Adiponectin is a mediator of the inverse association of adiposity with radiographic damage in rheumatoid arthritis




Recent reports have suggested that increasing adiposity may protect against radiographic damage in rheumatoid arthritis (RA). We explored the role of serum adipokines (adiponectin, resistin, and leptin) in mediating this association.


Patients with RA underwent total-body dual x-ray absorptiometry for measurement of total and regional body fat and lean mass, abdominal computed tomography for measurement of visceral fat area, and radiographs of the hands and feet scored according to the modified Sharp/van der Heijde (SHS) method. Serum levels of adipokines were measured and cross-sectional associations with radiographic damage were explored, adjusting for pertinent confounders. The associations of measures of adiposity with radiographic damage were explored with the introduction of adipokines into multivariable modeling as potential mediators.


Among the 197 patients studied, adiponectin demonstrated a strong association with radiographic damage, with the log SHS score increasing by 0.40 units for each log unit increase in adiponectin (P = 0.001) after adjusting for pertinent predictors of radiographic damage. Adiponectin independently accounted for 6.1% of the explainable variability in SHS score, a proportion comparable with rheumatoid factor, and greater than HLA–DRB1 shared epitope alleles or C-reactive protein levels. Resistin and leptin were not associated with radiographic damage in adjusted models. An inverse association between visceral fat area and radiographic damage was attenuated when adiponectin was modeled as a mediator. The association of adiponectin with radiographic damage was stronger in patients with longer disease duration.


Adiponectin may represent a mechanistic link between low adiposity and increased radiographic damage in RA. Adiponectin modulation may represent a novel strategy for attenuating articular damage.


Rheumatoid arthritis (RA) is a chronic systemic autoimmune disorder characterized by synovitis and progressive damage to articular cartilage and subchondral bone in a majority of affected individuals (1). Accumulated joint damage leads to deformity and contributes to dysfunction and disability (2), which in turn are major contributors to reduced quality of life. Several disease-related factors are useful for predicting those with more rapid progression of erosive disease. Among these, early erosions, ongoing synovial and systemic inflammation, and the presence of rheumatoid factor (RF), anti–cyclic citrullinated peptide (anti-CCP) antibodies, and HLA–DRB1 shared epitope alleles are well established (3, 4). However, even when these factors are considered there remains considerable variability in radiographic progression, suggesting that other unaccounted factors may contribute to articular damage. In this regard, an association of low body mass index (BMI) with an increased risk of progression of joint damage in RA has been reported (5–7), with the risk of radiographic progression declining as BMI increased in 2 studies (6, 7). Although rheumatoid cachexia, a surrogate for highly active RA, could potentially explain the association at the lowest end of the BMI spectrum, it fails to explain why the trend in protection against radiographic progression continues to increase with increasing adiposity, even into the overweight and obese categories.

Adipose tissue expresses proteins with hormonal properties (i.e., adipokines) that participate in energy homeostasis and metabolism and, more recently, have been shown to participate in inflammatory and immune regulation. Levels of one adipokine, resistin, were elevated in the serum and synovial fluid of patients with RA and were shown to induce de novo arthritis when injected into mice (8). Similarly, serum levels of the adipokine leptin were elevated in patients with RA (9) and were shown to induce interleukin-8 (IL-8) production in RA synoviocytes via the JAK/STAT pathway (10). Another adipokine, adiponectin, which shares sequence homology with complement C1q and tumor necrosis factor α (TNFα) (11), has potent antiinflammatory activity in vascular endothelium (12), and has been shown to be protective against atherosclerosis (13, 14). However, levels of this same adipokine are elevated in the serum (15) and synovial fluid (16) of patients with RA and function to increase the production of IL-6 by synovial fibroblasts via the NF-κB signaling pathway, suggesting proinflammatory activity in the joint (17). In contrast to an increase in circulating levels of resistin and leptin with increasing adiposity, adiponectin levels decrease as fat mass increases. This observation, coupled with its proinflammatory properties in the joint, make adiponectin the most attractive candidate among the adipokines for mediating, in whole or part, the association of decreasing BMI with increasing radiographic progression.

