Assessment of Sex-Specific Genetic and Environmental Effects on
Bone Mineral Density

Lillian B. Brown,1 Elizabeth A. Streeten,2 Alan R. Shuldiner,2,3 Laura A. Almasy,4 Patricia A. Peyser,1

and Braxton D. Mitchell2n

1Department of Epidemiology, University of Michigan School of Public Health, Ann Arbor, Michigan
2Division of Endocrinology, Diabetes and Nutrition, University of Maryland School of Medicine, Baltimore, Maryland

3Geriatric Research and Education Clinical Center (GRECC), Veterans Administration Medical Center, Baltimore, Maryland
4Department of Genetics, Southwest Foundation for Biomedical Research, San Antonio, Texas

Although it is widely accepted that genes contribute significantly to the variation in bone mineral density (BMD), the nature
of the genetic contribution is poorly defined. There are large gender differences in BMD, although whether sex-specific
genetic effects influencing variation in BMD contribute to these differences is not known. To address this issue, we studied
929 subjects from large families participating in the Amish Family Osteoporosis Study. Bone mineral density was measured
at the hip and spine by dual energy x-ray absorptiometry (DXA). We used variance decomposition procedures to partition
variation in BMD into genetic and environmental effects common to both sexes and to men and women separately. After
accounting for covariate effects, the heritability of BMD ranged from 0.63 to 0.72 in men and 0.80 to 0.87 in women.
The residual environmental variance in BMD at the spine, but not hip, was significantly higher in men than in women
(P o 0.05), reflecting a greater variance in BMD due to unexplained non-genetic factors in men. In contrast, there were no
significant differences between men and women in the magnitude of the genetic variance in BMD, nor did the genetic
correlation in BMD between men and women differ significantly from one. Overall, these analyses do not provide evidence
for sex-specific genetic effects, suggesting that many of the genes influencing variation in BMD should be detectable in both
men and women. & 2004 Wiley-Liss, Inc.

Key words: sex; bone mineral density; genetics; variance; interaction; heritability

Grant sponsor: National Institutes of Health; Grant numbers: RO1-AR46838, RO1-AG18728, RO1-HL69313, and M01 RR165001 awarded
to the University of Maryland General Clinical Research Center.
nCorrespondence to: Braxton D. Mitchell, PhD, Division of Endocrinology, Diabetes and Nutrition, University of Maryland School of
Medicine, 660 W. Redwood Street, Room 492, Baltimore, MD 21201. E-mail: bmitchel@medicine.umaryland.edu
Received 11 December 2003; Accepted 10 March 2004
Published online 14 May 2004 in Wiley InterScience (www.interscience.wiley.com)
DOI: 10.1002/gepi.20009

INTRODUCTION

Osteoporosis is a debilitating disease character-
ized by low bone mineral density (BMD) that is
associated with significant morbidity and mortal-
ity. According to the Third National Health and
Nutrition Examination Survey (NHANES III), 50%
of U.S. women over age 50 have low bone mass
and 20% of white postmenopausal women have
osteoporosis at the femoral neck (hip) [Looker
et al., 1997]. Osteoporotic fractures are one of the
most common causes of disability and contribu-
tors to medical care costs in many regions of the
world [Cummings and Melton, 2002]. Large
prospective studies have shown that almost all
types of fracture are increased in adults with low
BMD [Cummings et al., 1993]. Fractures result in

functional impairment, including impaired basic
activities of daily living, subsequent nursing home
care, the loss of ambulatory ability, and loss of the
ability to live independently [Cummings and
Melton, 2002; Gullberg et al., 1997].
Longitudinal, family, and twin studies suggest

that variation in BMD is strongly influenced by
genes. Women with a maternal history of hip
fracture have lower BMD than women without a
history of such fracture [Seeman et al., 1994] and
are themselves twice as likely to suffer a hip
fracture [Cummings et al., 1993; Zmuda et al.,
1999]. BMD is highly correlated between twin
pairs [Eisman, 1999; Christian et al., 1989; Pocock
et al., 1987] and among related family members
[Jouanny et al., 1995; Matkovic et al., 1990; Ferrari
et al., 1998; Lutz, 1986; Seeman et al., 1989].

Genetic Epidemiology 27: 153–161 (2004)

& 2004 Wiley-Liss, Inc.



Overall, twin and family studies suggest that after
considering covariates, 60–80% of variance in
bone density measurement is genetically deter-
mined [Eisman, 1999; Krall and Dawson-Hughes,
1993; Prentice, 2001].
The nature of the genetic contribution to BMD is

poorly defined. There are strong gender differ-
ences in BMD that contribute to a substantially
higher fracture risk among women than men.
Moreover, the gender differences are not constant
across the age range as, for example, in the
immediate post-menopausal period, when bone
loss is accelerated in women as compared to men
of the same age [Hunter and Sambrook, 2000].
Given the large genetic effects on BMD, it is
possible that different (although overlapping)
subsets of genes contribute to variation in BMD
in men and women and/or that the genetic
determinants of BMD may be modulated by sex-
specific hormonal, environmental, and nutritional
factors. Evidence for sex-specific genetic effects on
BMD, however, is scant. In the experimental
mouse model, Orwoll et al. [2001] examined peak
BMD in males and females from 18 different
inbred recombinant strains and observed signifi-
cant gender by strain interactions, with males
having higher BMD than females in some strains,
but lower in others. In subsequent quantitative
trait linkage (QTL) analysis, these researchers then
identified six distinct chromosomal regions linked
to variation in BMD in male mice only, five
regions linked in female mice only, and two linked
regions that were shared between the genders
[Orwoll et al., 2001]. Whether these QTLs will, in
fact, turn out to have differential effects on BMD
between male and female mice remains to be seen,
as does their potential relevance to BMD in
humans.
In humans, Naganathan and colleagues also

