Introduction

Telomeres are repetitive DNA–protein structures, comprising thousands of tandem repeats of the TTAGGG sequence, located at the ends of chromosomes. They serve to maintain genomic stability and determine cellular lifespan [1, 2]. With each cell division, telomere length progressively shortens because of the inability of DNA polymerase to fully replicate the 3’ end of the DNA strand. When these sequences reach a critical length, the cellular DNA damage machinery is activated, which, in turn, triggers cellular senescence [1]. Telomere length is commonly measured as leukocyte telomere length (LTL), which is relatively easy to obtain from blood samples and is highly correlated with telomere length in other tissues [3]. LTL considerably varies across individuals [3,4,5,6], including across those of the same chronological age [7]. LTL has been proposed as a biomarker of biological aging as it reflects the amount of cellular turnover within an individual.

Many epidemiological studies have evaluated LTL in relation to a range of aging-related outcomes, including all-cause mortality [8,9,10], various types of cancer [11, 12], neurodegenerative diseases [13,14,15,16], chronic kidney diseases [17, 18], diabetes [19,20,21], cardiometabolic risk factors [22,23,24], and cardiovascular diseases (CVDs) [7,8,9,10, 19, 21, 25,26,27]. Although associations of shorter LTL with cardiovascular risk factors and CVDs have been repeatedly shown, observational studies investigating the association between LTL with preclinical vascular phenotypes and markers of vascular function, including endothelial function [28, 29], hemodynamics [30,31,32,33,34,35,36], arterial stiffness [37] and blood pressure traits [23, 24, 27, 34, 38,39,40,41,42] have yielded inconsistent results. Evidence from in vitro studies suggests that telomere shortening may be part of the mechanistic pathway leading to microvascular and hemodynamic dysfunction [4, 43, 44]. Aging and senescence induced by telomere shortening reportedly cause endothelial and hemodynamic dysfunction [45,46,47], and inhibition of telomere shortening has been shown to restore endothelial function and prevent the progression of atherosclerosis [48,49,50]. A detailed assessment of the relation between LTL and quantitative and sensitive preclinical vascular phenotypes, especially microvascular and hemodynamic function, could substantially advance our understanding of the role of telomere length in the pathogenesis of CVDs [4].

LTL has a strong inherited genetic component in humans, with an estimated heritability ranging from 44 to 86% [51, 52]. Recently, a large genome-wide association study (GWAS) found 197 independent sentinel variants associated with LTL, accounting for 4.54% of the variance in LTL [53]. Of note, these variants were located in gene regions involved in telomere regulation, maintenance, as well as cellular aging and senescence. Large-scale Mendelian Randomization studies have suggested a causal relationship between LTL and CVDs [2, 5, 26, 27, 53]. However, to what extent inherited genetic variation influencing LTL relates to markers of vascular function remains largely unknown.

Leveraging existing GWAS findings, we created weighted polygenic risk scores (PRSs) of LTL as proxies for genetically predicted LTL. This enabled us to investigate whether and to what extent measured and genetically predicted LTL, as well as the difference between the two measures, i.e. telomere length independent of genetic predisposition, are associated with vascular function phenotypes, including microvascular function, hemodynamics, arterial stiffness, and blood pressure, in the general population over a wide age range. Furthermore, we performed epigenome-wide association studies of the three LTL measures, followed by gene enrichment and pathway analyses, to gain biological insights into the mechanisms underlying the effects of LTL on vascular function.

Methods

Study population

Our research was based on cross-sectional baseline data from the Rhineland Study, an ongoing single-center, population-based cohort study among people aged 30 years and older in Bonn, Germany. All individuals from the age of 30 years onwards living in two pre-defined recruitment areas are invited to participate in the study. Participants are predominantly German of Caucasian descent. The only exclusion criterion is an insufficient command of the German language to provide informed consent. A primary objective of the Rhineland Study is to identify determinants and markers of healthy aging through a deep-phenotyping approach. At baseline, participants complete an 8-h in-depth multi-domain phenotypic assessment, and various types of biomaterials (including blood, urine, stool, and hair) are collected. Approval to undertake the study was obtained from the ethics committee of the University of Bonn, Medical Faculty. We obtained written informed consent from all participants in accordance with the Declaration of Helsinki.

