Collagen antibody–induced arthritis evokes persistent pain with spinal glial involvement and transient prostaglandin dependency




Pain is one of the most debilitating symptoms reported by rheumatoid arthritis (RA) patients. While the collagen antibody–induced arthritis (CAIA) model is used for studying the effector phase of RA pathologic progression, it has not been evaluated as a model for studies of pain. Thus, this study was undertaken to examine pain-like behavior induced by anticollagen antibodies and to assess the effect of currently prescribed analgesics for RA. In addition, the involvement of spinal glia in antibody-induced pain was explored.


CAIA was induced in mice by intravenous injection of a collagen antibody cocktail, followed by intraperitoneal injection of lipopolysaccharide. Disease severity was assessed by visual and histologic examination. Pain-like behavior and the antinociceptive effect of diclofenac, buprenorphine, gabapentin, pentoxifylline, and JNK-interacting protein 1 were examined in mechanical stimulation experiments. Spinal astrocyte and microglia reactivity were investigated by real-time polymerase chain reaction and immunohistochemistry.


Following the induction of CAIA, mice developed transient joint inflammation. In contrast, pain-like behavior was observed prior to, and outlasted, the visual signs of arthritis. Whereas gabapentin and buprenorphine attenuated mechanical hypersensitivity during both the inflammatory and postinflammatory phases of arthritis, diclofenac was antinociceptive only during the inflammatory phase. Spinal astrocytes and microglia displayed time-dependent signs of activation, and inhibition of glial activity reversed CAIA-induced mechanical hypersensitivity.


CAIA represents a multifaceted model for studies exploring the mechanisms of pain induced by inflammation in the articular joint. Our findings of a time-dependent prostaglandin and spinal glial contribution to antibody-induced pain highlight the importance of using appropriate disease models to assess joint-related pain.

Rheumatoid arthritis (RA) is a chronic autoimmune disease that is characterized by joint inflammation and bone and cartilage destruction. Pain is the most commonly reported and debilitating problem. Indeed, RA patients often rank pain relief as one of their highest priorities for health improvement (1). While it may not be surprising that joint pain is a significant issue during active disease, recent studies show that pain continues to be a problem for a considerable number of RA patients with minimal disease activity (2, 3) and even for patients with sustained remission (4). This indicates that it is not only the inflammation but also additional factors that contribute to the persistence of pain in RA. Hence, it is critical that we increase our understanding of the mechanisms driving RA-mediated pain, in order to identify new treatment strategies.

The use of disease-relevant models has been instrumental in the identification of factors that drive disease progression and in the development of novel therapeutic agents for the treatment of RA. Explorations of the pain mechanisms in these models have the potential to provide new information that would help not only to elucidate how chronic joint inflammation affects the sensory nervous system, but also to identify new targets for pain relief. Collagen-induced arthritis (CIA), mediated by autoantibodies against type II collagen (CII), is the most widely used model for studying the pathologic processes of RA. Serum transfer from mice with CIA to naive mice (5) or injection of monoclonal antibodies against CII into naive mice will induce arthritis (6–9). Collagen antibody–induced arthritis (CAIA) thus resembles CIA and RA in several aspects, but has a more rapid and synchronized onset. In addition, the amount of antibodies needed to induce CAIA is known, providing the ability to titrate the degree of disease. This is advantageous in pain studies, because a pronounced sickness behavior that may mask the signs of pain-like behavior can be avoided by inducing a mild-to-moderate degree of arthritis. Although the CAIA model has been utilized frequently to investigate RA pathologic processes, it has not been used as a model of RA-induced pain. Thus, the first aim of this study was to explore the capacity of anticollagen antibodies to induce pain-like behavior in naive mice.

In the joint, all structures except for cartilage are innervated by pain-sensing (nociceptive) neurons. In the presence of physical or biochemical changes, such as edema or release of inflammatory mediators, e.g., prostaglandins or cytokines, the threshold for activation of peripheral nociceptive neurons is lowered, resulting in pain with movement or on palpation and generation of spontaneous joint pain (pain at rest) (10, 11). Prolonged excitation of the peripheral primary nociceptive neurons leads to sensitization of spinal cord dorsal horn neurons, enabling the neurons to respond more easily than would occur in a noninflamed state, which further amplifies the transmission of pain signals to the brain (12). Rapidly growing evidence suggests that non-neuronal cells in the spinal cord, i.e., astrocytes and microglia, are activated by persistent pain transmission and participate in the modulation of pain signaling (13). Such a notion is supported by studies in which agents that inhibit or disrupt glia function have been found to attenuate neuropathic and inflammatory pain (14–16). Thus, the second aim of this study was to examine whether spinal microglia and astrocytes are activated in response to CAIA, and whether these cells are functionally coupled to CAIA-induced pain behavior. In addition, we assessed the pharmacologic profile of CAIA-induced hypersensitivity in order to characterize the pain behavior in antibody-induced arthritis.



