Proteome‐Wide Mendelian Randomization Identifies Natriuretic Peptide‐B and Novel Proteins as Potential Regulators of Pulse Pressure in Humans

Abstract

Background

Large‐artery stiffness (LAS) significantly contributes to cardiovascular morbidity and death and is characterized by increased pulse pressure (PP). The biology underlying large‐artery stiffness in humans remains incompletely understood.

Methods and Results

We investigated associations between PP and circulating levels of 2941 proteins among 53 016 UK Biobank participants. Analyses were adjusted for age, sex, mean arterial pressure, body mass index and stroke volume. Interaction analyses assessed the effect modification by sex on these relationships. We evaluated causal associations between plasma protein levels and PP, using inverse variance–weighted Mendelian randomization as the main analysis and Bayesian colocalization as a sensitivity analysis. A 5% false discovery rate threshold was used to account for multiple comparisons. Measured levels of 871 proteins were significantly associated with PP when adjusting for age, sex, mean arterial pressure, and body mass index, and 61 remained significantly associated after further adjusting for stroke volume. Top associations included NPPB (natriuretic peptide B), thrombospondin‐2, paraoxonase‐2, and sclerostin. Genetic analyses indicated that genetically predicted levels for 16 proteins were significantly associated with PP after false discovery rate correction, including fibroblast growth factor 5 (β_IVW per SD change in protein levels=0.47 [95% CI, 0.34–0.61]), NPPB (β_IVW=−1.40 [95% CI, −1.85 to −0.95]), insulin‐like growth factor binding 3 (β_IVW=−1.143 [95% CI, −1.57 to −0.71]), and furin (β_IVW, 1.31 [95% CI, 0.88–1.73]).

Conclusions

Using complementary epidemiological approaches to triangulate findings, our study identifies novel proteins with a putative causal effect on PP. Notably, our findings identify NPPB with high statistical confidence. This may have potentially impactful implications given the current availability of Food and Drug Administration–approved medications to boost NPPB effects.

Nonstandard Abbreviations and Acronyms

ARCH

Active‐Controlled Fracture Study in Postmenopausal Women With Osteoporosis at High Risk

FRAME

Fracture Study in Postmenopausal Women With Osteoporosis

LAS

large‐artery stiffness

MR

Mendelian randomization

NPPB

natriuretic peptide B

PP

pulse pressure

pQTLs

protein quantitative trait loci

PWV

pulse wave velocity

SMOC2

SPARC‐related modular calcium binding 2

SV

stroke volume

Clinical Perspective.

What Is New?

• Our study identified hundreds of novel proteins associated with pulse pressure that may serve as biomarkers of large‐artery stiffness; our causal inference analyses identified 16 proteins with a putative causal effect on pulse pressure, warranting further investigation for their potential role as therapeutic candidates.

• The identification of novel biomarkers of pulse pressure highlights the importance of conducting proteomic screening and triangulating evidence from observational, genetic, and clinical evidence; this approach may guide in identifying novel therapeutic avenues and may facilitate the early diagnosis of large‐artery stiffness and downstream target organ damage.

What Are the Clinical Implications?

• Notably, our study identified natriuretic peptide B to be potentially causally associated with pulse pressure with a high level of statistical confidence; this may have important implications given the current availability of FDA‐approved medications to boost natriuretic peptide B effects.

The increasing prevalence of the aging population worldwide is an unprecedented demographic phenomenon expected to drive a growing wave of age‐related health care and economic burden globally. ¹ , ² , ³ Aging and multiple pathological conditions that are acquired over the lifetime lead to increased risk of large‐artery stiffness (LAS). ⁴ A clinical consequence of LAS is an increased systolic blood pressure during left ventricular ejection and a decreased diastolic pressure during diastolic runoff, resulting in a higher pulse pressure (PP). ⁵ Increased LAS and PP lead to excess pulsatility in the microvasculature, particularly in low‐resistance target organs such as the kidney and the brain, ⁵ and impacts left ventricular afterload and coronary perfusion pressure via its effects on pulsatile load. ⁵ As such, a higher PP, which is a direct consequence of increased LAS, has been shown to be associated with target‐organ damage (including worsening of kidney function, ⁶ , ⁷ , ⁸ , ⁹ cognitive decline, ⁵ , ⁸ and new‐onset diabetes ¹⁰ , ¹¹ , ¹² ), and is established as an independent risk factor for cardiovascular disease and all‐cause death. ¹³ Understanding the molecular basis of PP elevation may provide novel insights underlying LAS in humans, thereby leading to the improved diagnosis and treatment of LAS and its downstream clinical manifestations.

Genome‐wide association of plasma proteomics, applied among large‐population cohorts, ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ has paved the way for enhanced insights into the genetic regulation of the plasma proteome, providing a unique opportunity to evaluate the causal effects of individual proteins on the development and progression of various conditions. Mendelian randomization (MR) is a powerful method that enables the identification of causal effects in humans, overcoming some limitations of traditional observational studies such as residual confounding and reverse causality. MR leverages the naturally randomized allocation of genetic variants among the population as instrumental variables, analogous to treatment allocation in a randomized controlled trial. Under certain assumptions, this approach estimates the causal effects of exposures on outcomes. ¹⁸ Pairing plasma proteomics with human genetics for causal inferencing can provide important insights on the putative causal relationship between the human proteome and various traits and conditions. However, studies investigating the proteome in relation to the manifestation of LAS, including PP, have not been conducted.

In this study, we aimed to (1) evaluate the relationship between plasma proteins and PP in middle‐aged adults from the UK Biobank and (2) assess the putative causal associations of specific plasma proteins on PP using MR and Bayesian colocalization analysis.

Methods

The study design is illustrated in Figure 1. This study is reported using the Strengthening the Reporting of Observational Studies in Epidemiology–Mendelian Randomization guidelines ¹⁹ , ²⁰ (https://www.strobe‐mr.org) (Data S1). The UK Biobank study was approved by the UK Biobank's North West Multi‐Centre Research Ethics Committee. All participants provided written informed consent.

