Models of osteoarthritis: the good, the bad and the promising

Summary Osteoarthritis (OA) is a chronic degenerative disease of diarthrodial joints most commonly affecting people over the age of forty. The causes of OA are still unknown and there is much debate in the literature as to the exact sequence of events that trigger the onset of the heterogeneous disease we recognise as OA. There is currently no consensus model for OA that naturally reflects human disease. Existing ex-vivo models do not incorporate the important inter-tissue communication between joint components required for disease progression and differences in size, anatomy, histology and biomechanics between different animal models makes translation to the human model very difficult. This narrative review highlights the advantages and disadvantages of the current models used to study OA. It discusses the challenges of producing a more reliable OA-model and proposes a direction for the development of a consensus model that reflects the natural environment of human OA. We suggest that a human osteochondral plug-based model may overcome many of the fundamental limitations associated with animal and in-vitro models based on isolated cells. Such a model will also provide a platform for the development and testing of targeted treatment and validation of novel OA markers directly on human tissues.


Introduction
Osteoarthritis (OA) is a chronic degenerative disease of diarthrodial joints, predominantly affecting the spine and peripheral joints of the body, particularly the hands, hips, knees and feet. OA most commonly affects people over the age of forty, with the risk of disease increasing with age. OA is a complex heterogeneous disease with different clinical and biochemical phenotypes.
The cause(s) of OA are unknown, and many studies have suggested that the pathobiology of OA is far more complex than a simple cartilaginous or bone disease. It is now acknowledged that OA affects many joint structures, including degeneration of cartilage, abnormal bone remodelling and synovial inflammation 1,2 . Also, studies have shown that there is a complex interplay between the different joint components, making understanding of the degradative sequence of events involved in OA pathogenesis very difficult to dissect 3e5 .
The initial onset of OA disease is considered due to an imbalance between the cartilage degradation and repair process 6,7 . The exact sequence of events that trigger the onset of the disease is however widely debated throughout the literature. One hypothesis, suggests that secretion of pro-inflammatory cytokines into the synovial joint induces matrix metalloproteinases which cause the fragmentation and degradation of cartilage extracellular matrix leading to bone remodelling and synovitis 8e10 . Contrary to this theory, some studies suggest that subchondral bone remodelling and synovitis precede articular degeneration in the early stages of OA 8,11,12 . While other studies suggest that meniscal degeneration evolving through fibrillation of tissue and a decrease in the levels of type I and II collagen within the meniscus, act as a predisposing or contributing factor to OA progression 13,14 . In the later stages of OA, formation of subchondral cysts, subchondral sclerosis and osteophytes occur as a direct result of bone remodelling, cartilage degradation and synovitis 15e17 .
Treatment of OA is largely symptomatic due to insufficient understanding of aetiopathogenesis hindering the development of suitable disease-modifying drugs. This makes targeted treatment of OA a distinct challenge. Human OA tissue samples are usually collected for research once end stages of the disease have been reached, for example during joint replacement, by which time destructive changes in the joint are well established. This makes studying the early disease process very challenging 18 . OA pathology, particularly early OA, is therefore very difficult to study, and so researchers turn to in-vivo and ex-vivo preclinical animal models to investigate early pathological changes in OA. These models offer unique advantages as well as limitations for studying human OA. This article will review the different models used for -3D cell culture can be used to study the effects of cytokine stimulation and osmotic pressure, as well as the effects of physical injury and loading on tissue 23 -A matrix structure of collagens and proteoglycans favours phenotypically normal cartilage 28 Explant based models -Simple, cheap and easy to produce 23 -Explant models allow for the natural processes that occur within the extracellular matrix environment to be observed 26 -Cell death often occurs at the explant edge -Only a limited number of cells can be extracted from a single source -Limited tissue availability and significant inter-experimental variability 23 -Explant based models can be used to study the effects of cytokine stimulation and osmotic pressure, as well as the effects of physical injury and biomechanical loading on tissue 23,28 -Synovial tissue explants useful to study the role of the synovium in OA -Genetic models: PAR2 À/À , CD4 À/À , MMP17 À/À , Tenascin C À/-, Ddr2 À/À , SulPhatase À/-1/2, Syndecan 4 À/À , Fgf2 À/À , Mmp13 À/À , Hif2a +/-, GDF5 +/À , Osteopontin, Ptges1, Tnfrsf11b +/-, Runx 2 +/À , ADAMTS-5/4 À/À , Adamts5 À/À , ADAMTS4 À/À , MMP3 À/À , ICE À/-, IL-1b À/-, iNOS À/À35 -Transgenic models 2

Current models used in OA research
OA research models can be categorised into either ex-vivo or invivo models. Depending on the research question, different models can be used to address different aspects of OA development and progression. Each model has its advantages, yet it has become clear that no single model provides the opportunity to study the disease as a whole. The different models currently used in OA research are discussed below.

