Urinary Proteomic Profile of Arterial Stiffness Is Associated With Mortality and Cardiovascular Outcomes.

Background The underlying mechanisms of arterial stiffness remain not fully understood. This study aimed to identify a urinary proteomic profile to illuminate its pathogenesis and to determine the prognostic value of the profile for adverse outcomes. Methods and Results We measured aortic stiffness using pulse wave velocity (PWV) and analyzed urinary proteome using capillary electrophoresis coupled with mass spectrometry in 669 randomly recruited Flemish patients (mean age, 50.2 years; 51.1% women). We developed a PWV-derived urinary proteomic score (PWV-UP) by modeling PWV with proteomics data at baseline through orthogonal projections to latent structures. PWV-UP that consisted of 2336 peptides explained the 65% variance of PWV, higher than 36% explained by clinical risk factors. PWV-UP was significantly associated with PWV (adjusted β=0.73 [95% CI, 0.67-0.79]; P<0.0001). Over 9.2 years (median), 36 participants died, and 75 experienced cardiovascular events. The adjusted hazard ratios (+1 SD) were 1.46 (95% CI, 1.08-1.97) for all-cause mortality, 2.04 (95% CI, 1.07-3.87) for cardiovascular mortality, and 1.39 (95% CI, 1.11-1.74) for cardiovascular events (P≤0.031). For PWV, the corresponding estimates were 1.25 (95% CI, 0.97-1.60), 1.35 (95% CI, 0.85-2.15), and 1.22 (95% CI, 1.02-1.47), respectively (P≥0.033). Pathway analysis revealed that the peptides in PWV-UP mostly involved multiple pathways, including collagen turnover, cell adhesion, inflammation, and lipid metabolism. Conclusions PWV-UP was highly associated with PWV and could be used as a biomarker of arterial stiffness. PWV-UP, but not PWV, was associated with all-cause mortality and cardiovascular mortality, implying that PWV-UP-associated peptides may be multifaceted and involved in diverse pathological processes beyond arterial stiffness.

capillary electrophoresis coupled with mass spectrometry

carotid‐femoral pulse wave velocity

orthogonal projections to latent structures

pulse wave velocity

pulse wave velocity–derived urinary proteomic score

variable importance in the projection

Clinical Perspective

What Is New?

A prospective cohort study including 669 Flemish patients investigated the urinary proteomic profile for arterial stiffness determined by pulse wave velocity and assessed its associations with cardiovascular events and mortality.

The proteomic profile that was related to multiple pathological processes showed a high correlation with pulse wave velocity, stronger than conventional clinical risk factors.

The proteomic profile for arterial stiffness was superior to pulse wave velocity in terms of the prediction of adverse outcomes, especially for all‐cause mortality and cardiovascular mortality.

What Are the Clinical Implications?

The potential of using urinary proteomics fingerprinting biomarkers to recognize the underlying pathophysiological processes and to facilitate the search for personalized treatment targets is described.

Given the superior prognostic value, the proteomic profile could be used as an alternative surrogate of arterial stiffness for better prediction of adverse outcomes.

Arterial stiffness reflects the altered structural and functional properties of the arterial wall attributable to vascular aging. It has been demonstrated as a strong independent predictor of cardiovascular events and mortality. , , As vascular stiffening affects pulse wave transmission, aortic stiffness can be quantified by carotid‐femoral pulse wave velocity (cfPWV). A meta‐analysis of 17 prospective studies with 15 877 participants revealed that each 1‐m/s increase in aortic pulse wave velocity (PWV) was associated with 14% increased risk of total cardiovascular events, 15% increased risk of cardiovascular mortality, and 15% increased risk of all‐cause mortality. Although PWV is an intuitive metric of hemodynamic alterations attributable to arterial stiffness, it does not directly reveal the underlying mechanisms of arterial stiffness. From the perspective of personalized medicine, it is necessary to propose molecular biomarkers to complement PWV.

Urine contains diverse proteins and peptides that predominately originate from plasma, renal tubules, or lower urinary tract. Small‐to‐medium sized molecular proteins that leak from local tissues into blood circulation could be excreted in urine through glomerular filtration. Therefore, urinary proteins/peptides have the potential to show pathological changes underlying kidney diseases and various other disorders, including cardiovascular diseases. To date, the emerging urinary proteomics have been used for biomarker discovery, risk stratification, and elucidation of pathological mechanisms. , , , To the best of our knowledge, there is no published study of urinary proteomics in relation to arterial stiffness. Prior proteomic studies investigated the proteomic profile of arterial stiffness in plasma and arterial tissue, but were limited to a small sample size, a cross‐sectional study design, or only in young healthy adults. ,

Given these circumstances, we aimed to identify a urinary proteomic profile of arterial stiffness at baseline and to determine its prognostic value for adverse outcomes in a prospective population. In addition, we further discussed the pathological processes of the proteomic profile through pathway analyses.

METHODS

The authors declare that all supporting data are available within the article and its online supplemental material.

Study Population

This study was conducted in a family‐based population study, the FLEMENGHO (Flemish Study on Environment, Genes and Health Outcomes). The enrollment of the FLEMENGHO started from 1985 through 2004 with an initial participation rate of 78%. , The study was approved by the University of Leuven Ethics Committee. , All participants provided written informed consent. Participants underwent repeated follow‐up. From May 2005 until May 2010, there were 688 eligible participants who had undergone PWV measurements and urinary proteomics analysis. Because of the inadequate quality of PWV measurement, 19 were excluded. Hence, 669 participants were included.

