Blood pressure, proteomic vascular ageing, and incident cardiovascular disease

Abstract

Aims

Blood proteomic profiling may model vascular biological ageing with high precision. This study aimed to assess the association between blood pressure and proteomic vascular ageing, and its potential mediation role in the relationship between high blood pressure and incident cardiovascular events.

Methods and results

Among 45 387 UK Biobank participants, we developed a proteomic vascular ageing signature (vascular age gap) using 21 heart- and artery-enriched proteins via a LightGBM model. Associations with vascular age gap were evaluated for blood pressure categories and continuous systolic/diastolic blood pressure (SBP/DBP). Mediation analyses examined the role of proteomic vascular ageing in the link between high blood pressure and cardiovascular events. Compared to normotensive participants, the vascular age gap was significantly higher in those with high-normal blood pressure (0.122 ± 0.048 years, P = 0.011), Grade 1 hypertension (0.266 ± 0.049 years, P < 0.001), and Grade 2 hypertension (0.494 ± 0.071 years, P < 0.001). Non-linear associations were observed for both SBP and DBP, with thresholds near 120/80 mmHg. The associations were stronger in midlife adults. The vascular age gap may mediate 3.73–10.59% of the association between Grade 2 hypertension and cardiovascular events. No mediation was observed for peripheral artery disease in Grade 2 hypertension. The most influential proteins exhibited potential mediation effects, with TNFRSF11B and CRLF1 consistently contributing to mediation across blood pressure categories and outcomes.

Conclusion

High blood pressure is associated with accelerated proteomic vascular ageing, which potentially mediates its link to cardiovascular events. This supports vascular age gap as a modest but significant marker of blood pressure-related cardiovascular risk.

Lay summary

Measuring vascular aging using heart- and artery-enriched proteins offers a novel approach to understanding cardiovascular risk and may support clinical strategies for vascular aging assessment. Maintaining a blood pressure of ≤120/80 mm Hg is associated with the lowest levels of vascular aging. Proteomic vascular aging may partially mediate the relationship between high blood pressure and incident cardiovascular disease, with mediation effects ranging from 3.73% to 10.59%.

Graphical Abstract

Introduction

High blood pressure is the leading modifiable risk factor for cardiovascular disease (CVD).^1–3 Vascular ageing (e.g. arterial stiffness, dilation of elastic arteries), indicated by deterioration in arterial structure and function over time,⁴ may play a role in the link between high blood pressure and cardiovascular outcomes.^1,5–8 Evidence suggests that hypertension accelerates vascular ageing, as young people with hypertension have similarities in arterial changes to older people with normal blood pressure.⁷ Thus, deferring vascular ageing may be an effective strategy to prevent the incidence of CVD related to high blood pressure.

There is no universally accepted definition of vascular ageing, while various approaches and biomarkers measure different arterial characteristics.⁴ Proteomics techniques enable an innovative measure of ageing, the ‘proteomic age gap’, which reflects accelerated or decelerated biological ageing compared with chronological age. Building on this concept, a ‘vascular age gap’ can be estimated to reflect ageing of the vascular system specifically, by utilizing heart- and artery-enriched proteomic signatures.⁹ We hypothesize that high blood pressure contributes to CVD through accelerated proteomic vascular ageing. Understanding these associations may inform earlier and targeted strategies for preventing future CVD risks. Furthermore, the varying strength of association between high blood pressure and different types of CVD has suggested variations in underlying mechanisms.^2,3 Thus, quantifying the proportion of risk mediated by vascular proteomic profile may provide deeper biological insight into the differences between cardiovascular events at the molecular level.

In this study, we utilized a large proteomics dataset from the UK Biobank to estimate proteomic vascular ageing using heart- and artery-enriched proteins. We evaluated the associations between blood pressure measures and the vascular age gap and assessed its mediation role in the relationship between high blood pressure and incident CVD.

Methods

Study population

The UK Biobank is a prospective cohort with > 500 000 participants recruited throughout the UK, aged 40–69 years between 2006 and 2010. The study population included a subset of randomly sampled participants from the total UK Biobank population and had their blood sample processed with Olink proteomics assessment at baseline (n = 45 428). In brief, we included a total of 45 387 participants who had available measurements of systolic and diastolic blood pressure (SBP and DBP). The study was approved by the Northwest Multi-Centre Research Ethics Committee, and written informed consent was obtained from all participants. This study complies with the Declaration of Helsinki.¹⁰

Assessment of blood pressure

Blood pressure was measured twice at the seated position by a trained nurse using an automated Omron 705 IT electronic blood pressure monitor, or a mercury sphygmomanometer if the automated instruments were not available. Average SBP and DBP were calculated as the mean of double measurements, respectively. For those who reported antihypertensive medication use, we added 15 and 10 mm Hg to SBP and DBP to adjust treatment effects, respectively.^11,12 According to the 2020 International Society of Hypertension practice guidelines,¹³ adjusted blood pressure were classified into four categories: normal blood pressure (SBP < 130 mmHg, and DBP < 85 mmHg), high-normal blood pressure (SBP between 130 and 139 mmHg, and/or DBP between 85 and 89 mmHg), Grade 1 hypertension (SBP between 140 and 159 mmHg, and/or DBP between 90 and 99 mmHg), and Grade 2 hypertension (SBP ≥ 160 mmHg, and/or DBP ≥ 100 mmHg). We have also classified participants as having hypertension (Grade 1 or Grade 2 hypertension) or no hypertension (normal or high-normal blood pressure).

