Supplementary Information (doc 495K)

advertisement
1
Supplemental material
2
Gene by stress genome-wide interaction analysis and path
3
analysis identify EBF1 as a cardiovascular and metabolic
4
risk gene
5
Abanish Singh1,2,4, *, Michael A. Babyak1,4, Daniel K. Nolan2, Beverly H. Brummett1,4, Rong
6
Jiang1,4 , Ilene C. Siegler1,4, William E. Kraus2,3,6, Svati H. Shah2,3, Redford B. Williams1,4,
7
Elizabeth R. Hauser2,3,5
8
1
Behavioral Medicine Research Center, Duke University Medical Center, Durham, NC
9
2
Center for Human Genetics, Duke University Medical Center, Durham, NC
10
3
Department of Medicine, Duke University Medical Center, Durham, NC
11
4
Department of Psychiatry and Behavioral Sciences, Duke University Medical Center,
12
Durham, NC
13
5
14
Affairs Medical Center, Durham, NC, USA
15
6
16
*Correspondence:
17
Abanish Singh, Ph.D.
18
Duke University Medical Center, Duke Box 3445
19
905 LaSalle St. Durham, NC 27710
20
Tel: +1 919 681 2133 Fax: +1 919 684 0934
21
Email: abanish.singh@duke.edu
Durham Epidemiologic Research and Information Center, Durham Veterans
Duke Center for Living, Duke University Medical Center, Durham, NC
1
22
Study Populations and Phenotypes
23
Discovery Dataset: We obtained the publicly available Multi-Ethnic Study of Atherosclerosis
24
(MESA) Cohort GWAS dataset from dbGaP under the standard data user agreement and Duke
25
IRB approval. The MESA GWAS data set includes a total of 8,227 samples (3,790 males, 4,437
26
females), including White, Chinese American, Black, and Hispanic ancestries. We chose the
27
MESA dataset for testing the GxE interaction hypothesis based on the availability of a well-
28
characterized phenotype psychosocial stress chronic burden (i.e. chronic psychosocial stress) that
29
was quantified on an ordinal scale of 0 to 5 (0, 1, 2, 3, 4, and 5) based on questionnaires in five
30
domains including questions about ongoing serious health problems, serious health problems
31
with someone close, work related problems, financial strains, and difficulties in relationships.
32
The severity of stress was rated based on how stressful (not very, moderately or very stressful)
33
each problem was perceived to be by the participant, and classified as chronic for each of the
34
five domains if the problem was experienced for at least 6 months.
35
We were interested in the relationship between genes and stress in modifying CVD risk
36
factors and common carotid intimal-media thickness, a surrogate marker for atherosclerosis that
37
is a strong predictor of future vascular events. Thus, we included measures of central obesity,
38
hip circumference (HIPCM), body mass index (BMI), waist circumference (WAISTCM), along
39
with fasting glucose (GLUC), triglycerides (TRIG), low-density lipoprotein (LDL), high-density
40
lipoprotein (HDL), systolic blood pressure (SYSTBP), diastolic blood pressure (DIASTBP), type
41
II diabetes status (DIABSTAT), common carotid intimal-media thickness (CCIMT), and chronic
42
psychosocial stress (STRESS). Table 1 provides the summary of these traits in the MESA exam
43
1 data. A total of 5,805 individuals had both phenotype and quality controlled genotype data
44
available, comprised of 2,460 Whites (1,174 Male, 1,286 Female), 548 Chinese Americans (359
2
45
Male, 189 Female), 1,547 Blacks (721 Male, 826 Female), and 1,250 Hispanics (630 Male, 620
46
Female). For all the selected traits we checked for departure from normality using STATA SE
47
11.1 and performed a transformation to achieve approximate normality within each ancestry.
48
For Whites, we performed log transformation for waist circumference, triglyceride, HDL,
49
systolic and diastolic blood pressures; square root inverse transformation for BMI; inverse
50
transformation for glucose; and square root transformation for LDL. Hip circumference and
51
common carotid intimal-media thickness did not require transformation. Also in the GxE
52
GWAS for hip circumference, we performed an inverse transformation in Chinese Americans,
53
and square inverse transformation in Blacks and Hispanics.
