Abstract
Objective The objective of this study was to investigate the impact of 92 inflammatory proteins on the risk of cardiovascular disease (CVD) in patients with early rheumatoid arthritis (RA).
Methods This study included consecutive patients with early RA recruited between 1995 and 2002. Stored plasma samples were analyzed for 92 inflammatory proteins. CVD diagnoses were retrieved from national in-patient and cause-of-death registries. Statistical analyses were predesignated as hypothesis-driven or exploratory. For the latter, proteins were selected based on principal component analysis (ie, factor loading > 0.5 within main components). Potential predictors of CVD and coronary artery disease (CAD) were assessed using Cox regression.
Results Data on baseline levels of proteins and CVD were available for 163 patients. As hypothesized, levels of interleukin 17A (IL-17A) were associated with CVD (hazard ratio 1.35, 95% CI 1.02-1.78, adjusted for age, sex, hypertension, diabetes, smoking, and erythrocyte sedimentation rate [ESR]), although not significantly with CAD. Osteoprotegerin (OPG) levels were significantly associated with both outcomes, but only in crude models. No associations were observed for IL-6, tumor necrosis factor, monocyte chemotactic protein-1, or IL-8. In the exploratory analyses, MCP-3 in particular had significant associations with both outcomes in crude models.
Conclusion Circulating IL-17A at RA diagnosis predicted future CVD, although we cannot exclude the possibility that this finding is due to multiple testing. The association was independent of traditional CVD risk factors, and of ESR at the time of diagnosis. Further, OPG may be a predictor of CVD. We also identified some novel potential biomarkers for CVD in RA.
The risk of cardiovascular disease (CVD) is increased in patients with rheumatoid arthritis (RA) compared to the general population,1-3 evident already in the early stages of RA,4 and partly independent of traditional cardiovascular risk factors.5 Inflammation in RA has been associated with endothelial dysfunction, subclinical atherosclerosis,6,7 and the development of clinical CVD8,9 and it is considered a main contributor to CVD in RA. There are also proven links between systemic inflammation and the risk of CVD in the general population.10
Numerous inflammatory proteins generated by the immune system; endothelial and other synovial cells, such as tumor necrosis factor (TNF); interleukins (ILs); interferon-γ; chemokines such as monocyte chemotactic proteins (MCPs); matrix metalloproteinases (MMPs); and vascular endothelial growth factor A (VEGF-A) have been implicated in disease processes leading to synovial inflammation, joint destruction, and systemic inflammation in RA.11,12 Several of these inflammatory proteins are included in the panel used for the present study (Olink Target 96 Inflammation panel), and some have previously been associated with vascular pathology, and to a more limited extent, with CVD-related outcomes in RA.
TNF, which is central to the RA disease processes, contributes to CVD in RA, as demonstrated by reduced aortic stiffness and risk of CVD events in patients treated with TNF inhibitors.13,14 IL-6, a main driver of inflammation with involvement in the RA pathogenesis, has been associated with endothelial dysfunction and subclinical atherosclerosis in RA.15,16 In subjects without RA, both TNF and IL-6 have been associated with coronary artery disease (CAD).17
IL-17 is involved in key aspects of RA and is produced mainly by synovial Th17 cells, but also by peripheral blood mononuclear cells.18 In RA, IL-17 promotes inflammation, neoangiogenesis, and bone and cartilage destruction,11 and possibly atherosclerosis as well.18 Evidence from experimental animal models suggested proatherogenic effects.19 IL-17 has been identified in human atherosclerotic plaques, with higher expression in symptomatic and vulnerable plaques.20 Circulating IL-17 levels have been positively associated with acute coronary syndrome in non-RA subjects21 and in patients with RA with endothelial dysfunction and reduced arterial compliance,22 but also in 1 study with cardiovascular events.23 Further, IL-8 has been associated with reduced endothelial dysfunction and carotid intima-media thickness in patients treated with rituximab.24
