Review Article

Standardized Xenograft Models for Preclinical Cancer Research

65 views

⸱

DOI:

10.3791/71892

⸱

August 18th, 2026

1Altogen Labs R&D

Corresponding Authors: Dmitriy Ovcharenko <ovcharenko@altogenlabs.com>

In This Article

Summary

Xenograft models - cell line-derived (CDX), patient-derived (PDX), humanized, and autologous - remain the primary in vivo platform for cancer drug development; standardized reporting, rational platform selection, and AI-driven translation define current best practice.

Abstract

Xenograft models are the principal in vivo platform of preclinical oncology and the most established experimental link between cell culture and clinical investigation. From the carcinogen-exposed rabbit models of the early twentieth century through the current generation of humanized patient-derived xenograft (PDX) systems, these platforms have evolved in response to the demands of translational cancer research. This review critically examines the biological principles, methodological standards, and translational applications of the principal xenograft platforms in current use. Cell line-derived xenograft (CDX) models remain the most widely used and most cost-effective modality for preclinical efficacy testing, offering the reproducibility, scalability, and accessibility that have sustained their role across oncology drug development pipelines for decades. PDX models have emerged as the preferred platform for co-clinical trial design, predictive biomarker discovery, and personalized oncology applications, preserving the genomic landscape, intratumor heterogeneity, and histological architecture of the donor tumor across serial passages. The engraftment biology of PDX systems, including immunodeficient host strain selection, implantation site, tumor source, and passage biology, is reviewed, together with humanized and autologous humanized configurations that extend the platform to immune checkpoint inhibitors, bispecific T-cell engagers, and chimeric antigen receptor T (CAR-T) cell therapy evaluation. This review addresses preclinical-to-clinical translation as a function of immunological divergence, incomplete tumor microenvironment recapitulation, and standardization. Formal frameworks, including the PDX Model Minimal Information (PDX-MI) standard and the Minimal Information for Standardization of Humanized Mice (MISHUM), are examined alongside global biobank infrastructure and emerging AI-driven translational modeling approaches.

Introduction

Cancer remains a leading cause of mortality worldwide, with an estimated 28.4 million new diagnoses projected annually by 20401. Despite remarkable therapeutic advances, including molecularly targeted agents, immune checkpoint inhibitors, chimeric antigen receptor T (CAR-T) cell therapies, and antibody–drug conjugates, the majority of patients with advanced malignancies develop drug resistance or fail to respond to initial treatment, and overall five-year survival rates for many solid tumors remain unsatisfactory2. Cancer drug development continues to be defined by a difficult translational bottleneck. A large proportion of oncology agents with encouraging preclinical activity fail during clinical testing, indicating that many experimental systems still do not capture the biological determinants of patient response3,4. The need for in vivo preclinical systems that faithfully recapitulate the genomic, histopathological, and pharmacological properties of human cancer is well established. Xenograft models have remained central to this problem, and to its potential solution, because they represent the most established in vivo bridge between cell culture and the clinic. Xenograft models are created by implanting human tumor material into an immunologically permissive host, most commonly an immunodeficient mouse5. Two major forms dominate the field, cell line-derived xenografts (CDX) that are generated from established cancer cell lines and patient-derived xenografts (PDX) that are generated by direct engraftment of fresh surgical or biopsy material with no intervening long-term culture step6,7. Orthotopic, disseminated, subrenal capsule, organoid-derived, and humanized variants further extend this framework by changing implantation site, source material, or host immune status8,9,10.

Human malignancies are dynamic systems shaped by somatic mutation, epigenetic reprogramming, intratumor heterogeneity, and continuous co-evolutionary interactions between neoplastic and stromal cell populations6,11. While cancer cell lines were historically considered less effective predictors of clinical response3,4, this view has since been refined10,12,13. While cell lines do suffer from genomic drift, loss of stromal and immune components, and divergence from parental tumor heterogeneity, much of their apparent non-predictivity reflects study design (small panels, mismatched histologies, and absent molecular stratification) rather than intrinsic model failure14. PDX models partially address the principal limitations of cell line models (genomic drift, loss of heterogeneity, and absence of stroma) by engrafting fresh patient tumor fragments directly into immunodeficient mice, preserving donor histology, genomic architecture, and aspects of stromal organization across early passages, with orthotopic, disseminated, subrenal capsule, organoid-derived, and humanized variants extending the framework across implantation site, source material, and host immune status10. Population-scale PDX trials recapitulate clinical responses at the indication level12, and matched patient–PDX pairs show ~87% concordance with individual patient outcomes, although specificity in that analysis was 70%13.

The xenograft landscape extends further to encompass humanized platforms, in which PDX tumors are grown in hosts reconstituted with components of a human immune system9, alongside orthotopic models at anatomically cognate primary sites and metastatic variants14. At the same time, this review focuses on murine xenograft models; complementary systems such as chick chorioallantoic membrane (CAM) assays15, zebrafish embryo xenografts16, and patient-derived organoids17 can offer orthogonal advantages in speed, throughput, and scalability, positioning them within integrated multi-platform preclinical workflows. International PDX biobanking infrastructure has expanded substantially over the past decade18, and formal standardization frameworks - the PDX Model Minimal Information (PDX-MI) standard19 and the Minimal Information for Standardization of Humanized Mice (MISHUM)20, have been established; however, incomplete adoption in the published literature remains a structural barrier to clinical translation. Formal standardization frameworks have been established to harmonize reporting of clinical annotation, engraftment methodology, passage history, quality assurance, and immune reconstitution parameters. Despite these advances, no single xenograft platform is sufficient for every translational question. PDX models are inherently slow, resource-intensive, and subject to variable take rates, while humanized xenograft systems, although partially addressing the immune deficit of conventional PDX hosts, introduce additional variability through incomplete immune reconstitution9, inter-donor heterogeneity21, and graft-versus-host constraints22. Standardization, moreover, remains uneven: model establishment protocols, passage numbering conventions, authentication methods, tumor growth quantification, immune monitoring, endpoint definitions, and overall reporting quality vary substantially between institutions and publications23. This review examines xenograft models through the dual lens of standardization and practical model selection. Positioning CDX as the essential early in vivo workhorse and PDX as the higher-fidelity but more resource-intensive platform, we synthesize current evidence on model taxonomy, host strain biology, engraftment determinants, biobanking, genomic stability, quantitative study design, and reporting frameworks, and contextualize xenograft systems alongside syngeneic models, organoids, genetically engineered mouse models (GEMM), and rapid alternative assays. The central argument is that standardization should not be understood as an effort to consolidate studies into a single model class, but rather as a framework defining how each platform is generated, validated, reported, and rationally positioned within an integrated preclinical workflow (Figure 1).