54
Replication Dataset: We obtained the publicly available Framingham Cohort dataset under the
55
standard data user agreement and Duke IRB approval and used Generation 2 (or Offspring)
56
dataset for our replication analysis. The original Framingham Study started in 1948 with the
57
recruitment of 5,209 men and women between the ages of 30 and 62 from the town of
58
Framingham, Massachusetts. In 1971, 5,124 adult children and their spouses were enrolled in the
59
Offspring cohort in Framingham Study. The cohort is primarily white including 2,483 males and
60
2,641females. A total of 3,157 individuals had both phenotype and quality controlled genotype
61
data available, comprised of 1,515 males and 1,642 females.
62
We used selected replication phenotypes hip girth (HIPGIRTH), waist girth
63
(WAISTGIRTH) from exam 4, average of body mass index (BMI), glucose (GLUC), diabetes
64
status (DIABSTAT ) from exam 3 and 4 (hip girth and waist girth was not available in exam 3),
65
and common carotid IMT (CCIMT) from exam 6. We used a computed ordinal measure of
66
chronic psychosocial stress (STRESS) as the sum of indicators of financial strain, marital
67
problems, work related difficulties, and health problems of someone close from the exam 3
3
68
psychosocial data, employing a generalized stress calibration computational method (Singh et
69
al., unpublished). We did not use self health problem and disease status (such as CVD and
70
diabetes) to infer corresponding stress indicator as it would have been confounding in the
71
interaction analysis, unlike in MESA where it was based on a questionnaire. The stress measure
72
ranged from 0-3. Table S1 in Supplemental Material provides the summary of these variables
73
along with age and sex. We checked for departure from normality using STATA SE 11.1 and
74
performed a transformation to achieve approximate normality: log transformation for hip girth
75
and BMI, square root transformation for waist girth, cube inverse transformation for glucose and
76
inverse transformation for common carotid IMT. To adjust analysis for the family relatedness,
77
we implemented robust Huber-White familial clustering using family ID.
78
SNP Data and Quality Control Procedure
79
Discovery Dataset: We used the SNP data from the MESA SNP Health Association Resource
80
(SHARe). Funding for SHARe genotyping was provided by NHLBI Contract N02-HL-64278.
81
The genotyping was performed at Affymetrix (Santa Clara, California, USA) and the Broad
82
Institute of Harvard and MIT (Boston, Massachusetts, USA) using the Affymetrix Genome-Wide
83
Human SNP Array 6.0. The total number of SNPs was 909,622. We performed standard quality
84
control procedures as implemented in PLINKS1 v1.07. Our quality control criteria were:
85
minimum minor allele frequency (MAF) of 0.01 (1%), maximum missing genotypes of 0.05
86
(5%) per individual, maximum missing individuals per SNP of 0.05 (5%), and significant
87
departure from Hardy-Weinberg equilibrium (HWE) (P<0.00001). The 11,748 SNPs failed
88
frequency test, 6,755 SNPs failed missing individual SNPs, and 365,902 SNPs were excluded
89
based on HWE test. The final number of SNPs used in the GxE GWAS was 528,298. We
90
estimated principal components (PCs) based on 162,072 LD-pruned SNPs in each ethnic group
4
91
using EIGENSTRATS2 and used five PCs in Whites, five PCs in Chinese Americans, three PCs
92
in Blacks, and three PCs in Hispanics based on eigenvalue scree plots.
93
Replication Dataset: We used SNPs from the Framingham SNP Health Association Resource
94
(SHARe). Funding for Framingham SHARe Affymetrix genotyping was provided by NHLBI
95
Contract N02-HL-64278. SHARe Illumina genotyping was provided under an agreement
96
between Illumina and Boston University using Affymetrix Mapping250K (Nsp and Sty) Arrays
97
and Mapping50K (Hind240 and Xba240) Arrays. Although we used only a couple of selected
98
SNPs in our analysis, we performed similar quality control as described above before selecting
99
SNPs. There were a total of 549,782 SNPs, 20,825 SNPs were excluded based on HWE test,
100
37,756 SNPs failed missing individual test, and 66,965 SNPs failed frequency test. We estimated
101
principal components (PCs) based on 95,650 LD-pruned SNPs using EIGENSTRATS2 and used
102
five PCs based on eigenvalue scree plot.
