Experimental models of osteoarthritis: Advances, limitations and rational selection for translational research

Abstract

Osteoarthritis (OA) is a complex degenerative joint disease driven by mechanical overload, inflammation, metabolic disorders, and aging. Despite substantial advances in basic and translational research, no disease-modifying OA drugs (DMOADs) have been approved for clinical use, largely due to the translational gap between preclinical experimental models and clinical practice. Experimental models are essential tools for investigating OA pathophysiology and evaluating therapeutic strategies, among which animal models play an irreplaceable role in recapitulating the in vivo joint environment. This review comprehensively evaluates current OA experimental models, with a focus on animal systems, including surgically induced structural instability, mechanically induced models, chemical induction approaches, metabolism-related models, spontaneous aging models, and genetically modified strains. We assess each model's advantages, limitations, and appropriate applications, clarifying their ability to reflect specific OA subtypes. Moreover, we summarize recent progress in in vitro platforms, ex vivo tissue explants, and emerging technologies such as organ-on-a-chip and organoid models, which provide complementary insights to animal models. This review aims to establish a rational model selection framework aligned with OA pathological features, thereby bridging the preclinical-clinical translational gap and facilitating the development of effective therapeutics.

The translational potential of this article

This review systematically evaluates OA experimental models and analyzes their practical utility in translational research of different OA subtypes. It establishes a reasonable framework for model selection and describes how cutting-edge bioengineered platforms complement traditional in vitro and animal studies, providing a reference for improving the predictive reliability of preclinical research and advancing studies on personalized OA therapies.

Graphical abstract

1. Introduction

Osteoarthritis (OA) is a globally prevalent, disabling joint disease involving progressive dysfunction of the entire joint [1,2]. Pathological changes affect multiple joint tissues, including articular cartilage, synovium, subchondral bone, meniscus, infrapatellar fat pad, and their crosstalk, leading to chronic pain, stiffness, and irreversible functional impairment [[3], [4], [5]]. Although great efforts have been made in basic and translational research, no disease-modifying osteoarthritis drugs (DMOADs) are approved so far, highlighting a persistent disconnect between preclinical model findings and clinical efficacy [[6], [7], [8]]. A key cause of this translational gap is the biological heterogeneity of OA. Accumulating evidence shows OA is not homogeneous but comprises multiple phenotypes and molecular endotypes, driven by distinct risk factor combinations of mechanical overload, inflammation, metabolic disorders, genetic susceptibility, and aging [[9], [10], [11]]. This heterogeneity leads to obvious variability in disease trajectories, dominant pathological drivers, and treatment responses, complicating mechanistic research and therapy development.

Experimental models are indispensable for investigating OA pathophysiology and evaluating therapies under controlled conditions. In vitro systems, especially cell-based cultures, have been the cornerstone of mechanistic research. Due to their accessibility, scalability, and controllability, they are widely used to study molecular signaling, matrix catabolism, and drug responsiveness [12]. However, they can hardly recapitulate the complex joint microenvironment. Hence, ex vivo tissue explants and animal models have been developed to overcome in vitro reductionism by preserving tissue architecture or systemic interactions [13,14]. These models also have inherent drawbacks, namely significant interspecies differences in joint anatomy, biomechanics, lifespan, and immunity. More importantly, variability in model induction hinders comparability across studies and neglecting OA heterogeneity directly leads to high candidate therapy attrition in clinical translation [15], emphasizing the need to establish models that match specific subtypes of OA and to re-evaluate existing OA experimental models. Recent advances in bioengineering, microfabrication, and stem cell biology have addressed the needs by developing human-relevant microphysiological systems. Emerging organ-on-a-chip (OoC) and organoid models integrate multiple joint tissues with defined biochemical and biomechanical cues, providing a robust framework to explore OA heterogeneity and improve translational relevance [[16], [17], [18], [19], [20]].

In this review, we evaluate the current landscape of OA experimental models, from conventional in vitro and animal systems to emerging bioengineered platforms (Fig. 1). While previous reviews mainly classify OA experimental models according to technical platforms, species, and induction protocols, the current review establishes a phenotype- and research-oriented translational paradigm for rational model selection. A suitable model is selected based not only on its recapitulation of human OA across anatomical and molecular dimensions, but also on its compatibility with disease drivers, target tissues and therapeutic strategies for a specific preclinical research question. By systematically assessing each model's strengths, limitations, and optimal uses, we aim to establish a framework for rational model selection. We integrate conventional cell and explant cultures, small/large animal models, and innovative organ-on-a-chip and organoid systems into a stepwise framework to guide model selection and validation. This rational decision-making framework will reduce translational mismatches between preclinical outcomes and clinical trial cohorts, support subtype-driven OA research, enhance preclinical translational relevance, and facilitate development of effective DMOADs and personalized therapies for OA patients.

Fig. 1.

Experimental models for OA research. Conventional 2D and 3D cell culture models as well as ex vivo explant models are limited in replicating the complex in vivo microenvironment of OA. Animal models can be generated through diverse methodologies. The selection of species and models that accurately reflect OA subtype pathology plays a crucial role in preclinical OA research. Emerging advanced models, including organ-on-a-chip, multi-tissue chip, mechanically active chip, and organoids, provide new platforms for the heterogeneity research and drug development of OA.

2. In vitro and ex vivo models

2.1. Monolayer cell models

The monolayer model refers to the culture of a single cell population on polystyrene surfaces in nutrient media supplemented with essential growth and maintenance factors [21]. This model is simple to operate and cost-effective, and therefore represents a commonly used in vitro approach for OA research. By culturing cells as an adherent monolayer on flat plastic surfaces, it partially recapitulates aspects of the in vivo cellular microenvironment while allowing convenient observation of cellular morphology, proliferation, and differentiation. In OA studies, commonly used cell types in monolayer models include chondrocytes, synoviocytes, and mesenchymal stem cells (MSCs) [13]. Under in vitro culture conditions, these cells are able to respond to various stimuli, such as cytokines and growth factors, thereby mimicking cellular responses involved in the initiation and progression of OA.

Monolayer cell models, particularly those based on human cell lines, are widely used to screen anti-inflammatory or chondroprotective molecules by quantitatively assessing a panel of inflammatory and cartilage-degradation–related biomarkers [22,23]. Moreover, monolayer cell models provide a useful platform for investigating multiple signaling pathways. For example, these models are used to investigate the regulatory roles of microRNAs [24]. Monolayer cell models are also applied to evaluate gene transfection efficiency. For instance, the delivery of insulin-like growth factor 1 and transforming growth factor-β1 into chondrocytes or MSCs via gene transfection has been shown to significantly upregulate the expression of cartilage-specific markers [25]. Monolayer models can also simulate the crosstalk between different cell types in the joint microenvironment (e.g., the interaction between chondrocytes and synoviocytes), which plays a crucial role in the initiation and progression of OA inflammation and cartilage degradation.

However, monolayer cell models have several drawbacks that restrict their ability to simulate the in vivo pathological microenvironment of OA. The planar growth characteristic of monolayer cultures, coupled with the absence of mechanical stimuli, directly leads to chondrocyte dedifferentiation and altered cellular morphology [26]. Moreover, monolayer models lack three-dimensional (3D) cell-cell and cell-extracellular matrix (ECM) interactions, which are essential for maintaining the phenotypic stability of cells. Furthermore, monolayer cultures fail to replicate the in vivo nutrient gradients and metabolic microenvironment and the planar culture system cannot mimic the complex mechanical microenvironment of the joint.

2.2. Three-dimensional culture cell models

Compared to monolayer cell culture models, 3D cell culture technology can control the structure and microenvironment of cell compartments through integrating fluid circulation, growth factor delivery, and physical stimuli such as oxygen levels, external pressure, and fluidity, thereby providing a better simulation of the physiological and pathological states of joints [27]. 3D cell culture models, especially 3D spheroid culture models, have been widely used in the fields of cartilage tissue engineering, disease modeling, and drug screening. Using a 3D spheroid culture system, researchers can simulate the early condensation stage of chondrogenesis [28]. Moreover, a recent study revealed that cells in 3D culture models show better protective effects in response to inflammatory stimuli, which may be attributed to the physical barriers provided by the 3D structure, suggesting 3D culture may be more suitable for exploring the inflammatory endotype [29].

