. Projections indicate that the largest increase in CRC cases by 2040 will occur in developing regions. An alarming trend in high-income countries is the increasing burden of CRC in younger populations. In the mid-1990s only 10% of CRC cases in the United States occurred in adults under the age of 54. Currently, that proportion has doubled to 20%[3]. These epidemiological trends have led to the implementation of enhanced screening protocols and the development of intensive systemic regimens, including cytotoxic combinations, vascular endothelial growth factor (VEGF) inhibitors, epidermal growth factor receptor antibodies, and recently immune checkpoint inhibitors[4]. Despite these advances long-term outcomes for patients with advanced or metastatic CRC remain poor. The 5-year survival rate is 72% for regional disease and decreases to 12% in the presence of distant metastases[5].

Although the integration of molecular biomarkers has guided therapy decisions, their predictive accuracy is far from ideal. For example, cetuximab, an epidermal growth factor receptor-blocking antibody used as first-line treatment in metastatic CRC, is less effective in the presence of activating mutations in KRAS, NRAS, or BRAF genes[6]. Even among patients with wild-type tumors for these key oncogenes, at least 23% of patients do not achieve an objective response rate (ORR) to anti-epidermal growth factor receptor-based therapy[7] with another study reporting an even lower rate[8]. Recently, nivolumab plus ipilimumab received approval as a first-line treatment for microsatellite instability-high or mismatch repair-deficient metastatic CRC based on results from the phase III CheckMate 8HW trial. Although the study showed robust overall efficacy, 15% of patients still progressed within 3 months[9].

Since actionable biomarker frequencies reflect the diversity of epidemiological patterns, a personalized approach is necessary. KRAS gene mutations are present in nearly half of the CRC cases in the United States and Europe but in only 20% of patients in Malaysia[10]. Moreover, the prevalence can vary up to two-fold between regions within the same country[11,12]. Microsatellite instability shows a similar trend with microsatellite instability-high tumors accounting for 16% of CRCs in Norway[13] and only 6% in Japan[14]. These patterns underscore the need to align therapeutic strategies with the molecular landscape of each population and ultimately of each patient.

Most patients with advanced disease will cycle through several regimens. Evidence of the response to the first-line treatments is robust but drops sharply for later options in which no direct comparisons between therapies are available. First-line protocols achieve ORRs of 55%, whereas the ORR can drop to as low as 1%-2% in subsequent lines[15-17]. This steep decline highlights the need for treatment guidance tailored to the individual. Functional precision platforms, such as patient-derived xenografts (PDXs), can reproduce a patient’s tumor biology outside the body to allow clinicians to test multiple protocols in parallel. This approach could shift therapeutic selection in cases lacking a clear standard of care.

WHAT ARE PDXS?

As cancer research has advanced and patient outcomes have steadily improved, traditional in vitro, in vivo, and in silico methods have likewise evolved, ushering in the era of precision oncology in which therapies are matched to the unique molecular features of each tumor. In response, dynamic platforms for the functional assessment of tumor behavior and drug sensitivity profiling have been developed, among which animal PDX models gained popularity. PDXs are generated by implanting human primary or metastatic tumor cells or tissues into a live animal host, creating models that preserve the architecture, heterogeneity, genetic makeup, and overall complexity of human cancers[18].

Murine PDX (mPDX) models are widely regarded as the gold standard. They were created in 1969 when Rygaard and Povlsen[19] implanted colon carcinoma tissue from a patient into immunodeficient nude mice. Since then, mPDX models have consistently demonstrated a high degree of concordance between patient responses and those observed in vivo, reinforcing their value in translational research. Stratified drug testing across cancer subtypes and the establishment of large, diverse mPDX cohorts for preclinical evaluation are conducted in mPDX models[20,21]. Furthermore, recent advances in humanized mPDX models have broadened their application in immunotherapy research[22,23].

