Abstract
Background Epidemiological studies have shown that impaired lung function is common and associated with increased risk of cardiovascular disease. Increased levels of several inflammatory and cardiovascular disease-related plasma proteins have been associated with impaired lung function. The aim was to study the association between plasma proteomics and forced expiratory volume in 1 s (FEV1), forced vital capacity (FVC) and FEV1/FVC ratio.
Methods We used a discovery and replication approach in two community-based cohorts, EpiHealth and the Malmö Offspring Study (total n=2874), to cross-sectionally study 242 cardiovascular disease- and metabolism-linked proteins in relation to FEV1, FVC (both % predicted) and FEV1/FVC ratio. A false discovery rate of 5% was used as the significance threshold in the discovery cohort.
Results Plasma fatty acid-binding protein 4, interleukin-1 receptor antagonist, interleukin-6 and leptin were negatively associated with FEV1 and paraoxonase 3 was positively associated therewith. Fatty acid-binding protein 4, fibroblast growth factor 21, interleukin-1 receptor antagonist, interleukin-6 and leptin were negatively associated with FVC and agouti-related protein, insulin-like growth factor-binding protein 2, paraoxonase 3 and receptor for advanced glycation end products were positively associated therewith. No proteins were associated with FEV1/FVC ratio. A sensitivity analysis in EpiHealth revealed only minor changes after excluding individuals with known cardiovascular disease, diabetes or obesity.
Conclusions Five proteins were associated with both FEV1 and FVC. Four proteins associated with only FVC and none with FEV1/FVC ratio, suggesting associations mainly through lung volume, not airway obstruction. However, additional studies are needed to investigate underlying mechanisms for these findings.
Abstract
Several cardiovascular disease-linked proteins were associated with FEV1 and FVC, but not with FEV1/FVC ratio, suggesting that the relationships are mainly with lung volume, not airflow obstruction https://bit.ly/3j19Io8
Introduction
Impaired lung function is common and associated with increased risk of cardiovascular disease (CVD) and overall mortality [1–4]. The lung function trajectories leading to impaired lung function vary: some individuals achieve a normal peak value in young adulthood and then have an accelerated decline in lung function, others start from a lower peak value and have a physiological age-associated decline [5]. Therefore, it is difficult to predict which individuals are going to develop impaired lung function. New diagnostic tools are warranted [5].
CVD is a common comorbidity and cause of death among patients with COPD, with low-grade systemic inflammation suggested as an important mediator [6]. Previous studies have identified several inflammatory biomarkers associated with both impaired lung function and CVD, including C-reactive protein, blood neutrophil counts, blood eosinophil counts, fibrinogen and interleukin (IL)-6 [7–9]. However, there are likely many mechanisms linking impaired lung function and CVD, e.g., oxidative stress, extracellular matrix destruction and defective immunomodulation [10].
Knowledge of pathophysiological mechanisms leading to impaired lung function remains limited [5]. Proteomics detects and quantifies many proteins simultaneously and can indicate pathways involved in disease development [11]. Proteins associated with low spirometry values could increase knowledge of the continuum from normal to impaired lung function [12]. Thus, large proteomics studies could increase knowledge of impaired lung function and its association with CVD.
We have previously studied the association between forced expiratory volume in 1 s (FEV1) and proteomics, focusing on inflammatory processes [13]. The primary aim of this paper was to study associations between broader proteomics data, focusing on CVD-related proteins, and lung function in a large population. More specifically, we tested the associations between proteomics and spirometry data in two community-based cohorts in a discovery and replication approach. The proteomics data included cardiovascular and inflammation-associated proteins and proteins involved in metabolic processes, e.g., cellular stress or apoptosis. The analyses encompassed FEV1, forced vital capacity (FVC) and FEV1/FVC ratio.
Material and methods
More details on the material and methods are available in online supplementary file 1a and 1b.
Cohort descriptions
This study was based on two independent community-based Swedish cohorts, EpiHealth and the Malmö Offspring Study (MOS).