With the development of drug discovery and tissue repair fields, cell sheet technology emerges as a new scaffold-free 3D cell culture technique. Cell sheet technology can control cell adhesion and detachment through temperature-responsive surfaces, forming intact cell sheet structures. For example, Tan et al. provided preclinical evidence that a scaffold-free, chemically detachable chondrocyte sheet can support structurally and biochemically appropriate cartilage regeneration (Fig. 2), highlighting its translational potential for cartilage repair [30]. Moreover, cell sheet technique has advantages such as uniform cell distribution, retention of the extracellular matrix, easy transplantation, and stacking to form complex 3D structures, favoring the research of large, cell-heterogeneous, and matrix-poor tissues. It can also provide multifunctionality and customizability to study OA heterogeneity, especially in terms of cell type selection, culture condition setting, and experimental design optimization [31]. Hence, 3D cell culture models, ranging from traditional spheroid culture to advanced cell sheets, can recapitulate a more complex cellular microenvironment and structural characteristics of native joint tissues relative to 2D culture.

Fig. 2.

Cartilage cell sheets for OA research. (A) External and ultrastructural traits of fabricated chondrocyte patches. A1) Chondrocyte sheets situated on a 6-well insert before isolation. A2) The isolated sheet from A1. A3) The sheet from A1 spread out in a 6-cm culture dish. A4) Sheet placed on a 6-cm culture insert. A5) The isolated sheet from A4. A6) A triple-layered assembled construct. A7) Micrograph (40×) demonstrating cellular outgrowth from a sheet explant. A8) H&E staining demonstrate that both single-layer and multi-layer chondrocyte sheets can be easily cultivated on and harvested from customized inserts, and the detached cell sheets exhibited cell migration and proliferation properties. (B) Immunohistochemical and immunofluorescence staining of the chondrocyte sheets. B1, B4, B7) Staining for Collagen I. B2, B5, B8) Staining for Collagen II. B3, B6, B9) Staining for Aggrecan. The expression pattern (Collagen I-negative, Collagen II-positive, Aggrecan-positive) verifies the hyaline cartilage feature of the cell sheet. Reproduced and adapted with permission from Ref. [30].

2.3. Explants

Explant models are typically derived directly from joint tissues of animals or humans, such as cartilage, subchondral bone, and synovium, and maintain the original morphology and function of these tissues [13]. Explant models, especially cartilage explants derived from OA patients, are used to simulate the pathological environment of OA in vitro. It allows researchers to observe and analyze the occurrence and development of OA. For example, D'Oria et al. developed a 3D synovial explant culture system that preserved the architectural (Fig. 3A) and cellular features (Fig. 3B) of OA synovium [32]. Li et al. established a dose-controlled ex vivo inflammatory osteoarthritis model using human femoral head osteochondral explants stimulated with IL-1β and TNF-α, which reproduced OA-like catabolic, inflammatory, and matrix-degradative changes while maintaining tissue viability (Fig. 3C) [33]. In addition, Zevenbergen et al. utilized a bovine cartilage-on-cartilage explant model combined with DENSE MRI to quantify intratissue strain patterns around full-thickness chondral defects under compressive loading (Fig. 3D) [34].

Fig. 3.

Explant models for simulating OA microenvironment. (A) OA synovial tissue explants preserve both the histological characteristics (as illustrated by hematoxylin and eosin staining; up panel) and the cellular viability (as confirmed by the immunohistochemical staining of cleaved caspase 3; down panel). Reproduced and adapted with permission from Ref. [32]. (B) Flow cytometry analysis was conducted to evaluate the different cellular components in OA synovium explants at various time points. Reproduced and adapted with permission from Ref. [32]. (C) Osteochondral explants isolated from human femoral heads were utilized for the establishment of an ex vivo inflammatory OA model. C1) The osteochondral explants were obtained from the region between the dotted lines in a human femoral head. C2) Cartilage thickness of explants in the Day 0 group, control group, and groups exposed to different concentrations of IL-1β and TNF-α for 3 and 7 days. Reproduced and adapted with permission from Ref. [33]. (D) MRI protocols were employed to evaluate the mechanical response and compositional properties of intact and defect cartilage explants. Rapid acquisition with relaxation enhancement (RARE) images enabled morphological evaluation of cartilage geometry. Reproduced and adapted with permission from Ref. [34].

Explants can be used to test the permeation and distribution of drugs or drug delivery systems in cartilage, subchondral bone, and other tissues. The therapeutic effects of new treatment strategies can also be investigated under controlled conditions. McMasters et al. found that fluorescent dye-labeled drug-loaded hollow PEGylated poly (NIPAM) nanoparticles could effectively deliver a therapeutically active dose of drug to bovine cartilage explants and inhibit the expression of pro-inflammatory IL-6 after stimulation with IL-1β [35]. Geurts et al. established a human osteochondral explant model derived from knee and facet joint OA specimens that preserves cartilage–subchondral bone integrity and enables assessment of inflammatory and drug treatment responses ex vivo (Fig. 4A and B) [36]. A previous study used cartilage explants from OA patients to evaluate the targeting (Fig. 4C and D) and therapeutic effects of cartilage-targeting peptide and nanoparticle engineered MSCs on cartilage repair [37]. In another study conducted by Bhattacharjee et al., they used explant models to assess the therapeutic effects of a combination therapy of sulfated carboxymethyl cellulose (sCMC) and tissue metalloproteinase inhibitor 3 (Timp3) on OA (Fig. 4E and F) [38]. It demonstrated the value of explant models in assessing the effectiveness of drug combination therapies.

Fig. 4.

The application of cartilage explants for evaluating drug and stem cell efficacy. (A) Tissue viability was examined after explant culture of OA osteochondral samples obtained from knee (up) and facet joint (down). Explants were either kept untreated (control) or stimulated with 1 μg/mL lipopolysaccharide (LPS) to induce inflammation, either in the absence or presence of a drug intervention (10 μM TGF-β type I receptor inhibitor). Reproduced and adapted with permission from Ref. [36]. (B) The concentrations of bone metabolic and inflammatory proteins secreted from the explants, including pro-collagen type I (pro-Col-I), IL-6, and monocyte chemoattractant protein-1 (MCP-1), were measured using enzyme-linked immunosorbent assay. Reproduced and adapted with permission from Ref. [36]. (C) Schematic diagram of the functionalization of cartilage-homing peptides on the surface of MSCs via a click-chemistry reaction. Reproduced and adapted with permission from Ref. [37]. (D) Representative bioluminescence images of human knee OA cartilage explants following 3-day in vitro co-culture with the peptide functionalized MSCs. Reproduced and adapted with permission from Ref. [37]. (E) Representative safranin O and alcian blue staining of goat cartilage explants after IL-1β stimulation and treatment with sCMC, Timp3, or their combination. Reproduced and adapted with permission from Ref. [38]. (F) Representative immunofluorescence staining of goat cartilage explants showing OA-related inflammation and degradation markers, including NFκB, ADAMTS4, and MMP13. Reproduced and adapted with permission from Ref. [38].

Explants can also be applied to detect OA-related biomarkers. The surface plasmon resonance (SPR) immunosensor developed by Kausaite-Minkstimiene et al. successfully quantified human cartilage oligomeric matrix protein (COMP, an OA biomarker) in explants, demonstrating the promising applications of explants in biomarker detection [39]. Thus, explant models that preserve the native structure and function of joint tissues represent a useful ex vivo platform for investigating OA pathogenesis, evaluating therapeutic interventions, assessing drug delivery, and identifying OA-related biomarkers.

Taken together, in vitro and ex vivo models serve as the first tier for OA experimental models. Monolayer culture systems are ideal for dissecting signaling pathways, conducting rapid compound screening and optimizing gene delivery. However, they cannot reflect therapeutic effects in the intact joint due to their simplified structure. Three-dimensional cultures and cell-sheet constructs improve cell-matrix interactions, and provide more meaningful information for research into cartilage repair, inflammatory responses, and tissue engineering. Explants preserve native tissue architecture and patient-derived pathological features, while reducing the costs and ethical constraints associated with animal experiments. Thus, these in vitro and ex vivo models are recommended for studies focusing on in vitro molecular mechanisms, tissue-level responses, drug penetration or early therapeutic validation. Importantly, any positive outcomes must be further validated using models that recapitulate joint biomechanics, systemic metabolism, pain behaviour, and long-term structural progression of OA.

3. Animal models established by different factors

Current OA animal models simulate certain aspects of specific etiologies or pathogenic mechanisms observed in OA patients and cover a range of small animals such as mice, rats, rabbits, and guinea pigs, as well as large animals like dogs, goats, sheep, and horses [14,15]. These models include spontaneous models (naturally occurring models and transgenic models), induced models (surgically induced models and chemically induced models), and non-invasive models [40].