Zebrafish (Danio rerio) have been used as a model organism since Creaser’s early studies on embryogenesis in the 1930s[24]. George Streisinger later played a pivotal role in establishing zebrafish as a prominent model for genetic research[25]. Building on the foundational work of Lee et al[26] and Haldi et al[27] in developing zebrafish embryo xenografts, Fior et al[28] described the first zebrafish PDX (zPDX) model incorporating phenotypic drug-response screening and the first co-clinical trial[29] in CRC.

It is now established that zebrafish have notable conservation of the genetics involved in tissue and organ development and in key oncogenes and tumor suppressors. Approximately 70% of zebrafish genes have human orthologs, representing 82% of genes associated with human diseases[30]. The prolific reproduction, rapid and external development facilitating real-time visualization[31], and immune incompetency during early embryonic stages[32] are a few advantages of zPDX over mPDX[33-36]. A comparative overview of the advantages and disadvantages of these two PDX platforms is presented in Table 1, and the clinical relevance of each model is further explored in their specific chapters. Comprehensive protocols and detailed reviews outlining the process of PDX development for both animal models are extensively documented in the literature and fall beyond the scope of this discussion[37-40].

Table 1 Comparison of zebrafish embryo and murine patient-derived xenograft models in colorectal cancer.

Category	Zebrafish PDX	Mouse PDX
Host	Embryos are usually used (immune system not fully developed); multiple transgenic zebrafish lines available	Immunodeficient mice are required for xenografting (e.g., non-obese diabetic-severe combined immunodeficient-γ, nude)
Physiological alignment	70% synteny to humans - representing 82% of genes associated with human diseases; pharmacokinetics differ from mammals	90% synteny to humans; more predictive drug bioavailability and clearance
Engraftment site	Perivitelline space or yolk sac	Subcutaneous or orthotopic
Tumor material	1 mm to 3 mm fragment or cell suspension; no expansion needed for chemoprofile experiments; compatible with biopsies	3 mm to 5 mm fragment or dissociated cells; often needs expanded in culture
Tumor microenvironment and stroma fidelity	Can temporarily retain native tumor architecture and early immune interactions; models early metastatic steps	Maintains three-dimensional architecture and tumor heterogeneity in early passages but over time, human stroma replaced by mouse stroma
Temperature	28-37 °C (usually intermediate temperatures for both the graft and the fish to thrive are required)	37 °C
Time to results (chemoprofiling)	Days to weeks	Weeks to months; impractical for time sensitive clinical decisions
Sample throughput	High	Low
Xenograft monitoring and imaging	Variables usually monitored in studies - changes in tumor size, apoptosis, micro metastasis; stereo or confocal microscopy; compatible with automation for injection/imaging	Variables usually monitored in studies - tumor volume, metastasis, survival; calipers, bioluminescence, imaging (e.g., ultrasound, small-animal magnetic resonance imaging/computed tomography), postmortem pathology
Cost	Low	High
Ethical burden	No ethical approval required before 120 hours post fertilizations; better aligns with 3R principles; minimal tissue needed, reducing ethical concerns	Subject to strict animal welfare regulations; ethical constraints limit large-scale use; requires ethical approval and complex logistics (biosafety, permits)

PDX: Patient-derived xenograft.

MPDX

The mPDX models provide valuable prognostic information. One key observation is the ability of a patient’s tumor to engraft and grow in mice, indicating an aggressive tumor biology and poor outcome. Advanced tumor stage and poor differentiation, which are both associated with unfavorable prognosis, have been linked to higher PDX engraftment rates. Oh et al[41] established CRC mPDX and found that patients with stage III tumors that successfully engrafted had significantly lower disease-free survival than patients whose tumors did not engraft. Similarly, tumors from patients with stage IV CRC tended to engraft at a much higher frequency than those from patients with stage I CRC, and metastatic lesions had particularly high implantation success rates. In one (albeit small) series, 100% of CRC liver metastasis samples formed PDX tumors compared with 84% of primary CRC samples. Moreover, tumor growth was faster in metastasis-derived PDX than in those originating from primary tumors[42]. Other groups have correlated successful engraftment with adverse biological features, including high Ki67 expression, TP53 gene mutations, mismatch repair deficiency, lymphovascular and perineural invasion, high tumor nuclear grade, and elevated lactate dehydrogenase levels[43].

