Decoding drug resistance and apoptosis in biliary tract cancers using 3D tumor models

1 Introduction: biliary tract cancers—an unmet need

1.1 Biliary tract cancers (BTCs)

Biliary tract cancers are group of malignancies, primarily including intra-hepatic cholangiocarcinoma (iCCA), extrahepatic cholangiocarcinoma (eCCA), perihilar cholangiocarcinoma (pCCA) also called Klatskin tumors, distal cholangiocarcinoma (dCCA), gallbladder cancer (GBC) and ampulla of vater cancer (AoVC) (Søreide et al., 2025). pCCA is the most prevalent form of CCA followed dCCA and iCCA (Ilyas and Gores, 2013). GBC is the most common and aggressive type of all the BTCs and the vast majority are adenocarcinoma. It is characterized by local and vascular invasion, extensive regional lymph node metastases and distant metastases (Wistuba and Gazdar, 2004). BTCs exhibit significant geographic variation, with East Asia and South America, especially Chile and South Korea, experiencing the highest incidence and mortality rates. Europe and North America report lower rates. GBC is more prevalent in women, while iCCA is the primary cause of rising global mortality. Age significantly increases risk. In India, the incidence is moderate but remains a significant health problem in the northern Ganges belt due to environmental factors, chronic infections, and socio-economic disparities. Increased GBC rates in Indian women point to environmental and genetic influences (Baria et al., 2022). CCA risk factors span several categories, including general factors like older age, obesity, type II diabetes, alcohol use, and smoking, as well as congenital conditions such as Caroli’s disease and choledochal cysts. Additional risks arise from viral infections (HBV, HCV), chronic inflammatory states (PSC, hepatolithiasis, cirrhosis, IBD), parasitic infections (Clonorchis sinensis, Opisthorchis viverrini), and chemical exposures such as nitrosamines, vinyl chloride, and dioxins (Benavides et al., 2015). BTCs are aggressive malignancies often presenting with late or nonspecific symptoms, complicating early detection. Jaundice due to biliary obstruction is common in CCA, while GBC may remain asymptomatic until advanced stages (Vogel et al., 2023). Diagnosing BTCs is challenging as imaging methods do not reliably differentiate between benign and malignant lesions, and accurate histological classification necessitates a preoperative biopsy, which is often impractical due to the biliary tract’s complex anatomical location (Macarulla et al., 2025). Late diagnosis limits surgical options and even after surgery, recurrence rates are high with systemic therapy showing limited effectiveness, leading to a poor prognosis (Benson et al., 2023; Primrose et al., 2019). Overall, rising global mortality highlights a substantial unmet need for early detection, preventive strategies, and improved treatment options.

1.2 The clinical challenge of drug resistance

The only effective treatment of BTCs is radical surgical resection, but only 35% of patients have resectable disease at diagnosis, and postoperative recurrence occurs in 50%–70% of cases (Bridgewater et al., 2014; Komaya et al., 2017; Radtke and Königsrainer, 2016). The lack of sensitive and specific early detection biomarkers frequently results in late-stage diagnosis, while the anatomical inaccessibility of the biliary tree complicates preoperative biopsy acquisition, accurate histological classification, and surgical planning. Therefore, current standard-of-care for advanced BTCs relies adjuvant therapies. Gemcitabine–cisplatin plus S-1 (GCS)–has become the preferred chemotherapeutic regimen (Ioka et al., 2023), which has remained the frontline regimen for over a decade. First-line palliative treatment using gemcitabine, cisplatin, and nab-paclitaxel (GCNP) in patients with advanced BTCs has shown a higher response rate and longer survival (Gedela et al., 2024). Now gemcitabine-cisplatin frequently combined with immunotherapy following the positive results of TOPAZ-1 and KEYNOTE-966, where adding durvalumab or pembrolizumab produced modest but significant survival benefits (Kelley et al., 2023; Oh et al., 2022). However, the inherently immunosuppressive tumor microenvironment (TME) in BTCs limits the efficacy of immune checkpoint inhibitors even in tumors expressing PD-L1. Second-line therapy for BTCs is increasingly guided by targetable mutations, with FGFR2, IDH1, and HER2 inhibitors providing meaningful survival and response benefits. For patients without actionable mutations, chemotherapy remains the standard but offers only modest benefit, with low response rates and persistent primary resistance (Toledo et al., 2024). Despite these developments, drug resistance remains a major challenge. Drug resistance in BTCs is driven by an interplay of genetic, epigenetic, and microenvironmental factors. Key genetic alterations involve TP53 and KRAS mutations (Peng et al., 2023; Sanchon-Sanchez et al., 2023; Tannapfel et al., 2003), SHP2–YAP signaling (Buckarma et al., 2020), and modifications in DNA repair genes like BRCA1/2 (Kim et al., 2023), affecting apoptosis and drug sensitivity. Epigenetic factors include P16INK4a hypermethylation (Yu et al., 2018), histone changes, and dysregulated miRNAs (e.g., miR-21, miR-1249, miR-664b-3p, miR-3651, and miR-6087) (Carotenuto et al., 2020; Meng et al., 2006; Toyota et al., 2015), which support tumor survival. The tumor microenvironment further contributes through cytokine-mediated activation of PI3K/AKT and NF-κB pathways, as well as hypoxia-induced factors driving epithelial-mesenchymal transition (EMT) (Fabris et al., 2021; Okumura et al., 2018). These mechanisms underscore the complexity of resistance and the necessity for targeted combination therapies. Resistance also affects newer therapeutic modalities—targeted therapies (e.g., FGFR2 inhibitors, IDH1 inhibitors) commonly encounter secondary mutations, bypass signaling, or polyclonal resistance (Cleary et al., 2022; Goyal et al., 2017; Lau et al., 2019), while immunotherapies benefit only a small subset of patients due to low MSI-H prevalence and variable PD-L1 expression (Porro et al., 2025; Yang et al., 2024). Consequently, though the treatment landscape is expanding, BTCs remain highly refractory cancers, with resistance representing one of the central barriers to durable disease control and improved long-term outcomes.

Critically, the poor therapeutic outcomes in BTCs cannot be attributed solely to the limitations of 2D culture systems. Before a candidate drug reaches clinical trial, it typically passes through multiple preclinical filters, including patient-derived xenograft (PDX) models. PDX systems have been extensively used in BTCs research and preserve tumor architecture and stroma better than 2D cultures; however, they are costly, low-throughput, lack a human immune system, and require long establishment times (3–6 months). Beyond model systems, several independent translational barriers exist: (i) the absence of sensitive and specific early detection biomarkers, leading to late-stage diagnosis; (ii) the anatomical inaccessibility of the biliary tree, which complicates preoperative biopsy acquisition and surgical planning; (iii) an inherently immunosuppressive tumor microenvironment that limits immune checkpoint inhibitor efficacy even in PD-L1-expressing tumors; and (iv) the lack of approved second-line regimens for the majority of patients without actionable mutations (e.g., FGFR2, IDH1, HER2). These barriers operate in parallel with, not downstream of, in vitro model limitations. Consequently, while 3D tumor models address a critical gap in the preclinical screening pipeline, they do not represent a panacea for BTCs drug development failure.

These broader translational barriers highlight the need for improved preclinical systems, particularly 3D tumor models, to better recapitulate tumor heterogeneity, TME interactions, and therapy response in BTCs–while acknowledging that such models address only the in vitro stage of a multifaceted translational ecosystem.

1.3 Apoptosis evasion as a hallmark of BTCs

Apoptosis, programmed cell death executed by two major pathways, i.e., the extrinsic and intrinsic pathways, both of which have cell death as their ultimate goal. Among them, the extrinsic pathway, is activated by pro-apoptotic stimuli outside the cell, whereas the intrinsic pathway, is activated by the intrinsic mechanisms of the cell itself (Campbell and Tait, 2018). Evasion of apoptosis is fundamental to cancer pathogenesis (Hanahan and Weinberg, 2011). The Bcl-2 protein family is crucial in regulating apoptosis, consisting of pro-survival proteins (e.g., Bcl-2, Bcl-xL, Mcl-1), pro-apoptotic BH3-only proteins (e.g., BIM, BID, Puma), and effector proteins (BAX, BAK) that facilitate cell death mechanisms (Kaloni et al., 2023; Warren et al., 2019). Apoptosis evasion is a defining characteristic of CCA and results from a complex dysregulation of extrinsic, intrinsic, and immune-mediated apoptosis pathways. In the extrinsic pathway, CCA cells inhibit Fas/FasL and TNF-α/TNFR1 signaling, which are critical for activating caspase-8 and downstream caspase-3 to initiate apoptosis (Koessinger et al., 2022; Poulaki et al., 2001; Webster and Vucic, 2020). This is achieved through the aberrant upregulation of FasL and downregulation of Fas, mediated by NF-κB activation, allowing tumor cells to eliminate T and NK cells while avoiding self-apoptosis (Carnevale et al., 2017; Pan et al., 2007). Additionally, CCA cells suppress TNF-α–mediated apoptosis via exosome-mediated downregulation of TNF-α secretion (huang et al., 2016; Utaisincharoen et al., 2000). Impairment of intrinsic apoptosis also occurs; although mitochondrial damage usually activates BAX/BAK, leading to cytochrome c release and caspase-9/3 activation, CCA cells often overexpress anti-apoptotic Bcl-2 family members, such as Bcl-xL and Mcl-1, which prevent mitochondrial outer membrane permeabilization, contributing to chemoresistance (Hoffmeister-Wittmann et al., 2022; Kale et al., 2018; Okaro et al., 2001). Together, these alterations (detailed in Section 3.2) establish apoptosis evasion as a central hallmark of BTCs, driving tumor survival, drug resistance, and disease progression (Figure 1).

