Circulating Fibroblast Growth Factor-21 Levels in Rheumatoid Arthritis: Associations With Disease Characteristics, Body Composition, and Physical Functioning

MATERIALS AND METHODS

Study setting. The data collection methods for this study have been previously described²³. Briefly, RA subjects aged 18–70 years, with established disease and meeting 2010 American College of Rheumatology classification criteria, were recruited from the University of Pennsylvania and Philadelphia Veterans Affairs Medical Center (VAMC) rheumatology practices²⁴. Subjects with juvenile idiopathic arthritis (or another inflammatory arthritis), active cancer, a history of chronic diseases known to affect bone health, pregnancy, or who were unable to perform the muscle density or body composition assessments were excluded. The original pilot study was expanded to include a follow-up visit occurring 2–3 years later for most participants; 84 patients completed the study. The data collection procedures described below were conducted at both the initial visit and follow-up. The protocols were approved by the institutional review board (IRB) at University of Pennsylvania and the Philadelphia VAMC (approval number 01427 from the Corporal Michael J. Crescenz VAMC IRB), and informed consent was obtained from all participants.

Assessment of anthropometrics, race, and medical history. Weight and height were measured in light clothing and with shoes removed using a digital scale (Scaltronix) and stadiometer (Holtain Ltd.), respectively. BMI was calculated (kg/m²). Participants self-reported race (according to National Institute of Health categories), smoking status, and comorbidities (cardiac disease and diabetes). Comorbidities were confirmed in the medical record.

Whole-body dual-energy X-ray absorptiometry. Subjects underwent whole-body dual-energy X-ray absorptiometry (DXA) assessment using a Hologic densitometer (Delphi/Discovery Systems, Hologic, Inc.) to measure appendicular lean mass, total fat mass, and visceral adipose tissue area (VAT). Similar to the adjustment of weight for height to estimate BMI, body composition estimates were adjusted for height² to generate appendicular lean mass index (ALMI; kg/m²) and fat mass index (FMI; kg/m²). Measurement of VAT by DXA has been previously validated²⁵.

Peripheral quantitative computed tomography. Muscle density and muscle area measurements in the left lower leg were obtained by peripheral quantitative computed tomography (pQCT; Stratec XCT2000 12-detector unit, Orthometrix, Inc.) with a voxel size of 0.4 mm, slice thickness of 2.3 mm, and scan speed of 25 mm/s. All scans were analyzed with Stratec software version 6.00. Calf muscle and subcutaneous fat cross-sectional area (mm²) were assessed 66% proximal to the distal physis using threshold 40 mg/cm³ for fat/lean separation and 711 mg/cm³ for lean/bone separation. The pQCT measure of muscle density (mg/cm³) was used as a composite index of intra- and extramyocellular fat content, as previously described^26,27. Edge-detection and threshold techniques were used to separate tissues (fat, muscle, and bone) based on attenuation characteristics that are directly related to tissue composition and density^28,29. Images were filtered prior to being analyzed, using contour mode 3 (−101 mg/cm³) to find skin, and peel mode 2 (40 mg/cm³) to separate adipose and muscle/bone. Images were filtered subsequently with a combination 3 × 3, and double 5 × 5 kernel image filter that clearly defined the edge of the muscle using contour mode 31 (40 mg/cm³). All bone was identified using a threshold of 150 mg/cm³ and mathematically removed to generate results for muscle density. In our laboratory, the coefficient of variation (CV) for short-term precision has ranged from 0.5% to 1.6% for pQCT outcomes.

Dynamometric measurement of muscle strength. Muscle strength was assessed in several ways using the Biodex Multi-Joint System 3 Pro Dynamometer (Biodex Medical Systems Inc.). Peak isokinetic torque (ft-lbs) was measured in triplicate at the knee and lower leg (ankle). For lower leg (ankle) Biodex, we reported strength as peak isometric torque (ft-lbs) in dorsiflexion (with the foot placed in 20º plantarflexion) as previously described³⁰. Peak isometric torque in flexion and extension at the knee was also reported (ft-lbs). Hand-grip strength (kg) was measured using a hand-grip dynamometer (Takei Scientific Instruments Co., Ltd.). A clinically important decrease in hand grip strength has been previously defined as 6.5 kg, and a clinically meaningful change in leg extensor power in older adults has been defined as 9–10%^31,32.

Assessments of physical function and disability. Disability was assessed using the Health Assessment Questionnaire (HAQ), a widely used tool in RA. Briefly, 8 categories are assessed, including dressing and grooming, arising, eating, walking, hygiene, reach, and grip³³. The survey also includes questions about work status, symptoms, and overall health status. Physical function was assessed using the Short Physical Performance Battery (SPPB), a simple test to measure lower extremity function using tasks that mimic daily activities. It examines static balance, gait speed, and timed chair rises³⁴. SPPB testing was initiated later and was therefore measured in a smaller sample of participants (n = 69). Clinically important changes in HAQ and SPPB have been previously defined^35,36. Based on these prior data, this study defined an important worsening of HAQ as an increase of 0.2 and an important worsening of SPPB as a decrease of 1.

