Browse the corpus

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 (6).

ve 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 (6). 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 (8), 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 (9). 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.

as 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 (10). There is evidence that p38 phosphorylates c-Fos to enhance its transcriptional activity (11). OSM synergizes with IL-1 to increase the expression of MMPs in chondrocytes (12), 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.

P-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.

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. METHODS Model construction The model contains 3 separate submodels: the first describes the IL-1/JNK signaling pathway, the second describes the OSM/STAT-3 signaling pathway, and the third describes proMMP activation and the release of aggrecan and collagen (see Supplementary Figures 1–3, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38297/abstract). This modular approach allows new components to be easily added and allows existing modules to be reused in further models. The models were encoded in the Systems Biology Markup Language (SBML), a computer-readable format for representing biochemical networks (16), using the Python tool SBML-shorthand (17). The integrated model was deposited in the BioModels Database (18) and assigned the identifier MODEL1305280001. The main features and assumptions of the model are described below. Full details of the model species, parameters, and reactions are given in Supplementary Tables 1–3 (available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38297/abstract).

ed the identifier MODEL1305280001. The main features and assumptions of the model are described below. Full details of the model species, parameters, and reactions are given in Supplementary Tables 1–3 (available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38297/abstract). IL-1 signaling pathway module This module contains details of the signaling pathway from IL-1 binding to its receptor, leading to a cascade of phosphorylation events via IRAK-2, TRAF6, JNK, p38, and finally c-Jun. We assumed that phosphorylated c-Jun can form dimers that have low transcriptional activity for the following target genes: c-Jun, MMP-1, MMP-3, MMP-13, ADAMTS-4, protein phosphatase 4 (PP-4), dual-specificity protein phosphatase 16 (DUSP-16), and MAPK phosphatase 1 (MKP-1). We also included basal transcription of c-Jun, TIMP-1, and TIMP-3, which does not involve c-Jun. For species with available experimental data for both the mRNA and protein levels, we included transcription and translation reactions (c-Jun, MMPs, and ADAMTS-4); otherwise, we modeled protein synthesis as one reaction (DUSP-16, PP-4, and MKP-1).

ranscription of c-Jun, TIMP-1, and TIMP-3, which does not involve c-Jun. For species with available experimental data for both the mRNA and protein levels, we included transcription and translation reactions (c-Jun, MMPs, and ADAMTS-4); otherwise, we modeled protein synthesis as one reaction (DUSP-16, PP-4, and MKP-1). OSM signaling pathway module This module contains details of the signaling pathway from OSM binding to its receptor (OSMR), leading to phosphorylation of JAK-1 and STAT-3. We assumed that pSTAT-3 can be transported to the nucleus, where it may transcribe c-Fos, receptor-type protein tyrosine phosphatase T (PTPRT), and suppressor of cytokine signaling 3 (SOCS-3). If pSTAT-3 is dephosphorylated in the nucleus, it is transported back to the cytoplasm. SOCS-3 binds to OSMR in competition with OSM and so provides a negative feedback loop by which to stop this signaling pathway. The c-Fos protein is phosphorylated by p38 and dephosphorylated by either DUSP-16 or an unspecified phosphatase. Phosphorylated c-Fos can reversibly bind to phosphorylated c-Jun to form a complex, which represents the AP-1 transcription factor. We include transcription of the following genes by c-Fos/c-Jun heterodimers: c-Fos, c-Jun, MMP-1, MMP-3, MMP-13, ADAMTS-4, DUSP-16, PP-4, MKP-1, Sp-1, TIMP-1, and TIMP-3. PP-4 binds to TRAF6 to prevent the binding of TRAF6 to IRAK-2 and so provides an additional negative feedback loop to stop OSM signaling.

ion factor. We include transcription of the following genes by c-Fos/c-Jun heterodimers: c-Fos, c-Jun, MMP-1, MMP-3, MMP-13, ADAMTS-4, DUSP-16, PP-4, MKP-1, Sp-1, TIMP-1, and TIMP-3. PP-4 binds to TRAF6 to prevent the binding of TRAF6 to IRAK-2 and so provides an additional negative feedback loop to stop OSM signaling. Activation of proMMPs and degradation of aggrecan and collagen module We assumed that proMMP-1 and proMMP-3 are activated by a generic activator (MMP activator) and that MMP-3 also activates proMMP-1 and proMMP-13. Initially, aggrecan is bound to collagen and behaves as a complex, since it has been shown that aggrecan protects collagen from degradation. We modeled aggrecan degradation by ADAMTS-4, which results in an unprotected collagen molecule. Free collagen can then be degraded by MMP-1 or MMP-13. We assumed that MMP-13 has greater activity at cleaving collagen than MMP-1 but that there are much higher levels of MMP-1 compared to MMP-13, which means that both collagenases are important for collagen release (19). We also assumed that activated pools of MMPs are removed with a half-life of about 30 hours. It has been shown that TIMP-1 is more effective at inhibiting MMPs whereas TIMP-3 mainly inhibits ADAMTS (1), and this assumption is included in our model.

h means that both collagenases are important for collagen release (19). We also assumed that activated pools of MMPs are removed with a half-life of about 30 hours. It has been shown that TIMP-1 is more effective at inhibiting MMPs whereas TIMP-3 mainly inhibits ADAMTS (1), and this assumption is included in our model. Parameter values Parameters for the signaling pathways were chosen to fit experimental data where available. For example, we used published data on receptor binding affinities (20,21), time course data on kinase activity (22,23), data on phosphorylation and localization of STAT-3 (23), and time course data on SOCS-3 expression (24). We used data from our laboratory to fit the kinetics of c-Fos induction and the return of c-Fos to basal levels (see Supplementary Figure 4, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38297/abstract). As we wished to use stochastic as well as deterministic simulations, we needed to use low numbers of molecules so that multiple simulations could be performed in reasonable time frames. We therefore used values in the region of 0–200 for basal expression, with levels increasing to ∼500–30,000 after cytokine-induced up-regulation. In a deterministic simulation, the level of the mRNA or protein remains fixed under basal conditions, but in a stochastic simulation, the levels may fluctuate around the mean value (compare Supplementary Figure 5 [available on the Arthritis & Rheumatology web site at http://online library.wiley.com/doi/10.1002/art.38297/abstract] with Figure 1). The level of MMP-1 (mRNA) is often 10-fold higher than that of MMP-13 (25). Thus, assuming equal degradation rates for MMP-1 and MMP-13 mRNA, we set the transcription rate of MMP-1 at 10 times higher than that of MMP-13.

atology web site at http://online library.wiley.com/doi/10.1002/art.38297/abstract] with Figure 1). The level of MMP-1 (mRNA) is often 10-fold higher than that of MMP-13 (25). Thus, assuming equal degradation rates for MMP-1 and MMP-13 mRNA, we set the transcription rate of MMP-1 at 10 times higher than that of MMP-13. Figure 1 Simulation results showing the effect of interleukin-1 (IL-1) and/or oncostatin M (OSM) on the expression of matrix metalloproteinase 1 (MMP-1), MMP-13, and tissue inhibitor of metalloproteinases 1 (TIMP-1), using a simulated time period of 48 hours. Curves show the level of MMP-1 mRNA, MMP-13 mRNA, TIMP-1 mRNA, and collagen fragments. A, Effect of IL-1 alone. B, Effect of OSM alone. C, Effect of IL-1 plus OSM. Note that in B, MMP-1, MMP-13, and collagen levels are all zero; thus, the individual lines are not visible. Model validation The model was validated using experimental data that were not used in construction of the model (26,27). The details are given in Supplementary Figure 9 and Supplementary Table 7 (available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38297/abstract).

Figure 1 Simulation results showing the effect of interleukin-1 (IL-1) and/or oncostatin M (OSM) on the expression of matrix metalloproteinase 1 (MMP-1), MMP-13, and tissue inhibitor of metalloproteinases 1 (TIMP-1), using a simulated time period of 48 hours. Curves show the level of MMP-1 mRNA, MMP-13 mRNA, TIMP-1 mRNA, and collagen fragments. A, Effect of IL-1 alone. B, Effect of OSM alone. C, Effect of IL-1 plus OSM. Note that in B, MMP-1, MMP-13, and collagen levels are all zero; thus, the individual lines are not visible. Model validation The model was validated using experimental data that were not used in construction of the model (26,27). The details are given in Supplementary Figure 9 and Supplementary Table 7 (available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38297/abstract). Simulated treatments and interventions The integrated model was used to assess possible therapeutic interventions, since cartilage breakdown does not occur in any of the individual submodels. We mimicked the addition of cytokines and MMP activator by altering the initial values of IL-1, OSM, and/or MMP activator before running the simulations (see Supplementary Table 4). Details of the species and parameters used to simulate possible interventions are shown in Supplementary Table 5 (Supplementary Tables 4 and 5 available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38297/abstract).

before running the simulations (see Supplementary Table 4). Details of the species and parameters used to simulate possible interventions are shown in Supplementary Table 5 (Supplementary Tables 4 and 5 available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38297/abstract). Modeling tools We carried out all the deterministic simulations using the COPASI software tool (28). Stochastic simulations were carried out on a computer cluster using the Gillespie algorithm (direct method) (29) and code developed by staff members of Newcastle University (30). We analyzed the model output using the R statistical programming language, and created the graphs using the R package ggplot2 (31). Network figures were constructed in CellDesigner (32) using the standard Systems Biology Graphical Notation (SBGN) (33). Previous experimental data used for constructing model Human T/C-28a4 chondrocytes were stimulated with IL-1α (1 ng/ml), OSM (10 ng/ml), or IL-1α (1 ng/ml) plus OSM (10 ng/ml). The test reagents were added to the medium at the start of the experiment, and cells were harvested at different time points thereafter (4, 8, 12, 24, 48, and 72 hours) (4).

d for constructing model Human T/C-28a4 chondrocytes were stimulated with IL-1α (1 ng/ml), OSM (10 ng/ml), or IL-1α (1 ng/ml) plus OSM (10 ng/ml). The test reagents were added to the medium at the start of the experiment, and cells were harvested at different time points thereafter (4, 8, 12, 24, 48, and 72 hours) (4). RESULTS IL-1 and OSM have synergistic effects on MMP-1, MMP-13, and TIMP-1 expression and collagen release Simulations were performed for 48 hours (virtual time) to examine the kinetics of MMP-1, MMP-13, and TIMP-1 induction (Figure 1). No collagen release occurred at this time point. In the model, addition of IL-1 alone led to an increase in MMP-1 expression, which peaked at ∼12 hours, but there was no induction of TIMP-1 above basal levels (Figure 1A). The addition of OSM alone led to an early induction of TIMP-1, which peaked at ∼2–4 hours and then slowly returned to basal levels; however, there was no induction of MMP-1 or collagen release (Figure 1B).

MMP-1 expression, which peaked at ∼12 hours, but there was no induction of TIMP-1 above basal levels (Figure 1A). The addition of OSM alone led to an early induction of TIMP-1, which peaked at ∼2–4 hours and then slowly returned to basal levels; however, there was no induction of MMP-1 or collagen release (Figure 1B). Modeling the addition of IL-1 and OSM together led to a synergistic induction of MMP-1, which peaked at ∼8 hours, with levels being ∼20-fold higher than those seen with IL-1 alone (Figure 1C). There was also a lower induction of TIMP-1 than was modeled with OSM alone. The simulation output showed that the level of MMP-13 was ∼10 times lower than that of MMP-1 throughout the time course, as demonstrated experimentally (25). These results compare well to the experimental data whereby human T/C-28a4 chondrocytes were stimulated with IL-1α, OSM, or IL-1α plus OSM (see Methods for details) (4). Therefore, our model assumption that c-Jun homodimers have very low transcriptional activity compared to c-Fos/c-Jun heterodimers produces results that are consistent with the experimental data.

al data whereby human T/C-28a4 chondrocytes were stimulated with IL-1α, OSM, or IL-1α plus OSM (see Methods for details) (4). Therefore, our model assumption that c-Jun homodimers have very low transcriptional activity compared to c-Fos/c-Jun heterodimers produces results that are consistent with the experimental data. Simulated addition of MMP activator leads to collagen release in the presence of IL-1 plus OSM Further simulations were performed over a 14-day period to examine the kinetics of collagen release with and without the addition of MMP activator (Figure 2). In the absence of MMP activator, the model output showed that ∼0.03% of collagen was released (by MMPs) by day 14 due to low basal levels of MMP activator following the addition of IL-1 plus OSM (Figure 2A). However, aggrecan was cleaved by ADAMTS-4, with ∼87% released by day 14. If an activator of collagenases was added at the start of the simulation, then proMMPs were processed at a much greater rate, and by day 14, 10% of collagen was released in the model (Figure 2B). The rate of aggrecan release was not affected by the addition of MMP activator. Figure 2 Simulation results showing the effect of an MMP-activating protease on the activation of MMPs and on collagen release, using a simulated time period of 14 days. Curves show the levels of active MMP-1 protein, active MMP-13 protein (scaled by a factor of 10), and collagen fragments. In each simulation, both IL-1 and OSM were added. A, Effect without MMP activator. B, Effect with addition of MMP activator. See Figure 1 for definitions.

, using a simulated time period of 14 days. Curves show the levels of active MMP-1 protein, active MMP-13 protein (scaled by a factor of 10), and collagen fragments. In each simulation, both IL-1 and OSM were added. A, Effect without MMP activator. B, Effect with addition of MMP activator. See Figure 1 for definitions. The model was used to assess therapeutic intervention points for the prevention of cytokine-induced cartilage breakdown In order to assess the effects of potential therapeutic interventions, we used the model with IL-1 plus OSM and high levels of activating protease, such that collagen release occurred. We ran simulations for a time period of 14 days and compared the levels of aggrecan and collagen fragments with different simulated doses of each treatment/addition. IL-1 receptor antagonist is predicted to have only limited beneficial effects The model predicted that very high levels of IL-1R antagonist (IL-Ra) would need to be added for there to be any significant effect on collagen release (100 times more antagonist than IL-1 receptors). This is because even low levels of IL-1R binding to IL-1 could initiate signaling, which eventually led to downstream events that culminated in collagen release. Thus, the model predicted that this intervention may delay or slow down collagen release, but cannot prevent initiation of the disease process. It is therefore likely that such an intervention used alone would not be very beneficial (Figure 3A).

h eventually led to downstream events that culminated in collagen release. Thus, the model predicted that this intervention may delay or slow down collagen release, but cannot prevent initiation of the disease process. It is therefore likely that such an intervention used alone would not be very beneficial (Figure 3A). Figure 3 Simulation results for interventions with interleukin-1 receptor (IL-1R) or oncostatin M receptor (OSMR) antagonists, using a simulated time period of 14 days. Simulated conditions consisted of IL-1 plus OSM plus matrix metalloproteinase activator. Curves show the percentage of aggrecan and collagen degraded. A, Effect of IL-1R antagonist. B, Effect of OSMR antagonist. Arrows in A and B show the direction of increase in the ratio of receptor antagonist to receptor (1, 10, 100, 1,000). OSMR antagonist is predicted to be less effective than IL-1 receptor antagonist The model predicted that the addition of OSMR antagonists would be much less effective than IL-1R antagonists (Figure 3B). It was necessary to increase the ratio of OSMR antagonist by 107 to have the same effect as a ratio of 103 for IL-1Ra. This is because even low levels of OSMR binding to OSM can lead to the synergistic effect of IL-1 and OSM. The simultaneous addition of IL-1 and OSMR antagonists at the same concentration did not significantly reduce the levels of aggrecan and collagen fragments as compared to using IL-1Ra alone (see Supplementary Table 6, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38297/abstract).

s addition of IL-1 and OSMR antagonists at the same concentration did not significantly reduce the levels of aggrecan and collagen fragments as compared to using IL-1Ra alone (see Supplementary Table 6, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38297/abstract). JAK-1 inhibition is ineffective as an intervention to reduce collagen release In the model, JAK-1 phosphorylated only STAT-3, and so to simulate inhibition of this kinase, we varied the value of the parameter for STAT-3 phosphorylation (kphosSTAT3) from 0 (100% inhibition) to 0.005 (no inhibition). The effect of 100% inhibition was marked by a reduction in collagen release, from ∼10% to <1% on day 14, and a reduction in aggrecan release, from 87.5% to ∼1.0% (Figure 4A). However, 90% inhibition did not lead to any significant reduction in collagen release, and even 98% inhibition only reduced collagen fragments by <1%.

ffect of 100% inhibition was marked by a reduction in collagen release, from ∼10% to <1% on day 14, and a reduction in aggrecan release, from 87.5% to ∼1.0% (Figure 4A). However, 90% inhibition did not lead to any significant reduction in collagen release, and even 98% inhibition only reduced collagen fragments by <1%. Figure 4 Simulation results showing the effect of inhibiting JAK-1, JNK, or p38 activity, using a simulated time period of 14 days. Simulated conditions consisted of IL-1 plus OSM plus MMP activator. Curves show the percentage of aggrecan and collagen degraded. A, Effect of JAK-1 inhibition. B, Effect of p38 inhibition. Arrows show the direction of increase (0–100%, in steps of 10; kphoscFos = 5 × 10−7, 4.5 × 10−7, 4 × 10−7, …, 0) molecules−1 seconds−1). C, Effect of JNK inhibition. Arrows show the direction of increase (0–100%, in steps of 10; kphoscJun = 1 × 10−4, 9 × 10−5, 8 × 10−5, …, 0 molecules−1 seconds−1). D, Effect of JNK inhibition on IL-1 plus OSM–induced cartilage breakdown. Bovine nasal cartilage discs were cultured in the presence of IL-1 (0.5 ng/ml) plus OSM (10 ng/ml) in the presence or absence of the JNK inhibitor SP600125 (SP; 30 μM) or DMSO control. Medium was removed on day 7, and fresh reagents added. On day 14, the medium was removed, and the remaining cartilage was papain digested. The hydroxyproline assay was used to measure the release of collagen into the medium on day 7 and day 14. Values are the mean ± SEM of data accumulated from a minimum of 2 different experiments of a total of 4 experiments conducted. ∗∗ = P < 0.01 for cytokine treatment versus cytokine plus inhibitor treatment, by t-test. See Figure 1 for definitions.

d to measure the release of collagen into the medium on day 7 and day 14. Values are the mean ± SEM of data accumulated from a minimum of 2 different experiments of a total of 4 experiments conducted. ∗∗ = P < 0.01 for cytokine treatment versus cytokine plus inhibitor treatment, by t-test. See Figure 1 for definitions. Inhibition of p38 or JNK produces a moderate reduction in collagen release We simulated the inhibition of p38. In our model, p38 was induced by IL-1 due to phosphorylation by IRAK-2/TRAF6; however, it then acted in the OSM/STAT pathway by subsequently phosphorylating c-Fos, which bound phosphorylated c-Jun to form an AP-1 complex (Figure 5). We found that the model predicted that p38 inhibition had a much greater effect on collagen release than did JAK-1 inhibition (Figure 4B). Previously published data, which were not used in the model construction, demonstrated that p38 inhibition is effective at blocking IL-1–stimulated cartilage collagen release (34). JNK is phosphorylated by IRAK-2/TRAF6, and phosphorylated JNK subsequently phosphorylates c-Jun. The model predicted that inhibition of JNK, by reducing the rate of kphoscFos from 0 to 100%, led to a reduction in collagen release (Figure 4C), and similar to p38, this was a more effective intervention than JAK-1 inhibition (Figure 4A). In order to confirm this prediction, we conducted an in vitro experiment in which we cultured cytokine-stimulated cartilage in the presence of a JNK inhibitor, and showed that cartilage collagen release was almost completely blocked by the JNK inhibitor SP600125 (Figure 4D).

ervention than JAK-1 inhibition (Figure 4A). In order to confirm this prediction, we conducted an in vitro experiment in which we cultured cytokine-stimulated cartilage in the presence of a JNK inhibitor, and showed that cartilage collagen release was almost completely blocked by the JNK inhibitor SP600125 (Figure 4D). Figure 5 Simplified network diagram showing the involvement of JNK and p38 in the system. Interleukin-1 (IL-1) activates tumor necrosis factor receptor–associated factor 6 (TRAF6), which phosphorylates both p38 and JNK. JNK phosphorylates c-Jun, and p38 phosphorylates c-Fos, which has been up-regulated via the oncostatin M (OSM)/JAK-1/STAT-3 signaling pathway. Phosphorylated c-Fos binds to phosphorylated c-Jun to form the activator protein 1 (AP-1) complex. See Supplementary Figures 1 and 2 for diagrams showing all of the reactions (available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38297/abstract). Overexpression of TIMP-3 has a greater beneficial effect than overexpression of TIMP-1 To simulate overexpression of TIMP-1 or TIMP-3, we varied the initial amount of protein from 200 to 200,000 molecules, with 3 intervals on a logarithmic scale (Figure 6). TIMP-1 overexpression delayed the onset of aggrecan and collagen release but only slightly reduced the maximum amount of degradation by day 14 (Figure 6A). Interestingly, TIMP-3 overexpression led to a much greater delay in aggrecan release and decreased the amount of collagen release by day 14 (Figure 6B).

(Figure 6). TIMP-1 overexpression delayed the onset of aggrecan and collagen release but only slightly reduced the maximum amount of degradation by day 14 (Figure 6A). Interestingly, TIMP-3 overexpression led to a much greater delay in aggrecan release and decreased the amount of collagen release by day 14 (Figure 6B). Figure 6 Simulation results showing model predictions for the overexpression of TIMP-1 or TIMP-3 protein, using a simulated time period of 14 days. Simulated conditions consisted of IL-1 plus OSM plus MMP activator. Curves show the percentage of aggrecan and collagen degraded. A, Effect of TIMP-1 overexpression. B, Effect of TIMP-3 overexpression. Arrows in A and B show the direction of increase (2 × 102, 2 × 103, 2 × 104, 2 × 105 molecules). See Figure 1 for definitions. Stochastic effects are important and have implications for treatments It is known that stochastic effects are important in biologic systems, may partly explain the cellular heterogeneity observed in experimental systems, and similarly account for the observed differences in loss of cartilage collagen in patients, in that some patients experience rapid joint destruction while others have relatively slow progression. Factors such as age at disease onset, severity of symptoms, and response to treatments also further contribute to etiology. Therefore, we also ran the model using stochastic simulation.

artilage collagen in patients, in that some patients experience rapid joint destruction while others have relatively slow progression. Factors such as age at disease onset, severity of symptoms, and response to treatments also further contribute to etiology. Therefore, we also ran the model using stochastic simulation. We first compared the output for MMP and TIMP-1 expression over 48 hours with different cytokine treatments and found that although the mean behavior was similar to that of the deterministic model, there was large variability in the expression of MMP-1 mRNA when both IL-1 and OSM were added (see Supplementary Figure 5, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38297/abstract). The level of MMP-1 mRNA was plotted for 50 individual runs, and as can be seen in Supplementary Figure 5D, there was considerable variation in the amount of induction and in the timing of maximal induction. We also ran the model with IL-1 plus OSM with the addition of MMP activator for a simulated time of 14 days. The model predicted a lot of variability in the levels of active MMPs and, hence, variation also in the percentage of collagen released by day 14 (see Supplementary Figure 6). The mean values of 200 stochastic simulations were fairly similar to the deterministic output, although slightly higher values of active MMPs and collagen release were obtained with the stochastic model.

ctive MMPs and, hence, variation also in the percentage of collagen released by day 14 (see Supplementary Figure 6). The mean values of 200 stochastic simulations were fairly similar to the deterministic output, although slightly higher values of active MMPs and collagen release were obtained with the stochastic model. To examine the role of stochastic effects in possible treatments, we carried out stochastic simulation for the TIMP-1 and TIMP-3 overexpression intervention. The model predicted that TIMP-1 overexpression (×103 or more) may significantly delay collagen release but that there would be only a small reduction in the amount of collagen release by day 14 (see Supplementary Figure 7A, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38297/abstract). Conversely, TIMP-3 overexpression (×102 or more) both significantly delayed and reduced the amount of collagen release (see Supplementary Figure 7B). These results are consistent with the deterministic model and could imply that stochastic effects are not important. However, as in the case of no treatments, there was a lot of individual variation in collagen release (as evidenced experimentally [see refs.3–5]), which could be seen when examining individual simulations (see Supplementary Figure 8). However, with TIMP-3 overexpression, the variability declined with increasing levels of TIMP-3, suggesting that collagen release was consistently reduced by this intervention.

en release (as evidenced experimentally [see refs.3–5]), which could be seen when examining individual simulations (see Supplementary Figure 8). However, with TIMP-3 overexpression, the variability declined with increasing levels of TIMP-3, suggesting that collagen release was consistently reduced by this intervention. DISCUSSION 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 (4). 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.

of IL-1 and OSM on TIMP-1 expression (4). 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 (6). 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 (35). We then used the model to simulate various possible interventions, and we examined the effect of these on collagen release.

rized 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 (35). 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 (36), 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.

se in tissue from some patients but not from others (36), 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.

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 (34). 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.

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 (37). Direct inhibition of MMPs with low molecular weight inhibitors proved to be ineffective in patients, as off-target effects were identified (38).

sing strategy to investigate further. Our predictions are supported by experimental data showing limited benefit from overexpressing TIMP-1 in a mouse model of arthritis (37). Direct inhibition of MMPs with low molecular weight inhibitors proved to be ineffective in patients, as off-target effects were identified (38). 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.

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 (17). 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.

lding models is to capture the essential details without burdening the model with nonessential details (17). 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 (39). 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 (38). 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 (40).

sible 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 (40). In conclusion, there is a great need to increase our understanding of the molecular mechanisms involved in cartilage release and to develop new interventions (19). 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. AUTHOR CONTRIBUTIONS 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. Additional Supporting Information may be found in the online version of this article.

Phosphoinositide 3-kinases (PI3Ks) comprise a diverse family of lipid kinases involved in nearly all cellular functions as well as various disease processes ranging from cancer to metabolic and inflammatory diseases (1–6). The best known mammalian PI3Ks are the class I PI3K family members PI3Kα, PI3Kβ, PI3Kγ, and PI3Kδ (2,7). PI3Kα, PI3Kβ, and PI3Kδ mainly relay signals downstream from receptor or nonreceptor tyrosine kinases, whereas PI3Kγ is primarily involved in signal transduction by the βγ subunits of certain G protein–coupled receptors (2,7). The functional role of PI3Ks had initially been addressed by using general PI3K inhibitors such as wortmannin or LY294002. However, more recent studies using genetic deletion approaches and the recent development of isoform-specific PI3K inhibitors have revealed highly specific functions of the different isoforms in certain biologic processes, promising novel therapeutic strategies for various disease states (2,4). We previously demonstrated that PI3Kβ is required for arthritis development in the K/BxN serum–transfer model (8). Since that model mimics the myeloid cell–mediated effector phase of arthritis, PI3Kβ was most likely required in one (or more) myeloid lineage cell types.

The functional role of PI3Ks had initially been addressed by using general PI3K inhibitors such as wortmannin or LY294002. However, more recent studies using genetic deletion approaches and the recent development of isoform-specific PI3K inhibitors have revealed highly specific functions of the different isoforms in certain biologic processes, promising novel therapeutic strategies for various disease states (2,4). We previously demonstrated that PI3Kβ is required for arthritis development in the K/BxN serum–transfer model (8). Since that model mimics the myeloid cell–mediated effector phase of arthritis, PI3Kβ was most likely required in one (or more) myeloid lineage cell types. Osteoclasts are highly specialized bone-resorbing cells of myeloid hematopoietic cell origin (9–11) and are responsible for basal bone resorption as well as pathologic bone loss during inflammatory arthritis, bone metastasis, and postmenopausal osteoporosis. Their role in inflammatory arthritis is indicated by reduced arthritis-induced local bone resorption upon genetic or pharmacologic blockade of osteoclasts, both in experimental mice (12–14) and in patients with rheumatoid arthritis (15).

s during inflammatory arthritis, bone metastasis, and postmenopausal osteoporosis. Their role in inflammatory arthritis is indicated by reduced arthritis-induced local bone resorption upon genetic or pharmacologic blockade of osteoclasts, both in experimental mice (12–14) and in patients with rheumatoid arthritis (15). Osteoclast development and function are directed by a number of extracellular cues including macrophage colony-stimulating factor (M-CSF), RANKL, β3 integrin–mediated adhesive interactions, and immunoreceptor-like activation signals (9–11,16,17). Wortmannin and LY294002 inhibit both the development and the resorptive activity of osteoclasts (18–21). However, the role of the different PI3K isoforms in osteoclast development and function is poorly understood. These issues prompted us to analyze the role of PI3Kβ in primary in vitro osteoclast cultures and in in vivo bone homeostasis, using a combined genetic and pharmacologic approach. Our results indicate that PI3Kβ plays a major role in osteoclast development, osteoclast-mediated bone resorption, and in vivo bone homeostasis, likely due to its participation in the organization of the osteoclast cytoskeleton and the release of intracellular vesicles.

ng a combined genetic and pharmacologic approach. Our results indicate that PI3Kβ plays a major role in osteoclast development, osteoclast-mediated bone resorption, and in vivo bone homeostasis, likely due to its participation in the organization of the osteoclast cytoskeleton and the release of intracellular vesicles. MATERIALS AND METHODS Animals Pik3cbtm1.1Bvan/tm1.1Bvan (PI3Kβ−/−) mice carrying a homozygous deletion of exons 21–22 of Pik3b, the gene encoding the p110β catalytic subunit of PI3Kβ, were described previously (22) and were maintained in heterozygous form on a mixed C57BL/6:129Sv genetic background (backcrossed to C57BL/6 for ∼4 generations). Age- and sex-matched wild-type controls (mostly littermates) were obtained from the same colony. Transgenic mice ubiquitously expressing enhanced green fluorescent protein (EGFP)–tagged Lifeact (23) were provided by Dr. Michael Sixt (Institute of Science and Technology, Klosterneuburg, Austria) and were crossed with PI3Kβ+/+ mice to obtain Lifeact–EGFP–expressing PI3Kβ−/− mice. For pharmacologic studies, C57BL/6 mice were purchased from Charles River.

g enhanced green fluorescent protein (EGFP)–tagged Lifeact (23) were provided by Dr. Michael Sixt (Institute of Science and Technology, Klosterneuburg, Austria) and were crossed with PI3Kβ+/+ mice to obtain Lifeact–EGFP–expressing PI3Kβ−/− mice. For pharmacologic studies, C57BL/6 mice were purchased from Charles River. Due to the limited availability of PI3Kβ−/− mice, bone marrow cells for most in vitro osteoclast and macrophage cultures were obtained from bone marrow chimeras generated by transplanting PI3Kβ−/− (and parallel wild-type control) mouse bone marrow cells to lethally irradiated recipients, as previously described (24,25). No differences have been observed between osteoclast cultures derived from intact mice and those derived from corresponding bone marrow chimeras of either genotype (data not shown). Mice were kept in individually sterile ventilated cages (Tecniplast) in a conventional facility. All animal experiments were approved by the Semmelweis University Animal Experimentation Review Board.

Due to the limited availability of PI3Kβ−/− mice, bone marrow cells for most in vitro osteoclast and macrophage cultures were obtained from bone marrow chimeras generated by transplanting PI3Kβ−/− (and parallel wild-type control) mouse bone marrow cells to lethally irradiated recipients, as previously described (24,25). No differences have been observed between osteoclast cultures derived from intact mice and those derived from corresponding bone marrow chimeras of either genotype (data not shown). Mice were kept in individually sterile ventilated cages (Tecniplast) in a conventional facility. All animal experiments were approved by the Semmelweis University Animal Experimentation Review Board. In vitro culture and resorption assays In vitro osteoclast cultures were performed essentially as previously described (26). Wild-type or PI3Kβ−/− mouse bone marrow cells were first cultured in the presence of 10 ng/ml mouse M-CSF (PeproTech) for 2 days. Nonadherent cells (referred to as myeloid progenitors) were then plated at 2 × 105 cells/cm2 and cultured in the presence of 20 or 50 ng/ml recombinant mouse M-CSF and 20 or 50 ng/ml mouse RANKL (both from PeproTech) with media/cytokine changes every 2 days. Osteoclast morphology was tested 3 days later, using a commercial tartrate-resistant acid phosphatase (TRAP) staining kit (Sigma) and imaged using a Leica DMI6000B inverted microscope, and the number of osteoclasts (defined as TRAP-positive cells with ≥3 nuclei) was counted manually. The diameter of the cells was determined using ImageJ software (National Institutes of Health). For in vitro resorption assays, osteoclasts were cultured under similar conditions for 11 days on an artificial hydroxyapatite layer (BD BioCoat Osteologic slides) or on bovine cortical bone slices (Immunodiagnostic Systems), then processed according to the manufacturer's instructions, followed by imaging and determination of the resorbed area using ImageJ software.

cultured under similar conditions for 11 days on an artificial hydroxyapatite layer (BD BioCoat Osteologic slides) or on bovine cortical bone slices (Immunodiagnostic Systems), then processed according to the manufacturer's instructions, followed by imaging and determination of the resorbed area using ImageJ software. For osteoblast–osteoclast coculture experiments, calvariae of euthanized neonatal wild-type mice were digested by 0.1% collagenase and 0.25% trypsin–EDTA (both from Sigma) in 5 consecutive rounds. The cells isolated during the last 3 rounds were plated at 105 cells/well in 96-well plates and cultured for 2 days in the presence of 10 nM 1,25-dihydroxyvitamin D3 and 10 nM dexamethasone (both from Sigma). Wild-type or PI3Kβ−/− mouse bone marrow cells were then seeded onto the osteoblasts at 5 × 104 cells/well and cultured for 10 days with media changes every 2 days. TRAP expression was then determined as described above. Wild-type and PI3Kβ−/−mouse macrophages were generated by culturing myeloid progenitors in the presence of 50 ng/ml M-CSF without the addition of RANKL.

hen seeded onto the osteoblasts at 5 × 104 cells/well and cultured for 10 days with media changes every 2 days. TRAP expression was then determined as described above. Wild-type and PI3Kβ−/−mouse macrophages were generated by culturing myeloid progenitors in the presence of 50 ng/ml M-CSF without the addition of RANKL. Human osteoclasts were differentiated from peripheral blood mononuclear cells (PBMCs) from healthy volunteers. PBMCs were obtained by dextran sedimentation and centrifugation through Ficoll-Paque (GE Healthcare) as previously described (27). Mononuclear cells were washed and plated at 2 × 105 cells/cm2 to 24-well tissue culture plates or BD BioCoat Osteologic slides and cultured in the presence of 20 or 50 ng/ml recombinant human M-CSF and 20 or 50 ng/ml human RANKL (both from PeproTech) for 14 days with media/cytokine changes every 2 days. TRAP staining and resorption assays were performed as described above. Experiments on human cells were approved by the Semmelweis University Regional and Institutional Committee of Science and Research Ethics. For inhibitor studies, wortmannin (Sigma) and TGX221 (Cayman Chemical) were added either concomitantly with the intial RANKL treatment or 3 days later and then replaced, with media changes every 2 days. Vehicle control samples were treated with 0.1% DMSO or ethanol.

Human osteoclasts were differentiated from peripheral blood mononuclear cells (PBMCs) from healthy volunteers. PBMCs were obtained by dextran sedimentation and centrifugation through Ficoll-Paque (GE Healthcare) as previously described (27). Mononuclear cells were washed and plated at 2 × 105 cells/cm2 to 24-well tissue culture plates or BD BioCoat Osteologic slides and cultured in the presence of 20 or 50 ng/ml recombinant human M-CSF and 20 or 50 ng/ml human RANKL (both from PeproTech) for 14 days with media/cytokine changes every 2 days. TRAP staining and resorption assays were performed as described above. Experiments on human cells were approved by the Semmelweis University Regional and Institutional Committee of Science and Research Ethics. For inhibitor studies, wortmannin (Sigma) and TGX221 (Cayman Chemical) were added either concomitantly with the intial RANKL treatment or 3 days later and then replaced, with media changes every 2 days. Vehicle control samples were treated with 0.1% DMSO or ethanol. Detection of apoptosis For survival analysis, preosteoclasts obtained by culturing mouse myeloid progenitors for 2 days in the presence of 50 ng/ml M-CSF and 50 ng/ml RANKL were suspended by 0.25% trypsin–EDTA (Sigma) and either analyzed immediately or cultured for an additional 12 or 18 hours in α-minimum essential medium in the absence of serum/cytokines. Cells were stained with phycoerythrin-conjugated annexin V and 7-aminoactinomycin D (7-AAD) (Apoptosis Detection Kit; BD PharMingen) according to the manufacturer's instructions and analyzed on a BD FACSCalibur flow cytometer.

an additional 12 or 18 hours in α-minimum essential medium in the absence of serum/cytokines. Cells were stained with phycoerythrin-conjugated annexin V and 7-aminoactinomycin D (7-AAD) (Apoptosis Detection Kit; BD PharMingen) according to the manufacturer's instructions and analyzed on a BD FACSCalibur flow cytometer. For the TUNEL reaction, osteoclast cultures were stained with a Roche In Situ Cell Death Detection Kit, AP, according to the manufacturer's instructions. The number of TUNEL-positive cells was counted manually. Fluorescence microscopy For F-actin staining, mouse myeloid progenitors were cultured for the indicated time periods in the presence of 50 ng/ml M-CSF and 50 ng/ml RANKL with or without the indicated inhibitors, fixed with 4% paraformaldehyde, permeabilized with 0.1% Triton X-100 (Sigma), and stained with 1:400 Alexa Fluor 488–conjugated phalloidin (Invitrogen) and 1:1,000 DAPI (Invitrogen). After several washes, fluorescence was observed using a Leica DMI6000B inverted microscope.

NKL with or without the indicated inhibitors, fixed with 4% paraformaldehyde, permeabilized with 0.1% Triton X-100 (Sigma), and stained with 1:400 Alexa Fluor 488–conjugated phalloidin (Invitrogen) and 1:1,000 DAPI (Invitrogen). After several washes, fluorescence was observed using a Leica DMI6000B inverted microscope. For live imaging of osteoclast development, myeloid progenitors obtained from Lifeact–EGFP–transgenic wild-type or PI3Kβ−/− mice were cultured for the indicated time period in the presence of 50 ng/ml M-CSF and 50 ng/ml RANKL with or without the indicated inhibitors, and imaged using an Essen BioScience IncuCyte Zoom imaging system inside a tissue culture incubator or a Nikon BioStation IM-Q imaging system (Auro-Science Hungary). Videos were generated using IncuCyte Zoom Controller 2013A or Nikon BioStation IM software. For acidic vesicle staining, mouse myeloid progenitors cultured for 3 days in the presence of 50 ng/ml M-CSF and 50 ng/ml RANKL were incubated for 20 minutes with LysoTracker Red (Invitrogen) according to the manufacturer's instructions, then fixed, stained with DAPI, and imaged as described above.

Station IM software. For acidic vesicle staining, mouse myeloid progenitors cultured for 3 days in the presence of 50 ng/ml M-CSF and 50 ng/ml RANKL were incubated for 20 minutes with LysoTracker Red (Invitrogen) according to the manufacturer's instructions, then fixed, stained with DAPI, and imaged as described above. Analysis of gene expression To test gene expression changes, mouse myeloid progenitors were cultured for 0–3 days in the presence of 50 ng/ml M-CSF with or without 50 ng/ml RANKL, followed by RNA extraction and reverse transcription as previously described (26,28). Quantitative reverse transcription–polymerase chain reaction was then performed using TaqMan assays for the mouse genes Pik3ca (Mm00435673_m1), Pik3cb (Mm00659576_m1), Pik3cg (Mm00445038_m1), and Pik3cd (Mm00435674_m1) and the osteoclast-specific genes Acp5, Ctsk, Itgb3, Calcr, Nfatc1, and Tm7sf4 as previously described (26). Transcript levels relative to GAPDH were calculated using the comparative Ct method (26). Cathepsin K secretion Mouse myeloid progenitors were cultured for 3 days in the presence of 50 ng/ml M-CSF and 50 ng/ml RANKL and then stimulated with 50 ng/ml phorbol myristate acetate (Sigma) for 10 minutes. The supernatants were collected, and the proteins were precipitated with acetone. Whole-cell lysates were obtained using Triton X-100–based lysis buffer (29). Samples were subjected to sodium dodecyl sulfate–polyacrylamide gel electrophoresis and immunoblotted with mouse monoclonal antibodies against cathepsin K (E-7; Santa Cruz Biotechnology) with secondary reagents from GE Healthcare.

ith acetone. Whole-cell lysates were obtained using Triton X-100–based lysis buffer (29). Samples were subjected to sodium dodecyl sulfate–polyacrylamide gel electrophoresis and immunoblotted with mouse monoclonal antibodies against cathepsin K (E-7; Santa Cruz Biotechnology) with secondary reagents from GE Healthcare. Micro–computed tomography (micro-CT) analysis Trabecular bone structure and mineralization were tested by micro-CT analysis of the distal metaphysis of the femurs of age- and sex-matched wild-type and PI3Kβ−/− mice essentially as described previously (26). Micro-CT sections were acquired using a SkyScan 1172 micro-CT apparatus with an isometric voxel size of 4.5 μm, followed by reconstitution of a horizontal section 250 sections proximal to the distal growth plate and reconstitution of a 3-dimensional axial cylinder 700 μm in diameter expanding from 150 to 450 sections proximal to the distal growth plate, as well as calculation of quantitative micro-CT parameters using SkyScan NRecon and CT-Analyser software (both from SkyScan) as described previously (26).

e distal growth plate and reconstitution of a 3-dimensional axial cylinder 700 μm in diameter expanding from 150 to 450 sections proximal to the distal growth plate, as well as calculation of quantitative micro-CT parameters using SkyScan NRecon and CT-Analyser software (both from SkyScan) as described previously (26). Histomorphometric analysis Histomorphometry studies were performed on the distal metaphysis of the femurs of age-matched wild-type and PI3Kβ−/− male mice at ages 8–10 weeks. Bones were fixed, decalcified in 14% EDTA, embedded in paraffin, and sectioned and stained with TRAP, toluidine blue, and hematoxylin and eosin. Histomorphometric analysis was performed using a Zeiss Axioskop 2 microscope equipped with a video camera and an OsteoMeasure system (OsteoMetrics) according to international standards as described previously (30). Statistical analysis All experiments were performed ≥3 times (or on ≥3 individual mice), with comparable results. Statistical analysis was performed using Student's unpaired 2-sample t-test.

Histomorphometric analysis Histomorphometry studies were performed on the distal metaphysis of the femurs of age-matched wild-type and PI3Kβ−/− male mice at ages 8–10 weeks. Bones were fixed, decalcified in 14% EDTA, embedded in paraffin, and sectioned and stained with TRAP, toluidine blue, and hematoxylin and eosin. Histomorphometric analysis was performed using a Zeiss Axioskop 2 microscope equipped with a video camera and an OsteoMeasure system (OsteoMetrics) according to international standards as described previously (30). Statistical analysis All experiments were performed ≥3 times (or on ≥3 individual mice), with comparable results. Statistical analysis was performed using Student's unpaired 2-sample t-test. RESULTS PI3Kβ expression during osteoclast development We first tested the expression of the various PI3K isoforms during in vitro differentiation of mouse progenitors in the presence of 50 ng/ml M-CSF with (osteoclasts) or without (macrophages) 50 ng/ml RANKL. As shown in Figure 1A, the expression of PI3Kβ but not of the other PI3K isoforms was dramatically up-regulated during osteoclast differentiation. In contrast, no substantial changes were seen in parallel macrophage cultures (Figure 1B). Therefore, consistent with a recent report (31), PI3Kβ was dramatically and specifically up-regulated during osteoclast development.

not of the other PI3K isoforms was dramatically up-regulated during osteoclast differentiation. In contrast, no substantial changes were seen in parallel macrophage cultures (Figure 1B). Therefore, consistent with a recent report (31), PI3Kβ was dramatically and specifically up-regulated during osteoclast development. Figure 1 Expression of phosphoinositide 3-kinase β (PI3Kβ) during in vitro osteoclast development. Gene expression was analyzed in wild-type mouse myeloid precursors cultured for the indicated time periods in the presence of 50 ng/ml macrophage colony-stimulating factor with 50 ng/ml RANKL (osteoclasts) (A) or without RANKL (macrophages) (B). The expression of the Pik3ca, Pik3cb, Pik3cg, and Pik3cd (encoding the catalytic subunits of PI3Kα, PI3Kβ, PI3Kγ, and PI3Kδ, respectively) was determined by quantitative reverse transcription–polymerase chain reaction. Bars show the mean ± SEM from 3 independent experiments.

sts) (A) or without RANKL (macrophages) (B). The expression of the Pik3ca, Pik3cb, Pik3cg, and Pik3cd (encoding the catalytic subunits of PI3Kα, PI3Kβ, PI3Kγ, and PI3Kδ, respectively) was determined by quantitative reverse transcription–polymerase chain reaction. Bars show the mean ± SEM from 3 independent experiments. Inhibition of PI3Kβ blocks osteoclast development and function in vitro To assess the functional role of PI3Kβ in osteoclasts, we next tested the effect of the PI3Kβ inhibitor TGX221 on in vitro–generated osteoclasts. As shown in Figure 2A, the nonselective PI3K inhibitor wortmannin completely blocked development of osteoclasts (large multinucleated TRAP-positive cells) generated by culturing mouse myeloid progenitors in the presence of 50 ng/ml M-CSF and 50 ng/ml RANKL. TGX221 also dose-dependently inhibited osteoclast development by ∼90% at a concentration of 50 nM, a dose thought to specifically inhibit PI3Kβ (8,32,33) (Figure 2) (further information is available at semmelweis.hu/elettan/en/?p=664).

ring mouse myeloid progenitors in the presence of 50 ng/ml M-CSF and 50 ng/ml RANKL. TGX221 also dose-dependently inhibited osteoclast development by ∼90% at a concentration of 50 nM, a dose thought to specifically inhibit PI3Kβ (8,32,33) (Figure 2) (further information is available at semmelweis.hu/elettan/en/?p=664). Figure 2 Pharmacologic inhibition of phosphoinositide 3-kinase β (PI3Kβ) blocks development and function of murine and human osteoclasts. Shown are representative images and quantification of tartrate-resistant acid phosphatase (TRAP)–stained cell cultures of (A and C), and in vitro resorption pit formation by (B and D) wild-type mouse bone marrow–derived osteoclasts (A and B) and human blood mononuclear cell–derived osteoclasts (C and D) cultured for 3 days (A), 11 days (B), or 14 days (C and D) in the presence of 50 ng/ml macrophage colony-stimulating factor, 50 ng/ml RANKL, and the indicated concentrations of PI3K inhibitors or 0.1% vehicle. Resorption pits appear as lighter areas. Osteoclasts are defined as TRAP-positive cells with ≥3 nuclei. In vitro resorption is defined as the percentage of resorbed area. Bars show the mean ± SEM from 3 independent experiments. Images are representative of 8–20 (A), 4–9 (B), or 3–4 (C and D) independent experiments.

e. Resorption pits appear as lighter areas. Osteoclasts are defined as TRAP-positive cells with ≥3 nuclei. In vitro resorption is defined as the percentage of resorbed area. Bars show the mean ± SEM from 3 independent experiments. Images are representative of 8–20 (A), 4–9 (B), or 3–4 (C and D) independent experiments. We also tested the effect of TGX221 on the resorptive activity of osteoclasts. As shown in Figure 2B, 50 nM wortmannin abrogated the resorption of an artificial hydroxyapatite surface by murine osteoclast cultures generated using 50 ng/ml M-CSF and 50 ng/ml RANKL. TGX221 at 50 nM also nearly completely blocked the resorptive activity of osteoclasts under such conditions (Figure 2B). We next tested the effect of TGX221 on human osteoclasts differentiated from PBMCs in the presence of 50 ng/ml human M-CSF and 50 ng/ml human RANKL. Both wortmannin and TGX221 strongly reduced the number of osteoclasts, with 50 nM TGX221 causing ∼70% inhibition (Figure 2C) (further information is available at semmelweis.hu/elettan/en/?p=664). In addition, 50 nM wortmannin or 50 nM TGX221 dramatically inhibited the in vitro resorptive capacity of human osteoclasts on an artificial hydroxyapatite layer (Figure 2D). Results similar to those described above were obtained when the concentration of M-CSF and RANKL was reduced to 20 ng/ml (further information is available at semmelweis.hu/elettan/en/?p=664). Taken together, these findings show that PI3Kβ likely plays an important role in the in vitro development and resorptive function of both human and mouse osteoclasts.

ve were obtained when the concentration of M-CSF and RANKL was reduced to 20 ng/ml (further information is available at semmelweis.hu/elettan/en/?p=664). Taken together, these findings show that PI3Kβ likely plays an important role in the in vitro development and resorptive function of both human and mouse osteoclasts. In vivo bone homeostasis in PI3Kβ−/− mice To test the role of PI3Kβ in osteoclast biology and bone homeostasis using a genetic approach, we turned to the analysis of PI3Kβ−/− mice carrying a targeted deletion within the catalytic domain of PI3Kβ (22). We first analyzed trabecular bone structure using micro-CT analysis of the distal metaphysis of the femurs at ages 8–10 weeks. As shown in Figure 3A, significantly more trabeculae were seen in PI3Kβ−/− mice, both in representative single micro-CT slices and in 3-dimensional reconstitution of an axial cylinder. Quantification of the entire trabecular area (Figure 3B) revealed significantly increased percent bone volume/total volume (BV/TV) in PI3Kβ−/− mice in both females (P = 0.000029; n = 8) and males (P = 0.000084; n = 8), which was primarily due to increased trabecular number rather than increased thickness of the individual trabeculae. The increased BV:TV ratio was seen across all age groups tested (further information is available at semmelweis.hu/elettan/en/?p=664).

both females (P = 0.000029; n = 8) and males (P = 0.000084; n = 8), which was primarily due to increased trabecular number rather than increased thickness of the individual trabeculae. The increased BV:TV ratio was seen across all age groups tested (further information is available at semmelweis.hu/elettan/en/?p=664). Figure 3 Phosphoinositide 3-kinase β (PI3Kβ) is required for in vivo bone homeostasis in mice. A, Representative single micro–computed tomography (micro-CT) cross-sections (left) and 3-dimensional reconstitution of an axial cylinder (right) of the trabecular area of the distal femoral metaphysis of 9-week-old wild-type (WT) and PI3Kβ−/− female mice. B, Quantitative micro-CT analysis of the trabecular bone architecture of WT and PI3Kβ−/− female and male mice ages 8–10 weeks. C, Representative images of histomorphometric analysis of the trabecular area of WT and PI3Kβ−/− male mice at age 8 weeks. Original magnification × 10. Insets show enlarged view of tartrate-resistant acid phosphatase–stained sections with osteoclasts (arrows) and resorption pits (arrowheads). D, Histomorphometric analysis of the trabecular bone architecture and the number of osteoclasts (OC), the length of attachment of osteoclasts to the trabecular bone surface (OC bone contact length), and the depth of the resorption pits. Data were obtained from 8 (A and B) or 5 (C and D) mice per genotype. Bars show the mean ± SEM. * = P < 0.05; ** = P < 0.01; *** = P < 0.002; **** = P < 0.0004. BV/TV = bone volume/total volume; NS = not significant.

eoclasts to the trabecular bone surface (OC bone contact length), and the depth of the resorption pits. Data were obtained from 8 (A and B) or 5 (C and D) mice per genotype. Bars show the mean ± SEM. * = P < 0.05; ** = P < 0.01; *** = P < 0.002; **** = P < 0.0004. BV/TV = bone volume/total volume; NS = not significant. We also performed histologic and histomorphometric analysis of the trabecular bone of the distal femurs. Trabecular density was increased in PI3Kβ−/− mice (Figure 3C); the quantitative BV:TV ratio (P = 0.0059; n = 5) and trabecular number, but not trabecular thickness, were also increased in PI3Kβ−/− mice (Figure 3D). We next analyzed osteoclasts visible in the TRAP-stained histologic sections. There was a moderate but not statistically significant (P = 0.44; n = 5) reduction in the average number of osteoclasts per bone perimeter in PI3Kβ−/− mice (Figure 3D). Further analysis revealed that osteoclasts in PI3Kβ−/− mouse sections were more rounded (Figure 3C, inset) with significantly shorter bone contact length (P = 0.030; n = 30 osteoclasts) (Figure 3D). In addition, the depth of resorption pits was dramatically reduced in PI3Kβ−/− mouse sections (P = 0.0089; n = 30 osteoclasts) (Figures 3C [inset] and D). Taken together, these results indicate that PI3Kβ−/− mice have increased trabecular bone volume, likely caused by moderately reduced numbers and abnormal morphology/function of osteoclasts.

f resorption pits was dramatically reduced in PI3Kβ−/− mouse sections (P = 0.0089; n = 30 osteoclasts) (Figures 3C [inset] and D). Taken together, these results indicate that PI3Kβ−/− mice have increased trabecular bone volume, likely caused by moderately reduced numbers and abnormal morphology/function of osteoclasts. PI3Kβ deficiency impairs osteoclast development and function in vitro Next, we tested the effect of the PI3Kβ−/− mutation on in vitro osteoclast development and function. As shown in Figure 4A, PI3Kβ deficiency led to reduced numbers of osteoclasts generated in the presence of 50 ng/ml M-CSF and 50 ng/ml RANKL (P = 0.0015; n = 5), and the average diameter of those cells was even more dramatically reduced (P = 0.00083; n = 5). Similar results were obtained when the concentration of M-CSF and RANKL was reduced to 20 ng/ml (further information is available at semmelweis.hu/elettan/en/?p=664). Fluorescence labeling of DNA (Figure 4A) also revealed significantly decreased numbers of nuclei per osteoclast (P = 0.018; n = 6).

(P = 0.00083; n = 5). Similar results were obtained when the concentration of M-CSF and RANKL was reduced to 20 ng/ml (further information is available at semmelweis.hu/elettan/en/?p=664). Fluorescence labeling of DNA (Figure 4A) also revealed significantly decreased numbers of nuclei per osteoclast (P = 0.018; n = 6). Figure 4 Genetic deficiency of PI3Kβ leads to defective osteoclast development and function. Shown are representative images and quantification of tartrate-resistant acid phosphatase–stained cell cultures of (A and B) and in vitro resorption pit formation on artificial hydroxyapatite (C) or on bovine bone slices (D) by, WT and PI3Kβ−/− mouse bone marrow–derived osteoclasts cultured for 3 days (A), 10 days (B), or 11 days (C and D) in the presence of 50 ng/ml macrophage colony-stimulating factor and 50 ng/ml RANKL (A, C, and D) or with WT mouse calvarial osteoblasts (B). Bars show the mean ± SEM from 5–6 independent experiments. Images are representative of 8–10 (A), 3–4 (B), or 4–5 (C and D) independent experiments. * = P < 0.05; *** = P < 0.002; **** = P < 0.0004. See Figure 3 for definitions. We also tested osteoclast differentiation in osteoblast–osteoclast cocultures. We observed a moderate but not statistically significant (P = 0.12; n = 5) reduction of the number of osteoclasts differentiated from PI3Kβ−/− mouse bone marrow cells. However, the diameter of osteoclasts was dramatically reduced in PI3Kβ−/− mouse cell cultures (P = 0.00012; n = 5).

in osteoblast–osteoclast cocultures. We observed a moderate but not statistically significant (P = 0.12; n = 5) reduction of the number of osteoclasts differentiated from PI3Kβ−/− mouse bone marrow cells. However, the diameter of osteoclasts was dramatically reduced in PI3Kβ−/− mouse cell cultures (P = 0.00012; n = 5). We next tested the resorptive activity of PI3Kβ−/− mouse cell cultures. PI3Kβ deficiency strongly reduced the resorptive capacity of in vitro osteoclast cultures in the presence of 50 ng/ml M-CSF and 50 ng/ml RANKL, both on an artificial hydroxyapatite surface (P = 0.00042; n = 5) (Figure 4C) and on bovine bone slices (P = 6.2 × 10−6; n = 5) (Figure 4D). Results similar to those described above were obtained when the concentration of M-CSF and RANKL was reduced to 20 ng/ml (further information is available at semmelweis.hu/elettan/en/?p=664). Taken together, these findings demonstrate that genetic deficiency of PI3Kβ impairs in vitro development and resorptive function of osteoclasts.

to those described above were obtained when the concentration of M-CSF and RANKL was reduced to 20 ng/ml (further information is available at semmelweis.hu/elettan/en/?p=664). Taken together, these findings demonstrate that genetic deficiency of PI3Kβ impairs in vitro development and resorptive function of osteoclasts. PI3Kβ is not required for osteoclast-specific gene expression To better understand the role of PI3Kβ in osteoclasts, we next tested the expression of osteoclast-specific genes during differentiation of murine myeloid progenitors in the presence of 50 ng/ml M-CSF and 50 ng/ml RANKL (osteoclasts) or M-CSF alone (macrophages). As shown in Figure 5A, the expression of Acp5 (encoding TRAP), Ctsk (encoding cathepsin K), Itgb3 (encoding integrin β3 chain), Nfatc1 (encoding NF-ATc1), Calcr (encoding calcitonin receptor), and Tm7sf4 (encoding dendritic cell–specific transmembrane protein) was strongly increased during osteoclast differentiation but not during macrophage differentiation. PI3Kβ deficiency did not affect the expression of any of those genes (Figure 5A), indicating that PI3Kβ is not required for osteoclast-specific gene expression.

f4 (encoding dendritic cell–specific transmembrane protein) was strongly increased during osteoclast differentiation but not during macrophage differentiation. PI3Kβ deficiency did not affect the expression of any of those genes (Figure 5A), indicating that PI3Kβ is not required for osteoclast-specific gene expression. Figure 5 PI3Kβ is not required for up-regulation of osteoclast-specific gene expression and survival of osteoclasts. A, Gene expression in WT and PI3Kβ−/− mouse bone marrow–derived cells cultured for the indicated time periods in the presence of 50 ng/ml macrophage colony-stimulating factor (M-CSF) with 50 ng/ml RANKL (osteoclasts; OC) or without RANKL (macrophages; MΦ). The expression of Acp5, Ctsk, Itgb3, Nfatc1, Calcr, and Tm7sf4 (encoding tartrate-resistant acid phosphatase, cathepsin K, integrin β3 chain, NF-ATc1, calcitonin receptor, and dendritic cell–specific transmembrane protein, respectively) was determined by quantitative reverse transcription–polymerase chain reaction. B and C, Representative flow cytometric profiles (B) and quantification (C) of the binding of phycoerythrin (PE)–conjugated annexin V (apoptosis marker) and 7-aminoactinomycin D (7-AAD) (necrosis marker) to WT and PI3Kβ−/− mouse preosteoclasts (generated by culturing myeloid precursors in the presence of 50 ng/ml M-CSF and 50 ng/ml RANKL for 2 days) before (0 hours) or after serum and cytokine starvation for 12 or 18 hours. Surviving cells are defined as negative for both PE-conjugated annexin V and 7-AAD staining. D, Analysis of the number of TUNEL-positive cells in WT and PI3Kβ−/− mouse bone marrow–derived osteoclast cultures. Data are from 3 (A) or 5-6 (B–D) independent experiments. Bars show the mean ± SEM. See Figure 3 for other definitions.

g cells are defined as negative for both PE-conjugated annexin V and 7-AAD staining. D, Analysis of the number of TUNEL-positive cells in WT and PI3Kβ−/− mouse bone marrow–derived osteoclast cultures. Data are from 3 (A) or 5-6 (B–D) independent experiments. Bars show the mean ± SEM. See Figure 3 for other definitions. Normal survival and apoptosis of PI3Kβ−/− mouse osteoclast-lineage cells To test the role of PI3Kβ deficiency in survival and apoptosis of osteoclast-lineage cells, preosteoclasts were generated by culturing myeloid progenitors in 50 ng/ml M-CSF and 50 ng/ml RANKL for 2 days, followed by withdrawal of serum and cytokines for 12 or 18 hours. As shown in Figures 5B and C, ∼90% of the preosteoclasts before serum/cytokine withdrawal (0-hour samples in Figure 5B) were negative for the apoptosis marker annexin V and the necrosis marker 7-AAD, whereas serum/cytokine withdrawal triggered apoptosis and then necrosis of the cells. No difference in any of those processes was observed between cultures of cells from wild-type and PI3Kβ−/− mice (Figures 5B and C). We also tested apoptosis in regular osteoclast cultures by in situ TUNEL staining. As shown in Figure 5D, ∼12% of cells were TUNEL positive in cultures of osteoclasts from both wild-type and PI3Kβ−/− mice. Taken together, the observations suggest that PI3Kβ deficiency does not affect survival, apoptosis, or necrosis of osteoclast-lineage cells.

n regular osteoclast cultures by in situ TUNEL staining. As shown in Figure 5D, ∼12% of cells were TUNEL positive in cultures of osteoclasts from both wild-type and PI3Kβ−/− mice. Taken together, the observations suggest that PI3Kβ deficiency does not affect survival, apoptosis, or necrosis of osteoclast-lineage cells. PI3Kβ is required for actin ring formation We next tested whether PI3Kβ is required for the formation of the osteoclast actin ring, which is likely involved in sealing the resorption pit. A continuous F-actin ring was observed at the periphery of most wild-type mouse osteoclasts in the presence of 50 ng/ml M-CSF and 50 ng/ml RANKL (Figures 6A and C). In contrast, even multinucleated cells in PI3Kβ−/− mouse osteoclast cultures rarely showed a clear F-actin ring, and F-actin was instead accumulated in the cytoplasm and distinct patches at the cell periphery (P = 7.8 × 10−8; n = 6).

e osteoclasts in the presence of 50 ng/ml M-CSF and 50 ng/ml RANKL (Figures 6A and C). In contrast, even multinucleated cells in PI3Kβ−/− mouse osteoclast cultures rarely showed a clear F-actin ring, and F-actin was instead accumulated in the cytoplasm and distinct patches at the cell periphery (P = 7.8 × 10−8; n = 6). Figure 6 PI3Kβ is required for actin ring formation and cathepsin K secretion by murine osteoclasts. A and C (left), Representative fluorescence images (A) and quantification (C [left]) of WT and PI3Kβ−/− mouse myeloid precursors cultured in the presence of 50 ng/ml macrophage colony-stimulating factor and 50 ng/ml RANKL for 3 days and then stained with Alexa Fluor 488–conjugated phalloidin and DAPI. Arrows in A show the actin ring. Values in C (left) are the percentage of osteoclasts with closed actin rings at the cell periphery. B and C (right), Representative fluorescence images (B) and quantification (C [right]) of WT and PI3Kβ−/− mouse myeloid precursors cultured under conditions similar to those described in A and then stained with LysoTracker Red and DAPI. An acidic vesicle is marked with an arrow in B. Values in C (right) are the number of acidic vesicles. Boxed areas at left in A and B are shown at higher magnification at right. D, Immunoblot (left) and densitometric quantification (right) of cathepsin K (Ctsk) in the supernatant or whole-cell lysates (WCL) of WT and PI3Kβ−/− mouse bone marrow cells cultured as described in A. Fluorescence images are representative of 4–6 independent experiments; immunoblot is representative of 4 independent experiments. Bars show the mean ± SEM from 3–6 independent experiments. ** = P < 0.01; **** = P < 0.0004. See Figure 3 for other definitions.

PI3Kβ−/− mouse bone marrow cells cultured as described in A. Fluorescence images are representative of 4–6 independent experiments; immunoblot is representative of 4 independent experiments. Bars show the mean ± SEM from 3–6 independent experiments. ** = P < 0.01; **** = P < 0.0004. See Figure 3 for other definitions. We also assessed temporal changes of actin polymerization using transgenic mice expressing EGFP-tagged Lifeact, a short peptide specifically binding to fibrillar actin. As shown in Supplementary Videos 1 and 2 (available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38660/abstract) (further information is available at semmelweis.hu/elettan/en/?p=664), osteoclasts derived from Lifeact–EGFP–expressing wild-type mouse progenitor cells began to form continuous F-actin rings ∼2 days after addition of RANKL, and these remained visible until the osteoclasts succumbed to apoptosis. In contrast, osteoclast-like cells from Lifeact–EGFP–transgenic PI3Kβ−/− mice failed to form actin rings and instead showed patchy/dispersed Lifeact distribution. Taken together, these findings indicate that PI3Kβ plays an important role in the formation of the osteoclast actin ring.

osteoclasts succumbed to apoptosis. In contrast, osteoclast-like cells from Lifeact–EGFP–transgenic PI3Kβ−/− mice failed to form actin rings and instead showed patchy/dispersed Lifeact distribution. Taken together, these findings indicate that PI3Kβ plays an important role in the formation of the osteoclast actin ring. PI3Kβ−/− mouse osteoclasts retain acidic vesicles and fail to release cathepsin K During bone resorption, osteoclasts deliver and exocytose lysosome-related, cathepsin K–containing acidic vesicles to the ruffled border (34). Since prior studies proposed a role for PI3K activity in this process (31,35), we tested the distribution of acidic vesicles in PI3Kβ−/− mouse cell cultures in the presence of 50 ng/ml M-CSF and 50 ng/ml RANKL. While wild-type mouse osteoclasts contained very few small acidic vesicles, PI3Kβ−/− mouse cells were fully packed with such vesicles, suggesting that they cannot be discharged and therefore are retained in PI3Kβ−/− mouse cells (P = 0.0021; n = 5) (Figures 6B and C). Exocytosis of the matrix-degrading enzyme cathepsin K plays an important role in osteoclast-mediated bone resorption. As shown in Figure 6D, the PI3Kβ−/− mutation substantially reduced the amount of cathepsin K in the supernatant of osteoclast cultures (P = 0.0064; n = 3). Taken together, these results show that the absence of PI3Kβ leads to defective actin ring formation, accumulation of acidic vesicles, and impaired release of cathepsin K into the extracellular space.

mutation substantially reduced the amount of cathepsin K in the supernatant of osteoclast cultures (P = 0.0064; n = 3). Taken together, these results show that the absence of PI3Kβ leads to defective actin ring formation, accumulation of acidic vesicles, and impaired release of cathepsin K into the extracellular space. Inhibition of PI3Kβ after osteoclast formation blocks resorption and actin ring maintenance The above results suggested that PI3Kβ not only is involved in osteoclast development but also plays important functional roles in mature osteoclasts. To test that possibility, we investigated the effect of PI3Kβ inhibition after the formation of mature osteoclasts. As shown at semmelweis.hu/elettan/en/?p=664, adding 50 nM TGX221 to osteoclast cultures starting from 3 days after the initial RANKL treatment strongly inhibited resorption of an artificial hydroxyapatite layer (P = 0.0093; n = 3). As shown at semmelweis.hu/elettan/en/?p=664, treatment of osteoclasts with 50 nM TGX221 for 6 hours starting 3 days after the initial RANKL treatment (when mature osteoclasts with complete actin rings have already been formed) led to the disappearance of actin rings (P = 2.4 × 10−4; n = 3). Kinetic analysis of the effect of TGX221 using Lifeact–EGFP–expressing cells revealed that the actin rings began to resolve immediately after TGX221 treatment and became highly fragmented within a few hours (further information is available at semmelweis.hu/elettan/en/?p=664) (see Supplementary Videos 3 and 4, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38660/abstract). Taken together, these observations suggest that inhibition of PI3Kβ blocks the resorptive function and actin ring maintenance of in vitro osteoclast cultures even if performed several days after initial RANKL treatment.

Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38660/abstract). Taken together, these observations suggest that inhibition of PI3Kβ blocks the resorptive function and actin ring maintenance of in vitro osteoclast cultures even if performed several days after initial RANKL treatment. DISCUSSION Recent studies revealed highly specific functions of the various PI3K isoforms, triggering development of isoform-specific PI3K inhibitors for therapeutic purposes (2,36). Although prior studies suggested a role for PI3K activity in osteoclast development and bone resorption (18–21), little is known about the specific PI3K isoform(s) participating in those responses. Our initial gene expression studies (Figure 1) suggested a functional role for PI3Kβ in osteoclasts, which was tested using a detailed genetic and pharmacologic approach. Those studies revealed a critical role for PI3Kβ in in vitro development and resorbing function of primary human and mouse osteoclasts (Figures 2 and 4) and in in vivo bone homeostasis in experimental mice (Figure 3).

l role for PI3Kβ in osteoclasts, which was tested using a detailed genetic and pharmacologic approach. Those studies revealed a critical role for PI3Kβ in in vitro development and resorbing function of primary human and mouse osteoclasts (Figures 2 and 4) and in in vivo bone homeostasis in experimental mice (Figure 3). Reduced osteoclast-mediated bone resorption may be due to failure of osteoclast development or defective resorptive function of mature osteoclasts. While PI3Kβ appears to play a role in osteoclast development under certain conditions (Figures 2A and C, Figures 4A and B), the most dramatic in vitro effects of PI3Kβ inactivation were observed when testing resorptive function or cellular parameters related to bone resorption (Figures 2, 4, and 6). Osteoclasts from PI3Kβ−/− mice also showed reduced bone contact length and resorption pit depth in vivo (Figure 3D). Furthermore, addition of TGX221 to cultures with mature osteoclasts inhibited bone resorption and led to rapid disassembly of preexisting actin rings (further information is available at semmelweis.hu/elettan/en/?p=664). Those results suggest that the most critical role for PI3Kβ is to mediate the resorptive function of mature osteoclasts.

of TGX221 to cultures with mature osteoclasts inhibited bone resorption and led to rapid disassembly of preexisting actin rings (further information is available at semmelweis.hu/elettan/en/?p=664). Those results suggest that the most critical role for PI3Kβ is to mediate the resorptive function of mature osteoclasts. We recently reported the analysis of phospholipase Cγ2–deficient (PLCγ2−/−) mouse osteoclast cultures and bone morphology (26). It is interesting to note that while PI3Kβ−/− and PLCγ2−/− mice had similar increases in trabecular bone mass, PLCγ2−/− mouse cells showed a much more robust in vitro osteoclast developmental defect than PI3Kβ−/− mouse cells. Those results again suggest that the primary defect in PI3Kβ−/− mice lies in the bone-resorbing function of osteoclasts.

that while PI3Kβ−/− and PLCγ2−/− mice had similar increases in trabecular bone mass, PLCγ2−/− mouse cells showed a much more robust in vitro osteoclast developmental defect than PI3Kβ−/− mouse cells. Those results again suggest that the primary defect in PI3Kβ−/− mice lies in the bone-resorbing function of osteoclasts. The exact position of PI3Kβ in osteoclast signal transduction pathways is at present poorly understood. Osteoclast development and function are triggered through a large number of extracellular signals (9–11,16,17). Our initial experiments suggested that PI3Kβ is required for phosphorylation of Akt in response to M-CSF treatment (data not shown). However, a more detailed analysis of signaling by the various cell surface receptors and downstream signaling intermediates, as well as high-resolution imaging of phosphatidylinositol 3,4,5-trisphosphate generation in osteoclast cultures, would be needed to define the exact position of PI3Kβ in intracellular signaling of osteoclasts. Nevertheless, our findings indicate that PI3Kβ activation participates in reorganization of the actin cytoskeleton and release of cathepsin K–containing acidic vesicles (Figure 6) but not in gene expression changes or survival/apoptosis of the cells (Figure 5).

sition of PI3Kβ in intracellular signaling of osteoclasts. Nevertheless, our findings indicate that PI3Kβ activation participates in reorganization of the actin cytoskeleton and release of cathepsin K–containing acidic vesicles (Figure 6) but not in gene expression changes or survival/apoptosis of the cells (Figure 5). The results of prior studies using nonspecific PI3K inhibitors suggested a role for PI3Ks in osteoclast development and function (18–21). A few reported attempts to narrow the list of PI3K isoform(s) involved failed to reveal a major role for any of the specific PI3K isoforms, showed contradictory results, and were even inconsistent with a role for PI3Kβ (31,37,38). Our substantially more detailed experiments firmly identify PI3Kβ as a critical and likely predominant PI3K isoform involved in osteoclast development and function. There is substantial interest in the pharmaceutical industry in the development of isoform-specific PI3K inhibitors for the treatment of various human diseases (1,3–5). Based on its role in platelet activation (32,39–43), PI3Kβ has been proposed to be a suitable target for the treatment of thrombotic diseases (44). Based on the results presented here, PI3Kβ may also be a suitable therapeutic target in diseases characterized by excessive osteoclast-mediated bone resorption such as rheumatoid arthritis, osteoporosis, or cancer-induced focal bone loss.

n proposed to be a suitable target for the treatment of thrombotic diseases (44). Based on the results presented here, PI3Kβ may also be a suitable therapeutic target in diseases characterized by excessive osteoclast-mediated bone resorption such as rheumatoid arthritis, osteoporosis, or cancer-induced focal bone loss. We thank Edina Simon, Anna Tóth, and László Gölle for expert technical assistance; Michael Sixt for the Lifeact–EGFP–transgenic mice; Essen BioScience for the IncuCyte Zoom system; Auro-Science Hungary for the Nikon BioStation IM-Q system; Bence Szabó for micro-CT scanning; Birgit Niederreiter for slide preparation for histomorphometry; and Norbert Gyöngyösi for help with statistical analysis. AUTHOR CONTRIBUTIONS 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. Mócsai 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. Győri, Kulkarni, Stephens, Hawkins, Mócsai. Acquisition of data. Győri, Csete, Benkő, Dobó-Nagy. Analysis and interpretation of data. Győri, Csete, Benkő, Kulkarni, Mandl, Dobó-Nagy, Vanhaesebroeck, Stephens, Hawkins, Mócsai.

