Arthritis is a complex disease group that affects bone, cartilage, and the surrounding tissues in articular joints (1, 2). Arthritis is a major cause of disability and afflicts over 50M adults in the United States (3). There are currently few successful therapies for arthritis, most of which are palliative in nature such as physical therapy and non-steroidal anti-inflammatory drugs (4). There has been some progress with disease-modifying drugs that can prevent continued progression of arthritis, especially thanks to the use of monoclonal antibodies, however, none can completely reverse the effects (5). Therefore, continued study of disease progression and therapeutic efficacy is critical.

To design better solutions for the treatment of arthritis, it is important to acknowledge that an articular joint is a complex organ consisting of multiple tissues such as cartilage, bone, vasculature, and synovium (6). Articular cartilage is responsible for dispersing and distributing the load on the joint (7). The primary cells in cartilage are chondrocytes, which are sparsely distributed in a dense extracellular matrix (ECM) whose main characteristic components are collagen type II, glycosaminoglycans (GAGs), and aggrecan (1). Below cartilage and in stark contrast is the subchondral bone, innervated and vascularized, rich in minerals and collagen type I, and populated with osteocytes, osteoblasts, and osteoclasts (8). Bone also encapsulates the bone marrow cavity, which is rich in nutrients and stem cells (9, 10). The synovium, lining the articular joint capsule, contains macrophage-like cells and fibroblast-like cells that clear debris and secrete ECM components and cytokines (11, 12). In a healthy joint, the cartilage, bone, vasculature, and the joint space components undergo constant crosstalk to function as an articular unit. Nutrients from the joint space help support cartilage health and function. The vascularization drives bone remodeling (13), and in turn the bone supports cartilage structure and function (14). These processes can be compromised in arthritis, thus dysfunction in one tissue can negatively affect any of the others. This crosstalk plays an important role in the development and progression of arthritis in articular joints. Inflammatory cytokines in the joint space affect the phenotype of the resident cartilage and bone cells and propagate the disease (11, 15, 16). The cartilage responds to inflammatory cues with the degradation and progressive loss of critical ECM components such as GAGs and collagen type II (16–19). As a result, the subchondral bone experiences changes in mineralization and turnover, and local lesions appear (20–23). Functionally, the cartilage loses its mechanical integrity, and its erosion leads to bone-on-bone contact, which is associated with significant pain for patients. It is clear that when studying arthritic disorders or developing treatments, it is critical to consider tissue crosstalk in response to inflammatory signals and how they may affect each other as well as the ultimate clinical outcomes.

Given the complexity of the interactions in arthritic articular joints, animal models are commonly used to capture the full breadth of physiological functions (24–26). However, animal models have varying degrees of biological similarity compared to humans in both joint structure and immunological profile (27, 28). An example of differing joint mechanics is that human knees can reach full extension, whereas neither cow, sheep, goat, pig, dog, or rabbit knees can even surpass 20° extension (28). Immunologically, there are many differences in immune cell populations and responses between common animal models and humans. Specific examples relevant to this work include the toll-like receptor (TLR) pathway; Seok et al. found that in response to inflammation, activation of genes within the TLR pathway of a mouse model could not accurately predict the activation of genes in humans (29). Notably, mouse monocytes have been shown to lack a TLR4 response to lipopolysaccharide (LPS) (30). More broadly, Vijayan et al. found that mouse macrophages favored oxidative phosphorylation after LPS activation, whereas human macrophages reprogrammed to increase glycolysis (31). Another relevant example is that mouse endothelial cells have been shown to express P-selectin in response to tumor necrosis factor (TNF) and LPS, whereas human endothelial cells were nonresponsive (32). Considering the differences in immune response between human and animal models, it is not surprising that many therapeutics that are successfully tested in animals only have limited success in the clinic (5, 33). For example, mouse and rat studies of a disintegrin and metallopeptidase with thrombospondin (ADAMTS) type 5 inhibitor showed successful prevention of cartilage loss (34), but there was no statistically significant improvement in a human clinical trial (35). In fact, even a Food and Drug Administration (FDA) approved drug, anakinra, only shows modest efficacy in the clinic but poses a risk of infection (36, 37). Similarly, FDA-approved infliximab for TNF inhibition only has about a 50% success rate in clinic, and the response is subtle (38). Hence, while in vitro models lose in complexity compared to animals, they offer the unique possibility of using human cell sources, thus gaining biological similarity in that respect. Organ-on-a-chip and in vitro models have been advocated for as complementary tools alongside animal models. Additionally, animal models may be costly or inaccessible to many researchers around the globe, and there is increasing drive to reduce or replace the use of animals in scientific research (39, 40).