103
Structural equation path analysis
104
We were interested in the correlation structure among the related phenotypes when assuming a
105
particular model of interaction among chronic stress and genetic variation. In addition to
106
comparing genotypes on the magnitude of zero-order correlations among the variables under
107
study, we expanded our analysis of the correlations for MESA White and Framingham Offspring
108
Cohorts in Tables S9 and S10 by using them to estimate a path model. Path modelling provides
109
several pieces of information over and above the raw correlations. We used a structural equation
110
path analysis as implemented in MplusS3 version 6.11 to evaluate possible mediated causal
111
pathsS4 from stress through risk factors to CCIMT. Path analysis uses a series of simultaneous
112
equations to estimate possible mediating paths. For example, if it is hypothesized that the
113
association between x and y is mediated by the variable m, the mediated or indirect effect of x on
5
114
y can be estimated by the simultaneous equations y = b1*x + b2*m and m = b3*x, where b
115
represents a regression slope. The indirect effect can be represented graphically as: x → m → y,
116
where the directed (single-headed) arrows represent the regression slope between two variables
117
involved. The magnitude of a mediating effect is then calculated by taking the product of all
118
path coefficients along a given proposed path. In the example above, the indirect effect would be
119
the product of b2 and b3. The standard error for the product is then used to test the null
120
hypothesis that the product is zero.
121
For our analysis, the model can be expressed as three simultaneous equations: CCIMT =
122
g1*Glucose + g2*Hip Circumference + g3*Stress + g4*Age + g5*Male; Glucose=g6*Hip
123
Circumference + g7*Stress + g8*Age + g9*Male; and Hip Circumference =g10*Stress +
124
g11*Age + g12*Male. In each equation the term on the left is equivalent to the dependent
125
variable in a regression-type model, while the terms of the right are the predictor variables. The
126
terms g1 through g12 represent the association of the corresponding predictor with the dependent
127
variable, adjusted for all other variables in a given equation. For example, g1 is the slope
128
coefficient of the association between Stress and CCIMT, adjusted for Glucose, Hip
129
Circumference, Age, and Sex. (Stress, Hip Circumference, Glucose, and CCIMT were adjusted
130
for ancestry prior to analysis). The three path model equations also imply three mediated effects
131
from stress to CCIMT: stress chronic burden ⇒ hip circumference ⇒ CCIMT; stress chronic
132
burden ⇒ glucose ⇒ CCIMT; and stress chronic burden ⇒ hip circumference ⇒ glucose ⇒
133
CCIMT. The model was estimated simultaneously for the TT and CT/CC groups using 1000
134
bootstrap samples to generate standard errors.
135
6
136
137
Figures
138
Figure S1: Q-Q plot of GxE GWAS on Hip Circumference (HIPCM) in MESA Whites where
139
the environmental factor (E) was chronic psychosocial stress (STRESS). This plot demonstrated
140
a departure of the observed P-values from the assumed distribution (inflation factor 1.1524),
141
therefore, we performed genomic control as shown in Figure S2.
142
143
7
144
145
Figure S2: Genomic control Q-Q plot of GxE GWAS on Hip Circumference (HIPCM) in MESA
146
Whites where the environmental factor (E) was chronic psychosocial stress
147
(STRESS).
148
149
8
150
151
Figure S3: Manhattan plot of GxE GWAS on Hip Circumference (HIPCM) in MESA Chinese-
152
Americans where the environmental factor (E) was chronic psychosocial stress (STRESS).
153
154
9
155
156
Figure S4: Manhattan plot of GxE GWAS on Hip Circumference (HIPCM) in MESA Blacks
157
where the environmental factor (E) was chronic psychosocial stress (STRESS).
158
159
10
160
161
Figure S5: Manhattan plot of GxE GWAS on Hip Circumference (HIPCM) in MESA Hispanics
162
where the environmental factor (E) was chronic psychosocial stress (STRESS).
