Effects of cigarette smoking on cardiovascularrelated protein profiles in two community-based cohort studies

Transcription

1 Effects of cigarette smoking on cardiovascularrelated protein profiles in two community-based cohort studies Abstract Background: Cardiovascular diseases account for the largest fraction of smoking-induced deaths. Studies of smoking in relation to cardiovascular protein markers can provide novel insights into biological effects of smoking. Objectives: To study the associations between cigarette smoking and 80 protein markers known to be related to cardiovascular diseases in two community-based cohorts, the Prospective Study of the Vasculature in Uppsala Seniors (PIVUS, n=969, 50% women, all aged 70 years) and the Uppsala Longitudinal Study of Adult Men (ULSAM, n=717, all men aged 77 years). Material and Methods: Smoking status was self-reported and defined as current smoker, former smoker and never-smoker. Levels of the 80 proteins were measured using the proximity extension assay and analyzed in relation to smoking status in linear regression models. Results: We found 30 proteins to be significantly associated with current cigarette smoking in PIVUS (FDR<5%); and ten were replicated in ULSAM (P<0.05). Matrix metalloproteinase-12 (MMP-12), growth/differentiation factor 15 (GDF-15), urokinase plasminogen activator surface receptor (upar), TNF-related apoptosis-inducing ligand receptor 2 (TRAIL-R2), lectin-like oxidized LDL receptor 1 (LOX-1), hepatocyte growth factor (HGF), matrix metalloproteinase-10 (MMP-10) and matrix metalloproteinase-1 (MMP-1) were positively associated, while endothelial cell-specific molecule 1 (ESM-1) and interleukin-27 subunit alpha (IL27-A) showed inverse associations. Conclusion: The findings of the present study suggest that cigarette smoking may interfere with several essential parts of the atherosclerosis process, as evidenced by associations with protein markers representing endothelial dysfunction, inflammation, neointimal formation, foam cell formation and plaque instability.

2 Abbreviations CHF Congestive heart failure COPD Chronic obstructive pulmonary disease CRP C-reactive protein CVD Cardiovascular disease DAG Directed acyclic graph ECM Extracellular matrix ESM-1 Endothelial cell-specific molecule 1 FDR False discovery rate GDF-15 Growth/differentiation factor 15 HGF Hepatocyte growth factor IL27-A Interleukin-27 subunit alpha LOX-1 Lectin-like oxidized LDL receptor 1 MMP-1 Matrix metalloproteinase-1 MMP-10 Matrix metalloproteinase-10 MMP-12 Matrix metalloproteinase-12 oxldl Oxidized low-density lipoprotein PAD Peripheral artery disease PEA Proximity extension assay PIVUS Prospective Study of the Vasculature in Uppsala Seniors TRAIL-R2 TNF-related apoptosis-inducing ligand receptor 2 ULSAM Uppsala Longitudinal Study of Adult Men UPAR Urokinase plasminogen activator surface receptor VSMC Vascular smooth muscle cell 1

3 Introduction Smoking is a major threat to public health worldwide. It causes about 5 million deaths each year according to WHO, accounting for 12% of the total mortality in adults above 30 years of age (1). Additionally, second-hand smoking causes 600,000 deaths each year (2). Globally, the number of smoking-attributable deaths is estimated to increase to 8 million annually within the next two decades (1). The prevalence of daily tobacco smoking in the population above 15 years of age is estimated to 31% for men and 6% for women in the world (3). Also, the smoking prevalence varies widely between different regions of the world. The prevalence is estimated to as high as more than 50% for men in Armenia, Indonesia, Kiribati, Laos, Papua New Guinea, Russia and Timor-Leste, and as low as 1% or lower for women in Azerbaijan, Algeria, Cameroon, Eritrea, Ethiopia, Gambia, Lesotho, Libya, Morocco, Oman, Sri Lanka and Sudan (3). The diseases contributing to smoking-attributable mortality can be classified either as communicable or non-communicable (1). Among communicable diseases, tuberculosis and lower respiratory infections are the most common causes of death. Among non-communicable diseases, cardiovascular diseases (CVD), malignant neoplasms and respiratory diseases are the main causes of smoking-induced deaths out of which CVD accounts for the largest fraction of total smoking-induced deaths (1). In fact, CVD causes more smoking-induced deaths than all communicable diseases together, all malignant neoplasms together and all respiratory diseases together (1). Epidemiology of cardiovascular diseases attributable to smoking In the year 2000, 1.62 million cardiovascular deaths or 11% of all cardiovascular deaths around the world were estimated to be attributable to smoking. The majority (72%) of smoking-induced cardiovascular deaths occurred among men. Coronary heart disease was the most common cause of death and accounted for 54% of all smoking-induced cardiovascular deaths. The second most common cause of death was cerebrovascular disease which accounted for 25% of all smoking-induced cardiovascular deaths (4). In fact, all major forms of CVD have been shown to be causally associated with smoking. Other common diagnoses include congestive heart failure (CHF), peripheral artery disease (PAD) and aortic aneurysm (5). Because of the vast amount of cardiovascular deaths caused by smoking, it is important to understand the 2