Figure 1. Proteome‐wide association study of 2941 proteins for pulse pressure in the UK Biobank.

BMI indicates body mass index; MAP, mean arterial pressure; pQTLs, protein quantitative trait loci; and SV, stroke volume.

Study Population

The UK Biobank is a large population‐based prospective study, established to allow investigation of the genetic and nongenetic determinants of the diseases of middle and old age. Participants aged 40 to 69 years at the time of study enrollment and were recruited from 22 assessment centers across the United Kingdom from 2006 to 2010. Participants provided informed consent at recruitment. Details of the cohort have been previously described. ²¹ Data accession was facilitated under UK Biobank application number 81032.

Proteomics Measurements

Proteomic profiling was performed on blood plasma samples collected from 53 016 UK Biobank participants using the antibody‐based Olink Explore 3072 PEA. This platform measures 2923 unique proteins across the cardiometabolic, inflammation, neurology, and oncology panels. Details on NPX processing and quality control procedures have been previously described. ²²

Proteomics Association Analyses

In proteome‐wide analyses, we assessed the relationship between individual plasma protein biomarkers and PP using linear regression models. Considering that mean arterial pressure (MAP) and body mass index (BMI) are both associated with PP, ²³ , ²⁴ we conducted regression analyses in model 1 including age, sex, MAP, and BMI as covariates. Given that stroke volume (SV) can impact PP independently of LAS, we also built models (model 2) that further adjusted for SV measured via cardiac magnetic resonance imaging. Estimates for all proteins are standardized (expressed per SD increase, or 1‐point increase in the Z score). We performed linear regression interaction analyses adjusted for age, MAP, and BMI to assess the effect modification by sex on the relationship between circulating protein levels and PP. Correction for multiple testing was performed using the false discovery rate Benjamini–Hochberg method. ²⁵ Pathway enrichment analyses are described in detail in Data S1.

Genetic Analyses

Selection of Genetic Instruments

Using proteomics data from the UK Biobank (N=53 016), we first identified cis‐acting (within ±1 Mb of the protein‐encoding region) protein quantitative trait loci (pQTLs) for circulating protein levels of each of the identified proteins in the previous analyses. Proteins were analyzed if pQTLs with association P<5 × 10⁻⁸ were available. We restricted our analyses to cis‐pQTLs as they are more likely to have protein‐specific effects than trans‐pQTLs. ²⁶ The use of cis‐pQTLs also ensures that the MR assumption of “no horizontal pleiotropy” (ie, that the instrument is not related to the outcome other than via the exposure of interest) is not violated. Independent genetic instruments were selected using a clumping threshold r ²<0.001 using a random sample of 10 000 individuals of White British ancestry in the UK Biobank as reference panel. We estimated the strength of each instrument by calculating the F statistic. An F statistic >10 was indicative of adequate instrument strength. ²⁷

Two‐Sample MR

We conducted MR analyses to estimate the effect of genetically predicted plasma protein levels on PP. MR is a statistical approach that uses genetic variants as instrumental variables to proxy the causal effects of an exposure on an outcome of interest. Three fundamental assumptions underpin the effective use of genetic instruments to facilitate robust MR experiments. First, the genetic instrument(s) must be associated with the exposure of interest (ie, the relevance assumption). Second, no confounding factors influence the association between the genetic instrument(s) and outcomes of interest (ie, the independence assumption). Third, the genetic instrument is only related to the outcome via the exposure, ensuring the absence of pleiotropic effects that may bias MR estimates (ie, the exclusion restriction assumption). ¹⁸ For our outcome data set, we defined and calculated PP as the difference between systolic and diastolic blood pressure for 230 422 UK Biobank participants and conducted a genome‐wide association study of PP, excluding participants with proteomics data to avoid sample overlap in MR analyses, as described in detail in Data S1. As a sensitivity analysis, we then conducted the same analyses while accounting for antihypertensive medication use. Specifically, we adjusted systolic blood pressure by adding 10 mm Hg and diastolic blood pressure by adding 5 mm Hg when deriving PP in participants taking antihypertensive medication. We used a Wald ratio when 1 single‐nucleotide polymorphism was available for MR, ²⁸ and inverse variance–weighted MR when >1 single‐nucleotide polymorphism was available. ²⁹ We estimated the I ² statistic to detect heterogeneity among the MR estimates obtained from multiple genetic variants. We applied adjustments for multiple comparisons using the false discovery rate method. ²⁵ We report estimated β and 95% CIs for the putative effects of these proteins on PP. β values represent the changes in PP in mm Hg for every SD change in genetically predicted plasma protein levels.

We conducted Bayesian colocalization as a sensitivity analysis to confirm that each protein and PP shared the same causal variant. This is described in further detail in Data S1.

Results

General characteristics of study participants who had proteomics data in the UK Biobank cohort (N=53 016) are shown in Table 1.

Table 1.

Baseline Characteristics of UK Biobank Pharma Proteomics Project Included in This Study (N=53 016)

	Participants (N=53 016)
Age at blood draw, y	56.80±8.21
Female, n (%)	28 581 (53.89)
Body mass index, kg/m2	27.46±4.78
Stroke volume, mL	115.65±37.16
Blood pressure, mm Hg
Systolic blood pressure	139.66±19.73
Diastolic blood pressure	82.09±10.73
Mean arterial pressure	105.12±13.08
Pulse pressure	57.57±14.96
Smoking status, n (%)
Never	28 672 (54.06)
Previous	18 489 (34.87)
Current	5600 (10.56)
Race or ethnicity, n (%)
Asian	1134 (2.14)
Black	704 (1.33)
White	49 437 (93.22)
Mixed	349 (0.66)
Other	539 (1.02)
Medication use, n (%)
Cholesterol‐lowering medication	6054 (11.41)
Antihypertensive medication	6423 (12.11)
Blood biochemistry
Total cholesterol, mmol/L	4.52±0.96
LDL cholesterol, mmol/L	1.69±0.44
HDL cholesterol, mmol/L	1.29±0.32
Triglycerides, mmol/L	1.47 (1.04–2.14)
Creatinine, μmol/L	0.06±0.01

Continuous variables are summarized as mean±SD or median (interquartile range). Categorical variables are summarized as n (%). HDL indicates high‐density lipoprotein; and LDL, low‐density lipoprotein.