In this study, we explored the associations of serum adipokine concentrations with radiographic damage in a cross-sectional analysis of baseline data from an ongoing prospective observational study of patients with RA, controlling for other pertinent predictors of radiographic damage. We hypothesized that adiponectin would be a mediator of the association between adiposity and radiographic damage.


Study subjects.

Study subjects were participants in the Evaluation of Subclinical Cardiovascular Disease and Predictors of Events in RA study, a cohort study investigating the prevalence, progression, and risk factors for subclinical cardiovascular disease in RA. All subjects met the American College of Rheumatology (formerly the American Rheumatism Association) 1987 classification criteria for RA (18), were ages 45–84 years, and did not report any prior prespecified cardiovascular events or procedures, as previously described (19). Subjects weighing >300 pounds were excluded due to limitations of the imaging equipment used in the study.

The study was approved by the Institutional Review Board of the Johns Hopkins Hospital, with all subjects providing written informed consent prior to enrollment. Enrollment began in October 2004 and concluded in May 2006.

Assessments: outcome and independent variables.

Single-view, anteroposterior radiographs of the hands and feet were obtained and scored by a single trained radiologist blinded to patient characteristics (WMS) using the van der Heijde modification of the Sharp (SHS) method (20). The SHS score is a quantitative assessment of the presence and degree of marginal erosion and joint space narrowing (JSN) for selected joints in the hands, wrists, and feet. The SHS score ranges from 0–398 and is the sum of the erosion score (0–230) and the JSN score (0–168). Five subjects had incomplete radiographic assessments (3 had hand but not feet radiographs, and 2 had feet but not hand radiographs). For these subjects, the missing score (hand or foot) was imputed from the available hands and feet based on a regression equation relating the predicted hand or foot erosion and JSN scores derived from the remaining subjects in the cohort.

Body composition assessments.

Subjects underwent total-body dual x-ray absorptiometry (DXA) scanning on a Lunar Prodigy DXA scanner (GE/Lunar Radiation, Madison, WI). Prodigy software, version 05.60.003 (GE/Lunar Radiation) was utilized to analyze and measure fat, lean, and bone mass for the total body (minus the head) and per body region (arms, legs, and trunk). All subjects were scanned on the same DXA scanner. Quality control and calibration procedures were performed daily using standard procedures provided by the manufacturer.

A single cross-sectional image of the abdomen was obtained using computed tomography (CT) at the level of the interspace between the fourth and fifth lumbar vertebrae on a Toshiba Aquilon scanner (Toshiba America, Tustin, CA). A single trained reader (MA), blinded to the clinical characteristics of the participants, quantified the visceral fat area for all of the scans using the National Institutes on Aging Musculoskeletal Analysis Program. Briefly, the reader manually traces the borders of the subcutaneous and visceral abdominal cavities on the CT scan. Based on the attenuation of a selected area of fat from the subcutaneous compartment, tissue with similar fat attenuation in the visceral cavity is identified and quantified in square centimeters.

Measures of height were obtained using a wall-mounted stadiometer. Weight was measured with a Detecto platform (Webb City, MO) with subjects wearing light indoor clothing and no shoes. BMI was calculated as body weight (kg) divided by height (meters squared).

Demographic and lifestyle characteristics.

Age, sex, race, and highest level of education were assessed by self-report. Physical activity was assessed using the 7-Day Physical Activity Recall questionnaire (21). Current smokers were those reporting current smoking with >100 lifetime cigarettes smoked. Diabetes mellitus was defined as a fasting serum glucose ≥126 mg/dl or the use of diabetes medication.

RA disease characteristics.

RA disease duration was assessed by self-report from the date of diagnosis. Joint counts were performed by a single trained assessor. RA disease activity was calculated using the Disease Activity Score in 28 joints with C-reactive protein (CRP) level (22). The 21-item Stanford Health Assessment Questionnaire (23) was used to assess self-reported disability. Current and past use of glucocorticoids and biologic and nonbiologic disease-modifying antirheumatic drugs (DMARDs) was queried by detailed examiner-administered questionnaires.

Laboratory assessments.