1.4 Limitations of conventional 2D models

Two-dimensional (2D) culture models inadequately represent the complex architecture, cellular heterogeneity, and dynamic physiology of in vivo tumors. These models lack essential cell–cell and cell–extracellular matrix (ECM) interactions that are crucial for cellular differentiation, proliferation, signaling, and responses to stimuli (Pampaloni et al., 2007). Consequently, when cells isolated from tissue are switched to 2D conditions, they often experience morphological changes, loss of phenotypic diversity, and disruption of intracellular organization and signaling pathways, diminishing their functional relevance (von der Mark et al., 1977; Petersen et al., 1992). The enforced apicobasal polarity in adherent monolayers, along with constant plastic contact and uniform nutrient exposure, creates unnatural physiological conditions that can lead to atypical behaviours, particularly in mesenchymal cells, which negatively impacts their spreading, migration, and responsiveness to soluble cues (Baker and Chen, 2012; Ihalainen et al., 2015). Cells that grow adherently lose polarity due to disturbances in their interactions with the external environment, affecting their responses to phenomena like apoptosis (Mseka et al., 2007; Weaver et al., 2002; Meyers et al., 2006). 2D cultures allow cells unrestricted access to medium components like oxygen and nutrients, unlike in vivo cancer cells, which experience variable availability of these resources due to the complex architecture of tumor masses (Pampaloni et al., 2007). Additionally, 2D cultures can alter gene expression profiles and do not replicate the heterogeneous tumor microenvironment necessary for the maintenance of cancer-initiating cells or for accurately modeling therapeutic resistance. The absence of ECM’s biochemical composition and mechanical properties, which are vital for regulating homeostasis and influencing drug responses—including sensitivity, mode of action, and resistance—further challenges the predictive validity of preclinical drug screening (Nicolas et al., 2020). Therefore, the inherent inability of 2D cultures to recreate the structural, biochemical, and mechanical complexity of the tumor microenvironment fundamentally limits their reliability for studying apoptosis and drug resistance in cancer.

1.5 The rise of 3D tumor models

The success rate in antineoplastic drug discovery remains critically low, primarily because traditional screening methods fail to capture the complexity of tumors, which rely on the interplay between tumor cells and their surrounding microenvironment for growth and metastasis (Swann and Smyth, 2007). This limitation led to the development of 3D tumor models, a technology poised to dominate future cell culture methodologies despite existing technical challenges (Neal et al., 2018). Facilitated by advances in culture media, scaffolds, and gels, these models now effectively simulate the tumor microenvironment (Guan and Huang, 2022). Contemporary applications encompass a spectrum of platforms, ranging from scaffold-free systems like spheroids, organoids, and assembloids to scaffold-based hydrogels and complex microfluidic “tumor-on-a-chip” devices.

Importantly, the advocacy for 3D models presented in this review is not intended to imply that 2D culture limitations are the sole or primary cause of BTCs clinical trial failures. Rather, 3D systems address a specific, well-defined gap in the preclinical pipeline: the poor predictive validity of in vitro drug sensitivity assays. They do not replace rigorous in vivo pharmacokinetic, pharmacodynamic, and toxicity studies, nor do they address the independent clinical challenges of late diagnosis, anatomical inaccessibility, or systemic immunosuppression. The scope of this review is therefore limited to evaluating 3D models as superior in vitro platforms for mechanistic investigation of resistance and apoptosis, and as potential tools for improving the efficiency of drug candidate selection prior to animal testing.

In the context of apoptosis and drug resistance, 3D models offer a distinct advantage over monolayer cultures by recapitulating in vivo physiological conditions. These models establish nutrient and oxygen gradients that induce hypoxia, a critical driver of chemoresistance and altered apoptotic thresholds (Jensen and Teng, 2020). Furthermore, the enhanced cell-cell interactions within 3D structures influence signaling pathways, often resulting in the upregulation of anti-apoptotic proteins and the downregulation of pro-apoptotic markers (Imamura et al., 2015; Riedl et al., 2017). Unlike 2D cultures, where apoptosis is rare and drug sensitivity is artificially high, 3D models develop multilayered architectures with apoptotic cores and differential ATP production. Consequently, they demonstrate clinically relevant resistance profiles to agents such as carboplatin, paclitaxel, niraparib, dacarbazine, and cisplatin (El Mokbel et al., 2024; Fontoura et al., 2020). This enhanced predictive validity suggests that 3D data could significantly lower the failure rate of anticancer drugs in clinical trials (Nunes et al., 2019). Thus, the adoption of 3D culture systems is increasingly recognized as essential for improving the predictive accuracy of preclinical therapeutic screening. However, these models address only the in vitro stage of the translational pipeline and cannot replace rigorous in vivo pharmacokinetic, pharmacodynamic, and toxicity studies.

Specifically regarding BTCs, 3D models outperform 2D monolayers by preserving histological architecture, genetic heterogeneity, and physiological gradients (Saito et al., 2019). While 2D culture often exhibit hypersensitivity, 3D BTCs models accurately replicate the desmoplastic stroma and maintain resistance to standard chemotherapeutics like gemcitabine and cisplatin. Additionally, these models provide superior insights into invasion mechanisms by revealing the distinct localization of metastasis markers, such as galectin-3 (Sukphokkit et al., 2022). Consequently, the adoption of 3D culture systems represents an emerging paradigm shift to accurately predict therapeutic efficacy and overcome the historically high attrition rates associated with BTCs drug discovery. Nevertheless, 3D models do not overcome broader clinical challenges such as delayed diagnosis, limited surgical accessibility, or the immunosuppressive tumor microenvironment that independently contribute to poor BTCs outcomes. Rather than serving as a panacea, these systems should be viewed as tools that can improve translational efficiency by reducing the attrition of compounds progressing from in vitro to in vivo testing within the broader and multifaceted translational ecosystem (Figure 2).

2 3D tumor models as superior tools for BTCs research

2.1 Spheroids and aggregates

Spheroids and multicellular aggregates act as the definitive scaffold-free 3D architectures in BTCs research, serving as a physiological bridge between reductionist monolayer cultures and complex in vivo tumors. Unlike 2D monolayers, where cells are uniformly exposed to culture conditions, BTCs spheroids self-assemble into multilayered structures governed by discrete nutrient and oxygen gradients. This structural complexity facilitates the formation of necrotic cores and luminal architectures that mirror in vivo tumor heterogeneity (Sukphokkit et al., 2022). These gradients create hypoxic and metabolic niches (see Section 4.2) that promote aggressive phenotypic shifts, including EMT and altered Galectin-3 localization (Sukphokkit et al., 2022; Ciufolini et al., 2024; Phukhum et al., 2023). Furthermore, spheroid cultures functionally enrich for cholangiocarcinoma stem-like cells (CSCs). These models have revealed that CSCs actively remodel the immune microenvironment through a distinct secretome (IL13, OA, IL34), priming tumor-associated macrophages toward a pro-invasive phenotype (Raggi et al., 2017). This highlights the capacity of spheroids to elucidate stromal-mediated resistance mechanisms that remain obscured in homogeneous monolayer systems.

2.2 Patient-derived organoids (PDOs): a clinically relevant 3D model

Before discussing PDOs, it is important to note that patient-derived xenograft (PDX) models, generated by implanting patient tumor tissue into immunocompromised mice, have been extensively used in BTCs research. PDXs better preserve tumor architecture and stroma than 2D cultures and are valuable for therapeutic validation. However, they are costly, low-throughput, lack a human immune system, and require long establishment times (3–6 months). In contrast, PDOs offer faster establishment (2–4 weeks), higher throughput, lower cost, and compatibility with human immune co-cultures, making them strategically complementary rather than replacements for PDX models. PDOs are widely regarded as a leading preclinical model, though standardization challenges remain. By faithfully recapitulating the histological characteristics, genomic heterogeneity, and transcriptomic profiles of parental tumors, BTCs-PDOs offer a superior platform for decoding the mechanisms of chemoresistance and apoptosis. The generation of these high-fidelity models utilizes distinct methodologies depending on the source material. Standard protocols for tissue-derived organoids involve the enzymatic digestion of surgically resected specimens—typically iCCA or GBC—using a cocktail of collagenase (type I or IV) and dispase II to isolate viable epithelial cells. These cells are subsequently encapsulated in a basement membrane matrix, such as Matrigel, which mimics the extracellular matrix (ECM) scaffold, and cultured in a defined medium supplemented with essential niche factors including R-spondin 1, Epidermal Growth Factor (EGF), Hepatocyte Growth Factor (HGF), Noggin, and Wnt agonists to sustain stemness and proliferation (Maier et al., 2021; Qureshi et al., 2024).

A significant advancement in modeling accessibility is the establishment of Bile-Derived Organoids (BDOs). This minimally invasive approach processes bile collected via Endoscopic Retrograde Cholangiopancreatography (ERCP), isolating epithelial cells suspended in the fluid. BDOs demonstrate a superior derivation efficiency (88%) compared to tissue-derived methods (33%), attributed to the high viability of cells in fresh bile and significantly reduced stromal contamination. This method is particularly advantageous for patients with inoperable disease, allowing for longitudinal sampling and real-time monitoring of tumor evolution (Kinoshita et al., 2023). A critical challenge across both derivation methods is the potential overgrowth of non-neoplastic biliary epithelium or fibroblasts. However, this is effectively mitigated by exploiting the finite lifespan of normal cells (the Hayflick limit) or by tailoring media conditions to select for malignant phenotypes. For instance, the withdrawal of EGF or the addition of Nutlin-3a can select for cancer cells harboring specific driver mutations, such as KRAS activation or TP53 inactivation, ensuring the purity of the established cancer organoid lines (Saito et al., 2019).

Once established, BTCs-PDOs exhibit remarkable biological fidelity, serving as “avatars” of the patient’s original tumor. Comprehensive validation through histological staining and immunohistochemistry confirms the retention of key biliary markers, including Cytokeratin 7 (CK7), MUC1, and proliferation markers like Ki-67, preserving the original morphological architecture. Furthermore, deep molecular profiling via Whole Exome Sequencing (WES) and single-cell RNA sequencing (scRNA-seq) reveals that PDOs maintain the patient-specific genetic landscape, retaining essential oncogenic drivers (e.g., TP53, MUC4, KRAS) and recapitulating the intratumoral heterogeneity and signaling pathway activation (e.g., PI3K-AKT, MAPK) observed in the primary tissue (Yuan et al., 2022). Recent large-scale efforts have validated the robustness of this model; for instance, the establishment of a biobank of 61 BTCs-PDOs demonstrated that culture success positively correlates with intrinsic tumor stemness and proliferation indices (Ren et al., 2023). While some transcriptomic divergence may occur due to culture conditions, the functional pathways governing drug response and apoptosis remain intact, rendering PDOs an invaluable tool for precision oncology.