Associations Between Measured Protein Levels and PP

In regression models adjusted for age, sex, MAP, and BMI (model 1), we found 871 of 2923 proteins to be significantly associated with PP. The β estimates and P values are included in Table S1. A volcano plot showing the relationship between baseline plasma protein levels and PP is shown in Figure 2A.

Figure 2. Volcano plot representing proteins significantly associated with pulse pressure in observational analyses for (A) the model adjusted for age, sex, body mass index, and mean arterial pressure; and (B) the model adjusted for age, sex, body mass index, mean arterial pressure, and stroke volume.

The plot shows β estimates against the false discovery rate corrected log‐10 P value, to better visualize the importance of each biomarker in order of significance. The dashed line represents the 5% false discovery rate‐α threshold. ANGPT2 indicates angiopoietin‐2; APOM, apolipoprotein M; BCAN, brevican; CD38, cluster of differentiation 38; CD46, cluster of differentiation 46; CD59, cluster of differentiation 59; CD93, cluster of differentiation 93; CLEC14A, C‐type lectin domain family 14 member A; COL4A1, collagen type IV α1 chain; CTHRC1, collagen triple helix repeat containing 1; CXCL10, C‐X‐C motif chemokine ligand 10; EFNA1, ephrin‐A1; FABP4, fatty acid binding protein 4; GGH, γ‐glutamyl hydrolase; GIP, gastric inhibitory polypeptide; HNMT, histamine N‐methyltransferase; IGFBP3, insulin‐like growth factor binding protein 3; IL1B, interleukin‐1β; IL1R1, interleukin‐1 receptor type 1; IL1RN, interleukin‐1 receptor antagonist; KLK6, kallikrein‐related peptidase 6; LCAT, lecithin‐cholesterol acyltransferase; LDLR, low‐density lipoprotein receptor; LEP, leptin; LPL, lipoprotein lipase; MEGF10, multiple EGF‐like domains 10; MOG, myelin oligodendrocyte glycoprotein; MUC13, mucin 13; NPDC1, neural proliferation differentiation and control 1; NPPB, natriuretic peptide B; NPROBNP, N‐terminal pro‐B‐type natriuretic peptide; NPTXR, neuronal pentraxin receptor; NRP1, neuropilin 1; OMG, oligodendrocyte myelin glycoprotein; PECAM1, platelet and endothelial cell adhesion molecule 1; PLA2G7, phospholipase A2 group VII; PLA2G10, phospholipase A2 group X; PLAT, plasminogen activator tissue type; PLATBCAN, plasminogen activator tissue type and brevican; PRG, proteoglycan; PRKAB1, protein kinase AMP‐activated noncatalytic subunit β1; PRTG, protogenin; PSAP, prosaposin; PSG1, pregnancy‐specific β1‐glycoprotein 1; PTPRN2, protein tyrosine phosphatase receptor type N2; PTPRR, protein tyrosine phosphatase receptor type R; PYY, peptide YY; REN, renin; RGMA, repulsive guidance molecule A; RTBDN, retina and optic nerve development protein; SELE, selectin E; SLC9A3R2, solute carrier family 9 member 3 regulator 2; SMOC2, SPARC‐related modular calcium binding 2; SPARC, secreted protein acidic and rich in cysteine; SPOCK1, SPARC/osteonectin, Cwcv, and kazal‐like domains proteoglycan 1; SUSSD2, Sushi domain containing 2; SYT1, synaptotagmin 1; TAFA5, TAFA chemokine like family member 5; THBS2, thrombospondin 2; TNFSF10, tumor necrosis factor superfamily member 10; and VGF, VGF nerve growth factor inducible.

The top proteins positively associated with PP included NT‐proBNP (N‐terminal pro‐B‐type natriuretic peptide), NPPB (natriuretic peptide B), angiopoietin‐2, renin, glucose‐dependent insulinotropic polypeptide, and collagen type IV α1 chain. The top proteins negatively associated with PP included SPARC/osteonectin, Cwcv, and kazal‐like domains proteoglycan 1; phospholipase A2 group VII; TNF (tumor necrosis factor) superfamily member 10; oligodendrocyte myelin glycoprotein; and plasminogen activator, tissue type. A summary of top associations is shown in Table 2.

Table 2.

List of Top 20 Proteins Associated with Pulse Pressure with a Corrected P‐value <0.001 in the UK Biobank