Given the shortcomings of animal models discussed above, there has been a growing body of research on the use of in vitro models to study arthritis progression. However, many of the currently available in vitro models have yet to recapitulate the complexity of the multiple tissues in articular joints. It is a challenge to co-culture many different cell types in different matrices, perfuse them with different media to match each cell’s needs, and target the delivery of inflammatory agents. As mentioned earlier, there are important biochemical cues from the synovial fluid that lead to disease progression in cartilage, as well as between the cartilage and the subchondral bone (16, 41, 42). The inflammatory agents in the synovial fluid play a role in arthritis development (12), and the constant signaling between the synovium, cartilage, and bone is thought to sustain disease progression (43). Thus, a more veritable in vitro model should be able to capture the crosstalk between bone and cartilage in order to accurately depict disease progression. One approach has been using transwell plates to co-culture combinations of mesenchymal stem cells (MSCs), chondrocytes, and synoviocytes (44), and study the effect that each cell type has on the others in a healthy or diseased environment. However, transwells do not prevent media mixing and cells cultured in monolayers do not capture the three-dimensional (3D) structure of the joint tissues (45). Human cartilage is over 2mm thick, and subchondral bone is similar if not thicker. Diffusion of growth factors and cytokines across such relatively long distances compared to the scale of a cell, and through dense matrices critical to tissue function, is not frequently modeled in vitro. To account for this, Samavedi et al. adapted co-culture models in transwells to accommodate a hydrogel with encapsulated macrophages in one compartment and chondrocytes in a hydrogel in the other transwell compartment, to study crosstalk between the two cell types in a more biomimetic, 3D environment (46). The shortcoming of this model is that the two constructs do not interface, so the diffusion of signaling factors occurs through the common medium, whereas in vivo the signaling factors diffuse more directly through the tissue (47, 48). To account for direct tissue-tissue contact, biphasic constructs have successfully been created to model the cartilage-bone interface, however the whole construct was still cultured in a single common medium. This does not account for the specific cues each tissue needs or is exposed to in vivo, nor does it eliminate the confounding crosstalk factor of signaling diffusion through the medium rather than across the cartilage-bone interface (49, 50). So, while great progress has been made to model arthritis of the articular joint in vitro, there is much room for improvement.

In arthritis, the joint is often inflamed, and synovium inflammation, possibly associated with macrophage polarization to a pro-inflammatory phenotype, is thought to be a major driver of arthritic development. The synovial fluid directly interfaces with cartilage; however, few in vitro models can separately deliver inflammatory cues to either cartilage or bone. In this work, we adopted one of few approaches leveraging our previously described biphasic bioreactor that allows us to perfuse two different types of media through a biphasic construct without mixing (51, 52). Therefore, we perfused the cartilaginous region of a construct with chondrogenic medium, and the osseous region with osteogenic medium, separately. For our in vitro osteochondral construct, we generated cartilage and vascularized bone analogs as described in our previous work (41, 42). The constructs differentiated in the bioreactor independently while maintaining contact to allow crosstalk between the cartilaginous and vascularized osseous regions. Furthermore, the primary key benefit of the bioreactor is the possibility to perfuse the cartilage region only with inflammatory cues modeling an inflamed synovial fluid. We then probed the effects of the inflammatory agents on the cartilaginous layer, and the subsequent response of the vascularized osseous layer to an inflamed cartilage analog. While delivering synovial fluid in our bioreactor is beyond the scope of the study and not feasible due to limited access to samples, we perfused the cartilage region with inflammatory cytokines that are commonly found in synovial fluid of arthritic patients (11). Furthermore, rather than just delivering one or more inflammatory cytokines to the constructs, to better mimic some of the complexity of in vivo inflammation, we also tested the effect of macrophage conditioned medium (MCM) to represent the inflammatory signals secreted by synovial-resident macrophages during arthritis (15, 16). Thus, we compared the response to a combination of cytokines that are highly implicated in arthritis with the response to a pro-inflammatory macrophage conditioned medium that represents a complex biological cocktail more similar to synovial fluid (53, 54). We aimed to validate crosstalk in the proposed system and to determine how to best mimic in vivo inflammation. We hypothesized that MCM will evoke an inflammatory profile similar but not identical to that of the cytokine cocktail, and that there will be detectable responses in the vascularized-bone region after pro-inflammatory activation of the cartilage region.