Discovery cohort
The EpiHealth study was conducted in 2011–2018. The primary objective was to study how interactions between lifestyle factors and genotypes contributed to development of common disorders in humans (e.g., cancer, cardiovascular and respiratory diseases). For details, see www.epihealth.lu.se. Briefly, individuals living in Malmö and Uppsala, aged 45–70 years, were randomly selected from the Swedish population registry. A total of 25 104 answered an extensive online questionnaire and visited a test centre for blood sampling, spirometry and anthropometric measuring [14]. For this study, only participants from Uppsala were included, because they had proteomics measurements. All participants with measurements on proteomics, lung function and covariates were included in this study, and the final cohort comprised 2229 participants.
Replication cohort
The MOS was a population-based cohort study performed in 2013–2021 (n=5300). The aim was to map family traits as risk factors of chronic diseases based on gene–environment interaction. Participants include the children and grandchildren of participants in the Malmö Diet and Cancer Study – Cardiovascular Cohort; for details, see https://www.malmo-kohorter.lu.se/malmo-offspring-study-mos. Briefly, individuals aged ≥18 years and living in Skåne in southern Sweden were recruited using registry information from the Swedish Tax Agency. They were invited by post and visited a research centre for anthropometric measurement, spirometry and blood sampling. There were no exclusion criteria except understanding information in Swedish [15]. All participants with measurements on proteomics, lung function and confounders were included in this study, and the final cohort comprised 645 participants.
Measurements and questionnaires
Trained staff collected blood samples. Participants were fasting for ≥6 h for EpiHealth and overnight for MOS. The blood samples were immediately centrifuged and frozen at −80°C until analysis. Participant height (cm) and weight (kg) were measured. Information about smoking status and physical activity was obtained through questionnaires. Smoking status was divided into three categories: never-smokers, previous smokers and current smokers. Pack-years were calculated as numbers of cigarettes per day/20 × years of smoking. Physical activity was also divided into three categories: sedentary (physical activity ≤2 h/week (no sweating)), moderate (regular physical activity 1–2 × 30 min/week (sweating)) and high (regular physical activity ≥3 × 30 min/week (sweating)). CVD was assessed from questionnaires in EpiHealth and defined as any of the following self-reported conditions: previous myocardial infarction, stroke, heart failure, atrial fibrillation or angina pectoris. Similarly, diabetes, hypertension and hyperlipidaemia were questionnaire-assessed.
Spirometry
In EpiHealth, a simplified resting lung function test was performed using a MiniSpir spirometer (Medical International Research, Waukesha, WI, USA) and in MOS, a lung function test was performed using a MasterScope spirometer (JAEGER, Washington, USA). Spirometry was conducted in accordance with the instructions of the American Thoracic Society/European Respiratory Society [16]. At least three manoeuvres were performed, with the highest value of FEV1 and FVC recorded. The Global Lung Function Initiative (GLI) equations were used to calculate predicted FEV1 and FVC adjusted for age, sex, height and ethnicity [17], with FEV1/FVC ratios presented as absolute values.
Proteomics
A total of 276 preselected proteins were analysed in three OLINK Multiplex panels: CVD II, CVD III and metabolism. These assays are based on the proximity extension assay technology [18], in which each sample can bind to any of 92 oligonucleotide-labelled antibody probe pairs. The characteristics, validation and coefficients of variance for the panels are available online: https://www.olink.com/resources-support/document-download-center/.
The proteomics analyses were conducted by SciLifeLab (Uppsala, Sweden) in a random subgroup (n=2472) within the Uppsala portion of the EpiHealth cohort. Proteins with ≥15% of the values below limit of detection (LOD) were excluded (n=4 for CVD II, n=1 for CVD III and n=29 for metabolism). Valid proteomics data were obtained from 2360, 2466 and 2410 individuals using the CVD II, CVD III and metabolism panels, respectively. After exclusion of individuals with missing data on covariates, the final dataset encompassed 2229 individuals (figure 1).
Flowchart of subjects included in the analyses. MOS: Malmö Offspring Study.
OLINK Bioscience (Uppsala, Sweden) did the analyses in a subgroup of the MOS cohort (n=923; consecutive subjects from 6 March 2013 to 17 June 2015). Proteins with ≥15% of the values below LOD were excluded (n=7 for CVD II, n=4 for CVD III and n=20 for metabolism). Valid proteomics data were obtained from 915, 922 and 919 individuals using the CVD II, CVD III and metabolism panels, respectively. When the three proteomics data files were combined with the questionnaire data, the cohort comprised 750 individuals with complete data on lung function and proteomics. After exclusion of individuals with missing data on covariates, the final dataset encompassed 645 individuals (figure 1).
