Metabolomics reveals a link between homocysteine and lipid metabolism and leukocyte telomere length: the ENGAGE consortium

Telomere shortening has been associated with multiple age-related diseases such as cardiovascular disease, diabetes, and dementia. However, the biological mechanisms responsible for these associations remain largely unknown. In order to gain insight into the metabolic processes driving the association of leukocyte telomere length (LTL) with age-related diseases, we investigated the association between LTL and serum metabolite levels in 7,853 individuals from seven independent cohorts. LTL was determined by quantitative polymerase chain reaction and the levels of 131 serum metabolites were measured with mass spectrometry in biological samples from the same blood draw. With partial correlation analysis, we identified six metabolites that were significantly associated with LTL after adjustment for multiple testing: lysophosphatidylcholine acyl C17:0 (lysoPC a C17:0, p-value = 7.1 × 10−6), methionine (p-value = 9.2 × 10−5), tyrosine (p-value = 2.1 × 10−4), phosphatidylcholine diacyl C32:1 (PC aa C32:1, p-value = 2.4 × 10−4), hydroxypropionylcarnitine (C3-OH, p-value = 2.6 × 10−4), and phosphatidylcholine acyl-alkyl C38:4 (PC ae C38:4, p-value = 9.0 × 10−4). Pathway analysis showed that the three phosphatidylcholines and methionine are involved in homocysteine metabolism and we found supporting evidence for an association of lipid metabolism with LTL. In conclusion, we found longer LTL associated with higher levels of lysoPC a C17:0 and PC ae C38:4, and with lower levels of methionine, tyrosine, PC aa C32:1, and C3-OH. These metabolites have been implicated in inflammation, oxidative stress, homocysteine metabolism, and in cardiovascular disease and diabetes, two major drivers of morbidity and mortality.

Telomeres are located at the ends of chromosomes and protect against spontaneous DNA damage, thus preserving genomic integrity 1,2 . The progressive shortening of telomere length with each subsequent cell division underlies the so-called mitotic clock, i.e. the limited replicative capacity of a cell 3 . Replicative senescence and subsequent cell death occurs when the mean telomere length reaches a critical value and telomere length is therefore seen as a marker for biological age 4-6 . Short leukocyte telomere length (LTL) has been associated with age 5,7-9 and multiple age-related diseases such as cardiovascular disease [10][11][12][13][14][15] , diabetes 10,16,17 and dementia 18,19 . Short LTL has also been associated with mortality [20][21][22][23][24][25][26][27] , although not all studies support this association [28][29][30][31][32][33] . However, the biological mechanisms underlying the associations of LTL with age-related diseases and mortality are still largely unknown.
Longevity in humans has previously been associated with various metabolic traits in the elderly, including traits related to blood pressure and lipids, suggesting that changes at the metabolic level are key features in longevity [34][35][36] . Metabolic profiles have been associated with age and various age-related diseases, such as type 2 diabetes, atherosclerosis, cancer, and Alzheimer's Disease [37][38][39][40][41][42][43][44] . Only a few studies investigated the association of metabolic markers with LTL, all using untargeted metabolomics [45][46][47] . One study focused on metabolic markers involved in aging and early development in 6,055 individuals included in the TwinsUK registry 45 . Although a combined set of 22 metabolites was strongly correlated with age and age-related traits, the individual metabolites were not significantly associated with LTL 45 . Another study identified 19 metabolites associated with LTL in a small sample of American Indians (n = 423) 46 . The most recent study was conducted in 3,511 females from the TwinsUK registry, reporting associations of five metabolites with LTL. These include gamma-glutamyltyrosine, gamma-glutamylphenylalanine, 1-stearoylglycerophosphoinositol, 1-palmitoylglycerophosphoinositol, and 4-vinylphenol sulfate 47 .
In the current study, we used a standardized targeted metabolomics approach to investigate the association between LTL and serum metabolites of key biochemical pathways in the largest sample so far consisting of 7,853 individuals from seven independent population-based cohorts from Europe and Australia. We further aimed to disentangle the metabolic pathways that are represented by the metabolites significantly associated with LTL.

