Clinical Chemistry 58:11
1582–1591 (2012)
Proteomics and Protein Markers
Clinical and Genetic Correlates of Growth Differentiation
Factor 15 in the Community
Jennifer E. Ho,1,2,3† Anubha Mahajan,4† Ming-Huei Chen,1,5 Martin G. Larson,1,6 Elizabeth L. McCabe,5
Anahita Ghorbani,3 Susan Cheng,1,7 Andrew D. Johnson,1,2 Cecilia M. Lindgren,4 Tibor Kempf,8 Lars Lind,9
Erik Ingelsson,10 Ramachandran S. Vasan,1,11 James Januzzi,3 Kai C. Wollert,8 Andrew P. Morris,4† and
Thomas J. Wang1,3,12†*
BACKGROUND:
Growth differentiation factor 15 (GDF15),
a stress-responsive cytokine produced in cardiovascular
cells under conditions of inflammation and oxidative
stress, is emerging as an important prognostic marker in
individuals with and without existing cardiovascular disease (CVD). We therefore examined the clinical and genetic correlates of circulating GDF15 concentrations,
which have not been investigated collectively.
METHODS: Plasma GDF15 concentrations were measured in 2991 participants in the Framingham Offspring Study who were free of clinically overt CVD
(mean age, 59 years; 56% women). Clinical correlates
of GDF15 were examined in multivariable analyses. We
then conducted a genomewide association study of the
GDF15 concentration that included participants in
the Framingham Offspring Study and participants in
the PIVUS (Prospective Investigation of the Vasculature in Uppsala Seniors) study.
RESULTS: GDF15 was positively associated with age,
smoking, antihypertensive treatment, diabetes, worse
kidney function, and use of nonsteroidal antiinflammatory drugs (NSAIDs), but it was negatively associated with total cholesterol and HDL cholesterol. Clinical correlates accounted for 38% of interindividual
variation in the circulating GDF15 concentration,
whereas genetic factors accounted for up to 38% of the
1
Framingham Heart Study, Framingham, MA; 2 Center for Population Studies,
National Heart, Lung and Blood Institute, Bethesda, MD; 3 Cardiology Division,
Massachusetts General Hospital, Harvard Medical School, Boston, MA; 4 Wellcome Trust Centre for Human Genetics, University of Oxford, Oxford, UK;
5
School of Public Health and 6 Department of Mathematics and Statistics,
Department of Medicine, Boston University, Boston, MA; 7 Division of Cardiology, Brigham and Women’s Hospital, Harvard Medical School, Boston, MA;
8
Department of Cardiology and Angiology, Hannover Medical School, Hannover, Germany; 9 Department of Medicine and Uppsala Clinical Research
Centre, University of Uppsala, Uppsala, Sweden; 10 Department of Medical
Epidemiology and Biostatistics, Karolinska Institute, Stockholm, Sweden; 11 Division of Cardiology and Preventive Medicine, Department of Medicine, Boston
University, Boston, MA; 12 Cardiovascular Research Center, Massachusetts General Hospital, Harvard Medical School, Boston, MA.
†
These authors contributed equally to this work.
* Address correspondence to this author at: Cardiology Division, GRB-800, Massachusetts General Hospital, 55 Fruit St., Boston, MA 02114. Fax 617-7264105; e-mail [email protected].
1582
residual variability (h2 ⫽ 0.38; P ⫽ 2.5 ⫻ 10⫺11). We
identified 1 locus of genomewide significance. This locus, which is on chromosome 19p13.11 and includes the
GDF15 gene, is associated with GDF15 concentration
(smallest P ⫽ 2.74 ⫻ 10⫺32 for rs888663). Conditional
analyses revealed 2 independent association signals at this
locus (rs888663 and rs1054564), which were associated
with altered cis gene expression in blood cell lines.
CONCLUSIONS: In ambulatory individuals, both cardiometabolic risk factors and genetic factors play important roles in determining circulating GDF15 concentrations and contribute similarly to the overall
variation.
© 2012 American Association for Clinical Chemistry
Growth differentiation factor 15 (GDF15)13 is a member of the transforming growth factor ␤ superfamily,
and expression of the GDF1514 gene in cardiomyocytes, vascular smooth muscle cells, and endothelial
cells is strongly upregulated in response to oxidative
stress and inflammation (1 ). Increased GDF15 concentrations have been associated with an adverse prognosis in patients with acute coronary syndromes (2– 6 )
and chronic heart failure (7, 8 ). More recently, GDF15
has been associated with atherosclerosis, cardiovascu-
Received May 24, 2012; accepted August 30, 2012.
Previously published online at DOI: 10.1373/clinchem.2012.190322
Nonstandard abbreviations: GDF15, growth differentiation factor 15; CVD,
cardiovascular disease; CRP, C-reactive protein; FHS, Framingham Heart Study;
LV, left ventricular; PIVUS, Prospective Investigation of the Vasculature in
Uppsala Seniors (study); NSAID, nonsteroidal antiinflammatory drug; eGFR,
estimated glomerular filtration rate; BNP, B-type natriuretic peptide; MIP,
molecular inversion probe; SNP, single-nucleotide polymorphism; BMI, body
mass index; GEE, generalized estimating equation; hsTnI, cardiac troponin I
measured with a high-sensitivity assay.
14
Human genes: GDF15, growth differentiation factor 15; PGPEP1, pyroglutamylpeptidase I; LRRC25, leucine rich repeat containing 25; CRLF2, cytokine
receptor-like factor 2; LRRC31, leucine rich repeat containing 31.
13
Clinical and Genetic Correlates of GDF15
lar events, and mortality in community-dwelling
adults (9 –11 ). Thus, GDF15 appears to be an important biomarker of cardiovascular disease (CVD) that
confers prognostic information in addition to that of
established risk factors and other biomarkers, including natriuretic peptides and C-reactive protein (CRP)
(9, 11 ).
Considering the emerging role of GDF15 as a
prognostic biomarker, we sought to understand its genetic and clinical correlates. Although previous studies
have highlighted the correlation of selected clinical factors
with GDF15 expression, the contribution of genetic factors to interindividual variation in GDF15 concentrations
has not been evaluated comprehensively. To examine genetic correlates, we estimated heritability in a familybased cohort and conducted a genomewide association
study to explore the association of genetic loci with circulating GDF15 concentrations.