3.1. Structural damage models

Structural damage models, especially surgically induced models are the most widely used OA models. Anterior cruciate ligament transection (ACLT) is a common surgical method to induce OA models [41]. Through transecting the anterior cruciate ligament, this model can simulate the pathological changes of OA, including the degeneration of joint cartilage, synovitis, and the formation of osteophytes, which are highly similar to human OA. Another common surgical method is meniscectomy, especially the medial meniscal instability (DMM). The DMM model can induce mild to moderate OA lesions after the transection of the tibial ligament of the medial meniscus, making it an ideal choice for studying early OA [42]. A comparative study by Glasson et al. revealed differences between the ACLT and DMM models in inducing OA. The ACLT model can trigger severe OA structural changes, including the formation of joint capsule cartilage and significant erosion of subchondral bone. In contrast, the DMM model is relatively mild, mainly affecting the central weight-bearing area of the medial tibial plateau and medial femoral condyle, with no evidence of subchondral bone destruction observed throughout the observation period. Adebayo et al. studied the differences between the DMM and ACLT models in the kinematic properties of the mouse knee joint and the pathogenesis of OA. They found that under controlled tibial compressive loads, the kinematics of the DMM model joint were similar to those of a normal intact joint, while the ACLT model joint exhibited significant instability. Moreover, OA progression was rapid in the ACLT model, while it was relatively slow in the DMM model. Based on these findings, they concluded that the DMM model is more representative and relevant in simulating physiological joint kinematics and is a more suitable model for studying the occurrence of post-traumatic OA [43].

In rabbit models, posterior cruciate ligaments (PCL) transection model induces OA through the simple transection of the posterior cruciate ligament, with minimal impact on knee joint stability. Deng et al. constructed a rabbit PCL transection model to explore the morphological and histological changes of the medial meniscus after PCL transection. The results showed that the medial meniscus exhibited time-dependent degenerative changes at 4, 8, 16, and 24 weeks after PCL transection. The expression levels of TIMP-1, MMP-1, and MMP-13 were significantly increased on the PCL transected side relative to the control group [44]. The medial meniscus tear (MMT) method induces OA through meniscectomy combined with ligament injury. Huang et al. found that MMT model exhibited more severe cartilage structural damage and higher levels of catabolic gene expression at all observed time points relative to the ACLT model in rabbits [45]. In addition, the Hulth model induces OA by cutting the medial collateral ligament, ACL and PCL, and medial meniscus to disrupt joint stability, causing more severe cartilage damage. A modified Hulth method (combining with a focal cartilage defect, namely MHCD model) has been adopted in some studies. The MHCD model can induce OA pathological changes more rapidly and severely [46].

In addition to surgically induced models, there are three main non-invasive methods for inducing OA models with structural damage: intra-articular tibial plateau fracture (IATPF), cyclic articular cartilage tibial compression (CACTC), and tibial compression overload leading to ACL rupture [40]. The IATPF was the first non-invasive method developed, which involves fixing the bent knees of anesthetized mice and applying an impact force to cause a closed fracture of the joint [47]. This method allows researchers to change the severity of the lesion by adjusting the applied force, thereby simulating different degrees of traumatic injury. One of the main advantages is the capability of replicating the acute trauma caused by high-energy impacts and the resulting post-trauma OA [47], but this model is less applicable to OA cases caused by overuse and chronic injuries [14]. The CACTC model mainly involves applying axial loads to the posterior knee joint, causing the tibia to move forward relative to the femur, thereby simulating the impact of chronic overuse injuries on the development of OA [47]. This method allows for the cyclic application of loads over a period of time to study long-term injury effects, making it particularly suitable for chronic OA research [48]. The induction method of tibial compression overload leading to ACL rupture involves a one-time single overload applied to the knee, causing anterior subluxation of the tibia, thereby producing an injury. This method does not require invasive surgery while the pathology and development of OA are similar to those in ACLT model [49]. The traumatic load needs to be applied only once to produce the expected results [50]. Moreover, this injury leads to changes in bone microstructure and metabolites in synovial fluid, specifically manifested as a significant decrease in the proximal tibial bone volume fraction, and trabecular thickness, as well as metabolic dysregulation in the injured limb. These changes are also observed in the contralateral limb, indicating the presence of systemic skeletal and metabolic responses [51]. Overall, these models may be suitable for the study of structural phenotypes, and cartilage and bone damage/remodeling related endotypes.

3.2. Load-related models

Load-related OA models are usually induced by non-invasive methods, including repetitive loading models, single injury models, and exercise models [47]. Repetitive loading models apply controlled mechanical forces to joints to simulate the pressure experienced by joints during daily activities. For example, Poulet et al. successfully induced cartilage damage and OA-like lesions by applying controlled mechanical forces to the tibia of mice using a custom-made cup device [52]. Single injury models induce anterior cruciate ligament injury in the knee joint through a single episode of tibial compressive overload. This model is closer to surgical ligament transection models and has the advantages of being non-invasive, easy to implement, and highly reproducible [53]. Exercise models investigate the long-term effects of exercise on OA by controlling the amount of exercise in animals, such as training mice to run regularly and controllably on a treadmill with preset numbers of runs, speeds, and distances [54]. The long-term moderate-intensity running training can increase the incidence and severity of knee OA in mice, with this finding being more pronounced in transgenic OA-susceptible mice [55]. Non-invasive loading models avoid surgical intervention, enhance the reproducibility of the models, and reduce the potential impact of surgery on disease progression [47]. Hence, these models can be used to study load-bearing phenotypes, and cartilage and bone damage/remodeling endotypes.

3.3. Chemically induced models

The intra-articular injection of chemicals is also a common method to establish OA models. Injection of monosodium iodoacetate (MIA) can induce OA-like pathological changes and pain in the knee joint of mice and rats. The mechanism involves the inhibition of the activity of glyceraldehyde-3-phosphate dehydrogenase activity and the blockade of the glycolytic process, thereby triggering chondrocyte death, neovascularization, subchondral bone necrosis and collapse, and inflammatory responses [56]. Through precisely controlling the injection dose of MIA, researchers can flexibly induce OA-like pain responses of varying intensities, providing an experimental platform for pain management and the development of treatment strategies [57]. Similarly, the collagenase-induced OA (CIOA) model is also used in assessing therapeutic effects of drugs on OA-related pain and disease progression in mice [58]. Collagenase injection can simulate the early synovitis and pain development process of human OA in animals. Weber et al. explored the interaction between mechanical damage and inflammation on OA progression by injecting different doses of collagenase into the single joint of Wistar rats [59]. They showed that collagenase injection led to dose-dependent cartilage degeneration, bone resorption, and mechanical weakening of the anterior cruciate ligament, thereby causing persistent joint instability and synovitis. In addition, by injecting stromal cell derived factor 1 into the knee joint, rabbit inflammatory OA models can be induced [60]. Therefore, the chemically induced models are suitable for the research of the inflammatory phenotype and endotype.

3.4. Metabolism-related models

In metabolism-related models, the ovariectomized (OVX) model is an important research tool for OA. This model is constructed by surgically removing the ovaries to reduce estrogen levels, thereby simulating the physiological changes of menopausal women [61]. Estrogen is an important protective factor in preventing osteoporosis and OA. After ovariectomy, the level of estrogen decreases, so the incidence of OA significantly increases [62]. Mice after ovariectomy exhibit significant cartilage degeneration and subchondral bone destruction, including a significant reduction in cartilage volume, surface area, and thickness, along with the destruction of the subchondral bone and the loss of trabecular structure [63]. Ovariectomy is usually combined with surgically or chemically induced OA models to study the role of estrogen in the occurrence of OA and to explore treatments for OA caused by estrogen deficiency. For example, Tian et al. established a postmenopausal OA rat model using bilateral ovariectomy combined with intra-articular monosodium iodoacetate injection. They found that miR-203 was significantly upregulated, while ERα expression was downregulated. Dual-luciferase reporter assays confirmed that miR-203 directly binds to and negatively regulates ERα. The elevated miR-203 levels were correlated with increased expression of MMP-1, MMP-3, PGE2, and CTX-II. Moreover, inhibition of miR-203 alleviated cartilage degradation in this model [64].

Changes in dietary conditions are applied in the research of metabolism-related OA. A diet high in fat and high in sucrose can significantly accelerate the progression of OA in DMM model [65]. A previous study investigated the impact of metformin on knee OA in normal diet mice and high-fat diet-induced obese mice. We found that metformin can not only reduce apoptosis and catabolism of chondrocytes, and also suppress infiltration and pro-inflammatory polarization of synovial macrophages. Particularly, metformin has a greater protective effect on OA in high-fat diet-induced obese mice through reducing leptin secretion from adipose tissue [66]. These results emphasize the important role of specific dietary habits and metabolism in the development of OA heterogeneity.