provided evidence for the presence of sex-specific
effects on BMD by showing that the correlations
in BMD at the forearm and spine, as measured by
ultrasound, were higher in same sex dizygotic
(DZ) twin pairs than in opposite sex DZ twin pairs
[Naganathan et al., 2002]. The intraclass correla-
tion for the forearm BMD measurement was 0.40
in both female (n¼265 pairs) and male (n¼71
pairs) DZ twin pairs, compared to 0.19 in opposite
sex twin pairs (n¼82 pairs). Estimated correlations
between female DZ twin pairs, male DZ twin
pairs, and opposite sex DZ twin pairs at the spine
were 0.38, 0.56, and 0.17, respectively. Addition-
ally, linkage studies carried out in at least two
populations have reported stronger evidence for

linkage to QTLs influencing BMD in one sex than
in the other [Karasik et al., 2003; Kammerer et al.,
2003] although these results are difficult to
interpret since the hypothesis of sex-specific
linkage was not formally tested. Aside from these
few studies, the issue of sex-specific genetic
effects has not been extensively addressed.
The goal of our study is to explore more fully

potential sex-specific differences in the relative
influence of genetic and environmental factors on
BMD. To address this issue, we have analyzed
data collected from 929 Amish individuals from
large extended families. These families were
recruited through the Amish Family Osteoporosis
Study (AFOS), a study consisting of subjects
enrolled from very large families from Lancaster
County, Pennsylvania. Using variance decomposi-
tion procedures, we quantified genetic effects on
BMD that are common to both sexes and to men
and women separately. Specifically, we consid-
ered the following questions: (1) is the magnitude
of the genetic variation larger in one sex than in
the other?; (2) is the magnitude of the residual
environmental variation larger in one sex than in
the other?; and (3) is there evidence for sex-
specific genetic effects on BMD (i.e., is there a
subset of genes that influences variation in BMD
in both sexes jointly and another subset of genes
that influences variation in each sex separately)?

METHODS

SUBJECTS AND MEASUREMENTS

The AFOS began in 1997 with the goal of
identifying the genetic determinants of osteoporo-
sis. Individuals believed to be at risk for osteo-
porosis by virtue of their fracture history or prior
bone density measurements were recruited into
the study as index cases. These individuals were
recruited by word-of-mouth, a community-wide
mailing, advertisements in an Amish newspaper,
or by referral from local physicians. The diagnosis
of osteoporosis in these individuals was verified
by measurement of BMD using dual energy x-ray
absorptiometry (DXA). Individuals found to have
a T score of �2.5 or less in either the hip or spine
were designated as probands. We then invited the
probands’ spouses and all first-degree relatives
aged 20 years and over to participate in the study.
In addition, we recruited into the study the first-
degree relatives of any other examined individual
(e.g., spouses) having a T score of �2.5 or lower at
the spine or hip on our bone densitometry test.

Brown et al.154



Between the initiation of recruitment in 1997
and January 2002, a total of 943 individuals were
enrolled into the AFOS, including 57 probands
and their relatives. Of those enrolled, complete
information was obtained on 929 subjects, includ-
ing 573 women and 356 men. Using the extensive
genealogical records maintained on the Amish
[Beiler, 1988; Agarwala et al., 1998, 2001], these
individuals could be combined into a single
14-generation pedigree. Study participants
were evaluated at the Amish Research Clinic in
Strasburg, PA, by qualified nurses known to the
participants. A medical interview included past
medical history, family history of medical pro-
blems including fractures, and specific details
regarding previous fractures, history of medica-
tion use, and menstrual and reproductive history
for women. Height was measured using a
stadiometer and weight was recorded with the
participant in standard Amish clothing but with-
out shoes. The mineral content at the lumbar spine
and hip was measured by DEXA by a registered
nurse certified in bone densitometry (Hologic
4500W, Hologic, Inc., Bedford, MA). BMD was
determined by dividing the total bone mineral
content (g) by the projected area of the region
scanned (cm2). For this report, we have restricted
analysis of BMD to measures obtained at the
spine, femoral neck, and total hip. Total hip BMD
is defined as the sum of the bone mineral content
at the femoral neck, trochanter, and intertrochan-
ter sites divided by the total area of these three
sites.
The protocol for the AFOS was approved by the

Institutional Review Board at the University of
Maryland. Informed consent was obtained from
all subjects prior to participation.

