Original Article
Identification of Serum Metabolites Associated With Incident
Hypertension in the European Prospective Investigation
into Cancer and Nutrition–Potsdam Study
Stefan Dietrich, Anna Floegel, Cornelia Weikert, Tobias Pischon, Heiner Boeing, Dagmar Drogan
Downloaded from http://hyper.ahajournals.org/ by guest on June 15, 2017
Abstract—Metabolomics is a promising tool to gain new insights into early metabolic alterations preceding the development
of hypertension in humans. We therefore aimed to identify metabolites associated with incident hypertension using
measured data of serum metabolites of the European Prospective Investigation Into Cancer and Nutrition (EPIC)–Potsdam
study. Targeted metabolic profiling was conducted on serum blood samples of a randomly drawn EPIC-Potsdam subcohort
consisting of 135 cases and 981 noncases of incident hypertension, all of them being free of hypertension and not on
antihypertensive therapy at the time of blood sampling. Mean follow-up was 9.9 years. A validated set of 127 metabolites
was statistically analyzed with a random survival forest backward selection algorithm to identify predictive metabolites
of incident hypertension taking into account important epidemiological hypertension risk markers. Six metabolites were
identified to be most predictive for the development of hypertension. Higher concentrations of serine, glycine, and acylalkyl-phosphatidylcholines C42:4 and C44:3 tended to be associated with higher and diacyl-phosphatidylcholines C38:4
and C38:3 with lower predicted 10-year hypertension-free survival, although visualization by partial plots revealed
some nonlinearity in the above associations. The identified metabolites improved prediction of incident hypertension
when used together with known risk markers of hypertension. In conclusion, these findings indicate that metabolic
alterations occur early in the development of hypertension. However, these alterations are confined to a few members
of the amino acid or phosphatidylcholine metabolism, respectively. (Hypertension. 2016;68:00-00. DOI: 10.1161/
HYPERTENSIONAHA.116.07292.) Online Data Supplement
•
Key Words: glycine
■
hypertension
■
incidence
E
ssential hypertension is among the most important preclinical conditions of metabolic syndrome and affects
nearly 1 billion people worldwide.1,2 The risk to develop essential hypertension seems to be a function of age, triggered by
an unhealthy lifestyle with obesity, and physical inactivity as
major risk factors.1 Furthermore, dyslipidemia,3 inflammatory
processes,4 and oxidative stress5 have been closely linked to
this preclinical condition. Although many pathophysiological
mechanisms of hypertension have been elucidated, knowledge
is scarce about individual metabolic alterations promoting the
development of essential hypertension in healthy subjects or
subjects in early stages of this condition.
Application of metabolomics can contribute to fill this
gap and generate further insights into the pathogenesis of
hypertension development. Metabolites represent intermediates and end products of cellular processes and are substantial
for signaling, structuring of membranes, and catalytic activity.
Metabolic alterations associated with development of hypertension, therefore, may be present years before hypertension
■
metabolomics
■
phosphatidylcholines
■
serine
diagnosis. Hence, investigating metabolic profiles in prospective cohorts is a promising opportunity to improve our
knowledge of incident hypertension and to discover novel
biomarkers that elucidate early changes in potential pathways.
In the US cohort, of 204 metabolites, the metabolite
4-hydroxyhippurate and a metabolic sex steroids pattern were
associated with incident hypertension.6 Another US study using
metabolic profiling revealed an association of diacylglycerols, in
general, and of the 2 diacylglycerols 16:0/22:5 and 16:0/22:6, in
particular, with blood pressure (BP) and incident hypertension.7
However, to our knowledge, only a few prospective studies have
used metabolic profiling to investigate metabolic alterations
associated with incident hypertension,6,7 and thus further studies are necessary to elucidate this promising approach.
This study aimed to identify metabolites associated with incident hypertension using data of 127 serum metabolites (Biocrates
AbsoluteIDQ p150) determined within the European Prospective
Investigation Into Cancer and Nutrition (EPIC)–Potsdam study.
The statistical analyses were performed with random survival
Received February 4, 2016; first decision February 15, 2016; revision accepted May 4, 2016.
From the Department of Epidemiology, German Institute of Human Nutrition, Potsdam-Rehbruecke (DIfE), Nuthetal, Germany (S.D., A.F., H.B.,
D.D.); Department of Food Safety, Federal Institute for Risk Assessment, Berlin, Germany (C.W.); Institute for Social Medicine, Epidemiology, and
Health Economics, Charité University Medical Center, Berlin, Germany (C.W.); DZHK (German Center for Cardiovascular Research), Partner Site Berlin,
Germany (C.W., T.P.); Molecular Epidemiology Group, Max Delbrück Center for Molecular Medicine (MDC) Berlin-Buch, Berlin, Germany (T.P.);
Charité–Universiätsmedizin Berlin, Berlin, Germany (T.P.); and AOK Research Institute (WIdO), AOK Bundesverband, Berlin, Germany (D.D.).
The online-only Data Supplement is available with this article at http://hyper.ahajournals.org/lookup/suppl/doi:10.1161/HYPERTENSIONAHA.
116.07292/-/DC1.
Correspondence to Stefan Dietrich, Department of Epidemiology, German Institute of Human Nutrition (DIfE), Arthur-Scheunert-Allee 114–116, DE14558 Nuthetal, Germany. E-mail: [email protected]
© 2016 American Heart Association, Inc.
Hypertension is available at http://hyper.ahajournals.org
DOI: 10.1161/HYPERTENSIONAHA.116.07292
1
2 Hypertension August 2016
forest (RSF),8 a machine learning method specifically adapted
to prospective studies, for which a backward selection algorithm
was applied for an appropriate variable selection.
Methods
Study Sample
Downloaded from http://hyper.ahajournals.org/ by guest on June 15, 2017
The EPIC-Potsdam study included 27 548 women and men, aged
mainly 35 to 65 years who were recruited between 1994 and 1998
from the general population in Potsdam and surrounding areas.
The EPIC-Potsdam Study was approved by the ethical committee of the state of Brandenburg, Germany, following the principles
of the Helsinki Declaration for human rights, and each study participant gave written consent. At baseline participants filled in a
self-administered validated food frequency questionnaire, answered
personal computer–assisted interviews and questionnaires on medical conditions, diet and lifestyle, and underwent examinations. The
examinations were conducted by qualified medical staff following
standardized protocols and included anthropometric and 3 BP measurements and collection of blood samples (30 mL). Blood samples
were immediately fractionated, aliquoted into straws (0.5 mL), and
stored at −196°C until the measurement of serum metabolites.
This study was conducted in a subcohort embedded within the EPICPotsdam study. The subcohort was implemented for case–cohort analyses in 2005 and included 2500 randomly drawn participants.9 Baseline
blood serum samples of the subcohort members were used for targeted
metabolic profiling. After exclusion of participants with prevalent hypertension, diagnosed hypertension within first year of follow-up time, myocardial infarction and stroke at baseline, with missing or nonverified data
on incident hypertension, missing data on covariates and metabolites,
the analyzed study population included initially 1116 participants free of
high BP at baseline of which 135 participants developed hypertension
during follow-up. Among female cases, there were no pregnancies.
Assessment of Prevalent and Incident Hypertension
Participants with (1) systolic BP ≥140 mm Hg or diastolic BP ≥90
mm Hg, or both, as defined by the mean of the second and third measurements, (2) self-reported hypertension diagnosis, or (3) the use of
antihypertensive medication at baseline were classified as cases of
prevalent hypertension. Information about potential cases of incident
hypertension and hypertension-specific medication during the past
month was recorded every 2 to 3 years by self-administered questionnaires with response rates of 95% on average. For all self-reports, the
treating physician was contacted for verification and only confirmed
diagnosed cases by the treating physician were included in the analysis (International Classification of Diseases Tenth Revision: I10).
Assessment of Baseline Covariates
Body mass index was calculated as the ratio of weight (kg) to height
squared (m2). Education at attainment, physical activity, and smoking
were acquired by a standardized interview. For analyses, education at
attainment was categorized as no degree/vocational training, trade/
technical school, and university degree. Smoking behavior as never
smoker, former smoker, and current smoker (≤20 cigarettes/d and >20
cigarettes/d). To account for physical activity, the Improved Physical
Activity Index adjusted for sex and age was calculated as described previously.10 Dietary habits were assessed by the use of a validated food frequency questionnaire11 and used to calculate dietary approaches to stop
hypertension index as described previously.12 Alcohol intake from beverages was categorized into nonconsumer and consumer (women >0–6,
6–12, and >12 g/d and men >0–12, 12–24, and >24 g/d). Cases of prevalent type 2 diabetes mellitus were assessed during a standardized interview at baseline and verified by the treating physician. Diabetes mellitus
status was categorized as existence of type 2 diabetes mellitus or not.