Fasting serum and plasma samples were collected on the day of body composition analysis and stored at −70°C. Serum adipokines (adiponectin, resistin, and leptin) were measured by enzyme-linked immunosorbent assay (ELISA) at the Laboratory for Clinical Biochemistry Research (University of Vermont, Burlington, VT). CRP level was measured by nephelometry (Dade Behring, Deerfield, IL). RF was assessed by ELISA, with seropositivity defined at or above a level of 40 units. Anti-CCP antibody was assessed by ELISA, with seropositivity defined at or above a level of 60 units. Peptidyl argenine deiminase 4 (PAD4) antibodies were measured by immunoprecipitation using 35S methionine-labeled in vitro transcribed/translated products, as previously described (24). Autoradiograms were scanned and assigned a semiquantitative score based on densitometry, with 3+ immunoprecipitation defined as the level of seropositivity.

Genomic DNA was extracted from peripheral blood leukocytes and exon 2 of HLA–DRB1 sequenced for shared epitope alleles and identified according to accepted classification criteria (25).

Statistical analysis.

The distributions of all variables were examined. Means and SDs were calculated for all normally distributed continuous variables, and medians and interquartile ranges (IQRs) were calculated for continuous variables that were not normally distributed. For categorical variables, counts and percentages were calculated. Logarithmic transformation of variables was used for highly skewed variables (i.e., adipokine concentrations, visceral fat area, SHS scores, and CRP levels) when needed to satisfy the requirements of regression modeling.

The associations of sociodemographic, body composition, and RA disease and treatment variables with adipokine concentrations were explored using linear regression, first in simple (unadjusted) models and then in multivariate models retaining variables with associations in univariate modeling with significance at the P = 0.20 level or below. Simpler models were constructed by comparing Akaike's information criterion (AIC) values for complex versus simple models. The Shapiro-Wilk test was used to examine the normality of the modeled outcome variables across the extent of the independent variables. Variance inflation factors were calculated to ensure that variables with excessive collinearity were not co-modeled. Next, the associations of sociodemographic, body composition, RA disease and treatment variables, and adipokines with SHS scores were explored using the strategy outlined above. For the final model, the independent contributions of included variables to the total explainable variability of the model were calculated by dividing the change in the adjusted coefficient of determination (R2) for each variable when excluded from the full model by the adjusted R2 for the full model. Potential heterogeneities in the associations of adipokine concentrations with SHS scores across strata of patient characteristics were modeled in statistical interaction models and tested using analysis of covariance.

Statistical calculations were performed using Intercooled Stata 9 (StataCorp, College Station, TX). In all tests, a 2-tailed α of 0.05 was defined as the level of statistical significance.


Characteristics of the 197 study patients are summarized in Table 1. Whereas all 197 patients underwent anthropometric assessment, radiographs, and adipokine measurements, a subgroup of 131 patients (66.3%) underwent abdominal CT scanning for measurement of the visceral fat area. Patient characteristics for the subgroup with visceral fat measurements were not different from the entire cohort (Table 1). On average, the cohort was older (mean age 59 years), with a median RA disease duration of 9 years. Almost all patients (99%) had evidence of joint damage (total modified SHS score >0) on radiographs. The median modified total SHS score was 44 units; however, there was broad variability in the extent of radiographic damage among patients (absolute range 0–364 units; IQR 16–121 units).

Table 1. Characteristics of patients with RA*
CharacteristicAll patients (n = 197)Visceral fat subgroup (n = 131)
  • *

    Values are the median (interquartile range) unless otherwise indicated. RA = rheumatoid arthritis; kMET = kilo metabolic equivalents; BMI = body mass index; n/a = non-applicable; RF = rheumatoid factor; anti-CCP = anti–cyclic citrullinated peptide; PAD4 = peptidyl argenine deiminase 4; DAS28-CRP = Disease Activity Score in 28 joints with C-reactive protein level; HAQ = Health Assessment Questionnaire; SHS = modified Sharp/van der Heijde score; DMARDs = disease-modifying antirheumatic drugs.