ANALYTICAL METHODS

We carried out a series of statistical analyses
using a full pedigree-based variance component
approach for the purpose of partitioning variation
in BMD into selected components [Almasy and
Blangero, 1998]. In the basic model, the level of
BMD, y, for individual i was modeled as
yi¼mþ

P
biXij þgiþei, where m is the mean BMD,

Xij is the j-th covariate for the ith individual, bj is
its regression coefficient, and gi and ei represent
the random deviations from m for individual i that
are attributable to additive genetic and residual
error effects, respectively. The residual error
component includes true random error, measure-
ment error, and any non-additive genetic compo-

nents. The effects of gi and ei are uncorrelated and
normally distributed with mean zero and var-
iances sg

2 and se
2, respectively. The se

2 term can be
considered the residual environmental variance,
that is, the remaining environmental variance
after accounting for the effects of measured
covariates. Maximum likelihood methods were
used to simultaneously estimate the mean and
variances as well as the covariate and genetic
effects. All models included the following covari-
ates: age, age2, height, and BMI. These factors
were selected as covariates because they were
each independently associated with one or more
BMD measures in a sex-specific preliminary
analysis. The significance of particular compo-
nents can be assessed by comparing the likelihood
of a model with the component of interest
estimated to the likelihood of a model in which
the component effect is constrained to be zero.
The full and restricted models are then compared
by likelihood ratio test, which produces a test
statistic that is asymptotically distributed as a w2

distribution with the degrees of freedom equal to
difference in number of parameters between the
two models.
Using this variance component modeling

approach, we then considered various hypotheses
concerning the sex-specific variances in BMD.
Our initial goal was to compare the genetic
variance between the two sexes. We addressed
this issue in two ways, first by comparing
the ‘‘relative’’ proportion of the variance
attributable to genetic effects in men and
women separately and, second, by comparing
the absolute magnitude of the genetic effect. The
proportion of the total phenotypic variation in
BMD that could be attributable to additive genetic
effects (sg

2/sp
2) corresponds to ‘‘narrow’’ sense

heritability (h2) since it reflects the degree of
additive genetic variance only. We tested whether
the heritabilities in BMD differed between men
and women (i.e., h2men¼h2women) by comparing the
difference between the heritability estimates in the
two sexes with the estimated variance of the
difference.
Estimating the absolute magnitude of the

genetic variance required a further partitioning
of the variance. Following the approach of
Blangero and colleagues, [Blangero, 1993; Martin
et al., 2002b; Towne et al., 1993], we expanded the
basic variance component model to allow the
genetic variances in male and female BMD to
differ when men and women were considered
together. Briefly, the expected genetic covariance

Sex-Specific Effects on BMD 155



between a male and female relative pair is defined
as:

COVðGM; GFÞ ¼ 2FrG�sG�M�sG�F
where F is the coefficient of kinship between the
two individuals, rG is the genetic correlation
between the expressions of the trait in the two
sexes, and sG-M and sG-F are the genetic standard
deviations for men and women, respectively.
With the additional terms as defined above,

we constructed a general model that partitioned
variance in BMD into the following 13 terms: an
overall mean scaled to the value in males, a
coefficient corresponding to the effect of sex
(bsex), coefficients for age and sex*age (bage
and bage*sex), coefficients for age

2 and sex*age2

(bage
2 and bage

2 *sex), coefficients for height (bheight),
and body mass index (bBMI), male and female
genetic standard deviations (sG-M and sG-F), male
and female residual environmental standard
deviations (sE-M and sE-F), and the genetic
correlation between males and females (rG).
Genetic correlations reflect the degree to which
the genetic effect on BMD in men correlates with
the genetic effect on BMD in women [Falconer
and MacKay, 1994]. Interaction terms of sex with
age and age2 were included because of the well-
established differences between men and women
in the relationship between age and BMD.
This expanded model allowed us to test several

explicit hypotheses related to sex by gene interac-
tions. First, we considered if the magnitude of the
genetic effect was similar between the sexes by
testing whether the genetic standard deviations
were similar between men and women (i.e., H0:
sG-M¼ sG-F). Rejection of this hypothesis implies
that genes account for a larger proportion of the
variance in one sex than in the other. In similar
fashion, we tested if the magnitude of the residual
environmental effect was similar between the
sexes (i.e., H0: sE-M¼ sE-F). The third hypothesis
that we tested was whether the magnitude of the
genetic correlation was significantly less than one
(i.e., H0: rG¼1). A genetic correlation between
men and women that is significantly less than one
implies that a different gene or suite of genes
contributes to variance in BMD in men and
women. If there is a sex � gene interaction, then
the expectation is that rG o 1 and sG-M a sG-F.
As before, significance testing was conducted

using the likelihood ratio test. Specifically, we
compared likelihoods between models in which
values of sG-M and sG-F were allowed to differ
(full model) and in which they were constrained

to be the same (restricted model). Similarly, we
compared the likelihood between a model in
which rG was estimated (full model) to that in
which its value was constrained to be one
(restricted model). When testing models for which
the value of a particular parameter was con-
strained to a boundary value (e.g., rG¼1), the P
value was based on a 1

2
:1
2

mixture of a w1
2

distribution and a point mass at zero [Self and
Liang, 1987].