Ex-vivo models
Ex-vivo models can be categorised into monolayer culture, coculture, three-dimensional (3D) culture and explant-based culture. Each model has its advantages and disadvantages and so can be used to answer different questions in OA research.  identify due to the small anatomical size of mice 81 -The progression and process of disease is faster in mice than in humans (weeks rather than decades) 36 -The small size of mice makes surgically inducing OA more challenging 40 -Postoperative management of mice is difficult in surgically induced models 40,45 Rat -Rat cartilage is thicker than that of mice, so it is possible to induce partial and full-thickness cartilage defects 1,45 -Rats rarely experience post-operative infection so are useful animal models to surgically induce OA 1 -Rats are easily managed and require low maintenance costs 40,45 -It is easier to perform surgery in rats than in mice due to their larger size 40 -The full rat genome is available for study 40 -MMT, MCL transection and iodoacetate models useful to study pain 40,45 -Rat models useful in toxicology testing and studying cartilage restoration techniques 1,45 -Naturally occurring OA is uncommon in rats, variation in results is often observed between different strains of rat and disease severity varies with age, with older rats tending to present with more severe OA 18 -It is difficult to ascertain the skeletal maturity of rats 83 -Rats have greater volumes of highly vascularised adipose tissue and muscle in the medial knee region -Post-operative rats immediately resume load-bearing which accelerates joint degeneration 1 -Genetically engineered rat models are not available and postoperative management of rats is challenging 45 Guinea Pig -The guinea pig model has similar OA histopathology to disease in humans 84 -Guinea pigs are large enough that tissue samples can be easily collected for tests and the whole joint can be histologically sectioned 49 -Guinea pigs are easy to manage 40 -Naturally occurring guinea pig models are available and the disease pathogenesis is predictable and similar to that seen in humans 1,45 -Hartley guinea pigs can be used to study risk factors for OA such as BMI and age -Complete guinea pig genome available 85 -The weight of each guinea pig and whether they are housed alone or in pairs influences the severity of their OA 41,49 -Unlike in humans, guinea pigs resume load bearing postoperatively which accelerates joint degeneration 1 -The time to guinea pig skeletal maturity is fast 45 Cat -Cats are larger in size allowing for tissue and fluid collection 40 -The full cat genome is available 40 -Cats are difficult and costly to manage and there are ethical issues surrounding emotional attachment 40 -Cats display genetic variability between individuals 40 Rabbit -Naturally occurring OA is very common in rabbits 52 -Rabbit model useful in studying the efficacy of compounds 45 -Complete rabbit genome available 86 -Rabbits have a very different gait compared to humans and only rabbits over the age of eight or 9 months can be used to guarantee skeletal maturity 52 -The cartilage of rabbits is ten times thinner compared to humans, with a higher chondrocyte density and cartilage zonal layers that varies highly within the same joint 87,88 -The rabbit meniscus is more cellular, has less vascular penetration and can heal faster than the human menisci 89 -Rabbit cartilage can spontaneously heal and regenerate and there is no complete rabbit genome available for study 40  Caprine -Anatomically the caprine stifle joint is very similar to the human knee 64 -The caprine stifle joint is closest in size to the human knee joint, the larger size of the animal allows for tissue and fluid collection and goat cartilage thickness is close to that of humans 40 -Goats are cheap and easy to use in studies compared to most large animal models and they can be used to study cartilage repair 45 -Complete goat genome available 91 -Caprine cartilage thickness varies between individuals, the skeletal maturity of a goat is not reached until at least 2 years of age and cartilage healing capacity varies with a goat's age, with better capacity in younger animals 87,92 -Cartilage repair outcomes differ in the short and long term and so follow up is required to assess progress 83 -Naturally occurring OA in goats is uncommon 40 Ex-vivo models such as monolayer culture and co-culture are easier and cheaper to produce than 3D cell cultures and explantbased models. Monolayer cultures are also easy to produce on a large scale and avoid the challenges associated with culturing different cell types at different conditions. However, monolayer and co-cultures are limited in their use due to the fact that they isolate only one or two tissue components at a time. Many studies have shown that there is a strong interplaying network of communication between different joint components that help regulate and maintain a healthy joint, and so isolation of specific joint components hinders this communication 3,19,20 . For example, healthy articular cartilage is dependent upon the release of soluble factors by subchondral bone, and interactions between chondrocytes and synovial fluid ensures the flow of growth factors, regulatory peptides and nutrients between them 19,20 . When injured cartilage is co-cultured with synovium, a protective effect is produced on the synoviocytes 21 . Similarly, culture of subchondral bone and cartilage separately results in increased chondrocyte death and cartilage degradation as well as decreased protein content in culture media compared to when cultured together 3,19,22 . Explant models and 3D cell cultures allow for this inter-tissue communication and so are arguably more useful models available to OA researchers to reproduce natural in-vivo environments. Despite this, these models are more difficult to produce in terms of tissue volume and maintaining cell viability over extended periods of time. Some of the advantages, disadvantages and applications of various ex-vivo models used in OA research are summarised in Table I.