4 underlying biological mechanisms linking smoking to CVD in order to improve preventive and therapeutic strategies. Development of atherosclerosis Atherosclerosis is an inflammatory process that mainly involves large and medium-sized arteries. Most CVD is the result of atherosclerosis. The process starts with endothelial dysfunction which results in increased endothelial permeability. Monocytes are recruited to the sites of endothelial lesions, where they differentiate into macrophages in the subendothelial space. These macrophages can transform into foam cells by accumulating cholesterol and oxidized lipids (6, 7). Circulating lipoproteins in plasma, such as low-density lipoprotein, get trapped into arterial walls whereby they are no longer protected from oxidation (7). The oxidized lipoproteins cannot be recognized by regular lipoprotein receptors. Instead, they bind to so-called scavenger receptors which are not controlled by intracellular cholesterol level and the lipid uptake can accelerate without down-regulation (7). Thereby a chronic inflammatory state is generated which enhances and maintains recruitment of smooth muscle cells and activation of lymphocytes. An atherosclerotic plaque forms progressively and may ultimately lead to occlusion of the affected artery. The ongoing inflammatory process also destabilizes the plaque and may eventually result in plaque rupture followed by a thromboembolic event (6, 7). Constituents of cigarette smoke More than 4,000 compounds such as carbon monoxide, nicotine, reactive nitrogen species and reactive oxygen species have been found in the aerosol of cigarette smoke (8). Carbon monoxide reduces oxygen supply to the heart and other tissues (6). The effects of nicotine are primarily mediated by sympathetic activities including increased heart rate, increased cardiac contractility, transiently increased blood pressure and coronary artery constriction (6). However, most of the adverse effects of cigarette smoking on the development of CVD is thought to be caused by oxidizing chemicals, including reactive nitrogen species and reactive oxygen species (6, 8). These oxidants contribute to inflammation, endothelial dysfunction, oxidation of low-density lipoprotein and platelet activation, which are all essential parts of atherosclerosis and thrombosis (8). 3

5 Effects of cigarette smoke on the cardiovascular system Existing evidence indicates that cigarette smoke interferes with both atherosclerosis and thrombosis (6, 8). In the initial phase of atherosclerosis, cigarette smoke may induce endothelial dysfunction mainly by depressing vasodilatation and by enhancing prothrombotic activities (6, 8). Later during the plaque formation phase, the plaques of cigarette smokers have demonstrated higher lipid deposition, increased disruption of extracellular matrix, enhanced recruitment of inflammatory cells, increased intraplaque haemorrhage, as well as increased growth of atheromatous cores, compared to the plaques of non-smokers (8). All these characteristics make the plaques more vulnerable and more likely to rupture giving rise to thromboembolic events. Additionally, cigarette smoke has been shown to trigger platelet activation and aggregation, as well as an overall hypercoagulative state in the vascular system. Finally, cigarette smoke has been linked to impaired fibrinolysis and increased blood viscosity (8). Altogether, cigarette smoking contributes to increased risk for the development of CVD by accelerating the progression of atherosclerosis and thrombosis. In addition, cigarette smoking is responsible for increased mortality in acute events of CVD due to increased plaque vulnerability (8). Therefore, it is crucial to make global efforts for prevention and cessation of cigarette smoking. Previous studies have demonstrated reduced risks for myocardial infarction and coronary heart disease among former smokers compared to current smokers. However, the risks may remain slightly elevated, compared to never-smoker, for more than a decade after smoking cessation (5). Importantly, lower smoking cessation rates have been observed among people with low socioeconomic status, determined by education, income and occupational status (9). In order to optimize intervention strategies of smoking cessation, it is crucial to take socioeconomic status into consideration, and additional efforts are needed to target the economically disadvantaged people specifically. A case-control study based on more than 29,000 individuals found smoking to be the second most modifiable risk factor of myocardial infarction, after abnormal low-density lipoprotein: high-density lipoprotein ratio, and followed by psychosocial factors, diabetes mellitus, hypertension, abdominal obesity, consumption of alcohol, physical activity and consumption of fruits and vegetables (10). While statins have been established as part of primary and secondary prevention of CVD targeting abnormal lipid profiles (11), there is currently no 4

6 medical treatment to target effects of smoking, beside the treatments used for facilitating smoking cessation. Proteomic studies Effects of cigarette smoke on the cardiovascular system can be measured as alterations of protein structure and/or protein expression. The proteome is defined as the complete set of proteins present in an cell, tissue or organism at one time, and it can be analyzed by different technologies (12). Proteomic technologies make it possible to perform systematic investigations of the responses of relevant proteins to environmental stimuli (7). For the in-gel technologies, a mono- or two-dimensional polyacrylamide gel electrophoresis is used for protein separation. This is followed by mass spectrometry or tandem mass spectrometry used for sequence identification of selected proteins. There are also several non-gel technologies with capacity to add more proteins to the analysis than the in-gel technologies (12). The present study used a non-gel technology based on the proximity extension assay (PEA), which is a homogenous multiplex immunoassay with high sensitivity and high specificity (13). Usually, the main challenge for developing immunoassays is to avoid antibody cross-reactivity. The high specificity of PEA is achieved by the proximal requirement of the antibodies to give rise to a signal, and by using specific primers in the quantitative PCR detection. PEA remains highly specific even when targeting at up to 92 proteins simultaneously, which enables largescale screening of protein markers (13). Effects of smoking have previously been investigated in proteomic studies in cellular and animal models, as well as in the human setting (12). The human proteomic studies have mostly been conducted on tissues from the respiratory system such as bronchoalveolar lavage (14), and focused on lung diseases such as chronic obstructive pulmonary disease (COPD), emphysema and lung cancer. On the other hand, studies investigating effects of smoking on proteins involved in CVD are scarce. To the best of our knowledge, such proteomic studies focusing on CVD-related proteins in association to smoking status have only been performed in two small studies based on samples from platelets (15) and arterial wall biopsies (16). There is a lack of systematic analyses of smoking-associated alterations in CVD-related proteins based on serum or plasma from adequately sized cohorts based on the general population. 5