Osteoprotegerin (OPG) is a protein involved in bone remodeling and in function as a decoy receptor for RANK ligand (RANKL). Unopposed binding of RANKL to its receptor RANK induces osteoclastogenesis and bone resorption.25 The RANKL/RANK/OPG system can be modulated by hormones and growth factors, but also by immune cells and cytokines, such as TNF, IL-1, and IL-17, which are suspected to contribute to the development of bone erosions in RA.25 The RANKL/RANK/OPG system also appears to have physiological roles in the vasculature,26 and OPG has been identified in atherosclerotic lesions.27 OPG has been evaluated as a marker of atherosclerosis and CVD in patients with and without RA. Higher serum OPG levels have been associated with endothelial activation, atherosclerosis, and established CVD in RA,28-31 as well as correspondingly associated with the presence and severity of CAD in patients with stable chest pain in the absence of RA who are undergoing angiography, and with incident CVD in population-based health survey participants.32,33
MCP-1, also known as C-C motif chemokine ligand 2 (CCL2), is a chemokine that is secreted by leukocytes, but it is also anchored to the cell membrane on endothelial cells. In RA, upregulation of this protein leads to increased endothelial dysfunction and intima-media thickness, possibly through recruitment of leukocytes to the vessel wall.34
The extent to which the association of inflammation with CVD observed in RA is attributed to specific inflammatory and immunological mechanisms is unclear and needs further assessment. The aim of this study was to evaluate inflammation-associated proteins as potential biomarkers for future CVD development in a cohort of patients with early RA.
METHODS
Patients. An inception cohort of consecutive patients with early RA was investigated, as previously described.35,36 Patients were recruited between 1995 and 2002 from the rheumatology outpatient clinic of Skåne University Hospital Malmö, the only hospital serving the city of Malmö, or from the 4 rheumatologists in private practice in the area. The patients were diagnosed with RA by a specialist in rheumatology, fulfilled the 1987 American College of Rheumatology (ACR) classification criteria for RA,37 and had a duration of symptoms for ≤ 12 months at the time of inclusion (baseline). There were no additional exclusion criteria. All patients were managed according to usual care with no prespecified protocol for antirheumatic treatment. The patients were included before the current practice of treat-to-target38 was implemented, and before early treatment with biologic disease-modifying antirheumatic drugs (bDMARDs) came into widespread use. Data on treatment with bDMARDs at any time during the study period were obtained through linkage to a regional biologics register.39
Clinical assessment. Patients were followed according to a structured program and the same rheumatologist (C. Book; see acknowledgment) performed all the clinical examinations. Patient characteristics and disease activity variables were recorded as previously described,36 and blood samples were stored for subsequent analysis. Information on smoking status (current, previous, or never) was collected in a questionnaire filled out by patients at inclusion. Other traditional CVD risk factors at the time of inclusion were assessed by systematic case record reviews. The presence of hypertension, diabetes, or hyperlipidemia before RA diagnosis was defined by a corresponding diagnosis in the case records. For hyperlipidemia, only cases with elevated lipid levels in the case records were classified as having this exposure in the study.
Plasma proteomic biomarkers. Blood samples obtained at inclusion and stored at −80 °C were later analyzed using a protein profiling panel available from Olink, where a large number of proteins can be detected using a small blood sample. For this study, we used the Target 96 Inflammation panel, analyzing 92 inflammatory proteins. Plasma levels of proteins were evaluated by the proximity extension assay technique using a multiplex reagent kit (Olink Bioscience).40,41 For further information about the assays, see https://www.olink.com. A list of analyzed proteins and their measured levels at inclusion is shown in Supplementary Table S1 (available with the online version of this article).