All of the experiments were carried out in accordance with protocols approved by the local ethics committee for animal experiments in Sweden (Stockholms Norra Djurförsöksetiska Nämnd). Male (BALB/c × B10.Q)F1 (QB) mice (bred at the Department of Medical Biochemistry and Biophysics, Karolinska Institutet) and CBA and BALB/c mice (bred at B&K Universal) (all ages 10–12 weeks at the start of the experiments) were housed in standard cages (4–5 mice/cage) in a climate-controlled environment. The mice were maintained on a 12-hour light/dark cycle and provided with food and water ad libitum.

Induction of arthritis.

In QB mice, arthritis was induced by injecting the animals with 4 mg of an anti-CII arthritogenic cocktail (17) containing 4 monoclonal antibodies, which was administered intravenously (0.15–0.25 ml) on day 0. In CBA and BALB/c mice, 1.5 mg of an anti-CII arthritogenic cocktail (0.15 ml intravenously; Chondrex) containing 5 monoclonal antibodies was injected, since this cocktail induces CAIA in a broader range of mouse strains. Lipopolysaccharide (LPS) enhances the incidence and severity of the antibody-initiated disease (8). Therefore, 25 μg of LPS (serotype O55:B5; Sigma) in 100 μl physiologic saline was injected intraperitoneally 5 days after the antibody cocktail injection. Two control groups were included in the study. One group received saline intravenously on day 0 and saline intraperitoneally on day 5 (saline controls), and the other received saline intravenously on day 0 and 25 μg LPS intraperitoneally on day 5 (LPS controls).

Scoring of arthritis.

The development of joint inflammation in the fore and hind paws was monitored by visual inspection, as described previously (8). Briefly, 1 point was given for each inflamed toe or knuckle, and if the ankle/ wrist was inflamed, 5 points were given, resulting in a maximum possible arthritis score of 15 points per paw and 60 points per mouse.

Histopathologic evaluation.

Hind ankle joints from the saline control and CAIA groups were fixed in 4% paraformaldehyde solution for 48 hours, and then decalcified for 3–4 weeks in a solution containing EDTA, polyvinylpyrrolidone, and Tris HCl (Sigma), followed by dehydration and embedding in paraffin. Sagittal sections (5 μm) were stained with hematoxylin and eosin (Histolab). For each animal, 3 sections were analyzed for synovial inflammation (synovitis), bone erosion, and cartilage destruction, with a 3-grade scoring system for each aspect. For synovial inflammation, 0 = no inflammation, 1 = slight thickening of the synovial cell layer and/or some inflammatory cells in the sublining, 2 = moderate infiltration of the sublining, and 3 = marked-to-severe infiltration. For bone erosion, 0 = normal bone, 1 = small areas of resorption, 2 = more numerous areas of resorption, and 3 = full-thickness resorption areas in the bone. For cartilage destruction, 0 = normal cartilage, 1 = cartilage surface irregularities, 2 = minor-to-moderate loss of surface cartilage, and 3 = marked cartilage destruction and loss of surface cartilage. Joint sections were scored by 2 investigators (DBB and AB) who were blinded with regard to the group assignments.

Assessment of mechanical hypersensitivity and cold allodynia.

Animals were habituated to the testing environment prior to baseline testing. On test days, mice were placed in individual compartments on top of a wire-mesh surface and allowed to acclimatize prior to testing. Withdrawal thresholds were assessed with calibrated von Frey optiHair filaments (Marstock OptiHair) with a logarithmically incremental stiffness of 0.5, 1, 2, 4, 8, 16, and 32 mN (converted to grams [18]). A cutoff of 4 gm was applied in order to avoid tissue damage, and thus nociceptive responses in the suprathreshold range were not assessed. Each filament was pressed perpendicularly against the mid–hind paw and held for ∼3 seconds. A positive response was noted if the paw was withdrawn. The tactile threshold, i.e., the threshold of mechanical hypersensitivity at which there was a 50% probability of paw withdrawal, was calculated as previously described (18), with results expressed in grams. Withdrawal thresholds for both hind paws were determined and averaged.

Sensitivity to cold was assessed with the acetone drop test (19). Briefly, one drop of acetone was applied to the plantar surface of the hind paw, and the duration of the nocifensive response (biting, licking, or shaking of paw) was measured over 60 seconds. The test was repeated 3 times and the observations were averaged.

Drugs and injections.

