Osteoarthritis (OA) is a highly prevalent chronic disease that affects the joints and contributes significantly to a loss of physical function (Kloppenburg and Berenbaum, 2020). It is characterised by pain, sometimes linked to cartilage disintegration, bone remodelling and inflammation of the synovium. In the United Kingdom, the incidence of any type of OA in 2017 was 10.7%, with a 1-year consistent frequency of 6.8/10³ adults (≥20 years) and 40.5/10³ adults (≥45 years) (Swain et al., 2020; Allen et al., 2022). Pain associated disability as a result of OA represents a significant burden (22% of total ill health burden) in societal and personal costs (Versus Arthritis, 2019; Wolf et al., 2019). Internationally, hip and knee OA is the 11th highest cause of disability (GBD, 2017 Disease and Injury Incidence and Prevalence Collaborators, 2018). Knee OA (KOA) is the most prevalent type of OA and is increasing in incidence as obesity and life span increase (Wallace et al., 2017; Hunter and Bierma-Zeinstra, 2019). The latest musculoskeletal calculator (published in 2019) estimated that in 2012 18.2% of the English population over 45 years old had clinically diagnosed KOA, 1/3 of which was severe (Versus Arthritis, 2019).

KOA is historically categorised into two main types, primary OA- also known as idiopathic OA- which is characterised by natural cartilage degeneration during ageing as a result of unidentified reasons, and secondary OA which occurs as a result of known medical conditions or risk factors (Altman et al., 1986). However, with extensive research over the past 2 decades dedicated to understanding risk factors and aetiologies, primary OA has been further divided into genetic, aging and hormonal (estrogen-deficiency) subsets (Herrero-Beaumont et al., 2009). Secondary OA is classified into subsets like metabolic disorders, anatomical abnormalities, traumatic injury and inflammation (Arden and Nevitt, 2006).

Beyond these classic subsets centred on aetiological risk, recent studies propose use of clinical phenotype and endotype concepts to categorize this heterogeneous disease (Deveza and Loeser, 2018; Deveza et al., 2019; Mobasheri et al., 2019); a clinical phenotype refers to observable characteristics of an individual resulting from the interplay between genotype and environment (Mobasheri et al., 2019). Dell’Isola et al classified knee OA into six clinical phenotypes based on a qualitative meta-analysis of existing data from 24 studies: 1) long-lasting pain; 2) upregulation of inflammatory markers; 3) metabolic disorder (estrogen imbalance, dyslipidemia, diabetes, and obesity); 4) changes in bone and cartilage homeostasis; 5) altered joint mechanics (medial meniscus overload or varus malalignment) (Figure 1B); and 6) minor joint disease (slight clinical symptoms with chronic progression) (DellIsola et al., 2016). The complexity is confounded by these phenotypes not necessarily being discrete, with multiple phenotypes potentially co-existing (Bierma-Zeinstra and van Middelkoop, 2017). As a result, reaching a consensus on diagnosing each specific subgroup is challenging due to phenotype overlaps, meaning no specific biomarkers or tests can be uniquely applied. OA endotypes (or mechanistic phenotypes) are defined by distinct pathobiological molecular mechanisms and signaling pathways, making them identifiable by specific biomarkers (Anderson, 2008). This distinction is valuable for targeted anti-cytokine therapy. Inflammation, metabolism, cell senescence, and bone and cartilage endotypes are considered constituent molecular endotypes of each phenotype (Mobasheri et al., 2019).

KOA (Figure 1A) is insidious and is defined as the degeneration of the articular cartilage in the weight-bearing region, with structural changes of the surrounding bone, and joint inflammation, irrespective of the presence of clinical symptoms (Kraus et al., 2015; Lespasio et al., 2017) (Figure 1F). This joint degeneration is affected by a combination of mechanical and biological effects (Sandell and Aigner, 2001). Epidemiologically, OA risk factors can be categorised into individual-centred (or systemic) factors such as age, sex, race, genetic background, metabolic disorder, endocrine, diet, bone density, and joint-related factors such as different shapes of the articulating surfaces, muscle weakness, occupation/sports, and injury and joint overloading (Vina and Kwoh, 2018). Biomechanical factors are the predominant theme associated with these joint-related factors, and this is the focus of much recent KOA research (Glyn-Jones et al., 2015; van Tunen et al., 2016). There is also increasing evidence that OA onset is mechanically driven and the onset is strongly correlated with altered mechanical properties of subchondral bone (Day et al., 2001; Hayami et al., 2006).