3.5. Models induced by systemic factors

Specific strains of laboratory animals can spontaneously develop OA lesions without human intervention. Among them, the Dunkin Hartley guinea pig model, as the most commonly used guinea pig OA model to date, exhibits highly similar pathological features to human OA, especially in the progression of age-related OA [67]. The articular cartilage of guinea pigs gradually develops characteristic OA pathological changes, including structural damage and cartilage loss with age [68]. In another STR/Ort mouse strain, high levels of local and systemic inflammatory markers, including RAGE, AGE, HMGB1, etc., were detected in the joints, and the concentrations of multiple cytokines in the serum also increased as compared to OA-resistant mouse strains [69]. It is shown that targeted inhibition of aggrecanases can effectively prevent the degradation of joint cartilage and increase bone mass in STR/Ort mouse model, providing a tool for the development of new treatment strategies for OA and osteoporosis [70].

Very recently, tree shrew, as a new type of small mammal model, has emerged as a promising candidate for the study of OA and musculoskeletal disorders (MSDs) due to the close genetic relationship with primates. Specifically, after reaching skeletal maturity, the tree shrew naturally undergoes a joint degeneration process closely related to aging, showing high similarity to humans (Fig. 5) [71]. These spontaneous OA animal models are considered as ideal tools for studying OA subtypes caused by systemic factors.

Fig. 5.

Establishment of a progressive OA model in tree shrews. (A) Schematic illustration of an OA model in tree shrews by DMM surgery. (B) Representative histological staining of joint samples in sham and DMM-induced OA tree shrews at 4, 8, 12, and 16 weeks post-surgery. (C) Representative immunohistochemical staining of joint samples with COL2A1, aggrecan, and MMP13 in tree shrews at different time intervals. (D) OARSI score analysis in sham and OA tree shrews. (E) Quantification of immunohistochemical staining. Reproduced and adapted with permission from Ref. [71].

3.6. Transgenic models

Transgenic models play a significant role in the study of the functions of specific genes in OA. Through genetic engineering techniques, animal models with specific gene expression or deletion can be created. For example, Young and Kobayashi used Gdf5-Cre transgenic mouse model to knock out Piezo1 and Piezo2 in the joint. They induced DMM model to investigate whether the knockout of Piezo channels affects the expression of OA-related genes, providing insights into the role of Piezo1 and Piezo2 in OA pathology [72]. Collins et al. used a cartilage-specific Sirt6 knockout (Sirt6 cKO) mouse model and induced OA with DMM surgery to investigate the effects of cartilage-specific Sirt6 deficiency on the development of post-traumatic and age-related OA in mice [73]. They found that Sirt6-deficient mice exhibited more severe cartilage damage, osteophyte formation, and subchondral bone sclerosis. Similarly, Suo et al. constructed a Zmpste24-specific knockout mouse model, and revealed that the absence of Zmpste24 accelerates the senescence of chondrocytes and triggers the pathological process of osteoarthritis, such as cartilage wear, subchondral bone proliferation, and osteophyte formation, indicating that ZMPSTE24 plays a key role in maintaining chondrocyte health and preventing OA [74]. Thus, the transgenic models provide powerful tools for exploring the mechanisms of specific genes and discovering potential therapeutic targets in different OA subtypes.

3.7. Large-animal models for translational research

Large-animal models serve as a critical translational system that rodent models alone cannot replace. Their distinct features, including joint size, cartilage thickness, and gait mechanics, facilitate the evaluation of delivery strategies, implants, rehabilitation protocols, and long-term structural outcomes. For this reason, they are the late-stage validation models rather than high-throughput discovery platforms. The relevant OA phenotypes and therapeutic mechanisms should first be defined using small-animal systems before large animals are adopted.

Canine models have been developed for studying post-traumatic OA and for separating cartilage-intrinsic damage from gross joint instability. The Pond-Nuki model, originally established by ACLT in canines, reproduces progressive cartilage degeneration, osteophyte formation, and synovial responses that can be followed longitudinally [75,76]. Molecular analyses showed region-specific gene expression during early disease, supporting its value for mapping spatially heterogeneous OA responses [77]. By contrast, the canine femoral groove model developed by the Utrecht group produces a controlled cartilage injury without deliberately destabilizing the joint. This design induces steadily progressive OA-like changes and has been used to study cartilage damage, synovial involvement, and subchondral bone remodeling in a mechanically intact joint [[78], [79], [80]]. More recently, Yu and colleagues used a beagle canine OA model, together with rat experiments, to test chondrocyte membrane-coated nanoparticles for intra-articular retention and disease modification, illustrating how canine models can support translational assessment of drug delivery systems before clinical testing (Fig. 6A–C) [81].

Fig. 6.

Representative large-animal models used in translational OA research. (A) Schematic illustration of ACLT surgery and intra-articular treatment in a beagle canine model. (B) Representative macroscopic images of femoral and tibial articular surfaces from canine knee joints after treatment. (C) Representative histological and immunohistochemical images of canine femoral condyles and tibial plateaus, including toluidine blue (TB), safranin O/fast green (SO&FG), picrosirius red (PR), and COLII staining. Reproduced and adapted with permission from Ref. [81]. (D) Retroreflective kinematic markers and surface electrodes in a horse, illustrating the feasibility of gait, kinematic, and neuromuscular assessment in equine large-animal studies. Reproduced and adapted with permission from Ref. [82]. (E) Representative macroscopic images of knee joint cartilage in a Göttingen minipig ACLR model, showing variable degrees of cartilage surface alteration. (F) Representative MRI and corresponding safranin O-stained histological images from the minipig knee, illustrating the use of imaging-histology comparison in large-animal OA assessment. Reproduced and adapted with permission from Ref. [83].

Equine models offer additional strengths for studies that require whole joint, cartilage thickness, and arthroscopic accessibility closely mimicking clinical orthopaedic scenarios. The Colorado group has extensively used the equine carpal osteochondral fragment model to simulate post-traumatic OA after focal osteochondral injury. This model is well suited to arthroscopy, magnetic resonance imaging, synovial biomarker analysis, and controlled exercise or rehabilitation regimens (Fig. 6D) [82,84,85]. As athletic horses are also prone to spontaneous joint disorders, equine research bridges experimentally induced injury models and naturally occurring OA observed in clinical practice.

Sheep, goats, and porcine models represent powerful tools for osteochondral repair, biomaterial validation, and device-focused OA research. Focal osteochondral defects in ovine models recapitulate perilesional pathology analogous to human osteochondral injury, making them ideal for studying the interface between focal cartilage damage and wider joint degeneration [86]. Goat models have been used to evaluate biphasic scaffolds and other cartilage-bone repair strategies across long-term follow-up. They deliver comprehensive data on graft integration, subchondral reconstruction, and mechanical resilience under physiological loading [87]. Göttingen minipigs and other porcine models possess inherent advantages for knee surgery, imaging, and implant evaluation because of their joint size, cartilage architecture, and weight-bearing characteristics (Fig. 6E and F) [83]. Across these large animal models, studies yield the greatest translational value when endpoints extend beyond histological degeneration to include graft integration, load distribution, imaging biomarkers, gait analysis, and clinically applicable delivery systems [83].

Overall, animal models remain irreplaceable for testing hypotheses that involve whole-joint biology, systemic regulation, pain behavior, and long-term structural progression. Animal models should be selected for translational research aims, not technical convenience alone. Surgically induced and non-invasive injury models are well suited for studying post-traumatic or mechanically induced OA, while chemically induced models are ideal for assessments of inflammatory subtype. Metabolic, ageing-related, and genetically modified models are valuable when the therapeutic target is linked to systemic metabolism, senescence, endocrine status or a defined molecular pathway. Moreover, small animal models stand out in large-scale experiments and mechanistic exploration, while larger animals better recapitulate physiological joint loading, support imaging and surgical procedures, and enable reliable assessment of implant integration and long-term safety. Well-designed preclinical research should adopt these models in a sequential strategy: starting with mechanistic testing, moving on to phenotype-matched in vivo validation, and progressing to large-animal studies when needed prior to clinical translation.

4. Emerging experimental models

With the development of bioengineering technology, the emergence of OoC and organoids has shown great potential in OA [16,17]. Importantly, the US Food and Drug Administration (FDA) has recently approved that preclinical drug testing is no longer mandatory to use animal models [88]. Clinical trials for investigational therapies based entirely on preclinical efficacy data from OoC results combined with existing safety data without animal testing are undergoing [89]. This milestone breakthrough not only demonstrates the great potential of organoids/OoC systems in drug development but also paves the way for new strategies in OA research.

