Involvement of Peripheral and Spinal Tumor Necrosis Factor α in Spinal Cord Hyperexcitability During Knee Joint Inflammation in Rats

Authors


Abstract

Objective

Tumor necrosis factor α (TNFα) is produced not only in peripheral tissues, but also in the spinal cord. The purpose of this study was to address the potential of peripheral and spinal TNFα to induce and maintain spinal hyperexcitability, which is a hallmark of pain states in the joints during rheumatoid arthritis and osteoarthritis.

Methods

In vivo recordings of the responses of spinal cord neurons to nociceptive knee input under normal conditions and in the presence of experimental knee joint inflammation were obtained in anesthetized rats. TNFα, etanercept, or antibodies to TNF receptors were applied to either the knee joint or the spinal cord surface.

Results

Injection of TNFα into the knee joint cavity increased the responses of spinal cord neurons to mechanical joint stimulation, and injection of etanercept into the knee joint reduced the inflammation-evoked spinal activity. These spinal effects closely mirrored the induction and reduction of peripheral sensitization. Responses to joint stimulation were also enhanced by spinal application of TNFα, and spinal application of either etanercept or anti–TNF receptor type I significantly attenuated the generation of inflammation-evoked spinal hyperexcitability, which is characterized by widespread pain sensitization beyond the inflamed joint. Spinally applied etanercept did not reduce established hyperexcitability in the acute kaolin/carrageenan model. In antigen-induced arthritis, etanercept decreased spinal responses on day 1, but not on day 3.

Conclusion

While peripheral TNFα increases spinal responses to joint stimulation, spinal TNFα supports the generation of the full pattern of spinal hyperexcitability. However, established spinal hyperexcitability may be maintained by downstream mechanisms that are independent of spinal TNFα.

Pain in the joints as occurs in rheumatoid arthritis, osteoarthritis, and other rheumatic diseases is generated by neuronal mechanisms in the peripheral as well as central nociceptive system. Peripheral sensitization, the sensitization of peripheral nociceptors for stimuli, initiates the process of pain generation. However, nociceptive mechanisms in the central nervous system dramatically influence the phenotype of pain, such as its severity, persistence, localization, and spreading ([1]). Central processes include the mechanism of central sensitization (development of hyperexcitability of nociceptive spinal cord neurons), reduction of endogenous inhibitory mechanisms, such as diffuse noxious inhibitory control, and activation of the cortical and subcortical pain matrix ([1, 2]). Immune mediators, such as proinflammatory cytokines, seem to contribute to the orchestration of these neuronal events, thus providing a direct connection between the disease process and pain generation.

Tumor necrosis factor α (TNFα), a key mediator of human inflammatory joint diseases such as rheumatoid arthritis ([3, 4]), is also a peripheral pain mediator because it causes a persistent sensitization of nociceptive sensory neurons ([5]). Neutralization of TNFα has been shown to significantly reduce mechanical and thermal hyperalgesia in animal models of collagen-induced arthritis ([6]) and antigen-induced arthritis (AIA) ([7]), as well as in humans with rheumatoid arthritis ([8]). Antinociception was achieved well before the inflammation was attenuated ([8]) or even when inflammation was only moderately attenuated ([7]), thus underscoring effects of TNFα on the nociceptive system. TNFα and other cytokines are also produced by glial cells at different levels of the central nervous system ([9, 10]). Although activation of glial cells is not generally observed in inflammation ([11]), astroglia and microglia were found to be activated in the K/BxN serum–transfer model of arthritis ([12]) and in models of experimental osteoarthritis ([13]), which supports the need to clarify the role of cytokines produced by glial cells in joint pain.

In the present study, we focused on central sensitization. We first investigated whether peripheral TNFα was able to contribute to the generation of spinal sensitization and whether neutralization of peripheral TNFα (in the joint) could reduce spinal responses. Second, we examined whether TNFα was increased in the spinal cord upon peripheral inflammation and whether spinal TNFα was involved in the generation and maintenance of spinal sensitization upon inflammation, and if so, which TNF receptor was involved. Previous work has shown involvement of spinal TNFα in neuropathic pain states resulting from nerve injury ([14]), but the precise role of spinal TNFα in inflammatory pain states is less clear. Neutralization of spinal TNFα has been implicated in the control of inflammation in the joint ([15, 16]), which strongly suggests that spinal TNFα plays a functional role in arthritis. We obtained electrophysiologic recordings of nociceptive neuron activity in animals in vivo, which allowed us to directly monitor the neuronal changes evoked by peripheral and spinal TNFα.