Protein name	Symbol	Associations with PP
		β, mm Hg	Standard error
Model 1 (adjusted for age, sex, BMI, and MAP)
N‐terminal pro‐B‐type natriuretic peptide	NTPROBNP	0.103	0.003
Natriuretic peptide B	NPPB	0.075	0.003
Angiopoietin 2	ANGPT2	0.068	0.003
Renin	REN	0.063	0.003
Gastric inhibitory peptide	GIP	0.053	0.004
Collagen type IV α1 chain	COL4A1	0.053	0.003
SPARC/osteonectin, Cwcv, and kazal‐like domains proteoglycan 1	SPOCK1	−0.044	0.003
Phospholipase A2 group VII	PLA2G7	−0.045	0.003
Tumor necrosis factor superfamily member 10	TNFSF10	−0.045	0.003
Oligodendrocyte myelin glycoprotein	OMG	−0.043	0.003
Collagen triple helix repeat containing 1	CTHRC1	0.053	0.004
Plasminogen activator, tissue type	PLAT	−0.047	0.004
Brevican	BCAN	−0.043	0.003
Cluster of differentiation 93	CD93	0.042	0.003
C‐type lectin domain containing 14A	CLEC14A	0.041	0.003
Thrombospondin 2	THBS2	0.04	0.003
Peptide YY	PYY	0.043	0.003
Repulsive guidance molecule BMP co‐receptor A	RGMA	0.039	0.003
C‐X‐C motif chemokine ligand 10	CXCL10	−0.04	0.003
Low‐density lipoprotein receptor	LDLR	−0.037	0.003
Apolipoprotein M	APOM	−0.037	0.003
Cluster of differentiation 276	CD276	0.036	0.003
Model 2 (adjusted for age, sex, BMI, MAP, and SV)
N‐terminal pro‐B‐type natriuretic peptide	NTPROBNP	0.082	0.010
Brevican	BCAN	−0.059	0.009
C‐X‐C motif chemokine ligand 10	CXCL10	−0.055	0.009
Myelin oligodendrocyte glycoprotein	MOG	−0.055	0.009
Angiopoietin 2	ANGPT2	0.05	0.009
Oligodendrocyte myelin glycoprotein	OMG	−0.048	0.009
Plasminogen activator, tissue type	PLAT	−0.055	0.010
Natriuretic peptide B	NPPB	0.049	0.009
Leptin	LEP	−0.079	0.015
Thrombospondin‐2	THBS2	0.05	0.009
Kallikrein‐related peptidase	KLK6	−0.046	0.009
Protein tyrosine phosphatase receptor type N2	PTPRN2	−0.047	0.009
Interleukin 1 receptor antagonist	IL1RN	−0.054	0.011
Phospholipase A2 group 10	PLA2G10	0.044	0.009
Retbindin	RTBDN	−0.044	0.009
Histamine N‐methyltransferase	HNMT	−0.046	0.010
Collagen type IV α1 chain	COL4A1	0.043	0.009
Cluster of differentiation 38	CD38	−0.052	0.011
TAFA chemokine‐like family member 5	TAFA5	−0.045	0.01
Synaptotagmin 1	SYT1	−0.049	0.011
SPARC/osteonectin, Cwcv, and kazal‐like domains proteoglycan 1	SPOCK1	−0.039	0.009

The top 20 proteins that were significantly associated with pulse pressure for 2 linear regression models were selected for this table. Proteins were sorted by P value. BMI indicates body mass index; MAP, mean arterial pressure; and SV, stroke volume.

The top canonical pathways associated with PP included metabolic, inflammatory, and tissue remodeling pathways, including the liver X receptor/retinoid X receptor activation pathway, the retinoid metabolism and transport pathway, granulocyte adhesion and diapedesis, inhibition of matrix metalloproteases, and glycosaminoglycan metabolism (Figure S1A).

In regression models further adjusted for SV (model 2), we found 61 of 2923 proteins to be significantly associated with PP. The β estimates and P values are shown in Table S2. A volcano plot showing the relationship between plasma protein levels and PP is shown in Figure 2B. The top proteins positively associated with PP included NT‐proBNP, angiopoietin‐2, NPPB, thrombospondin‐2, phospholipase A2 group X, and collagen type IV α1 chain. The top proteins negatively associated with PP included brevican; C‐X‐C motif chemokine ligand 10; myelin oligodendrocyte glycoprotein; oligodendrocyte myelin glycoprotein; and plasminogen activator, tissue type (Table 2).

The top canonical pathways associated with PP included signaling by platelet‐derived growth factor, phospholipases, transcriptional regulation of white adipocyte differentiation, synthesis, secretion and diacylation of ghrelin, and collagen chain trimerization (Figure S1B).

Sex‐Specific Protein Associations With PP

We formally tested for effect modification by sex across all protein associations with PP in 2 regression models using protein–sex interaction terms in regression models. We also conducted sex‐stratified analyses to obtain sex‐specific effect estimates for differentially associated proteins. After correction for multiple testing, 184 proteins exhibited sex‐differential associations with PP in model 1 (Table S3). Sex‐stratified results for proteins are presented in Tables S4 and S5. Volcano plots for sex‐stratified analyses for multivariable models adjusting for age, BMI, and MAP are depicted in Figure S2. However, when SV was added as a covariate to the regression models (model 2), none of the proteins exhibited significant sex‐differential associations with PP (Table S6). Sex‐stratified results for proteins in model 2 are shown in Tables S7 and S8. Canonical pathway results are shown in Figure S3.

Two‐Sample MR Analyses

We conducted 2‐sample MR to investigate associations for all genetically predicted plasma proteins available in the Olink panel. Results are shown in Figure 3. Considering proteins that were significantly associated with PP from our previous observational analyses, 701 of 871 had available cis‐pQTLs for MR analyses. Genetically predicted levels for 22 proteins were significantly associated with PP after false discovery rate correction (Table 3). Thirteen of 22 had concordant directions of effect with estimates from measured levels. The top 5 proteins positively associated with PP included fibroblast growth factor 5, furin, α1‐3‐N‐acetylgalactosaminyltransferase and α1‐3‐galactosyltransferase, adrenomedullin, and cartilage oligomeric matrix protein. The top 5 proteins negatively associated with PP included NPPB, cluster of differentiation 46, IGFBP3 (insulin‐like growth factor binding 3), coagulation factor XIII B chain, and latent transforming growth factor βa binding protein 2. MR results investigating the full panel of proteins and PP are presented in Table S9. Results were consistent in sensitivity analyses considering PP adjusted for antihypertensive medication use (Figure S4, Table S10).

Figure 3. Volcano plot representing proteins significantly associated with pulse pressure in Mendelian randomization analyses.

