Blockade of NKG2D ameliorates disease in mice with collagen-induced arthritis: A potential pathogenic role in chronic inflammatory arthritis




To assess the role of the activating receptor NKG2D in arthritis.


Levels of NKG2D and its ligands were determined by fluorescence-activated cell sorting, real-time polymerase chain reaction, and immunohistochemistry in rheumatoid arthritis (RA) synovial membrane tissue and in paw tissue from arthritic mice. Arthritis was induced in DBA/1 mice by immunization with type II collagen, and mice were treated intraperitoneally with a blocking anti-NKG2D antibody (CX5) on days 1, 5, and 8 after clinical onset and were monitored for 10 days.


We demonstrated expression of NKG2D and its ligands on human RA synovial cells and extended this finding to the paws of arthritic mice. Expression of messenger RNA for the NKG2D ligand Rae-1 was up-regulated, and NKG2D was present predominantly on natural killer (NK) and CD4+ T cells, in arthritic paw cell isolates. NKG2D was down-modulated during the progression of collagen-induced arthritis (CIA). NKG2D expression in arthritic paws was demonstrated by immunohistochemistry. Blockade of NKG2D ameliorated established CIA, with significant reductions in clinical scores and paw swelling. Histologic analysis of arthritic joints from anti-NKG2D–treated mice demonstrated significant joint protection, compared with control mice. Moreover, anti-NKG2D treatment significantly reduced both interleukin-17 production from CD4+ T cells in arthritic paws and splenic NK cell cytotoxic effector functions in vivo and in vitro.


Our findings indicate that blockade of NKG2D in a murine model and in human explants has beneficial therapeutic potential that merits further investigation in RA.

Natural killer (NK) cells are lymphocytes of the innate immune system with both cytotoxic and cytokine-producing effector functions, regulated by a repertoire of cell receptors, including activating receptors NKG2D, CD16 (Fcγ receptor III), NKp30, NKp44, and NKp46; killer cell immunoglobulin-like receptor; and the inhibitory receptors CD94/NKG2A and LY49 (mouse) (for review, see ref.1). The role of NK cells in the defense against infections and cancer has been studied extensively (1), but their role in autoimmunity and chronic inflammation, where they have been shown to both contribute to and inhibit disease in animal models of autoimmunity (2–7), is less well elucidated. NK cells are present in rheumatoid arthritis (RA) synovium, produce interferon-γ (IFNγ) and tumor necrosis factor α (TNFα) following cytokine activation, and induce contact-dependent monocyte TNFα production (8–11).

Evidence from mouse models and human disease tissue indicates that signaling via the NKG2D receptor may be involved in autoimmunity (12–15). NKG2D is a type 2 transmembrane receptor expressed on NK, NK T, γ/δ T cell receptor (γ/δ TCR)+, and CD8+ cells and activated macrophages in humans and on NK and activated T cells in mice (for review, see ref.16). It associates with the signaling adaptor peptide DAP10 (an alternative splice variant can also associate with DAP12 in mice) and, following receptor engagement, induces cytotoxicity, cytokine production, and/or proliferation (16).

The ligands of NKG2D (MICA, MICB, ULBP1–4, and RAET1 in humans and Rae-1, H60, and Mult1 in mice) belong to a family of stress-induced glycoproteins with structural homology to class I major histocompatibility complex (MHC) proteins, whose expression is up-regulated under pathologic conditions (16). Proteolytic shedding of ligands has been reported in cancer and autoimmune diseases (13, 17), and upon binding NKG2D, ligands induce internalization and degradation (18). RA synoviocytes express MICA and MICB (13), and the expression of NKG2D on CD4+ T cells, which normally do not express NKG2D, is elevated in patients with RA, Crohn's disease, systemic lupus erythematosus, and tumors (13, 19–22). In those previous studies, NKG2D was preferentially expressed on CD4+CD28− T cells, raising the possibility that NKG2D is an alternative to CD28 for costimulation. In mice, blockade of NKG2D is beneficial both in models of colitis (23, 24) and in nonobese diabetes, where administration of a nondepleting anti-NKG2D monoclonal antibody (mAb), CX5, during the prediabetic stage prevented disease by impairing the clonal expansion of NKG2D-expressing cytotoxic CD8+ T cells (12).

In this study, we examined the expression of NKG2D and its ligand(s) in RA synovial tissue and demonstrated that blockade of NKG2D significantly reduced TNFα release. Expression of Rae-1, the NKG2D ligand, and NKG2D+ NK, CD4+, and γ/δ TCR+ T cells was observed in arthritic mouse paws. Blockade of NKG2D in established collagen-induced arthritis (CIA) with anti-NKG2D mAb (CX5) ameliorated disease, and protected against pathologic joint damage. This finding and reports describing a lack of efficacy of anti-CD4 therapy in established CIA led us to investigate the mechanism of action. Our results indicated that a significant decrease in interleukin-17 (IL-17) production from CD4+ T cells from mouse paws, as well as reduced splenic NK cell cytotoxic effector function, following CX5 treatment could contribute to reduced clinical signs of disease. Taken together, our data highlight a role for NKG2D in CIA and indicated that further studies of NKG2D as a potential therapeutic target in RA.



