Content uploaded by Donald W Bowden
Author content
All content in this area was uploaded by Donald W Bowden on Oct 14, 2014
Content may be subject to copyright.
Heritability of Body Composition Measured by
DXA in the Diabetes Heart Study
Fang-Chi Hsu,* Leon Lenchik,† Barbara J. Nicklas,‡ Kurt Lohman,* Thomas C. Register,§
Josyf Mychaleckyj,‡ Carl D. Langefeld,* Barry I. Freedman,¶ Donald W. Bowden,‡¶** and J. Jeffrey Carr*†
Abstract
HSU, FANG-CHI, LEON LENCHIK, BARBARA J.
NICKLAS, KURT LOHMAN, THOMAS C. REGISTER,
JOSYF MYCHALECKYJ, CARL D. LANGEFELD,
BARRY I. FREEDMAN, DONALD W. BOWDEN, AND
J. JEFFREY CARR. Heritability of body composition
measured by DXA in the Diabetes Heart Study. Obes Res.
2005;13:312–319.
Objective: The purpose of this study was to investigate the
heritability of body composition measured by DXA in the
Diabetes Heart Study (DHS).
Research Methods and Procedures: Participants were 292
women and 262 men (age, 38 to 86 years; BMI, 17 to 57
kg/m
2
) from 244 families. There were 492 white and 49
African-American sibling pairs. DXA measurements of per-
centage fat mass (FM), whole body FM, and lean mass
(LM), as well as regional measurements of trunk fat mass
(TFM) and appendicular lean mass (ALM), were obtained.
Heritability of FM, LM, and BMI were estimated using
Sequential Oligogenic Linkage Analysis Routines.
Results: After adjusting for age, gender, ethnicity, and
height, the heritability estimates of various compositional
attributes were %FM ⫽ 0.64, whole body FM ⫽ 0.71,
TFM ⫽ 0.63, whole body LM ⫽ 0.60, ALM ⫽ 0.66, and
BMI ⫽ 0.64 (all p ⬍ 0.0001). Additional adjustment for
diabetes status, smoking, dietary intake, and physical activ-
ity resulted in only minor changes in the heritability esti-
mates (h
ˆ
2
⫽ 0.63 to 0.72, all p ⬍ 0.0001). Furthermore,
heritability of TFM after additional adjustment for whole
body FM was significant (h
ˆ
2
⫽ 0.55, p ⬍ 0.0001), and
heritability of ALM after additional adjustment for whole
body LM was also significant (h
ˆ
2
⫽ 0.51, p ⬍ 0.0001).
Discussion: These data suggest that FM and LM measured
by DXA are highly heritable and can be effectively used in
designing linkage studies to locate genes governing body
composition. In addition, regional distribution of FM and
LM may be genetically determined.
Key words: body composition, heritability, type 2 dia-
betes, DXA
Introduction
Several studies of related individuals indicate that there is
a genetic component fundamental to overall body size, as
assessed by body weight and by BMI (1,2). However,
adverse health effects of excess fat tissue (adiposity) and/or
deficiency of muscle tissue (sarcopenia) have led to a grow-
ing interest in determining specific environmental and ge-
netic factors determining variation in amounts of these
tissues. In this regard, accurate quantification of the genetic
component of body composition is necessary for under-
standing the independent and/or shared genes contributing
to individual variation in relative and absolute amounts of
fat mass (FM) and lean mass (LM)
1
.
While heritability estimates for BMI range from 0.40 to
0.70, few data are available regarding the heritability of
specific body compartments. In particular, family studies
that used DXA, the current “gold-standard” for assessment
of quantities of FM and LM (3), are lacking. DXA provides
measurements of both whole body and regional quantities of
bone mineral content, FM, and LM using a three-compart-
ment model. Better understanding of the heritability of
individual compartments is essential for linkage and asso-
Received for review December 24, 2003.
Accepted in final form December 9, 2004.
The costs of publication of this article were defrayed, in part, by the payment of page
charges. This article must, therefore, be hereby marked “advertisement” in accordance with
18 U.S.C. Section 1734 solely to indicate this fact.
*Department of Public Health Sciences, †Department of Radiology, ‡Center for Human
Genomics, §Department of Pathology, ¶Department of Internal Medicine, and **Depart-
ment of Biochemistry, Wake Forest University School of Medicine, Winston-Salem, North
Carolina.
Address correspondence to Fang-Chi Hsu, Department of Public Health Sciences, Wake
Forest University School of Medicine, Medical Center Blvd., Winston-Salem, NC 27157.
E-mail: fhsu@wfubmc.edu
Copyright © 2005 NAASO
1
Nonstandard abbreviations: FM, fat mass; LM, lean mass; DM2, type 2 diabetes; TFM,
trunk fat mass; ALM, appendicular lean mass; DHS, Diabetes Heart Study; CV, coefficient
of variation; h
ˆ
2
, estimates of heritability.
312 OBESITY RESEARCH Vol. 13 No. 2 February 2005
ciation approaches to be used successfully to locate genes
affecting adiposity and sarcopenia.
