CSF-1 in Inflammatory and Arthritic Pain Development

Pain is one of the most severe symptoms associated with inflammatory/autoimmune diseases, ultimately leading to functional impairment, disability, and poor quality of life and is the major symptom for which patients with rheumatoid arthritis (RA), for example, want effective treatment. RA is an autoimmune-mediated, inflammatory joint disease characterized by an abundance of macrophages, with there being a positive correlation between synovial macrophage numbers and RA severity (1–3). RA pain is believed to arise from multiple mechanisms, involving inflammation, peripheral and central pain processing, and structural changes within the joint (4). Inflammatory cytokines and/or algesic mediators produced by immune cells, such as synovial macrophages, and/or by components of the nervous system are thought to be a major cause of such pain (5–9).

CSF-1, also known as M-CSF, is a member of the hematopoietic CSF family, is ubiquitously expressed in the steady state, and is a key regulator for the survival, proliferation, and differentiation of macrophage lineage populations, including osteoclasts (3, 10–14). The CSF-1R is a transmembrane tyrosine kinase (c-Fms) and is most highly expressed on the surface of cells of the monocyte/macrophage lineage (12, 15). Studies using neutralizing anti–CSF-1R mAb blockade have suggested that in the steady state in the adult, CSF-1 acts rather late in monocyte/macrophage development (16–19). CSF-1R signaling is also important in the brain, being expressed on microglia and possibly neurons, although the latter expression is debated (2, 20–22). The presence of another CSF-1R ligand, IL-34, broadens its biology but also makes interpretation of effects of its blockade more complicated (23, 24).

Aside from its role in homeostasis, there is supportive evidence that CSF-1/CSF-1R signaling is important in pathology (2, 12), for example, in models of arthritis (25–29), neurodegeneration (20), diabetes (30), multiple sclerosis (31), Alzheimer disease (32), systemic lupus erythematosus (33), and cancer (2). Clinical trials targeting CSF-1R or CSF-1 in inflammatory diseases and cancer are ongoing (2). Due in part to its prosurvival and/or mitogenic activity on macrophages, CSF-1 is one of the cytokines, along with, for example, GM-CSF and TNF, implicated in promoting accumulation of macrophages in the RA synovium (3, 34). High CSF-1 levels have been noted in the RA synovium (35), and anti–CSF-1R mAb can reduce inflammatory mediator production in RA synovial explant cultures (29). Studies using animal models of RA have indicated the involvement of CSF-1 and its receptor (25–29) in disease pathogenesis, perhaps by enhancing the accumulation of synovial macrophages, as well as by inducing osteoclastogenesis and bone destruction (3).

As regards involvement of CSF-1R/CSF-1 in pain development, a central role has been indicated in regulating neuropathic pain following peripheral nerve injury (36). A study by Thompson et al. (37), based on findings with an oral CSF-1R inhibitor, suggested a role for CSF-1/CSF-1R signaling in bone cancer pain, which could be largely driven by components of inflammatory and neuropathic pain, as well as by osteoclasts causing bone remodeling and resorption (38). Although an inflammatory role for CSF-1/CSF-1R signaling has been well established in arthritis models (25–29), its potential involvement in arthritic pain has received limited attention.

Given the above background literature, we reasoned that CSF-1R/CSF-1 blockade by mAb may be effective in suppressing inflammatory pain, including that associated with arthritis, and that CSF-1 administration may also generate such pain. Using zymosan-induced inflammatory and arthritic models, which we have shown to be CCL17, GM-CSF, and TNF dependent (39, 40), and models driven by GM-CSF and TNF, which we have shown to mediate both inflammatory and arthritic pain and disease via CCL17 (39, 40), we report for the first time, to our knowledge, that systemic mAb neutralization of CSF-1R and/or CSF-1 can prevent, but not ameliorate, the development such pain and disease and can do so with timing similar to that of the clinically important TNF (41). We also show for the first time, to our knowledge, that systemic administration of CSF-1 can induce pain and, in addition, can drive inflammatory arthritis pain and disease in an inflamed joint by a mechanism different from that employed by exogenous GM-CSF and TNF.