4.1. Organ-on-a-chip

OoCs can simulate organ functions in microfluidic systems and reproduce pathophysiological conditions with unprecedented precision [18]. Despite challenges in OoC development, such as cell sourcing, physiological relevance, high-throughput screening, and clinical validation, OoCs have shown promising potential in simulating joint inflammation, responses to mechanical stimuli, and genetic predisposition to OA, as well as assessing drug effects through different administration routes in OA. For example, Petta et al. designed a new microfluidic OoC-based OA model that contains two compartments: cartilage and synovium. These tissues were separated by a specially designed synovial fluid channel and a hyaluronic acid-based matrix material was used to construct the cartilage compartment. This model could simulate inflammatory OA by the addition of OA synovial fluid with inflammatory cytokines and degradative enzymes (Fig. 7A) [90]. Lin et al. developed an iPSC-based osteochondral tissue chip that mimics OA pathology under IL-1β stimulation in a dual-flow bioreactor, serving as a platform for drug screening, mechanism investigation, and personalized medicine [93]. Based on the microfluidic 3D chondrocyte culture-on-a-chip system developed by Rosser et al., which features a two-layer structure with a glass top containing inlets and a PDMS bottom with microfluidic channels, a semicircular tissue chamber (3 mm in diameter, 1 mm in height) for cell-laden hydrogel, and an adjacent rectangular medium channel (21.5 mm length × 1.0 mm width) to establish nutrient gradients, the platform enables the recapitulation of native cartilage microenvironment and serves as an animal-free in vitro model for studying osteoarthritis pathophysiology and treatment responses [94].

Fig. 7.

Organ-on-a-chip for OA research. (A) OoC-based OA model that contains cartilage and synovium. A1) Schematic diagram demonstrating how each component is recapitulated in the OoC. A2) Photos of the OoCs. Channels specific to the synovial and chondral compartments were highlighted in yellow and red, respectively. Reproduced and adapted with permission from Ref. [90]. (B) MOoC model of NVBchip. B1) The NVB-chip contains three distinct compartments. On the left, the nerve unit (blue) incorporates embryonic DRG explants as the source of neuronal cells. In the central channel, the vascular unit (purple) contains endothelial cells embedded within a hydrogel matrix. On the right, the bone unit (green), in which osteoclasts will be seeded. B2) Tile-scan microscopic image illustrating the microfeatures of the chip, including microchannels and micropillars. B3) Low-magnification tile-scan micrograph of the chip, displaying the elongated compartments and medium reservoirs. C4) Images depicting hydrogel loading (light blue) into the central channel (t = 60 s; left) and the application of aqueous solution (dark blue) to the lateral channel (t = 80 s; right). Reproduced and adapted with permission from Ref. [91]. (C) MOoC model combined with mechanical stimulation. C1) This MOoC system consists of three biologically independent chambers and is constructed from three superimposed layers: a cell culture layer (CCL), a mechanical actuation layer (MAL), and a glass coverslip. C2) Each culture chamber incorporates five channels, specifically two outer channels for culture medium (channels 1 and 5), two for cell-laden hydrogels (channels 2 and 4), and one for the injectable therapeutic agent (channel 3). B3) In the resting state, cell-laden hydrogels and the therapeutic product receive no mechanical stimulation. Upon pressurization of the MAL, they undergo confined mechanical compression of 30%. Reproduced and adapted with permission from Ref. [92].

4.2. Multi-tissue chips and mechanically active models

Multi-tissue chips or multi-organs-on-chips (MOoCs) are advanced forms of OoC system, containing multiple organ components connected through a microfluidic system to achieve interaction and communication between different tissues. Joint chips, as a multi-tissue organ chip platform, can integrate various cell types and tissues within the joint, such as cartilage, synovium, bone, and infrapatellar fat pad, to simulate the complex microenvironment of the joint [19]. Through techniques such as microfluidic technology, dynamic perfusion systems, and mechanical stimulation devices, joint chips can reproduce key factors such as cell-cell and cell-matrix interactions within the joint, as well as biomechanical stimulation [95].

In addition, different stimuli, such as inflammatory cytokines (simulating inflammatory subtype) and sodium palmitate (simulating metabolic subtype), can be added to the chips through microfluidic system for studying OA heterogeneity and classification. For example, one MOoC is designed for inflammatory subtypes, with its core simulating interactions between synovium, cartilage and immune cells. Patient-derived synovial fibroblasts and pro-inflammatory macrophages are co-cultured in the “synovium” chamber, which is connected to the “cartilage” chamber (containing chondrocytes or cartilage tissue blocks) via a “synovial fluid” channel containing hyaluronic acid. When the system is continuously perfused with medium containing TNF-α or IL-1β, the therapeutic effect of drugs toward inflammatory OA subtype can be observed by monitoring synovial inflammatory factor secretion, cartilage matrix degradation markers release, and immune cell infiltration. Neto et al. described a microfluidic-based neurovascularized bone chip (NVB chip) designed to model the mechanistic interactions between innervation and angiogenesis within an inflammatory bone niche. This NVB chip comprises three compartments: a left-side nerve unit containing embryonic dorsal root ganglion explants as a neuronal source, separated by microgrooves that allow only axon crossing; a central vascular unit where endothelial cells are seeded within a hydrogel and constrained by micropillars; and a right-side bone unit for osteoclast seeding. This integrated chip enables the establishment of a reproducible neurovascular network under inflammatory conditions, which may be used to serve as a experimental model of OA subchondral bone (Fig. 7B) [91]. For the metabolic subtype, a “fat-cartilage-bone” interaction model can be constructed by co-culturing fat tissue (derived from adipose stem cells of obese patients) with cartilage and bone tissue, perfusing with medium containing high concentrations of adipokines and adipokine receptors (leptin, soluble leptin receptor, resistin, or adiponectin) to simulate metabolic dysfunction [96]. The focus is on studying the effects of adipokines on chondrocyte metabolism, inflammatory responses, and bone remodeling [96,97].

Previously, a human mesenchymal stem cell-derived miniature joint system (miniJoint; Fig. 8) was generated [98]. The engineered osteochondral complex, synovial-like fibrous tissue, and adipose tissue were integrated to simulate the healthy and inflamed human knee joint. Joint inflammation was simulated by stimulating the synovial-like fibrous tissue with the pro-inflammatory cytokine IL-1β, which enabled us to reproduce the inflammatory and degenerative changes seen in human OA samples. The miniJoint system could be also used to evaluate the efficacy of nonsteroidal anti-inflammatory drugs and potential disease-modifying OA drugs, providing a promising platform for drug efficacy and toxicity assessment. In addition, mechanically active OA chip models introduce mechanical stimulation to the tissue components in OoCs or MOoCs to mimic the biomechanical signals present in native joints.

Fig. 8.

MiniJoint for OA drug screening. (A) The miniJoint is a microscale system recapitulating the human knee joint. A1) The miniJoint system incorporates engineered adipose tissues (AT), synovial-like fibrous tissues (SFT), and osteochondral tissues (OC), which are designed to mimic the adipose tissue, synovium, and cartilage-bone complex of the native human knee joint. Inter-tissue crosstalk was achieved through either diffusion (within the OC compartment) or fluid flow. A2) Microtissue modules were constructed through the photo-crosslinking process of GelMA hydrogel loaded with human bone marrow MSCs. A3) The miniJoint system was developed by incorporating differentiated microtissues into a chamber. 4) Tissue-specific culture medium streams (AM, FM, and OM; marked with yellow, pink, and blue, respectively) were perfused on the upper portion of the chip, while a shared common culture medium stream (SM; purple) was perfused at the bottom to replicate the physiological function of synovial fluid. A5) Photo of the miniJoint system. A6) A chronological schedule (with a maximum duration of 63 days) is provided, detailing the processes involved in constructing miniJoint cultures and establishing joint disease models. A7) The miniJoint is capable of generating four parallel replicates for each individual microtissue type. Reproduced and adapted with permission from Ref. [98] (B) Assessment of naproxen (NPX) therapeutic activity in the inflamed miniJoint model. B1) A schematic illustration and time schedule outlining the “systemic” administration of NPX in the miniJoint. Specifically, NPX was incorporated into all culture medium streams (denoted by black arrows) 3 days after the SFT tissue was treated with IL-1β. B2) Immunostaining micrographs exhibited a significant reduction in MMP-13 and IL-6 expression levels within the SFT microtissue following NPX intervention. B3) Safranin O staining and immunostaining revealed increased glycosaminoglycan retention and decreased MMP-13 expression in the OC-C region after NPX treatment. Reproduced and adapted with permission from Ref. [98].