Review and Perspective

Historical development and taxonomy

Yamagiwa and Ichikawa's 1915 demonstration of carcinogen-induced skin cancer in rabbits established the experimental tumor model concept24. Syngeneic models emerged in the mid-twentieth century with foundational lines, Lewis lung carcinoma, 4T1 mammary carcinoma, CT26 colorectal carcinoma, and B16 melanoma, that retain utility in the current immuno-oncology era25. Flanagan's 1966 discovery of the athymic nude mouse provided the first immunocompromised host capable of tolerating xenogeneic tissue26, enabling Rygaard and Povlsen to achieve the first successful human colorectal tumor engraftment in 196927. Subsequent strain development through SCID, NOD/SCID, and IL-2 receptor common gamma chain (Il2rg)-null backgrounds progressively eliminated T-cell, B-cell, and NK-cell barriers28, establishing the contemporary taxonomy: CDX, PDX, orthotopic and disseminated variants, organoid-derived xenografts, and humanized xenografts6.

The NCI-60 panel, established in the late 1980s as an in vitro screen of 60 human cancer cell lines spanning nine tissue types, institutionalized systematic drug evaluation and anchored cell line-based preclinical pharmacology for several decades, with hit compounds progressed into subcutaneous CDX studies in athymic nude mice using standardized endpoints such as %T/C (percentage of treated to control tumor volume), tumor growth delay, and drug-related deaths to enable inter-study comparison29. CDX models remain the most widely used and cost-effective platform for preclinical efficacy testing; subcutaneous CDX studies reach readout within 5–7 weeks (Figure 1) and occupy a central role in the drug development workflow for dose optimization, pharmacokinetic/pharmacodynamic (PK/PD) characterization, tumor growth inhibition (TGI) quantification, and survival endpoint assessment (Figure 2).

From a practical standpoint, xenograft model taxonomy is defined by three variables: tumor source, implantation site, and host background. CDX models rely on established cancer cell lines with a stable supply and extensive molecular and pharmacological annotation through resources such as the Cancer Cell Line Encyclopedia30. PDX models use freshly engrafted patient tumor fragments to preserve donor histology, genomic architecture, and intratumoral heterogeneity, although human stroma is progressively replaced by murine stroma with serial passaging10. Implantation site determines the biological context: subcutaneous engraftment prioritizes technical simplicity, reproducibility, and serial caliper measurement; orthotopic implantation restores the organ-specific microenvironment and supports clinically relevant patterns of spontaneous metastasis; intravascular injection (tail-vein, intracardiac, intrasplenic) generates experimental metastasis models that probe later steps of the metastatic cascade such as extravasation and colonization14; and subrenal capsule implantation exploits a highly vascularized site that markedly improves take rates for difficult-to-engraft tumors31.

CDX and PDX: Complementary roles in preclinical oncology

Accumulating evidence indicated that CDX models had limited ability to predict clinical drug response, prompting a reassessment of preclinical oncology methodology. Established cancer cell lines, maintained indefinitely in artificial culture, undergo progressive clonal selection, transcriptional divergence, and loss of stromal and immune context relative to the primary tumors from which they were derived. PDX models preserve donor histology, genomic architecture, and intratumoral heterogeneity across successive passages10. In 2006, a pancreatic cancer PDX platform provided the first demonstration of biomarker-driven drug response prediction32. In 2011, the 'xenopatient' platform assembled 85 molecularly annotated colorectal PDX models, recapitulated clinical cetuximab response patterns, and identified HER2 amplification as a targetable resistance mechanism - validated subsequently in the HERACLES trials33,34. A landmark 2015 study applied a 1 × 1 × 1 PDX clinical trial (PCT) design across ~1,000 models and 62 treatment regimens, reproducing genotype-response associations and nominating novel resistance mechanisms, establishing PDX-scale pharmacology as a credible bridge to clinical trial design12.

Despite these advances, CDX models remain a gold standard for early-stage preclinical in vivo evaluation of candidate therapeutics, because they are fast, experimentally clean, and cost-efficient: authenticated cell lines can be genetically manipulated, expanded at scale, and implanted in large matched cohorts with comparatively little biological noise, enabling interpretable efficacy, pharmacodynamic, and tolerability readouts on timelines impractical for PDX or humanized systems. Canonical lines such as HCT116, A549, and U8735 are anchored to decades of mechanistic literature and public omics resources, and orthotopic or disseminated CDX variants extend their utility to tissue tropism, metastatic colonization, and site-specific therapy without sacrificing operational advantages10. For hypothesis-driven questions centered on target dependence, dose-schedule optimization, comparative efficacy, or proof-of-mechanism, CDX remains the appropriate first in vivo platform. Its recognized limitations, including clonal drift, absent human stroma and immune compartments, and limited population-level predictive value, define its proper use as a rapid triage and mechanism platform rather than a surrogate for clinical outcome.

PDX model establishment: Tumor sources, implantation, and engraftment biology

Viable tumor material is obtained most commonly from surgical resection and also from core biopsy, malignant effusions, or marrow aspirates in hematologic malignancies31. Radical resection provides larger volumes of viable, architecturally intact tissue; biopsies nonetheless yield adequate material when directed to metastatic or unresectable disease31,36. Tissue must be transported in cold buffered medium and processed within ~24 h; warm ischemia exceeding 2 h and ex vivo intervals beyond 8–10 h progressively reduce engraftment efficiency. Institutional review board approval and written informed consent are mandatory.