Cardiovascular disease definitions, data sources, and outcome variables. Definitions of CVD in this study were based on International Classification of Diseases (ICD), 8th, 9th, and 10th revision codes. ICD codes from 1969 through 2019 were retrieved from the Swedish National Hospital Discharge Register and Causes of Death Register. In Sweden, reporting of underlying and contributing causes of death to the Cause of Death Register is mandatory.
The following ICD codes were used for definitions: 410-414 and I20-25 for CAD; 440-442, 443.90, 443.99, 443X, 444, I70-72, 173.9, and 174 for peripheral artery disease; and 433-436 (excluding 433.00, 433.99 and 434A), 437A, I63-66, I670, and I672 for cerebrovascular disease.
In the analyses on proteins as potential biomarkers, the primary outcome was first diagnosis of CVD (CAD, cerebrovascular disease, or peripheral artery disease) during the follow-up after RA diagnosis. Secondary outcomes were the first diagnosis of each respective CVD subcategory (CAD, cerebrovascular disease, and peripheral artery disease) during the follow-up after RA diagnosis. Patients with a registered diagnosis of CVD before inclusion in the study were excluded from the corresponding analyses of CVD and the respective CVD subcategory.
Statistical analysis. The relationship between potential biomarkers and outcomes of CVD and of each subcategory were examined in crude and age-sex adjusted Cox regression models. For potential biomarkers with an a priori hypothesis, analyses were additionally adjusted for hypertension, diabetes, smoking status, and erythrocyte sedimentation rate (ESR) to assess whether observed effects were independent of traditional CVD risk factors and the general inflammatory response, both known to influence CVD in RA. The above CVD risk factors were selected based on their significant associations with the outcomes in the current study (Supplementary Table S2, available with the online version of this article). ESR was chosen over C-reactive protein (CRP) since high-sensitivity CRP analysis was not available during a portion of the study period.36 Patients were censored at death or at the end of follow-up (December 31, 2019).
Statistical analyses were performed according to a prespecified study protocol that was finalized before obtaining the Olink data. Potential biomarkers with an a priori hypothesis (Supplementary Table S3, available with the online version of this article), conceived by the authors, were handled separately from analyses involving all potential biomarkers that were considered hypothesis-generating analyses.
For potential biomarkers with an a priori hypothesis, multiple testing was handled using the Holm correction for all P values. Both corrected and original P values, as well as CIs, are reported.
For the exploratory analyses considered to be hypothesis-generating, principal component analysis (PCA) was used to identify groups of proteins that explain the variance in the proteome, as previously described.42 Regression scores of identified components were saved and individually analyzed as potential predictors of the outcomes, as described above. Within components selected based on eigenvalues > 1 and a scree plot, proteins with a factor loading of > 0.50 were investigated as potential biomarkers of the outcomes using the same statistical methods as described above.
All data from Olink were presented in arbitrary units, and all nonnormally distributed variables were log-transformed using the natural logarithm. Variables were considered skewed when Shapiro-Wilks test for normality gave statistics of < 0.85. Variables with values 0.85-0.90 were graphically assessed.
Hazard ratio (HR) estimates for all potential biomarkers were presented as per SD of arbitrary units to facilitate comparison. Statistical analysis was performed using IBM SPSS version 28.0 (IBM Corp.).
Ethics statement. All patients gave their written informed consent for participation in the study, including data collection and inclusion in the database.
RESULTS
Patient characteristics. A total of 168 patients had available plasma samples from the time of inclusion and CVD data. Five were excluded due to insufficient sample quality, as described below; hence, 163 patients (median symptom duration: 8.0 [IQR 5.9-10.1] months) were included in the analyses of potential biomarkers. Patient characteristics at inclusion are shown in Table 1. The most frequently used DMARD was methotrexate, and 22.1% of patients were treated with a bDMARD at some time during the follow-up.
Characteristics of patients with early rheumatoid arthritis (N = 163) at inclusion.