The on-chip environment with in vivo-like biochemical and biophysical stimulation can more comprehensively reflect physiological and pathological processes in human joints. For example, Palma et al. developed the uBeat® MultiCompress platform, a microfluidic system that co-cultured human cartilage microtissues with injectable therapeutics under controlled and physiological mechanical compression (30% hyperphysiological strain) (Fig. 5C) [92]. This mechanically active model recapitulates key OA pathologies, including the upregulation of inflammatory genes (PTGS2, IL-6), matrix-degrading enzymes (MMP13), hypertrophic markers (COL10A1), and downregulation of matrix genes (ACAN). Using this platform, a novel injectable conjugate SYN321 was demonstrated to significantly reduce the expression of these pro-inflammatory and catabolic markers while promoting aggrecan deposition, thereby validating its anti-inflammatory and cartilage-protective effects in a human-relevant, mechanically stimulated environment. Overall, OoCs and advanced MOoCs systems can integrate multiple joint tissues and dynamic crosstalk under biochemical, metabolic, and mechanical cues, enabling precise recapitulation of joint pathophysiology, holding great promise for mechanistic studies, drug evaluation, and personalized therapy. These models effectively bridge simple cell cultures and in vivo animal studies, providing critical support for preclinical research in OA.

4.3. Organoids

Organoids are 3D tissues typically derived from stem cells via self-assembly and spatially restricted lineage commitment processes, thereby simulating the structure and function of native tissues/organs and providing a unique platform for studying the development and disease conditions of human organs. Recent advances have highlighted that the biological fidelity and translational relevance of organoids are highly dependent on multiscale construction strategies and systematic functional evaluation frameworks [20]. Shen et al. pointed out that organoids could be used to construct disease models to explore the pathogenesis of OA and serve as fillers for rapid cartilage repair [99]. Cho et al. found that TGF-β3-induced cartilage organoids can significantly promote cartilage regeneration in animal models and restore collagen II and aggrecan expression, achieving structure and function close to normal cartilage [100]. Similarly, Sun et al. prepared a cartilage organoid using composite hydrogel loaded with miR-24 microspheres and synovial MSCs. This organoid promoted chondrogenesis by reversing cellular senescence through the miR-24/TAOK1 signaling axis, and thus efficiently repaired cartilage defects and delayed joint degeneration in a rat model (Fig. 9A) [101]. Single-cell RNA sequencing further confirmed that this organoid remodeled chondrocyte homeostasis by regulating key metabolic pathways, including glycolysis and oxidative phosphorylation. Miao et al. developed biomimetic gradient-structured cartilage organoids (BGSC-organoids) based on decellularized cartilage extracellular matrix loaded with SOX9-overexpressing BMSC-derived extracellular vesicles [104]. They also constructed a cyclic mechanical stimulation-integrated OA-on-a-chip (OAOC) model to simulate cartilage mechanical microenvironment, confirming that BGSC-organoids achieved sustained release of chondroprotective factors and exhibited excellent mechanical tolerance via the Vimentin/14-3-3/FOXO3 pathway. Thus, BGSC-organoids effectively delayed articular cartilage and intervertebral disc degeneration in animal models. Single-cell sequencing revealed that BGSC-organoids elevated the proportions of cartilage progenitor and proliferative chondrocytes to form a stemness-maintaining CPC-ProC cycle, while 3D bioprinting optimized OAOC scaffold structure, overcoming the limitations of traditional OoC models and providing a biomimetic and regulable dual system for cartilage regeneration research.

Fig. 9.

Cartilage organoids with biomaterials and stimulus. (A) Cartilage based on composite hydrogels and synovial MSCs. A1) Schematic diagram showing the 3D-cultured cartilage organoids used for the treatment of cartilage damage in rats. A2) Scanning electron microscopy images of the miR-24-conjugated hydrogel (left panel). The miR-24-conjugated hydrogel exhibited good shape retention and the organoids were well-distributed within the hydrogel matrix (middle panel). Organoids were marked with yellow dotted circles (right panel). Reproduced and adapted with permission from Ref. [101]. (B) Bone marrow MSC-derived cartilage organoids in OA-related inflammatory microenvironment. B1) Schematic illustration of the establishment and functional characterization of the organoids. B2) Morphological characteristics and bright-field microscopic observations of the cartilage organoids. Reproduced and adapted with permission from Ref. [102]. (C) Generation of OA cartilage organoids within inflammatory environment. C1) A diagram illustrating the experimental procedure for generating OA cartilage organoids from human bone marrow MSCs exposed to inflammatory cytokines. C2) Histological examinations were performed on the organoids at day 14 and day 28 of the culture process. Reproduced and adapted with permission from Ref. [103].

Moreover, the OA microenvironment can be mimicked during the organoid culture. Gao et al. utilized 3D synovial organoids to recreate the OA microenvironment, successfully simulating inflammation, angiogenesis, and fibrosis, thereby modeling inflammatory OA characterized by chronic synovial inflammation and fibrotic lesions, providing a new tool for elucidating the pathogenesis of OA [105]. Strategies that utilize organoids to recapitulate pathological microenvironments provide an ideal platform for evaluating the penetration and distribution of nanomedicines in complex physiological contexts [106]. Zhang et al. found that BMSC-derived cartilage organoids can mimic cartilage structure and function in a 3D environment and have the potential to repair OA cartilage defects (Fig. 9B) [102]. However, in an inflammatory microenvironment, IL-1β-induced upregulation of IL-6, TNF-α, CCL2, and CXCL1 inhibits cell adhesion and ECM formation, significantly reducing transplantation success rates and repair efficacy. This model also reveals the negative effects of OA-related inflammatory mediators on organoids, providing a platform for screening interventions that enhance anti-inflammatory capabilities. Dönges et al. constructed a cartilage organoid model to simulate the pathological features of OA by exposing human bone marrow MSCs to an inflammatory environment. These organoids reproduced the degenerative, catabolic, and inflammatory characteristics of OA cartilage (Fig. 9C) [103]. This method can be extended to simulate specific OA subtypes. For example, continuous exposure to inflammatory cytokines during early differentiation can induce the formation of cartilage organoids with inflammatory subtype characteristics, exhibiting high expression of catabolic and pro-inflammatory proteins. Conversely, culturing precursor cells in an environment enriched with free fatty acids and insulin resistance inducers may generate organoids mimicking a metabolic subtype, characterized by lipid droplet accumulation, mitochondrial dysfunction, elevated endoplasmic reticulum stress markers, and potential increased sensitivity to lipotoxicity-induced damage. These disease-specific organoids induced through targeted induction can be used for high-throughput screening of candidate drugs targeting specific pathological pathways.

OoC and organoid technologies are newly developed platforms. They complement rather than replace traditional experimental models. A major advantage is that they can reconstruct human-like microenvironments with precise control over inflammatory, metabolic, mechanical and multicellular cues that are hard to isolate in animal experiments. OoC systems are useful for studying tissue crosstalk, perfusion, mechanical stimulation, and drug response under reproducible conditions. Organoids are flexible tools for patient-specific disease modelling, cartilage repair research, and phenotype-based drug screening. However, current systems are still limited to reproduce long-term disease evolution, systemic endocrine, and immune regulation, sensory innervation, and joint-level mechanical loading. Thus, these platforms work best for exploring molecular mechanisms, selecting promising therapies and producing human-relevant experimental data. They also help build more targeted translational workflows that avoid animal testing.

Gene editing technology is also used to construct organoids. Wei et al. developed a cartilage organoid system from human expanded pluripotent stem cells (hEPSCs) carrying a COL2A1^mCherry and COL10A1^eGFP double reporter for real-time monitoring of chondrogenesis and hypertrophy. They also screened 2040 FDA-approved drugs in this system and found that α-adrenergic receptor (α-AR) antagonists could promote chondrogenesis and repress hypertrophy (Fig. 10) [107]. This human cartilage organoid model has promising applications. Therefore, organoids have emerged as promising and personalized platforms for OA heterogeneity research by regulating the culture or induction conditions of organoids and gene editing of stem cells. Hence, organoids recapitulate the structure, function, and pathological microenvironment of native joint tissues, serving as robust models for OA pathogenesis study, cartilage regeneration, drug screening, and subtype-specific research. Combined with biomaterials, mechanical stimulation, and gene editing, they offer a customizable and translational platform for personalized OA therapy and disease modeling.

Fig. 10.