The plot shows β estimates against the false discovery rate corrected log‐10 P value, to better visualize the importance of each biomarker in order of significance. The dashed line represents the 5% false discovery rate‐α threshold. ABO indicates α1‐3‐N‐acetylgalactosaminyltransferase and α1‐3‐galactosyltransferase; ADM, adrenomedullin; CD46, cluster of differentiation 46; EFEMP1, EGF‐containing fibulin‐like extracellular matrix protein 1; F13B, coagulation factor XIII B subunit; FGF5, fibroblast growth factor 5; FOXO3, Forkhead box O3; HLA‐E, major histocompatibility complex, class I, E; IGFBP3, insulin‐like growth factor–binding protein 3; IL1B, interleukin‐1β; ITGAL, integrin subunit α L; KIF22, kinesin family member 22; KLKB1, kallikrein B1; LTBP2, latent transforming growth factor β binding protein 2; MANEAL, mannose endo‐α‐1,2‐mannosidase like; NPPB, natriuretic peptide B; OGN, osteoglycin; PECAM1, platelet and endothelial cell adhesion molecule 1; PRKAB1, protein kinase AMP‐activated noncatalytic subunit β1; PRSS53, serine protease 53; RABEPK, Rab9 effector protein with Kelch motif; SLC9A3R2, solute carrier family 9 member 3 regulator 2; SMAD3, SMAD family member 3; SMOC2, SPARC‐related modular calcium binding 2; SOST, sclerostin; SPARC, secreted protein acidic and rich in cysteine; TJAP1, tight junction–associated protein 1; UBE2L6, ubiquitin conjugating enzyme E2 L6; USP8, ubiquitin‐specific peptidase 8; YAP1, yes‐associated protein 1; and ZBTB17, zinc finger and BTB domain containing 17.

Table 3.

Protein Associations With Pulse Pressure From Mendelian Randomization and Linear Regression Analyses

Name	Category	Associations with pulse pressure from linear regression analyses				Mendelian randomization analyses
		β, mm Hg	SE	q	NSNPs	β, mm Hg	Lower 95% CI	Upper 95% CI	q
NPPB	Cardiometabolic	0.075	0.004	<0.001	1	−1.404	−1.850	−0.957	<0.001
IGFBP3	Cardiometabolic	−0.037	0.004	<0.001	3	−1.143	−1.573	−0.713	<0.001
EFEMP1	Cardiometabolic	0.040	0.004	<0.001	2	1.626	0.744	2.508	0.012
LTBP2	Cardiometabolic	0.037	0.004	<0.001	1	−1.634	−2.420	−0.848	0.003
COMP	Cardiometabolic	0.026	0.004	<0.001	4	0.755	0.381	1.129	0.005
KLKB1	Inflammation	−0.026	0.004	<0.001	3	−0.405	−0.618	−0.191	0.008
SMOC2	Inflammation	0.023	0.004	<0.001	6	−0.413	−0.621	−0.206	0.005
PRSS53	Neurology	−0.021	0.004	<0.001	9	−0.162	−0.255	−0.069	0.024
Sclerostin	Cardiometabolic	−0.021	0.004	<0.001	1	−2.459	−3.947	−0.971	0.0420
FURIN	Oncology	−0.020	0.004	<0.001	1	1.309	0.883	1.734	<0.001
TNFSF12	Inflammation	−0.016	0.003	<0.001	1	0.453	0.220	0.687	0.007
F13B	Inflammation	−0.015	0.004	<0.001	6	−0.408	−0.601	−0.216	0.002
FGF5	Inflammation	0.014	0.004	<0.001	8	0.479	0.340	0.618	<0.001
CD46	Cardiometabolic	0.013	0.004	<0.001	1	−2.722	−3.742	−1.703	<0.001
SLC9A3R2	Oncology	0.014	0.004	0.001	1	1.575	0.788	2.363	0.005
SMAD5	Oncology	−0.013	0.004	0.002	1	−3.694	−5.721	−1.667	0.013
UBE2L6	Cardiometabolic	−0.013	0.004	0.002	1	−0.669	−1.020	−0.318	0.008
ABO	Inflammation	0.012	0.004	0.006	8	0.217	0.145	0.289	<0.001
ADAMTS8	Oncology	0.010	0.003	0.010	5	−0.598	−0.908	−0.287	0.008
ITGAL	Inflammation	−0.011	0.004	0.020	1	3.742	1.442	6.042	0.045
Interleukin‐1β	Inflammation	−0.010	0.003	0.020	1	3.770	1.472	6.069	0.043
Adrenomedullin	Oncology	0.010	0.004	0.046	1	1.271	0.745	1.798	<0.001

Proteins that are shown in this table were significantly associated with pulse pressure in Mendelian randomization with an adjusted P value <0.05 after false discovery rate correction. Effect estimates from observational analyses are also shown. ABO indicates α1‐3‐N‐acetylgalactosaminyltransferase and α 1‐3‐galactosyltransferase; ADAMTS8, a disintegrin and metalloproteinase with thrombospondin motifs 8; CD46, cluster of differentiation 46; COMP, cartilage oligomeric matrix protein; EFEMP1, EGF‐containing fibulin‐like extracellular matrix protein 1; F13B, coagulation factor XIII B subunit; ITGAL, integrin alpha‐L; KLKB1, kallikrein B1; LTBP2, latent transforming growth factor β binding protein 2; PRSS53, serine protease 53; SLC9A3R2, solute carrier family 9 member 3 regulator 2; SMAD5, SMAD family member 5; SMOC2, SPARC‐related modular calcium binding 2; TNFSF12, tumor necrosis factor superfamily member 12; and UBE2L6, ubiquitin conjugating enzyme E2 L6.

Colocalization analyses suggested shared causal genetic variants between each of the proteins and PP (posterior probability for shared causal variants H₄>0.50), except for ADAMTS8 (a disintegrin and metalloproteinase with thrombospondin motifs 8), cartilage oligometric matrix protein, interleukin‐1β, integrin α‐L, TNF ligand superfamily member 12, and ubiquitin/ISG15‐conjugating enzyme E2 L6 (Table S11). Sensitivity analyses were indicative of support for the shared causal variant hypothesis at varying values of p₁₂ (Figure S5). Estimates for 16 proteins with evidence from MR and colocalization analyses supporting potentially causal associations with PP are shown in Figure 4. For all these proteins, sensitivity analyses were in support of the shared causal variant hypothesis at any prior value of the p₁₂ (Figure S5).