Synovial tissue and peripheral blood samples were obtained from patients who fulfilled the American College of Rheumatology criteria for RA (25) or osteoarthritis (OA) (26). Patients provided informed consent. Approval was obtained from the Riverside Research Ethics Committee (reference: 07/H0706/81). Blood samples obtained from healthy donors at the Kennedy Institute of Rheumatology were used as controls.

RA synovial membrane cell cultures.

Synovial samples were digested as previously described (27) and cultured with 1–30 μg/ml of anti-NKG2D mAb (R&D Systems), 30 μg/ml of rat IgG1 isotype control (BD Biosciences), or 10 μg/ml of anti-TNFα mAb (infliximab; Centocor) plus 10 μg/ml of IL-1 receptor antagonist (IL-1Ra; Abcam) in RPMI media with glutamine (PAA Laboratories) and 5% fetal calf serum (FCS; Biowest) at 2 × 105 cells/well in 96-well flat-bottomed culture plates (Becton Dickinson Falcon). After 24 hours, TNFα and IL-6 levels in the supernatants were determined by enzyme-linked immunosorbent assay (ELISA; BD PharMingen).

Coculture of arthritic synovial cells with 2B4 cells.

Cells isolated from both RA and OA synovial samples were cultured for 1 week by serial passage. The presence of NKG2D ligands on adherent cells was investigated in cocultures with 2B4 cells carrying green fluorescent protein (GFP)–conjugated nuclear factor of activated T cells (NF-AT) transfected with human NKG2D/DAP10-CD3ζ complex as previously described (28). Positive clones were selected by coculturing with cells expressing ULBP3, and their GFP expression was analyzed by flow cytometry. Cells were cocultured with 2B4 transfectants overnight in RPMI containing 5% FCS, and GFP expression was analyzed by flow cytometry.

Flow cytometric analysis.

Surface proteins on human or mouse cells were detected with antibodies (at 4°C for 30 minutes) purchased from BD Biosciences, eBioscience, and BioLegend, and fixed using Cytofix (BD Biosciences) at 4°C for 15 minutes. For mouse intracellular cytokine staining, cells (2 × 106/ml) were cultured in medium containing phorbol myristate acetate (20 ng/ml; Calbiochem, Merck Chemicals), ionomycin (1 μM; Calbiochem), and brefeldin A (12.5 μg/ml; Sigma-Aldrich) at 37°C for 3 hours, permeabilized in 0.5% saponin (for 30 minutes) and stained (at 4°C for 30 minutes) for IL-17 (BD Biosciences) and IFNγ (eBioscience). Data acquisition was done with a BD FACSCanto II (BD Biosciences), and Flow Jo software (Tree Star) was used for analysis.

Real-time polymerase chain reaction.

RNA was extracted from the paws of mice with CIA using an RNeasy mini kit (Qiagen). Rae-1 and control hypoxanthine guanine phosphoribosyltransferase (HPRT) were detected using a TaqMan Gene Expression Assay. Levels of Rae-1 messenger RNA (mRNA) (in relative expression units) in arthritic paws were calculated using the ΔΔCt method using a paw sample from a naive mouse as a calibrator after normalizing against HPRT.

Immunohistochemical staining for NKG2D.

Paws from mice with CIA (n = 3) and paws from mice without CIA (n = 3) were snap-frozen, cryosectioned (in 8-μm–thick sections), and fixed in paraformaldehyde. Incubations were performed with 1) Tris buffered saline (TBS)/3% skimmed milk/10% goat serum; 2) rat anti-mouse NKG2D (1 μg/ml; R&D Systems) or rat isotype-specific control antibody (1 μg/ml; BD Biosciences) in TBS/0.5% skimmed milk/7% goat/3% mouse sera; or 3) biotinylated goat anti-rat antibody (Jackson ImmunoResearch) in the buffer described above (for group 2). Sections were incubated using a Vectastain ABC peroxidase kit (Vector) followed by DuPont blocking reagent (TNB) (NEN PerkinElmer Life Science), amplified with biotin-conjugated tyramide (NEN PerkinElmer Life Science), and incubated with Vectastain. Chromogenic reaction was achieved with liquid permanent red (Dako), and counterstainining with hematoxylin.

CIA model.

Male DBA/1 mice (H-2q; Harlan) and C57BL/6 mice (H-2b; Charles River) were immunized with an emulsion of either purified bovine or chicken type II collagen (29) with Freund's complete adjuvant. Arthritic mice were randomized into different treatment groups, and all paws were monitored daily for 10 days. A clinical scoring system and hind paw thickness readings (in mm) were recorded as previously described (30). Data are presented as the mean ± SEM change in readings from day 1 of disease over a 10-day period.

Therapeutic blockade of NKG2D in CIA.