Juvenile idiopathic arthritis (JIA) is the most common form of autoimmune rheumatic disease in childhood, with a prevalence rate of 1/1,000 children under the age of 16 years (1). The successful introduction of therapies targeting tumor necrosis factor α (TNFα) has led to significant improvement in JIA outcomes. However, in one-third of patients the disease remains resistant or only partially responsive to anti-TNFα therapy, suggesting ongoing uncontrolled immunopathology that is independent of TNFα (2). The identification of a novel CD4+ T cell subset expressing interleukin-17 (IL-17) in a mouse model of arthritis led many to suggest that these cells (Th17 cells) have a role in human disease (3). We and others have demonstrated major enrichment of Th17 cells in the inflamed joints of children with JIA, with a correlation between the frequency of these cells and the severity of disease (4,5). It was therefore unexpected when data from studies of IL-17–deficient mice suggested that IL-17 was redundant for induction of autoimmunity in a mouse model of multiple sclerosis, and that granulocyte–macrophage colony-stimulating factor (GM-CSF) was instead necessary and sufficient for disease (6,7).

of disease (4,5). It was therefore unexpected when data from studies of IL-17–deficient mice suggested that IL-17 was redundant for induction of autoimmunity in a mouse model of multiple sclerosis, and that granulocyte–macrophage colony-stimulating factor (GM-CSF) was instead necessary and sufficient for disease (6,7). GM-CSF is similarly important in mouse models of arthritis and is found in high concentrations in the synovial fluid (SF) of patients with rheumatoid arthritis and JIA (8–10) It has widespread effects, promoting granulopoiesis and activating neutrophils, monocytes, and macrophages that contribute to joint inflammation and damage (9,10). Although GM-CSF is widely expressed in both stromal and hematopoietic compartments, recent murine studies suggest that GM-CSF from the hematopoietic compartment, particularly CD4+ T cells, is essential for disease (6,7,11). In mice, GM-CSF–secreting T cells are closely linked with the Th17 lineage downstream of retinoic acid receptor–related orphan nuclear receptor γt (RORγt) (murine homolog of RORC2), although data on transcriptional control of GM-CSF are conflicting (6,7). Activated human IL-17+ T cell clones produce GM-CSF (12), but the regulation of GM-CSF production in terms of response to the IL-12/IL-23 axis remains unknown, as does the exact relationship between GM-CSF– and IL-17–secreting cells. To date, evidence for the putative role of T cell–derived GM-CSF in autoimmune disease comes largely from murine studies. In the present study we examined this issue in human autoimmune arthritis.

response to the IL-12/IL-23 axis remains unknown, as does the exact relationship between GM-CSF– and IL-17–secreting cells. To date, evidence for the putative role of T cell–derived GM-CSF in autoimmune disease comes largely from murine studies. In the present study we examined this issue in human autoimmune arthritis. MATERIALS AND METHODS Patients and controls Samples studied were from 24 children who met the International League Against Rheumatism criteria for JIA (13) (21 with oligoarticular disease, 3 with polyarticular disease) and 13 adult healthy controls. Seventeen of the JIA patients were female and 7 were male; the median age was 10.8 years. The study was approved by the local ethical review committee, and full informed consent was obtained from patients/parents and control subjects.

oligoarticular disease, 3 with polyarticular disease) and 13 adult healthy controls. Seventeen of the JIA patients were female and 7 were male; the median age was 10.8 years. The study was approved by the local ethical review committee, and full informed consent was obtained from patients/parents and control subjects. Cell sorting and flow cytometry Peripheral blood mononuclear cells (PBMCs) and SF mononuclear cells (SFMCs) were isolated by density centrifugation. For analysis of T cell cytokine production, cells were cultured for 4 hours in the presence of 50 ng/ml phorbol myristate acetate (PMA), 500 ng/ml ionomycin, and 5 μg/ml brefeldin A (Sigma-Aldrich) before intracellular cytokine staining as previously described (4). Antibodies against the following human proteins were used: PC7-conjugated CD4 (Beckman Coulter), phycoerythrin-conjugated CD161 (eBioscience), fluorescein isothiocyanate– or v450-conjugated interferon-γ (IFNγ) (both from BD Biosciences), Alexa Fluor 488–conjugated IL-17A (eBioscience), and Alexa Fluor 647–conjugated GM-CSF (eBioscience). A Live/Dead discriminant dye was used according to the instructions of the manufacturer (Life Technologies). To capture cytokine-expressing cells, PBMCs or SFMCs were enriched for CD4+ T cells using magnetic beads (StemCell Technologies) and stimulated for 2 hours with PMA (10 ng/ml) and ionomycin (1 μg/ml). IL-17–secreting CD4+ T cells were detected according to the instructions of the manufacturer (Miltenyi Biotec) and sorted with a flow cytometer (FACSAria; BD PharMingen). The purity of sorted cells was assessed by detection of intracellular cytokines after overnight incubation in brefeldin A. Flow cytometric data were collected on an LSRII (BD PharMingen); a minimum of 1 × 105 events were collected. Data were analyzed using FlowJo (Tree Star).

a flow cytometer (FACSAria; BD PharMingen). The purity of sorted cells was assessed by detection of intracellular cytokines after overnight incubation in brefeldin A. Flow cytometric data were collected on an LSRII (BD PharMingen); a minimum of 1 × 105 events were collected. Data were analyzed using FlowJo (Tree Star). Polymerase chain reaction RNA was extracted using TRIzol, according to the instructions of the manufacturer (Life Technologies). Generation of complementary DNA, reverse transcription—polymerase chain reaction, and analysis by the Ct method were performed as previously described (14). Primers used were as follows: actin forward 5′-AGA-TGA-CCC-AGA-TCA-TGT-TTG-AG-3′, reverse 5′-AGG-TCC-AGA-CGC-AGG-ATG-3′; RORC2 forward 5′-GAC-CAC-CCC-CTG-CTG-AGA-A-3′, reverse 5′-GAC-ATG-CGG-CCG-AAC-TTG-A-3′; T-bet forward 5′-CCC-CAA-GGA-ATT-GAC-AGT-TG-3′, reverse 5′-GGG-AAA-CTA-AAG-CTC-ACA-AAC-3′; and GM-CSF QuantiTect primers (Qiagen). Levels of messenger RNA (mRNA) were normalized to those of β-actin mRNA. Cell culture Sorted cells were cultured in RPMI with 10% fetal calf serum (Life Technologies) in the presence of IL-2 alone or in combination with IL-12 (from R&D Systems) or IL-23 (eBioscience) (all at 10 ng/ml). On day 4, cytokine expression was detected by flow cytometry after restimulation with PMA and ionomycin in the presence of brefeldin A.

e cultured in RPMI with 10% fetal calf serum (Life Technologies) in the presence of IL-2 alone or in combination with IL-12 (from R&D Systems) or IL-23 (eBioscience) (all at 10 ng/ml). On day 4, cytokine expression was detected by flow cytometry after restimulation with PMA and ionomycin in the presence of brefeldin A. Statistical analysis Data were analyzed using GraphPad Prism 6. The significance of differences between group means was assessed by one-way or two-way analysis of variance, with Bonferroni correction for multiple comparisons. Correlations were assessed using Pearson's correlation coefficient. P values less than 0.05 were considered significant.

were analyzed using GraphPad Prism 6. The significance of differences between group means was assessed by one-way or two-way analysis of variance, with Bonferroni correction for multiple comparisons. Correlations were assessed using Pearson's correlation coefficient. P values less than 0.05 were considered significant. RESULTS Historically, synovial GM-CSF production has been linked with synovial fibroblasts and innate immune cells, chiefly monocytes and tissue macrophages (9). In experiments focusing on the mononuclear cell compartment, we analyzed unsorted SFMCs directly ex vivo and detected significantly higher expression of GM-CSF mRNA in SFMCs from patients with JIA than in PBMCs from the patients (Figure 1A). To compare the contribution of T cells and monocytes, we analyzed GM-CSF mRNA expression in sorted synovial monocytes and CD4+ T cells and found that the levels were comparable (Figure 1B), reinforcing the importance of T cell–derived GM-CSF within the joint. The frequency of GM-CSF–secreting T helper cells was quantified by flow cytometry and shown to be significantly enriched within the joint (mean frequency 24.1% of CD4+ T cells) compared to PBMCs from JIA patients or healthy controls (2.9% and 5.4%, respectively) (Figures 1C and D). The frequency of GM-CSF CD4+ T cells correlated directly with levels of GM-CSF protein in SF and also with levels of inflammation as determined by the erythrocyte sedimentation rate (ESR) (r2 = 0.91, P < 0.001), whereas the frequency of IFNγ+ T helper cells in SF did not correlate with the ESR (r2 = 0.16, P = 0.2) (Figures 1E and F).

f GM-CSF CD4+ T cells correlated directly with levels of GM-CSF protein in SF and also with levels of inflammation as determined by the erythrocyte sedimentation rate (ESR) (r2 = 0.91, P < 0.001), whereas the frequency of IFNγ+ T helper cells in SF did not correlate with the ESR (r2 = 0.16, P = 0.2) (Figures 1E and F). Figure 1 Enrichment of granulocyte–macrophage colony-stimulating factor (GM-CSF) expression within the arthritic joint. A and B, Expression of GM-CSF mRNA (normalized to mRNA for β-actin) in peripheral blood mononuclear cells (PBMCs) and synovial fluid mononuclear cells (SFMCs) from patients with juvenile idiopathic arthritis (JIA) (n = 3) (A) and in whole SFMCs, synovial CD14+ monocytes, and synovial CD4+ T cells from patients with JIA (n = 7) (B). Values are the mean ± SEM. C, Representative dot plots of GM-CSF expression, detected by flow cytometry, in JIA PBMCs and SFMCs following stimulation with phorbol myristate acetate and ionomycin in the presence of brefeldin A. All plots were gated on CD4+ T cells. D, GM-CSF expression in CD4+ T cells (detected as described in C) in healthy control (HC) PBMCs (n = 13) and in JIA PBMCs (n = 8) and SFMCs (n = 13). Symbols represent individual samples; horizontal lines show the mean. ∗∗ = P < 0.01 by one-way analysis of variance (ANOVA) with Bonferroni correction for multiple comparisons. E, Pearson's correlation between GM-CSF protein levels (measured by enzyme-linked immunosorbent assay) in synovial supernatants and the frequency of GM-CSF+ CD4+ T cells in paired SFMC samples. F, Pearson's correlation between the frequency of synovial interferon-γ–positive (IFNγ+) or GM-CSF+ CD4+ T cells and the serum erythrocyte sedimentation rate (ESR) (measured at the time of therapeutic arthrocentesis). G, Representative dot plots of GM-CSF and CD161 expression, detected by flow cytometry, in CD4+ T cells from JIA PBMCs and SFMCs. H, Proportion of the total CD4+ T cell population and the interleukin-17–positive (IL-17+), GM-CSF+, and IFNγ+ subsets expressing CD161 (detected as described in G) in JIA PBMCs (open bars) (n = 7) and SFMCs (solid bars) (n = 7). Values are the mean ± SEM. ∗∗ = P < 0.01 by two-way ANOVA with Bonferroni correction for multiple comparisons. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.38647/abstract.

ribed in G) in JIA PBMCs (open bars) (n = 7) and SFMCs (solid bars) (n = 7). Values are the mean ± SEM. ∗∗ = P < 0.01 by two-way ANOVA with Bonferroni correction for multiple comparisons. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.38647/abstract. In mice, GM-CSF expression is associated with the Th17 differentiation pathway (6). This led us to investigate whether GM-CSF secretion would be limited to T cells expressing CD161, a cell surface C-type lectin-like receptor that is closely associated with human Th17 cells and RORC2 (15). We found that T cells expressing GM-CSF were enriched for CD161 compared to the total CD4+ T cell population but had lower levels of CD161 expression than IL-17+ T cells (Figures 1G and H). Consistent with our previous data (14), the proportion of T cells expressing CD161 was increased in the joint compared to levels in PBMCs, and this was also the case for GM-CSF+ T cells (Figure 1H).

ared to the total CD4+ T cell population but had lower levels of CD161 expression than IL-17+ T cells (Figures 1G and H). Consistent with our previous data (14), the proportion of T cells expressing CD161 was increased in the joint compared to levels in PBMCs, and this was also the case for GM-CSF+ T cells (Figure 1H). Given that GM-CSF+ T cells were enriched within the CD161+ cell compartment, we predicted that they may coexpress IL-17. Surprisingly, only 10% of GM-CSF+ cells were found to be IL-17+ (Figures 2A and B). In contrast, GM-CSF expression showed much greater concordance with IFNγ expression, and this association was most prominent in synovial T cells; a mean of 80.1% of GM-CSF+ CD4+ T cells coexpressed IFNγ (Figures 2C and D). Based on these findings, we wondered whether GM-CSF could be up-regulated within the “ex-Th17” cell population. We have previously shown that Th17 cells within the joint undergo plasticity toward the Th1 phenotype while maintaining CD161 expression (14). To test whether GM-CSF expression was associated with ex-Th17 cells, we determined levels of GM-CSF in CD161+ and CD161− Th1 cell subsets. This confirmed that only CD161+ Th1 cells were significantly enriched for GM-CSF expression compared to the total CD4+ population (Figure 2E). To investigate the transcriptional profile of CD161+ Th1 cells, SFMCs were sorted by cytokine capture into Th17 cells, CD161+ Th1 cells, and CD161− Th1 cells. CD161+ Th1 cells had significantly higher GM-CSF and RORC2 expression than CD161− Th1 cells, but equivalent levels of the Th1 transcription factor T-bet (Figure 2F).

nvestigate the transcriptional profile of CD161+ Th1 cells, SFMCs were sorted by cytokine capture into Th17 cells, CD161+ Th1 cells, and CD161− Th1 cells. CD161+ Th1 cells had significantly higher GM-CSF and RORC2 expression than CD161− Th1 cells, but equivalent levels of the Th1 transcription factor T-bet (Figure 2F). Figure 2 GM-CSF+ T cells share features with ex-Th17 cells. A and C, Representative dot plots of GM-CSF and IL-17 expression (A) and GM-CSF and IFNγ expression (C), detected by flow cytometry, in CD4+ T cells from JIA PBMCs and SFMCs. B and D, IL-17 (B) and IFNγ (D) coexpression (detected as described in A and C) in GM-CSF+ CD4+ T cells from healthy control PBMCs (n = 7), JIA PBMCs (n = 10), and JIA SFMCs (n = 10). E, Proportion of SFMC T cell subsets expressing GM-CSF (n = 10). In B, D, and E, symbols represent individual samples; horizontal lines show the mean. F, Expression of mRNA for GM-CSF, retinoic acid receptor–related orphan nuclear receptor C2 (RORC2), and T-bet (Tbx21) (normalized to mRNA for β-actin), as assessed by quantitative polymerase chain reaction. Th17 and Th1 subsets were identified by cytokine capture assay and purified by flow cytometry. Values are the mean ± SEM (n = 3). ∗ = P < 0.05; ∗∗ = P < 0.01, by one-way ANOVA with Bonferroni correction for multiple comparisons. RU = relative units (see Figure 1 for other definitions). Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.38647/abstract.

es are the mean ± SEM (n = 3). ∗ = P < 0.05; ∗∗ = P < 0.01, by one-way ANOVA with Bonferroni correction for multiple comparisons. RU = relative units (see Figure 1 for other definitions). Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.38647/abstract. To test our hypothesis that Th17 cells undergo plasticity toward an IFNγ+GM-CSF+ phenotype, we used a cytokine capture assay to purify IL-17+ T helper cells and, as a control, IL-17− T helper cells, from healthy control PBMCs. Stimulation and culture of IL-17+ T cells in the presence of IL-2 down-regulated IL-17 and promoted plasticity toward an intermediate Th1/Th17 (IL-17+IFNγ+) and Th1 (IL-17−IFNγ+) phenotype (Figure 3A). We have previously demonstrated that IL-12 is significantly elevated in JIA SF compared to serum (14). To better understand the contribution of this cytokine to Th17 plasticity, we cultured cells in the presence of recombinant IL-12. This led to further differentiation toward Th1/Th17 and Th1 cells when compared to the results obtained after culture with IL-2 alone. IL-23, although important for murine Th17 plasticity, did not promote a Th1 phenotype (Figure 3C).

s cytokine to Th17 plasticity, we cultured cells in the presence of recombinant IL-12. This led to further differentiation toward Th1/Th17 and Th1 cells when compared to the results obtained after culture with IL-2 alone. IL-23, although important for murine Th17 plasticity, did not promote a Th1 phenotype (Figure 3C). Figure 3 Transition of Th17 cells toward a GM-CSF–expressing phenotype in response to IL-12. A and B, IL-17+ and IL-17− CD4+ T cells from healthy control PBMCs were sorted by flow cytometry following cytokine capture and analyzed for expression of IFNγ and IL-17 (A) or IFNγ and GM-CSF (B) at baseline and after culture for 4 days in the presence of IL-2 (alone or with IL-12) prior to restimulation with phorbol myristate acetate and ionomycin in the presence of brefeldin A and analysis for cytokine expression. Results of a representative experiment (of 5 performed) are shown. C, IL-17+ CD4+ T cells (gray bars) and IL-17− CD4+ T cells (black bars) from healthy control PBMCs were cultured as described in A (as well as with IL-2 plus IL-23), and the proportion of live cells expressing IFNγ but not IL-17 was determined. D and E, IL-17+ CD4+ T cells (gray bars) and IL-17− CD4+ T cells (black bars) from healthy control PBMCs were cultured as described in B (as well as with IL-2 plus IL-23), and the proportion of live cells expressing GM-CSF (D) or IFNγ and GM-CSF (E) was determined. Values in C–E are the mean ± SEM. ∗ = P < 0.05; ∗∗ = P < 0.01, by two-way ANOVA with Bonferroni correction for multiple comparisons. See Figure 1 for definitions.

e cultured as described in B (as well as with IL-2 plus IL-23), and the proportion of live cells expressing GM-CSF (D) or IFNγ and GM-CSF (E) was determined. Values in C–E are the mean ± SEM. ∗ = P < 0.05; ∗∗ = P < 0.01, by two-way ANOVA with Bonferroni correction for multiple comparisons. See Figure 1 for definitions. To track GM-CSF expression during Th17 plasticity, we stained cells for GM-CSF at baseline and after stimulation and culture in the presence of IL-12. At baseline, both the IL-17+ and the IL-17− populations exhibited minimal GM-CSF expression, whereas after activation, GM-CSF was up-regulated in IL-17+ cells significantly more than in IL-17− cells (Figures 3B and D). Most importantly, the IFNγ+GM-CSF+ T cell population identified in the joint was rapidly induced following culture of IL-17+ cells in the presence of IL-12 (Figure 3E). Our data demonstrate that plasticity toward a Th1 phenotype results not only in expression of IFNγ, but in concomitant up-regulation of GM-CSF expression. DISCUSSION GM-CSF has long been recognized as an inflammatory mediator in the joints of patients with inflammatory arthritis. Although this has been extensively studied during the last 2 decades (9), recent data from murine models of autoimmunity have led to a reappraisal of the role of T cell–derived GM-CSF and its importance in arthritis. These data, together with the emergence of biologic agents targeting this cytokine, make the study of T cell GM-CSF expression and its regulation in human arthritis both timely and important.

om murine models of autoimmunity have led to a reappraisal of the role of T cell–derived GM-CSF and its importance in arthritis. These data, together with the emergence of biologic agents targeting this cytokine, make the study of T cell GM-CSF expression and its regulation in human arthritis both timely and important. In the present study we demonstrated an enrichment of GM-CSF–producing T helper cells in the joints of patients with JIA and, importantly, the frequency of these cells was directly correlated with levels of GM-CSF protein in the joint and serum markers of disease activity. To our knowledge, this is the first study to show that GM-CSF–expressing T cells in human autoimmune disease have a phenotype associated with ex-Th17 cells, coexpressing CD161 and IFNγ. These ex-Th17 cells, although lacking IL-17 expression, continue to express high levels of the Th17 transcription factor RORC2 (5,14). It is possible that RORC2 continues to play a role in the maintenance of GM-CSF within the ex-Th17 population. Expression of T-bet was not elevated in synovial T cells that were enriched for GM-CSF (Figure 2F), which is consistent with results of mouse studies showing normal GM-CSF expression in T-bet–knockout animals (6,7).

possible that RORC2 continues to play a role in the maintenance of GM-CSF within the ex-Th17 population. Expression of T-bet was not elevated in synovial T cells that were enriched for GM-CSF (Figure 2F), which is consistent with results of mouse studies showing normal GM-CSF expression in T-bet–knockout animals (6,7). It has previously been shown that Th17 clones secrete GM-CSF (12), but, to our knowledge, this is the first study to demonstrate the role of IL-12 in driving Th17 plasticity toward a GM-CSF+ phenotype, and its possible role in arthritis. Interestingly, fate mapping studies have shown that IL-23, a member of the IL-12 cytokine family, has this effect in mice (16), while in humans IL-23 appears to stabilize the Th17 phenotype (17). The consequences of Th17 plasticity in arthritis were initially unclear, as the switch from IL-17 to IFNγ seemed unlikely to drive significant pathology given that administration of recombinant IFNγ to patients with rheumatoid arthritis has no adverse consequences (18). We propose that Th17 plasticity contributes to the persistence of joint inflammation by augmenting local levels of GM-CSF. The development of therapeutic antibodies targeting the GM-CSF receptor α-chain offers a valuable opportunity to test the importance of GM-CSF in JIA pathology, and early reports from rheumatoid arthritis studies are encouraging (19). Although our study has focused on Th17-related GM-CSF, Th17 cells are relatively rare in the joint and GM-CSF is also expressed by non–T cells, which suggests that Th17-independent GM-CSF could be important as well. Irrespective of the cellular source, receptor blockade would abrogate the downstream actions of GM-CSF and may be a viable therapeutic option in treatment-resistant JIA.

are relatively rare in the joint and GM-CSF is also expressed by non–T cells, which suggests that Th17-independent GM-CSF could be important as well. Irrespective of the cellular source, receptor blockade would abrogate the downstream actions of GM-CSF and may be a viable therapeutic option in treatment-resistant JIA. In conclusion, we have identified Th17 plasticity as a key driver of GM-CSF production within the arthritic joint. These results and the link between synovial GM-CSF enrichment and systemic inflammation strengthen the case for initiating trials of biologic agents that target the Th17 pathway, and in particular GM-CSF, in juvenile idiopathic arthritis. AUTHOR CONTRIBUTIONS 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. Nistala 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. Pesenacker, Bending, Wedderburn, Nistala. Acquisition of data. Piper, Pesenacker, Bending, Thirugnanabalan, Varsani, Nistala. Analysis and interpretation of data. Piper, Pesenacker, Bending, Varsani, Wedderburn, Nistala. We thank the patients for contribution of samples, their parents, and hospital staff who made this study possible.

Rheumatoid arthritis (RA) is a chronic autoimmune disorder characterized by synovial inflammation and resultant progressive joint damage. It has become increasingly evident that IL‐17–dependent responses play a central role in RA, with aberrant regulation of Th17 cells being implicated in disease onset and progression (1, 2). In particular, IL‐17 recruits neutrophils to the joint and induces secretion of proinflammatory cytokines by synovial fibroblasts, resulting in the promotion of osteoclastogenesis and hence, cartilage and bone destruction (3). Elevated numbers of Th17 cells have been found in patients with RA (4, 5), and a pathogenic role of IL‐17 in arthritis has been confirmed in animal models (6, 7). ES‐62, a phosphorylcholine (PC)–containing immunomodulator secreted by the filarial nematode Acanthocheilonema viteae (8), protects against collagen‐induced arthritis (CIA) in mice (9, 10) by down‐regulating IL‐17 responses, via targeting of an inflammatory cellular network involving dendritic cells, γ/δ T cells, and Th17 cells (11).

orylcholine (PC)–containing immunomodulator secreted by the filarial nematode Acanthocheilonema viteae (8), protects against collagen‐induced arthritis (CIA) in mice (9, 10) by down‐regulating IL‐17 responses, via targeting of an inflammatory cellular network involving dendritic cells, γ/δ T cells, and Th17 cells (11). Th17 cells also secrete IL‐22, a cytokine generally considered to be proinflammatory because of its coexpression with IL‐17 during in vitro differentiation of Th17 cells (12). However, there is increasing evidence that IL‐17 and IL‐22 are differentially regulated and often produced in vivo by different lymphocyte subsets. Thus, transforming growth factor β is not required, and IL‐6 is sufficient, to induce IL‐22 production by T cells (13)—unlike the case for IL‐17. However, the transcription factor aryl hydrocarbon receptor is essential for the production of IL‐22 (14) by CCR10+ “Th22” cells that can be discriminated from Th17 cells (15). IL‐22 is also produced by innate lymphocytes (lymphoid tissue–inducer cells, γ/δ T cells, and natural killer cells) (16), but the widely expressed IL‐22 receptor (IL‐22R1–IL‐10Rβ) is not usually expressed by hemopoietic cells (17). Thus, IL‐22 appears to provide a link between the immune system and other tissues to promote their innate immunity, in particular, to enhance antimicrobial defense and tissue repair (17, 18). Reflecting these pleiotropic effects, IL‐22 has been reported to exhibit both protective effects (hepatitis and inflammatory bowel disease) and pathogenic effects (psoriasis) (13, 19, 20, 21) in inflammatory disease.

mote their innate immunity, in particular, to enhance antimicrobial defense and tissue repair (17, 18). Reflecting these pleiotropic effects, IL‐22 has been reported to exhibit both protective effects (hepatitis and inflammatory bowel disease) and pathogenic effects (psoriasis) (13, 19, 20, 21) in inflammatory disease. In the context of RA, mice that are deficient in IL‐22 are less susceptible to CIA and/or develop less severe disease (22, 23). Moreover, levels of IL‐22 and Th22 cells have been found to be elevated in the periphery and synovia of RA patients (24, 25, 26), and IL‐22 has been shown to induce proliferation of synovial fibroblasts and promote RANKL production and osteoclastogenesis in vitro (27). We therefore investigated whether the protective effects of ES‐62 were also associated with targeting of such IL‐22 responses. Surprisingly, these studies revealed that IL‐22 can play dual pathogenic and protective roles in CIA and that ES‐62 harnesses the cytokine's antiinflammatory effects on synovial fibroblasts, to mediate its protection against joint destruction. In describing a novel mechanism by which a parasitic helminth–derived product acts to reduce autoimmune arthritis, these findings contribute to our fundamental understanding of IL‐22 immunobiology and identify novel therapeutic targets in inflammatory disease.

lasts, to mediate its protection against joint destruction. In describing a novel mechanism by which a parasitic helminth–derived product acts to reduce autoimmune arthritis, these findings contribute to our fundamental understanding of IL‐22 immunobiology and identify novel therapeutic targets in inflammatory disease. MATERIALS AND METHODS Mice Animals were maintained in the Biological Services Units at the University of Glasgow and the University of Strathclyde, in accordance with Home Office UK Licenses PPL60/4300, PPL60/3791, PIL60/12183, PIL60/12950, and PIL60/9576 and the respective ethics review boards of these universities. CIA was induced in 8–10‐week‐old male DBA/1 mice (Harlan Olac) by intradermal immunization with bovine type II collagen (MD Biosciences) in Freund's complete adjuvant (day 0) and by intraperitoneal (IP) administration in phosphate buffered saline (PBS) (day 21). Purified endotoxin‐free ES‐62 (2 μg/dose) or PBS was administered subcutaneously on days −2, 0, and 21 (9), and cells were recovered from draining lymph nodes (DLNs) and joints as previously described (11). Mice were treated with endotoxin‐free recombinant IL‐22 (rIL‐22; ImmunoTools) (1 μg/dose IP or 0.25 μg/dose footpad injection, twice weekly as indicated) or endotoxin‐free mouse IgG (Europa Bioproducts) (100 μg/dose IP twice weekly beginning on day 7) or anti–IL‐22 antibodies (28) purified from the AM22.1 hybridoma (100 μg/dose IP twice weekly as indicated) (hybridoma kindly provided by Dr. J. C. Renauld, Ludwig Institute for Cancer Research, Brussels, Belgium). There were no significant differences in articular scores between PBS‐treated mice with CIA that were treated with mouse IgG and those that were not treated with mouse IgG. Mice were monitored for clinical symptoms of arthritis, which were scored as 0 (normal), 1 (erythema), 2 (erythema plus swelling), 3 (extension of swelling), or 4 (loss of function); the overall disease score was the sum of the scores for each of the 4 limbs. The date mice were removed from the study varied slightly as euthanasia is required when a score of 10 is reached or clinical symptoms develop in all 4 limbs. Other than dial caliper (Kroeplin) analysis of paw thickness, all analyses were also performed on the day the animal was removed from the study.

of the 4 limbs. The date mice were removed from the study varied slightly as euthanasia is required when a score of 10 is reached or clinical symptoms develop in all 4 limbs. Other than dial caliper (Kroeplin) analysis of paw thickness, all analyses were also performed on the day the animal was removed from the study. Ex vivo analysis DLN cells (106/ml) were incubated with or without phorbol myristate acetate (PMA) (50 ng/ml)/ionomycin (500 ng/ml) (Sigma‐Aldrich) for 1 hour, followed by addition of 10 μg/ml brefeldin A (Sigma‐Aldrich) for 5 hours at 37°C with 5% CO2. Live cells were discriminated with Live/Dead Fixable Aqua Dye (Invitrogen), and phenotypic markers labeled using fluorescein isothiocyanate (FITC)–conjugated anti‐CD3 (BD PharMingen) and PerCP‐conjugated anti‐CD4 or biotinylated anti‐CD4 antibodies (detected with Alexa Fluor 450–conjugated streptavidin) (BD PharMingen) before the cells were fixed and permeabilized according to protocols recommended by the supplier (BioLegend). Cytokines were stained using phycoerythrin (PE)–Cy7–conjugated anti–interferon‐γ (anti‐IFNγ), allophycocyanin‐conjugated or PerCP–Cy5.5–conjugated anti–IL‐17A (BioLegend), or PE‐conjugated anti–IL‐22 antibodies (R&D Systems) for 30 minutes prior to analysis by flow cytometry, with gating according to appropriate isotype controls. Detection of biologically relevant IL‐22+ cells was validated by in vitro Th17/22 differentiation assays using 2 anti–IL‐22 antibodies (clone 140301 [R&D Systems] and clone 1H8PWSR [eBioscience]) with or without rIL‐22 pretreatment (10–40 μg/ml); essentially the same results were obtained with both clones when dead cells were excluded from the analyses. Cells were extracted from the joints of mice with CIA by collagenase digestion and incubated with 10 μg/ml brefeldin A for 5 hours at 37°C. Following live cell discrimination with Live/Dead Fixable Aqua Dye, cells were labeled with PerCP‐conjugated anti‐Gr1 and FITC‐conjugated anti‐CD11b antibodies (eBioscience) before being fixed and permeabilized for staining for IL‐17, IL‐22, and IFNγ expression.

incubated with 10 μg/ml brefeldin A for 5 hours at 37°C. Following live cell discrimination with Live/Dead Fixable Aqua Dye, cells were labeled with PerCP‐conjugated anti‐Gr1 and FITC‐conjugated anti‐CD11b antibodies (eBioscience) before being fixed and permeabilized for staining for IL‐17, IL‐22, and IFNγ expression. Cytokine measurement Levels of IL‐17A, IL‐22, and IL‐6 were determined by enzyme‐linked immunosorbent assay, according to the recommendations of the manufacturers (BioLegend, R&D Systems, and eBioscience, respectively). Synovial fibroblast explant cultures Single‐cell suspensions recovered from all 4 limbs by collagenase digestion (29) were pooled to avoid biasing results. Cells were cultured for 12 hours in Dulbecco's modified Eagle's medium (Invitrogen) containing 2 mM glutamine, 50 units/ml penicillin, 50 μg/ml streptomycin, and 10% fetal calf serum, after which nonadherent cells were removed. Adherent cells were cultured for 7 days and phenotyped using biotinylated anti‐CD54, PE–Cy7–conjugated anti‐CD106, and PerCP‐conjugated anti‐CD90.2 antibodies (eBioscience). Synovial fibroblasts were stimulated in vitro with murine rIL‐17 or rIL‐22. After 24 hours, supernatants were collected for analysis.

ere removed. Adherent cells were cultured for 7 days and phenotyped using biotinylated anti‐CD54, PE–Cy7–conjugated anti‐CD106, and PerCP‐conjugated anti‐CD90.2 antibodies (eBioscience). Synovial fibroblasts were stimulated in vitro with murine rIL‐17 or rIL‐22. After 24 hours, supernatants were collected for analysis. Immunofluorescence Preparation of tissue sections (7 μm), staining with hematoxylin and eosin, and detection of IL‐17 expression were performed as previously described (11). To detect IL‐22 expression, samples were stained for 12 hours at 4°C with a rat anti‐mouse IL‐22 antibody (rat IgG isotype control; R&D Systems) with DAPI counterstaining, followed by detection using a biotinylated goat anti‐rat IgG antibody and Alexa Fluor 647–conjugated streptavidin. Images were obtained using an LSM 510 Meta confocal laser coupled to an Axiovert 200 microscope (Zeiss) and analyzed with Zeiss LSM Image Browser software. Statistical analysis The significance of differences was determined by Student's unpaired 1‐tailed t‐test or by one‐way analysis of variance followed by Newman‐Keuls post hoc test. Articular scores were assessed by Mann‐Whitney test. P values less than 0.05 were considered significant.

Immunofluorescence Preparation of tissue sections (7 μm), staining with hematoxylin and eosin, and detection of IL‐17 expression were performed as previously described (11). To detect IL‐22 expression, samples were stained for 12 hours at 4°C with a rat anti‐mouse IL‐22 antibody (rat IgG isotype control; R&D Systems) with DAPI counterstaining, followed by detection using a biotinylated goat anti‐rat IgG antibody and Alexa Fluor 647–conjugated streptavidin. Images were obtained using an LSM 510 Meta confocal laser coupled to an Axiovert 200 microscope (Zeiss) and analyzed with Zeiss LSM Image Browser software. Statistical analysis The significance of differences was determined by Student's unpaired 1‐tailed t‐test or by one‐way analysis of variance followed by Newman‐Keuls post hoc test. Articular scores were assessed by Mann‐Whitney test. P values less than 0.05 were considered significant. RESULTS Differential regulation of IL‐17 and IL‐22 responses by ES‐62 in mice with CIA Compared to the PBS‐treated group, the incidence of CIA was significantly reduced in mice treated with ES‐62, as were the degree of hind paw swelling and clinical scores (even among ES‐62–treated animals that did develop CIA) (Figure 1A). In parallel with the clinical findings, the total number of DLN cells was significantly increased in mice with CIA treated with PBS, but not in those exposed to ES‐62 in vivo, compared to the number in naive mice without CIA (Figure 1B). However, while the proportions of DLN cells and CD4+ T cells that produced IL‐17 in response to ex vivo stimulation with either medium or PMA plus ionomycin were significantly reduced by exposure to ES‐62 in vivo, only the levels of unstimulated IL‐22–producing CD4+ T cells were significantly suppressed (Figures 1C and D). Moreover, whereas the ability of DLN and CD4+ cells to generate IL‐17 ex vivo was significantly increased by stimulation with PMA plus ionomycin (P < 0.01 for both), this was not the case with regard to production of IL‐22, as reflected by the finding that these cytokines were generated by distinct subsets of DLN and CD4+ cells (Figure 1B).

ability of DLN and CD4+ cells to generate IL‐17 ex vivo was significantly increased by stimulation with PMA plus ionomycin (P < 0.01 for both), this was not the case with regard to production of IL‐22, as reflected by the finding that these cytokines were generated by distinct subsets of DLN and CD4+ cells (Figure 1B). Figure 1 ES‐62 differentially targets interleukin‐17 (IL‐17)– and IL‐22–producing draining lymph node (DLN) cells in mice with collagen‐induced arthritis (CIA). A, Percent CIA incidence, mean ± SEM paw width, and mean ± SEM articular (clinical) score at various time points in a representative experiment (n = 7 phosphate buffered saline [PBS]–treated mice and 7 ES‐62–treated mice) (first through third panels), and mean ± SEM articular score (pooled from 6 independent experiments) at the time of removal from the experiment for all mice (fourth panel; n = 38 PBS‐treated mice and 36 ES‐62–treated mice) and mice that had developed CIA (fifth panel; n = 34 PBS‐treated mice and 24 ES‐62–treated mice). B, Number of DLN cells in naive mice (n = 9), PBS‐treated mice with CIA (n = 32), and ES‐62–treated mice with CIA (n = 25), and representative plots of intracellular IL‐22 and IL‐17 expression from experiments using medium‐treated or phorbol myristate acetate plus ionomycin (PMA/iono)–stimulated DLN cells and CD4+ T cells from PBS‐treated mice with CIA. C and D, Percentages of DLN cells (C) or CD4+ T cells (D) expressing IL‐17 or IL‐22 in PBS‐treated mice with CIA (n = 32) and ES‐62–treated mice with CIA (n = 24). In B (left panel), C, and D, each symbol represents an individual mouse; bars show the mean. ∗ = P < 0.05 (versus ES‐62 on the days indicated, in the second and third panels of A); ∗∗ = P < 0.01; ∗∗∗ = P < 0.001.

) expressing IL‐17 or IL‐22 in PBS‐treated mice with CIA (n = 32) and ES‐62–treated mice with CIA (n = 24). In B (left panel), C, and D, each symbol represents an individual mouse; bars show the mean. ∗ = P < 0.05 (versus ES‐62 on the days indicated, in the second and third panels of A); ∗∗ = P < 0.01; ∗∗∗ = P < 0.001. Although no differences could be detected at early time points, in accordance with previous observations in human patients (30) and the proposed pathogenic role of IL‐22 in the CIA model (22, 23), the mean serum level of IL‐22 was increased in PBS‐treated mice with CIA compared to naive controls at the last assessment (Figure 2A). Rather unexpectedly, ES‐62–treated mice exhibited an even higher mean serum level of IL‐22. As this group segregated into high and low IL‐22 producers, we investigated whether this was related to disease progression and found that IL‐22 levels (in both PBS‐treated and ES‐62–treated mice) correlated inversely with CIA severity (Figure 2A). These findings are in direct contrast to the association of serum IL‐17 levels with articular score and their significant reduction by in vivo exposure to ES‐62 (11).

lated to disease progression and found that IL‐22 levels (in both PBS‐treated and ES‐62–treated mice) correlated inversely with CIA severity (Figure 2A). These findings are in direct contrast to the association of serum IL‐17 levels with articular score and their significant reduction by in vivo exposure to ES‐62 (11). Figure 2 ES‐62–dependent up‐regulation of IL‐22 correlates with protection against CIA. A, Serum IL‐22 levels in naive mice (n = 8), PBS‐treated mice with CIA (n = 10), and ES‐62–treated mice with CIA (n = 8), and inverse correlation (Pearson's r = −0.7298, P < 0.0006) between serum IL‐22 levels and clinical scores in 18 PBS‐ or ES‐62–treated mice with CIA (each symbol represents the mean value from triplicate analyses in an individual mouse). B, Representative plots of intracellular IL‐17 and IL‐22 expression in pooled joint cells from PBS‐treated mice with CIA (n = 4) and ES‐62–treated mice with CIA (n = 4). C, Percentages of joint cells expressing IL‐17, interferon‐γ (IFNγ), and IL‐22 in PBS‐treated mice with CIA (n = 27, 17, and 24 for IL‐17, IFNγ, and IL‐22, respectively) and ES‐62–treated mice with CIA (n = 20, 17, and 21, respectively). In A (left panel) and C, each symbol represents an individual mouse; bars show the mean. ∗ = P < 0.05; ∗∗ = P < 0.01; ∗∗∗ = P < 0.001. See Figure 1 for other definitions.

eated mice with CIA (n = 27, 17, and 24 for IL‐17, IFNγ, and IL‐22, respectively) and ES‐62–treated mice with CIA (n = 20, 17, and 21, respectively). In A (left panel) and C, each symbol represents an individual mouse; bars show the mean. ∗ = P < 0.05; ∗∗ = P < 0.01; ∗∗∗ = P < 0.001. See Figure 1 for other definitions. Analysis of joint‐infiltrating cells revealed that IL‐17 and IL‐22 were also produced predominantly by distinct cell subsets within this population, although it was not clear whether these included Th cells due to our failure to detect expression of CD4 (presumably due to its cleavage during the collagenase extraction procedure). Nevertheless, a reduced proportion of the cells expressed IL‐17, and an increased proportion generated IL‐22, in the mice with CIA that had been exposed to ES‐62 in vivo (Figure 2B). Independent analysis of individual mice showed that, whereas exposure to ES‐62 resulted in significant suppression of the levels of IL‐17– and also IFNγ‐producing cells, the levels of IL‐22–generating cells were maintained and even increased, although this did not reach statistical significance (Figure 2C).

Figure 2B). Independent analysis of individual mice showed that, whereas exposure to ES‐62 resulted in significant suppression of the levels of IL‐17– and also IFNγ‐producing cells, the levels of IL‐22–generating cells were maintained and even increased, although this did not reach statistical significance (Figure 2C). Onset of arthritis was detected in a few mice prior to challenge with collagen. While exposure to ES‐62 reduced the incidence of this (11%, versus 23% in PBS‐treated mice), it did not significantly ameliorate pathology in the mice that developed disease. Perhaps consistent with this, ES‐62 did not suppress the (reduced) levels of IL‐17–producing cells in the joints of these mice. The differential IL‐17:IL‐22 ratios (0.49 and 0.21 in the PBS‐ and ES‐62–treated groups, respectively, postchallenge; 0.26 and 0.31, respectively, prechallenge) and the ES‐62 sensitivity observed in the 2 groups further supported the idea that resetting of the IL‐17/IL‐22 balance in the joint correlates with ES‐62–mediated protection and perhaps suggests that additional/alternative inflammatory parameters contributed to the pathogenesis in mice that developed arthritis prior to challenge.

2 sensitivity observed in the 2 groups further supported the idea that resetting of the IL‐17/IL‐22 balance in the joint correlates with ES‐62–mediated protection and perhaps suggests that additional/alternative inflammatory parameters contributed to the pathogenesis in mice that developed arthritis prior to challenge. ES‐62–mediated protection against CIA is dependent on IL‐22. In studies of the relationship of IL‐17 and IL‐22 expression with joint pathology in situ, we have shown that, while expression of IL‐17 was essentially absent in the joints of naive mice, increasing levels were observed throughout the progression of CIA and correlated with induction of joint pathology (11). In the present study we observed that in vivo exposure to ES‐62 suppressed both the expression of IL‐17 in the joints and the development of joint disease (Figure 3A). In contrast, IL‐22 was expressed in the joints of naive mice predominantly in the bone area (not bone marrow), but some IL‐22+ cells (∼10 μm) could also be detected in the synovium. Such IL‐22 expression was also evident in mice with CIA, but following an increase (within 2 weeks) during the disease initiation phase, it subsequently decreased over time. However, at 4 weeks some IL‐22+ cells were found around the periphery of the bone in mice with CIA, but not in naive or ES‐62–treated mice.

ovium. Such IL‐22 expression was also evident in mice with CIA, but following an increase (within 2 weeks) during the disease initiation phase, it subsequently decreased over time. However, at 4 weeks some IL‐22+ cells were found around the periphery of the bone in mice with CIA, but not in naive or ES‐62–treated mice. Figure 3 ES‐62 protection against CIA requires IL‐22. A, Joint sections from representative PBS‐treated mice with CIA, ES‐62–treated mice with CIA, and naive mice at weeks 2, 4, and 6 (articular scores 0, 4, and 8, respectively, at weeks 2, 4, and 6 in the PBS‐treated mice and 0, 2, and 2, respectively, in the ES‐62–treated mice). Sections were stained with hematoxylin and eosin (H&E) or for IL‐17 or IL‐22 (red) or nuclei (DAPI; blue). Isotype control sections were negative for IL‐17 and IL‐22. Joint structure is shown in the grayscale image. B = bone; AC = articular cavity; BM = bone marrow. Original magnification × 20 (1.9 zoom). B, Mean ± SEM articular scores in mice with CIA treated with PBS (n = 12), murine IgG (100 μg/dose; n = 11), or anti–IL‐22 (100 μg/dose) twice weekly from day 7 (early αIL‐22; n = 14) or from day 19 (late αIL‐22; n = 4), or treated intraperitoneally (IP) with PBS (n = 7) or recombinant IL‐22 (rIL‐22) (1 μg/dose; n = 7) twice weekly from day 7. ∗∗∗ = P < 0.001. C, Mean ± SEM articular scores in mice with CIA treated with PBS plus IgG (n = 13), ES‐62 plus IgG (n = 13), or ES‐62 plus anti–IL‐22 (n = 12) and with antibodies administered IP twice weekly from day 19 onward (100 μg/dose) and ES‐62 administered on days −2, 0, and 21 (2 μg/dose), and percent incidence of disease (score >1) by treatment group. ∗ = P < 0.05 (P values shown for specific days are versus treatment with ES‐62 plus anti–IL‐22). See Figure 1 for other definitions.

ies administered IP twice weekly from day 19 onward (100 μg/dose) and ES‐62 administered on days −2, 0, and 21 (2 μg/dose), and percent incidence of disease (score >1) by treatment group. ∗ = P < 0.05 (P values shown for specific days are versus treatment with ES‐62 plus anti–IL‐22). See Figure 1 for other definitions. Exposure to ES‐62 in vivo induced an inverse pattern of expression, with the parasite product suppressing the early peak of IL‐22 expression (Figure 3A), perhaps mirroring its inhibition of Th22 responses observed in the DLNs, but inducing strong expression at later time points consistent with its induction of IL‐22+ joint cells (peaking at week 4). These include synovium cells (∼20 μm; 20× digital magnification, 1.9 zoom) not seen in PBS‐treated mice with CIA. In terms of their bipolar, spindle shape and prominent secretory machinery (31) as evidenced by punctate IL‐22 staining, these cells are reminiscent of the IL‐22–producing fibroblast‐like synoviocytes recently reported to be protective against RA (32); hence, our results support the notion that ES‐62 mediates protection against joint pathology via an IL‐22–dependent mechanism.

etory machinery (31) as evidenced by punctate IL‐22 staining, these cells are reminiscent of the IL‐22–producing fibroblast‐like synoviocytes recently reported to be protective against RA (32); hence, our results support the notion that ES‐62 mediates protection against joint pathology via an IL‐22–dependent mechanism. To investigate the pathogenic role of IL‐22, mice were treated during the initiation phase (twice weekly from day 7 of disease, as preliminary experiments established that levels of IL‐22–producing DLN cells were elevated within 7–14 days) with either neutralizing anti–IL‐22 antibodies or rIL‐22 to determine whether these reagents could, respectively, block or promote development of CIA (Figure 3B). Exposure to neutralizing anti–IL‐22 antibodies essentially abrogated development of CIA, with no similar effect obtained with the use of irrelevant IgG. Administration of rIL‐22 tended to promote both disease onset and increased severity (Figure 3B); indeed, the number of limbs affected in the rIL‐22–treated cohort necessitated termination of these experiments before full pathology was established in the PBS group. In contrast, when neutralizing anti–IL‐22 antibodies were not administered until around the time of onset of pathology but prior to challenge with collagen (day 19), there was no significant disruption of the development of CIA (Figure 3B), supporting the notion that IL‐22 has a pathogenic role during the early, but not the later, phase of CIA.

zing anti–IL‐22 antibodies were not administered until around the time of onset of pathology but prior to challenge with collagen (day 19), there was no significant disruption of the development of CIA (Figure 3B), supporting the notion that IL‐22 has a pathogenic role during the early, but not the later, phase of CIA. As ES‐62 acted to maintain and/or enhance IL‐22 levels in serum, DLN cells, and joints during established disease (within 3–4 weeks), we also investigated whether administration of neutralizing anti–IL‐22 antibodies around the time of onset of joint pathology (day 19) would abrogate the protective effects of ES‐62. The results of these experiments (Figure 3C) indicated that the protective effects of ES‐62 were indeed dependent on IL‐22. ES‐62 and IL‐22 down‐regulate synovial fibroblast responses and suppress joint inflammation To investigate the mechanisms involved in the observed protection against CIA, we examined the effects of IL‐17, IL‐22, and ES‐62 on joint inflammation. Consistent with the suppression of pathogenic IL‐17 responses, the total number of infiltrating cells, and in particular, CD11b+Gr1+ neutrophils (Figure 4A), was significantly higher in the joints of mice with CIA treated with PBS relative to those exposed to ES‐62. Moreover, infiltrating cells from the ES‐62–treated mice showed significantly reduced levels of IL‐6 release (Figure 4A).

otal number of infiltrating cells, and in particular, CD11b+Gr1+ neutrophils (Figure 4A), was significantly higher in the joints of mice with CIA treated with PBS relative to those exposed to ES‐62. Moreover, infiltrating cells from the ES‐62–treated mice showed significantly reduced levels of IL‐6 release (Figure 4A). Figure 4 ES‐62 modulates cellular infiltration of the joints of mice with CIA. A, Numbers of infiltrating cells in the joints of PBS‐treated mice with CIA (n = 28) and ES‐62–treated mice with CIA (n = 22), numbers of CD11b+Gr1+ cells in the joints of PBS‐treated mice with CIA (n = 22) and ES‐62–treated mice with CIA (n = 17), and production of IL‐6, determined by enzyme‐linked immunosorbent assay, in joint cells (106/ml) from PBS‐treated mice with CIA (n = 11), ES‐62–treated mice with CIA (n = 12), and naive mice (n = 3) (each symbol represents the mean value from triplicate analyses in an individual mouse). B, Gating strategy for analysis of Gr1+CD11b+ cells isolated from the joints of PBS‐treated mice with CIA, and their expression of IL‐22 and interferon‐γ (IFNγ). C, Percentages of IFNγ+ and IL‐22+ CD11b+Gr1+ joint cells, and mean fluorescence intensity (MFI) of IL‐22 expression by CD11b+Gr1+ joint cells from PBS‐treated mice with CIA (n = 17 and 13 for IFNγ and IL‐22, respectively) and ES‐62–treated mice with CIA (n = 17 and 13, respectively). D, Inverse correlation (Pearson's r = −0.86, P = 0.0075) between the percentage of IL‐22+CD11b+Gr1+ cells and clinical scores in mice with CIA. (The same trend was found in 2 additional independent experiments.) In A and C, each symbol represents an individual mouse; bars show the mean. ∗ = P < 0.05. See Figure 1 for other definitions.

ation (Pearson's r = −0.86, P = 0.0075) between the percentage of IL‐22+CD11b+Gr1+ cells and clinical scores in mice with CIA. (The same trend was found in 2 additional independent experiments.) In A and C, each symbol represents an individual mouse; bars show the mean. ∗ = P < 0.05. See Figure 1 for other definitions. Most of the neutrophils (>70%) isolated from the joints of animals with CIA were able to produce IFNγ (Figure 4B), and while this proportion was not altered by exposure to ES‐62 (Figure 4C), their lower numbers would contribute to the observed reduction of IFNγ‐producing cells in the joint (Figure 2C). Some of the neutrophils appeared to express IL‐22 (Figure 4B), and the proportion of IL‐22+Gr1+CD11b+ cells and their levels of IL‐22 expression tended to be increased by ES‐62 (Figure 4C), although neither increase reached statistical significance. Nevertheless, such immunomodulation results in a shift in the balance of cytokines, with a relative reduction of IL‐6/IL‐17/IFNγ expression and increase in IL‐22 expression (Figures 2C, 4A, and 4C). Consistent with the ES‐62–mediated promotion of IL‐22 levels, there was a negative correlation between the proportion of this IL‐22+Gr1+CD11b+ subset of neutrophils and the severity of disease (Figure 4D), suggesting that ES‐62 promotes induction of an IL‐22+ neutrophil subset that infiltrates the joint to mediate protection against CIA.

ES‐62–mediated promotion of IL‐22 levels, there was a negative correlation between the proportion of this IL‐22+Gr1+CD11b+ subset of neutrophils and the severity of disease (Figure 4D), suggesting that ES‐62 promotes induction of an IL‐22+ neutrophil subset that infiltrates the joint to mediate protection against CIA. As hemopoietic cells generally do not express IL‐22R, we next investigated whether the IL‐22–dependent protective effects of ES‐62 reflected modulation of the function of synovial fibroblasts, which not only secrete cytokines (such as IL‐6) that recruit inflammatory cells (such as neutrophils) to the joint, but also act to directly mediate joint destruction by releasing proteolytic enzymes (matrix metalloproteinases) that degrade cartilage and by secreting factors (e.g., RANKL) that further contribute to bone destruction by activating osteoclastogenesis and promoting bone resorption (33, 34, 35). Consistent with this proposal, when explant cultures of synovial fibroblasts (CD90.2+CD54+CD106+) (Figure 5A) extracted from mice with CIA were incubated with rIL‐22 the production of IL‐6, rather than being stimulated, was inhibited to below basal levels (Figure 5B), in direct contrast with the results obtained with rIL‐17 incubation. Moreover, explant cultures of synovial fibroblasts from mice with CIA exposed to ES‐62 showed significantly reduced basal production of IL‐6 and were less responsive to rIL‐17 and rIL‐22 than fibroblast cultures from PBS‐treated mice (Figure 5C). Indeed, even though culture of synovial fibroblasts with IL‐22 in vitro suppressed their basal release of IL‐6, the production of this cytokine by cells from ES‐62–treated mice was still lower than that by cells from PBS‐treated mice with CIA. Collectively, these data indicate that ES‐62 may act via IL‐22 to suppress synovial fibroblast–mediated inflammation of the joints during established disease.

their basal release of IL‐6, the production of this cytokine by cells from ES‐62–treated mice was still lower than that by cells from PBS‐treated mice with CIA. Collectively, these data indicate that ES‐62 may act via IL‐22 to suppress synovial fibroblast–mediated inflammation of the joints during established disease. Figure 5 ES‐62 desensitizes IL‐17–mediated production of IL‐6 by synovial fibroblasts. A, CD90.2, CD54, and CD106 expression (black lines) by synovial fibroblasts from mice with CIA (gray lines represent isotype controls). Fibroblasts were cultured ex vivo for 7 days prior to in vitro stimulation. B, IL‐6 production by synovial fibroblasts from PBS‐treated mice with CIA, in response to recombinant IL‐17 (rIL‐17) or rIL‐22 (10 ng/ml). Three independent cultures of cells obtained by pooling mice (n = 7) within each group were performed. Values are the mean ± SEM (calculated using the mean result from the triplicate experiments). C, IL‐6 production over 24 hours by synovial fibroblast explant cultures (106/ml) from individual PBS‐treated mice with CIA or ES‐62–treated mice with CIA, in response to fresh medium, rIL‐17 (10 ng/ml), or rIL‐22 (10 ng/ml). For PBS‐treated mice, n = 12, 11, and 12 for fresh medium, rIL‐17, and rIL‐22, respectively; for ES‐62–treated mice, n = 14, 13, and 14, respectively. Each symbol represents the mean value from triplicate determinations (by enzyme‐linked immunosorbent assay) in an individual mouse; bars show the mean. ∗ = P < 0.05; ∗∗∗ = P < 0.001. See Figure 1 for other definitions.

h medium, rIL‐17, and rIL‐22, respectively; for ES‐62–treated mice, n = 14, 13, and 14, respectively. Each symbol represents the mean value from triplicate determinations (by enzyme‐linked immunosorbent assay) in an individual mouse; bars show the mean. ∗ = P < 0.05; ∗∗∗ = P < 0.001. See Figure 1 for other definitions. To directly investigate the potential dual roles of IL‐22 in joint inflammation, mice with CIA were administered rIL‐22 in the right footpads (PBS in the left footpads) twice weekly from day 7 in order to mimic local up‐regulation of IL‐22 during both the initiation and the effector phases of disease (Figure 6). Treatment with rIL‐22 initially increased the articular score (peak on day 25) before beginning to mediate some resolution of joint inflammation, resulting in suppression of the articular score relative to that in untreated mice with CIA by the end of the experimental period. These protective effects were more pronounced when rIL‐22 was first administered to the footpads around the time of disease onset (twice weekly from day 19) to mimic the elevated levels observed in mice with established CIA that had been exposed to ES‐62 in vivo. With this protocol there was significant reduction in pathology (Figure 6A). Although the inflammation in the footpads that received this “therapeutic” administration of rIL‐22 was also reduced relative to that in PBS‐treated footpads of the same animals, it was evident that the pattern of IL‐22 first promoting and then resolving joint inflammation was paralleled, albeit to a lesser extent and following a slight delay, in the PBS‐treated joints. This suggests that the effects of rIL‐22 were being transferred to these limbs, presumably via systemic effects of the injected rIL‐22 or perhaps by its functional suppression of “pathogenic” fibroblasts, which has been reported to mediate clinical spreading of arthritis between joints (35).

in the PBS‐treated joints. This suggests that the effects of rIL‐22 were being transferred to these limbs, presumably via systemic effects of the injected rIL‐22 or perhaps by its functional suppression of “pathogenic” fibroblasts, which has been reported to mediate clinical spreading of arthritis between joints (35). Figure 6 Recombinant IL‐22 (rIL‐22) modulates synovial fibroblast responses to IL‐17. A, Mice with CIA were injected twice weekly with rIL‐22 (0.25 μg/dose) in the right footpad or PBS (50 μl) in the left footpad (injection controls). Mice receiving no additional treatment were used as disease controls. IL‐22 injections were begun at the initiation phase of CIA (day 7 [early]) (n = 8) or around the time of onset of joint pathology (day 19 [late]) (n = 7). Values are the mean ± SEM (calculated using the mean clinical score of the 2 paws undergoing the same treatment in each individual mouse). B and C, Paws from individual mice with CIA receiving the indicated treatments, naive mice, and disease control mice with CIA were pooled to generate synovial fibroblast explant cultures representing the various treatment groups. Release of IL‐6 after 24 hours of treatment with medium alone (C) or with IL‐17 (B) was evaluated by enzyme‐linked immunosorbent assay. Values are the mean ± SEM (calculated using the mean of triplicate determinations of IL‐6 values from 3 independent cultures of cells from each treatment group). ∗ = P < 0.05; ∗∗ = P < 0.01; ∗∗∗ = P < 0.001 (P value shown for day 25 in the middle panel of A is versus untreated mice with CIA). See Figure 1 for other definitions.

e the mean ± SEM (calculated using the mean of triplicate determinations of IL‐6 values from 3 independent cultures of cells from each treatment group). ∗ = P < 0.05; ∗∗ = P < 0.01; ∗∗∗ = P < 0.001 (P value shown for day 25 in the middle panel of A is versus untreated mice with CIA). See Figure 1 for other definitions. Finally, to confirm that these pathogenic and protective effects of rIL‐22 were targeting synovial fibroblasts in the joint, we analyzed IL‐6 production by synovial fibroblast explant cultures from paws treated with rIL‐22 and from PBS‐treated control paws of the same animals. In accordance with their pathogenic status (35), synovial fibroblasts from mice with CIA exhibited enhanced IL‐6 production relative to those from naive mice (Figure 6B), and administration of rIL‐22 to the paw from the time of the initiation phase of CIA resulted in even higher production of IL‐6 by synovial fibroblasts. In contrast, when rIL‐22 administration did not begin until around the time of onset of joint pathology, the capacity to produce IL‐6 tended to be reduced toward the levels observed in naive mice (Figure 6B). Importantly, while synovial fibroblasts from ES‐62–treated mice exhibited reduced IL‐6 responses relative to those from mice treated with PBS (Figure 5C) or PBS plus IgG (Figure 6C), such desensitization was partially overcome in synovial fibroblasts derived from mice with CIA that were exposed to both ES‐62 and neutralizing anti–IL‐22 antibodies in vivo (Figure 6C), corroborating the antiinflammatory role of IL‐22 in ES‐62–mediated protection against joint inflammation.

r PBS plus IgG (Figure 6C), such desensitization was partially overcome in synovial fibroblasts derived from mice with CIA that were exposed to both ES‐62 and neutralizing anti–IL‐22 antibodies in vivo (Figure 6C), corroborating the antiinflammatory role of IL‐22 in ES‐62–mediated protection against joint inflammation. DISCUSSION ES‐62 protects against CIA by targeting priming of a complex IL‐17–producing cellular network that involves dendritic cells and γ/δ and CD4+ T cells, and also by acting directly on Th17 cells (11). In this study, we investigated whether ES‐62–mediated suppression of IL‐22 responses also contributed to its protective effects in CIA, as this Th17‐associated cytokine has similarly been implicated in CIA pathogenesis (22, 23). Although we confirmed the pathogenic role of IL‐22 in the initiation phase of CIA, we found, surprisingly, that ES‐62 treatment appeared to enhance IL‐22 responses following onset of disease. Moreover, during established disease, serum levels of IL‐22 correlated inversely with articular scores, and local administration of IL‐22 reduced joint inflammation. Furthermore, ES‐62–mediated protection against CIA could be blocked by administration of neutralizing anti–IL‐22 antibodies (from day 19). Collectively, these data suggest that IL‐22 has dual pro‐ and antiinflammatory roles in CIA, with early, systemic IL‐17 and IL‐22 (Th22) responses cooperating to drive pathogenesis, while later, IL‐22 acts at the site of inflammation to counterregulate IL‐17 proinflammatory signaling and promote resolution of joint disease.

ctively, these data suggest that IL‐22 has dual pro‐ and antiinflammatory roles in CIA, with early, systemic IL‐17 and IL‐22 (Th22) responses cooperating to drive pathogenesis, while later, IL‐22 acts at the site of inflammation to counterregulate IL‐17 proinflammatory signaling and promote resolution of joint disease. IL‐22 is involved in the host response to infectious diseases by promoting inflammation (36, 37) but, reflecting its tissue repair properties, it exerts both pro‐ and antiinflammatory actions in allergic and autoimmune inflammatory disorders (13, 19, 20, 38, 39, 40). Indeed, our results demonstrating dual pathogenic and inflammation‐resolving roles of IL‐22 are reminiscent of the findings in studies using models of ovalbumin‐induced airway hyperresponsiveness, in which IL‐22 appears to be essential for antigen sensitization (41) yet acts to resolve established airway inflammation. Likewise, in asthma patients, while serum IL‐22 levels are elevated and correlate positively with disease severity (41, 42), levels in bronchoalveolar lavage fluid correlate inversely with those of proinflammatory chemokines, and IL‐22 can inhibit the release of proinflammatory mediators by human bronchial epithelial cells (39). Results of other recent studies have called into question the idea that IL‐22 has a solely pathogenic role in arthritis, as antigen (methylated bovine serum albumin [BSA])–induced, IL‐17–mediated joint inflammation was found to occur independent of IL‐22 (43) and systemic administration of rIL‐22 was protective in late stages of disease in the CIA model (44). Further supporting the notion of an inflammation‐resolving role of this cytokine, IL‐22 has recently been shown to potentially modulate the IL‐23/IL‐17 inflammatory axis in RA patients by down‐regulating IL‐23 and IL‐17RC expression in fibroblast‐like synoviocytes (32).

in late stages of disease in the CIA model (44). Further supporting the notion of an inflammation‐resolving role of this cytokine, IL‐22 has recently been shown to potentially modulate the IL‐23/IL‐17 inflammatory axis in RA patients by down‐regulating IL‐23 and IL‐17RC expression in fibroblast‐like synoviocytes (32). Although full understanding of the mechanisms underlying the IL‐22–driven resolution of articular inflammation requires further investigation, the above findings are consistent with the notion that ES‐62 resets the balance of IL‐22/IL‐17 signaling in the inflamed synovium from proinflammatory toward desensitization of “pathogenic” synovial fibroblast responses, consequently reducing infiltration of effector cells and joint damage. Thus, and consistent with reports that IL‐22 promotes osteoclast differentiation from human monocytes via RANKL production by synovial fibroblasts in vitro (27), administration of rIL‐22 to the paws during the initiation phase of CIA resulted in enhanced basal and IL‐17–stimulated IL‐6 responses by synovial fibroblasts. In contrast, exposure to IL‐22 in vitro was found to suppress the levels of IL‐6 secreted by synovial fibroblasts derived from mice with established CIA, while synovial fibroblasts from ES‐62–treated mice with CIA, which exhibited elevated serum and joint levels of IL‐22 following disease onset, demonstrated desensitized basal and IL‐17–stimulated IL‐6 responses. These latter data are consistent with reports that Th17 cells from IL‐22–deficient mice induced synovial cells to produce higher levels of IL‐6 than those from wild‐type mice (23), suggesting that (aberrant) release of this proinflammatory cytokine may normally be limited by IL‐22. Our finding that neutralizing anti–IL‐22 antibodies could prevent both ES‐62–mediated desensitization of synovial fibroblast responses and protection against CIA strongly suggests that IL‐22 has a role in desensitizing synovial fibroblasts and promoting resolution of joint inflammation in established arthritis.

IL‐22. Our finding that neutralizing anti–IL‐22 antibodies could prevent both ES‐62–mediated desensitization of synovial fibroblast responses and protection against CIA strongly suggests that IL‐22 has a role in desensitizing synovial fibroblasts and promoting resolution of joint inflammation in established arthritis. A key question therefore relates to the trigger that switches IL‐22 from a pro‐ to an antiinflammatory cytokine in CIA. Our data suggest that this occurs around the time of onset of joint pathology (days 20–25) as in the early stages, both systemically and locally in the joint, IL‐17 and IL‐22 (in the relatively low levels at which they are present) appear to act cooperatively (and may indeed both be derived from Th17 cells as described for IL‐1–driven arthritis [45]) to promote pathogenesis, whereas in the IL‐22–mediated protection phase, IL‐17 levels are high and IL‐17 and IL‐22 appear to be produced by antagonistic cell populations. It is not clear what are the major cell producers of “antiinflammatory IL‐22” or their precise targets other than synovial fibroblasts; elucidation of this may help explain the failure of neutralizing anti–IL‐22 antibodies to exacerbate disease when administered systemically following the onset of pathology. Moreover, in addition to reducing the levels of neutrophils infiltrating the joints, it appears that ES‐62 may have modified such cells functionally to a “protective IL‐22–producing phenotype,” perhaps suggesting that the parasite product provides additional signals to rewire cells to become “protective” sources of IL‐22 and/or targets of the antiinflammatory action of IL‐22.

s infiltrating the joints, it appears that ES‐62 may have modified such cells functionally to a “protective IL‐22–producing phenotype,” perhaps suggesting that the parasite product provides additional signals to rewire cells to become “protective” sources of IL‐22 and/or targets of the antiinflammatory action of IL‐22. It is therefore of interest that infection with the helminth Trichuris trichiura has been reported to therapeutically alleviate ulcerative colitis (46) by a mechanism that is dependent on expansion of IL‐22–positive cells. This suggests that induction of the tissue repair properties of IL‐22 could be a mechanism evolved by worms to promote healing of wounds arising from their invasion, to prevent harmful pathology to the host and/or inflammatory responses that could result in their expulsion. Exploiting the ability of helminth products such as ES‐62 to induce such inflammation‐ and wound‐resolving responses to treat autoimmune disorders is therefore an attractive prospect. Toward this end, we have obtained preliminary data suggesting that PC (the active moiety of ES‐62), when conjugated to BSA, also significantly promotes generation of joint inflammation–resolving IL‐22. Indeed, studies of PC–BSA were the starting point for our development of drug‐like derivatives that mimic the ability of ES‐62 to suppress CIA by targeting pathogenic IL‐17 responses (47) and provide proof‐of‐concept that exploiting the actions of helminth‐derived immunomodulators may potentially open novel avenues for drug discovery in RA.

BSA were the starting point for our development of drug‐like derivatives that mimic the ability of ES‐62 to suppress CIA by targeting pathogenic IL‐17 responses (47) and provide proof‐of‐concept that exploiting the actions of helminth‐derived immunomodulators may potentially open novel avenues for drug discovery in RA. AUTHOR CONTRIBUTIONS 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. M. Harnett 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. Pineda, W. Harnett, M. Harnett. Acquisition of data. Pineda, Rodgers, Al‐Riyami. Analysis and interpretation of data. Pineda, Rodgers, Al‐Riyami, W. Harnett, M. Harnett.

Synovial joints allow movement, consist of multiple tissues and structures, and are considered to be skeletal organs. Osteoarthritis (OA) is a condition of multifactorial organ failure in which pathologic changes in the cartilage, bone, synovium, and other periarticular soft tissues interact (1). This structural failure causes inhibited function of the joint and, combined with chronic pain, results in debilitation and reduced quality of life. Indications of disease progression include destruction of the cartilage in combination with abnormal thickening of the subchondral bone and gross deformity of the affected joint (2). The quality of bone, as a material, may have been underestimated in the understanding of the etiology and progression of disease (3). The interaction between cartilage and subchondral bone and its role in the pathogenesis of OA require further examination, and may provide a mechanical basis for the cartilage degradation process (4,5). It is possible that not one single tissue is responsible (6). Clarification as to whether specific biomolecular changes occur in the onset of OA (7,8) is needed.

ndral bone and its role in the pathogenesis of OA require further examination, and may provide a mechanical basis for the cartilage degradation process (4,5). It is possible that not one single tissue is responsible (6). Clarification as to whether specific biomolecular changes occur in the onset of OA (7,8) is needed. To date, relatively little is known about the relationship between bone (material), structural competence, and the mechanobiologic interaction of bone and cartilage in the etiology of OA, particularly in the initiation and early stages of the disease. Two theories have been proposed. One theory postulates that OA is the result of altered impact mechanical loading, which induces bone adaptation with subchondral bone thickening, and leads to a stiffer structure (4). This new stiffer structure is less effective at absorbing shock, and therefore the distribution of force through the affected joint changes, leading to site-specific cartilage destruction. The “chemical” nature or quality of the newly synthesized bone as a material may be indistinguishable from the original bone tissue.

his new stiffer structure is less effective at absorbing shock, and therefore the distribution of force through the affected joint changes, leading to site-specific cartilage destruction. The “chemical” nature or quality of the newly synthesized bone as a material may be indistinguishable from the original bone tissue. Another theory postulates that an alteration in bone matrix chemistry occurs as part of the disease progression. This hypothesis is supported by studies of OA bone, which show that the tropocollagen molecules comprise 3 α1 chains (rather than 2 α1 chains and 1 α2 chain) (9–12). Bone with homotrimeric collagen has a lower modulus (4.3 GPa) than that with heterotrimeric collagen (4.5 GPa), thus being less able to support the overlying cartilage without a compensatory increase in structural thickness (9,13). In addition, homotrimers have higher water content (14). In this setting, subchondral sclerosis or structural thickening could be a compensatory attempt to provide support for cartilage in the presence of abnormal bone matrix. Conventional technologies, e.g., radiography, are used to assess the extent of damage in OA joints, particularly to identify joint space narrowing and thickened subchondral bone. Vibrational spectroscopic techniques, such as infrared or Raman spectroscopy, may also be applied. The main benefit of spectroscopic techniques over conventional techniques is the capability to detect the organic and inorganic phases of bone, effectively acquiring an overall biochemical signature, without destructive sample preparation (15).

ic techniques, such as infrared or Raman spectroscopy, may also be applied. The main benefit of spectroscopic techniques over conventional techniques is the capability to detect the organic and inorganic phases of bone, effectively acquiring an overall biochemical signature, without destructive sample preparation (15). Raman spectroscopy has been used to assess bone and pathologic changes in the tissue (16,17). Spectral analysis has shown that the hydroxyapatite:collagen ratio (mineral volume fraction), carbonate apatite:hydroxyapatite ratio (carbonate substitution), and amide III (protein conformation) are altered in subchondral bone from the hip joints of patients with OA (18), all of which are indicative of alterations in bone composition and the collagen secondary structure. In these studies, the nearby trabecular bone was not similarly affected, and analysis of samples from the most weight-bearing and least weight-bearing sites did not alter the outcome (18). The increased levels of homotrimeric collagen in the subchondral bone from OA femoral heads (8) may result in the alterations in the collagen secondary structure, as has been observed using Raman spectroscopy (18).

is of samples from the most weight-bearing and least weight-bearing sites did not alter the outcome (18). The increased levels of homotrimeric collagen in the subchondral bone from OA femoral heads (8) may result in the alterations in the collagen secondary structure, as has been observed using Raman spectroscopy (18). Raman spectroscopy is a technique that measures and quantifies the energetic changes in light (typically generated by a laser) scattered from materials. When light scatters from a sample, energy may be lost (or gained) by some photons and a shift in wavelength is observed to the red (energy loss) or the blue (energy gained) region. The shifts in wavelength of the photons are dependent on the chemicals within the material; therefore, Raman spectroscopy gives a chemical “fingerprint,” which, when analyzed, identifies the components present. The energy of the shifted light is plotted as a spectrum of the intensity of scattered light (y-axis) against the wave numbers (x-axis) (1 cm−1 = 1 × 107/ wavelength in nm). Spectral bands for molecular functional groups associated with mineralized tissue include phosphate, carbonate, amides I and III (indicators of secondary protein conformation), proline, hydroxyproline, and lipids (19–21). The heights/areas of the bands can be compared, e.g., the mineral:collagen ratio, thus providing information on the organic and inorganic phases of bone and the degree of mineralization of a given specimen. Multivariate analyses of the Raman spectra can discern subtle differences that cannot be identified by eye. For example, principal components analysis (PCA) may be applied to analyze the spread of data and identify segregation of the spectra (22,23).

phases of bone and the degree of mineralization of a given specimen. Multivariate analyses of the Raman spectra can discern subtle differences that cannot be identified by eye. For example, principal components analysis (PCA) may be applied to analyze the spread of data and identify segregation of the spectra (22,23). In this study, we explored the hypothesis that changes in bone matrix chemistry in the subchondral bone of the tibial plateau of patients with knee OA can be detected by Raman spectroscopy. Furthermore, the spectral distinction between OA and non-OA may be identified as changes in the organic phase (amide bands), which thereby result in a change in the mineral:collagen ratio. The load-bearing regions of both the medial (grossly affected) and lateral (not visually damaged at a macroscopic level) compartments of human tibial plateaus were probed independently, and samples were sex-matched (with a subset age-matched) to non-OA joints. The findings from Raman spectroscopy (overall biochemical signature) along with the findings obtained by peripheral quantitative computed tomography (pQCT) (mineral component) and the α-chain ratio (organic component) were assessed for correlations.

samples were sex-matched (with a subset age-matched) to non-OA joints. The findings from Raman spectroscopy (overall biochemical signature) along with the findings obtained by peripheral quantitative computed tomography (pQCT) (mineral component) and the α-chain ratio (organic component) were assessed for correlations. PATIENTS AND METHODS Samples Human tibial plateaus were acquired from patients with knee OA (n = 10) following the patients' provision of informed consent and ethics approval from the Royal National Orthopaedic Hospital, UK (ethics approval no. 08/H0304/78). The tibial plateaus were obtained from patients undergoing total knee replacement for established grade IV (Outerbridge classification) medial compartment OA (radiographic and macroscopic diagnosis). Control specimens were collected from non-OA patients undergoing various operations (due to different conditions of the proximal femur, with no evidence of tibial involvement, no OA of the knee joint, and without changes in clinically assessed load-bearing) requiring removal of the leg (n = 5). In addition, control samples were obtained from non-OA cadaveric specimens (n = 5, from 3 age-matched subjects) at Vesalius Clinical Training Centre, University of Bristol (ethics approval no. 08/H0724/34). The patients with OA (mean ± SD age 68 ± 15 years) and the non-OA subjects (mean ± SD age 75 ± 15 years for the cadaveric specimens and 30 ± 18 years for the amputees) were matched for sex and laterality. All non-OA specimens were examined by an orthopedic surgeon, and no visual appearance of OA was found, either macroscopically or radiographically. Figure 1A shows a typical OA specimen, and Figure 1B is the corresponding radiograph with evidence of clear joint space narrowing and thickening of the medial subchondral bone. Figures 1D and E show a non-OA equivalent specimen.

geon, and no visual appearance of OA was found, either macroscopically or radiographically. Figure 1A shows a typical OA specimen, and Figure 1B is the corresponding radiograph with evidence of clear joint space narrowing and thickening of the medial subchondral bone. Figures 1D and E show a non-OA equivalent specimen. Figure 1 Assessment of the tibial plateau from a patient with osteoarthritis (OA) compared to that from a non-OA control. Postoperative photographs (A and D) and preoperative radiographs (B and E) of the medial (M) and lateral (L) tibial plateaus from a representative patient with OA (A and B) and a non-OA control (D and E) are shown. The medial tibial plateau (C) and lateral tibial plateau (F) from a patient with OA were assessed by peripheral quantitative computed tomography (pQCT). In the pQCT images, the region of interest is demarcated as a green box. The red broken lines in A and D indicate the plane from which the pQCT measurements were obtained. Samples were frozen at −80°C within 2 hours of removal, for storage prior to analysis. Cadaveric specimens were kept in storage at −20°C at the Vesalius Clinical Training Centre within 48 hours of death, and subsequently transported frozen.