Methods cohort descriptions and measurements of LtL and metabolites. The cohorts included in this study
are the Cooperative Health Research in the Region of Augsburg (KORA) study, the Netherlands Twin Register (NTR), the Estonian Genome Center University of Tartu (EGCUT) study, the TwinsUK cohort, the Erasmus Rucphen Family (ERF) study, the Leiden Longevity Study (LLS), and the Queensland Institute of Medical Research (QIMR) study, all part of the ENGAGE consortium. Details on the individual cohorts as well as details on the LTL quantitative polymerase chain reaction measurements and the metabolites as measured with the AbsoluteIDQ TM p150 kit (BIOCRATES Life Sciences AG, Innsbruck, Austria) are provided in the Supplementary Materials. In summary, both LTL and metabolite concentrations were measured in the same laboratories according to a common protocol, using blood samples taken at the same time point. To ensure good data quality, each metabolite had to meet three criteria for inclusion in the study: 1) coefficient of variation (CV) not exceeding 25%; 2) less than 5% missing values; 3) median of metabolite concentrations above the limit of detection. This quality control was performed per cohort. Supplementary Table 1 describes the reasons for exclusion of metabolites from the analysis for each cohort. The metabolites measured include hexoses (H1), amino acids (AA), acyl-carnitines (AC), sphingomyelins (SMs), diacyl phosphatidylcholines (PC aa), acyl-alkyl-phosphatidylcholines (PC ae) and lysophosphatidylcholines (lysoPC).
Written informed consent was obtained from all study participants. Metabolite concentration values were natural log-transformed to attain a better approximation of the normal distribution. We performed partial correlation analysis per cohort, adjusting for age and sex, and if necessary for family relationships (model 1). In the extended model (model 2) we additionally adjusted for body mass index (BMI). We performed a sensitivity analysis to test the robustness of the results and repeated the meta-analysis excluding studies with low sample size or high mean LTL values.
A multiple testing-corrected statistical significance threshold for association of metabolite concentrations with LTL was defined at the meta-analysis level. We accounted for multiple testing by Bonferroni correction based on the effective number of independent variables (VeffLi) in the metabolite concentration data 48 (https:// neurogenetics.qimrberghofer.edu.au/matSpD/). The VeffLi value was determined using the correlation matrix of the quality controlled and log-transformed metabolomics data in the ERF and NTR cohorts, yielding a VeffLi (representing the number of independent metabolites) of 46 for both cohorts. This resulted in a Bonferroni corrected significance threshold of p-value < 1.1 × 10 −3 (=0.05/46). pathway analysis of the associated metabolites. For interpretation of the observed associations we followed two bioinformatics approaches in parallel. First, we employed a bioinformatics pipeline based on a workflow management software tool called "Taverna" (http://www.taverna.org.uk) 49 to determine if the metabolites associated with LTL share a network space within two reaction steps. We took as input all possible pairs of significantly associated metabolites. In this pipeline, all the reactions within a radius of two steps in the reaction space of the first metabolite were obtained from the KEGG database 50 . The second metabolite is searched against the substrates and the products of the reactions obtained in the previous step. The presence of the second metabolite in any of the reaction steps is an indication that the two metabolites participate in reactions within two steps of each other. In the final step, the path between the two metabolites is returned to the user. In order to prevent non-specific connections, an intermediate step filters out hub metabolites such as ATP, ADP, and NADP. Next, we used the function "heatmap.2" included in the R package "gplots" (https://CRAN.R-project.org/package=gplots) to prepare a heat map of the correlation of the individual metabolites with LTL in models 1 and 2. For this analysis, default functions for clustering were used (distance measure: euclidean distance). We also derived a correlation matrix for the significantly associated metabolites in ERF and visualized this in a correlogram using the R package "corrplot" 51 .