Baseline data of the first 4180 participants of the Rhineland Study with both genetic data and vascular phenotype data were used to assess the association between genetically predicted LTL and vascular phenotypes. In a subset of participants, both measured LTL and vascular phenotypes were available (n = 1828).

Measurement of leukocyte telomere length

LTL was manually measured using the quantitative polymerase chain reaction (qPCR) method adapted from the previously published original method [54]. Blood samples were collected between 7:00 to 9:45 in the morning from an antecubital or dorsal hand vein. Genomic DNA was extracted from buffy coat fractions of anti-coagulated blood samples using Chemagic DNA buffy coat kit (PerkinElmer, Germany) and stored at -80 °C before use. LTL was measured as the relative quantities (T/S ratio) of the telomeric TTAGGG repeat (T) and the single copy of a housekeeping gene, albumin (S). Each reaction contained 25 ng of DNA, 400 nM of the telomere length primers (tel-forward: ACA CTA AGG TTT GGG TTT GGG TTT GGG TTT GGG TTA GTGT; tel-reverse: TGT TAG GTA TCC CTA TCC CTA TCC CTA TCC CTA TCC CTA ACA) and 200 nM of the albumin primers (alb-forward: CGG CGG CGG GCG GCG CGG GCT GGG CGG AAA TGC TGC ACA GAA TCC TTG; alb-reverse: GCC CGG CCC GCC GCG CCC GTC CCG CCG GAA AAG CAT GGT CGC CTG TT) and 1 × SYBR Green PCR Mastermix (iTaq Universal SYBR Green Supermix). Seven concentrations of a reference DNA sample spanning a 128-fold range of DNA concentration (i.e. from 230 ng/ μl to 3.59 ng/ μl in twofold dilution steps) as well as the negative controls, were included in every run. The reactions were performed in triplicates for each sample using the 7900HT machine (Applied Biosystems). At the end of cycling, a dissociation curve was included to detect abnormal PCR products. A standard curve of each primer was assessed for quality control. Primers achieving 90–110% reaction efficiency and an R2 across the linear range > 0.99 were considered acceptable. No amplification in negative controls was acceptable. The Ct (cycle threshold) values that had a coefficient of variance of more than 1% of each sample were excluded from further analysis. The resulting T/S ratio was calculated for each well and the mean value of the triplicates was reported.

Based on DNA methylation levels, the relative proportion of twelve leukocyte subtypes, including basophils, eosinophils, neutrophils, monocytes, naïve B cells, memory B cells, naïve CD4T cells, memory CD4T cells, regulatory T cells, naïve CD8T cells, memory CD8T cells and natural killer cells, was derived using the “FlowSorted.BloodExtended.EPIC” R package, which is based on a reference-based deconvolution method described by Salas and colleagues [55]. To investigate the potential effects of cell type compositions of leukocytes on LTL, we further assessed the correlation between measured LTL and the twelve leukocyte subtypes.

Genetically predicted leukocyte telomere length

DNA was extracted from buffy coat samples and genotyped using Infinium Omni2.5Exome-8 BeadChip containing 2,612,357 SNPs and processed using GenomeStudio (version 2.0.5). Quality control of genotypes was performed using PLINK (version 1.9). Single-nucleotide polymorphisms (SNPs) exclusion criteria were Hardy–Weinberg disequilibrium (p < 1*10–6), minor allele frequency (< 0.01) and poor genotyping rate (< 99%) [56]. Samples with poor call rate (< 95%), abnormal heterozygosity, cryptic relatedness and gender mismatch were excluded. Since variation in population structure can cause systematic differences in allele frequencies, we used EIGENSTRAT (version 16000), which uses principal components (PCs) to detect and correct for variation in population structure [57]. Based on the EIGENSTRAT estimation, we excluded cases of non-Caucasian descent, retaining only participants from Caucasian descent for analysis. We used the 1000 Genomes phase 3 reference panel for the imputation of missing genotypes using impute2 (version 2). To include only SNPs with high imputation quality, we filtered the SNPs based on an info score metric > 0.3 [58].