MATERIALS AND METHODS

Animals and induction of joint inflammation

Adult male Wistar rats (weighing 250–400 gm), adult wild-type C57BL/6J mice, and adult TNFRI−/− and TNFRII–/– mice, both of which were backcrossed to C57BL/6J mice, were studied. Animals were housed under 12-hour light/12-hour dark conditions in a temperature-controlled environment, with unrestricted food and water.

In anesthetized rats, an acute inflammation in the knee joint was induced by injection of 70 μl of a 4% kaolin suspension (Sigma), followed by injection of 70 μl of a 2% carrageenan solution (Sigma), into the joint cavity ([17]). To trigger AIA ([18]), rats were immunized twice at a 1-week interval by subcutaneous injection of 1 mg/ml of methylated bovine serum albumin (mBSA; Sigma) dissolved in 0.9% NaCl supplemented with Freund's complete adjuvant (Sigma) containing 2 mg/ml of Mycobacterium tuberculosis strain H37Ra (Difco). Monarticular AIA in the left knee joint was induced by intraarticular injection of 70 μl of a 10-mg/ml preparation of mBSA after another 2 weeks.

In vivo recordings of spinal cord neuron activity

Rats were deeply anesthetized with 100–120 mg/kg of sodium thiopental intraperitoneally (Trapanal; Rotexmedica). The trachea was cannulated. Catheters filled with Tyrode's solution and heparin (heparin-natrium 25,000 units; Ratiopharm) were inserted into the carotid artery to control mean arterial blood pressure (usually, a mean ± SD of ∼100 ± 20 mm Hg). Body temperature was kept at 37°C using a feedback-controlled temperature constanter. Additional injections of sodium thiopental (12 mg/kg intraperitoneally) maintained deep anesthesia (stable blood pressure during noxious stimulation and absence of corneal reflexes).

The rat was fixed in a frame, laminectomy was performed to expose segments L1–L4, and the dura mater was opened. A trough (3 × 5 mm) was tightly sealed to the spinal cord surface and filled with 50 μl of Tyrode's solution. The activity of individual neurons was recorded extracellularly using glass-insulated carbon filaments. Neurons that responded to pressure applied to the ipsilateral left knee, but not to stroking and squeezing of the overlying skin, were selected for recording. Action potentials of a neuron were continuously monitored on an oscilloscope and saved on a PC for offline spike analysis. For stimulation, a manometer (Correx) was used to apply pressure to the mediolateral axis of the knee (at an angle of ∼30° left of the midline) at innocuous (1.9 N/40 cm2) and noxious (7.8 or 5.9 N/40 cm2) intensities. The ankle and paw were stimulated with defined clamps (1.1 N/20 cm2 for innocuous intensity and 5.8 N/20 cm2 for noxious intensity). Each stimulus lasted 15 seconds and was followed by 15 seconds without stimulation. The entire stimulation cycle was repeated every 5 minutes.

After establishing the baseline responses of the neurons to mechanical stimuli, recombinant rat TNFα (ImmunoTools) at the concentrations indicated below and/or 50 μl of a solution containing 25 μg/μl of etanercept (Enbrel; Pfizer) were either injected into the knee joint cavity or applied to the spinal trough at the recording site. Control groups received vehicle. Other experimental groups received a spinal injection of anti–TNF receptor type I (anti-TNFRI) antibody (10 μg/ml, MAB 625; R&D Systems), anti-IgG1 antibody (control; eBioscience), or anti-TNFRII antibody (MAB 226; R&D Systems).

For offline data analysis, action potentials were discriminated by shape and were counted using Spike/Spidi software. Mean neuronal responses to stimuli (action potentials/15 seconds) in consecutive intervals of 30 minutes were compared to a baseline of 30 minutes before application of the compound. The significance of changes in comparison to baseline was assessed within groups using Wilcoxon's matched pairs signed rank test. For intergroup comparisons, the Mann-Whitney U test was used. Bonferroni correction was applied if necessary. P values less than 0.05 were considered significant.

Measurement of TNFα.