Groups of arthritic mice received a single intraperitoneal injection of the cyclooxygenase (COX) inhibitor diclofenac (30 mg/kg intraperitoneally; Novartis), the partial opioid and nociception receptor agonist buprenorphine (0.1 mg/kg intraperitoneally; Schering-Plough), or the voltage-gated calcium channel blocker gabapentin (100 mg/kg intraperitoneally; Sigma) during the early phase (days 8–12) and late phase (days 26–30) of arthritis. A separate group of arthritic mice received intrathecal injections of pentoxifylline (10–30 μg intrathecally; Sigma), a glia inhibitor, or JNK-interacting protein 1 (JIP-1) (1–5 μg intrathecally; Tocris Bioscience), a JNK inhibitor, during the late phase (days 27–34) of arthritis. A minimum washout period of 2 days was allowed for the different drugs, and a maximum of 2 different drugs was tested in the same phase and same mouse. The investigator administering the injections (DBB) was blinded with regard to the drug treatments.

For intrathecal injections, the mice were anesthetized with isoflurane and the drugs were administered in 5 μl using a 25-μl Hamilton syringe, which was inserted into the intervertebral space between L5 and L6 (20). Changes in mechanical thresholds were measured 2 and 4 hours after drug injections. All drugs were dissolved in physiologic saline, which was used as a vehicle control. The doses of diclofenac, buprenorphine, and gabapentin used were based on those previously reported (21–23). Two different doses were used for pentoxifylline and JIP-1.


Mice were anesthetized with isoflurane, and then perfused with saline followed by 4% paraformaldehyde. After the mice were killed, lumbar spinal cords (L1–L6) were dissected, postfixed, and cryoprotected in 20% sucrose. Spinal cord sections were cryocut (30 μm) and incubated in 5% normal goat serum in 0.2% Triton X-100/phosphate buffered saline to block nonspecific binding. The sections were then incubated overnight with primary anti–ionized calcium–binding adapter molecule 1 (anti–IBA-1) antibodies (1:1,000; Wako) and anti–glial fibrillary acidic protein (anti-GFAP) antibodies (1:1,000; Invitrogen), in order to label microglia and astrocytes, respectively. The immunoreactivity was visualized using secondary antibodies conjugated to Alexa 488 or Alexa 594 (each 1:250; Invitrogen).

Images of 5 lumbar spinal cord sections from L1–L2, L3–L4, and L5–L6 were captured using a fluorescence microscope (Zeiss Axio Scope). The dorsal horn was manually outlined (lamina I–VI) and the integrated signal intensity was measured subsequent to background subtraction using ImageJ (NIH). The integrated signal intensities of the 3 × 5 sections and the 2 dorsal horns were averaged. An increase in signal intensity was interpreted as a sign of glia activation, as previously described (24). The signal intensity quantifications were performed by 2 investigators (JS and JL) who were blinded with regard to the group assignments.

Quantitative real-time polymerase chain reaction (PCR).

Flash-frozen lumbar spinal cords (L1–L6) were homogenized in TRIzol (Invitrogen) and RNA was extracted according to the manufacturer's protocol. The RNA was reverse transcribed to complementary DNA, and quantitative real-time PCR (GeneAmp 7000 Sequence Detection System; Applied Biosystems) was performed with hydrolysis probes, in accordance with the manufacturer's instructions, to determine the relative messenger RNA (mRNA) levels. Predeveloped specific primer/probe sets for GFAP (Mm00546086_m1), Cd11b (Mm00434455_m1), and hypoxanthine guanine phosphoribosyltransferase 1 (HPRT1) (Mm01545399_m1) (all from Applied Biosystems) were used to detect astrocyte, microglia, and reference genes, respectively. Threshold cycle values in each sample were used to calculate the number of cell equivalents in the test samples using the standard curve method (25). Data were normalized to the values for HPRT1, and expressed as a percentage of control values.

Western blotting.

Flash-frozen lumbar spinal cords (L1–L6) were homogenized in lysis buffer, and proteins were separated by gel electrophoresis (Invitrogen) and transferred to nitrocellulose membranes (Invitrogen) as described previously (26). After blocking of nonspecific binding sites, the membranes were incubated overnight with primary antibodies (for phosphorylated JNK [p-JNK] at 1:1,000 and β-actin at 1:10,000; Cell Signaling Technology), and then with secondary antibodies conjugated to horseradish peroxidase (1:7,500; Cell Signaling Technology). Immunopositive bands were detected using chemiluminescent reagents (SuperSignal; Pierce) and radiographic film. Signal intensity was measured using Bio-Rad Quantity One software. Immunopositive bands were normalized to the bands for β-actin, and results are expressed as the ratio of p-JNK signal intensity to β-actin signal intensity.

Statistical analysis.

For comparison of the changes in pain behavior and the effect of drug treatments, two-way analysis of variance (ANOVA) was used, followed by Bonferroni post hoc test. For assessment of the differences in results from immunohistochemistry, real-time PCR, and Western blotting, one-way ANOVA was used, followed by Bonferroni post hoc test. Arthritis scores were compared using the Kruskal-Wallis test, followed by Dunn's multiple comparison post hoc test, performed using GraphPad software. P values less than 0.05 were considered significant.