Genetically predicted leukocyte telomere length was calculated based on two GWASs. Our primary analyses were based on the GWAS by Codd et al., which included 472,174 UK Biobank participants and identified 197 variants associated with LTL at genome-wide significance (5*10–8) [53]. Importantly, these variants were located in gene regions with known roles in 1) telomere regulation: genes encoding components of the telomere SHELTRIN complexes, alternative lengthening of telomeres pathway and factors that post-translationally modify key telomere proteins; 2) telomerase regulation: genes encoding core components of proteins that regulate the assembly and activity of telomerase and genes involved in TERC stability, intracellular trafficking and processing which is important before telomerase assembly; 3) telomere maintenance: DNA replication, recombination, and repair. Therefore, these variants are likely to be causally associated with LTL [53]. Moreover, the 130 SNPs selected by Codd et al. for their Mendelian Randomization study, which we used for calculating PRSMRCodd, were non-pleiotropic. To remove potentially pleiotropic loci, Codd et al. investigated the identified variants for association with 558 traits using previously curated data [59]. For each variant, evidence of pleiotropy was defined as associations within at least three different domains, which led to the selection of 130 conditionally independent, uncorrelated, and non-pleiotropic genome-wide significant instruments [53]. Thus, the identified associations between genetically predicted LTL and vascular function are likely through only genetically determined telomere length. We calculated two weighted PRS for LTL: PRSGWSCodd included 150 variants that reached genome-wide significance, and PRSMRCodd included the MR instrument variants. Individual SNPs were coded for effect allele dosage associated with longer LTL, ranging from zero (no effect alleles) to two (two effect alleles). The published regression coefficient (beta) estimates representing the per-allele effect on normalized LTL were assigned as weights for each SNP (Table S1). As a sensitivity analysis, we also created PRSs based on a GWAS by Li et al., which included 78,592 individuals of European descent [5]. They identified 52 variants independently associated with LTL at a false discovery rate (FDR) < 0.05. Among these, 20 sentinel variants reached genome-wide significance (5*10–8); 47 out of the 52 identified SNPs were available in our genetic array. PRSFDRLi was calculated using 47 variants that were significant at FDR < 0.05 and PRSGWSLi was calculated using the 20 sentinel variants that reached genome-wide significance. These four PRSs were further standardized to have a mean of 0 and a standard deviation of 1 and were used in the analyses as proxies for genetically predicted LTL.

Difference between measured and genetically predicted LTL

We defined delta LTL (ΔLTL: ΔLTLMRCodd, ΔLTLGWSCodd, ΔLTLFDRLi, ΔLTLGWSLi) as the difference between measured and genetically predicted LTL for each participant, and estimated it as the residual remaining after regressing measured LTL on PRS of LTL, adjusting for batch information of measured LTL and the first 10 genetic PCs.

Assessment of microvascular function

Microvascular function was assessed as reactive skin hyperemia (RSH) with a laser Doppler flowmetry device (Moors, UK) using a local thermal heating protocol. Skin blood flow (SBF) was measured on the ventral surface of the forearm for a total of 26 min. After 2 min of baseline SBF measurement, the area of interest was heated up to 40 degrees Celsius with an integrated heating probe, and the temperature was kept constant until the end of the examination. The peak in baseline SBF is followed by a nadir and after approximately 20 min it reaches a plateau, which is associated with the nitric oxide production capacity of the endothelial cells [60]. RSH was calculated as the percentage increase in SBF from baseline to the last 2 min of plateau level ([(Plateau SBF—Baseline SBF) / Baseline SBF)] × 100).

Assessment of hemodynamic parameters

Hemodynamics was quantified as cardiac index (CI, L/min/m2), systemic vascular resistance index (SVRI, dynes/sec/cm5/m2) and stroke index (SI, mL/m2). Cardiovascular examinations were performed in temperature-controlled rooms after the acclimatization of the participants in the study centers. Pulsatile, resistive and flow-related hemodynamics were obtained beat-to-beat in the supine position after 5 min of rest with an impedance cardiography device, which registers simultaneously electrocardiography (ECG) signals and measures blood pressure at 2-min intervals. Briefly, cardiac output (CO, L/min) was computed as stroke volume (SV, mL) multiplied by heart rate (beat per minute). CI was computed as CO divided by body surface area (BSA, m2). SVRI was calculated as mean arterial pressure (MAP, mmHg) divided by CO, multiplied by 80. SI was computed as SV divided by BSA.