TNFα levels were measured in spinal cord supernatants. A spinal trough with a capacity of 120–140 μl was filled with Tyrode's solution (120 μl), which was then withdrawn after 20 minutes and immediately frozen at –20°C (see the Results section for the sampling protocol). To measure TNFα levels in synovial fluid, ∼70 μl of warm Tyrode's solution (37°C) was injected into each knee joint at 4 hours after the onset of ipsilateral inflammation, and using the same syringe, the fluid was withdrawn from the synovial cavity and immediately frozen at –20°C. To measure TNFα in spinal tissue, ∼5 mm of lumbar spinal cord was removed from normal rats, immunized rats, or rats with AIA (different time points after AIA onset). The meninges were removed, and the tissue was flushed with phosphate buffered saline (PBS) and frozen at –80°C. For each 10 mg of tissue, 100 μl of PBS (Invitrogen) plus 10 mM EDTA that had been freshly supplemented with protease inhibitor cocktail tablets (Roche) was added, and tissues were sonicated for 4 minutes (∼54W; Covaris and LGC Genomics). Tissue lysates were centrifuged and used for enzyme-linked immunosorbent assay (ELISA) after protein determination (bicinchoninic acid assay; Thermo Scientific). All samples were finally processed with a commercial enzyme immunoassay kit for rat TNFα (BDset OptEIA; eBioscience).

Sodium dodecyl sulfate–polyacrylamide gel electrophoresis and Western blotting of material from cultured dorsal root ganglion (DRG) neurons

DRGs were dissected from rats and mice and collected in Ham's F-12 medium (PAA). Tissue was digested for 1 hour with type II collagenase (Sigma) and for 10 minutes with trypsin (Sigma) at 37°C. DRG cells were suspended in Ham's F-12, dispersed by mechanical trituration, and seeded into poly-l-lysine (50 μg/ml)–coated 12-well plates in Ham's F-12 and 10% horse serum (Sigma) supplemented with 1 mM glutamine (Invitrogen), 1% penicillin/streptomycin (PenStrep; Invitrogen), and 10 ng/ml of nerve growth factor (ProSpec). Cells were used within 24 hours of culture and starved for 4 hours in pure Dulbecco's modified Eagle's medium before stimulation.

Cells were treated with 25 ng/ml of recombinant rat or mouse TNFα (ProSpec) for different time periods. The reaction was stopped on ice, medium was aspirated, and cells were scraped and harvested with lysis buffer (20 mM HEPES, 10 mM EGTA, 40 mM β-glycerophosphate, 2.5 mM MgCl2, 1% Nonidet P40, pH 7.4, freshly supplemented with protease inhibitor cocktail tablets [Roche]). After a freeze–thaw cycle, cell lysates were centrifuged, and 6× Laemmli loading buffer was added. Protein samples were separated on 12.5% polyacrylamide gels at 125V and transferred immediately to 2 overlaid PVDF membranes (Millipore). Immunoblotting with anti–phospho-NF-κB p65 (Ser536) (catalog no. 3033; Cell Signaling Technology) on the first PVDF membrane and anti–pan–NF-κB p65 (catalog no. 4764; Cell Signaling Technology) on the second PVDF membrane was performed. The signals were visualized with horseradish peroxidase–linked secondary antibodies (KPL) enhanced by a chemiluminescence reaction (Thermo Scientific) and viewed with a CCD camera system (Synoptics).

RESULTS

Recording of nociceptive neuron activity

We recorded the activity of nociceptive neurons in the deep dorsal horn at a depth of 787 ± 245 μm (mean ± SD) from the spinal cord surface. All neurons had mechanical receptive fields at the knee joint and the ankle. Some neurons also responded to muscle stimulation, and some showed weak responses to paw stimulation. Neurons responding to input from the skin overlying the knee were excluded. The activity of 1 neuron was recorded in each animal.

Spinal effects of TNFα injection into the knee joint

The experimental setup for examining the spinal effects of TNFα injection into the knee joint is shown in Figure 1A. After injection of 5 ng of TNFα into the knee joint, the responses of the spinal neurons to both innocuous and noxious stimulation of the knee joint increased significantly (P < 0.05 by Wilcoxon's matched pairs signed rank test) (Figure 1B). After vehicle injection, responses of the spinal neurons did not significantly change (Figure 1B). The responses to ankle stimulation after TNFα injection into the knee joint showed a tendency to increase, but the enhancement was not significant, and responses to paw stimulation displayed small increases that were only transiently significant (data not shown). Thus, peripheral application of TNFα significantly enhanced spinal responses to knee stimulation, matching the time course of the sensitization of Aδ fibers and C fibers of the knee joint to mechanical stimulation by TNFα ([5]), but the responses to ankle and paw stimulation showed only a tendency to increase.

Figure 1.