Development of inflammation and mechanical hypersensitivity in mice after induction of CAIA.

Arthritis scores increased significantly after injection of the collagen antibody cocktail and LPS, compared to the saline and LPS control groups, as has been reported previously (27), with the disease severity reaching maximum levels around days 8–10 in QB mice and days 12–20 in CBA and BALB/c mice (each P < 0.01 versus controls) (Figures 1A–C, top panels). The signs of arthritis were transient, and were no longer detectable by day 20 in QB mice (Figure 1A, top), day 36 in CBA mice (Figure 1B, top), and day 45 in BALB/c mice (Figure 1C, top). A strain-dependent distribution of joint inflammation was observed. No significant difference in arthritis scores between the fore and hind paws was observed in QB and BALB/c mice (each P > 0.05) (Figures 1A and C, top). In contrast, CBA mice displayed more severe signs of arthritis in the fore paws compared to the hind paws (P < 0.05) (Figure 1B, top).

Figure 1.

Disease severity and pain-like behavior in a collagen antibody–induced arthritis (CAIA) model. A–C, Top, Animals were injected with collagen antibody cocktail and lipopolysaccharide (LPS) (CAIA group), saline and LPS (LPS group), or saline twice (saline control) and evaluated by visual inspection for disease severity, assessed as the arthritis score (scale of 0–60, range 0–15/paw) among QB (A), CBA (B), and BALB/c (C) mice, presented as the score in the fore paws (CAIA [F]), hind paws (CAIA [H]), and in total (CAIA [F + H]) in mice with CAIA compared to saline controls. Bottom, Mechanical hypersensitivity of the hind paws was assessed in QB (A), CBA (B), and BALB/c (C) mice by application of von Frey filaments prior to injection and throughout the experimental period, in the CAIA (H), saline control, and LPS groups. D, QB mice in the CAIA (H), saline control, and LPS groups were also evaluated for cold allodynia, presented as duration of sensitivity to cold (in seconds) at various time points after injection. Values are the mean ± SEM of 5–10 mice/group. ∗ = P < 0.05; ∗∗ = P < 0.01; ∗∗∗ = P < 0.001, versus saline controls. # = P < 0.05 versus CAIA (H) group.

The CAIA groups developed significant mechanical hypersensitivity, whereas the LPS and saline control groups did not. Reductions in tactile paw withdrawal thresholds in mice with CAIA reached a maximum around days 8–18 (depending on the mouse strain) after injection (P < 0.01 versus saline controls) (Figures 1A– C, bottom panels), and these thresholds, surprisingly, remained at lower levels throughout the experiment, even though the arthritis scores declined and returned to baseline values after 2–4 weeks. We refer to this persistent pain behavior as postinflammatory, or late-phase, hypersensitivity. Of note, in the QB and CBA mice, CAIA-induced mechanical hypersensitivity was observed within 2–3 days after injection, prior to the appearance of signs of inflammation (Figures 1A and B, bottom). BALB/c mice were subjected to less frequent testing, because we have noted that these mice have a tendency to become sensitized if tested repeatedly.

Since studies have indicated that RA patients have an increased sensitivity to cold (28), we examined whether mice with CAIA displayed altered nocifensive behavior (biting, licking, or shaking of the paw) after acetone application. A modest transient increase in cold sensitivity was observed on days 8 and 12 in the CAIA group compared to the saline and LPS controls (P < 0.01) (Figure 1D).

Histopathologic changes in the ankle joints.

Histopathologic changes, including synovitis, bone erosion, and cartilage destruction, occurring in the ankle joints after the induction of CAIA were scored. Joints from saline-injected control mice showed no signs of histopathologic changes (Figures 2A–C). QB mice with CAIA displayed marked synovitis, bone erosion, and cartilage destruction on days 12 and 13 (inflammatory phase), and BALB/c mice with CAIA exhibited marked synovitis and bone erosion during this phase. By days 31 and 71 (late phase), a low degree of bone erosion was observed, but cartilage destruction in QB mice and synovitis and cartilage destruction in BALB/c mice were no longer detectable in the late phase (Figures 2A and C). CBA mice showed a modest increase in infiltration of inflammatory cells and a low degree of bone erosion on day 15 of CAIA (inflammatory phase), and a low degree of bone erosion was still detectable on day 47 (late phase) (Figure 2B). In subsequent studies, we focused on QB mice.

Figure 2.

Histopathologic changes in the joint after induction of collagen antibody–induced arthritis (CAIA). Left, Representative hematoxylin and eosin–stained images of the hind ankle joints of QB (A), CBA (B), and BALB/c (C) mice show synovial inflammation (synovitis) (white arrows), bone erosion (black arrows), and cartilage destruction (arrowheads) during the inflammatory or late phase of CAIA in comparison with saline-injected controls. Original magnification × 10. Right, The severity scores of synovitis, bone erosion, and cartilage destruction were assessed in the ankle joints of QB (A), CBA (B), and BALB/c (C) mice in the inflammatory phase (days 12, 13, or 15) and late phase (days 31, 47, or 71) of CAIA. Bars show the mean ± SEM of 3–5 mice/group.