RESULTS

Basic characteristics of the study population are
shown in Table I. The mean age of the 356 men
and 573 women was approximately 50 years. Men
were on average four to five inches taller than
women (mean 7 standard deviation; 67.2 7 2.7
vs. 62.8 7 3.5 in.) and 12 lbs heavier (171.2 7 38.0
vs. 159.5 7 49.3 lbs). BMD measurements at the
spine (L1-L4), femoral neck, and total femur were
significantly higher in men compared to women
(P¼0.003 at the spine; P o 0.001 at the femoral
neck and total hip). The overall (phenotypic)
variance in BMD was also significantly greater
in women than in men at all three BMD sites (P o
0.001 by the Levene test for equality of variances).
The numbers of relative pairs, both same sex

and total, who were phenotyped and included in
the analyses are shown in Table II. The sample
included 2,049 female-female pairs (212 mother-
daughter, 548 sister-sister, 673 aunt-niece, and 616
first cousin pairs) and 838 male-male pairs (103
father-son, 263 brother-brother, 257 uncle-nephew,
and 215 first cousin pairs). Overall, there were
2,887 same sex and 2,552 opposite sex relative
pairs, for a total of 5,439 total relative pairs,
including same sex and opposite sex pairs
combined.

TABLE I. Characteristics (mean 7 SD) of Amish Family
Osteoporosis Study participants

Variable Men (n¼356) Women (n¼573)
Age-adjusted

P value

Age (yrs) 50.2716.9 50.1724.6 0.60
Height (in) 67.272.7 62.873.5 o0.001
Weight (lb) 171.2738.0 159.5749.3 o0.001
BMI (kg/m2) 26.876.2 28.478.1 o0.001

BMD (g/cm2)
Spine (LI-L4) 0.96470.165 0.93970.205 0.003
Femoral neck 0.87970.150 0.83670.184 o0.001
Total femur 1.01170.156 0.92270.196 o0.001

Brown et al.156



The heritabilities of BMD at the spine, femoral
neck, and total femur were estimated after
accounting for the effects of the measured
covariates age, age2, height, and BMI. The residual
heritability of BMD ranged from 0.63 (spine) to
0.72 (total femur) in men and from 0.80 (femoral
neck) to 0.87 (spine and total hip) in women. At
each site, the estimated residual heritability in
BMD was significantly larger in women than in
men (P o 0.001 at all sites) (data not shown).
To gain insights into the factors contributing

to variation in BMD in men and women, we
partitioned the total variance in BMD into

components attributable to measured covariates
(e.g., age, height, and BMI), the additive effects of
genes, and to unmeasured, or residual, environ-
mental factors. Results of these analyses are
shown in Table III. In men, measured covariates
accounted for very little (6.3%) of the total
variation in spine BMD, and from 18 to 24% of
the variation in hip BMD. The additive effects of
genes accounted for 51 to 59% of the total
variation in BMD, in contrast to the 63 to 72% of
the residual variation in BMD as described above.
Thus, 35% of the total variation in spine BMD in
men could not be accounted for by genes or
measured covariates. In contrast, in women, the
measured covariates accounted for 32% of the
total variation in spine BMD and 52 to 53% of the
total variation in hip BMD. The additive effects of
genes accounted for an additional 59% of the
variation in spine BMD in women and 38 to 42%
of the variation in hip BMD. Thus, only 6 to 9% of
the total variation in BMD in women could
not be accounted for by genes and/or measured
covariates.
After estimating the proportion of the total

phenotypic variation attributable to genetic and
environmental effects in men and women, we
next tested several additional hypotheses, includ-
ing whether the magnitude of the genetic and
environmental variances differed between men
and women and whether the genetic correlation in
BMD differed between men and women. To
accomplish these goals, we performed a more
complete partitioning of BMD into its constituent
genetic and residual environmental components.
In these analyses, we allowed the genetic and
residual environmental variances in BMD be-
tween males and females to differ, and also
estimated the genetic and residual environmental
correlations in BMD between men and women.
Results from the full model, in which all para-
meters were estimated, are shown in Table IV.
Following estimation of the full set of model
parameters, we performed a series of nested tests

TABLE II. Number of relative pairs included in the
sample of 939 Amish subjects

Relative pair class Number

Parent-offspring
Mother-daughter 212
Father-son 103
All pairsa 575

Sibling-sibling
Sister-sister 548
Brother-brother 263
All pairsa 1,439

Avuncular
Aunt-niece 673
Uncle-nephew 257
All pairsa 1,778

Cousin-cousin (1st cousin only)
Female-female 616
Male-male 215
All pairsa 1,647

Total numbers of pairsb

Female pairs 2,049
Male pairs 838
Total same sex pairs 2,887
Totala 5,439

aIncluding same sex and opposite sex pairs.
bDoes not include grandparent-grandchild, grand avuncular, 2nd
cousins, and more distantly related pairs.

TABLE III. Components of variance for bone mineral density (BMD)

Men (n¼356) Women (n¼573)

BMD site Measured covariatesa Geneticb Residual environment Measured covariatesa Geneticb Residual environment

Spine 0.063 0.590 0.347 0.325 0.591 0.084
Femoral neck 0.235 0.513 0.252 0.531 0.378 0.091
Total femur 0.182 0.592 0.226 0.517 0.421 0.062

aMeasured covariates include age, age2, height, and BMI.
bNote that the genetic contribution to total phenotypic variation presented in this table differs from the heritability presented in the text,
which is estimated as the genetic contribution to residual (not total) phenotypic variation once covariate effects are accounted for.