Age, mean ± SD years59.4 ± 8.760.5 ± 8.9
Male sex, no. (%)79 (40.1)51 (38.9)
Caucasian, no. (%)169 (85.8)113 (86.3)
Any college, no. (%)148 (75.5)99 (75.6)
Current smoking, no. (%)23 (11.7)12 (9.2)
Diabetes, no. (%)12 (6.1)10 (7.6)
Exercise, kMET minutes/week0.91 (0–2.24)0.93 (0–2.55)
BMI, mean ± SD kg/m228.4 ± 5.328.5 ± 5.5
Total fat mass, mean ± SD kg29.9 ± 10.729.5 ± 10.8
Total lean mass, mean ± SD kg46.4 ± 11.546.3 ± 11.7
Truncal fat, mean ± SD kg16.2 ± 6.216.0 ± 6.1
Visceral fat area, mean ± SD cm2n/a119 ± 72
Adiponectin, mg/liter30.5 (19.1–40.3)32.0 (18.9–40.5)
Resistin, ng/ml16.0 (12.2–21.6)16.2 (12.3–22.4)
Leptin, ng/ml13.7 (6.5–26.4)14.4 (6.5–25.5)
RA duration, years9 (4–17)9 (5–17)
RF seropositivity, no. (%)129 (65.5)90 (68.7)
Anti-CCP seropositivity, no. (%)139 (70.9)91 (70.0)
Any shared epitope alleles present, no. (%)136 (69.7)92 (70.8)
PAD4 antibodies, no. (%)35 (17.9)23 (17.7)
DAS28-CRP, units3.6 (2.9–4.4)3.6 (2.8–4.5)
Swollen joints (0–42)7 (3–10)6 (3–10)
Tender joints (0–44)6 (2–13)6 (2–12)
CRP level, mg/liter2.6 (1.1–7.2)3.0 (1.1–7.8)
HAQ, units0.63 (0.13–1.25)0.63 (0.13–1.25)
Any radiographic damage, no. (%)195 (99.0)129 (98.5)
Total modified SHS score44 (16–121)47 (20–114)
Current prednisone, no. (%)76 (38.6)49 (37.4)
Cumulative prednisone, grams3.1 (0–9.1)2.8 (0–8.7)
Current nonbiologic DMARDs, no. (%)165 (84.2)110 (84.0)
Current methotrexate, no. (%)125 (63.4)80 (61.1)
Current biologic DMARDs, no. (%)89 (45.4)54 (41.2)
 Etanercept36 (18.4)22 (16.8)
 Infliximab22 (11.2)12 (9.2)
 Adalimumab27 (13.8)18 (13.7)
 Rituximab4 (2.0)2 (1.5)

Cross-sectional predictors of radiographic damage in patients with RA.

We considered patient characteristics, including serum adipokines, in a prediction model of the logarithmically transformed total modified SHS score (Table 2). Significant univariate predictors included the log of the serum adiponectin and resistin concentrations, age, any college education, body lean mass, RA duration, RF seropositivity, presence of any shared epitope alleles, PAD4 antibodies, swollen joint count, log CRP level, current and cumulative prednisone received, and current biologic DMARDs received (Table 2, univariate model). Leptin was not associated with SHS scores in either univariate or multivariate modeling. Simultaneous inclusion of univariate predictors into a multivariate model resulted in a final parsimonious model with 8 significant predictors: log adiponectin, age, RA duration, RF seropositivity, any shared epitope alleles, 3+ PAD4 antibodies, log CRP level, and current biologic DMARDs received (Table 2, multivariate model 2). Within this model, the log modified SHS score increased by 0.40 units for each increase in log adiponectin (P = 0.001). Resistin was no longer significantly associated with SHS scores in the final adjusted model. Together, these predictors accounted for 50% of the variability in total modified SHS score (i.e., the adjusted R2 was 0.501 for the final adjusted model).

Table 2. Crude and adjusted associations of patient characteristics with log total modified SHS score*
CharacteristicUnivariateR2Multivariate (complex)Multivariate (simplified)
  • *

    Potential predictors of radiographic damage scores tested but not found to contribute to model fit in either univariate or multivariate models included sex, race, smoking, diabetes mellitus, habitual exercise, BMI, total or truncal fat mass, and current nonbiologic DMARDs received. R2 = coefficient of determination. See Table 1 for additional definitions.

  • Includes only the covariate of interest.

  • Includes all the covariates with β coefficients in the column. Akaike's information criterion for the complex multivariate model minus the simpler multivariate model was 542.0 − 534.4 = 7.6 (P = 0.40), indicating that the fit of the simpler model was not significantly different from that of the more complex model.