Upon onset of arthritis, DBA/1 mice were treated intraperitoneally on days 1, 5, and 8 with 0.04, 0.2, or 0.8 mg/kg of a blocking, nondepleting anti-NKG2D mAb (CX5; eBioscience). C57BL/6 mice were treated with 0.8 mg/kg of CX5 on days 1, 5, and 8. An additional group of mice were injected intraperitoneally with 0.8 mg/kg of CX5 daily, while another group of control mice received affinity-purified rat IgG1, κ light chain isotype antibody (eBioscience) at 0.8 mg/kg on days 1, 5, and 8 after arthritis onset. Untreated arthritic mice were also included. All animal procedures, including determination of ethical humane end points, were performed in accordance with the principles of the Declaration of Helsinki and were conducted under the local project license Identifying and targeting mediators of inflammation (70/6533) as per Home Office regulations.

Histologic analysis of mouse hind feet.

Mouse hind feet were fixed in buffered formalin, embedded in paraffin, sagittally sectioned (5–6 μm), stained with hematoxylin and eosin (H&E), and scored by an observer (PFS) who was blinded with regard to treatment. Samples were scored on a scale of 0–3, where 0 = normal joint architecture; 1 = mild erosive focal changes to cartilage; 2 = moderate changes, with involved extensive synovial hyperplasia; and 3 = severe erosions with total joint destruction. Mouse joints scored as 0 or 1 were classified as protected, while those scored as 2 or 3 were classified as destroyed. Images were captured using a Leitz Dialux 22 camera and processed using Windows Spot Advance software, version 4.5.

Isolation of cells from mouse tissue.

Mice were euthanized, blood samples were collected in heparin (Sigma-Aldrich), and whole blood cell populations were obtained following erythrocyte lyses with red blood cell lysis buffer (Sigma-Aldrich). Individual spleens and draining inguinal lymph nodes were excised, and cell preparations were made in RPMI 1640 containing 10% FCS (Gibco Invitrogen). Paws from arthritic mice were skinned and digested with Liberase CI (0.42 mg/ml) and DNase I (1 mg/ml; Roche Diagnostics) for 90 minutes at 37°C.

Measurement of murine TNFα levels.

The viability of cells isolated by enzymatic digestion of individual arthritic paws (n = 16) from mice treated with isotype control or CX5 was assessed with trypan blue dye (0.45%; Sigma-Aldrich). Cells were plated for 24 hours, and supernatants were harvested. Adherent cells were freeze-thawed (4 cycles) and centrifuged to harvest lysates. All supernatants and lysates were assayed for murine TNFα levels (pg/ml) by sandwich ELISA (R&D Systems).

Cytotoxicity assay.

NK cells were isolated from splenocytes by positive selection using CD49b microbeads (Miltenyi Biotec). YAC-1 target cells were labeled with DDAO-SE (Molecular Probes) or CellVue Clariet (Sigma-Aldrich) and cultured with splenic NK cells in the presence of IL-2 for 16 hours at ratios of 10:1, 5:1, 2:1, 1:1, and 0:1. Intracellular staining for active caspase 3 (BD Biosciences) was performed to analyze the lytic activity of NK cells.

Statistical analysis.

Data were analyzed with Graph Pad Prism Software, version 5.01. All clinical data were analyzed by two-way analysis of variance (ANOVA). Histologic data were analyzed by the contingency table chi-square test for trend. In vitro data were analyzed by Student's t-test or one-way ANOVA with Bonferroni post hoc test for multiple comparisons.


Expression of NKG2D and its ligands in human arthritic tissue.

The presence of NKG2D ligands on adherent cells isolated from enzymatically dissociated RA and OA clinical tissue was analyzed using a GFP-conjugated 2B4 reporter cell line transfected with human NKG2D/DAP10-CD3ζ complex. As shown in Figure 1A, overnight coculture with transfectants resulted in an ∼10-fold increase in GFP expression in adherent cells from both patients with RA (n = 5) and patients with OA (n = 3) compared with NKG2D/DAP10-CD3ζ-transfected 2B4 cells cultured alone.

Figure 1.

NKG2D and ligand expression in arthritic tissue. A, Representative dot plots showing an increase in green fluorescent protein (GFP) expression in adherent cells from rheumatoid arthritis (RA) patients (n = 5) and osteoarthritis (OA) patients (n = 3). The 2B4-NF-AT-GFP parental reporter cell line expressing human NKG2D-DAP10/CD3ζ complex was incubated overnight with (left panels) or without (right panels) adherent cells from RA or OA synovial tissue. Detection of NKG2D ligands on synovial cells drives GFP expression on reporter cells, which were analyzed by flow cytometry. B, Percentage of NKG2D+CD4+ lymphocytes in peripheral blood lymphocytes (PBL) from healthy controls (HC) and patients with RA, synovial membrane (SM) from patients with OA and patients with RA, and synovial fluid (SF) from patients with RA. Cells were stained with anti-CD45, anti-CD4, and anti-NKG2D and analyzed by flow cytometry. Symbols represent individual subjects; horizontal lines show the median. C, Tumor necrosis factor α (TNFα) levels in RA synovial membrane cells cultured alone (cells only), with 1–30 μg/ml of anti-NKG2D monoclonal antibody, or with 30 μg/ml of isotype control antibody (IgG1) for 24 hours, expressed as a percentage of TNFα levels in cells cultured alone. TNFα levels in culture supernatants were determined by enzyme-linked immunosorbent assay. Bars show the mean ± SEM of pooled data from 3 patients. ∗∗∗ = P < 0.001 versus cells cultured alone and IgG1-treated cells, by one-way analysis of variance with Bonferroni post hoc test for multiple comparisons.