Figure 1 Assessment of the tibial plateau from a patient with osteoarthritis (OA) compared to that from a non-OA control. Postoperative photographs (A and D) and preoperative radiographs (B and E) of the medial (M) and lateral (L) tibial plateaus from a representative patient with OA (A and B) and a non-OA control (D and E) are shown. The medial tibial plateau (C) and lateral tibial plateau (F) from a patient with OA were assessed by peripheral quantitative computed tomography (pQCT). In the pQCT images, the region of interest is demarcated as a green box. The red broken lines in A and D indicate the plane from which the pQCT measurements were obtained. Samples were frozen at −80°C within 2 hours of removal, for storage prior to analysis. Cadaveric specimens were kept in storage at −20°C at the Vesalius Clinical Training Centre within 48 hours of death, and subsequently transported frozen. Peripheral QCT The medial and lateral compartments of the tibial plateaus were scanned using pQCT (XCT 3000; Stratec) to determine volumetric bone mineral density (vBMD) and thickness of the subchondral bone (Figures 1C and F). The x-ray beam width was 2 mm and the voxel size was fixed at 0.3 mm. Specimens were measured at 5-mm intervals across the tibial plateau. A measurement slice from the center of each compartment of the plateau (demarcated with the red broken lines in Figures 1A and D) was used to calculate the vBMD (in mg/cm3) of the subchondral bone; the regions of interest used were the same size and in the same location across all measurements (Figures 1C and F). Additionally, the thickness of the subchondral bone was measured (in mm).

au (demarcated with the red broken lines in Figures 1A and D) was used to calculate the vBMD (in mg/cm3) of the subchondral bone; the regions of interest used were the same size and in the same location across all measurements (Figures 1C and F). Additionally, the thickness of the subchondral bone was measured (in mm). Raman microspectroscopy Cylindrical cores (dimension ∼12 mm2 × height of plateau [range 10–20 mm]) were extracted from the tibial plateaus using an 11-gauge Jamshidi Crown bone marrow biopsy needle (Cardinal Health, France), with 2 samples obtained from each compartment, corresponding to the same sites analyzed by pQCT. The cores were turned on their side, and spectra were acquired directly from the subchondral bone (≤3 mm below the cartilage, if present) using an InVia Raman microspectrometer (Renishaw). This was equipped with an 830-nm laser, 300 mW at source. Calibration was performed every day on silicon, as silicon has a known Raman band (520.5 cm−1), and on polystyrene to measure bands in the same range as that of bone (wave numbers between 1000 cm−1 and 1030 cm−1). There was no difference across the values during the time of data collection. The spectra were acquired at a laser power of 2 mW for 1 second and 4 accumulations; 5 different spectra were acquired from the subchondral bone per core (total number of spectra = 400). The spatial resolution (at 50× objective) was 2 μm × 2 μm. There was no thermal heating or degradation of the samples at the laser powers utilized.

re acquired at a laser power of 2 mW for 1 second and 4 accumulations; 5 different spectra were acquired from the subchondral bone per core (total number of spectra = 400). The spatial resolution (at 50× objective) was 2 μm × 2 μm. There was no thermal heating or degradation of the samples at the laser powers utilized. Lipid removal The removal of lipids was required because lipids are strong light scatterers on Raman microspectroscopy, having spectral peaks that overlie some of the spectral peaks from bone matrix. Following established bone preparation techniques (24,25), the cores were washed in 5 ml of acetone on a roller for 1 hour at 37°C with constant agitation. The acetone was removed and replaced with fresh acetone twice. The cores were then rinsed in distilled H2O to remove the acetone. The spectral measurements were then repeated with the same parameters. The effect of lipid removal with acetone was validated by comparing peak ratios, namely the ratios of phosphate to amide I and phosphate to carbonate, before and after lipid removal (details available from the corresponding author upon request).

ne. The spectral measurements were then repeated with the same parameters. The effect of lipid removal with acetone was validated by comparing peak ratios, namely the ratios of phosphate to amide I and phosphate to carbonate, before and after lipid removal (details available from the corresponding author upon request). Polarization and orientation The laser used generates polarized light, and certain materials/molecular bonds that are preferentially orientated with respect to the laser polarization can be more or less efficient at scattering light. Therefore, it was important to assess the influence of polarization and orientation on these specimens. All spectra, as described above, were acquired without the polarizer and at the same orientation (within 10°). Additional spectra were acquired with a polarizer in place, i.e., the laser was fully polarized, under the same parameters and from the same location (within 1 μm) every 45° from 0° to 360°. Spectra were then processed as described below, and the mineralization ratios were calculated for the ratio of phosphate (η1; 960 cm−1) to amide I and the ratio of phosphate (η4; 588 cm−1) to amide III; of note, the former measure, being a symmetric mode, is more sensitive to polarization (26) (details available from the corresponding author upon request).

ed below, and the mineralization ratios were calculated for the ratio of phosphate (η1; 960 cm−1) to amide I and the ratio of phosphate (η4; 588 cm−1) to amide III; of note, the former measure, being a symmetric mode, is more sensitive to polarization (26) (details available from the corresponding author upon request). Type I collagen α-chain analysis Sample preparation Defatted cores of the subchondral bone were weighed and then decalcified in 10 ml 10% EDTA for 1 week on a roller at 4°C. The cores were washed with deionized water (to remove EDTA), and then reweighed and freeze-dried. Five milligrams from each sample was isolated for further processing. These cores were immersed in 1 ml 0.5M acetic acid with 25 μg pepsin (porcine gastric mucosa, 3,200–4,500 units/mg protein; Sigma) added at 0.5% weight/weight of 5 mg cores. Samples were left to digest with agitation for 2 days at 4°C. The supernatants were collected following centrifugation at 3,000 revolutions per minute for 30 minutes at 4°C, and freeze-dried overnight.

25 μg pepsin (porcine gastric mucosa, 3,200–4,500 units/mg protein; Sigma) added at 0.5% weight/weight of 5 mg cores. Samples were left to digest with agitation for 2 days at 4°C. The supernatants were collected following centrifugation at 3,000 revolutions per minute for 30 minutes at 4°C, and freeze-dried overnight. Sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE) The samples were dissolved in 0.5 ml of SDS-PAGE sample buffer (125 mM Tris [pH 6.8], 2% SDS, 10% glycerol, 0.01% bromphenol blue). SDS-PAGE was performed using 7.5% polyacrylamide resolving gel matrix and 4.5% stacking gel in a Mini-Protean II apparatus (Bio-Rad, UK). After electrophoresis, the gels were stained with Coomassie blue (0.005%) to visualize protein, and then destained with 10% acetic acid and 20% methanol solution. A type I collagen standard prepared from equine skin and molecular weight markers were run alongside the samples. Following staining and destaining, a digital image of the gel was obtained, and ImageJ software was used to quantify the intensities of the α1- and α2-chain bands (details available from the corresponding author upon request). The α1:α2 chain ratio was calculated using a correction factor (1.16) to account for the smaller α2 polypeptide chain.

staining, a digital image of the gel was obtained, and ImageJ software was used to quantify the intensities of the α1- and α2-chain bands (details available from the corresponding author upon request). The α1:α2 chain ratio was calculated using a correction factor (1.16) to account for the smaller α2 polypeptide chain. Statistical analysis Spectra were baseline corrected (750–1800 cm−1) using a third-order polynomial (Matlab; The Mathworks), with values normalized to the phosphate peak (960 cm−1). For multivariate analyses, PCA (unsupervised) and PCA-linear discriminant analysis (PCA-LDA; supervised) were performed to facilitate the identification of segregation of the spectra, based on variance, by forming linear combinations of the wave numbers and ranking them in order of variance (Matlab 2012a; The Mathworks) (27,28). The Raman spectra in each cohort were averaged (Figure 2). Subjectively, there were few spectral differences between the OA and non-OA samples; however, subtle differences are difficult to identify by observing the average spectra. A 3-dimensional scatter plot of the resulting scores obtained from the multivariate analysis (PCA) allows for differences to be identified objectively, because the distance between the scores (expressed as points; each point represents a spectrum) on the plot is proportional to the spectral, and therefore the biochemical, similarities.

catter plot of the resulting scores obtained from the multivariate analysis (PCA) allows for differences to be identified objectively, because the distance between the scores (expressed as points; each point represents a spectrum) on the plot is proportional to the spectral, and therefore the biochemical, similarities. Figure 2 Average intensity of the Ramen spectra (wave numbers 750–1800 cm−1) in A, non-osteoarthritis (non-OA) (purple) versus OA (orange) tibial specimens (difference shown in black), and in B, non-OA medial (black) versus non-OA lateral (blue) compartments, and OA medial (green) versus OA lateral (red) compartments. Univariate analysis was utilized for direct comparison of the bands of interest, i.e., the ratio of phosphate (960 cm−1) to amide I (1660 cm−1), the ratio of phosphate to proline (920 cm−1), the ratio of phosphate to carbonate (1070 cm−1), and the ratio of phosphate to hydroxyproline (885 cm−1 + 870 cm−1), which provides the relative amount of bioapatite to collagen (29). Peak heights were compared to determine the ratios. The α-chain ratios were analyzed using the Mann-Whitney U test, as a nonparametric equivalent to the independent-samples t-test (SPSS; IBM UK). One-way analysis of variance, followed by the Bonferroni post hoc test, was used to compare the density, thickness, and median spectral ratios across the multiple groups (Origin 8.6; OriginLab). Throughout the statistical testing, the number of subjects per cohort was noted.

e independent-samples t-test (SPSS; IBM UK). One-way analysis of variance, followed by the Bonferroni post hoc test, was used to compare the density, thickness, and median spectral ratios across the multiple groups (Origin 8.6; OriginLab). Throughout the statistical testing, the number of subjects per cohort was noted. RESULTS Density and thickness of the subchondral bone The density of the subchondral bone was significantly higher across the OA tibial plateaus (mean ± SD 950 ± 100 mg/cm3) compared to the non-OA samples (mean ± SD 820 ± 90 mg/cm3; P < 0.001) (Figure 3A). In comparing the tibial compartments, the medial subchondral bone had a significantly higher structural density than the lateral subchondral bone (mean ± SD 940 ± 94 mg/cm3 versus 884 ± 25 mg/cm3; P = 0.005) in all OA specimens, whereas only 3 of the 10 non-OA specimens showed a difference in density, albeit not significant (P = 0.2), between compartments. Moreover, the OA medial compartments were significantly denser than the non-OA medial compartments (P = 0.003) (Figure 3A). Figure 3 Results of peripheral quantitative computed tomography assessing A, the subchondral bone density and B, the subchondral bone thickness of each tibial compartment, comparing the medial (black squares) and lateral (blue circles) compartments of non-osteoarthritis (non-OA) tibial plateaus and the medial (green triangles) and lateral (red diamonds) compartments of OA tibial plateaus. * = P < 0.05.

hondral bone density and B, the subchondral bone thickness of each tibial compartment, comparing the medial (black squares) and lateral (blue circles) compartments of non-osteoarthritis (non-OA) tibial plateaus and the medial (green triangles) and lateral (red diamonds) compartments of OA tibial plateaus. * = P < 0.05. With regard to subchondral bone thickness, the medial compartment of the OA specimens was significantly thicker than the lateral compartment, and was significantly thicker in OA specimens compared to non-OA specimens (Figure 3B). The thickness of the non-OA medial and lateral compartments was comparable (mean ± SD 1.3 ± 0.2 mm). Raman spectral signatures Non-OA versus OA PCA revealed a large separation of spectra (confidence interval 0.95) between the non-OA and OA specimens (Figure 4A). However, there was no difference between the medial and lateral compartments within each cohort (Figure 4C).

With regard to subchondral bone thickness, the medial compartment of the OA specimens was significantly thicker than the lateral compartment, and was significantly thicker in OA specimens compared to non-OA specimens (Figure 3B). The thickness of the non-OA medial and lateral compartments was comparable (mean ± SD 1.3 ± 0.2 mm). Raman spectral signatures Non-OA versus OA PCA revealed a large separation of spectra (confidence interval 0.95) between the non-OA and OA specimens (Figure 4A). However, there was no difference between the medial and lateral compartments within each cohort (Figure 4C). Figure 4 Principal components analysis (PCA) of the Ramen spectra. A–C, Plot of PCA scores for the non-osteoarthritis (non-OA) tibial specimens compared to the OA tibial specimens (A), the corresponding PCA loadings plot (B), and the color-coded plot of PCA scores in the non-OA and OA medial and lateral compartments (color-coded to enable identification of the medial and lateral spectra) (C). D–F, Plot of PCA–linear discriminant analysis (PCA-LDA) scores for the non-OA and OA medial and lateral compartments (D), the corresponding PCA-LDA loadings plot (E), and the plot of PCA-LDA scores for the non-OA tibial specimens compared to the OA tibial specimens (F). The data in F are plotted along the x-axis, and split along the y-axis only for ease of visualization. The loadings plots (B and E) show the axes (linear combination of the variables) for the PCA-LDA and LDA analyses, respectively. In loadings plots, the larger the peak, the more influence it has on any separation in the scores plotted along a particular axis.

, and split along the y-axis only for ease of visualization. The loadings plots (B and E) show the axes (linear combination of the variables) for the PCA-LDA and LDA analyses, respectively. In loadings plots, the larger the peak, the more influence it has on any separation in the scores plotted along a particular axis. The spectral differences resolved from PCA can be identified by interpreting Figure 4B. The first axis (PC1) shows that the phosphate band (954 cm−1 and 966 cm−1) was different between the cohorts. The second axis (PC2) reveals that the second largest difference was in amide I (1668 cm−1 and 1685 cm−1) and the phosphate shoulder (941 cm−1). Finally, the third axis (PC3) shows that the next difference was related to broad differences across ∼1597 cm−1. Univariate analysis of the Raman spectral peaks (Table 1) revealed statistically significant differences between the cohorts for the phosphate:amide I ratio (P = 0.04) and the bioapatite:collagen ratio (P = 0.04), in the medial and lateral compartments combined. Table 1 Results of Raman spectral and biochemical analyses* Raman spectral analysis

Univariate analysis of the Raman spectral peaks (Table 1) revealed statistically significant differences between the cohorts for the phosphate:amide I ratio (P = 0.04) and the bioapatite:collagen ratio (P = 0.04), in the medial and lateral compartments combined. Table 1 Results of Raman spectral and biochemical analyses* Raman spectral analysis Phosphate:amide I Carbonate:phosphate Bioapatite:collagen Biochemical analysis, α1:α2 chain ratio Non-OA Medial compartment 13.12 (4.47) 0.17 (0.02) 8.5 (1.92) 2.0 (1.03–3.15):1.0 Lateral compartment 13.45 (3.36) 0.17 (0.01) 8.99 (2.98) 2.0 (1.44–3.62):1.0 P for comparison 1 1 1 0.5 OA Medial compartment 13.96 (2.98) 0.18 (0.02) 9.90 (2.11) 3.0 (1.58–8.08):1.0 Lateral compartment 14.70 (3.56) 0.18 (0.01) 10.7 (2.42) 2.3 (1.63–9.90):1.0 P for comparison 1 1 0.3 0.4 Medial + lateral compartments Non-OA 13.28 (3.86) 0.17 (0.02) 8.75 (2.38) 2.0 (1.03–3.62):1.0 OA 14.33 (3.47) 0.18 (0.02) 10.3 (2.27) 2.6 (1.58–9.90):1.0 P for comparison 0.04† 0.1 0.04† 0.2 Medial compartment P, non-OA vs. OA 0.4 0.3 0.2 0.03† * Results of the Raman spectral analysis are the median ratio (interquartile range) determined by univariate analysis for bone mineralization (ratio of phosphate to amide I), bone turnover (ratio of carbonate to phosphate), and the bioapatite:collagen ratio in osteoarthritis (OA) and non-OA tibial plateau specimens. Results of the biochemical analysis are the ratio of type I collagen α1 chain, expressed as the median (range), to α2 chain for each cohort. † P value is statistically significant at the 95% confidence level.

Phosphate:amide I Carbonate:phosphate Bioapatite:collagen Biochemical analysis, α1:α2 chain ratio Non-OA Medial compartment 13.12 (4.47) 0.17 (0.02) 8.5 (1.92) 2.0 (1.03–3.15):1.0 Lateral compartment 13.45 (3.36) 0.17 (0.01) 8.99 (2.98) 2.0 (1.44–3.62):1.0 P for comparison 1 1 1 0.5 OA Medial compartment 13.96 (2.98) 0.18 (0.02) 9.90 (2.11) 3.0 (1.58–8.08):1.0 Lateral compartment 14.70 (3.56) 0.18 (0.01) 10.7 (2.42) 2.3 (1.63–9.90):1.0 P for comparison 1 1 0.3 0.4 Medial + lateral compartments Non-OA 13.28 (3.86) 0.17 (0.02) 8.75 (2.38) 2.0 (1.03–3.62):1.0 OA 14.33 (3.47) 0.18 (0.02) 10.3 (2.27) 2.6 (1.58–9.90):1.0 P for comparison 0.04† 0.1 0.04† 0.2 Medial compartment P, non-OA vs. OA 0.4 0.3 0.2 0.03† * Results of the Raman spectral analysis are the median ratio (interquartile range) determined by univariate analysis for bone mineralization (ratio of phosphate to amide I), bone turnover (ratio of carbonate to phosphate), and the bioapatite:collagen ratio in osteoarthritis (OA) and non-OA tibial plateau specimens. Results of the biochemical analysis are the ratio of type I collagen α1 chain, expressed as the median (range), to α2 chain for each cohort. † P value is statistically significant at the 95% confidence level. Medial versus lateral compartments PCA-LDA was used to separate the 4 groups, comprising comparisons of the non-OA medial versus non-OA lateral compartments, and OA medial versus OA lateral compartments. The findings, shown in Figure 4D, confirm that there were differences between the non-OA and OA specimens, and that there were no differences between compartments within each cohort. The chemical components that showed the most differences in intensity between the non-OA and OA specimens, regardless of compartment (Figure 4B), were hydroxyproline (858 cm−1), C–C collagen backbone (941 cm−1), and phosphate (956 cm−1 and 966 cm−1).

at there were no differences between compartments within each cohort. The chemical components that showed the most differences in intensity between the non-OA and OA specimens, regardless of compartment (Figure 4B), were hydroxyproline (858 cm−1), C–C collagen backbone (941 cm−1), and phosphate (956 cm−1 and 966 cm−1). Medial non-OA versus medial OA, and lateral non-OA versus lateral OA Analysis of the non-OA medial compartment compared to the OA medial compartment (Figures 5A and B) revealed a spectral separation that could be attributed to differences in the intensity of phosphate (944 cm−1 and 953 cm−1), amide III (1275 cm−1), and a broad region across amide I (1650 cm−1). Analysis of the lateral compartment in non-OA specimens compared to OA specimens (Figures 5C and D) indicated that there was a biochemical difference, with more intracategory variance, in the non-OA specimens. The spectral bands contributing to the differences were the bands at 850 cm−1 and 910 cm−1, and the phosphate bands (963 cm−1 and 956 cm−1). Figure 5 Principal components analysis (PCA) showing the plot of PCA scores for the non-osteoarthritis (non-OA) and OA medial compartment (A) and the corresponding loadings plot (B), as well as the plot of PCA scores for the non-OA and OA lateral compartment (C) and the corresponding loadings plot (D).

Medial non-OA versus medial OA, and lateral non-OA versus lateral OA Analysis of the non-OA medial compartment compared to the OA medial compartment (Figures 5A and B) revealed a spectral separation that could be attributed to differences in the intensity of phosphate (944 cm−1 and 953 cm−1), amide III (1275 cm−1), and a broad region across amide I (1650 cm−1). Analysis of the lateral compartment in non-OA specimens compared to OA specimens (Figures 5C and D) indicated that there was a biochemical difference, with more intracategory variance, in the non-OA specimens. The spectral bands contributing to the differences were the bands at 850 cm−1 and 910 cm−1, and the phosphate bands (963 cm−1 and 956 cm−1). Figure 5 Principal components analysis (PCA) showing the plot of PCA scores for the non-osteoarthritis (non-OA) and OA medial compartment (A) and the corresponding loadings plot (B), as well as the plot of PCA scores for the non-OA and OA lateral compartment (C) and the corresponding loadings plot (D). Polarization Analyses of the results from the polarization/orientation tests showed that the Raman spectra were sensitive to sample orientation, but the variations were smaller than the differences between the samples (details available from the corresponding author upon request). Therefore, we can conclude that polarization had no effect on our findings.

from the polarization/orientation tests showed that the Raman spectra were sensitive to sample orientation, but the variations were smaller than the differences between the samples (details available from the corresponding author upon request). Therefore, we can conclude that polarization had no effect on our findings. Biochemical findings The OA specimens had a significantly higher α1:α2 chain ratio compared to the non-OA specimens (P = 0.03) (Table 1). The largest α1:α2 chain ratio in the non-OA specimens was 3.62:1.0 (lowest ratio 1.03:1.0; mean ratio 2.0:1.0). In the OA specimens, the largest α1:α2 chain ratio was 9.90:1.0 (lowest ratio 1.58:1.0; mean ratio 2.6:1.0).

had a significantly higher α1:α2 chain ratio compared to the non-OA specimens (P = 0.03) (Table 1). The largest α1:α2 chain ratio in the non-OA specimens was 3.62:1.0 (lowest ratio 1.03:1.0; mean ratio 2.0:1.0). In the OA specimens, the largest α1:α2 chain ratio was 9.90:1.0 (lowest ratio 1.58:1.0; mean ratio 2.6:1.0). DISCUSSION Results from the pQCT, Raman spectral, and collagen analyses all showed significant cohort-specific differences between the OA and non-OA samples. The results support our hypothesis that bone matrix changes found in OA specimens can be detected with Raman spectroscopy. An unexpected new finding is that this spectral difference between OA and non-OA tissue was discernible in both the lateral and medial compartments. The OA medial compartment was expected to be different, as all OA specimens were confirmed to have grade IV OA (30) in the medial compartment. The previously reported enhanced levels of homotrimeric type I collagen in OA bone were confirmed by our findings in the grossly affected OA medial compartment. However, elevated levels of homotrimeric collagen and associated spectral differences were also found in the grossly normal lateral compartment of the OA specimens, indicating that the whole joint is affected.

meric type I collagen in OA bone were confirmed by our findings in the grossly affected OA medial compartment. However, elevated levels of homotrimeric collagen and associated spectral differences were also found in the grossly normal lateral compartment of the OA specimens, indicating that the whole joint is affected. Damage and repair may enhance tissue turnover in individuals with OA, e.g., during weight-bearing. Previous studies have suggested that the high tissue turnover rate and proliferation lead to thickening of the subchondral bone (31). Differences in the lateral compartment suggest that underlying biochemical changes are taking place, leading to dissemination of gross pathologic alterations across the joint. Although the lateral compartment is not showing macroscopically visible/symptomatic changes toward OA, it is exhibiting small changes, suggestive of early signs of OA, or developing inherent differences in the bone matrix.

nges are taking place, leading to dissemination of gross pathologic alterations across the joint. Although the lateral compartment is not showing macroscopically visible/symptomatic changes toward OA, it is exhibiting small changes, suggestive of early signs of OA, or developing inherent differences in the bone matrix. OA of the knee is a painful condition that changes the loading of the limb, and therefore mechanobiologic factors may be a part of the changes observed. Furthermore, the OA cohort could be exhibiting an inherent difference in bone matrix chemistry, with associated material properties, which would predispose individuals to mechanobiologic joint degeneration. This theory is speculative, but it is known that there can be different responses of bone to increased and decreased mechanical loading (32), and that the material properties of bone across different (nonmutant) strains of a single species (of mice) can vary by up to 27% (33). The current findings may contribute evidence to suggest that individual variation is more complex than previously known, and therefore the findings warrant further investigation to determine whether they are inherent genetic differences or preclinical manifestations of disease. Confirmation of either finding would be significant, particularly in providing opportunities for early diagnosis.

variation is more complex than previously known, and therefore the findings warrant further investigation to determine whether they are inherent genetic differences or preclinical manifestations of disease. Confirmation of either finding would be significant, particularly in providing opportunities for early diagnosis. Subchondral bone thickness from a healthy joint is known to be in the average range of 1 mm (±SD 0.28) to 1.2 mm (±SD 0.41) for both compartments (34). Consistent with this, the mean ± SD measurement of subchondral bone thickness in this study, in each of the non-OA compartments and the OA lateral compartments, was 1.2 ± 0.23 mm. Comparatively, subchondral bone in the OA medial compartment was significantly thicker (3.0 ± 0.8 mm), which is consistent with previous observations in which OA subchondral bone was noted to be thicker (6). Interestingly, the density of the OA subchondral bone was increased in the lateral and medial compartments. The differences observed between non-OA and OA specimens could be attributed to the spectral signatures associated with phosphate, amide I, and phenylalanine tissue components. This suggests that there are changes in collagen that may affect mineralization of the subchondral bone. The spectral, and therefore biochemical, difference between OA and non-OA subchondral bone should be explored further to identify potential therapeutic targets for new pharmacologic treatments of early-stage OA and preclinical OA. Early-stage treatment could result in a delay or reduction in joint replacements.

l bone. The spectral, and therefore biochemical, difference between OA and non-OA subchondral bone should be explored further to identify potential therapeutic targets for new pharmacologic treatments of early-stage OA and preclinical OA. Early-stage treatment could result in a delay or reduction in joint replacements. The medial compartments in the OA and non-OA specimens could be distinguished by differences in the low–wave number shoulder of the phosphate peak and in the amide I peak. However, differences in the lateral compartments between OA and non-OA specimens were associated with hydroxyproline/proline, amide I, and centroid of the phosphate peak, supporting the theory that there are collagen changes in the bone. The hydroxylation of proline is an important contributor to the strength of collagen, as it forms hydrogen bonds and water bridges, which stabilize the triple helix (35,36). The distinguishable changes across the 2 compartments could identify a progression at the late stage of OA (medial) compared to the early stage of OA (lateral).

ion of proline is an important contributor to the strength of collagen, as it forms hydrogen bonds and water bridges, which stabilize the triple helix (35,36). The distinguishable changes across the 2 compartments could identify a progression at the late stage of OA (medial) compared to the early stage of OA (lateral). The higher mineralization ratio (phosphate:amide I) in OA compared to non-OA specimens was consistent with the pQCT density measurements, and suggests that a change/difference in the bone matrix composition precedes subchondral bone thickening. These findings support the theory that a material change to subchondral bone, rather than just an increase in volume, occurs. The second band ratio calculation that supports this finding is the bioapatite:collagen quantification, which shows that OA subchondral bone, as compared to non-OA specimens, has more mineral than collagen. This demonstrates a definable difference between the OA and non-OA specimens that can be resolved by Raman spectroscopy.

ond band ratio calculation that supports this finding is the bioapatite:collagen quantification, which shows that OA subchondral bone, as compared to non-OA specimens, has more mineral than collagen. This demonstrates a definable difference between the OA and non-OA specimens that can be resolved by Raman spectroscopy. Lipid removal was successful, as confirmed by a lack of lipid-specific peaks at 1300 cm−1 and 1750 cm−1(details available from the corresponding author upon request). This process was important, as lipids are strong Raman scatterers and their presence could mask the bone at the small spatial resolutions that were utilized. The polarization/orientation effects seen were smaller than the differences due to disease (results available from the corresponding author upon request). Furthermore, since all of the specimens were measured in the same orientation (within 10°), any impact from polarization effects would be minimized. A recent study of polarization effects on bone confirms the orientation-induced changes in the peaks. The bioapatite:collagen peak is the most reliable, as the proline and phosphate peaks are phase-matched peaks, i.e., they change in proportion to polarization/orientation (37). In addition, the validation study revealed that the OA specimens were more sensitive to polarization/orientation than the non-OA specimens.

ks. The bioapatite:collagen peak is the most reliable, as the proline and phosphate peaks are phase-matched peaks, i.e., they change in proportion to polarization/orientation (37). In addition, the validation study revealed that the OA specimens were more sensitive to polarization/orientation than the non-OA specimens. Considering the potential of Raman spectroscopy as a diagnostic tool, the recently developed technique of spatially offset Raman spectroscopy (SORS) (38) allows the acquisition of spectra from up to 4 mm below the surface, e.g., bone through skin, or through cartilage arthroscopically (39–43). As these approaches in a clinical setting would be minimally invasive and nondestructive, and would not involve ionizing radiation, they have the potential to be a powerful complementary technique to current technologies. SORS, as a noninvasive, nonionizing technique able to probe both the organic and inorganic phases of bone, has potential to be used as a diagnostic tool in vivo. The delivery of the laser and collection of the Raman photons further minimizes the effects of polarization/orientation associated with the laser, and the millimeter scale and depth measurement should reduce the contribution of lipids and probe a larger area of bone, thus minimizing the effect of heterogeneities within the bone structure.

laser and collection of the Raman photons further minimizes the effects of polarization/orientation associated with the laser, and the millimeter scale and depth measurement should reduce the contribution of lipids and probe a larger area of bone, thus minimizing the effect of heterogeneities within the bone structure. Results of the biochemical analyses showed that there was less homotrimeric collagen (ratio range 1.6:1 to 9.9:1) than that previously reported (ratio range 4:1 to 17:1) (8). There may be several reasons for this. First, this study used bone from tibial plateaus, whereas the levels previously reported used femoral heads (8). Second, the area measured in the femoral heads was a minimum of 1 cm below the cartilage (mostly epiphyseal trabecular bone), and therefore unlikely to include subchondral bone, so there cannot be a direct comparison between our data and those previously reported (8).

levels previously reported used femoral heads (8). Second, the area measured in the femoral heads was a minimum of 1 cm below the cartilage (mostly epiphyseal trabecular bone), and therefore unlikely to include subchondral bone, so there cannot be a direct comparison between our data and those previously reported (8). Studies have shown that homotrimeric type I collagen molecules have an increased hydroxyproline content and increased denaturation temperature compared to heterotrimeric molecules (14). An increased molecular stability is further supported by the finding that resistance to cleavage by collagenase is increased due to less efficient unwinding at the cleavage site (44). The α1 chain is less hydrophobic than the α2 chain, which may result in increased water content of the fibril and, consequently, increased spacing between collagen molecules (14). The increased distance between collagen molecules has been implicated in the reduction in immature collagen crosslink levels and decreased mechanical strength observed in type I homotrimer oim mice (14). Furthermore, because mineralization occurs preferentially in hydrophilic environments, this may explain the increased mineralization levels of the OA bone. Molecular simulation studies have suggested that homotrimeric collagen molecules are more flexible and form kinks more freely than do heterotrimeric molecules, providing an alternative explanation for the increased spacing and decreased intermolecular crosslinking in homotrimeric collagen (45). In previous studies, the presence of homotrimeric collagen was associated with a higher rate of collagen turnover (11). However, in our study, we found no evidence of increased matrix turnover.

an alternative explanation for the increased spacing and decreased intermolecular crosslinking in homotrimeric collagen (45). In previous studies, the presence of homotrimeric collagen was associated with a higher rate of collagen turnover (11). However, in our study, we found no evidence of increased matrix turnover. In summary, in the present study, differences were found between non-OA subchondral bone and the grossly affected medial compartment of OA subchondral bone. We were able to detect these changes using a nondestructive spectroscopic technique. Furthermore, these differences were comparable to that in the OA subchondral bone beneath macroscopically intact cartilage of the lateral compartment. There was a large significant difference between the profiles of the non-OA and OA specimens, which could be due to predisposition, increased turnover, or change in loading across the joint. The study results also indicated that the subchondral bone matrix chemistry across the whole joint was affected, possibly due to different stages through the progression of the disease, and thus the mineralization process was affected. This supports research previously reported in animal models (46,47).

ross the joint. The study results also indicated that the subchondral bone matrix chemistry across the whole joint was affected, possibly due to different stages through the progression of the disease, and thus the mineralization process was affected. This supports research previously reported in animal models (46,47). Our findings thus indicate that subchondral bone changes, or inherent differences, exist in both the medial and lateral compartments of the OA tibial plateau. Furthermore, changes can be found beneath visibly unaffected cartilage of the lateral compartment, indicating, for the first time, that the joint as a whole is predisposed to develop OA. Subchondral bone therefore represents a key target for therapeutic strategies (mechanical or pharmaceutical). The detection of bone matrix chemistry variations, by Raman spectroscopy coupled with multivariate analysis, in subchondral bone would facilitate the identification of clinical disease, including early molecular changes. It is possible that these changes may be inherent to the individual, and therefore a better understanding of the changes would enable identification of those at risk of OA.

coupled with multivariate analysis, in subchondral bone would facilitate the identification of clinical disease, including early molecular changes. It is possible that these changes may be inherent to the individual, and therefore a better understanding of the changes would enable identification of those at risk of OA. We wish to thank the Engineering and Physical Sciences Research Council for funding this study. In addition, we thank the Vesalius Clinical Training Centre, University of Bristol, for providing the cadaveric specimens and for the generosity of the donors. Thanks also go to the Royal National Orthopaedic Hospital, within the University College London Partnership, for supporting this study and providing patients. Finally, we thank Suzie Cro (statistician, Medical Research Council Clinical Trials Unit) for her statistical advice and input with regard to the data analysis. AUTHOR CONTRIBUTIONS 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. Kerns 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. Kerns, Buckley, Miles, Briggs, Keen, Parker, Goodship. Acquisition of data. Kerns, Gikas, Shepperd, Birch, McCarthy, Miles, Briggs. Analysis and interpretation of data. Kerns, Gikas, Buckley, Shepperd, Birch, McCarthy, Parker, Matousek, Goodship.

Rheumatoid arthritis (RA) is a common inflammatory joint disorder that may severely impair quality of life due to chronic, persistent pain and functional loss. In the early phase it is characterized by edema and tenderness around the affected joints, which are later accompanied by progressive, irreversible degeneration and bone remodeling. Immunologic factors that are predominantly involved in the inflammatory components of RA have been well described. In contrast, the mechanisms of chronic pain and sensitization, as well as the complexity of neurovascular and neuroimmune interactions, have received much less study, although the importance of neurogenic inflammation in the pathophysiologic processes of RA was described long ago (1). The relatively recent availability of biologic therapies has considerably improved the treatment of RA, but early intervention and identification of novel drug targets are imperative to prevent the disease from progressing to the less manageable late phase (2,3).

ophysiologic processes of RA was described long ago (1). The relatively recent availability of biologic therapies has considerably improved the treatment of RA, but early intervention and identification of novel drug targets are imperative to prevent the disease from progressing to the less manageable late phase (2,3). Pituitary adenylate cyclase–activating polypeptide (PACAP) belongs to the vasoactive intestinal polypeptide (VIP)/secretin/glucagon family (4). It is present in 27– and 38–amino acid–containing forms, the latter being predominant in most mammalian tissues. Its 3 G protein–coupled receptors (GPCRs) can be activated by both forms: PAC1 is specific for PACAP, whereas VPAC1 and VPAC2 are activated by both VIP and PACAP (5). In the last decade it was discovered that PACAP-27 also activates formyl peptide receptor–like 1 (FPRL1), a rhodopsin-like GPCR (6). Activation of the PAC1 receptor results in cAMP production, inositol trisphosphate turnover, increased intracellular calcium levels and phospholipase D activity. VPAC1/VPAC2 receptor activation induces primarily cAMP production (7).

AP-27 also activates formyl peptide receptor–like 1 (FPRL1), a rhodopsin-like GPCR (6). Activation of the PAC1 receptor results in cAMP production, inositol trisphosphate turnover, increased intracellular calcium levels and phospholipase D activity. VPAC1/VPAC2 receptor activation induces primarily cAMP production (7). PACAP is widely expressed in the nervous system (e.g., the brain, the superficial layer of the spinal dorsal horn, and capsaicin-sensitive sensory neurons) (7). It regulates nociceptive transmission in a complex manner: it is antihyperalgesic in several peripheral processes, but mainly pronociceptive centrally. Therefore, PACAP has been suggested to play a crucial role in central sensitization and the induction of chronic pain (7–9). We have provided evidence that peripherally administered PACAP-38 is antinociceptive in several models of acute pain, but induces sensitization of the knee joint primary afferents (10). It has been postulated that it activates VPAC1/VPAC2 receptors located on nociceptive nerve terminals within the articular capsule, thus increasing joint hypersensitivity similarly to VIP, which exerts its pronociceptive effect in osteoarthritis through the same pathway (11). PACAP has a potent vasodilatory effect through the PAC1 receptor and facilitates plasma leakage, edema formation, and leukocyte migration (12,13). We have also shown that PACAP has a crucial role in the long-term maintenance of neurogenic vasodilation on the periphery (14).

osteoarthritis through the same pathway (11). PACAP has a potent vasodilatory effect through the PAC1 receptor and facilitates plasma leakage, edema formation, and leukocyte migration (12,13). We have also shown that PACAP has a crucial role in the long-term maintenance of neurogenic vasodilation on the periphery (14). The role of PACAP in inflammation and immunoregulation has been studied extensively, and it has been suggested to be an important endogenous immunomodulator (7). Its receptors are widely distributed in the immune system: PAC1 is expressed on macrophages and monocytes, but not on lymphocytes (15). VPAC1 is present on both T and B lymphocytes, macrophages, and monocytes, whereas the VPAC2 receptor is expressed only on stimulated lymphocytes and macrophages (16). The FPRL1 receptor is expressed on phagocytic leukocytes, but to a lesser extent also on lymphocytes (17). PACAP-38 induces mast cell degranulation and histamine release, thereby contributing to edema formation (18). In human polymorphonuclear cells it also enhances respiratory burst, elastase, lactoferrin, and matrix metalloproteinase 9 release (19). Interestingly, it stimulates the activity of resting macrophages, but inhibits activated cells (16). PACAP-38 also increases the expression of several neutrophil activation markers, such as CD11b, CD63, and CD66b, suggesting that it has a role in neutrophil-mediated inflammatory pathways (20). Moreover, PACAP-27 specifically induces neutrophil chemotaxis through FPRL1 activation and induces phagocyte activation and Cd11b up-regulation (6). In a model of chronic autoimmune encephalomyelitis PACAP deficiency resulted in increased production of proinflammatory cytokines and chemokines, but decreased synthesis of antiinflammatory cytokines (21).

duces neutrophil chemotaxis through FPRL1 activation and induces phagocyte activation and Cd11b up-regulation (6). In a model of chronic autoimmune encephalomyelitis PACAP deficiency resulted in increased production of proinflammatory cytokines and chemokines, but decreased synthesis of antiinflammatory cytokines (21). These data demonstrate a surprisingly pleiotropic effect of PACAP on immune cells. This might be caused by changes in the receptor expression profile during inflammation, as inflammatory stimuli up-regulate several receptors (e.g., FPRL1 or VPAC1) and thereby alter the overall effect of PACAP (22).

duces neutrophil chemotaxis through FPRL1 activation and induces phagocyte activation and Cd11b up-regulation (6). In a model of chronic autoimmune encephalomyelitis PACAP deficiency resulted in increased production of proinflammatory cytokines and chemokines, but decreased synthesis of antiinflammatory cytokines (21). These data demonstrate a surprisingly pleiotropic effect of PACAP on immune cells. This might be caused by changes in the receptor expression profile during inflammation, as inflammatory stimuli up-regulate several receptors (e.g., FPRL1 or VPAC1) and thereby alter the overall effect of PACAP (22). A possible role of PACAP in bone metabolism has been suggested based on the presence of its receptors in the bone: osteoclasts express PAC1, and bone marrow cultures express VPAC1, VPAC2, and PAC1 (23). On desmal osteoblastic cell lines, mainly VPAC2 is expressed, and is even up-regulated during the differentiation process. However, little is known about the expression pattern in enchondral bone tissues. Both PACAP-38 and VIP block osteoblast differentiation by inhibiting alkaline phosphatase production and enhance bone resorption by stimulating interleukin-6 production, and thereby, osteoclast-activity (24,25). PACAP also inhibits thyroid hormone–stimulated osteocalcin synthesis in osteoblasts and decreases bone formation (26). It has considerably greater binding and activation at VPAC2 receptors on osteblastic cells than does VIP, emphasizing its importance in bone proliferation (27). It has recently been demonstrated that both PACAP-38 and VIP increase the RANKL:osteoprotegerin ratio through activation of VPAC2 on osteoblasts, which in turn promotes osteoclastogenesis (28). Despite the above-mentioned reports on the effects of PACAP on bone physiology, there is only one published study describing a chondroprotective effect of PACAP-38 in vitro; this effect was exerted mainly by increasing calcineurin levels (29).

ion of VPAC2 on osteoblasts, which in turn promotes osteoclastogenesis (28). Despite the above-mentioned reports on the effects of PACAP on bone physiology, there is only one published study describing a chondroprotective effect of PACAP-38 in vitro; this effect was exerted mainly by increasing calcineurin levels (29). In light of all of these divergent effects of PACAP in inflammatory, vascular, immune, and pain mechanisms as well as in bone turnover, we were interested in its potential role in arthritis. To investigate this we used a model of immune arthritis that has several well-established similarities to RA (30,31). Materials and Methods Animals and reagents Experiments were performed on 10–12-week-old PACAP gene–deficient mice on a CD1 background and their wild-type counterparts (PACAP+/+) (32). A total of 92 animals were studied.

In light of all of these divergent effects of PACAP in inflammatory, vascular, immune, and pain mechanisms as well as in bone turnover, we were interested in its potential role in arthritis. To investigate this we used a model of immune arthritis that has several well-established similarities to RA (30,31). Materials and Methods Animals and reagents Experiments were performed on 10–12-week-old PACAP gene–deficient mice on a CD1 background and their wild-type counterparts (PACAP+/+) (32). A total of 92 animals were studied. Because PACAP-27 and PACAP-38 are products of the same exon, the knockout animals lack both. Animals were bred in the Laboratory Animal House of the Department of Pharmacology and Pharmacotherapy at the University of Pécs, and were maintained in an ambient temperature of 24–25°C on a 12-hour light/dark cycle and provided with standard rodent chow and water ad libitum. As there were no differences in the parameters of interest between male and female animals, both sexes were used (except in the increasing temperature hot plate test, in which male mice cannot be used). The studies were approved by the Ethics Committee on Animal Research of the University of Pécs according to the Ethical Code of Animal Experiments (license no. BA 02/2000-2/2012) and complied with the recommendations of the International Association for the Study of Pain. All reagents were obtained from Sigma-Aldrich unless specified otherwise.

by the Ethics Committee on Animal Research of the University of Pécs according to the Ethical Code of Animal Experiments (license no. BA 02/2000-2/2012) and complied with the recommendations of the International Association for the Study of Pain. All reagents were obtained from Sigma-Aldrich unless specified otherwise. K/BxN serum–transfer arthritis model Sera obtained from transgene-positive (K/BxN) and -negative (BxN) mice were pooled separately and stored at −80°C as previously described (30,33). Arthritis was induced by intraperitoneal injection of arthritogenic (K/BxN) or control (BxN) serum on day 0. The amount administered was 150 μl for mice used in functional tests, as debilitating joint dysfunction produces gait abnormalities and limits functional nociception measurements. For in vivo imaging and histologic studies a single dose of 300 μl was used, to produce massive joint inflammation and cellular infiltration. For assessment of structural damage, 2 additional 150-μl booster injections were administered on days 10 and 20 to maintain longer-lasting inflammation and mimic the late, degenerative phase. Evaluation of disease severity and hind paw edema Arthritis severity, hyperemia, and paw volume were assessed daily by evaluation for edema and hyperemia (2 classic signs of inflammation) (33) and measurement of hind limb edema by plethysmometry (Ugo Basile) (34). The results were scored on a scale of 0–10 (0–0.5 = intact limb; 10 = most severe changes).

ind paw edema Arthritis severity, hyperemia, and paw volume were assessed daily by evaluation for edema and hyperemia (2 classic signs of inflammation) (33) and measurement of hind limb edema by plethysmometry (Ugo Basile) (34). The results were scored on a scale of 0–10 (0–0.5 = intact limb; 10 = most severe changes). In vivo fluorescence imaging of vascular leakage Indocyanine green (ICG) is a Food and Drug Administration–approved fluorescent cyanine near-infrared dye that, upon intravenous injection, rapidly binds to plasma proteins and remains in the healthy vasculature until excreted by the liver. It is suitable for use in imaging inflammatory hypervascularization and capillary leakage in arthritis both in preclinical models (35) and in patients with RA (36). To overcome its rapid clearance and stability problems when in aqueous solutions, non-ionic emulsifiers are used to encapsulate and stabilize ICG in micelles (37) and increase its plasma elimination half-life (38). Therefore, ICG (0.5 mg/kg) dissolved in a 5% (weight/volume) aqueous solution of Kolliphor HS 15, a macrogol-based surfactant (37), was injected intravenously into mice that had been anesthetized by intraperitoneal administration of 50 mg/kg sodium pentobarbital. Imaging (IVIS Lumina II optical imager; PerkinElmer) was performed 5, 10, 20, 30, and 60 minutes postinjection on days 0, 2, and 5, since vascular leakage is an early sign in this arthritis model. Imaging parameters were as follows: auto acquisition time (F-stop) 1, Binning 2, excitation 745 nm, emission filter >800 nm. Data were analyzed using Living Image software (PerkinElmer). Standardized regions of interest (ROIs) were drawn around the ankle joints. A calibrated unit of fluorescence, the radiant efficiency ([photons/second/cm2/steradian]/[μW/cm2]) originating from the ROIs was used for further analysis.

emission filter >800 nm. Data were analyzed using Living Image software (PerkinElmer). Standardized regions of interest (ROIs) were drawn around the ankle joints. A calibrated unit of fluorescence, the radiant efficiency ([photons/second/cm2/steradian]/[μW/cm2]) originating from the ROIs was used for further analysis. Measurements of mechano- and thermonociception The mechanical hyperalgesia of the hind paw was measured by dynamic plantar esthesiometry (Ugo Basile). Mechanonociceptive threshold was expressed in grams (34). The thermonociceptive threshold was determined using an increasing-temperature hot plate (IITC Life Sciences) heated increasingly from 30°C until the animal either exhibited nocifensive responses (lifting, shaking, or licking either paw) or the preset maximum (53°C) was attained (10). One conditioning and 2 control measurements were performed before arthritis induction. Assessment of motor performance and joint function To determine joint function and grasping ability, mice were placed on a horizontal wire grid, which was then turned over and maintained in this position for 20 seconds or until the animal fell. This is a simple but reliable method for functional analysis in this model (33). Potential impairment of locomotor coordination related to joint rigidity and consequent dysfunction were also studied using an accelerating rotarod (Ugo Basile) and expressed as the time spent on the wheel (14). Arthritis induction was preceded by 3 control measurements.

with the increased synovial hyperplasia in PACAP-deficient mice, suggests that PACAP might reduce FLS formation and exert a protective effect in chronic arthritis. This is also supported by the finding of pathologic bone neoformation in PACAP−/− mice, which can be at least partially attributed to FLS-derived mediators. The general limitations of experiments with knockout animals are that global gene deficiency can lead to potential phenotypic alterations during prenatal development and induce compensatory mechanisms. The possible effects of the flanking genes should also be taken into consideration. Discrepancies have been observed when comparing preclinical results obtained in knockout animals and clinical trial outcomes; for example, interferon-γ deficiency in mice was shown to induce arthritis, but clinical trials with interferon-γ resulted in only slight improvement in patients with RA (50).

thod for functional analysis in this model (33). Potential impairment of locomotor coordination related to joint rigidity and consequent dysfunction were also studied using an accelerating rotarod (Ugo Basile) and expressed as the time spent on the wheel (14). Arthritis induction was preceded by 3 control measurements. In vivo bioluminescence imaging of myeloperoxidase and NADPH oxidase activity Luminol (5-amino-2,3-dihydro-1,4-phthalazinedione) and lucigenin (bis-N-methylacridinium nitrate) are used to detect reactive oxygen species (ROS). Luminol bioluminescence detects mainly to neutrophil myeloperoxidase (MPO) activity, and lucigenin bioluminescence detects NADPH oxidase activity and extracellular superoxide production of macrophages (39,40). Na-luminol and lucigenin (Santa Cruz Biotechnology) were dissolved in phosphate buffered saline to form 20-mg/ml and 2.5-mg/ml stock solutions, respectively. Anesthetized mice were injected intraperitoneally with 150 mg/kg luminol (days 0, 1, 2, and 4) or 25 mg/kg lucigenin (days 0, 2, 6, and 10). Bioluminescence imaging was performed 10 minutes postinjection using an IVIS Lumina II. Acquisition time was 60 seconds, F-stop 1, Binning 8. Data were analyzed and ROIs were applied; luminescence was expressed as total radiance (photons/second/cm2/steradian) within the ROIs.

25 mg/kg lucigenin (days 0, 2, 6, and 10). Bioluminescence imaging was performed 10 minutes postinjection using an IVIS Lumina II. Acquisition time was 60 seconds, F-stop 1, Binning 8. Data were analyzed and ROIs were applied; luminescence was expressed as total radiance (photons/second/cm2/steradian) within the ROIs. In vivo investigation of metabolic activity by positron emission tomography/magnetic resonance imaging (PET/MRI) In vivo PET/MRI scans were obtained on day 4 using 2-18F-2-fluoro-2-deoxy-d-glucose (18F-FDG; 4 MBq per animal administered intravenously) and nanoScan PET/MRI (Mediso). The glucose analog 18F-FDG exhibits increased uptake kinetics in high-glucose–using cells of inflamed tissue. 18F-FDG accumulation adequately reflects inflammatory macrophage activity in acute arthritis, whereas it is less sensitive to neutrophils or T cells. In chronic arthritis, fibroblast proliferation and pannus formation are the main causes of radioisotope accumulation (41). MR images (gradient-echo–external averaging sequence) using a MultiCell imaging bed (Mediso) were overlaid on the PET scans. PET images were reconstructed with a Nucline Tera-Tomo PET algorithm (ordered-subsets expectation-maximization 3-dimensional reconstruction; Mediso) using 300-μm voxels. The standardized uptake values maximum (SUVmax) of 18F-FDG, i.e., the maximum values of the tissue radioactivity concentration divided by (injected activity/body mass) in the ROIs, were calculated in the ankle joints.

m (ordered-subsets expectation-maximization 3-dimensional reconstruction; Mediso) using 300-μm voxels. The standardized uptake values maximum (SUVmax) of 18F-FDG, i.e., the maximum values of the tissue radioactivity concentration divided by (injected activity/body mass) in the ROIs, were calculated in the ankle joints. In vivo micro–computed tomography (micro-CT) analysis of the tibiotarsal joint and bone structures Micro-CT imaging was performed on the same mice at every time point to minimize interindividual differences. The right ankles were repeatedly scanned (days 0, 7, 14, and 28) using the same settings and 17.5-μm voxel size, with a SkyScan 1176 micro-CT (Bruker). After reconstruction of the scans, bone structural changes were analyzed using CT Analyser software. Standardized ROIs were drawn around the periarticular region of the tibia and fibula, as well as the tibiotarsal and tarsometatarsal joints. In these ROIs bone volume (BV; μm3) and bone surface (BS; μm2) were calculated and expressed as the percentage of the standardized total volume (TV) of the ROI (% BV and BS density).

ndardized ROIs were drawn around the periarticular region of the tibia and fibula, as well as the tibiotarsal and tarsometatarsal joints. In these ROIs bone volume (BV; μm3) and bone surface (BS; μm2) were calculated and expressed as the percentage of the standardized total volume (TV) of the ROI (% BV and BS density). Histologic evaluation of joint inflammation Mice were killed on day 4 or day 28 to investigate acute and chronic alterations. They were transcardially perfused with 4% buffered paraformaldehyde, and ankle joints were fixed in the same buffer, decalcified and dehydrated, embedded in paraffin, sectioned (3–4 μm), and stained with Safranin O. The slides were evaluated semiquantitatively by a pathologist under blinded conditions. Synovial cell proliferation and mononuclear cell infiltration were scored from 0 (normal) to 3 (maximal severity) (34). Statistical analysis Results are expressed as the mean ± SEM. Statistical evaluation was performed using GraphPad Prism. Distribution of the data was tested by D'Agostino-Pearson or Shapiro-Wilk, test depending on the number of values. The significance of ICG imaging, clinical scoring, and plethysmometry results was evaluated by Kruskal-Wallis test, and functional results by repeated-measures analysis of variance (ANOVA). Wire grid performance was evaluated by log rank test, micro-CT results by two-way ANOVA, in vivo bioluminescence and PET imaging results by Student's t-test, and histopathologic scores by Mann-Whitney U test. P values less than 0.05 were considered significant.

results by repeated-measures analysis of variance (ANOVA). Wire grid performance was evaluated by log rank test, micro-CT results by two-way ANOVA, in vivo bioluminescence and PET imaging results by Student's t-test, and histopathologic scores by Mann-Whitney U test. P values less than 0.05 were considered significant. Results Reduced joint inflammation and edema in PACAP−/− mice After induction of arthritis in wild-type mice by administration of 150 μl arthritogenic serum, substantial hind paw edema developed, which peaked on day 3 at 40% and gradually decreased thereafter. In PACAP−/− animals, edema was present to a significantly lesser extent but the kinetics pattern was similar, with the maximum (20%) reached on day 5 (Figure 1A). Semiquantitative clinical scoring of edema and hyperemia showed similar results, but peak scores occurred on day 3−4 in wild-type mice and on day 5 in PACAP−/− mice. Arthritis severity scores were, however, significantly lower in the PACAP-deficient mice (Figure 1B).

with the maximum (20%) reached on day 5 (Figure 1A). Semiquantitative clinical scoring of edema and hyperemia showed similar results, but peak scores occurred on day 3−4 in wild-type mice and on day 5 in PACAP−/− mice. Arthritis severity scores were, however, significantly lower in the PACAP-deficient mice (Figure 1B). Figure 1 Edema, clinical severity, and inflammatory vascular leakage in PACAP+/+ (wild-type) and PACAP−/− mice in which arthritis was not induced (controls administered normal serum) or in which arthritis was induced by administration of K/BxN mouse serum. A and B, Plethysmometric determination of hind paw volumes (A) and clinical scores of disease severity (B) (n = 8−12 per group). C, Representative images of indocyanine green (ICG) fluorescence in the ankle joints, as an indicator of vascular permeability and leakage. Images were obtained 5 minutes and 60 minutes after intravenous injection of ICG (0.5 mg/kg). D, ICG fluorescence intensity ([photons/second/cm2/steradian]/[μW/cm2] [p/s/μW/cm2]) in the ankle joints 2 days and 5 days after induction of arthritis (n = 5−6 per group). Values in A, B, and D are the mean ± SEM. ∗ = P < 0.05; ∗∗ = P < 0.01; ∗∗∗ = P < 0.001 versus the respective control group. # = P < 0.05; ## = P < 0.01; ### = P < 0.001 versus the K/BxN serum−treated wild-type group. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.38772/abstract.

mean ± SEM. ∗ = P < 0.05; ∗∗ = P < 0.01; ∗∗∗ = P < 0.001 versus the respective control group. # = P < 0.05; ## = P < 0.01; ### = P < 0.001 versus the K/BxN serum−treated wild-type group. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.38772/abstract. Decreased inflammatory hyperemia and vascular leakage in PACAP−/− mice Before treatment, ICG fluorescence was similarly low in the ankle joints of PACAP+/+ and PACAP−/− mice, demonstrating negligible extravasation of the dye. Two days after induction of arthritis, the accumulation and fluorescence of ICG were notably increased in the ankle joints of wild-type mice both immediately after injection and 1 hour later, indicating hyperemia and plasma leakage (80% increase at 5 minutes, 170% at 1 hour postinjection compared to initial control data). In contrast, in PACAP-deficient animals the increase was significantly smaller (25% at 5 minutes, 65% after 1 hour). By day 5, ICG fluorescence increased even further in both wild-type mice (120% at 5 minutes, 320% after 1 hour) and PACAP−/− mice (60% and 130%, respectively) (Figures 1C and D).

o initial control data). In contrast, in PACAP-deficient animals the increase was significantly smaller (25% at 5 minutes, 65% after 1 hour). By day 5, ICG fluorescence increased even further in both wild-type mice (120% at 5 minutes, 320% after 1 hour) and PACAP−/− mice (60% and 130%, respectively) (Figures 1C and D). Diminished mechanical hyperalgesia in PACAP−/− mice There was no difference in mechano- or thermonociceptive thresholds between control PACAP+/+ and control PACAP−/− mice. In wild-type mice with arthritis, the mechanocociceptive threshold was decreased by 15–20% by day 5 after disease induction, but was normalized by day 9. Mechanical hyperalgesia did not develop in either arthritic PACAP−/− mice or normal serum–treated control PACAP−/− mice (Figure 2A). Arthritis did not result in a significant change in thermonociceptive thresholds in any of the groups (Figure 2B).

by 15–20% by day 5 after disease induction, but was normalized by day 9. Mechanical hyperalgesia did not develop in either arthritic PACAP−/− mice or normal serum–treated control PACAP−/− mice (Figure 2A). Arthritis did not result in a significant change in thermonociceptive thresholds in any of the groups (Figure 2B). Figure 2 Evaluation of nociceptive changes and motor impairment. A, Mechanonociceptive threshold in the hind paw (n = 8–12 per group). B, Thermonociceptive threshold measured using an increasing-temperature hot plate (n = 4–6 per group). C, Kaplan-Meier curve of the ability to hold onto a wire grid for 20 seconds (n = 8–12 per group). Animals were habituated in 3 control sessions prior to the induction of arthritis. D, Motor performance on an accelerating rotarod during 3 consecutive control measurements and following the induction of arthritis (n = 4–6 per group). Values in A, B, and D are the mean ± SEM. ∗∗ = P < 0.01; ∗∗∗ = P < 0.001 versus the respective control group. # = P < 0.05; ## = P < 0.01; ### = P < 0.001 versus the K/BxN serum–treated wild-type group.

ating rotarod during 3 consecutive control measurements and following the induction of arthritis (n = 4–6 per group). Values in A, B, and D are the mean ± SEM. ∗∗ = P < 0.01; ∗∗∗ = P < 0.001 versus the respective control group. # = P < 0.05; ## = P < 0.01; ### = P < 0.001 versus the K/BxN serum–treated wild-type group. Retained grasping ability of PACAP−/− mice The horizontal wire grid test revealed an abrupt decrease in grasping ability in wild-type mice; by day 4, only 30% of the mice could stay on the grid for 20 seconds. In contrast, 75–80% of the PACAP−/− animals could stay on the grid for this duration (Figure 2C). Motor performance on the rotarod gradually improved in all groups during the experimental period, demonstrating learning. The improvement was significant in the PACAP+/+ group but not in the PACAP-deficent group (Figure 2D). Decreased neutrophil MPO activity in the ankle joints of PACAP−/− mice in the early phase of arthritis and increased activity in the later phase Luminol bioluminescence imaging showing MPO activity in the inflamed joints of PACAP+/+ mice peaked in the initial phase of arthritis, reaching a maximum on day 1 and gradually decreasing thereafter. The initially diffuse activity in the hind paws rapidly declined and was concentrated in the tibiotarsal joints by day 4. In contrast, in PACAP−/− mice early MPO activity was significantly lower, but by day 4 it became significantly greater in the ankles (Figures 3A and B).

on day 1 and gradually decreasing thereafter. The initially diffuse activity in the hind paws rapidly declined and was concentrated in the tibiotarsal joints by day 4. In contrast, in PACAP−/− mice early MPO activity was significantly lower, but by day 4 it became significantly greater in the ankles (Figures 3A and B). Figure 3 Bioluminescence imaging of neutrophil-derived myeloperoxidase activity and macrophage-derived superoxide activity. A and B, Representative images of luminol activity obtained 10 minutes after intraperitoneal injection of Na-luminol (150 mg/kg) (A) and quantification of luminescence in diseased ankle joints (B). C and D, Representative images of lucigenin-based superoxide detection (C) and quantification of luminescence in diseased ankle joints 10 minutes after intraperitoneal injection of lucigenin (25 mg/kg) (D). Values in B and D are the mean ± SEM photons/second (p/s) (n = 4−6 per group). ∗ = P < 0.05 versus control PACAP+/+ mice; ## = P < 0.01. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.38772/abstract.

aperitoneal injection of lucigenin (25 mg/kg) (D). Values in B and D are the mean ± SEM photons/second (p/s) (n = 4−6 per group). ∗ = P < 0.05 versus control PACAP+/+ mice; ## = P < 0.01. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.38772/abstract. Lack of inflammatory superoxide production in PACAP−/− mice In PACAP+/+ mice, lucigenin bioluminescence, indicating the presence of extracellular superoxides, steadily increased, reaching a maximum on day 6. Its elevation occurred more slowly compared to that observed with luminol bioluminescence imaging, highlighting the differences between these markers. Superoxide generation in PACAP−/− mice remained similar to baseline and was significantly lower than that observed in wild-type mice (Figures 3C and D).

aximum on day 6. Its elevation occurred more slowly compared to that observed with luminol bioluminescence imaging, highlighting the differences between these markers. Superoxide generation in PACAP−/− mice remained similar to baseline and was significantly lower than that observed in wild-type mice (Figures 3C and D). Increased and accelerated periarticular osteophyte formation in PACAP−/− mice CT scanning revealed differences in the bone architecture of PACAP−/− mice even in the absence of arthritis. BV/TV in these mice was consistently increased in both the talocrural and distal periarticular regions of the tibia and fibula, although these increases were not significant. Bone surface density in these animals, expressed as BS/TV, was similar to that in wild-type mice. Arthritis did not substantially alter bone characteristics in PACAP+/+ mice. In contrast, in PACAP−/− mice it induced extensive, progressive osteophyte formation in the periarticular region of the tibia and fibula, which was apparent by day 14 (Figures 4A and B). These bone spurs had become compact, dense bone by day 28, leading to a prominent, significant increase in bone mass, even reaching 70% extra bone in some mice compared to controls (Figures 4C and D).

eophyte formation in the periarticular region of the tibia and fibula, which was apparent by day 14 (Figures 4A and B). These bone spurs had become compact, dense bone by day 28, leading to a prominent, significant increase in bone mass, even reaching 70% extra bone in some mice compared to controls (Figures 4C and D). Figure 4 Micro–computed tomography (micro-CT) analysis of structural changes in bone architecture. A and B, Representative 3-dimensional micro-CT reconstruction and histologic analysis of the tibiotarsal joints (A) and the periarticular region of the tibia and fibula (B). In the 4 images on the far right, bone spur formation (asterisks) can be seen in the 3-dimensional reconstructions (magnified views of the boxed areas), in an axial CT slice, and in a photomicrograph of the periarticular region of the tibia and fibula of a PACAP−/− mouse. Original magnification × 100. C and D, Changes in bone volume/total volume (BV/TV) and in bone surface (BS)/TV over time in the vicinity of the tibiotarsal joint (C) and the periarticular region of the tibia (D). Values are the mean ± SEM (n = 6 per group). ∗∗∗∗ = P < 0.0001 versus nonarthritic control PACAP−/− mice. # = P < 0.05; ## = P < 0.01; #### = P < 0.0001. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.38772/abstract.

rticular region of the tibia (D). Values are the mean ± SEM (n = 6 per group). ∗∗∗∗ = P < 0.0001 versus nonarthritic control PACAP−/− mice. # = P < 0.05; ## = P < 0.01; #### = P < 0.0001. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.38772/abstract. Decreased metabolic activity in the diseased ankle joints of PACAP−/− mice In the inflamed tibiotarsal joints of wild-type mice, metabolism was significantly increased on day 4, as shown by the elevated SUVmax values. In contrast, metabolism in the joints of arthritic PACAP-deficient animals did not differ from that observed in nonarthritic controls (Figures 5A and B). Figure 5 Positron emission tomography/magnetic resonance imaging (PET/MRI) of inflammatory metabolic burst. A, Representative 2-18F-2-fluoro-2-deoxy-d-glucose (18F-FDG; 4 MBq per animal) PET/MRI multimodal reconstruction images obtained 4 days after induction of arthritis. Areas with increased glucose uptake correspond to the intracapsular space containing tendons and joint capsule internal surface, as identified by both PET and MR signals. B, Quantitative evaluation of the standardized uptake value maximum (SUVmax) of 18F-FDG in the ankle joints. Values are the mean ± SEM (n = 3 per group). ∗∗ = P < 0.01 versus nonarthritic control PACAP+/+ mice. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.38772/abstract.

ion of the standardized uptake value maximum (SUVmax) of 18F-FDG in the ankle joints. Values are the mean ± SEM (n = 3 per group). ∗∗ = P < 0.01 versus nonarthritic control PACAP+/+ mice. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/doi/10.1002/art.38772/abstract. Increased synovial hyperplasia in PACAP−/− mice Histopathologic analysis of the joints revealed no difference between control serum–treated PACAP+/+ and PACAP−/− mice. In both groups the tibiotarsal joint, synovium, and periarticular connective tissue appeared normal (Figures 6A−C). Four days after arthritis induction there were prominent changes in the PACAP+/+ group, i.e., an irregular cartilage−bone border, enlarged synovium infiltrated with inflammatory cells, and massive infiltration of immune cells into the periarticular connective tissue with formation of mononuclear cell aggregates (Figure 6A–C). Similar changes were observed in the PACAP−/− mice, and the degree of synovial hyperplasia was greater than that in the wild-type mice whereas the degree of mononuclear cell infiltration was comparable (Figure 6D). By day 28 these acute inflammatory signs had decreased, but the cartilage−bone border became more irregular and the cartilage width was notably reduced in both groups (Figure 6A). The previously infiltrated synovial lining and connective tissue showed prominent collagen deposition, indicating the chronic stage of inflammation. The pathologic spurs in PACAP−/− mice demonstrated on the micro-CT scans were also identified by their irregularity, distinct staining, and notable vascularization seen on histologic analysis (Figure 6D). These features were absent in wild-type mice.

nt collagen deposition, indicating the chronic stage of inflammation. The pathologic spurs in PACAP−/− mice demonstrated on the micro-CT scans were also identified by their irregularity, distinct staining, and notable vascularization seen on histologic analysis (Figure 6D). These features were absent in wild-type mice. Figure 6 A–C, Representative photomicrographs of the ankle joints of control PACAP+/+ and PACAP−/− mice and PACAP+/+ and PACAP−/− mice after arthritis induction. Images of the tibiotarsal joints including cartilage lining (A), the synovium (B), and periarticular soft tissue (C) are shown. Blue staining in the day 28 images in B and C represents collagen deposition. Original magnification × 100 in A; × 200 in B and C. D, Left, Semiquantitative histopathologic scores of synovial hyperplasia and mononuclear cell infiltration on day 4. Values are the mean ± SEM (n = 4–6 per group). Right, Representative photomicrograph of a bone spur observed in a PACAP−/− mouse on day 28. Asterisk indicates the tibia. Arrows show irregular, cell-rich new bone (a) and a vessel entering the affected area (b).

ial hyperplasia and mononuclear cell infiltration on day 4. Values are the mean ± SEM (n = 4–6 per group). Right, Representative photomicrograph of a bone spur observed in a PACAP−/− mouse on day 28. Asterisk indicates the tibia. Arrows show irregular, cell-rich new bone (a) and a vessel entering the affected area (b). Discussion The primary and most important result of our study is the finding, for the first time, of evidence of a surprisingly pleiotropic effect of PACAP on different characteristics in a multifactorial transgenic disease model of RA. K/BxN mice display a spontaneous and progressive polyarthritis, with persistence of high-titer autoantibodies, mainly anti–glucose-6-phosphate isomerase, in the serum (30,31). This serum induces a similar, but transient, arthritis, mainly through inflammatory cytokine production, complement activation, neutrophil activation, and mast cell degranulation.

d progressive polyarthritis, with persistence of high-titer autoantibodies, mainly anti–glucose-6-phosphate isomerase, in the serum (30,31). This serum induces a similar, but transient, arthritis, mainly through inflammatory cytokine production, complement activation, neutrophil activation, and mast cell degranulation. In this model, which has translational significance, PACAP deficiency decreases vasodilation, plasma leakage, acute inflammatory cell accumulation, hyperalgesia, joint dysfunction, metabolic activity, and ROS generation, while facilitating later neutrophil activity, synovial cell proliferation, and pathologic bone formation. The second important message derived from our study, from a methodologic and translational standpoint, is that the K/BxN serum–transfer model is appropriate for investigating several early and late characteristics of RA using in vivo noninvasive imaging modalities. We adopted and modified in vivo functional and optical imaging techniques, as well as observational longitudinal experimental paradigms that help to identify key pathophysiologic mechanisms in inflammatory and degenerative joint diseases.

arly and late characteristics of RA using in vivo noninvasive imaging modalities. We adopted and modified in vivo functional and optical imaging techniques, as well as observational longitudinal experimental paradigms that help to identify key pathophysiologic mechanisms in inflammatory and degenerative joint diseases. The potent edema-forming and acute inflammatory actions of PACAP can be explained by its well-established vasoactive effects (12,42). Our in vivo plasma extravasation imaging results provide deeper insight into the core mechanisms responsible for this phenomenon and validate the functional data. Both VPAC and PAC1 receptors are likely to be involved in the vascular effects of PACAP: since nitric oxide might act as an agonist at the VPAC2 receptor, its level of synthesis can interact with VPAC-mediated pathways in vasodilation (43), and PAC1 receptor is of key importance in the induction of plasma leakage (13,44). The absence of arthritic hyperalgesia in PACAP−/− mice correlates well with previous results reported by our group and others, demonstrating that PACAP-38 has a pronociceptive role in peripheral pain conditions, in which sensitization mainly occurs in the spinal dorsal horn (8–10,14,45,46). This suggests the importance of PAC1 receptor activation in the central nervous system (CNS), since intrathecal administration of a PAC1 receptor antagonist had an antinociceptive effect in osteoarthritis (47).

n peripheral pain conditions, in which sensitization mainly occurs in the spinal dorsal horn (8–10,14,45,46). This suggests the importance of PAC1 receptor activation in the central nervous system (CNS), since intrathecal administration of a PAC1 receptor antagonist had an antinociceptive effect in osteoarthritis (47). The almost-normal grasping ability of arthritic PACAP−/− mice is likely due to milder joint swelling and pain. We did not find disease-related alterations in motor performance on the rotarod wheel in our experiments, similar to previously reported findings in a model of peripheral neuropathy (14). This can be explained by the fact that this test mainly reflects motor coordination related to CNS processes, and inflammation of the small joints in this model does not influence performance.

ance on the rotarod wheel in our experiments, similar to previously reported findings in a model of peripheral neuropathy (14). This can be explained by the fact that this test mainly reflects motor coordination related to CNS processes, and inflammation of the small joints in this model does not influence performance. We observed an interesting 2-phase pattern regarding the effect of PACAP deficiency on MPO activity during arthritis: in the early phase neutrophil activity is clearly and notably decreased in PACAP−/− mice, correlates well with our functional results but virtually contradicts the traditional view about the inhibitory effects of PACAP on immune cells (16). The explanations are likely to be 1) increased migration of immune cells into the inflamed tissue mainly due to the PACAP-38–driven capillary permeability increase through PAC1 receptor activation (12,18,44), and 2) the key importance of PACAP-27 in facilitating neutrophil chemotaxis and migration during the initial phase of the inflammation (6). Thus, we can conclude that these proinflammatory effects of PACAP lead to more rapid formation of the inflammatory microenvironment (19,20).

gh PAC1 receptor activation (12,18,44), and 2) the key importance of PACAP-27 in facilitating neutrophil chemotaxis and migration during the initial phase of the inflammation (6). Thus, we can conclude that these proinflammatory effects of PACAP lead to more rapid formation of the inflammatory microenvironment (19,20). In contrast, the increased luminol bioluminescence in the later phase suggests that PACAP inhibits neutrophil activity by suppressing proinflammatory cytokine production and increasing the expression of antiinflammatory mediators (16,21). This is consistent with earlier data showing that exogenous administration of PACAP reduces disease severity and inflammatory enzyme production in mice with collagen-induced arthritis (48). However, it is important to note that in the collagen-induced arthritis model there are profound T and B cell responses, whereas K/BxN serum–transfer arthritis depends mainly on myeloid lineages and develops similarly if T and B cells are absent (33). In our model the maximal inflammatory cell activity occurs on day 1 as shown by luminol bioluminescence, which is surprising since this is a very early phase of the disease, when functional changes are minimal. This suggests the importance of further studying this early period to elucidate the mechanisms that precede edema formation and functional loss.

ry cell activity occurs on day 1 as shown by luminol bioluminescence, which is surprising since this is a very early phase of the disease, when functional changes are minimal. This suggests the importance of further studying this early period to elucidate the mechanisms that precede edema formation and functional loss. The significantly lower level of extracellular ROS production in gene-deficient mice suggests that PACAP stimulates macrophage activity in arthritis and has distinct proinflammatory effects on phagocytes. The lack of increased inflammatory metabolic activity in the joints of PACAP−/− mice shown by 18F-FDG–PET can be attributed to decreased macrophage activity (41). The effects of PACAP on bone/cartilage metabolism and turnover have not previously been investigated in vivo under either normal or arthritic conditions. Our study provides substantial new evidence regarding its crucial role in bone pathophysiology. We demostrated that PACAP is a key regulator of bone turnover, particularly in inflammation. There are few, exclusively in vitro, data describing the ability of PACAP-38 to inhibit osteoblast precursors (24,25) and to diminish osteoclastogenesis (28). Taken together, the results of those studies and our present findings indicate that PACAP plays a crucial role in chondro- and osteogenesis by regulating bone formation and inhibiting pathologic osteophyte growth. Our finding of osteophyte formation in the late phase of disease indicates that this model has translational relevance regarding the degenerative complications of RA.

findings indicate that PACAP plays a crucial role in chondro- and osteogenesis by regulating bone formation and inhibiting pathologic osteophyte growth. Our finding of osteophyte formation in the late phase of disease indicates that this model has translational relevance regarding the degenerative complications of RA. Synovial hyperplasia is a characteristic and widely investigated feature of RA. It is due to the formation of disinhibited fibroblast-like synoviocytes (FLS) that express NF-κB and secrete several inflammatory cytokines and adhesion molecules (interleukin-6, CCL-2, vascular cell adhesion molecule 1, intercellular adhesion molecule 1). Consequently, there is an influx and accumulation of inflammatory cells, which produce cytokines, chemokines, and proteases (matrix metalloproteinases, cathepsins, etc.), leading to cartilage and bone destruction. It was recently demonstrated that in the K/BxN model, FLS develop and exhibit behavior similar to that in human RA (49). This, taken together with the increased synovial hyperplasia in PACAP-deficient mice, suggests that PACAP might reduce FLS formation and exert a protective effect in chronic arthritis. This is also supported by the finding of pathologic bone neoformation in PACAP−/− mice, which can be at least partially attributed to FLS-derived mediators.

nto consideration. Discrepancies have been observed when comparing preclinical results obtained in knockout animals and clinical trial outcomes; for example, interferon-γ deficiency in mice was shown to induce arthritis, but clinical trials with interferon-γ resulted in only slight improvement in patients with RA (50). In conclusion, PACAP deficiency has complex consequences with regard to several mechanisms related to RA. It decreases hyperemia, plasma leakage, and edema (most likely due to the lack of the potent vasodilating effect of PACAP) as well as functional impairment, and abolishes pain and sensitization. This in turn moderates the early migration of immune cells into the synovium and reduces metabolic activity. PACAP deficiency decreases early accumulation of neutrophils by slowing their extravasation from the vessels, but facilitates their function in the later phase. In addition, it decreases macrophage activity and ROS production and promotes inflammation-induced pathologic bone neoformation. We have provided experimental evidence of the important and complex regulatory function of PACAP. Identification of its targets and precise mechanisms might open future avenues for development of therapies aimed at both acute RA symptoms and chronic structural changes.

inflammation-induced pathologic bone neoformation. We have provided experimental evidence of the important and complex regulatory function of PACAP. Identification of its targets and precise mechanisms might open future avenues for development of therapies aimed at both acute RA symptoms and chronic structural changes. Author Contributions 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. Helyes 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. Botz, Németh, Szigeti, Horváth, Máthé, N. Kovács, Hashimoto, Reglődi, Szolcsányi, Pintér, Mócsai, Helyes. Acquisition of data. Botz, Bölcskei, Kereskai, M. Kovács, Szigeti, Horváth, Máthé, N. Kovács, Helyes. Analysis and interpretation of data. Botz, Bölcskei, Kereskai, Szigeti, Horváth, Máthé, N. Kovács, Hashimoto, Reglődi, Pintér, Helyes. Additional Disclosures Authors Máthé and N. Kovács are employees of CROmed Ltd. The authors are grateful to László Kollár for his valuable technical advice. We thank Anikó Perkecz and Nikolett Szentes for their expert technical assistance in the functional studies and histologic processing, Tamás Kiss for help in the micro-CT acquisitions and analysis, and Diane Mathis and Christophe Benoist for providing the KRN transgene-positive mice.