Results
General characteristics of the study populations are provided in Table 1. The study covers a wide age range, with the mean age of the participants ranging from 18.4 to 62.9 years. Most studies had approximately equal numbers of males and females, except for NTR (33% female) and TwinsUK (only females). Mean LTL ranged from 1.44 (LLS) to 3.58 (TwinsUK). BMI was on average between 25.2 and 27.6 kg/m 2 , but was unavailable at the time of metabolite and LTL assessment in the QIMR study.
We next conducted a sensitivity analysis to determine whether the analyses were driven by a single cohort. We removed two cohorts from the analysis: the TwinsUK cohort, which had a high mean LTL ( = . x 3 58 LTL ) and the QIMR cohort, which had a small sample size (N = 193) and was on average younger than the other cohorts. After excluding the TwinsUK cohort from the meta-analysis, all metabolites remained significantly associated with  Table 1. General characteristics of the study populations. Abbreviations: n = number of participants with data available on metabolites, telomere length, age, and sex; n_BMI = number of participants with data available on metabolites, telomere length, age, sex, and BMI; LTL = leukocyte telomere length; SD = standard deviation. LTL as a ratio of telomere repeat length to copy number of the single copy gene 36B4; Age in years; BMI in kg/m 2 .
To explore to which extent the various metabolites cluster, we constructed a heat map based on the correlation of each individual metabolite with LTL in both model 1 and model 2 (Fig. 1). The heat map shows two clusters of which one (hereafter referred to as "cluster 1") includes lysoPC a C17:0, PC ae C38:4, and a series of PC ae metabolites positively associated with LTL, while the second cluster (hereafter referred to as "cluster 2") includes methionine, tyrosine, PC aa C32:1, and a series of PC aa metabolites negatively associated with LTL. Figure 1 further shows that C3-OH is relatively dissimilar from all other metabolites. A correlogram of the six metabolites associated with LTL after correction for multiple testing is presented in Fig. 2 and shows a positive correlation of methionine with the three PC metabolites. LysoPC a C17:0 and PC ae C38:4 (cluster 1) are most strongly correlated in Fig. 2. Methionine and tyrosine are highly correlated with each other and both amino acids are correlated to PC aa C32.1 (Fig. 2).
Pathway analysis using the Taverna workflow showed that phosphatidylcholines (lysoPC a C17:0, PC aa C32:1, and PC ae C38:4) and methionine are involved in homocysteine metabolism. Homocysteine is the intermediate product in the conversion of the amino acid methionine to cysteine, a precursor of the antioxidant glutathione (Fig. 3). Briefly, PC is a precursor of choline which is oxidized to betaine. Betaine is used to convert homocysteine to methionine. Methionine is first converted to S-adenosylmethionine followed by demethylation to S-adenosylhomocysteine (SAH). Next, hydrolysis of SAH forms homocysteine, which can either be re-methylated into methionine (transmethylation cycle) or metabolized to cysteine (transsulfuration pathway) as shown in Fig. 3.

Discussion
When adjusting for false positive findings due to multiple testing, this study identified significant associations between LTL and six metabolites, which form two extended clusters. These metabolites include three phosphatidylcholines (lysoPC a C17:0, PC aa C32:1, PC ae C38:4), two amino acids (methionine, tyrosine), and one acylcarnitine (C3-OH). Longer LTL was associated with higher levels of lysoPC a C17:0 and PC ae C38:4, and with  www.nature.com/scientificreports www.nature.com/scientificreports/ lower levels of methionine, tyrosine, PC aa C32:1, and C3-OH. Pathway analysis highlighted a key role of the homocysteine metabolism.
A problem when comparing our findings to those in earlier studies is that the metabolites are platformspecific and thus we cannot replicate directly previous findings 43,44 . However, the PCs significantly associated with LTL in our study belong to the same metabolite family of glycerophospholipids 52 as two metabolites significantly associated with LTL in the study of Zierer et al. 47 : 1-stearoylglycerophosphoinositol and 1-palmitoylglycero-phosphoinositol. These metabolites are involved in fatty acid metabolism and particularly membrane composition in biological aging. The small study (N = 423) performed in American Indians from the Strong Family Heart Study also found associations of glycerophospholipids (e.g. glycerophosphoethanolamines, glycero-phosphocholine, and glycerophosphoglycerol) with LTL 46 .