7 Aims The primary aim was to study associations of CVD-related protein markers to current smoking by comparing protein profiles of current smokers to non-smokers. The secondary aims were to assess if associations were present also in former smokers and if there was evidence of a doseresponse effect, whether these associations differed by sex, and if they were similar when restricting analyses to individuals without prior CVD. Materials and Methods Study design This was a cross-sectional study in two community-based cohorts, the Prospective Study of the Vasculature in Uppsala Seniors (PIVUS) and the Uppsala Longitudinal Study of Adult Men (ULSAM). Study samples The PIVUS study has been described previously (17) and on the Internet ( All 70-year old individuals residing in Uppsala, Sweden, in 2001 to 2005 were eligible for the study. The invitation was sent to 2,025 individuals from the general population in a randomized order by mail, within two months of their 70 th birthday. There were 1,016 individuals (50.2 %) of the invited that participated and 510 (50.2 %) of them were women. We excluded 25 individuals due to unsuccessful proteomic profiling (not passing the quality control), 21 individuals due to C-reactive protein (CRP) levels above 20 mg/l (to exclude individuals with possible ongoing infection and other inflammatory processes) and one individual due to missing information of smoking status. Hence, the eligible sample size for the current study was 969 individuals. The ULSAM study has been described previously (18) and on the Internet ( The study was initiated in by inviting all men living in Uppsala, Sweden, born between 1920 and 1924 to investigation. The invitation letter was sent to 2,841 men and 2,322 (81.7 %) of them participated in the initial examination. The participants have been re-examined at multiple time points. The current study was based on the re-examination at age 77, when 748 of the original participants had died, and another 176 men were not eligible for other reasons. From 1997 to 2001, the remaining 1,398 6

8 individuals were re-invited, and 839 (60.0 %) of them participated. We excluded 77 individuals due to unsuccessful proteomic profiling (not passing the quality control), 21 individuals due to CRP levels above 20 mg/l and 24 individuals due to missing information of smoking status. Hence, the eligible sample size for the current study was 717 men. Assessment of clinical covariates The collection of clinical covariates was performed by similar and standardized methods in PIVUS and ULSAM. Basic information including medical history and regular medication was self-reported by answering a questionnaire. Other information including anthropometrical measurements, blood pressure and fasting blood samples were obtained by the investigators. All participants went through an overnight fast prior to the investigation, during which neither smoking nor medication was allowed. Traditional lipid variables and fasting blood glucose were measured by standard laboratory techniques. Waist and hip circumferences were measured in the supine position. Blood pressure was measured by a calibrated mercury sphygmomanometer to the nearest even mmhg after at least 30 min of rest. The mean of three and two recordings of blood pressure was used in PIVUS and ULSAM, respectively. Definition of smoking exposure The information of smoking status was self-reported through a questionnaire, and defined as current smoker, former smoker or never-smoker, in PIVUS and ULSAM. In both cohorts, an individual was defined as former smoker if he/she reported to ever have smoked regularly and stopped smoking in his/her lifetime. The definitions in ULSAM were based on information from both the age 70 and 77 investigations (as the age 77 investigation only included questions about current smoking and smoking in the past five years). In addition, smoking was assessed as pack years in PIVUS; one pack year was defined as an average consumption of one pack of cigarettes per day during one year. Proteomic analysis In the present study, 92 circulating proteins known to be related to cardiovascular diseases were analyzed simultaneously using a high-throughput technique, the OLINK Proseek Multiplex CVD I kit (Olink Bioscience, Uppsala, Sweden). The kit using PEA technology included 92 oligonucleotide-labeled antibody probe pairs allowed to bind to their respective target present in the sample (13, 19, 20). Compared to conventional multiplex immunoassays, the 7

9 PEA technique has an exceptionally high specificity as only correctly matched antibody pairs are able to generate a signal. PEA uses pairs of antibodies equipped with DNA reporter molecules. After correctly binding to their targets, the antibodies form new DNA amplicons each ID-barcoding their respective antigens. Further on, the quantification of the amplicons is performed by using a Fluidigm BioMark HD real-time PCR platform (20). The Proseek Multiplex CVD I has proved high reproducibility and repeatability. The mean intra-assay and inter-assay coefficients of variation are 8% and 12%, respectively, while the mean intersite variation is 15% (13). The output is presented as normalized protein expression data where a high value corresponds to a high protein concentration, but not an absolute value of the protein concentration. Further on, 12 proteins were removed from the analyses due to a call rate < 85% in PIVUS and/or ULSAM, i.e. less than 85% of the individuals had a valid measurement of the particular protein. These proteins were IL-4, melusin, BNP, Beta-NGF, SIRT2, NEMO, mamp, PTX3, NT-pro-BNP, MMP-7, cystatin-b and heat shock 27 kda protein. Thus, 80 proteins remained in the analyses of both study cohorts (listed in Table 2 and Supplementary Table 1). Individuals with excess missingness (more than 5% and 3% missing protein values in PIVUS and ULSAM, respectively, due to slightly different distributions of missingness in these cohorts) based on a histogram were excluded. Each protein was normalized by plate (by setting the mean=0 and standard deviation=1 within each plate) and further by storage time of each blood sample (correction based on the observed values and predicted values from a spline model). Statistics The distribution of protein values and clinical covariates were evaluated using the Shapiro-Wilk normality test and histograms. Non-normality distributed variables were transformed using 1/x 2 -transformation for fasting glucose in both PIVUS and ULSAM, while pack years, alcohol consumption and physical activity were analyzed as quartiles in PIVUS. All protein variables and other covariates were considered to be normally distributed. All covariates considered as confounders that had missing observations were imputed using multiple imputation methods. The imputations were based on all clinical characteristics in respective study (listed in Table 1). 8