The few studies that report estimates of the heritability of
FM indicate only a small genetic component, on the order of
25% (4 – 6). However, the heritability of %FM, ranging
from 0.54 to 0.76, was shown to be significant in Pima
Indians (7), North American whites (8), and a mixed eth-
nicity sample of young girls and their parents (6). Only one
of these studies (6) used DXA to measure FM. Twin and
family studies showed a stronger genetic component for
fat-free mass, ranging from 0.56 to 0.80 in various popula-
tions (9 –11). Although many important determinants of
body composition including age, sex, ethnicity, height,
weight, menopausal and disease [especially type 2 diabetes
(DM2)] status, and smoking are well established, only a few
of these studies have adjusted the heritability estimates for
these covariates. Adjustment for medications and lifestyle
factors such as diet and exercise habits has been even less
consistent.
The aims of this study were to use DXA measurements in
a family study of DM2 to determine the heritability of
%FM, whole body FM, and LM, as well as regional mea-
surements of trunk fat mass (TFM) and appendicular lean
mass (ALM), and to investigate how the heritability was
modified by covariates.
Research Methods and Procedures
Participants included individuals enrolled in the Diabetes
Heart Study (DHS). DHS is a family study of siblings
concordant for DM2, as well as unaffected family members,
designed to locate and identify genes contributing to mea-
sures of subclinical cardiovascular disease. Only families
with at least two siblings with diabetes were recruited for
the study. All DM2-affected participants had diabetes diag-
nosed after the age of 35 years, in the absence of history of
ketoacidosis, and of at least a 3-year duration. Subjects with
renal insufficiency (serum creatinine ⱖ 1.5 mg/dL or blood
urea nitrogen ⱖ 35 mg/dL) were excluded. Unaffected
siblings, similar in age to siblings with DM2, were also
recruited. Subjects were recruited from internal medicine
clinics and through community advertising. The study was
approved by the Institutional Review Board of the Wake
Forest University School of Medicine. All participants gave
informed consent.
Data for this study included all DHS participants with
complete food frequency and physical activity data. There
were a total of 554 participants (292 women and 262 men)
from 244 families. Pedigree size ranged from 1 to 10. There
were 492 white and 49 African-American sibling pairs, with
267 sibling pairs affected, 59 not affected, and 215 discor-
dant for DM2.
Participant examinations, conducted in the General Clin-
ical Research Center of Wake Forest University, included
interviews for medical history and health behaviors, anthro-
pometric measures, and fasting blood draws. Body weight
was recorded in lightly clothed, shoeless participants to the
nearest 0.1 kg; height was measured to the nearest 0.5 cm
using a stadiometer. Laboratory assays included fasting
glucose and hemoglobin A
1C
. Dietary intake was assessed
using Block food frequency questionnaire (12), and physi-
cal activity was measured using the Paffenbarger physical
activity questionnaire administered by trained interviewers
(13).
DXA measurements of %FM, whole body FM, and LM,
as well as TFM and ALM, were obtained using a fan-beam
scanner (Delphi A; Hologic, Waltham, MA). Whole body
DXA scans were obtained using manufacturer’s recommen-
dations for subject positioning, scan protocols, and scan
analysis. All scan printouts were reviewed by an expert
reader to ensure proper positioning and analysis. Artifacts
were noted and, when possible, excluded from analysis. FM
and LM were determined for the entire body and its subre-
gions. ALM was determined by adding LM for right arm,
left arm, right leg, and left leg. Coefficients of variation
(CVs) were 1.2%, whole body FM; 1.6%, TFM; 0.5%,
whole body LM; and 0.8%, ALM.
Spearman’s rank correlation coefficients were calculated
to estimate the magnitude of the association between con-
tinuous demographic covariates and measurements of body
composition. Demographic covariates included age, body
weight, height, and BMI. Partial correlation coefficients
were computed to adjust for potential common effects of
age, sex, and ethnicity on diabetes and lifestyle covariates as
well as measurements of body composition. Diabetes co-
variates were duration of diabetes, fasting glucose, and
hemoglobin A
1C
; lifestyle covariates were alcohol intake,
dietary intake, and physical activity. Secondary to the cor-
related data structure inherent in a study using siblings,
simple associations based on the correlation coefficient tests
were deemed invalid and reevaluated using the generalized
estimating equation procedure (14), which accounts for
familial correlation through a sandwich estimator of the
variance under exchangeable correlation. Associations be-
tween categorical covariates and measurements of body
composition were also determined using the generalized
estimating equation procedure by comparing whether the
means for different covariate groups were the same. The
categorical covariates included sex, ethnicity, medication
use, diabetes status, and smoking. All statistical analyses
were considered significant when p ⬍ 0.05. SAS software
(SAS Institute, Cary, NC) was used for the statistical anal-
yses.
To determine the contribution of genetic factors to body
composition, the data in family members were analyzed
using the Sequential Oligogenic Linkage Analysis Routines
software package (Southwest Foundation for Biomedical
Research) (15). Sequential Oligogenic Linkage Analysis
Routines perform a variance components analysis of family
Heritability of Body Composition, Hsu et al.