The metabolite most significantly associated with LTL in our study was lysoPC a C17:0. LysoPCs are formed through hydrolysis of PCs by phospholipase A2 53 and have pro-atherogenic and pro-inflammatory effects through impairment of endothelium-dependent vascular relaxation 54 , monocyte recruitment and macrophage proliferation 55,56 , and increased expression of adhesion molecules 57 . Previously, negative associations of lysoPC a C17:0 levels with high-sensitivity C-reactive protein (hsCRP), interleukin-6, insulin, and myocardial infarction have been found 58,59 . These results indicate the involvement of inflammation 58,59 . Inflammation and cardiovascular disease are related to telomere shortening 60,61 and our study brings to surface lysoPC a C17:0 as a novel key player.
We further identified a negative association of the two highly correlated amino acids methionine and tyrosine with LTL. Methionine is an essential amino acid and involved in multiple important biological processes necessary for normal growth and development in mammals, including protein synthesis, methylation, the transsulfuration pathway, and homocysteine metabolism 62 . Previous studies have shown that a methionine-restricted diet increased lifespan in rodents [63][64][65][66][67] . Tyrosine is a non-essential amino acid and a precursor for several catecholamines, including dopamine, as well as thyroid hormones (T 3 and T 4 ) 68,69 . Low thyroid hormone levels have been associated with increased lifespan in multiple animal models [70][71][72] and in humans [73][74][75] . Moreover, a role of tyrosine as developmental regulator and modulator of longevity has been described in Caenorhabditis elegans 76 . Tyrosine can also form a dipeptide with gamma-glutamate called gamma-glutamyltyrosine (http://www.hmdb.ca/metabolites/HMDB0011741), which was negatively associated with LTL in the TwinsUk cohort 47 . It is involved in the gamma-glutamyl cycle (as shown in Fig. 3) and indicates involvement of increased oxidative stress 47 , a factor related to LTL shortening 77,78 . Elevated blood levels of the amino acid tyrosine are seen in obese individuals 79,80 , and were found to be a novel risk factor for the development of diabetes 38,81 . Type 2 diabetes has been associated with shorter LTL 10,16,17 .
Both methionine and tyrosine are correlated to PC aa C32:1, which is the fourth metabolite significantly associated with LTL. PCs are the major phospholipids in cell membranes and lipoproteins 82,83 . They consist of a glycerol backbone with different fatty acid combinations that are linked by ester (a) or ether (e) bonds, resulting in either diacyl (aa) or acyl-alkyl (ae) PCs 84 . We observed a cluster of metabolites (cluster 2) negatively associated with LTL, including methionine, tyrosine, PC aa C32:1 and multiple other diacyl PCs. Various metabolites www.nature.com/scientificreports www.nature.com/scientificreports/ of this cluster including PC aa C32:1, PC aa C36:1, PC aa C38:3, and PC aa C40:5, have been associated with increased risk of type 2 diabetes 40 . The other PC that surpassed the significance threshold in model 1 was PC ae C38:4. However, PC ae C38:4 was nominally significant after including BMI in the model (model 2) and after excluding the younger QIMR study from the meta-analysis. PC ae C38:4 showed a positive association with LTL and clustered with lysoPC a C17:0 and a series of PC ae metabolites (cluster 1) that also show consistent effect across cohorts such as PC ae C40:3, PC ae C40:4, and PC ae C40:5 (FDR < 0.05). In line with this finding, PC ae's have been shown to have antioxidant properties, protecting lipids from oxidation 85,86 , and the metabolites in this cluster showed a reduced risk of type 2 diabetes 40 .