PWV Measurement

As the impaired compliance of stiffening aortic arteries accelerates the velocity of pulse wave propagation, the American Heart Association has endorsed the use of cfPWV to quantify aortic stiffness. PWV was calculated as the length divided by the transit time of pulse wave travel in the segment from the right carotid artery to the femoral artery. In this study, PWV was evaluated using the tonometry‐based SphygmoCor Device (AtCor Medical, West Ryde, New South Wales, Australia). The robustness of this device and the reproducibility have been demonstrated in previous studies. , For quality control, recordings were excluded when the deviation of pulse wave was >10%.

Urinary Proteomics Analysis

Detailed information on sample preparation, proteome analysis using capillary electrophoresis coupled with mass spectrometry (CE‐MS), data processing, and sequencing of the peptides has been described in previous publications and the Supplemental Methods (Data S1). , Briefly, CE‐MS was performed, using a P/ACE MDQ capillary electrophoresis system (Beckman Coulter, Fullerton, CA) coupled to a micrOTOF MS (Bruker Daltonics, Bremen, Germany). Mass spectral data were processed with MosaiquesVisu software, generating a raw list of peptides or proteins with molecular mass, migration time, and signal intensity that were calibrated using internal urinary standard peptides to assure the comparability between different data sets. , Identified peptides were assigned to the previously sequenced peptides from the Human Urinary Proteome Database by fragmenting peptides and matching the fragmentation spectra to the protein sequences from the database’s International Protein Index, the reference sequence database at National Center for Biotechnology Information, and the UniProt Knowledgebase. In the protein annotation, posttranslational modifications and specific mass spectra were considered. Peptides from different samples were considered the same when the deviations of their molecular weight and the migration time were <100 parts/million and <1 minute, respectively. As high‐dimensional data are prone to sparsity, multicollinearity with a high risk of overfitting, peptides were excluded when they were undetectable in >70% of the participants. Thus, of the 21 559 detected urinary peptides, 2336 were eventually analyzed.

Assessment of the Outcomes

The vital status of all the participants was annually ascertained through the Belgian Population Registry in Brussels until December 31, 2016. In cases of death, the causes were indicated with the International Classification of Diseases, Ninth Revision (ICD‐9) codes that were acquired from the Flemish Registry of Death Certificates. Information on nonfatal events was collected via either face‐to‐face follow‐up visits at the examination center with repeated administration of the standard questionnaire that had been used at baseline or a structured telephone interview. Coronary events included sudden death, myocardial infarction, acute coronary syndrome, new‐onset angina pectoris, ischemic cardiomyopathy, and coronary revascularization. Cardiac events included all coronary events and heart failure, new‐onset atrial fibrillation, life‐threatening arrhythmias and high‐degree atrioventricular block that required pacemaker implantation, and pulmonary heart disease. Cardiovascular events included cardiac events and were additionally considered aortic aneurysm, atrial embolism and revascularization of peripheral arteries, stroke, and transient ischemic attack. Physicians ascertained the diseases reported on the death certificates, in the questionnaires, and in the telephone interviews against the medical records of general practitioners or hospitals. Participants were censored when the first event within each category occurred.

Statistical Analysis

Statistical analysis was performed with SAS software, version 9.4 (SAS Institute, Cary, NC). Statistical significance was determined with a 2‐sided P value of 0.05. Means and proportions were compared via t test or ANOVA test, or Fisher test as appropriate. Because the high dimensionality and multicollinearity of the proteomic data remain unresolved for the conventional linear regression model, the orthogonal projections to latent structures (OPLS) method was used. As a variant of partial least squares method, OPLS is a supervised dimension reduction statistical method. It diminishes the data dimension by projecting the original data into a new space and constructing latent variables (components) to linearly model the dependent variables. Unlike partial least squares, OPLS initially filters the “noise” information that is irrelevant to dependent variables, then builds fewer latent variables. Hence, OPLS models are simpler and easier to interpret. , With a data matrix comprising of 2336 peptides and PWV in 669 participants, the OPLS analysis was conducted using the SIMCA software, version 14.1 (Umetrics, Sartorius‐Stedim, Sweden). Peptide data were scaled to unit variance and log transformed to obtain equal leverage of peptides. The default 7‐fold cross‐validation and 1000 permutations were used to assess model overfitting. The ANOVA of the cross‐validation residuals was used to calculate P value for model significance. The detailed information on the OPLS model can be found in the Supplemental Methods (Data S1). The model outputted PWV‐derived urinary proteomic score (PWV‐UP), equivalent to predictive values, to integrate the information carried by various urinary peptides. In the OPLS model, the importance of each peptide was assessed by the variable importance in the projection (VIP) to evaluate the association of a peptide with PWV. Peptides with a VIP >1.2 were considered significant for PWV. Besides, the OPLS model also provided the correlation coefficient for a single peptide as an alternative to assess the association between a given peptide and PWV. A figure visualized the VIPs and correlation coefficients of all studied peptides. The peptides in the top left quadrant and top right quadrant of the plot were inversely or positively associated with PWV, respectively.

Multivariable linear regression was used to investigate the association between PWV and PWV‐UP, with adjustment for potential clinical confounders. The proportions of the variance in PWV (R 2) explained by PWV‐UP, clinical variables, and a model with both of them were calculated. The estimate of a linear association was expressed as a β coefficient that indicated the change in PWV for per SD increment of PWV‐UP. Stepwise linear regression with backward selection was used for the selection of clinical confounders. The potential clinical confounders with a P value of <0.05 were retained in the multivariable linear regression model by the backward selection procedure. The following clinical variables were considered: sex, age, body mass index, heart rate, mean arterial pressure, serum total cholesterol, high‐density lipoprotein cholesterol, low‐density lipoprotein cholesterol, blood glucose, estimated glomerular filtration rate, current smoking, history of diabetes or cardiovascular diseases, hypertension, and being treated for hypertension. The collinearity of linear models was examined. The correlation of PWV‐UP and PWV with continuous clinical variables was determined by Pearson correlation. The correlation between a continuous variable and a dichotomous variable was assessed by the point biserial correlation coefficient.