The translational utility of BTCs-PDOs is most profound in the context of personalized drug testing and the elucidation of chemoresistance mechanisms. In vitro drug sensitivity assays performed on these organoids have revealed interpatient heterogeneity that in retrospective analyses closely parallels clinical outcomes. The largest such retrospective study, comprising 61 Asian BTCs patients treated with gemcitabine-cisplatin, reported a 92.3% concordance between PDO chemosensitivity and patient response, with ROC area under the curve (AUC) values of 0.92–0.99 (Ren et al., 2023). Beyond standard of care, PDOs facilitate drug repositioning and the identification of novel vulnerabilities in otherwise resistant phenotypes. For example, viability assays have identified specific sensitivities to antifungal drugs, MDM2 inhibitors, and Histone Deacetylase (HDAC) inhibitors (e.g., CUDC-907), offering new therapeutic avenues for refractory cases. Mechanistically, molecular profiling of resistant organoids has highlighted the upregulation of specific markers such as IGFBP1 and DNAJC12, contrasted with the activation of Fanconi anemia and TP53 pathways in sensitive phenotypes (Yuan et al., 2022; Ren et al., 2023). These promising results suggest that PDO may bridge bench-side molecular profiling with bedside clinical decision-making, ultimately refining therapeutic strategies to overcome drug resistance in biliary tract cancers, but several important caveats must be noted. First, the concordance data are limited to standard chemotherapy regimens (gemcitabine-cisplatin and 5-fluorouracil-cisplatin); no comparable evidence exists for targeted agents, ADCs, or immunotherapies. Second, organoid establishment success rates are variable (approximately 74% for tissue-derived samples, lower for small biopsies), and batch-to-batch variability in Matrigel and other matrix components may influence drug sensitivity readouts. Third, and most critically, prospective validation of PDO-guided treatment selection has not yet been performed in BTCs; all reported concordance is retrospective. With these qualifications, PDOs remain a powerful but still-developing preclinical tool, and their use for direct clinical decision-making should be considered investigational.

2.3 Advanced 3D systems (bioprinting and microfluidics)

The convergence of advanced 3D bioprinting and microfluidic technologies represents the current frontier in preclinical BTCs research. These approaches fundamentally revolutionize the ability to fabricate ‘Cholangiocarcinoma-on-a-Chip’ systems and multicellular constructs that faithfully recapitulate the dynamic and heterogeneous tumor microenvironment (TME) (de Barros et al., 2025). Unlike static scaffold systems, these advanced methodologies offer unprecedented methods for generating 3D models through techniques such as coaxial extrusion printing and microfluidic channel engineering, allowing for precise structural control over the spatial distribution of cellular and extracellular components. By evolving from simple synthetic polymer scaffolds to sophisticated assemblies incorporating liver-specific decellularized extracellular matrix (dECM) and vascularized channels, researchers can now simulate physiological fluid dynamics, bile acid gradients, and the mechanical stimuli of the biliary niche (Ashammakhi et al., 2019; Du et al., 2023; Justin et al., 2018). This structural fidelity is critical for investigating cell-cell interaction studies, as demonstrated by the development of PDMS-based microfluidic devices that reconstruct the desmoplastic stiffness of the cholangiocarcinoma (CCA) niche via tunable fibrin-collagen hydrogels (Polidoro et al., 2024). Such platforms facilitate the precise integration of controlled TME elements, including cancer-associated fibroblasts (CAFs), endothelial cells, and immune components, enabling analysis of bidirectional crosstalk that drives malignancy. Recent studies show that intricate signaling among CCA cells, CAFs, and remodeled ECM directly correlates with enhanced resistance to gemcitabine and cisplatin, surpassing the resistance profiles seen in 2D monolayers or 3D monocultures.

Furthermore, these technologies are instrumental in dissecting the mechanisms of apoptosis and drug resistance through the creation of personalized, patient-derived models. Advancements in bioprinting have enabled the fabrication of intrahepatic cholangiocarcinoma (ICC) models using composite hydrogels of gelatin, alginate, and Matrigel™ arranged in pre-designed grid architectures. These constructs not only maintain high cell survival and colony formation but also promote the expression of critical biomarkers associated with invasiveness, metastasis, and epithelial-mesenchymal transition (EMT), effectively mimicking the stem-like properties and drug resistance patterns found in vivo (Mao et al., 2020). The capacity for enhanced structural control is further exemplified by concentric cylindrical tetra-culture models of gallbladder carcinoma (GBC), where extrusion-based bioprinting is used to spatially segregate tumor cores from encapsulating stromal rings containing fibroblasts, macrophages, and endothelial cells. This specific architectural arrangement successfully induces intratumoral heterogeneity—revealing distinct glandular and squamous subtypes via single-cell RNA sequencing—and generates transcriptomic profiles enriched in pathways governing cytokine signaling and extracellular organization (Jin et al., 2024). Ultimately, by replicating patient-specific immune migration patterns, such as “hot” versus “cold” tumor phenotypes, and validating gene signatures associated with poor prognosis, these advanced 3D systems provide a robust, biologically relevant framework for decoding the complex interplay between spatial organization, TME elements, and therapeutic response in biliary tract cancers.

Table 1.

Scope and limitations of 3D tumor models in BTCs research.

2.4 Key TME components recapitulated in 3D BTCs models

Advanced 3D BTCs models have evolved beyond inert cell aggregates, utilizing biomimetic scaffolds to actively reconstruct reciprocal signaling between cancer cells and the tumor microenvironment (TME). These scaffolds are engineered to recapitulate two distinct categories of cues that drive invasion, epithelial-mesenchymal transition (EMT), and drug resistance:

2.4.1 Biochemical composition

Analysis of patient-derived iCCA scaffolds reveals a critical TME remodeling process: disassembly of the basement membrane (downregulation of laminins and type IV collagen) and accumulation of a desmoplastic matrix rich in periostin and fibrillar collagens (types I, III, and XII) (Carpino et al., 2019). To mimic this, modern models utilize decellularized ECM (dECM) and functionalized hydrogels. Specific ligand composition—such as laminin-derived peptides (IKVAV) versus fibronectin motifs (RGD)—directs proper morphogenesis and hemidesmosome formation (Baude et al., 2025).

2.4.2 Structural and mechanical anisotropy

Physical matrix properties are potent drivers of BTCs phenotype. In native tumors, type III collagen transitions from a reticular network to an aligned architecture, increasing stiffness and providing contact guidance for migration (Carpino et al., 2019). Advanced synthetic platforms (e.g., PEG, PIC, alginate) allow precise tuning of stiffness and viscoelasticity independent of biochemistry. The mechanobiological consequences of these matrix properties—including integrin signaling, YAP/TAZ activation, and drug resistance—are detailed in Section 4.3.

2.5 Comparative advantages of 3D over 2D models in BTCs

As summarized in Table 1, 3D models (spheroids and patient-derived organoids) offer superior translational fidelity over conventional 2D monolayers by recapitulating physiological oxygen/nutrient gradients, physical barriers from desmoplastic stroma, and preserved genetic heterogeneity, thereby demonstrating more realistic chemoresistance profiles and minimizing false-positive apoptosis signals typical of 2D-based drug discovery (Phukhum et al., 2023; Grigoreva et al., 2025; Heinzelmann et al., 2024; Jovanović et al., 2025; Kasuga et al., 2021; Montagner et al., 2025; Zhao et al., 2024).

TABLE 1. Comparative summary of 2D vs 3D models in key areas.

3 Unravelling apoptotic pathways and dysregulation in 3D

3.1 Overview of apoptosis

Apoptosis constitutes a fundamental, evolutionarily conserved mechanism of programmed cell death essential for maintaining tissue homeostasis and eliminating damaged or neoplastic cells (Mustafa et al., 2024). In the context of cancer, the evasion of apoptosis is a hallmark of tumorigenesis and a primary driver of chemoresistance (Hou et al., 2011). This process involves two distinct yet cross-communicating signaling cascades: the intrinsic (mitochondrial) pathway and the extrinsic (death receptor) pathway.

The intrinsic pathway initiates apoptosis through intracellular stressors like DNA damage and oxidative stress, regulated by Bcl-2 family proteins. Key proteins BAX and BAK facilitate mitochondrial outer membrane permeabilization (MOMP) and release cytochrome c, which forms the apoptosome with Apaf-1 and activates Caspase-9. The extrinsic pathway, on the other hand, is activated by extracellular signals binding to death receptors, leading to the formation of the Death-Inducing Signaling Complex (DISC) and the activation of Caspase-8. Both pathways converge at the execution phase, activating effector caspases (Caspase 3, 6, and 7) that cause cellular disassembly and characteristics of apoptosis like DNA fragmentation and membrane blebbing (Lossi, 2022; Kashyap et al., 2021). Because the intrinsic pathway integrates diverse stress signals and is tightly controlled by the Bcl-2 family, it is of particular relevance to chemoresistance in biliary tract cancers (BTCs) – the focus of the following sections.

3.2 Genetic and microenvironmental rewiring of the Bcl-2 family rheostat

3.2.1 The Bcl-2 family rheostat and mitochondrial priming

The transition from 2D monolayers to 3D architectures reveals a significant elevation in the apoptotic threshold, necessitating a more potent disruption of the Bcl-2 family rheostat to induce cell death. Comparative high-throughput screenings demonstrate that 3D spheroids exhibit drastically reduced cytotoxicity rates and significantly higher IC50 values—up to a 6-fold increase observed in lung cancer models—compared to their 2D counterparts. This heightened resistance is attributed to the protective spatial organization of the 3D niche, which buffers against stress signals. However, agents that successfully overcome this barrier do so by shifting the equilibrium between anti-apoptotic and pro-apoptotic proteins. Mechanistically, effective compounds induce a downregulation of Bcl-2 mRNA while concurrently upregulating BAX and p53. This shifts the ratio in favor of BAX translocation to the mitochondrial membrane, triggering the loss of mitochondrial membrane potential (ΔΨm) and subsequent Mitochondrial Outer Membrane Permeabilization (MOMP). This event facilitates the release of cytochrome c and the activation of the intrinsic apoptotic cascade, predominantly mediated by Caspase-9 rather than Caspase-8 (Agena et al., 2023; Mahmoud et al., 2025; Massaro et al., 2017).