Assessment of arterial stiffness

Arterial stiffness was quantified as total arterial compliance index (TACI, mL/mmHg/m2), aorta-femoral pulse wave velocity (PWV, m/s) and Ankle-Brachial Index (ABI). TACI was determined as dividing SV with brachial pulse pressure (PP, mmHg) and BSA. An integrated oscillometric femoral blood pressure cuff was used to determine aorta-femoral pulse wave velocity (PWV, m/s). The propagation time of the pulse wave was estimated as the delay between the opening of the aortic valve determined with impedance cardiography (ICG) waves and the arrival of the pulse wave to the mid-femoral cuff. PWV was calculated as the distance measured between the supra-sternal notch and the mid-femoral cuff divided by propagation time. Ankle-Brachial Index (ABI) was approximated as systolic blood pressure (SBP, mmHg) measured at the ankle divided by SBP measured at the upper arm on the same body-side, and measured separately for the left and right body sides. The lower value of ABI was used in the analyses in cases where ABI was lower than 1.40, otherwise, the higher value was used, as recommended previously [61].

Assessment of blood pressure

Systolic blood pressure (SBP, mmHg) and diastolic blood pressure (DBP, mmHg) were measured three times (separated by ten minutes intervals), using an oscillometric blood pressure device (Omron 705 IT). The measurements were performed while people were sitting in a resting chair in a quiet environment, and the average of the second and third measurements was used for further calculation. Mean arterial pressure (MAP) was calculated as (SBP + 2 × DBP)/3. Pulse pressure (PP) was defined as the difference between SBP and DBP.

Assessment of overall vascular health

An overall vascular health index was calculated as an average Z-score based on nine vascular phenotypes, including reactive skin hyperemia, cardiac index, systemic vascular resistance index, stroke index, total arterial compliance index, pulse wave velocity, ankle-brachial index, mean arterial pressure and pulse pressure.

Demographic and health variables

We included age, sex, and education level as demographic covariates. Education level was grouped as less than high school, high school, or higher. Smoking status was defined as current smokers or non-current smokers. Body mass index (BMI) was calculated as weight in kilograms divided by height in meters squared.

Statistical analysis

Data were summarized as mean ± standard deviation (SD) or counts with proportions, for continuous and categorical variables, respectively. All vascular phenotypes and LTL measures were standardized using z-scores before further analyses to enable a better comparison of the effect sizes across different physiological domains.

Association of measured LTL, genetically predicted LTL, ΔLTL with vascular phenotypes

We used multiple linear regression analyses to assess the association between measured LTL and each vascular phenotype. Models were adjusted for age, sex (women versus men), batch information of LTL, smoking status (current smokers versus non-current smokers) and BMI (model: vascular phenotype ~ measured LTL + age + sex + batch information of LTL + BMI + smoking status). As a previous study found that controlling for leukocyte compositions attenuated the association between LTL and cardiovascular risk factors by between 10 to 20% [23], we further adjusted for cell type proportions estimated from the same DNA samples, using a previously described method [55], as a sensitivity analysis.

To assess the association between measured and genetically predicted LTL, we first evaluated whether the previously reported genetic variants of LTL were associated with measured LTL in our cohort. Next, we examined the association between PRS and measured LTL. Third, we assessed the association between PRS and each vascular phenotype. All analyses were performed using multiple linear regression, adjusting for age, sex (women versus men), the first 10 genetic principal components (PCs) to account for population stratification, smoking status (current smokers versus non-current smokers) and BMI (model: vascular phenotype ~ PRS of LTL + age + sex + first 10 genetic PCs + BMI + smoking status). Finally, the association between ΔLTL and each vascular phenotype was assessed while adjusting for age, sex (women versus men), PRS, smoking status (current smokers versus non-current smokers) and BMI (model: vascular phenotype ~ ΔLTL + age + sex + BMI + smoking status + PRS of LTL). Our primary analyses were based on PRSGWSCodd and PRSMRCodd. As sensitivity analyses, we also perform the analyses with PRSFDRLi and PRSGWSLi.

To assess whether age and sex modified the associations between LTL and vascular phenotypes, we added interaction terms for the interaction between age and sex with vascular phenotype to the regression models. In case of significant interaction effects, additional sex-stratified analyses were performed.

All statistical analyses were performed using R version 3.5.2. All standardized effect estimates are reported with their 95% confidence intervals (CIs). We had very specific a priori hypotheses regarding the associations of telomere length with microvascular and hemodynamic function based on the literature [4, 43,44,45,46,47], therefore, we did not correct for multiple testing for the association between LTL measures and vascular phenotypes and set the level of statistical significance at P < 0.05.