Tumor material is processed into 1–3 mm3 fragments, preserving extracellular matrix architecture and spatial heterogeneity, or dissociated into single-cell suspensions for sorting and dose-normalized inoculation. Matrigel co-implantation is frequently used to support early vascularization; its effect is variable and should be reported as it alters the microenvironmental context31. Hormone supplementation, estradiol pellets for ER-positive breast tumors and dihydrotestosterone for prostate tumors, improves engraftment of hormone-dependent malignancies. Subcutaneous dorsal flank implantation offers caliper monitoring and operational simplicity; orthotopic implantation restores the organ-specific microenvironment; and subrenal capsule implantation maximizes take rates for difficult-to-engraft tumor types, including non-small cell lung cancer (NSCLC) and prostate cancer31.

Engraftment is strongly tumor-dependent. Aggressive, poorly differentiated, treatment-resistant, or metastatic tumors engraft more readily than indolent or well-differentiated ones; triple-negative breast cancers, pancreatic ductal adenocarcinomas, high-grade serous ovarian cancers, and acute leukemias consistently outperform luminal A breast tumors and other low-take entities, and successful engraftment itself predicts poor patient prognosis in triple-negative breast cancer and pancreaticobiliary malignancies10. Host strain is a second major determinant: highly immunodeficient strains such as NSG and NOG have supplanted nude and SCID mice for challenging or hematologic cases because of their broader permissiveness9.

By convention, the initial patient implantation is designated F0 (or P0), with serial passages labeled F1, F2, and onward. Growth kinetics typically stabilize after early passages, and most programs reserve formal pharmacology for F2 or F3 material to reduce inter-mouse variability, although no universal threshold exists; most programs define early passage as F2–F4, with material beyond F5 carrying increasing risk of subclonal outgrowth and transcriptional drift at a rate that is tumor-type dependent. Because passage number influences both reproducibility and biological drift, passage history, host strain, implantation site, and provenance should be recorded for every study in accordance with the PDX-MI reporting standard19.

Immunodeficient host strains

Athymic nude (Foxn1nu) mice lack mature T cells but retain B cells, NK cells, and innate immunity. CB-17 scid mice, carrying a loss-of-function mutation in Prkdc, lack T and B cells through defective V(D)J recombination but retain NK cells and develop "leaky" lymphocytes with age9. The scid mutation on the NOD/ShiLtJ background (NOD/SCID) added defects in innate immunity, including complement deficiency, impaired macrophage and dendritic cell function, and reduced NK activity. Addition of a null mutation in the IL-2 receptor common gamma chain (Il2rg), required for signaling by IL-2, IL-4, IL-7, IL-9, IL-15, and IL-21, eliminates functional NK cells and cripples residual innate immunity, yielding NSG (NOD.Cg-Prkdcscid Il2rgtm1wjI/SzJ) and NOG (NODShi.Cg-Prkdcscid Il2rgtm1sug); NSG carries a complete Il2rg knockout, whereas NOG expresses a truncated receptor that binds but cannot signal. NSG and NOG support engraftment of solid and hematologic malignancies and are the default hosts for PDX work. The NRG strain (NOD.Cg-Rag1tm1Mom Il2rgtm1wjI) replaces Prkdc with Rag1 knockout, preserving DNA-repair competence. The BRG strain (BALB/c-Rag2null Il2rgnull) provides a BALB/c-background, radiation-tolerant alternative that avoids the NOD-specific Sirpa allele. The SRG rat (Sprague-Dawley Rag2/Il2rg double knockout) supports cell lines and PDXs that grow poorly in NSG, including VCaP prostate cancer at >90% take rate, and enables serial blood and tissue sampling, orthotopic surgery, and physiologically relevant pharmacokinetics37.

Humanized xenograft systems reconstitute severely immunodeficient mice with human immune components before or concurrent with tumor engraftment38. The three principal formats are hu-PBMC (injection of adult peripheral blood mononuclear cells), hu-CD34 (intravenous human CD34⁺ hematopoietic stem cells (HSCs) after sublethal irradiation or busulfan conditioning), and BLT (co-implantation of fetal liver and thymus with CD34⁺ cells, yielding the most complete lymphoid development). These models support evaluation of immune checkpoint inhibitors, bispecific antibodies, cellular therapies, and immune-mediated resistance mechanisms38. Autologous or HLA-matched designs pair the tumor and immune compartments from the same donor. Reconstitution efficiency depends on donor source, host strain, conditioning regimen, and study duration; myeloid, stromal, and lymphoid compartments are typically incomplete or imbalanced: hu-PBMC models are deficient in NK cells, B cells, and myeloid lineages despite robust T-cell engraftment; hu-CD34 models support multilineage output but underrepresent conventional dendritic cells and tissue macrophages; and BLT models achieve superior thymic T-cell education but remain suboptimal for myeloid reconstitution. hu-PBMC mice achieve rapid T-cell reconstitution but develop xenogeneic graft-versus-host disease (GvHD) within three to 6 weeks through human T-cell recognition of murine MHC; NSG-MHC-I/II-null variants partially mitigate this22. hu-CD34 mice permit longer studies but require extended reconstitution and do not fully reproduce human myeloid, NK, or stromal biology.

Orthotopic and metastasis models

Subcutaneous xenografts, though the default first-line preclinical platform, place human tumor tissue in a biological context that diverges from the native organ microenvironment. Extracellular matrix composition, vascular architecture, oxygen tension, stromal populations, and paracrine signaling of the dorsal subcutaneous space differ from those of the breast, pancreas, peritoneum, colon, or lung10. Orthotopic implantation restores organ-specific microenvironment and typically supports enhanced vascularization and spontaneous metastasis that more closely recapitulate natural cancer progression, at the cost of greater surgical complexity and the need for imaging-based monitoring39.