NKG2D expression was determined by flow cytometry (Figure 1B). In a small number of RA patients, >10% of CD4+ T cells from peripheral blood were positive for the receptor. In addition, NKG2D+CD4+ T cells were detected in a cohort of RA synovial fluid samples. (A mean ± SEM of 11.2 ± 1.4% of CD4+ T cells in RA synovial fluid were NKG2D+.) As of yet, the clinical significance of elevated levels of NKG2D+CD4+ cells in synovial fluid is unknown and needs further investigation.

RA synovial membrane cells were cultured with anti-NKG2D mAb (1–30 μg/ml) or isotype control (30 μg/ml) for 24 hours, and TNFα and IL-6 production were measured by ELISA. A significant dose-dependent reduction in TNFα levels was observed (Figure 1C), with TNFα reduced by 64% (P < 0.001), 42% (P < 0.05), and 26% (P < 0.05) in individual experiments. Blocking of NKG2D had no effect on IL-6 release, whereas anti-TNFα mAb plus IL-1Ra (positive control) completely inhibited IL-6 production in all 3 experiments (data not shown).

Up-regulation of Rae-1 mRNA expression in the paws of arthritic mice.

Rae-1 mRNA levels (Figure 2A) were significantly higher in the paws of arthrtitic mice (mean ± SEM 12.6 ± 1.1 REU; n = 5) than in the paws of naive mice (mean ± SEM 1.5 ± 0.4 REU; n = 5) (P < 0.0001). Rae-1 protein could not be detected in arthritic paws by immunohistochemistry or flow cytometry, which may be due to technical reasons, proteolytic shedding, or masking by binding to NKG2D.

Figure 2.

NKG2D and ligand expression in the arthritic paws of mice with collagen-induced arthritis (CIA). A, Rae-1 mRNA expression in the paws of naive mice and mice with CIA. A sample from a naive mouse was used as a calibrator after normalizing against hypoxanthine guanine phosphoribosyltransferase (HPRT). Bars show the mean ± SEM relative expression units (n = 5 mice per group). P value versus naive mice was determined by Student's unpaired 2-tailed t-test. B, Immunohistochemical staining of paws from a naive control DBA/1 mouse and a mouse with CIA with an isotype-specific control antibody (rat IgG2b) (left panels) and a rat anti-NKG2D monoclonal antibody (right panels). NKG2D+ cells (red) were seen in the inflamed paw as solitary cells (arrows) and intermingled in a lymphoid aggregate (arrowheads). Note the absence of any NKG2D+ cells in the naive mouse. No immunoreactivity was seen with the isotype control antibody. Nuclei (blue) were counterstained with hematoxylin. Bar = 50 μm.

Presence of NKG2D+ cell infiltrates in the paws of mice with CIA and down-regulation of NKG2D expression in the late phase of CIA.

NKG2D+ cells were detected by immunohistochemistry in inflamed arthritic paws, as individually stained cells and as lymphoid aggregate foci in the synovium. In contrast, cryosections of the paws of naive control mice failed to show any NKG2D+ cells, while no immunoreactivity was seen in sections from either naive or arthritic mice stained with isotype antibody control (Figure 2B).

We were unable to characterize the phenotype of subpopulations of cells expressing NKG2D in arthritic paws by immunohistochemistry. Therefore, we sought to characterize surface NKG2D expression on CD4+, CD8+, CD49b+, and γ/δ TCR+ cells by flow cytometry (Figure 3) in arthritic mice with early CIA (days 1–3 after onset) and late CIA (days 7–10 after onset). In early CIA, the dominant lymphocyte subpopulations in paws were CD4+ T cells (3.1% of the total cells) and γ/δ T cells (3.3%), whereas CD49b+ NK cells comprised 1.6% and CD8+ T cells comprised a mere 0.2% of the total number of cells. (Results are available from the author upon request.) NKG2D was expressed by, on average, 87% of CD49b+ NK cells, 14.2% of CD4+ T cells, and 55% of γ/δ T cells in the blood, draining lymph nodes, and spleen of arthritic mice with early CIA, with no differences observed relative to naive mice (Figures 3A–C).

Figure 3.

Down-regulation of NKG2D expression in the late phase of collagen-induced arthritis (CIA). Cells were harvested from heparinized whole blood and the draining lymph nodes (LNs), spleen, and paws of naive mice, mice with early CIA (days 1–3 after onset), and mice with late CIA (days 7–10 after onset). Cells were stained for cell surface markers. The percentage of cells expressing NKG2D and the mean fluorescence intensity (MFI) of NKG2D were analyzed within a lymphocyte gate set on the lymph node population followed by gating on A, CD49b+ natural killer (NK) cells, B, CD4+ cells, and C, γ/δ T cell receptor (γ/δ T)+ cells. Bars show the mean ± SEM of data from 4 individual mice for blood, draining LNs, and spleen and from 3 sets each containing data from 3 pooled arthritic paws. Paws from naive mice did not contain enough cells to be included in the analysis. ∗∗∗ = P < 0.001 by Student's unpaired 2-tailed t-test.