Epigenome-wide association study of LTL measures and gene enrichment analyses

We investigated the associations between LTL measures (independent variable) and DNA methylation level (dependent variable) using multiple linear regression while adjusting for age, sex, batch effects, blood cell proportion, and the first 10 genetic PCs. FDR adjustment was applied to account for multiple comparisons: FDR adjusted q < 0.05 was considered epigenome-wide significant, while p < 1e-05 was considered to indicate suggestive significance.

Cytosine-phosphate-Guaniene sites (CpGs) showing associations with LTL measures at p < 1e-05 were searched in EWAS Catalog [62] and EWAS Atlas [63] to identify associated traits reported in previous EWAS. We also searched the known associations of the mapped gene for each CpG in previously published GWAS using the GWAS catalog [64]. We conducted further gene enrichment analyses using Gorilla [65] based on a significance-ranked gene list (i.e. from the lowest to the highest p-value of the corresponding CpG site) and summarized the results using the REViGO tool [66]. We also conducted Kyoto Encyclopedia of Genes and Genomes (KEGG) pathway analysis using the R missMethyl package [67].

Results

We included 4180 participants (56.2% women) with data on PRS of LTL and vascular phenotypes, with a mean age of 55.5 years (SD = 14.0 years, range 30 – 95 years). In the subset of 1828 participants with additionally measured LTL data, the mean age was 54.8 years (SD = 14.1 years, range from 30 – 95 years), and 56.8% were women. The two datasets did not statistically significantly differ regarding age, sex and BMI distribution. A summary of the characteristics of the study population is provided in Table 1.

Table 1 Characteristics of the study population
Full size table

Measured LTL was strongly associated with chronological age: measured LTL decreased 0.006 SD per year [95% CI: (-0.007, -0.005), p-value < 2e-1]. Compared to men, women’s measured LTL were 0.02 SD longer [95% CI: (0.05, 0.01), p-value = 0.04] (Figure S1). Measured LTL was weakly correlated with leukocyte subtypes (all r < 0.23, Figure S2).

The association between measured LTL and vascular phenotypes

Longer measured LTL was associated with better microvascular function and cardiac index: each SD increase in measured LTL was associated with 0.20 SD (95% CI: 0.03, 0.37) increase in reactive skin hyperemia, and 0.19 SD (95% CI: 0.01, 0.37) increase in cardiac index. Longer measured LTL was also associated with a lower systemic vascular resistance index, although this association did not reach statistical significance [-0.13 SD change (95% CI: -0.30, 0.04)]. Measured LTL was not significantly associated with arterial stiffness or blood pressure phenotypes (Fig. 1). The associations between measured LTL and microvascular and cardiac function remained significant after further adjustment of cell type proportions.

Fig. 1
figure 1

The association between leukocyte telomere length measurements and vascular phenotypes. Abbreviation: LTL, leukocyte telomere length; ΔLTL, difference between measured LTL and genetically predicted LTL; PRS, polygenetic risk score; FDR, false discovery rate; GWAS, genome-wide association study; CI, confidence interval; SD, standard deviation. PRSGWSCodd includes 150 variants at the genome-wide significance level (5*10–8) and PRSMRCodd includes 100 conditionally independent, uncorrelated and non-pleiotropic variants used as instruments for Mendelian Randomization (MR) in a previously study by Codd et al. [53]. PRSFDRLi includes 47 variants at false-discovery rate < 0.05 and PRSGWSLi of LTL includes 20 variants at genome-wide significance level (5*10–8). Models for each LTL measurement: Vascular phenotype ~ measured LTL + age + sex + batch information of LTL + body mass index + smoking status; Vascular phenotype ~ PRS of LTL (PRSMRCodd/PRSGWSCodd/ PRSFDRLi/ PRSGWSLi) + age + sex + first 10 genetic principal components + body mass index + smoking status; Vascular phenotype ~ ΔLTL (dLTLMRCodd/ dLTLGWSCodd/dLTLFDRLi/dLTLGWSLi) + age + sex + body mass index + smoking status + PRS of LTL

Full size image

The associations of the previously reported variants and PRS of LTL with measured LTL