C57BL/6 mice were originally from The Walter and Eliza Hall Institute (Kew, VIC, Australia). Transgenic mice were obtained as follows: RAG-1–deficient (Rag1^−/−) mice (42) and IFN regulatory factor-4–deficient (Irf4^−/−) mice (43) were from The Walter and Eliza Hall Institute and homozygous CCL17 mutant (Ccl17^E/E) mice, which have both CCL17 alleles replaced with enhanced GFP) (44), were from Prof. I. Förster (University of Bonn, Germany). Transgenic mice were all backcrossed onto the C57BL/6 background for at least 11 generations. Mice of both sexes (8–12 wk) were used. All experiments were performed in accordance with the ethics approval obtained from The University of Melbourne Animal Ethics Committee.

Inflammatory pain was induced by a single intraplantar (i.pl.; 10 μl) injection of zymosan (100 μg; Sigma-Aldrich, St. Louis, MO), GM-CSF (20 ng; PeproTech, NJ), TNF (20 ng; R&D Systems, Minneapolis, MN), or CSF-1 (6.65 μg; Chiron) into the left hind paw, whereas the contralateral hind paw received PBS (39); pain was monitored at 2, 4, and 6 h post i.pl. injection as before (39).

Arthritis was induced by an intraarticular (i.a.) injection of zymosan (300 μg; Sigma-Aldrich) in 10 μl of saline into the left joint, whereas the contralateral joint received 10 μl of saline as a control (39). Pain was monitored over a 7-d period using the validated incapacitance method as was done previously (39, 45–47) (see below). On day 7, mice were sacrificed, and knee joints were collected for histological analysis and/or cell population analysis using flow cytometry (FACS).

As before, mice were immunized with methylated BSA (mBSA) (Sigma-Aldrich), emulsified in CFA, intradermally in the base of the tail (39, 46, 48). Seven days later (day 0), arthritis was induced by an i.a. injection of mBSA into the left knee and saline into the right knee. Pain was monitored over a 3-d period.

mBSA/cytokine arthritis models were induced as before (39, 45, 49) by an i.a. injection of mBSA (200 μg in 0.9% saline) into the left knee and 0.9% saline into the right knee on day 0, followed by s.c. injection of saline (negative control), murine TNF (500 ng; R&D Systems), murine GM-CSF (500 ng; PeproTech), murine IL-1β (250 ng; R&D Systems), or human CSF-1 (6.65 μg; Chiron) on days 0–2. Mice were sacrificed on day 2, 4, or 7 post–mBSA injection to collect knee joints for histology or cell population analysis.

Mice were treated i.p. with 200 μl of anti-murine CSF-1 mAb (150 μg) (18) or its humanized IgG1 isotype control (150 μg) (MorphoSys, Planegg, Germany); anti-murine CSF-1R mAb (250 μg, AFS98; Dr. S. I. Nishikawa, Kobe, Japan) (18), anti-murine GM-CSF mAb (150 μg, 22E9.11) (50), anti-Ly6G mAb (150 μg) (46), or their IgG2a isotype control (250 μg); or anti–G-CSFR (150 μg, 5E2-VR81; CSL) (46), anti-murine TNF mAb (150 μg, XT-22) (40), or their IgG1 isotype control (150 μg) (Schering-Plough BioPharma, Palo Alto, CA) at various times as indicated for the different models (see 14Results section). In some experiments, anti–CSF-1R mAb and its IgG2a isotype control were given i.pl. (2 μg) at the same time as the indicated stimulus (see 14Results section). Mice were treated i.p. with the cyclooxygenase (COX) inhibitor indomethacin (1 mg/kg; Sigma-Aldrich) daily once pain was evident; pain was measured 1 h posttreatment (39, 45).

As a measure of pain, the differential distribution of weight over a 3-s period between the inflamed paw or limb relative to the noninflamed paw or limb was measured using an incapacitance meter (IITC Life Science, Woodland Hills, CA) as before (39, 45–47). Mice were acclimatized to the incapacitance meter on at least three occasions prior to the commencement of the experiment. Three measurements were taken for each time point and averaged.