In the categorical analysis, participants were grouped by the median of PWW‐UP and PWV, respectively. The crude cumulative incidence of all‐cause mortality was estimated across the groups using the Kaplan‐Meier method, and the cumulative incidence curves were compared using the log‐rank test. The Fine‐Gray subdistribution hazard model was used to calculate the crude incidence of cardiovascular events while taking the competing risk of noncardiovascular deaths into account. The difference of cumulative incidence of cardiovascular events between groups was examined by the Gray method. The adjusted 5‐year absolute risk was calculated with the baseline PWV‐UP and PWV, respectively, with adjustment of sex, age, smoking, mean arterial pressure, body mass index, plasma glucose, total cholesterol, estimated glomerular filtration rate, history of diabetes, and previous cardiovascular events. The prognostic values of PWV‐UP and PWV were assessed by multivariable‐adjusted Cox proportional hazard regression models. The proportional hazard assumption was examined by the Kolmogorov‐type supremum test. Furthermore, a competing risk analysis for cardiovascular events and cardiovascular mortality was performed with cause‐specific and subdistribution hazard models.

Pathway Analysis

To interpret the biological function of the urinary proteomics, enrichment pathway analysis was performed with the online Reactome Pathway Database, version 75 (https://reactome.org). Proteins with peptides having a VIP score of ≥1.2 were considered important and, thus, submitted for pathway analysis. P value of the annotated pathways was corrected with false discovery rate. A false discovery rate of <0.05 was considered statistically significant. As a complement to the Reactome Pathway analysis, the biological function of the significant peptides was also investigated in the Gene Ontology database, which was performed by using the ClueGO plug‐in of Cytoscape. Gene Ontology terms were determined by a 2‐sided hypergeometric statistical test. P value corrected by Bonferroni step down was set at 0.05, and the κ score threshold of 0.4 was considered as statistically significant.

RESULTS

Characteristics of Participants

Table 1 shows the characteristics of the study population. The age (±SD) of the 669 participants averaged 50.2±15.6 years. Among all the participants, 342 (51.1%) were women, 268 (40.1%) had hypertension, 25 (3.7%) had diabetes, and 39 (5.8%) had a history of cardiovascular diseases. The mean values were 25.8±3.6 kg/m2 for body mass index, 128.6±17.2/79.2±9.4 mm Hg for systolic/diastolic blood pressure, 59.2±9.0 beats/min for heart rate, 5.25±0.96 mmol/L for total cholesterol, and 3.21±0.85 for low‐density lipoprotein cholesterol. cfPWV averaged 7.56±2.02 m/s.

Table 1

Characteristics of 669 Participants

Characteristics	Total (n=669)
Participants with characteristic, n (%)
Women	342 (51.1)
Smoking history	241 (36.0)
Diabetes	25 (3.7)
Cardiovascular diseases	39 (5.8)
Hypertension	268 (40.1)
Treatment of hypertension	150 (22.4)
Statins	83 (12.4)
Antiplatelet drugs	64 (9.6)
Mean±SD or median (IQR) of characteristic
Age, y	50.2±15.6
Body mass index, kg/m2	25.8±3.6
Waist/hip ratio	0.9±0.1
Heart rate, beats/min	59.2±9.0
Systolic blood pressure, mm Hg	128.6±17.2
Diastolic blood pressure, mm Hg	79.2±9.4
Mean arterial pressure, mm Hg	112.1±13.4
Serum total cholesterol, mmol/L	5.25±0.96
HDL cholesterol, mmol/L	1.44±0.35
LDL cholesterol, mmol/L	3.21±0.85
Blood glucose, mmol/L	4.91±0.70
Serum creatinine, mg/dL	0.92±0.18
eGFR, mL/min per 1.73 m2	85.9±16.9
Urine microalbumin, mg/L	5.5 (4.1–7.5)
Pulse wave velocity, m/s	7.56±2.02

History of smoking refers to inhaling tobacco on a daily basis in the past; hypertension was an office blood pressure of ≥140 mm Hg systolic or ≥90 mm Hg diastolic or use of antihypertensive drugs; diabetes was use of antidiabetic drugs or fasting blood glucose of ≥126 mg/dL; eGFR is estimated using the Chronic Kidney Disease Epidemiology Collaboration creatinine equation. Body mass index was calculated by weight in kilograms divided by height in meters squared. eGFR indicates estimated glomerular filtration rate; HDL, high‐density lipoprotein; IQR, interquartile range; and LDL, low‐density lipoprotein.

Urinary Proteomic Profile

At baseline, the urinary proteomics data included 2336 peptides that were detectable in ≥30% of the participants (N=669). Of these peptides, 743 were sequenced and annotated to 111 parental proteins. Modeling the 2336 peptides with PWV, the generated PWV‐UP model explained 65.0% of the variance in PWV (equivalent to R 2: 0.65; 95% CI, 61.2–68.3; P<0.0001; Figure 1). The cross‐validation showed that PWV‐UP model explained 63.7% of the PWV variance in the leave‐out fold, which indicated a low risk of overfitting. This was also confirmed by the 1000 random permutations (P<0.05). The distribution of the PWV‐UP score is shown in Figure S1, and the average PWV‐UP score was 7.59±1.95 m/s.