Distribution of CVD, deaths, and follow-up time. There were 17 patients with a diagnosis of CVD and 11 with CAD before inclusion. These patients were excluded from the analyses of potential baseline predictors of respective outcomes. From inclusion to the end of follow-up (2019), a first-ever diagnosis of CVD occurred in 47 patients and CAD in 37. During a mean total follow-up time of 14 years, 89 out of the 163 patients died. In 31 patients, a defined CVD event was the cause of death. Since there were only 23 and 11 patients with incident peripheral artery disease and cerebrovascular disease, respectively, during the follow-up, analyses of predictors of these outcomes were not considered feasible.
Effect of traditional CVD risk factors on the risk of CVD and CAD during follow-up. Male sex, age, and hypertension were significantly associated with both CVD and CAD in crude and adjusted analyses (Supplementary Table S2, available with the online version of this article). Diabetes had a significant effect on both outcomes but only in crude analyses, whereas current smoking was significantly associated with CAD in both crude and adjusted analyses, but only in the adjusted analysis for CVD. Hyperlipidemia had no significant impact on either outcome (Supplementary Table S2).
Protein levels at inclusion and selection for further analysis. Measured plasma levels of examined proteins from blood samples taken at inclusion are listed in Supplementary Table S1 (available with the online version of this article). Although exact measurements were available for all evaluated proteins and cases, some proteins had individual values indicated as below the quality-assured level of detection (LOD) of the assay. Seventeen proteins had > 10% of their individual values below the LOD. Eight proteins had ≥ 50% of their individual values below the LOD and were excluded from analyses of these proteins as potential biomarkers of CVD and CAD (Supplementary Table S1). Blood samples from 5 cases did not pass the internal quality control, testing the reliability of the assay performance for these samples; therefore, these cases were excluded from all analyses.
Effect of biomarkers with a priori hypotheses on the risk of CVD and CAD during follow-up. Levels of IL-17A were significantly associated with CVD in the crude and age-sex adjusted models, whereas the associations did not reach significance when correcting for multiple testing (Table 2) or in models evaluating prediction of CAD (Table 3). The association between higher levels of IL-17A and CVD remained significant in analysis adjusted for age, sex, diabetes, hypertension, smoking status, and ESR (HR 1.35, 95% CI 1.02-1.78), with a similar trend for CAD (HR 1.25, 95% CI 0.90-1.75). OPG levels were significantly associated with CVD (HR 1.54, 95% CI 1.14-2.08) and with CAD (HR 1.63, 95% CI 1.19-2.23) in crude analysis and when correcting for multiple testing, but not in adjusted models (Table 2 and Table 3). There were no significant associations for IL-6, TNF, MCP-1, or IL-8 with CVD (Table 2) or CAD (Table 3). Results were similar in analyses adjusted for age, sex, and each traditional CVD risk individually (Supplementary Table S4 and Supplementary Table S5, available with the online version of this article).
Effect of biomarkers with a priori hypotheses on the risk of CVD.
Effect of biomarkers with a priori hypotheses on the risk of CAD.
PCA and effect of identified components on the risk of CVD and CAD during follow-up. Factor analysis revealed sufficient Kaiser-Meyer-Olkin measure of sampling adequacy of 0.892 and a Bartlett test of sphericity with a significance level of < 0.001. Seventeen components showed eigenvalue of > 1. When visualized in a scree plot, 5 components stood out as explaining the majority (57.1%) of the total variation in the dataset and were chosen for further evaluation (Supplementary Figure 1, available with the online version of this article). Forty-six biomarkers had a factor loading in any of the 5 components of > 0.5 and were chosen for exploratory analyses. These biomarkers and their respective localization in the components are listed in Supplementary Table S6 (available with the online version of this article). When analyzing the relation between the regression scores of each of the 5 components with the outcomes, component 4 was significantly and positively associated with CVD in the unadjusted model (crude HR 1.32, 95% CI 1.01-1.73), but not in the adjusted model or with CAD (Table 4).