Effects of peripheral application of tumor necrosis factor α (TNFα) on spinal responses to peripheral stimulation. A, Experimental setup. Innocuous (innoc.) and noxious (nox.) stimuli were applied to the knee, ankle, and paw, and the responses of spinal cord neurons to knee and ankle stimulation were recorded in vivo. Each stimulus lasted 15 seconds, followed by 15 seconds without stimulation. B, Number of action potentials (APs) in response to innocuous and noxious pressure on the knee and ankle joints before and after injection of either vehicle or TNFα into the knee joint (large arrow). ∗ with small arrow = P < 0.05 for comparison of each TNFα value from the indicated point to the end point versus responses before injection, by Wilcoxon's matched pairs signed rank test. C, Levels of TNFα in lavage fluid from the knee joints of healthy rats (control) and rats injected with kaolin/carrageenan into the knee joint 4 hours previously to induce acute inflammation. = P < 0.05 by Mann-Whitney U test. D, Number of APs (minus baseline) before and after injection of etanercept into the knee joint 7–11 hours after the initiation of acute joint inflammation. Baseline responses (mean ± SEM) were as follows: 323 ± 158 APs/15 seconds for innocuous knee stimulation, 687.6 ± 186.5 APs/15 seconds for noxious knee stimulation, 221 ± 89 APs/15 seconds for innocuous ankle stimulation, and 449 ± 107 APs/15 seconds for noxious ankle stimulation. with small arrow = P < 0.05 for innocuous versus noxious stimulation from the indicated point to the end point, by Wilcoxon's matched pairs signed rank test. Values in B–D are the mean ± SEM for the indicated number of animals.

Upon development of kaolin/carrageenan-induced inflammation in the knee joint, the TNFα concentration in the lavage fluid from the knee joint was found to increase by 757%, from a mean ± SEM of 9.4 ± 19.4 pg/ml before inflammation to 71.2 ± 37.3 pg/ml at 4 hours after injection of kaolin/carrageenan (Figure 1C). The injection of etanercept into the inflamed knee joint 7–11 hours after induction of inflammation partially reduced spinal mechanical hyperexcitability (Figure 1D). While the neuronal responses to noxious stimulation of the knee were reduced from an initial value of 687.6 ± 186.5 action potentials (APs)/15 seconds to 494.5 ± 169.2 APs/15 seconds (P < 0.05 by Wilcoxon's matched pairs signed rank test), the responses to innocuous stimulation of the knee and the responses to ankle stimulation were not altered (Figure 1D).

Effect of TNFα on cultured DRG neurons

In order to test whether peripheral TNFα effects are mediated by neuronal TNFRI or TNFRII, we applied TNFα to isolated DRG neurons and used Western blotting to examine which TNF receptor activated the NF-κB pathway, a typical signaling cascade ([19]). In primary cultures of rat DRGs (Figure 2A) and mouse DRGs (Figure 2B), TNFα induced an increase in activated NF-κB by phosphorylation of its serine residue 536 within 15 minutes. This NF-κB activation was not found in DRG neurons from TNFRI–/– mice (Figure 2C), but it was preserved in DRG neurons from TNFRII–/– mice (Figure 2D). Thus, this effect of TNFα depends on TNFRI. (Complete Western blots are available upon request from the corresponding author).

Figure 2.

Western blot analysis of the phosphorylation of NF-κB at 1 minute, 5 minutes, and 15 minutes after application of tumor necrosis factor α (TNFα) to cultured dorsal root ganglion (DRG) neurons in rats (A), wild-type (WT) mice (B), TNFRI–/– mice (C), and TNFRII–/– mice (D). The Western blot results were quantified and are shown as data columns under the individual blots. Values are the mean ± SEM of 3 cultures.

Effect of spinal application of TNFα in normal rats

The experimental setup for examining the effect of spinal application of TNFα in normal rats is shown in Figure 3A. Application of TNFα directly to the spinal cord surface caused a concentration-dependent increase in spinal neuron responses to mechanical stimulation of the knee and ankle joint (Figure 3B). The neuronal responses to noxious stimulation of the knee were significantly enhanced over baseline within 30 minutes after application of the lowest concentration of TNFα. Responses to innocuous stimulation of the knee and to both innocuous and noxious stimulation of the ankle significantly increased starting at 50 minutes after TNFα application (P < 0.05 by Wilcoxon's matched pairs signed rank test) (Figure 3B).

Figure 3.

