Initiating factors for the onset of OA: A systematic review of animal bone and cartilage pathology in OA

Abstract There is controversy over whether bone or cartilage is primarily involved in osteoarthritis (OA) pathogenesis; this is important for targeting early interventions. We explored evidence from animal models of knee OA by preforming a systematic review of PubMed, Scopus, and Web of Science for original articles reporting subchondral bone and cartilage pathology in animal models with epiphyseal closure. Extracted data included: method of induction; animal model; cartilage and bone assessment and method; meniscal assessment; skeletal maturity; controls; and time points assessed. Quality scoring was performed. The best evidence was synthesized from high‐quality skeletally mature models, without direct trauma to tissues of interest and with multiple time points. Altogether, 2849 abstracts were reviewed. Forty‐seven papers were included reporting eight different methods of inducing OA, six different species, six different methods of assessing cartilage, five different bone structural parameters, and four assessed meniscus as a potential initiator. Overall, the simultaneous onset of OA in cartilage and bone was reported in 82% of datasets, 16% reported bone onset, and 2% reported cartilage onset. No dataset containing meniscal data reported meniscal onset. However, using the best evidence synthesis (n = 8), five reported simultaneous onset when OA was induced, while three reported bone onset when OA occurred spontaneously; none reported cartilage onset. In summary, there is a paucity of well‐designed studies in this area which makes the conclusions drawn conjectures rather than proven certainties. However, within the limitation of data quality, this review suggests that in animal models, the structural onset of knee OA occurs either in bone prior to cartilage pathology or simultaneously.


| INTRODUCTION
Osteoarthritis (OA) is typically diagnosed and treated when pathology becomes detectable by conventional radiography, but the OA process begins long before this as evidenced by magnetic resonance imaging (MRI)-detectable structural pathology. 1 Therapies targeting tissue pathology at an earlier stage of disease (eg, after a major injury or in individuals at risk) might permit disease modification.
The concept of early OA in humans is imprecisely defined, and limited data exist on the timing and sequence of events that occur at the onset of human knee OA. OA is a syndrome of deterioration of synovial joints, characterized by multitissue pathology involving hyaline cartilage, subchondral bone, the calcified cartilage layer, the synovium and capsule, articular ligaments (in knees), and the menisci.
However, there remains considerable controversy over which tissue is key to initiate the OA process. 2 In humans, subchondral bone abnormalities, 3 cartilage lesions, 4 and meniscal damage 3,5 are all present in the knees of adults prior to radiographic OA. Nonetheless, the presence of these tissue lesions is associated with subsequent structural deterioration [6][7][8] and incident radiographic knee OA. 9 In light of ethical considerations surrounding human tissue sampling, animal models of OA are essential to advance the understanding of the temporal connection between cartilage pathology, subchondral bone structure, and menisci in the genesis and propagation of OA. Studying animal models is complex because in addition to the numerous different types of animals, there are numerous models of OA, some spontaneous and some induced. Often models are used to explore different mechanistic aspects of OA.
We aimed to perform a systematic review to evaluate animal studies of knee OA which reported data concerning subchondral bone structure along with cartilage and meniscal pathology at the onset of OA to assess the sequence of events occurring at the onset of knee OA.

| Search strategy and selection of studies
Three databases (PubMed, Web of Science, and Scopus) were searched up to April 2018. The keywords used were: bone, cartilage, OA, osteoarthrosis, arthrosis, arthritis, animal, and animals. Filters excluding articles not written in English, studying humans and review articles were applied and filters including animals and full articles were applied and were available for each database (Supporting Information A). Meniscus and synovium were not included in the search terms because earlier exploratory searches highlighted the relative paucity of studies including bone, cartilage, and meniscal analysis. Had meniscus been included in the search terms, only four studies would have been eligible for data extraction. Additionally, had synovium been included, no studies would have been eligible for data extraction. Therefore, only bone and cartilage search terms were used and we evaluated meniscal data when available as a part of the data extracted from the included studies.
Initially, titles and abstracts were screened and excluded if they: (a) used animals from the Muroidea superfamily; (b) used nonstandard animals (eg, python, sea lion, etc); (c) did not specify animal model; (d) did not present bone structure and cartilage degradation data; (e) conducted a surgical repair; (f) were conducted on joints other than the knee or stifle; (g) were conducting a study to see the effect of a given intervention; (h) induced OA through chemical injection of chondrotoxic or other proinflammatory substances; (i) did not induce OA; or (j) were a case study (n = 1). Additionally, articles were manually filtered for nonanimal, review articles, book chapters, and editorials. Full articles were then retrieved and the references screened for potential studies missed by the literature search.
We structured this review to exclude animal models in the Muroidea superfamily (mice and rats). Although mice and rats are commonly used for biomaterials testing at early stages of OA, their thin cartilage layer (3-5 cells thick) and the small nature of their joints 10-14 make it difficult to discern changes in bone structural parameters that are comparable to the human OA disease. Additionally, their lack of growth plate fusion makes it difficult to distinguish whether changes in bone structural parameters result solely from OA or from the influence of persistent growth plate patency. [15][16][17] Moreover, while these rodent models are useful for basic research because they are small, inexpensive, easy to manipulate on a genetic level, feature short reproduction cycles and lifespans, and produce meaningful and relevant data, 18 several articles and reviews highlight the poor correlation of rodent models with a variety of different human conditions by demonstrating failure to successfully translate rodent-based findings into successful clinical practice. [15][16][17] To remain focused and avoid ambiguity, we have decided to exclude these animal models. Guinea pigs, in contrast, were included in this review because OA in guinea pigs more closely resembles human knee OA pathogenesis 19 and because guinea pig growth plates fully fuse.