Besides the prognostic value of mPDX, clinical outcomes to systemic therapies are also provided. The mPDX have shown the effectiveness of treatment responses in CRC, allowing stratification across various systemic therapies. Maekawa et al[44] reported that mPDX models reliably mirrored patient outcomes to standard chemotherapy regimens like FOLFOX and FOLFIRI. Other larger studies involving multiple cancer types demonstrated a 96% sensitivity and 70% specificity in predicting responses across various treatment regimens[45].

Targeted therapies such as cetuximab and erlotinib have been studied in mPDX models. Bertotti et al[46] demonstrated that 10%-15% of cetuximab-refractory wild-type models carried focal amplification of human epidermal growth factor receptor 2 (HER2). Switching to HER2-directed treatment led to partial tumor regression of the mPDX models. Rivera et al[47] later confirmed that HER2 overexpression is frequent among cetuximab non-responders, reinforcing its role as a driver of resistance and as a druggable target for salvage therapy.

Immune checkpoint inhibitors have been evaluated using autologous humanized mouse models in which immunodeficient mice were engrafted with CRC tumor cells and peripheral blood mononuclear cells. These models replicated patient responses to the anti-programmed death-1 receptor agent nivolumab by engaging preexisting tumor-specific T cells. T cell receptor analysis confirmed patient-specific immune activity, and combination treatment with nivolumab and regorafenib in microsatellite stable CRC mPDX mirrored clinical efficacy[48]. This development is superior to allogeneic models that show only non-specific immune effects.

Remarkably, mPDX models created at the time of the patient’s initial surgery can remain therapeutically informative long after the patient’s disease has evolved. In a co-clinical series spanning several solid tumor types including CRC, mPDX established from early resections were re-challenged with a systemic regimen the patient subsequently received. Each time the xenograft recapitulated the clinical outcome[45]. Altogether mPDX models reveal tumors that are more aggressive and reliably predict treatment outcomes even beyond first-line therapy. This dual capacity makes them a valuable asset for guiding long-term clinical decisions in CRC management.

Despite their advantages mPDX models have limitation in certain situations. A major constraint is the extended time required to establish a xenograft. Recipient mice must be between 4-week-old and 8-week-old with most researchers waiting until at least week 6 before transplantation[49-51]. Following engraftment, tumor growth and additional profiling experiments may take up to 6 months[52]. Moreover, the establishment of xenografts requires special strains of immunodeficient mice, such as non-obese diabetic-severe combined immunodeficient or non-obese diabetic-severe combined immunodeficient gamma, to minimize immune interactions and improve engraftment success[53]. These mice strains lack a functional adaptive immune system and have severely limited innate immunity. As a result, they must be maintained in specialized facilities with sterile housing conditions, autoclaved supplies, sterilized food, and dedicated infrastructure distinct from standard laboratory animal units[54,55]. These limitations substantially increase the cost of establishing mPDX models.

ZPDX

A series of recent studies explored the potential of zPDX models to predict treatment responses in CRC after standard chemotherapy[28,29,56], targeted therapy[57], and radiotherapy[58]. The first study to evaluate the applicability of zPDX for chemosensitivity profiling in CRC was conducted by Fior et al[28] in 2017. Cell suspensions from resected colon adenocarcinomas were injected into zebrafish larvae. Consistent engraftment rates and preservation of key tumor features, including histoarchitectural organization, angiogenesis, and expression of human-specific markers, were observed. This study confirmed biological relevance and provided evidence that key histopathological and phenotypic traits of human tumors can be maintained in the zPDX model.