Materials and Methods
STUDY SAMPLES
The Framingham Heart Study (FHS) was initiated in
1948 to investigate risk factors for CVD. The children
(and their spouses) of the original participants were
enrolled in the Framingham Offspring Cohort in 1971
(12 ). GDF15 concentrations were measured during a
routine examination of the offspring cohort (1995–
1998). Of the 3532 eligible participants, we excluded
participants with missing biomarker measurements
(n ⫽ 82), heart failure (n ⫽ 38), left ventricular (LV)
systolic dysfunction as revealed by echocardiography
(defined as LV fractional shortening ⬍0.30 or mild or
greater LV systolic dysfunction by visual inspection)
(n ⫽ 302), or missing covariates (n ⫽ 60). We additionally excluded participants with previous myocardial infarction (n ⫽ 59) from the clinical-correlates
analyses, leaving 2991 participants for that analysis. All
participants provided written informed consent, and
the protocol was approved by the Institutional Review
Board at Boston University Medical Center.
For the genetic analyses, we used additional data
from 898 individuals enrolled in the Prospective Investigation of the Vasculature in Uppsala Seniors (PIVUS)
study, a community-based cohort of elderly individuals living in Sweden (see Data Methods in the Data
Supplement that accompanies the online version of
this article at http://www.clinchem.org/content/vol58/
issue11) (10 ).
examination were considered current smokers. Diabetes mellitus was defined as a fasting glucose concentration ⱖ126 mg/dL (ⱖ7.0 mmol/L) or the use of insulin
or oral hypoglycemic medications. The estimated glomerular filtration rate (eGFR) was calculated by means
of the Modification of Diet in Renal Disease equation
(13 ). Metabolic syndrome was defined as described
previously (14 ).
LABORATORY TESTING
Blood samples were collected after an overnight fast
and centrifuged immediately. Citrate-treated plasma
samples were stored at ⫺80 °C. GDF15 concentrations
in the FHS samples were measured with a precommercial, automated electrochemiluminescence immunoassay on a Cobas e 411 analyzer (Roche Diagnostics).
The assay has a limit of detection ⬍10 ng/L, a linear
measurement interval up to 20 000 ng/L, and interassay imprecisions of 2.3% and 1.8% at GDF15 concentrations of 1100 ng/L and 17 200 ng/L, respectively.
GDF15 values obtained with the electrochemiluminescence assay correlate closely with the values measured
with a previously described immunoradiometric assay
(r ⫽ 0.980; slope, 1.049; y intercept, ⫺136 ng/L; n ⫽ 45
samples with GDF15 concentrations of 567–13 334
ng/L) (15 ) that was used to measure GDF15 concentration in the PIVUS samples (10 ). The high-sensitivity
assay for CRP and the assay for B-type natriuretic peptide (BNP) were performed as previously described
(16 ). Troponin I was measured with a high-sensitivity
immunoassay (Erenna威 hsTnI; Singulex) (17 ).
GENOMEWIDE GENOTYPING AND IMPUTATION
The genotyping of FHS samples was conducted with
the Affymetrix 500K mapping array and the Affymetrix
50K gene-focused MIP (molecular inversion probe)
array. Genotypes from the Affymetrix 500K mapping
array were called with CHIAMO software. Imputation
of genotypes to the HapMap set of 2.5 ⫻ 106 singlenucleotide polymorphisms (SNPs) (CEU population,
release 22, build 36; http://hapmap.ncbi.nlm.nih.gov/)
was performed with MACH (version 1.0.15) (18 ).
Genomewide genotyping in the PIVUS population was
performed with the HumanOmniExpress BeadChip
(Illumina). Genotypes were called with GenCall software implemented in GenomeStudio. Imputation to
the same HapMap reference panel was performed with
IMPUTE version 2 (19 ).
STATISTICAL ANALYSES
CLINICAL ASSESSMENT
All FHS participants underwent routine medical history review, physical examination, and laboratory testing, as described previously (12 ). Participants regularly
smoking cigarettes during the year before the baseline
Healthy sample (FHS cohort). To examine reference
values, we studied a healthy subset of 1159 participants
without any major medical conditions, including coronary heart disease, heart failure, atrial fibrillation, diClinical Chemistry 58:11 (2012) 1583
abetes, hypertension, obesity [body mass index (BMI)
ⱖ30 kg/m2], valvular heart disease (systolic murmur
ⱖ3/6 in severity or any diastolic murmur), pulmonary
disease (forced expiratory volume in 1 s below the
lower reference limit), or serum creatinine ⱖ2.0 mg/dL
(ⱖ1768 ␮mol/L). Simple empirical estimates for the
2.5th, 10th, 50th, 90th, and 97.5th percentiles were examined by 10-year age and sex categories.
Clinical correlates (FHS cohort). GDF15 concentrations
were log-transformed because of the right-skewed distribution of the data. To examine the association of
GDF15 with clinical covariates, we used a forwardselection linear regression model, with a P value ⬍0.05
chosen for entry. Age and sex were forced into the
models; candidate covariates included systolic blood
pressure, diabetes, BMI, cigarette smoking, total and
HDL cholesterol, hypertension treatment, LV hypertrophy, atrial fibrillation, eGFR, and NSAID use.
NSAID use was included because these drugs are
known to induce GDF15 gene expression (20 ).
In secondary analyses, the association of GDF15
concentration with the metabolic syndrome was examined in age- and sex-adjusted models. Cardiovascular
risk factors were examined in aggregate with the Framingham CVD risk score (21 ). Generalized estimating
equations were used in secondary analyses to account
for familial correlations.
Heritability (FHS cohort). The heritability of GDF15
was estimated with variance component models that
use Sequential Oligogenic Linkage Analysis Routines
(22 ). Heritability estimates were age- and sex-adjusted
and multivariable-adjusted (age, sex, systolic blood
pressure, antihypertensive medication use, diabetes
mellitus, and smoking status).