Ethical permission
All individuals gave informed consent, also covering biobank storing. EpiHealth was approved by the Ethics Board in Uppsala (Dnr 2010/402) and MOS by the Regional Ethics Committee in Lund (Dnr 2012/594). This study was approved by the Ethics Board in Uppsala (Dnr 2019-03968).
Statistical analyses
Baseline continuous variables are presented as mean±SD and categorical variables as n (%). The EpiHealth cohort was used for discovery and the MOS cohort for replication.
For the discovery cohort, a total of 2229 subjects and 242 proteins on 28 plates were analysed. Proteins were used as independent variables and spirometry values (FEV1, FVC and FEV1/FVC ratio) as dependent variables.
Proteins with <15% of the values below LOD were imputed with LOD/sqrt(2). Then, the protein values were pre-processed by adjusting for storage time and plates to remove the potential non-random effects created thereby. The standardised residuals (z-scores) of each protein were used in further statistical analyses rather than the observed protein values.
The results were summarised as arithmetic means and β coefficients with 95% confidence intervals (95% CI). Multiple linear regressions were run repeatedly using one of the 242 proteins as a predictor, with age, sex (binary), body mass index (BMI), smoking status (three categories), pack-years and self-reported physical activity (three categories) as confounders. For sensitivity analyses, individuals with known CVD, diabetes or obesity (BMI ≥30) were excluded. This multivariable linear regression model was adjusted for hypertension, hyperlipidaemia and waist circumference (normal versus high) instead of BMI.
An interaction analysis was made for the effects of current versus never and previous versus never smoking on the association between each protein and lung function (FEV1, FVC and FEV1/FVC ratio) in the discovery cohort.
The p-values were adjusted for multiple testing using the Benjamini–Hochberg method [19] to control for a false discovery rate of 5%. Proteins that were significantly associated with lung function were tested in the replication cohort using an identical approach. In the replication study, a nominal significance threshold of 0.05 was used. All analyses were performed using R version 4.1.2.
Results
Baseline characteristics are shown in table 1. The participants in EpiHealth were older and exercised to a lesser extent than the individuals in MOS. A similar proportion of BMI ≥30 (obesity) was found in both cohorts, while the prevalence of central obesity appeared to be larger in EpiHealth. There was a difference in smoking patterns between the two cohorts with a slightly higher prevalence of current smokers in MOS. The prevalence of diabetes mellitus was low in both cohorts, while hypertension and hyperlipidaemia were slightly higher in EpiHealth than MOS. On the other hand, MOS had more individuals with known CVD than EpiHealth. Finally, more individuals had a normal spirometry in EpiHealth than in MOS. The prevalence of different pathological spirometry patterns is presented in table 1.
Baseline characteristics for the EpiHealth and Malmö Offspring Study (MOS) cohorts
Associations between proteomics and lung function – discovery cohort EpiHealth
In the discovery cohort, 19 of 242 proteins were significantly associated with FEV1 and 24 proteins with FVC. Of the 19 proteins, 14 were negatively associated with FEV1 (table 2). Of the 24 proteins, 17 were negatively associated with FVC (table 3). Two proteins, metalloproteinase inhibitor 4 (TIMP4) and hydroxy acid oxidase 1 (HAOX1), were associated with FEV1/FVC ratio (regression coefficient (95% CI) −0.76 (−1.13– −0.39) and 0.69 (0.33–1.05), p=0.02, respectively). All analyses were adjusted for age, sex, BMI, physical activity, smoking status (never-smokers versus previous smokers versus current smokers) and pack-years. For smoking status, never-smokers were used as the reference group.
Cross-sectional associations between plasma proteins and FEV1 % predicted in EpiHealth (discovery cohort) and MOS (replication cohort)
Cross-sectional associations between plasma proteins and FVC % predicted in EpiHealth (discovery cohort) and MOS (replication cohort)
Associations between proteomics and lung function – replication cohort MOS
Of the 19 proteins associated with FEV1 in the discovery cohort, five could be replicated: plasma fatty acid-binding protein 4 (FABP4), IL-1 receptor antagonist (IL-1RA), IL-6, leptin and paraoxonase 3 (PON3). Increased levels of four of these proteins were associated with lower FEV1, while increased levels of PON3 were associated with higher levels of FEV1 (table 2).
