Rheumatoid arthritis and osteoarthritis are both characterized by loss of extracellular matrix (ECM) in the cartilage of articular joints. Cartilage is maintained by chondrocytes that secrete ECM components, such as collagen and aggrecan. In both diseases, joint damage occurs as the cartilage matrix is destroyed by proteinases that are up-regulated by a variety of different stimuli. While ADAMTS-4 and ADAMTS-5 are mainly responsible for the degradation of aggrecan, collagen is degraded by the collagenases (matrix metalloproteinase 1 [MMP-1] and MMP-13). Tissue inhibitor of metalloproteinases (TIMPs) are endogenous inhibitors of MMPs, and TIMP-3 can also inhibit ADAMTS ([1, 2]). Aggrecan breakdown is reversible, but the irreversibility of collagen release makes its prevention key for developing effective therapies for arthritis. This requires detailed knowledge of the mechanisms involved in collagen breakdown. We have previously used cell and organ systems to examine the pathways that lead to the up-regulation of the collagenases following the addition of cytokines to chondrocytes ([3-5]). Since collagenases are initially synthesized in an inactive form, they require activators to be present in order to effect collagen release ().
In our in vitro models, we have used combinations of interleukin-1 (IL-1) and oncostatin M (OSM) to promote cartilage collagen breakdown; neither cytokine alone reproducibly leads to collagen cleavage ([5-7]). IL-1 is a proinflammatory cytokine that binds to the IL-1 receptor (IL-1R) and recruits IL-1R–associated kinase (IRAK) proteins, which are phosphorylated. This leads to recruitment of tumor necrosis factor receptor–associated factor 6 (TRAF6) proteins, which phosphorylate JNK. Activated JNK then phosphorylates c-Jun, which forms homodimers or binds c-Fos to form heterodimers, which form part of the activator protein 1 (AP-1) transcription factor. The c-Jun homodimers have low affinity for DNA (), whereas AP-1, which is composed of c-Fos and c-Jun, has high affinity for the promoter regions of many target genes, such as MMPs, phosphatases, ADAMTS, and the transcription factor Sp-1. Sp-1 inhibits TIMP-1 transcription by binding to a repressive element in the first intron of TIMP-1 (). Messenger RNA (mRNA) for c-Fos has a very short half-life, is not expressed under normal cellular conditions, and is only weakly expressed after stimulation with IL-1. Therefore, IL-1 stimulation alone will favor the formation of c-Jun homodimers, leading to lower levels of up-regulation of AP-1 target genes than those with IL-1 plus OSM stimulation.
OSM has antiinflammatory and proinflammatory roles, with signaling primarily via the JAK/STAT pathway (). There is evidence that p38 phosphorylates c-Fos to enhance its transcriptional activity (). OSM synergizes with IL-1 to increase the expression of MMPs in chondrocytes (), and since STAT proteins do not bind MMP promoters in chondrocytes, this synergy occurs through STAT stimulation of c-Fos expression, leading to changes in AP-1 composition that regulate MMP expression. It should be noted that c-Fos is regulated at the transcriptional level, whereas c-Jun is regulated post-translationally via phosphorylation. The pathways involved in collagen release are thus complex, involving cross-talk between different pathways and many feedback loops.
It has become increasingly recognized that systems modeling approaches are required to complement experimental work, and computational models have been widely used in the fields of cancer, cardiovascular diseases, and neurodegeneration; to date, this approach is not established in the study of musculoskeletal diseases ([13-15]). In this study, we used our existing in vitro data from cell and organ culture models to construct an in silico model of cartilage collagen breakdown following stimulation with IL-1 and OSM combinations. Our aim was to demonstrate how computational models can be developed using current knowledge of the system to highlight important gaps in knowledge, to test new hypotheses, and to predict outcomes for different therapeutic approaches.
- Top of page
- AUTHOR CONTRIBUTIONS
- Supporting Information
We developed a mathematical model of some of the pathways involved in cartilage degradation based on experimental data from human chondrocytes stimulated with the cytokines IL-1 and/or OSM. The model included sufficient components to explain the synergistic effects of IL-1 and OSM on MMP expression and the antagonistic effects of IL-1 and OSM on TIMP-1 expression (). The model was also validated using other data on components of the signaling pathways that are transiently activated in response to cytokines. Since arthritis is characterized by irreversible loss of ECM, we also wanted to include collagen and aggrecan degradation in the model.
The addition of cytokines by themselves is not sufficient for collagen breakdown and the release of collagen fragments, since MMPs are synthesized in an inactive form. Activation of proMMPs is mediated by proteases, and activation has been shown to be a key control point in terms of collagenolysis in arthritic cartilage (). Therefore, we included the addition of a collagenase-activating protease, which we named MMP activator, in our model. We parameterized the model so that when a collagenase-activating protease is included, ≥10% of collagen is degraded by day 14 after stimulation with IL-1 plus OSM, as seen with human cartilage explants (). We then used the model to simulate various possible interventions, and we examined the effect of these on collagen release.
Our model predicted that the use of receptor inhibitors may not be beneficial because antagonistic receptors do not totally stop the initial signal and because, following receptor activation, downstream signaling is rapid, leading to transcription of target genes such as ADAMTS and MMPs. This prediction was confirmed by published data showing that IL-1Ra inhibits collagen release in tissue from some patients but not from others (), which confirms that this may not be a useful approach by itself. OSMR antagonists have only recently been developed and have not yet been tested experimentally in a cartilage breakdown model, so these model predictions will be tested when reagents are available.