Although also C3-OH was found to be associated with LTL when adjusting for multiple testing, the association with LTL was only observed in ERF and LLS. In the other five studies, this metabolite did not pass the quality control. Therefore, this finding and other findings based on data of two studies only, such as PC aa C38:1, should be interpreted with care and more research, including alternative assessments of these metabolites, is needed. C3-OH is a metabolite of interest for further investigation as it is an acylcarnitine and involved in lipid transport as well as lipid and fatty acid metabolism (http://www.hmdb.ca/metabolites/HMDB0013125). Carnitine is mainly absorbed from the diet but can also be synthesized from the amino acids lysine and methionine 87 and is essential for energy metabolism as it transports fatty acids from the cytosol into the mitochondrion for β-oxidation 88,89 . Figure 2 shows indeed that C3-OH is correlated to methionine, as predicted 87 . Carnitine insufficiency has been implicated as a common trait of insulin-resistant states, including advanced age, genetic diabetes, and diet-induced obesity 90 . However, when clustering the correlations of the metabolites to LTL, we find that C3-OH does not cluster with other metabolites (Fig. 1).
Pathway analysis using the Taverna workflow revealed that both methionine and PCs are part of homocysteine metabolism. Our results give novel metabolic insights into the findings of previous studies that describe an increase in plasma homocysteine with age and shortening of LTL with increasing homocysteine levels 91,92 . Our study suggests that lysoPC a C17:0, PC aa C32:1, PC ae C38:4 as well as methionine and tyrosine are key metabolites in the link between the homocysteine pathway and telomere length. Homocysteine metabolism has been implicated in a wide range of age-related diseases, such as cardiovascular diseases 93,94 , dementia 95,96 , Alzheimer's disease 96,97 , diabetes and its associated vascular complications [98][99][100] , and in mortality [101][102][103][104] . Taking together the findings of our study with that of previous studies, the endothelium may be the tissue of interest. There is substantial evidence that homocysteine and lysoPC are involved in endothelial dysfunction 77,78 caused by inflammation and oxidative stress [105][106][107][108][109] . In cultured endothelial cells, homocysteine was also shown to accelerate telomere shortening and endothelial senescence 92,110 .
A major strength of this study is that both LTL and metabolites were measured centrally, using a standard protocol and blood samples taken at the same time point. Metabolite levels were quantified using the AbsoluteIDQ p150 kit (Biocrates Life Sciences, Innsbruck, Austria) that detects biologically relevant metabolites from four compound classes: acylcarnitines, amino acids, hexoses, and phosho-and sphingolipids. This method has been proven to be in conformance with FDA Guideline 'Guidance for Industry-Bioanalytical Method Validation www.nature.com/scientificreports www.nature.com/scientificreports/ (May 2001)' 111 , which implies proof of reproducibility within a given error range. At the same time, measuring metabolites with this specific platform may be considered also a limitation of our study as other metabolites might also be related to LTL.
In conclusion, using data from 7,853 individuals from seven independent cohorts, we found longer LTL associated with higher levels of lysoPC a C17:0 and PC ae C38:4, and with lower levels of methionine, tyrosine, PC aa C32:1, and C3-OH. These metabolites form two clusters, one including lysoPC a C17:0, PC ae C38:4, and a series of PC ae metabolites positively associated with LTL, while the second cluster includes methionine, tyrosine, PC aa C32:1, and a series of PC aa metabolites. These metabolites have been implicated in cardiovascular disease and diabetes, two major drivers of morbidity and mortality. The functional role of these metabolites involves inflammation and oxidative stress. Our pathway analysis links the metabolites to homocysteine metabolism, a pathway linked to cardiovascular disease, diabetes and many other age-related diseases.

Data Availability
All results generated during this study are included in this published article and its Supplementary Materials. The datasets analysed for each individual cohort can be requested by contacting the responsible Principal Investigator. Because of restrictions based on privacy regulations and informed consent of the participants, data cannot be made freely available in a public repository. For the Rotterdam Study data, requests should be directed towards the management team of the Rotterdam Study (secretariat.epi@erasmusmc.nl), which has a protocol for approving data requests.