Proteomics data and identification of heart- and artery-enriched proteins

Sample selection, data processing, and quality control of proteomics data have been described elsewhere.¹⁴ We included a total of 2916 out of 2924 proteins after excluding seven proteins with over 10% missingness, and one protein failing quality control. To identify heart- and artery-enriched proteins, we used the Gene Tissue Expression Atlas (GTEx) human tissue bulk RNA-seq database,¹⁵ and followed the definition proposed by the Human Protein Atlas: at least four times higher expression in a specific organ than in other organs.¹⁶ The method has been proven applicable in identifying organ-enriched proteins elsewhere, with 24 proteins identified as heart- or artery-enriched proteins based on the SomaScan assay.⁹ Based on the Olink assay, we finally identified 21 heart- or artery-enriched proteins (see Supplementary material online, Table S1), 13 of which overlapped with the 24-protein set,⁹ while the non-overlapped proteins were solely available in either assay (11 in SomaScan but not Olink, 8 in Olink but not SomaScan).

Proteomic vascular age and vascular age gap

The method of calculating vascular age gaps was adapted from Argentieri et al.,¹⁷ with customization on model complexity to avoid overfitting with far less features (21 vs. 2897 proteins). A gradient boosting machine learning model (LightGBM) was chosen to predict proteomic vascular age due to its strong predictive performance in the UK Biobank and external validation cohorts, as compared to standard regression and neural network approaches.¹⁷ In short, we trained a LightGBM model with 21 heart- and artery-enriched proteins to predict chronological age (see Supplementary material online, Figure S1). First, in the training set (70%, n = 31 799), the hyperparameters were tuned via five-fold cross-validation using a random search method in the Optuna module, with 100 trials to maximize the average R² of the model across all folds. Second, we narrowed the search space around the best values from random search and applied Bayesian optimization to further tune these parameters with 100 trials. As a comparison, the best models trained using random search and Bayesian optimization yielded almost identical performance (R² = 0.5927 vs. 0.5924) in the training set, suggesting a near-optimal state. Third, the performance of the best-tuned model was then tested in the testing set (30%, n = 13 629). Finally, proteomic vascular ages were estimated according to the best model. Across all the steps, the maximum run was set at 1000 with 50 early stopping rounds. The relative importance of individual proteins in the models was identified by SHAP (SHapley Additive exPlanations)¹⁸ values. Vascular age calculation was carried out using Python v.3.6.11.

To calculate the vascular age gap, we conducted a local regression between proteomic vascular age and chronological age with a fraction parameter set to two-third to estimate the true population mean vascular age, to account for the potential non-linearity. Vascular age gaps were then calculated as the difference between proteomic vascular age and the population mean proteomic vascular age.

Assessment of outcomes

We assessed coronary heart disease (CHD), stroke, heart failure, arterial fibrillation, peripheral artery disease (PAD), and CVD mortality based on the International Classification of Diseases, 9th and 10th Revision (ICD-9 and ICD-10) codes, and/or self-reported medical conditions (see Supplementary material online, Table S2). Incident CVD was a composite outcome mainly by CHD and stroke. Death and death date were obtained by reviewing death certificates held by the National Health Service. Incident cardiovascular events were defined based on the first diagnosis time of events. Cardiovascular disease mortality was defined as death due to any CVD (ICD-10: I00-I99). Follow-up time was calculated from the date of baseline assessment to the date of diagnosis, death, or the censoring date (31 December 2022), whichever occurred first.