Biomechanical factors play an important role in understanding the pathogenic mechanism of all forms of KOA and in helping to address clinical questions. For example, it is reported that frontal plane knee malalignment has the strongest relation with KOA development (Palazzo et al., 2016). The natural knee alignment is around 1° varus (Cooke et al., 2007) and the ideal realignment angle ranges from 1° to 13° valgus (Dugdale et al., 1992; Cooke et al., 2007; Tsukada and Wakui, 2017; Sethi et al., 2020). However, these data are all derived from retrospective studies, which are population-based and laboratory experiments to confirm these findings are lacking. Moreover, despite OA (Figure 1) being a progressive disease, symptomatic OA is usually only diagnosed in patients with clinical symptoms and imaging evidence at an advanced stage (Hayami et al., 2006), which is likely irreversible. Therefore, it is imperative to devote research efforts to detect biomechanical mechanisms at the initial and early stages of OA. This is key not only for studying disease aetiology, but would also represent a big step towards developing preventative treatment and in diagnosing pre-radiographic OA.

Experiments to investigate these biomechanical factors are difficult in human studies as the mechanical intervention and direct outcomes are hard to perform and observe independently (Wendler and Wehling, 2010), and the available human tissue diagnosed as OA is not in the early stage (Hayami et al., 2006). Ex vivo joint models cannot truly mimic musculoskeletal mechanics and the arthritic environment. Thus, in vivo experimental animal models are important in studying the onset and early stages of this disease, particularly when investigating biomechanical aspects (Kuyinu et al., 2016). The two main goals of using OA in vivo models are: 1) to understand the pathophysiology of the disease, especially in the early stages, and 2) to test the safety and efficacy of different therapeutics developed before entering into clinical trials (Kuyinu et al., 2016). Recent advances in measurement techniques for in vivo animal studies have enabled and improved the analysis of OA and reduced animal usage (Teeple et al., 2013). In the light of recent work on in vivo animal models and methodological advances on biomechanical aspects of KOA, the purpose herein is to conduct a narrative review on the use of preclinical in vivo models, and advanced measurement techniques focusing specifically on biomechanical aspects in the context of the identified human clinical sub-division of OA.

Biomechanical factors are defined herein as joint-related, OA-linked, features based on epidemiological statistics. These are subdivided into either those anatomical relationships influencing the joint, or those that are the consequence of functional joint usage (Glyn-Jones et al., 2015; van Tunen et al., 2016). The following section presents current evidence on biomechanical-related risk factors associated with OA.

Anatomical factors include those derived from individual joint morphology and limb alignment (Glyn-Jones et al., 2015), where these can be variations in the shape of the tibia, femur, or patella. The anatomy is the product of genetics and adaptation to use, and the reasons for the adaptive changes are not fully known, but are likely to align with Wolff’s law, which describes how bone tissue can be remodelled in response to prevailing mechanical environment (Chen et al., 2010). These morphological variations can, therefore, be created by habitual joint loading experiences, but will also then induce corresponding shifts in joint loading and so may initiate or accelerate OA (Neogi et al., 2013). The morphological variation can predict the onset of OA, a year earlier than visible radiographic changes (Neogi et al., 2013). Extensive variation, recognized as knee malalignment (i.e., varus or valgus alignment) is also a strong risk factor in OA onset and progression (Tanamas et al., 2009). It is defined as a shift from the collinear mechanical alignment of hip, knee and ankle to either a medialised (varus) or lateralised (valgus) loading distribution (Sharma et al., 2001). The alignment changes can be a result of localised morphological deviation. In the varus knee, the mechanical axis (from mid femoral head to mid ankle) will pass medial to the centre of the knee. This induces an adduction moment that increases the force on the medial tibiofemoral compartment; the opposite happens in the valgus knee (Tetsworth and Paley, 1994). This loading imbalance contributes to OA progression (Sharma et al., 2001), with a 1° varus deviation out of the natural physiological range of alignment, disproportionally increasing the medial load by 5% (Schipplein and Andriacchi, 1991; Halder et al., 2012), As such, varus and valgus malalignment (Figure 1B) is a known independent risk factor in the progression of medial and lateral OA (Tanamas et al., 2009). Of these, varus alignment is more common in the knee OA population and is significantly linked with OA development (Brouwer et al., 2007).