Orthotopic PDX techniques have expanded across tumor types. Breast cancer PDX is established in the mammary fat pad, supporting spontaneous lung and lymph node metastasis40. Colorectal cancer PDX implanted into the cecal or colonic wall recapitulates local invasion and liver metastasis39. Ovarian cancer PDX delivered intraperitoneally reproduces the peritoneal dissemination characteristic of high-grade serous disease41. Pancreatic ductal adenocarcinoma PDX implanted into the pancreas reproduces the desmoplastic stromal reaction and early local invasion32. Two principal metastasis modeling strategies are used: experimental metastasis models, in which tumor cells are delivered intravascularly to assess later steps of the metastatic cascade, such as extravasation and colonization; and spontaneous metastasis models, in which orthotopic primary tumors disseminate through the complete cascade14.

Standardization, authentication, and study design

Four reporting and study-design frameworks define xenograft experiments. PDX-MI specifies minimum metadata for patient-derived models, including donor information, tumor source, host strain, implantation site, passage number, engraftment rate, and histological and molecular characterization; short tandem repeat (STR) authentication is included under model quality control to confirm provenance and detect cross-contamination19. MISHUM extends analogous requirements to humanized systems, covering human hematopoietic stem cell or PBMC source, conditioning regimen, host strain, reconstitution efficiency verified by flow cytometry, and GvHD monitoring20. OBSERVE (Oncology Best-practices: Signs, Endpoints and Refinements for In Vivo Experiments), a 2024 EurOPDX/INFRAFRONTIER consensus, provides refinement and welfare guidance specific to rodent cancer models, complementing the broader PREPARE and ARRIVE 2.0 frameworks23,42. The 2024 NCI PDXNet consensus recommendations address study design, tumor growth analysis, and response categorization for PDX pharmacology experiments, and provide a public suite of analytical tools43.

These frameworks are not fully harmonized, and a modern xenograft study may need to meet the requirements of PDX-MI, MISHUM, OBSERVE, and PDXNet recommendations simultaneously. In practice, passage number, tumor measurement formula, host source, authentication method, and pre-specified analytic strategy are often omitted43. A unified reporting schema spanning CDX, PDX, and humanized systems remains an unmet need. Publicly accessible PDX repositories include the NCI Patient-Derived Models Repository (PDMR), the EurOPDX Data Portal, and PDX Finder, which collectively catalog several thousand models with associated genomic, histological, and pharmacological data to support model identification and acquisition18. CDX studies require routine STR profiling and contamination control, whereas PDX studies require linkage to the donor case, molecular confirmation of driver features where feasible, and documentation of human versus murine cellular content across passages19. Tumor volume estimation remains inconsistent between studies, with different formulas applied without explicit disclosure, altering treatment effect estimates and complicating cross-study comparison; standardization of growth metrics, response thresholds, censoring rules, and visualization formats is therefore as important as standardization of engraftment methods43.

Genomic fidelity, clonal evolution, and translational interpretation

PDX retains driver mutations and copy-number landscapes across early passages more faithfully than CDX, supporting its use for biomarker development, resistance studies, and co-clinical modeling44. However, mouse-specific tumor evolution, stromal replacement, and epigenetic reprogramming occur during passaging, and some apparent discordance reflects subclonal sampling of spatially heterogeneous tumors rather than model-induced drift45. Early passages provide stronger biological correspondence than extensively passaged material, though both are shaped by selective pressures of implantation site, host immunity, and prior treatment45. Two practical principles follow: PDX models should be characterized longitudinally rather than assumed static, and integration of sequencing, histopathology, transcriptional profiling, and treatment response is recommended when model selection carries major downstream consequences44.

Humanized mouse models and immuno-oncology applications

Humanized systems reconstituted with human PBMCs, CD34⁺ HSCs, or BLT tissue address the inability of conventional PDX hosts to model an intact human immune microenvironment. hu-PBMC models achieve rapid T-cell reconstitution but are constrained by xenogeneic GvHD within three to six weeks; hu-CD34 models require 8–16 weeks for multilineage reconstitution but support longer studies9. Introduction of human cytokine transgenes has further improved reconstitution of specific immune lineages. NSG-SGM3 mice (transgenic for human IL-3, GM-CSF, and SCF) support myeloid cell and mast cell development46. MISTRG mice (humanized for M-CSF, IL-3, GM-CSF, SIRPα, and thrombopoietin) and MISTRG6 mice (additionally humanized for IL-6) on a Rag2⁻/⁻ Il2rg⁻/⁻ background, produce the most comprehensive human myeloid, NK, and tissue-macrophage reconstitution currently available and support engraftment of adult bone-marrow-derived hematopoietic stem and progenitor cells that other strains fail to accommodate47. NOG-EXL mice (transgenic for human IL-3 and GM-CSF on the NOG background) support myeloid and mast cell reconstitution at lower cytokine levels than NSG-SGM3, which translates into improved long-term survival and less severe macrophage-activation syndrome, making them better suited to extended humanization and tumor studies. Humanized mice engrafted with human T cells enable modeling of cytokine release syndrome (CRS), permitting preclinical characterization of the safety profile of bispecific T-cell engagers, CD3-targeting antibodies, and related T-cell–activating biologics. No single strain is universally superior; selection is guided by the immune lineage of interest and study duration.