At termination, knee joints were collected, fixed, decalcified, and paraffin-embedded (45, 46, 48). Frontal sections (7 μm thick) were stained with H&E. For zymosan-induced arthritis (ZIA), cell infiltration, synovial hyperplasia, bone erosion, and proteoglycan loss (Safranin O/Fast green stain) were scored separately from 0 (normal) to 3 (severe) (39). For the mBSA/cytokine-induced arthritis models, histopathologic features, including cell infiltration, synovitis, pannus formation, cartilage damage, and bone erosion, were scored using an established scoring system ranging from 0 (normal) to 5 (severe), as previously described (39, 45, 49). The sum of these five features is presented as the total histologic score.

Synovial tissue cells were isolated from knee joints and analyzed by flow cytometry as before (46, 48, 51). Briefly, isolated cells from joint tissue were incubated with normal mouse serum (1:4 dilution) and stained using the following Abs: PerCP-conjugated CD45 (clone 30-F11), allophycocyanin-conjugated CD11b (M1/70) and V500-conjugated I-A/I-E (M5/114.15.2) (BD Biosciences), allophycocyanin-Cy7–conjugated Ly6G (1A8) and PE-Cy7–conjugated CD11c (N418) (BioLegend), PE-conjugated F4/80 (BM8) (Life Technologies), and the corresponding isotype controls. Cells were analyzed using a CyAn flow cytometer (Beckman Coulter). Compensation was acquired using single-stained samples, and analysis was performed using Kaluza 1.2 software (Beckman Coulter).

Quantitative PCR experiments were performed as described previously (39, 40). Briefly, total RNA was extracted from footpad skin or joint cells using an Isolate II RNA Mini Kit (Bioline) and revere transcribed using Tetro reverse transcriptase (Bioline). Quantitative PCR was performed using an ABI 7900HT sequence detection system (Applied Biosystems) and predeveloped TaqMan probe/primer combinations for murine Ccl17, GMCSF, Tnf, and ubiquitin C (Ubc) (Life Technologies). All samples were assayed in duplicate. Threshold cycle numbers were transformed to ∆ threshold cycle values, and the results were expressed relative to the reference gene, Ubc.

For pain measurements, two-way ANOVA compared the differences among three or more groups. For FACS analysis, statistical significance was determined using a two-tailed Student t test. For histologic scores, a two-way or Kruskal–Wallis one-way ANOVA was used to compare the differences. For gene expression, a two-tailed Student t test was used. GraphPad PRISM software package (La Jolla, CA) was employed to perform these statistical tests. A Bonferroni post hoc test was applied when appropriate. Data are expressed as the mean ± SEM with significance as indicated. A p value ≤0.05 was considered significant.

i.pl. zymosan administration is a well-studied inflammation and pain model (39, 40, 46, 52, 53). CSF-1 involvement in this inflammatory pain model was examined using neutralizing anti–CSF-1R (AFS98) and anti–CSF-1 mAbs (i.p., days −2 and −1) (18, 48). At 4 and 6 h post i.pl. zymosan injection, pain was evident in mice pretreated with isotype controls but not in those treated with anti–CSF-1R mAb or anti–CSF-1 mAb (Fig. 1A), indicating a CSF-1 dependence. As some indication of the mechanism governing the CSF-1R signaling dependence, and consistent with the dependence of this model on CCL17, GM-CSF, and TNF (39, 40), the mRNA expression levels of these cytokines at 6 h post i.pl. zymosan injection were lower in anti–CSF-1R mAb–treated mice compared with isotype-treated mice (Fig. 1B).