Of the 24 proteins associated with FVC in the discovery cohort, nine could be replicated: agouti-related protein (AGRP), FABP4, fibroblast growth factor 21 (FGF21), insulin-like growth factor-binding protein 2 (IGFBP2), IL-1RA, IL-6, leptin, PON3 and receptor for advanced glycation end products (RAGE). Increased levels of FABP4, FGF21, IL-1RA, IL-6 and leptin had negative associations with FVC (table 3). In contrast, increased levels of AGRP, IGFBP2, PON3 and RAGE were associated with higher FVC (table 3). No proteins were associated with FEV1/FVC ratio in the replication cohort (supplementary table S2).
Sensitivity analyses – EpiHealth
In EpiHealth, the sensitivity analyses showed that 11 out of 19 proteins were significantly associated with FEV1 after excluding individuals with known CVD, diabetes or obesity (BMI ≥30) and additional adjustment for hypertension, hyperlipidaemia and waist circumference (table 4). For FVC, using the same exclusion criteria 18 out of 25 proteins were significantly associated with FVC. In addition, five proteins became significantly associated with FVC only after exclusion of these individuals (table 5).
Sensitivity analysis in EpiHealth
Sensitivity analysis in EpiHealth
Interaction between smoking status and proteins
When we accounted for multiple testing, no significant interactions with smoking status could be seen in the discovery cohort.
Discussion
Principal findings
In this paper, we studied the associations between proteins known for involvement in CVD and metabolism and lung function assessed with spirometry. We used multivariable models in a discovery and replication approach and found that higher levels of four proteins (FABP4, IL-1RA, IL-6, leptin) were associated with lower FEV1, and higher levels of five proteins (FABP4, FGF21, IL-1RA, IL-6, leptin) were associated with lower FVC. Higher levels of four proteins (AGRP, IGFBP2, PON3, RAGE) were associated with higher FVC. Increased levels of PON3 were also associated with higher FEV1. No proteins were associated with FEV1/FVC ratio consistently in both the discovery and validation cohort, and we could not find any interaction with smoking status.
After exclusion of individuals with known CVD, diabetes or obesity from a sensitivity analysis in EpiHealth, 11 out of 19 proteins were statistically associated with FEV1, whereas 18 out of 24 proteins were significantly associated with FVC. In addition, five proteins were only significantly associated with FVC in the sensitivity analysis.
Comparison with literature, individual proteins and potential mechanisms
This study had some novel findings. PON3 was positively associated with both FEV1 and FVC, IGFBP2 was positively associated with FVC, and FGF21 was negatively associated with FVC.
PON3 is an antioxidant enzyme protecting against low-density lipoprotein oxidation and was associated with both FEV1 and FVC in our analyses, but only with FVC in our sensitivity analysis. To the best of our knowledge, PON3 has not previously been studied in the lung function context. However, increased PON3 levels have previously been reported in both coronary and peripheral artery disease [20]. A closely related protein, PON1, shares properties with PON3 and has been studied more extensively in relation to lung function. Lower levels have been associated with COPD [21], asthma and poorer asthma control [22]. Reduced levels of PON1 could indicate lower antioxidative protection [21]. Since both PON1 and PON3 seem to have protective roles, we speculate that the increased levels of PON3 in this study reflect how the body copes with increased oxidative stress and inflammation early in the disease process.
IGFBP2 was positively associated with FVC in both cohorts and in the sensitivity analysis in EpiHealth. IGFBP2 is a known tumour biomarker, expressed in many different types of cancers [23]. No previous study has reported any associations between increased levels of IGFBP2 and lung function in a community-based setting. However, Guiot et al. [24] reported higher levels of IGFBP2 in idiopathic pulmonary fibrosis, although no association with lung function was found in this patient group. The signal may differ between a community-based setting and subjects with idiopathic pulmonary fibrosis, which warrants further research.