The highest-fidelity configuration is the autologous humanized PDX, in which patient-matched hematopoietic cells and tumor tissue are engrafted together, eliminating MHC mismatch between immune and tumor compartments47. Chiorazzi and colleagues demonstrated in 2023 that autologous MISTRG6 PDX models generate human innate and adaptive immune populations that infiltrate the tumor microenvironment, recapitulate activation and exhaustion programs in CD8⁺ T cells, and produce a VEGF-A-driven pro-tumor myeloid signature whose inhibition abrogates enhanced tumor growth, validating this configuration as a platform for immunotherapy evaluation47. Phoon and colleagues extended this approach to melanoma, showing that autologous PBMC-humanized PDX models capture inter-patient variability in response to checkpoint blockade and engineered IL-2 cytokines48,49. A parallel autologous approach, adoptive transfer of ex vivo expanded patient-derived tumor-infiltrating lymphocytes (TILs) into mice bearing matched PDX tumors, has been used to model TIL therapy efficacy in melanoma and other solid tumors50. Interpretation of immunotherapy studies in humanized systems nonetheless requires caution: several immune checkpoint antibodies (e.g., anti-CTLA4, anti-PD1) have species-restricted cross-reactivity, dendritic cell function and antigen presentation remain incomplete in most humanized strains, and structures such as tertiary lymphoid organs and fully functional germinal centers are rarely reconstituted51.

CAR-T cell therapy and xenograft models

CAR-T cell therapy has multiple regulatory approvals for hematological malignancies, and xenograft evaluation in immunodeficient mice bearing CDX or PDX tumors is the standard preclinical platform for human CAR-T products prior to clinical translation52. The immunological deficit of NSG-class hosts excludes macrophage activation, dendritic cell cross-presentation, and regulatory T-cell–mediated suppression from the experimental system, and murine xenografts did not predict the cytokine-driven systemic toxicities later observed in patients52. A 2025 systematic review and meta-analysis of 422 clinical trials and 3,157 preclinical studies reported that preclinical efficacy signals in CAR-T xenograft models are largely homogeneous and antigen-independent and do not discriminate between CAR constructs that subsequently succeed or fail in clinical trials53. Fløe and colleagues documented in 2025 that species-level differences in cytokine signaling, antigen density, T-cell receptor repertoire, and tumor-immune crosstalk limit the translational capacity of CAR-T xenograft data, and that syngeneic models are employed in ~4% of published preclinical CAR-T studies52,53. Cytokine release syndrome (CRS) and immune effector cell-associated neurotoxicity syndrome (ICANS) are not reproduced in standard NSG xenografts. Mechanistic studies in NSG-SGM3 hosts indicated that CRS is driven primarily by monocyte-derived IL-1 and IL-6, with IL-1 and IL-6 differentially required for CRS versus neurotoxicity. Also, GM-CSF neutralization reduces CRS and neuroinflammation while preserving CAR-T antitumor activity. These pathways are captured in cytokine-transgenic and autologous humanized configurations rather than in conventional NSG-based xenografts. Syngeneic murine CAR-T models provide a complementary platform that permits evaluation of CAR-T interactions with an intact host immune system, at the cost of reliance on murine rather than human target biology. Xenograft-based evaluation supported the clinical translation of CD19-directed CAR-T therapy, where activity against B-ALL and B-NHL models in NSG hosts corresponded to durable clinical responses54.

Drug development, biomarker discovery, and AI-driven translation

PDX models aim to predict clinical drug responses with greater fidelity than CDX or GEMMs. The 2015 Novartis PDX clinical trial (PCT) of Gao and colleagues applied a 1 × 1 × 1 design across ~1,000 PDX models, 62 treatment regimens, and six indications and identified genotype-response associations together with established and novel resistance mechanisms12. Extension of the 1 × 1 × 1 concept to the individual-patient level is constrained by a 3–6 month establishment timeline, which typically positions PDX-guided decisions at second- or third-line therapy10. PDX studies have contributed to validated clinical biomarkers, including KRAS mutation as a predictor of anti-EGFR resistance in colorectal cancer, EGFR-activating and T790M mutations for NSCLC tyrosine kinase inhibitor selection, HER2 amplification in breast and gastric cancer, and platinum resistance in ovarian cancer10.

Large PDX drug response datasets have enabled machine learning approaches to drug response prediction. Yang and colleagues described TRANSPIRE-DRP in 2025, a deep-learning framework using unsupervised domain adaptation to translate PDX transcriptomic data into patient-level response predictions, evaluated across cetuximab, paclitaxel, and gemcitabine55. Tosca and colleagues published a translational modeling framework in 2024 that scaled PDX tumor growth rates allometrically to predict tumor volume doubling time and progression-free survival in untreated patient populations across eleven solid cancer types, with 91% of predicted tumor volume doubling time medians within 1.5-fold of clinical observations56. Unresolved challenges remain: biobank selection bias toward engraftable tumors, limited prospective clinical validation, and the absence of a defined regulatory pathway for AI-derived preclinical evidence33,55.

Xenograft platforms answer different translational questions at different development stages rather than competing for a single role. CDX models remain central for early mechanistic work and high-throughput screening; PDX models provide the genomic fidelity and interpatient heterogeneity required for biomarker validation and clinical response modeling; humanized and autologous humanized configurations address immunotherapy-specific questions that conventional xenografts cannot; and organoid-based platforms extend ex vivo throughput when tissue is limiting; these platforms collectively support preclinical evaluation of novel, established, and alternative cancer therapeutics57,58,59. AI and machine learning frameworks such as TRANSPIRE-DRP and translational PK/PD models now link PDX drug response data to patient-level response predictions, but their clinical utility depends on the quality and standardization of the underlying model annotation. The central unresolved issues are predictive validity, reporting reproducibility, biobank representativeness, and the regulatory status of AI-derived preclinical evidence. Progress will depend less on the introduction of new model categories than on disciplined use of existing ones, supported by harmonized metadata, longitudinal fidelity assessment, and transparent cross-platform integration.

Conclusions

Xenograft models have evolved from simple subcutaneous implants of established cell lines into a multifaceted experimental framework encompassing CDX and PDX models, orthotopic and metastasis variants, and humanized immune-reconstituted systems, including autologous configurations. This evolution has been driven by the progressive recognition that fidelity to human cancer biology, in terms of genomic landscape, intratumor heterogeneity, and tumor microenvironment, is a principal determinant of preclinical predictive value.