Interestingly, the proportion of NK cells expressing NKG2D in arthritic paws dropped significantly, by 16.5%, during the progression of CIA (P < 0.001) (Figure 3A). Similarly, the level of NKG2D expression (mean fluorescence intensity [MFI]) on NK cells, which was considerably higher in the paws compared to the blood, spleen, and lymph nodes of mice with early CIA, was decreased by 47.3% in the paws of mice with late CIA (P < 0.001) (Figure 3A), indicating a possible modulation of NKG2D+ NK cells during inflammation in CIA. In contrast, no significant changes in NKG2D expression (MFI) on CD4+ or γ/δ T cells were detected in the paws of mice with late CIA (Figures 3B and C).

Reduction of disease severity in CIA by treatment with anti-NKG2D antibody CX5.

Mice were treated with varying amounts of CX5, and NKG2D expression on NK cells from blood was analyzed longitudinally as a surrogate pharmacodynamic marker. A dose-response relationship was observed between the amount of injected CX5 and the residual level of accessible NKG2D expression on immune cells. Treatment with CX5 induced down-regulation of NKG2D, although a certain fraction of it was still expressed at the surface of the cell, albeit blocked by the injected CX5 antibody. (Results are available from the author upon request.) Furthermore, NKG2D expression was almost completely blocked by serum concentrations of CX5 >10 μg/ml. Pharmacokinetic/pharmacodynamic modeling was then used to predict the CX5 dosing regimens that would result in low, partial, or full blockade of NKG2D. (Results are available from the author upon request.)

Administration of CX5 intraperitoneally on days 1, 5, and 8 after the onset of arthritis significantly reduced disease severity, as evaluated by clinical score and paw thickness measurements, in both the DBA/1 and C57BL/6 mouse models (Figure 4). Amelioration of disease occurred in a dose-dependent manner, with the highest dose of 0.8 mg/kg being the most efficacious and pharmacokinetically active (Figures 4A and B). Administration of CX5 daily at 0.8 mg/kg was also protective and arrested disease progression in accordance with pharmacokinetic/pharmacodynamic predictions but conferred no additional clinical benefits over administration of 0.8 mg/kg of CX5 on days 1, 5, and 8 (Figures 4A and B).

Figure 4.

Dose-dependent response to treatment of established murine collagen-induced arthritis (CIA) with an anti-NKG2D monoclonal antibody (mAb). Clinical scores (A and C) and paw swelling (B and D) were determined in immunized DBA/1 mice with CIA (A and B) and immunized C57BL/6 mice with CIA (C and D) over a 10-day period and are shown as the change from the data obtained on day 1 of arthritis. Mice were treated intraperitoneally with 0.04 mg/kg (▾), 0.2 mg/kg (⧫), or 0.8 mg/kg (●) of an anti-NKG2D mAb, CX5, on days 1, 5 and 8. Alternatively, mice were treated daily with 0.8 mg/kg of CX5 (×). Rat IgG1, κ light chain mAb was administered as an isotype control antibody at 0.8 mg/kg (▴). Untreated arthritic mice (▪) were analyzed as a control. Values are the mean ± SEM (n = 7 mice per group). Broken lines show the baseline level (0.0). ∗∗∗ = P < 0.0001 versus untreated arthritic mice and isotype-treated control mice, by two-way analysis of variance.

No significant difference was noted in the average day of onset of arthritis, clinical score, or paw thickness readings (in mm) in the different treatment groups on day 1 of arthritis. (Data are available from the author upon request.) On day 5 of CIA, the mean ± SEM change in clinical score from day 1 was significantly reduced in DBA/1 mice treated with 0.8 mg/kg of CX5 (0.0 ± 0.3) compared to mice treated with isotype control (2.4 ± 0.6) (P < 0.0001) (Figure 4A). Similarly, the recorded mean ± SEM change in paw thickness from day 1 of arthritis was 0.3 ± 0.08 mm for untreated mice, 0.4 ± 0.07 mm for isotype-treated mice, 0.2 ± 0.09 mm for mice treated with 0.04 mg/kg of CX5, 0.16 ± 0.07 mm for mice treated with 0.2 mg/kg of CX5, and −0.03 ± 0.05 mm for mice treated with 0.8 mg/kg of CX5 (Figure 4B).

Similar results were obtained in C57BL/6 mice with established CIA. On day 10 after arthritis onset, the change in clinical score from day 1 was 0.07 ± 0.46 for CX5-treated mice compared with 3.6 ± 0.9 and 3.8 ± 0.6 for untreated mice and isotype-treated mice, respectively (Figure 4C). Paw thickness changes in C57BL/6 mice on day 10 were −0.03 ± 0.1 in CX5-treated mice, 0.54 ± 0.12 in untreated mice, and 0.53 ± 0.13 in isotype-treated mice (P < 0.0001 for CX5-treated mice versus both untreated mice and isotype-treated mice) (Figure 4D).