Reagents were purchased from Thermo Fisher Scientific (Waltham, MA) unless otherwise specified.

In this study, we developed an in vitro micro-physiological system to model cartilage-bone crosstalk in arthritic environments. To study the inflammatory response of the cartilage analog, we exposed isolated cartilage constructs in a plate to our pro-inflammatory macrophage conditioned medium and cytokine cocktail and compared to the control. Then, we exposed the cartilage region of an osteochondral construct to the same inflammatory conditions and compared the response of the coculture cartilage to the isolated cartilage to determine how the coculture affected the inflammatory response profile. Lastly, to probe crosstalk in our system, we analyzed which evidence of an inflammatory response was shown in the osseus region when the cartilage region was exposed to inflammatory conditions.

We first exposed isolated cartilage constructs to pro-inflammatory stressors to tease out the different responses between CMB and the MCM. In vivo during arthritis, cartilage degrades via aggrecanases and metalloproteinases, losing critical ECM components such as collagen type II and GAGs (17). These phenotypes and their associated gene expression are what we looked for when analyzing the response to inflammatory agents of our cartilage constructs. From RT-qPCR and biochemical assays we observed that both inflammatory conditions resulted in cartilage degradation, although the MCM and CMB conditions presented different inflammatory profiles. There was an increase in MMPs and ADAMTSs, and a decrease in Collagen II and aggrecan. The cartilage construct response to MCM seemed to have greater variability, suggesting a potential stronger response. Furthermore, immunofluorescent staining showed that MCM drove a more substantive decrease in collagen type II signal than the CMB condition. This suggests that the MCM comprises inflammatory agents beyond the inflammatory cytokines in the CMB that may contribute to arthritis initiation and progression. However, isolated cartilage constructs are a somewhat limited in vitro model when the whole articular joint is involved.

We then repeated the test of the same pro-inflammatory conditions within our biphasic bioreactor combining the cartilage constructs with a vascularized osseous construct in a vascularized osteochondral model. We assessed the impact of crosstalk with the adjacent vascularized-bone construct on the cartilage layers when exposed to a pro-inflammatory environment. We put particular care in including HUVECs within the bone scaffold, so that the construct was both vascularized and mineralized. In fact, previous work from our group when developing the vascularized-bone construct (41) showed that the addition of a vascular component to the bone analog supported more robust osteogenesis via paracrine signaling (65–67). In fact, in vivo, vasculature plays a critical role in bone development, maintenance, and regeneration (13), besides supplying oxygen and nutrients from blood. Recently, the Zelzar lab has even proposed that vasculature supports bone growth during development by acting as a collagen I template guiding osteoblast growth and mineralization (68). On the other hand, the MSCs included in our constructs acted as pericytes and stabilized the forming vessels (69), and the MSC-derived osteoblasts in the construct secreted VEGF to support endothelial cell growth and differentiation (70–72). This interconnected relationship between angiogenesis and osteogenesis is preserved in in vitro engineered bone constructs (73). Thus, we included vascularization in our model to ensure robust osteogenesis and a more realistic in vitro analog.