Effects of analgesic compounds on CAIA-induced mechanical hypersensitivity.

To further characterize the pain behavior in the CAIA model, we assessed the antinociceptive effects of diclofenac, buprenorphine, and gabapentin during the inflammatory (days 8–12) and late (days 26–30) phases of arthritis. During the early phase of ongoing inflammation, injection of the lumbar spinal cords with diclofenac, buprenorphine, or gabapentin significantly increased the tactile thresholds 2 and/or 4 hours after drug injection (P < 0.01 versus vehicle control) (Figure 3A). In the late phase, mechanical hypersensitivity was reversed at 2 hours and 4 hours after the injection of buprenorphine (P < 0.01 versus vehicle control) and at 2 hours after the injection of gabapentin (P < 0.01 versus vehicle control) (Figure 3B). In contrast, no significant analgesic effect was observed 2 or 4 hours following the injection of diclofenac during the late phase of arthritis (P > 0.05 versus vehicle control) (Figure 3B). No change in the degree of arthritis was observed after injection of the drug treatments (results not shown).

Figure 3.

Effects of diclofenac (top), buprenorphine (middle), and gabapentin (bottom) on collagen antibody–induced pain-like behavior. Each drug, or saline as vehicle control, was injected intraperitoneally during the inflammatory phase (days 8–12) (A) or late phase (days 26–30) (B) of collagen antibody–induced arthritis, and mechanical hypersensitivity (measured in grams and expressed as the threshold of 50% probability of paw withdrawal) was assessed at 2 and 4 hours after injection. The horizontal dotted line indicates the mean tactile threshold at baseline (3 gm). Values are the mean ± SEM of 5–8 mice/group. ∗∗ = P < 0.01; ∗∗∗ = P < 0.001, versus vehicle controls.

Spinal astrocyte and microglia activation.

To investigate whether glia display signs of increased activity in the spinal cord subsequent to the induction of CAIA, the expression levels of gene and protein markers for reactive astrocytes (GFAP) and microglia (Cd11b and IBA-1) were measured in the lumbar dorsal horn of QB mice. Whereas no change in GFAP mRNA levels was detected in the CAIA group on day 13 after arthritis induction (P > 0.05 versus saline controls), there was a significant increase in GFAP mRNA levels in the CAIA group on days 9 and 29 (P < 0.001 versus saline controls) (Figure 4A). In addition, spinal Cd11b mRNA levels after the induction of CAIA showed a pattern of change similar to that of the GFAP mRNA levels (Figure 4A).

Figure 4.

Astrocyte and microglia reactivity in the spinal cord following the induction of collagen antibody–induced arthritis (CAIA) in QB mice. A, The mRNA levels of glial fibrillary acidic protein (GFAP) (astrocyte marker) and Cd11b (microglia marker) were measured in lumbar spinal cords of mice with CAIA on days 9, 13, and 29 after arthritis induction. Values are the mean ± SEM of 4 mice/group, expressed as a percentage of the mRNA levels in controls (age-matched mice injected with saline). B and C, Representative images show the immunoreactivity of GFAP (astrocyte marker) (B) and ionized calcium–binding adapter molecule 1 (IBA-1) (microglia marker) (C) in lumbar dorsal horns of mice with CAIA, compared to age-matched, saline-injected controls, on days 13 and 29 after arthritis induction. Bars = 50 μm. D, The GFAP and IBA-1 signal intensity was quantified using densitometry (measured as the integrated intensity) in the lumbar dorsal horns of mice with CAIA, compared with saline-injected control mice, on days 13 and 29 after arthritis induction. Values are the mean ± SEM of 3 mice/group (15 sections per mouse). ∗ = P < 0.05; ∗∗∗ = P < 0.001, versus saline controls.

Levels of GFAP and IBA-1 protein expression were assessed by immunohistochemistry. Immunostaining revealed increased astrocyte reactivity (Figure 4B), but not microglia reactivity, on day 29 after the induction of CAIA in the lumbar dorsal horn sections from QB mice (Figure 4C). Quantification of the signal intensity of the respective markers confirmed this observation, showing a significant increase in GFAP signal intensity on day 29 (P < 0.05), but not on day 13 (P > 0.05), in the CAIA group compared to saline controls (Figure 4D). No difference in the IBA-1 signal intensity was observed in the lumbar dorsal horn sections from QB mice with CAIA compared to saline controls, on either day 13 or day 29 (each P > 0.05) (Figure 4D).