Five zPDX models were treated with FOLFOX chemotherapy, mirroring the adjuvant regimen administered to the corresponding patients, to test treatment predictability. Notably, the zPDX models that exhibited increased apoptosis in response to treatment were derived from patients who did not relapse during the follow-up, whereas non-responsive models corresponded to patients who experienced disease progression. This functional correlation, albeit limited by the retrospective design and small cohort, suggested that zPDX can offer early prediction of treatment outcomes. In a parallel sensitivity assay, zPDX models generated from tumors harboring KRAS or BRAF gene mutations showed no benefit from cetuximab, reflecting known genomic resistance patterns[6]. Even though the results are preliminary, zPDX may serve as a translationally relevant model for evaluating therapeutic sensitivity.

Subsequent work extended the application of zPDX models to anti-VEGF therapy, specifically bevacizumab. The zPDX models treated with bevacizumab-based regimens exhibited variable tumor phenotypes, including no therapeutic effect and features suggestive of prometastatic behavior. Some zPDX models showed modest tumor shrinkage accompanied by increased micrometastasis dissemination. Others showed complete resistance to therapy[57]. These divergent responses echo the paradoxical outcomes previously described for VEGF inhibition in which primary tumor stabilization may coincide with an increased risk of secondary determinations[59]. The strengths of that study include the treatment responses observed and the capacity to detect negative therapeutic outcomes that could inform clinical decision-making.

In two clinical cases in which patients progressed under bevacizumab, the corresponding zPDX models mirrored this resistance with a minimal apoptotic response and persistent or increased micrometastasis spread[57]. These observations suggest that zPDX models can expose hidden risks of different treatment options, especially in the absence of reliable biomarkers (e.g., anti-VEGF therapy)[60]. This work reinforces the functional resolution and relevance of zPDX models beyond binary response classifications.

A more recent study investigated the radio sensitizing potential of metformin in rectal cancer to demonstrate the adaptability of the zPDX platform in different therapeutic contexts[58]. This investigation focused on the neoadjuvant window and used both surgical resection and endoscopic biopsy. The zPDX models were treated with radiotherapy in combination with either 5-fluorouracil (5-FU) or metformin and assessed for apoptosis and micrometastasis dissemination. In tumors with distinct mutational profiles, metformin elicited radiosensitizing effects comparable to 5-FU. In some cases, it was associated with reduced micrometastatic burden. These effects were even observed in zPDX models that were derived from limited biopsy material, underscoring the feasibility of the platform in settings where tissue is scarce and timelines are constrained. Interestingly, one zPDX model derived from a previously irradiated tumor displayed resistance to both combinations, and this aligned with the patient’s refractory clinical course[58]. The zPDX model served as a proxy for therapeutic sensitivity and captured the consequences of prior treatment exposure for which a clinically relevant biomarker is not currently available.

A prospective co-clinical study published in 2024 marked a significant advancement in validating the clinical utility of the zPDX assay for CRC management[29]. In this trial zPDX models were generated for 55 patients receiving standard chemotherapy to replicate the patient’s therapeutic protocol. The test relied on early functional markers, particularly apoptosis and micrometastasis dissemination, to categorize tumors as sensitive or resistant. The assay achieved a positive predictive value of 91%, a negative predictive value of 90%, and an overall accuracy of 91%, with results available within 10 days. The data were refined by integration of the tumor stage and metastatic potential. Although these results are promising, the study was from a single-center with a limited patient cohort. Future larger, multi-center prospective trials must be conducted before routine clinical application.

The zPDX model also captures biological features that are not readily assessed by conventional staging or molecular profiling. While most zPDX models derived from early-stage tumors did not display micrometastases, a subset showed aggressive behavior that later correlated with relapse[29]. In a case involving a patient with synchronous primary and metastatic tumors, the zPDX model generated from the liver metastasis showed resistance to therapy, while those derived from the primary tumor were sensitive, mirroring the patient’s clinical outcome. These findings suggest that the model may reveal intrapatient heterogeneity and sampling site-specific therapeutic differences, particularly in advanced diseases. However, it is unknown whether this form of early stratification truly affects long-term outcomes. While progression-free survival was significantly longer in patients with sensitive zPDX, the study did not include overall survival as a predefined endpoint because the follow-up duration was insufficient to detect meaningful differences. Nonetheless, the ability to initially avoid ineffective therapies, especially in patients with limited therapeutic windows or fragile health status, may be a benefit that is not fully captured by survival curves alone.