Genomewide association study, replication, and metaanalysis (FHS and PIVUS cohorts). The associations of
genetic variants with GDF15 concentration in the FHS
cohort were tested with an additive genetic model that
uses linear mixed-effects models to accommodate pedigree data. We adjusted for age, sex, systolic blood pressure, antihypertensive medication use, diabetes mellitus, and smoking status. Results were considered
significant genomewide at P values ⬍5 ⫻ 10⫺8. In secondary analyses, we tested the NSAID ⫻ SNP interaction term for hits of genomewide significance within
the FHS population.
We then performed a separate genomewide association study with the PIVUS cohort to replicate the
FHS findings with an independent cohort and to metaanalyze the results. Owing to differences between the
FHS and PIVUS populations in GDF15 distribution,
we determined an inverse normal transformation to be
the most appropriate for metaanalysis of the genetic
1584 Clinical Chemistry 58:11 (2012)
data. Genomewide association analyses were performed with the PIVUS cohort by means of an additive
model and linear regression, with adjustment for the
same covariates as the FHS cohort analyses. PIVUS
analyses were also adjusted for 2 principal components
to adjust for population stratification. We used fixed
effects with inverse variance weighting in the metaanalysis of the results for the 2 cohorts. Heterogeneity
in allelic effects between the FHS and PIVUS cohorts
were assessed by means of the Cochran Q statistic. Imputed results were filtered for a minor-allele frequency
⬍0.01, and the imputed ratio or information score was
examined for quality control. A ratio ⬎0.4 was considered acceptable. The genomic control parameter ␭ was
1.05 in the FHS cohort, 1.01 in the PIVUS cohort, and
1.03 in the metaanalysis; therefore, analyses were not
adjusted by genomic control to account for residual
stratification within or between cohorts. We conducted conditional analyses by conditioning on each of
the most significant SNPs for GDF15 in the FHS cohort
(rs749451), the PIVUS cohort (rs1054564), and the
metaanalysis (rs888663). We performed genomewide
association analyses separately in each study: We accounted for each SNP alone and in 2-SNP combinations,
and we metaanalyzed results to determine whether SNPs
of genomewide significance represented independent signals. We also performed secondary analyses in which we
adjusted for BMI (correlated with the GDF15 concentration in the PIVUS cohort but not in the FHS cohort) and
adjusted metaanalysis results for 2 principal components
to account for stratification within or between cohorts.
IN SILICO ASSOCIATION OF GENETIC VARIANTS AND CLINICAL
TRAITS
The association of 3 independent variants of genomewide significance (rs888663, rs749451, and rs1054564)
with lipid traits was examined in a previously published
genomewide association study of ⬎100 000 individuals of European descent (23 ). These GDF15 variants
were searched against a collected database of expression SNPs to investigate any association with levels of
cis gene expression across different tissue types (see
Data Methods in the online Data Supplement).
Results
STUDY SAMPLE
Characteristics of the overall sample of 2991 FHS participants are summarized in Table 1. The mean age was
59 years, and 56% of the participants were women. The
median GDF15 concentration in the general FHS sample was 1020 ng/L (interquartile range, 803–1362 ng/L)
in men and 1017 ng/L (interquartile range, 809 –1297
ng/L) in women.
Clinical and Genetic Correlates of GDF15
Table 1. Characteristics of 2991 FHS and 898 PIVUS participants.a
FHS
Men (n ⴝ 1316)
Women (n ⴝ 1675)
59 (10)
59 (10)
70 (0.2)
70 (0.2)
Systolic blood pressure, mmHg
130 (17)
127 (20)
146 (22)
153 (23)
Diastolic blood pressure, mmHg
78 (9)
74 (9)
79 (10)
78 (10)
BMI, kg/m2
28 (4)
27 (6)
27 (4)
27 (5)
11
9
12
9
Age, years
Diabetes mellitus, %
Women (n ⴝ 450)
28
24
29
30
Smoker, %
15
16
11
13
Total cholesterol, mg/dLb
200 (36)
212 (38)
200 (37)
222 (37)
HDL cholesterol, mg/dLb
44 (12)
58 (16)
54 (14)
64 (16)
1180 (637)
1153 (615)
1257 (421)
1162 (390)
91 (42)
89 (43)
84 (21)
78 (21)
9
13
NA
NA
eGFR, mL 䡠 min⫺1 䡠 (1.73 m2)⫺1
NSAID use, %
b
Men (n ⴝ 437)
Antihypertensive treatment, %
GDF15, ng/L
a
PIVUS
Values are presented as the mean (SD) unless otherwise indicated. NA, data not available.
To convert cholesterol concentrations to millimoles per liter, multiply by 0.02586.
The GDF15 concentration in the subset of apparently healthy individuals free of major medical comorbidities (n ⫽ 1159; Table 2) varied with age but not by
sex. The median GDF15 concentration was 901 ng/L.
For men 30 –39 years of age, the median and 90th percentile GDF15 concentrations were 651 ng/L and 1197
ng/L, respectively, whereas the corresponding values
for men 70 –79 years of age were 1389 ng/L and 2030
ng/L. Within the apparently healthy sample, 21% of the
participants had a GDF15 concentration above the pre-
viously established upper reference limit of ⬎1200
ng/L (15 ).
CLINICAL CORRELATES OF GDF15 IN THE FHS COHORT
In multivariable analyses, men and women had similar
GDF15 concentrations. GDF15 concentration was positively associated with age, diabetes, antihypertensive
treatment, smoking, and NSAID use but was negatively
associated with total cholesterol, HDL cholesterol, and
eGFR (Table 3). The R2 value for this model was 0.38.
Table 2. GDF15 concentrations for 1159 apparently healthy FHS participants without major cardiovascular,
renal, or pulmonary disease.