Cartilage organoids developed using gene editing technique. (A) Establishment and characterization of hypertrophic cartilage organoids derived from COL2A1mCherry and COL10A1eGFP human expanded pluripotent stem cells. A1) Upper panel: A schematic illustration depicting the directed differentiation process of hEPSCs into hypertrophic cartilaginous organoids. Middle panel: Bright-field micrographs captured at typical time points during differentiation. Lower panel: Representative fluorescent images (red and green channels) obtained throughout the hypertrophic maturation stage. A2) Representative immunofluorescence images of cryosections prepared from organoids at day 42 of culture. COL2-H2B-mCherry (displayed in red) was expressed throughout the entire organoid structure. In contrast, COL10-H2B-eGFP (shown in green) was mainly localized at the marginal region of the organoid. A3) A schematic diagram illustrating the zonal distribution pattern of different cell types within the organoid. A4) Representative high-magnification immunofluorescence images corresponding to the hypertrophic zone, middle zone, and deep zone of the organoid. Reproduced and adapted with permission from Ref. [107]. (B) SLPI deletion in cartilage organoids inhibits the hypertrophic tendency in vivo and promotes cartilage repair. B1) A schematic illustration and chronological timeline outlining the implantation of chondrocyte pellets in an osteochondral defect model established in NSG mice. B2) Representative macroscopic views (left panel) and staining results of articular cartilage in the three experimental groups. B3) Immunofluorescence staining was performed to identify the implanted chondrocytes within the repaired tissue. The repaired chondral defect area was highlighted by dotted lines. B4) Representative immunofluorescence images showing (from left to right) ACAN, COL1, IHH, RUNX2, and MMP13 expression in the three groups. Reproduced and adapted with permission from Ref. [107].

5. Translational pipelines for OA model selection

A practical model-selection framework should guide the combination of experimental systems along the entire translational pathway. Since OA therapies differ substantially in their mechanisms of action, target tissues, and intended clinical benefits, a single model is insufficient to support the full progression from mechanistic research to clinical testing [19,21,108]. Instead, preclinical studies should adopt staged model sequences in which early-stage platforms address mechanistic questions, intermediate systems verify effects at the tissue level, and more complex in vivo or human-relevant models assess outcomes most relevant to the target patient population.

For small molecules and biological agents, translational pipelines typically start with cultured human chondrocytes, synoviocytes or co-culture systems to confirm target engagement and modulation of inflammatory or catabolic pathways. Candidate treatments are then tested in cartilage, synovial or osteochondral explants to evaluate matrix protection, tissue penetration, and biomarker release within a native extracellular matrix [32,33,36]. In vivo validation should be tailored to the primary therapeutic goal. DMM, ACLT, and MMT models are suitable for assessing structural deterioration after joint injury, while CIOA and MIA models are more meaningful when inflammation, pain or neuroimmune responses serve as core endpoints [[109], [110], [111], [112], [113], [114], [115]]. Imaging findings, soluble biomarkers, and histological scores should all align with the key disease driver, rather than being applied as generic outcome measures.

Interventions targeting pain and physical function also demand careful model selection. Structural endpoints alone cannot fully evaluate treatments designed for analgesia or functional recovery. Models incorporating MIA, CIOA or post-traumatic injury may be chosen based on whether the study focuses on acute nociception, inflammatory pain or the interplay between structural damage and pain [56,58,[113], [114], [115]]. Behavioral readouts, including weight-bearing asymmetry, gait analysis, mechanical sensitivity, and spontaneous activity, should be interpreted alongside synovitis severity, subchondral bone alterations, neural markers, and imaging features. Incorporating neural components via OoC systems, dorsal root ganglion–joint cell co-cultures or neurovascular chips can further clarify the mechanistic link between joint pathology and pain [91,113,115].

Gene and cell therapies require a more comprehensive testing sequence. Their efficacy depends not only on biological activity, but also on delivery efficiency, long-term persistence, biosafety, and tissue integration. Initial cell culture and organoid models are ideal for examining transgene expression, differentiation stability, cellular senescence, hypertrophy, and inflammatory responses [24,99,101,107]. Explants and OoC platforms can further evaluate local retention, matrix interaction, and crosstalk between synovium, bone, and adipose tissue [92,98]. Animal studies remain important for assessing biodistribution, immune reactions, ectopic tissue formation, pain behavior, and long-term repair outcomes. For these therapeutic modalities, large-animal models become especially valuable when joint anatomy, surgical accessibility, imaging resolution, and loading conditions related to rehabilitation affect translational interpretation [110,116,117].

Biomaterials, nanomedicines, and orthopaedic devices should follow a distinct testing workflow. Material characteristics including release kinetics, degradation profile, injectability, and mechanical properties must be fully investigated before evaluating biological efficacy [15,116,117]. Ex vivo cartilage and osteochondral explants are useful to test tissue penetration, graft integration, and local inflammatory reactions of biomaterials. Small-animal defect or OA models can generate early evidence of tissue repair or disease modification using biomaterials [[33], [34], [35]]. However, devices, implants, and load-bearing biomaterials require later validation in large animals because these models better recapitulate clinically relevant surgical handling, joint-scale biomechanics, longitudinal imaging, implant fixation, and safety assessment.

Finally, translational pipelines need to be designed with clinical patient stratification in mind. Preclinical imaging, biomarkers, histological signatures and response profiles from organoids or OoCs can help identify patient subgroups most likely to respond to the tested mechanism. For instance, therapies developed for metabolic OA should be paired with adipokine, inflammatory, and metabolic indices. By contrast, repair-focused interventions should rely on imaging, biomechanical, and integration-related endpoints [[116], [117], [118], [119], [120]]. In this regard, model selection is not only a technical decision for preclinical work, but also acts as a strategy to align biological mechanisms, study endpoints, and future clinical trial cohorts (Table 1).

Table 1.

Practical framework for rational selection of OA experimental models.

Clinical scenario/phenotype-endotype	Dominant pathological features	Key tissues/compartments	Intervention types	Suggested validation sequence	Core readouts	Time course and complexity	Role in translation	Selection rationale	References
Post-traumatic/structural instability OA	Mechanical instability, cartilage erosion, subchondral bone remodeling and osteophyte formation.	Cartilage, meniscus, synovium and subchondral bone.	Disease-modifying small molecules, biologics, gene therapies, biomaterials and devices.	Use cell models for target screening, explants for matrix-level confirmation, DMM/ACLT/MMT models for whole-joint structural efficacy, and larger-animal models when load-bearing repair, imaging or implant mechanics must be tested.	OARSI score, cartilage thickness, COL2A1, ACAN, MMP13, ADAMTS5, micro-CT/MRI and gait.	Medium to long; complexity rises with surgical induction, imaging and biomechanical testing.	Early mechanism testing, in vivo structural validation and late-stage mechanical or imaging assessment.	Best aligned with interventions intended to slow mechanically driven structural progression.	[109,110]
Metabolic/obesity-related OA	Low-grade systemic inflammation, adipokine imbalance, altered lipid metabolism and subchondral bone changes.	Cartilage, synovium, infrapatellar fat pad/adipose tissue and subchondral bone.	Metabolic modulators, anti-inflammatory agents, biologics and nanomedicines.	Begin with adipocyte-joint cell co-culture or fat-cartilage-bone chips, then use diet-induced or combined HFD plus injury models to test systemic-joint interactions; larger-animal studies are reserved for long-term structural or mechanical questions.	Leptin, adiponectin, IL-6, TNF-alpha, cartilage catabolic markers, synovitis, body weight, glucose/lipid indices and imaging.	Medium to long; interpretation depends on both joint pathology and systemic metabolic state.	Endotype modelling and prioritization of therapies that modify metabolic inflammation rather than cartilage alone.	Useful when OA progression is expected to reflect systemic metabolic dysfunction as well as local joint injury.	[[118], [119], [120]]
Inflammatory OA	Synovitis, cytokine-driven matrix catabolism, immune-cell activation and pain sensitization.	Synovium, cartilage, immune cells, subchondral bone and sensory nerve-associated compartments.	Anti-inflammatory small molecules, biologics, gene therapies and nanomedicines.	Test pathway activity in cytokine-stimulated joint cells, confirm tissue responses in synovium-cartilage explants, and use CIOA or MIA models when inflammation, pain and structural endpoints need to be assessed together.	IL-1beta, TNF-alpha, IL-6, MMPs, ADAMTSs, synovitis score, pain behavior and inflammatory biomarkers.	Short to medium; immune-integrated chips and behavioral readouts add technical complexity.	Target validation and linkage of anti-inflammatory effects with structural and symptomatic outcomes.	Appropriate for interventions directed at synovitis, cytokine signaling or neuroimmune pain mechanisms.	[111,112]
Ageing-related OA	Cellular senescence, mitochondrial dysfunction, impaired matrix turnover and slow spontaneous degeneration.	Ageing chondrocytes, cartilage matrix, synovium, subchondral bone and whole-joint microenvironment.	Senolytics, mitochondrial modulators, biologics and gene therapies.	Use senescent chondrocyte or MSC systems to test mechanism, organoid or explant models to examine matrix ageing, and spontaneous or naturally ageing animal models for chronic structural progression.	p16, p21, SASP factors, mitochondrial function, ROS, cartilage degeneration, osteophytes and subchondral bone remodeling.	Long; spontaneous phenotypes improve face validity but increase time and cost.	Assessment of therapies aimed at senescence, tissue resilience and slow disease modification.	Most relevant when the intended indication is gradual, age-associated OA rather than acute injury-driven disease.	[[121], [122], [123]]
Pain-dominant OA	Nociceptive sensitization, synovitis, subchondral bone innervation and neurovascular remodeling.	Sensory neurons/DRG, synovium, subchondral bone, cartilage and neurovascular units.	Analgesics, anti-inflammatory agents and neuroimmune modulators.	Use neuron-joint cell co-culture or neurovascular chips for pathway screening, then select MIA, CIOA or surgical models according to whether rapid pain behavior or structural-pain coupling is the priority.	von Frey response, weight-bearing asymmetry, gait, spontaneous activity, CGRP, NGF, synovitis and subchondral bone changes.	Short to medium; behavioral testing requires parallel structural assessment to avoid overinterpreting pain signals.	Symptom-oriented efficacy testing and selection of endpoints for analgesic or function-focused interventions.	Needed when pain relief or functional improvement is a primary therapeutic claim.	[[113], [114], [115]]
Cartilage repair/biomaterial-device scenario	Focal cartilage or osteochondral defects, mechanical mismatch, implant integration and local inflammation.	Cartilage, osteochondral unit, subchondral bone, synovial interface and implant-material interface.	Biomaterials, hydrogels, cell therapies, nanomedicines and orthopaedic implants/devices.	Characterize the material first, then test cartilage or osteochondral integration ex vivo; use small defect models for biological repair signals and sheep, goat, dog or horse models for surgical handling, joint-scale mechanics, imaging and safety.	Mechanical properties, degradation, release kinetics, integration, MRI/CT, histology, gait and local immune response.	Medium to long; high complexity because surgical feasibility, mechanics, imaging and safety must be assessed together.	Stepwise material/device qualification before clinically oriented repair or implant studies.	Most appropriate for interventions whose performance depends on joint-scale loading, implantation and long-term integration.	[116,117]