Bone and cartilage work together in vivo during both homeostasis and disease, so we expected the addition of the vascularized-bone construct to alter the response of the cartilage construct to inflammation. BMPs secreted by osteoblasts and osteocytes (74) have been shown to protect cartilage from inflammation (75). In contrast, Hopkins et al. showed that damaged subchondral bone tissue inhibited cartilage GAG production in vitro (76). The vascularization in the bone also affects cartilage. Endothelial cells that support osteogenesis may also activate the receptor activator of nuclear factor-κB ligand (RANKL) pathway that regulates cartilage resorption (77) and secrete MMPs that remodel both bone and cartilage. Nagao et al. showed that targeting VEGF, a critical factor in osteogenesis and angiogenesis coupling, attenuates osteoarthritis progression in cartilage (78). Furthermore, vasculature has been shown to activate the phosphatidylinositol 3-kinase and protein kinase B (PI3K/Akt) and BMP pathways (42, 67, 79, 80), which help chondrogenic differentiation both in vivo during development, and in vitro. Whether the vascularized-bone construct ameliorated or worsened the cartilage construct response to inflammation, we anticipated the crosstalk between the two regions to change in inflammation. Interestingly, we saw the secretion of MMPs was greater in the CMB condition compared to the MCM, which differed from the isolated cartilage results. This suggests that crosstalk between the bone and cartilage does in fact contribute to cartilage inflammatory response, and that factors secreted by the vascularized-bone construct regulate the cartilage construct’s inflammatory response profile. This reinforces the concept that an isolated cartilage construct is hardly sufficient to recapitulate the dynamics occurring in arthritis, and that a robust in vitro model should include the critical articular tissues of vascularized-bone.

This also raises interest in the difference between MCM and CMB, and how they elicit different responses in the cartilage construct and the vascularized-bone construct. When deciding what cytokines to include in the combined cytokine cocktail, it was critical to focus on cytokines most implicated in arthritis (54, 81–83). Among these, IL-1β, IL-6, and TNF-α play a substantive role in both rheumatoid arthritis and osteoarthritis (82, 83). Inflammation in the synovium is the driver of cytokine production in joint disease (81), with synovial fibroblasts producing IL-6 (23) during arthritis progression, whereas macrophages produce IL-1β and TNF-α (11, 54). In the design of the CMB, we started with the concentrations of these three cytokines in the MCM and multiplied tenfold. Arthritis development in vivo occurs over decades with physiological levels of these cytokines, so we then chose to consistently use supraphysiological levels to ensure a response in our in vitro model in experimentally reasonable time frames. Supraphysiological levels of cytokine are often necessary to elicit a response in 3D cultures due to cytokine interaction with and diffusion through the gel (84, 85), and the protective nature of 3D culture (86, 87). The tenfold multiplication placed the cytokine concentrations in the ranges found most commonly in literature for 3D studies (55, 88, 89) and kept it proportional to the levels in our MCM to maintain consistency. We chose to use a combination of the cytokines instead of testing them individually to represent the in vivo environment more closely. Interestingly, there is evidence of crosstalk between the IL-6 and IL-1β pathways (90, 91), and IL-6 has been shown to regulate the IL-1β and TNF-α signaling (92). The cytokines work synergistically to drive disease and should then be studied together. While the cytokines chosen in this study have been shown to have the strongest effect on inflammation, there are of course far more than 3 cytokines that contribute to arthritis. The modularity and convenience of our system means that many different cytokines could be tested. Moving forward, one might consider teasing out the role of other cytokines hypothesized to have a role in arthritis such as IL-10, IL-12, and IL-8, as well as chemokines such as CCL20 and CCL2 (23, 81, 83).

This very same reasoning is also why we chose to test pro-inflammatory macrophage conditioned medium as an inflammatory agent. In vivo, synovial macrophages release many pro-inflammatory signals that are major drivers of arthritis. While there has been significant research into the secretome of synovial macrophages, the specifics of its composition and effects are still being decoded (18, 93–95). Nevertheless, we used macrophage conditioned medium to test if it may better represent in our model the in vivo arthritic conditions, driving a response more analogous to in vivo inflammation. Although, macrophages polarized in vitro differ from macrophages in vivo (96), we hypothesized that the macrophage conditioned medium would still be sufficient to drive a response in cartilage. We saw that the MCM induces inflammation similar to—and in some cases greater than—that of the CMB condition. For simplicity, using a combination of cytokines successfully induced an inflammatory response in vitro, however, the MCM more severely depleted extracellular collagen II protein expression. Since loss of critical ECM components is a hallmark of cartilage degradation in arthritis, this suggests that the MCM can derive an inflammatory response that is more functional and clinically relevant. As for why the MCM may drive a greater inflammatory response, pro-inflammatory macrophages secrete an abundance of proteins (97) that account for many more signaling pathways than those activated by TNF-α, IL-6, and IL-1β. We then hypothesized that the other cytokines secreted by the macrophages, such as IL-12 or IL-23, or chemokines, such as CXC chemokine ligand 1 (CXCL1) and CXCL2 (98, 99), contributed to further response by the construct. Another hypothesis is that pro-inflammatory macrophages secreted nitric oxide (NO) (100), and NO has long been associated with degradation of articular cartilage in arthritis (101). Ideally, the MCM would be a more accurate representation of the synovial fluid in the joint space than the cytokine cocktail, but further secretory analysis of the MCM, and especially the synovial fluid of arthritic patients would be necessary to support this. At this point, we know that synovial fluid from arthritic patients contain cytokines like IL-1, IL-6 (82), and IL-20 (83), and proteases like MMP1 and MMP3 (102). Pro-inflammatory macrophages similarly express MM1 and MMP3 (103), and we confirmed here that the conditioned media contains cytokines IL-1β, IL-6, and TNF-α. There are clearly similarities between conditioned media from pro-inflammatory macrophages and arthritic synovial fluid, but more research is required to precisely determine the full extent of this similarity.