ZIA, initiated upon i.a. zymosan, is a widely used macrophage-dependent (54) monoarticular arthritis model. To determine whether CSF-1 is also required for ZIA pain and disease development, the neutralizing anti–CSF-1R and anti–CSF-1 mAbs were again used. Mice were treated with the mAbs on days −2, 0, and 3 to ensure sufficient coverage, with arthritis being induced on day 0. On day 1 following ZIA induction, mice treated with isotype controls showed pain, which was evident until day 4, but mice treated with anti–CSF-1R or anti–CSF-1 mAb (Fig. 2A) did not, again indicating a CSF-1 dependence. Data for TNF involvement, using an anti-TNF mAb and its isotype control, are included by way of comparison (Fig. 2A). Histologically, mice treated with anti–CSF-1R or anti–CSF-1 mAb showed significantly less arthritis (monitored by cell infiltration), synovial hyperplasia, proteoglycan loss (as determined by the amount of red staining in the cartilage by the Safranin O/Fast green stain), and bone erosion (39) than mice treated with their respective isotype controls (Fig. 2B).

We also determined the effect of anti–CSF-1R mAb and anti–CSF-1 mAb blockade on the infiltrating leukocyte numbers in the ZIA synovium using flow cytometry. Notwithstanding the challenges faced in defining categorically mononuclear phagocyte system populations by surface marker expression (18, 55), we identified the different synovial myeloid populations, using a gating strategy similar to that previously published (46, 48), and defined them as follows: leukocytes (CD45⁺), neutrophils (CD11b⁺Ly6G⁺), and macrophages (CD11b⁺F4/80⁺MHCII^+/−). All of these populations were reduced in number upon anti–CSF-1R or anti–CSF-1 treatment (Fig. 2C).

Again, regarding the mechanism and consistent with ZIA dependence on CCL17, GM-CSF, and TNF (39, 40), in ZIA joints, the mRNA expression levels of these cytokines were significantly lower in anti–CSF-1R mAb–treated mice compared with isotype-treated mice (Fig. 2D), similar to those seen in the footpad following i.pl. zymosan. To gain more insight into how and when endogenous CSF-1 is acting, we also determined whether the blockade of CSF-1R or CSF-1 had an effect on ZIA pain or disease once initiated. Interestingly, such neutralization of CSF-1R or CSF-1 (i.p. mAb on days 1, 3, and 5) did not ameliorate ZIA pain (Supplemental Fig. 1A) or disease (Supplemental Fig. 1B). Similar pain data for such TNF blockade are again included as a comparison (Supplemental Fig. 1A).

In contrast to ZIA, we previously reported that, unlike anti–GM-CSF or anti–GM-CSFRα mAb blockade, neither anti–CSF-1R mAb nor anti–CSF-1 mAb blockade prevented Ag-induced arthritis (AIA) development as measured by histology (48). We therefore checked whether the same applied for AIA pain. AIA mice were treated (days −2 and 0) with anti–CSF-1R mAb, anti–CSF-1 mAb, or isotype controls with no effect on pain development in this T lymphocyte–dependent model (Supplemental Fig. 2A); delayed neutralization of CSF-1R or CSF-1 (i.p., days 2 and 4) also did not ameliorate AIA pain (data not shown) or disease (histology on day 7) (48). In another T lymphocyte– and GM-CSF–dependent monoarticular arthritis model, the so-called mBSA/IL-1 model (i.a. mBSA day 0, s.c. IL-1β days 0–2) (56, 57), we found previously that neither anti–CSF-1R mAb nor anti–CSF-1 mAb reduced arthritis severity (57). We therefore tested whether CSF-1 activity was required for pain induction in this model. As for the AIA model, the degree of pain developed by mBSA/IL-1 mice treated (days −2 and 0) with anti–CSF-1R mAb or anti–CSF-1 mAb was similar to that seen in mBSA/IL-1 mice pretreated with their respective isotype controls (Supplemental Fig. 2B); again, no effect on arthritis severity (day 7) was noted (Supplemental Fig. 2C).

Collectively, these data indicate that CSF-1, like TNF, is required for the development of zymosan-induced inflammatory pain and for ZIA pain and optimal disease, and blockade of CSF-1 signaling results in lower Ccl17, GMCSF, and Tnf mRNA expression at the inflammatory site. In contrast, CSF-1 is not required for the development of AIA and mBSA/IL-1 pain and disease.

i.pl. GM-CSF also induces acute inflammatory pain in a TNF- and CCL17-dependent manner (39, 40). Anti–CSF-1R mAb treatment (days −2 and −1) prevented GM-CSF–induced pain development (Fig. 3A). Upon blockade of CSF-1R signaling, the increased Ccl17 and Tnf mRNA expression was reduced at 6 h (data not shown).