Assessment of Serum Metabolite Concentrations
Baseline blood serum samples were analyzed to determine metabolite concentrations by using AbsoluteIDQ p150 Kits (Biocrates Life
Scienes AG, Innsbruck, Austria), which is based on flow injection
analysis tandem mass spectrometry technique.13 Analysis was done
by the Genome Analysis Center at the Helmholtz Zentrum München.
Please refer the study by Römisch-Margl et al13 for analytic details.
Metabolites with concentrations below the detection limit or high
analytic variance (n=36) were excluded,14 leaving the following 127
quantified metabolites for statistical analyses (Table S1 in the onlineonly Data Supplement): hexose (sum of 6-carbon monosaccharides
without distinction of isomers), 14 amino acids, 14 spingomyelins,
17 acylcarnitines (Cx:y; with x indicating carbon atoms and y indicating double bonds), and 81 glycerophospholipids (37 acyl-alkyl-, 34
diacyl-, and 10 lyso-phosphatidylcholines).
Random Survival Forest
RSF computes a forest of decision trees based on bootstrap samples,
which can be used to select most predictive variables for event time of
interest.8 For computation of the decision trees, random node splitting
is used and a node is split by a variable, which maximize the survival
differences between daughter nodes determined by a log rank statistic. Please refer the study by Ishwaran et al8 for detailed description
of the RSF method.
The predictiveness of a variable for time until event can be determined by a ranking method called minimal depth.15 To compute the
minimal depth of a variable, the distance from the root node to the
node at which a variable splits first in a decision tree is determined
and is then averaged over all bootstrap decision trees. Variables that
split near the root in the decision trees are more predictive regarding
time until the event and result in lower minimal depth values.
The prediction accuracy of an RSF model is determined by the
RSF prediction error rate, which is based on Harrell concordance index (C-index).8 To calculate the RSF error rate, out of back samples,
representing observations not included in the respective bootstrap
samples, are used and dropped down the decision tree computed by
the respective bootstrap samples. According to Harrell C-index the
probability is than estimated, that within a randomly selected pair
of cases, the case with the shorter follow-up time has the worst predictive outcome.8 The RSF error rate is conform to 1-C-index with
values between 0 and 1, where lower prediction error rate values are
corresponding to RSF models with more precise prediction accuracy.8
Statistical Analysis
Baseline characteristics of noncases and cases of incident hypertension are presented as means and SD for continuous variables and as
frequencies for categorical variables. Age, body mass index, sex,
Improved Physical Activity Index, dietary approaches to stop hypertension index, alcohol intake from beverages, smoking behavior,
education at attainment, and prevalent diabetes mellitus were used
as covariates.
For metabolite selection, the following RSF backward selection algorithm, recently suggested for variable selection in the context of Random
Forest16 and subsequently adapted to RSF, was used: (1) compute an
RSF with the data set that contains the covariates and the metabolites
to be tested, (2) remove the metabolite with the worst minimal depth
rank from the data set, (3) use the data set with all the covariates and the
remaining metabolites to compute a new RSF, (4) repeat steps 2 and 3
till only 1 metabolite remains, and (5) choose the set of metabolites with
the smallest predicted error rate. The RSF backward selection algorithm
was applied on data consisting of covariates and all 127 metabolites
to identify the most predictive metabolites.
After the variable selection procedure, data of identified metabolites
and covariates were used to compute a new RSF model. The RSF model was further used to calculated partial plots of identified metabolites.17
Partial plots represent the effect of each metabolite on predicted 10year hypertension-free survival after accounting for the average effects
of the other selected metabolites and the covariates. Finally, to proof
whether the selected metabolites improve RSF prediction error rates
4 different RSF models were computed and compared. The following
data sets were used to compute the respective RSF models: (1) covariates only, (2) covariates and selected metabolites, (3) all metabolites,
and (4) all metabolites and covariates. Hundred repetitions of each RSF
Dietrich et al Metabolites Associated With Incident Hypertension 3
model were computed and used to calculate mean and 95% confidence
intervals of the RSF error rate for each RSF model, respectively. To
determine if the identified metabolites represents a connected metabolic pathway, a previously computed metabolite network18 derived by
using Gaussian graphic modeling was highlighted with the identified
metabolites. Gaussian graphic modelings are undirected probabilistic
graphs, nodes represent variables, and edges the conditional dependence between variables.19 Statistical analysis were performed with the
statistic software R (version 3.0.0), the R-package randomForestSRC
(version 1.2), and SAS version 9.4. The RSF parameter number of trees
and number of node splits were fixed at 10 000 and 10.
Results
Downloaded from http://hyper.ahajournals.org/ by guest on June 15, 2017
Baseline characteristics of the EPIC-Potsdam subcohort sample are presented in Table 1. Of 127 analyzed metabolites,
a set of 6 metabolites resulted in the smallest RSF prediction error rate during the variable selection process and was
most predictive for incident hypertension (Figure 1). Ranked
by the minimal depth measure, identified metabolites were
serine, acyl-alkyl-phosphatidylcholines C42:4, C44:3, diacylphosphatidylcholines C38:4, glycine, and diacyl-phosphatidylcholines C38:3. Serine was the most predictive metabolite for
incident hypertension, whereas diacyl-phosphatidylcholines
C38:3 was less predictive. The identified metabolites seem to
represent no common metabolic pathway (Figure S1).
From the minimal depth ranking (Figure 1), it was also
found that the covariates body mass index, age, and Improved
Physical Activity Index had an even smaller minimal depth
than serine, whereas the other covariates had higher minimal
depth than the 6 selected metabolites. This suggests that the
identified metabolites are more predictive than some traditionally used risk markers of hypertension. No metabolite from the
acylcarnitines, sphingomyelins, and lyso-phosphatidylcholines
class were selected to be predictive for incident hypertension. In
addition, the covariate sex was ranked with a relative high minimal depth value suggesting low information content about the
prediction of incident hypertension because of sex differences.
As visualized in the partial plots (Figure 2), nonlinear
associations between concentrations of identified metabolites and predicted 10-year hypertension-free survival were
observed. Higher concentrations of the metabolites serine,
acyl-alkyl-phosphatidylcholines C42:4, C44:3, and glycine
were associated with a higher 10-year predictive hypertension-free survival, suggesting lower risk to develop hypertension within 10 years of follow-up for individuals with the
respective metabolite concentrations. In contrast, higher concentrations of the metabolites diacyl-phosphatidylcholines
C38:4 and C38:3 were associated with a lower 10-year predictive hypertension-free survival. However, because of the
nonlinear nature of the observed associations, the partial plots
show a peak at given metabolite concentrations which were
associated with the highest 10-year predictive hypertensionfree survival. In particular, for serine, glycine, and diacylphosphatidylcholines C38:4 peaks at ≈100, 250, and 100
µmol/L were, respectively, observed to be associated with
the highest predicted 10-year hypertension-free survival.
An RSF model computed based on data including only
covariates resulted in a prediction error rate of 0.3168
(Table 2). A supplement of the covariate data with the 6 identified metabolites resulted in a more precise computed RSF
model regarding prediction of incident hypertension with an
improved prediction error rate of 0.2789. However, RSF models computed based on data of all 127 metabolites alone or
together with the covariates resulted in RSF models with >9%
worsened prediction error rates (error rate 0.4444 and 0.3747)
compared with an RSF model computed based on data of
selected metabolites and covariates.
Discussion
This study is one of the largest prospective cohorts using targeted
metabolomics to investigate possible associations between 127
serum metabolites and incident hypertension. By application of
an RSF backward selection procedure, 6 metabolites were identified to be most predictive for the development of hypertension
within a follow-up time of ≈10 years. As demonstrated by 4 different RSF models, the 6 identified metabolites may contribute
to an improvement of the prediction of incident hypertension
when used together with known epidemiological risk factors of
hypertension. The comparison of the prediction error rate of 4
different RSF models revealed that, especially noise metabolites were removed by the RSF backward selection process
resulting in the identification of the most predictive metabolites.
Moreover, the visualization by partial plots revealed nonlinear
associations between concentrations of identified metabolites
and predicted 10-year survival, indicating possible diagnostic
cut points for further research.