Sex-Specific Effects on BMD 157



in which we constrained values of selected
parameters, which enabled us to test, first,
whether the magnitude of the genetic variance in
BMD differed between men and women; second,
whether the magnitude of the residual environ-
mental variance in BMD differed between men
and women; and third, whether the genetic
correlation in BMD between male and female
relative pairs differed from one. With respect to
the first hypothesis, we observed that the genetic
SD did not differ significantly between men and
women (sG-M vs. sG-F: spine: 1.07 vs. 1.21; femoral
neck: 0.92 vs. 0.91; total hip: 0.96 vs. 0.96; P 4 0.30
for all). In contrast, we observed marginal
evidence (P¼0.046) for a larger residual environ-
mental SD in men than in women for spine BMD
(0.77 vs. 0.39), although not for BMD at either of
the two hip sites. With respect to the third
hypothesis, we observed that the genetic correla-
tion between men and women did not differ
significantly from one (rG¼0.91, 0.90, and 0.97 for
spine, femoral neck, and total hip). Thus, our
analyses provided modest evidence for a sex by
residual environment interaction at one of the
BMD sites (spine), but no evidence for a gene by
sex interaction. Nor was there evidence for sex-
specific genetics effects on BMD.

DISCUSSION

The hypothesis that genetic and/or residual
environmental effects on BMD might differ
between men and women is motivated in part
by the well-known gender differences that are
observed in BMD. Typically, men have higher
BMD than women, and women experience a
significantly higher rate of bone loss around the
time of menopause compared to men of similar
age. There are many possible explanations for
these sex differences, including the sex differ-
ences present in levels of androgens and estro-
gens, which are thought to influence bone
development and/or decrease bone resorption
[Vanderschueren and Bouillon, 1995; Gallagher,
2003]. Genes may potentially play a significant
role in influencing variation in steroid production
and regulation, including the changes in these
processes that occur over time.
Evaluating the gender-specificity of genetic

effects on BMD is not straightforward. Although
the hypothesis that a particular allele is more
strongly associated with BMD in one sex than the
other is easily tested in the context of conventional

TABLE IV. Model parameters estimated from variance
partitioning of bone mineral density (BMD)

Parameter

Spine BMD
( � 10, in
g/cm2)

Neck BMD
( � 10, in
g/cm2)

Total Hip BMD
( � 10, in
g/cm2)

m 9.27 8.41 9.85
b(age) �0.02 �0.04 �0.03
b(sex) 0.24 0.03 �0.51
b(agensex) �0.03 �0.01 �0.01
b(age2) �0.001 0.001 �0.004
b(age2nsex) �0.001 �0.001 �0.001
b(height) 0.08 0.09 0.09
b(BMI) 0.09 0.11 0.13
sG-M 1.07 0.92 0.96
sG-F 1.21 0.91 0.96
sE-M 0.77 0.65 0.62
sE-F 0.39 0.48 0.43
rG 0.91 0.90 0.97

LL
(full model)

�593.86 �438.17 �442.87

Hypothesis 1:
H0: genetic variance equal between the sexes
Parameterization: sG-M¼sG-F

LL
(nested model):

�594.39 �438.18 �442.88

w1
2 1.06 0.01 0.02

P 0.30 0.91 0.89
Conclusion Accept H0 Accept H0 Accept H0

Hypothesis 2:
H0: environmental variance equal between the sexes
Parameterization: sE-M¼sE-F

LL
(nested model):

�595.84 �438.84 �443.68

w1
2 3.96 1.32 1.62

P 0.046 0.25 0.20
Conclusion: Reject H0 Accept H0 Accept H0

Hypothesis 3:
H0: genetic correlation similar between males and females
Parameterization: rG¼1

LL
(nested model):

�594.27 �438.56 �442.93

w1
2 0.82 0.78 0.10

P 0.18 0.19 0.38
Conclusion Accept H0 Accept H0 Accept H0

aModel parameters: m¼mean BMD; b¼regression coefficients
(for age, agensex, age2, age2nsex, height, and body mass index);
sG-M¼genetic SD in males; sG-F¼genetic SD in females;
sE-M¼environmental SD in males; sE-F¼environmental SD in
females; rG¼genetic correlation in BMD between men and
women. BMI¼body mass index; BMD¼bone mineral density;
LL¼log-likelihood.
P values for Hypothesis 3 based on a 1

2
:1
2
mixture of a w1

2 and a point
mass of zero.