GDF15, ng/L
Percentile
Sex
Male
Female
Age group, years
n
2.5th
10th
50th
90th
97.5th
30–39
19
473
483
651
1197
1498
40–49
121
535
571
726
1081
1250
50–59
197
574
655
863
1414
1838
60–69
104
723
805
1115
1711
2197
70–79
27
893
1084
1389
2030
5006
30–39
14
520
543
745
1007
1085
40–49
180
488
561
762
1172
1574
50–59
275
557
668
890
1269
1786
60–69
167
652
784
1025
1553
1833
70–79
55
845
939
1277
1717
2562
Clinical Chemistry 58:11 (2012) 1585
Table 3. Clinical correlates of GDF15 in 2991
FHS participants.a
Clinical characteristic
Estimated
coefficient
SE
P
Age, per 10 years
0.211
0.006 ⬍0.0001
Sex, men vs women
0.006
0.013
Diabetes, yes vs no
0.142
0.020 ⬍0.0001
Hypertension treatment, yes vs
no
0.073
0.014 ⬍0.0001
0.64
0.220
0.016 ⬍0.0001
Total cholesterol, per 38 mg/dL
(per 0.98 mmol/L)
⫺0.028
0.006 ⬍0.0001
HDL cholesterol, per 16 mg/dL
(per 0.41 mmol/L)
⫺0.024
0.007
eGFR, per 42 mL 䡠 min⫺1 䡠
(1.73 m2)⫺1
⫺0.032
0.006 ⬍0.0001
Smoking, yes vs no
NSAID use, yes vs no
a
0.050
0.018
0.0003
0.006
Regression coefficients indicate the increase in log GDF15 concentration in
the presence vs absence of the trait for dichotomous variables, and per
1-SD increase as noted in continuous variables. The following variables
were not significant (P ⬎ 0.05) in the forward-selection model: systolic
blood pressure, BMI, LV hypertrophy, and atrial fibrillation.
In secondary analyses, the GDF15 concentration
was significantly associated with the metabolic syndrome in age- and sex-adjusted analyses (P ⬍ 0.0001).
GDF15 concentration increased across quartiles of the
Framingham CVD score (Fig. 1A) and was positively
correlated with the risk score in age- and sex-adjusted
analyses (P ⬍ 0.0001).
Secondary analyses also showed that the positive
association of NSAID use with GDF15 concentration
appeared to be independent of inflammation, because
the correlation persisted within the group of apparently healthy individuals (P ⫽ 0.006), after exclusion of
participants with inflammatory arthritis (P ⫽ 0.01),
and after adjustment for CRP (P ⫽ 0.01). Secondary
analyses that adjusted for relatedness within our sample with generalized estimating equation (GEE) models
did not change the results appreciably. Inclusion of troponin I measured with a high-sensitivity assay (hsTnI),
BNP, or CRP in the clinical-correlates analysis did not
change our findings materially (see Table 1 in the online Data Supplement), although all 3 biomarkers were
modestly correlated with GDF15 concentration
[hsTnI, r ⫽ 0.19 (P ⬍ 0.0001); BNP, r ⫽ 0.29 (P ⬍
0.0001); CRP, r ⫽ 0.25 (P ⬍ 0.0001)].
GENETIC CORRELATES OF GDF15
The age- and sex-adjusted heritability of GDF15 in
the FHS cohort was 0.38 (P ⫽ 2.5 ⫻ 10⫺11), and the
multivariable-adjusted heritability was 0.30 (P ⫽
4.8 ⫻ 10⫺8).
1586 Clinical Chemistry 58:11 (2012)
Fig. 1. Association of mean circulating GDF15 concentrations with clinical and genetic determinants.
Data are presented as the mean and SE. (A), Mean circulating GDF15 concentrations according to Framingham risk
score quartile. (B), Circulating GDF15 concentrations by
rs888663 genotype. GDF15 concentrations were estimated
(back-transformed from log-transformed data). Data are for
the FHS cohort.
Nine SNPs had significant associations genomewide (P ⬍ 5 ⫻ 10⫺8) in the FHS cohort (Table 4; for
details see Table 2 in the online Data Supplement).
These SNPs were located in noncoding regions near the
GDF15 and PGPEP1 (pyroglutamyl-peptidase I) genes
(see Table 3 in the online Data Supplement for pairwise
linkage disequilibrium data). The most significantly associated SNP (rs749451) had a P value of 1.9 ⫻ 10⫺31
and explained 2.5% of the residual phenotypic variance
in GDF15 concentration. Three of these 9 SNPs
(rs1054564, rs1227731, rs3195944) also had genome-
Clinical and Genetic Correlates of GDF15
Table 4. Variants of genomewide significance associated with GDF15 in the FHS and PIVUS cohorts.a
Chrb
19
SNP
rs888663
Position
(NCBI 36.3)
Location relative
to gene
Nearest
gene
18345922
Downstream
PGPEP1
Allele
(major/minor) MAF
T/G
FHS P
PIVUS P
Metaanalysis
P
0.19
7.12 ⫻ 10⫺31
1.01 ⫻ 10⫺3
2.74 ⫻ 10⫺32
⫺30
⫺4
19
rs3746181
18338017
3⬘ UTR
PGPEP1
G/A
0.19
2.72 ⫻ 10
4.08 ⫻ 10
6.22 ⫻ 10⫺32
19
rs1363120
18343304
Downstream
PGPEP1
G/C
0.19
2.38 ⫻ 10⫺30
9.23 ⫻ 10⫺4
1.50 ⫻ 10⫺31
0.43
⫺31
⫺2
1.21 ⫻ 10⫺29
⫺13
8.17 ⫻ 10⫺28
⫺13
19
19
rs749451
rs1054564
18340647
18360815
3⬘ UTR
3⬘ UTR
PGPEP1
GDF15
C/T
G/C
0.14
1.94 ⫻ 10
⫺18
2.14 ⫻ 10
2.49 ⫻ 10
19
rs1227731
18358903
Intronic
GDF15
G/A
0.14
2.32 ⫻ 10
2.53 ⫻ 10
8.22 ⫻ 10⫺28
19
rs3195944
18337711
3⬘ UTR
PGPEP1
A/G
0.13
1.16 ⫻ 10⫺17
2.24 ⫻ 10⫺9
6.16 ⫻ 10⫺25
0.24
⫺13
⫺1
1.91 ⫻ 10⫺11
⫺1
3.12 ⫻ 10⫺6
19
19
rs17725099
rs1043063
18343358
18341171
Downstream
3⬘ UTR
PGPEP1
PGPEP1
G/A
C/T
0.35
⫺18
5.17 ⫻ 10
2.59 ⫻ 10
⫺9
4.58 ⫻ 10
1.74 ⫻ 10
4.29 ⫻ 10
a
In the online Data Supplement, see Table 2 for estimated coefficients and SEs, Table 3 for pairwise linkage disequilibrium values between SNPs, Fig. 1 for a
quantile– quantile plot, and Fig. 2 for a Manhattan plot of the genomewide association study.
b
Chr, chromosome; MAF, minor-allele frequency; UTR, untranslated region.
wide associations with GDF15 in the PIVUS sample.