163
164
11
165
166
Figure S6: LocusZoomS5 plot of chromosome 5 (156mb-160mb) region around the EBF1 gene
167
which includes the five most significant EBF1 SNPs of MESA Whites GxE GWAS.
168
169
170
12
171
172
Figure S7: The plot shows age (years) group-wise mean and SE bars of raw values of CCIMT
173
(mm) in Framingham Offspring cohort and MESA White in panel A, and number of individuals
174
in each age group for the two datasets in panel B.
175
176
13
177
Figure S8: Mean hip circumference stratified by chronic stress and rs4704963:T>C carrier status
178
in the four ethnic groups of the MESA dataset for samples having QC genotype and phenotype
179
data.
180
14
181
Tables
182
Table S1: Summary of phenotypic traits in Framingham Offspring for the samples having
183
quality-controlled genotype data.
Trait
Mean (SD)
Min – Max
AGE (Years)
47.76 (9.89)
18 – 77
SEX (M / F)
47.99% / 52.01%
Binary (1/2)
HIPGIRTH (Inches)
40.03 (3.84)
30 – 64
WAISTGIRTH (Inches)
35.22 (5.71)
22 – 57
BMI (kg/m^2)
26.49 (4.63)
15 – 60.37
GLUC (mg/dl)
94.55 (20.63)
64 – 306
DIABSTAT (N/D*)
87.76% / 12.24%
Binary (0/1)
CCIMT (mm)
0.59 (0.18)
0.29 –3.23
STRESS (Score)
0.84 (0.79)
0–3
*Normal/Disease
184
15
185
186
Table S2: The GxE P-values (P-val) and MAF of ten SNPs with the lowest P-values (top hits)
187
from MESA Whites GxE GWAS on Hip Circumference (HIPCM). The Table also shows GxE
188
P-values and MAF of these ten SNPs in other ancestries – i.e. Chinese American, Blacks, and
189
Hispanics. The environment (E) component in this GWAS was chronic psychosocial stress
190
(STRESS). SNPs significant at Bonferroni correction threshold are shown in bold. All genomic
191
locations were numbered based on Genome Reference Consortium Human Build 37 patch
192
release 10 (GRCh37.p10).
rs4704963
Gene
Type
POS
AND
ALLELE
CHANGE
CHR
SNP
193
Whites
P-val
Chinese
MAF P-val MAF
Blacks
Hispanics
P-val
MAF
P-val MAF
5 g.158247378T>C Intronic
EBF1
7.14E-09 0.07
0.66
0.07
0.87
0.02
0.87
0.05
rs17056278 5 g.158252438C>G Intronic
EBF1
7.14E-09 0.07
0.66
0.07
0.87
0.02
0.74
0.05
rs17056298 5 g.158268679C>G Intronic
EBF1
1.30E-08 0.06
0.52
0.06
0.90
0.02
0.51
0.05
rs10077799 5 g.158272083T>C Intronic
EBF1
1.71E-08 0.06
0.51
0.06
0.65
0.05
0.66
0.06
rs17056318 5 g.158285953T>C Intronic
EBF1
2.33E-08 0.07
0.52
0.05
0.53
0.02
0.91
0.06
1.15E-07 0.48
0.61
0.48
0.15
0.29
0.26
0.36
-
0.001 0.051
0.01
0.36
0.04
rs686708
6 g.117091683T>C Upstream GPRC6A
rs1536274
14 g.67824636C>T
Intronic
ATP6V1D 3.37E-07 0.07
rs10872796 6 g.104685134T>G Intergenic N/A
4.93E-07 0.28
0.17
0.01
0.22
0.21
0.88
0.29
rs10092697 8 g.57960820C>T
8.71E-07 0.34
0.43
0.37
0.96
0.46
0.60
0.33
1.13E-06 0.29
0.45
0.29
0.10
0.26
0.12
0.31
rs7763618
Intergenic N/A
6 g.148621930A>C Intronic
SASH1
194
16
195
196
Table S3: Ten SNPs with the lowest P-values (top hits) from GxE GWAS on Hip
197
Circumference (HIPCM) in MESA Chinese Americans. The environment (E) component in this
198
GWAS was chronic psychosocial stress (STRESS). All genomic locations in this and
199
subsequent Talbles were numbered based on Genome Reference Consortium Human Build 37
200
patch release 10 (GRCh37.p10).