Two of the identified metabolites were the biochemically
closely related amino acids serine and glycine. In general, serine and glycine are known to act as neurotransmitters, to be
important for the catalytic function of many enzymes and to
represent essential elements of many lipids (eg, sphingolipids, ceramides, and glycerophospholipids). As far as known,
no study has until now demonstrated an association of serum
blood levels of serine and glycine with incident hypertension.
Nevertheless, the observed results are supported by recent
research indicating that supplementation with serine as well
as glycine may have BP-lowering effects.20–23 Moreover, antiinflammatory and antioxidant properties were attributed to
serine and glycine,20,23–26 which may contribute to protective
effects regarding the development of hypertension.
In this study, a nonlinear positive association of serine
concentrations with 10-year hypertension-free survival was
observed, whereby serine concentrations of 100 µmol/L were
most protective regarding 10-year hypertension-free survival. This is confirmed by a study of Mishra et al20 describing serine-induced vasodilation in endothelium-intact vessels
as a result of increasing serine concentrations. As shown by
this previous study, the endothelium-dependent vasodilation
was probably caused by serine promoted K+ efflux from the
endothelium, independent of endothelial nitric oxide release.20
Moreover, elevation of antioxidant agents by administering
serine to human endothelial cell culture were reported reflecting cytoprotective and antioxidant effects of serine.24 It was
also demonstrated that dietary serine intake lowers plasma
homocysteine concentrations. Homocysteine is a known risk
factor for cardiovascular diseases and arterial endothelial dysfunction confirming the present results.27,28
The second identified amino acid glycine is known to
reduce oxidative stress because of an enhancement of the
4 Hypertension August 2016
Table 1. Baseline Characteristics of the Subcohort
Embedded Within EPIC-Potsdam*
Characteristics
Noncases
(n=981)
Cases of Incident
Hypertension (n=135)
Age (y)
46.9 (8.4)
51.7 (8.7)
Women, %
71.2
71.7
BMI (kg/m2)
24.3 (3.5)
26.6 (4.2)
Systolic BP (mm Hg)
117.4 (10.0)
125.1 (8.8)
Diastolic BP (mm Hg)
76.8 (6.7)
80.7 (5.8)
No degree/vocational
training, %
33.8
39.1
Trade/technical school, %
25.5
30.4
University degree, %
40.7
30.4
Never, %
47.5
56.5
Former, %
30.2
25.4
Current, %
22.2
18.1
Among smokers: number of
cigarettes/d
12.4 (8.3)
14.4 (10.6)
IPAI
37.1 (4.4)
34.9 (4.4)
Alcohol intake from beverages
(g/d)
12.8 (14.7)
11.2 (14.4)
DASH index
16.0 (4.9)
15.7 (4.8)
Education
Smoking status
Downloaded from http://hyper.ahajournals.org/ by guest on June 15, 2017
Prevalent T2DM, %
1.6
1.5
Follow-up time (y)
10.4 (1.9)
6.5 (2.6)
BMI indicates body mass index; BP, blood pressure; DASH, dietary approaches
to stop hypertension; EPIC, European Prospective Investigation into Cancer and
Nutrition; IPAI, improved physical activity index; and T2DM, type 2 diabetes mellitus.
*Presented are mean (SD) for continuous variables or percentages for
categorical variables
bioavailability of nitric oxide.22,23 As seen in the partial plot
of glycine, an increase of glycine concentrations of ≈150 to
250 µmol/L was closely associated with an improvement
of 10-year hypertension-free survival time. A recent study
showed that glycine concentrations of around 200 µmol/L are
required to activate glycine-gated chloride channels which,
inter alia, occur in macrophages, monocytes, and endothelial
cells.29–32 Furthermore, it was shown that glycine may exhibit
anti-inflammatory effects, in particular, with regard to endothelial cells.25 In addition, reduced glycine concentration levels in urine of women with preeclampsia were previously also
detected.33 However, preeclampsia is linked with a period of
pronounced hormonal and metabolic changes and findings
may thus not be comparable with our sample of the general
population.
Remarkably, no other analyzed amino acid than serine
and glycine was associated with the development of hypertension, to date. In accordance with the present results, the
Rotterdam study also showed no association between incident
hypertension and several amino acids, such as arginine, lysine,
glutamic acid, cysteine, and tyrosine.34 In addition, serine and
glycine were not considered by the Rotterdam study.34
Figure 1. Selected metabolites that are most predictive for
incident hypertension and important risk marker of hypertension
ranked by the minimal depth. Metabolites were selected using
a random survival forest backward algorithm. Metabolites
with lower minimal depth values are more predictive regarding
incident hypertension. a indicates acyl; aa, diacyl; ae, acylalkyl; AC, acylcarnitine; DASH, Dietary Approaches to Stop
Hypertension; IPAI, improved physical activity index; M, men; PC,
phosphatidylcholine; T2D, type 2 diabetes; and W, women.
Although a large number of lipids of different classes were
measured in this study, only 4 phosphatidylcholines metabolites (acyl-alkyl-phosphatidylcholines C42:4, C44:3, diacylphosphatidylcholines C38:3, and diacyl-phosphatidylcholines
C38:4) were identified to be predictive to a certain extend for
incident hypertension during the 10 years of follow-up time.
The partial plots suggest a protective role of the 2 acyl-alkylphosphatidylcholines C42:4 and C44:3 with increasing concentrations while only a certain concentration range of the
diacyl-phosphatidylcholines C38:4 was protective regarding
incident hypertension. In this context, it is interesting to note
that recent studies have assigned a possibly anti-inflammatory
role under different conditions (eg, oxidative stress and ulcerative colitis) to phosphatidylcholines.35–38 It was shown that
phosphatidylcholines are able to inhibit the upregulation of
the inflammatory cytokines tumor necrosis factor-α and
interleukin-6 and the actin-assembly in phagosomes and
macrophages.35,37 Moreover, phosphatidylcholines seem to
be required for lipoprotein assembly and hepatic secretion
of triglyceride-rich very low–density lipoprotein particles
as well as high-density lipoprotein particles.39 Through a
vinyl-ether bond in the acyl-alkyl-phosphatidylcholines,
acyl-alkyl-phosphatidylcholines may be capable to act as
blood antioxidants to protect lipoproteins from oxidation.40
Moreover, in a previous study, it was shown that acylalkyl-phosphatidylcholines were positively correlated with
high-density lipoprotein-cholesterol.9 These previous findings
may support the observed protective association between acylalkyl-phosphatidylcholines and predicted 10-year hypertension-free survival of this study and may point to a role as blood
antioxidants of the identified acyl-alkyl-phosphatidylcholines.
Of note, the metabolites acyl-alkyl-phosphatidylcholines
C42:4, diacyl-phosphatidylcholines C38:3 as well as glycine
Dietrich et al Metabolites Associated With Incident Hypertension 5
Downloaded from http://hyper.ahajournals.org/ by guest on June 15, 2017
Figure 2. Partial plots of the selected metabolites including partial values (gray points) ±2 SE (dashed gray lines). Values on the vertical
axis represent predicted 10-year hypertension-free survival for a given metabolite after adjusting for all other variables (covariates and
selected metabolites). A lower predicted 10-year hypertension-free survival means a higher risk to develop hypertension within 10 years
of follow-up time in EPIC-Potsdam study. a indicates acyl; aa, diacyl; ae, acyl-alkyl; and PC, phosphatidylcholine.
were previously also found to be associated in the same direction with incident type 2 diabetes mellitus,9 suggesting a potential role of identified metabolites in the metabolic syndrome.
However, the results obtained suggest that the early stage
of hypertension is accompanied by only a few alteration of
investigated serum lipid metabolism. Although hypertension is
closely linked to dyslipidemia, to date, only a limited number of
prospective cohorts exists that investigated blood parameters in
lipid metabolism in relation to incident hypertension.6,7 A recent
study in the US-Hispanic population7 used serum lipidomic profiling including phosphatidylcholines to elucidate associations
with incident hypertension. In the study by Kulkarni et al,7 of
measured phosphatidylcholines, only the diacyl-phosphatidylcholines C34:4 was significantly associated with an increased
diastolic BP but not with incident hypertension. In addition, of
Table 2. Calculated RSF Error Rates of Different RSF Models
Using 100 Repetitions*
RSF model
RSF Error rate, Mean (95% CI)
6 selected metabolites+covariates*
0.2789 (0.2788–0.2790)
Only covariates*
0.3168 (0.3168–0.3169)
127 metabolites+covariates*
0.3747 (0.3746–0.3748)
Only 127 metabolites
0.4444 (0.4443–0.4447)
BMI indicates body mass index; CI, confidence interval; DASH, Dietary
Approaches to Stop Hypertension; IPAI, improved physical activity index; and
RSF, random survival forest.