A key functional measure of apoptotic readiness is “mitochondrial priming” – the proximity of a cell to the MOMP threshold. Priming is determined by the balance between pro-apoptotic BH3-only proteins (activators like BIM, BID, PUMA; sensitizers like BAD, NOXA, HRK) and anti-apoptotic proteins (Bcl-2, Bcl-xL, Mcl-1, BFL-1, Bcl-W) (López et al., 2025). Quantitative BH3 profiling, which measures cytochrome c release from isolated mitochondria exposed to synthetic BH3 peptides, has been applied to organoids (Fraser et al., 2019). These studies reveal that drug resistance exhibit “low priming” – they require higher concentrations of activator BH3 peptides (BIM, PUMA) to trigger MOMP, correlating with high Mcl-1 and Bcl-xL expression but not with Bcl-2 (Bhatt et al., 2020; Colak et al., 2014). This low-priming state may be a functional consequence of the 3D architecture of tumor.

3.2.2 Genetic determinants of the apoptotic threshold

Specific recurrent genetic alterations directly rewire the Bcl-2 family rheostat.

• TP53 loss-of-function mutations (30%–50% of cholangiocarcinomas, up to 80% of gallbladder cancers) disable transcription of pro-apoptotic BH3-only proteins PUMA, NOXA, and BIM (Diepstraten et al., 2024; Wang Z. et al., 2022).
• KRAS mutations (22%–53% of perihilar CCA, 9%–17% of intrahepatic CCA) inhibit apoptosis primarily through activation of the PI3K/AKT and MAPK signaling pathways, which suppress pro-apoptotic proteins such as Bad, Bim, Par-4, and caspase-9 while enhancing cell survival signaling. Additionally, KRAS-driven NF-κB activation and upregulation of anti-apoptotic proteins including Bcl-2, Bcl-xL, and Mcl-1 further promote tumor cell survival and resistance to apoptosis (Ferreira et al., 2022; Werner et al., 2015; Zaanan et al., 2015).

3D tumor models are essential for studying apoptosis resistance in biliary tract malignancies, as the 3D microenvironment fundamentally reshapes the Bcl-2 apoptotic rheostat in a genotype-dependent manner. These models may reveal that simultaneous targeting of anti-apoptotic proteins Bcl-xL and Mcl-1 required to induce MOMP, cytochrome c release, and effective therapeutic apoptosis, particularly in KRAS-mutant.

3.2.3 BH3-only network dysregulation

Beyond anti-apoptotic proteins, organoids exhibit profound rewiring of the BH3-only network, where BIM is sequestered by Bxl-xL and PUMA neutralised by Mcl-1, creating a “double-safe” anti-apoptotic state. BH3 profiling further reveals distinct Mcl-1 and Bcl-xL dependencies based on differential peptide sensitivities, highlighting the ability of 3D models to preserve native MOMP thresholds. In the specific context of biliary tract cancers (BTCs), very few studies investigated apoptosis in 3D model. patient-derived organoids (PDOs) of cholangiocarcinoma exhibit a distinct expression profile characterized by high levels of anti-apoptotic Bcl-xL and Mcl-1, coupled with relatively low Bcl-2 expression. Research indicates that structural complexity correlates with resistance; mature, branching organoids display significantly broader chemoresistance to standard taxanes and gemcitabine compared to simple cystic forms. Crucially, Bcl-xL functions as the primary checkpoint preventing MOMP in these chemoresistant models. Selective inhibition of Bcl-xL (e.g., via A-1155463 or DT2216) synergizes with gemcitabine to dismantle this blockade, resulting in widespread organoid fragmentation and cleaved caspase-3 activation (Mi et al., 2025).

3.3 Caspase activation and execution phase

The transition from 2D monolayers to 3D tumor models reveals a unique “multicellular resistance” (MCR) phenotype, primarily defined by the dysregulation of the apoptotic execution phase (Plati et al., 2008). In 2D cultures, chemotherapeutics like cisplatin and gemcitabine trigger rapid cleavage of pro-caspase-3 and -7. In contrast, 3D spheroids and organoids often show impaired or delayed activation of these caspases. This diminished response results from intrinsic changes in the 3D tumor microenvironment (TME), not just drug penetration limitations. Studies suggest that 3D architecture enhances cell-matrix interactions and increases anti-apoptotic Bcl-2 family members, such as Bcl-xL, which inhibit pro-apoptotic factors and prevent mitochondrial outer membrane permeabilization (MOMP), essential for caspase activation (Wang et al., 2016). Thus, even if initiator caspases (e.g., Caspase-8 or -9) are activated, a higher threshold for activating executioner Caspases-3 and -7 causes “apoptotic abortion” or sublethal signaling. This indicates that the inadequacy of caspase-dependent cell death observed in 3D models more accurately reflects the resistant nature of clinical BTCs, in contrast to 2D systems, where lack of stromal and spatial cues can lead to an overestimation of cytotoxic effects (Sukphokkit et al., 2022; Montagner et al., 2025) (Figure 3) .

3.4 Investigating ferroptosis and necroptosis in 3D

The main obstacle to treating BTCs is resistance to apoptosis, therefore research has shifted to focussing on non-apoptotic controlled cell death (RCD) pathways, particularly ferroptosis and necroptosis. Direct BTCs-specific evidence for these pathways in 3D models remains limited; the following discussion integrates findings from other gastrointestinal malignancies where the pathophysiology (desmoplasia, KRAS mutations, hypoxia) overlaps substantially with BTCs, with extrapolations clearly indicated.

Ferroptosis, driven by iron-dependent lipid peroxidation and regulated by the GPX4/SLC7A11 axis (Wang WJ. et al., 2021), and necroptosis, mediated by the RIPK1/RIPK3/MLKL cascade (Niu et al., 2022), offer critical vulnerabilities in drug-resistant tumors. However, traditional 2D monolayers frequently produce discordant drug sensitivity profiles because they fail to replicate the metabolic gradients, oxidative stress, and iron metabolism inherent to the native tumor microenvironment (TME). Advanced 3D systems, including spheroids and organoids, better recapitulate these features and therefore provide more physiologically relevant platforms for studying non-apoptotic cell death in BTCs.

Emerging BTCs-specific studies have already demonstrated the utility of these models. Functional genomic analyses revealed differential dependency on the ferroptosis regulator GPX4 between intrahepatic and extrahepatic cholangiocarcinoma subtypes, highlighting subtype-specific ferroptosis vulnerabilities that are more faithfully preserved in physiologically relevant 3D architectures (Lee et al., 2025). Similarly, patient-derived cholangiocarcinoma organoids demonstrated that combining surufatinib (SUR) with photodynamic therapy (PDT) effectively induced ferroptosis and suppressed tumor proliferation (Huang et al., 2024).

Recent methodological innovations have further refined the study of these pathways in 3D tumor models. The ‘3DELTA’ assay, a high-throughput, non-destructive methodology, utilizes Sytox DNA-intercalating dyes to distinguish ferroptosis from necroptosis in intact spheroids. Validated with the GPX4 inhibitor ML-162 and pathway-specific inhibitors like Fer-1 and Nec-1s, this system has shown that 3D spheroids exhibit significantly higher resistance to ferroptosis than monolayers, a phenotype correlated with increased matrix accumulation and spheroid maturity (Demuynck et al., 2020). Mechanistically, 3D models are superior for filtering out artifactual survival mechanisms seen in 2D cultures; for example, while 2D cultures often mount a protective autophagic response (upregulation of ATG16L2, ATG9A, SQSTM/p62) to shield against cell death, 3D architectures suppress this compensatory autophagy, thereby sensitizing tumors to Heme Oxygenase-1 (HO-1) mediated ferroptosis (Chipurupalli et al., 2023). Although these studies were performed primarily in non-BTCs models, they provide important methodological and mechanistic insights potentially relevant to BTCs biology.

Several mechanistic observations from other gastrointestinal malignancies may also inform future BTCs investigations because these tumors share common features such as KRAS-driven metabolic rewiring, oxidative stress adaptation, hypoxic niches, and desmoplastic stromal architecture. For example, pancreatic cancer organoids demonstrated that KRAS-mutant cells exploit Ferroptosis Suppressor Protein 1 (FSP1) to evade ferroptotic death (Müller et al., 2023), while hepatocellular carcinoma models identified the prefoldin RPB5 interactor (URI) as a mediator of tyrosine kinase inhibitor resistance (Ding et al., 2023). Similarly, colorectal cancer organoids have highlighted that upregulating RIPK1 can sensitize cells to necroptosis via IAP antagonists like tolinapant, a susceptibility regulated by Fragile X messenger ribonucleoprotein 1 (FMR1) (Crawford et al., 2021; Di Grazia et al., 2021). Collectively, these extrapolated findings–which await direct confirmation in BTCs-specific 3D platforms–underscore that 3D models are indispensable for unravelling how BTCs cells evade non-apoptotic death, providing a predictive tool for screening ferroptosis-inducing agents and targeting drug-resistant subclones (Zou et al., 2026).

4 Mechanisms of drug resistance decoded by 3D models

4.1 Drug distribution gradients and penetration

Unlike monolayer cultures that offer unrestricted drug access, 3D tumor models recapitulate the severe mass transport barriers and desmoplasia characteristic of Biliary Tract Cancers (BTCs), which are primary determinants of therapeutic failure (Minchinton and Tannock, 2006). The dense architecture and Extracellular Matrix (ECM) of spheroids prevent convective flow, forcing therapeutics to rely on diffusion; this process is significantly impeded by steric hindrance, matrix stiffness, and electrostatic interactions, particularly for positively charged agents or nanocarriers larger than 50 nm (Bordeleau et al., 2017; Holle et al., 2016; Jain and Stylianopoulos, 2010; Laurent and Ogston, 1963; Lieleg et al., 2009; Stylianopoulos et al., 2010; Tang et al., 2013). These physical constraints create concentric physiological gradients of oxygen and nutrients (Nunes et al., 2019; Langhans, 2018), resulting in a “diffusion-consumption mismatch” where peripheral cells metabolize drugs before they reach the core in cytotoxic concentrations (El Harane et al., 2023). This phenomenon leads to metabolic zoning—exemplified by the peripheral restriction of active metabolites like SN-38 despite core penetration of the prodrug irinotecan—effectively creating “sanctuary sites” for tumor survival (LaBonia et al., 2016). Consequently, core cells exposed to sublethal drug concentrations or hypoxia enter a quiescent (G0) state or activate adaptive stress pathways such as the Unfolded Protein Response (UPR) and autophagy, thereby evading apoptosis and priming the tumor for recurrence (Imamura et al., 2015; Chouaib et al., 2017; Ji et al., 2024; Wenzel et al., 2014). Recent advances in imaging, such as Light Sheet Fluorescence Microscopy (LSFM), have further validated these models by revealing the precise spatial heterogeneity of drug distribution that traditional microscopy often obscures (Yadav et al., 2025). Thus, by capturing the spatial heterogeneity and mass transport limitations absent in monolayer cultures, 3D models provide an indispensable framework for decoding how limited penetration creates sanctuary sites that drive residual disease, clinical recurrence and drug resistance.