Effect of identified components from the PCA on the risk of CVD and CAD.
Effect of potential biomarkers selected for exploratory analysis on CVD and CAD during follow-up. In the exploratory analyses, 7 new proteins showed significant associations with CVD, CAD, or both (Table 5). MCP-3 (also known as CCL7) significantly predicted CVD (HR 1.31, 95% CI 1.02-1.69) and CAD (HR 1.33, 95% CI 1.00-1.77) in crude models, but not in the adjusted models (Table 5). CUB domain-containing protein 1 (CDCP1) showed significant associations with both CVD (HR 1.57, 95% CI 1.17-2.09) and CAD (HR 1.36, 95% CI 1.01-1.83) in crude analyses, but not in the adjusted analyses (Table 5). VEGF-A, hepatocyte growth factor (HGF), protein S100-A12 (EN-RAGE), leukemia inhibitory factor receptor (LIF-R), and CCL25 showed significant associations in crude models with CVD but not with CAD (Table 5). In addition, as in the a priori hypothesis-testing analyses, OPG (which had a factor loading of 0.73 in component 4) was predictive of CVD and CAD in crude analyses. Further adjustment for traditional CVD risk factors had no major impact on associations above (Supplementary Table S4 and Supplementary Table S5, available with the online version of this article).
Biomarkers from the exploratory analyses with significant associations with CVD and/or CAD.
DISCUSSION
In this cohort of patients with early RA, circulating IL-17A at time of RA diagnosis was a significant predictor of future CVD. As hypothesized, OPG may be predictive of CVD and CAD. OPG was also identified as a predictor of CVD and CAD in the hypothesis-generating analyses, underlining its role as an important biomarker in this context. We identified several inflammatory proteins that may be used as potentially novel biomarkers of CVD and CAD.
Our results from the present study indicate that circulating levels of IL-17A at time of RA diagnosis are predictive of future CVD development, and that this effect is independent of age, sex, hypertension, diabetes, smoking, and ESR levels. This may indicate a mechanism leading to CVD in RA that is separate from the effects of traditional CVD risk factors and the general inflammatory response on CVD. Our findings are consistent with the results from the only other such study in RA, where levels of IL-17A were higher in patients who later experienced a CVD event.23 These findings, together with the known involvement of IL-17A in atherosclerosis19,20 and its association with CVD in patients without RA,21 suggest that IL-17A may have utility as a prognostic marker of CVD in RA. Further, therapeutic targeting of IL-17 may have potential for reducing the risk of CVD development in RA and possibly other high-risk groups.
Mechanistically, IL-17A enhances the effect of TNF, and their combination leads to activation of endothelial cells, upregulated expression of MMPs, and increased apoptosis in vascular smooth muscle cells, which may aggravate atherosclerosis and increase plaque instability.19,43 Although IL-17 inhibition reduces disease activity in RA, the effect size is modest and its use in RA is limited.44 The theoretically appealing use of IL-17 in conjunction with TNF inhibitors has been tested in a clinical trial of a dual antibody targeting both TNF and IL-17A in patients with RA, but did not have better efficacy than TNF inhibition alone.45 Another study showed that add-on IL-17 inhibition in patients with insufficient response to certolizumab pegol led only to transient improvement of disease activity.46 Whereas effects of IL-17 inhibition on endothelial dysfunction, carotid intima-media thickness, and noncalcified coronary plaques have been described in psoriasis,47-49 its potential use as an antiatherosclerotic agent has not yet been studied in RA.
OPG was a significant predictor of both CVD and CAD in unadjusted analyses but not when adjusting for age and sex. Previous studies in RA that have also taken into consideration that OPG increases with age have shown OPG to be associated with coronary artery calcification,30 carotid atherosclerosis,28 and established CVD.31 In a study that observed an association of baseline levels of OPG with carotid intima-media thickness at 11 years, the significance was lost when age was included in the model.29 OPG appears to be a prognostic marker of atherosclerosis and CVD, but this may be partly due to its correlation with age, a strong risk factor for CVD.