  • §

    For multivariable modeling, only the swollen and tender joint count and CRP level components of the DAS28-CRP were considered.

log adiponectin, mg/liter per unit0.618< 0.0010.0700.3920.0040.4010.001
log resistin, ng/ml per unit1.074< 0.0010.0940.0780.70  
log leptin, ng/ml per unit0.1210.23     
Age, per year0.047< 0.0010.0850.0220.0190.0200.019
Any college−0.4700.0380.017−0.1520.39  
Total lean mass, per kg−0.0210.0150.0250.0010.93  
RA duration, per year0.073< 0.0010.3140.051< 0.0010.055< 0.001
RF seropositivity0.775< 0.0010.0680.6480.0010.5390.001
Anti-CCP antibody seropositivity0.3740.0870.010−0.3300.090  
Any shared epitope alleles present0.6390.0030.0410.3790.0250.3820.017
PAD4 antibodies1.254< 0.0010.1180.6040.0040.5400.006
Swollen joints, per joint0.0350.0720.0120.0200.174  
Tender joints, per joint0.0070.50     
log CRP level, per log mg/liter0.264< 0.0010.0630.1560.0100.1540.007
Current prednisone0.5490.0060.0340.0340.85  
Cumulative prednisone0.0330.0010.0490.0100.25  
Current biologic DMARDs0.3540.0730.0120.3980.0090.4570.002
R2   0.497 0.501 

Within the context of this adjusted model, the greatest contributor to explainable variability was RA duration, which independently accounted for 68% of variability (Table 3). Outside of RA duration, adiponectin independently accounted for 6.1% of the explainable variability of the model, a proportion similar in magnitude to that of RF (6.3%) and biologic agents currently received (6.8%), and greater than that of age (2.6%), presence of any shared epitope alleles (2.0%), PAD4 antibodies (4.1%), or CRP level (3.8%). Based on the AIC, inclusion of adiponectin into the model significantly improved model fit (P = 0.0008).

Table 3. Contributions to the explainable variability and model fit for predictors of log total modified SHS score*
PredictorTotal association, %Independent association, %AIC§DifferenceP
  • *

    See Table 1 for definitions.

  • Proportion of the total explainable variability (coefficient of determination: R2) contributed by the predictor in the univariate model expressed as a proportion of the total variability explained by all the predictors from the final model (R2 = 0.501).

  • Proportion of the total explainable variability contributed by the predictor in the final adjusted model. The sum of the total associations is >100% due to overlapping associations of the predictors in the unadjusted models. The sum of the independent associations = 100%.

  • §

    The change in Akaike's information criterion (AIC) for the nested model that includes all of the predictors in the table, excluding the indicated covariate. AIC for the full model = 537.36. Higher AIC values for the model when the predictor is excluded indicate better fit for the model that includes the predictor.

  • In the AIC between the simpler model, excluding the predictor of interest, and the full model, including the predictor of interest. Higher values indicate a greater independent contribution to the full model.

RA duration62.368.0589.3752.01< 0.0001
RF seropositivity13.66.3548.5211.160.0004
Presence of any shared epitope alleles8.22545.127.760.014
PAD4 antibodies23.64.1545.818.450.005
log CRP level12.63.8543.366.000.006
Biologic DMARDs2.46.8552.0014.640.0013

Higher adjusted SHS scores for both erosions and JSN were observed with increasing adiponectin concentration (Figure 1). Trends for the incremental increase in the log erosion and log JSN scores per log unit increase in adiponectin were similar: 0.373 (P = 0.003) and 0.346 (P = 0.014), respectively.

Figure 1.

Adjusted mean modified Sharp/van der Heijde scores according to quartile of serum adiponectin. Markers reflect adjusted means; ranges indicate 95% confidence intervals for the estimates of the mean. Analyses adjusted for age, disease duration, rheumatoid factor seropositivity, presence of HLA–DRBI shared epitope alleles, peptidyl argenine deiminase 4 antibodies, log C-reactive protein level, and biologic disease-modifying antirheumatic drugs received.