4.2 Hypoxia-driven resistance

A definitive advantage of 3D tumor models over monolayer cultures is their ability to recapitulate physiological oxygen gradients, a critical feature for understanding drug resistance in biliary tract cancers. As BTCs spheroids or organoids expand beyond 200–500 µm, limited diffusion establishes a distinct hypoxic core that mimics the avascular regions of solid tumors, stabilizing Hypoxia-Inducible Factors (HIFs) such as HIF-1 alpha and HIF-2 alpha which are otherwise degraded in normoxic conditions (Imamura et al., 2015; Abbasi-Malati et al., 2024; Kirsh et al., 2023; Sutherland, 1988; Zhu et al., 2025). This hypoxic microenvironment acts as a master regulator of therapeutic failure by orchestrating a multidimensional resistance phenotype. Primarily, stabilized HIF-1 alpha upregulates ATP-binding cassette transporters, including P-glycoprotein (P-gp/ABCB1) and MRP1 (ABCC1), which actively efflux chemotherapeutics like gemcitabine, cisplatin, and doxorubicin, thereby reducing intracellular drug accumulation (Montagner et al., 2025; Doublier et al., 2012; Wu et al., 2021; XIE et al., 2013). Beyond efflux, hypoxia drives profound metabolic reprogramming; proteo-metabolomic profiling in cholangiocarcinoma spheroids reveals a “Warburg-like” shift toward anaerobic glycolysis via the upregulation of GLUT-1, GAPDH, and LDHA, alongside SNAT2-mediated amino acid adaptation and autophagy, enabling survival under nutrient deprivation (Phukhum et al., 2023; Morotti et al., 2019; Semenza, 2012). These 3D models further elucidate that hypoxia bypasses standard apoptotic pathways (e.g., Bcl-2/survivin) by activating alternative survival signals, such as the VEGF/bFGF-mediated RAF/MEK/ERK and AKT axes, while simultaneously impairing DNA damage repair mechanisms (Begg and Tavassoli, 2020; Bhandari et al., 2019; Karakashev and Reginato, 2015; Kim et al., 2011; Lugano et al., 2019; Sarkar et al., 2020). Moreover, the hypoxic niche enriches quiescent Cancer Stem Cell (CSC) populations—marked by CD24, CD133, and Nanog expression—and facilitates epithelial-mesenchymal transition (EMT) via mechanisms like miR-210 regulation, resulting in a phenotype that is significantly more refractory to treatment than normoxic cells (Semenza, 2012; Ham et al., 2016; Singh et al., 2018; Tang et al., 2018; Zhi et al., 2024). Ultimately, these complex interactions, including resistance to apoptosis-inducing agents, are observable only in 3D architectures, confirming their necessity for decoding hypoxia-dependent resistance vectors (Däster et al., 2017). Thus, the ability to model hypoxia-induced signaling networks affirms the superiority of 3D systems, establishing them as critical tools for unraveling the multi-layered resistance mechanisms that limit the efficacy of current BTCs therapies.

4.3 Extracellular matrix (ECM) and physical shielding

Beyond the compositional ECM features listed in Section 2.4, the physical and mechanical properties of the matrix—particularly stiffness and viscoelasticity—activate discrete mechanotransduction pathways that drive drug resistance. In Biliary Tract Cancers (BTCs) and similar solid tumors, Cancer-Associated Fibroblasts (CAFs) drive the excessive deposition and Lysyl Oxidase (LOX)-mediated cross-linking of type I collagen, creating a stiff, high-density matrix often described as a “natural capsid” or physical shield (Affo et al., 2017; Henke et al., 2020; Minini and Fouassier, 2023; Schütze et al., 2015). This structural rigidity elevates interstitial fluid pressure and forms a stromal barrier that compresses vasculature and physically impedes the diffusion of chemotherapeutic agents such as gemcitabine, cisplatin, and doxorubicin, a phenomenon elucidated through 3D co-cultures and patient-derived scaffolds (El-Derby et al., 2024).

Beyond simple exclusion, 3D studies demonstrate that matrix stiffness (>100 MPa in tumors vs. 0.19 MPa in normal tissue) acts as a distinct survival signal; the binding of tumor cells to this rigid ECM triggers integrin clustering (notably β1) and activates downstream mechanotransduction cascades including FAK, Src, PI3K/AKT, and MAPK (Broutier et al., 2016; Halder et al., 2005; Hodkinson et al., 2006; Xiao and Ge, 2012). This signaling reactivates the YAP/TAZ pathway and upregulates anti-apoptotic proteins like Bcl-2, conferring “anoikis resistance” and supporting metabolic rewiring toward oxidative phosphorylation (Ciufolini et al., 2024). Furthermore, varying matrix densities have been shown to regulate the switch between dormancy and invasion through hypoxia-inducible factors (HIF-1 alpha) and gene expression changes (CAV1, TGFB1), highlighting the dual role of the ECM in Cell Adhesion-Mediated Drug Resistance (CAM-DR) (Dogan et al., 2024; Pezeshki et al., 2025).

However, although detailed characterization has been valuable for decoding mechanisms in other solid tumors, there remains a significant lack of studies specifically investigating ECM-mediated resistance using dedicated 3D tumor models of BTCs. Thus, bridging this translational gap with physiologically relevant 3D models is critical to transforming our understanding of the ECM from a passive scaffold into a distinct, druggable target for overcoming chemoresistance in BTCs.

4.4 Cell state plasticity and heterogeneity

The 3D tumor microenvironment (TME) is increasingly recognized as a critical driver of acquired therapeutic resistance by recapitulating physiological gradients of oxygen, nutrients, pH, and drug diffusion that are absent in conventional 2D cultures (Costa et al., 2016; Phi et al., 2018; Senrung et al., 2023; Valente et al., 2023). Across multiple solid tumor models—including colorectal, pancreatic, gastric, and endometrial cancers—these microenvironmental pressures have been shown to promote intratumoral heterogeneity and dynamic cell state plasticity through the formation of hypoxic and nutrient-deprived niches characterized by HIF-1α, BNIP3, and CA-IX activation. Such conditions induce slower-cycling or quiescent cellular states that intrinsically reduce the efficacy of anti-proliferative therapies. Studies from colorectal and other epithelial cancer organoid systems further suggest that hypoxic 3D environments enrich drug-tolerant persister (DTP) cells and cancer stem-like cells (CSCs), which evade apoptosis through metabolic rewiring, chromatin remodeling, enhanced DNA repair, and adaptive stress-response signaling (Chitcholtan et al., 2013; Chu et al., 2024; De Conti et al., 2021; Kurppa et al., 2020; Mikubo et al., 2021; Plaks et al., 2015; Prager et al., 2019).

Mechanistically, evidence from colorectal and intestinal cancer models indicates that these adaptive states may be regulated through Wnt, Notch, Hedgehog, TGF-β, and YAP-associated signaling pathways, enabling reversible phenotypic transitions into dormant or regenerative cell populations, including LGR5+p27+ and MEX3A+ persister-like states (Álvarez-Varela et al., 2022; Batlle and Clevers, 2017; Coppo et al., 2023; Han et al., 2020; Ohta et al., 2022; Qin et al., 2023; Safa, 2020; Vermeulen et al., 2010; Zhou et al., 2021). Although these mechanisms have not yet been comprehensively validated in BTCs-specific 3D systems, their extrapolation to BTCs is biologically plausible given the shared features of gastrointestinal malignancies, including hypoxia-driven plasticity, stromal desmoplasia, epithelial–mesenchymal transition (EMT), and KRAS/TGF-β-associated signaling rewiring.

Partial BTCs-specific support for these concepts has emerged from recent single-cell RNA sequencing analyses of clinical BTCs biopsies, which identified distinct malignant cell states, including endothelial-like tumor populations and mesenchymal hubs interacting with immunosuppressive monocytes, highlighting lineage plasticity as a potential contributor to therapeutic resistance (Park et al., 2024). Similarly, studies across multiple tumor systems have demonstrated that the 3D architecture promotes stromal co-evolution, including fibroblast activation into cancer-associated fibroblasts (CAFs) and macrophage polarization toward immunosuppressive or pro-angiogenic phenotypes that reinforce CSC-associated programs involving OCT4, SOX2, and CD44v expression (Bonnans et al., 2014; Chen et al., 2020; De Val and Black, 2009; Emami Nejad et al., 2021; Lu et al., 2011; Shibuya, 2011; Ukai et al., 2020). While analogous stromal plasticity is increasingly suspected in BTCs, direct mechanistic validation using BTCs-specific co-culture organoid or immune-competent 3D models remains limited.

By accurately mimicking these complex immune-malignant interactions and resistance pathways, 3D models allow for the identification of vulnerabilities in resistant subpopulations—such as sensitivity to WEE1 or CHK1 inhibitors—providing a robust framework to prevent recurrence (Davies et al., 2023; Huang R. et al., 2015; Melia et al., 2025; Pérez-G et al., 2023; Yi et al., 2021). Future studies integrating BTCs organoids with stromal and immune components, lineage tracing, and single-cell multi-omics will therefore be essential to determine whether these adaptive resistance programs are conserved in biliary tract cancers and therapeutically targetable.