Effects of tumor necrosis factor α (TNFα) applied to the spinal cord. A, Experimental setup. TNFα or TNFα plus etanercept (etan) was applied to the surface of the spinal cord. Innocuous (innoc.) and noxious (nox.) stimuli were then applied to the knee, ankle, and paw, and the responses of spinal cord neurons to knee and ankle stimulation were recorded in vivo. Each stimulus lasted 15 seconds, followed by 15 seconds without stimulation. B, Number of action potentials (APs; minus baseline) in response to innocuous and noxious pressure on the knee and ankle joints before and after application of the indicated concentrations of TNFα to the spinal cord. ∗ with arrow = P < 0.05 for innocuous versus noxious stimulation from the indicated point to the end of the protocol, by Wilcoxon's matched pairs signed rank test. Baseline responses (mean ± SEM) were as follows: 140 ± 50 APs/15 seconds for innocuous knee stimulation, 393 ± 103 APs/15 seconds for noxious knee stimulation, 131 ± 31 APs/15 seconds for innocuous ankle stimulation, and 422 ± 95 APs/15 seconds for noxious ankle stimulation. C, Number of APs (minus baseline) in response to innocuous and noxious pressure on the knee and ankle joints before and after application of TNFα and subsequent application of TNFα plus etanercept to the spinal cord. The absolute numbers of APs in time blocks (each column represents 30 minutes) for innocuous and noxious stimulation of the knee are shown at the right. = P < 0.05 versus control, by Wilcoxon's matched pairs signed rank test followed by Bonferroni correction. D, Number of APs (minus baseline) in response to innocuous and noxious knee pressure on the knee and ankle joints before and after application of TNFα plus etanercept and subsequent application of TNFα to the spinal cord. The absolute numbers of APs in time blocks (each column represents 30 minutes) for innocuous and noxious stimulation of the knee are shown at the right. Values are the mean ± SEM for the indicated number of animals.

Spinal application of 1 ng/μl of TNFα caused a significant increase in spinal neuron responses, reaching a stable maximum after 1 hour (Figure 3C). During subsequent coapplication of TNFα (1 ng/μl) and etanercept (12.5 μg/μl), responses were slightly, but not significantly, decreased within 1 hour (Figure 3C). In contrast, pretreatment with etanercept prevented the increase of spinal neuron activity by TNFα (Figure 3D). Thus, spinal application of TNFα robustly increases the responses to mechanical joint stimulation. Pretreatment with etanercept can prevent this effect, but posttreatment only slightly reduces the TNFα-induced responses within the first hour after application.

Contribution of spinal TNFα to the generation of spinal hyperexcitability

In order to test whether endogenous spinal application of TNFα contributes to the generation of inflammation-evoked spinal hyperexcitability, we assessed the development of spinal hyperexcitability in control rats and in rats in which etanercept was applied to the spinal cord before induction of kaolin/carrageenan-induced joint inflammation. The experimental protocol is shown in Figure 4A. After establishing responses to mechanical stimulation during a control period of 30 minutes, either vehicle or etanercept was administered to the spinal cord. Then, kaolin/carrageenan was injected into the knee joint, and the recordings were continued for a further 4 hours.

Figure 4.

Role of tumor necrosis factor α (TNFα) during the development of kaolin/carrageenan-induced inflammation. A, Experimental protocol for recordings (see C and D) and for obtaining samples of spinal cord supernatants (see B). K/C = injection of kaolin/carrageenan into the knee; a with shaded arrow = mechanical stimulation protocol for the leg; b with solid line = replacement of vehicle by etanercept in the active treatment group; thin vertical arrows = sampling of supernatants in the vehicle group. B, TNFα concentrations in spinal cord supernatants obtained before triggering kaolin/carrageenan-induced inflammation (shaded bars; no stim. = vehicle without mechanical stimulation, and control = vehicle with mechanical stimulation) and during the development of kaolin/carrageenan-induced inflammation (open bars) in vehicle-treated rats at the indicated times, as measured by enzyme-linked immunosorbent assay. C, Comparison of neuronal responses to stimulation of the knee and ankle joints of vehicle-treated rats (control) and etanercept-treated rats during the development of inflammation following kaolin/carrageenan injection. D, Comparison of neuronal responses to stimulation of the knee and ankle joints of IgG1 isotype–treated rats (IgG1 isotype control) and anti-TNF receptor type I (anti-TNFRI) antibody (Ab)–treated rats during the development of kaolin/carrageenan-induced inflammation. In C and D, the mean number of action potentials (APs)/15 seconds at baseline was subtracted from all values, the result of which was a normalization to zero. * = P < 0.05 versus baseline, by Wilcoxon's matched pairs signed rank test; + = P < 0.05 between treatment groups, by Mann-Whitney U test; arrows indicate significant difference from the indicated time point to the end point. Values are the mean ± SEM for the indicated number of animals.

In the vehicle control experiments, we also removed fluid samples from the spinal cord surface, as indicated in Figure 4A. During the period of mechanical stimulation of the leg, we found a slight increase in the TNFα concentration, but the amount of TNFα was not further increased during development of knee inflammation (Figure 4B).