FGF21 was significantly associated with FVC in both cohorts and in the sensitivity analysis in EpiHealth. FGF21 has not previously been reported in the lung function context. FGF21 seems to be involved in protecting against CVD, with many potential pathways, e.g., endothelial function, inflammation and oxidative stress [25]. However, both lower and higher values of FGF21 have been associated with overall and cardiovascular mortality [26]. To summarise, we can merely speculate why increased FGF21 levels were associated with lower lung volume in the present study.
Some additional proteins from the analyses are worth mentioning in this section. Unsurprisingly, many of the proteins associated with lower spirometry values in this study are involved in inflammatory pathways. IL-6, for example, is a well-known marker of systemic inflammation and has previously been associated with lower FEV1 and FVC in otherwise healthy individuals [27]. However, despite IL-6's close association with inflammation, these results are not necessarily evidence of inflammation as damaged lung epithelial cells can increase IL-6 without chronic systemic inflammation [28]. Given this situation, it is hardly surprising that IL-6 levels are elevated, but we cannot be entirely certain how much inflammation and/or damaged endothelial cells contributes. The association between increased levels of IL-6 and lower FEV1 and FVC respectively was seen in both cohorts and in the sensitivity analysis in EpiHealth.
IL-1RA is a member of the IL-1 cytokine family and a natural inhibitor of the proinflammatory activities of IL-1α and IL-1β, modulating a variety of IL-1-related immune and inflammatory responses. IL-1RA completely inhibits the activity of IL-1α and IL-1β by binding to receptor IL-1R1 and preventing its association with the co-receptor IL-1RAP for signalling [29]. Lowered IL-1RA levels have been found in asthma [30] and COPD [31], causing increased IL-1 activity due to the lack of adequate anti-inflammatory regulation. Therefore, we might expect an association between lower IL-1RA levels and lower lung function based on these results from people with established airway disease. However, our results were the opposite, and we speculate that higher levels of IL-1RA could reflect how the body copes with a heightened inflammatory state early in the disease process. Interestingly, the association was not significantly associated with FEV1 in the sensitivity analysis, whereas the association with FVC was unchanged.
Leptin is a hormone secreted mainly by human adipose tissue cells, proportionally to the adipose tissue depot. Formerly thought to only modulate body weight through decreased appetite, leptin has other important roles in the body, including both the immune and the endocrine system [32]. We previously reported an association between leptin and FEV1 [13]. In this study, we replicated this association and extended it to FVC. The association does not seem to be driven by increased BMI, as the findings were consistent after BMI adjustment in our primary analysis, and after exclusion of individuals with obesity in our sensitivity analysis. Our results are supported by previous studies that leptin is not just a proxy for obesity. For example, Sin et al. [33] showed two decades ago that higher leptin levels were associated with lower FEV1 in non-obese individuals, and recently Peters et al. [34] have shown that in individuals with asthma, insulin resistance rather than obesity was associated with lower lung function values suggesting that higher insulin levels rather than high BMI are an important driver for lung function decline. Further studies need to establish which factors associated with obesity are really the drivers behind the accelerated decline in lung volume and airway obstruction.
FABP4 or adipocyte protein 2 is not only expressed in adipocytes, but also in other cells such as macrophages. Increased levels are associated with several disease states, e.g., insulin resistance, type 2 diabetes and hypertension. FABP4 increased with decreasing lung function in COPD-stable patients [35] and with decreasing FEV1 in our previous analyses [13]. The association between FABP4 and lower lung function could be mediated through matrix metalloproteinase (MMP9), as suggested in ischaemic stroke animal models [36]. However, MMP9 was not associated with any of the spirometry values in this study. Moreover, FABP4 was associated with FEV1 and FVC also after excluding individuals with known CVD, diabetes or obesity.
In our primary analysis, a higher level of RAGE was associated with higher FEV1 and FVC in EpiHealth, but the association could only be replicated for FVC. RAGE is highly expressed in the lungs, and our results contrasted with those of previous studies that linked increased levels of RAGE to respiratory diseases, e.g., COPD and asthma [37]. We can only speculate as to why increased levels of a proinflammatory protein were associated with lung volume but not airway obstruction in this study. However, it is important to note that the significant association for FEV1 and FVC in EpiHealth disappeared in the sensitivity analysis (tables 4 and 5).