Figure 1

Associations of pulse wave velocity (PWV) with PWV‐derived urinary proteomic score (PWV‐UP) and clinical variables.

A, Scatterplot of PWV and PWV‐UP. The solid line represents the regression line. The band with 2 solid lines indicates the 95% confidence limits of the regression line, and the transparent band refers to the 95% prediction limits of the regression model. The gray dotted line represents the identity line that had a slope of 1. B, Scatterplot of PWV and the predicted values of the clinical variables. The clinical variables included age, sex, heart rate, mean arterial pressure, blood glucose, urine microalbumin, and current smoking. C, Scatterplot of PWV and the predicted value of PWV‐UP and the clinical variables together. D, Bar plot illustrating the proportion of explained variance in PWV by 3 models. The error bars denote the upper 95% confidence limit.

In the OPLS model, the significance of peptides was reflected by VIPs and correlation coefficients, as shown in the volcano plot (Figure 2). Of the 743 sequenced peptides with parental protein, 276 (37.1%) peptides with a VIP threshold of ≥1.2 from 37 proteins significantly contributed to PWV‐UP. The correlation coefficients, VIPs, sequences, mass, and migration time of the 276 peptides were listed in Table S1. The top proteins above the VIP cutoff were 15 different types of collagens and 22 kinds of distinct proteins. Most of them were predominantly extracellular matrix (ECM) proteins, such as collagen I, collagen III, titin, mucin‐2, and cadherin 1. Notably, the peptides from collagen I and III accounted for 79.3% of these significant peptides, and most of them were inversely associated with PWV, as shown in Figure 2. Apart from extracellular structural proteins, other top proteins involved in diverse processes, including cell adhesion (titin, mucin, protocadherin‐9, and cadherin 1), cell‐protein interaction (vesicular integral‐membrane protein), lipid metabolism (apolipoprotein A‐I and A‐VI), vascular calcification (MGP [matrix Gla protein], osteopontin, and collagen 2), coagulation and fibrinolysis (fibrinogen, α‐1‐antitrypsin, α‐1‐antichymotrypsin, and plasminogen), and anti‐inflammation (annexin A1). The Reactome pathway analysis annotated to 62 biological processes that were mainly related to the ECM turnover, collagen formation and degradation, cell‐ECM interaction, cell adhesion, signal transduction, immune‐related pathways, platelet activation and hemostasis, inflammation, and lipid metabolism (Table S2). The top 20 pathways are shown in Figure 3. The functional annotations of the significant peptides revealed 137 Gene Ontology terms of biological processes that related to collagen fibril organization, platelet aggregation, lipoprotein oxidation, and immune response (Figure 3 and Table S3).

Figure 2

Volcano plot of significant peptides of pulse wave velocity–derived urinary proteomic score (PWV‐UP).

Of the 743 peptides with parental protein, 276 peptides (37 proteins) significantly contributed to PWV‐UP with a variable importance in the projection (VIP) ≥1.2 as the threshold. Peptides in the top left quadrant (202) and top right quadrant (74) of the plots were inversely or positively associated with pulse wave velocity, respectively. ANXA1 indicates annexin A1; APOA, apolipoprotein A; APOC, apolipoprotein C; CHGB, secretogranin‐1; COL, collagen; LMAN2, vesicular integral‐membrane protein VIP36; MGP, matrix Gla protein; PCDH, protocadherin‐9; PGRMC1, membrane‐associated progesterone receptor component; PKD1L2, polycystic kidney disease protein 1‐like 2; PLG, plasminogen; POTEF, POTE ankyrin domain family member F; S100A9, protein S100‐A9; SERPINA1, α1‐antitrypsin; SERPINA3, α1‐antichymotrypsin; and TMSB4X, thymosin β4.

Figure 3

Bioinformatic analysis for the significant proteins in the urinary proteomic profile.

A, Top 20 Reactome pathways annotated by the urinary proteomic profile. The bar length of every pathway indicates the corresponding −log10 (false discovery rate). B, The network of enriched Gene Ontology (GO) biological processes. GO terms are clustered and colored by different functional groups. Each dot represents a GO term with adjusted P<0.05. Dot size indicates their corresponding adjusted P value, whereas the edge width and transparency suggest the κ score. ECM indicates extracellular matrix; MET, mesenchymal‐epithelial transition factor; NCAM1, neural cell adhesion molecule 1; PDGF, platelet‐derived growth factor receptor; and PTK, protein‐tyrosine kinase.

Association With PWV

The adjusted association of PWV with PWV‐UP score was assessed by multivariable linear regression. With the backward selection procedure on covariates, the following clinical variables were retained as potential confounders: age, heart rate, mean arterial pressure, blood glucose, and current smoking (P≤0.006; Table S4). Sex forced into the model for clinical relevance. With adjustments of these confounders, per 1‐SD increment (1.95 m/s) in the PWV‐UP was associated with 0.73‐m/s increase in PWV (95% CI, 0.67–0.79 m/s; P<0.0001). With further adjustment for estimated glomerular filtration rate and urine microalbumin, the β coefficient of PWV‐UP (0.73 [95% CI, 0.67–0.79]) barely changed. No multicollinearity was detected through collinearity diagnostics, as the highest score of the variation inflation factor was 2.43. Furthermore, compared with the proportion of variance in PWV explained by PWV‐UP (65.0%), the effect of the clinical variables, including urine microalbumin, was 35.2% (95% CI, 29.1%–39.8%; Figure 1). The proportion was slightly improved to 67.0% (equivalent to R 2: 0.67; 95% CI, 62.9–69.7%) when the clinical variables were added into the PWV‐UP model.