SNP
CHR
POS & ALLELE
Type
Gene
MAF
P-val
rs907349
12
g.51789449C>T
Intronic
SLC4A8
0.28
2.50E-05
rs2180086
14
g.94444787A>G
5Upstream
ASB2
0.013
4.06E-05
rs3749107
2
g.109539774G>A
Intronic
EDAR
0.012
5.09E-05
rs4905125
14
g.94442455C>T
Intronic
ASB2
0.014
7.73E-05
rs4939458
11
g.60564804A>G
Intronic
MS4A10
0.144
1.07E-04
rs10897113
11
g.60556218G>A
Intronic
MS4A10
0.141
1.13E-04
rs10775622
19
g.17230199C>T
Intronic
MYO9B
0.30
1.46E-04
rs654680
11
g.63361442G>C
Intronic
PLA2G16
0.49
1.51E-04
rs13158501
5
g.15261054G>A
Non-cod intronic
RP1-137K24.1
0.043
1.63E-04
rs1790361
11
g.71118568A>C
Non- cod intronic
AP002387.1
0.16
1.64E-04
201
17
202
203
Table S4: Ten SNPs with the lowest P-values (top hits) from GxE GWAS on Hip
204
Circumference (HIPCM) in MESA Blacks. The environment (E) component in this GWAS was
205
chronic psychosocial stress (STRESS).
206
SNP
CHR
POS & ALLELE
Type
Gene
MAF
P-val
rs4427965
2
g.214802528C>T
Intronic
SPAG16
0.20
1.12E-05
rs12443351
15
g.98673011G>A
Intergenic
N/A
0.064
1.16E-05
rs3748358
14
g.21789794G>A
Intronic
RPGRIP1
0.22
2.21E-05
rs12265226
10
g.25670435G>A
Intronic
GPR158
0.19
2.51E-05
rs6052830
20
g.4729789G>C
Intergenic
N/A
0.25
2.58E-05
rs12867694
13
g.43902183C>T
Intronic
ENOX1
0.014
2.82E-05
rs6037950
20
g.4731022C>T
Intergenic
N/A
0.24
2.82E-05
rs17071864
3
g.64939606T>C
Non-cod intronic
CTD-2230D16.1
0.12
3.03E-05
rs12061877
1
g.147264252T>C
Intergenic
N/A
0.12
3.20E-05
rs2022211
2
g.36801857T>C
Intronic
FEZ2
0.40
3.44E-05
207
18
208
209
Table S5: Ten SNPs with the lowest P-values (top hits) from GxE GWAS on Hip
210
Circumference (HIPCM) in MESA Hispanics. The environment (E) component in this GWAS
211
was chronic psychosocial stress (STRESS).
212
213
SNP
CHR
POS & ALLELE
Type
Gene
MAF
P-val
rs7094310
10
g.6791034C>G
Intergenic
N/A
0.036
1.68E-06
rs11254430
10
g.6795454G>A
Intergenic
N/A
0.036
2.49E-06
rs564309
1
g.228585562C>A
Intronic
TRIM11
0.064
4.32E-06
rs7863100
9
g.18640590T>C
Intronic
ADAMTSL1
0.22
8.44E-06
rs13115750
4
g.182480575G>C
Intergenic
N/A
0.28
8.71E-06
rs2571342
12
g.94317979C>T
Non-cod intronic
RP11-1060G2.1
0.33
1.43E-05
rs2827258
21
g.23521786A>G
Intergenic
N/A
0.075
1.78E-05
rs17084244
18
g.69017186A>G
Intergenic
N/A
0.028
1.90E-05
rs2888734
12
g.103846180G>T
Intronic
C12orf42
0.14
2.27E-05
rs17281463
20
g.40278357G>C
Intergenic
N/A
0.020
2.58E-05
214
19
215
216
Table S6 : The SNP term (G) and gene-by-stress interaction term (GxE) terms P-values of the 5
217
most significant GxE GWAS SNPs (Table 3) in MESA Whites hip circumference (HIPCM),
218
waist circumference (WAISTCM), body mass index (BMI), fasting glucose (GLU), type II
219
diabetes status (DIABSTAT), common carotid intimal-media thickness (CCIMT), high-density
220
lipoprotein (HDL), systolic blood pressure (SYSTBP), diastolic blood pressure (DIASTBP),
221
low-density lipoprotein (LDL), and triglycerides (TRIG).