Protection of mouse joints by anti-NKG2D antibody CX5.

Administration of CX5 protected against destructive changes to mouse joint architecture (Figure 5). In mice that received 0.8 mg/kg of CX5, 41 (82%) of 50 joints were protected and 9 (18%) of 50 joints were damaged, compared to the 30 (61%) of 49 joints that were protected and 19 (39%) of 49 joints that were damaged in mice treated with isotype control (P = 0.022) (Figure 5A). Daily administration of CX5 also significantly protected joints against damage, with 32 (87%) of 37 joints protected and 5 (13%) of 37 joints damaged (P = 0.009 compared to isotype control) (Figure 5A). H&E-stained joint sections from isotype-treated mice exhibited extensive synovial hyperplasia and damage to joint architecture that was not observed in CX5-treated mice (Figure 5B).

Figure 5.

Joint histology and interleukin-17 (IL-17) profile of mice treated with anti-NKG2D monoclonal antibody. Hind paws from mice in all treatment groups were stained with hematoxylin and eosin. A, Histologic changes, represented as the percentage of joints protected (open bars) or destroyed (solid bars). Data were analyzed by chi-square test for trend. B, Images of proximal phalangeal joints from a mouse treated with isotype control (IgG1) and a mouse treated with 0.8 mg/kg of CX5. e = erosions; sy = synovium; c = cartilage; b = bone; js = joint space. Arrows show erosion (left panel) and cartilage (right panel). Original magnification × 120. C, Viability of paw cells, determined by trypan blue staining. Cells were cultured and the levels of tumor necrosis factor α (TNFα) in supernatants were determined by enzyme-linked immunosorbent assay. Individual symbols represent cells isolated from a single arthritic paw of a treated mouse; horizontal lines show the median; error bars show the interquartile range. ∗ = P = 0.04 versus isotype-treated mice, by Student's unpaired 2-tailed t-test. D, Proportion of IL-17 cells in the blood, draining lymph nodes (LNs), and paws of mice (on day 10 of arthritis) following treatment with 0.8 mg/kg of CX5 (n = 5) or IgG1 control (n = 5). Cells were stained for CD45, CD4, γ/δ T cell receptor (TCR), and intracellular IL-17 after stimulation with phorbol myristate acetate and ionomycin with brefeldin A block. Analyses were performed on CD45+-gated lymphocytes. Bars show the mean ± SEM. ∗ = P < 0.05; ∗∗ = P < 0.01; ∗∗∗ = P < 0.001, by one-way analysis of variance with Bonferroni post hoc test for multiple comparisons.

Dissociated arthritic paws from CX5-treated mice showed a significant reduction in viable cell counts (median [range] 1.71 [0.6–3.12]) compared to paws from isotype-treated mice (2.75 [0.96–12] (P = 0.04) (Figure 5C). No significant differences in TNFα levels in culture supernatants (Figure 5C) or cell lysates (data not shown) were observed between the 2 treatment groups.

Reduction in the frequency of CD4+ IL-17–producing cells in the paws of arthritic mice after anti-NKG2D treatment.

We explored the phenotype and intracellular cytokine expression profile of the CD45+-gated lymphocyte population from blood, draining lymph nodes, and paw cell isolates from CX5-treated mice versus isotype-treated mice. A significant reduction in the frequency of CD4+ cells (53%) and γ/δ T cells (57%) was noted in the paws of CX5 mice versus isotype-treated mice (P < 0.05 for CD4+ cells and P < 0.001 for γ/δ T cells). However, no significant reduction in the frequency of either CD8+ T cells or CD49b+ NK cells was observed (data not shown).

Isotype-treated mice had a significantly higher proportion of CD4+ IL-17+ T cells in the paws (mean ± SEM 18.0 ± 1.5%) compared to blood (0.35 ± 0.10%) or lymph nodes (1.15 ± 0.15%) (P < 0.001) (Figure 5D). Importantly, the frequency of IL-17–producing CD4+ T cells was significantly reduced in the paws of arthritic mice treated with CX5 (by 46%; 9.8 ± 1.6%) (P < 0.05 versus paws of isotype-treated mice) (Figure 5D).

Similarly, a significant increase in the frequency of γ/δ TCR+ IL-17+ T cells was noted in the paws (mean ± SEM 63.6 ± 3.8%) compared to blood (11.8 ± 4.5%) (P < 0.001) and lymph nodes (27.8 ± 3.0%) (P < 0.01) of isotype-treated mice. While the frequency of these cells was reduced by 21% in the paws of CX5-treated mice (50.1 ± 6.6%), the difference failed to reach statistical significance (Figure 5D). We did not record any differences in the profiles of IFNγ-expressing cells in CX5-treated mice versus isotype-treated mice (data not shown).