Statistical analysis

We first summarized baseline characteristics by blood pressure categories, with continuous variables described as mean [standard deviation (SD)], and categorical variables as frequencies (percentages). Generalized linear regression models were used to examine the relationships between blood pressure categories and vascular age gaps with normal blood pressure as the reference group. In Model 1, we adjusted chronological age and sex. In Model 2, we further adjusted for race/ethnicity (White or non-White), Townsend deprivation index (<median or ≥ median), obesity [body mass index (BMI) > 30, or BMI ≤ 30 kg/m²], estimated glomerular filtration rate (eGFR) (≤60, 60–90, or ≥90 mL/min/1.73 m²), smoking status (never smokers or ever smokers), alcohol consumption (moderate alcohol consumption or not), physical activity (active or inactive), diet (healthy or unhealthy), diabetes (yes or no), high cholesterol (yes or no), and history of CVD (yes or no). Detailed information about the assessment of covariates was described in the Supplemental Methods (Supplementary material online, Table S3). We also examined the associations between SBP and DBP with vascular age gaps using a restricted cubic spline with five knots at the 5th, 27.5th, 50th, 72.5th, and 95th percentiles of blood pressure measures (reference at 120 mmHg, and 80 mmHg). To evaluate the independent associations and avoid potential collinearity, we additionally adjusted the residuals of regressing SBP on DBP, or vice versa, for SBP and DBP, respectively. To further evaluate whether and how chronological age modifies the association between blood pressure categories and proteomic vascular ageing, sliding-window analysis was conducted. The window was defined with a width of 10 years starting from 39 years old; and shifted forward by 1 year at a time. A total of 22 windows were generated to cover the range of chronological age. In each window, we assessed the associations between blood pressure categories and vascular age gaps. The false discovery rate (FDR) was adjusted for multiple comparisons.

Mediation analysis was performed using the R package regmedint¹⁹ to estimate the proportion of the association between blood pressure categories and incident CVD that was potentially mediated by the vascular proteomic profile. In mediation analysis, blood pressure categories were dummy coded with the normal blood pressure as the reference. We followed a two-step method: first, we fitted mediator models between vascular age gaps and blood pressure categories using the generalized linear regression model, adjusting for all covariates; second, we fitted outcome models to evaluate the associations of blood pressure categories and risk of outcomes using Cox proportional hazard models, adjusting for all covariates and vascular age gap. We also estimated the proportion of the association between hypertension and incident CVD mediated by the vascular proteomic profile. For each type of cardiovascular event, corresponding baseline events were excluded, and the history of other events were adjusted as a covariate. For individual proteins, we selected the five most influential proteins (ELN, LTBP2, TNFRSF11B, BMP10, and CRLF1) in the LightGBM model and assessed their mediated proportions following the same process.

Sensitivity analyses were conducted to test the robustness of our findings. First, we conducted a complete case analysis after excluding participants with missing covariates (n = 35 349). Second, we repeated the analysis among participants without any baseline CVD (n = 41 064).

All statistical analyses were done with RStudio version 4.2.2. A two-tailed P value of < 0.05 was considered statistically significant.

Results

Study population and proteomic vascular age

Of 45 387 participants, 13 655 (30.1%), 13 550 (29.9%), 13 980 (30.8%), and 4202 (9.3%) had normal blood pressure, high-normal blood pressure, Grade 1 hypertension, and Grade 2 hypertension at baseline, respectively (Table 1). Participants with higher levels of blood pressure tended to be older, and men. Participants with higher blood pressure were more likely to have higher BMI, lower eGFR, unfavourable lifestyles (ever smoking, non-moderate alcohol consumption, physically inactive, and unhealthy diet), and prevalent health conditions (diabetes, high cholesterol, and CVD).

Table 1.

Baseline characteristics of the study population by blood pressure category^a

Characteristics	Blood pressure category
	Normal blood pressure (n = 13 655)	High-normal blood pressure (n = 13 550)	Grade 1 hypertension (n = 13 980)	Grade 2 hypertension (n = 4 202)
Chronological age, years	53.3 (8.1)	57.0 (8.1)	59.0 (7.5)	60.3 (7.0)
Men	4 589 (33.6%)	6 639 (49.0%)	7 287 (52.1%)	2 330 (55.4%)
White race/ethnicity	12 585 (92.7%)	12 673 (93.9%)	13 077 (94.0%)	3 880 (92.7%)
Townsend deprivation index	−1.1 (3.2)	−1.2 (3.2)	−1.2 (3.2)	−1.2 (3.2)
Systolic blood pressure, mmHg	118.2 (8.0)	137.7 (11.2)	148.7 (12.3)	166.1 (14.2)
Diastolic blood pressure, mmHg	73.3 (6.3)	82.6 (6.4)	86.8 (8.5)	94.7 (11.8)
Anti-hypertensive medication use	313 (2.3%)	2 098 (15.5%)	4 924 (35.2%)	2 606 (62.0%)
Body mass index, kg/m2	25.7 (4.2)	27.7 (4.6)	28.4 (4.9)	29.1 (5.1)
eGFR, mL/min/1.73 m2	98.2 (14.1)	94.2 (14.7)	92.0 (15.0)	90.4 (15.5)
Never smoker	7 696 (56.6%)	7 302 (54.1%)	7 347 (52.8%)	2 109 (50.4%)
Moderate alcohol consumption	7 563 (55.4%)	7 139 (52.7%)	7 105 (50.8%)	2 111 (50.2%)
Physical active	6 646 (58.5%)	6 497 (59.3%)	6 453 (58.2%)	1 882 (57.3%)
Healthy diet	5 001 (38.4%)	4 729 (36.3%)	4 813 (35.7%)	1 435 (35.4%)
Diabetes	341 (2.5%)	746 (5.5%)	1 113 (8.0%)	389 (9.3%)
High cholesterol	1 277 (9.4%)	2 560 (18.9%)	3 762 (26.9%)	1 407 (33.5%)
History of CVD	796 (5.8%)	1 243 (9.2%)	1 710 (12.2%)	574 (13.7%)