CDX models remain central to preclinical oncology as the most widely used and cost-effective in vivo platform for early-phase drug evaluation, providing reproducible, scalable data for dose optimization, PK/PD characterization, tumor growth inhibition assessment, and survival endpoint evaluation. PDX models provide the fidelity required for biomarker discovery, co-clinical trial design, and pharmacological applications that depend on preservation of donor genomic and histological characteristics. The immunodeficiency hierarchy - from nude and SCID through NSG, NOG, NRG, and BRG, with the emerging SRG rat, reflects progressive refinement of engraftment permissiveness. Humanized and autologous configurations extend these platforms to immune checkpoint inhibitors, bispecific antibodies, and CAR-T cell therapy evaluation, though species-level immunological limitations remain.

Improving preclinical translation requires rigorous study design and standardized reporting. PDX-MI, MISHUM, OBSERVE, and the NCI PDXNet consensus recommendations provide the necessary reporting frameworks; consistent adoption by researchers, journals, and funding agencies remains the critical determinant of impact. The trajectory of xenograft science converges on four priorities: standardized reporting across all platforms, rational platform selection aligned with the translational question, continued refinement of humanized model engineering, and AI-driven translation frameworks that translate large-scale PDX datasets into clinically actionable response predictions.

figure-results-1
Figure 1: Xenograft mouse models: establishment-to-readout timelines with comparative advantages and limitations. Timelines from initiation to experimental readout are shown across the principal preclinical platforms used in oncology drug development, grouped by host immune status. The upper panel covers xenograft models in immunodeficient hosts: subcutaneous and orthotopic or disseminated cell line-derived xenografts (CDX), subcutaneous patient-derived xenografts (PDX), subrenal capsule PDX (PDX-SRC), patient-derived orthotopic xenografts (PDOX), organoid-derived xenografts (ODX), and humanized PDX (huPDX). The lower panel covers immunocompetent and autochthonous platforms: syngeneic models, genetically engineered mouse model–derived allografts (GDA), and genetically engineered mouse models (GEMM). For each platform, experimental phases are color-coded (acclimatization, implant receipt, organoid culture, engraftment and growth, passage, humanization, and treatment/readout; or breeding and tumor latency/treatment/readout for autochthonous models), with approximate total durations indicated in weeks. Platform-specific advantages and limitations are summarized in the right-hand panel. Timelines represent standard operating procedures implemented at Altogen Labs and are consistent with published ranges in the field (https://altogenlabs.com/xenograft-models); actual durations vary by tumor type, host strain, study design, and engraftment kinetics. BLI, bioluminescence imaging; GvHD, graft-versus-host disease; HSC, hematopoietic stem cell; IO, immuno-oncology; MRI, magnetic resonance imaging; NSCLC, non-small cell lung cancer; PBMC, peripheral blood mononuclear cell; SC, subcutaneous; TME, tumor microenvironment; TNBC, triple-negative breast cancer. Please click here to view a larger version of this figure.

figure-results-2
Figure 2: Integrated preclinical oncology workflow: from in vitro efficacy through Investigational New Drug (IND)-enabling studies. Schematic of the tiered preclinical drug development pipeline, organized across four sequential stages. In vitro efficacy encompasses target engagement (IC50, KD), antiproliferative activity across cell line panels including the NCI-60 and disease-specific lines, patient-derived cells (PDC), mechanistic and pathway assays, three-dimensional models, and patient-derived tumor organoids (PDTO), migration and invasion assays, cytokine and immune profiling, angiogenesis and cancer stem cell assays, clonogenic assays, selectivity profiling, immune co-culture assays, resistance modeling, biomarker identification, target expression profiling, pharmacodynamic biomarker readouts, and combination studies. In vivo CDX covers subcutaneous, orthotopic, and disseminated or systemic CDX for dose optimization, PK/PD characterization, tumor volume and TGI assessment, partial and complete response classification (PR/CR), survival endpoints (OS, EFS, TTE), multimodal imaging (BLI/IVIS, MRI, PET/CT), tolerability readouts including body weight loss (BWL) and clinical signs, molecular endpoints (IHC, ELISA, WES, flow cytometry, RNA-seq, metastasis counts), and circulating tumor cell (CTC) enumeration. Advanced in vivo platforms include subcutaneous and orthotopic PDX, subrenal capsule (SRC) PDX, humanized PDX, syngeneic models for immuno-oncology benchmarking, GEMM and GDA, ODX, non-rodent xenograft models, hollow-fiber assays (HFA), PDX mouse clinical trials (MCT), and resistance/relapse models. IND-enabling studies comprise single- and repeat-dose GLP toxicology, toxicokinetics, safety pharmacology, local tolerance, genotoxicity, immunotoxicity, reproductive and developmental toxicity, first-in-human (FIH) dose projection, cytokine release evaluation, and phototoxicity assessment on a risk basis. Parallel chemistry, manufacturing, and controls (CMC) and absorption, distribution, metabolism, and excretion/pharmacokinetics (ADME/PK) characterization include in silico PBPK and IVIVE modeling, solubility and stability, permeability and P-glycoprotein (P-gp) efflux, cytochrome P450 (CYP) inhibition and induction, metabolic assays, in vivo PK, plasma protein binding, exposure–response modeling, tissue distribution, human PK prediction, metabolite identification, and drug–drug interaction (DDI) risk assessment. Please click here to view a larger version of this figure.

Disclosures

The author is employed by Altogen Labs (Austin, TX, USA), a contract research organization that provides preclinical oncology research services. Figure 1 and Figure 2 are original author-generated schematics. The figures incorporate timelines and workflows derived from Altogen Labs standard operating procedures. No conflict of interest declared.