OBESITY RESEARCH Vol. 13 No. 2 February 2005 313
data where the total phenotypic (e.g., whole body FM)
variation is partitioned into genetic and nongenetic sources
of variation. To minimize the bias associated with shared
environmental factors, the estimates of heritability (h
ˆ
2
) were
based on all available family data and were controlled for
covariates related to body composition. The measurements
of body composition were transformed to approximate the
distributional assumptions of the analysis if necessary. The
significance of the heritability estimates was obtained by
likelihood ratio tests, where the likelihood of the model in
which heritability was estimated was compared with the
likelihood of the model in which the heritability was con-
strained to zero. Twice the difference in the natural loga-
rithmic likelihoods yielded a test statistic that was asymp-
totically distributed as a 1/2:1/2 mixture of a
2
variable
with 1 degree of freedom and a point mass at zero (16).
A series of models were developed that incorporated an
increasing number of covariates to determine the extent that
genetic factors contribute to variation in body composition
independently of the measured risk factors. For univariate
analysis, each of the following covariates was examined
independently: age, sex, ethnicity, height, weight, BMI,
menopausal status, diabetes status, duration of diabetes,
serum glucose, hemoglobin A
1C
, smoking, alcohol use, di
-
etary intake, and physical activity. For multivariate analysis,
the most important models were as follows. First, we ex-
amined the combined effect of age, sex, ethnicity, and
height. Second, we added the combined effect of comorbid
factors, such as diabetes status. Third, we added the com-
bined effect of lifestyle factors, such as smoking, dietary
intake, and physical activity. Fourth, we further adjusted for
the use of medications, including insulin, glucocorticoids,
thyroid hormone, and estrogen. Fifth, for regional measure-
ments of body composition, TFM, and ALM, we further
adjusted for whole body FM and LM.
Results
Study Sample
Table 1 shows the characteristics of the study sample.
There were 262 men and 292 women, ranging in age from
38 to 86 years. Most of the women (92%) were postmeno-
pausal. Seventy-nine participants (14%) were African
American. One hundred sixteen participants (23%) were
being treated with insulin, 63 (14%) with estrogen, 30 (7%)
with glucocorticoids, 241 (44%) with statins, 2 (0.4%) with
testosterone, and 58 (13%) with thyroid hormone (data not
shown). The average dietary intake was 1661 ⫾ 731 kcal/d
(SD). The average physical activity level was 567 ⫾ 1018
kcal/wk.
Table 2 summarizes the whole body and regional body
composition for the study sample. Whole body LM and
ALM were lower in women than in men (p ⬍ 0.0001).
Whole body FM, TFM, %FM, and BMI were higher in
women than in men (p ⬍ 0.005).
Association of Body Composition with Potential
Covariates
Table 3 shows Spearman correlations of various possible
covariates with body composition. All measurements of
body composition were inversely associated with age except
%FM. All measurements were positively associated with
body weight. LM was positively associated with height, but
FM and BMI were not. All measurements were positively
associated with BMI. There were no ethnic differences in
any measurements. Those who took insulin medication had
higher FM, whole body LM, and BMI compared with those
who did not take insulin medication after adjusting for age,
sex, and ethnicity (p ⬍ 0.05; data not shown).
Diabetes Factors
After adjusting for age, sex, and ethnicity, measurements
of body composition were not associated with duration of
diabetes. TFM, whole body LM, and BMI were positively
associated with fasting glucose. Whole body FM, TFM,
whole body LM, and BMI were significantly associated
with hemoglobin A
1C
. Averages of %FM, whole body FM,
TFM, whole body LM, and BMI were higher for those who
had diabetes than for those who did not have diabetes (p ⬍
0.005; data not shown).
Lifestyle Factors
After adjusting for age, sex, and ethnicity, measurements
of body composition were not associated with alcohol in-
take. Whole body LM and BMI were positively associated
with dietary intake. Measurements of FM were negatively
associated with physical activity, but ALM was positively
associated. For the three smoking groups, former smokers
had the highest averages of body composition except ALM,
current smokers had the lowest values, and nonsmokers had
intermediate values (p ⬍ 0.005; data not shown).
Heritability of Body Composition
Table 4 shows heritability estimates for body composi-
tion. In the unadjusted model, the heritability estimates
ranged from 0.48 for %FM to 0.76 for whole body FM.
Heritability estimates remained significant in the univariate
analyses after adding potential covariates (i.e., age, sex,
ethnicity, height, weight, BMI, menopausal status, medica-
tion use, diabetes status, duration of diabetes, serum glu-
cose, hemoglobin A
1C
, smoking, dietary intake, alcohol
intake, and physical activity) to the model one at a time
(data not shown). The proportions of phenotypic variance
caused by sex, height, and weight were higher than those
caused by other covariates. Sex adjustment increased heri-
tability estimates for the measurements of LM, whereas
height adjustment lowered the heritability estimates for the
measurements of FM compared with other covariates.
Heritability of Body Composition, Hsu et al.