Metaanalysis also revealed 8 SNPs near GDF15 with
genomewide significance, with rs888663 being the
most significantly associated SNP (P ⫽ 2.7 ⫻ 10⫺32);
Fig. 1B displays GDF15 concentrations according to
rs888663 genotype. The regional-association plot in
Fig. 2 demonstrates that all SNPs with genomewide
significance were well within 100 kb of the GDF15
gene.
Secondary analyses that adjusted for the first 2
principal components did not substantively alter metaanalysis results (see Table 4 in the online Data Supplement), and adjustment for BMI did not change the
results meaningfully. There was no statistically significant interaction between NSAID use or eGFR and the
results for the loci with genomewide significance
within the FHS sample (P ⬎ 0.05 for all). Adjustment
for eGFR did not attenuate heritability estimates (h2 ⫽
Fig. 2. Regional-association plot of metaanalysis loci associated with circulating GDF15 concentrations.
CEU, specific population within the HapMap database; cM/Mb, centimorgans per megabase.
Clinical Chemistry 58:11 (2012) 1587
0.36; P ⫽ 1.0 ⫻ 10⫺9) or associations with genomewide
significance.
Conditional analyses accounting for rs888663,
rs749451, and rs1054564 revealed 2 potentially independent signals: 1 signal from rs888663/rs749451 (r 2 ⫽
0.43 for the SNPs with each other), and 1 signal from
rs1054564 (see Table 5 in the online Data Supplement).
When conditioned on these 2 signals, no additional
SNPs reached genomewide significance. Although the
directionality of the association between SNP genotype
and GDF15 concentration was consistent between
studies, there was nominal evidence (P ⬍ 0.05, Cochran Q statistic) of heterogeneity in allelic effects between the FHS and PIVUS results, despite the similarity
of the 2 populations in allele frequencies (see Table 2 in
the online Data Supplement for allele frequencies;
rs888663, P ⫽ 0.04; rs749451, P ⫽ 0.0005; rs1054564,
P ⫽ 0.001).
Putative functional variants. All 3 SNPs associated with
GDF15 (rs888663, rs749451, and rs1054564) were associated with cis gene expression in lymphocyte,
monocyte, and adipose tissue cell lines (24 –26 ). Specifically, rs749451 and rs888663 were associated with
PGPEP1 gene expression (P ⫽ 1.32 ⫻ 10⫺11 and 4.86 ⫻
10⫺4, respectively). This gene encodes pyroglutamyl
peptidase I and abuts GDF15 on chromosome 9p13.11.
Both rs1054564 and rs888663 were associated with cis
expression of the LRRC25 (leucine rich repeat containing 25) gene, which is also on 9p13.11 (P ⫽ 3.61 ⫻
10⫺66 and 7.81 ⫻ 10⫺7, respectively). In addition,
rs1054564 was associated with trans expression of the
CRLF2 (cytokine receptor-like factor 2) gene, located
on chromosome Xp22.3 (P ⫽ 1.61 ⫻ 10⫺11), and the
LRRC31 (leucine rich repeat containing 31) gene,
which is on chromosome 3q26.2 (P ⫽ 2.87 ⫻ 10⫺11).
In silico association with clinical traits. Given the crosssectional association of GDF15 and both total cholesterol and HDL cholesterol, we pursued an in silico investigation of the 2 independent GDF15 variants
within a published genomewide association study of
lipid traits (23 ). Our genomewide-significant GDF15
SNP rs1054564 was associated with HDL cholesterol
(P ⫽ 0.025). Specifically, the allele related to higher
GDF15 concentrations was associated with lower HDL
cholesterol, a finding consistent with the association
found in the clinical correlates of GDF15.
Discussion
In this report, we describe the clinical and genetic correlates of GDF15 in the community. Our findings demonstrate that higher circulating GDF15 concentrations
are associated with increasing cardiometabolic risk fac1588 Clinical Chemistry 58:11 (2012)
tors in individuals without overt CVD. Our findings
also suggest that genetic factors play an important role
in determining GDF15 concentrations. Additive genetic effects may explain up to 38% of the phenotypic
variation in GDF15 concentration, which is comparable to the proportion of the variation explained by clinical factors. Our genomewide association studies
showed that specific variants near the GDF15 gene on
chromosome 19p13.11 were strongly associated with
GDF15 concentration. Furthermore, GDF15 variants
were associated with gene expression in published
databases.
Experimental studies have shown that the GDF15
gene is expressed in human atherosclerotic plaque (27 )
and cardiac myocytes after an ischemic insult (28 ). Under these conditions, GDF15 appears to provide protection against cardiac injury via antiinflammatory
(29 ), antiapoptotic (28 ), and antihypertrophic (30 )
pathways. Clinical studies of individuals with existing
CVD (2– 8 ) and of community-based populations (9 –
11 ) have largely shown higher GDF15 concentrations
to be associated with adverse outcomes, although
whether GDF15 is a mediator or a marker of CVD remains unclear. It may be that GDF15 is similar to the
natriuretic peptides, which have protective biological
effects but are increased in individuals at risk for CVD,
presumably reflecting a response to increased hemodynamic stress (31 ). Accordingly, understanding the
clinical and genetic correlates of GDF15 may elucidate
pathways underlying the association of GDF15 with
CVD.