In-vivo models
Many animal models in at least eighteen different species have been developed to study established pathological features of OA such as pain, synovitis, cartilage degeneration and bone remodelling. Animal models used in OA research (see Table II) can be categorised into either induced or spontaneous models. Induced models refer to models where OA disease (or OA like features) have been induced either chemically or surgically. On the other hand, spontaneous models are subcategorised into naturally occurring and genetically modified models that develop OA.
Smaller animal models of OA such as mice, rats, rabbits and guinea pigs are much easier, quicker, cheaper and more readily available than larger animal models such as horses, pigs and dogs 2,40 . Smaller animals can be handled and housed with greater ease than larger models, but due to their smaller size, tissue samples extracted are much smaller and therefore tend to differ to a greater extent in their anatomical and histological structure when compared to humans 18 . Larger animal models therefore provide many advantages over the use of smaller animal models in terms of their greater anatomical similarity to the human model. A dog's articular cartilage for example is half the thickness of a humans, whereas that of a mouse is a minimum of 70 times thinner 2,18 . Additionally, a wider range of tests can be performed on larger animals, such as repeated synovial fluid collection and imaging. They also have a longer life span allowing for slower disease progression and time to end stage OA, as seen in humans. Whilst slow progressive models most accurately reflect human OA, they are however more expensive and time consuming to conduct. There are also greater ethical considerations around the use of larger animal models such as non-human primates and canines 41 . Based on this, some animal models are therefore better suited to OA research than others, such as the canine, caprine, bovine and porcine models. Some of the advantages and disadvantages of various animal species used in OA research are summarised in Table III.

Challenges presented by current OA models
At present, there is no gold standard animal model used in OA research. Differences in size, anatomy, histology (specifically Non-human primates -Non-human primates have similar anatomy, genetics, biology, behaviour and physiology to humans 2 -The pathology of OA and the relationship between age and disease severity is very similar to in humans 18,76,77 -The larger size of non-human primates allows for tissue and fluid collection and the full primate genome is available for some species 40 -Non-human primates are expensive and ethically difficult to keep, for example chimpanzees display depression and post-traumatic stress disorder on a similar scale to that of humans 97 -Non-human primates have a long-life span and so a long disease pathogenesis time scale which is both time consuming and costly 2 -There are difficulties in obtaining adequate subject numbers for studies 1 -Housing and management of non-human primates is challenging 40 cartilage thickness) biomechanics and physiology makes translatability between animal models and human disease very difficult 52,83 . Challenges are posed by species-specific differences in disease pathology and progression, as well as normal joint homeostasis, specifically the repair processes that occur within the joint. Different OA induction methods used in certain animal species also sometimes results in differences in OA presentation. There is therefore a need to reduce experimental variability and increase the reliability of data interpretation. Furthermore, different animal models have been shown to represent different stages of the disease more effectively, making selection of an animal model that completely reflects natural human disease challenging. To add to this, it has become clear that there is much dispute as to what defines OA and which molecules are associated with the disease. This is in part hindered by the fact that current ex-vivo models do not allow for the important inter-tissue communication between different joint components required for natural OA disease processes to be studied. To gain a better understanding of the mechanisms of joint damage in OA, specifically the exact sequence of events and interactions between different joint components, it has been suggested that more focus should be placed on developing 3D cell cultures and explant-based models that allow for these important interactions, providing interesting opportunities for researchers to develop their understanding of OA.
In the ideal animal model, the disease must be induced reliably, with 100% penetrance, and within a suitable time frame, and yet still present with disease characteristics that are comparable to the human condition. Disease progression in the animal model should also allow for the examination of all stages of disease to ensure full detection of any therapeutic effects. The animal must be inexpensive, easy to house and manage but also be of large enough size to allow for a full range of analysis techniques to be performed. The animal must also be anatomically, biomechanically and histologically similar to humans 40 . Some animal species, such as the canine, caprine, bovine and porcine models, are therefore better suited to OA research than others.