Functional factors that are a consequence of challenging joint usage include previous knee injury, poor muscle function, reduced proprioceptive acuity, activities (occupational activity and sports activity), and joint laxity (Bennell et al., 2012; Richmond et al., 2013; Glyn-Jones et al., 2015; van Tunen et al., 2016). Knee injury mainly includes knee ligament ruptures, sprains or meniscal tears (Thomas et al., 2017), which carry a high risk of post-traumatic OA (PTOA) onset compared to uninjured individuals (Thomas et al., 2017). There is some evidence of the high OA incidence in athletes being mainly due to joint injury (Bennell et al., 2012). Meniscal and anterior cruciate ligament (ACL) injuries each account for one-quarter of all knee injuries (Thomas et al., 2017). Ligament injuries can cause changes in joint mechanics, including translation and shear, leading to overload (Figure 1E) (Kernozek et al., 2013). Simply, meniscal tears decrease the contact area between cartilaginous surfaces thus increasing contact stresses during load transfer (Figure 1D) (Masouros et al., 2008). Muscle weakness (Figure 1C), including athrogenic muscle inhibition, was previously thought to be a secondary factor that results from progression of OA damage, but recent population-based evidence shows that muscle dysfunction can be a primary risk of OA initiation, preceding knee pain and muscle atrophy (Slemenda et al., 1997). Quadriceps weakness specifically is a better predictor of joint pain and joint space narrowing than radiographic evidence (Palmieri-Smith et al., 2010).

Impaired knee proprioception refers to a deteriorated accuracy in position and motion sense, caused by impaired articular mechanoreceptors (Knoop et al., 2011). Reduced proprioceptive accuracy has been found in OA patients and this causes a decrease in knee stabilization and joint movement coordination, both of which are directly related to mechanics, and has a significant association with knee pain (Knoop et al., 2011; van Tunen et al., 2018). These factors are linked, as the loss of muscle strength can be attributed to afferent sensory dysfunction (i.e., impaired proprioception) and the subsequent reduction in efferent neuron stimulation (Slemenda et al., 1997). Although there is currently no robust evidence that dysfunctional proprioception causes OA onset, it has been proposed as a putative OA risk factor (Roos et al., 2011). Activities (e.g., physical, occupational and sporting) are controversial factors for knee OA, as cartilage homeostasis is known to rely upon moderate loading but could change to catabolic metabolism in high intensity and long duration, abnormal mechanics (Griffin and Guilak, 2005). Heavy occupational lifting, frequent deep knee flexion (squatting, kneeling) and sports activity (at elite level) with significant impact, twisting, turning and running have a demonstrated higher OA risk (Richmond et al., 2013). In contrast, there is no evidence that physical activity (moderate sports) and recreational sports activity, without suffering injury, increase the risk of OA (Bennell et al., 2012; Richmond et al., 2013). Some large studies have found that ice hockey and soccer players, wrestlers and weightlifters have a much higher incidence of OA than endurance sport athletes (Kujala et al., 1994; Tveit et al., 2012). This aligns with the notion that there is a biomechanical tolerance level in weight-bearing joints; this level is not currently known (Thelin et al., 2006). However, there could simply be a higher incidence of injury in those sports in which higher biomechanical loads are engendered.

Joint laxity–loss of stability with deficiencies in the primary soft-tissue stabilisers such as the cruciate ligaments, secondary stabilisers such as the capsule, or in alignment–is another biomechanical-related OA risk factor (van Tunen et al., 2016). Evidence has shown that patients with medial knee OA always have laxity, usually in the varus-valgus direction (van Tunen et al., 2018). All of these factors are related to biomechanics (loading, motion) and in general, alterations in these factors will change knee joint loading (van Tunen et al., 2016). Acute or long-term overuse changes in loading will subsequently induce the structural, biological and mechanical level changes and present as clinical symptoms.