| Data analysis
Histological and imaging structural bone and cartilage data were assessed to determine which tissue degenerated first or whether they degenerated simultaneously. Cartilage and bone involvement CASPER-TAYLOR ET AL.

| 1811
were defined as pathology differing from baseline or control.
Datasets were stratified according to whether subchondral bone and cartilage structural pathology occurred concurrently (simultaneous onset) or in series with bone first (bone onset) or cartilage first (cartilage onset). A meta-analysis was precluded because of the heterogeneity of the datasets which were included.

| Quality analysis
The quality of each dataset was determined using a combination of several quality scoring systems that were modified to assess dataset quality and potential bias in study design and reporting of cartilage degradation and subchondral bone structural data. [20][21][22][23][24] The checklist was comprised of 11 criteria (Supporting Information B). Each criterion received a score of 1 if it was reported satisfactorily and received a 0 if it was not. The maximum score was 11. Scores were tabulated by two reviewers independently (MC and AB) and then compared. Discrepancies in scoring were discussed until consensus was reached. Datasets were then rated as having high (9-11), medium (6)(7)(8), or low (0-5) quality based on their final quality score.
Best evidence synthesis was derived from selecting high-quality studies, using models without direct trauma to tissues of interest (ie, meniscectomy, groove model, subchondral fracture, and osteochondral fragment), with skeletally mature animals and tissue data at multiple time points.

| Statistical analysis
Statistical analysis was not preformed since there were not sufficient best evidence synthesis datasets. The analysis would have been underpowered, and, thus, would have provided no meaningful insight.

| Study selection and characterization
The PRISMA flowchart describing study selection is presented in Figure 1.
The search returned 4004 articles. Replicate removal resulted in 2849 articles being included. Inclusion/exclusion criteria were applied and they were manually filtered to exclude nonanimal, review, and editorial articles, as well as book chapters. After reviewing the reference sections of the resulting articles, an additional 12 articles were added and screened. In total, 47 articles were included (Supporting Information C) and from these, we were able to extract 49 datasets.

| Data extraction from selected studies
A summary of key characteristics of these datasets is presented in Table 1 and the extracted data from each dataset are presented in Table S1 and Supporting Information C. The majority of datasets used

| Quality analysis
There were 17 datasets of high quality, 29 datasets were of medium quality, and 3 were of low quality. Individual scores are presented in Supporting Information D. The mean quality score was 7.6 (range, [5][6][7][8][9][10][11] which was considered to be "medium" quality. Half of the datasets demonstrating bone onset were of high quality (3/6), one was of medium quality, and two were of low quality. The one dataset demonstrating cartilage onset was a high-quality study. Most datasets demonstrating simultaneous onset were of medium quality 28/42. There was one of low quality and 13/42 was of high quality. This data are summarized in  All datasets reporting bone onset were from spontaneous models of OA and all datasets reporting simultaneous onset were from the ACLT induced model of OA. The one dataset from the medial release model did not meet the criteria for best evidence synthesis and so is not included in this analysis. The results of the best evidence synthesis are summarized in Table 2.

| Spontaneous
There were 19 datasets that used a spontaneous animal model to investigate OA and this included 15 guinea pig models, 3 horse models, and 1 cat model. Three datasets were of high quality, 14 were of medium quality, and 2 were of low quality. Twelve datasets

| Groove model
There were four datasets that used the groove model to induce OA and this included three dog models and one sheep model. Two datasets were of high quality and two were of medium quality. There were no lowquality datasets. Two datasets used multiple time points and two used single time points. All four datasets used skeletally mature animals.
Overall, all four datasets reported simultaneous onset, none reported either bone or cartilage onset. There was one dataset ranked as high quality that was in skeletally mature animals and used multiple time points. However, because this model causes direct trauma to cartilage, datasets from this model were not included in the best evidence synthesis analysis. This dataset reported simultaneous onset.