Figure 4. Two‐sample Mendelian randomization effect estimates for association between 16 genetically instrumented circulating plasma protein levels and pulse pressure.

Only associations between genetically predicted proteins and pulse pressure with statistically significant estimates (q<0.05) and with evidence of colocalization (posterior probability H₄>0.50) are reported. Estimates and 95% CIs are reported in mm Hg per 1 SD increase in genetically predicted plasma protein levels. ABO indicates α1‐3‐N‐acetylgalactosaminyltransferase and α1‐3‐galactosyltransferase; ADM, adrenomedullin; CD46, cluster of differentiation 46; EFEMP1, EGF‐containing fibulin‐like extracellular matrix protein 1; F13B, coagulation factor XIII B subunit; FGF5, fibroblast growth factor 5; IGFBP3, insulin‐like growth factor–binding protein 3; KLKB1, kallikrein B1; LTBP2, latent transforming growth factor β binding protein 2; NPPB, natriuretic peptide B; PRSS53, serine protease 53; SLC9A3R2, solute carrier family 9 member 3 regulator 2; SMAD5, SMAD family member 5; SMOC2, SPARC‐related modular calcium binding 2; and SOST, sclerostin.

Discussion

This is the first study to broadly assess proteomic associations of PP in an older community‐based population sample. Among 53 016 UK Biobank participants, we identified 871 proteins significantly associated with PP, including NT‐proBNP, NPPB, angiopoietin‐2; renin; glucose‐dependent insulinotropic polypeptide; phospholipase A2 group VII; SPARC/osteonectin, Cwcv, and kazal‐like domains proteoglycan 1; TNF superfamily member 10; oligodendrocyte myelin glycoprotein; plasminogen activator, tissue type; and collagen type IV α1 chain. Many of these proteins exhibit high biologic plausibility based on their known functions. Furthermore, we performed genetic analyses, providing supporting evidence of a causal effect of 16 proteins on PP, including the peptide NPPB, a well‐known hemodynamic regulator that can be enhanced with currently available therapeutic agents. Adjusted analyses did not identify sex‐specific effects, suggesting that these associations are present in both men and women. Our findings provide novel insights into the mechanisms of increased PP and LAS and prioritize candidate therapeutic targets for intervention.

Associations Between NPPB, NT‐proBNP, and PP

Our findings identified NPPB as the most significant protein to be associated with PP in both observational and MR analyses. We note that the association between genetically predicted NPPB and PP was inverse, whereas the association between NT‐proBNP and PP was direct. This is readily explained by the generally favorable biologic actions of NPPB, which contrast with the adverse epidemiologic and clinical significance of elevated NT‐proBNP levels (which usually coincide with conditions that increase the stress on cardiomyocytes). NT‐proBNP is a fragment of pro‐B‐type natriuretic peptide (BNP), the precursor molecule of BNP, and has been previously associated with adverse cardiovascular outcomes. A previous proteome‐wide association study by Azzo et al showed that increased NT‐proBNP levels in heart failure with preserved ejection fraction are associated with pathways related to fibrosis and matrix metalloproteinases ³⁰ , which are known regulators of arterial stiffening. ³⁰ , ³¹ Several studies previously investigated the role of BNP and NT‐proBNP in arterial stiffness in smaller studies. ³² , ³³ , ³⁴ , ³⁵ , ³⁶ , ³⁷ , ³⁸ Yambe et al showed that PP and brachial–ankle pulse wave velocity (PWV) were significantly positively associated with plasma BNP levels independent of age in a healthy Japanese population. ³² In the Rotterdam study population, Rutten et al demonstrated that plasma NT‐proBNP levels were significantly positively correlated with carotid–femoral PWV. ³⁸ Moreover, in a renal transplant population, plasma NT‐proBNP level was a significant positive predictor of cardio–ankle vascular index independent of conventional cardiovascular risk factors. ² Similarly, Liu et al also found NT‐proBNP to be significantly associated with brachial–ankle PWV in a patient population with Takayasu arteritis. ³⁵ In the Framingham Heart Study population, another multivariable regression done by Levy et al, exhibited a significant correlation of BNP with carotid–femoral and carotid–radial PWV as well as carotid PP across both men and women independent of conventional cardiovascular risk factors including prior CVD. ³⁴ BNP was negatively associated with carotid–femoral PWV in women, but positively associated with it in men, indicating a differential sex‐specific association, in accordance with our observational findings that revealed a sex‐differential association of NPPB and NT‐proBNP with PP, with a positive relationship observed for NT‐proBNP in males. However, it is important to note that (1) PP is highly dependent on stroke volume, which is known to differ between men and women; (2) BNP is a well‐known regulator of intravascular volume, which can impact stroke volume; and (3) observational associations cannot be taken as surrogates of causal effects. We performed adjustment for SV measured via cardiac magnetic resonance imaging, which represents the gold standard method for SV measurements. Importantly, although the association between NT‐proBNP and NPPB and PP persisted in these analyses, sex differences in these associations were no longer present when SV was considered, indicating that these depend on sex differences in SV. We also performed MR and genetic colocalization analyses to assess the potential causal effect of NPPB on PP. In agreement with the known beneficial effects of NPPB on the vasculature, we observed a negative association between genetically predicted NPPB and PP. This suggests that the positive observational association represents a counterregulatory association that may be triggered by the effects of PP and pulsatile afterload on the left ventricle, leading to an increased release of NPPB. These findings illustrate the utility of MR and genetic colocalization analyses to reduce or eliminate confounding, allowing for a better assessment of potentially causal associations.