Correlation With Clinical Variables

The correlations between PWV, PWV‐UP, and clinical variables are shown in Table 2. Higher PWV‐UP and PWV were positively correlated with age, body mass index, systolic and diastolic blood pressure, mean arterial pressure, total cholesterol, low‐density lipoprotein cholesterol, blood glucose, and urine microalbumin, but were inversely correlated with estimated glomerular filtration rate (P<0.0001). Although PWV‐UP and PWV were modestly correlated with heart rate (r=−0.02 and r=0.05, respectively), the correlations were in opposite directions. A history of diabetes, cardiovascular disease, or hypertension was correlated with higher PWV‐UP and PWV (P≤0.009). PWV‐UP and PWV had no significant correlation with high‐density lipoprotein cholesterol and sex.

Table 2

Correlation Between PWV‐UP, PWV, and Clinical Variables in 669 Flemish Patients at Baseline

Variable	PWV‐UP		PWV
	r	P value	r	P value
Age	0.61	<0.0001	0.52	<0.0001
Body mass index	0.24	<0.0001	0.21	<0.0001
Heart rate	−0.02	<0.0001	0.05	<0.0001
Systolic blood pressure	0.49	<0.0001	0.48	<0.0001
Diastolic blood pressure	0.16	<0.0001	0.23	<0.0001
Mean arterial pressure	0.46	<0.0001	0.46	<0.0001
Total cholesterol	0.15	0.0001	0.15	0.0001
HDL cholesterol	−0.07	0.071	−0.04	0.26
LDL cholesterol	0.16	<0.0001	0.16	<0.0001
Blood glucose	0.26	<0.0001	0.26	<0.0001
eGFR	−0.44	<0.0001	−0.36	<0.0001
Urine microalbumin	0.19	<0.0001	0.16	<0.0001
Sex (men, women)	−0.03	0.51	−0.07	0.070
Current smoking (0, 1)	−0.001	0.97	0.02	0.52
Diabetes (0, 1)	0.11	0.005	0.10	0.009
Cardiovascular diseases (0, 1)	0.27	<0.0001	0.20	<0.0001
Hypertension (0, 1)	0.45	<0.0001	0.40	<0.0001

The r value was calculated by Pearson correlation for 2 continuous variables, whereas the correlation between a continuous variable and a dichotomous variable was assessed by the point biserial correlation coefficient. eGFR indicates estimated glomerular filtration rate; HDL, high‐density lipoprotein; LDL, low‐density lipoprotein; PWV, pulse wave velocity; PWV‐UP, PWV‐derived urinary proteomic score; and r, correlation coefficient.

Longitudinal Association With Mortality and Cardiovascular Risk

The median follow‐up was 9.2 years (5th–95th percentile, 6.1–10.7 years). Of the 669 participants, 36 died, including 10 (27.8%) from cardiovascular deaths and 15 (41.7%) of cancer. The total of 75 cardiovascular events included 51 cardiac events. The detailed information is presented in Table S5. Using Kaplan‐Meier survival function estimates and the log‐rank test, the cumulative incidence of all‐cause mortality and cardiovascular events was significantly higher at the upper half of PWV‐UP (P≤0.0002; Figure 4). A similar result was observed in PWV (P<0.0001; Figure 4). With adjustment of potential confounders, the absolute 10‐year risk of all‐cause and cardiovascular events increased with the level of baseline PWV‐UP and PWV, but the independent effect of PWV was not significant (P≥0.053; Figure S2). In univariate Cox regression analysis, a higher PWV‐UP or PWV was significantly associated with an increased risk of all‐cause mortality, cardiovascular mortality, cardiovascular events, cardiac events, coronary events, and stroke (P≤0.025; Table 3). The association of PWV with all the outcomes lost significance after adjustment for PWV‐UP (P≥0.24). After adjusting for PWV, PWV‐UP was still significantly associated with all outcomes (P≤0.042), except for stroke (P=0.87). With the adjustment of the potential confounders for per‐SD increment of PWV‐UP, hazard ratio (HR) was 1.46 (95% CI, 1.08–1.97) for all‐cause mortality, 2.04 (95% CI, 1.07–3.87) for cardiovascular mortality, and 1.39 (95% CI, 1.11–1.74) for cardiovascular events (P≤0.031; Table 3). The adjusted association of PWV‐UP with cardiac events, coronary events, and stroke did not reach significance (P≥0.30). The associations of PWV‐UP with all‐cause mortality, cardiovascular mortality, and cardiovascular events remained significant when additionally adjusting for urine microalbumin, with the adjusted HRs of 1.46 (95% CI, 1.08–1.97) for all‐cause mortality, 1.94 (95% CI, 1.01–3.73) for cardiovascular mortality, and 1.34 (95% CI, 1.06–1.68) for cardiovascular events (P≤0.048). However, PWV was only significantly associated with cardiovascular events after adjustment of cofounders (HR, 1.22 [95% CI, 1.02–1.47]; P=0.033). The adjusted associations of PWV‐UP and PWV with cardiovascular events were confirmed as well when considering the competing risk of death for other causes (subdistribution and cause‐specific HRs of PWV‐UP: 1.21 [95% CI, 1.04–1.40] and 1.39 [95% CI, 1.11–1.74]; subdistribution and cause‐specific HRs of PWV: 1.34 [95% CI, 1.08–1.73] and 1.22 [95% CI, 1.02–1.47]). Similarly, cardiovascular mortality remained higher in participants with higher PWV‐UP in the competing analysis (subdistribution and cause‐specific HRs: 1.89 [95% CI, 1.12–3.17] and 2.04 [95% CI, 1.07–3.87]). All models met the proportional hazard assumption (P≥0.15). In addition, they are not particularly linked to arterial stiffness because arterial stiffness might not be particularly related to arrhythmias and peripheral vascular diseases.