Brown et al.158



genetic association studies, such analyses consider
only the effect of an individual gene. Linkage
studies conducted in both mice [Orwoll et al.,
2001] and humans [Karasik et al., 2003; Kammerer
et al., 2003] have revealed the presence of QTLs
and/or regions of linkage that are more strongly
linked in one sex than in the other, although none
of these studies has explicitly tested whether
evidence for linkage could be excluded in the
other sex. To our knowledge, only one other study
has formally sought to identify gender-specific
genetic effects in humans, with these researchers
inferring the presence of such effects on the basis
that the correlation in BMD was significantly
higher in same sex DZ twins than in opposite sex
DZ twins [Naganathan et al., 2002]. However,
these data are also consistent with several other
interpretations, including the presence of a com-
mon environmental component that was more
frequently shared among same sex compared to
opposite sex twin pairs.
Interestingly, the Amish population does not

appear to be at particularly high risk for fracture,
and, in fact, rates of non-traumatic hip fracture are
somewhat lower in the Amish than in their non-
Amish counterparts, an observation consistent
with mean BMD in the Amish also being slightly
higher [Streeten et al., 2004]. Nevertheless, the
unique attributes of the Old Order Amish (OOA)
make this population an attractive one for
attempting to dissect out the genetic contributions
to phenotypic variation. Amish families typically
tend to be very large, so that there are a large
number of sibling relative pairs available for
analysis. Moreover, the Amish have a strong
interest in their genealogies, and accurate record-
keeping dating back many generations allows the
large Amish families to be linked into a single
pedigree. Finally, the relatively homogenous en-
vironment of the OOA may allow for a more clear
elucidation of the genetic factors contributing
to BMD.
Comparison of trait heritabilities between men

and women, whether expressed as the contribu-
tion of genes to total trait variability or residual
trait variability, provides only limited insight into
possible gender differences in genetic effects. The
reason for this is that heritabilities are proportions,
and their values are influenced both by the
magnitude of the genetic variance and the
contribution of residual environmental factors to
the total variance. We, thus, tested several addi-
tional hypotheses, including whether the magni-
tude of the genetic variance differed between men

and women and whether the genetic correlation in
BMD differed between men and women. The fact
that the magnitude of the genetic standard
deviation did not differ significantly between
men and women indicates only that the contribu-
tion of genes to the variance in absolute terms is
relatively similar between men and women in this
study group; this test does not address whether it
is the same gene or suite of genes that influences
variation in BMD in both sexes. However, this
latter possibility (i.e., that the same subset of genes
influences variation in BMD in men as in women)
is consistent with the finding that the estimated
genetic correlation did not differ significantly
from one.
Our results do not support the existence of

sex-specific genetic effects on BMD. However,
this conclusion does not preclude the possibility
that there might be individual loci that have
sex-specific effects on BMD; rather the net
contribution of sex-specific genetic effects
across all loci is likely to be small. Two additional
caveats should be considered. First, this study
population consists of families ascertained on
probands with a history of hip fracture and/or
osteoporosis. Thus, the mean BMD in the study
population is slightly lower than what would
have been observed in randomly ascertained
families. However, how this selection scheme
might impact detection of sex-specific genetic
effects on BMD is unclear. A second, and perhaps
more limiting, issue is that we have also
pooled subjects across a wide range of ages for
our analyses. While pooling subjects of
different ages produces large numbers of same
sex relative pairs suitable for analysis, this
strategy may dilute sex-specific genetic and
residual environmental effects whose expression
is age-dependent. In this regard, a more compre-
hensive evaluation of this issue might include
analysis of 20–40-year-old men and women in
order to evaluate sex-specific genetic and envir-
onmental influences on acquisition of peak bone
mass, or alternatively, to restrict analysis to post-
menopausal women and older men to evaluate
potential sex-specific genetic and environmental
influences on bone loss. Unfortunately, our study
sample was not sufficiently large to consider sex-
specific analyses in young and old groups
separately as there were only 120 males and 179
females aged 20–40 years and 102 males and 169
females aged 60 years and over. These totals
provided an insufficient number of relative pairs
in these strata to obtain the robust partitioning of

Sex-Specific Effects on BMD 159



the variances required to estimate sex-specific
genetic effects.
Despite the unique ancestral history of the

Amish and their relative protection against hip
fracture, it seems reasonable to expect that the
basic conclusions reached by this study pertaining
to sex-specific genetic effects should be general-
izable to other populations. The OOA in Lancaster
County are descendents of migrants from Western
Europe and are thus derived from the same
overall gene pool as the overall U.S. and European
Caucasian populations. Perhaps more impor-
tantly, the analytic approach we followed empha-
sizes the additive effects of genes. Although
generalizability of these findings in the Amish
would be limited if there were Amish-specific
mutations in genes that had very large effects on
BMD, we are not aware that this is the case. In our
initial survey we did, in fact, identify several
individuals with very low BMD who upon
subsequent evaluation were found to carry muta-
tions in COL1A2, the gene encoding type 1
collagen alpha 2 and responsible for osteogenesis
imperfecta. However, these individuals were
excluded from the current analysis.
It is possible that there may be gender differ-

ences in genetic effects influencing other aspects
of bone morphology not captured by BMD. Male-
female differences in a variety of indices related to
hip geometry and structure have been reported
[Beck et al., 1992; Kaptoge et al., 2003], many of
which are also related to hip fracture indepen-
dently of BMD [Gluer et al., 1994; Partanen et al.,
2001; Beck et al., 1996; Karlsson et al., 1996].
Genetic influences on some of these measures
have been reported [Jian et al., 2004; Liu et al.,
2004; Koller et al., 2003], although the sex-
specificity of these effects has not yet been
investigated.
Detection and identification of sex-specific

genetic effects when they exist is a worthwhile
pursuit because such effects may provide impor-
tant insights into disease etiology (e.g., genes
expressed in only one sex or the other may operate
on sex-specific risk factors) and also because
appropriately modeling these effects should the-
oretically enhance our ability to identify the genes
underlying them. For example, knowledge of sex-
specific genetic effects may motivate one’s choice
of candidate genes, or improve ability to localize
individual loci through linkage analysis by ex-
plicit modeling of sex-specific linkage compo-
nents [Martin et al., 2002a]. The absence of
evidence for sex-specific genetic effects on BMD

in our study offers the hope that at least many of
the loci influencing variation in BMD should be
detectable in both men and women.