CVD, cardiovascular disease; eGFR, estimated glomerular filtration rate.

^aData are presented as mean (SD) or n (%).

The best-tuned model of proteomic vascular age had an average R² of 0.5962 in the testing set. Supplementary material online, Figure S2 shows the relative importance of individual proteins in the model, with ELN ranked as the most important protein. Proteomic vascular ages had a high correlation with chronological ages (r = 0.77, P < 0.0001). The vascular ageing level, indicated by a proteomic ageing biomarker specific to the vascular system (vascular age gap), showed no correlation with chronological age (see Supplementary material online, Table S4, r < 0.001, P = 0.88).

Associations between blood pressure and vascular age gap

Significant associations were observed between blood pressure categories and vascular age gaps (Table 2). Specifically, vascular ageing was accelerated for 0.122 ± 0.048 (P = 0.011) years in high-normal blood pressure than normal blood pressure; for Grade 1 and Grade 2 hypertension, vascular ageing was accelerated for 0.266 ± 0.049 and 0.494 ± 0.071 years (all P < 0.001) after adjusting for all covariates, respectively. The associations remained significant in sensitivity analyses, with slightly higher coefficients, suggesting the robustness of the associations between blood pressure categories and vascular age gaps (see Supplementary material online, Table S5). Significant and non-linear associations between both SBP and DBP and vascular age gaps were found (Figure 1). There was a significantly positive association when SBP was over 120 mmHg (the reference point) while no apparent relationship was found before the reference value (Figure 1A, P for non-linear = 0.002). We found a similar positive association between DBP and vascular age gaps, which became pronounced beyond the reference point of 80 mmHg (Figure 1B, P for non-linear < 0.001).

Table 2.

The association between blood pressure categories and vascular age gap

	Blood pressure category
	Normal blood pressure	High-normal blood pressure		Grade 1 hypertension		Grade 2 hypertension
		β (SD)a	P-value	β (SD)a	P-value	β (SD)a	P-value
Model 1b	Ref	0.106 (0.049)	0.03	0.242 (0.05)	<0.001	0.446 (0.072)	<0.001
Model 2c	Ref	0.122 (0.048)	0.011	0.266 (0.049)	<0.001	0.494 (0.071)	<0.001

CVD, cardiovascular disease; eGFR, estimated glomerular filtration rate; SD, standard deviation.

^aUnit in years.

^bModel 1: chronological age and sex.

^cModel 2: model 1 + race, Townsend deprivation index, obesity, eGFR, smoking status, moderate alcohol consumption, regular physical activity, healthy diet, diabetes, high cholesterol, and CVD history.

Figure 1.

The association between vascular age gaps and continuous blood pressure measures. (A) systolic blood pressure; (B) diastolic blood pressure. All model adjusted for chronological age, sex, race, Townsend deprivation index, obesity, estimated glomerular filtration rate, smoking status, moderate alcohol consumption, regular physical activity, healthy diet, diabetes, high cholesterol, and history of cardiovascular disease. The residuals of regressing systolic blood pressure on diastolic blood pressure (or vice versa) were additionally adjusted for systolic blood pressure and diastolic blood pressure.

Significant associations (FDR < 0.05) were found between Grade 2 hypertension and vascular age gaps from 44 to 65 years old, with the coefficients decreasing at around 56 years old (Figure 2). Similarly, Grade 1 hypertension was significantly associated with vascular ageing from 44 to 57 years old, while the coefficients started to decrease earlier at around 51 years old. High-normal blood pressure and normal blood pressure showed no difference in vascular age gaps (see Supplementary material online, Table S6).

Figure 2.

Sliding-window analysis of the association between blood pressure categories and vascular age gaps. X axis indicates the median chronological age of windows, and the Y axis indicates the regression coefficient of blood pressure categories on vascular age gap, with normal blood pressure as the reference. The circle marks indicate a statistically significant regression coefficient (FDR < 0.05). All model adjusted for chronological age, sex, race, Townsend deprivation index, obesity, estimated glomerular filtration rate, smoking status, moderate alcohol consumption, regular physical activity, healthy diet, diabetes, high cholesterol, and history of cardiovascular disease.