Figure 4

Cumulative incidence of all‐cause mortality and cardiovascular events in 669 participants.

The participants were divided into 2 groups according to the median of pulse wave velocity–derived urinary proteomic score (PWV‐UP) (A and B) and the median of pulse wave velocity (PWV) (C and D). The colored bands represent the SE. P values express the significance of the log‐rank test or the Gray method for the difference across the groups.

Table 3

Adjusted HRs Associated With Baseline PWV‐UP and PWV

End points	Events/at risks	PWV‐UP		PWV
		HR (95% CI)	P value	HR (95% CI)	P value
All‐cause mortality	36/669
Unadjusted		2.11 (1.74–2.56)	<0.0001	1.49 (1.31–1.69)	<0.0001
Adjusting for PWV or PWV‐UP		2.49 (1.80–3.43)	<0.0001	0.86 (0.66–1.12)	0.26
Adjusting for covariates		1.46 (1.08–1.97)	0.014	1.25 (0.97–1.60)	0.085
Cardiovascular mortality	10/669
Unadjusted		2.74 (1.96–3.84)	<0.0001	1.61 (1.32–1.96)	<0.0001
Adjusting for PWV or PWV‐UP		3.53 (2.17–5.75)	<0.0001	0.79 (0.53–1.17)	0.24
Adjusting for covariates		2.04 (1.07–3.87)	0.031	1.35 (0.85–2.15)	0.21
Cardiovascular events	75/669
Unadjusted		2.03 (1.76–2.35)	<0.0001	1.52 (1.38–1.67)	<0.0001
Adjusting for PWV or PWV‐UP		2.12 (1.66–2.70)	<0.0001	0.96 (0.79–1.17)	0.68
Adjusting for covariates		1.39 (1.11–1.74)	0.004	1.22 (1.02–1.47)	0.033
Cardiac events	51/669
Unadjusted		1.85 (1.55–2.20)	<0.0001	1.40 (1.24–1.59)	<0.0001
Adjusting for PWV or PWV‐UP		2.18 (1.60–2.96)	<0.0001	0.85 (0.65–1.12)	0.24
Adjusting for covariates		1.17 (0.87–1.59)	0.30	0.98 (0.72–1.33)	0.88
Coronary events	23/669
Unadjusted		1.70 (1.28–2.26)	0.0003	1.40 (1.15–1.70)	0.001
Adjusting for PWV or PWV‐UP		1.73 (1.02–2.93)	0.042	0.98 (0.64–1.51)	0.94
Adjusting for covariates		0.89 (0.53–1.51)	0.67	0.88 (0.52–1.48)	0.62
Stroke	13/669
Unadjusted		1.68 (1.15–2.45)	0.007	1.37 (1.04–1.80)	0.025
Adjusting for PWV or PWV‐UP		1.82 (0.92–3.63)	0.087	0.92 (0.51–1.66)	0.78
Adjusting for covariates		0.79 (0.42–1.48)	0.45	0.72 (0.38–1.38)	0.32

HRs express the risk per SD increment in PWV‐UP (1.95 m/s) or PWV (2.02 m/s). Covariates included baseline characteristics, including sex, age, smoking, diabetes, history of cardiovascular events, mean arterial pressure, body mass index, plasma glucose, total cholesterol, and estimated glomerular filtration rate. HR indicates hazard ratio; PWV, pulse wave velocity; and PWV‐UP, PWV‐derived urinary proteomic score.

DISCUSSION

This study developed a PWV‐UP score that was significantly associated with aortic stiffness, as measured by cfPWV, independent of clinical confounders. Besides, PWV‐UP contributed more to the variation of PWV than multiple clinical variables did. Similar to PWV, PWV‐UP was positively correlated with clinical risk factors, such as age, systolic blood pressure, diabetes, and hypertension. Notably, PWV‐UP was associated with all‐cause mortality, cardiovascular mortality, and cardiovascular events, with the adjustment of potential confounders, whereas PWV was only associated with cardiovascular events. The pathway analysis revealed that peptides included in PWV‐UP were involved in multiple biological processes, such as collagen turnover, cell adhesion, inflammation, and lipid metabolism.