When evaluating individual SNPs that had previously been associated with LTL by Codd et. al [53], we could replicate the genes (loci) with established roles in telomere biology: TERC (rs2293607) and TERT (rs2853677, rs138895564, rs79717857) regulate the formation and activity of telomerase; STN1/OBFC (rs9419958, rs10748858), RTEL1 (rs2259797, rs8114049, rs187577818) and CTC1 (rs75664430) regulate the telomere structure; SAMHD1 (rs6030416) and TYMS (rs111811424) regulate the nucleotide metabolism (Table S1a). Regarding other variants, although the p-values did not reach statistical significance, the direction of the associations and the magnitude of the estimates were quite similar to those reported previously. The replication results on prior GWAS by Li et al. [5] were quite similar: we replicated the previously reported top variants, including rs10936600 (TERC), rs2853677 (TERT), rs9419958 (OBFC1), rs75691080 (STMN3), with the same direction and even larger effect sizes (Table S1b).

All PRSs were associated with longer measured LTL in our cohort: each SD increase in PRSs was associated with around 0.14 SD increase (95% CI: 0.10, 0.19) in measured LTL. Taken together, these findings support the reliability of our LTL measurements and further validate the derived genetic instruments.

The association between genetically predicted LTL and vascular phenotypes

Longer genetically predicted LTL was associated with a better cardiac function (0.04 SD increase (95% CI: 0.01, 0.07) in cardiac index per SD increase in PRSGWSCodd and PRSMRCodd). In the sensitivity analysis, an increase in PRSGWSLi, but not PRSFDRLi, was associated with a better cardiac index, indicating that the PRS calculated using the genome-wide significant sentinel variants associated with LTL was informative and included less noise. There was no significant association of genetically predicted LTL with arterial stiffness traits and blood pressure (Fig. 1). The associations between genetically predicted LTL and vascular phenotypes were virtually identical in the subset of 1828 participants with measured LTL data.

To explore which variants drove the associations between PRS and cardiac index, we further investigated the associations of the individual genetic variants with cardiac index. Alleles mapping to distinct genes, including CDA, SLC16A4, ACYP2, LINC01122, SMC4, POT1 and STN1 (OBFC1), were associated with cardiac index (Figure S3). These genes are involved in DNA damage repair and nucleotide metabolism, which both play a vital role in cellular senescence [2, 5].

The association between ΔLTL and vascular phenotypes

Longer measured than genetically predicted LTL (higher ΔLTL) was associated with better microvascular and cardiac function independent of genetically predicted LTL: each SD increase in ΔLTL was associated with 0.07 SD increase (95% CI: 0.02, 0.12) in reactive skin hyperemia, 0.08 SD increase (95% CI: 0.02, 0.13) in cardiac index, and 0.06 SD decrease (95% CI: -0.12, -0.01) in systemic vascular resistance index. ΔLTL was not associated with arterial stiffness and blood pressure traits. Age and sex did not significantly modify the association of LTL measures with endothelial function and cardiac index (all interaction P values > 0.10). Overall, both the magnitudes and directions of the associations with vascular phenotypes were consistent in measured, genetically predicted LTL and ΔLTL (Fig. 1).

In addition, we found that ΔLTL was associated with the overall vascular health index (Fig. 1). However, measured LTL and genetically predicted LTL were not associated with the overall vascular health index, which indicates that primarily the non-genetically determined LTL contributes to overall vascular health.

EWAS of LTL measures and gene enrichment analyses

We identified 5 CpGs associated with measured LTL (Fig. 2a), 8 CpGs associated with genetically predicted LTL (Fig. 2b), and 27 CpGs associated with ΔLTL (Fig. 2c) at p < 1e-05 level.

Fig. 2
figure 2

Epigenome-wide association results of leukocyte telomere length measurements. Manhattan plots of the epigenome-wide association study (EWAS) results for (a) measured leukocyte telomere length and (b) genetically predicted leukocyte telomere length and (c) delta leukocyte telomere length. The x-axis depicts sites ordered by chromosomal position with the respective -log10 p-value on the y-axis. The red dash horizontal line represents the level of significance at p-value < 1e-05

Full size image

The CpGs we found associated with measured LTL have previously been linked to cardiac function (i.e. electrocardiogram morphology, QT interval, QT dynamics), and the CpGs associated with genetically predicted LTL have been linked to blood pressure traits (Table 2a and 2b). The CpGs associated with ΔLTL have been linked to environmental exposure (i.e. air pollution, nitrogen dioxide exposure, HIV infection), lifestyle factors (i.e. alcohol consumption, vitamin B12 supplement) and mortality in previous EWAS. Moreover, the mapped genes have been previously associated with CVDs (i.e. coronary artery disease, myocardial infarction, ischemic stroke) and other aging-related phenotypes (i.e. cancer, chronic kidney disease, type 2 diabetes, Alzheimer’s diseases, hand grip strength, cognition, neurofibrillary tangles, and cortical thickness) (Table 2c).