Although biomechanical risks can be independent, direct-causative factors in some cases of knee OA, all OA phenotypes have a mechanical component, and all risk factors can induce shifts in mechanical stress in chondrocyte pericellular matrix (Brandt et al., 2009; Zhao et al., 2020). For example, obesity is an important OA risk factor; not only because the systemic inflammation arising from adipokines can promote local joint inflammation (Berenbaum et al., 2013; Robinson et al., 2016), but because obesity can also interact with confounding biomechanical factors, evidenced by less frequent cyclic physiologic joint loading (lowering risk) and altered gait and increased peak knee force (which both increase risk) (Aaboe et al., 2011; Issa and Griffin, 2012). At the molecular and cellular level, cartilage degradation is generally affected by biological and mechanical stimuli (Sandell and Aigner, 2001) and many risk factors induce OA by combined inflammation and stress-induced mechanotransduction signaling pathways in chondrocytes (Guilak, 2011). Recent advances in our grasp of chondrocyte mechano-signaling make the mechanism by which biomechanical factors may contribute to OA clearer. It is now evident that the pericellular matrix of the cartilage chondrocyte transmits the physiological mechanical stimulus (including abnormal stress, compression, fluid pressure) to cell surface mechanoreceptors (ion channels, integrins) which convert these outside physical signals to intracellular signals, to regulate a series of downstream pathways (Guilak, 2011; Gilbert and Blain, 2018; Zhao et al., 2020). This suggests that this biomechanical stress may be critical in the initiation and progression of OA, independently of the influence of inflammation. Therefore, although there is tantalising information, there is still no consensus in the understanding of the biomechanical factors underlying OA initiation and progression (Guilak, 2011; Saxby and Lloyd, 2017; Hunt et al., 2020). Some experts propose that any efforts to subdivide OA should largely consider the abnormal mechanical stress loading on the joint, and advise subclassifying OA based on mechanical abnormalities (Radin et al., 1991; Deveza et al., 2019). This highlights the need to accurately define the various mechanical factors, and the abnormal mechanical stress, contributing to the different phenotypes of OA. To identify these factors and to address this fundamental gap of mechanical stress mechanism, preclinical models should be developed that are clinically representative of biomechanically induced OA (McCoy, 2015; Kuyinu et al., 2016; Serra and Soler, 2019).

As OA is a multifactorial disease driven by genetic, biological, and biomechanical factors (Glyn-Jones et al., 2015), there are several different in vivo models that provide a means to study its various distinct features. Each model replicates a unique OA aetiology and pathogenesis that reflects the specific mechanism of interest as well as the target uses for drug treatment (Cope et al., 2019). Generally, small mammals (nearly always rodents), including mice, rats, rabbits and guinea pigs, are considered for primary basic research (Serra and Soler, 2019), such as the study of OA pathology/pathophysiology, and assessing the effectiveness of different therapeutics (Kuyinu et al., 2016). Larger animals (dog, horse, goat, cattle) develop the disease more slowly, are more expensive, less reproducible and harder to handle (Serra and Soler, 2019), yet can be more relevant to disease heterogeneity and the anatomy, dimensions and biomechanics of human joints (Lampropoulou-Adamidou et al., 2014; Meeson et al., 2019b). Therefore, they are more widely used in translational preclinical studies, where the clinical progression of the disease and treatment are being considered.

These in vivo OA models are characteristically classified as: spontaneous, which include naturally occurring for uncertain/unknown reasons and genetically modified models, and induced models which can be provoked by chemical or surgical intervention (Table 1). Recently, some non-invasive induced models have been developed as alternatives to surgical models (Christiansen et al., 2015; Poulet, 2016). These OA models are typically categorised in terms of classical primary and secondary OA types. Spontaneous models characterise to a certain extent the natural progression of human primary OA with or without genetic modification. Accordingly, induced models are developed as a result of replicating specific known risk factors of secondary OA and these models are usually surgically induced, and sometimes less appropriately chemically induced (Vincent et al., 2012; McCoy, 2015). The induced methods can stimulate acute inflammation and/or alter the joint mechanics (Serra and Soler, 2019).