Death receptor 3 (DR-3; also known as TRAMP, lymphocyte-associated receptor of death, WSL-1, Apo-3, TR3, and tumor necrosis factor receptor superfamily member 25 [TNFRSF25]) is the closest relative to TNFR type I (TNFRI; TNFRSF1), one of the major ligands for TNFα, the archetypal “master regulator” of inflammation (1). Like TNFRI, DR-3 has an intracellular death domain that can recruit downstream effectors of apoptosis (2–7) but can also activate the transcription factor NF-κB, inducing immune activation and differentiation (8,9). It has a single TNFSF ligand, TNF-like protein 1A (TL1A; TNFSF15) (9,10), that is closely related in structure to TNFα (11).

an intracellular death domain that can recruit downstream effectors of apoptosis (2–7) but can also activate the transcription factor NF-κB, inducing immune activation and differentiation (8,9). It has a single TNFSF ligand, TNF-like protein 1A (TL1A; TNFSF15) (9,10), that is closely related in structure to TNFα (11). In recent years, the DR-3/TL1A axis has emerged as a key regulator of inflammation and autoimmunity in its own right, with in vivo studies of transgenic mice deficient for DR-3 or TL1A and those overexpressing TL1A or dominant-negative forms of DR-3 providing compelling evidence for an essential role of the DR-3/TL1A axis in many models of inflammatory and autoimmune disease (12–22). In contrast to TNFRI, much of the function of DR-3 has been attributed to its expression on T cells and natural killer T cells and its role in driving the accumulation or maintenance (23) of Teff cell numbers at sites of pathology, irrespective of their lineage. Consistent with this, DR-3 has also been shown to be essential for the development of efficient T cell immunity to certain bacterial and viral pathogens (24,25) and, in some cases, becoming essential for host survival (25). DR-3 expression is not restricted to lymphoid cells. In cells of myeloid lineage, in vitro DR-3 signaling can influence cytokine release (26) and myeloid cell differentiation (27), while in nonhematopoietic cells in vivo, DR-3 is expressed on neurones controlling neuronal innervation (28) or can be triggered on tubular epithelial cells to regulate responses to renal inflammation and injury (29,30).

in vitro DR-3 signaling can influence cytokine release (26) and myeloid cell differentiation (27), while in nonhematopoietic cells in vivo, DR-3 is expressed on neurones controlling neuronal innervation (28) or can be triggered on tubular epithelial cells to regulate responses to renal inflammation and injury (29,30). Rheumatoid arthritis (RA) is a chronic inflammatory disease characterized by immune cell infiltration into the joints, which eventually leads to destruction of cartilage and bone. Proinflammatory cytokines such as TNFα and interleukin-6 (IL-6) are critical for controlling the pathogenic process (31,32). A role for the DR-3/TL1A pathway has been proposed in RA, because the frequency of DR-3 gene duplication is higher in patients with RA compared with healthy individuals (33). In contrast, TL1A levels are increased in RA serum (34), synovial fluid, and synovial tissue, and the expression of TL1A can be induced by immune complex–stimulated monocytes in RA (35). This has been borne out in in vivo studies demonstrating that DR3−/− mice with experimental inflammatory arthritis are resistant to bone erosion, while treatment with antagonistic antibodies was protective in wild-type (WT) mice (21). Mechanistically, this effect has been attributed to the control of multiple late events in the arthritis disease process, from effector Th17 cell development (36) and differentiation of macrophages into osteoclasts (21) to the potential action of TL1A on osteoblasts (37).

ies was protective in wild-type (WT) mice (21). Mechanistically, this effect has been attributed to the control of multiple late events in the arthritis disease process, from effector Th17 cell development (36) and differentiation of macrophages into osteoclasts (21) to the potential action of TL1A on osteoblasts (37). In the current study, we investigated the in vivo role of the DR-3/TL1A pathway in early events in antigen-induced arthritis (AIA), uncovering previously overlooked functions of this proinflammatory pathway that have an impact on neutrophil recruitment and cartilage degradation. MATERIALS AND METHODS Animals DR3−/− mice and their age-matched DR3+/+ (WT) littermates (ages 6–12 weeks) were used in the experiments; these mice were derived from a mouse colony with heterozygous DR-3 expression that was founded from mice provided by Cancer Research UK (38). AIA was generated in male mice only. All procedures were approved by the local Research Ethics Committee and were performed in accordance with Home Office–approved licenses PPL 30/1999, 30/2361, and 30/2480. Generation of murine AIA AIA was generated as previously described (39). Briefly, mice were subcutaneously immunized with 1 mg/ml of methylated bovine serum albumin (mBSA) and Freund's complete adjuvant (CFA), in conjunction with an intraperitoneal injection of heat-inactivated Bordetella pertussis toxin. A booster immunization of BSA and CFA was administered 1 week later. Arthritis was induced in the hind right knee joint via an intraarticular injection of 10 mg/ml mBSA, given 21 days after the initial immunizations.

(CFA), in conjunction with an intraperitoneal injection of heat-inactivated Bordetella pertussis toxin. A booster immunization of BSA and CFA was administered 1 week later. Arthritis was induced in the hind right knee joint via an intraarticular injection of 10 mg/ml mBSA, given 21 days after the initial immunizations. Assessment of cartilage degradation The mice were killed on day 3 or day 21 after the induction of arthritis, for assessment of inflammatory and pathologic changes within the joint. For in vitro assays, whole murine patellae were incubated with neutrophil lysates for 3 days. All samples were then fixed in neutral buffered formalin and decalcified with formic acid (10%) for 2 weeks at 4°C, prior to embedding in paraffin. Serial sections (7 μm thick) were obtained, deparaffinized, and stained with Safranin O–fast green or toluidine blue, both of which are cationic stains that dye the acidic proteoglycan present in cartilage tissue red or purple. Total cartilage depth was then measured under 40× magnification using a line-graduated scale. The depth of cartilage depletion was determined by measuring to the “tideline” created by the absence of Safranin O–fast green or toluidine blue staining (Figure 2), and a percentage relative to the total cartilage depth was generated. Five points on the femoral head were measured to give an overall score for each joint.

e depth of cartilage depletion was determined by measuring to the “tideline” created by the absence of Safranin O–fast green or toluidine blue staining (Figure 2), and a percentage relative to the total cartilage depth was generated. Five points on the femoral head were measured to give an overall score for each joint. Immunohistochemical analysis Expression of the target ligand/receptor was detected using anti-rat, anti-rabbit, or anti-goat horseradish peroxidase (HRP)–diaminobenzidine (DAB) staining kits (R&D Systems), depending on the primary antibody, and according to the manufacturer's instructions. Briefly, sections were rehydrated, and endogenous peroxidase activity was blocked. Antigen unmasking was achieved by incubating the sections in prewarmed trypsin–EDTA (0.1%) in phosphate buffered saline (PBS) for 30 minutes at 37°C. Following the blocking steps, the sections were incubated overnight with 4 μg/ml of rat anti–Ly-6G (Invitrogen), goat anti–matrix metalloproteinase 9 (anti–MMP-9; Santa Cruz Biotechnology), rabbit anti-CXCL1 (Clontech), goat biotinylated anti–DR-3 (R&D Systems), or isotype controls diluted in PBS followed by biotinylated secondary antibody, according to the manufacturers' instructions. Sections were counterstained with hematoxylin, dehydrated, and mounted in DPX. Positive staining was visualized using a streptavidin–HRP conjugate and DAB chromogen that stained positive areas brown. Images were captured using a digital camera (Olympus N457 or Canon EOS 100D), and the proportion of brown pixels within a particular area was measured using Adobe Photoshop CS3.5. Five randomly selected areas were used to generate scores for each sample.

RP conjugate and DAB chromogen that stained positive areas brown. Images were captured using a digital camera (Olympus N457 or Canon EOS 100D), and the proportion of brown pixels within a particular area was measured using Adobe Photoshop CS3.5. Five randomly selected areas were used to generate scores for each sample. In vitro cell culture Human monocytes were obtained from peripheral blood using density-gradient centrifugation to purify mononuclear cells, followed by isolation with anti-CD14 microbeads according to the manufacturer's instructions (Miltenyi Biotec). Macrophages were then generated by 7-day culture in RPMI 1640 supplemented with 10% heat-inactivated fetal calf serum and macrophage colony-stimulating factor (20 ng/ml; R&D Systems). Human neutrophils, skin, and synovial fibroblasts were isolated as previously described (40–42). Ethics approval for all human experiments was obtained from the Bro Taf Health Authority (Cardiff, Wales, UK) prior to commencement of the study. Murine bone marrow–derived macrophages were generated from bone marrow extracted from the femurs of DR3−/− and WT mice, as previously described (21). Cells were cultured with or without recombinant TL1A or murine soluble DR-3 (R&D Systems) at the concentrations indicated, sometimes with additional stimuli such as interferon-γ (IFNγ) (26), lipopolysaccharide, or antigen/antibody complexes (35). Supernatants were collected over a 24-hour period, and the concentrations of enzymes, chemokines, or cytokines were measured as indicated.

luble DR-3 (R&D Systems) at the concentrations indicated, sometimes with additional stimuli such as interferon-γ (IFNγ) (26), lipopolysaccharide, or antigen/antibody complexes (35). Supernatants were collected over a 24-hour period, and the concentrations of enzymes, chemokines, or cytokines were measured as indicated. Enzyme-linked immunosorbent assays (ELISAs) ELISAs for murine CXCL1 and human MMP-9 were performed according to the instructions of the manufacturer (R&D Systems). Statistical analysis Cartilage degradation and staining readouts were percentages; therefore, Mann-Whitney nonparametric U tests were used for statistical analysis. Student's t-tests were used in the analyses of protein concentrations determined by ELISAs. Analyses were performed using GraphPad Prism version 4. P values less than or equal to 0.05 were considered significant, and P values less than or equal to 0.01 were considered highly significant.

used for statistical analysis. Student's t-tests were used in the analyses of protein concentrations determined by ELISAs. Analyses were performed using GraphPad Prism version 4. P values less than or equal to 0.05 were considered significant, and P values less than or equal to 0.01 were considered highly significant. RESULTS DR-3 expression in inflamed joints Although the DR-3/TL1A pathway has been shown to be essential in the development of inflammatory arthritis in mice, and that antagonism of this pathway can ameliorate disease (21), relatively little is known about the expression patterns of members of this pathway in the joint. Here, we investigated DR-3 expression early in the inflammatory process by staining joint sections with a polyclonal antibody. As expected, synovial membrane sections from the joints of WT mice showed minimal isotype staining, and synovial membrane sections from the joints of DR3−/− mice showed minimal anti–DR-3 staining (Figure 1A) (mean ± SEM 5.5 ± 1.0% and 2.8 ± 0.8%, respectively). In contrast, strong signals for DR-3 were recorded in synovial membrane sections (20.6 ± 3.5%) and anterior fat pad sections (20.2 ± 1.7%) from the joints of WT mice, 3 days after the generation of AIA (13,21). The DR-3 signal was low or absent in sections obtained from both of these areas in lateral control knees, in which mBSA had not been injected (Figure 1A). These data showed that DR-3 is primarily absent in healthy joints (some low-level expression may be present in the synovial membrane) but is significantly and rapidly increased by the inflammatory process induced by injection of mBSA (Figure 1B).

n lateral control knees, in which mBSA had not been injected (Figure 1A). These data showed that DR-3 is primarily absent in healthy joints (some low-level expression may be present in the synovial membrane) but is significantly and rapidly increased by the inflammatory process induced by injection of mBSA (Figure 1B). Figure 1 Death receptor 3 (DR-3) expression in the joints of mice with antigen-induced arthritis. Arthritis was induced in DR3-knockout (DR3-KO; DR3−/−) mice and their DR3+/+ (wild-type [WT]) littermates, and the joints were prepared, sectioned, and stained for DR-3 as described in Materials and Methods. Antigen (methylated bovine serum albumin [mBSA]) was administered into the right knee to induce localized inflammatory arthritis. A, Representative high-magnification (40×) photomicrographs showing (from top to bottom) isotype staining in a right knee section from a WT mouse, anti– DR-3 staining in a right knee section from a DR3−/− mouse, anti–DR-3 staining in a right knee section from a WT mouse, and anti–DR-3 staining in a left knee section (contralateral negative control) from a WT mouse. Bars = 45 μm. B, Quantification of positive staining, as measured by the percentage of positive pixels within a particular area. Values are the mean ± SEM (n = 4–5 mice per group). P values were determined by Mann-Whitney U test.

nti–DR-3 staining in a left knee section (contralateral negative control) from a WT mouse. Bars = 45 μm. B, Quantification of positive staining, as measured by the percentage of positive pixels within a particular area. Values are the mean ± SEM (n = 4–5 mice per group). P values were determined by Mann-Whitney U test. Protection against early cartilage degradation in the joints of DR3−/− mice To examine the functional significance of this increase in DR-3 expression, we investigated cartilage degradation at both early (day 3) and late (day 21) time points following generation of AIA. Consistent with a previous report (21), DR3−/− mice showed significant protection against cartilage destruction compared with their WT littermates on day 21 (mean ± SEM 11 ± 7% versus 50 ± 6%; P = 0.006), as measured by proteoglycan staining with Safranin O–fast green (Figures 2A and B). Unexpectedly, this pattern was also observed early in the inflammatory process on day 3 after generation of AIA (17 ± 5% in WT mice and 2 ± 1% in DR3−/− mice; P = 0.03), as measured by staining with Safranin O–fast green or toluidine blue (Figures 2C and D). Thus, the DR-3/TL1A pathway contributes to the development of early pathologic features of inflammatory arthritis prior to exerting an effect on Teff cell development and osteoclastogenesis in murine models of inflammatory arthritis (21,36).

d by staining with Safranin O–fast green or toluidine blue (Figures 2C and D). Thus, the DR-3/TL1A pathway contributes to the development of early pathologic features of inflammatory arthritis prior to exerting an effect on Teff cell development and osteoclastogenesis in murine models of inflammatory arthritis (21,36). Figure 2 Cartilage depletion in the joints of mice with antigen-induced arthritis (AIA). Arthritis was induced in WT and DR3−/− mice, and the joints were prepared, sectioned, and stained as described in Materials and Methods. A, Representative Safranin O–fast green–stained joint sections from WT and DR3−/− mice, 21 days after generation of AIA. B, Quantification of cartilage depletion in WT and DR3−/− mice on day 21. C, Representative Safranin O–fast green–stained (top row) and toludine blue–stained (bottom row) joint sections from WT and DR3−/− mice, 3 days after generation of AIA. D, Quantification of cartilage depletion in WT and DR3−/− mice on day 3, as measured by Safranin O–fast green staining. In A and C, bars = 60 μm. In B and D, each data point represents a single mouse; horizontal lines show the mean. P values were determined by Mann-Whitney U test. See Figure 1 for other definitions.

A. D, Quantification of cartilage depletion in WT and DR3−/− mice on day 3, as measured by Safranin O–fast green staining. In A and C, bars = 60 μm. In B and D, each data point represents a single mouse; horizontal lines show the mean. P values were determined by Mann-Whitney U test. See Figure 1 for other definitions. Concentrations of MMP-9 and ADAMTS-5 in the joints of DR3−/− mice early in the course of AIA In an attempt to determine how DR-3 so rapidly contributes to joint degradation, we investigated the level of cartilage-destroying enzymes within the joints of DR3−/− and WT mice in early AIA. MMP-9, a gelatinase that degrades type IV and type V collagen and has an established role in cartilage degradation during RA (43), was chosen because of previous reports that MMP-9 release could be induced from the myeloid cell line THP-1 in vitro by either crosslinking of DR-3 (44) or the action of IFNγ and TL1A (26). Consistent with the observed reductions in cartilage depletion, MMP-9 levels were significantly reduced in the joints of DR3−/− mice with AIA (mean ± SEM 2.3 ± 0.4%) compared with the levels in WT mice (4.4 ± 0.7%; P = 0.03) 3 days after the initiation of AIA (Figures 3A and C). This was primarily attributable to the presence of MMP-9 within infiltrating cells in the fat pad (Figures 3A and C), but MMP-9 was also detected in chondrocytes from the joints of WT and DR3−/− mice (additional information is available from the corresponding author). In contrast, the levels of ADAMTS-5, the major aggrecanase in mouse cartilage (45), in the joints of DR3−/− mice were not different from the levels in WT mice (additional information is available from the corresponding author). Therefore, at this time point (day 3), levels of MMP-9, but not ADAMTS-5, were dependent on the presence of DR-3.

ls of ADAMTS-5, the major aggrecanase in mouse cartilage (45), in the joints of DR3−/− mice were not different from the levels in WT mice (additional information is available from the corresponding author). Therefore, at this time point (day 3), levels of MMP-9, but not ADAMTS-5, were dependent on the presence of DR-3. Figure 3 Matrix metalloproteinase 9 (MMP-9) expression in the joints of mice with antigen-induced arthritis and MMP-9 production in vitro. Arthritis was induced in WT mice and DR3−/− mice, and the joints were prepared, sectioned, and stained as described in Materials and Methods. A and B, Representative joint sections from a WT mouse (A) and a DR3−/− mouse (B, 3 days after induction of arthritis, stained for MMP-9. Arrowheads highlight areas of positive brown staining. Bars = 60 μm. C, Quantification of MMP-9 expression in WT and DR3−/− mice. D, MMP-9 production in cultures of rheumatoid arthritis fibroblast-like synoviocytes (RA FLS), healthy skin fibroblasts (fibs), human fetal foreskin fibroblasts (HFFF), macrophages, and neutrophils treated with the indicated stimuli. In C and D, each symbol represents a single mouse (C) or a single subject (D); horizontal lines show the mean. P values were determined by Mann-Whitney U test (C) and Student's t-test (D). TL1A = tumor necrosis factor–like molecule 1A; IFNγ = interferon-γ (see Figure 1 for other definitions).

ith the indicated stimuli. In C and D, each symbol represents a single mouse (C) or a single subject (D); horizontal lines show the mean. P values were determined by Mann-Whitney U test (C) and Student's t-test (D). TL1A = tumor necrosis factor–like molecule 1A; IFNγ = interferon-γ (see Figure 1 for other definitions). Neutrophils as a major source of MMP-9 In order to determine the potential source of DR-3–dependent MMP-9, cell lines representing stromal and infiltrating cell types within the inflamed joint were established and tested for MMP-9 production in response to TL1A. These included fibroblasts derived from multiple sources (RA synovium, healthy skin, or fetal foreskin), macrophages, and neutrophils. As expected, fibroblasts produced only small amounts of MMP-9 (on a per-cell basis), with skin fibroblasts and fetal foreskin fibroblasts producing significantly more than RA synovial fibroblasts (mean ± SEM 0.15 ± 0.01, 0.15 ± 0.02, and 0.05 ± 0.01 ng/hour/million cells, respectively). In contrast, primary macrophages produced ∼20 times more MMP-9 (3.1 ± 0.9 ng/hour/million cells) than fibroblasts, and neutrophils generated ∼800 times more MMP-9 (126 ± 31 ng/hour/million cells) than fibroblasts (Figure 3D).

ts (mean ± SEM 0.15 ± 0.01, 0.15 ± 0.02, and 0.05 ± 0.01 ng/hour/million cells, respectively). In contrast, primary macrophages produced ∼20 times more MMP-9 (3.1 ± 0.9 ng/hour/million cells) than fibroblasts, and neutrophils generated ∼800 times more MMP-9 (126 ± 31 ng/hour/million cells) than fibroblasts (Figure 3D). Although this production was significant, neutrophils contained even larger (17-fold) intracellular stores of MMP-9, as shown by testing lysed cultures by ELISA (Figure 4A). Such lysates were also highly capable of degrading articular cartilage in vitro (Figure 4B). However, although DR-3 was observed on the surface of neutrophils (Figure 4C), and general activation using fMLP significantly increased the production of MMP-9 by neutrophils, TL1A did not stimulate release of MMP-9 (Figure 4D). In addition, TL1A and fMLP activation had no significant effect on the release of the neutrophil collagenase MMP-8, which was observed at concentrations ∼50-fold less than those of MMP-9 in neutrophil culture supernatants (Figure 4D). Indeed, TL1A did not increase MMP-9 release from any of the cultured cells examined (Figure 3D). Thus, although neutrophils were the likeliest source of cartilage-depleting MMP-9 in the joints of mice with AIA, TL1A does not appear to elevate levels of MMP-9 by directly inducing production.

re supernatants (Figure 4D). Indeed, TL1A did not increase MMP-9 release from any of the cultured cells examined (Figure 3D). Thus, although neutrophils were the likeliest source of cartilage-depleting MMP-9 in the joints of mice with AIA, TL1A does not appear to elevate levels of MMP-9 by directly inducing production. Figure 4 Matrix metalloproteinase 9 (MMP-9) production by neutrophils. A, Concentration (conc) of MMP-9 in neutrophil culture supernatants (S/N) and lysates. Values are the mean ± SEM. B, Left, Representative whole murine patellae sections incubated with control media or neutrophil lysates. Arrows indicate the tidemark used to determine cartilage degradation. Bars = 60 μm. Right, Percentage of cartilage degradation. C, Histogram showing death receptor 3 (DR-3) expression on neutrophils, as determined by flow cytometric analysis. D, MMP-8 and MMP-9 production by neutrophils following in vitro activation. In B and D, each data point represents a single culture; horizontal lines show the mean. P values were determined by Mann-Whitney U test (B) or Student's t-test (D). TL1A = tumor necrosis factor–like molecule 1A; APC = allophycocyanin.

nalysis. D, MMP-8 and MMP-9 production by neutrophils following in vitro activation. In B and D, each data point represents a single culture; horizontal lines show the mean. P values were determined by Mann-Whitney U test (B) or Student's t-test (D). TL1A = tumor necrosis factor–like molecule 1A; APC = allophycocyanin. Impaired neutrophil infiltration into the joints of DR3−/− mice early in the course of AIA We hypothesized that DR-3 could control early pathologic changes in the joint by increasing the number of infiltrating innate effector cells bearing MMP-9, which would include macrophages and neutrophils. Previous studies have indicated that there is no difference between the level of macrophages in the joints of DR3−/− and WT mice with AIA, early or late in the disease course, as measured by F4/80 staining (21). The predominant cell type involved in early infiltration into the joints of mice with AIA are neutrophils, which can be observed as soon as 6 hours after antigen injection (46). Thus, we stained the joints of mice with AIA that were killed on day 3 for the neutrophil marker Ly-6G. The joints of DR3−/− mice showed significantly less Ly-6G staining compared with their WT counterparts (mean ± SEM 1.3 ± 0.5% versus 5.3 ± 0.9%; P = 0.001) (Figures 5A and B), and this was primarily associated with cellular infiltration into the fat pad (Figures 5C and D). Thus, the accumulation or maintenance of neutrophil numbers in the joint early after the generation of AIA was dependent on DR-3 expression.

terparts (mean ± SEM 1.3 ± 0.5% versus 5.3 ± 0.9%; P = 0.001) (Figures 5A and B), and this was primarily associated with cellular infiltration into the fat pad (Figures 5C and D). Thus, the accumulation or maintenance of neutrophil numbers in the joint early after the generation of AIA was dependent on DR-3 expression. Figure 5 Expression of the neutrophil marker Ly-6G in the joints of mice with antigen-induced arthritis. Arthritis was induced in WT and DR3−/− mice, and the joints were prepared, sectioned, and stained as described in Materials and Methods. A, Representative low-magnification photomicrographs of joint sections from WT and DR3−/− mice stained for Ly-6G, 3 days after induction of arthritis. Arrowheads highlight staining in the synovial membrane (red) or fat pad (yellow). B, Quantification of Ly-6G expression in the joints of WT and DR3−/− mice. C, Representative high-magnification photomicrographs of fat pad sections from the joints of WT and DR3−/− mice. Arrowheads highlight staining of infiltrating cells. D, Quantification of Ly-6G expression in fat pad and synovial membrane sections obtained from the right knees of WT and DR3−/− mice. Bars in A and C = 60 μm. In B and D, each data point represents a single mouse; horizontal lines show the mean. P values were determined by Mann-Whitney U test. See Figure 1 for definitions.

Quantification of Ly-6G expression in fat pad and synovial membrane sections obtained from the right knees of WT and DR3−/− mice. Bars in A and C = 60 μm. In B and D, each data point represents a single mouse; horizontal lines show the mean. P values were determined by Mann-Whitney U test. See Figure 1 for definitions. Reduced expression of the neutrophil chemoattractant CXCL1 in the joints of DR3−/− mice Several potential mechanisms could explain the reduced expression of CXCL1 in the joints of DR3−/− mice. The most obvious, considering DR-3 contains a death domain, is an alteration in neutrophil survival. However, in vitro experiments indicated that TL1A had no significant effect on neutrophil death, as measured by staining with fluorescein isothiocyanate–labeled annexin V/7-aminoactinomycin D and flow cytometric evaluation, with or without activating stimuli (additional information is available from the corresponding author). Another possible explanation is that DR-3 controlled neutrophil recruitment. A number of chemokines have been reported to attract neutrophils, but the release of human IL-8 from the macrophage-like cell line THP-1 has previously been shown to be triggered in response to TL1A following IFNγ priming (47). We therefore stained the joints of mice with AIA for the murine ortholog of IL-8, CXCL1 (also known as murine keratinocyte-derived chemokine). DR3−/− mouse joints showed significantly less staining for CXCL1 than joints from WT mice (mean ± SEM 11 ± 2% and 26 ± 4%, respectively; P = 0.006) (Figures 6A and C). These data are consistent with the hypothesis that a reduction in production of neutrophil attractants such as CXCL1, rather than any affect on survival or lifespan, causes the decrease in neutrophil infiltration in DR3−/− mice early in the inflammatory process of AIA.

± 4%, respectively; P = 0.006) (Figures 6A and C). These data are consistent with the hypothesis that a reduction in production of neutrophil attractants such as CXCL1, rather than any affect on survival or lifespan, causes the decrease in neutrophil infiltration in DR3−/− mice early in the inflammatory process of AIA. Figure 6 CXCL1 expression in the joints of mice with antigen-induced arthritis. Arthritis was induced in WT and DR3−/− mice, and the joints were prepared, sectioned, and stained as described in Materials and Methods. A and B, Representative joint sections from a WT mouse (A) and a DR3−/− mouse (B), 3 days after induction of arthritis, stained for CXCL1. Bars = 60 μm. C, Quantification of CXCL1 expression in WT and DR3−/− mice. Each data point represents a single mouse; horizontal lines show the mean. P values were determined by Mann-Whitney U test. See Figure 1 for definitions. DISCUSSION The DR-3/TL1A pathway has recently emerged as a potential therapeutic target in inflammatory arthritis, the antagonism of which could impair the mechanisms that are controlled by this pathway. These include the development of effector CD4+ Th17 cells (36), macrophage differentiation into osteoclasts (21), and osteoblast function (37), all of which influence late events in the inflammatory arthritis disease process through an impact on bone turnover. Here, we show that DR-3 also controls early stages of the pathogenic process by regulating the initial damage to cartilage that occurs prior to the events described above.

, and osteoblast function (37), all of which influence late events in the inflammatory arthritis disease process through an impact on bone turnover. Here, we show that DR-3 also controls early stages of the pathogenic process by regulating the initial damage to cartilage that occurs prior to the events described above. To our knowledge, DR-3 expression patterns in the joint have not been previously described, and only one study has shown the presence of its TNFSF ligand TL1A in the joints of patients with RA (35). In the current study, we show that DR-3 is present only at low levels in unchallenged contralateral joints but is up-regulated following injection of arthritis-inducing antigen (Figure 1). The simplest interpretation of these observations is that local antigen-driven signals induce up-regulation of DR-3; however, a degree of caution is required. The strongest DR-3 signals came from the areas just below the synovial membrane and from within the fat pad, but these signals localized to the same areas that stained infiltrating neutrophils using Ly-6G (Figure 4). The fact that DR-3 was detected on the surface of human neutrophils (Figure 4) and has also been observed on macrophage-like cell lines and primary macrophages (26), means the extent to which increasing DR-3 signals can be attributed to induction of expression on stromal cells versus its surface expression on infiltrating cells cannot yet be judged. Interestingly, a more general diffuse signal throughout the joints of WT mice with AIA was also observed (Figure 1) and would be consistent with the presence of soluble DR-3.

3 signals can be attributed to induction of expression on stromal cells versus its surface expression on infiltrating cells cannot yet be judged. Interestingly, a more general diffuse signal throughout the joints of WT mice with AIA was also observed (Figure 1) and would be consistent with the presence of soluble DR-3. At least 3 murine splice variants have been described, including a soluble form lacking a transmembrane region (48), the expression of which is differentially regulated by activation (7,25). The function of these different splice variants is still poorly understood, but soluble DR-3 should buffer the action of TL1A. In mice, this may be particularly significant, because there is no known murine homolog for human decoy receptor 3 (DcR-3), which is described as an additional soluble decoy ligand for 3 TNFSF members (TL1A, FasL, and LIGHT) (49), and its level has also been shown to be increased in the serum of patients with RA (34). Human DcR-3 also binds murine TL1A, FasL, and LIGHT (49), and it is interesting that its systemic application in a murine model of collagen-induced arthritis (CIA) resulted in amelioration of disease associated with inhibition of effector CD4+ T cells and B cells (50). This is consistent with studies by our group and other investigators demonstrating the role of DR-3 in AIA and CIA (21,36), as is the ability of DcR-3 to inhibit osteoclastogenesis in vitro (51), but neither study determined the dominant pathway(s) through which DcR-3–dependent inhibition occurred. These differences between species and the complexity of TNFSF and TNFRSF interactions are clearly areas that should be studied further in inflammatory diseases such as RA.

to inhibit osteoclastogenesis in vitro (51), but neither study determined the dominant pathway(s) through which DcR-3–dependent inhibition occurred. These differences between species and the complexity of TNFSF and TNFRSF interactions are clearly areas that should be studied further in inflammatory diseases such as RA. Although many MMPs, including MMP-1, MMP-2, MMP-3, MMP-9, and MMP-13, have been associated with the destruction of cartilage, tendon, and bone in RA (43), the current study focused on MMP-9 because of several previous in vitro studies demonstrating its TL1A-driven release from macrophage-like cell lines (26,44). MMP-9 is primarily a gelatinase but also targets type IV collagen and is believed to further degrade extracellular matrix after the action of type I and II collagenases such as MMP-1 and MMP-13. Increasing levels of serum MMP-9 correlate with RA severity (52,53), while MMP-9–deficient mice show resistance to antibody-induced arthritis (54). Our data showed for the first time that the absence of DR-3 is associated with a significant decrease in MMP-9 expression at very early stages in the development of AIA. Interestingly, the absence of DR-3 was not associated with a change in the levels of ADAMTS-5, an aggrecanase responsible for cartilage degradation in osteoarthritis (55), suggesting that DR-3 signaling differentially regulates the levels of some (e.g., MMP-9) but not other cartilage-destroying enzymes at this early time point in the AIA process. Further study at later time points, when more effector Th17 cells would be present in the joint, would be required to determine whether the reported synergistic induction of ADAMTS-5 from macrophages by TL1A and IL-17 occurs in AIA (56).

ot other cartilage-destroying enzymes at this early time point in the AIA process. Further study at later time points, when more effector Th17 cells would be present in the joint, would be required to determine whether the reported synergistic induction of ADAMTS-5 from macrophages by TL1A and IL-17 occurs in AIA (56). We also discovered that neutrophils produced ∼40 times more MMP-9 in culture on a per-cell basis than macrophages, with an additional capacity to produce ≥600 times more MMP-9 due to high intracellular stores (Figures 3 and 4). Neutrophils also produce MMP-8, although this collagenase was generated at ∼50-fold lower concentrations than MMP-9 in our in vitro cultures and was not significantly induced by activation or exogenous TL1A (Figure 4). This does not preclude a role for MMP-8 in cartilage destruction in inflammatory arthritides, but suggests that there may be a hierarchy of MMP production from neutrophils, several of which could contribute to cartilage destruction.

cultures and was not significantly induced by activation or exogenous TL1A (Figure 4). This does not preclude a role for MMP-8 in cartilage destruction in inflammatory arthritides, but suggests that there may be a hierarchy of MMP production from neutrophils, several of which could contribute to cartilage destruction. These data are consistent with the results of several studies showing that neutrophils are a primary source of MMP-9 in diseases requiring breakdown of tissue, such as coronary heart disease (57) or stroke (58), although their potential to contribute significantly to MMP-9 levels in the inflamed joint has not previously been described. Instead, it has been suggested that macrophages are the primary source of MMP-9 in RA (59,60). The potential role of neutrophils in the early pathogenesis of RA seems to have mostly been ignored, probably because patients often present with later-stage disease, when joint damage has already occurred and swelling has resolved. Historically, however, it has been estimated that the turnover of neutrophils is extremely high in inflamed joints (61), with the main role for neutrophils in models of inflammatory arthritis being attributed to the supply of proinflammatory leukotrienes (62,63).

t damage has already occurred and swelling has resolved. Historically, however, it has been estimated that the turnover of neutrophils is extremely high in inflamed joints (61), with the main role for neutrophils in models of inflammatory arthritis being attributed to the supply of proinflammatory leukotrienes (62,63). Intriguingly, we failed to reproduce the previous in vitro findings of TL1A-driven MMP-9 release, although this may in part have been attributable to our use of primary cells, which may require additional signals for priming. Kang and colleagues demonstrated these effects using THP-1 cells, which also required priming with interferon-γ (26). Instead, the role of the DR-3/TL1A pathway at this early stage in the development of inflammatory arthritis in murine AIA seems to be the production of chemokines that attract neutrophils into the inflamed joint. In humans, IL-8 (CXCL8) is considered to be the primary neutrophil chemoattractant and has itself been reported to induce MMP-9 release (64,65). Mice, however, do not have a CXCL8 homolog, with CXCL1 (keratinocyte-derived chemokine) being considered the murine functional ortholog of IL-8. To our knowledge, there are no studies investigating whether CXCL1 can trigger MMP-9 release, but it is interesting to note that studies of human neutrophils have suggested that signaling through CXCR2, and not CXCR1, induces IL-8–dependent MMP-9 release (65).

e) being considered the murine functional ortholog of IL-8. To our knowledge, there are no studies investigating whether CXCL1 can trigger MMP-9 release, but it is interesting to note that studies of human neutrophils have suggested that signaling through CXCR2, and not CXCR1, induces IL-8–dependent MMP-9 release (65). CXCL1 levels were reduced in the absence of DR-3 (Figure 6), but to date, we have been unable to confirm the exact source of DR-3–dependent CXCL1 in the joints of mice with AIA. Neutrophils, macrophages, and epithelial cells have all been reported to release CXCL1 (66,67). The pattern of more Ly-6G–positive neutrophils in the fat pad but not around the synovial membranes in the joints of WT mice (Figure 5) would be consistent with a DR-3–independent source of CXCL1 from stromal cells, with further CXCL1 being provided by infiltrating cells in a DR-3–dependent manner. However, our in vitro experiments in bone marrow–derived macrophages from DR3−/− and WT mice have shown both increases and decreases in DR-3–dependent CXCL1 production triggered by the addition of TL1A (data not shown). This is likely to reflect the intrinsic plasticity of macrophages, coupled with the effects of DR-3/TL1A signaling impacting on target cells at different stages of differentiation. This has been previously observed with CD4+ T cells, in which TL1A inhibits the differentiation of naive cells to Th17 cells but maintains the numbers of these Teff cells once they are committed to the IL-17–producing lineage (23).

of DR-3/TL1A signaling impacting on target cells at different stages of differentiation. This has been previously observed with CD4+ T cells, in which TL1A inhibits the differentiation of naive cells to Th17 cells but maintains the numbers of these Teff cells once they are committed to the IL-17–producing lineage (23). The description of a reduction in the accumulation of neutrophils in the joints of DR3−/− mice 3 days after the generation of AIA is novel. Previous studies have suggested that cellular infiltration at this time point was not different between DR3−/− mice and WT mice (21), but in those studies only macrophage infiltration was investigated in any detail, using staining for F4/80. Here, we used Ly-6G as a stain, with the microscopic study of Ly-6G–positive cells showing morphologic characteristics of neutrophils (data not shown). In so doing, we also identify neutrophils as a major source of MMP-9 early in the course of AIA and highlight a novel function for DR-3, namely, the recruitment of neutrophils to inflamed joints. It is clear that the DR-3/TL1A pathway regulates multiple functions relating to the development of inflammatory and autoimmune disease, and further study is required to determine how antagonism of this pathway may be used as a potential treatment in the future.

namely, the recruitment of neutrophils to inflamed joints. It is clear that the DR-3/TL1A pathway regulates multiple functions relating to the development of inflammatory and autoimmune disease, and further study is required to determine how antagonism of this pathway may be used as a potential treatment in the future. AUTHOR CONTRIBUTIONS 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. Wang 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. Wang, Newton, Williams. Acquisition of data. Wang, Newton, Hayward, Clark, Collins, Perks, Singh, Twohig, Williams. Analysis and interpretation of data. Wang, Newton, Williams.

Systemic lupus erythematosus (SLE) is characterized by high titers of autoantibodies, typically against nuclear antigens. These autoantibodies generate immune complex–mediated inflammation in the kidneys, skin, joints, and cardiovascular system, with glomerulonephritis being a major contributor to resultant morbidity 1. Inflammation in the kidney is driven by cross-talk among immunoglobulin (Fc receptor [FcR]), complement, and Toll-like receptors (TLRs), resulting in the production of cytokines and infiltration of proinflammatory cells, which perpetuate chronic inflammation and organ damage 1–3. Studies in interleukin-23 (IL-23)–deficient mice suggest that the IL-23/IL-17 axis promotes such kidney inflammation 4, and, perhaps reflecting this, expanded populations of Th17- and IL-17–producing CD3+CD4−CD8− T cells are observed in the kidneys of both lupus-prone mice and patients with SLE 5. Moreover, IL-17 has been reported to act in concert with BAFF to promote B cell survival and (auto)antibody production 4–6. Consistent with the central role of B cells in the pathogenesis of SLE, increased expression of BAFF correlates with disease activity in SLE, and overexpression of BAFF promotes SLE-like pathology in mouse models, even in the absence of T cells. Specific targeting of this cytokine has proved effective in suppressing pathology, in both mouse models and human patients 3, and indeed, belimumab (an anti-BAFF monoclonal antibody) is the first SLE-specific treatment to be granted Food and Drug Administration approval in the past 50 years, although disappointingly, disease activity was reduced only in a limited number of patients during phase III trials 3,7.

use models and human patients 3, and indeed, belimumab (an anti-BAFF monoclonal antibody) is the first SLE-specific treatment to be granted Food and Drug Administration approval in the past 50 years, although disappointingly, disease activity was reduced only in a limited number of patients during phase III trials 3,7. Autoimmune inflammatory disorders appear to be increasingly prevalent in the developed world. As suggested by the hygiene hypothesis 8, this may reflect reduced exposure to infection, particularly by parasitic helminths (worms), which would normally shape and balance immune responses to limit pathology and promote tissue repair 9,10. Consistent with this notion, in experimental models of autoimmune disease, infection with helminths was shown to be protective 9,10, and this has generated interest in the potential for exploiting worm-based immunomodulation for the treatment of these inflammatory disorders in humans. Although clinical trials involving infection with live parasites have shown some promise in terms of therapeutic benefit to patients with autoimmune inflammatory disease 11, infection with pathogens is clearly not an ideal therapeutic strategy; thus, much recent attention has focused on the idea of developing novel drugs based on the individual helminth molecules (or their antiinflammatory targets) that promote parasite survival by limiting the inflammatory response of the host in a safe manner 9. In this study, we investigated whether ES-62, an immunomodulator secreted by the filarial nematode Acanthocheilonema viteae 9, protects against pathology in the MRL/lpr mouse model of SLE.

heir antiinflammatory targets) that promote parasite survival by limiting the inflammatory response of the host in a safe manner 9. In this study, we investigated whether ES-62, an immunomodulator secreted by the filarial nematode Acanthocheilonema viteae 9, protects against pathology in the MRL/lpr mouse model of SLE. MATERIALS AND METHODS Animal models Animals were bred and/or maintained in the Biological Services Units at the University of Glasgow and the University of Strathclyde, in accordance with Home Office UK Licences PIL60/9576, PIL60/11671, PIL60/12183, PIL60/12950, PPL60/3580, PPL60/4492, PPL60/4300, and PPL60/3810 and the ethics review boards of these universities. Although lupus-like pathology develops in MRL/Mp mice within 12–18 months, the Fas deficiency in the MRL/Mplpr/lpr (MRL/lpr) strain accelerates disease, with these mice developing (within 4 months) high-titer antinuclear antibodies (ANAs), glomerulonephritis, and arthritis-like footpad inflammation as well as the splenomegaly/lymphadenopathy typical of autoimmune lymphoproliferative syndrome 6,12.

as deficiency in the MRL/Mplpr/lpr (MRL/lpr) strain accelerates disease, with these mice developing (within 4 months) high-titer antinuclear antibodies (ANAs), glomerulonephritis, and arthritis-like footpad inflammation as well as the splenomegaly/lymphadenopathy typical of autoimmune lymphoproliferative syndrome 6,12. Kidney damage, as evidenced by proteinuria, was monitored twice weekly using Multistix (Siemens) and, where indicated, arthritis was scored at the time of culling 13,14. In addition, some mice were tested for renal function as evidenced by serum creatinine and blood urea nitrogen (BUN) levels, using relevant detection kits (Arbor Assays KB02-H1 and K024-H1; Tebu-Bio). The mice were treated twice weekly with phosphate buffered saline (PBS; 100 μl subcutaneously from 7 to 21 weeks of age), purified ES-62 (2 μg in 100 μl PBS subcutaneously from 7 to 21 weeks of age) 15, mouse IgG (Europa Bioproducts) (100 μg in 100 μl PBS intraperitoneally from 7 to 21 weeks of age), or neutralizing anti–IL-22 (AM22.1; 100 μg in 100 μl PBS intraperitoneally from 12 to 21 weeks of age) 16 or anti–IL-17A (MM17F3; 100 μg in 100 μl PBS intraperitoneally from 7 to 12 weeks of age) 17, monoclonal antibodies (kindly provided by Drs. Jean-Christophe Renauld and Jacques Van Snick, Ludwig Institute for Cancer Research, Belgium), or alternatively, with recombinant IL-22 (rIL-22; 1 μg in 100 μl PBS intraperitoneally from 12 to 21 weeks of age) or rIL-17A (PeproTech) (1 μg in 100 μl PBS intraperitoneally from 12 to 21 weeks of age). ES-62 inhibited proteinuria similarly in male and female MRL/lpr mice, and proteinuria levels for MRL/lpr mice treated with PBS and MRL/lpr mice treated with PBS plus IgG were not significantly different. The absence of endotoxin from these reagents was confirmed using an Endosafe Kit (Charles River) 15. Splenic B cells obtained from ES-62– or PBS-treated MRL/lpr mice at 21 weeks were purified by negative selection using anti-CD43–labeled magnetic beads (Miltenyi Biotec) (>90% B220+CD3− B2 cells) and transferred into the tail vein of recipient 7-week-old MRL/lpr mice (5 × 106 cells in 100 μl sterile PBS). Intravenously administered PBS (100 μl) was used as a control.

ed MRL/lpr mice at 21 weeks were purified by negative selection using anti-CD43–labeled magnetic beads (Miltenyi Biotec) (>90% B220+CD3− B2 cells) and transferred into the tail vein of recipient 7-week-old MRL/lpr mice (5 × 106 cells in 100 μl sterile PBS). Intravenously administered PBS (100 μl) was used as a control. Ex vivo analysis Blood samples were obtained by cardiac puncture, and red blood cells were lysed prior to flow cytometric analysis. Cells from the spleens or from popliteal, inguinal, axial, and brachial lymph nodes (LNs; 106/ml) were resuspended in RPMI medium containing 2 mMl-glutamine, 1 mM sodium pyruvate, 100 units/ml penicillin, 100 μg/ml streptomycin, and 1% nonessential amino acids (RPMI complete medium) supplemented with 50 μM 2-mercaptoethanol and 10% heat-inactivated fetal calf serum (all from Invitrogen). Dissection of kidneys and generation of kidney supernatants enriched in interstitial fluid were performed as described previously 18. Following red cell lysis, renal and hematopoietic kidney cells were analyzed by flow cytometry using a gating strategy based on forward scatter versus side scatter exclusion of dead cells/cell debris, exclusion of doublets, and selection of live cells as discriminated using a Live/Dead Fixable Aqua Dead Cell Stain kit (Invitrogen).

cell lysis, renal and hematopoietic kidney cells were analyzed by flow cytometry using a gating strategy based on forward scatter versus side scatter exclusion of dead cells/cell debris, exclusion of doublets, and selection of live cells as discriminated using a Live/Dead Fixable Aqua Dead Cell Stain kit (Invitrogen). For analysis of intracellular cytokine production, cells were incubated with medium or 50 ng/ml phorbol myristate acetate (PMA) plus 500 ng/ml ionomycin (plus 10 μg/ml lipopolysaccharide [Escherichia coli O111:B4] for B cell responses) for 1 hour before the addition of 10 μg/ml brefeldin A (Sigma-Aldrich) for a further 5 hours at 37°C with 5% CO2 13. Cells were stained with Live/Dead Fixable Aqua Dead Cell Stain to allow exclusion of dead cells from the analysis following permeabilization, using the solutions and protocols provided by BioLegend. B cell populations 13 were analyzed using the following phenotypic markers: Brilliant Violet 421–conjugated anti-B220, phycoerythrin (PE)–conjugated anti-CD138, Alexa Fluor 700– or PE–Cy7–conjugated anti-CD19; PE-conjugated anti-CD1d; PE–Cy7–conjugated anti-CD23, PE–Cy7–conjugated anti-CD43, PerCP–Cy5.5–conjugated anti-IgD, allophycocyanin (APC)–Cy7–conjugated anti-IgM, APC-conjugated anti-CD16/32, PerCP–Cy5.5–conjugated anti-CD80, APC-conjugated anti-CD206 (all from BioLegend), and eFluor 450–conjugated anti-CD21 (eBioscience) 13.

ugated anti-CD1d; PE–Cy7–conjugated anti-CD23, PE–Cy7–conjugated anti-CD43, PerCP–Cy5.5–conjugated anti-IgD, allophycocyanin (APC)–Cy7–conjugated anti-IgM, APC-conjugated anti-CD16/32, PerCP–Cy5.5–conjugated anti-CD80, APC-conjugated anti-CD206 (all from BioLegend), and eFluor 450–conjugated anti-CD21 (eBioscience) 13. For the identification of plasmablasts and plasma cells, a dump channel (PerCP) identifying CD11c, CD11b, CD4, CD8, F4/80, and Gr-1 (and CD3, when indicated) markers was used to facilitate exclusion of non-B CD138+ cells 13. Intracellular analysis involved staining with APC-conjugated anti–IL-6, APC-conjugated anti–IL-10 (BioLegend) 13, or anti-myeloid differentiation factor 88 (anti-MyD88) (Abcam) and fluorescein isothiocyanate (FITC)–conjugated goat anti-rabbit IgG (Vector) 14. Data were acquired using BD FACSCalibur and BD LSR II flow cytometers (BD Biosciences) and analyzed using FlowJo software (Tree Star) 13. MyD88 expression in the kidney (30 μg/sample) was additionally assessed by Western blotting using anti-MyD88 (ab2068; Abcam) and densitometric analysis using ImageJ software (National Institutes of Health) 14.

FACSCalibur and BD LSR II flow cytometers (BD Biosciences) and analyzed using FlowJo software (Tree Star) 13. MyD88 expression in the kidney (30 μg/sample) was additionally assessed by Western blotting using anti-MyD88 (ab2068; Abcam) and densitometric analysis using ImageJ software (National Institutes of Health) 14. The levels of cytokines (IL-17A [BioLegend], IL-17E and IL-17F [eBioscience], and IL-22 [R&D Systems]) in serum and kidney supernatants were analyzed by enzyme-linked immunosorbent assay 14,19. ANAs were visualized using HEp-2 slides (Antibodies Inc.) and FITC-conjugated anti-mouse IgG (Vector). Quantitative analysis was performed by determining the end point dilutions of serum from individual mice (102–105 log dilutions); the final dilution at which intracellular fluorescence was detectable was recorded. ANA reactivity was visualized using an Axiovert S100 fluorescence microscope (Zeiss).

-mouse IgG (Vector). Quantitative analysis was performed by determining the end point dilutions of serum from individual mice (102–105 log dilutions); the final dilution at which intracellular fluorescence was detectable was recorded. ANA reactivity was visualized using an Axiovert S100 fluorescence microscope (Zeiss). Kidney pathology Kidneys were fixed in formalin (24 hours at 4°C), treated with 30% sucrose (24 hours at 4°C), embedded in Tissue-Tek OCT medium, and snap-frozen in liquid nitrogen. Sections (7 μm) were stained with Harris' hematoxylin and eosin (Sigma-Aldrich) and imaged using an Olympus BX41 camera with Cell software. Hypercellularity was assessed by analyzing 20 glomerular cross-sections per kidney. Deposition of C3a and IgG was detected using rat anti-mouse C3 (11H9: Abcam)/Alexa Fluor 647–conjugated goat anti-rat IgG (Invitrogen) or rabbit anti-mouse IgG (The Jackson Laboratory)/Alexa Fluor 488–conjugated goat anti-rabbit IgG (Invitrogen), respectively, and visualized using an EVOS fluorescence microscope (Life Technologies). Differences between PBS- and ES-62–treated groups were detected using 1:10 and 1:25 dilutions (but not dilutions of ≥1:50) of the primary antibodies. Statistical analysis Proteinuria data were analyzed by two-way analysis of variance with the Bonferroni post hoc test, and experimental data were analyzed by Student's t-test. Nonparametric data were analyzed using the Mann-Whitney test.

Kidney pathology Kidneys were fixed in formalin (24 hours at 4°C), treated with 30% sucrose (24 hours at 4°C), embedded in Tissue-Tek OCT medium, and snap-frozen in liquid nitrogen. Sections (7 μm) were stained with Harris' hematoxylin and eosin (Sigma-Aldrich) and imaged using an Olympus BX41 camera with Cell software. Hypercellularity was assessed by analyzing 20 glomerular cross-sections per kidney. Deposition of C3a and IgG was detected using rat anti-mouse C3 (11H9: Abcam)/Alexa Fluor 647–conjugated goat anti-rat IgG (Invitrogen) or rabbit anti-mouse IgG (The Jackson Laboratory)/Alexa Fluor 488–conjugated goat anti-rabbit IgG (Invitrogen), respectively, and visualized using an EVOS fluorescence microscope (Life Technologies). Differences between PBS- and ES-62–treated groups were detected using 1:10 and 1:25 dilutions (but not dilutions of ≥1:50) of the primary antibodies. Statistical analysis Proteinuria data were analyzed by two-way analysis of variance with the Bonferroni post hoc test, and experimental data were analyzed by Student's t-test. Nonparametric data were analyzed using the Mann-Whitney test. RESULTS Suppression of proteinuria in MRL/lpr mice by ES-62 Proteinuria, a surrogate for glomerular vascular permeability, inflammation, and kidney damage, was detected in MRL/lpr mice, but not in MRL/Mp mice, by 16 weeks of age and increased thereafter, indicating that progressive kidney damage was occurring (Figure 1A). Treatment of the MRL/lpr mice with ES-62 reduced the level of proteinuria (Figure 1A) and the associated incidence of disease (>3 mg/ml protein: 0% in MRL/Mp mice, 100% in PBS-treated MRL/lpr mice, and 22.2% in ES-62–treated MRL/lpr mice). Moreover, additional parameters of renal function were also tested in some mice (proteinuria at 21 weeks: 20 mg/ml in all PBS-treated mice and mean ± SEM 0.98 ± 0.36 mg/ml in ES-62–treated mice), and this showed that ES-62 reduced serum creatinine levels both prior to and during established proteinuria (12 weeks and 21 weeks, respectively), although at 21 weeks, the reduction did not reach significance (Figure 1A).

weeks: 20 mg/ml in all PBS-treated mice and mean ± SEM 0.98 ± 0.36 mg/ml in ES-62–treated mice), and this showed that ES-62 reduced serum creatinine levels both prior to and during established proteinuria (12 weeks and 21 weeks, respectively), although at 21 weeks, the reduction did not reach significance (Figure 1A). Figure 1 ES-62 suppresses proteinuria and antinuclear antibody (ANA) production in MRL/lpr mice. A, Proteinuria was measured twice weekly in male MRL/Mp mice (n = 15) and MRL/lpr mice treated with either phosphate buffered saline (PBS) (n = 16) or ES-62 (n = 19), administered subcutaneously twice weekly from 7 to 21 weeks of age. Serum creatinine levels and arthritis articular scores were also determined in some of the individual MRL/lpr mice treated with PBS or ES-62 examined for these disease parameters. B, Left, Immunofluorescence imaging of HEp-2 cell staining by ANAs in the serum of PBS-treated and ES-62−treated MRL/lpr mice. Original magnification × 63. Middle, ANA levels in the serum of individual PBS-treated (n = 6) and ES-62−treated (n = 6) MRL/lpr mice, as measured by end point dilution analysis. Right, Immunofluorescence imaging showing weak detection of ANAs in kidney supernatant (KS) at a dilution of 102. Original magnification × 63. C, Proportions of plasmablast-like CD138+B220lowCD19+ B cells in the kidneys at 12 and 21 weeks (first and second panels), CD138+B220−CD19− plasma cells in the kidneys at 12 weeks (third panel), and plasmablasts and plasma cells in blood at 12 weeks in mice treated with PBS or ES-62 (fourth and fifth panels). Samples from the relevant treatment groups were pooled. In the fourth and fifth panels, values in the gates are the percentage of Dump−CD138+ cells; values in the boxes are the percentage of live cells. D, Levels of interleukin-17A (IL-17A) and BAFF in serum (first and second panels, respectively), levels of IL-17A in kidney supernatant (third panel) at the time of culling, as measured by enzyme-linked immunosorbent assay, and proteinuria (fourth panel) in MRL/lpr mice. Proteinuria was measured twice weekly in mice that received twice-weekly intraperitoneal injections of mouse IgG (100 μg in 100 μl PBS from 7 to 21 weeks, n = 16) or anti–IL-17A (100 μg in 100 μl PBS from 7 to 12 weeks, n = 8) or recombinant IL-17A (rIL-17A; 1 μg in 100 μl PBS from 12 to 21 weeks, n = 6). Values for proteinuria are the mean ± SEM and in A are collated from 3 independent experiments.

toneal injections of mouse IgG (100 μg in 100 μl PBS from 7 to 21 weeks, n = 16) or anti–IL-17A (100 μg in 100 μl PBS from 7 to 12 weeks, n = 8) or recombinant IL-17A (rIL-17A; 1 μg in 100 μl PBS from 12 to 21 weeks, n = 6). Values for proteinuria are the mean ± SEM and in A are collated from 3 independent experiments. In A (second and third panels), B (middle panel), C (first, second, and third panels), and D (first, second, and third panels), each symbol represents an individual mouse; bars show the mean. ∗ = P < 0.05; ∗∗ = P < 0.01; ∗∗∗ = P < 0.001. Black asterisks indicate PBS-treated MRL/lpr mice versus MRL/Mp mice. Blue asterisks indicate PBS-treated versus ES-62–treated MRL/lpr mice or rIL-17A–treated versus murine IgG–treated mice. In contrast, ES-62 did not significantly reduce BUN levels (data not shown), but this presumably reflected that these levels were still in the normal range (∼35 mg/dl), because typically these levels do not become elevated in MRL/lpr mice until ∼24–32 weeks of age 20–22. However, both the severity of arthritis (Figure 1A) and the incidence of arthritis (72.7% in the PBS-treated mice and 8.3% in the ES-62–treated mice) were suppressed in the ES-62–treated MRL/lpr mice examined. Moreover, although systematic survival analysis was precluded due to ethical constraints, exposure to ES-62 promoted the survival of MRL/lpr mice (66.7% of PBS-treated mice and 92.9% of ES-62–treated mice) over the 21-week time course of the experiments.

mice) were suppressed in the ES-62–treated MRL/lpr mice examined. Moreover, although systematic survival analysis was precluded due to ethical constraints, exposure to ES-62 promoted the survival of MRL/lpr mice (66.7% of PBS-treated mice and 92.9% of ES-62–treated mice) over the 21-week time course of the experiments. ES-62–induced suppression of ANA production in MRL/lpr mice ES-62 did not significantly modulate the levels of total IgG1, IgG2a, or IgM in the serum of MRL/lpr mice (data not shown). In contrast, ES-62 inhibited the production of ANAs, as measured in the serum of MRL/lpr mice, both prior to (12 weeks; data not shown) and during established disease (21 weeks). Similarly, the levels of ANA weakly detected in the kidney supernatants were also reduced following exposure to ES-62 (Figure 1B). Consistent with this suppression of pathogenic autoantibody production, the numbers (data not shown) and proportions of the plasmablast-like CD138+B220lowCD19+ B cells that may represent short-lived plasma cells 13 were reduced in the kidneys of mice exposed to ES-62, at both 12 weeks and 21 weeks (Figure 1C). These cells are associated with disease flares in patients with SLE and are the likely source of pathogenic anti–double-stranded DNA (anti-dsDNA) IgG2a and IgG3 autoantibodies 23. In contrast, the percentage of long-lived CD138+B220−CD19− plasma cells 13, which have been reported to be responsible for producing anti-RNA and anticardiolipin antibodies 23, was reduced in the kidneys prior to (Figure 1C) but not during established disease (mean ± SEM 1.8 ± 0.6% in PBS-treated mice and 2.4 ± 0.6% in ES-62–treated mice). At 21 weeks, however, the proportions of both plasmablasts and plasma cells were reduced in the blood of ES-62–treated mice relative to PBS-treated mice (Figure 1C).

d in the kidneys prior to (Figure 1C) but not during established disease (mean ± SEM 1.8 ± 0.6% in PBS-treated mice and 2.4 ± 0.6% in ES-62–treated mice). At 21 weeks, however, the proportions of both plasmablasts and plasma cells were reduced in the blood of ES-62–treated mice relative to PBS-treated mice (Figure 1C). Association between ES-62–induced suppression of B cell responses in MRL/lpr mice and resetting the balance between effector and regulatory B cells To address the mechanisms underpinning the suppression of ANA production and the reduced levels of CD138+B220lowCD19+ plasmablast-like B cells, we first investigated the effect of in vivo exposure to ES-62 on BAFF and IL-17 expression, because these cytokines have been proposed to synergize and promote (auto)antibody production 4,5. This investigation revealed that ES-62 did not suppress the levels of either cytokine in the serum or kidney supernatants of MRL/lpr mice (mean ± SEM BAFF levels in kidney supernatants 4,823 ± 333 pg/ml in PBS-treated mice and 5,084 ± 1,543 pg/ml in ES-62–treated mice) (Figure 1D), nor did the parasite product decrease the levels in kidney supernatants of IL-17E (mean ± SEM 949 ± 127 pg/ml in PBS-treated mice and 1,067 ± 153 pg/ml in ES-62–treated mice) or IL-17F (333 ± 61 pg/ml in PBS-treated mice and 433 ± 104 pg/ml in ES-62–treated mice), the latter of which was recently correlated with disease activity in SLE 24. Moreover, neutralizing anti–IL-17 antibodies did not block ANA production (data not shown) or development of proteinuria (Figure 1D). Indeed, ES-62 tended to promote IL-17 production in the kidney during established proteinuria, and consistent with this, administration of rIL-17 from 12 weeks onward partially suppressed proteinuria (incidence 40%) (Figure 1D).

es did not block ANA production (data not shown) or development of proteinuria (Figure 1D). Indeed, ES-62 tended to promote IL-17 production in the kidney during established proteinuria, and consistent with this, administration of rIL-17 from 12 weeks onward partially suppressed proteinuria (incidence 40%) (Figure 1D). Mice treated with ES-62 displayed increased total numbers of splenic CD19+ B cells (mean ± SEM 61.5 ± 7.5 × 106 in PBS-treated mice [n = 11] and 90.02 ± 9.2 × 106 in ES-62–treated mice [n = 18]) and follicular 1 B cells (CD19+CD93−CD21intermediateCD23+IgDhighIgMlow) (Figure 2A) but not T cells (results not shown), perhaps suggesting that the reduced plasmablast differentiation reflected induction of a hyporesponsive phenotype of B cells. Consistent with this, expression of CD80 on splenic B cells was down-regulated, while that of Fcγ receptor IIb was up-regulated (Figure 2B) in ES-62–treated MRL/lpr mice. Furthermore, following ex vivo stimulation, the levels of IL-6–producing splenic B cells, which are proposed to be an important driver of autoimmunity in mice 25, were reduced by in vivo exposure to ES-62 (Figure 2C). This was reflected by a reduction in the IL-6 messenger RNA levels (mean ± SEM relative quantity value 0.67 ± 0.13) observed in splenic CD19+CD3− B cells purified from ES-62–treated MRL/lpr mice when normalized to those from PBS-treated mice.

toimmunity in mice 25, were reduced by in vivo exposure to ES-62 (Figure 2C). This was reflected by a reduction in the IL-6 messenger RNA levels (mean ± SEM relative quantity value 0.67 ± 0.13) observed in splenic CD19+CD3− B cells purified from ES-62–treated MRL/lpr mice when normalized to those from PBS-treated mice. Figure 2 ES-62 modulates effector B cells by targeting myeloid differentiation factor 88 (MyD88). A, Total numbers of splenic follicular 1 (Fo1) B cells (CD19+CD93−CD21intermediateCD23+IgDhighIgMlow) in individual 21-week-old PBS- and ES-62–treated MRL/lpr mice, as determined by flow cytometry. B, Expression of CD80 (first and second panels) and Fcγ receptor IIb (FcγRIIb) (third and fourth panels) on CD19+ splenic B cells from PBS- and ES-62–treated MRL/lpr mice. C, Expression of IL-6–producing splenic B cells from MRL/lpr mice following in vivo exposure to PBS or ES-62. D, Intracellular levels of MyD88 in CD19+ and Dump–CD3–CD138+ B cells from PBS- and ES-62–treated MRL/lpr mice. In A (first panel), B (second and fourth panels), C (second panel), and D (middle panel), each symbol represents an individual mouse; bars show the mean. ∗ = P < 0.05. MFI = mean fluorescence intensity (see Figure 1 for other definitions).

8 in CD19+ and Dump–CD3–CD138+ B cells from PBS- and ES-62–treated MRL/lpr mice. In A (first panel), B (second and fourth panels), C (second panel), and D (middle panel), each symbol represents an individual mouse; bars show the mean. ∗ = P < 0.05. MFI = mean fluorescence intensity (see Figure 1 for other definitions). Similarly, such purified splenic B cells from ES-62–treated mice produced less interferon-γ than those from PBS-treated control mice (mean ± SEM 284 ± 104 versus 540 ± 51 pg/ml) in ex vivo cultures. However, surface expression of IgD and IgM was not modulated (data not shown); thus, ES-62 does not simply induce anergy resulting from down-regulation of the B cell receptor. Intriguingly, given the abrogation of ANA responses reported in MRL/lpr mice with MyD88-deficient B cells 26, B cells, including CD138+ B cells (Figure 2D) from ES-62–treated MRL/lpr mice, exhibited reduced levels of MyD88.

thus, ES-62 does not simply induce anergy resulting from down-regulation of the B cell receptor. Intriguingly, given the abrogation of ANA responses reported in MRL/lpr mice with MyD88-deficient B cells 26, B cells, including CD138+ B cells (Figure 2D) from ES-62–treated MRL/lpr mice, exhibited reduced levels of MyD88. In contrast, the levels of B cells with the capacity to produce IL-10, particularly CD19+CD21+CD23+ B cells analogous to those proposed to exhibit regulatory function in MRL/lpr mice and SLE 27,28, were increased in the spleen and kidney in ES-62–treated mice (Figure 3A). Moreover, the levels of CD19+CD21+CD23+ B cells were increased in the blood of ES-62–treated mice (Figure 3B). Such “regulatory” B cells have been reported to mediate their protective effects, at least in part, via the induction of Treg cells, particularly IL-10–producing CD4+ T (Tr1) cells, in MRL/lpr mice 27,28. Consistent with the proposed protective role of Treg cells in SLE 29, the proportion of FoxP3+CD4+ Treg cells and IL-10+CD4+Tr1 cells in the LNs of MRL/lpr, but not MRL/Mp, mice declined with age (although the absolute numbers increased), with kinetics that correlated with the initiation and progression of proteinuria. Treatment with ES-62, however, did not increase the levels of Treg or Tr1 cells in the LNs (Figure 3C), spleens (data not shown), or kidneys (Figure 3D) of MRL/lpr mice.

p, mice declined with age (although the absolute numbers increased), with kinetics that correlated with the initiation and progression of proteinuria. Treatment with ES-62, however, did not increase the levels of Treg or Tr1 cells in the LNs (Figure 3C), spleens (data not shown), or kidneys (Figure 3D) of MRL/lpr mice. Figure 3 ES-62 induces IL-10–producing B cells but not T cells. A, IL-10 production by spleen and kidney cells from PBS- and ES-62–treated 21-week-old MRL/lpr mice stimulated ex vivo, as determined by flow cytometry. B, Proportions of CD19+CD21+CD23+ B cells in the blood of PBS- and ES-62–treated MRL/lpr mice at 21 weeks. C, Proportions of FoxP3+ Treg cells (left) and IL-10+ Tr1 cells (right) in the lymph nodes (LNs) of MRL/Mp mice (n = 3), ES-62–treated MRL/lpr mice (n = 3–5), and PBS-treated MRL/lpr mice (n = 5) at all time points. Values are the mean ± SEM. D, Levels of Treg cells and Tr1 cells in the kidneys of MRL/lpr mice at 21 weeks. In A (third and fourth panels) and D, each symbol represents an individual mouse; bars show the mean. ∗ = P < 0.05. See Figure 1 for other definitions.

), and PBS-treated MRL/lpr mice (n = 5) at all time points. Values are the mean ± SEM. D, Levels of Treg cells and Tr1 cells in the kidneys of MRL/lpr mice at 21 weeks. In A (third and fourth panels) and D, each symbol represents an individual mouse; bars show the mean. ∗ = P < 0.05. See Figure 1 for other definitions. ES-62–induced protection against kidney damage correlates with antagonism of IL-22 responses and is mimicked by transfer of B cells from ES-62–treated MRL/lpr mice Perhaps surprisingly, given the striking inhibition of proteinuria, histologic analysis of kidneys from ES-62–treated MRL/lpr mice (Figure 4A) did not reveal any substantial modulation of glomerular hypercellularity, as confirmed by counting cells within individual glomeruli (mean ± SEM 63.6 ± 4.8 in PBS-treated mice [n = 26] and 64.4 ± 7.4 in ES-62–treated mice [n = 18]). Exposure to ES-62 did, however, reduce IgG and C3 deposition in the kidneys (Figure 4B) and modulated the phenotype of the infiltrating cell population, selectively reducing the proportion of CD3+ T cells, Lin− CD127+ innate lymphoid cells (ILCs), and CD11b+ cells (Figure 4C) while increasing the levels of antiinflammatory F4/80highCD11c−CD206+Ly-6G+ M2 macrophages (Figure 4D) (mean ± SEM 2 ± 0.4% of live cells in PBS-treated mice and 5.5 ± 2.5% of live cells in ES-62–treated mice) associated with protection, which appear to be depleted in SLE 30.

cells (ILCs), and CD11b+ cells (Figure 4C) while increasing the levels of antiinflammatory F4/80highCD11c−CD206+Ly-6G+ M2 macrophages (Figure 4D) (mean ± SEM 2 ± 0.4% of live cells in PBS-treated mice and 5.5 ± 2.5% of live cells in ES-62–treated mice) associated with protection, which appear to be depleted in SLE 30. Figure 4 ES-62 modulates cellular infiltration and deposition of IgG and C3 in the kidney. A, Glomerular hypercellularity (top) and cellular infiltration (bottom) in hematoxylin and eosin−stained kidney sections from 21-week-old MRL/Mp and MRL/lpr mice treated with PBS or ES-62. Original magnification × 40 (top); × 10 (bottom). B, Deposition of IgG and C3 in the kidneys of MRL/lpr mice treated with PBS or ES-62, as detected using rabbit anti-mouse IgG/Alexa Fluor 488−conjugated goat anti-rabbit IgG and rat anti-mouse C3/Alexa Fluor 647−conjugated goat anti-rat IgG, respectively. Original magnification × 40. C, Proportions of CD3+ T cells, Lin−CD127+ innate lymphoid cells, and CD11b+ cells in the kidneys of MRL/lpr mice treated with PBS or ES-62. Each symbol represents an individual mouse; bars show the mean. D, Levels of CD206+Ly-6G+ M2 macrophages in the kidneys of MRL/lpr mice treated with PBS or ES-62. The values shown represent the percentage of cells in the CD206+Ly-6G+ gate; when indicated, however, these were further analyzed as F4/80+CD11c− cells as a proportion of live cells. ∗ = P < 0.05. See Figure 1 for definitions.

vels of CD206+Ly-6G+ M2 macrophages in the kidneys of MRL/lpr mice treated with PBS or ES-62. The values shown represent the percentage of cells in the CD206+Ly-6G+ gate; when indicated, however, these were further analyzed as F4/80+CD11c− cells as a proportion of live cells. ∗ = P < 0.05. See Figure 1 for definitions. The cytokine IL-22, which can promote barrier integrity and wound repair 31, was recently reported to stimulate kidney regeneration after acute injury by acting on tubular epithelial cells 32. In the current study, however, ES-62 suppressed the levels of IL-22 in kidney supernatants (Figure 5A). Thus, as IL-23 promotes IL-22 responses, and because of the increasing recognition of pathogenic roles for IL-22 in autoimmune disorders 33,34 including SLE 35–38, we investigated whether IL-22 production in the kidney was associated with SLE pathogenesis. Administration of rIL-22 significantly accelerated and exacerbated the development of proteinuria. In contrast, neutralization of this cytokine suppressed proteinuria, although treatment with anti–IL-22 did not prevent glomerular hypercellularity or ANA production (Figure 5B). Exposure of MRL/lpr mice to ES-62 plus anti–IL-22 resulted in no significant differences between this combination treatment and protocols with either anti–IL-22 or ES-62 plus IgG alone (data not shown). Instead, although expression of MyD88 was up-regulated in kidney cells from MRL/lpr mice relative to MRL/Mp mice, it was reduced in kidney cells from ES-62–treated MRL/lpr mice and anti–IL-22–treated MRL/lpr mice, and rIL-22 appeared to maintain (if not substantially increase) MyD88 levels (Figure 5C). Collectively, these data are consistent with the notion that IL-22 plays a pathogenic role in promoting MyD88-dependent inflammation and vascular barrier permeability in the MRL/lpr mouse and suggest that this cytokine activity may be targeted by ES-62 to mediate some of its protective effects in the kidney.

igure 5C). Collectively, these data are consistent with the notion that IL-22 plays a pathogenic role in promoting MyD88-dependent inflammation and vascular barrier permeability in the MRL/lpr mouse and suggest that this cytokine activity may be targeted by ES-62 to mediate some of its protective effects in the kidney. Figure 5 ES-62 suppresses pathogenic IL-22 production and myeloid differentiation factor 88 (MyD88) expression in the kidney. A, IL-22 levels in kidney supernatant derived from PBS- and ES-62−treated MRL/lpr mice, as measured by enzyme-linked immunosorbent assay. B, Left, Proteinuria in MRL/lpr mice that received twice-weekly intraperitoneal injections of rIL-22 (1 μg in 100 μl PBS from 12 to 21 weeks of age, n = 10), anti−IL-22 (100 μg in 100 μl PBS from 12 to 21 weeks of age, n = 9), or mouse IgG (100 μg in 100 μl PBS from 7 to 21 weeks of age, n = 16). The IgG data are the same as those shown in Figure 1D, as the matched IL-17 and IL-22−modulated groups were analyzed in parallel cohorts to promote the 3 Rs (replace, reduce, refine). Right, Glomerular hypercellularity in hematoxylin and eosin (H&E)−stained kidney sections (top) and ANA production in serum (102 dilution) (bottom) from MRL/lpr mice treated with PBS or anti−IL-22. Original magnification × 40 (top); × 63 (bottom). C, Left, MyD88 expression in kidney cells from MRL/Mp and MRL/lpr mice at 12 weeks of age (top) and from PBS- and ES-62−treated MRL/lpr mice at 21 weeks of age (bottom), as determined by flow cytometric analysis. Middle, MyD88 expression in the kidneys of individual PBS- or ES-62−treated MRL/lpr mice. Right, Western blots showing MyD88 levels in kidney protein lysates derived from MRL/lpr mice treated with IgG, anti−IL-22, or rIL-22 (top), and densitometric analysis of MyD88/β-actin expression, normalized to IgG control (bottom). D, Proteinuria (left), glomerular hypercellularity and ANA production (middle), and IL-22 expression (right) in recipient MRL/lpr mice in which splenic B cells harvested from MRL/lpr mice were transferred. Purified splenic CD43− B cells from 21-week-old MRL/lpr mice treated with ES-62 (n = 4) or PBS (n = 4) were transferred (5 × 106 cells pooled in 100 μl sterile PBS) into the tail veins of 7-week-old MRL/lpr mice; intravenous (IV) PBS was used as a control. Proteinuria in the recipient mice (6 received PBS, 8 received PBS-treated B cells, and 8 received B cells from ES-62–treated mice) was measured twice weekly.