References

  1. Sung H, Ferlay J, Siegel RL, et al. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2021;71(3):209-249.
  2. Islami F, Wiese D, Schafer EJ, Sung H, Jemal A. Stage-specific cancer survival in Black and White persons by urbanicity of county of residence, United States, 2015-2021. Cancer. 2025:e70073.
  3. Johnson JI, Decker S, Zaharevitz D, et al. Relationships between drug activity in NCI preclinical in vitro and in vivo models and early clinical trials. Br J Cancer. 2001;84(10):1424-1431.
  4. Voskoglou-Nomikos T, Pater JL, Seymour L. Clinical predictive value of the in vitro cell line, human xenograft, and mouse allograft preclinical cancer models. Clin Cancer Res. 2003;9(11):4227-4239.
  5. Jung J. Human tumor xenograft models for preclinical assessment of anticancer drug development. Toxicol Res. 2014;30(1):1-5.
  6. Liu Y, Wu W, Cai C, et al. Patient-derived xenograft models in cancer therapy: technologies and applications. Signal Transduct Target Ther. 2023;8:160.
  7. Okada S, Vaeteewoottacharn K, Kariya R. Application of highly immunocompromised mice for the establishment of patient-derived xenograft (PDX) models. Cells. 2019;8(8):889.
  8. Blanchard Z, Brown EA, Ghazaryan A, et al. PDX models for functional precision oncology and discovery science. Nat Rev Cancer. 2025;25(3):153-166.
  9. Shultz LD, Brehm MA, Garcia-Martinez JV, et al. Humanized mice for immune system investigation: progress, promise and challenges. Nat Rev Immunol. 2012;12(11):786-798.
  10. Hidalgo M, Amant F, Biankin AV, et al. Patient-derived xenograft models: an emerging platform for translational cancer research. Cancer Discov. 2014;4(9):998-1013.
  11. Dobrolecki LE, Airhart SD, Alferez DG, et al. Patient-derived xenograft (PDX) models in basic and translational breast cancer research. Cancer Metastasis Rev. 2016;35(4):547-573.
  12. Gao H, Korn JM, Ferretti S, et al. High-throughput screening using patient-derived tumor xenografts to predict clinical trial drug response. Nat Med. 2015;21(11):1318-1325.
  13. Izumchenko E, Paz K, Ciznadija D, Sloma I, Katz A, Vasquez-Dunddel D, et al. Patient-derived xenografts effectively capture responses to oncology therapy in a heterogeneous cohort of patients with solid tumors. Ann Oncol. 2017;28(10):2595-2605.
  14. Khanna C, Hunter K. Modeling metastasis in vivo. Carcinogenesis. 2005;26(3):513-523.
  15. Ribatti D. The chick embryo chorioallantoic membrane (CAM) assay. Reprod Toxicol. 2017;70:97-101.
  16. Fazio M, Ablain J, Chuan Y, et al. Zebrafish patient avatars in cancer biology and precision cancer therapy. Nat Rev Cancer. 2020;20(5):263-273.
  17. Drost J, Clevers H. Organoids in cancer research. Nat Rev Cancer. 2018;18(7):407-418.
  18. Conte N, Mason JC, Halmagyi C, et al. PDX Finder: a portal for patient-derived tumor xenograft model discovery. Nucleic Acids Res. 2019;47(D1):D1073-D1079.
  19. Meehan TF, Conte N, Goldstein T, et al. PDX-MI: minimal information for patient-derived tumor xenograft models. Cancer Res. 2017;77(21):e62-e66.
  20. Stripecke R, Münz C, Schuringa JJ, et al. Innovations, challenges, and minimal information for standardization of humanized mice. EMBO Mol Med. 2020;12(7):e8662.
  21. Brehm MA, Wiles MV, Greiner DL, et al. Generation of improved humanized mouse models for human infectious diseases. J Immunol Methods. 2014;410:3-17.
  22. King MA, Covassin L, Brehm MA, et al. Human peripheral blood leucocyte non-obese diabetic-severe combined immunodeficiency interleukin-2 receptor gamma chain gene mouse model of xenogeneic graft-versus-host-like disease and the role of host major histocompatibility complex. Clin Exp Immunol. 2009;157(1):104-118.
  23. Percie du Sert N, Hurst V, Ahluwalia A, et al. The ARRIVE guidelines 2.0: updated guidelines for reporting animal research. Br J Pharmacol. 2020;177(16):3617-3624.
  24. Sajjad H, Imtiaz S, Noor T, et al. Cancer models in preclinical research: a chronicle review of advancement in effective cancer research. Anim Models Exp Med. 2021;4(2):87-103.
  25. Mosely SI, Prime JE, Sainson RC, et al. Rational selection of syngeneic preclinical tumor models for immunotherapeutic drug discovery. Cancer Immunol Res. 2017;5(1):29-41.
  26. Flanagan SP. "Nude", a new hairless gene with pleiotropic effects in the mouse. Genet Res. 1966;8(3):295-309.
  27. Rygaard J, Povlsen CO. Heterotransplantation of a human malignant tumour to "Nude" mice. Acta Pathol Microbiol Scand. 1969;77(4):758-760.
  28. Chen J, Liao S, Xiao Z, et al. The development and improvement of immunodeficient mice and humanized immune system mouse models. Front Immunol. 2022;13:1007579.
  29. Shoemaker RH. The NCI60 human tumour cell line anticancer drug screen. Nat Rev Cancer. 2006;6(10):813-823.
  30. Barretina J, Caponigro G, Stransky N, et al. The Cancer Cell Line Encyclopedia enables predictive modelling of anticancer drug sensitivity. Nature. 2012;483(7391):603-607.
  31. Liu M, Yang X. Patient-derived xenograft models: current status, challenges, and innovations in cancer research. Genes Dis. 2025;12(5):101520.
  32. Rubio-Viqueira B, Jimeno A, Cusatis G, et al. An in vivo platform for translational drug development in pancreatic cancer. Clin Cancer Res. 2006;12(15):4652-61.