The Dunkin Hartley guinea pig is the most used spontaneous model to develop naturally occurring, age-related OA. This has a high incidence and occurs at a reasonably young age compared to humans (Jimenez et al., 1997). It has similar histopathology to the human disease and the histological changes can normally be observed at 3 months of age (Kraus et al., 2010). The Str/ort mouse, derived from selective breeding with susceptibility to OA, is a well-recognized spontaneous OA model where >85% of all male mice have very mild OA lesions in the medial tibial plateau from 10 weeks of age (Mason et al., 2001; Staines et al., 2017a). Similarly, OA can arise spontaneously in rabbits, where natural disease can be found in >50% by the sixth year of age (Arzi et al., 2012). Spontaneous dog OA has also been proposed as a model, as OA can have high prevalence (>20%) over 1-year of age, and dogs have more human-like anatomy and OA heterogeneity than rodents (Meeson et al., 2019a). Genetically modified models, mainly in mice, are designed to knock down, knockout, knock-in or mutate specific genes to breed strains with modified OA susceptibility (Little and Hunter, 2013).

The induced methods aim to generate joint destabilization and increase joint contact force, and/or create an intra-articular inflammation, and so consequently alter cell metabolism and induce an OA lesion (Teeple et al., 2013). Chemically induced models mostly include intra-articular injection of an agent to trigger acute local inflammation, extracellular matrix degradation and chondrocyte death (Bendele, 2001). The chemical agents generally used in induced animal models of ‘OA-like’ joint pain are papain (Sukur et al., 2016), quinolone (Rodrigues-Neto et al., 2016), collagenase (Lee et al., 2020), carrageenan (Korotkyi et al., 2019), Freund’s adjuvant (Robin, 2017) and, most commonly, sodium mono-iodoacetate (MIA) (Adeyemi and Olayaki, 2017). Surgically induced models mainly use intra-articular invasive surgery to generate joint instability or create chondral defects to induce OA. PTOA is the most widely studied secondary OA model, commonly used to analyse mechanical aspects of OA. Currently, the most often used surgical models are destabilization of the medial meniscus (Arzi et al., 2012)), anterior cruciate ligament transection (ACLT; (Lampropoulou-Adamidou et al., 2014), collateral ligament transection (Bendele, 2001) other meniscal injury (meniscectomy (Ali et al., 2018) or medial meniscal tear/transection (Brederson et al., 2018; Pucha et al., 2020; Temp et al., 2020)), and chondral/osteochondral defects (Mastbergen et al., 2006; Flanigan et al., 2010; McNulty et al., 2012; Figueroa et al., 2014; Serra et al., 2014), or ovariectomy (oestrogen deficiency causing subchondral osteoblast changes) (Xu et al., 2019). Additionally, a novel non-articular invasive method was raised again recently, using tibial osteotomy (TO) to induce a varus tibial malalignment and joint mechanical overload (Britzman et al., 2018); this is more representative of the human primary chronic overloading OA condition, or a means of replicating the susceptibility to OA based on anatomical deviation, without any intra-articular damage or destabilization of internal joint mechanics (Britzman et al., 2018).

Although surgical models better mimic the pathogenic mechanisms than chemical models and have a shorter progression period than spontaneous models, they involve aseptic procedures, and the results are highly dependent on surgical skill. This means that ensuring repeatability and minimising variability of these invasive models is difficult. To address these issues, some studies create PTOA by using non-invasive methods through applying an external mechanical load to the relevant joint without disrupting the skin or the joint capsule (Christiansen et al., 2015) through either single-impact injury or repetitive loading (Poulet, 2016). There are five main types of non-surgically induced model: 1) intra-articular tibial plateau fracture (Lewis et al., 2011; Schenker et al., 2014; Kimmerling et al., 2015); 2) ACL rupture through tibial compression overload (Lockwood et al., 2014; Khorasani et al., 2015); 3) knee immobilization in an extended position (Videman, 1982) 4) cyclic articular cartilage tibial compression (Poulet et al., 2011; Melville et al., 2015); and 5) moderate running exercise. Compared to severe surgical PTOA models, these models are more likely to represent real biomechanical conditions in OA development.