314 OBESITY RESEARCH Vol. 13 No. 2 February 2005
Table 4 also shows heritability estimates for body compo-
sition adjusted for covariates. First, adjustment for age, sex,
and ethnicity increased the heritability estimate for %FM,
whole body LM, and ALM (h
ˆ
2
⫽ 0.64, 0.73, and 0.80, respec
-
tively). Second, additional adjustment for height lowered the
heritability estimates for whole body LM and ALM (h
ˆ
2
⫽ 0.60
and 0.66, respectively). Third, additional adjustment for dia-
betes status, smoking, dietary intake, and physical activity
resulted in similar heritability estimates for all measurements
of body composition (h
ˆ
2
⫽ 0.64, 0.72, 0.64, 0.63, 0.67, and
0.64 for %FM, whole body FM, TFM, whole body LM, ALM,
and BMI, respectively). Note that further adjustment for the
use of medications, including insulin, glucocorticoids, thyroid
hormone, and estrogen, also resulted in similar heritability
estimates (h
ˆ
2
⫽ 0.64, 0.71, 0.63, 0.56, 0.64, and 0.65 for %FM,
whole body FM, TFM, whole body LM, ALM, and BMI,
respectively; data not shown). There were 95 participants with-
out complete medication data, so the sample size for the model
including medication adjustment is different from the others.
The comparison between the models with and without medi-
cation adjustment may not be fair. Fourth, for the two regional
measurements of body composition, TFM had heritability of
Table 1. Characteristics of the study sample
Characteristic
Men
关n ⴝ 262;
mean ⴞ SD
or%(n)兴
Women
关n ⴝ 292;
mean ⴞ SD
or%(n)兴
Total (n ⴝ 554)
Mean ⴞ SD
or%(n)
Range or
%(n)
Age (years) 61.8 ⫾ 8.6 61.9 ⫾ 8.8 61.9 ⫾ 8.7 38 to 86
Ethnicity (% African Americans) 11.8 (31) 16.4 (48) 14.3 (79)
Weight (kg) 94.3 ⫾ 16.0 85.6 ⫾ 19.2 89.7 ⫾ 18.3 44.3 to 150.5
Height (cm) 175.6 ⫾ 6.9 161.7 ⫾ 5.9 168.3 ⫾ 9.5 122.8 to 195.2
Duration of diabetes (years) 11.1 ⫾ 7.6 10.1 ⫾ 7.0 10.6 ⫾ 7.3 1to40
Diabetes status 86.6 (227) 78.4 (229) 82.3 (456)
Laboratory
Fasting glucose (mM) 141.7 ⫾ 58.7 138.5 ⫾ 58.4 140.0 ⫾ 58.5 16 to 423
Hemoglobin A
1C
(%)
7.3 ⫾ 1.6 7.2 ⫾ 1.9 7.3 ⫾ 1.7 4.6 to 21.8
Lifestyle
Smoking current (%) 19.5 (51) 14.4 (42) 16.8 (93)
Smoking past (%) 60.7 (159) 28.8 (84) 43.9 (243)
Smoking never (%) 19.9 (52) 56.9 (166) 39.4 (218)
Alcohol intake (% kcal/d) 1.1 ⫾ 3.0 0.4 ⫾ 2.6 0.7 ⫾ 2.8 0 to 36.6
Dietary intake (kcal/d) 1805.3 ⫾ 735.2 1532.2 ⫾ 704.4 1661.4 ⫾ 731.3 501.5 to 4531.3
Physical activity (kcal/wk) 696.5 ⫾ 1277.9 450.6 ⫾ 690.1 566.8 ⫾ 1018.1 0 to 10,272
Sample size was 554 except for duration of diabetes (n ⫽ 452), fasting glucose (n ⫽ 553), and hemoglobin A
1C
(n ⫽ 551).
Table 2. Body composition measurements for the study sample
Men
(n ⴝ 262; mean ⴞ SD)
Women
(n ⴝ 292; mean ⴞ SD)
Total (n ⴝ 554)
Mean ⴞ SD Range
%FM (%) 27.55 ⫾ 5.29 41.22 ⫾ 5.29 34.75 ⫾ 8.64 11.17 to 52.31
Whole body FM (kg) 26.49 ⫾ 8.44 35.66 ⫾ 11.56 31.32 ⫾ 11.18 7.18 to 67.29
TFM (kg) 14.96 ⫾ 5.06 18.34 ⫾ 6.24 16.74 ⫾ 5.95 2.72 to 36.40
Whole body LM (kg) 66.25 ⫾ 8.46 49.40 ⫾ 8.04 57.37 ⫾ 11.77 29.59 to 93.93
ALM (kg) 28.94 ⫾ 4.31 20.74 ⫾ 4.10 24.61 ⫾ 5.87 11.02 to 40.33
BMI (kg/m
2
)
30.6 ⫾ 5.1 32.7 ⫾ 7.0 31.7 ⫾ 6.3 16.7 to 57.2
Heritability of Body Composition, Hsu et al.