Previous research has linked GDF-15 to both COPD [38] and COPD with concomitant subclinical coronary artery disease [39]. In a previous study in community-based cohorts, we reported an association between higher levels of GDF-15 and lower FEV1 [13]. In this study, we did not find any consistent association between GDF-15 and lung function in both cohorts, but we found an association with FEV1 in EpiHealth, and the signal was also there in the sensitivity analysis (table 4). EpiHealth encompassed a much larger cohort and participants were of a narrower age range and older than participants in MOS. It is possible that GDF-15 levels are associated with lung function later in the disease process, and no signal could therefore be found in the younger MOS cohort.
Clinical implications and future directions
The effect sizes of the associations between proteins and lung function indicate that they are unlikely to be of clinical importance in a community-based population. This is also expected because the overall goal of this study was not to find a biomarker for use in clinical practice, but rather to find early signals of associations that could have pathophysiological implications in the future. Moreover, it is likely that several proteins might relate to the same pathophysiological pathway, for instance by causing poorer antioxidative protection. As several of the proteins are involved in CVD and metabolism, some could be interesting in analyses of common disease mechanisms. Lastly, it should be kept in mind that finding individual proteins and implementing them into clinical practice is not an easy task, for example, >100 biomarkers were in clinical use in 2015, yet over 150 000 articles on biomarkers have been published [40]. Still, it is necessary to continue the search for proteins that could be of clinical use in this cardiopulmonary context. Especially since there seems to be a point of no return in the disease process, making early discovery of the individuals at risk the only way forward with today's treatment standards [41].
Strengths and limitations
Study strengths include the use of a discovery and replication approach using two different study cohorts and including many proteins. Moreover, we did a sensitivity analysis after excluding individuals with known CVD, obesity and diabetes. The main limitation is the cross-sectional design. A prospective cohort study with repeated measurements of lung function and proteomics data would provide both more information about fluctuations over time and stronger evidence of potential causal relationships.
When testing the associations between many proteins using several different measurements of lung function, there is an increased risk of spurious associations. However, we adjusted our results for a false discovery rate of 5% in the discovery cohort and nominal p-values in the replication cohort. The rationale was that we wanted to balance the risks of Type 1 and Type 2 errors and did not want to be overly conservative.
Conclusions
Several CVD-linked proteins were associated with FEV1 and FVC but not with FEV1/FVC ratio, suggesting that the relationships are mainly with lung volume, not airflow obstruction. That increased levels of several proteins are associated with better lung function warrants further studies.
Supplementary material
Supplementary Material
Please note: supplementary material is not edited by the Editorial Office, and is uploaded as it has been supplied by the author.
Table S1a 00321-2022.tableS1a
Table S1b 00321-2022.tableS1b
Footnotes
Provenance: Submitted article, peer reviewed.
Support statement: We thank the Swedish Research Council for supporting the strategic research network Epidemiology for Health and the EpiHealth screening cohort. The Malmö Offspring Study was funded by the Research Council of Sweden (grant 521-2013-2756), the Heart and Lung Foundation (grant 20150427) and funds obtained from Region Skåne (ALF) for P.M. Nilsson. In addition, funding has been obtained from Ernhold Lundströms Stiftelse, donations to the Skåne University Hospital and the European Research Council (for M. Orho-Melander). Funding information for this article has been deposited with the Crossref Funder Registry.
Conflict of interest: A. Rydell has received lecturing fees from AstraZeneca. K. Lisspers has received lecturing fees from AstraZeneca, Novartis and Boehringer Ingelheim; and payments for advisory board membership from AstraZeneca, Boehringer Ingelheim and GlaxoSmithKline. S. Zaigham has received a research grant from Magnus Bergvalls Stiftelse. P.M. Nilsson has received lecturing fees from Novartis, Novo Nordisk, Amgen and Boehringer Ingelheim. J. Ärnlöv has received lecturing fees from AstraZeneca and Novartis, and served on advisory boards for AstraZeneca and Boehringer Ingelheim on subjects unrelated to this study. The remaining authors have nothing to disclose.
- Received June 30, 2022.
- Accepted December 4, 2022.
- Copyright ©The authors 2023
This version is distributed under the terms of the Creative Commons Attribution Non-Commercial Licence 4.0. For commercial reproduction rights and permissions contact permissions{at}ersnet.org