Vascular age gap as a potential mediator

With a median follow-up of 13.76 years, we documented 4821 CVD, 3595 CHD, 1176 stroke, 1889 heart failure, 3278 atrial fibrillation, 769 PAD, and 1004 CVD mortality. The results of mediation analysis, including total effects, direct effects, indirect effects, and proportion mediated by mediator, are shown in Table 3. The total effects indicate hazard ratios of incident cardiovascular events comparing blood pressure categories to normal blood pressure. The direct effects represent the primary effects of higher blood pressure categories, independent of vascular ageing. The indirect effects suggest the effect of high blood pressure on the incidence of cardiovascular events through accelerated vascular ageing. The proportion mediated outlines the proportion of total effects mediated by proteomic vascular ageing.

Table 3.

Mediation analysis for blood pressure categories, vascular age gaps, and incident CVD

Blood pressure categories	Outcomes	Total effect		Direct effect		Indirect effect		Proportional mediated, %
		HR (95%CI)	P-value	HR (95%CI)	P-value	HR (95%CI)	P-value	Estimate	P-value
High-normal blood pressure vs. normal blood pressure	CVD mortality	1.04 (0.85–1.28)	0.684	1.04 (0.84–1.27)	0.729	1.006 (0.997–1.016)	0.172	15.45 (−59.61–90.50)	0.687
	CVD	1.31 (1.19–1.44)	<0.001	1.31 (1.19–1.44)	<0.001	1.004 (0.999–1.007)	0.074	1.49 (−0.19–3.16)	0.082
	CHD	1.29 (1.16–1.44)	<0.001	1.29 (1.16–1.43)	<0.001	1.002 (0.999–1.006)	0.141	1.07 (−0.40–2.54)	0.152
	Stroke	1.22 (1.01–1.48)	0.041	1.22 (1.01–1.48)	0.046	1.005 (0.998–1.013)	0.15	2.89 (−1.68–7.45)	0.215
	Heart failure	1.14 (0.98–1.33)	0.082	1.14 (0.98–1.32)	0.096	1.006 (0.999–1.013)	0.107	4.67 (−2.65–11.99)	0.211
	Atrial fibrillation	1.05 (0.94–1.17)	0.354	1.05 (0.94–1.17)	0.402	1.005 (0.999–1.012)	0.129	10.03 (−13.69–33.75)	0.407
	PAD	1.06 (0.84–1.33)	0.643	1.05 (0.83–1.33)	0.669	1.004 (0.998–1.011)	0.209	7.93 (−26.62–42.48)	0.653
Grade 1 hypertension vs. normal blood pressure	CVD mortality	1.21 (0.99–1.47)	0.059	1.18 (0.97–1.44)	0.094	1.022 (1.010–1.033)	<0.001	34.67 (23.86–45.49)	<0.001
	CVD	1.50 (1.37–1.65)	<0.001	1.49 (1.35–1.64)	<0.001	1.010 (1.005–1.015)	<0.001	12.23 (−0.60–25.06)	0.062
	CHD	1.43 (1.28–1.59)	<0.001	1.41 (1.27–1.58)	<0.001	1.009 (1.004–1.015)	<0.001	2.93 (1.39–4.47)	<0.001
	Stroke	1.60 (1.33–1.93)	<0.001	1.58 (1.31–1.91)	<0.001	1.012 (1.005–1.020)	0.001	3.11 (1.29–4.93)	0.001
	Heart failure	1.38 (1.20–1.60)	<0.001	1.36 (1.17–1.57)	<0.001	1.019 (1.010–1.029)	<0.001	3.19 (1.02–5.35)	0.004
	Atrial fibrillation	1.19 (1.07–1.32)	0.001	1.17 (1.05–1.30)	0.003	1.016 (1.008–1.023)	<0.001	6.74 (2.71–10.77)	0.001
	PAD	1.19 (0.94–1.49)	0.143	1.17 (0.93–1.47)	0.171	1.011 (1.003–1.019)	0.006	9.63 (2.82–16.43)	0.006
Grade 2 hypertension vs. normal blood pressure	CVD mortality	1.55 (1.23–1.97)	<0.001	1.49 (1.18–1.89)	0.001	1.039 (1.019–1.060)	<0.001	10.57 (3.78–17.36)	0.002
	CVD	1.69 (1.50–1.91)	<0.001	1.67 (1.47–1.88)	<0.001	1.016 (1.007–1.025)	<0.001	3.94 (1.74–6.15)	<0.001
	CHD	1.51 (1.32–1.74)	<0.001	1.49 (1.30–1.71)	<0.001	1.013 (1.004–1.021)	0.003	3.73 (1.11–6.34)	0.005
	Stroke	1.98 (1.58–2.49)	<0.001	1.92 (1.53–2.41)	<0.001	1.032 (1.015–1.050)	<0.001	6.31 (2.70–9.91)	0.001
	Heart failure	1.49 (1.25–1.79)	<0.001	1.45 (1.21–1.74)	<0.001	1.031 (1.016–1.047)	<0.001	9.19 (3.74–14.64)	0.001
	Atrial fibrillation	1.33 (1.16–1.52)	<0.001	1.29 (1.13–1.48)	<0.001	1.027 (1.014–1.040)	<0.001	10.59 (4.23–16.94)	0.001
	PAD	1.22 (0.91–1.63)	0.185	1.19 (0.89–1.60)	0.233	1.020 (1.003–1.037)	0.019	11.00 (−6.10–28.1)	0.207
Hypertension vs. no hypertension	CVD mortality	1.34 (1.16–1.54)	<0.001	1.3 (1.13–1.51)	<0.001	1.025 (1.016–1.034)	<0.001	9.65 (4.46–14.84)	<0.001
	CVD	1.39 (1.3–1.49)	<0.001	1.38 (1.29–1.48)	<0.001	1.01 (1.006–1.014)	<0.001	3.57 (2.16–4.98)	<0.001
	CHD	1.3 (1.21–1.41)	<0.001	1.29 (1.2–1.39)	<0.001	1.01 (1.006–1.013)	<0.001	4.09 (2.22–5.96)	<0.001
	Stroke	1.63 (1.42–1.87)	<0.001	1.61 (1.4–1.84)	<0.001	1.013 (1.007–1.019)	<0.001	3.24 (1.62–4.86)	<0.001
	Heart failure	1.35 (1.22–1.5)	<0.001	1.32 (1.19–1.47)	<0.001	1.022 (1.015–1.03)	<0.001	8.37 (4.79–11.96)	<0.001
	Atrial fibrillation	1.19 (1.1–1.29)	<0.001	1.17 (1.09–1.26)	<0.001	1.017 (1.012–1.023)	<0.001	10.57 (5.3–15.85)	<0.001
	PAD	1.25 (1.06–1.47)	0.009	1.23 (1.04–1.45)	0.015	1.015 (1.008–1.023)	<0.001	7.57 (1.48–13.66)	0.015

