Abstract
Objective While general population studies have shown inverse associations between physical activity and common inflammatory biomarkers, the effects of physical activity on inflammatory gene expression and signaling pathways in rheumatoid arthritis (RA) remain unknown. We aimed to determine whether physical activity independently associates with expression of inflammatory genes among people with RA.
Methods This was a prospective observational study of adults with RA. Physical activity was measured by quantitative actigraphy over 7 consecutive days, and peripheral blood collected during the same time period was used for RNA sequencing followed by differential gene expression, pathway, and network analyses.
Results Actigraphy and RNA sequencing data were evaluated in 35 patients. The cohort had a mean age of 56 (SD 12) years, and was 91% female, 31% White, 9% Black, 9% Asian, and 40% Hispanic. We found 767 genes differentially expressed (adjusted P < 0.1) between patients in the greatest vs lowest physical activity tertiles, after adjusting for sex, age, race, and ethnicity. The most active patients exhibited dose-dependent downregulation of several immune signaling pathways implicated in RA pathogenesis. These included CD40, STAT3, TREM-1, interleukin (IL)-17A, IL-8, Toll-like receptor, and interferon (IFN) signaling pathways. Upstream cytokine activation state analysis predicted reduced activation of tumor necrosis factor-α and IFN in the most active group. In sensitivity analyses, we adjusted for RA disease activity and physical function and found consistent results.
Conclusion Patients with RA who were more physically active had lower expression of immune signaling pathways implicated in RA pathogenesis, even after adjusting for disease activity, suggesting that physical activity may confer a protective effect in RA.
Rheumatoid arthritis (RA) is a systemic autoimmune disease that affects 1.5 million Americans and is characterized by joint and systemic inflammation, resulting in chronic pain, functional limitations, and premature cardiovascular disease (CVD).1 While existing therapeutics such as biologic disease-modifying antirheumatic drugs (bDMARDs) reduce disease severity, they often fail to adequately control symptoms and confer a significant risk of adverse drug events, including increased risk of life-threatening infections. Given these limitations, there is an unmet need for nonpharmacologic strategies to augment existing medical management of RA, improve long-term clinical outcomes, and reduce disease-related symptoms.
While historically it was thought that exercise might exacerbate rheumatic conditions, that precept has since been disproven.2 Driving this shift were clinical trials of resistance and aerobic exercise programs for people with RA, in which the exercise treatment interventions were found to be safe and resulted in unchanged or improved disease activity scores compared to controls.3,4 As an example, a previous exercise pilot trial found that older adults with RA randomized to a 10-week high-intensity interval walking intervention had a 38% reduction in disease activity at the end of the intervention.5
The mechanisms responsible for the therapeutic effect of exercise in rheumatic diseases are incompletely understood. One hypothesis is that skeletal muscle-secreted myokines confer antiinflammatory effects following episodes of physical activity.6 This hypothesis is supported by several studies in the general population that demonstrated a clear association between higher levels of physical activity, reductions in proinflammatory cytokine signaling, and lower systemic inflammation.7-10 Though studies in the general population have observed favorable effects of exercise on immune function, how physical activity influences systemic inflammation in the context of immune-mediated inflammatory diseases such as RA remains poorly understood and represents an important question in need of further investigation.
We conducted a prospective observational cohort study to investigate mechanistic relationships between physical activity and systemic inflammatory gene expression in patients with RA. Using quantitative actigraphy, whole blood transcriptomics, and detailed clinical phenotyping, we asked whether physical activity independently associates with expression of inflammatory genes implicated in the pathogenesis of RA.
METHODS
Study design, clinical cohort, and ethics statement. We studied participants enrolled in an ongoing prospective observational cohort study of patients with RA: “Sleep disturbance in rheumatoid arthritis: phenotypes, causes, and impact” (RAZZ; R01 AR069616). Patients were enrolled in RAZZ between 2018 and 2021 after being recruited from an extensive network of community, academic, and safety net rheumatology clinics in the San Francisco Bay Area, including the University of California San Francisco (UCSF) and Zuckerberg San Francisco General Hospital rheumatology clinics. Inclusion criteria for the parent cohort included age ≥ 18 years, a physician’s diagnosis of RA, and English or Spanish fluency. All patients in the parent cohort with actigraphy and transcriptomic data were eligible for inclusion in the current analysis. However, as a result of the unexpected occurrence of the coronavirus disease 2019 (COVID-19) pandemic beginning in early 2020, which precluded in-person nonessential clinical research studies, we were only able to obtain blood samples for transcriptomic data generation on the subset of participants with baseline visits before the pandemic.