SNP→
PHENO↓
rs4704963:T>C
G / GxE P-val
rs17056278:C>G
G / GxE P-val
rs17056298:C>G
G / GxE P-val
rs10077799:T>C
G / GxE P-val
rs17056318:T>C
G / GxE P-val
HIPCM
0.0154 / 7.14E-09
0.0154 / 7.14E-09
0.01907 / 1.30E-08
0.02305 / 1.72E-08
0.0318 / 2.33E-08
WAISTCM
0.1853 / 3.47E-05
0.1853 / 3.47E-05
0.2332 / 6.00E-05
0.2658 / 7.52E-05
0.3357 / 9.67E-05
BMI
0.398 / 0.0007209
0.398 / 0.0007209
0.4618 / 0.000936
0.5348 / 0.001239
0.6487/ 0.001658
GLU
0.06293 / 0.00015
0.0629 / 0.00015
0.04528 / 7.32E-05
0.1132 / 0.000244
0.1012/ 0.00114
DIABSTAT
0.941 / 0.01723
0.941 / 0.01723
0.777 / 0.009345
0.9641 / 0.01706
0.9173 / 0.03232
CCIMT
0.7167 / 0.2518
0.7167 / 0.2518
0.9524 / 0.1704
0.9638 / 0.173
0.9228 / 0.222
HDL
0.626 / 0.14
0.626 / 0.14
0.4397 / 0.1061
0.4807 / 0.1168
0.4696 / 0.2769
SYSTBP
0.7522 / 0.2128
0.7522 / 0.2128
0.5462 / 0.1164
0.6196 / 0.1351
0.4142 / 0.07616
DIASTBP
0.6969 / 0.4003
0.6969 / 0.4003
0.5799 / 0.3279
0.6096 / 0.3432
0.5609 / 0.5064
LDL
0.1057 / 0.348
0.1057 / 0.348
0.1825 / 0.5857
0.2046 / 0.619
0.2482 / 0.7256
TRIG
0.6309 / 0.6295
0.6309 / 0.6295
0.5398 / 0.6925
0.429 / 0.7873
0.5734 / 0.8064
222
20
223
224
Table S7: SNP term (G), chronic stress term (E) and gene-by-stress interaction term (GxE) beta
225
(β) and P-values for GxE GWAS lead SNP rs4704963:T>C in MESA Whites phenotypes: waist
226
circumference (WAISTCM), body mass index (BMI), fasting glucose (GLU), type II diabetes
227
status (DIABSTAT), common carotid intimal-media thickness (CCIMT), high-density
228
lipoprotein (HDL), systolic blood pressure (SYSTBP), diastolic blood pressure (DIASTBP),
229
low-density lipoprotein (LDL), and triglycerides (TRIG).