(n = 4) were transferred (5 × 106 cells pooled in 100 μl sterile PBS) into the tail veins of 7-week-old MRL/lpr mice; intravenous (IV) PBS was used as a control. Proteinuria in the recipient mice (6 received PBS, 8 received PBS-treated B cells, and 8 received B cells from ES-62–treated mice) was measured twice weekly. Kidney sections were stained with H&E for analysis of glomerular hypercellularity, and serum analyzed for ANA production. IL-22 was measured in the kidneys of mice from one of these experiments. Values for proteinuria are the mean ± SEM (n = number of relevant individual mice pooled from 2 independent experiments). In A, C (middle), and D (right), each symbol represents an individual mouse; bars show the mean. ∗ = P < 0.05; ∗∗ = P < 0.01; ∗∗∗ = P < 0.001. MFI = mean fluorescence intensity (see Figure 1 for other definitions).

inuria are the mean ± SEM (n = number of relevant individual mice pooled from 2 independent experiments). In A, C (middle), and D (right), each symbol represents an individual mouse; bars show the mean. ∗ = P < 0.05; ∗∗ = P < 0.01; ∗∗∗ = P < 0.001. MFI = mean fluorescence intensity (see Figure 1 for other definitions). Finally, in order to confirm that the protection against proteinuria afforded by the helminth product was attributable to ES-62 resetting effector B cell responses, we investigated the effect of adoptively transferring purified splenic B cells harvested from MRL/lpr mice treated with either PBS or ES-62 into recipient 7-week-old MRL/lpr mice. Strikingly, the transfer of B cells from ES-62–treated MRL/lpr mice was sufficient to provide significant protection against the development of proteinuria in the recipient mice despite, as was observed with ES-62, no substantial improvement in glomerular hypercellularity (Figure 5D). However, transfer of B cells from ES-62–treated mice did suppress pathogenic ANA production (Figure 5D) (mean ± SEM reciprocal end point dilutions 68,500 ± 19,956 in mice treated with intravenous PBS, 55,000 ± 25,981 in mice that received B cells from MRL/lpr mice treated with PBS, and 5,500 ± 5,196 in mice that received B cells from MRL/lpr mice treated with ES-62) and reduced IL-22 levels in kidney supernatants (Figure 5D), indicating that modulation of IL-22 responses is secondary to ES-62–mediated resetting of the balance between effector and regulatory B cells.

MRL/lpr mice treated with PBS, and 5,500 ± 5,196 in mice that received B cells from MRL/lpr mice treated with ES-62) and reduced IL-22 levels in kidney supernatants (Figure 5D), indicating that modulation of IL-22 responses is secondary to ES-62–mediated resetting of the balance between effector and regulatory B cells. DISCUSSION ES-62 significantly suppresses pathogenic ANA production (but not total IgG or IgM production) and consequent deposition of IgG and C3a in the kidneys of MRL/lpr mice. In addition, ES-62 reduces IL-22 levels and modulates the phenotype of the kidney cellular infiltrate, effects that collectively result in suppressed development of proteinuria (Figure 6), a biomarker of the kidney inflammation and damage that are the major causes of mortality in the MRL/lpr mouse model of SLE 12. ES-62 appears to prevent ANA generation by down-regulating the expression of B cell MyD88; the notion that this homeostatic regulation of effector B cell responsiveness by the helminth product is relevant to the observed protection is validated by studies in which purified splenic B cells from ES-62–treated MRL/lpr mice similarly suppress ANA production and also development of proteinuria in recipient MRL/lpr mice. Moreover, and consistent with our findings, studies by other investigators using the MRL/lpr 26 and Lyn−/− 39 mouse models of SLE have shown that complete deletion of MyD88 in B cells is sufficient to abrogate ANA production (but not that of total IgG), proteinuria, and glomerulonephritis, and as with ES-62, reduces the number of plasmablasts while increasing the numbers of follicular and total splenic B cells.

6 and Lyn−/− 39 mouse models of SLE have shown that complete deletion of MyD88 in B cells is sufficient to abrogate ANA production (but not that of total IgG), proteinuria, and glomerulonephritis, and as with ES-62, reduces the number of plasmablasts while increasing the numbers of follicular and total splenic B cells. Figure 6 Schematic representation of the action of ES-62. ES-62 desensitizes hyperresponsive B cells in MRL/lpr mice, resulting in a lower level of plasmablasts and, consequently, suppression of antinuclear antibody (ANA) production and IgG and C3a deposition in the kidney. This population of hyporesponsive effector B cells has a higher frequency of B cells that can potentially reset the balance between effector cells and regulatory cells via the production of interleukin-10 (IL-10) and/or other regulatory B (Breg) cell mechanisms. In addition to reducing the levels of pathogenic ANAs, this hyporesponsive phenotype is associated with a reduction in pathogenic IL-22 responses in the kidney, which is reflected, at least in part, by down-regulation of myeloid differentiation factor 88 (MyD88) expression in kidney cells. ES-62 may also act directly to down-regulate MyD88 in kidney cells and in this way act to desensitize the pathogenic cross-talk among Toll-like receptors, Fc receptors, and complement receptors that results in glomerular vascular permeability, inflammation, and kidney damage.

ctor 88 (MyD88) expression in kidney cells. ES-62 may also act directly to down-regulate MyD88 in kidney cells and in this way act to desensitize the pathogenic cross-talk among Toll-like receptors, Fc receptors, and complement receptors that results in glomerular vascular permeability, inflammation, and kidney damage. The IL-23/IL-17A axis has been proposed to play a pathogenic role in SLE 4,5, but ES-62 did not suppress IL-17 production, and neutralizing antibodies to this cytokine did not abrogate disease progression. Although anti–IL-17 treatment has been shown to suppress the levels of anti-dsDNA antibodies and proteinuria in the MRL/lpr mouse model, this occurred during the initiation phase (∼12 weeks) 6, and, consistent with our observations that the number of plasma cells decreased at this stage, ES-62 suppressed the levels of IL-17A–producing CD4+ and γ/δ T cells in LNs at 9 weeks (data not shown). Moreover, although levels of IL-17A have been widely shown to be elevated in patients with SLE, only a few studies have shown that IL-17A levels correlate with the SLE Disease Activity Index (SLEDAI) 40–48, with slightly more studies concluding either that IL-17A levels do not correlate with the SLEDAI or that these levels are actually inversely correlated with the disease score 49–60.

vated in patients with SLE, only a few studies have shown that IL-17A levels correlate with the SLE Disease Activity Index (SLEDAI) 40–48, with slightly more studies concluding either that IL-17A levels do not correlate with the SLEDAI or that these levels are actually inversely correlated with the disease score 49–60. Of note, IL-23 can also promote IL-22 responses 61, a finding that perhaps reconciles some of these contradictory data relating to the role of the inflammatory IL-23/IL-17 axis in SLE. Pertinently, we observed that IL-17 and IL-22 are produced by distinct populations of cells in the MRL/lpr mouse, with typically at the time of culling only 1–2% of IL-22+ cells in LNs producing IL-17 in response to ex vivo stimulation with PMA/ionomycin. Moreover, although exposure to ES-62 results in significantly lower levels of IL-22+ but not IL-17+ CD3+B220+CD4−CD8− (double-negative) T cells, it induces higher levels of IL-17+ but not IL-22+ ILCs, suggesting that targeting of particular cellular sources of IL-17 and IL-22 allows ES-62 to modulate the distinct, potentially counterregulatory, effector functions of these cytokines in the MRL/lpr mouse (data not shown).

+B220+CD4−CD8− (double-negative) T cells, it induces higher levels of IL-17+ but not IL-22+ ILCs, suggesting that targeting of particular cellular sources of IL-17 and IL-22 allows ES-62 to modulate the distinct, potentially counterregulatory, effector functions of these cytokines in the MRL/lpr mouse (data not shown). Thus, while not ruling out the possibility that IL-17A may promote initiation of pathogenesis, our data clearly suggest that this cytokine can resolve inflammation during established disease, a hypothesis that is consistent with the proposed dual roles of IL-17 in initiating and resolving kidney disease in experimental crescentic glomerulonephritis 62. In contrast, and consistent with its role in linking the regulation of inflammatory responses and barrier tissue homeostasis 31,34, IL-22 appears to promote pathogenic effector cell mechanisms in the kidney. Thus, although a disease-causing role for IL-22 has yet to be established unequivocally in SLE, our data resonate with reports that this cytokine is associated with pathogenesis 36,37 and may be predictive of specific pathologies in certain patients with SLE 35.

romote pathogenic effector cell mechanisms in the kidney. Thus, although a disease-causing role for IL-22 has yet to be established unequivocally in SLE, our data resonate with reports that this cytokine is associated with pathogenesis 36,37 and may be predictive of specific pathologies in certain patients with SLE 35. Although deletion of MyD88 signaling in B cells is sufficient to abrogate lupus nephritis in the MRL/lpr and Lyn−/− mouse models 26,39, MyD88 deficiency in dendritic cells (DCs) can also confer some protection and highlights the cooperation between B cells and DCs 26,63, and potentially other (nonhematopoietic) cells, in the development of SLE-like pathologies. Interestingly, therefore, MyD88-dependent cooperation between myeloid and endothelial cells was recently proven to be key to the promotion of the vascular inflammation and atherosclerosis associated with metabolic syndrome 64. Thus, because MyD88 is likely to be a key player in integrating cross-talk among TLR, FcR, and complement receptors, in concert with the reduction in IgG and C3a deposition, our finding that ES-62 and anti–IL-22 down-regulate MyD88 expression in kidney cells provides an effective mechanism for protecting against kidney inflammation. In addition, and relating to the IL-22–mediated regulation of barrier function and vascular inflammation alluded to above, IL-1β was recently shown to act on endothelial cells to stimulate an NF-κB–independent, MyD88/ADP ribosylation factor (ARF) nucleotide binding site opener (ARNO)/ARF 6–dependent pathway of vascular permeability that contributes to vasculitis in autoimmune disease 65. Therefore, disruption of the cooperative interplay resulting from partial down-regulation of MyD88 signaling in B cells and kidney cells would act to break the persistent cycle of inflammation and vascular permeability resulting in kidney damage in SLE without fully immunosuppressing the host.

in autoimmune disease 65. Therefore, disruption of the cooperative interplay resulting from partial down-regulation of MyD88 signaling in B cells and kidney cells would act to break the persistent cycle of inflammation and vascular permeability resulting in kidney damage in SLE without fully immunosuppressing the host. The ability of ES-62 to suppress pathogenic B cell and effector cell responses appears to be associated with a homeostatic resetting of the effector cell–to–regulatory cell balance, as indicated by the reduction in the frequency of pathogenic plasmablasts and the increased number of CD19+CD23+CD21+ B cells with the capacity to produce IL-10, a phenotype reminiscent of the regulatory B cells proposed to be defective in MRL/lpr mice 28 and SLE 27. Interestingly, there is increasing evidence that the dampening of inflammatory responses by some helminth products may reflect, at least in part, activation of regulatory B (“Breg”) cell function 66,67. Indeed, patients with multiple sclerosis who have helminthic infection exhibit less severe disease, and such protection appears to be associated with elevated levels of IL-10–producing B cells 68. Moreover, in a mouse model of asthma, IL-10–producing B cells induced by the trematode helminth Schistosoma mansoni suppressed disease 69,70, and in addition, B cells from mice infected with the gastrointestinal nematode Heligmosomoides polygyrus protected against the development of both allergic airway inflammation and autoimmune inflammation in an experimental model of autoimmune encephalomyelitis 71. Nevertheless, recent genetic studies suggest that regulatory B cells do not counterregulate pathogenic effector B cell responses in the MRL/lpr mouse in an IL-10–dependent manner 72, and although these data may reflect their dysfunctional phenotype and loss in this mouse strain, our preliminary in vitro data also suggest that while B cells from ES-62–treated MRL/lpr mice inhibit T cell responses, they may do so in an IL-10–independent manner.

es in the MRL/lpr mouse in an IL-10–dependent manner 72, and although these data may reflect their dysfunctional phenotype and loss in this mouse strain, our preliminary in vitro data also suggest that while B cells from ES-62–treated MRL/lpr mice inhibit T cell responses, they may do so in an IL-10–independent manner. These findings reflect the increasing recognition that regulatory B cells can exploit a variety of mechanisms to limit chronic inflammation 73, and in any case, expanded populations of such regulatory B cells can transfer protection in MRL/lpr mice 27,28, suggesting that regardless of their mode of action, they could be exploited therapeutically. To date, however, therapies aimed at targeting effector B cells and/or resetting the balance between effector and regulatory cells have been disappointing in clinical trials in SLE; therefore, the need remains for the development of new and safer therapies to achieve this. Thus, exploiting the targets identified by parasitic helminths, which appear to have evolved such homeostatic actions as regulating proinflammatory MyD88 signaling in several cell types including B cells, as a general and safe mechanism to dampen hyperinflammatory responses may provide an alternative blueprint for the development of novel biologic agents or drugs to treat SLE.

ths, which appear to have evolved such homeostatic actions as regulating proinflammatory MyD88 signaling in several cell types including B cells, as a general and safe mechanism to dampen hyperinflammatory responses may provide an alternative blueprint for the development of novel biologic agents or drugs to treat SLE. AUTHOR CONTRIBUTIONS 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. M. M. Harnett 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. Rodgers, McGrath, Pineda, W. Harnett, M. M. Harnett. Acquisition of data. Rodgers, McGrath, Pineda, Al-Riyami, Rzepecka, Lumb. Analysis and interpretation of data. Rodgers, McGrath, Al-Riyami, Lumb, W. Harnett, M. M. Harnett. We thank Drs. Agnes Boitelle and Dorothy Kean for their contributions to our pilot studies.

Pain is the most dominant and impairing symptom associated with rheumatoid arthritis (RA). Patients may experience pain before clinical signs permit confirmation of the diagnosis of RA 1; thus, pain is present at the predisease stage as well as during the disease stage of RA. Treatment of RA pain with nonsteroidal antiinflammatory drugs (NSAIDs) results in modest efficacy and may produce side effects 2. Improved understanding of the specific mechanisms of RA‐associated pain will enable us to identify new strategies for analgesia. Collagen‐induced arthritis (CIA) is a clinically relevant model of RA. The immunopathogenesis of CIA involves both B and T lymphocyte responses, with the production of type II collagen–specific antibodies that bind to cartilage in the joints 3. The resulting pathogenesis shares several pathologic features with RA, including synovial hyperplasia, inflammatory cell infiltration, and cartilage degradation 3. However, only a few studies (by our group and others) have investigated the mechanisms underlying pain in this model in either mice 4 or rats 5.

e joints 3. The resulting pathogenesis shares several pathologic features with RA, including synovial hyperplasia, inflammatory cell infiltration, and cartilage degradation 3. However, only a few studies (by our group and others) have investigated the mechanisms underlying pain in this model in either mice 4 or rats 5. Although cartilage is not innervated, inflammation of the synovial membrane and bone alterations can lead to the sensitization of primary afferent fibers (nociceptors) that innervate the joints and tissue outside the joints (peripheral sensitization) and respond to noxious stimuli 6. All nociceptors contain glutamate, while the peptidergic subpopulation of nociceptors also contains substance P (SP) and calcitonin gene‐related peptide (CGRP) 7, 8 and is especially enriched in the joint 9. Increased input from such sensitized afferent fibers, whose cell bodies are located in the dorsal root ganglia (DRGs), can lead to an augmented release of glutamate, SP, and CGRP from their central terminals in the spinal cord 10, whereby increased activation of specific receptors in dorsal horn neurons results in amplification of signaling (central sensitization) 6. Along with neurons, spinal microglia are known to respond to increased neuronal activity and play modulatory roles by releasing pronociceptive mediators 11, 12, 13, 14. Central sensitization can contribute to secondary hyperalgesia in joint‐adjacent tissue (hind paw), as many spinal cord neurons receive convergent inputs from skin and deep tissues 6, 13.

are known to respond to increased neuronal activity and play modulatory roles by releasing pronociceptive mediators 11, 12, 13, 14. Central sensitization can contribute to secondary hyperalgesia in joint‐adjacent tissue (hind paw), as many spinal cord neurons receive convergent inputs from skin and deep tissues 6, 13. Our recent work demonstrated that the early stages of CIA are associated with increased nocifensive behavior prior to the appearance of clinical signs of the disease, whereas at later stages nocifensive responses are present concomitant with significant hind paw swelling and enhanced spinal microglial response 5. The weak correlation between pain and swelling in the early stages of CIA mimics the clinical situation 1 and suggests that mechanisms other than overt inflammation contribute to pain at this stage. Thus, in this study we evaluated whether nociceptive sensory neurons innervating the joint and adjacent tissues are recruited and activated during the development of pain and inflammation in CIA and contribute to associated spinal mechanisms. MATERIALS AND METHODS Animals Experiments were performed in 70 female adult Lewis rats weighing 180–200 gm (Charles River UK). Experimental study groups were randomized, and assessments were performed under blinded conditions. All experiments were undertaken with approval of the UK Home Office.

Our recent work demonstrated that the early stages of CIA are associated with increased nocifensive behavior prior to the appearance of clinical signs of the disease, whereas at later stages nocifensive responses are present concomitant with significant hind paw swelling and enhanced spinal microglial response 5. The weak correlation between pain and swelling in the early stages of CIA mimics the clinical situation 1 and suggests that mechanisms other than overt inflammation contribute to pain at this stage. Thus, in this study we evaluated whether nociceptive sensory neurons innervating the joint and adjacent tissues are recruited and activated during the development of pain and inflammation in CIA and contribute to associated spinal mechanisms. MATERIALS AND METHODS Animals Experiments were performed in 70 female adult Lewis rats weighing 180–200 gm (Charles River UK). Experimental study groups were randomized, and assessments were performed under blinded conditions. All experiments were undertaken with approval of the UK Home Office. CIA As described previously 5, 4 mg/ml bovine type II collagen (MD Bioproducts) was dissolved in 0.1M acetic acid and emulsified with 1 mg/ml Freund's complete adjuvant (CFA; BD Biosciences). Rats were anesthetized with isoflurane (Abbott) and injected intradermally at the base of the tail with 400 μg collagen/200 μl CFA or 200 μl CFA emulsion (control rats).

vine type II collagen (MD Bioproducts) was dissolved in 0.1M acetic acid and emulsified with 1 mg/ml Freund's complete adjuvant (CFA; BD Biosciences). Rats were anesthetized with isoflurane (Abbott) and injected intradermally at the base of the tail with 400 μg collagen/200 μl CFA or 200 μl CFA emulsion (control rats). Macroscopic assessment of arthritis Rats were scored on a scale of 0–3 per hind paw (0–6 per rat) 5. Ankle swelling, the earliest visible sign of arthritis, was scored as 1. Thereafter, footpad swelling occurred and was scored as 2. Subsequent swelling of 1 or more digits resulted in a score of 3. The thickness of each hind paw was measured using a thickness gauge (Mitutoyo) and expressed in mm. Clinical signs of arthritis and body weight were monitored prior to immunization and then throughout the disease process.

ng occurred and was scored as 2. Subsequent swelling of 1 or more digits resulted in a score of 3. The thickness of each hind paw was measured using a thickness gauge (Mitutoyo) and expressed in mm. Clinical signs of arthritis and body weight were monitored prior to immunization and then throughout the disease process. Behavior indicating pain Mechanical and thermal hypersensitivity in the hind paws was assessed as an indicator of secondary hyperalgesia remote from the inflamed ankle joint. Changes in hind paw mechanical withdrawal thresholds were assessed by applying a series of calibrated von Frey filaments (0.4–15 gm; North Coast Medical) to the plantar surface of the hind paw according to the “up‐down” method 15. On each day of testing, animals were habituated for 15 minutes in individual transparent Plexiglas boxes with a wire mesh bottom in a temperature‐controlled room (22°C). Calibrated von Frey filaments were applied to the plantar surface of the hind paw for 3–5 seconds or until the paw was withdrawn. Mechanical thresholds of the left and right paws were assessed alternately. Each test started with application of the 2‐gm filament. Once a withdrawal response to a von Frey hair was established, the paw was retested starting with the filament below the one that elicited a withdrawal and subsequently with the remaining filaments in descending force sequence until no withdrawal occurred and then in ascending force sequence until a response was observed. This up‐down method was continued until the “k” value could be calculated (between 4 and 9 applications of the von Frey hairs). From this value, 50% withdrawal thresholds were calculated.

aments in descending force sequence until no withdrawal occurred and then in ascending force sequence until a response was observed. This up‐down method was continued until the “k” value could be calculated (between 4 and 9 applications of the von Frey hairs). From this value, 50% withdrawal thresholds were calculated. Heat hyperalgesia was assessed with a Plantar test apparatus (Ugo Basile) as previously described 16. Rats were habituated for 15 minutes to individual Plexiglas chambers placed on a glass floor in a temperature‐controlled room (22°C). A beam of radiant heat was applied onto the plantar surface of each hind paw, and paw withdrawal latencies were recorded. Withdrawal latencies were monitored 3 times in each paw alternately, and the latencies of both paws were averaged on each measurement day. At least 1 minute was allowed between consecutive measurements in the same paw, and a cutoff latency time of 20 seconds was used to avoid tissue damage. Baseline measurements were obtained prior to collagen immunization, and withdrawal latencies were then recorded on several days throughout the disease process.

nt day. At least 1 minute was allowed between consecutive measurements in the same paw, and a cutoff latency time of 20 seconds was used to avoid tissue damage. Baseline measurements were obtained prior to collagen immunization, and withdrawal latencies were then recorded on several days throughout the disease process. Intrathecal drug treatment Two weeks prior to collagen immunization, intrathecal cannulae were implanted at the lumbar level in the spinal cord. While animals were under anesthesia, a small laminectomy was performed at the sixth thoracic vertebra and a flexible cannula was inserted under the dura mater and glued in place, such that the tip rested at the lumbar enlargement. The opposite end of the cannula was placed subcutaneously, and 11 days after immunization an osmotic minipump (Alzet model 2001) was connected to the cannula 5. CGRP receptor antagonist CGRP8–37 (1 mg/ml; Bachem) was delivered at 24 μl/day for 7 days (from day 11 to day 18 postimmunization), resulting in a dose of 24 μg/day as reported previously 17. Histology Ankle joints were excised from rats with CIA and control rats that were perfuse‐fixed on days 7 and 18 postimmunization. The joints were postfixed for 1 week in the perfusion fixative, decalcified for 3 days in 10% formic acid with 0.5M trisodium citrate, and embedded in paraffin. Longitudinal sections (7 μm) were cut from the center of the ankle joint in the sagittal plane and stained with hematoxylin and eosin. Sections were examined under light microscopy for cellular infiltration, synovitis, and structural integrity.

s in 10% formic acid with 0.5M trisodium citrate, and embedded in paraffin. Longitudinal sections (7 μm) were cut from the center of the ankle joint in the sagittal plane and stained with hematoxylin and eosin. Sections were examined under light microscopy for cellular infiltration, synovitis, and structural integrity. Retrograde neuronal labeling To identify the cell bodies of ankle joint afferent neurons, 5 μl of 2% Fluoro‐Gold (FG; Fluorochrome) was injected intraarticularly using a Hamilton syringe into the ankle joint of the rats 7 days prior to the isolation of DRGs, as previously described 18. As intraarticular injection of FG does not produce bilateral labeling and collagen immunization affects both joints, we performed intraarticular injections of Fluoro‐Gold in both ankle joints.

ly using a Hamilton syringe into the ankle joint of the rats 7 days prior to the isolation of DRGs, as previously described 18. As intraarticular injection of FG does not produce bilateral labeling and collagen immunization affects both joints, we performed intraarticular injections of Fluoro‐Gold in both ankle joints. Tissue processing On days 4, 7, and 18 postimmunization, naive rats, rats with CIA, and control rats were anesthetized with pentobarbital and transcardially perfused with 0.9% saline solution followed by 4% paraformaldehyde with 1.5% picric acid in 0.1M phosphate buffer (pH 7.4). L4 and L5 DRGs and lumbar spinal cords were excised, postfixed for 4 hours in the perfusion fixative (4°C), cryoprotected in 20% sucrose in phosphate buffer (0.1M, 4°C) for 48 hours, and frozen in OCT embedding compound (VWR International). DRG (15 μm) and spinal cord (20 μm) sections were cryostat‐cut and thaw‐mounted onto Superfrost Plus microscope slides (VWR International). In preliminary experiments all lumbar DRGs (L1–L6) were excised from naive rats and processed in the same way in order to elucidate the segmental distribution of DRG neurons innervating the ankle joint.

cord (20 μm) sections were cryostat‐cut and thaw‐mounted onto Superfrost Plus microscope slides (VWR International). In preliminary experiments all lumbar DRGs (L1–L6) were excised from naive rats and processed in the same way in order to elucidate the segmental distribution of DRG neurons innervating the ankle joint. Immunohistochemistry Slides containing every sixth section of L4 and L5 DRGs from rats previously injected or not injected with FG were blocked with 1% normal goat serum (Jackson ImmunoResearch) for 1 hour and then incubated overnight with solutions of primary antibodies (sheep anti‐CGRP [1:800; Enzo], monoclonal rabbit anti–phospho‐p44/42 MAPK [ERK‐1/2] [p‐ERK] [1:400; Cell Signaling Technology], and/or Anti‐βIII Tubulin monoclonal antibody [1:1,000; Promega]) followed by incubation for 2 hours with solutions of appropriate secondary antibodies (Alexa Fluor 350 or Alexa Fluor 488 conjugated [Molecular Probes] or Cy3 conjugated [Jackson ImmunoResearch]). To ensure that the intraarticular injection of FG was correctly located, some FG‐labeled sections from each injected rat were incubated with fluorescein isothiocyanate–conjugated isolectin B4 (10 μg/ml; Sigma) for 1 hour, as it has been described that isolectin B4 staining is absent or expressed at very low levels in DRG neurons innervating joints 18, 19, 20. All antibodies were prepared in phosphate buffered saline (PBS) with 0.5% normal goat serum and 0.1% Triton X‐100 (Sigma).

ate–conjugated isolectin B4 (10 μg/ml; Sigma) for 1 hour, as it has been described that isolectin B4 staining is absent or expressed at very low levels in DRG neurons innervating joints 18, 19, 20. All antibodies were prepared in phosphate buffered saline (PBS) with 0.5% normal goat serum and 0.1% Triton X‐100 (Sigma). Slides containing every sixth section of lumbar (L4 and L5) spinal cord were incubated overnight with a solution of the primary antibody rabbit anti–ionized calcium–binding adapter molecule 1 (anti–IBA‐1) (1:1,000; Wako) followed by incubation for 2 hours with a solution of the appropriate secondary antibody (Alexa Fluor 488 conjugated). All antibodies were prepared in PBS with 0.1% Triton X‐100. In control experiments, primary antibody was omitted from some sections of spinal cords and DRGs; this resulted in complete abolition of staining. Slides were coverslipped with Vectashield mounting medium (Vector), and images were captured using a Zeiss Axioplan 2 fluorescence microscope.

Slides containing every sixth section of lumbar (L4 and L5) spinal cord were incubated overnight with a solution of the primary antibody rabbit anti–ionized calcium–binding adapter molecule 1 (anti–IBA‐1) (1:1,000; Wako) followed by incubation for 2 hours with a solution of the appropriate secondary antibody (Alexa Fluor 488 conjugated). All antibodies were prepared in PBS with 0.1% Triton X‐100. In control experiments, primary antibody was omitted from some sections of spinal cords and DRGs; this resulted in complete abolition of staining. Slides were coverslipped with Vectashield mounting medium (Vector), and images were captured using a Zeiss Axioplan 2 fluorescence microscope. Immunofluorescence analysis FG‐labeled DRG neurons were easily identifiable and considered to be FG labeled when cell profiles were clearly distinguishable from the faint, homogeneous staining of unlabeled neurons. To study the entire neuronal population in DRGs, sections not labeled with FG were used and neurons were identified by the specific neuronal marker βIII Tubulin. The number of FG‐labeled or βIII Tubulin–labeled neurons immunoreactive for CGRP and p‐ERK was counted (8–12 sections per rat and 4 sections per rat, respectively) and expressed as the percentage of total FG‐labeled or βIII Tubulin–labeled neurons. In spinal cord sections, IBA‐1+ profiles (as indicative of the number of microglial cells) were counted within fixed areas (6 in total) of 20 × 104 μm2 on the lateral, central, and medial dorsal horns (3 sections per rat) 5, 21. Bilateral DRGs and dorsal horns were included in the analysis, and data were averaged for each rat.

l cord sections, IBA‐1+ profiles (as indicative of the number of microglial cells) were counted within fixed areas (6 in total) of 20 × 104 μm2 on the lateral, central, and medial dorsal horns (3 sections per rat) 5, 21. Bilateral DRGs and dorsal horns were included in the analysis, and data were averaged for each rat. Release of CGRP from dorsal horn slices On day 7 or day 18 postimmunization, horizontal lumbar dorsal horn slices (400 μm thick) with dorsal roots attached were obtained from the lumbar spinal cord of rats as previously described 22. One slice was obtained from each rat, mounted in the central compartment of a 3‐compartment chamber, and continuously superfused (1 ml/minute) with oxygenated Krebs solution (118 moles/liter NaCl, 4 moles/liter KCl, 1.2 moles/liter MgSO4, 1.2 moles/liter KH2PO4, 25 moles/liter NaHCO3, 2.5 moles/liter CaCl2, and 11 moles/liter glucose) containing 0.1% bovine serum albumin and 20 μg/ml bacitracin to minimize peptide degradation. The dorsal roots were placed in the lateral compartments and immersed in mineral oil to avoid dehydration. Before, during, and after electrical stimulation, 8‐ml fractions of superfusates were collected at 8‐minute intervals from the central compartment in glass tubes containing 0.1M acetic acid (VWR International). The release of CGRP was evoked by electrical stimulation of the dorsal roots for 8 minutes under conditions mimicking the transmission of noxious stimuli (C fiber strength; 20V, 0.5 msec, 10 Hz). Three 8‐ml fractions were collected before stimulation, one 8‐ml fraction was collected during electrical stimulation. and four 8‐ml recovery fractions were collected after stimulation.

he dorsal roots for 8 minutes under conditions mimicking the transmission of noxious stimuli (C fiber strength; 20V, 0.5 msec, 10 Hz). Three 8‐ml fractions were collected before stimulation, one 8‐ml fraction was collected during electrical stimulation. and four 8‐ml recovery fractions were collected after stimulation. To quantify CGRP content, superfusate samples were partially purified and desalted using Sep‐pak C18 reverse‐phase silica gel cartridges (Waters) previously conditioned with acetonitrile (VWR International) and 0.1% trifluoroacetic acid (TFA; VWR International). The peptide was then eluted from columns using a solution of acetonitrile/TFA (80:20). Eluates were dried by evaporation under nitrogen, reconstituted in 300 μl of sample buffer, and assayed for CGRP content by enzyme‐linked immunosorbent assay (standards 3.9–500 pg; SPI Bio). Data are expressed as the percentage of CGRP content in basal fractions. Statistical analysis Differences between values in behavioral tests were analyzed with two‐way repeated‐measures analysis of variance (ANOVA) followed by Tukey's test. Immunohistochemistry and release data were analyzed by two‐way ANOVA followed by Tukey's test or by Student's t‐test as appropriate. Data are reported as the mean ± SEM. Differences between means were considered statistically significant when P values were less than 0.05.

s of variance (ANOVA) followed by Tukey's test. Immunohistochemistry and release data were analyzed by two‐way ANOVA followed by Tukey's test or by Student's t‐test as appropriate. Data are reported as the mean ± SEM. Differences between means were considered statistically significant when P values were less than 0.05. RESULTS While both clinical scores and paw swelling were apparent beginning 12–14 days after collagen immunization (Figure 1A), hind paw mechanical thresholds decreased significantly on day 7 postimmunization, decreasing further concomitant with establishment of clinical signs (days 14–18) (Figure 1B). Hind paw thermal thresholds decreased alongside the appearance of clinical signs (days 14–18) (Figure 1B). Histologic examination of the ankle joint revealed infiltration of inflammatory cells on day 7, which was exacerbated by day 18 (Figures 1C and D). Figure 1 Collagen‐induced arthritis (CIA) is associated with development of hind paw inflammation and behavior indicating pain. A, Clinical scores and hind paw thickness. B, Hind paw mechanical and thermal thresholds. C and D, Photomicrographs of hematoxylin and eosin–stained ankle joints (C) and synovium (D). Values are the mean ± SEM (8–12 rats per group). * = P < 0.05; ** = P < 0.01; *** = P < 0.001 versus controls, by two‐way repeated‐measures analysis of variance followed by Tukey's test. Bars = 100 μm. PWT = paw withdrawal threshold; PWLT = paw withdrawal latency time.

d eosin–stained ankle joints (C) and synovium (D). Values are the mean ± SEM (8–12 rats per group). * = P < 0.05; ** = P < 0.01; *** = P < 0.001 versus controls, by two‐way repeated‐measures analysis of variance followed by Tukey's test. Bars = 100 μm. PWT = paw withdrawal threshold; PWLT = paw withdrawal latency time. We hypothesized that increasing recruitment of sensory afferent fibers, which would be activated at the site of inflammation in the ankle joints and the hind paws, might contribute to the escalation of mechanical and thermal hypersensitivity associated with CIA. The numbers of neuronal cell bodies from the ankle joint retrograde‐labeled with Fluoro‐Gold (FG+) were quantified in naive rats, control (CFA emulsion–injected) rats, and rats with CIA. In naive rats, ∼85% of retrograde‐labeled neuronal cell bodies from the ankle joint were in the L4 and L5 DRGs, and the remaining 15% were in L3 DRGs (Figures 2A and B). FG+ neurons represented a mean ± SEM of 2 ± 0.4% and 4 ± 0.3% of the total population of L4 and L5 neurons (n = 6 rats), respectively. In rats with CIA, the number of FG+ neurons in the L4 and L5 DRGs on days 4, 7, and 18 following collagen immunization was similar to the number in naive rats and control rats (Figure 2C). In DRGs from naive rats, 60% of FG+ neurons had small‐size cell bodies, while 29% had medium‐size cell bodies and 9% had large‐size cell bodies (Figure 2D). This cell size distribution was not altered by CIA (Figure 2D). Thus, the majority of sensory neurons innervating the ankle joint were derived from the L4 and L5 DRGs and represented a small group of DRG neurons that remained unaltered by the progression of CIA.

ell bodies and 9% had large‐size cell bodies (Figure 2D). This cell size distribution was not altered by CIA (Figure 2D). Thus, the majority of sensory neurons innervating the ankle joint were derived from the L4 and L5 DRGs and represented a small group of DRG neurons that remained unaltered by the progression of CIA. Figure 2 Development of collagen‐induced arthritis (CIA) induces no changes in numbers of ankle joint dorsal root ganglion (DRG) neurons retrograde‐labeled with Fluoro‐Gold (FG+). A, FG+ neurons detected in ipsilateral (right) but not contralateral (left) L5 naive rat DRGs. Bar = 100 μm. B, Distribution of FG+ neurons in naive rat L1–L6 DRGs (n = 6 rats). Circles represent individual values for each rat; horizontal lines indicate the mean. C and D, Number (C) and cell size distribution (D) of FG+ neurons in L4 and L5 DRGs during CIA development (n = 6 rats per group). Small size = 0–600 μm2; medium size = 600–1,200 μm2; large size = >1,200 μm2. Values are the mean ± SEM.

ircles represent individual values for each rat; horizontal lines indicate the mean. C and D, Number (C) and cell size distribution (D) of FG+ neurons in L4 and L5 DRGs during CIA development (n = 6 rats per group). Small size = 0–600 μm2; medium size = 600–1,200 μm2; large size = >1,200 μm2. Values are the mean ± SEM. To investigate further the population of neurons innervating the ankle joint and the impact of CIA on them, expression of CGRP and p‐ERK, as markers for nociceptive neurons and neuronal activity, respectively, was quantified in FG‐labeled neurons and in the entire L4 and L5 DRG neuronal population (innervating sites outside the joint). At the earliest time point (day 4), there was no difference in the number of FG+ neurons that expressed CGRP in DRGs from naive or control rats or rats with CIA (Figure 3A). By day 7 following collagen immunization, there was a significant increase in the number of FG+CGRP+ DRG neurons compared to controls (Figure 3A). This increase in CGRP expression remained significant on day 18 following injection of collagen (Figures 3A and B). When the DRG neuronal population was considered as a whole (sensory neurons innervating both ankle and non‐ankle joints), the number of CGRP+ neurons was unaltered on days 4 and 7 following collagen immunization (Figure 3C), but was significantly increased on day 18 compared to controls (Figures 3C and D). Nonpeptidergic isolectin B4–positive neurons represented only 4.1 ± 0.9% of FG+ neurons (mean ± SEM) in naive DRGs (n = 4 rats) and remained unaltered in DRGs obtained 4, 7, and 18 days after collagen immunization (5.1 ± 1.0%, 7.1 ± 2.0%, and 6.5 ± 1.3%, respectively; n = 4 rats per group).

controls (Figures 3C and D). Nonpeptidergic isolectin B4–positive neurons represented only 4.1 ± 0.9% of FG+ neurons (mean ± SEM) in naive DRGs (n = 4 rats) and remained unaltered in DRGs obtained 4, 7, and 18 days after collagen immunization (5.1 ± 1.0%, 7.1 ± 2.0%, and 6.5 ± 1.3%, respectively; n = 4 rats per group). Figure 3 Development of CIA is associated with an increase in calcitonin gene‐related peptide (CGRP)–expressing neurons in lumbar DRGs. A, Quantification of CGRP+FG+ L4 and L5 DRG neurons during CIA development (n = 4–6 rats per group). B, CGRP+ neurons in L5 DRGs retrograde‐labeled with FG in control rats and rats with CIA at 18 days. Arrows indicate FG+CGRP+ neurons; arrowheads indicate FG+ neurons; chevrons indicate CGRP+ neurons. C, Quantification of CGRP+ neurons in whole L4 and L5 DRGs during CIA development (n = 4 rats per group). D, CGRP+ neurons in control rats and rats with CIA at 18 days. Values are the mean ± SEM. ** = P < 0.01; *** = P < 0.001 versus controls, by two‐way analysis of variance followed by Tukey's test. Bars = 100 μm. See Figure 2 for other definitions.

ns in whole L4 and L5 DRGs during CIA development (n = 4 rats per group). D, CGRP+ neurons in control rats and rats with CIA at 18 days. Values are the mean ± SEM. ** = P < 0.01; *** = P < 0.001 versus controls, by two‐way analysis of variance followed by Tukey's test. Bars = 100 μm. See Figure 2 for other definitions. We also quantified the expression of p‐ERK in FG+ and FG+CGRP+ neurons. On day 4, there were no differences in p‐ERK expression in FG+ neurons among the 3 groups of rats (Figure 4A). By day 7, both FG+ and FG+CGRP+ neurons showed a significant increase in p‐ERK expression, which was further increased on day 18 following collagen immunization (Figures 4A and C). When the whole neuronal population in L4 and L5 DRGs was considered, there was a significant increase in p‐ERK expression in neurons on both day 7 and day 18 following collagen immunization (Figures 4B and D). However, p‐ERK expression in FG+CGRP+ neurons reached significance only on day 18 following collagen immunization compared to controls (Figure 4B).

L4 and L5 DRGs was considered, there was a significant increase in p‐ERK expression in neurons on both day 7 and day 18 following collagen immunization (Figures 4B and D). However, p‐ERK expression in FG+CGRP+ neurons reached significance only on day 18 following collagen immunization compared to controls (Figure 4B). Figure 4 CIA development is associated with increased p‐ERK in lumbar DRGs. A, Quantification of FG+p‐ERK+ neurons and FG+ calcitonin gene‐related peptide–positive (CGRP+) p‐ERK+ neurons in L4 and L5 DRGs during CIA development. B, Quantification of p‐ERK+ neurons and p‐ERK+CGRP+ neurons in the whole neuronal population in L4 and L5 DRGs during CIA development. In A and B, n = 4–6 rats per group. C, Phospho‐ERK+ and CGRP+ neurons in L5 DRGs retrograde‐labeled with FG in control rats and rats with CIA at 18 days. Arrows indicate FG+p‐ERK+CGRP+ neurons; arrowheads indicate FG+CGRP+ neurons; chevrons indicate CGRP+p‐ERK+ neurons. D, Phospho‐ERK+ and CGRP+ neurons in the whole neuronal population in L4 and L5 DRGs in control rats and rats with CIA at 18 days. Values are the mean ± SEM. ** = P < 0.01; *** = P < 0.001 versus controls, by two‐way analysis of variance followed by Tukey's test. Bars = 100 μm. See Figure 2 for other definitions.

rons. D, Phospho‐ERK+ and CGRP+ neurons in the whole neuronal population in L4 and L5 DRGs in control rats and rats with CIA at 18 days. Values are the mean ± SEM. ** = P < 0.01; *** = P < 0.001 versus controls, by two‐way analysis of variance followed by Tukey's test. Bars = 100 μm. See Figure 2 for other definitions. These data indicate that beginning 7 days from collagen immunization, CGRP‐expressing sensory neurons start to be activated in the ankle joint. Activation of ankle joint afferents is then followed by activation of neurons that innervate tissues outside the joints with progression of CIA. As CGRP is a marker of nociceptors, these data provide a correlation between sensory neuron phenotype and increased nocifensive behavior on both day 7 and day 18 of CIA.

oint. Activation of ankle joint afferents is then followed by activation of neurons that innervate tissues outside the joints with progression of CIA. As CGRP is a marker of nociceptors, these data provide a correlation between sensory neuron phenotype and increased nocifensive behavior on both day 7 and day 18 of CIA. In order to establish whether nociceptor activation in CIA resulted in increased input from the central terminals of these neurons in the spinal cord, we quantified the release of CGRP evoked by activity of primary afferent fibers. In the preparation of isolated dorsal horn with dorsal roots attached, obtained from control rats and from rats 7 and 18 days following collagen immunization, CGRP levels in dorsal horn superfusates were significantly increased following electrical stimulation of primary afferent fibers at noxious intensity, returning to basal levels after stimulation (Figure 5A). However, CGRP release in dorsal horns from rats with CIA at 18 days was significantly higher than that in dorsal horns from control rats at 18 days and from rats with CIA at 7 days (Figure 5A). Thus, despite the increase in retrograde‐labeled p‐ERK+CGRP+ neurons from the ankle of rats with CIA at 7 days, activity‐induced release of CGRP from their central terminals in the dorsal horn remained at control levels. Indeed, for enhanced activity‐evoked release of CGRP from the central terminals of nociceptors to occur, as was evident 18 days following induction of CIA, not only joint afferents but also neurons innervating tissues outside the joints must be recruited. In a further attempt to determine signs of enhanced central activity 7 days following collagen injection, we examined spinal microglia and observed significant microgliosis in dorsal horns from rats with CIA compared to those from control rats at 7 days (Figure 5B), with further enhancement in dorsal horns from rats with CIA at 18 days (Figure 5B).

ine signs of enhanced central activity 7 days following collagen injection, we examined spinal microglia and observed significant microgliosis in dorsal horns from rats with CIA compared to those from control rats at 7 days (Figure 5B), with further enhancement in dorsal horns from rats with CIA at 18 days (Figure 5B). Figure 5 Enhanced release of calcitonin gene‐related peptide (CGRP) and significant microgliosis in the dorsal horn during collagen‐induced arthritis (CIA). A, Activity‐evoked release of CGRP from primary afferent fibers in the dorsal horn of control rats at 18 days and rats with CIA at 7 days and 18 days (n = 7 rats per group). Mean ± SEM basal peptide content was 7.8 ± 0.27 pg/8 ml (n = 9 rats). R1–R4 = recovery fractions. B, Left and top, Ionized calcium–binding adapter molecule 1 (IBA‐1)–positive profiles in control rats at 18 days and in rats with CIA at 7 days and 18 days. Bottom right, Quantification of IBA‐1+ profiles in the dorsal horn of control rats and rats with CIA during CIA development (n = 4–6 rats per group). Values are the mean ± SEM. * = P < 0.05; ** = P < 0.01; *** = P < 0.001 versus controls, by two‐way analysis of variance followed by Tukey's test. ### = P < 0.001 versus basal levels, by two‐way analysis of variance followed by Tukey's test. Bar = 100 μm.

d rats with CIA during CIA development (n = 4–6 rats per group). Values are the mean ± SEM. * = P < 0.05; ** = P < 0.01; *** = P < 0.001 versus controls, by two‐way analysis of variance followed by Tukey's test. ### = P < 0.001 versus basal levels, by two‐way analysis of variance followed by Tukey's test. Bar = 100 μm. Finally, in order to evaluate whether increased CGRP release from primary afferent fibers in the dorsal horn contributed to mechanical hypersensitivity and microgliosis, we intrathecally delivered a CGRP antagonist between days 11 and 18 of CIA. CGRP8–37 significantly attenuated mechanical hypersensitivity (Figure 6A) but not thermal hypersensitivity (Figure 6B) and produced a significant attenuation of microgliosis in the dorsal horn (Figure 6C). Figure 6 Intrathecal (IT) administration of a CGRP antagonist attenuates established mechanical hypersensitivity in CIA. A and B, Established mechanical hypersensitivity (A) but not heat hypersensitivity (B) attenuated by 7‐day intrathecal administration of CGRP8–37 (24 μg/day) (n = 9 rats per group). C, Quantification of IBA‐1+ profiles in the L4 and L5 dorsal horns at 18 days (n = 4 rats per group). Values are the mean ± SEM. ** = P < 0.01; *** = P < 0.001 versus vehicle (Veh)–treated controls, by two‐way repeated‐measures analysis of variance followed by Tukey's test (A and B) or by Student's t‐test (C). PWT = paw withdrawal threshold; PWLT = paw withdrawal latency time (see Figure 5 for other definitions).

oup). Values are the mean ± SEM. ** = P < 0.01; *** = P < 0.001 versus vehicle (Veh)–treated controls, by two‐way repeated‐measures analysis of variance followed by Tukey's test (A and B) or by Student's t‐test (C). PWT = paw withdrawal threshold; PWLT = paw withdrawal latency time (see Figure 5 for other definitions). DISCUSSION The main findings of this study are that before the appearance of overt clinical signs of CIA, mechanical hypersensitivity (allodynia) of the hind paw develops concomitant with inflammatory cell infiltration in the joint, activation of ankle joint nociceptors, and mild spinal microgliosis, but not paw swelling. However, with establishment of the disease, paw swelling and thermal hypersensitivity (hyperalgesia) accompany mechanical allodynia and, besides articular nociceptors, a significant number of primary afferent neurons innervating tissues outside the joint are activated. Furthermore, central changes in the dorsal horn of the spinal cord are readily detectable, including significant microgliosis and enhanced release of the pronociceptive peptide CGRP from nociceptor central terminals. Such release of CGRP is functionally relevant, as the delivery of a CGRP receptor antagonist to the lumbar spinal cord attenuates both mechanical allodynia and spinal microgliosis. Thus, similar to other forms of chronic pain 23, allodynia in CIA is likely to be maintained by a combination of central and peripheral mechanisms.

ease of CGRP is functionally relevant, as the delivery of a CGRP receptor antagonist to the lumbar spinal cord attenuates both mechanical allodynia and spinal microgliosis. Thus, similar to other forms of chronic pain 23, allodynia in CIA is likely to be maintained by a combination of central and peripheral mechanisms. In the present study and in a previous study 5, we observed the development of allodynia before the onset of clinical signs of CIA in Lewis rats. Similarly, QB mice displayed allodynia before signs of arthritis in collagen antibody–induced arthritis 24. However, in DBA/1 mice, CIA allodynia developed along with the onset of clinical signs, although that study largely assessed pain in association with inflammation 4. These discrepancies highlight the importance of considering species and strain differences when selecting the most appropriate RA model for pain studies.

However, in DBA/1 mice, CIA allodynia developed along with the onset of clinical signs, although that study largely assessed pain in association with inflammation 4. These discrepancies highlight the importance of considering species and strain differences when selecting the most appropriate RA model for pain studies. We observed that a small percentage of sensory neurons innervated the rat ankle joint, and their cell bodies were located in the L4 and L5 DRGs, consistent with previous findings in mice 25. These DRGs also contain neurons innervating additional sites including skin and muscle of the hind paw and the knee joint 18, 26, 27. The number of sensory neurons innervating the ankle joint was not altered by inflammation at any stage of CIA, but the expression of both the nociceptive peptide CGRP and the activation marker p‐ERK increased with the severity of disease, demonstrating that CIA is associated not only with the activation of these sensory neurons in the ankle joint, but also with the activation of neurons innervating areas outside this joint. For instance, the synovium of the ankle joint is highly innervated by CGRP‐expressing fibers, with enhanced expression of CGRP reported in both the joint and DRGs following adjuvant‐induced arthritis 28. Furthermore, as phosphorylation of ERK occurs in sensory neurons following peripheral noxious stimulation, the activation of this intracellular pathway is indicative of nociceptive neuron activation 29, 30. The increase in nociceptive sensory neuron activation, together with the development of allodynia and the exacerbation of the disease reported herein, provide a mechanism that links pain and inflammation. However, given that nociceptive sensory neuron activation and the development of allodynia occurred in the absence of paw swelling, our data also suggest that pain hypersensitivity in the early phase of CIA precedes macroscopic inflammation.

disease reported herein, provide a mechanism that links pain and inflammation. However, given that nociceptive sensory neuron activation and the development of allodynia occurred in the absence of paw swelling, our data also suggest that pain hypersensitivity in the early phase of CIA precedes macroscopic inflammation. Joint afferents are highly responsive to mechanical stimuli 6; thus, sensitization of these fibers likely drives the central changes that underlie the early development of mechanical allodynia during CIA. Nevertheless, even though sensory neurons up‐regulated nociceptive peptide expression and displayed the presence of activation markers in the early development of allodynia, their activity was not sufficient to detect increased CGRP peptide release from their central terminals in the spinal dorsal horn. It is possible that the sensitivity of our assay does not allow for detection of small changes in CGRP content that might have occurred. In any case, alongside CGRP, the occurrence of glutamate and SP release from the central terminals of primary afferent fibers in the dorsal horn may contribute to behavioral allodynia 7. In order for primary afferent enhanced input to become detectable in our experimental system, the disease had to reach advanced phases with evident inflammation and swelling in the joints. Microglia appeared to be a more sensitive indicator of spinal cord changes in CIA than increased primary afferent input, as microgliosis was detected concomitant with the manifestation of allodynia and before the onset of clinical signs.

had to reach advanced phases with evident inflammation and swelling in the joints. Microglia appeared to be a more sensitive indicator of spinal cord changes in CIA than increased primary afferent input, as microgliosis was detected concomitant with the manifestation of allodynia and before the onset of clinical signs. Our data suggest that pain development in CIA involves not only inflammation and changes in the joint, but also alterations in central pain processing that are dependent on pathologic inflammatory mechanisms. Thus, subtle collagen‐induced changes in the joint cartilage lead to an inflammatory milieu, including enhanced cytokine and prostaglandin release, which results in the sensitization of sensory neurons locally and mild microgliosis in the spinal cord during the early phase of CIA allodynia. Spinal microgliosis gradually increases with escalating activation of nociceptive neurons by inflammatory mediators in the joint with CIA. These mechanistic data are particularly relevant in the context of CIA as a model of RA, in view of the suggestion that activation of central interleukin‐1 signaling contributes to fatigue in RA patients in whom the most likely source of this cytokine are the microglia 31. Peripheral and central changes grow in both incidence and significance in the later phases of CIA allodynia and hyperalgesia, which are characterized by an evident increase of noxious input from the periphery to the spinal cord. In CIA, central changes such as microgliosis and increased afferent input play a critical role in mediating allodynia as demonstrated by the analgesic effect of a centrally delivered CGRP receptor antagonist. Nevertheless, the small though significant microglial inhibition through putative microglial CGRP receptors 32, 33 is not likely to be the primary mechanism of this antagonist, and attenuation of microgliosis is probably a reflection of neuronal receptor blockade 34, 35.

a centrally delivered CGRP receptor antagonist. Nevertheless, the small though significant microglial inhibition through putative microglial CGRP receptors 32, 33 is not likely to be the primary mechanism of this antagonist, and attenuation of microgliosis is probably a reflection of neuronal receptor blockade 34, 35. While we observed no changes in paw swelling following CGRP antagonist treatment (data not shown), spinal mechanisms can regulate the extent of peripheral inflammation in arthritis models. For example, spinal manipulation of microglial signaling by inhibiting p38 MAPK reduces joint inflammation and disease progression 36, with levels of spinal cytokines such as tumor necrosis factor α (TNFα) regulating neuronal responses to mechanical stimulation of the inflamed joint 37. It is relevant that neutralization of TNFα in RA patients inhibits pain before reducing inflammation in the joint 38, possibly through inhibition of central sensitization. In our study, the mechanisms underlying CIA thermal hyperalgesia, which develops concomitant with the onset of swelling and is unaltered by a centrally delivered CGRP antagonist, remain to be established. These mechanisms may reside peripherally and may arise from activation of different subsets of peripheral and spinal neurons 39. Indeed, while changes in mechanical sensitivity are a major feature of central sensitization, peripheral sensitization plays a major role in the alteration of heat sensation 13.

established. These mechanisms may reside peripherally and may arise from activation of different subsets of peripheral and spinal neurons 39. Indeed, while changes in mechanical sensitivity are a major feature of central sensitization, peripheral sensitization plays a major role in the alteration of heat sensation 13. In conclusion, while RA results from immune‐mediated inflammation and damage to the joint, pain in RA may not simply be the result of joint inflammation. Indeed, pain may be experienced by patients before clinical signs permit confirmation of the diagnosis 1. We suggest that pain therapy in RA can be improved by developing treatment strategies that take into account the association of NSAIDs with molecules that stop central changes, for example, inhibitors of microglial cell reactivity and CGRP receptor antagonists or antibodies. Most of the currently available CGRP antagonists, which were developed for migraine, do not readily enter the central nervous system (CNS) 40. However, whether CNS penetration is critical for their antimigraine effect is still under debate 40. Our study suggests that CNS‐penetrating CGRP antagonists, when available, may be effective analgesics in RA. AUTHOR CONTRIBUTIONS 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. Malcangio 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.

hors 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. Malcangio 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. Chapman, Malcangio. Acquisition of data. Nieto, Clark, Grist. Analysis and interpretation of data. Nieto, Clark, Malcangio. ACKNOWLEDGMENTS We wish to thank Thomas Pitcher and Carl Hobbs for technical support.

Osteoarthritis (OA) is a major cause of long-term disability (1). Traditionally characterized by structural degradation of one or more synovial joints, OA is typically associated with pain, swelling, and stiffness (2). Reduced manual dexterity and inability to perform tasks due to symptoms affect quality of life and ability to function independently (3). Levels of pain and disability experienced by patients often do not correspond to the level of tissue damage observed in the affected joints (4), suggesting involvement of central nervous system mechanisms in the generation of OA-related pain (5). Given the prevalence of OA and an increasingly aging population, there is a pressing need for more effective treatments (6,7). Many commonly prescribed medications are ill suited for long-term use due to side effects including tolerance, dependency, and gastrointestinal complications (8,9). While pain is always a personal experience and subject to multiple psychosocial factors, there has long been a desire to complement subjective reports with reliable markers of pain perception and treatment effectiveness (10,11).

to side effects including tolerance, dependency, and gastrointestinal complications (8,9). While pain is always a personal experience and subject to multiple psychosocial factors, there has long been a desire to complement subjective reports with reliable markers of pain perception and treatment effectiveness (10,11). Neuroimaging methodologies offer the potential to meet these desires (12). Functional magnetic resonance imaging (fMRI) provides a safe, noninvasive, repeatable indirect measure of neural activity suitable for use in crossover trial designs (12). Reports have suggested that fMRI could demonstrate analgesic effects in populations of patients with clinical pain, adding value to conventional behavioral end points and facilitating decision-making early in drug development, reducing risk, development times, and cost (12). Experimental fMRI studies have provided useful insights into the neural mechanisms of drug action (for review, see ref.(13)). In order to fulfill this potential, fMRI must depict the neural correlates of the response to analgesics in patients experiencing persistent pain. Preliminary studies in OA have shown promise in establishing neural correlates of response to analgesics (14,15), but to date these studies have been underpowered or lacked appropriate placebo control. In this study we investigated the potential of fMRI as a “fit-for-purpose” measurement technique in development of analgesics for persistent pain.

ve shown promise in establishing neural correlates of response to analgesics (14,15), but to date these studies have been underpowered or lacked appropriate placebo control. In this study we investigated the potential of fMRI as a “fit-for-purpose” measurement technique in development of analgesics for persistent pain. We sought to determine the ability of fMRI to detect changes in brain functioning following a 1-week course of naproxen in individuals with OA of the first carpometacarpal (CMC) joint. Naproxen is an established, effective nonsteroidal antiinflammatory drug (NSAID) labeled for treatment of OA-related symptoms in the UK (http://www.emc.medicines.org.uk). We aimed to detect naproxen-induced changes in brain activity using a functionally relevant task that evoked pain. We hypothesized that task-evoked brain activity would be reduced following administration of naproxen, compared to placebo. In particular, we hypothesized that a network of brain regions previously associated with pain perception would display significant reductions in activity following naproxen administration, during the task load that induced the highest rating of pain intensity. Patients and Methods Design This study was a placebo-controlled, double-blind, 2-period crossover design undertaken in participants with OA of the first CMC joint of the right hand. Ethics Ethical approval was obtained from The Joint South London and Maudsley and The Institute of Psychiatry NHS Research Ethics Committee (reference no. 10/H0807/10). Written informed consent was obtained from all participants.

Patients and Methods Design This study was a placebo-controlled, double-blind, 2-period crossover design undertaken in participants with OA of the first CMC joint of the right hand. Ethics Ethical approval was obtained from The Joint South London and Maudsley and The Institute of Psychiatry NHS Research Ethics Committee (reference no. 10/H0807/10). Written informed consent was obtained from all participants. Participants Participants were recruited from multiple sources, including national radio, local and national print magazines, a university e-mail circular, and referrals from hospital rheumatology and physiotherapy departments. All participants visited on 4 occasions: screening, familiarization, and 2 scanning sessions (1 following each treatment period). All had a confirmed diagnosis of OA of the first CMC joint of the right hand according to American College of Rheumatology criteria (16) and were right-hand dominant with a background pain intensity score of ≥4 on an 11-point (0–10) numerical rating scale (NRS) at screening or randomization. Normal exclusion criteria for MRI applied (e.g., height and weight criteria, presence of internal metal, claustrophobia). Several other major exclusion criteria were also applied, including severe pain elsewhere in the body that might impair the assessment of OA-related pain, history of other severe acute or chronic medical or psychiatric conditions, previous nonresponse to naproxen for pain relief, increased risk of adverse reactions to NSAIDs, inability to conform to lifestyle guidelines, and use of prohibited medications (including analgesics other than NSAIDs, paracetamol, or codeine) prior to and during study participation. Detailed inclusion/exclusion criteria as well as screening, blinding, and randomization information are provided in Supplementary Materials, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38987/abstract.

acetamol, or codeine) prior to and during study participation. Detailed inclusion/exclusion criteria as well as screening, blinding, and randomization information are provided in Supplementary Materials, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38987/abstract. A total of 189 participants (131 women, 58 men) completed a preliminary telephone interview; 36 met the inclusion criteria for screening, and 23 were randomized into the study. Four participants were excluded after randomization, due to radiologic abnormalities, drug-related adverse reactions, and an MRI hardware failure. Nineteen participants (18 women, 1 man) ages 50–80 years (mean ± SD 60.72 ± 6.44 years) were included in the final analysis set. All participants who completed the study reported compliance with the dosing regimen (Figure 1). (Approximately 30% of all prospects for the study were male, but only 18% were screened. Exclusions in men were predominantly health related, for example, severe pain elsewhere in the body, other types of hand pain [e.g., repetitive strain disorder/trigger finger], contraindicating medical/surgical histories, head trauma, and lifestyle guideline exclusions.) Figure 1 Overview of study procedures and experimental design.

A total of 189 participants (131 women, 58 men) completed a preliminary telephone interview; 36 met the inclusion criteria for screening, and 23 were randomized into the study. Four participants were excluded after randomization, due to radiologic abnormalities, drug-related adverse reactions, and an MRI hardware failure. Nineteen participants (18 women, 1 man) ages 50–80 years (mean ± SD 60.72 ± 6.44 years) were included in the final analysis set. All participants who completed the study reported compliance with the dosing regimen (Figure 1). (Approximately 30% of all prospects for the study were male, but only 18% were screened. Exclusions in men were predominantly health related, for example, severe pain elsewhere in the body, other types of hand pain [e.g., repetitive strain disorder/trigger finger], contraindicating medical/surgical histories, head trauma, and lifestyle guideline exclusions.) Figure 1 Overview of study procedures and experimental design. Procedure All participants completed psychometric and pain-related questionnaires at each session, including the short form McGill Pain Questionnaire (MPQ) (17), the Beck Depression Inventory II (BDI-II) (18), the Spielberger State-Trait Anxiety Inventory (STAI) (19), and the Patient-Rated Wrist and Hand Evaluation (PRWHE) (20).

All participants completed psychometric and pain-related questionnaires at each session, including the short form McGill Pain Questionnaire (MPQ) (17), the Beck Depression Inventory II (BDI-II) (18), the Spielberger State-Trait Anxiety Inventory (STAI) (19), and the Patient-Rated Wrist and Hand Evaluation (PRWHE) (20). Screening At the initial session patients underwent a medical and comprehensive blood screening, and medication history was ascertained. Urine drug screen and alcohol breath tests were administered at every session to ensure no confounding drug or alcohol use. Participants also underwent a clinical hand examination by a specialist physiotherapist. Familiarization During the familiarization and measurement sessions, each participant underwent a “mock scan” in a simulated MRI environment designed to reduce feelings of anxiety and claustrophobia. During the mock scan, participants received training on the task they would perform during actual fMRI sessions, including a complete “trial run” and participant-specific calibration of the task.

articipant underwent a “mock scan” in a simulated MRI environment designed to reduce feelings of anxiety and claustrophobia. During the mock scan, participants received training on the task they would perform during actual fMRI sessions, including a complete “trial run” and participant-specific calibration of the task. Medication dosing All participants underwent down-titration of existing analgesic medications over a period of up to 28 days. Following down-titration, all participants received a 14-day paper diary in which to record perceived pain intensity every morning and evening using an 11-point (0–10) NRS and to record any observations or changes relating to their hands or general health. In the first 7 days, all participants were administered placebo. In the remaining 7 days, participants were randomly administered either naproxen (500 mg twice a day) or matched placebo. Following the first fMRI session, all participants received a further 7 days of placebo and then either naproxen or placebo (crossed over from period 1) as their second period of dosing. Naproxen has well-established pharmacokinetics, with peak plasma concentration 2–4 hours after ingestion and a plasma half-life of 12–15 hours (http://www.emc.medicines.org.uk).

ipants received a further 7 days of placebo and then either naproxen or placebo (crossed over from period 1) as their second period of dosing. Naproxen has well-established pharmacokinetics, with peak plasma concentration 2–4 hours after ingestion and a plasma half-life of 12–15 hours (http://www.emc.medicines.org.uk). Functional MRI scanning sessions and followup The fMRI task was designed to evoke OA-related pain (see below). After completion of the scanning session, medication containers and pain diaries were given for the second dosing period. The final fMRI scanning session was identical to the first, except that no further treatment was dispensed. Six additional telephone checks were performed during the study to ensure well-being and compliance, and a followup telephone interview was conducted ∼2 weeks after the final fMRI session.

e second dosing period. The final fMRI scanning session was identical to the first, except that no further treatment was dispensed. Six additional telephone checks were performed during the study to ensure well-being and compliance, and a followup telephone interview was conducted ∼2 weeks after the final fMRI session. Task The task involved squeezing a key-shaped pressure device between the right thumb and index finger (lateral pinch) of the affected hand, designed to mimic everyday activities difficult for individuals with OA hand pain, such as gripping or handling small objects (3). The lateral pinch device was an MRI-compatible pressure transducer, to which participants applied varying levels of pressure. The device was calibrated for each participant during the familiarization session to determine the maximum voluntary contraction (MVC) for each individual, measured in kilopascals. MVC was defined as the average of 3 lateral pinches at each participant's maximum effort. During the 23-minute event-related fMRI task, participants were cued to perform the lateral pinch (squeeze) at each of 3 different levels (10%, 40%, and 70%) of their MVC. The target force to be applied and visual feedback of the participant's own applied force were provided via a visual target and moving vertical line on a custom computerized scale projected onto a screen located at the participant's feet.

l pinch (squeeze) at each of 3 different levels (10%, 40%, and 70%) of their MVC. The target force to be applied and visual feedback of the participant's own applied force were provided via a visual target and moving vertical line on a custom computerized scale projected onto a screen located at the participant's feet. In each trial, participants were instructed to squeeze the device until they reached the displayed target and then perform an isometric hold for 1 second. The time required to reach the desired effort was recorded to account for variability in motor task duration. Each trial cycle lasted for 24 seconds. Stimulus onsets were jittered within the first 6 seconds of each trial. In addition, 15 (5 per level of MVC) computerized, horizontal visual analog scale (VAS) assessments of pain intensity were presented pseudorandomly throughout the experiment, during which participants rated their perceived pain intensity during the last-performed squeeze. Participants performed VAS assessments using a joystick in their left hand, anchored at the left with “no pain” and at the right with “worst pain imaginable.” The VAS was displayed for 9 seconds to allow adequate response time, followed by a 12-second intertrial interval. In total, 60 trials were presented in each experimental run—15 per percentage of MVC and 15 VAS ratings. Illustrations of the pinch device, target display, and computerized VAS are provided in Supplementary Figure 1, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38987/abstract.

trials were presented in each experimental run—15 per percentage of MVC and 15 VAS ratings. Illustrations of the pinch device, target display, and computerized VAS are provided in Supplementary Figure 1, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38987/abstract. Behavioral data analysis Pain diary and psychometric data were analyzed using SPSS version 19. Differences between treatments on the PRWHE, STAI, BDI-II, and short form MPQ were assessed using paired t-tests. In-scanner behavioral analyses In-scanner VAS ratings were analyzed using repeated-measures analysis of variance, with MVC level (10%/40%/70%) and treatment (naproxen/placebo) as factors.

Behavioral data analysis Pain diary and psychometric data were analyzed using SPSS version 19. Differences between treatments on the PRWHE, STAI, BDI-II, and short form MPQ were assessed using paired t-tests. In-scanner behavioral analyses In-scanner VAS ratings were analyzed using repeated-measures analysis of variance, with MVC level (10%/40%/70%) and treatment (naproxen/placebo) as factors. MRI data acquisition MRI data were acquired on a 3T Signa HDx whole-body scanner (General Electric) fitted with an 8-channel, phased-array receive-only head coil. A total of 465 whole-brain T2*-weighted interleaved axial volumes were acquired using echo-planar imaging. A high-resolution T1-weighted 3-dimensional (3-D) spoiled gradient-recalled acquisition in the steady state (SPGR) sequence was also acquired for interparticipant registration. Finally, we sought to investigate whether treatment-related variability in the blood oxygen level–dependent (BOLD) response (21) might be explained by its relationship to underlying changes in global cerebral blood flow (CBF)—a possibility given the modulatory effect of naproxen on prostaglandin production (22), which in turn modulates vascular tone, blood pressure, and global CBF. We assessed global CBF using a pulsed continuous arterial spin–labeled (pCASL) perfusion MRI sequence acquired immediately prior to fMRI at each session (23,24). Acquisition order and parameters were identical in both scanning sessions. Full details regarding all sequence parameters are provided in Supplementary Materials, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38987/abstract.

y prior to fMRI at each session (23,24). Acquisition order and parameters were identical in both scanning sessions. Full details regarding all sequence parameters are provided in Supplementary Materials, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38987/abstract. Functional MRI data preprocessing Preprocessing was conducted using Statistical Parametric Mapping software (SPM8) (http://www.fil.ion.ucl.ac.uk/spm/software/spm8/). Preprocessing included motion correction, coregistration of the 3-D SPGR image to the mean functional image, segmentation, and normalization to Montreal Neurological Institute (MNI152) template space. Spatial smoothing using a full-width half-maximum Gaussian kernel of 8 mm, interleaved slice timing correction, and high-pass temporal filtering (100 seconds) were performed using FMRI Expert Analysis Tool (FEAT) version 5.98 (http://www.fmrib.ox.ac.uk/fsl).

n to Montreal Neurological Institute (MNI152) template space. Spatial smoothing using a full-width half-maximum Gaussian kernel of 8 mm, interleaved slice timing correction, and high-pass temporal filtering (100 seconds) were performed using FMRI Expert Analysis Tool (FEAT) version 5.98 (http://www.fmrib.ox.ac.uk/fsl). Functional MRI statistical analysis Statistical analysis of fMRI data was carried out using FEAT. A generalized linear model for each participant and session was constructed to perform the voxelwise time series statistical analysis using FSL-FILM. Event timings at 10%, 40%, and 70% MVC were convolved with a double-gamma hemodynamic response function and its temporal derivative. Temporal autocorrelation correction was used (25), and estimated motion parameters were included as confound regressors. Individual participant first-level contrast of parameter estimates (COPE) images, under both naproxen and placebo conditions, were derived for 10%, 40%, and 70% MVC. An intermediate second-level fixed-effects analysis was performed to average the 10%, 40%, and 70% MVC COPE images to take forward for higher-level analysis.

ividual participant first-level contrast of parameter estimates (COPE) images, under both naproxen and placebo conditions, were derived for 10%, 40%, and 70% MVC. An intermediate second-level fixed-effects analysis was performed to average the 10%, 40%, and 70% MVC COPE images to take forward for higher-level analysis. Higher-level group fMRI statistical analysis Higher-level analysis was carried out using FSL-FLAME stage 1 with robust outlier detection (26). A whole-brain mixed-effects analysis was performed, using individual participants' COPE images derived from the second level as inputs, to compute voxelwise paired t-tests comparing differences in evoked brain activity between naproxen and placebo conditions. An additional confound regressor for medication administration order was included in the model. Z (Gaussianized T/F) statistic images were thresholded using clusters determined by Z >2.3 and a (Gaussian random field–corrected) cluster significance threshold of P = 0.05 (27). Global CBF analysis A measure of global CBF in each session was defined as the mean of all gray matter voxels within the 3-D pCASL volume. Differences in global CBF between naproxen and placebo sessions were assessed using a paired t-test in SPSS version 19. The significance threshold was set at P < 0.05.

Higher-level group fMRI statistical analysis Higher-level analysis was carried out using FSL-FLAME stage 1 with robust outlier detection (26). A whole-brain mixed-effects analysis was performed, using individual participants' COPE images derived from the second level as inputs, to compute voxelwise paired t-tests comparing differences in evoked brain activity between naproxen and placebo conditions. An additional confound regressor for medication administration order was included in the model. Z (Gaussianized T/F) statistic images were thresholded using clusters determined by Z >2.3 and a (Gaussian random field–corrected) cluster significance threshold of P = 0.05 (27). Global CBF analysis A measure of global CBF in each session was defined as the mean of all gray matter voxels within the 3-D pCASL volume. Differences in global CBF between naproxen and placebo sessions were assessed using a paired t-test in SPSS version 19. The significance threshold was set at P < 0.05. Region of interest (ROI) analyses Anatomic ROIs in MNI space were derived from Harvard-Oxford Cortical/Subcortical and Juelich Histological probabilistic atlases available within FSL. Based on a priori information regarding brain activation related to pain (28), ROIs were created for the thalamus, primary and secondary somatosensory cortices, hippocampal formation, anterior and posterior insula, anterior and posterior cingulate cortices, and amygdala in both hemispheres. Probabilistic ROIs were thresholded to include voxels more than 20% likely to represent their indicated anatomic location, and were then binarized to become mask images.

tosensory cortices, hippocampal formation, anterior and posterior insula, anterior and posterior cingulate cortices, and amygdala in both hemispheres. Probabilistic ROIs were thresholded to include voxels more than 20% likely to represent their indicated anatomic location, and were then binarized to become mask images. For each participant, the mean value across all voxels within the ROI mask was computed as a summary measure of the BOLD response under naproxen and placebo conditions. Values were extracted from first-level 70% MVC COPE images, as we hypothesized that this would induce the highest reported levels of pain intensity. For each ROI hemisphere, summary measure data were analyzed using a paired t-test in SPSS version 19. Right and left hemisphere data were analyzed separately as their variances differed between hemispheres. Mean values for each treatment and corresponding standard errors were plotted for ROIs demonstrating significant treatment differences. (A supplementary voxelwise analysis of treatment effects using the 70% MVC class only is provided as part of Supplementary Figure 2, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38987/abstract.)

ere plotted for ROIs demonstrating significant treatment differences. (A supplementary voxelwise analysis of treatment effects using the 70% MVC class only is provided as part of Supplementary Figure 2, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38987/abstract.) Pearson product-moment correlation coefficients were computed in SPSS version 19 to assess the relationship between treatment-induced changes in fMRI squeeze task–related VAS pain intensity and treatment-induced changes in mean BOLD responses in selected bilateral ROIs (thalamus, primary and secondary somatosensory cortices, anterior and posterior insula) previously reported to relate to the intensity of perceived pain (29,30). Treatment differences (Δ = naproxen − placebo) were derived for each participant. The significance threshold for all ROI analyses was set at P < 0.05. Due to the exploratory nature of these analyses, we did not correct for multiple comparisons. Post hoc sample size calculations to detect significant treatment differences (alpha level of 0.05, 2-tailed, power of 80%) in each ROI in each hemisphere were derived using G*Power version 3.12 (31).

yses was set at P < 0.05. Due to the exploratory nature of these analyses, we did not correct for multiple comparisons. Post hoc sample size calculations to detect significant treatment differences (alpha level of 0.05, 2-tailed, power of 80%) in each ROI in each hemisphere were derived using G*Power version 3.12 (31). Results Behavioral data (psychometry and pain diaries) As determined using an NRS, daily pain intensity after 7 days of treatment with naproxen was significantly reduced compared to treatment with placebo (t[18] = −3.2, P = 0.005), confirming the known analgesic properties of naproxen (Figure 2A). The VAS, present pain intensity (PPI), and sensory subsections of the short form MPQ demonstrated significant reductions after administration of naproxen compared to placebo (VAS rating t[18] = −4.34, P < 0.001; PPI rating t[18] = −3.31, P = 0.004; sensory rating t[18] = −3.34, P = 0.004). No differences were observed for affective ratings on the short form MPQ (t[18] = −1.42, P = 0.17). Measures of wrist and hand pain and function (the PRWHE) were significantly improved following administration of naproxen compared to placebo (t[18] = −2.08, P = 0.05). No significant differences between treatments were observed for state anxiety (the STAI-S) (t[18] = −1.61, P = 0.12) or depression (the BDI-II) (t[18] = −0.77, P = 0.45), both of which were in the normal range. Psychometric data are summarized in Table1.

ration of naproxen compared to placebo (t[18] = −2.08, P = 0.05). No significant differences between treatments were observed for state anxiety (the STAI-S) (t[18] = −1.61, P = 0.12) or depression (the BDI-II) (t[18] = −0.77, P = 0.45), both of which were in the normal range. Psychometric data are summarized in Table1. Figure 2 A, Mean ± SEM daily pain intensity determined using a numerical rating scale (NRS) in patients receiving active treatment (light gray; placebo on days 1–7, naproxen on days 8–14) compared to patients receiving placebo (dark gray; placebo on days 1–14). B, Mean ± SEM visual analog scale (VAS) scores during the lateral pinch task. During the scanning visits, participants with painful osteoarthritis of the thumb were cued to perform a lateral pinch (squeeze) at each of 3 different levels (10%, 40%, and 70%) of their maximum voluntary contraction (MVC). Table 1 Summary of demographic characteristics and pain and psychometric assessments in the 19 study patients

Figure 2 A, Mean ± SEM daily pain intensity determined using a numerical rating scale (NRS) in patients receiving active treatment (light gray; placebo on days 1–7, naproxen on days 8–14) compared to patients receiving placebo (dark gray; placebo on days 1–14). B, Mean ± SEM visual analog scale (VAS) scores during the lateral pinch task. During the scanning visits, participants with painful osteoarthritis of the thumb were cued to perform a lateral pinch (squeeze) at each of 3 different levels (10%, 40%, and 70%) of their maximum voluntary contraction (MVC). Table 1 Summary of demographic characteristics and pain and psychometric assessments in the 19 study patients Variable, session* Range Mean ± SD t(18) P† Age, years 52.00–72.00 60.72 ± 6.44 – – Pain duration, years 0.50–17.00 4.03 ± 4.05 – – Baseline pain, 0–10 – – Screening 2.00–7.00 4.53 ± 1.43 Familiarization 1.00–8.00 4.21 ± 1.81 MPQ VAS rating −4.34 <0.001 Naproxen 9.00–57.00 23.68 ± 11.6 Placebo 8.00–82.00 39.47 ± 20.12 MPQ PPI rating −3.31 0.004 Naproxen 0.00–2.00 0.89 ± 0.65 Placebo 1.00–2.00 1.52 ± 0.51 MPQ sensory rating −3.34 0.004 Naproxen 1.00–17.00 5.73 ± 3.94 Placebo 3.00–30.00 9.68 ± 7.11 MPQ affective rating −1.42 0.17 Naproxen 0.00–3.00 0.21 ± 0.71 Placebo 0.00–7.00 0.73 ± 1.72 PRWHE −2.08 0.05 Familiarization 11.50–68.00 42.39 ± 15.45 Naproxen 5.50–65.50 28.97 ± 15.08 Placebo 14.00–73.50 35.92 ± 17.56 Spielberger STAI state anxiety rating −1.61 0.12 Naproxen 36.00–80.00 47 ± 11.36 Placebo 36.00–84.00 50.52 ± 13.14 BDI-II depression rating −0.77 0.45 Screening 0.00–23.00 7.05 ± 6.41 Naproxen 0.00–22.00 6.05 ± 5.93 Placebo 0.00–26.00 7.05 ± 6.6 * MPQ = McGill Pain Questionnaire; VAS = visual analog scale; PPI = present pain intensity; PRWHE = Patient-Rated Wrist and Hand Evaluation; STAI = State-Trait Anxiety Inventory; BDI-II = Beck Depression Inventory II.