Association of adiposity with radiographic damage

We next explored the ability of adiponectin to cross-sectionally mediate the association of adiposity with radiographic damage (Table 4). BMI, total fat, and truncal fat (measured with DXA) were not associated with SHS scores in univariate or multivariate modeling. After adjusting for confounders of radiographic damage (age, RA duration, RF seropositivity, any shared epitope alleles, 3+ PAD4 antibodies, log CRP level, and current biologic DMARDs received), a strong inverse association between log visceral fat area and log SHS score was observed, with the log SHS score decreasing by −0.354 units for each unit increase in log visceral fat area (P = 0.008). As expected, serum adiponectin concentration decreased as visceral fat area increased (data not shown). Introducing log adiponectin into the model as a possible mediator attenuated the magnitude of the association of visceral fat with SHS score by almost half, diminishing its level of significance (P = 0.17) (Table 4, model 3). In this final adjusted model, the log SHS score increased by 0.409 units for each unit increase in log adiponectin (P = 0.017).

Table 4. Crude and adjusted associations of visceral fat area with log total modified SHS score*
CharacteristicModel 1Model 2§Model 3
  • *

    See Table 1 for definitions.

  • Analyses include only those patients in the subgroup that underwent abdominal computed tomography scanning for visceral fat assessment (n = 131).

  • Unadjusted. Includes log visceral fat area as the only covariate.

  • §

    Includes adjustment for age, RA disease duration, RF seropositivity, the presence of any shared epitope alleles, 3+ PAD4 antibodies, biologic DMARDs currently received, and log CRP level.

  • Includes adjustment for all model 2 confounders with log adiponectin included as a possible mediator of the association between log visceral fat area and log total SHS score.

log visceral fat area, per log unit−0.2760.089−0.3540.008−0.1980.17
log adiponectin, per mg/liter    0.4090.017

Modification of the association of adiponectin with radiographic damage by level of RA characteristics

Finally, we explored potential differences in the association of log adiponectin with log SHS score according to strata of other RA characteristics (Table 5). There was no association between log adiponectin and log SHS score noted in RA subjects with early disease (disease duration <2 years, n = 16; P = 0.45) in contrast to the strong positive association in patients with longer standing disease (P < 0.001). This heterogeneity by strata of RA duration was statistically significant (P = 0.034). Otherwise, the association of log adiponectin with log SHS score was not significantly modified by sex, age group (defined by a median split at 60 years), presence of diabetes mellitus, RF, shared epitope alleles, anti-CCP antibody, PAD4 antibody status, biologic DMARDs received, or current glucocorticoid use.

Table 5. Adjusted associations of log adiponectin with log SHS score according to the strata of select characteristics*
Potential effect modifierModifier presentModifier absentP
  • *

    Adjusted for age, RA duration, RF seropositivity, the presence of any shared epitope alleles, 3+ PAD4 expression, biologic DMARDs currently received, and log CRP level. See Table 1 for definitions.

  • For the interaction of the modifier on the association of log adiponectin with the log modified SHS score, P > 0.05 indicates that the association of log adiponectin with log SHS score does not statistically differ according to the 2 strata of the modifier.

  • Disease duration <2 years (n = 16).

Male sex0.4690.0080.3110.110.54
Age ≥ the 60-year median0.5110.0070.3050.0590.41
Diabetes mellitus1.0650.0770.3730.0030.26
Early RA−0.2860.480.618< 0.0010.034
RF present0.2730.0740.6040.0020.17
Anti-CCP present0.3180.0280.5420.0100.37
Any shared epitope alleles0.4420.0020.2770.240.54
Current biologic agent use0.5700.0030.2910.0610.25
Current prednisone use0.340.0640.440.0060.66


In this study exploring the associations of adipokines with radiographic damage in RA, we observed a strong cross-sectional association between increasing serum adiponectin concentration and both radiographic erosions and JSN, even after accounting for recognized predictors of radiographic damage. Significant associations between other adipokines (resistin and leptin) and radiographic damage were not observed. Adiponectin accounted for the variability in radiographic damage scores to an extent equal to that of RF, and greater than that of other well-recognized predictors of radiographic damage (such as shared epitope alleles and CRP level), and was shown to be a mediator of the inverse association of visceral fat area with radiographic damage. A recent study (26) identified an association between adiponectin levels and radiographic damage. Our investigation extends this observation by estimating the contribution of adiponectin levels to radiographic damage in relation to other recognized predictors and by exploring the potential for adiponectin to mediate the association of adiposity with joint damage.