OBESITY RESEARCH Vol. 13 No. 2 February 2005 315
Table 3. Correlation between body composition measurements and covariates
Characteristic
%FM
[r (p value)]*
Whole body FM
[r (p value)]*
TFM
[r (p value)]*
Whole body LM
[r (p value)]*
ALM
[r (p value)]*
BMI
[r (p value)]*
Age ⫺0.04 (0.4366) ⫺0.18 (⬍0.0001) ⫺0.21 (⬍0.0001) ⫺0.21 (⬍0.0001) ⫺0.22 (⬍0.0001) ⫺0.25 (0.0001)
Weight (kg) 0.18 (⬍0.0001) 0.68 (⬍0.0001) 0.74 (⬍0.0001) 0.74 (⬍0.0001) 0.78 (⬍0.0001) 0.84 (⬍0.0001)
Height (cm) ⫺0.62 (⬍0.0001) ⫺0.20 (⬍0.0001) ⫺0.1 (0.0066) 0.76 (⬍0.0001) 0.75 (⬍0.0001) ⫺0.12 (0.0003)
BMI (kg/m
2
)
0.54 (⬍0.0001) 0.86 (⬍0.0001) 0.87 (⬍0.0001) 0.43 (0.0001) 0.41 (⬍0.0001) 1.00
Duration of diabetes (years) ⫺0.01 (0.6024)† 0.00 (0.7914)† 0.03 (0.7225)† ⫺0.02 (0.9266)† ⫺0.06 (0.2994)† 0.02 (0.7493)†
Fasting glucose (mM) 0.07 (0.5270)† 0.14 (0.0815)† 0.21 (0.0033)† 0.16 (0.0044)† 0.09 (0.1076)† 0.14 (0.0341)†
Hemoglobin A
1C
(%)
0.08 (0.3266)† 0.15 (0.0081)† 0.23 (⬍0.0001)† 0.20 (⬍0.0001)† 0.05 (0.1380)† 0.18 (0.0029)†
Alcohol intake 0.01 (0.1500)† 0.03 (0.1875)† 0.01 (0.0967)† 0.03 (0.4117)† 0.00 (0.5978)† ⫺0.02 (0.2421)†
Dietary intake (kcal/d) 0.06 (0.1716)† 0.07 (0.0546)† 0.06 (0.1467)† 0.09 (0.0200)† 0.01 (0.2507)† 0.08 (0.0251)†
Physical activity (kcal/wk) ⫺0.11 (0.0010)† ⫺0.11 (0.0006)† ⫺0.10 (0.0020)† ⫺0.07 (0.2293)† 0.09 (0.0035)† ⫺0.10 (0.0762)†
* Spearman correlation coefficients; generalized estimating equation p values (adjusted for relatedness).
† Adjusted for age, sex, and ethnicity.
Table 4. Heritability estimates for body composition
Covariates
%FM
[h
ˆ
2
(SE)]
Whole body FM
[h
ˆ
2
(SE)]
TFM
[h
ˆ
2
(SE)]
Whole body LM
[h
ˆ
2
(SE)]
ALM
[h
ˆ
2
(SE)]
BMI
[h
ˆ
2
(SE)]
None 0.48 (0.11) 0.76 (0.11) 0.69 (0.11) 0.49 (0.10) 0.51 (0.10) 0.69 (0.11)
Age, sex, ethnicity 0.64 (0.11) 0.74 (0.11) 0.63 (0.11) 0.73 (0.10) 0.80 (0.10) 0.60 (0.11)
Age, sex, ethnicity, height 0.64 (0.11) 0.71 (0.11) 0.63 (0.11) 0.60 (0.11) 0.66 (0.11) 0.64 (0.11)
Age, sex, ethnicity, height, diabetes status 0.64 (0.11) 0.73 (0.11) 0.67 (0.11) 0.64 (0.11) 0.68 (0.11) 0.67 (0.11)
Age, sex, ethnicity, height, diabetes status,
smoking, dietary intake, physical activity 0.64 (0.11) 0.72 (0.10) 0.64 (0.11) 0.63 (0.11) 0.67 (0.11) 0.64 (0.11)
Age, sex, ethnicity, height, diabetes status,
smoking, dietary intake, physical
activity, whole body FM 0.55 (0.13)
Age, sex, ethnicity, height, diabetes status,
smoking, dietary intake, physical
activity, whole body LM 0.51 (0.12)
p ⬍ 0.0001.
Heritability of Body Composition, Hsu et al.
316 OBESITY RESEARCH Vol. 13 No. 2 February 2005
0.55 with additional adjustment for whole body FM (p ⬍
0.0001), and ALM had heritability of 0.51 with additional
adjustment for whole body LM (p ⬍ 0.0001).
Heritability of Body Composition by Sex
Table 5 shows heritability estimates for body composi-
tion by sex. After adjusting for age, ethnicity, height, dia-
betes status, smoking, dietary intake, and physical activity,
men had heritability estimates ranging from 0.53 to 0.70,
and women had heritability estimates ranging from 0.60 to
0.96. Although it seems that women had higher heritability
compared with men except for %FM, the differences were
not significant when considering the standard errors asso-
ciated with the point estimates.