We have established a GM-CSF–driven, lymphocyte-independent, monoarticular arthritis model (i.a. mBSA day 0, s.c. GM-CSF days 0–2), thereby potentially enabling downstream mediators of GM-CSF action to be defined (39, 40); we have reported that pain and disease in this model are TNF and CCL17 dependent (39, 40). We used this model to determine whether CSF-1R signaling was required for GM-CSF–driven pain and disease development. Treatment [days −2 and 0, as before (40)] of mBSA/GM-CSF mice with anti–CSF-1R mAb prevented the development of pain seen in isotype-treated mice (Fig. 3B) as well as the development of optimal disease (Fig. 3C). Once again, Ccl17 and Tnf mRNA expression were reduced in mBSA/GM-CSF joints from mice treated with anti–CSF-1R mAb compared with isotype control (Fig. 3D). Delayed blockade of CSF-1R (day 2) again did not reverse the pain (Supplemental Fig. 3A) or ameliorate the arthritis (Supplemental Fig. 3B).

Thus, similar to zymosan-induced inflammatory pain and ZIA and like TNF, CSF-1R signaling is required for the development of GM-CSF–induced inflammatory pain and of GM-CSF–driven arthritic pain and disease.

There are several reports demonstrating that i.pl. TNF induces inflammatory pain (58–61), and we have recently found this pain to be GM-CSF and CCL17 dependent (40). CSF-1R neutralization also prevented i.pl. TNF-induced inflammatory pain development (Fig. 4A). Upon blockade of CSF-1R signaling, the increased Ccl17 and GMCSF mRNA expression (6 h) was reduced (data not shown).

We have also established a TNF-driven, lymphocyte-independent, monoarticular arthritis model (i.a. mBSA day 0, s.c. TNF days 0–2), likewise enabling mediators downstream of TNF to be potentially defined (40); we have found this model to be GM-CSF and CCL17 dependent (40). As for the mBSA/GM-CSF model, pain was not evident (Fig. 4B) and arthritis did not progress (Fig. 4C) following anti–CSF-1R mAb administration (days −2 and 0); delayed CSF-1R blockade (day 2) was again ineffective (Supplemental Fig. 3C, 3D).

Thus, similar to GM-CSF–induced inflammatory pain and the mBSA/GM-CSF arthritis model, CSF-1R signaling is required for the development of TNF-induced inflammatory pain and of TNF-driven arthritis pain and disease.

The i.pl. injection of CSF-1 itself was sufficient to induce inflammatory pain within 4 h (Fig. 5A).

Similar to the mBSA/IL-1 (49, 62), mBSA/GM-CSF (39, 45), and mBSA/TNF (40) arthritis models, CSF-1 has also been shown to exacerbate the mild synovitis seen following i.a. mBSA, with the degree of synovitis peaking on day 7 before resolving by day 21 (49); arthritic pain, however, was not assessed in that study. Mice injected with mBSA/CSF-1 (i.a. mBSA day 0, s.c. CSF-1 d 0–2) displayed significant pain from 6 h until day 4 (Fig. 5B) [i.e., with rapid kinetics similar to that observed in the mBSA/TNF arthritis model (Fig. 4B)]. Arthritis severity was increased over days 2–7 compared with mBSA/saline mice and then declined (Fig. 5C).

We also examined the mechanistic requirements of such CSF-1–driven arthritis pain and disease.