Table 2 CpG sites associated with leukocyte telomere length measurements and other EWAS and GWAS traits
Full size table

Genes, whose methylation status was associated with ΔLTL, were enriched in gene sets involved in the vascular endothelial signaling pathway (i.e. regulation of vascular endothelial growth factor receptor signaling pathway, cell response to vascular endothelial growth factor stimulus and regulation of Wnt signaling pathway), regulation of phospholipid metabolic process and regulation of toll-like receptor 4 signaling pathway (Fig. 3 and Table S2). Gene set was only enriched in the regulation of peptidase activity for measured LTL and regulation of membrane lipid distribution for genetically predicted LTL (Table S2). KEGG pathway analysis revealed that genes linked to ΔLTL were related to vascular smooth muscle contraction. However, all the pathways did not survive after multiple comparison corrections (Table S3).

Fig. 3
figure 3

Gene ontology enrichment of nearest genes associated with delta leukocyte telomere length. Red bubble corresponds to p-value < 1e-5, pink bubble corresponds to p-value < 1e-3. Highly similar GO terms are linked by edges in the graph, where the line width of the edges indicates the degree of similarity. The placement (position and distance) of the nodes indicates the similarity of the GO terms, which is determined by a ‘force-directed’ layout algorithm used in REVIGO tool [66] that aims to keep the more similar nodes closer together

Full size image

Discussion

Using a hypothesis-driven approach, we systematically examined the associations of three LTL measures [i.e. measured LTL, genetically predicted LTL, and the difference between the two (ΔLTL)] with sensitive quantitative markers of vascular function in the general population. We found that genetically predicted LTL was only associated with the cardiac index. Measured LTL and ΔLTL were consistently associated with microvascular function and hemodynamic traits, but not with markers of arterial stiffness or blood pressure. The consistent associations of measured and genetically predicted LTL with cardiac index indicate that longer LTL is likely to be causally related to better cardiac function. Importantly, ΔLTL was associated with microvascular function and cardiac index independent of genetically predicted LTL. Of note, ΔLTL was more strongly associated with microvascular function and cardiac index than genetically predicted LTL. This suggests that telomere shortening itself, independent of genetic predisposition, contributes to cardiovascular dysfunction [68, 69]. Additionally, genes, whose methylation levels were associated with ΔLTL, were enriched in pathways related to vascular endothelial growth factor function and have been linked to environmental exposure, CVDs as well as mortality.

We found that both longer measured LTL as well as higher ΔLTL, were associated with better microvascular function. These findings support the notion that telomere shortening could potentially cause microvascular dysfunction, which is an early feature of atherosclerosis and vascular diseases [70]. Although experimental studies have suggested that telomere function is a crucial determinant of microvascular function [44, 49, 71,72,73,74], and some studies have shown that telomere length is related to other subclinical markers of atherosclerosis [35, 75, 76], only a few clinical and epidemiological studies have investigated the association of telomere length with microvascular function [28, 29]. One cross-sectional study in 102 patients with a history of cerebrovascular diseases found shortened telomeric 3’-overhang (G-tail), but not telomere length, to be associated with microvascular function [29]. Another cross-sectional study from the LIPGENE cohort including 88 patients with metabolic syndrome also found that microvascular function, through high oxidative stress, was associated with shorter telomere length [28]. Our study, with a larger sample size and a wide age range among community-dwelling adults, not only confirms and substantially extends these previous findings but also provides evidence for a causal connection between shorter telomere length and microvascular dysfunction at the population level.