Discussion
The heritability of anthropometric measures, including
BMI, has been widely discussed (1,2). In contrast, the
heritability of body composition determined by DXA has
not been well studied. In this paper, we show that %FM,
whole body FM, TFM, whole body LM, and ALM were all
highly heritable (adjusted h
ˆ
2
⫽ 0.63 to 0.72) after adjusting
for age, sex, ethnicity, height, diabetes status, smoking,
dietary intake, and physical activity. Thus, these traits can
be used effectively in linkage studies designed to locate
genes for body composition and regional fat and lean mass
distribution. Furthermore, although it is tempting to con-
clude from the stratified results that women have higher
heritability compared with men, the large SEs (⬃0.20 for all
of the measurements) associated with point estimates do not
support this contention.
The heritability estimates for %FM, whole body FM, and
TFM (h
ˆ
2
⫽ 0.48, 0.69, and 0.76, respectively) were in the
range of commonly reported heritability estimates (from
0.30 to 0.90) (17,18) from prior twin and family studies.
Although some studies have used computed tomography or
underwater weighing to measure body fat (4,8,17,19,20),
only two studies have used whole body FM measured by
DXA (5,6), and neither of these studied diabetes-affected
individuals. One study in 112 female white twin pairs re-
ported that 65% of variance in FM was attributable to
genetic factors using univariate model-fitting analysis (5).
The other study in 101 girls and their biological parents
reported that 50% of the variance in percentage body fat
was accounted for by genetic factors (6).
Our heritability estimates were adjusted for a number of
confounding factors explaining 65%, 31%, and 24% of the
variance of %FM, whole body FM, and TFM, respectively.
These confounding factors are likely a subset of a large
number of potentially interrelated factors influencing vari-
ation in body composition and accounted for ⬎50% of the
non-genetic variation in FM. Importantly, however, in our
sample population, genetic factors accounted for ⬎50% of
both whole body FM and TFM.
Table 5. Heritability estimates for body composition by sex
Covariates
%FM
[h
ˆ
2
(p)]
Whole body FM
[h
ˆ
2
(p)]
TFM
[h
ˆ
2
(p)]
Whole body LM
[h
ˆ
2
(p)]
ALM
[h
ˆ
2
(p)]
BMI
[h
ˆ
2
(p)]
Men
None 0.72 (⬍0.0001) 0.71 (⬍0.0001) 0.71 (⬍0.0001) 0.63 (0.0004) 0.78 (⬍0.0001) 0.66 (0.0003)
Age, ethnicity, height 0.69 (0.0001) 0.71 (⬍0.0001) 0.67 (0.0001) 0.49 (0.0053) 0.61 (0.0007) 0.54 (0.0025)
Age, ethnicity, height, diabetes status,
smoking, dietary intake, physical activity 0.66 (0.0001) 0.70 (0.0001) 0.65 (0.0002) 0.53 (0.0054) 0.68 (0.0004) 0.56 (0.0029)
Women
None 0.65 (0.0007) 0.93 (⬍0.0001) 0.77 (⬍0.0001) 0.99 (⬍0.0001) 0.99 (⬍0.0001) 0.85 (⬍0.0001)
Age, ethnicity, height 0.64 (0.0007) 0.85 (⬍0.0001) 0.71 (⬍0.0001) 0.92 (⬍0.0001) 0.96 (⬍0.0001) 0.78 (⬍0.0001)
Age, ethnicity, height, diabetes status,
smoking, dietary intake, physical activity 0.60 (0.0012) 0.81 (⬍0.0001) 0.69 (0.0001) 0.96 (⬍0.0001) 0.94 (⬍0.0001) 0.72 (⬍0.0001)
Heritability of Body Composition, Hsu et al.
OBESITY RESEARCH Vol. 13 No. 2 February 2005 317
There were large interindividual differences in the distri-
bution or location of body fat at any level of whole body fat.
Specific patterning of fat distribution may be genetically
influenced independently of whole body FM. We examined
this hypothesis by estimating the heritability of TFM after
adjustment for whole body FM. The heritability of TFM
after this adjustment was 0.55 (p ⬍ 0.0001). Thus, we found
strongly significant evidence that TFM measured by DXA
was independently influenced by genetic factors.