10 Linear regression models were used for all analyses. In the primary analysis, smoking status defined as current smoker or non-smoker (i.e. former or never smoker) was used as the independent variable, and the 80 proteins were used as dependent variables in separate models (one per protein marker). The PIVUS study was a priori designated discovery cohort as it was larger and had equal proportions of males and females. For protein markers with a false discovery rate (FDR; estimated using the Benjamini & Hochberg method(21)) < 5% in PIVUS, we attempted replication in ULSAM. FDR is defined as the expected proportion of rejected null hypotheses that are falsely rejected (21). Protein markers showing a nominally significant P-value (P<0.05) in ULSAM were considered as successfully replicated. This replication strategy has previously been shown to be conservative (22) and indeed, the risk of false positive findings in the replication stage (validated false discovery rate, vfdr) was calculated to 0.4% using these methods (15), that are also available as an online application ( The analyses of the proteomic data were performed in an age- and sex-adjusted model (only age in ULSAM as all participants were males), and a multivariable-adjusted model, in PIVUS and ULSAM, respectively. The multivariable-adjusted models were considered as main models, both when deciding which protein markers to pursue for replication, and which to annotate as significantly associated. For the multivariable-adjusted model, confounders were considered using a directed acyclic graph (DAG; Figure 1) informed by literature search. The fully adjusted model in PIVUS included age, sex, alcohol consumption, education level, body mass index (BMI), total daily energy intake, physical activity, fasting glucose level, history of CVD, chronic obstructive pulmonary disease (COPD) and diabetes mellitus as covariates. CVD was defined as myocardial infarction, stroke and/or congestive heart failure. In ULSAM, the same covariates were used, except sex (all were males) and total daily energy intake (unavailable). 9

11 Figure 1. Covariates in multivariable-adjusted models (PIVUS and ULSAM)* The green lines connect the mediators with the exposure (smoking) and the outcome (protein level), which means that CRP, triglycerides and HDL mediate effects of smoking on protein levels. The pink lines connect the confounders with the exposure (smoking) and the outcome (protein level), which means that the confounders interfere with smoking and protein levels by independent pathways. * Total daily energy intake was not included as covariate in multivariable-adjusted model in ULSAM (as that information was unavailable). Abbreviations: CVD, cardiovascular disease; COPD, chronic obstructive pulmonary disease. In addition, four sets of secondary analyses were performed based on proteins that were significantly replicated in the primary analysis. First, current smokers and former smokers were compared to never-smokers, respectively, to investigate whether smoking-induced effects remained after smoking cessation. Second, the dose-response relationships of smoking with protein levels in current and former smokers within PIVUS was assessed using pack years in quartiles as the independent variable and protein levels as dependent variables in separate models. Third, sex-stratified analyses were performed within PIVUS to look at gender-specific 10

12 effects. Finally, a sensitivity analysis was performed where all individuals with previous CVD were excluded to indicate whether the associations were driven by existing atherosclerotic disease. All analyses were performed using STATA 14.1 (College Station, TX, USA). Ethical considerations This study aimed at gaining new knowledge about effects of cigarette smoking on the cardiovascular system at proteome level. The data was based on blood samples, self-reported information and physical examination performed by the investigators. Already collected blood samples were analyzed for proteins in this study and other required biomarkers including lipid levels and fasting glucose. A random identification number was generated for each participant. The analyses were performed at group level, and no individual information was studied specifically. Thus, potential harms to the integrity of the participants were minimal. All participants in PIVUS and ULSAM gave informed written consent. Therefore, the potential benefits of this study were considered to outweigh the potential harms. Both studies were approved by the Ethics Committee of X and the ethical review numbers are: PIVUS: X ULSAM: X 11