In vehicle-treated rats, the development of knee inflammation significantly increased the responses to innocuous and noxious stimulation of the knee, and in parallel, the responses to innocuous and noxious stimulation of the ankle were enhanced (Figure 4C). In the etanercept group, the responses to knee and ankle stimulation did not significantly increase, although the swelling at the knee joint was similar as in the control group. The comparison of the control group with the etanercept group showed significant differences for both stimulation sites and both types of stimulation (Figure 4C). Thus, endogenous TNFα contributes to the development of spinal hyperexcitability.

Using the same approach, we explored which TNF receptor is involved in the development of spinal hyperexcitability. For this purpose, we applied antibodies against TNFRI and TNFRII to the spinal cord. Compared to the effect of its corresponding immunoglobulin isotype (IgG1 control), the anti-TNFRI antibody reduced the responses of spinal cord neurons to noxious stimulation of the knee and ankle joints starting at 2 hours after induction of inflammation (Figure 4D). Application of an anti-TNFRII antibody to the surface of the spinal cord did not prevent increases in the responses to stimulation of the leg during the development of inflammation: for responses before versus 4 hours after kaolin/carrageenan induction, the values were 17 ± 10 versus 97 ± 56 APs/15 seconds for innocuous stimulation of the knee, 79 ± 45 versus 590 ± 341 APs/15 seconds for noxious stimulation of the knee, 39 ± 22 versus 117 ± 67 APs/15 seconds for innocuous stimulation of the ankle, and 87 ± 50 versus 197 ± 114 APs/15 seconds for noxious stimulation of the ankle (n = 3 animals per group). Application of neither anti-TNFRI nor anti-TNFRII showed an effect of its own in normal rats (data not shown).

We next applied etanercept to the spinal cords of rats in which kaolin/carrageenan-induced inflammation in the joint had been initiated 7–11 hours before the recordings (i.e., inflammation and spinal hyperexcitability were already established). Spinal application of 12.5 μg/μl of etanercept did not significantly reduce the neuronal responses to mechanical stimulation of the knee, ankle, or paw. Responses before versus 2 hours after etanercept were 141 ± 32 versus 117 ± 24 APs/15 seconds for innocuous stimulation of the knee and 360 ± 71 versus 374 ± 79 APs/15 seconds for noxious stimulation of the knee. Thus, within 2 hours, etanercept did not reduce inflammation-evoked hyperexcitability in the rat model of kaolin/carrageenan-induced inflammation.

Effect of spinal application of etanercept in rats with AIA

Because spinal application of etanercept had no effect on established hyperexcitability and inflammation in the rat model of kaolin/carrageenan-induced inflammation, we also assessed the spinal effect of etanercept on AIA. This inflammation is immune-mediated and has an acute phase (days 1–3) followed by a chronic phase of inflammation ([20]). Recordings were performed on day 1 (24 hours after induction of inflammation) or day 3 of AIA. After application of vehicle to the spinal cord, the responses to mechanical stimulation of the leg did not change significantly (Figure 5A), but after spinal application of etanercept on day 1 of AIA, the responses to knee, ankle, and paw stimulation were consistently reduced. In contrast, etanercept did not alter the neuronal responses on day 3 after onset of inflammation.

Figure 5.

Effect of spinal application of vehicle or etanercept in rats with antigen-induced arthritis (AIA). A, Normalized responses (action potentials [APs]/15 seconds minus baseline) to innocuous and noxious stimulation of the knee and ankle joints on day 1 after induction of AIA in vehicle-treated rats and on day 1 and day 3 in etanercept-treated rats. Baseline responses (mean ± SEM) were as follows: in the vehicle-treated group, 402 ± 104 APs/15 seconds for innocuous knee stimulation, 868 ± 97 APs/15 seconds for noxious knee stimulation, 599 ± 94 APs/15 seconds for innocuous ankle stimulation, and 824 ± 139 APs/15 seconds for noxious ankle stimulation; in the etanercept-treated group on day 1, 606 ± 94 APs/15 seconds for innocuous knee stimulation, 1,183 ± 128 APs/15 seconds for noxious knee stimulation, 495 ± 126 APs/15 seconds for innocuous ankle stimulation, and 781 ± 153 APs/15 seconds for noxious ankle stimulation, and on day 3, 242 ± 123 APs/15 seconds for innocuous knee stimulation, 494 ± 106 APs/15 seconds for noxious knee stimulation, 272 ± 132 APs/15 seconds for innocuous ankle stimulation, and 467 ± 102 APs/15 seconds for noxious ankle stimulation. Values are the mean ± SEM for the indicated number of animals. B, Levels of TNFα in the lumbar spinal cord (L1–L4) of normal animals, preimmunized (immun.; methylated bovine serum albumin–treated) animals, and in animals with AIA on days 1, 3, 7, and 21 after induction (n = 7 animals per group). Solid line traces the mean across groups.