Genetic factors explained about 49% of the total variance
for whole body LM and about 51% for ALM. One study in
353 postmenopausal white twin pairs reported a heritability
estimate of 0.52 for whole body LM measured by DXA (9),
which is consistent with our study. Although our study
included both men and women, most affected with diabetes,
the genetic components for LM measured by DXA were
similar. After adjusting for age, sex, and ethnicity, the
heritability estimates for LM increased ⬃20%. Because this
adjustment reduces the remaining unexplained phenotypic
variance, the genetic contribution to LM becomes more
apparent. After further adjustment for height, the heritability
estimates for LM were ⬃0.60. LM is highly associated with
height (r ⫽ 0.76 and 0.75 for whole body LM and ALM,
respectively), which is also under strong genetic control
(21). It is possible that the genetic component of LM with-
out further adjusting for height may simply reflect a genetic
component of body size. Additional adjustment for diabetes
status, smoking, dietary intake, and physical activity re-
sulted in similar heritability estimates (h
ˆ
2
⫽ 0.63 and 0.67 for
whole body LM and ALM, respectively). These high estimates
suggest that the genetic component for LM is not explained
solely by the genetic component of body size and is indeed
highly heritable. Note that the confounding factors explained
68% and 66% variance of whole body LM and ALM, respec-
tively (data not shown), and accounted for most of the non-
genetic variation of LM. Furthermore, the heritability estimate
of ALM after adjusting additionally for whole body LM was
0.51 (p ⬍ 0.0001). This suggests that ALM is still heritable
even after adjustment for whole body LM.
There are three major strengths of this study. First, her-
itability of body composition in elderly populations affected
by chronic disorders has not been well studied. Many such
populations are being used in genetic epidemiology research
related to other disorders. In particular, DM2 is an increas-
ingly prevalent condition that has a broad range of clinical
consequences and, as such, is of particular interest to ge-
neticists. Demonstration of substantial heritability of body
composition in families with DM2 provides a strong ratio-
nale for investigating genetic influences on both FM and
LM in diabetics. Note that the ascertainment of DM2 may
bias heritability estimates upward because body composi-
tion is a risk factor for diabetes. Thus, these estimates may
not be directly applicable to the nondiabetic population.
Second, although BMI, body weight, waist circumfer-
ence, and waist-to-hip ratio are commonly used as measures
of adiposity, the use of DXA allows a more accurate mea-
surement of %FM, whole body, and trunk FM. Better char-
acterized phenotypes can improve heritability estimates (6).
Furthermore, Faith et al. (22) suggested that there might be
a substantial genetic contribution to FM but not BMI. This
implies that BMI might be a useful but insufficient measure
of FM for mapping genes, and more precise measurements
of body composition would be more appropriate. In our
study, BMI had a reasonably high heritability estimate (h
ˆ
2
⫽ 0.64), and whole body FM had an even higher heritability
estimate (h
ˆ
2
⫽ 0.72), although the difference was not sig
-
nificant based on the overlap confidence interval for heri-
tability. BMI is still a useful and convenient measurement.
However, DXA measurements constitute a different pheno-
type than BMI, and they are also heritable.
Last, despite known determinants of body composition,
including age, sex, ethnicity, height, weight, BMI, and
menopausal status, few studies have adjusted their calcu-
lated heritability estimates for these covariates. Adjustment
for medications and lifestyle factors such as smoking, diet,
and exercise has been even less consistent. This study
provides a rigorous computation of heritability estimates
after adjusting for these determinants. Furthermore, inclu-
sion of the determinants in the model reduces the pheno-
typic variance, thereby increasing our ability to estimate the
genetic contribution to the variation in body composition.
One limitation of the heritability estimates reported in our
study is that they do not delineate between shared genes and
shared environment. When common environment is a po-
tential risk factor, the genetic component may be overesti-
mated by the heritability estimate. Another limitation is that,
in using a traditional sibling-pair design, we can only esti-
mate heritability using sibling correlations. Thus, the esti-
mated heritability may include both dominant and epistatic
effects, potentially inflating the “true” estimate.
Direct comparison of the heritability estimates between
studies is difficult. Different ascertainment schemes, study
designs, methods of parameter estimation, and population-
specific environmental contributions to the phenotypic vari-
ance can affect heritability estimates. Therefore, differing
heritability estimates for a phenotype occur even when the
genetic variance estimates in the different populations are
similar (23). Similar heritability estimates in the different
populations do not provide evidence for the same genes in
the expression of a trait, nor do dissimilar heritability esti-
mates provide evidence for the exclusion of the same genes
in the expression of a trait (24).
We have shown that the heritability estimates of these
measurements were not statistically different in men com-
pared with women. We did not have sufficient power, based
on the small sample size of African Americans, to perform
Heritability of Body Composition, Hsu et al.
318 OBESITY RESEARCH Vol. 13 No. 2 February 2005
ethnically stratified analyses. We will be able to address this
question in the future with recruitment of additional Afri-
can-American families.
In summary, we have shown that body composition is
highly heritable. Different genes may contribute to the ex-
pression of FM and LM. Future linkage and association
studies to identify the genetic factors underlying the varia-
tion in body composition may ultimately improve strategies
for the prevention and treatment of obesity and sarcopenia.
Acknowledgments
This study was supported by NIH Grants R01 AR48797
(J.J.C.) and R01 HL67348 (D.W.B.) and, in part, by General
Clinical Research Center of the Wake Forest University
School of Medicine Grant M01 RR07122. The authors
acknowledge the cooperation of our participants; the con-
tributions of our study recruiters, Bonnie Dryman, Sue Ann
Backus, and Jennie Locklear, as well as DXA technicians;
and the helpful comments from Dr. Lynne E. Wagenknecht,
which improved the quality of this work.