*The RSF error rate is conform to 1 C-index, lower values corresponding to
RSF models with more precise prediction accuracy. The covariates included in the
RSF models were age, BMI, sex, IPAI, DASH index, alcohol intake from beverages,
smoking behavior, education at attainment, and prevalent type 2 diabetes mellitus.
all the measured metabolites, 1 phosphatidylethanolamin metabolite (C40:6) and 2 diacylglycerol metabolites (DG 16:0/22:5
and DG 16:0/22:6) were associated with incident hypertension. However, phosphatidylcholines with >40 C-atoms were
not measured in the study by Kulkarni et al,7 and diacylglycerols and phosphatidylethanolamins were not measured in this
study, which allows only a limited comparison with this study.
In a further study by Zheng et al,6 the association between 204
metabolites and incident hypertension was investigated resulting in the identification of the metabolite 4-hidroxyhippurate
and a sex steroids pattern. Again, only a limited comparability is
given, because, except for some amino acids, different metabolites were measured. Serine was not measured in the study by
Zheng et al,6 but glycine. However, glycine was not associated
with incident hypertension risk in this study.
In addition, metabolomics was also applied to retrospectively
identify metabolites associated with prevalent hypertension and
BP.41,42 Graessler et al41 reported altered blood plasma levels of acylalkyl-phosphatidylcholines in hypertensive German men, relative
to the control group of normotensive men. Notably, identified acylalkyl-phosphatidylcholines (C36:4, C36:5, C38:4–C38:6) were
highly unsaturated, as the 3 identified phosphatidylcholines metabolites in this study which may be beneficial to enhance membrane
fluidity. However, none of the metabolites, reported by Graessler et
al,41 were associated with incident hypertension in this study, suggesting differences in metabolite composition of individuals with
incident and prevalent hypertension. Recently, of 280 metabolites,
15 metabolites were identified by Menni et al42 to be independently
associated with BP. However, of identified metabolites, only hexadecanedioate, a dicarboxylic acid, showed concordant association
with BP in 2 replication cohort (KORA and Hertfordshire). In
6 Hypertension August 2016
Downloaded from http://hyper.ahajournals.org/ by guest on June 15, 2017
summary, the small number of studies also highlights the need for
further research to gain a deeper insight into possible association
between metabolites and hypertension development.
The strength of this study is the application of targeted metabolomics in a well-described population-based prospective cohort
with strictly standardized study protocols and a long follow-up
time. Furthermore, the investigated metabolites have been previously validated and those metabolites below the detection limit
and with high analytic variance were excluded.14 With exception
of acyl-alkyl-phosphatidylcholines C42:4, the identified metabolites showed good reliability during a 4-month period14 (Table S2).
The exploratory data analysis of complex metabolomic data using
traditional statistical regression approaches is accompanied by
false-positive detection because of high number of correlated variables, which brings problems such as multiple hypotheses testing,
decreased statistical power, and increased risk of multicollinearity. The machine learning method RSF was specifically developed
for statistical analysis of complex, right-censored survival data.8
RSF is completely data driven, reduces overfitting by bootstrapping and the interruption of intercorrelation structures by random
node splitting allows reliable variable selection in the presence of
multicollinearity.8 Indeed, RSF has been successfully applied to
identify risk factors of cancer and cardiovascular diseases43,44 and
thus RSF seems appropriate for metabolite selection. Moreover,
nonlinear associations between identified metabolites and predicted survival time were visualized allowing definition of possible clinical thresholds and cut points in further scientific studies.
Indeed, this study also has some limitations. Although we
included established hypertension risk factors, our findings
were obtained from an observational study. As such, we cannot
exclude the possibility that our findings have been influenced by
additional factors that were not included. Moreover, given the
observational nature of our study, causality of present findings
cannot be proven. Nevertheless, the observed nonlinear associations in the partial plots are useful to derive possible cut points
for future studies. A major limitation of this study is that our findings cannot be validated in an external cohort because there is no
similar prospective study of sufficient size with available measurements of Biocrates metabolites that have validated end point
of incident hypertension. Although previous studies support the
biological plausibility of our findings, validation in external prospective cohorts and proof of causality are necessary to confirm
the observed results and to translate them into clinical practice.
In conclusion, the analysis of targeted metabolic data
in the EPIC-Potsdam study provided new insights about
metabolic alterations that occur early in the development of
hypertension. However, these alterations are confined to a
few chemically related members of amino acid or phosphatidylcholine metabolism, respectively. Similar studies of this
type are essential to elucidate metabolic predictors of hypertension, as interindividual and population-based metabolic
differences and their technical measurements may lead to a
modified selection of metabolites in other studies.
Perspectives
This study successfully used targeted metabolic profiling to
identify 6 metabolites associated with incident hypertension.
The gained insights enhance our knowledge about metabolic
alterations that influence the development of hypertension. On
the basis of this study, further research can be carried out with
the objective to confirm the archived results and to improve
individual clinical treatment and prevention strategies.
Acknowledgments
We thank the Human Study Centre of the German Institute of Human
Nutrition Potsdam-Rehbruecke for data collection, the data hub for
the processing, the participants for the provision of the data, and
Manuela Bergmann for the contribution to the study design and leading the underlying processes of data generation.
Sources of Funding
This work was supported by a grant from the Federal Ministry of
Education and Research, Germany to the German Center for Diabetes
Research (grant number 01GI0922).
Disclosures
None.
References
1. World Health Organization. Global Status Report on Noncommunicable
Diseases 2010. Geneva, Switzerland: World Health Organization; 2010.
2. Grundy SM, Cleeman JI, Daniels SR, Donato KA, Eckel RH, Franklin BA,
Gordon DJ, Krauss RM, Savage PJ, Smith SC Jr, Spertus JA, Costa F; American
Heart Association; National Heart, Lung, and Blood Institute. Diagnosis and
management of the metabolic syndrome: an American Heart Association/
National Heart, Lung, and Blood Institute Scientific Statement. Circulation.
2005;112:2735–2752. doi: 10.1161/CIRCULATIONAHA.105.169404.
3. Halperin RO, Sesso HD, Ma J, Buring JE, Stampfer MJ, Gaziano JM.
Dyslipidemia and the risk of incident hypertension in men. Hypertension.
2006;47:45–50. doi: 10.1161/01.HYP.0000196306.42418.0e.
4. Harrison DG, Guzik TJ, Lob HE, Madhur MS, Marvar PJ, Thabet SR, Vinh
A, Weyand CM. Inflammation, immunity, and hypertension. Hypertension.
2011;57:132–140. doi: 10.1161/HYPERTENSIONAHA.110.163576.
5.González J, Valls N, Brito R, Rodrigo R. Essential hypertension and
oxidative stress: New insights. World J Cardiol. 2014;6:353–366. doi:
10.4330/wjc.v6.i6.353.
6.Zheng Y, Yu B, Alexander D, Mosley TH, Heiss G, Nettleton JA,
Boerwinkle E. Metabolomics and incident hypertension among blacks: the
atherosclerosis risk in communities study. Hypertension. 2013;62:398–
403. doi: 10.1161/HYPERTENSIONAHA.113.01166.
7. Kulkarni H, Meikle PJ, Mamtani M, Weir JM, Barlow CK, Jowett JB,
Bellis C, Dyer TD, Johnson MP, Rainwater DL, Almasy L, Mahaney
MC, Comuzzie AG, Blangero J, Curran JE. Plasma lipidomic profile signature of hypertension in Mexican American families: specific
role of diacylglycerols. Hypertension. 2013;62:621–626. doi: 10.1161/
HYPERTENSIONAHA.113.01396.
8. Ishwaran H, Kogalur UB, Blackstone EH, Lauer MS. Random survival
forests. Ann Appl Stat. 2008; 2: 841–860.
9. Floegel A, Stefan N, Yu Z, et al. Identification of serum metabolites associated with risk of type 2 diabetes using a targeted metabolomic approach.
Diabetes. 2013;62:639–648. doi: 10.2337/db12-0495.
10. Wientzek A, Vigl M, Steindorf K, Brühmann B, Bergmann MM, Harttig
U, Katzke V, Kaaks R, Boeing H. The improved physical activity index for
measuring physical activity in EPIC Germany. PLoS One. 2014;9:e92005.
doi: 10.1371/journal.pone.0092005.
11. Bohlscheid-Thomas S, Hoting I, Boeing H, Wahrendorf J. Reproducibility
and relative validity of energy and macronutrient intake of a food frequency questionnaire developed for the German part of the EPIC project. European Prospective Investigation into Cancer and Nutrition. Int J
Epidemiol. 1997;26 (suppl 1):S71–S81.