Regarding synovial cell populations, FACS analysis at days 2, 4, and 7 indicated that in the mBSA-injected joint, s.c. CSF-1, compared with s.c. saline, increased total leukocyte (CD45⁺), neutrophil (CD11b⁺Ly6G⁺), and macrophage (predominantly CD11b⁺Ly6G⁻F4/80⁺MHCII⁻CD11c⁻) numbers, which were maximal at day 2 (data not shown). As for the mBSA/GM-CSF (39) and mBSA/TNF (40) models, mBSA/CSF-1 pain and disease were independent of T and B lymphocytes because they were similar in wild type (WT) and Rag1^−/− mice (Fig. 6A, 6B, respectively). Even though synovial neutrophil numbers also increased, their depletion upon administration of anti-Ly6G (46) failed to prevent pain or disease development in the mBSA/CSF-1 model (Supplemental Fig. 4A, 4B, respectively); in contrast, for both the mBSA/GM-CSF and mBSA/TNF models, anti-Ly6G depletion blocked both pain and disease development (Supplemental Fig. 4C–E). G-CSF is a key cytokine for the regulation of neutrophil development, activation, and trafficking, particularly during an inflammatory reaction (2, 12); neutralizing anti–G-CSFR mAb (46) in the mBSA/CSF-1 arthritis model also failed to alter pain and disease, indicating independence from G-CSF signaling (Supplemental Fig. 4F, 4G, respectively).

The pain in the mBSA/GM-CSF and mBSA/TNF models required COX activity, indicating eicosanoid dependence (40, 45); however, as seen in Fig. 6C and 6D, respectively, both mBSA/CSF-1 pain and disease were not altered by the COX inhibitor indomethacin.

Pain and disease progression in the mBSA/GM-CSF and mBSA/TNF arthritis models require the chemokine, CCL17, whose formation in turn is dependent upon the transcription factor IRF4 (39, 40). However, CCL17 is not required in the mBSA/CSF-1 model because pain and disease were similar in gene-deficient (Ccl17^E/E) mice to those observed in WT mice (Fig. 6E, 6F, respectively). Surprisingly, given this and our prior data (39, 40), mice treated with anti–GM-CSF mAb or anti-TNF mAb (days −2 and 0) did not develop mBSA/CSF-1 arthritis pain (Fig. 7A, 7C, respectively) or disease (Fig. 7B, 7D, respectively). Using Irf4^−/− mice, and paralleling the Ccl17^E/E mouse data (Fig. 6E, 6F), IRF4 was found not to be required (Fig. 7E, 7F), suggesting that the contributions of GM-CSF and TNF in this model are not via this transcription factor.

We found above that i.p. CSF-1R/CSF-1 blockade suppressed the onset of pain and disease in three arthritis models driven, respectively, by zymosan, GM-CSF, and TNF as well as the pain initiated by i.pl. injection of these stimuli, indicating a CSF-1 requirement. This is the first report, to our knowledge, showing that neutralization of CSF-1R or CSF-1 systemically can alleviate pain, indicating a peripheral action for CSF-1. Supporting this peripheral site of action, we also report for the first time, to our knowledge, that systemically administered CSF-1 could induce arthritic and inflammatory pain. In the arthritis models, as for TNF neutralization, the blockade with the neutralizing mAbs was only effective if they were administered early with respect to the stimulus. Our neutralizing mAb approach is a continuation of the one adopted previously by us in studies comparing the roles of CSF-1 and GM-CSF in homeostasis and pathology (2, 12, 18, 48, 49). Many studies have used oral antagonists of CSF-1R (c-Fms; CD115) kinase activity but run the risk of nonspecificity, as has been discussed (2). We again compared the effects of anti–CSF-1R and anti–CSF-1 mAbs because the existence of another CSF-1R ligand, namely IL-34, clouds interpretation of approaches targeting only the receptor (23, 24).