| Meniscectomy
There were four datasets that used meniscectomy to induce OA and this included three rabbit models and one guinea pig model. Two datasets were of high quality and two were of medium quality. There were no datasets of low quality. Three datasets used multiple time points and one used a single time point. Three of the datasets used skeletally mature animals and one used skeletally immature animals.
Overall, three datasets reported simultaneous onset, none reported bone onset, and one reported cartilage onset. No datasets that were high quality were in skeletally mature animals and had multiple time points, so no datasets from this group were included in the best evidence synthesis. There were two datasets that were ranked medium quality that were in skeletally mature animals and used multiple time points. These datasets both reported simultaneous onset.

| Overloading
There were three datasets that used overloading to induce OA and this included two rabbit models and one horse model. One dataset was of high quality, one was of medium quality, and one was of low quality. Two datasets had multiple time points and one used a single time point. All three datasets used skeletally mature animals. Overall, two datasets reported simultaneous onset and one reported bone onset. There was one dataset ranked as high quality that was in skeletally mature animals but used only one time point. This dataset reported simultaneous onset. However, because the high-quality dataset had a singular time point, the datasets from this model are not included in the best evidence synthesis analysis.

| Subchondral fracture
One dataset used subchondral fracture to induce OA. It used the dog model, in skeletally mature animals, was ranked of medium quality, and used one time point. It reported a simultaneous onset. The dataset from this model was not included in the best evidence synthesis analysis.

| Medial release
One dataset used medial release as a method to induce OA. It used the dog model, used skeletally mature animals, was ranked of high quality, and only used one time point. It reported a simultaneous onset. The dataset from this model was not included in the best evidence synthesis analysis.

| Osteochondral fragment
One dataset used osteochondral fragments as a method to induce OA. It used the horse model, used skeletally mature animals, was ranked of high quality, and used multiple time points. It reported a simultaneous onset.
However, because this model causes direct trauma to bone, the datasets from this model are not included in the best evidence synthesis analysis.

| Analysis of the initiation of OA from studies with available meniscal data
We found four datasets that included meniscal pathology assessment in addition to bone and cartilage assessments. One induced model, a dog In summary, from these four datasets, OA structural onset appears to occur simultaneously in the cartilage, bone, and meniscus although available data to draw this conclusion are limited and questionable.

| DISCUSSION
This is the first systematic literature review to explore the time sequence of tissue involvement in animal models of knee OA. We incorporated robust quality scoring and performed the best evidence synthesis. Overall, the high-quality evidence base was small, thus any conclusions drawn are subject to some uncertainty and there is a need for further investigation by studies designed specifically for the purpose of uncovering the sequence of events in animal knee OA.
Nonetheless, our findings suggest that the structural onset of OA occurs either simultaneously or in the subchondral bone. After the best evidence synthesis, all spontaneous models reported bone onset, while the induced models all reported simultaneous onset; none reported cartilage onset.
There is further evidence that supports bone onset and comes from studies including both spontaneous and induced animal models.
The studies could not be included in this review because of exclusion criteria and/or also the presence of human knee data which was excluded in the original database search. In one such study, using the spontaneous DH guinea pig model, bone pathology was observed at 10 weeks 26 before the onset of cartilage disease at 3 months Unfortunately, definitive human studies are lacking. There is a longitudinal study using serial knee MRI measurements in humans without symptoms of knee OA which reported that a higher baseline tibial bone area and osteophytes were associated with greater knee cartilage loss over 2 years. 32,33 However, at baseline between 13% and 28% of participants had a cartilage defect in the tibiofemoral compartment. 4 No best evidence synthesis dataset reported cartilage onset in There are important limitations to this study. Since the datasets included data from numerous animal models with a variety of induction methods, our analysis was performed cross-species and cross-induction methods. Differences within animal models include differing joint mechanics (eg, rabbits load knees mainly in the lateral compartment, whereas most other animals load knees more in the medial compartment), and different exercise regimes postsurgery (eg, rabbits and guinea pigs are allowed free cage movement, whereas generally dogs were allowed free pen movement with additional daily walks).  As stated previously, another limitation was some data were collected before epiphyseal closure, which confounds results concerning onset. There is a high potential that prior to epiphyseal closure the bone properties are changing. One example is that an increase in BV/BT has been shown prior to architectural adaptation in growing bone. 13 Lastly, the sample sizes included in final analyses were typically small, both in the number of animals used and/or the number of samples assessed. Smaller sample sizes inherently increase the risk of bias.

Differences in induction methods include destabilization vs
In conclusion, this systematic review indicates that in animal studies