The tissue from the lumbar spinal cord was removed at different stages of AIA and evaluated for TNFα content by ELISA. The amount of TNFα per milligram of tissue at each time point is shown in Figure 5B (n = 7 rats per time point). Overall, there were no significant differences in the TNFα content of the samples.

DISCUSSION

The present study revealed that peripheral as well as spinal application of TNFα causes an increase in the responses of spinal cord neurons to mechanical stimulation of the knee joint, showing that TNFα at both sites facilitates mechanonociception. In contrast, intraarticular administration of etanercept reduced spinal responses to noxious stimulation of the inflamed joint, and the development of spinal hyperexcitability after induction of joint inflammation was attenuated by spinal application of either etanercept or anti-TNFRI antibody. However, spinal application of etanercept reduced established spinal hyperexcitability only on day 1 of AIA, but not on day 3, and not in the kaolin/carrageenan model, suggesting that neutralization of spinal TNFα during persistent inflammation does not unequivocally reduce established spinal hyperexcitability.

Injection of TNFα into the knee joint increased the spinal responses to knee stimulation over the same time course as it sensitizes joint afferent nerve fibers for mechanical stimuli ([5]), but it did not result in a significant parallel increase in the responses of spinal cord neurons to ankle stimulation. Thus, the increased responses mainly reflect the increased peripheral input from the knee to the spinal cord, but it does not indicate a state of significant spinal hyperexcitability. The latter condition would be characterized by an increased response to both knee and ankle stimulation upon knee inflammation and an increased response to paw stimulation and an expansion of receptive fields toward the paw ([17, 21]) (Figure 4). In contrast, injection of etanercept into the inflamed joint reduced only the response to noxious stimulation of the knee, the site of inflammation, but there was no parallel response to mechanical stimulation of the ankle. Thus, peripheral TNFα is less potent than peripheral interleukin-6 (IL-6), together with its soluble receptor sIL-6R, in generating spinal hyperexcitability, although injection of IL-6/sIL-6R into the knee joint was shown to sensitize only articular C fibers ([22]), whereas TNFα sensitizes articular C fibers as well as articular Aδ fibers ([5]). One possible explanation for this is that the total number of fibers activated by IL-6/sIL-6R is much higher than the total number of fibers expressing TNF receptors ([16]), or the nociceptive fibers with TNF receptors have a lower potential to induce spinal sensitization than other fibers have.

TNFα stimulation of isolated DRG neurons rapidly phosphorylated NF-κB. This TNFα effect was determined to be mediated by TNFRI because activation of TNFRI specifically phosphorylates NF-κB on Ser residue 536 ([19]) and because it was abolished in DRGs from TNFRI–/– mice but not TNFRII–/– mice. In addition, it suggests that TNFα stimulation triggers long-term effects in DRG neurons because NF-κB is a transcription factor. Furthermore, NF-κB can evoke protein–protein interactions at sodium channels ([23]). It should be noted, however, that this particular approach did not selectively address TNF receptors in sensory afferent fibers in the joint because DRGs contain sensory neurons that supply different organs. Nevertheless, the data suggest that the neuronal effects of TNFα in the joint are likely to be mediated by TNFRI.

Spinally applied TNFα significantly increased the responses of spinal neurons to joint stimulation, showing responsiveness of spinal neurons to TNFα. The attenuation of the generation of spinal hyperexcitability during developing inflammation by etanercept and anti-TNFRI antibody provides clear electrophysiologic evidence that endogenous spinal TNFα is involved in the development of spinal hyperexcitability. Interestingly, however, this was not reflected by a significant increase in soluble TNFα in spinal cord supernatants, which is necessary for TNFRI activation ([24]). This may be the result of technical and biologic causes. Diffusion from the site of release to the surface may be low, and there is profound internalization of TNFα, with even immunoactive tissues generally showing low levels of TNFα ([25, 26]). In the peripheral tissue, inflammation may cause shedding of TNFRI, as seen in patients with rheumatoid arthritis ([27]). The ELISA may not recognize soluble complexes of TNFα and TNFRI. Furthermore, mediators that further the development of inflammation (e.g., prostaglandins) may simultaneously inhibit the production of TNFα ([28]). In fact, levels of prostaglandin E2 in the spinal cord are rapidly increased upon the development of kaolin/carrageenan-induced inflammation ([29]) and contribute to the generation of spinal hyperexcitability ([17]). On the other hand, the spinal content of TNFα was neither up-regulated nor down-regulated during the course of AIA.