We recently found (39) that GM-CSF–induced inflammatory pain and GM-CSF–driven arthritis pain and disease (mBSA/GM-CSF model) require IRF4 and CCL17. We now show that the proinflammatory action of GM-CSF in these models requires CSF-1 as an intermediary. Despite the fact that TNF is a key cytokine in many human inflammatory disorders, its mode of action in these conditions is still unknown. As for the GM-CSF–driven regulation, we provided evidence above that CSF-1 can also play a part in exogenous TNF regulation of pain and arthritis. Thus, CSF-1 can play a critical role in the pain and arthritic disease elicited by two cytokines that are pivotal to RA pathogenesis (2, 63). As regards the possible mechanism(s) for the effectiveness of early CSF-1R blockade in the joint, reduced synovial myeloid cell numbers were noted in ZIA, along with a reduction in expression of Tnf, GMCSF, and Ccl17 mRNA at the site of inflammation consistent with the requirement for these cytokines in this model (39, 40). It may be that CSF-1 is needed only when joint tissue levels of TNF or GM-CSF are sufficient; such levels could fluctuate even during progression of a chronic disease (e.g., during an RA “flare”). We have recently found that there is a requirement for GM-CSF in the same exogenous TNF-driven model (40), again indicating a link between the activities of the two CSFs in question. It has been proposed that the successful treatment of the TNF transgenic mouse arthritis model with a TNF inhibitor was due to decreased migration of monocytes and/or increased synovial macrophage apoptosis (64); based on our finding, perhaps CSF-1 could be involved in regulating these functions. Also, and encouragingly for any clinical applications arising from our above findings, the benefit of early but not late CSF-1 blockade parallels that of anti-TNF blockade for ZIA (Fig. 2) (40) and mBSA/GM-CSF models (Fig. 3) (40), given that TNF blockade is quite successful in many human inflammatory conditions (63). Consistent with this similarity, both anti-TNF mAb (65, 66) and anti–CSF-1R mAb blockade (29) reduce inflammatory mediator production in RA synovial explant cultures.

Systemic administration of CSF-1 can exacerbate collagen-induced arthritis (25) and a monoarticular mBSA-primed arthritis model (49). We have taken advantage of the latter model by showing that pain is induced rapidly. Regarding mechanism, and even though the effects of systemic administration of a ligand may not necessarily be informative about its endogenous role (2, 12), increased myeloid cell infiltration and local macrophage proliferation (49) were observed in the mBSA-injected joint in the mBSA/CSF-1 model, consistent with our proposal above on the role of endogenous CSF-1 in increasing myeloid cell numbers in inflammatory arthritis, as well as previously in certain other inflammatory reactions (2, 3, 18). Lymphocytes, neutrophils, and G-CSFR signaling appear not to be involved in this CSF-1–driven arthritis model. For neutrophils, at least, this was found to be in contrast to both GM-CSF– and TNF-driven arthritis pain and disease (46). In accord with the inability of CSF-1 to induce PGE₂ in murine macrophages (67) and human monocytes (68), pain and arthritis in the mBSA/CSF-1 model, again unlike the pain in the mBSA/GM-CSF and mBSA/TNF models (39, 40), were not modified by COX inhibition indicating a lack of eicosanoid involvement. Interestingly, despite synovial tissue Ccl17 mRNA expression being increased in the mBSA/CSF-1 model (data not shown) in a similar fashion to the zymosan-, GM-CSF–, and TNF-driven models (39, 40), there was no requirement for CCL17 or IRF4. Given we have previously shown that CSF-1 is unable to upregulate CCL17 or IRF4 in monocytes/macrophages (39, 69, 70), this suggests that both the increased tissue Ccl17 mRNA expression in the mBSA/CSF-1 model and the decreased tissue Ccl17 mRNA expression seen following CSF-1R blockade in the zymosan-, GM-CSF–, and TNF-driven models are not direct. Intriguingly, however, there was a requirement for GM-CSF and TNF in the mBSA/CSF-1 model. Although this may explain the increased Ccl17 expression (i.e., via GM-CSF), it does not explain why CCL17 is not required; the reason for this is currently being investigated.

Although the TNF-driven (mBSA/TNF) and the CSF-1-driven (mBSA/CSF-1) arthritis models seem to function via different pathways, TNF and CSF-1 interact with the mBSA-injected joint with the same rapid kinetics. How systemically administered CSF-1 upregulates the pain and arthritis in the mBSA-“primed” (inflamed) joint, presumably by acting on a monocyte/macrophage population(s) (2, 12, 15, 24), and with a GM-CSF and TNF requirement, remains to be elucidated. It has been proposed that CSF-1 could be used as a therapeutic agent in tissue repair (15); the pain induction in the mBSA/CSF-1 model and the arthritis progression in this and in other arthritis models (25, 49) indicate that such therapeutic applications should proceed with caution. We have previously invoked CSF-1, GM-CSF, and TNF as being part of an inflammatory cytokine “network” with possible implications for regulating macrophage numbers and function at a site of chronic inflammation: for example, the RA synovium (3, 12, 34). From the above, these concepts can be extended to include the regulation of inflammatory pain. Our data confirm that even though CSF-1 and GM-CSF are both prosurvival/mitogenic factors for monocytes/macrophages (2, 3, 12, 18, 34) and, as mentioned, their activities can be linked, they have very different functions overall (2, 3, 12, 18, 69, 71).