  33. Bertotti A, Migliardi G, Galimi F, et al. A molecularly annotated platform of patient-derived xenografts ("xenopatients") identifies HER2 as an effective therapeutic target in cetuximab-resistant colorectal cancer. Cancer Discov. 2011;1(6):508-23.
  34. Sartore-Bianchi A, Lonardi S, Martino C, et al. Pertuzumab and trastuzumab emtansine in patients with HER2-amplified metastatic colorectal cancer: the phase II HERACLES-B trial. ESMO Open. 2020;5(5):e000911.
  35. Ovcharenko D, Chitjian C, Kashkin A, et al. Two dichloric compounds inhibit in vivo U87 xenograft tumor growth. Cancer Biol Ther. 2019;20(9):1281-1289.
  36. Hernandez MC, Bergquist JR, Leiting JL, et al. Patient-derived xenografts can be reliably generated from patient clinical biopsy specimens. J Gastrointest Surg. 2019;23(4):818-824.
  37. Noto FK, Sangodkar J, Adedeji BT, et al. The SRG rat, a Sprague-Dawley Rag2/Il2rg double-knockout validated for human tumor oncology studies. PLoS One. 2020;15(10):e0240169.
  38. De La Rochere P, Guil-Luna S, Decaudin D, et al. Humanized mice for the study of immuno-oncology. Trends Immunol. 2018;39(9):748-763.
  39. Hoffman RM. Patient-derived orthotopic xenografts: better mimic of metastasis than subcutaneous xenografts. Nat Rev Cancer. 2015;15(8):451-2.
  40. Okano M, Oshi M, Butash A, et al. Orthotopic implantation achieves better engraftment and faster growth than subcutaneous implantation in breast cancer patient-derived xenografts. J Mammary Gland Biol Neoplasia. 2020;25(1):27-36.
  41. Scott CL, Becker MA, Haluska P, et al. Patient-derived xenograft models to improve targeted therapy in epithelial ovarian cancer treatment. Front Oncol. 2013;3:295.
  42. De Vleeschauwer SI, van de Ven M, Oudin A, et al. OBSERVE: guidelines for the refinement of rodent cancer models. Nat Protoc. 2024;19(9):2571-2596.
  43. Meric-Bernstam F, Lloyd MW, Koc S, et al. Assessment of patient-derived xenograft growth and antitumor activity: the NCI PDXNet consensus recommendations. Mol Cancer Ther. 2024;23(7):924-938.
  44. Woo XY, Giordano J, Srivastava A, et al.; PDXNET Consortium; EurOPDX Consortium. Conservation of copy number profiles during engraftment and passaging of patient-derived cancer xenografts. Nat Genet. 2021;53(1):86-99.
  45. Hynds RE, Huebner A, Pearce DR, et al. Representation of genomic intratumor heterogeneity in multi-region non-small cell lung cancer patient-derived xenograft models. Nat Commun. 2024;15(1):4653.
  46. Wunderlich M, Chou FS, Sexton C, et al. Improved multilineage human hematopoietic reconstitution and function in NSGS mice. PLoS One. 2018;13(12):e0209034.
  47. Chiorazzi M, Martinek J, Krasnick B, et al. Autologous humanized PDX modeling for immuno-oncology recapitulates features of the human tumor microenvironment. J Immunother Cancer. 2023;11(7):e006921.
  48. Phoon YP, Lopes JE, Pfannenstiel LW, et al. Autologous human preclinical modeling of melanoma interpatient clinical responses to immunotherapeutics. J Immunother Cancer. 2024;12(4):e008066.
  49. Tsimafeyeu I, Makhov P, Ovcharenko D, et al. A novel anti-FGFR1 monoclonal antibody OM-RCA-01 exhibits potent antitumor activity and enhances the efficacy of immune checkpoint inhibitors in lung cancer models. Immunooncol Technol. 2024;23:100725.
  50. Jespersen H, Lindberg MF, Donia M, et al. Clinical responses to adoptive T-cell transfer can be modeled in an autologous immune-humanized mouse model. Nat Commun. 2017;8(1):707.
  51. Morton JJ, Bird G, Refaeli Y, et al. Humanized mouse xenograft models: narrowing the tumor-microenvironment gap. Cancer Res. 2016;76(21):6153-6158.
  52. Fløe LVB, Pedersen MG, Møller BK. Animal models in preclinical evaluation of CAR-T cell therapy: advantages and limitations. Biochim Biophys Acta Rev Cancer. 2025;1880(6):189455.
  53. Andreu-Sanz D, Gregor L, Carlini E, et al. Predictive value of preclinical models for CAR-T cell therapy clinical trials: a systematic review and meta-analysis. J Immunother Cancer. 2025;13(6):e011698.
  54. Duncan BB, Dunbar CE, Ishii K. Applying a clinical lens to animal models of CAR-T cell therapies. Mol Ther Methods Clin Dev. 2022;27:17-31.
  55. Yang J, Wu T, Xue F, et al. TRANSPIRE-DRP: a deep learning framework for translating patient-derived xenograft drug response to clinical patients via domain adaptation. J Transl Med. 2025;23(1):1340.
  56. Tosca EM, Ronchi D, Rocchetti M, et al. Predicting tumor volume doubling time and progression-free survival in untreated patients from patient-derived-xenograft (PDX) models: a translational model-based approach. AAPS J. 2024;26(5):92.
  57. Ovcharenko D, Mukhin D, Ovcharenko G. Alternative cancer therapeutics: unpatentable compounds and their potential in oncology. Pharmaceutics. 2024;16(9):1237.
  58. Altogen Labs. Xenograft models in oncology: CDX, PDX, organoids, and humanized immuno-oncology platforms. Austin (TX): Altogen Labs; https://altogenlabs.com/xenograft-models/
  59. Sirotkin AV, Ovcharenko D, Mlyncek M. Identification of protein kinases that control ovarian hormone release by selective siRNAs. J Mol Endocrinol. 2010;44(1):45-53.

Reprints and Permissions

Request permission to reuse the text or figures of this JoVE article

Request Permission

Tags

Patient Derived XenograftCell Line XenograftTumor EngraftmentImmunodeficient MiceTumor MicroenvironmentTranslational OncologyBiomarker DiscoveryCAR T Therapy

Related Articles