We then looked at the effect of the inflammatory agents on the vascularized-bone analog to determine if the cartilage construct’s inflammation would initiate a response in the vascularized-bone construct. It is important to emphasize that the vascularized-bone construct was never directly perfused with inflammatory agents, and the effect of inflammatory agents onto the vascularized-bone region would be mediated by the cartilage construct. In vivo, during arthritic disruption, inflammation dysregulates the RANKL signaling and produces inflammatory cytokines that cause osteoclast-mediated bone erosion (20). Furthermore, IL-1β blocks mineralization and disrupts osteoblasts (20), and results in strong secretion of MMPs by both osteocytes and endothelial cells in arthritic environments (10). Therefore, we expected to see a decrease in bone mineralization agents and an increase in MMP secretion in our vascularized-bone region in response to inflammation. Notably, we did see an increase in MMP1, 3, and 13 gene expression and protein secretion in the vascularized-bone construct in both CMB and MCM inflammatory conditions. We also saw a decrease in osteogenic genes in the CMB condition. Unexpectedly, we observed a slight increase in RUNX2 and IBSP in the MCM group, implying there may be something in the MCM that promotes expression of those genes. One potential explanation is that inflamed and resorbed bone release BMPs (104), and BMPs can trigger both RUNX2 and subsequent IBSP upregulation (105). Since the vascularized-bone construct presented an inflammatory response profile even when MCM or CMB were delivered only to the cartilage construct, this suggests that the inflamed cartilage construct signals to the bone construct and propagates the inflammation and disease. Thus, we confirmed that our in vitro model allows for crosstalk between the cartilage analog and the bone analog, recapitulating some aspects of the crosstalk that occurs in vivo during disease.

As an even closer model to in vivo circumstances, one might consider using explants in the biphasic bioreactor, and several research groups have created successful ex vivo models using tissue explants (106, 107). However, animal explants suffer from the same limitation as the use of animal cells in terms of lower biological similarity to humans. Human explants would be ideal, but they are more challenging to source in sufficient number for screening and can more easily be obtained by joint replacement surgeries, hence, from joints that present varying and inhomogeneous degrees of tissue degeneration. Therefore, creating a biomimetic osteochondral construct with human cell sources is a good option to create a biologically similar human analog while allowing for significant replicates appropriate for the future screening of therapeutics.

Overall, the results of our study showed that our micro-physiological system is a valid model of crosstalk between bone and cartilage in an arthritic environment. We showed that macrophage conditioned media generated an inflammatory response profile in the cartilage constructs that may be more clinically relevant compared to a cytokine cocktail, suggesting that the latter may not capture the full extent of the inflammatory profile of the synovium in arthritis. Furthermore, we showed that the combination with a vascularized-bone analog in an osteochondral construct modulated the cartilage analog’s response to inflammatory agents, pointing to crosstalk between the cartilaginous and vascularized-osseous components. Notably, the main benefit of our model is the ability to deliver agents to one compartment within the biphasic bioreactor to study the effect of crosstalk between the two tissues. This unique feature could be leveraged to explore the effects of specific molecules in modeling or modulating arthritis, or to test the efficacy of therapeutics in a good throughput system prior to preclinical animal studies.