How might endogenous CSF-1 be regulating pain and arthritis peripherally and when is it likely to be important in this context? Emerging roles for CSF-1R and its ligands in the nervous system have recently been reported (2, 20–22). There is disagreement as to whether there is CSF-1R expression in the neuronal lineage, with developmental control probably being important (2, 19–22). CSF-1 has been proposed to act centrally on microglia to control nerve injury–induced neuropathic pain and to therefore be a target in this condition (36, 72). In lupus-prone mice, the development of chronic pain is associated with the activation of microglia and overexpression of CSF-1 in the spinal cord (73). Our i.p. mAb blockade data above suggest that CSF-1 can also act peripherally to regulate pain, probably by its action on macrophages in tissues and possibly in nerves (74). In this connection, local i.a. CSF-1R blockade–reduced joint swelling and pain are consistent with peripheral ligand activity (75), and important cross-talk between CSF-1–activated muscularis macrophages and enteric neurons has been identified (2, 76).

The three arthritis models studied above in which early CSF-1R blockade was effective are lymphocyte independent, whereas other monoarticular models in which no effect on pain or arthritis was observed, namely the AIA and mBSA/IL-1 models, are T lymphocyte dependent (48, 56), suggesting a possible explanation. However, mAb-mediated CSF-1R or CSF-1 blockade is effective in the systemic, T lymphocyte–dependent collagen-induced arthritis model, although there is debate as to in what stage CSF-1 might be acting (25–27, 29). Our findings above on early blockade appear similar to those showing that the anti–CSF-1R mAb (AFS98) used above prevents the initial events of atherogenesis but does not reduce the size of advanced lesions in apolipoprotein E–deficient mice (77). There is evidence that CSF-1 acts rather late in the development of the monocyte/macrophage lineage, with a key activity likely being on regulating the numbers of resident tissue macrophages and more mature blood monocytes rather than less mature “inflammatory” monocytes (2, 16–18). We have proposed as a possible mechanism that at a site of inflammation, CSF-1 neutralization depletes resident tissue macrophages, thereby governing overall myeloid and lymphoid cell infiltration (3, 18), a proposal consistent with our findings in the ZIA model and the suggested role of phagocytosis of zymosan by such macrophages in inflammation (78, 79); a similar mechanism may also apply to the poorly soluble mBSA (80–82) in the joint. In this connection, depletion of resident pancreatic islet macrophages by the AFS98 mAb reduced diabetes in the NOD mouse (30) and suppressed diabetic nephropathy in db/db mice by targeting macrophage-mediated injury (83). Our concept invoking the significance of CSF-1 action on resident tissue macrophage numbers and/or function (3, 18) offers an explanation for the efficacy of early CSF-1R/CSF-1 blockade, and our findings above clarify how and when CSF-1 might function as a critical proinflammatory mediator.

Our findings above have important ramifications for the clinical trials in inflammatory diseases mainly targeting CSF-1R with oral antagonists and that so far are not particularly encouraging (2). It could be that CSF-1 is only relevant in patients with conditions in which the driving inflammatory insults in tissues are acute and/or periodic and involve resident tissue macrophages. Moreover, our findings also suggest that CSF-1 can be targeted peripherally with an mAb strategy to control pain and can be critical to the pathogenic roles of the clinically important cytokines, TNF and GM-CSF (2, 41, 84). These ideas, based in part on our current findings, perhaps could be incorporated into the selection and design of future anti–CSF-1R/CSF-1 clinical trials.

We thank members of the Melbourne Brain Centre Parkville Flow Cytometry Facility for assistance.

The authors have no financial conflicts of interest.