An alternative zPDX model for CRC based on the implantation of intact tumor fragments into embryos was developed. In the prospective study responses to standard chemotherapy regimens including 5-FU, FOLFOX, FOLFIRI and FOLFOXIRI were evaluated in PDX derived from 36 patients. The model enabled short-term evaluation of tumor volume changes. In a subset of metastatic cases, the in vivo response aligned with the clinical outcomes in 6/8 patients[56]. Although this study was designed primarily as a methodological proof-of-concept, it supports the feasibility of using tissue-based xenografts for individualized drug testing. By preserving tumor architecture and performing response evaluation criteria in solid tumors-based assessments, this approach contributed to the growing diversity of zPDX strategies aimed at therapy stratification in CRC.

Taken together, these studies highlight the versatility of the zPDX model in understanding therapy-specific tumor behavior. Despite methodological differences and limited clinical integration, the model is a promising functional tool for treatment stratification. Its future value depends on standardization and prospective validation across defined clinical settings.

CONCLUSION

PDX models are powerful tools for predicting therapy outcomes in CRC. They can bridge the translational gap between laboratory findings and clinical application. Both mPDX and zPDX systems offer unique advantages for personalized oncology. The mPDX models are still the gold standard due to their well-established record, ability to maintain tumor architecture and molecular fidelity over serial passages. Their ability to reflect patient-specific responses across standard chemotherapies, targeted agents, and immunotherapies is an obvious advantage. The utility of mPDX extends beyond drug screening by allowing the accumulation of important prognostic information.

The zPDX models offer complementary strengths. Their rapid engraftment time, high-throughput potential, and capacity to utilize small biopsy samples make them particularly attractive for functional profiling. Studies have shown that zPDX can preserve key histopathological features and accurately predict response to chemotherapy, anti-angiogenic agents, and radiotherapy. Prospective co-clinical studies confirmed their predictive reliability with some demonstrating an alignment with the actual patient response over 90% in less than 10 days. The zPDX is a potential tool to assist rapid clinical decision-making. It is important that future studies confirming this technique are conducted in single-center cohorts and will then require larger multicenter prospective trials to assess the robustness and universality before zPDX can be integrated in routine clinical practice.

The mPDX and zPDX techniques will advance considerably from integration with next-generation technologies. The zPDX models equipped with fluorescent probes and specialized transgenic lines will allow investigators to watch angiogenesis, immune-cell infiltration, and clonal competition unfold in real time at a single-cell resolution. The resulting images could then be analyzed by machine learning algorithms to quantify tumor metrics and correlate in vivo dynamics with drug response. Artificial intelligence is increasingly applied to the analysis of large-scale biological datasets to identify new indications for existing compounds. These agents could be screened in PDX models to validate in vivo efficacy, accelerating therapeutic discovery in resistant or rare tumor subtypes. Liquid-biopsy assays also add a complementary layer of precision. Qualitative and quantitative circulating tumor DNA profiling stratifies patients by relapse risk and prioritizes those at high risk who would benefit from zPDX testing. Viable circulating tumor cells can be grafted directly to form xenografts to enable drug testing on the most aggressive clones and to extend functional precision oncology to frail patients who cannot undergo surgical biopsy.

Ultimately, the question a clinician would ask is: Can PDX models be leveraged to meaningfully improve patient outcomes, prolong overall survival, increase ORRs, identify those most (or least) likely to benefit when biomarkers give little guidance, and direct effective therapy choices in data-poor settings while sparing patients unnecessary toxicity? The evidence so far is encouraging, but larger prospective trials are still needed to standardize and validate these protocols and to determine the best-use scenarios.

ACKNOWLEDGEMENTS

The authors gratefully acknowledge institutional support from the University of Medicine and Pharmacy “Grigore T Popa”, Iasi, Romania.