While both the GFAP mRNA and GFAP protein markers of astrocyte reactivity showed elevated levels in the late phase of CAIA, there was a discrepancy in these parameters with regard to microglia reactivity. In order to explore whether microglia show more robust signs of increased activity in mice that have a pronounced, longer-lasting joint inflammation, CBA mice were assessed for GFAP and IBA-1 immunoreactivity. In addition, the effect of LPS injection on glia activation was examined. Changes in spinal GFAP immunoreactivity in CBA mice with CAIA paralleled the observations made in QB mice (Figures 5A and C). However, CBA mice with CAIA displayed increased spinal IBA-1 immunoreactivity, in both the inflammatory and late phases of arthritis (Figure 5B), which was confirmed by quantification of the IBA-1 signal intensity in the dorsal horn (P < 0.05 versus saline controls) (Figure 5C). Intraperitoneal injection of LPS did not increase the spinal GFAP or spinal IBA-1 immunoreactivity (P > 0.05 versus saline controls) (Figures 5A–C).

Figure 5.

Astrocyte and microglia reactivity in the spinal cord following the induction of collagen antibody–induced arthritis (CAIA) in CBA mice. A and B, Representative immunohistochemical images show results of staining for glial fibrillary acidic protein (GFAP) (astrocyte marker) (A) and ionized calcium–binding adapter molecule 1 (IBA-1) (microglia marker) (B) in lumbar dorsal horns of mice with CAIA, compared to age-matched mice injected with lipopolysaccharide (LPS) or saline (control), on days 15 and 47 after arthritis induction. Bars = 100 μm. C, The GFAP and IBA-1 signal intensity was quantified using densitometry (measured as the integrated intensity) in the lumbar dorsal horns of mice with CAIA and LPS-injected mice, with results expressed as a percentage of the values in the saline-injected controls (set at 100). Values are the mean ± SEM of 5 mice/group (15 sections per mouse). ∗ = P < 0.05; ∗∗∗ = P < 0.001, versus saline controls.

Assessment of the role of glia and JNK in CAIA-induced late-phase pain behavior.

We next examined whether inhibition of glia activity could attenuate the CAIA-induced pain behavior in QB mice. Intrathecal injection of pentoxifylline, a glia inhibitor (in astrocytes and microglia), at 30 μg, but not 10 μg, resulted in a reversal of the mechanical hypersensitivity of mice with CAIA at 2 and 4 hours after drug administration, during the late phase of arthritis (days 27–34) (P < 0.01 versus vehicle control) (Figure 6A).

Figure 6.

Effects of pentoxifylline (PTX) and JNK-interacting protein 1 (JIP-1) on collagen antibody–induced mechanical hypersensitivity. A and B, Mice with collagen antibody–induced arthritis (CAIA) were injected intrathecally with either PTX (A) or JIP-1 (B), in different doses dissolved in saline, or with saline alone as vehicle control, in the late phase of inflammation (days 27–34), and tactile thresholds of mechanical hypersensitivity, measured (in grams) as the threshold of 50% probability of paw withdrawal, were assessed at 2 and 4 hours after injection. The horizontal dotted line indicates the mean tactile threshold at baseline (3 gm). Values are the mean ± SEM of 5–8 mice/group. C and D, Phosphorylation of JNK-1 (p-JNK-1) and JNK-2 was quantified (C) and assessed by Western blotting (D) in lumbar spinal cords of mice with CAIA, compared to age-matched, saline-injected controls, on days 13 and 29 after arthritis induction. In C, values are the mean ± SEM of 4 mice/group, presented as the ratio of p-JNK signal intensity to β-actin signal intensity. In D, lanes between day 13 and day 29 have been digitally removed from the membrane picture, and JNK-immunopositive bands are normalized to those for β-actin. ∗ = P < 0.05; ∗∗ = P < 0.01; ∗∗∗ = P < 0.001, versus saline controls.

Previous studies have indicated that JNK is activated in spinal cord astrocytes subsequent to the onset of peripheral inflammation, and the inhibition of spinal JNK attenuates inflammatory pain (29). Therefore, we sought to test whether spinal JNK plays a role in CAIA-induced hypersensitivity. Intrathecal injection of the JNK inhibitor JIP-1 at 5 μg, but not 1 μg, reversed the mechanical hypersensitivity of mice with CAIA at 4 hours after administration (P < 0.05 versus vehicle control) (Figure 6B).

In order to determine which JNK cellular type is activated in the spinal cord of the mice, we used 3 different p-JNK antibodies (from Santa Cruz Biotechnology, Cell Signaling Technology, and Promega). Although no positive signal was detected by immunohistochemistry, we were able to observe increased levels of JNK phosphorylation in the spinal cord using Western blotting. A pronounced increase in phosphorylated JNK-1, but not JNK-2, was detected on day 29 in the CAIA group compared to saline controls (Figures 6C and D).