Proteomic studies on human arterial stiffness are scarce. Lyck Hanssen et al enrolled 19 patients in their study for coronary artery bypass grafting and measured their cfPWV before the surgery. The patients were classified into the high‐PWV group (N=10) and the low‐PWV group (N=9), according to the PWV threshold of 10 m/s. The left internal mammary arterial tissue pieces were collected during the surgery and analyzed using liquid chromatography–mass spectrometry. Among 418 proteins, 28 were differentially expressed between groups (P<0.05 without multiple testing correction) and mainly consisted of ECM proteins. Part of the tissue proteomic portraits was consistent with our urinary proteomic profile, such as collagen IV, apolipoprotein A‐1, and protein S100. Pettersson‐Pablo et al performed a plasma proteomic study on 834 healthy young adults (aged 18–26 years). They measured 92 inflammatory proteins with proximity extension assay (OLINK Proteomics, Uppsala, Sweden) and arterial stiffness determined by cfPWV and augmentation index. They used the unsupervised dimension reduction statistical method, principal component analysis to integrate the correlated proteins. However, of the 4 constructed components, only 1 component that consisted of proteins related to hemostasis was significantly associated with PWV, and the model that included all the components and clinical risk factors explained only 5.7% of the variance of PWV. This indicated that the mechanisms underlying arterial stiffness entangle multifactorial molecules, and inflammatory proteins are part of it. Of note, there were several differences between previous studies and the current study. First, the recruited participants had different clinical settings. The participants in previous studies were diagnosed with atherosclerosis or were healthy young adults, whereas our participants were from the general population with a wide age range. Having a high prevalence of hypertension (40%), our population was exposed to a higher risk of arterial stiffness compared with the young adults. Second, aortic arteries are not easily accessible samples; therefore, alternative samples were used (muscular artery tissue versus plasma versus urine). As urine and blood samples are easier to collect, the latter 2 studies had a relatively large sample size (N=19 versus N=834 versus N=669). Although we investigated the proteomic patterns in different types of specimens, the findings were complementary to one another. Because the reference value of a normal cfPWV is still controversial, Pettersson‐Pablo et al and the current study both modeled continuous PWV with proteomics data to identify potential relevant peptides. Previous study adopted the unsupervised method (principal component analysis), whereas we used the supervised method (OPLS). However, in the previous study, the constructed components were poorly related with PWV. In contrast, PWV‐UP was significantly associated with PWV, independent of the confounding effect of the clinical variables. Our study also confirmed that arterial stiffness is poorly correlated with conventional clinical variables. In particular, we found that PWV‐UP was remarkably superior to the clinical variables in terms of the explained proportion of PWV variance. This was probably attributable to the multifunctional peptides included in PWV‐UP, as revealed in the pathway analysis. These peptides might be involved in various mechanisms of arterial stiffness.

More important, we demonstrated that PWV‐UP was associated with adverse outcomes. Over a follow‐up period of 9.2 years, PWV‐UP was associated withall‐cause mortality, cardiovascular mortality, and cardiovascular events, whereas PWV itself was only associated with cardiovascular events. This appears contradictory because PWV‐UP was developed from PWV. The predictive value of PWV‐UP was presumably inherited from PWV and constrained within the prediction limit of PWV, but it outperformed PWV. The reasonable explanation for this observation might be that the multifaceted peptides of PWV‐UP were also involved in other pathogenesis processes. For instance, MGP, a significant protein in PWV‐UP, is a small protein that inhibits vascular calcification. Because vascular calcification is considered one of the main pathogenesis variables of arterial stiffness, circulating inactive MGP has been demonstrated to be associated with aortic stiffness. , However, apart from accelerating aortic aging, vascular calcification is also a significant risk factor of mortality and cardiovascular events. , Consistent with this, Liu et al conducted a prospective study on 2318 Flemish people with 14.1 years of follow‐up and discovered that a doubling of the plasma inactive MGP was independently associated with a 15% increased risk of mortality. The association of PWV‐UP with adverse outcomes might be partly mediated by the pathological mechanism of arterial stiffness. In fact, previous studies have demonstrated that urinary proteomics‐derived classifiers could be independent predictors of cardiovascular events and coronary artery diseases. , In 791 participants (mean age, 51.2 years; 50.6% women), HF1, a classifier of 85 urinary peptides for left ventricular diastolic dysfunction, was associated with the risk of cardiovascular events after a follow‐up period of 6.1 years, instead of systolic blood pressure. Although clearly designed experiments on the role of the various proteins and their interaction with the pathogenesis of arterial stiffness are required, we hypothesized that the association of PWV‐UP and adverse outcomes is mediated by these proteins through the enriched pathways, such as fibrosis, inflammation, calcification, and cell‐ECM interactions. The constitution of urinary proteomics is not fixed, but can be modified with dynamic pathological processes, attributable to which PWV‐UP has the potential to monitor the development of arterial stiffness and provide hints for personalized treatment targets.

The urinary proteome mostly consists of endogenous peptides and low‐molecular‐weight proteins derived from larger precursor proteins and protein degradation without additional manipulation (eg, proteolytic digestion). These substantial peptides in urine provide a window into interpreting the biological functions of these endogenous peptides, their precursor proteins, and the process of degradation under various situations. Different modified collagen fragments in urine, for example, are considered markers for diabetes and diabetic nephropathy. This desirable information would be buried in the analysis at the protein level to some extent. Besides, the bottom‐up proteomics is typically measured at the peptide level, then estimates protein abundance by multiple peptide intensities, and this combining process could also introduce extra error. , With the advent of deep learning, the dimensionality and multicollinearity extended by the analysis at the peptide level might be an acceptable trade‐off for additional information. Furthermore, the definition of a polypeptide of CE‐MS was not merely based on the amino acid sequence of the precursor, but considered its mass, the migration time in capillary electrophoresis. Several distinct proteins with different posttranslational modifications might produce peptides with same sequence, but their molecule mass and the migration time in capillary electrophoresis could be altered with posttranslational modifications. The posttranslational modifications can be informative and useful for some diseases (eg, advanced glycation end products for uremia). Besides, calibrating mass and migration time enables CE‐MS to consistently detect the same peptides from different samples. The obtained peptide data from CE‐MS have been validated and showed great reproducibility in respect to intra‐day variation (0.7%–1.6%) and inter‐day variation (1.3%–5.7% for the top ten abundant peptides).