Our identification of the association of genetically predicted NPPB and PP, suggesting a potentially causal effect, has important implications, given the current availability of Food and Drug Administration–approved medications to boost NPPB effects in humans, based on neprilysin inhibition. Interestingly, Mitchell et al demonstrated that treatment with sacubitril–valsartan, compared with enalapril, resulted in a significant reduction of aortic characteristic impedance (a key hemodynamic determinant of PP influenced by aortic root stiffness and size) as well as a reduction in brachial PP among participants with heart failure with reduced ejection fraction. ³⁶ Whereas Myhre et al showed that sacubitiril–valsartan did not lead to significant changes in plasma concentrations of BNP and NT‐proBNP in participants with heart failure with reduced ejection fraction, it did show a marked increase in urinary concentration of cGMP, which may be indicative of increased natriuretic peptide signaling. ³⁷ Hence, the difference in effect of sacubitril–valsartan as compared with enalapril on aortic root characteristic impedance and PP shown by Mitchell et al may be mediated by enhanced natriuretic peptide signaling. Further studies should be performed regarding the role of NPPB as a regulator of PP and NPPB as a candidate therapeutic target to ameliorate LAS and reduce PP.

Other Proteomic Associations With PP

We also found associations between genetically predicted levels of multiple proteins and PP, supporting putative causal associations. We identified several novel proteins associated with PP in humans, including α1‐3‐N‐acetylgalactosaminyltransferase and α1‐3‐galactosyltransferase, NPPB, IGFBP3, cluster of differentiation 46, epidermal growth factor–containing fibulin‐like extracellular matrix protein 1, coagulation factor XIII B chain, fibroblast growth factor 5, furin, latent transforming growth factor β binding protein 2, sclerostin, serine protease 53, SPARC‐related modular calcium binding 2 (SMOC2), and SMAD family member 5.

IGFBP3 was negatively associated with brachial PP in our analyses. IGFBP3 is a key protein in the insulin‐like growth factor pathway and plays a role in regulating various biological actions of the insulin‐like growth factors, including cellular proliferation, differentiation, and increased metabolic activity. ³⁹ There is conflicting literature on the role of IGFBP3 in the progression of arterial stiffness and disease. A genome‐wide study of PP nested in the Framingham Heart Study (N=8478) revealed a connection between PP and genetic loci associated with IGFBP3, suggesting a potential role for IGFBP3 in vascular stiffness. ⁴⁰ Moreover, IGFBP3 was found to act as a vascular protective protein following vascular injury, promoting revascularization by increasing endothelial nitric oxide synthase expression in human endothelial progenitor cells, leading to increased nitric oxide generation. ⁴¹ Our MR and genetic colocalization analyses support a potential causal role for IGFBP3 on PP in humans, which will require further investigation.

Sclerostin was also negatively associated with PP. Sclerostin is a secreted glycoprotein that reduces osteoblastic bone formation by inhibiting canonical Wnt/β‐catenin signaling. ⁴² Consequently, the increased availability of sclerostin may disrupt the Wnt signaling cascade. Previous studies have shown that Wnt signaling pathways play a role in various processes related to vascular aging. ⁴² These pathways have also been linked to arterial stiffness in men of African descent, a high‐risk population for hypertensive disease. ⁴³ Moreover, observational studies in humans have revealed a correlation between elevated serum sclerostin levels and carotid intima‐media thickness, vascular calcification, and arterial stiffness. ⁴⁴ , ⁴⁵ Genetic studies in humans have demonstrated that variants associated with reduced arterial sclerostin expression were linked to an increased risk of cardiovascular events. ⁴⁴ Our findings suggest a causal role for sclerostin on PP regulation, and further studies should be performed to assess whether it represents a suitable therapeutic target to reduce aortic wall calcification and stiffness. Importantly, sclerostin also regulates bone formation and resorption. An important implication of our findings relates to the current availability of an Food and Drug Administration–approved monoclonal antibody sclerostin inhibitor for the treatment of osteoporosis (romosozumab), as well as the ongoing development of novel small‐molecule inhibitors. ⁴⁶ Interestingly, romosozumab administration was associated with an increased risk of cardiovascular events (cardiac ischemia, heart failure, cerebrovascular events) in the ARCH (Active‐Controlled Fracture Study in Postmenopausal Women With Osteoporosis at High Risk) trial, in which it was compared with alendronate, although this was not seen in the larger placebo‐controlled FRAME (Fracture Study in Postmenopausal Women With Osteoporosis) trial. ⁴⁷ Whether PP increases with sclerostin inhibition, and whether this is related to an increased cardiovascular risk in this context should be examined in future studies.

Furin is a subtilisin/kex2p‐like endoprotease enzyme involved in the processing of several precursor proteins, ⁴⁸ including the cleaving of pro‐B‐type natriuretic peptide into NT‐proBNP. Furin is involved in lipid metabolism, blood pressure regulation, and atherosclerotic plaque formation ⁴⁹ and has recently been implicated in SARS‐CoV‐2 transmission. ⁵⁰ In individuals with type 2 diabetes, furin was shown to be a better predictor of cardiovascular outcomes than BNP. ⁵¹ Mouse model studies have also shown that furin inhibition decreased vascular remodeling and the progression of atherosclerosis, potentially through the modulation of matrix metallopeptidase‐2. ⁵² Additionally, Mitchell et al identified the FURIN gene locus as a susceptibility locus for arterial stiffness, measured by carotid–femoral PWV, in the Framingham study. ⁵³ Our MR and colocalization analyses highlighted a positive association between genetically predicted furin levels and increased pulse pressure, the mechanism of which remains unknown. Whether furin inhibition leads to a decrease in arterial stiffness and subsequent end‐organ damage in humans should be further investigated.