During the period of data collection for this analysis, study visits were conducted in-person. At each study visit, we collected detailed clinical information, assessed RA disease activity, and downloaded actigraphy data collected between study visits. In addition, peripheral blood was collected and stored at −80 °C for subsequent RNA sequencing (RNA-seq). The UCSF institutional review board (IRB) approved all procedures (UCSF IRB protocol #17-21790), and all research was performed in accordance with the Declaration of Helsinki and UCSF IRB guidelines and regulations. Written informed consent was obtained from all participants.
Physical activity. Physical activity was assessed objectively using the GT9X ActiGraph Link device (ActiGraph, Inc.). This ActiGraph device resembles a small watch and contains a validated tri-axial accelerometer and integrated gyroscope and magnetometer, which collectively record absolute subject movement. Participants wore the ActiGraph on their wrists for 7 consecutive 24-hour periods (1 full week) before returning them for data analysis. ActiGraph physical activity data are provided as counts per minute (CPM), which are a result of aggregating postfiltered raw accelerometer data over 1-minute intervals (epochs) and are used to define the time awake spent in sedentary activity (0-99 CPM), light activity (100-1951 CPM), moderate activity (1952-5724 CPM), vigorous activity (5725-9498 CPM), and very vigorous activity (≥ 9499 CPM) based on established cutpoints that correspond to metabolic equivalent (MET) levels.11 Participants were categorized using the highest, middle, and lowest tertiles for percentage of time spent in at least moderately intense physical activity across the entire RA cohort, resulting in 3 physical activity groups: the least active (labeled “inactive”), the most active (labeled “active”), and an intermediate group.
RA-specific disease factors. Age of diagnosis was obtained by self-report. Disease activity was assessed with the Rheumatoid Arthritis Disease Activity Index, a validated patient-reported instrument.12 Participants were also queried regarding current treatment with glucocorticoids (GCs)—including dosage and frequency—as well as other immunomodulatory medications.
Other variables. Sociodemographic data were collected, including sex, age, race, ethnicity, and educational attainment (categorized as high school graduate or less vs those with additional education). Physical function was measured using the Patient-Reported Outcomes Measurement Information System physical function scale.13 Height and weight were measured during the baseline in-person visit, and BMI was calculated as weight in kilograms divided by height in meters squared. Participants provided information on health-related behaviors (eg, smoking) and comorbidities such as CVD, diabetes mellitus, asthma, and cancer.
RNA-seq. Whole blood was collected and stored at −80 °C. Following RNA extraction (Zymo Pathogen Magbead Kit) and DNAse treatment, human globin and ribosomal RNA was depleted using FastSelect (Qiagen) according to described methods.14 RNA was then fragmented and underwent library preparation using the NEBNext Ultra II RNAseq Kit (New England Biolabs) as previously described.14 Libraries underwent 146 nucleotide paired-end Illumina sequencings on a Novaseq 6000 (Illumina) instrument.
Gene expression data processing and quality control. Following demultiplexing, sequencing reads were aligned against the human genome (NCBI GRC h38) using STAR1 to extract gene counts. Samples retained in the dataset had ≥ 3.0 × 105 gene counts, and the median across all samples was 5.8 × 105 gene counts. Data merging and normalization across different activity groups were performed. Gene counts were normalized with the median of ratios method using R package DEseq2. For covariates, one-hot encoding was applied for categorical variables and min-max scaling was performed for numeric covariates. As an additional quality control measure, genes expressed in < 30% of the patients in each group were filtered. In total, 19,260 genes were kept for the subsequent analysis.