12. Fung TT, Chiuve SE, McCullough ML, Rexrode KM, Logroscino G, Hu
FB. Adherence to a DASH-style diet and risk of coronary heart disease
and stroke in women. Arch Intern Med. 2008;168:713–720. doi: 10.1001/
archinte.168.7.713.
13. Römisch-Margl W, Prehn C, Bogumil R, Röhring C, Suhre K, Adamski
J. Procedure for tissue sample preparation and metabolite extraction for
high-throughput targeted metabolomics. Metabolomics. 2012; 8: 133–
142.doi: 10.1007/s11306-011-0293-4) co.
14. Floegel A, Drogan D, Wang-Sattler R, Prehn C, Illig T, Adamski J, Joost
HG, Boeing H, Pischon T. Reliability of serum metabolite concentrations
Dietrich et al Metabolites Associated With Incident Hypertension 7
Downloaded from http://hyper.ahajournals.org/ by guest on June 15, 2017
over a 4-month period using a targeted metabolomic approach. PLoS One.
2011;6:e21103. doi: 10.1371/journal.pone.0021103.
15. Ishwaran H, Kogalur UB, Chen X, Minn AJ. Random survival forests for
high-dimensional data. Stat Anal Data Mining. 2011; 4: 115–132. doi:
10.1002/sam.10103.
16. Díaz-Uriarte R, Alvarez de Andrés S. Gene selection and classification of
microarray data using random forest. BMC Bioinformatics. 2006;7:3. doi:
10.1186/1471-2105-7-3.
17. Ishwaran H, Kogalur UB. Random survival forest for R. R News. 2007;
7(2): 25–31.
18. Floegel A, Wientzek A, Bachlechner U, Jacobs S, Drogan D, Prehn C,
Adamski J, Krumsiek J, Schulze MB, Pischon T, Boeing H. Linking diet,
physical activity, cardiorespiratory fitness and obesity to serum metabolite networks: findings from a population-based study. Int J Obes (Lond).
2014;38:1388–1396. doi: 10.1038/ijo.2014.39.
19. Krämer N, Schäfer J, Boulesteix AL. Regularized estimation of largescale gene association networks using graphical Gaussian models. BMC
Bioinformatics. 2009;10:384. doi: 10.1186/1471-2105-10-384.
20.Mishra RC, Tripathy S, Desai KM, Quest D, Lu Y, Akhtar J,
Gopalakrishnan V. Nitric oxide synthase inhibition promotes endothelium-dependent vasodilatation and the antihypertensive effect of L-serine.
Hypertension. 2008;51:791–796. doi: 10.1161/HYPERTENSIONAHA.
107.099598.
21.Mishra RC, Tripathy S, Quest D, Desai KM, Akhtar J, Dattani ID,
Gopalakrishnan V. L-Serine lowers while glycine increases blood pressure in chronic L-NAME-treated and spontaneously hypertensive rats.
J Hypertens. 2008;26:2339–2348.
22. El Hafidi M, Pérez I, Zamora J, Soto V, Carvajal-Sandoval G, Baños G.
Glycine intake decreases plasma free fatty acids, adipose cell size, and
blood pressure in sucrose-fed rats. Am J Physiol Regul Integr Comp
Physiol. 2004;287:R1387–R1393. doi: 10.1152/ajpregu.00159.2004.
23.Díaz-Flores M, Cruz M, Duran-Reyes G, Munguia-Miranda C, LozaRodríguez H, Pulido-Casas E, Torres-Ramírez N, Gaja-Rodriguez O,
Kumate J, Baiza-Gutman LA, Hernández-Saavedra D. Oral supplementation with glycine reduces oxidative stress in patients with metabolic syndrome, improving their systolic blood pressure. Can J Physiol Pharmacol.
2013;91:855–860. doi: 10.1139/cjpp-2012-0341.
24. Maralani MN, Movahedian A, Javanmard ShH. Antioxidant and cytoprotective effects of L-Serine on human endothelial cells. Res Pharm Sci.
2012;7:209–215.
25.Hasegawa S, Ichiyama T, Sonaka I, Ohsaki A, Okada S, Wakiguchi
H, Kudo K, Kittaka S, Hara M, Furukawa S. Cysteine, histidine and
glycine exhibit anti-inflammatory effects in human coronary arterial endothelial cells. Clin Exp Immunol. 2012;167:269–274. doi:
10.1111/j.1365-2249.2011.04519.x.
26. Leipnitz G, da Silva Lde B, Fernandes CG, Seminotti B, Amaral AU,
Dutra-Filho CS, Wajner M. d-Serine administration provokes lipid oxidation and decreases the antioxidant defenses in rat striatum. Int J Dev
Neurosci. 2010;28:297–301. doi: 10.1016/j.ijdevneu.2010.03.002.
27. Verhoef P, Steenge GR, Boelsma E, van Vliet T, Olthof MR, Katan MB.
Dietary serine and cystine attenuate the homocysteine-raising effect of
dietary methionine: a randomized crossover trial in humans. Am J Clin
Nutr. 2004;80:674–679.
28. Cao C, Hu J, Dong Y, Zhan R, Li P, Su H, Peng Q, Wu T, Lei L, Huang X,
Wu Q, Cheng X. Gender differences in the risk factors for endothelial dysfunction in Chinese hypertensive patients: homocysteine is an independent risk factor in females. PLoS One. 2015;10:e0118686. doi: 10.1371/
journal.pone.0118686.
29.Yamashina S, Konno A, Wheeler MD, Rusyn I, Rusyn EV, Cox AD,
Thurman RG. Endothelial cells contain a glycine-gated chloride channel.
Nutr Cancer. 2001;40:197–204. doi: 10.1207/S15327914NC402_17.
30. Froh M, Thurman RG, Wheeler MD. Molecular evidence for a glycine-gated
chloride channel in macrophages and leukocytes. Am J Physiol Gastrointest
Liver Physiol. 2002;283:G856–G863. doi: 10.1152/ajpgi.00503.2001.
31. McCarty MF, DiNicolantonio JJ. The cardiometabolic benefits of glycine:
Is glycine an ‘antidote’ to dietary fructose? Open Heart. 2014;1:e000103.
doi: 10.1136/openhrt-2014-000103.
32. Zhong Z, Wheeler MD, Li X, Froh M, Schemmer P, Yin M, Bunzendaul
H, Bradford B, Lemasters JJ. L-Glycine: a novel antiinflammatory, immunomodulatory, and cytoprotective agent. Curr Opin Clin Nutr Metab
Care. 2003;6:229–240. doi: 10.1097/01.mco.0000058609.19236.a4.
33.Austdal M, Skråstad RB, Gundersen AS, Austgulen R, Iversen AC,
Bathen TF. Metabolomic biomarkers in serum and urine in women
with preeclampsia. PLoS One. 2014;9:e91923. doi: 10.1371/journal.
pone.0091923.
34. Altorf-van der Kuil W, Engberink MF, De Neve M, van Rooij FJ, Hofman A,
van’t Veer P, Witteman JC, Franco OH, Geleijnse JM. Dietary amino acids
and the risk of hypertension in a Dutch older population: the Rotterdam
Study. Am J Clin Nutr. 2013;97:403–410. doi: 10.3945/ajcn.112.038737.
35. Jung YY, Nam Y, Park YS, Lee HS, Hong SA, Kim BK, Park ES, Chung
YH, Jeong JH. Protective effect of phosphatidylcholine on lipopolysaccharide-induced acute inflammation in multiple organ injury. Korean J
Physiol Pharmacol. 2013;17:209–216. doi: 10.4196/kjpp.2013.17.3.209.
36.Al-Orf SM. Effect of oxidized phosphatidylcholine on biomarkers of
oxidative stress in rats. Indian J Clin Biochem. 2011;26:154–160. doi:
10.1007/s12291-010-0064-4.
37.Treede I, Braun A, Sparla R, Kühnel M, Giese T, Turner JR, Anes
E, Kulaksiz H, Füllekrug J, Stremmel W, Griffiths G, Ehehalt R.
Anti-inflammatory effects of phosphatidylcholine. J Biol Chem.
2007;282:27155–27164. doi: 10.1074/jbc.M704408200.
38. Eros G, Varga G, Váradi R, Czóbel M, Kaszaki J, Ghyczy M, Boros M.
Anti-inflammatory action of a phosphatidylcholine, phosphatidylethanolamine and N-acylphosphatidylethanolamine-enriched diet in carrageenaninduced pleurisy. Eur Surg Res. 2009;42:40–48. doi: 10.1159/000167856.