The relationship of both measured and genetically predicted LTL with hemodynamic indices indicates that telomere shortening may be a biologically important factor that contributes to the age-related decline in heart function. Experimental studies highlighted the important role of cardiac telomere length in heart development, function and disease [77, 78]. Decreases in telomere length in cardiomyocytes induced apoptosis and heart disease [79, 80]. However, there are only a few epidemiological studies assessing the association between LTL and hemodynamic traits. Of note, we found that LTL-related genetic variant mapping to Protection of Telomeres 1 (POT1) was associated with cardiac index. The POT1 protein is one of the six core proteins forming the terminal t-loop shelterin complex of telomere and it is essential for telomere length maintenance and regulation [81]. Previous experimental studies have shown that disruption of POT1 function accelerates telomere shortening, increases apoptosis, and initiates an ATR-dependent DNA damage response [81, 82], which potentially activates immune signaling and chronic inflammation [83]. Moreover, depletion of POT1 has been linked to phagocytosis and nitric oxide generation, which is involved in endothelial dysfunction [84]. Applying a population-based approach in which we leveraged new genetic findings, our findings support a causal role for shorter telomere length in the pathogenesis of cardiac dysfunction across the adult life course.

The molecular processes through which telomere length affects microvascular and cardiac function are thus far poorly understood. Our EWAS and subsequent gene enrichment analyses showed that genes, whose methylation levels were associated with ΔLTL, were enriched in pathways related to vascular endothelial growth factor function. Interestingly, the methylation status of these CpGs has been previously linked to environmental exposures, lifestyle factors as well as mortality. Moreover, the mapped genes associated with ΔLTL have been related to CVDs and other aging-related phenotypes. These findings have profound implications for our understanding of cardiovascular senescence and suggest that counteracting telomere shortening via non-genetic factors, including nutrition [85, 86], physical activity [69, 87], and sleep [88, 89], may improve microvascular and cardiac function, preventing CVDs independent of the genetic basis of LTL variation.

We found little evidence for an association between LTL measures, arterial stiffness, and blood pressure parameters. Prior studies have observed inconsistent associations between measured or genetically predicted LTL and arterial stiffness and blood pressure traits, with several null results reported [23, 32, 38, 42, 90, 91]. Studies with a larger sample size identified associations between measured/genetically predicted longer LTL with higher blood pressure, but not with arterial stiffness [24, 53]. Moreover, these 197 identified independent genomic variants associated with LTL increased the amount of explained variance in LTL [53]. However, there is an incomplete understanding of the biological underlying mechanism of telomere length with regard to blood pressure traits.

Our study has both strengths and limitations. Strengths of our study include the wide coverage of the vascular phenotypes, the inclusion of individuals across a wide age range from the general population, as well as the availability of estimates of both measured and genetically predicted LTL. This enabled internal validation of the PRS of LTL in our cohort, as well as the derivation of another metric, ΔLTL, which allowed estimation of the difference between measured LTL and genetically predicted LTL in relation to vascular phenotypes, independent of genetically predicted LTL. Indeed, we were able to replicate the previously identified top hits related to LTL and were able to construct PRSs of LTL based on SNPs with clear relevance to telomere biology, further supporting the biological plausibility of our findings. The availability of robust PRSs of LTL allowed us to provide evidence for a causal role of telomere length in cardiovascular senescence and dysfunction. Moreover, we performed an EWAS of three LTL measures (i.e. measured LTL, genetically LTL, and ΔLTL), which allowed us to further explore the mechanistic insights underlying the role of telomere length in microvascular and cardiac function. A limitation of our study is the lack of longitudinal follow-up data, which precluded the assessment of causality. Moreover, LTL was only measured in a subset of participants, however, there was no significant difference between two datasets. Furthermore, some SNPs previously identified to be associated with LTL were not available on our genetic arrays; however, PRSs created based on two GWAS showed similar results with vascular phenotypes.

In conclusion, both measured and ΔTL were consistently associated with microvascular and cardiac function. The association between validated PRSs of LTL and cardiac index supports a causal role for telomere shortening in the pathogenesis of cardiac dysfunction. Importantly, genes, whose methylation levels were associated with ΔLTL, were involved in vascular endothelial growth factor function pathways. These findings implicate telomere shortening in the mechanistic pathways underlying cardiovascular dysfunction and CVDs. Combined with prior studies, our data provide evidence that lifestyle interventions may not only reduce the risk of CVDs but also other age-related disorders [92, 93]. These findings also suggest that besides lifestyle interventions that can slow telomere shortening, the development of (pharmacological) treatments that target telomere shortening [94, 95], might improve microvascular and cardiac function independent of the genetic basis of LTL variation, preventing CVDs, and potentially also other age-related diseases.