The OA model classifications previously mentioned (spontaneous and induced) are rooted in the conventional view (primary and secondary OA) that distinguishes various aetiologies (intervention methods) (Table 1) (Esdaille et al., 2022). Notably, most pharmacotherapies derived from these animal models are symptomatic, primarily addressing the clinical phase of the disease rather than the pre-arthritis phase. A recent review article analyzed gene expression data across different OA models, emphasizing how discrepancies between models can lead to divergent conclusions regarding targeted intervention therapies (Soul et al., 2020). Therefore, it is crucial to clarify the commonalities among these models and understand their relationship to each etiopathogenesis and the six distinct OA clinical classifications currently adopted. A recent systematic review defined these clinical phenotypes as those related to: 1. chronic pain, 2. inflammation, 3. metabolic syndrome, 4. bone and cartilage homeostasis, 5. mechanical overload, and 6. minor joint disease phenotypes (DellIsola et al., 2016). This section summarises how these animal models relate to the clinical classifications.

Types 1 and 2. Responses to chronic pain and inflammation are usually conducted using chemically induced models, which induce cartilage degeneration based on triggering the inflammatory mechanism rapidly (Serra and Soler, 2019). The MIA method induces widespread cell death across many joint tissues leading to intense pain and is typically used to study ‘joint’ pain (Pitcher et al., 2016). These models should not be used to study the pathogenic mechanism of OA in humans (Vincent et al., 2012).

Type 3. Metabolic OA phenotype is usually characterized by surgical (ovariectomy) or spontaneous OA models. These models are appropriate to study diabetes, obesity and other metabolic disturbances (estrogen imbalance) causing OA (Xu et al., 2019).

Type 4. Bone and cartilage homeostasis phenotype presents an OA subgroup with significant changes in cartilage and bone metabolism (DellIsola et al., 2016), categorised to atrophic (few osteophytes with severe joint space narrowing) and hypertrophic (large osteophytes with little joint space narrowing) subsets (Roemer et al., 2012). The ACLT surgical model is usually used to study this phenotype, but the cyclic articular cartilage tibial compression models are also used. Bone osteophytes can be rapidly induced in experimental canine OA as early as a few days after ACL surgical induction (Gilbertson, 1975), and the subchondral and trabecular bone remodeling can be observed at around 2 months (Lavigne et al., 2005).

Type 5. The mechanical overloading phenotype which accounts for up to 22% of incidence (DellIsola et al., 2016), is usually modelled by surgical disruption of joint biomechanics, such as meniscectomy or ligament transection, causing joint instability. These models can simulate a human injury event: so-called PTOA models (Little and Hunter, 2013). However, as shown in a population study, PTOA comprises only approximately 12% of symptomatic OA (Brown et al., 2006). Therefore, it could be contended that there is an overuse of the PTOA models in representing the chronic overloading OA phenotype.

Type 6. Ageing or minor joint disease phenotypes are best studied by naturally occurring models (Lampropoulou-Adamidou et al., 2014).

The mechanical overloading phenotype itself has been shown to arise from 1) long-term overuse or cyclic small loads or 2) a short-term high load (impact) (Gardiner et al., 2016) and these two are suggested to have distinct pathophysiology progression (Little and Hunter, 2013). For example, a comparative study developed in guinea pigs with or without ACLT surgical intervention showed that there are several differences in expression of OA biomarkers between these two mechanical overload phenotypes (Wei et al., 2010; Rice et al., 2020). A later study characterised the subgroups of the mechanical overload phenotype into two: one having severe medial OA and varus malalignment with a high prevalence of injury history (55%), and the other having evidence of lateral OA and valgus alignment with lower body mass index (BMI) (DellIsola et al., 2016). Therefore, current PTOA models only represent the short impact (injury-history) subgroup, and there should be another type of biomechanically induced OA model which occurs in a relatively idiopathic, non-traumatic and slow-progressive condition to present the chronic overloading OA that is considered responsible for primary OA (DellIsola et al., 2016). The studies more representative of this form of chronic mechanical overloading OA are the non-invasive cyclic overloading model (Ko et al., 2013) and the extra-articular TO surgical model (Britzman et al., 2018). These will be described in more detail in section 4.2.