5 Translational impact and precision oncology

5.1 Targeting anti-apoptotic proteins

Three-dimensional tumor models have been valuable for dissecting the elevated apoptotic threshold in biliary tract cancers (BTCs), revealing that standard chemotherapies often require targeted potentiation to overcome resistance. Transcriptomic profiling of patient-derived branching organoids identified Bcl-xL and Mcl-1 as dominant drivers of gemcitabine resistance; targeting Bcl-xL with the degrader DT2216 restored caspase-3 activity and enhanced chemosensitivity in these complex models (Mi et al., 2025). Upstream modulation of these pathways has also shown promise: farnesoid X receptor (FXR) agonists sensitized BTCs to cisplatin by upregulating SHP to inhibit STAT3-mediated Bcl-xL expression (Wang et al., 2016), and miR-451a replacement suppressed stemness in 3D spheroids by downregulating Bcl-2 via the MIF-PI3K/AKT axis (Obata et al., 2023). Finally, 3D models have helped elucidate microenvironmental resistance mechanisms, such as the link between obesity and chemoresistance: adipocyte-derived leptin upregulates Mcl-1 to sustain mitochondrial function, a protective effect reversed by the Mcl-1 inhibitor MIM1 (Wang WJ. et al., 2021). Together, these findings support the use of 3D platforms for designing precision regimens that combine metabolic or upstream signal modulators with BH3-mimetics to breach the apoptotic threshold in refractory BTCs.

5.2 Targeting the tumor microenvironment to sensitize tumors

To effectively dismantle protective niches within the TME, preclinical models must accurately recapitulate stromal-mediated signaling that drives therapeutic resistance in desmoplastic malignancies such as BTCs. As detailed in Section 4.3, ECM mechanics and mechanotransduction (e.g., FAK/Src, YAP/TAZ) create physical and signaling barriers to drug penetration. Recent studies using 3D co-culture spheroids and bioprinted hydrogels have demonstrated that targeting these barriers can re-sensitize tumors to apoptosis.

In cholangiocarcinoma-specific GelMA-based bioprinted models, gemcitabine and cisplatin IC50 values closely align with those obtained from Matrigel-cultured organoids and demonstrate strong concordance with the corresponding patients’ clinical treatment responses, supporting the platform’s utility for personalized drug sensitivity testing (Lu et al., 2026). Organoid models of pancreatic cancer identified focal adhesion kinase (FAK) as a central regulator of this resistance; FAK inhibition (VS-4718) remodeled the stroma and restored apoptotic susceptibility (cleaved caspase-3), a finding translated to Phase I clinical trials (Lander et al., 2022). Similar efficacy has been observed in liver and colorectal models by targeting stromal crosstalk via receptor tyrosine kinases (Chiew et al., 2017; Liu et al., 2020; Truong et al., 2019), the CXCR4-CXCL12 axis (Truong et al., 2019), or TGF-β-mediated stiffness (Dominijanni et al., 2020).

Strategies focused on stromal “normalization”—such as ATRA to induce quiescence in stellate cells (Carapuça et al., 2016) or prolyl hydroxylase inhibitors to modulate hypoxic pathways (Hu et al., 2022) —have also disrupted pro-survival niches, enhancing delivery and cytotoxicity of apoptotic stimuli. Collectively, these findings support the use of complex 3D systems to identify TME-disrupting agents that re-sensitize refractory BTCs tumors to apoptosis-inducing therapies.

5.3 Combination therapies for overcoming resistance

The high attrition rate of combination therapies in clinical trials is largely attributable to the failure of 2D monolayers to replicate the spatial barriers and metabolic heterogeneity that govern drug synergy in vivo. Three-dimensional models, particularly patient-derived organoids (PDOs) and co-cultures, serve as valuable “go/no-go” gatekeepers for validating synergistic regimens by enforcing physiological diffusion gradients and preserving genomic fidelity. For example, PDOs retain a high percentage of non-synonymous mutations found in primary tumors (Maier et al., 2021), and large-scale biobanks have demonstrated strong concordance between PDO drug sensitivity profiles and clinical outcomes (Yáñez-Bartolomé et al., 2023). Unlike 2D cultures, which often yield false-positive signals, 3D models can identify patient-specific synergies that conventional screening may miss.

As summarized in Table 2, 3D models have validated several combination strategies that overcome resistance in BTCs, including BH3-mimetics with chemotherapy, mTOR inhibitors with gemcitabine, and TME-modulating approaches such as Smac mimetics and FGFR/HDAC inhibitor combinations. Importantly, co-culture systems reveal that cancer-associated fibroblasts (CAFs) significantly alter drug sensitivity profiles, necessitating 3D environments to detect resistance mechanisms linked to specific gene signatures (Lau et al., 2024; Schrom et al., 2023; Shen et al., 2024). While these models excel at validating cytotoxic and targeted agents, future iterations must integrate immune components to effectively screen combinatorial immunotherapies (e.g., durvalumab). The transition to biologically relevant 3D co-cultures represents an important shift in preclinical validation for refractory biliary tract cancers.

TABLE 2. Therapeutic strategies and targets validated using 3D BTCs models.

5.4 Integration with high-throughput screening (HTS)

The integration of 3D biliary tract cancers (BTCs) models into high-throughput screening (HTS) platforms marks a critical evolution in precision oncology, bridging the translational gap between in vitro assays and clinical outcomes. The primary success of this transition is the establishment of patient-derived organoids (PDOs) that demonstrate high physiological fidelity; for instance, a biobank of 61 BTCs PDOs achieved a 92.3% concordance with patient clinical responses to gemcitabine and cisplatin, with machine learning algorithms (AUC >0.90) further enhancing predictive capability (Ren et al., 2023). However, scaling these models for automated HTS faces physical hurdles, primarily the handling of viscous extracellular matrix (ECM) scaffolds like Matrigel which complicate liquid dispensing, and the optical thickness of necrotic cores that hinders standard imaging (Somasekhar and Griffiths, 2023; Yan et al., 2021).

To overcome these barriers, bioengineering has introduced novel platforms that standardize spheroid formation and analysis. Image-based cytometry (e.g., Celigo) has been validated to filter false positives by simultaneously quantifying spheroid size and ATP viability, revealing distinct drug response profiles compared to 2D controls (Kessel et al., 2017). Furthermore, to address the high costs and variability of commercial plates, researchers have developed “stamp-like” agarose microwells and paper-based microfluidics (“Spheromatrix”). These low-cost innovations allow for the uniform culture of scarce biopsy samples, the retention of metabolic gene expression (e.g., CYP3A4), and even the cryopreservation of “ready-to-use” biobanks that retain drug sensitivity (Glia et al., 2025; Liao et al., 2019).

Advanced platforms are now pushing the boundaries of complexity by incorporating the tumor microenvironment (TME) and combinatorial testing. For example, 384-pillar plate technologies allow for the simultaneous mapping of radiation doses and drug concentrations, identifying synergistic radiosensitizers (Lee et al., 2022). Similarly, extrusion-based 3D bioprinting has successfully generated “tetra-culture” models (tumor, endothelial, fibroblast, macrophage) that replicate the physical barriers of the TME, yielding IC50 values that correlate closely with clinical databases (Jin et al., 2024). Despite these technological triumphs, biological challenges persist; specifically, the variable establishment rates of organoids (∼74% success) and the difficulty of maintaining stoichiometric ratios of stromal cells over long-term culture continue to complicate the standardization required for routine clinical decision support.

The integration of 3D microphysiological systems (MPS) and organ-on-chip (OoC) platforms marks a transformative advancement in HTS for BTCs, overcoming the predictive limitations of static 2D cultures by replicating the dynamic physical and biochemical cues of the TME. These systems utilize biocompatible materials, such as polydimethylsiloxane (PDMS), to mimic laminar fluid flow, vascular perfusion, and desmoplasia-induced stiffness, thereby accurately modeling drug delivery gradients and stromal resistance often observed in vivo (Heylman et al., 2014). To enhance physiological relevance, advanced platforms like the Microphysiological System Chip Platform (MSCP) enable the “daisy-chaining” of multiple organ tissues—including liver, heart, and tumor models by using a pneumatic microvalve array to switch dynamically between “organ formation” and “blood flow” modes—to simulate systemic circulation and metabolic interplay. This multi-organ configuration is critical for evaluating pharmacodynamics; studies have shown that upstream absorption by normal organoids alters the effective drug concentration reaching the tumor, thereby providing a more accurate prediction of in vivo efficacy and off-target toxicities, such as cardiomyopathy and myelosuppression, compared to isolated static cultures (Heylman et al., 2014; Zhu et al., 2024). 3D MPS with automated computational analysis have potential to evaluate drug efficacy and toxicity simultaneously. For instance, a dual-chamber closed-capillary circuit microfluidic platform, fabricated via stereolithography, has been utilized to co-culture cancer cell lines (such as SKBR3 and U87) with cardiomyocytes in collagen hydrogels, creating a physiologically relevant model for assessing chemotherapy-induced cardiotoxicity. This system integrates an Automated Image Analysis Python Script (AIAPS) to rapidly quantify 3D fluorescence z-stack images, moving beyond manual observation to high-throughput data extraction by quantify 3D morphological changes and drug sensitivity with superior precision compared to manual methods (Borrego et al., 2025).

Finally, the incorporation of PDOs into these high-throughput systems facilitates a personalized medicine approach. Hepatobiliary organoids retain the histological architecture and transcriptomic profiles of original tumors, enabling the reliable screening of FDA-approved libraries to identify high-efficacy compounds like gemcitabine and paclitaxel (Maier et al., 2021; Broutier et al., 2017; Li et al., 2019; Wang Z. et al., 2021; Wu et al., 2023). When combined with microfluidic technologies, these platforms allow for the automation of metabolic monitoring via liquid chromatography–mass spectrometry and the discovery of novel therapeutic targets, such as the inhibition of the CRIPTO/GRP78 pathway to reverse sorafenib resistance (Gu et al., 2023; Harrison et al., 2021; Karkampouna et al., 2018; Zhang et al., 2023; Zhou et al., 2019). Despite these capabilities, the broader pharmaceutical adoption of MPS remains limited by a lack of rigorous standardization and reproducibility, challenges that must be addressed to fully realize the potential of these platforms in regulatory evaluation and clinical translation (Moon et al., 2023), to transform 3D HTS platforms from experimental novelties into reliable clinical assets for decoding drug resistance and precision oncology.

5.5 Emerging therapeutic modalities and 3D models

The therapeutic landscape of biliary tract cancers (BTCs) has expanded beyond gemcitabine–cisplatin chemotherapy to include immune checkpoint inhibitors (ICIs), antibody–drug conjugates (ADCs), FGFR2-and IDH1-targeted therapies, and emerging cell-based immunotherapies. In this context, 3D tumor models have become important translational platforms for evaluating therapeutic responses, modelling resistance mechanisms, and guiding personalized treatment strategies. Patient-derived organoids, tumor spheroids, human precision-cut tumor slices (hPCTS), and multicellular bioprinted systems better recapitulate tumor architecture, stromal interactions, extracellular matrix (ECM) composition, and immune heterogeneity than conventional 2D cultures (Calvisi et al., 2023; Lancaster and Knoblich, 2014).