39.Cole LK, Vance JE, Vance DE. Phosphatidylcholine biosynthesis and
lipoprotein metabolism. Biochim Biophys Acta. 2012;1821:754–761. doi:
10.1016/j.bbalip.2011.09.009.
40.Wallner S, Schmitz G. Plasmalogens the neglected regulatory and
scavenging lipid species. Chem Phys Lipids. 2011;164:573–589. doi:
10.1016/j.chemphyslip.2011.06.008.
41.Graessler J, Schwudke D, Schwarz PE, Herzog R, Shevchenko A,
Bornstein SR. Top-down lipidomics reveals ether lipid deficiency in blood
plasma of hypertensive patients. PLoS One. 2009;4:e6261. doi: 10.1371/
journal.pone.0006261.
42.Menni C, Graham D, Kastenmüller G, et al. Metabolomic identification of a novel pathway of blood pressure regulation involving
hexadecanedioate. Hypertension. 2015;66:422–429. doi: 10.1161/
HYPERTENSIONAHA.115.05544.
43.Hsich E, Gorodeski EZ, Blackstone EH, Ishwaran H, Lauer MS.
Identifying important risk factors for survival in patient with systolic heart
failure using random survival forests. Circ Cardiovasc Qual Outcomes.
2011;4:39–45. doi: 10.1161/CIRCOUTCOMES.110.939371.
44. Omurlu IK, Ture M, Tokatli F. The comparisons of random survival forests
and Cox regression analysis with simulation and an application related to
breast cancer. Expert Systems With Applications. 2009; 36: 8582–8588.
doi:10.1016/j.eswa.2008.10.023.
Novelty and Significance
What Is New?
• This is one of the first studies using targeted metabolomics in a prospective
cohort (European Prospective Investigation into Cancer and Nutrition [EPIC]Potsdam) to identify metabolites associated with incident hypertension.
What Is Relevant?
• Higher concentrations of serine, glycine, and the acyl-alkyl-phosphati-
dylcholines C42:4 and C44:3 tended to be associated with higher and
diacyl-phosphatidylcholine C38:4 with lower predicted 10-year hypertension-free survival.
• Nonlinear associations between concentrations of identified metabolites
and predicted 10-year hypertension-free survival time were observed.
Summary
This study indicates that metabolic alterations occur early in the
development of hypertension. However, these alterations are confined to a few members of the amino acid or phosphatidylcholine
metabolism, respectively.
Identification of Serum Metabolites Associated With Incident Hypertension in the
European Prospective Investigation into Cancer and Nutrition−Potsdam Study
Stefan Dietrich, Anna Floegel, Cornelia Weikert, Tobias Pischon, Heiner Boeing and Dagmar
Drogan
Downloaded from http://hyper.ahajournals.org/ by guest on June 15, 2017
Hypertension. published online May 31, 2016;
Hypertension is published by the American Heart Association, 7272 Greenville Avenue, Dallas, TX 75231
Copyright © 2016 American Heart Association, Inc. All rights reserved.
Print ISSN: 0194-911X. Online ISSN: 1524-4563
The online version of this article, along with updated information and services, is located on the
World Wide Web at:
http://hyper.ahajournals.org/content/early/2016/05/31/HYPERTENSIONAHA.116.07292
Data Supplement (unedited) at:
http://hyper.ahajournals.org/content/suppl/2016/05/31/HYPERTENSIONAHA.116.07292.DC1
Permissions: Requests for permissions to reproduce figures, tables, or portions of articles originally published
in Hypertension can be obtained via RightsLink, a service of the Copyright Clearance Center, not the Editorial
Office. Once the online version of the published article for which permission is being requested is located,
click Request Permissions in the middle column of the Web page under Services. Further information about
this process is available in the Permissions and Rights Question and Answer document.
Reprints: Information about reprints can be found online at:
http://www.lww.com/reprints
Subscriptions: Information about subscribing to Hypertension is online at:
http://hyper.ahajournals.org//subscriptions/
Online Supplement
IDENTIFICATION OF SERUM METABOLITES ASSOCIATED WITH INCIDENT
HYPERTENSION IN THE EPIC-POTSDAM STUDY
Running title: Metabolites associated with incident hypertension
Authors: Stefan Dietrich1, Anna Floegel1, Cornelia Weikert2-4, Tobias Pischon4-6, Heiner
Boeing1, Dagmar Drogan1,7
Affiliations:
1
Department of Epidemiology, German Institute of Human Nutrition, Potsdam-Rehbruecke
(DIfE), Arthur-Scheunert-Allee 114-116, 14558 Nuthetal, Germany
2
Department of Food Safety, Federal Institute for Risk Assessment, Berlin, Germany
3
Institute for Social Medicine, Epidemiology and Health Economics, Charité University
Medical Center, Berlin, Germany
4
DZHK (German Center for Cardiovascular Research), partner site Berlin
5
Molecular Epidemiology Group, Max Delbrück Center for Molecular Medicine (MDC)
Berlin-Buch, Robert-Rössle-Str. 10, 13125 Berlin, Germany
6
Charité - Universiätsmedizin Berlin, Germany
7
AOK Research Institute (WIdO), AOK Bundesverband, Berlin, Germany
Corresponding Author: Stefan Dietrich, Department of Epidemiology, German Institute of
Human Nutrition (DIfE), Arthur-Scheunert-Allee 114-116, DE-14558 Nuthetal, Germany,
phone: +49(0)33200 88-2717 fax: +49 (0)33200 88-2721, e-mail: [email protected]
References:
1. Floegel A, Drogan D, Wang-Sattler R, et al. Reliability of Serum Metabolite
Concentrations over a 4-Month Period Using a Targeted Metabolomic Approach. Plos One
2011;6: e21103.
Table S1: Summary results of metabolites in the control group and incident hypertension
group
Metabolite
C0
C10
C10:2
C14:1
C14:2
C16
C16:2
C18
C18:1
C18:2
C2
C3
C3-DC-M / C5OH
C5-DC / C6OH
C7-DC
C8:1
C9
Arginine
Glutamine
Glycine
Histidine
Methionine