Inhibition of the cytotoxic effector function of purified splenic NK cells both ex vivo and in vitro following anti-NKG2D treatment.

There was a significant reduction in the ex vivo killing capacity of purified splenic NK cells from CX5-treated mice (27.8 ± 2.0%) versus isotype-treated mice (40.2 ± 4.9%) (P < 0.05) at the effector:target cell ratio of 10:1 (Figure 6A). Cytotoxicity was significantly inhibited (up to 39% at the effector:target cell ratio of 10:1) in in vitro cocultures of purified splenic NK cells from naive mice with target YAC-1 cells preincubated with 0.1–10 μg/ml of CX5 (P < 0.05) (Figure 6B).

Figure 6.

Inhibition of the killing capacity of splenic NK cells by anti-NKG2D monoclonal antibody treatment. A and B, Percent specific lysis in NK cells isolated from the spleens of mice with CIA treated with 0.8 mg/kg CX5 or IgG1 antibody (n = 5 mice per group) (A) and in NK cells isolated from the spleens of naive mice and preincubated with 0.1 μg/ml CX5 (triangles), 1 μg/ml CX5 (diamonds), 10 μg/ml CX5 (circles), or 10 μg/ml IgG1 antibody (squares) (B) and cocultured with YAC-1 target cells. All cocultures were performed in triplicate at the indicated effector:target ratios for 16 hours in the presence of interleukin-2 (50 ng/ml). The lytic activity of NK cells was determined by fluorescence-activated cell sorting (FACS) staining and analysis for caspase 3 expression in target cells. Values in the left panel of A and in B are the mean ± SEM. Symbols in the right panel of A represent individual mice; horizontal bars show the mean. The broken line in B shows the baseline response level. ∗ = P < 0.05 versus IgG1 antibody, by Student's t-test. C, Percentage of NKG2D+ cells and NKG2D MFI in spleen cells from naive mice. Spleen cells were cultured for 20 hours at 37°C (solid bars) or incubated for 30 minutes at 4°C (open bars) in the absence or presence of CX5 (0.01–30 μg/ml) or IgG1 isotype control (30 μg/ml), then stained for CD49b and NKG2D using the MI6 clone. The MFI of NKG2D on CD49b+ NK cells and the percentage of NK cells expressing NKG2D were analyzed by FACS. Data were normalized to untreated control. ∗ ∗ = P < 0.01; ∗∗∗ = P < 0.001 versus IgG1 isotype control, by one-way analysis of variance with Bonferroni post hoc test for multiple comparisons. See Figure 3 for other definitions.

Down-modulation of NKG2D expression by anti-NKG2D antibody.

The percentage of NKG2D+ NK cells and the MFI of the receptor (as detected with an alternative epitope recognizing the anti-NKG2D antibody, MI6) were significantly reduced (P < 0.001) following overnight (but not 30-minute) incubation with CX5 (Figure 6C). No reduction in the percentage of CD49b+ cells was observed (data not shown), indicating that CX5 was not depleting NK cells.


There is emerging evidence that the innate NK-activating receptor NKG2D (16, 31, 32) is involved in autoimmunity (12–15). In this study, we demonstrated the presence of NKG2D ligands in arthritic tissue, and confirmed the previously reported (13) increase in NKG2D+CD4+ T cells in peripheral blood and synovial fluid from RA patients. We also showed that neutralization of NKG2D in RA synovial membrane cell cultures ex vivo significantly inhibited spontaneous TNFα release.

To further explore the role of NKG2D, we used a CIA model. Higher levels of mRNA for Rae-1, the NKG2D ligand, were expressed in arthritic paws, in which immunoreactive NKG2D+ cells were also detected. Interestingly, analysis of cell isolates from the paws of mice with early disease revealed a predominant increase in NKG2D-bearing NK and CD4+ cells versus those from the paws of mice with late disease, heralding the onset of disease, followed by a reduction in MFI levels in the late phase of arthritis. These findings indicate the exciting twin possibilities of receptor engagement and down-modulation as disease progressed, and the presence of ligands at the site of inflammation, as previously described (18). Our results thus highlight the significance of a contributory role of the NKG2D receptor on immunoreactive cells in the pathology of established arthritis.

Blockade of NKG2D with a nondepleting antibody, CX5, in established CIA was efficacious, with amelioration of disease and joint protection. Paws from CX5-treated mice had significantly lower viable cell counts, possibly indicating that cellular trafficking into the site of inflammation was reduced. Furthermore, we observed a significant reduction in the levels of IL-17 in CD4+ cells, and to a lesser extent in γ/δ T cells, in arthritic paws from CX5-treated mice; this is especially relevant since both cell types are known to play key pathogenic roles in CIA (33, 34). We believe that the reduction in the clinical signs of disease could be partially due to reduced IL-17 production in CD4+ Th17 cells and possibly to functional inactivation of NKG2D+CD4+ cells upon treatment. In contrast, IFNγ-producing cells were not affected, indicating that blockade of NKG2D affected pathogenic CD4+ cells of Th17 but not Th1 lineage in CIA.