13 Results The clinical characteristics of the study samples are presented in Table 1. In PIVUS, 11% of the participants were current smokers (12% of women and 10% of men) and 41% were former smokers (40% of women and 54% of men). ULSAM participants had similar proportions of current smokers (8%) and former smokers (41%). Table 1. Clinical characteristics of the study samples* Variable PIVUS ULSAM N=969 N=717 Age 70.2 (0.17) 77.5 (0.76) Sex (% females) 50 0 Current smokers (%) 11 8 Former smokers (%) HDL cholesterol (mmol/l) 1.5 (0.4) 1.3 (0.3) LDL cholesterol (mmol/l) 3.4 (0.9) 3.5 (0.9) Serum triglycerides (mmol/l) 1.3 (0.6) 1.4 (0.7) Lipid lowering medication (%) BMI (kg/m 2 ) 27.0 (4.2) 26.3 (3.5) Waist circumference (cm) 91 (11) 96 (10) Waist/hip-ratio 0.90 (0.07) 0.93 (0.06) Systolic blood pressure (mmhg) 150 (23) 151 (20) Diastolic blood pressure (mmhg) 79 (10) 81 (10) Antihypertensive treatment (%) Fasting glucose (mmol/l) 5.3 (1.5) 5.9 (1.3) Prevalent diabetes (%) 8 8 History of myocardial infarction (%) 7 13 History of stroke (%) 3 10 History of congestive heart failure (%) 4 7 History of COPD (%) 8 4 * Values are presented as means (standard deviations) or proportions. History of asthma is included. Abbreviations: PIVUS, Prospective Study of the Vasculature in Uppsala Seniors; ULSAM, Uppsala Longitudinal Study of Adult Men; COPD, chronic obstructive pulmonary diseases. Primary analyses In PIVUS, 30 protein markers were found to be significantly associated with current cigarette smoking at a FDR < 5% (corrected overall critical P-value < ; Table 2). The rest of protein markers (50 proteins) with FDR > 5% are listed in Supplementary Table 1. 12

14 Table 2. Proteins significantly associated (FDR<5%) with current smoking in PIVUS (N=969) Age- and sex-adjusted Multivariable-adjusted Biomarkers N Beta 95% CI P-value N Beta 95% CI P-value Matrix metalloproteinase , E , E (MMP-12) Growth/differentiati on factor 15 (GDF , E , E-12 15) Urokinase plasminogen activator surface , E , E-11 receptor (upar) Interleukin-6 (IL-6) , E , E-10 TNF-related apoptosis-inducing ligand receptor , E , E-10 (TRAIL-R2) Lectin-like oxidized LDL receptor 1 (LOX-1) , E , E-09 Stem cell factor (SCF) , E , E-06 Vascular endothelial growth factor D , E , E-06 (VEGF-D) Hepatocyte growth factor (HGF) , E , E-06 Spondin-1 (SPON1) , E , E-06 Endothelial cellspecific molecule , E , E-06 (ESM-1) Prolactin (PRL) , E , E-05 Matrix metalloproteinase- 10 (MMP-10) , E , E-05 Kallikrein-6 (KLK6) , , C-X-C motif chemokine , , (CXCL16) Adrenomedullin (AM) , , T-cell immunoglobulin and mucin domain , , (TIM-1) C-X-C motif chemokine 1 (CXCL1) , , Leptin (LEP) , E , Ctd. 13

15 Interleukin-27 subunit alpha (IL , , A) Tumor necrosis factor ligand superfamily member , , (TNFSF14) Cathepsin D (CTSD) , , Galectin-3 (Gal-3) , , Monocyte chemotactic protein , , (MCP-1) Vascular endothelial growth factor A (VEGF-A) , , Myoglobin (MB) , , Galanin peptides (GAL) , , TNF-related activation-induced , , cytokine (TRANCE) Tumor necrosis factor receptor superfamily member , , (FAS) Matrix metalloproteinase-1 (MMP-1) , , Adjusted for age, sex, alcohol consumption, education level, body mass index (BMI), total daily energy intake, physical activity, fasting glucose level, history of cardiovascular disease (myocardial infarction, congestive heart failure and/or stroke), chronic obstructive pulmonary disease and diabetes mellitus. Abbreviations: FDR, false discovery rate; N, number; Beta, beta coefficient; CI, confidence interval. When attempting to replicate these findings in ULSAM, ten protein markers remained significantly associated at a nominal P-value < These were matrix metalloproteinase-12 (MMP-12), growth/differentiation factor 15 (GDF-15), urokinase plasminogen activator surface receptor (upar), TNF-related apoptosis-inducing ligand receptor 2 (TRAIL-R2), lectin-like oxidized LDL receptor 1 (LOX-1), hepatocyte growth factor (HGF), matrix metalloproteinase-10 (MMP-10) and matrix metalloproteinase-1 (MMP-1) (positively associated), and endothelial cell-specific molecule 1 (ESM-1) and interleukin-27 subunit alpha (IL27-A) (inversely associated; Table 3). 14

16 Secondary analyses The ten proteins associated with smoking in our primary analyses were further characterized in four sets of secondary analyses. First, when comparing former smokers to never-smokers, MMP-12 and MMP-1 reached nominal significance (P<0.05) in PIVUS, but could not be replicated in ULSAM (Table 4). Revisiting all 80 proteins, none showed a significant association with previous smoking in the multivariable-adjusted model in PIVUS (FDR< 5%; data not shown). 15