References
1. Bouchard C, Perusse L, Rice T, Rao DC. The genetics of
human obesity. In: GA Bray, C Bouchard, and WPT James,
eds. Handbook of Obesity. New York: Marcel Dekker; 1998,
pp. 157–90.
2. Rice T, Perusse L, Bouchard C, Rao DC. Familial aggre-
gation of body mass index and subcutaneous fat measures in
the longitudinal Quebec family study. Genet Epidemiol. 1999;
16:316 –34.
3. Albanese CV, Diessel E, Genant HK. Clinical applications
of body composition measurements using DXA. J Clin Den-
sitom. 2003;6:75– 85.
4. Bouchard C. Heredity and the path to overweight and obesity.
Med Sci Sports Exerc. 1991;23:285–91.
5. Nguyen TV, Howard GM, Kelly PJ, Eisman JA. Bone
mass, lean mass, and fat mass: same genes or same environ-
ments? Am J Epidemiol. 1998;147:3–16.
6. Treuth MS, Butte NF, Ellis KJ, Martin LJ, Comuzzie AG.
Familial resemblance of body composition in prepubertal girls
and their biological parents. Am J Clin Nutr. 2001;74:529 –33.
7. Sakul H, Pratley R, Cardon L, Ravussin E, Mott D, Bogardus
C. Familiality of physical and metabolic characteristics that pre-
dict the development of non-insulin-dependent diabetes mellitus
in Pima Indians. Am J Hum Genet. 1997;60:651– 6.
8. Rice T, Daw EW, Gagnon J, et al. Familial resemblance for
body composition measures: the HERITAGE Family Study.
Obes Res. 1997;5:557– 62.
9. Arden NK, Spector TD. Genetic influences on muscle
strength, lean body mass, and bone mineral density: a twin
study. J Bone Miner Res. 1997;12:2076 –81.
10. Seeman E, Hopper JL, Young NR, Formica C, Goss P,
Tsalamandris C. Do genetic factors explain associations be-
tween muscle strength, lean mass, and bone density? A twin
study. Am J Physiol. 1996;270:E320 –7.
11. Abney M, McPeek MS, Ober C. Broad and narrow herita-
bilities of quantitative traits in a founder population. Am J
Hum Genet. 2001;68:1302–7.
12. Block G, Woods M, Potosky A, Clifford C. Validation of a
self-administered diet history questionnaire using multiple
diet records. J Clin Epidemiol. 1990;43:1327–35.
13. Paffenbarger RS Jr, Wing AL, Hyde RT. Physical activity
as an index of heart attack risk in college alumni. Am J
Epidemiol. 1995;142:889 –903.
14. Liang KY, Zeger SL. Longitudinal data analysis using gen-
eralized linear models. Biometrika. 1986;73:13–22.
15. Almasy L, Blangero J. Multipoint quantitative-trait linkage analysis
in general pedigrees. Am J Hum Genet. 1998;62:1198 –211.
16. Self SG, Liang KY. Asymptotic properties of maximum
likelihood ratio tests under nonstandard conditions. J Am Stat
Assoc. 1987;82:605–10.
17. Perusse L, Despres JP, Lemieux S, Rice T, Rao DC, Bou-
chard C. Familial aggregation of abdominal visceral fat lev-
els: results from the Quebec Family Study. Metabolism. 1996;
45:378 – 82.
18. Perusse L, Chagnon YC, Weisnagel J, Bouchard C. The
Human Obesity Gene Map: the 1998 update. Obes Res. 1999;
7:111–29.
19. Chagnon YC, Borecki IB, Perusse L, et al. Genome-wide
search for genes related to the fat-free body mass in the
Quebec family study. Metabolism. 2000;49:203–7.
20. Chagnon YC, Rice T, Perusse L, et al. Genomic scan for genes
affecting body composition before and after training in Cauca-
sians from HERITAGE. J Appl Physiol. 2001;90:1777– 87.
21. Arden NK, Baker J, Hogg C, Baan K, Spector TD. The
heritability of bone mineral density, ultrasound of the calca-
neus and hip axis length: a study of postmenopausal twins.
J Bone Miner Res. 1996;11:530 – 4.
22. Faith MS, Pietrobelli A, Nunez C, Heo M, Heymsfield SB,
Allison DB. Evidence for independent genetic influences on
fat mass and bone mass index in a pediatric twin sample.
Pediatrics. 1999;104:61–7.
23. Mitchell BD, Kammerer CM, Blangero J, et al. Genetic and
environmental contributions to cardiovascular risk factors in
Mexican Americans: the San Antonio Family Heart Study.
Circulation. 1996;94:2159 –70.
24. Mahaney MC, Blangero J, Comuzzie AG, VandeBerg JL,
Stern MP, MacCluer JW. Plasma HDL cholesterol, triglyc-
erides, and adiposity: a quantitative genetic test of the conjoint
trait hypothesis in the San Antonio Family Heart Study. Cir-
culation. 1995;92:3240 – 8.
Heritability of Body Composition, Hsu et al.
OBESITY RESEARCH Vol. 13 No. 2 February 2005 319