CHD, coronary heart disease; CI, confidence interval; CVD, cardiovascular disease; eGFR, estimated glomerular filtration rate; HR, hazard ratio; PAD, peripheral artery disease.

All model adjusted for age, sex, race, Townsend deprivation index, obesity, eGFR, smoking status, moderate alcohol consumption, regular physical activity, healthy diet, diabetes, high cholesterol, and other CVD history.

None of the associations were mediated by vascular age gaps when comparing high-normal blood pressure to normal blood pressure. Vascular age gaps might explain 3.94% (95%CI: 1.74–6.15%; P < 0.001) of the total effect of Grade 2 hypertension on CVD. As for CHD, the vascular age gap might mediate ~ 3% of the total effects of higher blood pressure categories. For stroke, significant mediation by vascular age gap was observed for Grade 1 hypertension [3.11% (95%CI: 1.29–4.93%; P = 0.001)] and Grade 2 hypertension [6.31% (95%CI: 2.70–9.91%; P = 0.001)] compared to normal blood pressure. The effects of elevated blood pressure on the incidence of heart failure might also be mediated through vascular age gap, when blood pressure reaches Grade 1 hypertension [3.19% (95%CI: 1.02–5.35%; P = 0.004)] and Grade 2 hypertension [9.19% (95%CI: 3.74–14.64%; P = 0.001)]. The largest mediated proportion by vascular age gap was found for atrial fibrillation by 10.59% (95%CI: 4.23–16.94%; P = 0.001) when comparing Grade 2 hypertension to normal blood pressure. Unlike other outcomes manifesting in heart or brain, the mediation by proteomic vascular ageing was only statistically significant for PAD when comparing Grade 1 hypertension to normal blood pressure (P = 0.006) but not Grade 2 hypertension, potentially due to the limited number of incident events. When comparing hypertension vs. no hypertension, all the associations with incident cardiovascular events were significantly mediated by vascular age gaps, with mediated proportions ranging from 3.24% to 13.3% across subtypes. In sensitivity analyses (see Supplementary material online, Tables S7 and S8), observed patterns of mediation by vascular age gaps were consistent.

Figure 3 shows the mediation proportion by individual proteins for the associations between blood pressure categories and incident cardiovascular events. TNFRSF11B (tumour necrosis factor receptor superfamily member 11B; Osteoprotegerin) and CRLF1 (cytokine receptor-like factor 1) might consistently mediate the effects of high blood pressure on CVD, CHD, and heart failure across all blood pressure categories. Additionally, TNFRSF11B might also mediate the effect of elevated blood pressure on incident stroke. In contrast, BMP10 showed minimal mediation effects, and LTBP2 might play a role only in the context of Grade 2 hypertension. ELN, although the top-ranking protein in the LightGBM model, appeared to mediate only selected blood pressure-CVD associations. When comparing hypertension vs. no hypertension, individual proteins might mediate the associations with incident CVD more consistently, except BMP10.