Ornithine
Phenylalanine
Proline
Serine
Threonine
Tryptophan
Tyrosine
Valine
xLeu
PC aa C28:1
PC aa C30:0
PC aa C32:0
PC aa C32:1
PC aa C32:2
PC aa C32:3
PC aa C34:1
PC aa C34:2
PC aa C34:3
PC aa C34:4
PC aa C36:0
PC aa C36:1
PC aa C36:2
PC aa C36:3
Biochemical name
DL-Carnitine
Decanoyl-L-carnitine
Decadienyl-L-carnitine
Tetradecenoyl-L-carnitine
Tetradecadienyl-L-carnitine
Hexadecanoyl-L-carnitine
Hexadecadienyl-L-carnitine
Octadecanoyl-L-carnitine
Octadecenoyl-L-carnitine
Octadecadienyl-L-carnitine
Acetyl-L-carnitine
Propionyl-L-carnitine
Methylmalonyl-L-carnitine /
Hydroxyvaleryl-L-carnitine
Glutaryl-L-carnitine /
Hydroxyhexanoyl-L-carnitine
Pimelyl-L-carnitine
Octenoyl-L-carnitine
Nonayl-L-carnitine
Arginine
Glutamine
Glycine
Histidine
Methionine
Ornithine
Phenylalanine
Proline
Serine
Threonine
Tryptophan
Tyrosine
Valine
Leucine/Isoleucine
PC diacyl C28:1
PC diacyl C30:0
PC diacyl C32:0
PC diacyl C32:1
PC diacyl C32:2
PC diacyl C32:3
PC diacyl C34:1
PC diacyl C34:2
PC diacyl C34:3
PC diacyl C34:4
PC diacyl C36:0
PC diacyl C36:1
PC diacyl C36:2
PC diacyl C36:3
Non-cases of incident
hypertension
Cases of incident
hypertension
Mean (95% CI)
µM
Mean (95%CI)
µM
34.80 (34.24 - 35.35)
0.25 (0.24 - 0.26)
0.05 (0.05 - 0.05)
0.19 (0.19 - 0.19)
0.03 (0.03 - 0.03)
0.11 (0.11 - 0.11)
0.01 (0.01 - 0.01)
0.05 (0.05 - 0.05)
0.14 (0.14 - 0.14)
0.07 (0.06 - 0.07)
6.70 (6.52 - 6.88)
0.35 (0.34 - 0.36)
0.03 (0.03 - 0.03)
36.21 (34.7 - 37.73)
0.25 (0.23 - 0.27)
0.05 (0.05 - 0.05)
0.20 (0.19 - 0.21)
0.03 (0.03 - 0.03)
0.12 (0.11 - 0.12)
0.01 (0.01 - 0.01)
0.05 (0.05 - 0.05)
0.15 (0.14 - 0.15)
0.07 (0.06 - 0.07)
7.02 (6.54 - 7.49)
0.37 (0.35 - 0.40)
0.03 (0.03 - 0.03)
0.02 (0.02 - 0.02)
0.02 (0.02 - 0.03)
0.03 (0.03 - 0.04)
0.10 (0.10 - 0.11)
0.05 (0.05 - 0.05)
104.00 (102.59 - 105.41)
580.98 (575.08 - 586.89)
265.96 (260.89 - 271.04)
93.68 (92.66 - 94.69)
29.04 (28.54 - 29.54)
97.70 (96 - 99.41)
55.30 (54.57 - 56.03)
211.59 (207.01 - 216.17)
117.50 (115.79 - 119.22)
98.55 (96.69 - 100.41)
80.11 (79.34 - 80.88)
78.63 (77.22 - 80.05)
285.36 (280.87 - 289.85)
205.58 (201.54 - 209.61)
3.54 (3.49 - 3.60)
5.35 (5.24 - 5.46)
14.22 (14.02 - 14.41)
15.40 (14.83 - 15.98)
5.25 (5.13 - 5.38)
0.60 (0.59 - 0.60)
228.71 (224.88 - 232.53)
431.03 (425.2 - 436.85)
18.03 (17.68 - 18.37)
2.25 (2.20 - 2.31)
2.39 (2.34 - 2.44)
55.19 (54.25 - 56.13)
274.31 (270.76 - 277.86)
149.19 (146.98 - 151.4)
0.04 (0.03 - 0.04)
0.11 (0.10 - 0.11)
0.06 (0.05 - 0.06)
105.95 (102.13 - 109.77)
573.77 (557.72 - 589.83)
253.02 (238.93 - 267.11)
91.67 (88.87 - 94.47)
28.00 (26.82 - 29.18)
95.24 (90.98 - 99.51)
54.27 (52.48 - 56.06)
203.98 (192.74 - 215.22)
110.36 (105.93 - 114.78)
91.44 (86.86 - 96.01)
78.92 (76.79 - 81.05)
79.39 (75.91 - 82.87)
286.94 (275.10 - 298.78)
205.4 (194.98 - 215.81)
3.72 (3.56 - 3.87)
5.46 (5.20 - 5.71)
14.29 (13.8 - 14.78)
16.12 (14.57 - 17.67)
5.45 (5.15 - 5.74)
0.61 (0.59 - 0.64)
226.62 (216.49 - 236.74)
420.86 (406.45 - 435.28)
17.98 (17.08 - 18.89)
2.43 (2.29 - 2.58)
2.43 (2.30 - 2.55)
55.53 (52.88 - 58.19)
272.27 (262.34 - 282.20)
148.41 (142.41 - 154.42)
PC aa C36:4
PC aa C36:5
PC aa C36:6
PC aa C38:0
PC aa C38:1
PC aa C38:3
PC aa C38:4
PC aa C38:5
PC aa C38:6
PC aa C40:2
PC aa C40:3
PC aa C40:4
PC aa C40:5
PC aa C40:6
PC aa C42:0
PC aa C42:1
PC aa C42:2
PC aa C42:4
PC aa C42:5
PC aa C42:6
PC ae C30:0
PC ae C30:1
PC ae C30:2
PC ae C32:1
PC ae C32:2
PC ae C34:0
PC ae C34:1
PC ae C34:2
PC ae C34:3
PC ae C36:0
PC ae C36:1
PC ae C36:2
PC ae C36:3
PC ae C36:4
PC ae C36:5
PC ae C38:0
PC ae C38:1
PC ae C38:2
PC ae C38:3
PC ae C38:4
PC ae C38:5
PC ae C38:6
PC ae C40:1
PC ae C40:2
PC ae C40:3
PC ae C40:4
PC ae C40:5
PC ae C40:6
PC ae C42:1
PC ae C42:2
PC ae C42:3
PC ae C42:4
PC ae C42:5
PC ae C44:3
PC diacyl C36:4
PC diacyl C36:5
PC diacyl C36:6
PC diacyl C38:0
PC diacyl C38:1
PC diacyl C38:3
PC diacyl C38:4
PC diacyl C38:5
PC diacyl C38:6
PC diacyl C40:2
PC diacyl C40:3
PC diacyl C40:4
PC diacyl C40:5
PC diacyl C40:6
PC diacyl C42:0
PC diacyl C42:1
PC diacyl C42:2
PC diacyl C42:4
PC diacyl C42:5
PC diacyl C42:6
PC acyl alkyl C30:0
PC acyl alkyl C30:1
PC acyl alkyl C30:2
PC acyl alkyl C32:1
PC acyl alkyl C32:2
PC acyl alkyl C34:0
PC acyl alkyl C34:1
PC acyl alkyl C34:2
PC acyl alkyl C34:3
PC acyl alkyl C36:0
PC acyl alkyl C36:1
PC acyl alkyl C36:2
PC acyl alkyl C36:3
PC acyl alkyl C36:4
PC acyl alkyl C36:5
PC acyl alkyl C38:0
PC acyl alkyl C38:1
PC acyl alkyl C38:2
PC acyl alkyl C38:3
PC acyl alkyl C38:4
PC acyl alkyl C38:5
PC acyl alkyl C38:6
PC acyl alkyl C40:1
PC acyl alkyl C40:2
PC acyl alkyl C40:3
PC acyl alkyl C40:4
PC acyl alkyl C40:5
PC acyl alkyl C40:6
PC acyl alkyl C42:1
PC acyl alkyl C42:2
PC acyl alkyl C42:3
PC acyl alkyl C42:4
PC acyl alkyl C42:5
PC acyl alkyl C44:3
214.94 (211.5 - 218.38)
30.87 (29.93 - 31.81)
1.33 (1.30 - 1.36)
3.17 (3.11 - 3.22)
0.68 (0.66 - 0.71)
51.66 (50.84 - 52.48)
110.17 (108.44 - 111.91)
56.30 (55.37 - 57.23)
100.04 (98.21 - 101.87)
0.38 (0.37 - 0.38)
0.58 (0.57 - 0.59)
3.80 (3.73 - 3.86)
10.61 (10.41 - 10.80)
30.42 (29.80 - 31.03)
0.62 (0.61 - 0.64)
0.31 (0.31 - 0.32)
0.21 (0.20 - 0.21)
0.20 (0.20 - 0.20)
0.42 (0.41 - 0.42)
0.70 (0.69 - 0.71)
0.42 (0.41 - 0.43)
0.41 (0.39 - 0.42)
0.14 (0.14 - 0.14)
2.86 (2.82 - 2.90)
0.72 (0.71 - 0.74)
1.83 (1.80 - 1.87)
10.21 (10.05 - 10.36)
13.75 (13.52 - 13.98)
8.82 (8.67 - 8.97)
0.78 (0.76 - 0.80)
9.44 (9.29 - 9.59)
17.62 (17.35 - 17.89)
9.75 (9.60 - 9.90)
18.95 (18.63 - 19.27)
11.36 (11.17 - 11.55)
2.25 (2.20 - 2.29)
1.24 (1.22 - 1.26)
2.21 (2.17 - 2.24)
4.75 (4.68 - 4.82)