Our analyses highlighted potential putative causal effects of several inflammatory proteins on PP, including TNF ligand superfamily member 12, interleukin‐1β, and SMOC2, among others. Our findings are in line with previous epidemiological and animal studies, that have highlighted roles for interleukin‐1 and TNFα include the regulation of vascular structure and function, potentially through the inhibition of subendothelial release of nitric oxide, and increased endothelin‐1 release. ⁵⁴ Excessive activation of the TNF ligand superfamily member 12 pathway, which is upstream of TNFα, has been associated with chronic inflammation, fibrosis, and angiogenesis. ⁵⁵ Interleukin‐1β and TNFα were also shown to be higher in individuals with higher LAS measured by PWV. ⁵⁶ SMOC2 encodes a secreted modular protein that is involved in the regulation of cell–matrix interactions and is implicated in preclinical studies as a contributor to the regulation of fibrotic disorders, potentially through the modulation of inflammatory pathways. It has been suggested that SMOC2 plays a role in tissue remodeling, as it is found to be upregulated in injured aortic vessel wall in animal studies. SMOC2 inhibition was associated with the mitigation of vascular smooth muscle cells proliferation, migration, and extracellular matrix degradation. ⁵⁷ Our findings suggest that SMOC2 may be a candidate target for aortic tissue remodeling and a therapeutic strategy for LAS and its downstream consequences. Our findings encourage further investigation of inflammatory biomarkers such as TNF ligand superfamily member 12 and interleukin‐1β as therapeutic targets for atherosclerosis‐independent conditions, including the mitigation of increased PP and LAS risk.

Strengths and Limitations

Our study has notable strengths. Given the high power and comprehensive proteomics platform used to conduct our analyses, our study provided a unique opportunity to identify and report multiple novel proteins associated with PP. To our knowledge, this is the largest study to comprehensively investigate the associations of 2923 proteins with PP, and their interactions with sex, in the UK Biobank, as well as their putative causal effect as assessed using an MR framework. The combination of observational analyses and state‐of‐the‐art genetic epidemiology methods increases the robustness of our findings. Nonetheless, our findings should be interpreted in the context of their limitations. Notably, the study was conducted in a predominantly European ancestry population, thereby precluding generalization to other ancestry groups. Considering the reported ethnic disparities in the development of LAS, whereby individuals of African descent are reported to exhibit more premature vascular aging compared with individuals of European descent, ⁵⁸ , ⁵⁹ extending proteomic and genomic research to diverse populations is crucial to inform the development of more effective and equitable therapeutic strategies. Furthermore, our observational analyses were limited to participants of the UK Biobank with a mean age of 56 to 57 years, resulting in a larger proportion of postmenopausal women. Since menopause may increase age‐related LAS and PP, this age limitation may have biased association estimates between plasma protein levels and PP. As a result, we may have overlooked relevant proteins that were not considered in MR analyses. Future studies should therefore consider a broader set of proteins in a more age‐inclusive population group. Additionally, SV is an important determinant of PP in addition to arterial compliance. The development and general availability of well‐validated genetic instruments for carotid–femoral PWV (the reference metric of LAS) should be pursued for this purpose. Whereas we adjusted for SV when appropriate, we note that PP is influenced by the left ventricular ejection pattern, in addition to arterial compliance. Moreover, not all compliance in the arterial system is provided by large arteries, and therefore we cannot exclude some contribution of medium‐size (muscular artery) compliance to our findings.

Furthermore, MR analyses rely on several assumptions that impact the reliability of causal inference, which we aimed to overcome through a series of complementary methods. First, we restricted MR analyses to cis‐acting genetic instruments to limit horizontal pleiotropy. We also implemented colocalization analyses as sensitivity to test whether MR assumptions were violated at each locus. Of note, the number of genetic instruments and instrument strength varied across proteins. Instruments with more variants are favored in terms of statistical power to detect genetic protein–disease associations, potentially leading to more precise estimates. We still ensured that the genetic instruments used for the MR analyses were robust and validated our findings using sensitivity analyses. Finally, MR instruments are used as a proxy of lifelong genetic effects and the MR thereby assumes time‐fixed effects, which may differ from pharmacologic interventions that may occur over shorter time periods with different potency. Our analyses assume linearity in the genetic model and linearity in the protein‐to‐outcome effect. Violations of these assumptions could lead to inefficiencies but are unlikely to generate false‐positive results. Further investigation, including more specific drug‐target analyses and validation in model systems is required to establish the causal mechanisms and effects of protein inhibition or stimulation on PP. Prioritized candidate therapeutic targets remain to be evaluated in experimental studies.

Conclusions

Our study identifies multiple novel proteins with a putative causal effect on PP and unraveled biological sex differences in the manifestation of PP. Notably, our findings identify NPPB with a high level of statistical confidence, which could have important implications given the current availability of Food and Drug Administration–approved medications to boost NPPB effects. Our findings will lead to further investigations regarding the biology and clinical role of the identified candidate therapeutic targets.

Sources of Funding

J.A.C. is supported by National Institutes of Health grants R01‐HL 121510, R33‐HL‐146390, R01HL153646, R01‐AG058969, 1R01‐HL104106, P01‐HL094307, R03‐HL146874, and R56‐HL136730.

Disclosures

Dr Chirinos has recently consulted for Bayer, Sanifit, Fukuda‐Denshi, Bristol Myers Squibb, JNJ, Edwards Life Sciences, Merck, and the Galway‐Mayo Institute of Technology. He received University of Pennsylvania research grants from the National Institutes of Health, Fukuda‐Denshi, Bristol Myers Squibb, and Microsoft. He is named as an inventor in a University of Pennsylvania patent for the use of inorganic nitrates/nitrites in heart failure with preserved ejection fraction. He has received research device loans from Atcor Medical, Fukuda‐Denshi, Uscom, NDD Medical Technologies, Microsoft, and MicroVision Medical. Dr Gill acknowledges support by the British Heart Foundation Centre of Research Excellence at Imperial College London (RE/18/4/34215). S. Burgess is supported by the Wellcome Trust (225 790/Z/22/Z) and the United Kingdom Research and Innovation Medical Research Council (MC_UU_00002/7). The remaining authors have no disclosures to report.

Supporting information

Data S1

Figures S1–S5

Tables S1–S11