Figure 3.

Mediation proportions of proteins in the association between blood pressure categories and cardiovascular events.

Discussion

In the current study, elevated blood pressure was cross-sectionally associated with accelerated proteomic vascular ageing. Both high SBP and DBP demonstrated non-linear associations with vascular age gaps, with ≤120/80 mmHg as the optimal SBP/DBP for maintaining the smallest vascular age gaps. Finally, proteomic vascular ageing significantly mediated the relationship between Grade 2 hypertension and most incident cardiovascular events except for PAD, with varying mediation proportions ranging from 3.73% to 10.59%.

Comparison with previous studies

To our knowledge, this is the first study to evaluate the associations between high blood pressure and proteomic vascular ageing. Compared to established ageing clocks based on epigenomic or metabolomic data, proteomic ageing may offer greater translational potential, as proteins represent the most frequent drug targets and key functional intermediaries between genes and disease.^20–22 Moreover, by leveraging heart- and artery-enriched proteins, our vascular age gap model provides an system-specific measure of biological ageing, which may better capture vascular-related ageing processes than whole-body ageing clocks. Compared to traditional vascular ageing measures, the proteomic signature offers notable advantages. Recommended clinical measures primarily reflect structural and functional abnormalities, such as atherosclerosis [e.g. coronary artery calcification (CAC)] and arteriosclerosis (e.g. pulse wave velocity).^23,24 However, these measures may not capture the underlying molecular drivers of vascular ageing.⁴ In contrast, proteomic ageing signatures reflect a broader range of biological processes. As such, proteomic vascular ageing may provide complementary or additive information to traditional measures. For example, while CAC accompanies the development of advanced atherosclerosis,²⁵ proteomic signatures may detect subclinical vascular dysfunction that precedes structural changes. Future research should compare and integrate proteomic and clinical vascular ageing measures to assess their combined predictive power and evaluate whether they reflect distinct biological pathways. This may enhance early detection, enable more precise risk stratification, and inform tailored cardiovascular prevention strategies.

Hypertension and biological ageing

Biological ageing and hypertension share common mechanisms, such as inflammation, oxidative stress, and endothelial dysfunction, creating an interconnected relationship where ageing promotes hypertension, and hypertension accelerates vascular ageing.⁵ Unlike the extensively studied associations where ageing causes hypertension,^26,27 how high blood pressure accelerates biological vascular ageing is less understood. The associations between high blood pressure and proteomic vascular ageing might be modified by chronological age, with stronger associations in midlife and weaker associations in older age. For example, controlling Grade 2 hypertension before the age of 55 could decelerate vascular ageing by up to 0.8 years; however, these benefits are reduced by half, to < 0.4 years, after the age of 60. This age-dependent trend may be explained by the inverse U-shaped trajectories of SBP and DBP observed in the two decades before death,²⁸ where older people are more likely to be misclassified into lower blood pressure categories due to natural reductions in SBP and DBP, thus attenuating the associations. Despite the weaker associations in older age, the consistent significance of Grade 2 hypertension underscores the importance of maintaining lower blood pressure levels throughout the lifespan. Consistently, a previous study reported similar age-dependent trajectories of the association between blood pressure and vascular ageing acceleration, with the critical age at ~ 64 years old.²⁹ Taken together, these findings suggest the benefits of early blood pressure management as well as lifetime persistence in managing high blood pressure.

Our findings regarding the relationships between SBP/DBP and vascular ageing acceleration provide insight into optimal blood pressure targets for vascular ageing management. Targeting an SBP of < 120 mmHg might yield the most effective protection against vascular ageing, consistent with results from the Systolic Blood Pressure Intervention Trial (SPRINT).³⁰ Lowering DBP to < 80 mmHg could be a reasonable target for vascular ageing management; however, more aggressive targets (e.g. <60 mmHg) require cautious evaluations because of increased risk of CVD and mortality with isolated low DBP.^31,32 This underscores the need for balanced and individualized blood pressure management strategies.

Implications of mediation effects

Intriguingly, mediation analyses revealed a statistically significant, although modest (<10%), mediation of proteomic vascular ageing in the association between high blood pressure and cardiovascular events. These modest effects may still have meaningful clinical relevance given the widespread prevalence of hypertension and its impact on public health.³³ Moreover, the proteomic vascular age gap likely reflects biological processes such as inflammation, endothelial dysfunction, and vascular remodelling, pathways not directly targeted by conventional antihypertensive medications.³⁴ The development of cardiovascular disease is multifactorial and involves complex pathophysiological processes. The mediation effects, though modest, suggest that proteomic vascular ageing captures a potentially distinct and complementary mechanism in the development of cardiovascular events. Interventions aimed at slowing or reversing molecular vascular ageing may enhance the benefits of blood pressure control and offer additional strategies for cardiovascular risk reduction.