Several prior investigations have consistently demonstrated that adiponectin is elevated in the serum (15) and synovial fluid (16) of patients with RA. This finding has been somewhat puzzling because adiponectin expression by adipocytes is antagonized by inflammatory cytokines that are typically elevated in RA (i.e., TNFα and IL-6) (27), and that adiposity has been shown to be increased in patients with RA (28). Under these circumstances, adiponectin would be expected to be reduced, rather than increased, in patients with RA.

What are the potential implications of higher adiponectin levels for patients with RA? Although adiponectin has been shown to exert potent antiinflammatory effects on the vasculature and in fat and muscle tissue, antagonizing atherogenesis and improving insulin sensitivity (13), its activity in the joint may be proinflammatory, as suggested by several recent in vitro studies. Tang et al (17) demonstrated an increase in IL-6 production in cultured synovial fibroblasts from RA and osteoarthritis (OA) patients when stimulated with increasing concentrations of adiponectin. Upregulation of IL-6 was induced by adiponectin stimulated signaling through the NF-κB pathway. In another study, Luo et al (29) demonstrated that stimulation with adiponectin induced RANKL and inhibited osteoprotegerin in cultured human osteoblasts, leading to increased osteoclast formation. Taken together, these studies provide circumstantial evidence that adiponectin may have proinflammatory activity in the joint, inducing the terminal mechanisms involved in erosive joint damage. Indeed, the effect may not be specific to RA, as adiponectin levels were higher in patients with erosive versus nonerosive OA of the hands in a recent report (30).

Three prior studies have identified an association between increasing BMI (a surrogate for adiposity) and lower rates of radiographic progression in patients with RA (5–7). This association is seemingly paradoxical, as adipose tissue is a potent source of cytokines (31) and was associated with increased systemic inflammation in patients with RA (32). Thus, one might hypothesize that increasing adiposity would result in higher, rather than lower, rates of radiographic progression. However, a key feature of adiponectin physiology is that circulating levels diminish as adiposity increases, with highest levels noted in individuals with the lowest fat mass (33). In this light, considering that adiponectin may have detrimental effects on the joint, adiponectin becomes an excellent candidate to mediate the inverse relationship between increasing adiposity and radiographic damage observed in prior RA studies. Two of the studies identified a protective effect of obesity only in patients seropositive for RF (6) or anti-CCP antibodies (7), a heterogeneity not detected in our study. Differences in study design (prospective design and enrolling only early patients with RA in the prior studies, and cross-sectional design with early and established patients in our study) may account for the discrepancy in findings.

Adiponectin expression is suppressed to the greatest extent in visceral fat (fat located within the visceral abdominal cavity) under a pathophysiologic mechanism that is not well defined (34). As expected, we observed in our patients with RA that adiponectin levels were the highest in the patients with the lowest visceral fat area. These patients also demonstrated higher radiographic damage scores, even after accounting for multiple pertinent confounders. When adiponectin and visceral fat were comodeled, the association of increasing visceral fat area with radiographic damage was subsumed in part by adiponectin, suggesting that adiponectin functions as an intermediate in the pathway between visceral fat and radiographic damage (35). These mediating effects of adiponectin provide further clues to demystifying the apparent paradoxical association between increasing BMI and protection from radiographic progression noted in RA (6, 7). We did not, however, observe significant associations between other measures of adiposity (e.g., BMI, total fat by DXA) and radiographic damage. However, adiponectin tends to be more strongly correlated with visceral fat than BMI or DXA measures of fat. Longitudinal assessments of radiographic damage are underway to explore the ability of measures of adiposity and adipokines to predict radiographic change scores over time.