As in previous studies (15 ), GDF15 concentration
and age showed a strong association, which was quite
pronounced even in apparently healthy adults. Using
the previously studied upper reference limit of 1200
ng/L (15 ), we found that ⬍10% of adults 40 – 49 years
of age met the criteria for “abnormal” GDF15 concentrations. In contrast, ⬎50% of apparently healthy
adults 70 –79 years of age were classified as having an
increased GDF15 concentration when the same cutoff
was used. This marked increase in GDF15 concentration with age may be related to a higher burden of subclinical CVD, even within an ostensibly healthy population. GDF15 concentrations also have been shown to
be increased in a number of advanced cancers (32 ), but
it is less likely that occult malignancies could be related
to the age-related increase in GDF15 concentration.
Lastly, owing to enhanced production upon kidney injury or disease, age-related renal dysfunction may lead
to higher GDF15 concentrations, although age was a
correlate even after we adjusted for eGFR in our study.
In considering the potential clinical utility of GDF15 as
a biomarker, it will be important to account for the
robust association between age and GDF15 concentra-
Clinical and Genetic Correlates of GDF15
tion, an effect greater than any other clinical trait in
apparently healthy adults.
We found no difference between the sexes in
GDF15 concentration, in contrast to prior studies that
examined individuals with acute coronary syndromes
(3, 5, 6, 33–35 ), heart failure (8 ), or older age
(9, 10 )—all of which demonstrated higher GDF15
concentrations in men than in women. It is possible
that sex differences in CVD severity or subtype may
have contributed to the previously observed differences in GDF15 concentrations.
We observed a strong association between higher
GDF15 concentrations and cardiometabolic risk factors, including diabetes, hypertension, smoking, and
low HDL, which precede the onset of overt CVD. Our
findings are similar to those for older participants in
the PIVUS study (10 ), as well as those in the Rancho
Bernardo Study (9 ). Although the mechanisms by
which GDF15 may modulate risk are not well understood, recent animal and clinical studies have shown
that GDF15 is produced in adipocytes and may act as
an adipokine (36 ). Circulating GDF15 has been associated with insulin resistance in obese individuals (37 ).
Interestingly, higher GDF15 concentrations were associated with lower total cholesterol and LDL cholesterol
in the community, as others have found (9, 10 ). Experimental data suggest that oxidized LDL can induce
GDF15 in atherosclerotic lesions (27 ), and different
LDL subtypes might be differentially associated with
GDF15.
Lastly, we have shown that NSAID use is associated with higher circulating GDF15 concentrations, an
association that has not previously been described.
GDF15 is also known as an NSAID-activated gene, and
experimental studies have shown NSAIDs to induce its
expression, a process that appears to be independent of
cyclo-oxygenase or prostanoids (20 ). Inflammatory conditions such as rheumatoid arthritis have been associated
with GDF15 increases (38 ), but the association of NSAID
use with GDF15 concentration persisted even after we excluded individuals with rheumatoid arthritis and adjusted
for CRP as a marker of inflammation.
Our study is the first to report the heritability of
GDF15 and suggests that circulating GDF15 concentrations are in part genetically determined. We conducted a metaanalysis of 2 community-based cohorts
and found 8 SNPs in the region of the GDF15 gene that
are associated with circulating concentrations on a
genomewide-significant basis. Two of these SNPs appear to be independent signals, according to our conditional analyses. A missense variant in the promoter
region in pairwise linkage disequilibrium with one of
our loci with genomewide significance (rs4808793 and
rs1054564, respectively; r 2 ⫽ 0.40) has previously been
associated with increased transcriptional activity and
higher circulating GDF15 concentrations, as well as
with favorable echocardiographic traits in hypertensive
Chinese individuals (39 ). Two of the SNPs with
genomewide significance in our analysis (rs888663 and
rs1054564) were previously associated with GDF15
concentration in a cohort of prostate cancer patients
(40 ), a result that also lends support to our findings.
Although we have shown that genetic factors play
an important role in circulating GDF15 concentrations, the exact mechanism by which the 2 potentially
independent signals modulate GDF15 expression remains unknown. We observed nominal evidence of
heterogeneity between the 2 studies in the effect of genetic variants, which could potentially be attributed to
clinical differences in age and cardiovascular risk factors. Alternatively, this pattern of heterogeneity in allelic effects and independent association signals could
reflect differences between the 2 populations with respect to underlying and unobserved rare causal variants. Further studies are needed to explore factors that
might explain this heterogeneity. Interestingly, 3
GDF15 variants with genomewide significance were
strongly associated with cis gene expression of PGPEP1
and LRRC25, genes within 100 kb of GDF15. There is
no known relationship between PGPEP1, LRRC25, and
GDF15. Because these genetic variants are intergenic,
they might upregulate expression via a promoter or
other distal elements that may affect GDF15 expression
itself.
One notable finding was that the C allele of one of
the GDF15 loci (rs1054561) was associated with both
higher GDF15 concentrations and lower HDL cholesterol concentrations. It is possible that genetically increased GDF15 concentrations could lead to lower
HDL, but further prospective studies are needed to explore this association.
Several limitations of our study deserve mention.
GDF15 concentrations in our sample of healthy participants were slightly higher than those reported by
Kempf and colleagues, who studied a sample of elderly
Swedish individuals. Several aspects of the Framingham cohort support the generalizability of our findings, including the community-based design and the
rigorous characterization of participants who have
been followed since the early 1970s. We cannot exclude
the possibility that laboratory variation, unmeasured
differences in the study cohorts, and/or undetected disease contributed to differences in the study findings.
The cross-sectional nature of our study limits inferences of causality with respect to clinical correlates;
thus, GDF15 may both contribute to and be a marker
of cardiometabolic risk. Future studies involving larger
sample sizes may be able to identify genetic variants
outside of the GDF15 locus that might play an important role in determining GDF15 concentrations. Lastly,
Clinical Chemistry 58:11 (2012) 1589
our sample consisted of white participants of European
ancestry, limiting the generalizability of our findings to
other racial and ethnic groups.
In summary, we have demonstrated that GDF15
is associated with cardiometabolic risk factors and
that genetic factors play an important role in determining GDF15 concentrations. Importantly, using 2
community-based cohorts, we identified genetic variants in the region of the GDF15 gene on chromosome
19p13.11 that influence circulating GDF15 concentrations. Two variants were associated with altered gene
expression in different blood cell lines, and one was
associated with lower HDL concentrations. Further
studies are required to elucidate how genetic factors
regulate GDF15 expression and how these mechanisms
relate ultimately to the development of CVD.