CX5 treatment did not reduce spontaneous ex vivo TNFα release from paw cells. However, since cellular infiltration was reduced in the paws of treated animals, this could translate into lower systemic levels of the cytokine. This key observation sets up the possibility of trials of synergistic therapy with anti-TNFα and anti-NKG2D in CIA.

Although CX5 treatment had no effect on the levels of IFNγ in NK cells in arthritic paws, ex vivo studies showed that splenic NK cells isolated from CX5-treated mice had a reduced ability to kill target cells. This result was mirrored in vitro, upon addition of CX5 to naive splenic NK cells, indicating other possible mechanisms by which CX5 treatment could work in CIA. Treatment, leading to engagement and down-modulation of NKG2D on NK cells (as was seen with MI6 staining in vitro), results in a decrease in the innate effector-killing ability of NK cells, which directly or indirectly could result in the amelioration of disease in a manner yet unknown and which merits further investigation. It is reasonable to speculate here that reduced cytotoxicity leads to a reduction in proinflammatory mediators, including damage-associated molecular pattern molecules, which signal through TLRs and have recently been shown to be involved in CIA disease pathogenesis (35, 36).

Our study clearly indicates a role of NK cells in inducing arthritis, and a reduction in their effector functions could be a way of dampening their pathogenic role in CIA. Conversely, Lo et al recently observed that depletion of NK cells in CIA exacerbated and enhanced the onset of arthritis (3), and postulated a suppression of pathogenic Th17 cells mediated by IFNγ from NK cells (37). The role of NK cells in RA and CIA as well as other autoimmune diseases is thus not fully understood (38), with evidence that they can both contribute to (8–11) or protect against (39, 40) inflammation.

It is possible that NK cells play different roles during different stages of disease (38). Ligands for both activating and inhibitory receptors can be up-regulated under stress or by viral infections (41). Killing by NK cells is known to occur when cells down-regulate class I MHC due to stress or viral infections, thereby up-regulating stress ligands and becoming targets (1). Ligands are also known to be up-regulated in the presence of a chronic inflammation mediator pool in autoimmune diseases such as RA (40, 42, 43). Thus, the interplay between activating versus inhibitory and killer receptors on NK cells is as yet unexplored and is particularly interesting with regard to inflammatory/autoimmune status. It could hold the key to the role played by NK cells in autoimmune diseases such as RA, and merits further studies in CIA.

We suggest that there is a likely involvement of NKG2D in modulating inflammation in RA through ligand-dependent activation of blood-derived lymphocytes, following chronic inflammatory cytokine stress–induced up-regulation of ligands on synovial cells. This hypothesis is supported by our observation that blockade of NKG2D in the CIA model ameliorated disease. However, the precise mechanism of this significant amelioration has been only partially elucidated through our results; this opens up the possibility of interesting future studies. We noted that NK, CD4+, and γ/δ T cells predominantly expressed the receptor, which was modulated during the course of CIA. Further, our results suggested that cells were targeted at the site of ongoing inflammation (in the paws and not in the draining lymph nodes), since down-regulation of NKG2D was observed on the paw lymphocytes only. Most importantly, in the CIA model used in this study, we established the crucial finding that treatment with CX5 reduced IL-17–expressing CD4+ cells significantly, and reduced γ/δ TCR+ T cells to a lesser extent, thereby influencing the contribution of these pathogenic Th17 cells to the establishment of arthritis, while having no effect on NK cell production of IFNγ, which is known to protect against arthritis.

Taken together, our results suggest a pathogenic role of NKG2D-bearing cells in CIA and highlight the importance of NKG2D in RA. However, it is clear that interactions between NKG2D and its ligands depend on several factors, such as cell specificity, the expression profiles of ligands, and receptor distribution. Targeting NKG2D-mediated interactions in RA is thus a legitimate strategy to consider in designing future therapeutic interventions in arthritis.


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. Sumariwalla 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. Andersson, Sumariwalla, McCann, Amjadi, Chang, McNamee, Tornehave, Haase, Agersø, Stennicke, Ahern, Ursø, Trowsdale, Feldmann, Brennan.

Acquisition of data. Andersson, Sumariwalla, McCann, Amjadi, Chang, McNamee, Tornehave, Haase, Agersø, Stennicke, Ahern, Ursø, Trowsdale, Feldmann, Brennan.

Analysis and interpretation of data. Andersson, Sumariwalla, McCann, Amjadi, Chang, McNamee, Tornehave, Haase, Agersø, Stennicke, Ahern, Ursø, Trowsdale, Feldmann, Brennan.


All authors from Novo Nordisk contributed to the study design, data collection and analysis, and writing of the manuscript. Publication of the article was not contingent upon approval by Novo Nordisk.


We thank Ms Renee Best for processing synovial samples and Ms Pauline Bakker, Ms Marthe Reox, and Dr. Maggie Larche for collection of blood samples. We thank the staff of the Biological Services Unit for the care and maintenance of mice. We thank Mr. David Essex and Mr. Steen Kryger for help with immunohistology and Ms Gena Mellett and Ms Sandra Lock for help in preparing the manuscript.