Disease measures, inflammatory markers, and medical history. Medication use was determined by self-report and confirmed in the medical record. Erythrocyte sedimentation rate (ESR) was performed using the Westergren method. C-reactive protein (CRP) levels were measured using a Fixed Point Immuno Rate Assay. Creatinine was measured in the clinical laboratory, and estimated glomerular filtration rate (eGFR) was determined using the Modification of Diet in Renal Disease Study equation. Disease activity was quantified using the Disease Activity Score in 28 joints based on CRP (DAS28-CRP) and the modified DAS28 (M-DAS28)³⁷. The M-DAS28 is a composite measure that includes the CRP, swollen joint count, and evaluator global score, and has been validated to correlate strongly with synovitis on magnetic resonance imaging (MRI) and radiographic damage progression. It was included to avoid bias related to incorporation of the patient global score in the DAS28-CRP, which is closely correlated with physical functioning scores (HAQ). Commercially available ELISAs (R&D Systems) were used to measure FGF-21 (CV = 2.4–3.4%) and adiponectin (CV = 3–5%). A V-Plex Plus Proinflammatory Panel 1 kit (Meso Scale Discovery) was used to measure tumor necrosis factor (TNF), interleukin 1 (IL-1), IL-6, and interferon-γ (IFN-γ). A multibiomarker disease activity (MBDA) test (Vectra DA; Crescendo Bioscience) was used to measure levels of 12 serum markers: vascular cell adhesion molecule I (VCAM-I), epidermal growth factor (EGF), vascular endothelial growth factor (VEGF), IL-6, TNF receptor type I (TNF-RI), matrix metalloproteinase 1 (MMP-1), MMP-3, bone glycoprotein 39 (YKL-40), leptin, resistin, serum amyloid A (SAA), and CRP. The MBDA score, a validated test of RA activity, was also analyzed^38,39. IL-6 was measured in both the Meso Scale Discovery kit and the MBDA test. The primary analysis used the value from the MBDA test (results were similar for both). Cytokine assays were run on serum samples stored at –80ºC that had never been previously thawed. Radiographs of the hands and feet were performed, and Sharp/van der Heijde (SvdH) scores were determined by a trained radiologist (EGT).

Statistical analysis. Measures of muscle density and muscle area were converted to age-, sex-, and race-specific Z-scores based on distributions among a reference population, as has been previously described^2,5,40,41. Z-scores represent the number of SD above or below the predicted value for a healthy control of the same age, sex, and race. Body composition (ALMI and FMI) measures were converted to Z-scores relative to a national reference population (National Health and Nutrition Examination Survey).

FGF-21 levels at baseline were first log-adjusted to normalize the distribution, and then a standardized FGF-21 value was constructed from the log-adjusted values, with mean and SD set to 0 and 1, respectively. The minimum measurable FGF-21 value was 31.3 pg/mL, so all measurements at this level and below were converted to 15.65 pg/mL for analysis (n = 10). A similar procedure for standardization was used for other cytokines, adipokines, and inflammatory markers with a skewed distribution (specifically IL-1, IL-6, IFN, TNF, ESR, CRP, leptin, EGF, VEGF, SAA, MMP-1, MMP-3, TNF-RI, and YKL-40).

Associations between FGF-21 levels and demographics, disease characteristics, and other serum markers were first assessed using individual linear regression analyses at baseline. All variables found to be moderately associated with FGF-21 levels (P < 0.20) were tested together in a multivariable model. A stepwise deletion process was then used to eliminate covariates with P > 0.10. The final model included standard demographics (age, sex) and all variables with P < 0.10 in the multivariable model after stepwise deletion.

Associations between selected outcome measures at baseline (muscle density and area Z-scores, strength outcomes, and physical functioning assessments) and FGF-21 levels at baseline were assessed using 2 linear regression models: 1 only adjusting for age and sex, and 1 also incorporating adjustments for important identified covariates. In each of these models, FGF-21 was tested as both a continuous and categorial variable (quartiles), comparing the highest quartile to all other quartiles. Models using muscle strength measures as outcomes included an adjustment for height.

To assess associations between FGF-21 and other known cytokines, adipokines, and inflammatory markers, linear regression analyses were performed at baseline between FGF-21 and each individual marker. These models were then tested including an adjustment for the previously identified covariates.

Linear regression analyses were conducted evaluating associations between baseline FGF-21 and per-year rates of change of each outcome measure over follow-up. The regression coefficients represent the difference in the rate of change in outcome per year associated with an FGF-21 level of 1 SD higher at baseline. To assess the clinical importance of these relationships, logistic regression models estimating the risk of clinically meaningful worsening in function, strength, and imaging outcomes were also performed. In the primary analysis, we considered death as a worsening in these outcomes; however, we also performed sensitivity analyses wherein participants who died were excluded from the analysis.

Analyses were performed with STATA 15.1 (StataCorp). Since a Bonferroni correction for multiple comparisons was considered overly conservative in the context of related outcomes, we evaluated instead for consistency across analyses.