Author Contributions: All authors confirmed they have contributed to
the intellectual content of this paper and have met the following 3 requirements: (a) significant contributions to the conception and design,
acquisition of data, or analysis and interpretation of data; (b) drafting
or revising the article for intellectual content; and (c) final approval of
the published article.
Authors’ Disclosures or Potential Conflicts of Interest: Upon manuscript submission, all authors completed the author disclosure form.
Disclosures and/or potential conflicts of interest:
Employment or Leadership: None declared.
Consultant or Advisory Role: J. Januzzi, Roche and Critical
Diagnostics.
Stock Ownership: None declared.
Honoraria: T.J. Wang, Roche.
Research Funding: E. Ingelsson, Swedish Research Council, Swedish
Heart and Lung Foundation, Swedish Foundation for Strategic Research, Royal Swedish Academy of Sciences, Swedish Diabetes Foundation, Swedish Society of Medicine, and Novo Nordisk Fonden; J.
Januzzi, Roche, Siemens, BG Medicine, and Thermo Fisher Scientific; K.C. Wollert, GDF15 test kits from Roche Diagnostics; A.P.
Morris, Wellcome Trust (WT081682, WT064890, and WT090532).
Expert Testimony: None declared.
Patents: T. Kempf, European Patent No. 2 047 275 B1; K.C. Wollert,
European Patent No. 2 047 275 B1.
Role of Sponsor: The funding organizations played no role in the
design of study, choice of enrolled patients, review and interpretation
of data, or preparation or approval of manuscript.
References
1. Kempf T, Wollert KC. Growth-differentiation
factor-15 in heart failure. Heart Fail Clin 2009;5:
537– 47.
2. Bonaca MP, Morrow DA, Braunwald E, Cannon
CP, Jiang S, Breher S, et al. Growth differentiation
factor-15 and risk of recurrent events in patients
stabilized after acute coronary syndrome: observations from PROVE IT-TIMI 22. Arterioscler
Thromb Vasc Biol 2011;31:203–10.
3. Kempf T, Bjorklund E, Olofsson S, Lindahl B,
Allhoff T, Peter T, et al. Growth-differentiation
factor-15 improves risk stratification in STsegment elevation myocardial infarction. Eur
Heart J 2007;28:2858 – 65.
4. Khan SQ, Ng K, Dhillon O, Kelly D, Quinn P,
Squire IB, et al. Growth differentiation
factor-15 as a prognostic marker in patients
with acute myocardial infarction. Eur Heart J
2009;30:1057– 65.
5. Eggers KM, Kempf T, Lagerqvist B, Lindahl B,
Olofsson S, Jantzen F, et al. Growthdifferentiation factor-15 for long-term risk prediction in patients stabilized after an episode of
non-ST-segment-elevation acute coronary syndrome. Circ Cardiovasc Genet 2010;3:88 –96.
6. Wollert KC, Kempf T, Peter T, Olofsson S, James S,
Johnston N, et al. Prognostic value of growthdifferentiation factor-15 in patients with non-STelevation acute coronary syndrome. Circulation
2007;115:962–71.
7. Kempf T, von Haehling S, Peter T, Allhoff T,
Cicoira M, Doehner W, et al. Prognostic utility of
growth differentiation factor-15 in patients with
chronic heart failure. J Am Coll Cardiol 2007;50:
1054 – 60.
8. Anand IS, Kempf T, Rector TS, Tapken H, Allhoff
T, Jantzen F, et al. Serial measurement of growthdifferentiation factor-15 in heart failure: relation
to disease severity and prognosis in the Valsartan
1590 Clinical Chemistry 58:11 (2012)
9.
10.
11.
12.
13.
14.
15.
Heart Failure Trial. Circulation 2010;122:1387–
95.
Daniels LB, Clopton P, Laughlin GA, Maisel
AS, Barrett-Connor E. Growth-differentiation
factor-15 is a robust, independent predictor of
11-year mortality risk in community-dwelling
older adults: the Rancho Bernardo Study. Circulation 2011;123:2101–10.
Lind L, Wallentin L, Kempf T, Tapken H, Quint A,
Lindahl B, et al. Growth-differentiation factor-15
is an independent marker of cardiovascular dysfunction and disease in the elderly: results from
the Prospective Investigation of the Vasculature
in Uppsala Seniors (PIVUS) study. Eur Heart J
2009;30:2346 –53.
Rohatgi A, Patel P, Das SR, Ayers CR, Khera A,
Martinez-Rumayor A, et al. Association of growth
differentiation factor-15 with coronary atherosclerosis and mortality in a young, multiethnic
population: observations from the Dallas Heart
Study. Clin Chem 2012;58:172– 82.
Kannel WB, Feinleib M, McNamara PM, Garrison
RJ, Castelli WP. An investigation of coronary
heart disease in families. The Framingham Offspring Study. Am J Epidemiol 1979;110:281–90.
Levey AS, Bosch JP, Lewis JB, Greene T, Rogers N,
Roth D. A more accurate method to estimate
glomerular filtration rate from serum creatinine: a
new prediction equation. Modification of Diet in
Renal Disease Study Group. Ann Intern Med
1999;130:461–70.
Grundy SM, Cleeman JI, Daniels SR, Donato KA,
Eckel RH, Franklin BA, et al. Diagnosis and management of the metabolic syndrome: an American Heart Association/National Heart, Lung, and
Blood Institute scientific statement. Circulation
2005;112:2735–52.
Kempf T, Horn-Wichmann R, Brabant G, Peter T,
Allhoff T, Klein G, et al. Circulating concentra-
16.
17.
18.
19.
20.
21.
22.
23.
24.
tions of growth-differentiation factor 15 in apparently healthy elderly individuals and patients with
chronic heart failure as assessed by a new immunoradiometric sandwich assay. Clin Chem 2007;
53:284 –91.
Wang TJ, Gona P, Larson MG, Tofler GH, Levy D,
Newton-Cheh C, et al. Multiple biomarkers for
the prediction of first major cardiovascular events
and death. N Engl J Med 2006;355:2631–9.