Differential gene expression analysis. Differentially expressed genes were identified using the R package DEseq2.15 Sex, age, race, and ethnicity were included as covariates in the linear model. Next, to address the possibility of confounding from inability to participate in physical activity among patients with more severe disease, we conducted 2 sensitivity analyses in which we adjusted for disease activity and physical function in addition to the aforementioned demographic variables. Finally, to correct for the potential batch effect, the R package sva16 was used to calculate a surrogate variable, which was also integrated into the differential expression linear model as a covariate. Independent hypothesis weighting was used as a multiple testing procedure and the significance of differential expression was defined as an adjusted P (Padj) < 0.1.
Pathway analysis. Ingenuity pathway analysis (IPA)17 as carried out on differentially expressed genes with a P < 0.1 ranked by log2 fold change. Significant IPA results were defined as those with a score absolute value > 2 or an overlap P < 0.05. The top 3 up- and downregulated canonical pathways based on score, as well as all pathways related to immunity and inflammation with > 1 and overlap P < 0.05, were included in Figure 1. Also in Figure 1 are upstream regulating cytokines with an > 2 and overlap P < 0.05. Complete IPA results are provided in the Supplementary Materials (available with the online version of this article).
Network analysis. The network analysis was performed using STRING v.11 (https://string-db.org). The network matrix was exported as a .tsv file and the plot was recreated using Cytoscape (https://cytoscape.org). Each protein was visualized as a node and each potential protein–protein interaction was visualized as an edge. The size of each node reflects its degree centrality metrics. Figure 2A represents one of multiple network maps. Complete network map data are provided in Supplementary Data File 3 (available with the online version of this article).
In silico analysis of cell type proportions. Cell type proportions were estimated from bulk host transcriptome data using the CIBERSORT X algorithm following previously described methods.14 The estimated proportions were compared between the 3 patient groups using a Mann-Whitney-Wilcoxon test (2-sided) with Bonferroni correction.
Data availability. Gene counts are available under Gene Expression Omnibus accession number GSE179302. All code is available at Github (https://github.com/drychkov/PA_in_RA).
RESULTS
There were 35 adults in the RAZZ cohort with complete actigraphy (physical activity), RNA-seq (gene expression), and clinical data. The cohort had a mean age of 56 years (SD 12), was 91% female, and self-reported the following racial and ethnic identities: 31% White, 9% Black, 9% Asian, and 40% Hispanic (Table). The mean disease duration was 13 years (SD 11) and 71% of the cohort was seropositive for rheumatoid factor and/or anticyclic citrullinated peptide antibody. Fifty-four percent of participants were taking methotrexate, 51% were treated with a bDMARD, and 26% were treated with systemic GCs. Only 13% of the participants were taking oral GCs > 7.5 mg of prednisone equivalent per day.
Consistent with prior studies of physical activity behavior among people with RA,18 the cohort was relatively sedentary; none of the patients engaged in vigorous physical activity during the 7-day period of actigraphy monitoring. The percentage of time spent in moderate physical activity was 18% in the most active group compared to 4.5% in the least active group (Supplementary Table S1, available with the online version of this article). Though the study was not powered to detect statistically significant demographic or clinical differences between the physical activity groups, we found that patients in the active group were younger (mean age = 50 yrs) compared to those in the intermediate (mean age 56 yrs) and inactive groups (mean age = 63 yrs; P = 0.04; Table).
Differential gene expression analysis comparing the most vs least physically active groups (based on activity tertile) identified 767 genes at a Padj < 0.1 (Figure 1A). Principal component analysis demonstrated clear separation based on physical activity groups (Figure 1B), and pathway analysis (Methods) revealed that the most physically active patients exhibited downregulation of diverse innate and adaptive immune signaling pathways implicated in the pathogenesis of RA, including CD40, signal transducer and activator of transcription 3 (STAT3), triggering receptor expressed on myeloid cells (TREM-1), interleukin (IL)-17A, IL-8, Toll-like receptor (TLR), and type I interferon (IFN) signaling (Figure 1C; Supplementary Data Files 1 and 2A, available with the online version of this article).19 Complete IPA results are provided in the Supplementary Data File 2. Prediction of upstream cytokine activation states from the transcriptomic data suggested inhibition of type I, II, and III IFNs, and activation of epoetin, among the most physically active patients (Figure 1D; Supplementary Data File 2B).