The origin of the adiponectin implicated in the joint damage observed in our study remains unclear. Adipocytes are abundant in the joint (36) and may express adiponectin that acts locally in a paracrine fashion. Another possibility is that circulating adiponectin produced by remote adipocytes acts in the joint as an endocrine hormone. A prior study in patients with RA showed that serum adiponectin concentration was higher than synovial fluid concentration (16), with significant correlation noted in adiponectin levels between the 2 sources. This finding suggests that the source of articular adiponectin may be peripheral adipose rather than a local source. Inhibition of adiponectin could represent a potential therapeutic target in RA, and identifying its source may have relevant therapeutic implications. Given its antiinflammatory effects on the vasculature and on metabolic pathways, systemic adiponectin inhibition may have undesirable adverse cardiovascular consequences, whereas intraarticular delivery of an adiponectin inhibitor could theoretically limit its proinflammatory effects on the joint.

We also explored whether the association of adiponectin with radiographic damage was modified by levels of RA characteristics. Our finding that adiponectin was more strongly associated with radiographic damage in patients with established disease compared with those with early disease could indicate that there is a physiologic difference in the pathologic activity of adiponectin at different stages of RA. The difference could, however, reflect the relative insensitivity of radiographs in early disease, in which a lag in the emergence of erosions is often noted (37). A more sensitive imaging technique, such as magnetic resonance imaging, could assess whether there is a true interaction of RA duration on the association of adiponectin with radiographic damage. We did not detect a statistical difference in the association of adiponectin with radiographic damage between men and women. However, because adiponectin tends to be higher in women (38), this could make women at greater risk for adiponectin-associated radiographic damage. Importantly, we did not see differences in the association of adiponectin with radiographic damage according to autoantibody or shared epitope status. Adiponectin was as strongly associated with radiographic damage in patients receiving biologic DMARDs as those who were not receiving biologic agents, suggesting that cytokine inhibition may be not be a successful way of limiting potential adiponectin effects on the joint. The current literature is divergent on the effect of biologic agents on circulating adiponectin, with studies demonstrating increased levels (39–41), no change in levels (42), and even decreased levels (43) after administration of TNF inhibitors. Thus, further study is required to appreciate the role of treatment on the association of adiponectin with articular damage.

Some notable limitations to our study should be acknowledged. The analyses were cross-sectional, and strong claims on causality cannot be made. Additionally, because our cohort was not followed from disease onset, disentangling the effects of other radiographic predictors and treatment effects over time from the association of adiponectin with joint damage is difficult. We did not measure synovial fluid adiponectin levels, nor did we quantify articular fat mass. Knowing these factors proximal to the outcome of radiographic damage could validate and strengthen the associations observed. Finally, we excluded participants exceeding 300 pounds due to limitations of the imaging equipment used. Because these patients would be expected to have higher visceral fat and lower adiponectin, inclusion of this group could have provided additional confirmation of trends in the observed associations.

In conclusion, we identified a robust cross-sectional association between serum adiponectin levels and radiographic damage in patients with RA, suggesting that this adipokine may be a pathogenic mediator of the seemingly paradoxical relationship between increasing adiposity and protection from radiographic damage in RA. These findings suggest that targeting adiponectin may be beneficial in preventing articular damage in patients with RA. However, recognizing the beneficial effects of adiponectin on cardiovascular risk, any effect on radiographic progression of systemically reducing adiponectin could be accompanied by an increase in cardiovascular disease. Conversely, attempting to improve the elevated cardiovascular risk of patients with RA by reducing visceral fat (with diet and exercise, for example) and increasing adiponectin levels could serve to undesirably accelerate joint damage.


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 submitted for publication. Dr. Giles 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. Giles, Bathon.

Acquisition of data. Giles, Allison, Scott, Bathon.

Analysis and interpretation of data. Giles, Allison, Bingham, Scott, Bathon.


We would like to thank the Johns Hopkins Bayview Medical Center General Clinical Research Center and staff for providing support for the DXA scanning used in this study. We are indebted to the dedication and hard work of the Evaluation of Subclinical Cardiovascular Disease and Predictors of Events in Rheumatoid Arthritis staff: Marilyn Towns, Michelle Jones, Patricia Jones, Marissa Hildebrandt, and Shawn Franckowiak. Drs. Uzma Haque, Clifton Bingham III, Carol Ziminski, Jill Ratain, Ira Fine, Joyce Kopicky-Burd, David McGinnis, Andrea Marx, Howard Hauptman, Achini Perera, Peter Holt, Alan Matsumoto, Megan Clowse, Gordon Lam, and others generously recommended their patients for this study.