As for the other potential biomarkers with an a priori hypothesis in the present study, TNF, IL-6, IL-8, and MCP-1 were not significantly associated with CVD or CAD. Although treatment with TNF inhibition has been shown to reduce the risk of CVD in RA,14 previous studies to date have shown no direct associations for levels of these proteins with the development of clinical CVD, but only with endothelial dysfunction and surrogate measures of atherosclerosis.15,16,24,34 However, in subjects without RA, levels of TNF and IL-6 have been associated with CAD events.17
In our exploratory analyses we identified several proteins that may have some prognostic ability for CVD or CAD, but none displayed significant associations in the adjusted analyses. MCP-3 and CDCP1 showed significant associations for both CVD and CAD in crude analyses. Although they have not been previously studied in relation to CVD in RA, MCP-3 has been suggested to be a contributor to atherosclerosis.50 VEGF-A, HGF, EN-RAGE, LIF-R, and CCL25 may be novel biomarkers for CVD and need further studying.
The association between component 4 in our PCA and CVD appeared to be largely driven by the effects of OPG and LIF-R, which had high factor loadings within this component. Like OPG and LIF-R, component 4 lost significance after adjustment for age, suggesting that it reflects proteins related to aging.
Limitations in our study include the relatively small sample size, which affects statistical power for the multivariable analyses. Further, the patients were included just before or shortly after the introduction of bDMARDs for the treatment of RA, and they were classified according to the 1987 ACR criteria. The results of this study may not apply to patients diagnosed according to more recent algorithms; in particular, those with ready access to biologics who are treated according to a treat-to-target strategy.38
Another limitation is that data on potential predictors were available only at baseline, so longitudinal evaluation of the effects of these factors was not possible. Also, inherent to measuring only circulating levels of proteins, our study was unable to identify the origins of protein expression and whether they represent RA-associated inflammatory activity, preclinical atherosclerotic processes, or both.
Strengths of our study include the structured longitudinal follow-up of an inception cohort from a defined catchment area. Therefore, selection bias is not a major issue in this study, and the results could be generalized to patients with RA seen in clinical practice.
Circulating IL-17A at RA diagnosis predicted future CVD, although we cannot exclude that this finding was due to multiple testing. The observed association was independent of traditional CVD risk factors and levels of ESR at the time of diagnosis, suggesting that other mechanisms may contribute to development of CVD in patients with RA. Our results suggest that the utility of IL-17A as a predictive biomarker for CVD should be further investigated, although the practical implications for risk stratification and therapy are presently unclear. In contrast to our hypotheses, TNF, IL-6, IL-8, and MCP-1 were not associated with CVD in this study, whereas OPG may have some prognostic value in cardiovascular risk assessment. Lastly, we also identified some novel potential biomarkers for CVD in RA.
ACKNOWLEDGMENT
Christina Book, MD, PhD, initiated this project and performed a major part of the data collection. She passed away before the preparation of this manuscript.
Footnotes
This work was supported by the Swedish Research Council (grant no. 2015-02228), the Swedish Rheumatism Association (grant no. R-664091), Lund University (grant no. ALFSKANE-446501), and Region Skåne (grant no. 2022-1222).
CT has received consulting fees from Roche; speaking fees from AbbVie, BMS, Nordic Drugs, Pfizer, and Roche; and an unrestricted grant from BMS. The remaining authors declare no conflicts of interest relevant to this article.
- Accepted for publication April 12, 2024.
- Copyright © 2024 by the Journal of Rheumatology
REFERENCES
DATA AVAILABILITY
The datasets generated and/or analyzed during the current study are not publicly available due to Swedish legislation (Personal Data Act), but a limited and fully anonymized dataset containing the individual patient data that support the main analyses is available from the corresponding author on reasonable request.