As the PTOA surgical model cannot cover all the mechanical overloading phenotypes, here, we define a new term, biomechanically induced animal models (biomechanical model for short) to present all the mechanical overloading phenotypes of OA, including those that mimic ‘primary’ mechanically induced slow-progressive OA associated with chronic overloading which is naturally occurring after long-term lower levels of increased loading (mechanical overuse), and ‘secondary’ PTOA that develops in response to acute high impact overloading or trauma. Unlike genetically modified or chemically induced models, which primarily focus on altering genetic expressions or chemical interactions within the joint, biomechanically induced models specifically recreate the mechanical conditions contributing to OA development. The main underpinning rationale is to alter the joint mechanical challenges based on changing the loading distribution, or creating joint instability or joint defects. The following section will focus on these biomechanical aspects identified as key in animal models of biomechanical mechanisms in OA development.

This section summarises recent advances in the development of mechanical overloading models. It also highlights that there remains a gap in ensuring that these mimic with sufficient precision the clinical condition in terms of the biomechanical aspects. We hypothesise that there are two key considerations when studying mechanically-driven OA, which are: 1) how to develop a mechanical model without inflammatory sequelae, and 2) how to induce OA without significant and immediate experimental trauma. Appropriate tackling of these concerns will enable the discrimination of changes indicative of early-stage OA in both the presence and absence of such potential injury-related onset, enabling comprehensive identification of mechanisms underpinning slow-progressive disease.

There are three key diagnostic tools for biomechanically induced OA: biochemical, biomechanical and imaging markers to measure the biological, functional and structural OA changes (Andriacchi, 2012). Typically, microscopic tools (biochemical) are usually used in small animals, and macroscopic tools (biomechanical and imaging) are used in large animals and humans (Kuyinu et al., 2016). Characterizing biomechanical models and outcomes relies on measuring biomechanical factors, but biomechanical changes are very subtle in animal models, in particular in rodents due to their size and quadrupedalism, and a single mechanical factor might respond to both mechanical and inflammatory stimuli and cause multiple marker changes. This subtlety hinders early detection and characterization of this mechanically driven disease in animal models. Therefore, advanced diagnostic tools are needed for animal models in order to improve our understanding of this disease.

Knee OA is a multifactorial complex disease which is not yet fully understood. As a mechanically-driven or mediated disease, biomechanical factors are involved in all OA phenotypes. The aim of this review was to give a narrative description of preclinical in vivo models and advanced measurement tools to address these biomechanical factors in the context of clinical phenotypes. We hypothesised that there are two key considerations when studying mechanically driven OA, which are: 1) how to develop a mechanical model without inflammatory sequelae and 2) how to induce OA without significant experimental trauma and so enable the detection of changes indicative of early-stage OA in the absence of such sequelae.

We first summarised the evidence for the association between biomechanical factors with knee OA, including anatomical factors (joint morphology and limb alignment) and functional factors (knee injury, poor muscle function, reduced proprioceptive acuity, activities, and joint laxity). KOA has been categorised into two phenotypes (primary and secondary OA) and several subgroups based on aetiology and risk factors, yet they are frequently overlapping and OA in many individuals likely reflects a combination of many risk factors. Cohort case studies and meta-analyses have helped to narrow these down into six phenotypes based on the clinical characteristics (DellIsola et al., 2016). As abnormal intra-articular stress is thought to be a major determinant and key feature of OA pathology in all clinical phenotypes initiated by multiple factors, the mechanical mechanism has an overwhelming importance in OA pathophysiology and treatment (Brandt et al., 2009), suggesting that all OA might share a common final pathway linking the mechanical and biological process to cause joint lesions. Treating this single causative factor (mechanical stress) might prevent the disease process in all OA phenotypes. Thus, the focus is to identify broad biomechanical principles, but more work needs to be done to characterize the effect of single biomechanical factors on OA progression when considering the complex interaction between mechanical and biological factors.

Animal models may play a critical role in identifying biomechanical factors in OA which currently still rely on results from retrospective cohort studies or theoretical biomechanical studies. A major limitation for OA model research is that the OA resulting from different aetiologies are often lumped together within the same model despite the significant heterogeneity. As the disease characteristics of OA differ between phenotypes and the phenotype-specific treatment is needed, so animal models must take these into account.