PHENO
Transformation
Beta(G)
P-val(G)
Beta(E)
P-val(E)
Beta(GxE)
P-val(GxE)
-2.06
-1.50E-02
1.54E-02
1.85E-01
4.31E-01
9.19E-03
2.58E-02
6.24E-04
2.98
2.96E-02
7.14E-09
3.47E-05
BMI
Square Root Inverse 1.15E-03
3.98E-01
-1.06E-03
6.16E-04
-2.79E-03
7.21E-04
GLU
Inverse
2.45E-04
6.29E-02
-5.10E-05
8.81E-02
-3.02E-04
1.50E-04
DIABSTAT Identity
9.74E-01
9.41E-01
1.34
2.30E-04
1.48
1.72E-02
CCIMT
Identity
5.19E-03
7.17E-01
4.57E-03
1.59E-01
9.89E-03
2.52E-01
HDL
Logarithmic
1.04E-02
6.26E-01
-4.66E-03
3.35E-01
-1.90E-02
1.40E-01
SYSTBP
Logarithmic
-3.89E-03
7.52E-01
4.52E-04
8.72E-01
9.28E-03
2.13E-01
DIASTBP
Logarithmic
-4.35E-03
6.97E-01
-5.95E-04
8.15E-01
5.68E-03
4.00E-01
LDL
Square Root
-1.90E-01
1.06E-01
-4.40E-02
1.01E-01
6.67E-02
3.48E-01
TRIG
Logarithmic
2.12E-02
6.31E-01
2.84E-02
4.60E-03
1.29E-02
6.30E-01
HIPCM
Identity
WAISTCM Logarithmic
230
21
231
232
Table S8: Replication analysis in the Framingham Offspring Cohort: The Table includes P-
233
values of the SNP term (G P-val) and gene-by-stress interaction term (GxE P-val) in the
234
interaction model on analysis for six phenotypes under study ( Figure 1B) from Framingham
235
Offspring Cohort, i.e. Hip Girth, Waist Girth, body mass index (BMI), fasting glucose (Gluco),
236
type II diabetes status (Diab Status), and common carotid intimal-media thickness (CCIMT), and
237
three of the five most significant MESA GxE GWAS (Table 3) SNPs. The Table also includes
238
P-values of the SNP-only Model (SNP P-val) i.e. the conventional model with no interaction
239
term for these SNPs.
rs17056278:C>G
Phenotype
Transformation
rs17056298:C>G
rs17056318T>C
SNP
G
GxE
SNP
G
GxE
SNP
G
GxE
P-val
P-val
P-val
P-val
P-val
P-val
P-val
P-val
P-val
HipGirth
Logarithmic
0.308
0.097
0.007
0.230
0.169
0.012
0.152
0.215
0.008
WaistGirth
Square Root
0.099
0.272
0.001
0.063
0.406
0.002
0.034
0.502
0.001
BMI
Logarithmic
0.152
0.354
0.011
0.148
0.322
0.008
0.111
0.315
0.003
Gluco
Cube Inverse
0.887
0.522
0.343
0.740
0.597
0.335
0.507
0.752
0.312
Diab Status
Identity
0.351
0.442
0.041
0.318
0.467
0.039
0.261
0.522
0.039
CCIMT
Inverse
0.571
0.603
0.960
0.475
0.624
0.887
0.362
0.655
0.647
240
241
22
242
243
Table S9: Comparison of correlations between the study variables from Figure 1B pathway
244
model for TT and CT/CC genotype groups of EBF1 SNP rs4704963:T>C using MESA Whites
245
data. aP-value for null hypotheses that rho=0, bP-value calculated using Fisher’s z-
246
transformation for null hypothesis rho1-rho2=0.
TT Group
CT/CC Group
Difference
Var 1
Var 2
Rho1
a
P-value
Rho2
a
P-value
b
P-value
STRESS
HIPCM
0.06
0.003
0.27
<.0001
<.0001
STRESS
WAISTCM
0.02
0.383
0.20
<.0001
0.002
STRESS
BMI
0.06
0.004
0.20
<.0001
0.019
STRESS
GLUC
-0.03
0.172
0.08
0.153
0.068
STRESS
DIABSTAT
0.06
0.009
0.12
0.038
0.323
STRESS
CCIMT
-0.08
0.0004
-0.05
0.415
0.609
HIPCM
BMI
0.85
<.0001
0.87
<.0001
0.180
HIPCM
GLUC
0.24
<.0001
0.20
<.0001
0.515
HIPCM
DIABSTAT
0.14
<.0001
0.20
0.001
0.352
HIPCM
CCIMT
0.12
<.0001
0.04
0.447
0.181
WAISTCM
HIPCM
0.77
<.0001
0.82
<.0001
0.059
WAISTCM
BMI
0.86
<.0001
0.87
<.0001
0.479
WAISTCM
GLUC
0.39
<.0001
0.36
<.0001
0.565
WAISTCM
DIABSTAT
0.21
<.0001
0.25
<.0001
0.517
WAISTCM
CCIMT
0.24
<.0001
0.16
0.005
0.133
BMI
GLUC
0.34
<.0001
0.29
<.0001
0.389
BMI
DIABSTAT
0.19
<.0001
0.24
<.0001
0.388
BMI
CCIMT
0.18
<.0001
0.10
0.070
0.174
GLUC
DIABSTAT
0.36
<.0001
0.45
<.0001
0.091
GLUC
CCIMT
0.21
<.0001
0.30
<.0001
0.095
DIABSTAT
CCIMT
0.13
<.0001
0.23
<.0001
0.091
247
248
23
249
250
Table S10: Comparison of correlations between the study variables from Figure 1B pathway
251
model for CC and GC/GG genotype groups of EBF1 SNP rs17056278:C>G using Framingham
252
Offspring data. aP-value for null hypotheses that rho=0, bP-value calculated using difference in
253
likelihoods under robust maximum likelihood (MLR) estimation, where null hypothesis is rho1-
254
rho2=0. Difference tests are adjusted for family clustering.