77 0.45 Screening 0.00–23.00 7.05 ± 6.41 Naproxen 0.00–22.00 6.05 ± 5.93 Placebo 0.00–26.00 7.05 ± 6.6 * MPQ = McGill Pain Questionnaire; VAS = visual analog scale; PPI = present pain intensity; PRWHE = Patient-Rated Wrist and Hand Evaluation; STAI = State-Trait Anxiety Inventory; BDI-II = Beck Depression Inventory II. † Two-tailed. In-scanner behavioral data We sought to investigate differences between naproxen and placebo treatments in in-scanner self-reported pain measures. VAS scores were significantly reduced following administration of naproxen compared to placebo (F[1,18] = 10.61, P = 0.004) (Figure 2B). Participants' levels of squeeze (10%, 40%, and 70% MVC) induced a significant increase in VAS scores (Greenhouse-Geisser–corrected F[1.115,20.07] = 12.83, P < 0.001) (Figure 2B). There was no interaction between levels of squeeze and treatment. Task performance at 70% MVC evoked the highest perceived pain intensity scores (mean ± SEM VAS score 37.61 ± 4.64 for placebo and 25.379 ± 4.06 for naproxen), providing an appropriate rationale for its use in the ROI analysis. Neuroimaging Effects of naproxen on global CBF Differences between naproxen and placebo conditions in global CBF were not significant (t[18] = −1.3, P = 0.2), suggesting that BOLD signal treatment differences were unlikely to be caused by generalized vascular effects.

In-scanner behavioral data We sought to investigate differences between naproxen and placebo treatments in in-scanner self-reported pain measures. VAS scores were significantly reduced following administration of naproxen compared to placebo (F[1,18] = 10.61, P = 0.004) (Figure 2B). Participants' levels of squeeze (10%, 40%, and 70% MVC) induced a significant increase in VAS scores (Greenhouse-Geisser–corrected F[1.115,20.07] = 12.83, P < 0.001) (Figure 2B). There was no interaction between levels of squeeze and treatment. Task performance at 70% MVC evoked the highest perceived pain intensity scores (mean ± SEM VAS score 37.61 ± 4.64 for placebo and 25.379 ± 4.06 for naproxen), providing an appropriate rationale for its use in the ROI analysis. Neuroimaging Effects of naproxen on global CBF Differences between naproxen and placebo conditions in global CBF were not significant (t[18] = −1.3, P = 0.2), suggesting that BOLD signal treatment differences were unlikely to be caused by generalized vascular effects. Whole-brain voxelwise analysis Whole-brain analyses of task performance fMRI data were performed. Using the average of all levels of load (10%, 40%, and 70% MVC) as an input for each participant in each session, significant reductions in BOLD signal following treatment with naproxen, compared to placebo, were observed in a distributed network of brain regions (see Supplementary Table 1 and Supplementary Figure 2, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.38987/abstract). Several of these brain regions were those that we had identified a priori as being important in the cerebral representation of pain. We observed a cluster in the left primary somatosensory cortex and a smaller cluster in the right primary somatosensory cortex. A further large cluster was identified in the thalamus bilaterally that extended anteriorly into bilateral caudate nuclei and the left nucleus accumbens and inferiorly into the left midbrain in an area approximating that of the substantia nigra. Further reductions were observed in the right insula, extending to the temporal lobe, and bilaterally in the amygdalae. We also observed reductions in the right and left frontal lobes and occipital cortices, extending inferiorly into the cerebellum and the posterior aspect of the pons. We did not observe any relative BOLD signal increases following naproxen administration compared to placebo.

oral lobe, and bilaterally in the amygdalae. We also observed reductions in the right and left frontal lobes and occipital cortices, extending inferiorly into the cerebellum and the posterior aspect of the pons. We did not observe any relative BOLD signal increases following naproxen administration compared to placebo. ROI analysis Significant naproxen-mediated reductions in BOLD signal intensity were identified in the amygdala, hippocampal formation, thalamus, primary somatosensory cortex, and posterior cingulate cortex ROIs bilaterally. Further reductions in the right hemisphere only were observed in the anterior and posterior insula cortices and in the secondary somatosensory cortex. Figure 3 illustrates the mean values in significant ROIs at placebo and naproxen sessions. Results of the pairwise t-tests for the treatment differences are detailed in Table2. Figure 3 A priori region of interest analysis. Adjacent to each image are corresponding left (L) and right (R) hemisphere mean ± SEM blood oxygen level–dependent (BOLD) responses for patients receiving placebo (bright red/bright blue) and patients receiving naproxen (light red/light blue). ∗ = P < 0.05. S2 = secondary somatosensory cortex; S1 = primary somatosensory cortex. Table 2 A priori ROI analysis*

Figure 3 A priori region of interest analysis. Adjacent to each image are corresponding left (L) and right (R) hemisphere mean ± SEM blood oxygen level–dependent (BOLD) responses for patients receiving placebo (bright red/bright blue) and patients receiving naproxen (light red/light blue). ∗ = P < 0.05. S2 = secondary somatosensory cortex; S1 = primary somatosensory cortex. Table 2 A priori ROI analysis* ROI Mean response for placebo treatment Mean response for naproxen treatment Treatment difference Standard error of treatment difference t(18) P for t(18)† Pearson's r (n = 19)‡ P for Pearson's r† Sample size§ ACC (L) 132.81 78.08 54.73 43.57 1.256 0.225 – – 97 ACC (R) 134.38 85.97 48.41 41.75 1.160 0.261 – – 112 AMY (L) 100.57 6.97 93.60 28.53 3.281 0.004 – – 16 AMY (R) 127.35 18.40 108.95 33.85 3.219 0.005 – – 17 HF (L) 57.72 −33.81 91.54 33.65 2.720 0.014 – – 23 HF (R) 84.44 −17.79 102.23 38.32 2.668 0.016 – – 23 aINS (L) 189.50 129.33 60.18 39.16 1.537 0.142 0.109 0.657 66 aINS (R) 256.77 172.95 83.82 33.18 2.527 0.021 −0.188 0.440 26 pINS (L) 84.95 30.50 54.45 40.08 1.359 0.191 0.275 0.255 83 pINS (R) 91.15 −15.19 106.34 33.72 3.154 0.005 −0.015 0.952 18 PCC (L) 80.31 −31.21 111.51 49.16 2.268 0.036 – – 31 PCC (R) 79.04 −37.22 116.26 47.20 2.463 0.024 – – 27 S1 (L) 489.29 368.04 121.25 55.15 2.198 0.041 0.590 0.008 33 S1 (R) 158.06 72.13 85.93 39.10 2.198 0.041 0.484 0.036 33 S2 (L) 190.13 136.42 53.71 41.09 1.307 0.208 0.646 0.003 90 S2 (R) 177.96 88.88 89.09 33.38 2.668 0.016 0.585 0.008 23 TH (L) 181.76 40.95 140.81 42.21 3.336 0.004 0.460 0.048 16 TH (R) 161.75 36.81 124.94 40.57 3.080 0.006 0.427 0.070 18 * Shown are analyses of the following regions of interest (ROIs) in left (L) and right (R) hemispheres: anterior cingulate cortex (ACC), amygdala (AMY), hippocampal formation (HF), anterior insula (aINS), posterior insula (pINS), posterior cingulate cortex (PCC), primary somatosensory cortex (S1), secondary somatosensory cortex (S2), and thalamus (TH).

ing regions of interest (ROIs) in left (L) and right (R) hemispheres: anterior cingulate cortex (ACC), amygdala (AMY), hippocampal formation (HF), anterior insula (aINS), posterior insula (pINS), posterior cingulate cortex (PCC), primary somatosensory cortex (S1), secondary somatosensory cortex (S2), and thalamus (TH). † Two-tailed. ‡ Pearson's product-moment correlation coefficient for relationships between treatment-related differences in the magnitude of the blood oxygen level–dependent signal and differences in perceived pain intensity measured by in-scanner visual analog scale recordings. § Post hoc sample size calculations (alpha level of 0.05, power of 80%) for each ROI. Exploring the relationship between treatment-induced changes in task-related VAS pain intensity and mean BOLD responses demonstrated positive correlations in the primary and secondary somatosensory cortices bilaterally and in the left thalamus (contralateral to the affected hand). Representative plots of these relationships (Figure 4) illustrate that reductions in BOLD response following naproxen treatment were associated with reductions in perceived pain intensity. Post hoc calculations of the sample size needed to detect significant treatment effects varied considerably between ROIs, ranging from a minimum of 16 participants (left amygdala and left thalamus) to a maximum of 112 participants (right anterior cingulate cortex) (Table2).

with reductions in perceived pain intensity. Post hoc calculations of the sample size needed to detect significant treatment effects varied considerably between ROIs, ranging from a minimum of 16 participants (left amygdala and left thalamus) to a maximum of 112 participants (right anterior cingulate cortex) (Table2). Figure 4 Scatterplots showing the relationship between the change (naproxen condition − placebo condition) in BOLD responses within a particular region of interest (y-axis) and the change in visual analog scale (VAS) scores collected during the task (x-axis). Plus signs represent individual patients. See Figure 3 for other definitions. Discussion We have demonstrated the sensitivity of BOLD fMRI to detect the effects of naproxen, an analgesic of known efficacy, in a group of individuals with pain secondary to OA of the CMC joint. Local reductions in BOLD signal were identified in brain regions commonly associated with the experience of pain (28), several of which we hypothesized a priori to show changes in response to analgesia. Changes in treatment-related VAS indices of pain intensity correlated with changes in BOLD response in several of these brain regions. These findings demonstrate that BOLD fMRI adds value to conventional self-report measures, offering a mechanistic understanding of both pain and treatment response (11).

e to analgesia. Changes in treatment-related VAS indices of pain intensity correlated with changes in BOLD response in several of these brain regions. These findings demonstrate that BOLD fMRI adds value to conventional self-report measures, offering a mechanistic understanding of both pain and treatment response (11). Reductions in BOLD response were identified in a distributed network of brain regions following administration of naproxen, compared to placebo. These effects were identified in a real-world sample of patients with painful OA, within the constraints of a fully blinded crossover experimental design. Given the location of reductions in BOLD response following naproxen treatment and the relationship in the thalamus and primary and secondary somatosensory cortices between changes in BOLD signal intensity and perceived pain, we infer that our fMRI findings represent an analgesic effect of naproxen on OA-related evoked pain. The absence of a significant treatment effect on global CBF provides confidence that treatment-derived BOLD signal changes were most likely driven via mechanisms of neurovascular coupling, rather than via a more generalized global increase in CBF (32). We recommend acquisition of imaging data using multiple fMRI modalities, alongside conventional behavioral end points, in order to best understand pain and analgesic responses (11).

changes were most likely driven via mechanisms of neurovascular coupling, rather than via a more generalized global increase in CBF (32). We recommend acquisition of imaging data using multiple fMRI modalities, alongside conventional behavioral end points, in order to best understand pain and analgesic responses (11). Our results build on those of earlier studies of experimental, acute, and persistent pain conditions treated with various analgesic medications, in which treatment effects have been demonstrated by fMRI (e.g., see refs.(33–35)). For the first time, we have demonstrated a robust effect of naproxen on evoked stimulation of pain in the CMC joint with OA. Previous attempts to use pain secondary to OA as a clinical model to assess analgesic effects by fMRI are difficult to interpret, given that these effects were examined in so few patients. The intention of Parks et al (15) to perform a larger fMRI study to evaluate the treatment effects of valdecoxib, a cyclooxygenase 2 inhibitor, was hampered by the withdrawal of this agent from the US market. In a subgroup of only 6 participants, those investigators reported a significant effect of treatment on behavioral indices of background/spontaneous pain but not evoked pain. Interrelationships between drug concentrations, perceived indices of spontaneous pain, and BOLD activity were also described, but caution should be exercised given their derivation from so few participants. While Baliki et al (14) reported treatment effects in their comparison of the effects of lidocaine on chronic low back pain (n = 7) and evoked knee OA pain (n = 5) in the medial prefrontal cortex and thalamus, respectively, the authors acknowledge limitations in that study. Due to small sample sizes, statistical handling of the data limited treatment-related inferences to the study samples rather than to the clinical populations that they represented, and the study had no placebo arm or a control for potential treatment order effects.

ly, the authors acknowledge limitations in that study. Due to small sample sizes, statistical handling of the data limited treatment-related inferences to the study samples rather than to the clinical populations that they represented, and the study had no placebo arm or a control for potential treatment order effects. Our work adds new knowledge to the field by demonstrating an observable treatment effect in a larger cohort of participants, in a study that was appropriately double blinded and placebo controlled. Post hoc power calculations from this study using a priori–defined ROIs indicated that a minimum of 16 individuals with OA of the CMC joint are needed in order to identify a treatment effect on evoked pain. A sample size of 16 was sufficient only in the thalamus and amygdala contralateral to the affected hand. We included these calculations to guide others in powering future studies, but we recognize that sample sizes may change due to factors including body site, severity of OA pain, paradigm design, and analgesic effect size. It is an unrealistic expectation that treatment effects will modulate brain responses either in all brain regions or in equal magnitude across them. We have shown that changes in brain response relate to perceived pain relief; however, we speculate that the magnitude, direction, and spatial location of these changes are likely to represent a composite generalized “analgesic effect” and response characteristics unique to the analgesic under assessment. In light of increasing evidence of the impact of persistent pain states on both brain structure and function (36,37), use of anatomically defined ROIs, while unbiased (38), may also partially account for variability in required sample sizes across ROIs. However, given the broad acknowledgment that pain is a multifaceted experience underpinned by a complex network of brain regions (39), it is of course overly simplistic to ascertain treatment effects from single ROIs in isolation.

ed (38), may also partially account for variability in required sample sizes across ROIs. However, given the broad acknowledgment that pain is a multifaceted experience underpinned by a complex network of brain regions (39), it is of course overly simplistic to ascertain treatment effects from single ROIs in isolation. A critical quality provided by the use of neuroimaging end points is the value that they add to conventional behavioral indices, providing mechanistic correlates of persistent pain (24) and treatment effects. Previous studies (15,40) have shown perturbed affective responses to pain in OA; these have recently been theorized to be maladaptive assessments of and responses to pain as a threat to homeostatic body functioning (24). Subtle variations in the location of supraspinal pain responses in these studies may reflect heterogeneity in clinical pain states, but together they add weight to a developing view that persistent OA pain may be, at least in part, maintained by changes in central as well as peripheral nervous system functioning (14,15,24,40). Neuroimaging reports have also described altered functioning of endogenous pain control mechanisms in individuals with painful OA. A recent fMRI study (5) showed the presence of neuropathic-like symptoms in patients with OA, using the PainDETECT questionnaire (41). When fMRI was combined with quantitative sensory testing, hyperalgesic responses to punctate stimuli were associated with significant periaqueductal gray matter activity, leading the authors to interpret that the periaqueductal gray matter has a role in central sensitization in at least some patients with painful hip OA. Similarly, our own work examining background pain in individuals with painful hand OA, compared to healthy controls, showed differential CBF between groups in the periaqueductal gray matter and other descending modulatory systems in the midbrain and brainstem (24).

ization in at least some patients with painful hip OA. Similarly, our own work examining background pain in individuals with painful hand OA, compared to healthy controls, showed differential CBF between groups in the periaqueductal gray matter and other descending modulatory systems in the midbrain and brainstem (24). These findings provide important insights for interventions aimed at the reduction of OA-related pain, which currently remain focused on reducing or eliminating potential peripheral generators of nociception, for example, joint replacement surgery (42,43). Outcomes of these interventions are not always satisfactory for individuals (for review, see ref.(44)), which may be due in part to supraspinal pain mechanisms remaining essentially uncharacterized and untreated. In this work, we did not find a modulatory effect of naproxen in these regions. Type II errors aside, these null findings suggest the working hypothesis that the analgesic effect of naproxen does not modulate descending pain control system responses to evoked pain. Specifically designed new studies will be needed to test this.

work, we did not find a modulatory effect of naproxen in these regions. Type II errors aside, these null findings suggest the working hypothesis that the analgesic effect of naproxen does not modulate descending pain control system responses to evoked pain. Specifically designed new studies will be needed to test this. In this study, we combined a “gold standard” pharmacologic MRI analgesic design methodology with a novel functional task and a mock scan protocol. This design aims to account for additional influences on BOLD response, including treatment order, task demands, anxiety, mood, and placebo responses. Psychometric data on anxiety and depression were stable across sessions, reducing the likelihood that they influenced the fMRI results. Our chosen task differs from conventional evoked-response paradigms in that pain responses are elicited with an accompanying motor component. However, in light of recent comments that pain may represent an actual or perceived threat to the body (45,46), we speculate that experimental paradigms that promote genuine pain-provoking actions, rather than those that elicit responses to on/off noxious stimulation, may represent a more comprehensive cerebral fingerprint of pain response in OA (47).

nts that pain may represent an actual or perceived threat to the body (45,46), we speculate that experimental paradigms that promote genuine pain-provoking actions, rather than those that elicit responses to on/off noxious stimulation, may represent a more comprehensive cerebral fingerprint of pain response in OA (47). In summary, our study demonstrates the sensitivity of BOLD fMRI to detect the mechanisms underlying treatments of known efficacy in OA. The locations of the effects were specific and physiologically plausible, and treatment-related changes in VAS scores were related to changes in BOLD response in brain regions underpinning sensory discriminative aspects of the pain experience. These results demonstrate the enticing potential of fMRI as an adjunct to self-report for detection of early signals of efficacy of novel pharmacologic and nonpharmacologic treatments, in small numbers of individuals with persistent pain (see refs.(12) and(48)), which in turn may provide them with reduced pain and increased quality of life. Author Contributions 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. Howard 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. Sanders, Huggins, Vennart, Massat, Williams, Howard. Acquisition of data. Sanders, Krause, Choy, Williams, Howard.

Author Contributions 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. Howard 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. Sanders, Huggins, Vennart, Massat, Williams, Howard. Acquisition of data. Sanders, Krause, Choy, Williams, Howard. Analysis and interpretation of data. Sanders, Krause, O'Muircheartaigh, Thacker, Huggins, Vennart, Massat, Choy, Howard. Role of the Study Sponsor This project was an academic–industrial collaboration between King's College London and the study sponsor, Pfizer Global Research and Development, UK. Pfizer and King's College London scientists worked in collaboration on the following areas: study design, data analysis, interpretation of data, and the writing of the manuscript. All data collection was performed by King's College London scientists only. Pfizer approved the content of the manuscript prior to submission. We would like to thank Chris Andrew for assistance in the design and development of the task and Dr. Fernando Zelaya for his valuable contributions.

B cell–targeted monoclonal antibodies (mAb) are increasingly being explored for use in the treatment of autoimmune diseases such as rheumatoid arthritis (RA) and systemic lupus erythematosus (SLE). Rituximab (RTX), a chimeric anti‐CD20 mAb, is licensed for the treatment of RA and is used extensively off‐label for the treatment of refractory SLE. However, RTX induces incomplete B cell depletion in some individuals with RA 1 and SLE 2, which may at least partly explain the poor clinical response noted in some individuals 3, 4. A long duration of B cell depletion in RA (using an extra dose of RTX) 5 and SLE patients is associated with better clinical response 6. Hence, enhancing B cell depletion may improve treatment efficacy, and understanding the mechanisms of resistance in RA and SLE is of clear clinical importance. B cell–depletion studies in lupus‐prone mice suggest disease‐specific mechanisms of resistance to anti‐CD20 mAb 7, but the precise mechanisms of resistance to RTX in patients with RA and SLE remain elusive.

treatment efficacy, and understanding the mechanisms of resistance in RA and SLE is of clear clinical importance. B cell–depletion studies in lupus‐prone mice suggest disease‐specific mechanisms of resistance to anti‐CD20 mAb 7, but the precise mechanisms of resistance to RTX in patients with RA and SLE remain elusive. Administration of anti‐CD20 mAb can evoke 3 main cytotoxic effector mechanisms, antibody‐dependent cell‐mediated cytotoxicity (ADCC), complement‐dependent cytotoxicity (CDC), and direct cell death 8. The association between Fcγ receptor type IIIa (FcγRIIIa) genotype and clinical response and/or the degree of B cell depletion in RA 9 and SLE 10 suggests that the ADCC‐type FcγR‐dependent systems (including antibody‐dependent cellular phagocytosis) are the main RTX effector mechanisms in vivo, in both RA and SLE, as previously noted for some B cell malignancies 11, 12, 13.

) genotype and clinical response and/or the degree of B cell depletion in RA 9 and SLE 10 suggests that the ADCC‐type FcγR‐dependent systems (including antibody‐dependent cellular phagocytosis) are the main RTX effector mechanisms in vivo, in both RA and SLE, as previously noted for some B cell malignancies 11, 12, 13. Anti‐CD20 mAb can be categorized into 2 types based on whether they redistribute CD20 into lipid rafts and consequently evoke different effector mechanisms 8, 14. Type I mAb include RTX, ofatumumab (2F2, a fully human IgG1), and ocrelizumab (a humanized IgG1). Both ofatumumab and ocrelizumab have been shown to be effective in treating patients with RA 15, 16. Type II mAb include tositumomab (anti‐B1, a mouse IgG2a) and obinutuzumab (GA101, a glycoengineered human IgG1). Type II mAb have been shown to be more efficient than type I at depleting B cells in preclinical models 17 and in patients with B cell malignancies 18, resulting in improved clinical efficacy in chronic lymphocytic leukemia 19. However, whether type II mAb are more effective at depleting B cells from patients with RA and SLE is not known.

n shown to be more efficient than type I at depleting B cells in preclinical models 17 and in patients with B cell malignancies 18, resulting in improved clinical efficacy in chronic lymphocytic leukemia 19. However, whether type II mAb are more effective at depleting B cells from patients with RA and SLE is not known. The improved efficacy of type II mAb is attributed to the observation that normal and malignant B cells internalize type I mAb more rapidly than type II, a mechanism regulated by the inhibitory Fcγ receptor IIb (FcγRIIb) on B cells 20. Internalization of mAb reduces the ability to activate FcγR‐dependent ADCC functions 21, including phagocytosis 20, 22, and so is thought to be detrimental for target‐cell depletion, with the expression of FcγRIIb on target lymphoma cells being associated with a poor clinical response to RTX 20, 23. However, whether similar resistance mechanisms are operant in RA and SLE is not known.

nt ADCC functions 21, including phagocytosis 20, 22, and so is thought to be detrimental for target‐cell depletion, with the expression of FcγRIIb on target lymphoma cells being associated with a poor clinical response to RTX 20, 23. However, whether similar resistance mechanisms are operant in RA and SLE is not known. In autoimmune conditions, internalization of B cell–targeted antigen–mAb‐complexes may have a beneficial immunomodulatory effect, as discussed elsewhere 24. For example, epratuzumab, an anti‐CD22 mAb, is rapidly internalized after binding to its target antigen CD22, and so anti‐CD22 mAb–conjugated toxins are used to treat B cell malignancies 25. Unconjugated anti‐CD22 mAb may have utility in autoimmune situations by facilitating endocytosis of CD22 and modulating B cell receptor (BCR) signaling 26, in addition to eliciting modest ADCC 27. Epratuzumab appears to be effective in SLE 28. B cells from patients with SLE likely internalize anti‐CD22 mAb 29; whether FcγRIIb regulates this internalization is not known.

mmune situations by facilitating endocytosis of CD22 and modulating B cell receptor (BCR) signaling 26, in addition to eliciting modest ADCC 27. Epratuzumab appears to be effective in SLE 28. B cells from patients with SLE likely internalize anti‐CD22 mAb 29; whether FcγRIIb regulates this internalization is not known. In this study, we found that the slower‐internalizing type II anti‐CD20 mAb depleted B cells from patients with RA and SLE more efficiently than either type I anti‐CD20 or anti‐CD22 mAb and that internalization influenced the efficiency of depletion. We also found that the extent of internalization of rituximab was highly variable between patients, was regulated by FcγRIIb, and was inversely correlated with its cytotoxicity in whole blood B cell–depletion assays. Blocking of FcγRIIb inhibited the internalization of type I anti‐CD20 mAb, with variable levels of internalization noted between different B cell subpopulations; being least for postswitched (IgD–CD27+) memory cells and IgD– cells. Internalization of type I anti‐CD20, but not anti‐CD22, mAb was partially inhibited by stimulation with anti‐IgM, which suggests independent roles for the BCR and FcγRIIb in facilitating the internalization of type I mAb. PATIENTS AND METHODS Patients and healthy blood donors Ethical approval for the study was obtained from the National Research Ethics Committee. Whole blood samples from all participants were obtained with their informed consent, adhering to the Declaration of Helsinki.

Supporting information Supplementary Figure 1. Dose‐response experiments. We determined the optimal concentration of mAbs (0.01, 0.1, 1 and 10 μg/mL) in four Independent experiments using blood from normal healthy controls and using non‐glycomodified versions of GA101 (GA101gly) to directly assess the effects of type I versus II without the influence of afucosylation. Whole blood samples were incubated with or without RTX or GA101gly at 0.01, 0.1, 1 and 10 μg/ml and percentage B cell death measured by flow cytometric analysis after 24 h and mean of triplicate wells was used. Cytotoxicity of RTX and GA101gly were compared in healthy controls (n=4). Rituximab (RTX) lyses B cells less efficiently than GA101gly in all four samples at all four concentrations tested. The results are the means and SD. Supplementary Figure 2. Differential expression of IgD and FcγRIIb in B cell subpopulations. (A) Similar to a previous report,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

neS8qaW1tdW5vbG9neS9waHlzaW9sb2d5PC9rZXl3b3JkPjxrZXl3b3JkPkNhbGNpdW0vbWV0YWJvbGlzbTwva2V5d29yZD48a2V5d29yZD5GZW1hbGU8L2tleXdvcmQ+PGtleXdvcmQ+SHVtYW5zPC9rZXl3b3JkPjxrZXl3b3JkPipJbW11bm9sb2dpYyBNZW1vcnk8L2tleXdvcmQ+PGtleXdvcmQ+THVwdXMgRXJ5dGhlbWF0b3N1cywgU3lzdGVtaWMvKmltbXVub2xvZ3k8L2tleXdvcmQ+PGtleXdvcmQ+THltcGhvY3l0ZSB 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

1bWJpYSBVbml2ZXJzaXR5IE1lZGljYWwgQ2VudGVyLCBOZXcgWW9yaywgTlkgMTAwMzIsIFVTQS4gbWNtMjEyM0Bjb2x1bWJpYS5lZHU8L2F1dGgtYWRkcmVzcz48dGl0bGVzPjx0aXRsZT5TZWxlY3RpdmUgZHlzcmVndWxhdGlvbiBvZiB0aGUgRmNnYW1tYUlJQiByZWNlcHRvciBvbiBtZW1vcnkgQiBjZWxscyBpbiBTTEU8L3RpdGxlPjxzZWNvbmRhcnktdGl0bGU+SiBFeHAgTWVkPC9zZWNvbmRhcnktdGl0bGU+PGFsdC1 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1 we found that the mean fluorescence intensity (MFI) of FcγRIIb varied between B cell subpopulations in SLE.

In this study, we found that the slower‐internalizing type II anti‐CD20 mAb depleted B cells from patients with RA and SLE more efficiently than either type I anti‐CD20 or anti‐CD22 mAb and that internalization influenced the efficiency of depletion. We also found that the extent of internalization of rituximab was highly variable between patients, was regulated by FcγRIIb, and was inversely correlated with its cytotoxicity in whole blood B cell–depletion assays. Blocking of FcγRIIb inhibited the internalization of type I anti‐CD20 mAb, with variable levels of internalization noted between different B cell subpopulations; being least for postswitched (IgD–CD27+) memory cells and IgD– cells. Internalization of type I anti‐CD20, but not anti‐CD22, mAb was partially inhibited by stimulation with anti‐IgM, which suggests independent roles for the BCR and FcγRIIb in facilitating the internalization of type I mAb. PATIENTS AND METHODS Patients and healthy blood donors Ethical approval for the study was obtained from the National Research Ethics Committee. Whole blood samples from all participants were obtained with their informed consent, adhering to the Declaration of Helsinki. Demographic features of the RA and SLE patients are summarized in Supplementary Tables 1 and 2, respectively, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.39167/abstract. The median age of the 3 study groups was 31 years (range 22–60 years) in the healthy controls, 52 years (range 24–79 years) in the RA patients, and 39 years (range 21–76 years) in the SLE patients. All RA patients were positive for rheumatoid factor and/or anti–cyclic citrullinated peptide antibodies. Peripheral blood was collected into tubes containing lithium heparin. Peripheral blood mononuclear cells (PBMCs) were separated using Ficoll‐Paque density‐gradient centrifugation, and B cells were isolated from the PBMCs by negative selection using either a human B cell enrichment kit (StemCell Technologies) or human B cell isolation kit II (Miltenyi Biotec).

containing lithium heparin. Peripheral blood mononuclear cells (PBMCs) were separated using Ficoll‐Paque density‐gradient centrifugation, and B cells were isolated from the PBMCs by negative selection using either a human B cell enrichment kit (StemCell Technologies) or human B cell isolation kit II (Miltenyi Biotec). Antibodies and reagents AT10, which binds both FcγRIIa and FcγRIIb 30, was produced in‐house. Rituximab was a gift from the Southampton General Hospital Pharmacy, and tositumomab was a gift from Prof T. Illidge (University of Manchester, Manchester, UK). Glycosylated GA101 with an unmodified Fc portion (GA101Gly) and ofatumumab were produced in‐house from patented published sequences in Chinese hamster ovary or 293F cells; therefore, their carbohydrate structures may differ from mAb in clinical use. Alexa Fluor 488 and anti–Alexa Fluor 488 were purchased from Invitrogen. The mAb were labeled with Alexa Fluor 488 according to the manufacturer's (Invitrogen) instructions.

patented published sequences in Chinese hamster ovary or 293F cells; therefore, their carbohydrate structures may differ from mAb in clinical use. Alexa Fluor 488 and anti–Alexa Fluor 488 were purchased from Invitrogen. The mAb were labeled with Alexa Fluor 488 according to the manufacturer's (Invitrogen) instructions. Flow cytometry The following fluorochrome‐conjugated mAb (all from Becton Dickinson) were used for flow cytometry: CD3 (allophycocyanin), CD19 (phycoerythrin [PE]–Cy7 or PerCP–Cy5.5), CD20 (fluorescein isothiocyanate), CD32 (PE), CD45 (PE), and IgD (Brilliant Violet 421). Flow cytometry was performed using a Becton Dickinson LSRFortessa cell analyzer. Lymphocyte populations were identified using forward‐ and side‐scatter characteristics and CD45 positivity. B cells were identified as CD19+ or CD20+ and T cells as CD3+. To account for interexperimental variation, the mean fluorescence intensity (MFI) of CD20 and FcγRIIb was determined as the ratio of the MFI of CD20/FcγRIIb to the MFI of the isotype control.

ng forward‐ and side‐scatter characteristics and CD45 positivity. B cells were identified as CD19+ or CD20+ and T cells as CD3+. To account for interexperimental variation, the mean fluorescence intensity (MFI) of CD20 and FcγRIIb was determined as the ratio of the MFI of CD20/FcγRIIb to the MFI of the isotype control. Whole blood B cell–depletion assay The whole blood B cell–depletion assay was performed as described previously 31. Briefly, 100 μl of freshly drawn whole blood was incubated in the presence or absence of mAb at 37°C in an atmosphere of 5% CO2. Samples were harvested after 24 hours and stained with anti‐CD3, anti‐CD19, and anti‐CD45 and then incubated for another 30 minutes before lysing the red blood cells with BD PharmLyse. Ten thousand events were acquired in the lymphocyte gate per sample, and the data were analyzed by flow cytometry using FlowJo software using the protocol shown in Figure 1A. The percentage of B cell depletion with mAb was defined as the cytotoxicity index (CTI) and was determined using the following formula: CTI of mAb = 100 – [(100/B cell:T cell ratio in sample without antibody) × (B cell:T cell ratio in sample with antibody)]. The percentage B cell depletion in the sample without antibody is set at 0. The mean values for triplicate wells were used to calculate the CTI.

CTI) and was determined using the following formula: CTI of mAb = 100 – [(100/B cell:T cell ratio in sample without antibody) × (B cell:T cell ratio in sample with antibody)]. The percentage B cell depletion in the sample without antibody is set at 0. The mean values for triplicate wells were used to calculate the CTI. Figure 1 Whole blood B cell–depletion assay. A, Whole blood samples were incubated with or without rituximab (RTX) or glycosylated GA101 with an unmodified Fc portion (GA101Gly) for 24 hours, and the percentage of B cell death was determined by flow cytometry. Ten thousand events gated on lymphocytes were acquired per sample. B cells were identified as CD19+ and T cells as CD3+. B, Cytotoxicity (percentage B cell depletion) of RTX was significantly lower than that of GA101Gly in samples from healthy controls (n = 9), patients with rheumatoid arthritis (RA; n = 26) and patients with systemic lupus erythematosus (SLE; n = 50). Cytotoxicity achieved by both RTX and GA101Gly was significantly lower in SLE patients than in healthy controls or in RA patients. Data are shown as box plots. Each box represents the interquartile range. Lines inside the boxes represent the median. Whiskers represent the range. ∗∗ = P < 0.005; ∗∗∗ = P < 0.0001.

= 50). Cytotoxicity achieved by both RTX and GA101Gly was significantly lower in SLE patients than in healthy controls or in RA patients. Data are shown as box plots. Each box represents the interquartile range. Lines inside the boxes represent the median. Whiskers represent the range. ∗∗ = P < 0.005; ∗∗∗ = P < 0.0001. Surface fluorescence–quenching assay The surface fluorescence–quenching assay was performed as described previously 22. Briefly, 2–4 × 105 B cells were incubated with Alexa Fluor 488–labeled mAb in a volume of 5 μg/ml at 37°C for 6 hours. As we had observed differences in the ability of different IgG isotypes to activate FcγRIIb 20, all mAb used were either human or mouse IgG1, which give equivalent activity in internalization assays with anti‐CD20 mAb 24. Samples were then harvested, washed twice, and incubated for 30 minutes at 4°C with PE–Cy7–labeled anti‐CD19 in the presence or absence of anti–Alexa Fluor 488 quenching antibody (Invitrogen). After washing, samples were analyzed by flow cytometry. We investigated internalization in the following B cell subpopulations: naive (IgD+CD27–), preswitched (IgD+CD27+), postswitched (IgD–CD27+), and double‐negative (IgD–CD27–) cells. Samples were stained with PE–Cy7–labeled anti‐CD19, BV421‐labeled IgD, or PE‐labeled CD27 after incubation with Alexa Fluor 488–labeled mAb.

We investigated internalization in the following B cell subpopulations: naive (IgD+CD27–), preswitched (IgD+CD27+), postswitched (IgD–CD27+), and double‐negative (IgD–CD27–) cells. Samples were stained with PE–Cy7–labeled anti‐CD19, BV421‐labeled IgD, or PE‐labeled CD27 after incubation with Alexa Fluor 488–labeled mAb. The effect of FcγRIIb on internalization of mAb was investigated by comparing the MFI of FcγRIIb in samples with and those without prior incubation with AT10 at 50 μg/ml for 30 minutes before the addition of Alexa Fluor 488–labeled mAb. The effect of B cell activation on internalization was investigated by stimulating isolated B cells with anti‐IgM F(ab′)2 at 25 μg/ml for 30 minutes or for 6 hours before incubating with Alexa Fluor 488–labeled mAb. Statistical analysis Statistical analyses were performed with GraphPad Prism software version 5.0. Paired‐t test or Mann‐Whitney U test was used to compare groups as appropriate. Spearman's rank correlation r2 was used to analyze correlations between parameters.

The effect of FcγRIIb on internalization of mAb was investigated by comparing the MFI of FcγRIIb in samples with and those without prior incubation with AT10 at 50 μg/ml for 30 minutes before the addition of Alexa Fluor 488–labeled mAb. The effect of B cell activation on internalization was investigated by stimulating isolated B cells with anti‐IgM F(ab′)2 at 25 μg/ml for 30 minutes or for 6 hours before incubating with Alexa Fluor 488–labeled mAb. Statistical analysis Statistical analyses were performed with GraphPad Prism software version 5.0. Paired‐t test or Mann‐Whitney U test was used to compare groups as appropriate. Spearman's rank correlation r2 was used to analyze correlations between parameters. RESULTS Double the efficiency of B cell depletion by type II mAb versus type I mAb The autologous whole blood B cell–depletion assay is a comprehensive method for assessing mAb cytotoxicity in vitro, as it accounts for all 3 effector mechanisms evoked by mAb: ADCC, CDC, and direct cell death. This assay was previously used to show that type II mAb are more efficient at lysing B cells from healthy control subjects and from patients with B cell malignancies 18, 31. However, whether type II mAb are more efficient at lysing B cells from patients with autoimmune disease is not known.

CDC, and direct cell death. This assay was previously used to show that type II mAb are more efficient at lysing B cells from healthy control subjects and from patients with B cell malignancies 18, 31. However, whether type II mAb are more efficient at lysing B cells from patients with autoimmune disease is not known. Since SLE patients often have lymphopenia, making extensive assays difficult, we initially determined the optimal concentration of mAb (0.01, 0.1, 1, and 10 μg/ml) required for the assay using blood from healthy controls. In these assays, we used nonglycomodified versions of GA101 to directly assess the effects of type I versus type II mAb without the influence of afucosylation. Independent experiments were performed in whole blood samples from 4 healthy control subjects. The mean percentage of cell death was used to assess the cytotoxicity of the mAb and to determine the optimum dose. We found that GA101Gly was significantly more efficient at lysing B cells than rituximab was in all 4 samples at all 4 concentrations tested (see Supplementary Figure 1, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.39167/abstract). Therefore, we used 1 μg/ml for subsequent experiments.

was significantly more efficient at lysing B cells than rituximab was in all 4 samples at all 4 concentrations tested (see Supplementary Figure 1, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.39167/abstract). Therefore, we used 1 μg/ml for subsequent experiments. The cytotoxicity index from the autologous whole blood B cell–depletion assay (1 μg/ml) was calculated in whole blood samples from 9 healthy control subjects, 26 patients with RA, and 50 patients with SLE (Figure 1B). GA101Gly (type II mAb) was found to be significantly more efficient than RTX (type I mAb) at lysing B cells in vitro in samples from all study groups. The mean ± SD CTI for GA101Gly versus RTX was 63 ± 11 versus 36 ± 18 in healthy controls (P = 0.005), 54 ± 16 versus 27 ± 16 in RA patients (P < 0.0001), and 38 ± 15 versus 17 ± 12 in SLE patients (P < 0.0001). There was no significant difference between the CTI for RTX in healthy controls and RA patients, whereas the CTI for RTX was significantly lower in SLE patients as compared with healthy controls (P = 0.008) and RA (P = 0.01). Similarly, there was no significant difference in the CTI for GA101Gly between healthy controls and RA patients, whereas it was significantly lower in SLE patients as compared with healthy controls (P = 0.0006) and with RA patients (P < 0.0001) (Figure 1B). The median ratio of the CTI for GA101Gly to the CTI for RTX was 1.5, 1.7, and 2.5, for healthy controls, RA patients, and SLE patients, respectively.

controls and RA patients, whereas it was significantly lower in SLE patients as compared with healthy controls (P = 0.0006) and with RA patients (P < 0.0001) (Figure 1B). The median ratio of the CTI for GA101Gly to the CTI for RTX was 1.5, 1.7, and 2.5, for healthy controls, RA patients, and SLE patients, respectively. We did not find a correlation between the CTI for RTX and the distribution of relative frequencies of B cell subpopulations (data not shown and Supplementary Table 3, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.39167/abstract.) or serum complement C3 levels (data not shown). Taken together, these results suggested that type II mAb are more effective at depleting B cells in each subset of study subjects and that B cells from SLE patients are less susceptible to lysis by RTX and by GA101Gly, indicating an inherent resistance mechanism. Next, we wanted to investigate whether the difference in the B cell–lysing potential of RTX and GA101Gly was also applicable to additional type I and type II mAb. We therefore compared the CTI of 2 other mAb: ofatumumab (2F2) and tositumomab (B1), representing type I and type II mAb, respectively (Figure 2A). Again, we found that type II mAb were significantly more efficient than type I mAb at lysing B cells in all samples examined from patients with RA (n = 3) and SLE (n = 10). We noted a significant hierarchy in the efficiency of CTI of the mAb, with GA101Gly > B1 >2F2 > RTX and with a >2‐fold difference in the CTIs for GA101Gly versus 2F2 (Figure 2A).

b were significantly more efficient than type I mAb at lysing B cells in all samples examined from patients with RA (n = 3) and SLE (n = 10). We noted a significant hierarchy in the efficiency of CTI of the mAb, with GA101Gly > B1 >2F2 > RTX and with a >2‐fold difference in the CTIs for GA101Gly versus 2F2 (Figure 2A). Figure 2 Efficiency of B cell lysis by type I monoclonal antibody (mAb) as compared with type II mAb and with anti‐CD22 mAb. A, Whole blood samples were incubated with or without 1 μg/ml of either rituximab (RTX; IgG1), ofatumumab (2F2; IgG1), tositumomab (B1; mouse IgG2a), or glycosylated GA101 with an unmodified Fc portion (GA101Gly; IgG1). After 24 hours, the percentage of B cell death was measured by flow cytometry. The cytotoxicity of type I and type II mAb was compared in patients with rheumatoid arthritis (n = 3) and systemic lupus erythematosus (SLE; n = 10). Type I mAb lysed B cells less efficiently than did type II mAb, with a cytotoxicity index for RTX < 2F2 < B1 < GA101Gly. Values are the mean of triplicate wells. Each line represents an individual sample. B, In SLE patients (n = 4), the cytotoxicity index of anti‐CD22mAb was significantly lower than that of RTX and GA101Gly. ∗ = P < 0.05; ∗∗ = P < 0.005; ∗∗∗ = P < 0.0001.

e II mAb, with a cytotoxicity index for RTX < 2F2 < B1 < GA101Gly. Values are the mean of triplicate wells. Each line represents an individual sample. B, In SLE patients (n = 4), the cytotoxicity index of anti‐CD22mAb was significantly lower than that of RTX and GA101Gly. ∗ = P < 0.05; ∗∗ = P < 0.005; ∗∗∗ = P < 0.0001. As anti‐CD22 mAb have also been reported to deplete B cells, albeit weakly 27, we examined their activity in the assay. We found that the CTI for anti‐CD22 mAb was found to be significantly lower than that for anti‐CD20 mAb, with a CTI hierarchy of anti‐CD22 < RTX < GA101Gly (n = 4) (Figure 2B). This may be at least partly due to the differences in internalization of the mAb, as noted previously for B cell malignancies 24. Both B cell–intrinsic and B cell–extrinsic factors may account for the apparent resistance of SLE B cells to depletion. Malignant B cell expression of CD20 32 and FcγRIIb 20 correlated with susceptibility to deletion by anti‐CD20 mAb; however, we did not find a correlation between the expression of CD20 and FcγRIIb or between their relative expression (ratio of the MFI of CD20 to the MFI of FcγRIIb) and the CTI for RTX or GA101Gly in all groups examined (data not shown). This may reflect the relatively small difference in B cell expression of CD20 and FcγRIIb between these RA and SLE study patients (data not shown), in contrast to that reported for patients with B cell malignancies 20, 32.

0 to the MFI of FcγRIIb) and the CTI for RTX or GA101Gly in all groups examined (data not shown). This may reflect the relatively small difference in B cell expression of CD20 and FcγRIIb between these RA and SLE study patients (data not shown), in contrast to that reported for patients with B cell malignancies 20, 32. Efficiency of depletion influenced by internalization of rituximab Given the large variability in depletion afforded by RTX in SLE and a superior efficacy of type II mAb in the whole blood B cell–depletion assays, we next examined whether internalization of mAb might explain the greater resistance of SLE B cells to depletion. Internalization was assessed using the surface fluorescence–quenching assay using isolated B cells from 5 healthy controls, 16 patients with RA, and 22 patients with SLE. In all groups, a significantly greater percentage of GA101Gly than RTX was accessible on the cell surface. The median percentage of surface‐accessible mAb after 6 hours of incubation for GA101Gly versus RTX in the healthy controls, RA patients, and SLE patients was 67 versus 57, 69 versus 55 (P < 0.005), and 74 versus 47 (P < 0.005), respectively (Figure 3A). Thus, internalization of mAb was a notable feature of B cells from healthy controls as well as from patients with RA and SLE.

HM+PC91cmxzPjxjdXN0b20yPjIxMTgzOTA8L2N1c3RvbTI+PGVsZWN0cm9uaWMtcmVzb3VyY2UtbnVtPjEwLjEwODQvamVtLjIwMDUxNTAzPC9lbGVjdHJvbmljLXJlc291cmNlLW51bT48bGFuZ3VhZ2U+ZW5nPC9sYW5ndWFnZT48L3JlY29yZD48L0NpdGU+PC9FbmROb3RlPn==1 we found that the mean fluorescence intensity (MFI) of FcγRIIb varied between B cell subpopulations in SLE. Naïve cells expressed significantly lower levels when compared with other B cell subpopulations with a hierarchy of expression: naïve < double negative < post‐switched < pre‐switched cells. Post‐switched memory cells (MCs) expressed FcγRIIb to a similar level as pre‐switched MCs and double negative cells. The horizontal line represents the median; the box, interquartile range; the whiskers, 10‐90th percentile; and the dots represent outliers. (B) Naïve cells expressed significantly higher levels of IgD compared with pre‐switched cells, the results represent the mean and SD, in contrast to the expression of FcγRIIb (A).

s of incubation for GA101Gly versus RTX in the healthy controls, RA patients, and SLE patients was 67 versus 57, 69 versus 55 (P < 0.005), and 74 versus 47 (P < 0.005), respectively (Figure 3A). Thus, internalization of mAb was a notable feature of B cells from healthy controls as well as from patients with RA and SLE. Figure 3 Internalization of rituximab (RTX) to a highly variable extent, impairing its efficiency of depletion. A, Internalization was assessed by surface fluorescence–quenching assay, which revealed that a greater percentage of glycosylated GA101 with an unmodified Fc portion (GA101Gly) than RTX was accessible for quenching in samples from healthy controls (n = 5), rheumatoid arthritis patients (n = 16), and systemic lupus erythematosus (SLE) patients (n = 22). Each line represents an individual sample. ∗ = P < 0.05; ∗∗∗ = P < 0.0001. B, Spearman's rank correlation analysis showed a significant correlation between the percentage of surface‐accessible RTX and the percentage of B cell depletion in patients with SLE (n = 22), as assessed by whole blood B cell–depletion assay. C, The relative cytotoxicity and the ratio of the cytotoxicity index (CTI) for GA101Gly to the CTI for RTX between SLE patients with >65% surface‐accessible RTX and SLE patients with <40% surface‐accessible RTX was 2‐fold and 4‐fold, respectively, and correlated with the percentage of surface‐accessible RTX, by Spearman's rank correlation analysis. mAbs = monoclonal andibodies.

ty index (CTI) for GA101Gly to the CTI for RTX between SLE patients with >65% surface‐accessible RTX and SLE patients with <40% surface‐accessible RTX was 2‐fold and 4‐fold, respectively, and correlated with the percentage of surface‐accessible RTX, by Spearman's rank correlation analysis. mAbs = monoclonal andibodies. Interestingly, we noted a correlation between surface‐accessible RTX and the CTI for RTX (Spearman's r2 = 0.5, P < 0.05) (Figure 3B) and between the relative CTI for GA101Gly and the relative CTI for RTX (Spearman's r2 = 0.6, P < 0.05) in samples from SLE patients, but not those from healthy controls or RA patients (data not shown). The relative potency of RTX compared with GA101Gly also differed, such that in samples with >65% surface‐accessible RTX (n = 5), the mean difference in relative potency was 2‐fold, whereas in samples with <40% surface‐accessible RTX (n = 5), the mean difference in relative potency was 4‐fold (Figure 3C). We found no significant correlations between surface‐accessible GA101Gly and the CTI for GA101Gly in all groups examined (data not shown). The results therefore suggest that internalization of RTX contributes to its inferior efficiency of depletion, as assessed by whole blood B cell–depletion assay, in B cells from SLE patients.

o significant correlations between surface‐accessible GA101Gly and the CTI for GA101Gly in all groups examined (data not shown). The results therefore suggest that internalization of RTX contributes to its inferior efficiency of depletion, as assessed by whole blood B cell–depletion assay, in B cells from SLE patients. FcγRIIb facilitation of rituximab internalization RTX internalizes as part of a tripartite complex with CD20 and FcγRIIb 20, but B cell expression of FcγRIIb may be altered in SLE 33. We therefore investigated whether FcγRIIb also regulated the internalization of mAb in samples from RA and SLE patients and whether FcγRIIb internalized to a greater extent with RTX than with GA101Gly. Isolated B cells from 3 healthy controls, 9 RA patients, and 9 SLE patients were incubated for 6 hours in the presence or absence of 5 μg/ml of mAb. A significant difference in the mean fluorescence intensity of FcγRIIb (P < 0.005 for each comparison) was seen in all 3 groups (Figures 4A and B), with RTX having the greatest internalization.

3 healthy controls, 9 RA patients, and 9 SLE patients were incubated for 6 hours in the presence or absence of 5 μg/ml of mAb. A significant difference in the mean fluorescence intensity of FcγRIIb (P < 0.005 for each comparison) was seen in all 3 groups (Figures 4A and B), with RTX having the greatest internalization. Figure 4 Fcγ receptor type IIb (FcγRIIb) regulation of the internalization of rituximab (RTX). A, The mean fluorescence intensity (MFI) of FcγRIIb was significantly lower in samples incubated with RTX than in those incubated with glycosylated GA101 with an unmodified Fc portion (GA101Gly) in all samples from healthy control (HC) subjects (n = 3), rheumatoid arthritis (RA) patients (n = 9), and systemic lupus erythematosus (SLE) patients (n = 9), suggesting that RTX was internalized along with FcγRIIb. Each line represents an individual sample. B, The MFI of FcγRIIb in samples incubated with monoclonal antibodies (mAb) compared with that in samples without antibodies, expressed as a percentage of untreated samples, revealed significantly lower values in samples incubated with RTX as compared with those incubated with GA101Gly in all 3 study groups. Values are the mean ± SD. C, Blocking of FcγRIIb using AT10 (an anti‐FcγRII mAb) inhibited the internalization of RTX to a greater extent than that of GA101Gly in SLE patients (n = 11). Each line represents an individual sample. ∗ = P < 0.05; ∗∗ = P < 0.005; ∗∗∗ = P < 0.0001.

ubated with GA101Gly in all 3 study groups. Values are the mean ± SD. C, Blocking of FcγRIIb using AT10 (an anti‐FcγRII mAb) inhibited the internalization of RTX to a greater extent than that of GA101Gly in SLE patients (n = 11). Each line represents an individual sample. ∗ = P < 0.05; ∗∗ = P < 0.005; ∗∗∗ = P < 0.0001. We then examined whether blocking FcγRIIb inhibited internalization in samples from 11 patients with SLE, with or without prior incubation with AT10 (an FcγRII‐specific mAb) 30. Internalization of both RTX and GA101Gly was inhibited by AT10. However, accessible RTX was greater in samples incubated with AT10 as compared with samples without AT10 (median 61% versus 51%, respectively) whereas this difference was only modest for GA101Gly (median 78% versus 74%) (Figure 4C). Intriguingly, despite blocking FcγRIIb, the median surface‐accessible RTX was lower than that of GA101Gly (61% versus 78%). Although there was no direct correlation between the degree of inhibition of internalization with AT10 or between the MFI of FcγRIIb and the fold difference between the CTI for RTX and the CTI for GA101Gly, we noted that in the 2 samples with the greatest inhibition of mAb internalization, the CTI for GA101Gly was >4‐fold higher than that for RTX, whereas the mean for the cohort was a 2‐fold difference in CTI between the 2 mAb (data not shown). Thus, FcγRIIb facilitated the internalization of type I CD20 mAb and reduced the efficiency of deletion, albeit to a variable extent.

tion of mAb internalization, the CTI for GA101Gly was >4‐fold higher than that for RTX, whereas the mean for the cohort was a 2‐fold difference in CTI between the 2 mAb (data not shown). Thus, FcγRIIb facilitated the internalization of type I CD20 mAb and reduced the efficiency of deletion, albeit to a variable extent. Disparity in internalization of B cell–targeted mAb In addition to CD20, mAb targeting other B cell surface antigens are being explored for use in SLE, including CD19 34 and CD22 28. We therefore investigated whether the differences in the CTIs for type I and type II CD20 mAb and anti‐CD22 mAb (Figure 2B) were due to a disparity in internalization and whether FcγRIIb regulated their internalization. The median percentages of surface‐accessible mAb were 67%, 51%, 73%, 22%, and 76% for anti‐CD19, type I anti‐CD20 (RTX), type II anti‐CD20 (GA101Gly), anti‐CD22, and anti‐CD38 mAb, respectively (Figure 5). Furthermore, similar to our observations in malignant B cells 24, SLE B cells also displayed a remarkable degree of internalization of anti‐CD22 mAb, greater than that seen with RTX, whereas the other mAb (anti‐CD19, anti‐CD38, and GA101Gly) were internalized to a lesser degree. In contrast to the hierarchy of depletion with mAb, with anti‐CD22 < RTX < GA101Gly (Figure 2B), we noted a reverse hierarchy of the extent of internalization, with anti‐CD22 > RTX > GA101Gly. However, only internalization of anti‐CD20 mAb was consistently inhibited by AT10 and was therefore FcγRIIb‐dependent.

degree. In contrast to the hierarchy of depletion with mAb, with anti‐CD22 < RTX < GA101Gly (Figure 2B), we noted a reverse hierarchy of the extent of internalization, with anti‐CD22 > RTX > GA101Gly. However, only internalization of anti‐CD20 mAb was consistently inhibited by AT10 and was therefore FcγRIIb‐dependent. Figure 5 Disparity in the internalization of monoclonal antibodies (mAb) and inhibition by Fcγ receptor type IIb (FcγRIIb). There was a high rate of internalization of anti‐CD22 mAb, as assessed by the surface fluorescence–quenching assay. Internalization of rituximab (RTX; a type I anti‐CD20 mAb) and, to some extent, anti‐CD19 mAb, showed remarkable variability between samples, whereas internalization of anti‐CD38 mAb and glycosylated GA101 with an unmodified Fc portion (GA101Gly) was consistently low. Internalization of only type I and type II anti‐CD20 mAb, but not the other mAb, was significantly inhibited by anti‐FcγRII mAb (AT10). Each symbol represents an individual sample; horizontal lines show the median. ∗∗ = P < 0.005; ∗∗∗ = P < 0.0001.

ed GA101 with an unmodified Fc portion (GA101Gly) was consistently low. Internalization of only type I and type II anti‐CD20 mAb, but not the other mAb, was significantly inhibited by anti‐FcγRII mAb (AT10). Each symbol represents an individual sample; horizontal lines show the median. ∗∗ = P < 0.005; ∗∗∗ = P < 0.0001. Influence of IgD and B cell activation on the internalization of type I mAb We next examined whether there were any differences in internalization between B cell subpopulations in samples from 5 patients with SLE. In all cases, postswitched memory cells internalized significantly less RTX than did naive, preswitched, and double‐negative cells (P < 0.05 for each comparison) (Figure 6A). For GA101Gly, a significant difference was noted between postswitched cells and naive and double‐negative cells before blocking with AT10 and only in naive cells after blocking with AT10.

AUTHOR CONTRIBUTIONS 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. Reddy 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 Reddy, Isenberg, Glennie, Cragg, Leandro. Acquisition of data Reddy. Analysis and interpretation of data Reddy, Cambridge, Isenberg, Cragg, Leandro. Supporting information Supplementary Figure 1. Dose‐response experiments. We determined the optimal concentration of mAbs (0.01, 0.1, 1 and 10 μg/mL) in four Independent experiments using blood from normal healthy controls and using non‐glycomodified versions of GA101 (GA101gly) to directly assess the effects of type I versus II without the influence of afucosylation. Whole blood samples were incubated with or without RTX or GA101gly at 0.01, 0.1, 1 and 10 μg/ml and percentage B cell death measured by flow cytometric analysis after 24 h and mean of triplicate wells was used. Cytotoxicity of RTX and GA101gly were compared in healthy controls (n=4). Rituximab (RTX) lyses B cells less efficiently than GA101gly in all four samples at all four concentrations tested. The results are the means and SD. Supplementary Figure 2. Differential expression of IgD and FcγRIIb in B cell subpopulations.

ternalized significantly less RTX than did naive, preswitched, and double‐negative cells (P < 0.05 for each comparison) (Figure 6A). For GA101Gly, a significant difference was noted between postswitched cells and naive and double‐negative cells before blocking with AT10 and only in naive cells after blocking with AT10. Figure 6 Effect of IgD and B cell activation on the internalization of anti‐CD20 monoclonal antibodies (mAb) in B cell subpopulations. A, Rituximab (RTX) was internalized to a significantly lesser extent by postswitched (post‐s) memory cells (IgD–CD27+) than by the other B cell subpopulations, both before and after blocking with AT10. A significantly greater percentage of glycosylated GA101 with an unmodified Fc portion (GA101Gly) was accessible on postswitched cells than on naive (IgD+CD27–) or double‐negative (DN; IgD–CD27–) cells before blocking with AT10 and only on naive cells after blocking with AT10. Preswitched (pre‐s) cells were defined as IgD+CD27+ B cells. B, A greater percentage of RTX was accessible on the surface of IgD– B cells than on IgD+ B cells. No such difference was noted for GA101Gly. C, Internalization of RTX was not inhibited by B cell activation with anti–IgM F(ab′)2 in CD19+ B cells as compared with no treatment (NT). D, A greater percentage of RTX was accessible at 6 hours in IgD+ B cells, but not IgD– B cells, from samples incubated with anti‐IgM F(ab′)2 as compared with untreated samples. No such difference was noted for anti‐CD22 mAb. In A and B, each symbol represents an individual sample; horizontal lines show the median. In C and D, each line represents an individual sample. ∗ = P < 0.05; ∗∗ = P < 0.005. NS = not significant.

amples incubated with anti‐IgM F(ab′)2 as compared with untreated samples. No such difference was noted for anti‐CD22 mAb. In A and B, each symbol represents an individual sample; horizontal lines show the median. In C and D, each line represents an individual sample. ∗ = P < 0.05; ∗∗ = P < 0.005. NS = not significant. We also examined differences in internalization between B cell subpopulations based on the expression of IgD (IgD+ or IgD–), CD27 (CD27+ or CD27–), and CD38 (CD38low or CD38++). Again, in samples incubated with RTX, there was significantly greater internalization of RTX in IgD+ B cells than IgD– B cells, and internalization was inhibited by AT10 (Figure 6B). This may be partly due to the differential expression of FcγRIIb and IgD (Supplementary Figures 2A and B, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.39167/abstract). No such findings were observed in B cell subpopulations based on the expression of CD27 (Supplementary Figure 3A, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.39167/abstract) or CD38 (Supplementary Figure 3B). In samples incubated with GA101Gly, no differences were observed with any B cell subpopulations. We found no differences in the internalization of anti‐CD22 mAb between B cell subpopulations (n = 3) (data not shown). Thus, internalization of type I mAb, but not type II mAb or anti‐CD22 mAb, was significantly lower in postswitched cells and IgD– B cells overall.

erences were observed with any B cell subpopulations. We found no differences in the internalization of anti‐CD22 mAb between B cell subpopulations (n = 3) (data not shown). Thus, internalization of type I mAb, but not type II mAb or anti‐CD22 mAb, was significantly lower in postswitched cells and IgD– B cells overall. It has previously been reported that the expression of FcγRIIb differs between B cell subpopulations in SLE patients 33 and may therefore account for the disparity in internalization. We confirmed that the expression of FcγRIIb varied between B cell subpopulations in SLE (Supplementary Figure 2A, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.39167/abstract), with naive cells < double‐negative cells < postswitched cells < preswitched cells. In contrast to the expression of FcγRIIb, the expression of IgD on naive cells was greater than that in preswitched cells (Supplementary Figure 2B). This finding was especially of interest in conjunction with the finding that internalization of mAb was greatest in the IgD+ B cells rather than the IgD– B cells.

d cells. In contrast to the expression of FcγRIIb, the expression of IgD on naive cells was greater than that in preswitched cells (Supplementary Figure 2B). This finding was especially of interest in conjunction with the finding that internalization of mAb was greatest in the IgD+ B cells rather than the IgD– B cells. Given that internalization of mAb was higher in the IgD+ B cells and that the outcome of BCR engagement leading to either signaling or internalization has previously been shown to be mutually exclusive and dependent on the phosphorylation of tyrosine‐based motifs 35, we investigated whether B cell activation inhibited internalization of RTX or anti‐CD22 mAb. Isolated B cells were incubated with or without 25 μg/ml of anti‐IgM F(ab′)2 for 0.5 or 6 hours. Internalization of RTX, but not anti‐CD22 mAb, was inhibited by B cell activation with anti‐IgM F(ab′)2 only in IgD+ B cells (P < 0.05), but not IgD− B cells, at 6 hours (Figures 6C and D). Taken together, these results suggest independent roles for FcγRIIb and the BCR in regulating the internalization of RTX, but not GA101Gly or anti‐CD22 mAb.

t not anti‐CD22 mAb, was inhibited by B cell activation with anti‐IgM F(ab′)2 only in IgD+ B cells (P < 0.05), but not IgD− B cells, at 6 hours (Figures 6C and D). Taken together, these results suggest independent roles for FcγRIIb and the BCR in regulating the internalization of RTX, but not GA101Gly or anti‐CD22 mAb. DISCUSSION Rituximab treatment was first used at our center for the treatment of RA 36 and SLE 37. Although efficacy was demonstrated in seropositive RA 38 and despite encouraging results in several open studies 39, 2 randomized clinical trials failed to show efficacy in SLE 40, 41. We have previously discussed whether several factors, including trial design, may have contributed to the apparent lack of efficacy in these trials 42. A key factor is that rituximab fails to induce complete depletion in some patients with RA 3, 43 and SLE 4, which is associated with a poor treatment response. We have previously shown that serum rituximab levels vary remarkably in both RA and SLE patients, are higher in RA patients than in SLE patients, and are higher in patients with well‐depleted B cells than in those without, but only in RA patients and not SLE patients 44.

ch is associated with a poor treatment response. We have previously shown that serum rituximab levels vary remarkably in both RA and SLE patients, are higher in RA patients than in SLE patients, and are higher in patients with well‐depleted B cells than in those without, but only in RA patients and not SLE patients 44. Together, these findings suggested disease‐specific mechanisms of resistance to depletion, in particular for SLE, and enhancing depletion may improve clinical response. Furthermore, a better understanding of resistance mechanisms may guide selection of appropriate B cell–depleting agents. Our goal in the present study was thus to compare the in vitro efficiency of RTX and alternative CD20 mAb and to explore potential resistance mechanisms in RA and SLE. We used whole blood B cell–depletion assays to compare the type I mAb RTX and ofatumumab with the type II mAb tositumomab (B1) and GA101Gly and showed that the type II mAb were significantly more effective at depleting B cells from patients with RA and SLE. Owing to its murine IgG2a isotype, tositumomab would be expected to be less efficient at recruiting CDC and ADCC in humans as compared with the human IgG1 isotype of rituximab; however, the type II nature appears to offset the murine isotype effect, resulting in superior cytotoxicity to that of RTX in the whole blood B cell–depletion assay.

G2a isotype, tositumomab would be expected to be less efficient at recruiting CDC and ADCC in humans as compared with the human IgG1 isotype of rituximab; however, the type II nature appears to offset the murine isotype effect, resulting in superior cytotoxicity to that of RTX in the whole blood B cell–depletion assay. A wide variability in the efficiency of depletion was observed in the case of RTX in SLE patients, which correlated with the level of internalization. This suggested that internalization of RTX is a probable “resistance mechanism” in patients with SLE and may explain its variability in depletion 45. The activity of the 2 types of mAb demonstrated in vitro may not reflect their activity in vivo. However, alterations of the immune system in patients with SLE, such as defective phagocytosis 46 and natural killer cell function 47, may explain why even type II anti‐CD20 mAb failed to achieve B cell depletion in SLE patients that was comparable to that in RA patients and healthy controls. Furthermore, type I CD20 mAb (RTX) induces CDC, whereas type II mAb are poor inducers of CDC 8, and thus, the efficiency of type I CD20 mAb may be compromised in conditions with defects in complement function, such as SLE 48. We are currently investigating this possibility.

to that in RA patients and healthy controls. Furthermore, type I CD20 mAb (RTX) induces CDC, whereas type II mAb are poor inducers of CDC 8, and thus, the efficiency of type I CD20 mAb may be compromised in conditions with defects in complement function, such as SLE 48. We are currently investigating this possibility. The development of human antichimeric antibodies is more common in SLE patients 2, and this limits the repeated use of RTX. GA101, a fully humanized, Fc‐engineered type II CD20 mAb that has been shown to achieve better patient outcomes than RTX in chronic lymphocytic leukemia 19, which is now a Food and Drug Administration–approved indication 49, may potentially overcome, at least in part, these resistance mechanisms. Also under exploration for use in SLE are mAb that target B cell surface proteins other than CD20, including anti‐CD19 34 and anti‐CD22 mAb 50, which are aimed at depleting B cells and/or modulating their function. We found a differential internalization of these mAb in B cells from patients with SLE, with rapid internalization of anti‐CD22 mAb that was unaffected by FcγRIIb and with variable internalization and regulation of anti‐CD19 mAb by FcγRIIb. Internalization of mAb results in lower amounts of mAb on the target cell surface being accessible to immune effector cells 22, thereby compromising their cytotoxicity, particularly in SLE, which showed rapid internalization of RTX.

y FcγRIIb and with variable internalization and regulation of anti‐CD19 mAb by FcγRIIb. Internalization of mAb results in lower amounts of mAb on the target cell surface being accessible to immune effector cells 22, thereby compromising their cytotoxicity, particularly in SLE, which showed rapid internalization of RTX. Interestingly, internalization of RTX, but not anti‐CD22 mAb, was variable across B cell subpopulations, being low in postswitched (IgD–CD27+) memory cells and IgD– cells, which suggests reduced intrinsic resistance to depletion. Also, internalization of RTX, but not anti‐CD22 mAb, was independently inhibited both by blocking of FcγRIIb and by B cell activation in IgD+ B cells. Taken together, these results suggest that FcγRIIb and BCR activation influence the internalization of type I anti‐CD20 mAb, but not anti‐CD22 mAb. Thus, distinct mechanisms operate to facilitate the internalization of different mAb. The differences in internalization between antigen‐specific mAb may be related to the constitutive endocytosis of the target antigen, as for CD22 51, or to the redistribution of CD20 into lipid rafts after incubation with RTX 22. Knowledge of the factors that influence internalization of mAb could be exploited to refine B cell–targeting strategies in autoimmune diseases such as RA and SLE.

elated to the constitutive endocytosis of the target antigen, as for CD22 51, or to the redistribution of CD20 into lipid rafts after incubation with RTX 22. Knowledge of the factors that influence internalization of mAb could be exploited to refine B cell–targeting strategies in autoimmune diseases such as RA and SLE. In conclusion, our results provide strong preclinical evidence for considering the use of mechanistically different type II CD20 mAb such as GA101 as alternative B cell–depleting agents for the treatment of RA and SLE. We have also identified distinct mechanisms of internalization of rituximab and its regulation, which may explain the variability in B cell depletion noted in patients with SLE. AUTHOR CONTRIBUTIONS 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. Reddy 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 Reddy, Isenberg, Glennie, Cragg, Leandro. Acquisition of data Reddy. Analysis and interpretation of data Reddy, Cambridge, Isenberg, Cragg, Leandro.

The horizontal line represents the median; the box, interquartile range; the whiskers, 10‐90th percentile; and the dots represent outliers. (B) Naïve cells expressed significantly higher levels of IgD compared with pre‐switched cells, the results represent the mean and SD, in contrast to the expression of FcγRIIb (A). Supplementary Figure 3. Internalization of anti‐CD20 monoclonal antibodies (mAbs) in B cell subpopulations. (A) B cell subpopulations were categorized based on the expression of CD27 and CD38. B cell subpopulations were characterized based on the expression of CD27: CD27+ or CD27‐; or the expression of CD38: CD38lo or CD38++. Surface fluorescence quenching assay was performed using enriched B cells from patients with systemic lupus erythematosus (SLE) (n=5). There was no significant difference between CD27+ and CD27‐ subpopulations in the amount of internalization of RTX or GA101gly. The horizontal line represents the median. (B) Similarly, there was no significant difference between in internalization of RTX or GA101gly between CD38lo or CD38++ B cell subpopulations. Supplementary Table 1. Demographics of patients with Rheumatoid Arthritis Supplementary Table 2. Demographics of patients with Systemic Lupus Erythematosus Supplementary Table 3. Efficiency of anti‐CD20 mAbs and frequency of B cell phenotypes of patients with Rheumatoid Arthritis and Systemic Lupus Erythematosus Click here for additional data file.

Supplementary Table 1. Demographics of patients with Rheumatoid Arthritis Supplementary Table 2. Demographics of patients with Systemic Lupus Erythematosus Supplementary Table 3. Efficiency of anti‐CD20 mAbs and frequency of B cell phenotypes of patients with Rheumatoid Arthritis and Systemic Lupus Erythematosus Click here for additional data file. ACKNOWLEDGMENTS The authors wish to thank Pauline Buck, Samantha Moore, Lindsay Kidd, Nicola Whitbread, Nicola Daly, Emma Ross, and Dean Heathcote for their help with obtaining blood samples from the study participants.

Rheumatoid arthritis (RA) is a systemic inflammatory condition that primarily affects synovial joints. It is characterized by persistent synovitis and destruction of bone and cartilage. RA affects ∼1% of the adult population, with a higher prevalence in the population over 60 years of age (2%) and a 3‐fold higher incidence in women 1. While the cause of the disease remains incompletely understood, it is known that proinflammatory cytokines play a role in its pathogenesis by sustaining inflammation, which leads to joint destruction 2. Key cytokines in the development of RA include tumor necrosis factor (TNF), interleukin‐1β (IL‐1β), and IL‐6. These cytokines can stimulate the production of matrix metalloproteinase (MMP) enzymes, destroying the extracellular matrix and leading to cartilage and bone damage 3. Collagenases MMP‐1 and MMP‐13 play a significant role in RA, as they are shown to be the rate‐limiting step in the process of collagen degradation 4.