Abbreviations: OA, osteoarthritis; DMM, destabilization of the medial meniscus; ACLT, anterior cruciate ligament transection; MMT, medial meniscus tear; HFD, high-fat diet; CIOA, collagenase-induced osteoarthritis; MIA, monosodium iodoacetate; DRG, dorsal root ganglion; OARSI, Osteoarthritis Research Society International.

6. Limitations and future perspectives

Despite substantial advances in experimental modeling of OA, inherent limitations still persist across different types of models. The 3D culture models improve cellular interactions and microenvironmental relevance compared with monolayer systems, yet remain limited by insufficient mechanical support, poor shape control, and challenges in experimental reproducibility [124]. Explant models preserve native tissue architecture and enable the assessment of OA heterogeneity and therapeutic responses; however, tissue extraction may induce outer-layer cell death, accurately reproducing pathological OA microenvironments ex vivo remains difficult, and strict control of culture conditions is required to maintain tissue viability, all of which compromise reliability and reproducibility [108,125]. Animal models continue to serve as indispensable tools for investigating OA mechanisms and phenotypes, although no single model can fully recapitulate the complexity of human OA or reliably predict clinical drug responses [15,126,127]. Structural damage–based models are influenced by operator-dependent variability in surgical procedures, while non-invasive models are limited in their ability to quantify in vivo shear stress, thereby restricting their interpretability primarily to cartilage and bone damage–related endotypes [128]. Moreover, models induced by systemic factors are affected by strain-dependent differences in OA incidence and lesion severity, further complicating cross-study comparability and reproducibility [129,130].

Emerging OoC platforms offer improved control over microenvironmental parameters and enable partial reconstruction of OA-related pathological processes; nevertheless, most current designs incorporate a limited number of joint components, often lacking synovium, immune cells, bone tissue, or macrophages, and are constrained by short culture durations, immature matrices, small sample sizes, and insufficient mechanical or hypoxic stimulation, which collectively restrict their capacity to model whole-joint pathology and long-term disease progression [90,93,94,131,132]. Even more advanced multi-tissue and mechanically active joint chips have yet to integrate sensory innervation, a critical component in OA pathophysiology, thereby limiting their utility in studying pain mechanisms, neuro-immune interactions, and the evaluation of analgesic therapies; additionally, fully recapitulating the complex extracellular matrix composition, dynamic biomechanical cues, and systemic endocrine or metabolic signaling remains technically challenging. Organoid-based models offer patient-relevant platforms for investigating OA heterogeneity and cartilage repair, but are still constrained by immune rejection risks in xenograft settings, potential long-term tumorigenicity, insufficient vascularization, limited structural complexity, and batch-to-batch variability, together with the lack of standardized protocols for large-scale production and quality control. Furthermore, current cartilage organoids remain unable to fully reproduce the complex mechanical loading conditions and prolonged pathological evolution characteristic of native joints, and long-term culture may induce cellular phenotypic and functional drift, necessitating continuous monitoring.

Addressing these challenges will require interdisciplinary collaboration, technological innovations, such as gene editing and OoC integration, and clearly defined staged development strategies, ultimately enabling the scalable generation of functionally mature bone and cartilage organoids with clinical translational potential [133]. Future research should prioritize integrative and multi-scale modeling strategies that synergize the strengths of existing platforms. A pivotal direction is the development of composite, multi-tissue models (e.g., next-generation joint-on-a-chip systems) that integrate cartilage, synovium, bone, and neural/immune components within mechanically active, perfusion-enabled microenvironments to better simulate joint pathophysiology [19,95,98]. These systems should incorporate patient-derived cells and tissues to establish personalized models that capture the heterogeneity of OA subtypes, enhancing clinical relevance [90,98]. Moreover, future OoCs should also investigate the crosstalk between joint and other organs in the body. The establishment of body-on-a-chip (BoC) system can facilitate studies on how other organ systems, such as gut microbiome and central/peripheral system, affect OA development and drug responses. Furthermore, AI-enabled organoids hold considerable potential for advancing OA research through more precise disease modeling and personalized therapeutic exploration [134]. In addition, advances in gene editing and stem cell engineering will enable the creation of disease-specific organoids and transgenic models for more accurate pathophysiological recapitulation and mechanistic discovery [72,73,107]. The convergence of bioengineering, computational modeling, and translational science promises to yield human-relevant and predictive platforms. These next-generation models will be crucial for accelerating the discovery of effective therapeutics and validating them in a patient-specific context [88,89,133].

7. Conclusions

Increasing evidence indicates that OA, similar to other chronic non-communicable diseases, comprises multiple phenotypes and endotypes driven by distinct mechanical, metabolic, inflammatory, and aging-related pathways. Current experimental models of OA inevitably have limitations in reproducing the decades-long and fluctuating disease course observed in humans, where periods of progression, clinical exacerbation, and relative quiescence coexist. Interspecies differences in joint anatomy, physiology, and biomechanics further complicate direct translational comparisons with the human condition. Consequently, no single experimental model can accurately capture the full spectrum of human OA. Given the gap between existing research models and the natural progression of OA, investigators and clinicians must judiciously select models appropriate for preclinical studies and conduct precise clinical trials on specific OA subtypes. Notably, the progress in OA research models, especially OoCs and organoids, opens up a new avenue for translational research and new drug development. The FDA recently announced plans to phase out animal research in drug testing, making animal models no longer mandatory for preclinical drug testing [88]. Clinical trials for investigational therapies based on preclinical efficacy data from OoC results combined with existing safety data, without animal testing, are undergoing [135]. Future breakthroughs in BoCs and organoids will further unleash their great potential in drug development, paving the way for new strategies for OA research and treatment.

Author contributions

Jianxiong Shu: Writing - original draft, Investigation, Visualization.

Taiyuan Huang: Writing - original draft, Investigation.

Zhaoran Wu: Writing - original draft, Investigation.

Junliang Lu: Writing - original draft, Visualization.

Zhong Alan Li: Conceptualization, Funding acquisition, Writing - review & editing.

Changhai Ding: Conceptualization, Funding acquisition, Supervision, Writing - review & editing.

Yao Lu: Conceptualization, Funding acquisition, Project administration, Supervision, Writing - review & editing.

Funding

This study was supported by the National Key Research and Development Program of China (No. 2023YFE0209700), Guangdong Basic and Applied Basic Research foundation (No. 2024B1515020015), and the Mainland–Hong Kong Joint Funding Scheme (MHKJFS) of the Innovation and Technology Commission, Hong Kong S.A.R., China (to RST and ZAL, project #MHP/101/23).

Declaration of competing interest

None.