Interestingly, several of the frequently used models of RA have not been adapted for studies of pain. Therefore, studies of arthritis-related pain in a K/BxN serum transfer arthritis model, in which naive mice were injected intraperitoneally with polyclonal serum containing anti–glucose-6-phospate isomerase (anti-G6PI) antibodies, were recently performed (23). The present study is the first to utilize the CAIA model, involving monoclonal antibody binding to well-defined CII epitopes that are shared between humans and rodents, as a tool for evaluating pain behavior and delineating changes in the nociceptive system during arthritic disease progression. We found that, in contrast to mice injected with LPS alone, mice injected with the antibody cocktail followed by LPS displayed robust mechanical hypersensitivity and mild cold allodynia, concomitant with the onset of joint inflammation. Surprisingly, whereas cold allodynia followed the course of inflammation, the reduced tactile thresholds did not return to baseline levels even at 2 weeks after the visible and histologic signs of arthritis had disappeared.

The hypersensitivity observed in the mice during ongoing joint inflammation was found to be prostaglandin dependent, while that in the postinflammatory phase was not, indicating that pain signaling in different stages of arthritis in this model is driven by different mechanisms. Furthermore, assessment of spinal glial reactivity also revealed a phase-dependent pattern of activation, and our study provides evidence that astrocytes have an important role in the late phase of CAIA-induced hypersensitivity.

In an earlier study, we observed that anti-CII antibodies induced significant changes in the cartilage, independent of inflammatory mediators or cells, both in vitro (30) and in vivo (31). Of note, in the present study, we found that collagen antibody–induced pain-like behavior was observed prior to LPS injection and preceded the onset of inflammation, in the 2 mouse strains in which hypersensitivity was assessed prior to the development of visibly detectable inflammation. These findings are of clinical relevance, because arthralgia (joint pain without inflammation) and the presence of serum autoantibodies can predate the development of RA by years (32). Ongoing work in our laboratory is aimed at dissecting the mechanisms that initiate the early hypersensitivity in the CAIA model.

All 3 mouse strains displayed differing degrees of fore paw and hind paw inflammation, as measured by scoring of histologic features. The arthritis scores in the fore and hind paws were equal between QB and BALB/c mice, while a more prominent arthritic involvement of the fore paws was observed in CBA mice. Nevertheless, the level of pain hypersensitivity was equally pronounced in the hind paws in all 3 mouse strains.

In this study, LPS was injected intraperitoneally 5 days after the injection of the collagen antibody cocktail, since LPS synchronizes the onset of disease and increases the incidence and severity of CAIA (8). Although previous studies have shown that systemic and intrathecal injection of LPS induces pain-like behavior (14, 33, 34) and enhances glia activity in the central nervous system (35), the dose of LPS used in the present study was not sufficient to alter either nociceptive thresholds or spinal glial reactivity.

Models in which animals are immunized to produce antibodies against, for example, CII (36) or bovine serum albumin (antigen-induced arthritis) (37) generate long-lasting pain-like behavior. Since there is a continuous replacement of antibodies by autologous B cells in these models, the inflammatory process is chronic, and therefore it is not surprising that the nociceptive thresholds also remain low (36). Both the CAIA and K/BxN models are based on systemic injection of preformed antibodies that gravitate to the joint (17, 23, 33). Because the antibodies are not replaced, the inflammation resolves over time, and accordingly, we did not detect inflammation by visual or histologic measures in the late phase.

The CAIA model shares many of the characteristics of the K/BxN model. It is noteworthy that even though anti-G6PI and anti-CII antibodies have a different antigen specificity, and thus bind different proteins in the joint, pain behavior in both models persists for more than 6 weeks, even though signs of arthritis are only present for 2–4 weeks. This is in contrast to the observations during soft tissue inflammation (i.e., induced by subcutaneous injection of irritant), for which the most common outcome is that the resolution of pain is concurrent with the resolution of the inflammation (38). Our study findings suggest that an episode of joint inflammation has long-lasting impact on the nociceptive sensory nervous system, and that the presence of antibodies and/or the localization of inflammation to the joint are critical factors in the development of persistent joint pain.

Importantly, our findings can be translated to the clinical situation. The prognosis of RA has been dramatically improved since the introduction of biologic drugs, a type of therapy that halts the progression of RA by blocking the effect of cytokines, T cells, or B cells. However, despite the attainment of improvements in disease control and disease remission, pain is still a major issue in RA patients (2). Hence, the CAIA model constitutes a promising animal model in which new aspects of RA-induced chronic pain can be explored.