ACKNOWLEDGMENTS

We thank Drs. Richa Agarwala and Alejandro
Schaffer for their assistance in pedigree construc-
tion using the Amish Genealogical Database. We
also thank Dr. Holmes Morton and Caroline
Morton for their support in establishing and
maintaining ties with the Amish community.
Finally, we gratefully acknowledge our Amish
liaisons and field workers and the extraordinary
cooperation and support of the Amish community
without which these studies would not have been
possible. This work was supported in part by the
University of Maryland General Clinical Research
Center, Grant M01 RR165001, General Clinical
Research Centers Program, National Center for
Research Resources (NCRR), NIH.

REFERENCES

Agarwala R, Biesecker LG, Hopkins KA, Francomano CA,
Schäffer AA. 1998. Software for constructing and verifying
pedigrees within large genealogies and an application to the
Old Order Amish of Lancaster County. Genome Res 8:211–221.

Agarwala R, Schäffer AA, Tomlin JF. 2001. Towards a complete
North American Anabaptist genealogy II: analysis of
inbreeding. Hum Biol 73:533–545.

Almasy L, Blangero J. 1998. Multipoint quantitative-trait linkage
analysis in general pedigrees. Am J Hum Genet 62:1198–1211.

Beck TJ, Ruff CB, Scott WW, Jr., Plato CC, Tobin JD, Quan CA.
1992. Sex differences in geometry of the femoral neck with
aging: a structural analysis of bone mineral data. Calcif.Tissue
Int 50:24–29.

Beck TJ, Ruff CB, Mourtada FA, Shaffer RA, Maxwell-Williams K,
Kao GL, Sartoris DJ, Brodine S. 1996. Dual-energy X-ray
absorptiometry derived structural geometry for stress fracture
prediction in male U.S. Marine Corps recruits. J Bone Miner Res
11:645–653.

Beiler K. 1988. Fisher family history: descendants and history of
Christian Fisher (1757–1838), 3rd ed. Lancaster, PA: Eby’s
Quality Printing.

Blangero J. 1993. Statistical genetic approaches to human
adaptability. Hum Biol 65:941–966.

Christian JC, Yu PL, Slemenda CW, Johnston CC, Jr. 1989.
Heritability of bone mass: a longitudinal study in aging male
twins. Am J Hum Genet 44:429–433.

Cummings SR, Black DM, Nevitt MC, Browner W, Cauley J,
Ensrud K, Genant HK, Palermo L, Scott J, Vogt TM. 1993. Bone
density at various sites for prediction of hip fractures. The Study
of Osteoporotic Fractures Research Group [see comments].
Lancet 341:72–75.

Cummings SR, Melton LJ. 2002. Epidemiology and outcomes of
osteoporotic fractures. Lancet 359:1761–1767.

Eisman JA. 1999. Genetics of osteoporosis. Endocr Rev 20:788–804.

Brown et al.160



Falconer D.S., MacKay T.F.C. 1994. Quantitative genetics, 4th ed.
Essex, England: Longman Group Lmt.

Ferrari S, Rizzoli R, Slosman D, Bonjour JP. 1998. Familial
resemblance for bone mineral mass is expressed before
puberty. J Clin Endocrinol Metab 83:358–361.

Gallagher JC. 2003. Effect of estrogen on bone. In: Favus MJ,
editor. Primer on the metabolic bone diseases and disorders of
mineral metabolism, 5th ed. Washington, DC: American Society
for Bone and Mineral Research. Chapter 53, p 327–330.

Gluer CC, Cummings SR, Pressman A, Li J, Gluer K, Faulkner KG,
Grampp S, Genant HK. 1994. Prediction of hip fractures from
pelvic radiographs: the study of osteoporotic fractures. The
Study of Osteoporotic Fractures Research Group. J Bone Miner
Res 9:671–677.

Gullberg B, Johnell O, Kanis JA. 1997. World-wide projections for
hip fracture. Osteoporos Int 7:407–413.

Hunter DJ, Sambrook PN. 2000. Bone loss. Epidemiology of bone
loss. Arthritis Res 2:441–445.

Jian WX, Long JR, Deng HW. 2004. High heritability of bone size at
the hip and spine in Chinese. J Hum Genet 49:87–91.

Jouanny P, Guillemin F, Kuntz C, Jeandel C, Pourel J. 1995.
Environmental and genetic factors affecting bone mass.
Similarity of bone density among members of healthy
families. Arthritis Rheum 38:61–67.