6. These cytokines can stimulate the production of matrix metalloproteinase (MMP) enzymes, destroying the extracellular matrix and leading to cartilage and bone damage 3. Collagenases MMP‐1 and MMP‐13 play a significant role in RA, as they are shown to be the rate‐limiting step in the process of collagen degradation 4. In recent years, the availability of biologic drugs has revolutionized the field of RA treatment. Nonetheless, the disease continues to be linked to severe pain, depression, and impaired function, with 20–40% of patients failing to respond to current therapy 5, 6. The cost of treating RA with biologic agents is far higher than the cost of “conventional” disease‐modifying antirheumatic drugs (DMARDs) and continues to be linked to negative consequences of organ toxicity 7. Targeting TNF with monoclonal antibodies such as adalimumab (Humira; AbbVie) and infliximab (Remicade; Janssen Biologics), either alone or in combination with other DMARDs, has become the gold standard for RA therapy 8. While TNF has a highly deleterious effect in inflammatory joint diseases, it plays a vital role in the body's defenses against infection 9. In the immune response to Mycobacterium tuberculosis, TNF plays a critical protective role, leading to macrophage activation, cell recruitment, granuloma formation, and maintenance of granuloma integrity 10, 11, 12. Thus, systemic blockade of TNF increases the risk of tuberculosis infection and reactivation in patients with latent disease, as compared to alternative DMARD therapy 13. Although the exact mechanism behind the high number of nonresponders to anti‐TNF biologic therapy is not clear 14, it is plausible to hypothesize that a lack of efficacy may be due to suboptimal TNF blockade at sites of inflammation, which cannot be improved by increased systemic administration because of potential general toxicity when taken at doses higher than those recommended 15, 16.

anti‐TNF biologic therapy is not clear 14, it is plausible to hypothesize that a lack of efficacy may be due to suboptimal TNF blockade at sites of inflammation, which cannot be improved by increased systemic administration because of potential general toxicity when taken at doses higher than those recommended 15, 16. A possible solution would be to develop new agents with dual specificity, in which one domain targets molecules expressed at high levels (e.g., cell adhesion molecules) at the site of disease (the synovium in the case of RA) and the other domain targets the cytokine of interest (e.g., TNF). Bispecific agents, such as dual variable‐domain immunoglobulins (DVD‐Ig) 17, could theoretically deliver higher local concentrations with lower systemic exposure. In this format, the variable domains of 2 distinct monoclonal antibodies are linked, creating a tetravalent, dual‐targeting single agent. While it has been shown that viable DVD‐Ig molecules can be identified through optimization of an antibody pair, antibody variable domain orientation, and linkers, an ongoing limitation of the technology is the lower binding affinity observed by the “inner domain” as compared to the “outer domain.” Several bodies of work have investigated the possibility of increasing the viability of the inner domain by using variable linkers, and it has been suggested that each antibody needs to be optimized individually in terms of inner/outer domain arrangement and linker length construction to derive the best molecule 18, 19.

everal bodies of work have investigated the possibility of increasing the viability of the inner domain by using variable linkers, and it has been suggested that each antibody needs to be optimized individually in terms of inner/outer domain arrangement and linker length construction to derive the best molecule 18, 19. By reversing the concept, the intrinsic reduced activity of the inner domain represents an advantage in the design of an antibody‐based “prodrug”‐targeting molecule, with an outer domain capable of targeting the inflamed synovium and an inner domain that binds TNF. We hypothesized that by varying the linker length, it would be possible to reduce the binding affinity of the inner domain to TNF in its circulating unbound form while maintaining the specificity of the outer domain for the target of interest. This, coupled with further engineering of the linker to contain an MMP‐cleavable sequence, would allow a fully functional antibody to be released and to act locally at the site of inflammation. Such a molecule, which we identify as an activatable DVD (aDVD), would have the benefits of reducing systemic toxicity while increasing the therapeutic dosage available at sites of disease, thus improving its therapeutic index.

w a fully functional antibody to be released and to act locally at the site of inflammation. Such a molecule, which we identify as an activatable DVD (aDVD), would have the benefits of reducing systemic toxicity while increasing the therapeutic dosage available at sites of disease, thus improving its therapeutic index. MATERIALS AND METHODS Cloning and expression of DVD antibodies The sequences of the variable regions of anti‐human ICAM‐1 antibody and human TNF have been previously described 20, 21. Sequence data management was performed using Serial Cloner 2.6. Variable sequences were generated by gene synthesis (GenScript) and combined into various constructs using overlapping‐extension polymerase chain reaction (PCR) 22. The PCR products were cloned into the AbVec‐hIgG1 and AbVec‐hIgK vectors 23 using the restriction sites Age I–Sal I and Age I–Bsi WI, respectively. Clones were sequence‐verified prior to protein expression.

(GenScript) and combined into various constructs using overlapping‐extension polymerase chain reaction (PCR) 22. The PCR products were cloned into the AbVec‐hIgG1 and AbVec‐hIgK vectors 23 using the restriction sites Age I–Sal I and Age I–Bsi WI, respectively. Clones were sequence‐verified prior to protein expression. Twenty‐four hours before transfection, vectors encoding the heavy and light chains of the DVD antibody were transfected into HEK 293T cells in Dulbecco's modified Eagle's medium (DMEM) containing 10% fetal bovine serum (FBS), 100 units/ml of penicillin, 100 μg/ml of streptomycin, and 0.5 mg/ml of Geneticin. Transfection was performed with JetPrime reagent (Polyplus) according to the manufacturer's protocol. The antibodies were purified from the supernatant via affinity chromatography using protein A–Sepharose CL‐4B (GE Healthcare). DVD antibodies were biotinylated using an EZ‐Link Sulfo‐NHS‐SS biotinylation kit (Thermo‐Fisher Scientific) according to the manufacturer's protocol. MMP enzymatic digestion Antibodies were incubated at 37°C at a concentration of 100 μg/ml with 35 units of recombinant MMP‐1 enzyme (Enzo Life Sciences) in 50 mM Tris, 0.15M NaCl, 10 mM CaCl2, 50 mM ZnCl2, and 0.02% Brij35. Antibodies used for kinetic analysis were digested for 1 hour at 37°C. Digestion with RA synovial fluid (SF) and RA serum was performed by incubating 500 ng of biotinylated antibody in 200 μl of fluid at 37°C for 24–72 hours in the presence of 20 μM GM6001 (MMP inhibitor).

l, 10 mM CaCl2, 50 mM ZnCl2, and 0.02% Brij35. Antibodies used for kinetic analysis were digested for 1 hour at 37°C. Digestion with RA synovial fluid (SF) and RA serum was performed by incubating 500 ng of biotinylated antibody in 200 μl of fluid at 37°C for 24–72 hours in the presence of 20 μM GM6001 (MMP inhibitor). Protein characterization Protein purity and molecular weight were assessed by resolution in sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS‐PAGE) reducing gels using Mini‐Protean 4–20% TGX gels (Bio‐Rad) followed by Sypro Ruby protein gel stain according to the manufacturer's instructions. Western blot analysis of antibodies digested with RA SF and serum was performed via nitrocellulose transfer. Biotinylated antibody heavy and light chains were detected using streptavidin–horseradish peroxidase (HRP).

X gels (Bio‐Rad) followed by Sypro Ruby protein gel stain according to the manufacturer's instructions. Western blot analysis of antibodies digested with RA SF and serum was performed via nitrocellulose transfer. Biotinylated antibody heavy and light chains were detected using streptavidin–horseradish peroxidase (HRP). Quantification of anti‐TNF activity Enzyme‐linked immunosorbent assay (ELISA) of anti‐TNF activity was performed in 96‐well plates (Thermo‐Fisher Scientific) that had been coated overnight at 4°C with 100 ng/ml of TNF in phosphate buffered saline (PBS). Plates were blocked for 2 hours at room temperature with PBS/2% bovine serum albumin and then incubated with serial dilutions of DVD antibody. Bound antibodies were detected with HRP‐conjugated anti‐human IgG antibody (Jackson Immunotools). Plates were then incubated with tetramethylbenzidine substrate (GE Healthcare), and reactions were stopped with 1N H2SO4. Optical absorption was measured at 450 nm. The 20% effective concentration (EC20) was calculated using a dose‐response nonlinear‐fit curve in GraphPad Prism v5.

n IgG antibody (Jackson Immunotools). Plates were then incubated with tetramethylbenzidine substrate (GE Healthcare), and reactions were stopped with 1N H2SO4. Optical absorption was measured at 450 nm. The 20% effective concentration (EC20) was calculated using a dose‐response nonlinear‐fit curve in GraphPad Prism v5. Inhibition of TNF‐induced cytotoxicity was conducted in the L929 cell line. Briefly, 3 × 104 cells were seeded in 96‐well plates for 18 hours at 37°C in 100 μl of DMEM supplemented with 10% FBS, 100 units/ml of penicillin, 100 μg/ml of streptomycin, and a 10 μM concentration of the MMP inhibitor GM6001. The medium was then replaced with 100 μl of complete medium with 1 μg/ml of actinomycin D and 0.45 ng/ml of either TNF (Sigma) or TNF plus the antibody of interest (1:2 serial dilutions), and this was incubated for 24 hours at 37°C. We added 500 μg/ml of thiazolyl blue tetrazolium bromide in PBS (Sigma) to the wells and incubated them for 3 hours at 37°C. Medium was then removed, and the cells were resuspended in 100 μl of 90% isopropanol 10% DMSO for 15 minutes. Optical absorption was measured at 595 nm. The percentage of viable cells was calculated as follows: (OD595 nm × 100)/OD595 nm of sample without TNF. The 20% inhibition concentration (IC20) was determined using a dose‐response nonlinear‐fit curve in GraphPad Prism v5.

100 μl of 90% isopropanol 10% DMSO for 15 minutes. Optical absorption was measured at 595 nm. The percentage of viable cells was calculated as follows: (OD595 nm × 100)/OD595 nm of sample without TNF. The 20% inhibition concentration (IC20) was determined using a dose‐response nonlinear‐fit curve in GraphPad Prism v5. Surface plasmon resonance (SPR) SPR experiments were performed with a Biacore T200 instrument using HBS‐P+ as the running and dilution buffer (GE Healthcare Bio‐Sciences). BIAevaluation software version 2.0 (GE Healthcare) was used for data processing. For determination of the binding kinetics, mouse anti‐human IgG (GE Healthcare) was covalently coupled to a CM5 Sensor Chip (GE Healthcare). Human antibody or DVD antibody was captured, and various concentrations of interaction partner protein were injected over the flow cell at a flow rate of 30 μl/minute. A double‐reference subtraction was performed using buffer alone. Kinetic rate constants were obtained by curve fitting according to a 1:1 Langmuir binding model.

ibody or DVD antibody was captured, and various concentrations of interaction partner protein were injected over the flow cell at a flow rate of 30 μl/minute. A double‐reference subtraction was performed using buffer alone. Kinetic rate constants were obtained by curve fitting according to a 1:1 Langmuir binding model. Immunohistochemical analysis Formalin‐fixed paraffin‐embedded tissue sections were dewaxed, and antigens were retrieved after 10 minutes of boiling in citrate buffer, pH 6 (Dako). Slides were stained with 10 μg/ml of biotinylated DVD antibodies for 1 hour at room temperature and were visualized with streptavidin–HRP complex using 3,3′‐diaminobenzidine chromogen (Dako). Rabbit anti–ICAM‐1 IgG (Abcam) was detected with HRP‐conjugated goat anti‐rabbit IgG (Jackson Immunotools). Mouse anti–von Willebrand factor (Dako) and mouse anti‐CD31 (R&D Systems) were used to identify human vascular endothelial cells, which were revealed with HRP‐conjugated goat anti‐mouse IgG (Santa Cruz Biotechnology). Sections were counterstained with hematoxylin, mounted with Depex mounting medium (Dako), and acquired with a CellSens imaging system (Olympus).

d mouse anti‐CD31 (R&D Systems) were used to identify human vascular endothelial cells, which were revealed with HRP‐conjugated goat anti‐mouse IgG (Santa Cruz Biotechnology). Sections were counterstained with hematoxylin, mounted with Depex mounting medium (Dako), and acquired with a CellSens imaging system (Olympus). RESULTS Design and cleavage of an activatable dual variable‐domain antibody To create a bispecific antibody format with therapeutic activity in RA and targeting capacity for the inflamed synovium, the gold standard for anti‐TNF biologic drugs, adalimumab, was coupled with an ICAM‐1–targeting antibody, using an adaptation of the well‐established DVD‐Ig format 17. The construct we describe contains the anti–ICAM‐1 VL and VH domains linked to the light chain and heavy chain, respectively, of the anti‐TNF drug adalimumab via a small peptide linker (Figure 1A). To create a DVD bispecific antibody with impaired binding capacity for the internal variable domain, a series of linkers with various lengths and amino acid compositions were designed to test for the desired activity (Figure 1A). The long linker was derived from a natural linker found in human IgG antibodies and was previously described in the context of the DVD‐Ig format 24. Reducing the linker length can substantially alter the kinetic properties of the internal binding domain 24.

ns were designed to test for the desired activity (Figure 1A). The long linker was derived from a natural linker found in human IgG antibodies and was previously described in the context of the DVD‐Ig format 24. Reducing the linker length can substantially alter the kinetic properties of the internal binding domain 24. Figure 1 Structure and characterization of activatable dual variable‐domain (aDVD) antibodies. A, Schematic representation of the general structure of DVD constructs with an anti–intercellular adhesion molecule 1 (anti–ICAM‐1) outer domain linked to the anti–tumor necrosis factor (anti‐TNF) antibody adalimumab. Variable heavy (VH), constant heavy (CH), variable light (VL), and constant light (CL) chain regions are indicated. Linker length and amino acid composition are summarized in the table, with the matrix metalloproteinase (MMP)–cleavable sequence in boldface and the cutting position marked with a slash. HC = heavy chain; LC = light chain. B, Time course of aDVD antibody cleavage with recombinant MMP as resolved by sodium dodecyl sulfate–polyacrylamide gel electrophoresis. The gel shows a gradual conversion from DVD heavy chain to IgG heavy chain as a result of cleavage and removal of the outer anti–ICAM‐1 variable region. Similar processing was detected on the light chain (results not shown). C, Western blot analysis under reducing conditions. The time‐dependent increase in IgG heavy chain content due to cleavage of the biotinylated aDVD antibody carrying the short MMP‐cleavable linker, following incubation with rheumatoid arthritis (RA) synovial fluid (SF) at 37°C, can be seen. No cleavage was detected at 72 hours for the antibody carrying the scrambled MMP linker, nor was any cleavage detected upon incubation with RA serum or incubation in the presence of the MMP inhibitor GM6001.

‐cleavable linker, following incubation with rheumatoid arthritis (RA) synovial fluid (SF) at 37°C, can be seen. No cleavage was detected at 72 hours for the antibody carrying the scrambled MMP linker, nor was any cleavage detected upon incubation with RA serum or incubation in the presence of the MMP inhibitor GM6001. We hypothesized that short linkers could impair the accessibility of the ligand to the internal domain in such a way that could be reverted upon cleavage of the internal linker, thus forming an activatable DVD prodrug. The remaining 4 linkers contained an MMP‐cleavable site (PLGLWA) 25, either alone or in the presence of G4S‐flanking regions, and a scrambled MMP‐cleavable sequence (AGPLLW). To test the ability of the MMP enzyme to access, cleave, and activate the internal anti‐TNF domain, the aDVD constructs where incubated with physiologically relevant concentrations of recombinant MMP enzyme. Analysis of the digested aDVD constructs by reduced SDS‐PAGE (Figure 1B) showed rapid processing of the aDVD carrying the MMP cleavable site, with the formation of molecular weight products consistent with an IgG format.

constructs where incubated with physiologically relevant concentrations of recombinant MMP enzyme. Analysis of the digested aDVD constructs by reduced SDS‐PAGE (Figure 1B) showed rapid processing of the aDVD carrying the MMP cleavable site, with the formation of molecular weight products consistent with an IgG format. Incubation of the aDVD construct carrying the short MMP linker (PLGLWA) with SF from RA patients also showed time‐dependent activation of the construct, confirming the processing capacity under physiologic conditions (Figure 1C). Activation using SF was less efficient than that observed with recombinant protein. This may be due to saturation of MMP activity in ex vivo assays, which one would not anticipate to occur in vivo during chronic inflammation, where MMP expression in the synovial tissue is expected to be higher than in the surrounding SF 26. Additionally, the cleavage could be inhibited by the MMP inhibitor GM6001, while no cleavage could be detected for the aDVD carrying the scrambled MMP linker, further confirming MMP‐mediated activation of the constructs.

where MMP expression in the synovial tissue is expected to be higher than in the surrounding SF 26. Additionally, the cleavage could be inhibited by the MMP inhibitor GM6001, while no cleavage could be detected for the aDVD carrying the scrambled MMP linker, further confirming MMP‐mediated activation of the constructs. Impaired binding of aDVD to TNF and rescue by MMP cleavage To be effective as a targeting prodrug, it is important that the aDVD molecules retain their ability to bind to their target antigen via the outer binding domain, while the inner domain remains shielded. The binding of aDVD molecules to ICAM‐1 (outer domain) and TNF (inner domain) was investigated via ELISA. The uncut aDVD molecules retained their binding to ICAM‐1 to the same extent as the parent anti–ICAM‐1 antibody (data not shown). However, before MMP cleavage, the molecules showed a 275‐fold reduction in binding to TNF as compared to adalimumab IgG. Binding to TNF was fully rescued for all the constructs following MMP cleavage (Figure 2A).

es retained their binding to ICAM‐1 to the same extent as the parent anti–ICAM‐1 antibody (data not shown). However, before MMP cleavage, the molecules showed a 275‐fold reduction in binding to TNF as compared to adalimumab IgG. Binding to TNF was fully rescued for all the constructs following MMP cleavage (Figure 2A). Figure 2 Anti‐TNF activity of DVD antibodies. A, TNF binding capacity for DVD antibodies (Ab) (scrambled MMP and long linker) and aDVD antibodies (MMP, G4S‐MMP‐G4S, and [G4S]2‐MMP‐[G4S]2), as determined by enzyme‐linked immunosorbent assay. Reduced binding capacity was detected for uncut aDVD constructs, while full potency could be restored upon cleavage with MMP enzyme, as compared with adalimumab IgG. B, Neutralization of TNF‐induced cytotoxicity in the L929 cell functional assay. Neutralization of cytotoxicity was impaired in uncut aDVD antibodies, with stronger impairment for ICAM‐MMP‐adalimumab antibody. Potencies similar to that of adalimumab IgG were obtained following MMP enzymatic digestion. Results in A are expressed as the 20% effective concentration (EC20) and results in B as the 20% inhibition concentration (IC20), corresponding to the dose necessary to obtain 20% of the activity. Values are the mean ± SEM. See Figure 1 for other definitions.

limumab IgG were obtained following MMP enzymatic digestion. Results in A are expressed as the 20% effective concentration (EC20) and results in B as the 20% inhibition concentration (IC20), corresponding to the dose necessary to obtain 20% of the activity. Values are the mean ± SEM. See Figure 1 for other definitions. We used the L929 assay to assess the ability of the aDVD constructs to inhibit ligand binding to its receptor and to prevent TNF‐induced cytotoxicity. The ability of the uncleaved aDVD construct to block and inhibit TNF was severely impaired, consistent with the binding data obtained by ELISA. The uncleaved aDVD antibodies showed up to a 132‐fold increase in the IC20 as compared to adalimumab IgG, while cleavage with MMP completely rescued the inhibitory capacity (Figure 2B). As expected, the short MMP‐cleavable linker (PLGLWA) was characterized by a greater TNF binding impairment and was further validated using SPR.

aved aDVD antibodies showed up to a 132‐fold increase in the IC20 as compared to adalimumab IgG, while cleavage with MMP completely rescued the inhibitory capacity (Figure 2B). As expected, the short MMP‐cleavable linker (PLGLWA) was characterized by a greater TNF binding impairment and was further validated using SPR. We also compared the affinities of the restricted and processed forms of the aDVD molecule (Table 1). Binding of TNF to the uncleaved molecule was greatly impaired, as demonstrated by a 365‐fold reduction in the affinity constant (K D). Observation of the kinetics of binding indicated that the difference in affinity was predominantly driven by a reduction in the association rate constant (K a), which was 189‐fold lower than that for the uncleaved molecule, whereas the dissociation rate constant (K d) was largely unchanged (Figure 3A and Table 1). This result demonstrated that blocking of the external domain acts primarily by inhibiting association through steric hindrance; however, once bound, the antibody retains similar binding characteristics, indicating that the internal domain has not been modified and remains fully functional. Importantly, cleaved aDVD molecules showed not only identical affinity, but also identical component kinetics of binding to the parent adalimumab antibody (Table 1). In both cases, the binding kinetics were in good agreement with previously reported data 27.

al domain has not been modified and remains fully functional. Importantly, cleaved aDVD molecules showed not only identical affinity, but also identical component kinetics of binding to the parent adalimumab antibody (Table 1). In both cases, the binding kinetics were in good agreement with previously reported data 27. Figure 3 Analysis of the antigen‐binding kinetics of aDVD. A, Surface plasmon resonance sensorgrams showing the binding kinetics of adalimumab IgG, aDVD, and recombinant MMP–cleaved aDVD to TNF. The reduced binding capacity of aDVD for TNF could be reverted following digestion with MMP enzyme, restoring its full binding potential as compared to adalimumab IgG (see Table 1 for kinetic measurements). TNF concentrations were 20 nM (red), 8 nM (yellow), 3.2 nM (green), 1.28 nM (blue), and 0.512 nM (purple). B, Dynamic binding kinetics for TNF and ICAM‐1. When the aDVD antibody had been saturated with TNF of limited binding capacity, the second antigen was injected, showing retention of the ICAM‐1 specificity in the presence of TNF. On‐chip digestion of the construct with recombinant MMP enzyme was sufficient to cleave the antibody, releasing the outer domain and the coupled ICAM‐1 antigen. Injection of TNF at the end highlights the restored binding potency of the internal anti‐TNF domain. See Figure 1 for definitions. Table 1 Surface plasmon resonance kinetic measurements of aDVD antibodiesa Molecule Substrate K a, 1/msec K d, 1/second K D, pM

Figure 3 Analysis of the antigen‐binding kinetics of aDVD. A, Surface plasmon resonance sensorgrams showing the binding kinetics of adalimumab IgG, aDVD, and recombinant MMP–cleaved aDVD to TNF. The reduced binding capacity of aDVD for TNF could be reverted following digestion with MMP enzyme, restoring its full binding potential as compared to adalimumab IgG (see Table 1 for kinetic measurements). TNF concentrations were 20 nM (red), 8 nM (yellow), 3.2 nM (green), 1.28 nM (blue), and 0.512 nM (purple). B, Dynamic binding kinetics for TNF and ICAM‐1. When the aDVD antibody had been saturated with TNF of limited binding capacity, the second antigen was injected, showing retention of the ICAM‐1 specificity in the presence of TNF. On‐chip digestion of the construct with recombinant MMP enzyme was sufficient to cleave the antibody, releasing the outer domain and the coupled ICAM‐1 antigen. Injection of TNF at the end highlights the restored binding potency of the internal anti‐TNF domain. See Figure 1 for definitions. Table 1 Surface plasmon resonance kinetic measurements of aDVD antibodiesa Molecule Substrate K a, 1/msec K d, 1/second K D, pM Adalimumab IgG TNF 2.416 × 106 1.192 × 10−4 49.3 aDVD ICAM‐MMP‐adalimumab Cleaved TNF 2.047 × 106 1.087 × 10−4 53.1 Uncleaved TNF 1.08 × 104 1.83 × 10−4 18,000 Infliximab IgG TNF 4.04 × 106 2.5 × 10−4 61.8 aDVD ICAM‐MMP‐infliximab Cleaved TNF 2.98 × 106 3.66 × 10−4 123 Uncleaved TNF 3.8 × 103 6 × 10−4 158,000 a Activatable dual variable‐domain (aDVD) antibody constructs targeting intercellular adhesion molecule 1 (ICAM‐1) and the anti–tumor necrosis factor (anti‐TNF) antibodies adalimumab and infliximab were designed with matrix metalloproteinase (MMP)–cleavable linkers as described in Materials and Methods. K a = association rate constant; K d = dissociation rate constant; K D = affinity constant.

rcellular adhesion molecule 1 (ICAM‐1) and the anti–tumor necrosis factor (anti‐TNF) antibodies adalimumab and infliximab were designed with matrix metalloproteinase (MMP)–cleavable linkers as described in Materials and Methods. K a = association rate constant; K d = dissociation rate constant; K D = affinity constant. In order to target proteins in a disease setting, the aDVD needs to maintain its cleavage capacity in the presence of both targeting and effector antigens, as is likely to be the scenario in the cytokine‐rich environment of the inflamed synovium. This is particularly pertinent as the aDVD is still capable of binding to TNF with a slow dissociation rate, which could conceivably block the cleavage site by steric hindrance (Figure 3A). In order to observe whether the aDVD molecule could be cleaved and activated in this environment, the molecule was immobilized on an SPR sensor chip, and saturating concentrations of TNF were injected, followed by ICAM‐1, prior to MMP cleavage on the sensor surface (Figure 3B). TNF showed the same restricted level of binding as had previously been demonstrated with the uncut material. ICAM‐1, however, was capable of binding to the molecule in the presence of TNF, as demonstrated by the observed change in response units, which was of the same magnitude as ICAM‐1 injected onto free antibody at the same concentration (data not shown).

binding as had previously been demonstrated with the uncut material. ICAM‐1, however, was capable of binding to the molecule in the presence of TNF, as demonstrated by the observed change in response units, which was of the same magnitude as ICAM‐1 injected onto free antibody at the same concentration (data not shown). In the presence of both saturating concentrations of ICAM‐1 and TNF, MMP enzyme was injected over a period of 30 minutes. Following chip cleavage and a period of stabilization to remove unbound material from the chip surface, the chip was rechallenged with TNF. Postcleavage, the TNF binding capacity was rescued, as demonstrated by the enhanced change in response units, which was measured at the same level as the injected concentration on the unrestricted antibody (Figures 3A and B).

lization to remove unbound material from the chip surface, the chip was rechallenged with TNF. Postcleavage, the TNF binding capacity was rescued, as demonstrated by the enhanced change in response units, which was measured at the same level as the injected concentration on the unrestricted antibody (Figures 3A and B). Tissue‐specific targeting using the aDVD antibody platform One of the key characteristics of the aDVD format is the ability to present the anti‐TNF therapeutic function in a prodrug format that can be activated following encounter with MMP enzymes at the site of arthritic inflammation. The presence of the outer variable‐domain–targeting ICAM‐1, an integrin overexpressed in inflammatory conditions such as RA 28, 29, would allow the preferential accumulation of antibody in the target tissue, facilitating the encounter with proteolytic enzymes. MMP cleavage causes the removal of the anti–ICAM‐1 external domain, resulting in loss of ICAM‐1 specificity (data not shown). To test the ability of the aDVD to retain tissue‐targeting capacity when in full conformation, we examined its reactivity with the microvasculature in samples of human synovium obtained from 3 RA patients, 3 osteoarthritis patients, and a patient without arthritis (normal synovium) via immunohistochemistry (Figure 4).

To test the ability of the aDVD to retain tissue‐targeting capacity when in full conformation, we examined its reactivity with the microvasculature in samples of human synovium obtained from 3 RA patients, 3 osteoarthritis patients, and a patient without arthritis (normal synovium) via immunohistochemistry (Figure 4). Figure 4 Human synovial tissue reactivity of aDVD antibodies. The reactivity of the ICAM‐1‐MMP‐adalimumab construct with RA, osteoarthritis (OA), and nonarthritic (normal) human synovial tissues was examined using immunohistochemistry. Bound biotinylated aDVD antibodies were detected using streptavidin–horseradish peroxidase complex and compared to the staining pattern of anti–ICAM‐1 IgG. The presence of blood vessels was depicted using anti–von Willebrand factor (anti‐vWF) antibody in combination with anti‐CD31 antibody. Biotinylated human IgG was used as the negative control. Bars = 100 μm. See Figure 1 for other definitions. The ICAM‐MMP‐adalimumab aDVD was able to selectively target the human inflamed synovium from both the RA patients and the OA patients with similar efficacy as compared to an anti–ICAM‐1 IgG antibody. Importantly, no detectable reactivity was identified in the normal synovium sample from a patient without arthritis. The specificity for arthritic synovium further strengthens the potential of the aDVD for use in targeted drug delivery in RA.

OA patients with similar efficacy as compared to an anti–ICAM‐1 IgG antibody. Importantly, no detectable reactivity was identified in the normal synovium sample from a patient without arthritis. The specificity for arthritic synovium further strengthens the potential of the aDVD for use in targeted drug delivery in RA. Furthermore, the aDVD format may represent a flexible platform for targeted delivery of prodrugs that can be easily adapted to other cytokines and to other disease conditions with a simple exchange of the outer targeting domain.

OA patients with similar efficacy as compared to an anti–ICAM‐1 IgG antibody. Importantly, no detectable reactivity was identified in the normal synovium sample from a patient without arthritis. The specificity for arthritic synovium further strengthens the potential of the aDVD for use in targeted drug delivery in RA. Furthermore, the aDVD format may represent a flexible platform for targeted delivery of prodrugs that can be easily adapted to other cytokines and to other disease conditions with a simple exchange of the outer targeting domain. Improvement in the structural design of aDVD molecules with knowledge of the molecular interactions Since the reduced binding of aDVD molecules can be mediated by blocking of the internal domain, we hypothesized that further inhibition of binding could be predicted through knowledge of the interaction between TNF and the inner domain antibody. The crystal structures of adalimumab and infliximab in complex with TNF have recently been reported 30, 31. Crystal structure data showed that adalimumab bound to trimeric TNF via a broader binding interface, with a total buried surface area of 2,540 Å2 31, while infliximab bound to the TNF trimer via a reduced binding interface of 1,977 Å2 (Figure 5A). Additionally, adalimumab engages the TNF trimer through interactions with 2 monomers of the trimer, while the binding of infliximab is mediated almost exclusively through the loop region of a single TNF monomer. We therefore predicted that the smaller interaction surface area in the infliximab–TNF complex would translate to a binding interface that would be more readily blocked by the outer domain. To test this, an aDVD molecule was engineered with infliximab as the inner binding domain and was tested for binding and functionality.

erefore predicted that the smaller interaction surface area in the infliximab–TNF complex would translate to a binding interface that would be more readily blocked by the outer domain. To test this, an aDVD molecule was engineered with infliximab as the inner binding domain and was tested for binding and functionality. Figure 5 Molecular interactions as determinants of aDVD antibody (Ab) inhibitory properties. A, Schematic representation of the interaction of infliximab and adalimumab with TNF. There is a reduced contact area (red) for infliximab as compared to that for adalimumab, which may predict an increased binding inhibition in the aDVD format. Adapted, with permission, from ref. 31 (© 2013 The American Society for Biochemistry and Molecular Biology). B, TNF binding capacity for the aDVD ICAM‐MMP‐infliximab construct. Binding capacities before and after MMP cleavage were compared to that of infliximab, as determined by enzyme‐linked immunosorbent assay. A 2,500‐fold inhibition for the infliximab aDVD construct compared to infliximab was observed, with rescue of the binding upon cleavage of the construct with MMP. C, Neutralization of TNF‐induced cytotoxicity in the L929 cell functional assay. There was a complete loss of function of the ICAM‐MMP‐infliximab construct as compared to infliximab. Results in B are expressed as the 20% effective concentration (EC20) and results in C as the 20% inhibition concentration (IC20). Values are the mean ± SEM. D, Binding kinetics of aDVD ICAM‐MMP‐infliximab and infliximab to TNF. Results show the binding of 20 nM TNF to infliximab and ICAM‐MMP‐infliximab coupled to the sensor surface at the same density. See Figure 1 for other definitions.

nd results in C as the 20% inhibition concentration (IC20). Values are the mean ± SEM. D, Binding kinetics of aDVD ICAM‐MMP‐infliximab and infliximab to TNF. Results show the binding of 20 nM TNF to infliximab and ICAM‐MMP‐infliximab coupled to the sensor surface at the same density. See Figure 1 for other definitions. Infliximab bound to TNF with an EC20 of 0.004 nM, while the aDVD‐infliximab bound with an EC20 of 21.6 nM. Once processed by MMP cleavage the antibody demonstrated binding that was comparable to the original infliximab antibody (Figure 5B). The fold difference between cleaved and uncleaved aDVD‐infliximab antibody was 3,000, which was 10‐fold higher than the difference measured for the aDVD‐adalimumab construct. The ability of the antibody to inhibit TNF binding to its receptor in L929 cell functional studies was also greatly diminished, as no anti‐TNF functionality could be detected for the uncleaved aDVD‐infliximab over the range of concentrations tested, while activity was fully rescued after linker cleavage (Figure 5C). SPR data demonstrated a 2,500‐fold reduction in the K D that was predominantly driven by a reduction in the K a, with an observed 1,000‐fold reduction in the association constant (Figure 5D and Table 1). These data further demonstrated the flexibility with which different binding moieties can be introduced and highlighted the fact that the molecular interactions between parent antibody and the target antigen can be used to design aDVD molecules with more‐potent blocking capacity.

ation constant (Figure 5D and Table 1). These data further demonstrated the flexibility with which different binding moieties can be introduced and highlighted the fact that the molecular interactions between parent antibody and the target antigen can be used to design aDVD molecules with more‐potent blocking capacity. DISCUSSION The last 2 decades have marked a substantial revolution in the treatment paradigm for RA. The advent of biologic agents has provided a new avenue for the successful treatment of RA. However, there remain a considerable number of patients who do not respond to the available therapies and in whom a treatment‐free remission is rarely achieved 5, 6. RA also represents one of the most lucrative markets for pharmaceutical companies, and in 2013, the 3 top‐selling drugs were all biologic agents marketed for the treatment of RA (see EvaluatePharma, which is available online at http://www.evaluategroup.com/). Despite the obvious success of the current treatments, little effort has been invested in improving the safety profiles of the available therapeutic alternatives.

top‐selling drugs were all biologic agents marketed for the treatment of RA (see EvaluatePharma, which is available online at http://www.evaluategroup.com/). Despite the obvious success of the current treatments, little effort has been invested in improving the safety profiles of the available therapeutic alternatives. The identification of tissue‐specific or overexpressed antigens in the inflamed synovium may provide a solution to these concerns and allow the development of therapeutic agents for tissue‐specific drug delivery 32. Several candidates are now being considered for arthritic synovium targeting, including the oncofetal extradomain A of fibronectin. Indeed, a single‐chain antibody fragment that targets extradomain A has recently entered clinical evaluation as an scFv‐IL‐10 fusion protein for the treatment of RA (ClinicalTrials.gov ID NCT02076659). An alternative strategy could involve the use of bispecific antibody to combine tissue targeting with therapeutic function. To date, no bispecific antibody has been clinically evaluated for use in RA; however, 2 antibodies, catumaxomab and blinatumomab, have been recently approved for cancer treatment, and several constructs are currently in clinical trials 33.

se of bispecific antibody to combine tissue targeting with therapeutic function. To date, no bispecific antibody has been clinically evaluated for use in RA; however, 2 antibodies, catumaxomab and blinatumomab, have been recently approved for cancer treatment, and several constructs are currently in clinical trials 33. The bispecific antibody format DVD‐Ig has shown potential as a versatile platform for dual‐antigen targeting 17. One of the crucial aspects necessary for a conventional bispecific antibody is the capacity for real synergistic activity between the 2 binding moieties. Differences in binding affinities may result in targeting skewed toward one antigen and, as a consequence, suboptimal therapeutic activity. This is probably the main drawback when combining tissue‐targeting moieties with existing therapeutic domains, which are usually characterized by very high affinities. Interestingly, the size and composition of the linker between the outer and inner variable domain of DVD‐Ig antibodies has been shown to significantly affect the kinetic activity of the inner region 19. We hypothesized that by reducing the linker length, we could selectively impair antigen accessibility to the internal domain in a reversible manner, via the presence of an MMP‐cleavable site within the linker. By placing the therapeutic moiety on the inner region, we could obtain 3 important effects: 1) binding capacity skewed toward tissue targeting provided by the outer variable domain, 2) inhibition of systemic engagement of the inner therapeutic binding region, and 3) selective activation of the therapeutic antibody at the site of local inflammation. This construct would therefore provide a tissue‐specific delivery of an antibody prodrug.

oward tissue targeting provided by the outer variable domain, 2) inhibition of systemic engagement of the inner therapeutic binding region, and 3) selective activation of the therapeutic antibody at the site of local inflammation. This construct would therefore provide a tissue‐specific delivery of an antibody prodrug. We developed an activatable DVD‐like construct with an anti–ICAM‐1 outer domain (for targeting the inflamed synovium) linked to the anti‐TNF antibody adalimumab. The linkers we designed contained an MMP‐cleavable sequence and were readily cleaved by the proteolytic MMP‐1 enzyme, both in recombinant form and in physiologic form in human RA SF, providing insight for efficient in vivo antibody activation. Although MMP overexpression is generally associated with inflammation, angiogenesis, and wound repair, different tissues/conditions are characterized by increased expression of specific MMP subgroups. In the context of inflammation, elevated levels of MMPs 2, 7, 8, and 9 have been reported in experimental autoimmune encephalomyelitis 34, 35, 36, while MMPs 3 and 9 have been associated with cutaneous inflammation 37. In inflammatory arthritis, overexpression of MMPs 1, 3, 9, and 13 has been correlated with disease progression and joint damage 26, 38, 39. MMP levels in RA SF and serum were found to be significantly higher than those in healthy controls, with SF levels being several hundred‐fold higher than serum levels 39, 40.

n 37. In inflammatory arthritis, overexpression of MMPs 1, 3, 9, and 13 has been correlated with disease progression and joint damage 26, 38, 39. MMP levels in RA SF and serum were found to be significantly higher than those in healthy controls, with SF levels being several hundred‐fold higher than serum levels 39, 40. As synoviocytes in the lining layer represent the predominant source of MMPs in the arthritic synovium 41, the synovial tissue can be expected to display a greater concentration of MMP‐1 and MMP‐3 than the associated SF and peripheral blood. The slower antibody cleavage rate observed when using RA SF and the lack of activation in RA sera may result in an advantage in vivo, where only the antibody that is actively accumulated in the arthritic synovial tissue may be held long enough for efficient MMP cleavage. Furthermore, in the context of infections such as tuberculosis (a major risk of treatment with anti‐TNF), the main driver of cartilage degradation is MMP‐9, with low secreted levels of MMP‐1 (100‐fold lower than in RA SF) 42, 43. This may result in increased safety because of a reduced risk of unwanted antibody activation in other tissues, and/or it may result in the presence of concomitant infections.

ment with anti‐TNF), the main driver of cartilage degradation is MMP‐9, with low secreted levels of MMP‐1 (100‐fold lower than in RA SF) 42, 43. This may result in increased safety because of a reduced risk of unwanted antibody activation in other tissues, and/or it may result in the presence of concomitant infections. Characterization of the binding of TNF by ELISA and inhibition of the biologic activity of TNF in L929 cell assays proved that short linkers could substantially restrict antigen accessibility and binding capacity. The binding, however, could be fully resolved upon MMP cleavage. Among the 3 MMP‐containing linkers, the short PLGLWA linker showed the highest binding inhibition, reasonably due to the vicinity of the outer domain and increased steric hindrance. Indeed, measurement of the binding kinetics by SPR showed impaired antigen access to the inner domain, resulting in a slower K a. Once bound, however, the K d kinetics for TNF were unchanged, suggesting unaltered functionalities of adalimumab in the aDVD format. This was further confirmed by the identical kinetics of the cleaved aDVD and the parent adalimumab antibody. This effect was more pronounced when we used variable regions with smaller binding interface (e.g., infliximab).

ics for TNF were unchanged, suggesting unaltered functionalities of adalimumab in the aDVD format. This was further confirmed by the identical kinetics of the cleaved aDVD and the parent adalimumab antibody. This effect was more pronounced when we used variable regions with smaller binding interface (e.g., infliximab). The change of paradigm in the treatment of RA with the consequent huge costs to national health systems has spurred the identification of new clinical tools for stratification of patients in order to improve therapeutic effectiveness and reduce adverse effects as well as costs 44. The next step in increasing the therapeutic potency and the safety profile would be the development of tissue‐specific (not only inflammation‐specific) targeting agents 32. With regard to RA synovium, our group of investigators has previously identified peptides and scFv antibody fragments with selective specificity for the synovium 45, 46, 47. The incorporation of the scFv A7 in the aDVD format, as demonstrated herein for ICAM‐1, will provide specificity for human arthritic synovium to the molecule, with great potential for tissue‐specific drug delivery in RA.

tified peptides and scFv antibody fragments with selective specificity for the synovium 45, 46, 47. The incorporation of the scFv A7 in the aDVD format, as demonstrated herein for ICAM‐1, will provide specificity for human arthritic synovium to the molecule, with great potential for tissue‐specific drug delivery in RA. These constructs have the potential to increase safety because of their selective activation in situ and their inhibited systemic cytokine engagement and to increase potency because of their enhanced therapeutically relevant concentrations achieved at the site of active disease. We can believe that the development of tissue‐targeting prodrugs such as the aDVD described herein could represent optimum flexible platforms for the rational design of therapeutic agents with significant impact on RA. AUTHOR CONTRIBUTIONS 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. Pitzalis 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. Onuoha, Ferrari, Sblattero, Pitzalis. Acquisition of data. Onuoha, Ferrari. Analysis and interpretation of data. Onuoha, Ferrari, Sblattero, Pitzalis. ACKNOWLEDGMENT The authors thank Dr. Zhiyong Lou for granting permission to adapt Protein Data Bank atomic coordinates for adalimumab and infliximab and for helpful discussion.

Rheumatoid arthritis (RA) is an autoimmune disease characterized by the presence of disease‐specific anti–citrullinated protein antibodies (ACPAs) 1. Because ACPAs can be detected in patients with RA several years before the diagnosis is made 2, it is now thought that RA‐related autoimmunity may be initiated outside the joint, in sites such as the lungs and the periodontium 3, 4. Smoking is a known risk factor for RA 3, 5. There is accumulating evidence that the ACPA response results from smoking‐induced inflammation of the lung, resulting in increased expression of citrullinated proteins 6, 7. Periodontitis, which often is cited as one of the most common inflammatory diseases, is also a risk factor for RA 8, and patients with periodontitis have increased levels of antibodies against the uncitrullinated forms of RA autoantigens 9, 10.

the lung, resulting in increased expression of citrullinated proteins 6, 7. Periodontitis, which often is cited as one of the most common inflammatory diseases, is also a risk factor for RA 8, and patients with periodontitis have increased levels of antibodies against the uncitrullinated forms of RA autoantigens 9, 10. Bronchiectasis (BR) has been recognized as a risk factor for RA since publication of the classic studies by Walker nearly 50 years ago 11. He observed that among 516 patients with RA, 2.5% had symptoms of antecedent BR compared with 0.3% of 300 patients with degenerative joint disease. Similar findings have been observed in other cohorts of patients with RA 12. Importantly, in a more recent study, RA developed in 2 patients with BR over 12 months of followup 13. Although it would be difficult to confidently calculate the relative risk in these studies, it would be fair to conclude that BR is a potent risk factor for RA in a minority of patients. Similar to other severe chronic bacterial infections, BR has been known for decades to be associated with a high frequency of rheumatoid factors (RFs) 14, 15, suggesting that chronic bacterial infection of the lung could lead to autoimmunity in RA. However, there are no published studies of the fine specificity of ACPAs in BR, and the potential mechanisms of citrulline‐specific autoimmunity induced by bacterial infection have not been studied in BR.

oid factors (RFs) 14, 15, suggesting that chronic bacterial infection of the lung could lead to autoimmunity in RA. However, there are no published studies of the fine specificity of ACPAs in BR, and the potential mechanisms of citrulline‐specific autoimmunity induced by bacterial infection have not been studied in BR. In this study, we used BR as a model to study the evolution of the ACPA response induced by severe chronic bacterial infection, as 2 cross‐sectional “snapshots” at the beginning and the end of development of the ACPA response, in patients with BR and BR patients in whom RA later develops. To assess whether BR could be a model for the induction of autoimmunity in RA, we measured the levels of autoantibodies to both citrullinated and uncitrullinated peptides in a well‐documented group of BR patients without RA, using healthy subjects and patients with asthma as controls. To examine the ACPA response in patients with established disease, we measured the levels of these autoantibodies in BR patients with concomitant RA (BR/RA) and in RA patients without any lung disease.

des in a well‐documented group of BR patients without RA, using healthy subjects and patients with asthma as controls. To examine the ACPA response in patients with established disease, we measured the levels of these autoantibodies in BR patients with concomitant RA (BR/RA) and in RA patients without any lung disease. PATIENTS AND METHODS Serum samples from patients and control subjects Serum samples from 122 patients with BR, 50 patients with BR/RA, 50 RA patients without lung disease, 87 patients with asthma, and 79 healthy control subjects were obtained from several centers across the UK 13. All of the patients with BR were adults (age >18 years) with high‐resolution computed tomography (HRCT)–proven symptomatic non–cystic fibrosis BR and a history of ≥2 respiratory infections per year. HRCT was performed by chest radiologists who were not involved in the study, and all patients in the BR cohorts were receiving followup care from a respiratory consultant (a physician with an interest in BR). Patients with coexistent intestinal lung disease, asthma, emphysema, allergic bronchopulmonary aspergillosis, a history of or current tuberculosis, or those in whom any other lung disease was observed by HRCT or reported in respiratory clinic notes (referencing electronic letters and patients’ paper notes taken at the respiratory clinic) were excluded from the study. All BR patients also underwent a musculoskeletal examination by a rheumatologist and were excluded if they had a history of inflammatory joint pain, inflammatory arthritis, or any synovitis 13. All RA patients met the American College of Rheumatology/European League Against Rheumatism 2010 criteria for RA 16. RF was present in significantly greater numbers of BR/RA patients compared with patients with RA alone 13. Multicenter ethics approval was obtained at all participating centers (Integrated Research Application System approval no. 12324). Recruitment was performed between May 2012 and May 2013.

2010 criteria for RA 16. RF was present in significantly greater numbers of BR/RA patients compared with patients with RA alone 13. Multicenter ethics approval was obtained at all participating centers (Integrated Research Application System approval no. 12324). Recruitment was performed between May 2012 and May 2013. Antibody measurements The levels of IgM RFs were measured with an Hitachi Modular P analyzer and were defined as positive or negative according to the cutoff levels recommended by the manufacturer. All of the ACPA tests were performed in the same laboratory by a single investigator (A‐MQ). Anti–cyclic citrullinated peptide 2 (anti–CCP‐2) antibody levels were measured using a commercial enzyme‐linked immunosorbent assay (ELISA) (Diastat; Eurodiagnostica) according to the instructions of the manufacturer. Antibodies to immunodominant peptides from 3 established RA autoantigens, citrullinated α‐enolase peptide 1 (CEP‐1; amino acids [aa] 4–21 [KIHA‐cit‐EIFDS‐cit‐GNPTVE]) 17, Cit‐vimentin (aa 59–74 [VYAT‐cit‐SSAV‐cit‐L‐cit‐SSVP]), and Cit‐fibrinogen β‐chain (aa 36–52 [NEEGFFSA‐cit‐GHRPLDKK]) 10 were also measured by in‐house ELISAs, as previously described 10. For all assays, arginine‐containing control peptides were run in parallel. For the immunodominant peptides and control peptides, we calculated the cutoff for positivity as the ninety‐fifth percentile of the value for normal controls, and all serum samples with values above this cutoff were considered positive. A standard curve was included in all plates coated with citrullinated peptides and those coated with uncitrullinated vimentin, and results were expressed as arbitrary units. For assays without standard curves, the results were expressed as the optical density (OD).

ith values above this cutoff were considered positive. A standard curve was included in all plates coated with citrullinated peptides and those coated with uncitrullinated vimentin, and results were expressed as arbitrary units. For assays without standard curves, the results were expressed as the optical density (OD). Inhibition/competition assays Inhibition/competition experiments were performed in the liquid phase. Six anti–arginine‐containing α‐enolase peptide 1 (anti–REP‐1)–positive serum samples from patients with BR and 6 anti–CEP‐1–positive samples from patients with BR/RA were chosen. Sera were incubated for 2 hours at room temperature and overnight at 4°C in buffer alone as control or in 1 mg/ml of either REP‐1 or CEP‐1. The mixtures were centrifuged at 16,200g for 20 minutes, the BR supernatants were transferred to REP‐1–coated plates and the BR/RA supernatants were transferred to CEP‐1–coated plates, and assays were performed as described above. All results were expressed as the OD at 450 nm. Statistical analysis The Mann‐Whitney nonparametric test (for unmatched groups) was used to compare differences between antibody responses in the cohorts of serum samples. Spearman's nonparametric correlations between data sets were assessed. Calculations were performed using GraphPad software.

Inhibition/competition assays Inhibition/competition experiments were performed in the liquid phase. Six anti–arginine‐containing α‐enolase peptide 1 (anti–REP‐1)–positive serum samples from patients with BR and 6 anti–CEP‐1–positive samples from patients with BR/RA were chosen. Sera were incubated for 2 hours at room temperature and overnight at 4°C in buffer alone as control or in 1 mg/ml of either REP‐1 or CEP‐1. The mixtures were centrifuged at 16,200g for 20 minutes, the BR supernatants were transferred to REP‐1–coated plates and the BR/RA supernatants were transferred to CEP‐1–coated plates, and assays were performed as described above. All results were expressed as the OD at 450 nm. Statistical analysis The Mann‐Whitney nonparametric test (for unmatched groups) was used to compare differences between antibody responses in the cohorts of serum samples. Spearman's nonparametric correlations between data sets were assessed. Calculations were performed using GraphPad software. RESULTS Demographic characteristics of the study participants The demographic characteristics of the patients and control subjects are shown in Table 1. The patients were well matched for sex, with a preponderance of women in all groups, although the median age of subjects in the control group was ∼7 years younger than that of patients with BR, patients with BR/RA, and patients with RA. Ever‐smoking was reported by 39% of the BR patients without RA, 42% of the control subjects, 43% of the patients with asthma, and 56% of the patients with RA only. The frequency of RF was significantly increased in patients with BR (25%) compared with healthy controls (10%), as previously observed 14, 15. As expected, highly significant increases in RF positivity were observed in both the RA group and the BR/RA group compared with controls (Table 2).

of the patients with RA only. The frequency of RF was significantly increased in patients with BR (25%) compared with healthy controls (10%), as previously observed 14, 15. As expected, highly significant increases in RF positivity were observed in both the RA group and the BR/RA group compared with controls (Table 2). Table 1 Demographic characteristics of the healthy control subjects and patientsa n Age, median (IQR) years Female sex, % Ever‐smoker, % Healthy controls 79 60 (17) 74 42 Patients Asthma 87 50 (27) 79 43 BR 122 66 (13) 65 39 BR/RA 50 68 (12) 72 42 RA 50 66 (15) 72 56 a There were no significant differences in age between patients with bronchiectasis (BR), BR patients with rheumatoid arthritis (BR/RA), and patients with RA. Patients with asthma were significantly younger than subjects in all other groups. Healthy control subjects were significantly younger then patients in the BR/RA and RA groups. There were no significant differences in sex or smoking status between all groups, with the exception of the group with asthma, in which the percentage of female patients was significantly higher than that in the group with BR. Table 2 Serum antibody positivity in the healthy control subjects and patientsa Antibody Healthy controls (n = 79) Patients Asthma (n = 87) BR (n = 122) BR/RA (n = 50) RA (n = 50) Rheumatoid factor 10 16 25b 82c 52c CCP‐2 0 1 5 88c 48c CEP‐1 4 1 7 60c 24d Cit‐vimentin 4 3 7 56c 20d Cit‐fibrinogen 4 0 12b 74c 40c

n Age, median (IQR) years Female sex, % Ever‐smoker, % Healthy controls 79 60 (17) 74 42 Patients Asthma 87 50 (27) 79 43 BR 122 66 (13) 65 39 BR/RA 50 68 (12) 72 42 RA 50 66 (15) 72 56 a There were no significant differences in age between patients with bronchiectasis (BR), BR patients with rheumatoid arthritis (BR/RA), and patients with RA. Patients with asthma were significantly younger than subjects in all other groups. Healthy control subjects were significantly younger then patients in the BR/RA and RA groups. There were no significant differences in sex or smoking status between all groups, with the exception of the group with asthma, in which the percentage of female patients was significantly higher than that in the group with BR. Table 2 Serum antibody positivity in the healthy control subjects and patientsa Antibody Healthy controls (n = 79) Patients Asthma (n = 87) BR (n = 122) BR/RA (n = 50) RA (n = 50) Rheumatoid factor 10 16 25b 82c 52c CCP‐2 0 1 5 88c 48c CEP‐1 4 1 7 60c 24d Cit‐vimentin 4 3 7 56c 20d Cit‐fibrinogen 4 0 12b 74c 40c REP‐1 4 3 19d 12 0 Vimentin 4 3 16b 20d 2 Fibrinogen 4 2 9 6 8 a Values are the percent. BR = bronchiectasis; BR/RA = BR with rheumatoid arthritis (RA); CCP‐2 = cyclic citrullinated peptide 2; CEP‐1 = citrullinated α‐enolase peptide 1; REP‐1 = arginine‐containing α‐enolase peptide 1. b P < 0.05 versus controls, by Fisher's exact test. c P < 0.001 versus controls, by Fisher's exact test. d P < 0.01 versus controls, by Fisher's exact test.

REP‐1 4 3 19d 12 0 Vimentin 4 3 16b 20d 2 Fibrinogen 4 2 9 6 8 a Values are the percent. BR = bronchiectasis; BR/RA = BR with rheumatoid arthritis (RA); CCP‐2 = cyclic citrullinated peptide 2; CEP‐1 = citrullinated α‐enolase peptide 1; REP‐1 = arginine‐containing α‐enolase peptide 1. b P < 0.05 versus controls, by Fisher's exact test. c P < 0.001 versus controls, by Fisher's exact test. d P < 0.01 versus controls, by Fisher's exact test. Non–citrulline‐specific ACPAs in BR patients without RA Six of the BR patients (5%) were anti–CCP‐2 antibody positive. Although the rates of anti–CCP‐2 positivity were not significantly different between BR patients and controls (Table 2), the titers of anti–CCP‐2 antibodies in BR patients were significantly increased compared with those in both healthy controls (P < 0.001) and patients with asthma (P < 0.01) (Figure 1). The frequencies of antibody positivity to the “specific” citrullinated peptides were slightly increased in patients with BR compared with controls (for CEP‐1, 7% versus 4%; for Cit‐vimentin, 7% versus 4%; for Cit‐fibrinogen, 12% versus 4%); only the difference in the frequency of antibodies to Cit‐fibrinogen reached statistical significance (P < 0.05) (Table 2). When comparing the titers of antibody binding, the levels of antibodies to all 3 peptides were significantly increased in BR patients compared with patients with asthma and, in the case of both anti‐vimentin and anti–Cit‐fibrinogen, compared with healthy controls (Figure 2).

atistical significance (P < 0.05) (Table 2). When comparing the titers of antibody binding, the levels of antibodies to all 3 peptides were significantly increased in BR patients compared with patients with asthma and, in the case of both anti‐vimentin and anti–Cit‐fibrinogen, compared with healthy controls (Figure 2). Figure 1 Anti–cyclic citrullinated peptide 2 (anti–CCP‐2) levels in healthy control subjects, patients with asthma, patients with bronchiectasis (BR), BR patients with rheumatoid arthritis (BR/RA), and patients with RA. Each symbol represents an individual subject; horizontal lines show the median. NS = not significant. ∗∗ = P < 0.01; ∗∗∗ = P < 0.001; ∗∗∗∗ = P < 0.0001. Figure 2 Levels of anti–citrullinated α‐enolase peptide 1 (anti–CEP‐1) and anti–arginine‐containing α‐enolase peptide 1 (anti–REP‐1) (A), anti–Cit‐vimentin (anti‐cVim) and anti‐vimentin (B), and anti–Cit‐fibrinogen (anti‐cFib) and anti‐fibrinogen (C) in healthy control subjects, patients with asthma, patients with BR, patients with BR/RA, and patients with RA. Each symbol represents an individual subject; horizontal lines show the median. ∗ = P < 0.05; ∗∗ = P < 0.01; ∗∗∗ = P < 0.001; ∗∗∗∗ = P < 0.0001. See Figure 1 for other definitions.

ib) and anti‐fibrinogen (C) in healthy control subjects, patients with asthma, patients with BR, patients with BR/RA, and patients with RA. Each symbol represents an individual subject; horizontal lines show the median. ∗ = P < 0.05; ∗∗ = P < 0.01; ∗∗∗ = P < 0.001; ∗∗∗∗ = P < 0.0001. See Figure 1 for other definitions. The frequencies of antibodies to the arginine‐containing control peptides from the specific citrullinated antigens were also increased in patients with BR compared with healthy controls (for REP‐1, 19% versus 4%; for vimentin, 16% versus 4%; for fibrinogen, 9% versus 4%) (Table 2). In addition, the titers of anti–REP‐1 and anti‐vimentin were significantly increased in patients with BR compared with both healthy controls and patients with asthma (Figure 2). The lack of citrulline‐specific ACPA responses in the BR patients was further suggested by the strong correlations between antibodies to each of the citrullinated peptides compared with their uncitrullinated variants (for anti–CEP‐1 versus anti–REP‐1, r = 0.502 [P < 0.0001]; for anti–Cit‐vimentin versus anti–vimentin, r = 0.725 [P < 0.0001]; for anti–Cit‐fibrinogen versus anti‐fibrinogen, r = 0.798 [P < 0.0001]) (Figure 3).

lations between antibodies to each of the citrullinated peptides compared with their uncitrullinated variants (for anti–CEP‐1 versus anti–REP‐1, r = 0.502 [P < 0.0001]; for anti–Cit‐vimentin versus anti–vimentin, r = 0.725 [P < 0.0001]; for anti–Cit‐fibrinogen versus anti‐fibrinogen, r = 0.798 [P < 0.0001]) (Figure 3). Figure 3 Correlations between the levels of antibodies to specific anti–citrullinated protein antibodies and their arginine‐containing control peptides in patients with BR, patients with BR/RA, and patients with RA. A, Anti–citrullinated α‐enolase peptide 1 (anti–CEP‐1) responses versus anti–arginine‐containing α‐enolase peptide 1 (anti–REP‐1) responses. B, Anti–Cit‐vimentin (anti‐cVim) responses versus anti‐vimentin responses. C, Anti–Cit‐fibrinogen (anti‐cFib) responses versus anti‐fibrinogen responses. See Figure 1 for other definitions. The citrulline specificity of the antibody response in patients with BR was also examined by performing absorption experiments with the 2 enolase peptides, CEP‐1 and REP‐1, in 6 of the serum samples from BR patients. The anti–REP‐1 response in these sera was absorbed partially and almost equally with both REP‐1 and CEP‐1 (Figures 4A and B), suggesting that antibodies were cross‐reactive for both peptides and further confirming their lack of citrulline specificity.

ptides, CEP‐1 and REP‐1, in 6 of the serum samples from BR patients. The anti–REP‐1 response in these sera was absorbed partially and almost equally with both REP‐1 and CEP‐1 (Figures 4A and B), suggesting that antibodies were cross‐reactive for both peptides and further confirming their lack of citrulline specificity. Figure 4 Antibody responses in anti–REP‐1–seropositive patients with bronchiectasis (BR) and in anti–CEP‐1–seropositive BR patients with rheumatoid arthritis (BR/RA) in the presence of either REP‐1 or CEP‐1, as determined by enzyme‐linked immunosorbent assay–based competition/inhibition assays. A and B, Anti–REP‐1 reactivity in the presence of REP‐1 and CEP‐1 in patients with BR. C and D, Anti–CEP‐1 reactivity in the presence of REP‐1 and CEP‐1 in patients with BR/RA. Each symbol represents a different patient; lines connect the individual data points. See Figure 2 for other definitions. Citrulline specificity and ACPA levels in patients with BR/RA Compared with the patients with BR, those with BR/RA had a highly citrulline‐specific ACPA response to each antigen tested (Figures 1 and 2). In addition, no significant correlations between the antibody responses to the citrullinated peptides and the arginine‐containing control peptides were observed in patients with BR/RA, with the exception of the correlation between anti–Cit‐vimentin and anti‐vimentin (r = 0.455 [P < 0.0001]) (Figure 3).

igures 1 and 2). In addition, no significant correlations between the antibody responses to the citrullinated peptides and the arginine‐containing control peptides were observed in patients with BR/RA, with the exception of the correlation between anti–Cit‐vimentin and anti‐vimentin (r = 0.455 [P < 0.0001]) (Figure 3). Although the percentage of ever‐smokers was reduced among patients with BR/RA compared with that among patients with RA alone (42% versus 56%; P = 0.06), the rate of seropositivity in BR/RA patients compared with RA patients was significantly increased for each ACPA tested: for anti–CCP‐2, 88% versus 48%; for anti–CEP‐1, 60% versus 24%; for anti–Cit‐vimentin, 56% versus 20%; and for anti–Cit‐fibrinogen, 74% versus 40% (Table 2) (all P < 0.01). The citrulline specificity of the ACPA response in patients with RA was confirmed by the lack of correlation with the response to the arginine‐containing peptides (Figures 2 and 3) and by complete absorption of the anti–CEP‐1 response in 6 BR/RA serum specimens only with the CEP‐1 peptide (Figures 4C and D). Taken together, these data showed that the citrulline specificity of ACPAs in patients with BR/RA was increased compared with that in patients with BR alone, and its magnitude was increased compared with that in RA patients without any lung disease.

BR/RA serum specimens only with the CEP‐1 peptide (Figures 4C and D). Taken together, these data showed that the citrulline specificity of ACPAs in patients with BR/RA was increased compared with that in patients with BR alone, and its magnitude was increased compared with that in RA patients without any lung disease. DISCUSSION Our study showed significantly elevated levels of ACPAs in patients with BR, although the antibody response was not citrulline specific. We suggest that after tolerance has been breached, subsequent long‐term exposure to citrullinated antigens in the inflamed lung results in the spreading of epitopes recognized by B cells to citrullinated peptides. Epitope spreading in patients with BR in whom RA develops may well be reflected by the markedly citrulline‐specific responses to CCP‐2, CEP‐1, Cit‐fibrinogen, and Cit‐vimentin observed in the BR/RA patients. We also observed an increased frequency of RF positivity in patients with BR, but this has been well described in previous studies 14, 15. Furthermore, RFs can also be found in patients with other autoimmune and nonautoimmune conditions, as well as in healthy individuals (for review, see ref. 18). The increased titers and frequencies of both ACPAs and RF in patients with BR/RA compared with patients with RA may also reflect the potency of BR in exacerbating the autoimmune response in RA, once it has developed.

r autoimmune and nonautoimmune conditions, as well as in healthy individuals (for review, see ref. 18). The increased titers and frequencies of both ACPAs and RF in patients with BR/RA compared with patients with RA may also reflect the potency of BR in exacerbating the autoimmune response in RA, once it has developed. Our data suggest that this effect could not be explained by cigarette smoking, which is a well‐established risk factor for RA. In our study the percentage of ever‐smokers was actually lower among BR patients compared with controls, although it was increased slightly among the RA patients without lung disease. However, these data must be interpreted with caution, because “ever smoking” is not regarded as a particularly robust measure of tobacco exposure 19.

the percentage of ever‐smokers was actually lower among BR patients compared with controls, although it was increased slightly among the RA patients without lung disease. However, these data must be interpreted with caution, because “ever smoking” is not regarded as a particularly robust measure of tobacco exposure 19. The largely arginine‐specific autoantibody response that we observed in patients with BR was remarkably similar to that observed in patients with periodontitis 10. This pattern of response was also recently observed in a study by Brink et al 20, in which antibodies against arginine‐containing peptides (including REP‐1, vimentin [aa 60–75], and fibrinogen β‐chain [aa 36–52]) from RA autoantigens were observed before the development of citrulline‐specific ACPAs in serum samples obtained from patients with RA years before symptoms of RA developed. Importantly, that study also included sequential samples. However, it was entirely cross‐sectional, and further prospective investigations will need to be carried out in BR patients at risk of RA in order to confirm the evolution of the citrulline specificity of ACPAs in patients with BR in whom BR/RA subsequently develops.

tantly, that study also included sequential samples. However, it was entirely cross‐sectional, and further prospective investigations will need to be carried out in BR patients at risk of RA in order to confirm the evolution of the citrulline specificity of ACPAs in patients with BR in whom BR/RA subsequently develops. The results of the inhibition assay showed that in BR patients the antibody response to the enolase peptides CEP‐1 and REP‐1 not only is cross‐reactive and lacking citrulline specificity but also may be of low avidity, thereby explaining the incomplete absorption of anti–REP‐1 activity by the REP‐1 peptide. Although we did not examine antibody avidity in this study, a longitudinal study of patients with presymptomatic RA showed that antibody avidity gradually increased as the onset of joint disease approached 21. Therefore, if certain bacterial infections lead to the initiation of an ACPA response, our study and others 9, 10, 20, 21 suggest that infections such as periodontitis, BR, and others yet to be defined induce a low‐titer, low‐avidity, non–citrulline‐specific ACPA response in the early phases of tolerance breakdown, which subsequently evolves into the higher‐avidity, higher‐titer, and highly citrulline‐specific responses that characterize RA.

gest that infections such as periodontitis, BR, and others yet to be defined induce a low‐titer, low‐avidity, non–citrulline‐specific ACPA response in the early phases of tolerance breakdown, which subsequently evolves into the higher‐avidity, higher‐titer, and highly citrulline‐specific responses that characterize RA. A limitation of this study is that the different patient groups were selected from different regions of the UK and were not matched for age. The large number of BR/RA patients in our study does not reflect the prevalence of symptomatic BR in a random population of RA patients, which is probably ∼3% 11, 12. The patients with RA alone were also selected for this study based on the absence of lung disease, and this bias may well reflect the low prevalence of ACPAs in the RA group. However, these results are entirely consistent with a meta‐analysis of several studies 22 and a recent large study, which confirmed a strikingly low frequency (55%) of ACPAs in 230 RA patients without lung disease 23. In contrast, the more heightened reactivity to citrullinated antigens in patients with BR/RA could represent a certain type of disease in which reactivity to these citrullinated antigens is dominant and perhaps related to the airway disease. Demoruelle et al observed that 76% of anti‐CCP–positive subjects without arthritis had airway abnormalities, including BR, and that articular RA developed in 2 of these subjects within a year 24, and Fischer et al reported that 63% of patients with lung disease without RA had “moderate to high” anti‐CCP positivity 25.

Demoruelle et al observed that 76% of anti‐CCP–positive subjects without arthritis had airway abnormalities, including BR, and that articular RA developed in 2 of these subjects within a year 24, and Fischer et al reported that 63% of patients with lung disease without RA had “moderate to high” anti‐CCP positivity 25. Even though the current study cannot be regarded as an epidemiologic study, we have demonstrated in this substantial population of well‐documented patients that BR could be an unusual but potent model for the induction of autoimmunity in RA by bacterial infection in the lung. To prove that this is a true causal relationship, longitudinal followup of patients with BR is required, and such studies are under way within the UK Clinical Research Network Study Portfolio. AUTHOR CONTRIBUTIONS 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. Venables 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. Perry, Kelly, De Soyza, Eggleton, Hutchinson, Venables. Acquisition of data. Quirke, Perry, Cartwright, Kelly, De Soyza, Eggleton, Hutchinson. Analysis and interpretation of data. Quirke, Perry, De Soyza, Eggleton, Hutchinson, Venables.

AUTHOR CONTRIBUTIONS 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. Venables 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. Perry, Kelly, De Soyza, Eggleton, Hutchinson, Venables. Acquisition of data. Quirke, Perry, Cartwright, Kelly, De Soyza, Eggleton, Hutchinson. Analysis and interpretation of data. Quirke, Perry, De Soyza, Eggleton, Hutchinson, Venables. ACKNOWLEDGMENTS We thank Dr. Gill Baker and colleagues at the National Institute for Health Research Exeter Clinical Research Facility for selecting and providing ethically approved serum samples from healthy control subjects. We also acknowledge National Institute for Health Research Clinical Research Network support to the recruiting centers via the UK Clinical Research Network Study Portfolio (http://public.ukcrn.org.uk/search/StudyDetail.aspx?StudyID=12324).

Osteoarthritis (OA), typified by degenerative loss of cartilage integrity and joint space narrowing, is a leading cause of pain, disability, and shortening of adult working life throughout the world 1, 2, 3. Unfortunately, at present there is no approved treatment that can modify the disease progression, resulting in limited therapeutic options for patients 4. In attempting to identify novel therapeutics, inflammation is increasingly being recognized as an important driver of OA cartilage pathology. Histologic analysis, ultrasound, and magnetic resonance imaging have all demonstrated evidence of synovitis in OA joints 5, 6, 7, with increased cellular infiltration of activated B cells and T lymphocytes. Indeed, synovitis is reported not only in established OA, but also at the onset of OA, being present in patients with only minimal radiographic signs of the disease 8. Several proinflammatory cytokines are elevated in the synovial fluid of OA joints compared to normal healthy joints 9, and cytokine stimulation of ex vivo cartilage tissue mimics the pathologic changes observed within the OA joint 9, 10. However, the key regulators of the cellular inflammatory response in cartilage tissue are not well defined.

ytokines are elevated in the synovial fluid of OA joints compared to normal healthy joints 9, and cytokine stimulation of ex vivo cartilage tissue mimics the pathologic changes observed within the OA joint 9, 10. However, the key regulators of the cellular inflammatory response in cartilage tissue are not well defined. There is now overwhelming evidence that the microRNA (miRNA) family of short noncoding RNAs can regulate the inflammatory response 11, 12. Indeed, our group previously identified differentially expressed miRNAs in human OA cartilage tissue that mediated the production of matrix metalloproteinase 13 (MMP‐13) and tumor necrosis factor (TNF) 13, suggesting a role of miRNAs in regulating inflammation and OA pathology 13. Importantly, RNA sequencing (RNAseq) has now identified multiple families of long noncoding RNAs (lncRNAs), which include antisense RNAs, pseudogenes, and long intergenic noncoding RNAs (lincRNAs) 14, 15. Of interest, earlier reports suggest that these lncRNAs may also be central regulators of biologic processes 16, 17, 18, 19, including the inflammatory response 20. In support of those findings, we recently identified lncRNAs that were differentially expressed upon lipopolysaccharide (LPS)–induced activation of the human innate response and demonstrated that these regulated interleukin‐1β (IL‐1β) and IL‐8 production 21.

esses 16, 17, 18, 19, including the inflammatory response 20. In support of those findings, we recently identified lncRNAs that were differentially expressed upon lipopolysaccharide (LPS)–induced activation of the human innate response and demonstrated that these regulated interleukin‐1β (IL‐1β) and IL‐8 production 21. Currently, little is known about the expression and functional role of lncRNAs in OA joint tissue. Their potential importance is indicated in a recent report by Fu et al 22, who identified ∼4,700 lncRNAs that were differentially expressed in cartilage from patients with knee OA (compared with controls) using a microarray‐based approach. Although that preliminary study did not examine the function of these lncRNAs, another recent study has identified a lincRNA located upstream of the gene PTGS2 (cyclooxygenase 2 [COX‐2]). This was shown to be increased in phorbol myristate acetate– and LPS‐stimulated monocytes and to positively regulate COX‐2 expression 23 by binding to, and relieving the action of, the repressive p50 component of the NF‐κB complex 23. As a result of this action, the lincRNA was renamed p50‐associated COX‐2–extragenic RNA (PACER). Importantly, COX‐2 is a key regulator of the arachidonic acid pathway and subsequent prostaglandin E2 production 24, which is a putative mediator of inflammation and pain in OA cartilage tissue 25, 26. Given these observations, and the key role of inflammation in OA cartilage pathology, we hypothesized that lncRNAs, including PACER, are central regulators of the inflammatory response in cartilage tissue.

ostaglandin E2 production 24, which is a putative mediator of inflammation and pain in OA cartilage tissue 25, 26. Given these observations, and the key role of inflammation in OA cartilage pathology, we hypothesized that lncRNAs, including PACER, are central regulators of the inflammatory response in cartilage tissue. The aim of this study was therefore to perform RNAseq in order to identify lncRNAs that are associated with the inflammatory response in primary human OA chondrocytes isolated from the articular cartilage of patients with hip OA. We then proceeded to assess their potential involvement in OA by examining the expression of several “inflammation‐associated” lncRNAs (including PACER) in human articular cartilage from patients with and those without hip or knee OA, profiling their expression in response to multiple proinflammatory cytokines and determining the functional effect of modulating the expression of an inflammation‐associated lncRNA on the chondrocyte inflammatory response.

ncluding PACER) in human articular cartilage from patients with and those without hip or knee OA, profiling their expression in response to multiple proinflammatory cytokines and determining the functional effect of modulating the expression of an inflammation‐associated lncRNA on the chondrocyte inflammatory response. PATIENTS AND METHODS Patients and tissue samples Following ethics approval (UK National Research Ethics Committee 14/ES/1044), patients with hip OA (mean ± SEM age 69 ± 3 years; n = 9), patients with knee OA (age 70 ± 3 years; n = 12), and patients with fracture of the neck of the femur without OA (age 74 ± 2 years; n = 6) were recruited prior to elective joint replacement surgery at either The Royal Orthopaedic Hospital (Birmingham, UK) or Russell's Hall Hospital (Dudley, UK). Patients with hip OA had Kellgren/Lawrence (K/L) grades 27 of 3 or 4, patients with knee OA all had K/L grades of 4, and patients with fracture of the neck of the femur had K/L grades of 0. Cartilage from femoral condyles (from knee OA patients) and femoral heads (from hip OA patients) was collected. Ethics approval was also obtained (Derby Research Ethics Committee 1 [11/H0405/2]) to collect non‐OA knee cartilage from postmortem donors (mean ± SEM age 74 ± 5 years; n = 4) (Kings Mill Hospital, Sutton‐in‐Ashfield, UK) with no history of joint pain or evidence of cartilage fibrillation based on chondropathy assessment 28. Consent was obtained from all patients or families. Patient demographic data are provided in Supplementary Table 1, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.39520/abstract. A protocol was in place to ensure that samples were all handled in the same way and processed in the same timeframe. For tissue processing, upon separation of cartilage from bone tissue, the cartilage was immediately snap‐frozen in liquid nitrogen.

site at http://onlinelibrary.wiley.com/doi/10.1002/art.39520/abstract. A protocol was in place to ensure that samples were all handled in the same way and processed in the same timeframe. For tissue processing, upon separation of cartilage from bone tissue, the cartilage was immediately snap‐frozen in liquid nitrogen. Isolation of primary chondrocytes from articular cartilage Articular cartilage was separated from the subchondral bone using a scalpel and digested using filter‐sterilized collagenase IIA (2 mg/ml; Sigma‐Aldrich) for 5 hours at 37°C. Digested cartilage was then filtered by passing through a 40‐μm cell strainer (BD Biosciences), and the filtrate was centrifuged. Primary chondrocytes were then resuspended in growth media (Dulbecco's modified Eagle's medium supplemented with 10% fetal calf serum [FCS], penicillin [100 units/ml], streptomycin [100 μg/ml], l‐glutamine [2 mM], nonessential amino acids [5% volume/volume] [all from Life Technologies], and amphotericin [2 μg/ml; Sigma‐Aldrich]) and grown to 70–80% confluence before being used in subsequent studies.

gle's medium supplemented with 10% fetal calf serum [FCS], penicillin [100 units/ml], streptomycin [100 μg/ml], l‐glutamine [2 mM], nonessential amino acids [5% volume/volume] [all from Life Technologies], and amphotericin [2 μg/ml; Sigma‐Aldrich]) and grown to 70–80% confluence before being used in subsequent studies. RNAseq analysis Primary hip OA chondrocytes (n = 3 patients) were left unstimulated or stimulated with IL‐1β (1 ng/ml) for 4 hours in 0.1% FCS culture media in the absence of antibiotics and amphotericin. Total RNA was isolated using TRIzol reagent (Life Technologies), further purified (RNeasy column; Qiagen), and the RNA integrity number (RIN) was assessed (Agilent Bioanalyzer). All RIN values were >7, and 260:280 ratios (measured by NanoDrop) were >1.7. Ribosomal RNA was removed using Ribozero (Epicentre Technologies), and RNAseq (100‐bp paired‐end, stranded sequencing) was performed on an Illumina HiSeq 2000 sequencer. Subsequent analysis was undertaken using Tophat2/Cufflinks with alignment against the hg19 reference genome (Figure 1A). LncRNAs were identified using Cufflinks and then compared with known lncRNAs previously annotated in Gencode version 19 and the Human LincRNAs Catalog 29. CuffDiff was used to compare control and IL‐1β–treated cells to identify differentially expressed transcripts (false discovery rate [FDR] <0.05, fold change >2, and change in fragments per kilobase of transcript per million mapped reads [FPKM] >1). Sequence data are available through the GEO database under series number GSE74220.

s used to compare control and IL‐1β–treated cells to identify differentially expressed transcripts (false discovery rate [FDR] <0.05, fold change >2, and change in fragments per kilobase of transcript per million mapped reads [FPKM] >1). Sequence data are available through the GEO database under series number GSE74220. Figure 1 Regulation of long noncoding RNA (lncRNA) expression by interleukin‐1β (IL‐1β) in human osteoarthritis (OA) chondrocytes. A, Pipeline for predicting lncRNAs from Cufflinks‐assembled transfrags. FPKM = fragments per kilobase of transcript per million mapped reads. B, Release of IL‐6 from primary human hip OA chondrocytes left unstimulated or stimulated with IL‐1β for 4 hours or 24 hours, as measured by enzyme‐linked immunosorbent assay. IL‐6 release indicates activation of the inflammatory response. Bars show the mean ± SEM. ∗∗∗ = P < 0.001. C, Volcano plot displaying differentially expressed mRNAs (n = 3 IL‐1β–stimulated hip OA chondrocytes and 3 unstimulated hip OA chondrocytes.). D, Pathway analysis of differentially expressed mRNAs. FDR = false discovery rate. E, Overlap of lncRNAs in OA chondrocytes, Gencode version 19, and the HumanBodyMap catalogs. F, Breakdown of differentially expressed lncRNAs based on positional classifications.