Given the role of pathologic inflammation in RA, we next sought to more rigorously evaluate the proinflammatory genes downregulated in the most active group by performing a network connectivity analysis. This revealed relationships between genes related to type 1 IFN signaling (eg, MX1, IFI44L, IFIT1) and inflammasome signaling (eg, IL-1RN, NOD2), as well as other cytokine signaling pathways (eg, IL-6R, IL17RA; Figure 2A, Supplementary Data File 3, available with the online version of this article). To more deeply characterize the relationships between these inflammatory genes and physical activity, we evaluated their expression across all 3 physical activity groups. Intriguingly, a dose-dependent correlation between physical activity and reduced expression of proinflammatory genes was observed (Figure 2B).
We considered that differences in immune cell populations might underlie the observed differentially expressed genes, and thus performed in silico cell type deconvolution. We found no statistically significant differences in predicted proportions between the highest and lowest physical activity groups, but we observed a trend toward decreased monocytes in the most active group (Supplementary Figure S1, Supplementary Data File 4, available with the online version of this article).
Last, we considered that differences in the ability to engage in exercise among patients with more active disease and/or more joint damage might explain our findings. To test for this possibility, we conducted 2 sensitivity analyses in which disease activity (sensitivity analysis 1) and physical function (sensitivity analysis 2) were included as covariates. The sensitivity analysis in which we adjusted for disease activity revealed consistent results compared to the main analysis, including 674 differentially expressed genes (Padj < 0.1) and similar downregulation of IFN and other proinflammatory cytokine signaling pathways among the most active patients (Supplementary Data File 5, available with the online version of this article). The sensitivity analysis adjusted for physical function yielded 332 differentially expressed genes, including those most statistically significant and biologically relevant from the primary analysis (eg, IFI44L, MX1), as well as similar pathway analysis results (Supplementary Data File 6).
DISCUSSION
Randomized controlled trials of exercise interventions for RA have found that exercise improves joint pain, fatigue, and disease activity,3,4,20-23 and decreases the risk of progressive joint damage,20 but the mechanisms underlying these apparent benefits have remained in question. In this study, we used quantitative actigraphy and whole blood transcriptomics to show that physical activity moderates inflammatory signaling among people with RA.
Studies in the general population have found that higher levels of physical activity associate with lower inflammatory biomarkers, including C-reactive protein and tumor necrosis factor (TNF)-α, even after adjusting for excess adiposity.7-10 Our results are consistent with these observations, as we observed an association between physical activity and attenuation of several proinflammatory signaling pathways implicated in RA pathogenesis, including CD40, STAT3, TREM-1, IL-17A, IL-8, TLR, and type I IFN.19 Further, our gene expression data demonstrated lower TNF activation among the most active relative to the least active patients in this RA cohort.
Unexpectedly, we found that IFN signaling was inversely correlated with physical activity. This finding has clinical significance given that IFN gene expression in peripheral blood has been associated with RA autoantibody production, development of chronic RA in patients with early inflammatory arthritis, poorer response to initial therapy, and nonresponse to rituximab (RTX).24 Additionally, previous single-cell RNA-seq studies have shown a correlation between peripheral blood mononuclear cell IFN gene expression and infiltration of synovium with plasma cells in people with RA.25 We also found that physical activity was associated with significantly attenuated TLR and IL17RA signaling, which have been shown to contribute to unbalanced production of cytokines in RA.26 We considered that disease severity or lower physical function might represent potential confounders, but in sensitivity analyses that included disease activity and physical function as covariates, we found consistent results.
Intriguingly, the association of physical activity with reduced expression of several proinflammatory genes was dose dependent. The threshold of moderate activity employed in ActiGraph scoring is quite modest at approximately 2.0 METs, which corresponds to home activities such as standing to wash dishes, food shopping, or walking at a ≤ 2.0-mph pace. The most active group spent an average of 4.3 hours per day in these moderate intensity activities, compared to the least active group’s 1.1 hours per day. The dose-dependent relationship between moderate activity and attenuation of proinflammatory gene expression suggests that greater exposure to physical activity could confer a larger effect on inflammatory signaling—and, therefore, greater potential therapeutic benefit—in this patient population.