Because intrahepatic cholangiocarcinoma (iCCA) is frequently an immune “cold” tumor with marked immunoresistance, 3D co-culture and bioprinted models incorporating patient-derived tumor cells, cancer-associated fibroblasts (CAFs), immune cells, and ECM components have emerged as valuable tools to study tumor–immune interactions, CAF-mediated immune exclusion, and mechanisms of resistance to ICIs (Beri, 2023; Cantallops Vilà et al., 2024). Tumor–immune organoid co-cultures containing patient-matched immune cells have enabled modelling of immunotherapy responses in cholangiocarcinoma, including T-cell- and PBMC-mediated cytotoxicity, thereby supporting their use for personalized ICI testing (Dijkstra et al., 2018; Polak et al., 2024; Zhou et al., 2022). Similarly, model like hPCTS preserve native immune populations within the tumor microenvironment and provide physiologically relevant platforms for ex vivo therapeutic evaluation, although they remain limited in modelling systemic immune trafficking and advanced unresectable disease (Liu et al., 2022; Nizamoglu et al., 2023).

Beyond BTCs, several studies in other tumor types demonstrate the broader applicability of 3D platforms for antibody- and immune-based therapy development, approaches that can be adapted to BTCs models. Spheroid systems have been used to assess monoclonal antibody penetration and efficacy, including daratumumab-mediated spheroid shrinkage in lymphoma and rituximab-driven antibody-dependent phagocytosis in immune-enriched co-cultures (Foxall et al., 2020; Vidal-Crespo et al., 2020). Similar 3D co-culture platforms have also enabled investigation of NK- and T-cell cytotoxicity, cetuximab-mediated antibody-dependent cellular cytotoxicity (ADCC), and ADC-based combination therapies (Courau et al., 2019; Pece et al., 2022; Varesano et al., 2018). Microfluidic organotypic spheroid systems and air–liquid interface organoids have further supported ex vivo assessment of PD-1/PD-L1 blockade, T-cell activation, cytokine responses, and nivolumab-mediated anti-tumor activity, highlighting the utility of these platforms for immunotherapy optimization (Deng et al., 2018; Jenkins et al., 2018; Neal et al., 2018). Hydrogel-based 3D systems have additionally facilitated evaluation of bispecific antibodies and immune-cell infiltration within biomimetic matrices (van Diest et al., 2021). Such 3D stromal-rich BTCs models are particularly relevant for evaluating ADC penetration and payload delivery across the dense desmoplastic microenvironment characteristic of cholangiocarcinoma. Furthermore, metabolically relevant 3D BTCs models may help investigate how hypoxia, nutrient gradients, and stromal crosstalk influence sensitivity to FGFR2-and IDH1-targeted therapies.

Dynamic bioreactor-based 3D cultures and organotypic explant systems further extend these applications by enabling long-term modelling of tumor–stroma interactions, angiogenesis, and therapeutic responses. In hepatobiliary malignancies, organotypic bioreactor cultures of hepatocellular carcinoma and cholangiocarcinoma explants have been used to predict responses to anti-angiogenic therapies such as bevacizumab (Guzzeloni et al., 2022). Emerging technologies including bioprinting and microfluidic platforms are increasingly recognized as promising systems for studying ICIs, ADCs, targeted agents, and combination therapies in BTCs, although limitations related to cost, throughput, and incomplete systemic immune representation (Kotecki et al., 2021; Zhang et al., 2016).

5.6 Unravelling molecular heterogeneity

The integration of 3D culture systems—particularly patient-derived organoids (PDOs) and bioprinted scaffolds—with high-resolution multi-omics offers an unprecedented platform for dissecting the intricate molecular heterogeneity of Biliary Tract Cancers (BTCs). Unlike monolayer cultures, these models preserve the genomic fidelity and spatial architecture necessary to identify rare, therapy-resistant subpopulations, such as quiescent cancer stem cells or clones with specific metabolic adaptations (Broutier et al., 2017; Huang L. et al., 2015; Lin et al., 2022; Tang et al., 2022). By applying single-cell transcriptomics (scRNA-seq) and proteomics, researchers can now deconvolute bulk tumor signals to map cell-type-specific responses, including the identification of immunosuppressive Treg populations or distinct spatial autophagy patterns (Gong et al., 2023; Lee et al., 2024). This approach successfully derives organoid-based gene signatures that predict clinical responses to standard regimens like gemcitabine and cisplatin, validating these models as “clinical avatars” for precision oncology (Cesari et al., 2023; Chen et al., 2024; Yao et al., 2020).

Beyond cellular composition, the convergence of 3D models with spatial multi-omics technologies (e.g., Smart-seq3D, MALDI-MSI, CODEX) allows for the dissection of the “molecular geography” of the tumor microenvironment (TME). These tools resolve continuous transcriptional gradients—such as hypoxic core survival versus peripheral stress—and visualize drug distribution within the 3D architecture, which is critical for understanding differential drug sensitivity (Cougnoux et al., 2025; Hunter et al., 2021; Wang and Hummon, 2021). Notably, 3D reconstruction of cholangiocarcinoma samples has revealed complex branching growth patterns and “spatial subclones” driven by somatic copy number variations (CNVs), which exhibit differential sensitivities to targeted therapies like mTOR inhibitors within the same tumor block (Mo et al., 2024).

Furthermore, advanced analysis of the “colocatome” highlights how specific stromal niches, such as tumor-adjacent fibroblasts, spatially organize to induce resistance phenotypes (Bouchard et al., 2025; De Zuan et al., 2024; Gong et al., 2024). Recent innovations in rapid micro-organosphere (MOS) generation and HLA-neoantigen profiling further bridge the translational gap, enabling the assessment of patient-specific drug responses and immunotherapy biomarkers within clinically relevant timeframes (Demmers et al., 2020; Ding et al., 2022; Moncada et al., 2020; Wang W. et al., 2022). Ultimately, this coupling of longitudinal multi-omics with 3D modeling provides a robust framework for discovering novel resistance mechanisms and refining patient stratification with greater predictive fidelity than traditional assays (Di Mauro and Arbore, 2024; Qu et al., 2021; Tian et al., 2023). Collectively, these multi-dimensional insights underscore the transformative potential of omics-integrated 3D platforms to decode the spatial logic of chemoresistance, providing a robust framework for identifying novel biomarkers and realizing the promise of high-resolution precision oncology (Figure 4).

Figure 4 was generated using Google Gemini AI.

6 Future directions and challenges

6.1 Incorporation of the immune component

Current patient-derived organoid (PDO) models are limited by their inability to replicate the dynamic vascularization and immune infiltration characteristic of the desmoplastic biliary tract cancers (BTCs) tumor microenvironment (TME). To address this, the field is pivoting towards microfluidic “organoid-on-a-chip” systems that integrate physiological fluid flow, shear stress, and multicellular co-cultures to accurately model drug delivery and immune evasion (Ucci et al., 2024).

Recent studies utilizing these advanced platforms have successfully decoupled specific resistance mechanisms:

6.1.1 Immune-modulated resistance

Co-culture models have revealed that tumor-associated macrophages (TAMs) actively remodel the therapeutic landscape, enhancing viability and inducing resistance to standard regimens (e.g., cisplatin, gemcitabine, 5-FU) via the secretion of cytokines such as IL-6, TGF-beta, and VEGF (Guo et al., 2024). Similarly, the inclusion of monocytes and stromal cells has been shown to drive tumor growth and invasiveness (Haque et al., 2022).

6.1.2 Vascular and stromal dynamics

A pioneering “Cholangiocarcinoma-on-a-chip” featuring endothelialized channels and cancer-associated fibroblasts (CAFs) demonstrated that while the tumor secretome recruits T-cells, CAFs actively facilitate immune exclusion (a “cold” tumor phenotype) through immunosuppressive signaling (Polidoro et al., 2024). Furthermore, the introduction of vascular perfusion has been confirmed to yield differential drug response profiles compared to static administration, highlighting the necessity of flow dynamics for predictive accuracy. By overcoming the static limitations that obscure the role of vascular dynamics in drug delivery, these advanced systems are essential for the next-generation of research (Lai Benjamin et al., 2020).

Ultimately, the standardization of these vascularized, immune-competent chips is essential to simulate complex neuro-endocrine-immune regulatory networks and to screen combinatorial therapies in a personalized setting (Du et al., 2025; Günther et al., 2022; Wang J. et al., 2025; Wang Lxiao et al., 2025) and the widespread adoption of these dynamic systems is a prerequisite for moving beyond simple cytotoxicity screening toward a systemic understanding of how the TME orchestrates antiapoptotic activity and resistance.

6.2 Standardization and validation

The translation of patient-derived 3D Biliary Tract Cancers (BTCs) models from the bench to the bedside is currently hindered by the lack of uniform methodologies, which compromises their reliability as predictive tools for precision oncology. Unlike established cell lines, patient-derived organoids (PDOs) and spheroid cultures exhibit significant heterogeneity in establishment rates and growth kinetics, often driven by variations in isolation techniques, matrix composition (e.g., batch-to-batch variability in Matrigel), and culture media formulations significantly contribute to batch-to-batch heterogeneity and inter-institutional inconsistency (Bose et al., 2022; Brancato et al., 2020; Kopper et al., 2019). Furthermore, undefined media components can exert selective pressures that may lead to the loss of intratumoral heterogeneity or the competitive outgrowth of non-malignant cholangiocytes, failing to accurately represent the genomic landscape of the parental tumor (Calvisi et al., 2023; Fujii et al., 2016). To ensure these models serve as faithful clinical “avatars,” it is imperative to move beyond simple morphological observation and adopt rigorous, multi-omic quality assessment frameworks—comparing genomic, transcriptomic, and histological features directly against the original patient tissue (Foo et al., 2022). This standardization must extend to drug screening parameters, such as exposure timing and organoid size, which currently vary widely (Driehuis et al., 2020; Mullenders et al., 2019; Tiriac et al., 2018). The integration of automation technology to minimize handling variability (Jiang et al., 2020; Rosenbluth et al., 2020) and advanced assessment tools like optical metabolic imaging (Walsh et al., 2017) offers a promising path toward reproducibility. Recent biobanking efforts have demonstrated the efficacy of such standardized approaches, achieving high generation rates and strong concordance with clinical outcomes (AUC >0.93) (Ren et al., 2023). Ultimately, widespread clinical utility depends on the community adopting consensus-driven “experimental model sheet” criteria and validating these models through prospective trials that account for diverse patient metadata (Calvisi et al., 2023). Therefore, establishing a universally accepted framework for standardization and validation is not merely a technical necessity, but the critical prerequisite for transforming 3D BTCs models into reliable clinical tools capable of accurately predicting therapeutic response and overcoming drug resistance.