Further analysis of individual proteins revealed biological plausibility. TNFRSF11B, known for its involvement in vascular calcification and atherosclerosis,^35,36 showed consistent mediation effects across blood pressure categories. While its causal role in humans remains uncertain, prior studies have shown its involvement in inflammation, endothelial dysfunction, and vascular remodelling, and animal models suggest it may be a potential therapeutic target.^37–39 In contrast, LTBP2, associated with cardiac fibrosis and right ventricular dysfunction,^40,41 displayed notable mediation effects for heart failure and atrial fibrillation in Grade 2 hypertension, though its mechanistic role remains less understood and is likely correlational. These findings underscore the value of using a composite proteomic vascular ageing signature, which integrates multiple pathways and biological processes, rather than relying on individual proteins. At the same time, they point to specific proteins such as TNFRSF11B and LTBP2 as potential biomarkers or therapeutic targets that warrant further investigation. Given the exploratory nature of our mediation analysis, interpretations of individual protein effects should remain cautious, avoiding causal inference without further mechanistic evidence.

It is noteworthy that the contributions of proteomic vascular ageing differed between PAD and other cardiovascular events, potentially indicating heterogeneity in ageing pathways across vascular territories. Although our proteomic signature effectively captures mediation for central cardiovascular outcomes (e.g. CHD and stroke), it may be less sensitive to ageing processes relevant to peripheral arteries. This aligns with prior pathological evidence showing that PAD is more often characterized by thrombosis and medial calcification, sometimes in the absence of extensive atherosclerosis, whereas coronary disease typically involves atherosclerotic plaque instability.^42,43 While these differences do not directly establish proteomic mechanisms, they support the possibility of regional vascular ageing heterogeneity that may be reflected at the molecular level. Alternatively, our findings may be limited to the proteomic panel’s ability to capture relevant biology for PAD, or reduced power due to fewer incident PAD cases. Future research should aim to develop territory-specific proteomic ageing signatures that can better capture vascular ageing across the full spectrum of cardiovascular conditions.

Strengths and limitations

The major strengths of this study lie in its novel approach to studying vascular ageing at the molecular level, powerful machine learning methods, and a large general population. The study’s limitations should be acknowledged. First, while our proteomic vascular ageing measure shows promise, further validation is needed in populations with broader ethnic and geographic diversity. Although related proteomic ageing methods have demonstrated generalizability across different populations from the UK, China, and Finland,¹⁷ vascular-specific ageing methods warrant similar testing. Second, our study used cross-sectional data, which limits causal interpretation between high blood pressure and proteomic vascular ageing. Longitudinal analyses are needed to examine whether elevated blood pressure precedes acceleration in proteomic vascular ageing and whether changes in vascular age gap predict future cardiovascular risk. Such validation could also inform optimal timing for interventions. Third, the current set of vascular-enriched proteins was derived from the Olink platform. Further studies comparing vascular ageing estimates across platforms and identifying a consensus protein panel would enhance the robustness and reproducibility of this approach. Fourth, our vascular age model relied on a limited panel of 21 heart- and artery-enriched proteins, which may overlook important ageing-related pathways. Future studies could refine this panel by incorporating functionally validated proteins across diverse vascular ageing mechanisms. Fifth, while lifestyle factors such as smoking may influence vascular ageing through inflammatory and proteomic pathways,^44–46 their dynamic effects over time were not assessed in this study, highlighting the need for longitudinal research to capture these changes and their clinical implications. Sixth, we classified hypertension according to the 2020 International Society of Hypertension guidelines,¹³ which differ slightly from the European Society of Hypertension guidelines⁴⁷ by merging Grade 3 hypertension (SBP ≥ 180 mmHg or DBP ≥ 110 mmHg) into Grade 2. Although this approach simplifies the classification, the very small number of participants with extremely high BP and the similar clinical management recommended for individuals with BP ≥ 160/100 mmHg support this choice and make it unlikely to affect the study conclusions.

In conclusion, high blood pressure was associated with accelerated proteomic vascular ageing, and proteomic vascular ageing appeared to partly mediate the observed association between high blood pressure and cardiovascular events. While causality cannot be inferred from this observational study, these findings highlight the potential relevance of vascular ageing in the pathway linking blood pressure to cardiovascular risk and suggest that the proteomic vascular age gap warrants further investigation as a biomarker in intervention studies.

Supplementary Material

Supplementary material is available at European Journal of Preventive Cardiology.

Data availability

UK Biobank is an open-access database, available by application.