Todd J, Freese B, Lu A, Held D, Morey J, Livingston R, Goix P. Ultrasensitive flow-based immunoassays using single-molecule counting. Clin Chem
2007;53:1990 –5.
Li Y, Abecasis GR. MACH 1.0: rapid haplotype
reconstruction and missing genotype inference.
Am J Hum Genet 2006:2290.
Howie BN, Donnelly P, Marchini J. A flexible and
accurate genotype imputation method for the
next generation of genome-wide association
studies. PLoS Genet 2009;5:e1000529.
Baek SJ, Wilson LC, Lee C-H, Eling TE. Dual
function of nonsteroidal anti-inflammatory drugs
(NSAIDs): inhibition of cyclooxygenase and induction of NSAID-activated gene. J Pharmacol Exp
Ther 2002;301:1126 –31.
D’Agostino RB Sr, Vasan RS, Pencina MJ, Wolf
PA, Cobain M, Massaro JM, Kannel WB. General
cardiovascular risk profile for use in primary care:
the Framingham Heart Study. Circulation 2008;
117:743–53.
Almasy L, Blangero J. Multipoint quantitativetrait linkage analysis in general pedigrees. Am J
Hum Genet 1998;62:1198 –211.
Teslovich TM, Musunuru K, Smith AV, Edmondson AC, Stylianou IM, Koseki M, et al. Biological,
clinical and population relevance of 95 loci for
blood lipids. Nature 2010;466:707–13.
Goring HH, Curran JE, Johnson MP, Dyer TD,
Charlesworth J, Cole SA, et al. Discovery of ex-
Clinical and Genetic Correlates of GDF15
25.
26.
27.
28.
29.
pression QTLs using large-scale transcriptional
profiling in human lymphocytes. Nat Genet 2007;
39:1208 –16.
Zeller T, Wild P, Szymczak S, Rotival M, Schillert
A, Castagne R, et al. Genetics and beyond—the
transcriptome of human monocytes and disease
susceptibility. PLoS One 2010;5:e10693.
Greenawalt DM, Dobrin R, Chudin E, Hatoum IJ,
Suver C, Beaulaurier J, et al. A survey of the
genetics of stomach, liver, and adipose gene
expression from a morbidly obese cohort. Genome Res 2011;21:1008 –16.
Schlittenhardt D, Schober A, Strelau J, Bonaterra
GA, Schmiedt W, Unsicker K, et al. Involvement
of growth differentiation factor-15/macrophage
inhibitory cytokine-1 (GDF-15/MIC-1) in oxLDLinduced apoptosis of human macrophages in
vitro and in arteriosclerotic lesions. Cell Tissue
Res 2004;318:325–33.
Kempf T, Eden M, Strelau J, Naguib M, Willenbockel
C, Tongers J, et al. The transforming growth
factor-␤ superfamily member growth-differentiation
factor-15 protects the heart from ischemia/reperfusion injury. Circ Res 2006;98:351– 60.
Kempf T, Zarbock A, Widera C, Butz S, Stadtmann
A, Rossaint J, et al. GDF-15 is an inhibitor of
leukocyte integrin activation required for survival
30.
31.
32.
33.
34.
35.
after myocardial infarction in mice. Nat Med
2011;17:581– 8.
Xu J, Kimball TR, Lorenz JN, Brown DA, Bauskin
AR, Klevitsky R, et al. GDF15/MIC-1 functions as
a protective and antihypertrophic factor released
from the myocardium in association with SMAD
protein activation. Circ Res 2006;98:342–50.
Ho JE, Wang TJ. Growth differentiation factor 15:
a canary in a coal mine? Clin Chem 2012;58:3–5.
Bauskin AR, Brown DA, Kuffner T, Johnen H, Luo
XW, Hunter M, Breit SN. Role of macrophage
inhibitory cytokine-1 in tumorigenesis and diagnosis of cancer. Cancer Res 2006;66:4983– 6.
Eggers KM, Kempf T, Allhoff T, Lindahl B, Wallentin
L, Wollert KC. Growth-differentiation factor-15 for
early risk stratification in patients with acute chest
pain. Eur Heart J 2008;29:2327–35.
Kempf T, Sinning J-M, Quint A, Bickel C, Sinning
C, Wild PS, et al. Growth-differentiation factor-15
for risk stratification in patients with stable and
unstable coronary heart disease: results from the
AtheroGene study. Circ Cardiovasc Genet 2009;
2:286 –92.
Wollert KC, Kempf T, Lagerqvist B, Lindahl B,
Olofsson S, Allhoff T, et al. Growth differentiation
factor 15 for risk stratification and selection of an
invasive treatment strategy in non ST-elevation
36.
37.
38.
39.
40.
acute coronary syndrome. Circulation 2007;116:
1540 – 8.
Ding Q, Mracek T, Gonzalez-Muniesa P, Kos K,
Wilding J, Trayhurn P, Bing C. Identification of
macrophage inhibitory cytokine-1 in adipose tissue and its secretion as an adipokine by human
adipocytes. Endocrinology 2009;150:1688 –96.
Vila G, Riedl M, Anderwald C, Resl M, Handisurya
A, Clodi M, et al. The relationship between insulin resistance and the cardiovascular biomarker
growth differentiation factor-15 in obese patients. Clin Chem 2011;57:309 –16.
Brown DA, Moore J, Johnen H, Smeets TJ,
Bauskin AR, Kuffner T, et al. Serum macrophage
inhibitory cytokine 1 in rheumatoid arthritis: a
potential marker of erosive joint destruction. Arthritis Rheum 2007;56:753– 64.
Wang X, Yang X, Sun K, Chen J, Song X, Wang H,
et al. The haplotype of the growth-differentiation
factor 15 gene is associated with left ventricular
hypertrophy in human essential hypertension.
Clin Sci (Lond) 2010;118:137– 45.
Brown DA, Lindmark F, Stattin P, Balter K, Adami
HO, Zheng SL, et al. Macrophage inhibitory cytokine 1: a new prognostic marker in prostate
cancer. Clin Cancer Res 2009;15:6658 – 64.
Clinical Chemistry 58:11 (2012) 1591