Daily administration of a COX inhibitor reduces the arthritis severity in the CAIA model (39). In contrast, one intraperitoneal bolus injection of diclofenac, a COX inhibitor commonly used for pain relief in RA, is sufficient to attenuate CAIA-induced pain behavior during the inflammatory phase, but not at the postinflammatory phase, indicating that prostaglandins play an important nociceptive role during ongoing inflammation. Buprenorphine, used for the treatment of moderate-to-severe pain, and gabapentin, approved for the treatment of neuropathic pain, had analgesic effects during both phases. These observations are in accordance with the results of previous work showing that although gabapentin is antinociceptive both during and after inflammation, prostaglandin dependency is lost in the late-phase hypersensitivity of mice in the K/BxN model (23).

Thus, our study findings strongly suggest that the mechanisms that drive mechanical hypersensitivity in the inflammatory and late phases of antibody-induced pain in these models are different. COX inhibition has a very limited effect on neuropathic pain (40). In addition, activation of spinal astrocytes is associated with peripheral nerve injury (13), and we found that, independent of mouse strain and the antibody cocktails used, astrocyte reactivity was increased in the late phase. These observations, together with the antiallodynic effect of gabapentin, suggest the possibility that neuropathic pain mechanisms may be involved in the late phase of arthritis in the CAIA model. Although it was not examined in this study, expression of a nerve injury marker, activating transcription factor 3 protein, is induced in the dorsal root ganglia of mice in the K/BxN and CIA models (23, 36). This provides support for the notion that antibody-induced inflammation, even when transient, may have an impact on nociceptive sensory nerves that resembles nerve damage.

Cold allodynia is a frequent feature in different types of nerve injury–induced pain (41, 42). A classic view, albeit controversial (43), is that RA patients experience increased joint pain during cold weather. Thus, we assessed whether mice with CAIA displayed altered nocifensive behavior after acetone application. Interestingly, in the CAIA model, cold allodynia occurred at the same time as joint inflammation and, in contrast to the mechanical hypersensitivity, did not persist beyond the signs of arthritis. We did not assess cold allodynia in the K/BxN model, but did observe transient heat hypoalgesia (23). Therefore, it appears as if nociceptive neurons that are activated by either cold stimuli or heat stimuli are affected only in the short term, correlating with the extent of inflammation, in models of antibody-induced pain. The mechanistic explanation for the different temporal profiles of CAIA-induced mechanical and cold allodynia warrants further studies.

In contrast to nerve injury models, short-lasting models of subcutaneous soft tissue inflammation do not always drive microglia activation (44). Of note, in this study, QB mice did not display morphologic signs of microglia activation, but CBA mice did. Since the degree of arthritis was higher and the inflammatory episode lasted longer in CBA mice, it is possible that there is a threshold at which peripheral inflammation leads to spinal microglia reactivity. In accordance with such a notion, astrocyte and microglia activation–associated mRNA levels in QB mice were increased during both the early and late phases of arthritis in the model, indicating that peripheral inflammation initiates changes in glia-associated factors, but that this does not always lead to changes in glia-related protein expression.

In contrast to spinal microglia, the temporal profile of astrocyte reactivity was similar in the 2 strains. Astrocyte activation has frequently been reported to occur later than microglia activation in experimental models of pain, and it has been postulated that astrocytes, rather than microglia, play a more pronounced role in the long-term aspect of chronic pain (24). Of importance, even though we did not determine which cell type, if not both, contributes to spinal sensitization in the CAIA model, intrathecal injection of the glia inhibitor pentoxifylline attenuated CAIA-mediated late-phase pain behavior, indicating a functional coupling between increased glia reactivity and arthritis-induced mechanical hypersensitivity.

Activation of MAPKs plays an important role in the regulation of spinal pain processing (45). Phosphorylation of the MAPK JNK in the spinal cord was pronounced in the late phase of arthritis in the CAIA model. Although we failed to determine the cellular location of JNK activation, previous studies have shown that peripheral nerve injury and inflammation activate JNK in spinal astrocytes (29, 46, 47). Intrathecal injection of the JNK inhibitor JIP-1 partially reversed the late-phase hypersensitivity, indicating that spinal JNK participates in the modulation of arthritis-induced pain transmission.

In summary, the results of the present study show that the CAIA model represents a multifaceted model for studies aimed at exploring the pain mechanisms in conditions of RA-induced pain, during both active and controlled states of the disease. Our findings of a time-dependent prostaglandin and glial involvement in arthritis-induced pain highlight the importance of using appropriate disease models when investigating the pathologic processes related to joint pain.


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. Svensson 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. Bas, Lundberg, Holmdahl, Svensson.

Acquisition of data. Bas, Su, Sandor, Agalave, Lundberg, Codeluppi, Baharpoor.

Analysis and interpretation of data. Bas, Su, Sandor, Agalave, Lundberg, Codeluppi, Baharpoor, Nandakumar, Holmdahl, Svensson.


We appreciate the technical assistance provided by Emma Mondoc, Christina Christianson, and Joshua Gregory, and the critical reading and editing of the manuscript provided by Ada Delaney.