17 Table 3. Proteins significantly associated with current smoking in PIVUS and ULSAM* PIVUS (N=969) ULSAM (N=717) Age- and sex-adjusted Multivariable-adjusted Age-adjusted Multivariable-adjusted Protein Beta 95% CI P-value Beta 95% CI P-value Beta 95% CI P-value Beta 95% CI P-value markers MMP , E , E , E , GDF , E , E , , UPAR , E , E , E , E-04 TRAIL-R , E , E , , LOX , E , E , , HGF , E , E , , ESM , E , E , , MMP , E , E , , IL27-A , , , E , E-04 MMP , , , , * Listing only proteins that were significantly associated in the discovery sample (PIVUS) at false discovery rate < 5% (corresponding to an alpha of ), and successfully replicated in ULSAM with the same effect direction and P<0.05. Adjusted for age, sex, alcohol consumption, education level, body mass index (BMI), total daily energy intake, physical activity, fasting glucose level, history of cardiovascular disease (myocardial infarction, congestive heart failure and/or stroke), asthma, chronic bronchitis, chronic obstructive pulmonary disease and diabetes mellitus. Adjusted for the same variables as in PIVUS except sex (as all were men) and total energy intake (as that information was unavailable). Abbreviations: PIVUS, Prospective Study of the Vasculature in Uppsala Seniors; ULSAM, Uppsala Longitudinal Study of Adult Men; Beta, beta coefficient; CI, confidence interval; MMP-12, Matrix metalloproteinase-12; GDF-15, Growth/differentiation factor 15; UPAR, Urokinase plasminogen activator surface receptor; TRAIL-R2, TNF-related apoptosis-inducing ligand receptor 2; LOX-1, Lectin-like oxidized LDL receptor 1; HGF, Hepatocyte growth factor; ESM- 1, Endothelial cell-specific molecule 1; MMP-10, Matrix metalloproteinase-10; IL27-A, Interleukin-27 subunit alpha; MMP-1, Matrix metalloproteinase-1. 16

18 Table 4. Associations of current and former smoking compared to never-smoking* Previous smoking vs never smoking PIVUS (N=969) Current smoking vs never smoking ULSAM (N=717) Previous smoking vs never Current smoking vs never smoking smoking Protein markers Beta 95% CI P-value Beta 95% CI P-value Beta 95% CI P-value Beta 95% CI P-value MMP , , E , , GDF , , E , , UPAR , , E , , TRAIL-R , , E , , LOX , , E , , HGF , , E , , ESM , , E , , MMP , , E , , IL27-A , , , , MMP , , , , * Assessing only proteins significantly associated and replicated in the main models. Adjusted for age, sex, alcohol consumption, education level, body mass index (BMI), total daily energy intake (not in ULSAM, as that information was unavailable), physical activity, fasting glucose level, history of cardiovascular disease (myocardial infarction, congestive heart failure and/or stroke), chronic obstructive pulmonary disease and diabetes mellitus. Abbreviations: Beta, beta coefficient; CI, confidence interval; MMP-12, Matrix metalloproteinase-12; GDF-15, Growth/differentiation factor 15; UPAR, Urokinase plasminogen activator surface receptor; TRAIL-R2, TNF-related apoptosis-inducing ligand receptor 2; LOX-1, Lectin-like oxidized LDL receptor 1; HGF, Hepatocyte growth factor; ESM-1, Endothelial cell-specific molecule 1; MMP-10, Matrix metalloproteinase-10; IL27-A, Interleukin-27 subunit alpha; MMP-1, Matrix metalloproteinase-1.

19 When assessing dose-response of smoking on protein levels in current and former smokers, only GDF-15 was nominally significant (P<0.05) in the main model (multivariable-adjusted), while MMP-12, GDF-15, upar and HGF were significant in the ageand sex-adjusted model (P<0.05; Table 5). Table 5. Associations of smoking pack years with proteins in the PIVUS cohort* Age- & sex-adjusted (N=332) Multivariable-adjusted (N=326) Protein markers Beta 95% CI P-value Beta 95% CI P-value MMP , , GDF , , UPAR , , TRAIL-R , , LOX , , HGF , , ESM , , MMP , , IL27-A , , MMP , , * Assessing only proteins significantly associated and replicated in the main models. Only current and former smokers were included in the analyses. One pack year was defined as an average consumption of one pack of cigarettes per day during one year. Beta represents the change in protein marker per quartile of pack years. Adjusted for age, sex, alcohol consumption, education level, body mass index (BMI), total daily energy intake, physical activity, fasting glucose level, history of cardiovascular disease (myocardial infarction, congestive heart failure and/or stroke), chronic obstructive pulmonary disease and diabetes mellitus. Abbreviations: Beta, beta coefficient; CI, confidence interval; MMP-12, Matrix metalloproteinase-12; GDF-15, Growth/differentiation factor 15; UPAR, Urokinase plasminogen activator surface receptor; TRAIL-R2, TNF-related apoptosis-inducing ligand receptor 2; LOX-1, Lectin-like oxidized LDL receptor 1; HGF, Hepatocyte growth factor; ESM-1, Endothelial cell-specific molecule 1; MMP-10, Matrix metalloproteinase-10; IL27-A, Interleukin-27 subunit alpha; MMP-1, Matrix metalloproteinase- 1. Sex-stratified analyses were performed to assess the role of sex on the association between current smoking and protein levels in PIVUS. All ten proteins except MMP-1 were significantly associated (P<0.05) with smoking in women with similar beta coefficients as in the primary analysis. In men, all proteins except IL27-A were significantly associated (P<0.05) showing similar beta coefficients (Table 6). None of the ten proteins showed significantly different associations with current smoking between men and women in a formal sex-interaction test (P>0.05). In the sensitivity analysis excluding all individuals with manifest CVD, all ten proteins were confirmed to be significantly associated with current smoking in PIVUS (FDR<5%). All of them were successfully replicated in ULSAM (P<0.05; Table 7).