In contrast, spinal levels of TNFα protein and messenger RNA were reported to be significantly enhanced on day 21 of Freund's adjuvant–induced arthritis ([30]). Interestingly, patients with rheumatoid arthritis were not reported to display higher TNF levels in the cerebrospinal fluid as compared with the controls (whereas IL-1β was strongly increased and IL-6 was weakly increased), which also shows that spinal TNFα production may not significantly change upon peripheral inflammation ([31]). A significant increase in the levels of IL-1β was correlated with fatigue, but not with mechanical hypersensitivity ([31]).

The data suggest that spinal TNFα modifies as a constantly present cofactor the effects of other mediators, which prevail under the particular conditions. In isolated spinal cord slices, TNFα has been shown to increase the frequency of spontaneous excitatory postsynaptic potentials as well as AMPA- and N-methyl-d-aspartate (NMDA)–induced currents in dorsal horn neurons ([32-34]). In fact, both AMPA and NMDA receptors for the glutamate transmitter are crucially involved in the generation of spinal hyperexcitability in the kaolin/carrageenan model of inflammation ([35]). TNFα can also down-regulate glutamate transporters in astrocytes and delay the clearance of glutamate from the synaptic cleft ([36]), and it can facilitate excitation by inhibiting superficial inhibitory GABAergic dorsal horn neurons ([37, 38]). In addition, TNFα may activate microglia, which then release a number of pronociceptive mediators ([10]). We previously found that spinal IL-6 was significantly increased within 3–4 hours following kaolin/carrageenan injection into the knee joint ([21]) and that spinal IL-6 contributed to the development of inflammation-evoked spinal hyperexcitability ([21]). We are currently investigating whether the effects of TNFα are mediated in part by the activation of IL-6 release.

Notably, the ability of spinally applied etanercept to reduce the spinal state of hyperexcitability was dependent on the particular situation. In normal rats, the spinal coapplication of etanercept and TNFα after previous application of TNFα did not significantly reduce the spinal responses. The spinal application of etanercept during established kaolin/carrageenan–induced inflammation (7–11 hours following knee injection) did not reduce the responses to stimulation of the inflamed joint, whereas intraarticular injection of etanercept reduced the spinal responses. However, the spinal application of etanercept in rats with AIA significantly and consistently reduced the spinal responses on day 1, but not day 3, of the disease. Interestingly, at both time points, the pain behavior is similarly strong ([16]). Intrathecal application of etanercept with minipumps was shown to strongly attenuate mechanical hyperalgesia at the inflamed knee joint on days 1 and 3 of AIA, but in those experiments, the application of etanercept was continuous and may have started briefly before the onset of inflammation ([16]). The failure of spinal etanercept to reduce the enhanced spinal responses may indicate that established effects induced by TNFα are persistent and therefore not reversible by neutralizing TNFα, or it may indicate that TNFα is not an important mechanism of hyperexcitability at this particular phase of neuronal plasticity. Currently, we cannot differentiate between these mechanisms. Apart from its involvement in the generation of pain, spinal TNFα is involved in the control of inflammation in the joint because neutralization of TNFα reduces the severity of arthritis ([15, 16]).

In summary, this study shows that not only does peripheral TNFα play an important role in the generation of joint pain upon development of joint inflammation, but there is also a significant concomitant effect of spinal TNFα that supports the development of the full picture of central sensitization. In established inflammation and spinal hyperexcitability, however, only the peripheral application of etanercept consistently reduced the spinal responses, whereas the effect of spinal etanercept was dependent on the particular model of inflammation and the stage of inflammation. We therefore believe that the successful reduction of pain via neutralization of TNFα in the presence of established inflammation ([8]) most likely results from a reduction in peripheral levels of TNFα and a decrease in input from the peripheral nervous system. However, it would be interesting to know whether TNFα-neutralizing compounds reach the central nervous system and whether the central TNFα effects are also neutralized in patients with pain due to joint inflammation.

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. Ebersberger 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. Schaible, Ebersberger.

Acquisition of data. König, Zharsky, Möller.

Analysis and interpretation of data. König, Zharsky, Möller, Schaible, Ebersberger.

Acknowledgments

The authors thank Mrs. Konstanze Ernst and Mrs. Gabi Weigand for technical assistance.

Ancillary