14.40 (14.20 - 14.60)
18.46 (18.19 - 18.72)
8.21 (8.07 - 8.34)
1.47 (1.45 - 1.50)
2.22 (2.18 - 2.25)
1.17 (1.16 - 1.19)
2.39 (2.36 - 2.42)
4.02 (3.97 - 4.07)
5.54 (5.45 - 5.62)
0.35 (0.34 - 0.35)
0.69 (0.67 - 0.70)
0.87 (0.86 - 0.89)
0.99 (0.97 – 1.00)
2.44 (2.40 - 2.47)
0.11 (0.11 - 0.11)
219.15 (209.28 - 229.01)
33.27 (30.45 - 36.10)
1.43 (1.34 - 1.52)
3.16 (3.01 - 3.30)
0.65 (0.58 - 0.71)
55.48 (52.83 - 58.13)
116.47 (111.35 - 121.6)
58.56 (55.75 - 61.38)
102.82 (97.67 - 107.96)
0.37 (0.35 - 0.39)
0.58 (0.54 - 0.62)
3.98 (3.78 - 4.18)
11.28 (10.67 - 11.88)
32.86 (31.02 - 34.7)
0.59 (0.56 - 0.61)
0.30 (0.29 - 0.31)
0.21 (0.20 - 0.22)
0.20 (0.19 - 0.21)
0.42 (0.40 - 0.44)
0.73 (0.70 - 0.76)
0.41 (0.39 - 0.43)
0.40 (0.37 - 0.43)
0.14 (0.14 - 0.15)
2.85 (2.73 - 2.96)
0.73 (0.69 - 0.76)
1.84 (1.76 - 1.92)
10.13 (9.73 - 10.54)
13.34 (12.72 - 13.96)
8.40 (7.96 - 8.84)
0.73 (0.70 - 0.77)
9.52 (9.12 - 9.91)
17.1 (16.34 - 17.85)
9.42 (9.00 - 9.84)
19.20 (18.33 - 20.07)
11.70 (11.11 - 12.28)
2.29 (2.18 - 2.41)
1.24 (1.18 - 1.30)
2.16 (2.06 - 2.26)
4.78 (4.58 - 4.98)
14.32 (13.77 - 14.88)
18.45 (17.72 - 19.18)
8.40 (8.04 - 8.77)
1.44 (1.38 - 1.51)
2.27 (2.17 - 2.37)
1.15 (1.11 - 1.19)
2.30 (2.21 - 2.39)
3.92 (3.77 - 4.06)
5.51 (5.28 - 5.74)
0.34 (0.33 - 0.36)
0.68 (0.65 - 0.71)
0.82 (0.79 - 0.86)
0.91 (0.87 - 0.95)
2.30 (2.22 - 2.38)
0.10 (0.09 - 0.11)
PC ae C44:4
PC ae C44:5
PC ae C44:6
lysoPC a C14:0
lysoPC a C16:0
lysoPC a C16:1
lysoPC a C17:0
lysoPC a C18:0
lysoPC a C18:1
lysoPC a C18:2
lysoPC a C20:3
lysoPC a C20:4
lysoPC a C28:1
SM OH C14:1
SM OH C16:1
SM OH C22:1
SM OH C22:2
SM OH C24:1
SM C16:0
SM C16:1
SM C18:0
SM C18:1
SM C20:2
SM C24:0
SM C24:1
SM C26:0
SM C26:1
H1
PC acyl alkyl C44:4
PC acyl alkyl C44:5
PC acyl alkyl C44:6
lysoPC acyl C14:0
lysoPC acyl C16:0
lysoPC acyl C16:1
lysoPC acyl C17:0
lysoPC acyl C18:0
lysoPC acyl C18:1
lysoPC acyl C18:2
lysoPC acyl C20:3
lysoPC acyl C20:4
lysoPC acyl C28:1
Hydroxysphingomyelin C14:1
Hydroxysphingomyelin C16:1
Hydroxysphingomyelin C22:1
Hydroxysphingomyelin C22:2
Hydroxysphingomyelin C24:1
Sphingomyelin C16:0
Sphingomyelin C16:1
Sphingomyelin C18:0
Sphingomyelin C18:1
Sphingomyelin C20:2
Sphingomyelin C24:0
Sphingomyelin C24:1
Sphingomyelin C26:0
Sphingomyelin C26:1
Hexose
0.4 (0.39 - 0.41)
1.85 (1.82 - 1.88)
1.23 (1.21 - 1.25)
5.60 (5.54 - 5.66)
108.73 (107.15 - 110.31)
3.23 (3.16 - 3.31)
2.00 (1.97 - 2.04)
29.43 (28.92 - 29.94)
19.61 (19.24 - 19.97)
36.81 (35.93 - 37.70)
2.40 (2.35 - 2.45)
6.28 (6.16 - 6.40)
0.63 (0.62 - 0.65)
7.46 (7.33 - 7.59)
3.86 (3.79 - 3.93)
14.46 (14.22 - 14.70)
12.36 (12.15 - 12.57)
1.58 (1.55 - 1.61)
114.01 (112.55 - 115.47)
17.22 (16.99 - 17.46)
24.34 (23.97 - 24.72)
11.70 (11.51 - 11.89)
0.69 (0.67 - 0.71)
25.28 (24.89 - 25.68)
49.32 (48.5 - 50.15)
0.23 (0.23 - 0.24)
0.54 (0.53 - 0.55)
4616 (4545 - 4687)
0.37 (0.36 - 0.39)
1.71 (1.64 - 1.79)
1.13 (1.08 - 1.18)
5.68 (5.54 - 5.83)
109.93 (106.04 - 113.82)
3.34 (3.13 - 3.54)
2.03 (1.93 - 2.13)
30.50 (29.21 - 31.79)
18.79 (17.94 - 19.63)
33.69 (31.54 - 35.83)
2.44 (2.33 - 2.56)
6.38 (6.05 - 6.70)
0.62 (0.58 - 0.65)
7.64 (7.29 - 7.99)
3.94 (3.75 - 4.13)
14.74 (14.08 - 15.4)
12.53 (11.94 - 13.12)
1.62 (1.53 - 1.70)
112.99 (109.01 - 116.97)
17.59 (16.91 - 18.28)
24.91 (23.82 – 26.00)
11.93 (11.35 - 12.51)
0.73 (0.66 - 0.80)
25.42 (24.29 - 26.56)
47.77 (45.55 – 50.00)
0.23 (0.22 - 0.25)
0.54 (0.52 - 0.57)
4669 (4467 - 4871)
* Abbreviations: aa, diacyl; ae, acyl alkyl; C, Carbon; CI, confidence interval; PC,
Phosphatidylcholine.
Table S2: Intraclass-correlation coefficient of the six identified metabolites predictive for
incident hypertension.*
Metabolite
ICC (95% CI)
Serine
0.61 (0.47-0.72)
PC ae C42:4
0.75 (0.65-0.83)
PC ae C44:3
0.47 (0.30-0.61)
PC aa C38:4
0.70 (0.59-0.79)
Glycine
0.68 (0.55-0.77)
PC aa C38:3
0.52 (0.37-0.65)
1
*Results were previously presented by Floegel et al.
Abbreviations: aa, diacyl; ae, acyl alkyl; C, carbon; CI, confidence interval; ICC, intraclasscorrelation coefficient; PC, phosphatidylcholine.
Table S3: number of non-cases and case of incident hypertension listed for given metabolite
concentration ranges.
Metabolite concentration
[µmol/L]
Serine
50 - 90
90.1 - 110
110.1 - 150
150.1 - 200
else
PC ae C42:4
0.5 - 0.75
0.75 - 1.0
1.0 - 2.0
else
PC ae C44:3
>0.05 - 0.08
>0.08 - 0.10
>0.10 - 0.12
>0.12 - 0.15
>0.15 - 0.20
else
PC aa C38:4
0 - 50
50 - 90
90 - 110
110 - 150
>150
Glycine
0-140
>140-200
>200-300
>300-400
>400
PC aa C38:4
0-30
>30-60
>60
Number of non-cases
(mean follow-up time)
Number of cases
(mean follow-up time)
120 (10.0)
136 (10.1)
614 (10.5)
98 (10.8)
13 (10.4)
31 (6.1)
20 (7.2)
73 (6.7)
10 (7.0)
1 (4.65)
133 (9.9)
417 (10.4)
424 (10.5)
7 (10.7)
34 (6.6)
57 (6.9)
43 (6.5)
1 (4.7)
161 (10.3)
223 (10.5)
252 (10.2)
236 (10.6)
75 (10.5)
34 (10.3)
24 (6.7)
48 (6.7)
17 (7.5)
25 (6.6)
13 (5.5)
8 (6.5)
6 (11.1)
203 (10.3)
343 (10.4)
351 (10.4)
78 (10.5)
1 (4.7)
29 (7.5)
25 (6.5)
66 (6.6)
14 (5.6)
11 (9.8)
164 (10.5)
550 (10.4)
188 (10.4)
68 (10.3)
5 (4.9)
33 (6.5)
66 (7.1)
23 (6.9)
8 (4.3)
21 (10.6)
720 (10.4)
240 (10.2)
2 (5.0)
91 (7.0)
42 (5.6)
Figure S1: Serum metabolite network derived by gaussian graphic modelling. Each node represents one metabolite and each edge between two nodes
represents the partial correlation between two metabolites mutually adjusted for all other metabolites. Metabolites that were identified to be associated with
incident hypertension are colour coded. Abbreviations: a, acyl; aa, diacyl; ae, acyl-alkyl; PC, phosphatidylcholine.