The gene expression differences associated with physical activity in our study were in the range of those observed in studies of RA treatment interventions. For instance, a study by van Baarsen et al that evaluated gene expression differences before and after treatment with infliximab in 33 patients with RA found 1623 differentially expressed genes at a Padj < 0.05 with a median log2 fold change of 0.86,27 including some of the same genes found to be influenced by physical activity in our study (eg, IL17R). This compared to 206 differentially expressed genes at a Padj < 0.05 and log2 fold change of 2.1 in our cohort. Another observational study that compared 14 patients with RA before and 6 months post-RTX treatment identified 0 differentially expressed genes at a Padj < 0.05 but 124 with at least a 2-fold change in expression,28 including many of the same IFN-related genes that we observed (eg, IFI44L). Further work is needed to understand the comparative effect size of physical activity on inflammatory gene expression with respect to pharmacologic treatment interventions for RA.
This study has limitations. Because this was an observational study, we were able to identify associations between physical activity and inflammatory gene expression, but we cannot prove causation or directionality. Only a subset of patients in the RAZZ cohort had complete data for both actigraphy measures and transcriptomics, which limited the sample size for the current analysis. Additionally, though we included a validated patient-reported assessment of disease activity that has been shown to strongly correlate with the Clinical Disease Activity Index,29 the study would have been enhanced by including a physician evaluation of disease activity. Finally, we were unable to assess the effect of vigorous physical activity because no patient engaged in vigorous physical activity during the study. Despite these limitations, we believe the findings remain noteworthy as this is the first study, to our knowledge, to assess physical activity in the context of RA using whole blood transcriptomics. The level of physical activity observed in this cohort realistically reflects the activity patterns observed in this patient demographic,18 and if anything, our results may have been more pronounced in a cohort with a broader range of activity levels.
Given the limitations of existing pharmacologic treatments for RA and the need for nonpharmacologic adjunctive strategies, our findings have important clinical implications. Despite major advances in treatment, including the advent of bDMARDs, RA remains a chronic, incurable condition, and fewer than half of patients treated with immunosuppressive DMARDs achieve disease remission.30,31 Further, adverse drug events from RA medications are common, with an estimated incidence of 15 per 100 patient-years,32 leading many people with RA to express a preference for nonpharmacologic treatment approaches.33,34 Taken together, the prevalence of persistent symptoms despite treatment, risk of side effects from DMARDs, and patient preference for lower-risk treatment approaches emphasize the critical need for adjunctive nonpharmacologic interventions for RA. Our results suggest that physical activity interventions have the potential to not only improve overall health and mitigate the risk of important comorbidities among people with RA, but may also attenuate pathologic inflammatory signaling and disease activity.
This work provides an important foundation for future studies, including additional research to further characterize the biological mechanisms linking physical activity to inflammatory signaling using proteomic, metabolomic, and single-cell RNA-seq approaches. Our study also highlights the need for a randomized clinical trial to assess the direct effect of exercise on blood transcriptomic markers, and to compare gene expression among patients with RA across a greater spectrum of physical activity. Last, a key outstanding question is whether our findings extend to other autoimmune rheumatic diseases such as systemic lupus erythematosus.
In summary, among a representative cohort of patients with RA, we found a striking and dose-dependent association between moderate physical activity and attenuated expression of genes involved in both innate and adaptive immune signaling, even after adjusting for disease severity and other covariates. These findings provide the first mechanistic evidence to support a disease-modifying effect of physical activity in RA.
Footnotes
This work is supported by the National Institutes of Health (NIH)/National Institute of Arthritis and Musculoskeletal and Skin Diseases (NIAMS) R01 AR069616, NIH/National Heart, Lung, and Blood Institute (NHLBI) K23HL138461-01A1, and the Rheumatology Research Foundation. This study was also supported by the NIH/NIAMS P30 Center for the Advancement of Precision Medicine in Rheumatology at UCSF (P30AR070155).
S.L. Patterson and S. Sun are co–first authors and contributed equally to this work. C.R. Langelier and M. Sirota are co–senior authors and contributed equally to this work.
The authors declare no conflicts of interest relevant to this article.
- Accepted for publication June 17, 2022.
- Copyright © 2022 by the Journal of Rheumatology