OA chondrocytes and 3 unstimulated hip OA chondrocytes.). D, Pathway analysis of differentially expressed mRNAs. FDR = false discovery rate. E, Overlap of lncRNAs in OA chondrocytes, Gencode version 19, and the HumanBodyMap catalogs. F, Breakdown of differentially expressed lncRNAs based on positional classifications. Analysis of lncRNA expression in primary chondrocytes and articular cartilage by quantitative reverse transcriptase–polymerase chain reaction (qRT‐PCR) Articular hip and knee cartilage was snap‐frozen in liquid nitrogen and pulverized using a 6770 Freezer/mill (Spex Sample Prep). Total RNA was extracted from both powdered cartilage and primary chondrocytes using TRIzol and further purified using RNeasy columns. RIN values were >7, and 260:280 ratios were >1.7. Custom primers and FAM‐labeled probes were designed using Primer Express 3 software (Life Technologies) for qRT‐PCR. The qRT‐PCR was performed from 25 ng of total RNA in a one‐step reaction (QuantiFast One‐Step RT‐PCR kit; Qiagen) using a Roche LightCycler 480 II. The relative expression of lncRNAs was determined using the ΔΔCt method, following normalization to 18S RNA. GAPDH expression (relative to 18S RNA) was comparable between non‐OA and OA cartilage in both hip and knee samples (see Supplementary Figure 1, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.39520/abstract).

using the ΔΔCt method, following normalization to 18S RNA. GAPDH expression (relative to 18S RNA) was comparable between non‐OA and OA cartilage in both hip and knee samples (see Supplementary Figure 1, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.39520/abstract). Inhibition of lincRNA expression in human chondrocytes using locked nucleic acid (LNA) GapmeRs The human chondrocyte cell line TC28, which was previously characterized by Goldring et al 30, and provided to us as a gift from AstraZeneca, was transfected with either LNAs targeting CILinc01 or CILinc02 (30 nM) or with LNA control (30 nM) using Lipofectamine 2000 (Life Technologies). Following 24‐hour transfection, cells were stimulated (in 0.1% FCS culture media in the absence of antibiotics and amphotericin) for either 4 hours or 24 hours with IL‐1β (1 ng/ml). Supernatants were collected for subsequent cytokine analysis with Luminex. Cells were lysed with RLT (Qiagen) for subsequent RNA extraction to examine knockdown of CILinc01 expression.

mulated (in 0.1% FCS culture media in the absence of antibiotics and amphotericin) for either 4 hours or 24 hours with IL‐1β (1 ng/ml). Supernatants were collected for subsequent cytokine analysis with Luminex. Cells were lysed with RLT (Qiagen) for subsequent RNA extraction to examine knockdown of CILinc01 expression. Analysis of cytokine production in human chondrocyte supernatants Supernatants from human chondrocyte–transfected cells and media controls were assayed for the concentration of 17 human proinflammatory cytokines using a human cytokine 17‐plex immunoassay (Bio‐Plex Pro; Bio‐Rad). The interassay variability is <15%; intraassay variability is <10%. Cross‐reactivity is <1%, and the dynamic range is between 1 and 2,500 pg/ml. Briefly, nondiluted chondrocyte cell culture supernatants were incubated with a magnetic Bio‐Plex bead cocktail consisting of beads specific for IL‐1β, IL‐2, IL‐4, IL‐5, IL‐6, IL‐7, IL‐8, IL‐10, IL‐12 (p70), IL‐13, IL‐17, granulocyte colony‐stimulating factor (G‐CSF), granulocyte–macrophage colony‐stimulating factor, interferon‐γ, monocyte chemotactic protein 1, macrophage inflammatory protein 1β (MIP‐1β), and TNF. A Bio‐Plex Pro Wash Station was used to wash the beads between incubation steps using the wash buffer supplied with the kit. A biotinylated secondary antibody was added, and quantification was carried out using a streptavidin–phycoerythrin substrate with fluorescence detected on a Bio‐Plex 200 System (Bio‐Rad/Luminex).

A Bio‐Plex Pro Wash Station was used to wash the beads between incubation steps using the wash buffer supplied with the kit. A biotinylated secondary antibody was added, and quantification was carried out using a streptavidin–phycoerythrin substrate with fluorescence detected on a Bio‐Plex 200 System (Bio‐Rad/Luminex). Statistical analysis Data were analyzed using SPSS software. Analysis of variance was performed throughout, followed by Fisher's least significant difference post hoc test, where appropriate. In all cases, data are presented as the mean ± SEM, and P values less than 0.05 were considered significant.

A Bio‐Plex Pro Wash Station was used to wash the beads between incubation steps using the wash buffer supplied with the kit. A biotinylated secondary antibody was added, and quantification was carried out using a streptavidin–phycoerythrin substrate with fluorescence detected on a Bio‐Plex 200 System (Bio‐Rad/Luminex). Statistical analysis Data were analyzed using SPSS software. Analysis of variance was performed throughout, followed by Fisher's least significant difference post hoc test, where appropriate. In all cases, data are presented as the mean ± SEM, and P values less than 0.05 were considered significant. RESULTS RNAseq transcriptome profile of primary human OA chondrocytes in response to stimulation with IL‐1β IL‐1β stimulation of primary human hip OA chondrocytes (n = 3 patients) induced a rapid release of IL‐6 protein that peaked at 4 hours and remained elevated at 24 hours (Figure 1B). IL‐1β stimulation also induced a significant increase in the release of MMP‐13 at 24 hours (see Supplementary Figure 2, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.39520/abstract). Analysis of RNAseq data for Gencode‐annotated messenger RNAs (mRNAs) showed that 499 protein‐coding genes were differentially expressed upon IL‐1β stimulation (382 up‐regulated and 117 down‐regulated) (Figure 1C and Supplementary Table 2, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.39520/abstract). As expected, the up‐regulated genes from this set were significantly enriched (FDR <0.05) in Kyoto Encyclopedia of Genes and Genomes pathways involved in the inflammatory response (Figure 1D). There were no significantly enriched pathways in down‐regulated genes. This initial evaluation therefore demonstrated rapid and widespread induction of inflammatory gene expression following IL‐1β stimulation of human chondrocytes.

lopedia of Genes and Genomes pathways involved in the inflammatory response (Figure 1D). There were no significantly enriched pathways in down‐regulated genes. This initial evaluation therefore demonstrated rapid and widespread induction of inflammatory gene expression following IL‐1β stimulation of human chondrocytes. Identification of novel lncRNAs in chondrocytes by RNAseq Using the computational analysis pathway described in Figure 1A, we identified 983 lncRNAs in human chondrocytes, which could be divided into 642 lincRNAs, 124 antisense RNAs, and 217 pseudogenes (see Supplementary Table 3, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.39520/abstract). Of these assembled genes, 158 lincRNAs and 25 antisense RNAs had not previously been identified in Gencode version 19 or HumanBodyMap lncRNA (Figures 1E and F). As previously reported 14, 15, the mean FPKM, length, and exon number for lncRNAs were smaller than those for mRNAs (mean FPKM 4.7 for lncRNAs and 29.6 for mRNAs, mean length 1.2 kb for lncRNAs and 2.8 kb for mRNAs, and mean exon number 3.6 for lncRNAs and 16.4 for mRNAs).

19 or HumanBodyMap lncRNA (Figures 1E and F). As previously reported 14, 15, the mean FPKM, length, and exon number for lncRNAs were smaller than those for mRNAs (mean FPKM 4.7 for lncRNAs and 29.6 for mRNAs, mean length 1.2 kb for lncRNAs and 2.8 kb for mRNAs, and mean exon number 3.6 for lncRNAs and 16.4 for mRNAs). Based on sequencing in ∼400 human cell types including chondrocytes, the FANTOM project has recently released an atlas of 43,011 enhancer regions that are characterized by bidirectional transcription of single‐exon efference RNAs (eRNAs) 31. Interestingly, we found that <4% of our identified lncRNAs overlapped with putative eRNA regions (see Supplementary Table 3, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.39520/abstract). Furthermore, visual inspection and the fact that our transcripts were unidirectional and multiexonic indicated that these lncRNAs did not represent eRNAs.

tative eRNA regions (see Supplementary Table 3, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.39520/abstract). Furthermore, visual inspection and the fact that our transcripts were unidirectional and multiexonic indicated that these lncRNAs did not represent eRNAs. Induction of widespread changes in lncRNA expression by IL‐1β stimulation Following IL‐1β stimulation, we identified 125 lncRNAs that were differentially expressed (P < 0.05), including 93 lincRNAs (74%), 13 antisense RNAs (11%), and 19 pseudogenes (15%) (see Supplementary Table 4, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.39520/abstract). Of these, we observed 106 up‐regulated and 19 down‐regulated lncRNAs, of which 37 (30%) were novel lncRNAs. Using the Integrative Genomics Viewer (Broad Institute), the transcription start sites (TSS) for the majority of the 92 differentially expressed lincRNAs were found to be genomically located <5 kb from the TSS of a coding mRNA (Figures 2A and B). Previously, we have referred to these as mRNA‐flanking lincRNAs 21, and it has been suggested that they may regulate the expression of the nearby mRNA. In support of this notion, we found a significant positive correlation between the fold change in expression of an mRNA‐flanking lincRNA and the fold change in expression of its nearest coding mRNA (Figure 2C). In addition, detailed examination of these differentially expressed mRNA‐flanking lincRNAs identified one as being PACER (Table 1). As previously described, PACER is located upstream of the PTGS2 (COX‐2) gene, is transcribed in a bidirectional manner from the same promoter region, and is known to positively regulate PTGS2 expression 23 (Figure 2A). However, whether this is true of other mRNA‐flanking lincRNAs remains to be elucidated.

ER (Table 1). As previously described, PACER is located upstream of the PTGS2 (COX‐2) gene, is transcribed in a bidirectional manner from the same promoter region, and is known to positively regulate PTGS2 expression 23 (Figure 2A). However, whether this is true of other mRNA‐flanking lincRNAs remains to be elucidated. Figure 2 Location and expression of chondrocyte inflammation–associated long intergenic noncoding RNAs (lincRNAs) in relation to their nearest protein‐coding gene. A, Integrative Genomics Viewer plots showing the mapping data and relative locations of the long noncoding RNA (lncRNA) and protein‐coding genes in hip osteoarthritis (OA) chondrocyte samples left unstimulated (control) and samples stimulated with interleukin‐1β (IL‐1β) for 4 hours. Colors represent the direction of first read. Red blocks represent forward (positive) strand; blue blocks represent reverse (negative) strand; gray blocks represent reads of unknown status. PACER = p50‐associated cyclooxygenase 2–extragenic RNA. B, Dot plot showing the distances between transcription start sites (TSS) of novel lincRNAs and the TSS of their nearest protein‐coding gene. C, Pearson's correlation between absolute fold change in expression of lincRNAs and absolute fold change of their nearest expressed protein‐coding gene in primary human OA chondrocytes after stimulation with IL‐1β for 4 hours. Table 1 Human chondrocyte inflammation–associated lincRNAs and their expression in human OA and non‐OA cartilage tissuea

Figure 2 Location and expression of chondrocyte inflammation–associated long intergenic noncoding RNAs (lincRNAs) in relation to their nearest protein‐coding gene. A, Integrative Genomics Viewer plots showing the mapping data and relative locations of the long noncoding RNA (lncRNA) and protein‐coding genes in hip osteoarthritis (OA) chondrocyte samples left unstimulated (control) and samples stimulated with interleukin‐1β (IL‐1β) for 4 hours. Colors represent the direction of first read. Red blocks represent forward (positive) strand; blue blocks represent reverse (negative) strand; gray blocks represent reads of unknown status. PACER = p50‐associated cyclooxygenase 2–extragenic RNA. B, Dot plot showing the distances between transcription start sites (TSS) of novel lincRNAs and the TSS of their nearest protein‐coding gene. C, Pearson's correlation between absolute fold change in expression of lincRNAs and absolute fold change of their nearest expressed protein‐coding gene in primary human OA chondrocytes after stimulation with IL‐1β for 4 hours. Table 1 Human chondrocyte inflammation–associated lincRNAs and their expression in human OA and non‐OA cartilage tissuea LincRNA LncRNA number Position Nearest gene (kb to TSS) Fold change in expression after IL‐1β stimulation, log2 PTGS2‐lincRNA (PACER) XLOC_081995 chr9:21682903‐21689760 PTGS2 (0.188) 3.1b CILinc01 XLOC_043077 chr6:143267747‐143280112 HIVEP2 (1.409) 6.0b CILinc02 XLOC_078832 chr8:79717154‐79798424 IL‐7 (0.604) 7.9b CILinc03 XLOC_080615 chr8:90627962‐90765918 RIPK2 (4.056) 2.6b

LincRNA LncRNA number Position Nearest gene (kb to TSS) Fold change in expression after IL‐1β stimulation, log2 PTGS2‐lincRNA (PACER) XLOC_081995 chr9:21682903‐21689760 PTGS2 (0.188) 3.1b CILinc01 XLOC_043077 chr6:143267747‐143280112 HIVEP2 (1.409) 6.0b CILinc02 XLOC_078832 chr8:79717154‐79798424 IL‐7 (0.604) 7.9b CILinc03 XLOC_080615 chr8:90627962‐90765918 RIPK2 (4.056) 2.6b CILinc04 XLOC_048072 chr21:43188194‐43194760 RIPK4 (0.928) 3.7b CILinc05 XLOC_072067 chr6:138175998‐138186493 TNFAIP3 (1.857) 1.8c CILinc06 XLOC_076579 chr7:80553659‐80558813 SEMA3C (7.138) 1.4b CILinc07 XLOC_048423 chr21:28984539‐29019990 ADMATS5 (681.158) 5.0b a LincRNA = long intergenic noncoding RNA; OA = osteoarthritis; lncRNA = long noncoding RNA; TSS = transcription start site; IL‐1β = interleukin‐1β; PACER = p50‐associated COX‐2–extragenic RNA. b = P < 0.001 versus unstimulated control chondrocytes. c = P < 0.01 versus unstimulated control chondrocytes. Differential expression of inflammation‐associated lincRNAs in human hip OA and knee OA cartilage We next wished to further characterize the expression of PACER as well as 7 additional chondrocyte inflammation–associated lincRNAs (named CILinc01–CILinc07) that were selected based on being significantly induced in response to IL‐1β stimulation (Table 1) and their nearest coding mRNA being a gene with purported evidence of a role in either inflammation or OA pathology (e.g., IL‐7 and ADAMTS‐5, respectively).

ocyte inflammation–associated lincRNAs (named CILinc01–CILinc07) that were selected based on being significantly induced in response to IL‐1β stimulation (Table 1) and their nearest coding mRNA being a gene with purported evidence of a role in either inflammation or OA pathology (e.g., IL‐7 and ADAMTS‐5, respectively). We initially determined the potential clinical relevance of these chondrocyte inflammation–associated lincRNAs by measuring their expression in human OA hip cartilage compared to non‐OA hip cartilage. All 8 lincRNAs were found to be significantly down‐regulated in OA hip cartilage (n = 9 patients) compared to non‐OA hip cartilage (n = 6 patients) (Figure 3A). The lincRNAs PACER, CILinc01, and CILinc02 were also significantly down‐regulated (>2‐fold) in OA knee cartilage (n = 12) compared to non‐OA knee cartilage (n = 4) (Figure 3B). Figure 3 Expression of chondrocyte inflammation–associated long intergenic noncoding RNAs (lincRNAs) in human hip osteoarthritis (OA) and knee OA cartilage compared to non‐OA cartilage. Graphs show the relative expression of 8 chondrocyte inflammation–associated lincRNAs, as determined by quantitative reverse transcriptase–polymerase chain reaction, in A, OA hip femoral head articular cartilage (n = 9 patients) compared to non‐OA hip femoral head articular cartilage (n = 6 patients), and B, OA knee cartilage (n = 12 patients) compared to non‐OA knee cartilage (n = 4 patients). Bars show the mean ± SEM. ∗ = P < 0.05; † = P < 0.01; ‡ = P < 0.001, by one‐way analysis of variance. PACER = p50‐associated cyclooxygenase 2–extragenic RNA.

ed to non‐OA hip femoral head articular cartilage (n = 6 patients), and B, OA knee cartilage (n = 12 patients) compared to non‐OA knee cartilage (n = 4 patients). Bars show the mean ± SEM. ∗ = P < 0.05; † = P < 0.01; ‡ = P < 0.001, by one‐way analysis of variance. PACER = p50‐associated cyclooxygenase 2–extragenic RNA. Rapid, transient induction of lincRNAs by multiple proinflammatory cytokines Based on their induction in response to IL‐1β stimulation, and their differential expression in both hip OA and knee OA cartilage, we next examined the time course of expression of PACER, CILinc01, and CILinc02 in primary OA chondrocytes in response to a panel of proinflammatory cytokines implicated in the pathogenesis of OA. Following stimulation with IL‐1β, TNF, visfatin, and leptin, we observed a rapid and time‐dependent induction of expression of all 3 lincRNAs (Figure 4A). Of note, stimulation with either TNF or leptin led to peak lincRNA expression at ∼2 hours, which had dropped toward baseline levels by 24 hours. Stimulation with IL‐1β or visfatin led to a slightly more prolonged induction of lincRNA expression, with peak induction of CILinc01 and CILinc02 between 4 and 6 hours (Figure 4A). Of note, stimulation with IL‐1β for 4 hours also led to significant (P < 0.001) induction of the expression of mRNA for the closest coding genes to PACER, CILinc01, and CILinc02, namely, PTGS2, HIVEP2, and IL‐7, respectively (Figure 4B).

n, with peak induction of CILinc01 and CILinc02 between 4 and 6 hours (Figure 4A). Of note, stimulation with IL‐1β for 4 hours also led to significant (P < 0.001) induction of the expression of mRNA for the closest coding genes to PACER, CILinc01, and CILinc02, namely, PTGS2, HIVEP2, and IL‐7, respectively (Figure 4B). Figure 4 Rapid and transient induction of chondrocyte inflammation–associated long intergenic noncoding RNAs (lincRNAs) in primary human osteoarthritis (OA) chondrocytes in response to proinflammatory cytokines. A, Time course of lincRNA expression in primary human hip OA chondrocytes over 24 hours following exposure to either tumor necrosis factor (TNF; 1 ng/ml), leptin (100 ng/ml), visfatin (100 ng/ml), or interleukin‐1β (IL‐1β; 1 ng/ml). Symbols and error bars indicate the mean ± SEM. ∗ = P < 0.05; † = P < 0.01; ‡ = P < 0.001 versus time 0, by two‐way analysis of variance with a least significant difference post hoc test. B, Primary OA chondrocyte expression of PTGS2, HIVEP2, and IL‐7 genes in response to 4 hours of stimulation with IL‐1β (1 ng/ml). Bars show the mean ± SEM. ‡ = P < 0.001 versus unstimulated samples, by two‐way analysis of variance with a least significant difference post hoc test. C, Expression of p50‐associated cyclooxygenase 2–extragenic RNA (PACER), CILinc01, and CILinc02 in primary non‐OA chondrocytes isolated from cartilage from patients with fracture of the neck of the femur (#NOF) and from postmortem (PM) cartilage and in primary OA chondrocytes isolated from hip and knee cartilage. Expression of lincRNAs and genes was determined by quantitative reverse transcriptase–polymerase chain reaction and is shown as fold change compared to control. Bars show the mean ± SEM from 3 independent experiments. ∗ = P < 0.05; † = P < 0.01; ‡ = P < 0.001 versus unstimulated samples, by two‐way analysis of variance with a least significant difference post hoc test.

quantitative reverse transcriptase–polymerase chain reaction and is shown as fold change compared to control. Bars show the mean ± SEM from 3 independent experiments. ∗ = P < 0.05; † = P < 0.01; ‡ = P < 0.001 versus unstimulated samples, by two‐way analysis of variance with a least significant difference post hoc test. We then assessed whether PACER, CILinc01, and CILinc02 were also present in non‐OA chondrocytes and whether stimulation of these cells with IL‐1β would also induce their expression. PACER, CILinc01, and CILinc02 were expressed in both non‐OA knee chondrocytes (isolated from postmortem cartilage) and non‐OA hip chondrocytes (isolated from patients with fracture of the neck of the femur). Furthermore, 4 hours of IL‐1β stimulation of both non‐OA knee and non‐OA hip chondrocytes led to a significant increase in expression of each of the 3 lincRNAs (Figure 4C).

knee chondrocytes (isolated from postmortem cartilage) and non‐OA hip chondrocytes (isolated from patients with fracture of the neck of the femur). Furthermore, 4 hours of IL‐1β stimulation of both non‐OA knee and non‐OA hip chondrocytes led to a significant increase in expression of each of the 3 lincRNAs (Figure 4C). Negative regulation of the IL‐1β–stimulated production of proinflammatory cytokines in human chondrocytes by CILinc01 and CILinc02 Given the association of CILinc01 and CILinc02 with the IL‐1β chondrocyte inflammatory response, and their down‐regulation in OA cartilage tissue, we speculated that CILinc01 and CILinc02 might mediate the production of proinflammatory cytokines. To test this hypothesis, we examined the effect of knockdown of CILinc01 and CILinc02 expression on the human chondrocyte inflammatory response. For these experiments, we used the human chondrocyte TC28 cell line, which when incubated in low serum (0.1% FCS) without stimulation expressed type II collagen (see Supplementary Figures 3A and B, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.39520/abstract). Similar to the findings in primary chondrocytes, IL‐1β stimulation of TC28 cells induced a rapid release of IL‐6 protein (Supplementary Figure 3C) and induction of MMPs and proinflammatory cytokines (Supplementary Figure 3D). TC28 cells were transfected with either LNA GapmeRs targeting CILinc01 or CILinc02, or a nontargeting control LNA GapmeR. Following 24 hours of transfection, cells were stimulated with IL‐1β for 4 hours in order to provoke an inflammatory response. Similar to our findings in primary human chondrocytes, 4 hours of exposure of the TC28 chondrocyte cell line to IL‐1β led to a significant induction of expression of CILinc01 and CILinc02. The IL‐1β–induced expression of CILinc01 was significantly reduced (by 63%) in chondrocytes transfected with an anti‐CILinc01 LNA GapmeR, and CILinc02 expression was significantly reduced (by 74%) in cells transfected with an anti‐CILinc02 GapmeR, compared to an LNA control sequence (Figure 5A).

CILinc01 and CILinc02. The IL‐1β–induced expression of CILinc01 was significantly reduced (by 63%) in chondrocytes transfected with an anti‐CILinc01 LNA GapmeR, and CILinc02 expression was significantly reduced (by 74%) in cells transfected with an anti‐CILinc02 GapmeR, compared to an LNA control sequence (Figure 5A). Figure 5 Long intergenic noncoding RNAs (lincRNAs) modulate the interleukin‐1β (IL‐1β)–stimulated induction of proinflammatory cytokines in human chondrocytes and are suppressed by IKK‐2 inhibition. A, Knockdown of IL‐1β–stimulated CILinc01 and CILinc02 human chondrocyte TC28 cells using LNA GapmeRs. TC28 cells were transfected overnight either with locked nucleic acids (LNAs) targeting CILinc01 or targeting CILinc02 or with a nontargeting control (NTC) LNA GapmeR. Following transfection, cells were left unstimulated or stimulated with IL‐1β (1 ng/ml) for 4 hours. Bars show the mean ± SEM. ∗ = P < 0.05; ‡ = P < 0.001, versus nontargeting control LNA–transfected cells. B and C, Concentration of cytokines (pg/ml) in supernatants from human chondrocytes transfected with either B, CILinc01 LNA or C, CILinc02 LNA and then stimulated with 1 ng/ml of IL‐1β for 4 hours. Bars show the mean ± SEM (n = 3 samples per group). ∗ = P < 0.05; † = P < 0.01, versus nontargeting control LNA–transfected cells. D, Suppression of the IL‐1β–stimulated induction of CILinc01 and CILinc02 in primary human OA chondrocytes preincubated with the IKK‐2 inhibitor TPCA‐1 (10 μM). Bars show the mean ± SEM (n = 3 samples per group). ∗ = P < 0.05; ‡ = P < 0.001, by one‐way analysis of variance. TNF = tumor necrosis factor; G‐CSF = granulocyte colony‐stimulating factor; MIP‐1β = macrophage inflammatory protein 1β.

n primary human OA chondrocytes preincubated with the IKK‐2 inhibitor TPCA‐1 (10 μM). Bars show the mean ± SEM (n = 3 samples per group). ∗ = P < 0.05; ‡ = P < 0.001, by one‐way analysis of variance. TNF = tumor necrosis factor; G‐CSF = granulocyte colony‐stimulating factor; MIP‐1β = macrophage inflammatory protein 1β. We then investigated the effect of CILinc01 and CILinc02 knockdown on the inflammatory response, by measuring the secretion of a panel of 17 proinflammatory cytokines in response to 4 hours of IL‐1β stimulation of human chondrocytes (see Supplementary Table 5, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art.39520/abstract). Knockdown of CILinc01 expression significantly enhanced the IL‐1β–stimulated production of IL‐6, IL‐8, TNF, MIP‐1β, and G‐CSF (Figure 5B), while knockdown of CILinc02 expression significantly enhanced the IL‐1β–stimulated production of IL‐6 (Figure 5C). Since previous studies have shown that NF‐κB activity can regulate the expression of lncRNAs, we then also examined the effect of pharmacologic inhibition of IKK‐2 on the IL‐1β–stimulated production of CILinc01 and CILinc02. To this end we used TPCA‐1, a known IKK‐2 inhibitor 20, 32, 33. In cell‐free enzymatic assays, TPCA‐1 displays 22‐fold selectivity for IKK‐2 over IKK‐1 and a >550‐fold selectivity over other kinases, including MAP kinases and JNK kinases 32, though a recent study showed that in non–small cell lung cancer cell lines TPCA‐1 also inhibited STAT‐3 phosphorylation 34. Preincubation of primary chondrocytes with TPCA‐1 (10 μM) significantly reduced the induction of both CILinc01 and CILinc02 that occurred after 4 hours of stimulation with IL‐1β. (Figure 5D).

32, though a recent study showed that in non–small cell lung cancer cell lines TPCA‐1 also inhibited STAT‐3 phosphorylation 34. Preincubation of primary chondrocytes with TPCA‐1 (10 μM) significantly reduced the induction of both CILinc01 and CILinc02 that occurred after 4 hours of stimulation with IL‐1β. (Figure 5D). DISCUSSION This study is the first to use RNAseq to determine the profile of lncRNA expression in primary human OA chondrocytes and has resulted in the cataloging of 983 lncRNAs, including members of the lincRNA, antisense RNA, and pseudogene families. Importantly, we have identified 158 lincRNAs and 25 antisense RNAs that are absent from Gencode version 19 35 and the HumanBodyMap lncRNA catalog 29, and might therefore be unique to chondrocytes and have a cell‐specific function. In addition, this study is the first to examine the changes in lncRNA levels that are associated with the inflammatory response in human chondrocytes. In this regard, 125 lncRNAs were differentially expressed upon IL‐1β stimulation of human OA chondrocytes. Of relevance, Fu et al 22 recently showed a catalog of 4,714 lncRNAs found by microarray analysis to be differentially expressed in knee OA patients compared to non‐OA cartilage. In our RNAseq chondrocyte analysis, if we included lncRNAs with a P value of less than 0.05 (rather than an FDR optimized q of <0.05), which was the inclusion criterion used by Fu et al 22, 7 of these lncRNAs (namely, ENST00000426066, ENST00000369884, ENST00000419463, ENST00000421237, ENST00000412485, ENST00000455607, and ENST00000418242) were differentially expressed in chondrocytes upon IL‐1β stimulation. This relatively low number of lncRNAs in common is likely due to differences in conditions (IL‐1β stimulation of chondrocytes versus end‐stage cartilage disease comparison) and methodologic approach (sequencing versus microarrays). As an example, the microarray studies by Fu et al 22 would not have detected changes in PACER, CILinc01, and CILinc02 since these are novel transcripts for which there are no microarray probes. Despite these differences, we speculate that these shared lncRNAs might have a function in OA, which would warrant further investigation.

le, the microarray studies by Fu et al 22 would not have detected changes in PACER, CILinc01, and CILinc02 since these are novel transcripts for which there are no microarray probes. Despite these differences, we speculate that these shared lncRNAs might have a function in OA, which would warrant further investigation. Importantly, there is now evidence that lncRNAs regulate in cis local mRNA expression 21, 36. Indeed, among those lncRNAs differentially expressed upon IL‐1β stimulation was the lincRNA PACER 23, which is located adjacent to and upstream of the gene PTGS2 (COX‐2) and has been shown to regulate PTGS2 production 23. As shown in the present study, PACER appears to be transcribed from the same promoter regions as PTGS2, which results in bidirectional production of both coding and noncoding RNA. Significantly, the majority of the inflammation‐associated lncRNAs we identified were found to be mRNA flanking, several of which (including PACER) were located close to genes relevant to either inflammation or cartilage biology, which could be indicative of a functional role in OA.

on of both coding and noncoding RNA. Significantly, the majority of the inflammation‐associated lncRNAs we identified were found to be mRNA flanking, several of which (including PACER) were located close to genes relevant to either inflammation or cartilage biology, which could be indicative of a functional role in OA. Given these observations, we selected PACER and 7 additional inflammation‐associated lincRNAs and proceeded to investigate their potential clinical relevance by comparing their expression in articular hip and knee cartilage obtained from both OA and non‐OA patients. Notably, all 8 of the inflammation‐associated lincRNAs were found to be significantly down‐regulated in hip OA cartilage, while only PACER, CILinc01, and CILinc02 were also down‐regulated in knee OA cartilage. This could indicate that these lincRNAs perform protective roles in preventing inflammation‐mediated cartilage degeneration, but also suggests that there are anatomic site–specific differences in OA cartilage at the level of lncRNA expression. Indeed, a recent report described epigenetic differences between knee and hip OA cartilage based on DNA methylation analyses 37, and previous studies have shown differences in dysregulated mRNA transcripts and pathways between knee OA and hip OA cartilage 38. It should be noted that there were differences in K/L grade between our hip OA and knee OA patients. However, all of our knee OA samples were K/L grade 4, while our hip OA samples were either K/L grade 3 or K/L grade 4, so it would appear unlikely that the differential expression of all 8 inflammation‐associated lncRNAs in hip OA cartilage was due to differences in OA severity.

between our hip OA and knee OA patients. However, all of our knee OA samples were K/L grade 4, while our hip OA samples were either K/L grade 3 or K/L grade 4, so it would appear unlikely that the differential expression of all 8 inflammation‐associated lncRNAs in hip OA cartilage was due to differences in OA severity. Subsequent experiments demonstrated that PACER, CILinc01, and CILinc02 were induced in OA chondrocytes by multiple proinflammatory cytokines, which have been reported to be elevated in either OA sera or OA synovial fluid (TNF, visfatin, and leptin as well as IL‐1β). Importantly, the induction of chondrocyte lincRNA expression in response to multiple proinflammatory stimuli was rapid and transient, as might be expected if they were key regulators of the inflammatory response in joint cartilage.

ted in either OA sera or OA synovial fluid (TNF, visfatin, and leptin as well as IL‐1β). Importantly, the induction of chondrocyte lincRNA expression in response to multiple proinflammatory stimuli was rapid and transient, as might be expected if they were key regulators of the inflammatory response in joint cartilage. Given that PACER has previously been shown to positively regulate PTGS2 production 23 and that PTGS2 is associated with inflammation, we were initially surprised to discover that PACER was down‐regulated in hip OA cartilage. However, there are reports that PTGS2 expression in OA synovial tissue is significantly lower in late OA compared to early OA 39, suggesting it may play a different role in established human OA. Furthermore, PTGS2 has also been implicated as having an antiinflammatory functional role 40, since the release of prostaglandin D2 (PGD2) and its breakdown product PGDJ2 are associated with the resolution of inflammation 41. Indeed, in stark contrast to their efficacy in blocking proinflammatory responses, inhibitors of COX‐2 have been shown to delay the resolution of inflammation 42. Therefore, the decreased expression of PACER we observed in human hip OA cartilage could represent a pathologic reduction in the ability of the cartilage tissue to resolve aberrant inflammation.

ficacy in blocking proinflammatory responses, inhibitors of COX‐2 have been shown to delay the resolution of inflammation 42. Therefore, the decreased expression of PACER we observed in human hip OA cartilage could represent a pathologic reduction in the ability of the cartilage tissue to resolve aberrant inflammation. The functional significance of our finding that CILinc02 is down‐regulated in human hip OA cartilage is unclear. Studies in rheumatoid arthritis suggest that IL‐7 (the nearest coding gene to CILinc02) contributes to inflammation 43 and mediates the production of TNF 44, while in OA, IL‐7 has been reported to induce MMP‐13 and proteoglycan loss from cartilage, suggesting that it may promote cartilage degeneration 45. However, we did not detect expression of the IL‐7 gene in either OA or non‐OA hip cartilage samples.

contributes to inflammation 43 and mediates the production of TNF 44, while in OA, IL‐7 has been reported to induce MMP‐13 and proteoglycan loss from cartilage, suggesting that it may promote cartilage degeneration 45. However, we did not detect expression of the IL‐7 gene in either OA or non‐OA hip cartilage samples. Functional studies to determine the roles of CILinc01 and CILinc02 showed that knockdown of their expression in human chondrocytes significantly increased the IL‐1β–stimulated production of several proinflammatory cytokines, including IL‐6, suggesting that CILinc01 and CILinc02 may negatively regulate the chondrocyte inflammatory response. It is significant, therefore, that we found decreased expression of CILinc01 and CILinc02 in both knee OA and hip OA cartilage compared to normal healthy cartilage, since this could indicate that down‐regulation of CILinc01 and CILinc02 in human articular cartilage leads to an inability to regulate inflammation in the joint. Of interest, the nearest coding gene to CILinc01 is HIVEP2 (also known as Schnurri‐2), which has previously been reported to be a negative regulator of allergic airway inflammation via repression of NF‐κB activity 46, as well as being implicated in mediating chondrocyte differentiation 47. Therefore, it is conceivable that the observed effects of CILinc01 on chondrocyte cytokine production are mediated via repression of NF‐κB activity through modulation of HIVEP2 gene expression. Of note, stimulation of primary chondrocytes with an IKK‐2 inhibitor blocked the IL‐1β–stimulated production of both CILinc01 and CIlinc02, suggesting that NF‐κB activity may regulate their expression in chondrocytes.

production are mediated via repression of NF‐κB activity through modulation of HIVEP2 gene expression. Of note, stimulation of primary chondrocytes with an IKK‐2 inhibitor blocked the IL‐1β–stimulated production of both CILinc01 and CIlinc02, suggesting that NF‐κB activity may regulate their expression in chondrocytes. In conclusion, these data signify that CILinc01 and CILinc02 may play an important physiologic role in regulating the pathologic response to inflammation within the OA joint, and that its down‐regulation in both knee and hip OA cartilage could contribute to inflammation‐driven cartilage degeneration. Clearly, future studies to determine the mode of action of CILinc01 and CILinc02 as well as other chondrocyte inflammation–associated lincRNAs in mediating OA cartilage pathology and inflammation are warranted and may lead to the identification of novel targets for the development of therapeutic agents. AUTHOR CONTRIBUTIONS 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. Jones 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 Pearson, Lindsay, Jones. Acquisition of data Pearson, Philp, Heward, Roux, Walsh, Davis, Lindsay, Jones. Analysis and interpretation of data Pearson, Philp, Heward, Roux, Davis, Lindsay, Jones. Supporting information

AUTHOR CONTRIBUTIONS 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. Jones 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 Pearson, Lindsay, Jones. Acquisition of data Pearson, Philp, Heward, Roux, Walsh, Davis, Lindsay, Jones. Analysis and interpretation of data Pearson, Philp, Heward, Roux, Davis, Lindsay, Jones. Supporting information Supplementary Figure 1. Expression of GAPDH in (A) non‐OA hip (n=6) and OA hip (n=9), and (B) non‐OA knee (n=4) and OA knee (n=12) cartilage tissue. Expression was normalised to 18S using the ΔΔCt method. Click here for additional data file. Supplementary Figure 2. Release of MMP13 following 4h and 24h IL‐1β (1ng/m1) stimulation of primary OA chondrocytes as measured by ELISA. Bars represent mean ± SEM (n=3). ‡= P < 0.001, significantly different from control. Click here for additional data file.

Supplementary Figure 1. Expression of GAPDH in (A) non‐OA hip (n=6) and OA hip (n=9), and (B) non‐OA knee (n=4) and OA knee (n=12) cartilage tissue. Expression was normalised to 18S using the ΔΔCt method. Click here for additional data file. Supplementary Figure 2. Release of MMP13 following 4h and 24h IL‐1β (1ng/m1) stimulation of primary OA chondrocytes as measured by ELISA. Bars represent mean ± SEM (n=3). ‡= P < 0.001, significantly different from control. Click here for additional data file. Supplementary Figure 3. (A) Relative expression of COL2A1 in freshly isolated primary human chondrocytes (P0) and TC28s. (B) Western blot showing COL2A1 (∼55 kDa) in freshly isolated primary chondrocytes (P0) and passage 2 (P2) chondrocytes and TC28s. Mouse monoclonal anti‐COL2A1 (Sigma SAB1403684) was used at 1:1000 dilutiion. (C) IL‐6 cytokine release in response to 4h IL‐1β stimulation of primary OA chondrocytes and TC28s, as measured by ELISA. (D) Induction of pro‐inflammatory cytokines and MMPs in response to 4h IL‐1β stimulation of TC28s. Bars represent mean ± SEM, *=P < 0.05, †=P < 0.01, significantly different from control. Click here for additional data file. Supplementary Table 1. Patient demographics. X‐ray radiographs were assessed by a clinician to determined KL grade. X‐ray radiographs were not available for post‐mortem subjects. Click here for additional data file. Supplementary Table 2: Differentially expressed protein coding transcripts upon 4h IL‐1β stimulation of primary chondrocytes Click here for additional data file. Supplementary Table 3 Click here for additional data file.

Supplementary Table 1. Patient demographics. X‐ray radiographs were assessed by a clinician to determined KL grade. X‐ray radiographs were not available for post‐mortem subjects. Click here for additional data file. Supplementary Table 2: Differentially expressed protein coding transcripts upon 4h IL‐1β stimulation of primary chondrocytes Click here for additional data file. Supplementary Table 3 Click here for additional data file. Supplementary Table 4 Click here for additional data file. Supplementary Table 5. Human Bio‐plex 17‐plex data following 4h IL‐1β stimulation of human TC28 chondrocytes transfected with LNA GAPmeRs. Fold change is shown for those analytes which were significantly different (P < 0.05) relative to the non‐targeting control LNA. Data which were not statistically significant are shown as ‘NS’ whilst analytes which were not detectable are shown as ‘n.d’. IL‐1β (1ng/ml) was used as the cell stimulant and thus was discounted from these data. Click here for additional data file. ACKNOWLEDGMENTS The authors acknowledge all study participants, research staff at The Royal Orthopaedic Hospital NHS Foundation Trust (Birmingham, UK), Russell's Hall Hospital (Dudley, UK) and Kings Mill Hospital (Sutton in Ashfield, UK) for obtaining consents and screening, and the orthopaedic surgeons Drs. David Dunlop, Matthew Revell, and Sohail Quraishi.

Osteoarthritis (OA) is a degenerative joint disease and a health care burden throughout the world. Characterized by articular cartilage loss, subchondral bone thickening, and osteophyte formation, OA causes much pain and disability. Its underlying molecular mechanisms are, nevertheless, not fully understood; indeed, even the precipitating pathology is still a matter of debate. As such, there is an ever‐growing need for an effective disease‐modifying treatment. Canine hip dysplasia is a hereditary predisposition to the development of degenerative OA and is more common in certain breeds, in particular larger breeds which tend to grow more rapidly 1. While no direct link has been made between growth dynamics and OA, recent murine and human studies have prompted speculation that articular cartilage chondrocytes may undergo a transition from their inherently stable phenotype to a more transient one characteristic of the chondrocytes in the growth plate 2, 3, 4, 5, 6, 7, 8, 9.

direct link has been made between growth dynamics and OA, recent murine and human studies have prompted speculation that articular cartilage chondrocytes may undergo a transition from their inherently stable phenotype to a more transient one characteristic of the chondrocytes in the growth plate 2, 3, 4, 5, 6, 7, 8, 9. The epiphyseal growth plates are responsible for long bone development (endochondral ossification) and growth, which is secured by growth plate chondrocytes undergoing differentiation, maturation, hypertrophy, and death, resulting in mineralization of the cartilage matrix 10, 11, 12, 13. Transience of growth plate cartilage chondrocytes is thus a crucial attribute. However, this is in sharp contrast with the inherent stability of articular cartilage chondrocytes, in which these dynamic events must be restricted to assure life long articular integrity and joint function. Interlinks between these apparently discordant phenotypes are not fully understood, and whether switching in these behaviors may contribute to the structural demise of articular cartilage in OA joints has not yet been established 13, 14, 15. However, based on the common embryology of cartilage and bone, along with recent evidence supporting distinct origins of growth plate and articular cartilage chondrocytes, it is not surprising that this hypothesis has been controversial 16, 17, 18. Regardless, an exploration of the mechanisms controlling changes that chondrocytes undergo during their transition through the various stages of endochondral ossification may help to decipher those that underlie pathologic ossification in OA.

it is not surprising that this hypothesis has been controversial 16, 17, 18. Regardless, an exploration of the mechanisms controlling changes that chondrocytes undergo during their transition through the various stages of endochondral ossification may help to decipher those that underlie pathologic ossification in OA. The STR/Ort mouse is a well‐established, natural model of OA, with disease resembling that in humans. Mice develop articular cartilage lesions on the medial tibial plateau, with subchondral bone thickening and expected degenerative changes in other joint tissues beginning at ∼18 weeks of age, coincident with attainment of skeletal maturity 19, 20, 21, 22. CBA mice, the closest available parental strain, show, in contrast, very low spontaneous OA susceptibility 21, 23. We therefore aimed to establish whether an aberrant deployment of the transient chondrocyte phenotype is observed in STR/Ort mouse joints and whether this can be attributed to modified growth dynamics underpinned by an inherent endochondral growth defect. MATERIALS AND METHODS Animals Male STR/Ort mice (bred in‐house) and CBA mice (Charles River) were used in all experiments. All procedures complied with the Animals (Scientific Procedures) Act 1986 and local ethics committee guidelines. Meta‐analysis of microarray data Gene ontology classification, on Affymetrix mouse gene microarray profiling of articular cartilage that we had performed previously 22, was carried out using DAVID (http://david.abcc.ncifcrf.gov/) 24.

MATERIALS AND METHODS Animals Male STR/Ort mice (bred in‐house) and CBA mice (Charles River) were used in all experiments. All procedures complied with the Animals (Scientific Procedures) Act 1986 and local ethics committee guidelines. Meta‐analysis of microarray data Gene ontology classification, on Affymetrix mouse gene microarray profiling of articular cartilage that we had performed previously 22, was carried out using DAVID (http://david.abcc.ncifcrf.gov/) 24. RNA extraction RNA was extracted from the knee joint articular cartilage of STR/Ort and CBA mice at ages 8–10 weeks, 18–20 weeks, and ≥40 weeks (n = 3 joints per strain per age group), as previously described 22. Multiplex quantitative reverse transcriptase–polymerase chain reaction (qRT‐PCR) A GeXP multiplex qRT‐PCR assay was designed for the following gene targets: Ank, Dmp1, Enpp1, Mepe, Opn (Spp1), Phex, and Sost (see Supplementary Table 1, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art39508/abstract). Target‐specific reverse transcription was performed as previously described 25, 26, using 50 ng of total RNA. Immunohistochemistry Immunohistochemical analysis was performed on 6‐μm coronal sections using anti‐sclerostin antibody (1:100 dilution; R&D Systems), anti–matrix metalloproteinase (anti–MMP‐13) antibody (1:200 dilution; Abcam), anti‐Col10a1 antibody (1:500 dilution; provided by Professor R. Boot‐Handford, University of Manchester), or anti‐MEPE antibody (1:200 dilution; provided by Professor P. Rowe, University of Kansas Medical Center, Kansas City, Kansas).

stems), anti–matrix metalloproteinase (anti–MMP‐13) antibody (1:200 dilution; Abcam), anti‐Col10a1 antibody (1:500 dilution; provided by Professor R. Boot‐Handford, University of Manchester), or anti‐MEPE antibody (1:200 dilution; provided by Professor P. Rowe, University of Kansas Medical Center, Kansas City, Kansas). Articular cartilage and growth plate zone analysis. Multiple toluidine blue–stained coronal sections (n => 6) from the joints of 4 individual mice per strain in each age group were used to measure the width of joint compartments and growth plate zones based on established cell morphology 27. Joint imaging by micro–computed tomography (micro‐CT) Mouse joints were scanned using a laboratory source for 5μ voxels and at a synchrotron for 1μ voxels. The laboratory scans were performed using a SkyScan 1172 x‐ray microtomograph to evaluate cortical and trabecular bone geometry. The synchrotron radiation microtomography was performed at Diamond Light Source on the Diamond‐Manchester Branchline I13‐2 with projections being reconstructed and a procedure developed to characterize each individual bridge and map its location on the tibial joint surface 28, 29, 30 (see Supplementary Methods, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art39508/abstract). Metatarsal organ cultures Metatarsal bones (day 15 of embryogenesis) were cultured for up to 7 days 31, 32. The total length of the bone through the center of the mineralizing zone and the length of the central mineralization zone were determined using Image J software.

Joint imaging by micro–computed tomography (micro‐CT) Mouse joints were scanned using a laboratory source for 5μ voxels and at a synchrotron for 1μ voxels. The laboratory scans were performed using a SkyScan 1172 x‐ray microtomograph to evaluate cortical and trabecular bone geometry. The synchrotron radiation microtomography was performed at Diamond Light Source on the Diamond‐Manchester Branchline I13‐2 with projections being reconstructed and a procedure developed to characterize each individual bridge and map its location on the tibial joint surface 28, 29, 30 (see Supplementary Methods, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art39508/abstract). Metatarsal organ cultures Metatarsal bones (day 15 of embryogenesis) were cultured for up to 7 days 31, 32. The total length of the bone through the center of the mineralizing zone and the length of the central mineralization zone were determined using Image J software. Sclerostin enzyme‐linked immunosorbent assay (ELISA). Serum sclerostin levels in CBA and STR/Ort mice at ages 8–10 weeks, 18–20 weeks, and ≥40 weeks (n = 4 for each strain at each age) were measured using a mouse/rat sclerostin ELISA kit (R&D Systems). Statistical analysis Data were analyzed by one‐way analysis of variance, Student's t‐test, or a suitable nonparametric test using GraphPad Prism 6 and following normality checks. All data are expressed as the mean ± SEM.

Sclerostin enzyme‐linked immunosorbent assay (ELISA). Serum sclerostin levels in CBA and STR/Ort mice at ages 8–10 weeks, 18–20 weeks, and ≥40 weeks (n = 4 for each strain at each age) were measured using a mouse/rat sclerostin ELISA kit (R&D Systems). Statistical analysis Data were analyzed by one‐way analysis of variance, Student's t‐test, or a suitable nonparametric test using GraphPad Prism 6 and following normality checks. All data are expressed as the mean ± SEM. RESULTS Retention of calcified cartilage thickness despite articular cartilage loss and subchondral bone thickening in STR/Ort mice We first sought to determine temporospatial patterns of changing joint architecture in STR/Ort mice. Consistent with previous findings 33, we found that young STR/Ort mice had thicker medial tibial articular cartilage than age‐matched CBA controls (P < 0.001). As STR/Ort mice aged, the medial tibial articular cartilage became thinner, with concomitant thickening of subchondral bone, neither of which occurred in CBA mice (P < 0.001) (Figures 1A and B). Despite this, there was no change in calcified cartilage thickness in STR/Ort mice (Figure 1B), and further analysis revealed similar changes in the medial femur (see Supplementary Figure 1, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art39508/abstract). Lateral condyles in STR/Ort mice were unaffected by such age‐related structural modifications. The lateral tibia also exhibited greater articular cartilage thickness, possibly as compensation for altered mechanical loads (see Supplementary Figure 1).

ite at http://onlinelibrary.wiley.com/doi/10.1002/art39508/abstract). Lateral condyles in STR/Ort mice were unaffected by such age‐related structural modifications. The lateral tibia also exhibited greater articular cartilage thickness, possibly as compensation for altered mechanical loads (see Supplementary Figure 1). Figure 1 A and B , Thickness of uncalcified cartilage, calcified cartilage, and subchondral bone in the medial tibia of CBA mice (A) and STR/Ort mice (B) at 8–10 weeks, 18–20 weeks, and ≥40 weeks of age. Ten measurements per section were obtained in >6 sections per mouse (n = 4 mice per age group for each strain). Results are presented as the average percent of the thickness of each zone measured from articular surface to subchondral bone. C–F , Immunolabeling for matrix metalloproteinase 13 (C and D) and for type X collagen (E and F) in the medial tibia (C and E) and lateral tibia (D and F) of STR/Ort mice prior to the onset of osteoarthritis. Arrows indicate positive staining. Images are representative of results in 3 individual mice. G and H , GeXP multiplex quantitative reverse transcription–polymerase chain reaction analysis of mRNA for Enpp1 (G) and Ank (H) in the articular cartilage of CBA and STR/Ort mice at 8–10 weeks, 18–20 weeks, and ≥40 weeks of age. Bars show the mean ± SEM (n = 3 joints per sample; n = 3 samples per age group per strain). ∗∗ = P < 0.01; ∗∗∗ = P < 0.001, versus CBA mice except where indicated otherwise. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/journal/doi/10.1002/art.39508/abstract.

age. Bars show the mean ± SEM (n = 3 joints per sample; n = 3 samples per age group per strain). ∗∗ = P < 0.01; ∗∗∗ = P < 0.001, versus CBA mice except where indicated otherwise. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/journal/doi/10.1002/art.39508/abstract. Transient chondrocyte behaviors in the articular cartilage of STR/Ort mice prior to OA onset We next examined whether the predisposition to age‐related subchondral bone thickening and loss of articular cartilage in STR/Ort mice with OA was linked to the expression of markers of the transient chondrocyte phenotype. Initially, we used the DAVID functional annotation clustering tool to identify biologic functions enriched within differentially expressed genes in articular cartilage from mice with OA (STR/Ort mice ages >18 weeks) compared to unaffected mice (CBA mice ages 8–40 weeks and STR/Ort mice ages 8–10 weeks) 22. Within the major gene ontology classifications, there was significant up‐regulation of endochondral bone growth (P < 0.01) (Table 1 and Supplementary Table 2, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art39508/abstract). No gene ontology classifications associated with skeletal development and ossification were found to be significantly down‐regulated. Table 1 Gene ontology analysis of genes up‐regulated in osteoarthritic mouse articular cartilage compared to nonosteoarthritic mouse articular cartilagea Term % change P Bone development 3.1 1.0 × 10−5 Ossification 2.8 2.8 × 10−5

Transient chondrocyte behaviors in the articular cartilage of STR/Ort mice prior to OA onset We next examined whether the predisposition to age‐related subchondral bone thickening and loss of articular cartilage in STR/Ort mice with OA was linked to the expression of markers of the transient chondrocyte phenotype. Initially, we used the DAVID functional annotation clustering tool to identify biologic functions enriched within differentially expressed genes in articular cartilage from mice with OA (STR/Ort mice ages >18 weeks) compared to unaffected mice (CBA mice ages 8–40 weeks and STR/Ort mice ages 8–10 weeks) 22. Within the major gene ontology classifications, there was significant up‐regulation of endochondral bone growth (P < 0.01) (Table 1 and Supplementary Table 2, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art39508/abstract). No gene ontology classifications associated with skeletal development and ossification were found to be significantly down‐regulated. Table 1 Gene ontology analysis of genes up‐regulated in osteoarthritic mouse articular cartilage compared to nonosteoarthritic mouse articular cartilagea Term % change P Bone development 3.1 1.0 × 10−5 Ossification 2.8 2.8 × 10−5 Bone mineral formation 1.7 1.6 × 10−4 Skeletal system development 4 3.4 × 10−4 Cartilage development 1.7 6.10 × 10−3 a Genes showing a change of >1.5 fold (n = 491) were analyzed with DAVID.

Table 1 Gene ontology analysis of genes up‐regulated in osteoarthritic mouse articular cartilage compared to nonosteoarthritic mouse articular cartilagea Term % change P Bone development 3.1 1.0 × 10−5 Ossification 2.8 2.8 × 10−5 Bone mineral formation 1.7 1.6 × 10−4 Skeletal system development 4 3.4 × 10−4 Cartilage development 1.7 6.10 × 10−3 a Genes showing a change of >1.5 fold (n = 491) were analyzed with DAVID. These endochondral ossification gene ontologies in STR/Ort mouse OA articular cartilage led us to examine protein expression of well‐established chondrocyte hypertrophy markers in OA development. Immunohistochemistry demonstrated positive MMP‐13 labeling in both superficial and deep articular chondrocytes in the joints of STR/Ort mice prior to OA onset (Figures 1C and D). Consistent with previous findings 35, an expected pattern of type X collagen expression was observed in the unaffected condyles of STR/Ort mouse joints, with immunolabeling restricted to hypertrophic chondrocytes (Figure 1F). Intriguingly, type X collagen immunolabeling was observed throughout the medial condylar articular cartilage matrix in 8–10‐week‐old STR/Ort mice, before histologically detectable OA (Figure 1E). Multiplex gene analysis of genes associated with matrix mineralization confirmed significant elevation in Enpp1 and Ank messenger RNA (mRNA) isolated from articular cartilage from 8–10‐week‐old STR/Ort mice, compared to CBA mice (P < 0.001) (Figures 1G and H). Interestingly, these increases were diminished upon OA onset (P < 0.01 for STR/Ort mice at 18–20 weeks versus age‐matched CBA mice for Enpp1) (Figures 1G and H). These findings suggest that OA‐prone regions of the STR/Ort mouse cartilage exhibit an inherent instability defect in articular chondrocytes.

gures 1G and H). Interestingly, these increases were diminished upon OA onset (P < 0.01 for STR/Ort mice at 18–20 weeks versus age‐matched CBA mice for Enpp1) (Figures 1G and H). These findings suggest that OA‐prone regions of the STR/Ort mouse cartilage exhibit an inherent instability defect in articular chondrocytes. Accelerated growth, dysfunctional growth plate morphology, and matrix mineralization in STR/Ort mice To more fully define this inherent endochondral chondrocyte defect, we examined the growth trajectories of 1–8‐week‐old STR/Ort and CBA mice. We found that STR/Ort mice weigh less than CBA mice (P < 0.05) until 4–5 weeks of age, corresponding with the attainment of sexual maturity, when they overtake CBA mice (Figure 2A). Consistent with this finding, analysis of embryonic longitudinal growth and mineralization in vitro revealed aberrant rates of growth and endochondral ossification in STR/Ort mouse metatarsal bones 31, 32 (Figures 2C–F), involving both a slowing of metatarsal growth (Figure 2E) and marked reductions in the mineralized portion of the element (Figure 2F).

sis of embryonic longitudinal growth and mineralization in vitro revealed aberrant rates of growth and endochondral ossification in STR/Ort mouse metatarsal bones 31, 32 (Figures 2C–F), involving both a slowing of metatarsal growth (Figure 2E) and marked reductions in the mineralized portion of the element (Figure 2F). Figure 2 A, Weight of CBA and STR/Ort mice from 1 to 8 weeks of age (n = 5 or more mice per group). B, Length of the tibia in CBA and STR/Ort mice at 3, 6, and 8 weeks of age (n = 6 mice per group). C and D, Digital images of metatarsal bones from STR/Ort mice on day 15 of embryogenesis (C) and CBA mice (D) that were cultured for 7 days and measured. E and F, Growth rate (E) and percent change in mineralization zone (MZ) length (F) of metatarsal bones of CBA and STR/Ort mice (n = 8 or more mice per group). G, Growth plate zone length in CBA and STR/Ort mice at 3, 6, and 8 weeks of age. Ten measurements per section were obtained in >6 sections per mouse (n = 4 mice per age group for each strain). PZ = proliferative zone; HZ = hypertrophic zone. H–K, Immunohistochemical analysis of type X collagen in a 3‐week‐old CBA mouse (H), a 6‐week‐old CBA mouse (I), a 3‐week‐old STR/Ort mouse (J), and a 6‐week‐old STR/Ort mouse (K). L–O, Immunohistochemical analysis of MMP‐13 in a 3‐week‐old CBA mouse (L), a 6‐week‐old CBA mouse (M), a 3‐week‐old STR/Ort mouse (N), and a 6‐week‐old STR/Ort mouse (O). Images are representative of results in 3 individual mice. Values in A, B, E, F, and G are the mean ± SEM. ∗ = P < 0.05; ∗∗ = P < 0.01; ∗∗∗ = P < 0.001. Original magnification × 10 in H–O. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/journal/doi/10.1002/art.39508/abstract.

e representative of results in 3 individual mice. Values in A, B, E, F, and G are the mean ± SEM. ∗ = P < 0.05; ∗∗ = P < 0.01; ∗∗∗ = P < 0.001. Original magnification × 10 in H–O. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/journal/doi/10.1002/art.39508/abstract. Consistent with this accelerated growth phenotype, STR/Ort mice also had shorter tibiae than CBA mice at 3 weeks of age (P < 0.05), which seemed to be reversed, since tibia length at 6 weeks of age was not significantly different from that of CBA mice (Figure 2B). This was further supported by comparisons of 8‐week‐old mice, when the tibia was longer in STR/Ort mice than in age‐matched CBA mice (P < 0.01) (Figure 2B). Micro‐CT analysis showed significantly enhanced cortical and trabecular parameters in 6‐week‐old STR/Ort mouse tibiae, with higher percent differences in bone volume/total volume (12%; P < 0.01), cortical thickness (23%; P < 0.05), cortical area (23%; P < 0.001), polar moment of inertia (46%; P < 0.05), trabecular pattern factor (27%; P < 0.05), and structure model index (13%; P < 0.05) compared to age‐matched CBA mice (see Supplementary Table 3, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art39508/abstract). In comparison, there were no significant differences at 3 weeks of age (Supplementary Table 3).

05), and structure model index (13%; P < 0.05) compared to age‐matched CBA mice (see Supplementary Table 3, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art39508/abstract). In comparison, there were no significant differences at 3 weeks of age (Supplementary Table 3). Growth plate zone analysis showed a significantly enlarged proliferating zone of chondrocytes in both 3‐week‐old and 6‐week‐old STR/Ort mice (P < 0.001 and P < 0.05, respectively) (Figure 2G). This was not apparent in 8‐week‐old STR/Ort mice when compared to age‐matched CBA mice (Figure 2G). Despite the lack of any differences in the size of hypertrophic chondrocyte zones, immunolabeling for type X collagen showed the expected localization in CBA mouse growth plates, limited exclusively to the hypertrophic zone and underlying metaphyseal bone at both 3 weeks and 6 weeks of age (Figures 2H and I). In contrast, STR/Ort mice at both ages showed considerably greater and more widely dispersed type X collagen expression, which extended additionally into the proliferative chondrocyte zone (Figures 2J and K). This disrupted distribution of growth plate zone markers was also evident for MMP‐13 in the growth plates of STR/Ort mice (Figures 2L–O).

at both ages showed considerably greater and more widely dispersed type X collagen expression, which extended additionally into the proliferative chondrocyte zone (Figures 2J and K). This disrupted distribution of growth plate zone markers was also evident for MMP‐13 in the growth plates of STR/Ort mice (Figures 2L–O). Link between OA in STR/Ort mice and modifications in the MEPE/sclerostin axis Molecular mechanisms controlling endochondral ossification may help identify those involved in OA. We have previously shown MEPE, a member of the small integrin‐binding ligand N‐linked glycoprotein (SIBLING) family, to be a negative regulator of growth plate chondrocyte matrix mineralization 31. Examination of MEPE expression by multiplex analysis showed significantly higher levels of Mepe mRNA in STR/Ort mouse articular cartilage than CBA mouse articular cartilage (P < 0.001) (Figure 3A). Immunolabeling for MEPE showed differential expression across the tibia of the STR/Ort mouse joints, with a distinct lack of positive MEPE protein labeling in the medial (affected) aspects both prior to and during OA progression (Figure 3C). This is in contrast to the lateral (unaffected) aspect of the STR/Ort mouse joints and throughout all aspects of CBA mouse joints, where labeling for MEPE was observed throughout the depth of the articular cartilage tissue (Figures 3C and D).

edial (affected) aspects both prior to and during OA progression (Figure 3C). This is in contrast to the lateral (unaffected) aspect of the STR/Ort mouse joints and throughout all aspects of CBA mouse joints, where labeling for MEPE was observed throughout the depth of the articular cartilage tissue (Figures 3C and D). Figure 3 A and B, GeXP multiplex quantitative reverse transcription–polymerase chain reaction (qRT‐PCR) analysis of mRNA for Mepe (A) and Phex (B) in the articular cartilage of CBA and STR/Ort mice at 8–10 weeks, 18–20 weeks, and ≥40 weeks of age (n = 3 joints per sample; n = 3 samples per age group per strain). C, Immunohistochemical analysis of MEPE in the medial (affected) and lateral (unaffected) tibial condyles of STR/Ort mice prior to and at the onset of osteoarthritis. D, Immunohistochemical analysis of MEPE in the tibial condyles of a CBA mouse. Images are representative of results in 3 individual mice. E and F, GeXP multiplex qRT‐PCR analysis of mRNA for Dmp1 (E) and Opn (F) in the articular cartilage of CBA and STR/Ort mice at 8–10 weeks, 18–20 weeks, and ≥40 weeks of age (n = 3 joints per sample; n = 3 samples per age group per strain). Bars in A, B, E, and F show the mean ± SEM. ∗ = P < 0.05; ∗∗∗ = P < 0.001, versus CBA mice except where indicated otherwise. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/journal/doi/10.1002/art.39508/abstract.

s per sample; n = 3 samples per age group per strain). Bars in A, B, E, and F show the mean ± SEM. ∗ = P < 0.05; ∗∗∗ = P < 0.001, versus CBA mice except where indicated otherwise. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/journal/doi/10.1002/art.39508/abstract. Assessment of the MEPE regulator Phex revealed elevated mRNA levels in articular cartilage from 8–10‐week‐old STR/Ort mice compared to age‐matched CBA mice (P < 0.001) (Figure 3B). In STR/Ort mice, this expression significantly decreased with OA onset (P < 0.001 for STR/Ort mice at 18–20 weeks versus STR/Ort mice at 8–10 weeks) (Figure 3B). Identical age‐related expression patterns were found for Dmp1 mRNA, another SIBLING family member (P < 0.001) (Figure 3E). Analysis of mRNA levels of osteopontin (Opn), a SIBLING family member with shared roles in biomineralization, showed significantly higher levels in aged STR/Ort mice compared to young STR/Ort mice (P < 0.05) and age‐matched CBA mice (P < 0.05), resembling patterns of Mepe expression (Figure 3F). Taken together, these findings suggest a regulatory role for the SIBLING family of proteins in OA development in these mice.

ization, showed significantly higher levels in aged STR/Ort mice compared to young STR/Ort mice (P < 0.05) and age‐matched CBA mice (P < 0.05), resembling patterns of Mepe expression (Figure 3F). Taken together, these findings suggest a regulatory role for the SIBLING family of proteins in OA development in these mice. We next sought to examine the temporal expression of another important regulator of MEPE expression, the Wnt signaling inhibitor sclerostin (Sost) 36. Our analyses showed greater levels of Sost mRNA in articular cartilage from 8–10‐week‐old STR/Ort mice than that from age‐matched CBA mice (P < 0.05) (Figure 4A), with levels significantly decreasing with OA onset (P < 0.01 for STR/Ort mice at 8–10 weeks versus STR/Ort mice at 18–20 weeks) (Figure 4A). Despite this, no differences in circulating serum sclerostin concentrations were observed in these mice at any age (Figure 4B), indicating solely local effects. Consistent with this finding, sclerostin immunolabeling showed a clear enrichment in cells at the osteochondral interface in unaffected regions of STR/Ort mouse joints (Figure 4C). In contrast, STR/Ort mice with OA showed suppression of positive sclerostin labeling of regions of subchondral bone thickening underlying those with compromised articular cartilage integrity (Figure 4D).

clear enrichment in cells at the osteochondral interface in unaffected regions of STR/Ort mouse joints (Figure 4C). In contrast, STR/Ort mice with OA showed suppression of positive sclerostin labeling of regions of subchondral bone thickening underlying those with compromised articular cartilage integrity (Figure 4D). Figure 4 A, GeXP multiplex quantitative polymerase chain reaction analysis of mRNA for Sost in the articular cartilage of CBA and STR/Ort mice at 8–10 weeks, 18–20 weeks, and ≥40 weeks of age (n = 3 joints per sample; n = 3 samples per age group per strain). B, Serum sclerostin levels in CBA and STR/Ort mice at 8–10 weeks, 18–20 weeks, and ≥40 weeks of age (n = 4 mice per age group for each strain). Bars in A and B show the mean ± SEM. C and D, Immunohistochemical analysis of sclerostin in the lateral (unaffected) tibial condyles (C) and medial (affected) tibial condyles (D) in STR/Ort mice at the onset of osteoarthritis. Arrows in C indicate sclerostin immunolabeling. Asterisk in D indicates subchondral bone thickening. Images are representative of results in 3 individual mice. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/journal/doi/10.1002/art.39508/abstract.

of osteoarthritis. Arrows in C indicate sclerostin immunolabeling. Asterisk in D indicates subchondral bone thickening. Images are representative of results in 3 individual mice. Color figure can be viewed in the online issue, which is available at http://onlinelibrary.wiley.com/journal/doi/10.1002/art.39508/abstract. Link between premature growth plate closure in STR/Ort mice and OA development To directly test whether longitudinal growth, growth plate fusion, and OA exhibit interrelationships in STR/Ort mice, we developed a novel protocol for quantifying bony bridges formed across the entire murine tibia epiphysis during growth plate fusion (see Supplementary Methods, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art39508/abstract) (Figures 5A–C). Applying this novel method to examine growth plate closure in STR/Ort mice and CBA mice at 8 weeks of age and ≥40 weeks of age revealed a dramatically (10‐fold) greater total number of bridges in 8‐week‐old STR/Ort mice (mean ± SEM 137 ± 10) than in CBA mice (mean ± SEM 14 ± 10) (P < 0.001) (Figures 5D, E, and H) (see Supplementary Figure 2, available on the Arthritis & Rheumatology web site at http://onlinelibrary.wiley.com/doi/10.1002/art39508/abstract). This enriched growth plate bridging was apparent in all aspects of STR/Ort mouse tibiae (P < 0.05) (Figure 5H). Although still evident in aged STR/Ort mice (≥40 weeks), the enriched bone bridging was much less pronounced (mean ± SEM 295 ± 72 in STR/Ort mice and 266 ± 53 in CBA mice) (Figures 5F, G, and I and Supplementary Figure 2). Mean areal bridge densities were also greater in STR/Ort mice at both ages (P < 0.01) (Figure 5J). These intriguing data reveal an accelerated cartilage–bone transition in the growth plate which, taken together with our findings described above, support the notion of an inherent endochondral defect in both the articular and growth plate cartilage in STR/Ort mice.

in STR/Ort mice at both ages (P < 0.01) (Figure 5J). These intriguing data reveal an accelerated cartilage–bone transition in the growth plate which, taken together with our findings described above, support the notion of an inherent endochondral defect in both the articular and growth plate cartilage in STR/Ort mice. Figure 5 Development of a 3‐dimensional quantification method for growth plate bridging. A, Three‐dimensional representation of an entire joint from an STR/Ort mouse at ≥40 weeks of age. B, Three‐dimensional representation of the growth plate cartilage (yellow) underneath the tibial joint surface (shown in gray in A). C, Three‐dimensional representation of bridges crossing the growth plate underneath the tibial joint. Crosses indicate bony bridges identified by an observer. D–G, Location and areal density of bridges across the growth plate projected on the tibial joint surface in an STR/Ort mouse at 8 weeks of age (D), a CBA mouse at 8 weeks of age (E), an STR/Ort mouse at ≥40 weeks of age (F), and a CBA mouse at ≥40 weeks of age (G). H and I, Number of bridges per tibia in CBA and STR/Ort mice at 8 weeks of age (H) and ≥40 weeks of age (I). The lateral and medial segments and anterior and posterior segments were split in order to examine whether bridging is balanced during fusion. J, Areal density (d) of bridges, defined as the number of bridges per 256 μm × 256 μm window. Bars in H–J show the mean ± SEM (n = 3 mice per group). ∗ = P < 0.05; ∗∗ = P < 0.01; ∗∗∗ = P < 0.001, versus CBA mice except where indicated otherwise.

were split in order to examine whether bridging is balanced during fusion. J, Areal density (d) of bridges, defined as the number of bridges per 256 μm × 256 μm window. Bars in H–J show the mean ± SEM (n = 3 mice per group). ∗ = P < 0.05; ∗∗ = P < 0.01; ∗∗∗ = P < 0.001, versus CBA mice except where indicated otherwise. DISCUSSION Our data reveal changes in the articular cartilage of STR/Ort mouse knee joints consistent with an aberrant deployment of endochondral processes. This is associated with inherent longitudinal growth modifications, disrupted growth plate morphology, premature growth plate fusion, and aberrant bone formation and matrix mineralization prior to OA onset. These data indicate that, at least in the spontaneous human‐like OA seen in STR/Ort mice, growth‐related endochondral ossification abnormalities may forecast mechanisms of OA development in articular cartilage.

logy, premature growth plate fusion, and aberrant bone formation and matrix mineralization prior to OA onset. These data indicate that, at least in the spontaneous human‐like OA seen in STR/Ort mice, growth‐related endochondral ossification abnormalities may forecast mechanisms of OA development in articular cartilage. There are certainly some intriguing previously published data on the expression of endochondral ossification markers that support this notion 14. Type X collagen is a marker of chondrocyte hypertrophy that is usually found in the growth plate and is unique to the calcified cartilage in normal joints 36. Expression of type X collagen mRNA transcripts, as examined by in situ hybridization, has, however, been observed throughout articular cartilage in both young STR/Ort mice (at 9 weeks of age) and older STR/Ort mice (at ≥41 weeks of age) 35. This is the first study to provide evidence of associated type X collagen protein expression in these mice. Consistent with our findings, an additional marker of chondrocyte hypertrophy, MMP‐13, has been detected in the calcified cartilage chondrocytes of STR/Ort mice at both young and old ages, at levels greater than those in age‐matched CBA mice 37. Similarly, higher expression levels of several other MMPs (MMP‐2, MMP‐3, MMP‐7, MMP‐9, and membrane type 1 MMP) were observed in the tibial articular chondrocytes of the STR/Ort mouse 37. Indeed, many of these MMPs were also shown to be significantly increased in our previous microarray study, further highlighting their likely role as key players in cartilage degradation in OA 22.

2, MMP‐3, MMP‐7, MMP‐9, and membrane type 1 MMP) were observed in the tibial articular chondrocytes of the STR/Ort mouse 37. Indeed, many of these MMPs were also shown to be significantly increased in our previous microarray study, further highlighting their likely role as key players in cartilage degradation in OA 22. The STR/Ort mouse growth plate has remained relatively unexamined with, to our knowledge, only one published report describing phenotypic changes associated with aging. Chambers et al 35 describe type X collagen mRNA expression localized to hypertrophic chondrocytes, as expected, in both young CBA mice and young STR/Ort mice. However, in the older mice, no expression of type X collagen mRNA was observed, despite the preservation of type II collagen mRNA throughout the depth of the thinned growth plate cartilage 35. The results of the present study indicate aberrant expression of type X collagen and MMP‐13 additionally in the growth plate of young STR/Ort mice. STR/Ort mice also display an increased zone of proliferative chondrocytes, based on well‐established cell morphologic features 27. These results may seem counterintuitive, but they highlight the fact that there is clearly an inherent endochondral defect in STR/Ort mice, which may also precipitate OA pathogenesis.

mice. STR/Ort mice also display an increased zone of proliferative chondrocytes, based on well‐established cell morphologic features 27. These results may seem counterintuitive, but they highlight the fact that there is clearly an inherent endochondral defect in STR/Ort mice, which may also precipitate OA pathogenesis. Molecular mechanisms controlling endochondral ossification may help identify those involved in OA. Effective control of the Wnt signaling pathway is certainly proving critical in regulating both the extent of OA joint pathology 38 and growth plate chondrocyte behavior, and the data in the present study corroborate this.

Molecular mechanisms controlling endochondral ossification may help identify those involved in OA. Effective control of the Wnt signaling pathway is certainly proving critical in regulating both the extent of OA joint pathology 38 and growth plate chondrocyte behavior, and the data in the present study corroborate this. Genetic and microarray analyses have been performed in STR/Ort mice in order to better elucidate the etiology of their OA 39, 40, 41, 42. Jaeger and colleagues identified a quantitative trait locus (QTL) associated with articular cartilage degeneration on chromosome 8 of the STR/Ort mouse 39. This, however, was not corroborated in a more recent QTL analysis in which STR/Ort mice were backcrossed with the C57BL/6N strain 43. This QTL was therefore suggested to be a recessive trait among the polygenetic factors in OA in STR/Ort mice 19, 43. Instead, the authors identified a QTL for the OA phenotype that is mapped to chromosome 4 43. Chromosome 8 was, however, revisited, and fine‐mapping of the OA QTL in a more recent study revealed Wnt‐related genes associated with altered chondrogenesis, including dickkopf 4 (Dkk4), secreted Frizzled‐related protein 1 (Sfrp1), and fibroblast growth factor 1 (Fgfr1) 38, 42. While a number of genes, including Wnt‐related genes, have been implicated in OA by association studies in human populations, there is a distinct lack of functional data to support a causative link between these associated genes and OA.