CC Group
GC/GG Group
Difference
Var 1
Var 2
Rho1
a
P-value
Rho2
a
P-value
b
P-value
STRESS
HIPCM
-0.01
0.582
0.10
0.047
0.028
STRESS
WAISTCM
-0.02
0.382
0.09
0.066
0.034
STRESS
BMI
-0.01
0.53
0.07
0.133
0.093
STRESS
GLUC
-0.04
0.034
0.01
0.885
0.327
STRESS
DIABSTAT
-0.06
0.001
0.07
0.115
0.030
STRESS
CCIMT
0.02
0.252
-0.02
0.67
0.388
HIPCM
BMI
0.83
<.0001
0.84
<.0001
0.599
HIPCM
GLUC
0.29
<.0001
0.19
<.0001
0.069
HIPCM
DIABSTAT
0.21
<.0001
0.17
<.0001
0.612
HIPCM
CCIMT
0.13
<.0001
0.13
0.01
0.845
WAISTCM
HIPCM
0.74
<.0001
0.74
<.0001
0.708
WAISTCM
BMI
0.84
<.0001
0.84
<.0001
0.689
WAISTCM
GLUC
0.44
<.0001
0.35
<.0001
0.076
WAISTCM
DIABSTAT
0.29
<.0001
0.28
<.0001
0.922
WAISTCM
CCIMT
0.25
<.0001
0.26
<.0001
0.776
BMI
GLUC
0.39
<.0001
0.33
<.0001
0.203
BMI
DIABSTAT
0.27
<.0001
0.26
<.0001
0.849
BMI
CCIMT
0.21
<.0001
0.23
<.0001
0.687
GLUC
DIABSTAT
0.44
<.0001
0.40
<.0001
0.688
GLUC
CCIMT
0.26
<.0001
0.13
0.018
0.029
DIABSTAT
CCIMT
0.15
<.0001
0.08
0.187
0.320
255
24
256
257
REFERENCE:
258
S1 Purcell S, Neale B, Todd-Brown K et al.: PLINK: a toolset for whole-genome association
259
and population-based linkage analysis. American Journal of Human Genetics 2007; 81:
260
560–575.
261
S2 Price AL, Patterson NJ, Plenge RM, Weinblatt ME, Shadick NA Reich D: Principal
262
components analysis corrects for stratification in genome-wide association studies. Nature
263
Genetics 2006; 38: 904–909.
264
265
266
267
268
S3 Muthen LK Muthen BO: Mplus User’s Guide. Sixth Edition., Muthen & Muthen, Los
Angeles, CA, 1998-2010.
S4 Streiner DL: Finding our way: an introduction to path analysis. Can J Psychiatry 2005; 50:
115–122.
S5 Pruim RJ*, Welch RP*, Sanna S, Teslovich TM, Chines PS, Gliedt TP, Boehnke M,
269
Abecasis GR, Willer CJ. (2010) LocusZoom: Regional visualization of genome-wide
270
association scan results. Bioinformatics 2010 September 15; 26(18): 2336.2337.
25
Download