Notably, most of the significant peptides in PWV‐UP were fragments of collagen I and III. As primary constituents of ECM, they provide an architectural framework and maintain a high tensile strength for vessels. In contrast, elastic fibers offer elasticity and resilience to arteries and are abundant in elastic arteries, especially in aortic arteries. With vascular aging, the mechanical properties are altered: the elastic fibers are damaged and fragmented, and the collagen fibers accumulated excessively to adapt to the mechanical overloading. , In the volcano plot, PWV was inversely correlated with most collagen I and III peptides, which implies that arterial stiffness is correlated with fewer degraded collagen products. This is consistent with the unbalanced collagen turnover of arterial stiffness, which is also reflected by the serum collagen I turnover markers. In 80 patients with chronic heart failure, cfPWV was inversely associated with serum level of carboxy‐terminal telopeptide of collagen type I (a marker of collagen degradation), but was positively correlated with prometalloproteinase‐1, which was a marker of matrix metalloproteinase‐1 production. Similar findings were also found in subjects with hypertension. , Stakos et al measured free amino‐terminal propeptides of procollagen type I (a maker of collagen synthesis), carboxy‐terminal telopeptide of collagen type I, and matrix metalloproteinase‐1 in 72 patients with hypertension and 27 normotensive individuals. They showed that cfPWV was positively associated with the amino‐terminal propeptides of procollagen type I/carboxy‐terminal telopeptide of collagen type I ratio and matrix metalloproteinase‐1. Although matrix metalloproteases are mobilized to degrade the disproportionate collagen fibers, the additional cross‐link forged by advanced glycation end products prevents collagen from degrading to a certain extent. The urinary peptides of collagen I and III might be the biomarkers of the dynamic process of unbalanced collagen turnover in arterial stiffness.

In addition to MGP, osteopontin was another inhibitory protein of vascular calcification. Osteopontin was initially identified in osteoblasts to moderate mineralization in bones by inhibiting crystal growth. It is not detectable in the normal vascular walls, but highly expressed in the calcification sites of atherosclerosis plaque. Similarly, PWV‐UP also included collagen II, which is the primary fibrillar protein component in cartilage tissue and generally associated with cartilage and skeletal disorders. However, a previous study that collected 97 aorta samples from human subjects with sudden death reported that the expression of collagen type II was significantly higher around the sites of calcium depositions on the arterial wall and was positively associated with the grade of atherosclerosis. In addition, we also found several lipoproteins in PWV‐UP. Apolipoproteins A‐I and IV are the primary proteins of the high‐density lipoprotein, and are involved in lipid metabolism, reverse cholesterol transport, and protection against atherosclerosis. , A study investigated the relationship between PWV and the ratio of apolipoprotein B/A‐1 in 1252 subjects with metabolic syndrome. It was observed that the ratio significantly increased when PWV increased. Furthermore, PWV‐UP also included proteins that are involved in coagulation and platelet activation, such as fibrinogen, α‐1‐antitrypsin, α‐1‐antichymotrypsin, and plasminogen. For instance, fibrinogen, a key coagulation factor and acute‐phase protein, has been known to associate cardiovascular risks and has been reported to be correlated with aortic stiffness. , Other studies found that arterial stiffness is also associated with platelet activation and aggregation. , , Consistent with our findings, in previous study, the arterial stiffness index β and intima‐media thickness of the carotid artery of 517 participants were significantly associated with the expression level of P‐selectin related to platelet adhesion of the activated platelets.

Our findings add to the increasing evidence that arterial stiffness is not only hemodynamic alterations but concerns multifactorial molecules with potentially deleterious influence on mortality. This could explain why arterial stiffness is hardly explained by single clinical risk factors or biomarkers, but PWV‐UP that combined >2000 peptides can achieve better. This combination of individual biomarkers also boosted the prognostic performance of arterial stiffness. These findings suggest that PWV may not be sufficient to stratify the risk of mortality related to arterial stiffness, and the PWV‐derived proteomic biomarker might be a complementary approach.

Strengths of the current study included a relatively large sample size, standardized PWV measurement and quality control, well‐characterized participants, which reduced the confounding effect, a prospective study design, and long‐term follow‐up. However, our findings also have several potential limitations. As an observational study, potential residual confounding may exist; thus, the association was not subjected to causality inference. However, in this study, we collected a wide range of clinical risk factors, including those from biochemical tests, to eliminate the potential confounding effects. The specific roles of the PWV‐UP‐associated proteins in the pathogenesis of arterial stiffness and their effects on adverse outcomes require further validation in experimental studies. Last, the urinary proteomic profiling was developed from the general population from Flemish region. Therefore, the findings should be cautiously generalized to other ethnicities and clinical settings before further verification.

In conclusion, this study showed, for the first time, that the PWV‐derived urinary proteomic profile was significantly associated with aortic stiffness in the general population, independent of the conventional clinical risk factors, including age, sex, heart rate, mean arterial pressure, blood glucose, and current smoking. Over a 9‐year follow‐up period, PWV‐UP, but not PWV itself, predicted mortality and cardiovascular mortality, which implies that the peptides included in PWV‐UP might be involved in multiple pathophysiological mechanisms and not just limited to arterial stiffness. PWV‐UP offers the possibility of using urinary proteomics as a personalized biomarker of arterial stiffness and as a marker of adverse outcomes.

Sources of Funding

The current study was supported by the European Union (HEALTH‐F7‐305507 HOMAGE), the European Research Council (Advanced Researcher Grant 2011‐294713‐EPLORE and Proof‐of‐Concept Grant 713601‐uPROPHET), the European Research Area Net for Cardiovascular Diseases (JTC2017‐046‐PROACT), and KU Leuven (STG‐18‐00379), which currently support the Studies Coordinating Centre in Leuven.

Disclosures

The authors declare no conflicts of interest.

Data S1. Supplemental Methods

Tables S1–S5

Figures S1–S2

References 54, 55, 56, 57, 58, 59, 60, 61

Click here for additional data file.