20 Table 6. Associations of current smoking protein levels by gender in the PIVUS cohort* Women (N=487) Men (N=482) Age- and sex-adjusted Multivariable-adjusted Age-adjusted Multivariable-adjusted Protein markers Beta 95% CI P-value Beta 95% CI P-value Beta 95% CI P-value Beta 95% CI P-value MMP , E , E , E , E-06 GDF , E , E , E , E-06 UPAR , E , E , E , E-06 TRAIL-R , E , E , E , E-05 LOX , E , E , E , E-06 HGF , , E , , ESM , , E , , MMP , E , E , , IL27-A , , , , MMP , , , E , * Assessing only proteins significantly associated and replicated in the main models. Adjusted for age, alcohol consumption, education level, body mass index (BMI), physical activity, fasting glucose level, history of cardiovascular disease (myocardial infarction, congestive heart failure and/or stroke), chronic obstructive pulmonary disease and diabetes mellitus. Abbreviations: Beta, beta coefficient; CI, confidence interval; MMP-12, Matrix metalloproteinase-12; GDF-15, Growth/differentiation factor 15; UPAR, Urokinase plasminogen activator surface receptor; TRAIL-R2, TNF-related apoptosis-inducing ligand receptor 2; LOX-1, Lectin-like oxidized LDL receptor 1; HGF, Hepatocyte growth factor; ESM-1, Endothelial cell-specific molecule 1; MMP-10, Matrix metalloproteinase-10; IL27-A, Interleukin-27 subunit alpha; MMP-1, Matrix metalloproteinase-1.

21 Table 7. Associations of current smoking with protein levels among individuals without previous CVD* PIVUS (N=857) ULSAM (N=543) Age- and sex-adjusted Multivariable-adjusted Age-adjusted Multivariable-adjusted Protein markers Beta 95% CI P-value Beta 95% CI P-value Beta 95% CI P-value Beta 95% CI P-value MMP , E , E , E , E-04 GDF , E , E , , UPAR , E , E , , TRAIL-R , E , E , , LOX , E , E , , HGF , E , E , , ESM , E , E , , MMP , E , E , , IL27-A , , , E , E-04 MMP , , , , * Assessing only proteins significantly associated and replicated in the main models. CVD defined as prior myocardial infarction, stroke and/or heart failure. Adjusted for age, sex, alcohol consumption, education level, body mass index (BMI), total daily energy intake, physical activity, fasting glucose level, history of chronic obstructive pulmonary disease and diabetes mellitus. Adjusted for the same variables as in PIVUS except sex (as all were men) and total energy intake (as that information was unavailable). Abbreviations: Beta, beta coefficient; CI, confidence interval; MMP-12, Matrix metalloproteinase-12; GDF-15, Growth/differentiation factor 15; UPAR, Urokinase plasminogen activator surface receptor; TRAIL-R2, TNF-related apoptosis-inducing ligand receptor 2; LOX-1, Lectin-like oxidized LDL receptor 1; HGF, Hepatocyte growth factor; ESM-1, Endothelial cell-specific molecule 1; MMP-10, Matrix metalloproteinase-10; IL27-A, Interleukin-27 subunit alpha; MMP-1, Matrix metalloproteinase-1.

22 Discussion Principal findings In the present study, associations of smoking with levels of 80 circulating CVD-related proteins were studied in two Swedish population-based cohorts. Using a conservative approach including strict correction for multiple testing and independent replication, ten proteins were identified to be significantly associated with cigarette smoking after taking possible confounders into account. Eight of the proteins were positively associated with smoking, while two were inversely associated. Pathophysiology Our findings support the notion that smoking via a range of mechanisms can promote initiation and progression of atherosclerosis, as well as plaque rupture giving rise to thromboembolic events. Besides, several of the smoking-associated proteins in the current study have also been linked to a wide range of diseases such as cancer (23) and COPD (24), reflecting other aspects of smoking-induced health effects. The underlying biological mechanisms by which smoking interferes with the atherosclerosis process include endothelial dysfunction, inflammation, neointimal formation, foam cell formation and plaque instability (Figure 2). The mechanisms are discussed below as supported by existing literature. Endothelial dysfunction Smoking is known to contribute to endothelial dysfunction, which is considered to be important in the initial phase of atherosclerosis (8). In the present study, three of the smoking-associated proteins have previously been suggested to be involved in endothelial dysfunction, namely GDF-15, LOX-1 and ESM-1. GDF-15 is a member of the transforming growth factor-ß cytokine superfamily (25). In a previous study of PIVUS participants, elevated levels of GDF- 15 were associated to reduced endothelium-dependent vasodilatation (25). LOX-1 is the predominant scavenger receptor in endothelial cells binding oxidized low-density lipoproteins (oxldl) (26). Binding of oxldl to LOX-1 results in vasoconstriction, endothelial cell apoptosis and endothelial progenitor cell apoptosis and dysfunction (26). Higher levels of ESM- 1, also called endocan, have been shown to be associated with worse endothelial function as evidenced by flow-mediated vasodilatation (27). However, while all other proteins were