Simulated blocking of JAK-1 activity, one of the first kinases in the signaling pathways, was not effective unless 100% inhibition was achieved. With 100% inhibition, there was no phosphorylation of STAT-3, so there was no up-regulation of c-Fos. This means that there were only low levels of ADAMTS, MMPs, and MMP activator, which are insufficient to cause collagen release. However, 100% inhibition is unlikely in the clinical setting, and so this intervention may not be very beneficial if administered alone. The prediction for the effect of JAK-1 inhibition on collagen release is yet to be tested in an experimental setting.
Interestingly, the model predicted that inhibition of p38 or JNK activity would be much more effective, as this decreased the amount of phosphorylated c-Jun and c-Fos, respectively, and so inhibited the formation of AP-1 transcription factor complexes. Inhibiting p38 was predicted to be slightly more effective than inhibiting JNK, as this reduced the formation of the more transcriptionally active AP-1 complex, consisting of c-Fos/c-Jun heterodimers. Previous experimental data (not used in the construction of the model) confirmed that p38 inhibition reduces collagen release in a bovine model of cartilage breakdown (). In this study, we experimentally explored the effectiveness of JNK inhibition, as predicted by the model, and showed that this was indeed effective. However, the JNK inhibitor we used is not entirely specific, so it is not possible to conclude that these effects were solely due to JNK inhibition, although the results confirm the usefulness of computational models to predict effective interventions.
The model was also used to mimic the overexpression of TIMP-1 or TIMP-3 protein. Although we assumed that TIMP-1 mainly inhibits MMPs, whereas TIMP-3 mainly inhibits ADAMTS-4, the model predicted that TIMP-3 overexpression would have a greater effect on reducing collagen release (Figure 5). This was due to our assumption that aggrecan protects collagen from degradation, and the delay in aggrecan release meant that collagen was not accessible for degradation during the time period when MMP-1 and MMP-13 are most active. Therefore, the model suggested that targeting aggrecan release, especially if the intervention is performed at the appropriate time window, is a promising strategy to investigate further. Our predictions are supported by experimental data showing limited benefit from overexpressing TIMP-1 in a mouse model of arthritis (). Direct inhibition of MMPs with low molecular weight inhibitors proved to be ineffective in patients, as off-target effects were identified ().
Although we mainly used deterministic simulations in this study, stochastic effects are an important consideration in biologic systems. Our model predicted that the response to IL-1 plus OSM was variable in terms of the levels of active MMPs and collagen release (see Supplementary Figures 6 and 7, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38297/abstract). Stochastic simulations for TIMP-1 and TIMP-3 overexpression generated average behaviors that were similar to those in the deterministic model. TIMP-3 overexpression was much more effective, significantly delaying and reducing collagen release. Although individual simulation results exhibited considerable variability, this was reduced with increasing amounts of TIMP-3 overexpression, which suggests that this treatment is effective at reducing collagen release.
Our model represents a substantial contribution to the development of a systems approach to ECM breakdown, using cartilage as a reference tissue. This tissue is ideal for modeling studies, as it contains a single cell type. The current model is comprehensive, but we used a modeling approach that is very amenable to adding further details and making modifications as subsequent experimental data and new hypotheses emerge. For example, we are aware that there is cross-talk between signaling pathways, that other cytokines can initiate cartilage breakdown, and that other pathways or levels of control (e.g., the role of noncoding RNAs such as microRNAs and their effect on mRNA stability) are implicated, none of which were included in the model. As new experimental data become available, our model can be extended and refined. However, the guiding principle for building models is to capture the essential details without burdening the model with nonessential details (). For example, we modeled protein synthesis of some proteins (DUSP-1, MMP activator, MKP-1, PP-4, PTPRT, and Sp-1) as one step, omitting details of transcription, where we did not have data concerning mRNA levels.
The predictions generated by the present model are interesting in that intervention at the level of the receptors had little effect. This is supported by the fact that treatment of rheumatoid arthritis patients with IL-1Ra showed only modest beneficial effects (). Increasing the level of TIMP-1 was equally ineffective, which confirms the data generated when direct inhibition with MMP inhibitors proved to be ineffective in patients and affected other tissues of the joint as well (). The results with TIMP-3 could suggest that inhibition of the ADAMTS family could be effective in patients. Although this treatment would target aggrecan-degrading enzymes, a reduction in aggrecan release also helps to prevent irreversible collagen release, since collagen is inaccessible to MMPs when protected by aggrecan. Interventions that prevent the transcription of collagenases, particularly by interfering with JNK signaling pathways, had a much greater effect, and we have validated this prediction experimentally, confirming that this pathway may represent tractable therapeutic targets ().
In conclusion, there is a great need to increase our understanding of the molecular mechanisms involved in cartilage release and to develop new interventions (). We have shown that computer modeling is an ideal tool to assist in these processes, and there is great potential for future developments of this approach.
- Top of page
- AUTHOR CONTRIBUTIONS
- Supporting Information
All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. Proctor had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study conception and design. Proctor, Macdonald, Rowan, Cawston.
Acquisition of data. Macdonald, Milner, Cawston.
Analysis and interpretation of data. Proctor, Macdonald, Milner, Rowan, Cawston.