6.3 Microfluidic “lab-on-a-chip/organ-on-a-chip” systems

Microfluidic “lab-on-a-chip (LoC)/organ-on-a-chip (OoC)” platforms represent a transformative evolution in BTCs research, overcoming the limitations of static 3D models by simulating dynamic mechanical forces—such as fluid shear stress, interstitial pressure, and laminar flow—that critically influence cytoskeletal reorganization, epithelial-to-mesenchymal transition (EMT), and drug response (Bhatia and Ingber, 2014; Mani et al., 2019; Sleeboom et al., 2018). Unlike static cultures, these systems enable precise spatiotemporal control over drug gradients and pharmacokinetic profiles, essential for decoding chemoresistance mechanisms often obscured in standard assays (Lanz et al., 2017). For example, bioprinted tri-culture models incorporating hepatic parenchyma have demonstrated CYP3A4-mediated metabolic activation of prodrugs like cyclophosphamide, revealing apoptotic responses and p53 downregulation undetectable in monocultures (Liu et al., 2023). Furthermore, recent “cholangiocarcinoma-on-a-chip” advancements have reconstructed the desmoplastic and vascularized TME, showing that CAF-mediated matrix stiffening and collagen deposition correlate with Gemcitabine-Cisplatin resistance via YAP/TAZ signaling, while simultaneously orchestrating immune evasion through cytokine modulation (IL-6, IL-10, CXCL12) (Polidoro et al., 2024; Lu et al., 2018; Sun et al., 2020). These platforms also facilitate the study of the metastatic cascade, including intravasation and organotropism, effectively validating the “seed and soil” hypothesis (Aleman and Skardal, 2019; Bittner et al., 2020; Mi et al., 2019; Xu et al., 2016). Moving forward, the integration of high-throughput omics, the transition from PDMS to bio-based materials like silk fibroin (Carvalho et al., 2018), and the utilization of patient-derived “avatars”—potentially isolated directly from bile via novel immunomagnetic microfluidic systems (Truong et al., 2019; Gout et al., 2025; Hung et al., 2017)— will be pivotal for bridging the translational gap in precision oncology, offering a physiologically relevant platform to decipher the dynamic interplay between hemodynamic forces, stromal interactions, and drug sensitivity in biliary tract cancers.

6.4 Ethical and logistical considerations

The translation of patient-derived organoids (PDOs) from preclinical research to clinical application in biliary tract cancers (BTCs) is currently restricted by a convergence of logistical, technical, and ethical hurdles. A fundamental challenge remains the acquisition of high-quality biospecimens; while surgical resection provides ample tissue, most BTCs patients present with unresectable disease, necessitating reliance on low-yield biopsies or endoscopic ultrasound-guided fine-needle aspiration (Yanaidani et al., 2024). Although bile-derived organoids (BDOs) offer a minimally invasive “liquid biopsy” alternative with higher establishment rates, their utility is tempered by non-malignant contamination and the need for time-consuming enrichment protocols that must align with the narrow therapeutic window of aggressive disease (Kinoshita et al., 2023; Zhao et al., 2025). Beyond acquisition, the lack of standardized manufacturing, quality control, and pharmacoeconomic models complicates the scalability and regulatory validation required for “n-of-1″ precision medicine (Alexander et al., 2021; He et al., 2014; Podgórski et al., 2023; Ravi et al., 2023; Schuh and Funk, 2019; Shim, 2023). Furthermore, navigating the intellectual property landscape and varying international standards poses significant legal barriers (3D Printing in Medicine, 2023; Wilbanks, 2013). Finally, the establishment of “living biobanks” introduces complex ethical dilemmas regarding the immortalization of donor identity, necessitating the evolution of consent frameworks from static to “dynamic” models and the development of adaptive governance to manage data privacy and commercialization concerns (Boers et al., 2016; Bredenoord et al., 2017). Thus, realizing the full potential of BTCs organoids for precision medicine necessitates navigating these complex intersections of law, ethics, and logistics to establish a standardized, patient-centric clinical pipeline.

6.5 Integration with artificial intelligence (AI)

The Integration of AI and 3D tumor modeling can bring transformative paradigm shift in decoding the complex mechanisms of drug resistance and apoptosis in BTCs. While 3D systems such as organoids, spheroids, and organs-on-chip (OoC) provide superior physiological relevance compared to 2D cultures, their inherent morphological heterogeneity and the massive, high-dimensional datasets they generate—ranging from multi-omics to real-time sensor metrics—necessitate the use of advanced machine learning (ML) and deep learning (DL) algorithms for standardized, high-throughput analysis (Dai et al., 2023; Lee et al., 2023). Deep learning architectures like Convolutional Neural Networks (CNNs) and U-Net are now being utilized to automate label-free viability assessments, single-cell segmentation, and the tracking of subtle phenotypic changes invisible to the human eye, such as metabolic reprogramming and invasive front dynamics (Chen et al., 2021; Dave et al., 2025). Emerging platforms like HCS-3DX integrate AI with Light-Sheet Fluorescence Microscopy to bridge the gap between morphological variability and standardized screening. This approach utilizes AI not only for post-acquisition analysis—such as single-cell segmentation and classification via Multi-Layer Perceptron (MLP) algorithms—but also for the pre-selection and handling of spheroids. By employing AI-guided tools (e.g., SpheroidPicker), researchers can standardize cultures by selecting morphologically homogeneous samples, thereby overcoming reproducibility issues inherent in manual handling (Diosdi et al., 2025). Similarly, “SMART” (Spheroid Monitoring and AI-based Recognition Technique) system, which utilizes improved CNN to automate the detection of tumor spheroid boundaries. This AI-driven approach replaces inconsistent manual measurements with precise automated recognition and introduces novel invasiveness metrics, such as the Excess Perimeter Index (EPI) and the Multiscale Entropy Index (MSEI). Unlike traditional unidirectional measurements, these indices capture complex morphological changes, providing a more physiologically relevant assessment of tumor behavior and drug response in both established cell lines and primary tumor cultures (Chen et al., 2021). Furthermore, AI is optimizing the biofabrication workflow in 3D bioprinting by predicting bioink properties and providing real-time monitoring of structural fidelity, which is critical for consistent drug screening (Bracco et al., 2025; Jin et al., 2023). These computational strategies, including generative models like Generative Adversarial Networks (GANs) for data augmentation are paving the way for “digital twins” and “patient-on-a-chip” systems, where patient-specific genomic data are coupled with ex vivo drug sensitivity patterns to simulate individualized therapeutic responses, potentially bypassing traditional animal testing (Baniukiewicz et al., 2019; Caballero et al., 2020; Gangwal and Lavecchia, 2025; Li et al., 2023; Serrano et al., 2024). High-performance models like TransOrGAN and DeepDILI are already demonstrating superior accuracy in predicting transcriptomic mappings and drug-induced liver injury, respectively (Li et al., 2023; Li et al., 2020) Despite this potential, significant challenges persist regarding the “black-box” nature of deep learning, which complicates regulatory approval, as well as the risk of algorithmic bias and the requirement for massive, high-quality datasets to prevent overfitting (Alzubaidi et al., 2023; Bonner et al., 2022; Timilsina et al., 2025). For successful transition these tools into the clinic, future efforts must prioritize interpretable AI models and rigorous validation against patient-matched clinical outcomes (Benning et al., 2020; Sadeghi et al., 2024; Tang et al., 2024) to accelerate the timeline for drug discovery, screening and precision medicine in BTCs.

7 Conclusion

The transition from 2D monolayers to 3D tumor models in biliary tract cancers (BTCs) research signifies a crucial advancement, addressing the limitations of traditional flat cell cultures. These limitations include the absence of spatial architecture, physiological nutrient and oxygen gradients, cell–extracellular matrix (ECM) interactions, and multicellular signalling, all of which contribute to the poor predictive validity of 2D-based drug screening and the high failure rates of antineoplastic agents. Models such as spheroids, patient-derived organoids (PDOs), and “tumor-on-a-chip” systems enhance our understanding of apoptosis and drug resistance, accurately reflecting the multicellular resistance seen in clinical settings by revealing interactions among the extracellular matrix, metabolic stress, and Bcl-2 family signaling. These 3D systems facilitate the exploration of alternative cell death pathways, such as ferroptosis and necroptosis, uncovering vulnerabilities in drug-resistant cancer subpopulations. Looking ahead, the translational potential of 3D models in precision oncology is significant. PDO biobanks that maintain the genomic and histological integrity of tumors are vital for personalized medicine, high-throughput screening, and predicting treatment responses. By combining these models with advanced profiling and microfluidic technologies, researchers aim to enhance drug efficacy, explore combination therapies, and identify new targets within the TME. However, 3D models are not a standalone solution and address only the preclinical modelling phase of the translational pipeline. Their ultimate clinical impact will depend on integration with animal studies, including patient-derived xenograft (PDX) models, biomarker discovery, improved clinical trial design, and healthcare system innovations. Despite ongoing challenges in standardization and immune integration, the adoption of 3D culture systems marks an important shift for preclinical drug screening, potentially bridging laboratory discoveries with clinical applications when combined with other translational tools. Whether this will translate into improved survival rates for BTCs patients remains to be demonstrated in prospective clinical studies. (Figure 5).

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. During the preparation of this manuscript, the authors used ChatGPT and Copilot for language polishing, grammar checking, and literature search organisation. Google Gemini (generative AI) was used to create Figures 4. No generative AI was used to generate scientific content or conclusions. The authors take full responsibility for the accuracy and integrity of the manuscript and all figures.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher’s note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: Daniel Nieto, University of A Coruña, Spain

Reviewed by: Kasturi Mahadik, Centre for Cellular and Molecular Biology (CCMB), India

Pavan Kumar Chintamaneni, Gandhi Institute of Technology and Management School of Pharmacy, India