23 associated in the direction expected by prior literature, ESM-1 was significantly lower in current smokers. This may reflect compensatory mechanisms following smoking exposure, or just highlight the limitation of any cross-sectional observational study, which cannot definitely disentangle causal mechanisms. Figure 2. Possible mechanisms connecting smoking and atherosclerosis highlighted by our findings Inflammation Inflammation is a major contributor to the initiation, as well as the progression of atherosclerosis (28). Among the ten smoking-associated proteins in the present study, GDF-15, LOX-1, ESM-1 and upar are known to promote inflammation, while IL-27 is known as an anti-inflammatory cytokine. GDF-15 promotes chemotaxis, apoptosis and necrosis of macrophages within atherosclerotic plaques (29). LOX-1 facilitates recruitment of monocytes to activated endothelium by up-regulating chemokines and adhesion molecules (26). ESM-1 increases endothelial permeability, stimulates production of pro-inflammatory cytokines by endothelial cells and facilitates recruitment of leukocytes by up-regulating adhesion molecules on activated endothelium (30). UPAR is the receptor of urokinase plasminogen activator, contributing to monocyte adhesion, macrophage infiltration and increased oxidative stress in macrophages (31). We observed an inverse association between smoking and IL-27A, also known as IL-27, which inhibits production of TNF and IL-17A by Th17 cells. TNF and IL-17A

24 up-regulate various chemokines synergistically and thereby facilitate the recruitment of inflammatory cells to atherosclerotic lesions (32). Neointimal formation Neointimal formation and subsequent vascular remodeling are important mechanisms leading to lumen narrowing in atherosclerosis, and is mainly carried out by proliferation and migration of vascular smooth muscle cells (VSMCs) (31). Four out of the ten smoking-associated proteins in the present study are known to be involved in this process. UPAR mediates adhesion and migration of VSMCs (31). LOX-1 stimulates proliferation of VSMCs (26), while ESM-1 stimulates both proliferation and migration of VSMCs (30). TRAIL-R2 is one of two death receptors of TRAIL, which promotes survival, proliferation and migration of VSMCs (33). TRAIL was also examined in the present study, and while the association with smoking did not reach out stringent significance criteria, it was indeed close to nominally significant in both PIVUS and ULSAM. Foam cell formation Monocytes entering the arterial intima become macrophages that in turn can accumulate oxldl by scavenger receptors, and gradually transform into foam cells to become an essential contributor to plaque growth (7). Among findings in the current study, upar, TRAIL-R2 and LOX-1 are known to promote foam cell formation, while IL-27 has shown inhibitory effect on this process. UPAR stimulates lipid uptake, as well as cholesterol biosynthesis in macrophages (31). TRAIL-R2 mediates lipid internalization in macrophages by up-regulating the scavenger receptor SR-AI, but not LOX-1 (34). During inflammation, LOX-1 is up-regulated in macrophages, and it also mediates oxldl uptake in dendritic cells and SMC-differentiated foam cells (26). On the other hand, IL-27 (inversely associated in our study) induces cholesterol efflux from macrophages, and therefore inhibits foam cell formation (35). Plaque instability Beside plaque formation, smoking is also known to contribute to plaque instability (8). In the current study, three matrix metalloproteinases (MMPs) were identified to be associated with smoking, namely MMP-1, MMP-10 and MMP-12. MMP-1 and MMP-10 are also known as collagenase-1 and stromelysin-2, respectively (36), while MMP-12 is known to degrade elastin (37). By degrading different components in the extracellular matrix (ECM), these proteins may promote ECM remodeling which in turn can result in plaque destabilization (36, 37). UPAR

25 mediates the conversion of plasminogen to plasmin leading to degradation of ECM directly or indirectly by activation of MMPs (31). Moreover, plaque neovascularization may contribute to plaque instability by facilitating infiltration of inflammatory cells and intraplaque hemorrhage (8). Among our findings, ESM-1 is known to be an indicator of angiogenesis (30), while HGF is a potent angiogenic factor (38). Comparisons with previous studies Among the ten proteins, GDF-15 (25), upar (39), MMP-10 (40), LOX-1 (41) and HGF (42) have previously been shown to be associated with smoking patterns in population-based studies. We are only aware of two prior proteomic studies assessing smoking in relation to proteins with high relevance for CVD, and both of these studies were performed in considerably smaller sample sizes than the present. Faarvang and colleagues performed profiling of 45 proteins in arterial walls in 11 active smokers and 13 never-smokers, and found differences in extracellular matrix proteins between smokers and non-smokers implicating a role for smoking in vascular remodeling (16). Della Corte and colleagues performed mass spectrometry-based proteomics of platelets from 16 smokers and 16 non-smokers, all apparently healthy. Their analyses showed different levels of several proteins related to inflammation and apoptosis, suggesting that smoking may interfere with platelet aggregation and cytoskeleton remodeling (15). The current study is the first to investigate the relationship between smoking and many circulating protein markers relevant for CVD in a large number of individuals from the general population. It is difficult to directly compare our findings of protein markers to the two previous proteomic studies, since smoking-associated proteins identified in those studies were not analyzed in the current study. However, a comparison of our findings with the two previous proteomic studies shows overlapping pathophysiology regarding two important biological mechanisms linking smoking to atherosclerosis. These are ECM remodeling (as evidenced by decreased levels of collagen α1, collagen α2 and decorin in the study conducted by Faarvang et al, and increased levels of MMP-1, MMP-10 and MMP-12 in the current study) and inflammation (as evidenced by down-regulation of heat shock protein 60 among other proteins in the study conducted by Della Corte et al, and up-regulation of GDF-15, LOX-1, ESM-1, upar and down-regulation of IL-27 in the current study).