A promising model for OA
OA is a disease of the whole joint and therefore the gold standard model for human OA must allow for key communication between different tissues of the joint. An osteochondral (cartilage on bone) model may overcome many of the current challenges and limitations of the various models discussed above. The use of osteochondral plugs provides a model incorporating the key joint tissues affected in OA, maintaining the important interactions between these tissues as seen in human disease. There are a few studies that have used osteochondral plugs from bovine, equine and human samples as ex-vivo models of OA 98e100 . These osteochondral plugbased models appear very promising and so further studies should be encouraged as the basis for developing a gold standard model for OA. In the model design, cytokines such as interleukin-1 beta (IL-1b) and tumour necrosis factor alpha (TNF-a) may be used as the best method for inducing OA (cartilage damage) and tissue responses that closely replicate the natural disease, as these cytokines are known to contribute to the inflammatory effect of the synovium in the model. However, OA is a complex disease and many cytokines and chemokines have been shown to be expressed in OA synovium and detected in synovial fluid. Therefore, in the model design, investigators should consider using synovial tissue and/or synovial fluid from patients with active disease to induce OA in the osteochondral plugs. Osteochondral plugs can be harvested from joint surfaces, such as the femoral condyle, tibial plateau and patella. Methods of plug extraction available for use include different sizes of graft harvester, biopsy punch, mosaicplasty osteotome, diamond tipped cylindrical cutter or surgical trephine burr. Plugs can be cultured in serum-free culture medium such as Dulbecco's Modified Eagle Medium F-12 (DMEMF-12, Invitrogen, USA) or a-Minimum Essential Medium (a-MEM, 22,561 Gibco, The Netherlands) for up to 57 days with significant cell viability 98 . Indeed, an early study reported that >99% chondrocyte viability can be maintained in the untraumatized areas at the centre of the osteochondral plugs 100 .
The availability of an osteochondral plug-based model, particularly a human tissue-based one, would be invaluable in screening of new disease-modifying osteoarthritis drugs (DMOADs). Currently, there are many DMOADs under different stages of development. Early studies of these drugs were in animal models, but the availability of this new model will provide an opportunity to directly test these drugs on human tissues. An osteochondral plug system like this may also be used for discovery of novel markers for OA.

Conclusion
It is clear that we are limited in our understanding of OA because we do not have a suitable model that accurately reflects natural human OA. Whilst animal models provide crucial information about disease mechanisms, none of the current models used in OA research recreate the natural in-vivo environment and allow the whole disease process to be studied. Differences in anatomy and biomechanics also makes translatability to the human model a distinct challenge. It is also important to consider the cost and ease of management of using animal models in research. Whilst smaller animal models provide many benefits in terms of availability, handling and management, larger animal models such as canines and pigs are not only more comparable to humans physiologically but also in their progression to disease. Their larger size also allows for performance of a broader range of analysis techniques and investigations.
The models currently used in OA research each have their advantages and disadvantages; however, it has become clear that there are consistent problems with all of these models that hinders our ability to understand the pathogenesis of OA. The only way to achieve greater understanding of the pathological processes that underpin OA is to produce a 'gold standard' model for OA. Development of a consensus model will provide greater understanding of the specific stages and interactions involved in OA pathogenesis, as well as a model that can be used to compare data findings between different research groups, test pre-clinical drugs and identify and test possible biomarker targets directly on OA joint tissue. An osteochondral plug-based model could be a "promising" new model for OA, able to provide a reliable throughput model for proof of concept and mechanistic studies, aiding the discovery of targeted OA therapy. The model would also provide an opportunity to reduce the financial, ethical and time restraints associated with using animals in research, shifting OA research to embody the principle of the 3Rs; replacement, reduction and refinement of animal use in research.