In this review we build on recent work that classifies OA into six phenotypes, of which the mechanical overloading phenotype accounts for the highest proportion of OA incidences (DellIsola et al., 2016). We propose in this article that this overloading phenotype include distinct subcategories of the PTOA subtype that occurs after trauma and, separately, a new ‘primary’ chronic subtype that represents mechanical overloading phenotype after long-term knee overuse without known major injuries. Recent studies that compare post-traumatic OA and non-traumatic OA would seem to support this proposal. Some cohort studies have indeed found that radiological structural changes in the medial and lateral compartments are equally distributed in PTOA patients but are primarily in the medial compartment in non-traumatic patient (Swärd et al., 2010). Also, non-traumatic OA patients have higher quadriceps and lateral hamstring electromyography, and higher knee adduction angles and moments compared to PTOA (Robbins et al., 2019).

Accordingly, in vivo OA models need to be subcategorised and established to explicitly represent certain clinical phenotypes. Our review classifies the main models in the literature in the context of the clinical phenotypes. The comparative analysis of OA models in Table 1 underscores differences in utility for disease studies and the fidelity with which they replicate the OA process. We found that there are advances in modelling the PTOA phenotype through, for example, low impact arthroscopic insults, non-invasive ACL rupture or DMM. These have proven successful in reproducing subchondral bone changes that indicate the early stage of OA. Although there are models with external cyclic compression, extra-articular osteotomy and running exercise, a primary chronic overloading phenotype is, however, less well-modelled. Whilst invasive models are intuitively closer to PTOA and non-invasive models may be closer to the primary chronic overloading phenotype, this direct correspondence should not be assumed. Full analysis, including measures of inflammatory and biomechanical markers needs to be conducted to test these potential assumptions. If models could be developed that minimised or even stopped an acute inflammation stage and so slowed down the OA progression, then this would enable early-stage OA to be detected and analysed. Moreover, ensuring repeatability and reliability to minimize variability for these biomechanical models (both invasive and non-invasive) remains a challenge. Although these models more accurately replicate clinical conditions and thereby guide more relevant therapeutic developments, they often lack the rigorous standardization seen in chemically-induced models, which are preferred for their speed and cost-effectiveness in preliminary drug tests but fail to accurately reflect the disease’s true pathogenic processes or the underlying joint damage. Consequently, standardization is particularly vital for the widespread adoption of biomechanical models. There is a critical need for standardized protocols to enhance reproducibility across studies and boost the translational potential of research findings from biomechanical preclinical models into different subtypes of human OA.

Furthermore, for an animal model to succeed, well-defined variables and outcomes are crucial. These could include biochemical and imaging markers, mechanical measures and histological data. Advanced genetic and proteomic data can also be used for more precise phenotyping (endotypes) of these models. Lack of reliable diagnostic tools poses another concern in OA model development. This might partly explain the current poor translation from preclinical animal studies to human practice where diagnostics are entirely reliant on the presence of symptom and radiology evidence, and regulators follow this approach. Our review found that detailed quantitative biomechanical and imaging tools are not routinely used and no studies have used all of the most advanced tools simultaneously. There use in combination with other bio(chemical) markers would enable a full understanding of OA instigation and progression. Such an integrative multiscale experimental and computational framework has been applied in some large animal and human studies, and the application of this approach to small animals would facilitate many benefits due to cost, availability and genetic manipulation.

This review has identified a gap in the description of clinical OA phenotypes when applied to in vivo studies, particularly with reference to the non-traumatic chronic overloading phenotype and so we propose that the overloading phenotype include distinct subcategories of the PTOA subtype that occurs after trauma and, separately, a new ‘primary’ chronic overloading subtype that represents mechanical overloading OA phenotype after long-term knee overuse without known major injuries. This new ‘primary’ chronic overloading subtype is less well-modelled in the literature and we recommend enhanced efforts to address this. However, although invasive models are intuitively closer to PTOA and non-invasive models may be closer to the primary chronic overloading phenotype, this direct correspondence should not be assumed. Furthermore, alignment between OA onset and progression mechanisms in this ‘primary’ chronic overloading subtype and those contributing to the OA that arises in spontaneous animal models will better define their utility in translational studies.

Advances in biochemical, biomechanical and imaging biomarkers, combined with the opportunities that such new small animal models provide would enable the better development of early diagnosis of the most prevalent form of knee OA.