Kammerer CM, Schneider JL, Cole SA, Hixson JE, Samollow PB,
O’Connell JR, Perez R, Dyer TD, Almasy L, Blangero J, Bauer
RL, Mitchell BD. 2003. Quantitative trait loci on chromosomes
2p, 4p, and 13q influence bone mineral density of the forearm
and hip in Mexican Americans. J Bone Miner Res 18:2245–2252.

Kaptoge S, Dalzell N, Loveridge N, Beck TJ, Khaw KT, Reeve J.
2003. Effects of gender, anthropometric variables, and aging on
the evolution of hip strength in men and women aged over 65.
Bone 32:561–570.

Karasik D, Cupples LA, Hannan MT, Kiel DP. 2003. Age, gender,
and body mass effects on quantitative trait loci for bone mineral
density: the Framingham Study. Bone 33:308–316.

Karlsson KM, Sernbo I, Obrant KJ, Redlund-Johnell I, Johnell O.
1996. Femoral neck geometry and radiographic signs of
osteoporosis as predictors of hip fracture. Bone 18:327–330.

Koller DL, White KE, Liu G, Hui SL, Conneally PM, Johnston CC,
Econs MJ, Foroud T, Peacock M. 2003. Linkage of structure at
the proximal femur to chromosomes 3, 7, 8, and 19. J Bone
Miner Res 18:1057–1065.

Krall EA, Dawson-Hughes B. 1993. Heritable and life-style
determinants of bone mineral density. J Bone Miner Res 8:1–9.

Liu PY, Qin YJ, Recker RR, Deng HW. 2004. Evidence for a major
gene underlying bone size variation in the Chinese. Am J
Human Biol 16:68–77.

Looker AC, Orwoll ES, Johnston CC, Jr., Lindsay RL, Wahner HW,
Dunn WL, Calvo MS, Harris TB, Heyse SP. 1997. Prevalence of
low femoral bone density in older U.S. adults from NHANES
III. J Bone Miner Res 12:1761–1768.

Lutz J. 1986. Bone mineral, serum calcium, and dietary intakes of
mother/daughter pairs. Am J Clin Nutr 44:99–106.

Martin LJ, Cole SA, Hixson JE, Mahaney MC, Czerwinski SA,
Almasy L, Blangero J, Comuzzie AG. 2002a. Genotype by
smoking interaction for leptin levels in the San Antonio Family
Heart Study. Genet Epidemiol 22:105–115.

Martin LJ, Mahaney MC, Almasy L, MacCluer JW, Blangero J,
Jaquish CE, Comuzzie AG. 2002b. Leptin’s sexual dimorphism
results from genotype by sex interactions mediated by
testosterone. Obes Res 10:14–21.

Matkovic V, Fontana D, Tominac C, Goel P, Chesnut CH, III. 1990.
Factors that influence peak bone mass formation: a study of
calcium balance and the inheritance of bone mass in adolescent
females. Am J Clin Nutr 52:878–888.

Naganathan V, Macgregor A, Snieder H, Nguyen T, Spector T,
Sambrook P. 2002. Gender differences in the genetic factors
responsible for variation in bone density and ultrasound. J Bone
Miner Res 17:725–733.

Orwoll ES, Belknap JK, Klein RF. 2001. Gender specificity in the
genetic determinants of peak bone mass. J Bone Miner Res
16:1962–1971.

Partanen J, Jamsa T, Jalovaara P. 2001. Influence of the upper
femur and pelvic geometry on the risk and type of hip fractures.
J Bone Miner Res 16:1540–1546.

Pocock NA, Eisman JA, Hopper JL, Yeates MG, Sambrook PN,
Eberl S. 1987. Genetic determinants of bone mass in adults. A
twin study. J Clin Invest 80:706–710.

Prentice A. 2001. The relative contribution of diet and genotype to
bone development. Proc Nutr Soc 60:45–52.

Seeman E, Hopper JL, Bach LA, Cooper ME, Parkinson E, McKay
J, Jerums G. 1989. Reduced bone mass in daughters of women
with osteoporosis. N Engl J Med 320:554–558.

Seeman E, Tsalamandris C, Formica C, Hopper JL, McKay J. 1994.
Reduced femoral neck bone density in the daughters of
women with hip fractures: the role of low peak bone
density in the pathogenesis of osteoporosis. J Bone Miner Res
9:739–743.

Self SG, Liang KY. 1987. Asymptotic properties of maximum
likelihood estimators and likelihood ratio tests under non-
standard conditions. J Am Stat Assoc 82:605–610.

Streeten EA, McBride DJ, Lodge AL, Pollin TI, Stinchcomb DG,
Agarwala R, Schaffer AA, Shapiro JR, Shuldiner AR, Mitchell
BD. 2004. Reduced incidence of hip fracture in the Old Order
Amish. J Bone Miner Res 19:308–313.

Towne B, Blangero J, Siervogel RM. 1993. Genotype by sex
interaction in measures of lipids, lipoproteins, and
apolipoproteins. Genet Epidemiol 10:611–616.

Vanderschueren D, Bouillon R. 1995. Androgens and bone. Calcif
Tissue Int 56:341–346.

Zmuda JM, Cauley JA, Ferrell RE. 1999. Recent progress in
understanding the genetic susceptibility to osteoporosis. Genet
Epidemiol 16:356–367.

Sex-Specific Effects on BMD 161