Osteoporosis is a common disease characterized by an [604862]

Osteoporosis is a common disease characterized by an
increased propensity to fracture owing to decreased
bone mass and bone quality1. Over 9 million oste –
oporotic fractures occur worldwide, and of these
approximately 2 million occur per year in the United
States, incurring US$17 billion in direct costs annu –
ally: a burden that is projected to increase by 50%
by 2025 (REF.  2). Therefore, osteoporotic fractures are
common and inflict a substantial economic, social and
clinical burden.
Clinically, osteoporosis is diagnosed when a patient
presents with a fracture that has resulted from mini –
mal trauma (such as a fall from standing height).
Preventive therapies exist for fractures, and so indi –
viduals are screened through measurement of bone
mineral density (BMD). In combination with clinical
risk factors, osteoporosis is commonly diagnosed, for
the purposes of preventive therapy, through measure –
ment of BMD3. Thus, although low BMD is only a risk
factor for fracture, much like hypertension is a risk for
myocardial infarction, therapeutic decisions that are
aimed at preventing fracture are often based on BMD
measurement.
BMD has a high heritability: estimates lie between
50 and 85%4–6. Although BMD is the most important
clinical risk factor for osteoporotic fracture, apart from
age and sex, most individuals who develop osteoporo –
tic fractures do not have BMD-defined osteoporosis7,8,
suggesting that factors that are unrelated to BMD have
a strong impact on the risk of fracture. Osteoporotic fractures themselves have moderate heritability, with
estimates of 54% and 68% for wrist and hip fractures,
respectively, in peri-menopausal women9,10. However,
this heritability appears to decrease dramatically with
age, such that after 79 years, estimates of the heritability
for hip fractures drop to 3%10.
Linkage studies have identified loci for BMD11–14,
but these findings have not been replicated between
studies, and when nine linkage studies were meta-
analysed, involving a total of 11,842 subjects, no loci
were associated15 (reviewed elsewhere16). Therefore, it
seemed that the allelic architecture of BMD would not
be amenable to the large effect sizes that are required
for linkage studies. This led investigators to seek
osteoporosis loci by focusing on candidate regions,
but again most of these studies have not been subse –
quently replicated by larger and more rigorous studies
with systematic phenotypic and genotypic definitions
across cohorts17. By contrast, genome-wide association
studies (GW ASs) have enjoyed considerable success
in identifying replicated loci that are associated with
osteoporosis.
In this Review, we will first introduce the various
GW AS designs that have allowed the identification of
osteoporosis loci that will serve to provide insights into
pathophysiologic mechanisms. Then, novel drug tar –
gets and the ability to identify people who are at risk of
fracture through genotypic profiles will be discussed.
Finally, the possible future directions for research into
the genetics of osteoporosis will be presented.1Departments of Medicine,
Human Genetics,
Epidemiology and
Biostatistics, Lady Davis
Institute for Medical
Research, Jewish General
Hospital, McGill University,
Montreal, Quebec H3T 1E2,
Canada.
2T win Research and Genetic
Epidemiology, King’s College
London, London, UK.
Correspondence to J.B.R.
e-mail: brent.richards@
mcgill.ca
doi:10.1038/nrg3228Genetics of osteoporosis from
genome-wide association studies:
advances and challenges
J. Brent Richards1,2, Hou-Feng Zheng1 and Tim D. Spector2
Abstract | Osteoporosis is among the most common and costly diseases and is increasing
in prevalence owing to the ageing of our global population. Clinically defined largely
through bone mineral density, osteoporosis and osteoporotic fractures have reasonably
high heritabilities, prompting much effort to identify the genetic determinants of this
disease. Genome-wide association studies have recently provided rapid insights into the
allelic architecture of this condition, identifying 62 genome-wide-significant loci. Here, we
review how these new loci provide an opportunity to explore how the genetics of
osteoporosis can elucidate its pathophysiology, provide drug targets and allow for
prediction of future fracture risk. DISEASE MECHANISMSREVIEWS
576 | AUGUST 2012 | VOLUME 13 www.nature.com/reviews/genetics
© 2012 Macmillan Publishers Limited. All rights reserved

deCODE genetics
An Icelandic company that
specializes in the identification
of genetic risk factors for
disease.GWASs for BMD
Before GW ASs, studies of osteoporosis involved link –
age and candidate gene association studies. Given the
failure of linkage studies, researchers turned their
focus to candidate gene studies. This candidate-based
strategy was comprehensively tested by assessing all
candidate genes implicated in osteoporosis and/or
fracture in a large consortium of five population-
based studies involving 19,195 participants17. In this
study, only 9 of the 150 previously tested candidate
loci harboured common variants that are associated
with BMD, and only four loci harboured common
variants for fracture. Although this study was limited
insofar as it was able to test only SNPs that were cap –
tured by HapMap SNP proxies, these findings suggest
that most previously studied candidate genes were not
replicated in a well-powered study with standardized
phenotyping and genotyping. An exception to this
arose from candidate gene studies for low-density
lipoprotein receptor-related protein 5 ( LRP5 ) for BMD
and fracture18,19.
Whereas candidate gene and linkage studies
have not contributed substantial gains in osteoporo –
sis genetics, GW ASs have identified 62 loci that are
genome-wide-significant (which, for the purpose of
this Review, is defined as P < 5 × 10−8) for BMD at either
the lumbar spine or the femoral neck. This Review
focuses on GW ASs that use SNP data from >200,000
SNPs and that include a replication sample (we there –
fore do not intend to review copy-number-variant- or
pathway-based analyses using GW AS data). To date, 14
such GW ASs have been published for BMD, which is
the clinically relevant measure of osteoporosis. TABLE 1
summarizes all of these GW ASs, and TABLE 2 describes
the loci identified.Initial GWASs. In 2008, two GW ASs were published
using discovery phase data from the TwinsUK/
Rotterdam20 and deCODE Genetics21 studies. Together,
these independent studies identified five loci that
were associated at a genome-wide-significant level
(P < 5 × 10−8) with BMD (TABLES 1 ,2). In addition, LRP5 ,
zinc finger and BTB domain containing 40 ( ZBTB40 )
and spectrin, beta, non-erythrocytic 1 ( SPTBN1 ) were
associated with a risk of osteoporotic fracture.
deCODE Genetics identified two new loci that were
genome-wide-significant for BMD, both of which had
significant effects on the risk of fracture22. The first
GW AS reported for BMD in children identified the SP7
locus, which encodes the transcription factor osterix, as
being associated with BMD, and replication was subse –
quently achieved in three additional adult populations.
Interestingly, variants in osterix were also associated
with height in children23.
Multi-ethnic studies. Multi-ethnic studies permit the
identification of alleles that are shared across populations.
The first multi-ethnic BMD GW AS used Europeans as
a discovery population, with follow-up in Europeans,
Asians and subjects of African ancestry from Tobago. This
particular GW AS identified genome-wide-significant
variants in ADAM metallopeptidase with thrombospon –
din type 1 motif, 18 ( ADAMTS18 )24, but this locus was
not found to be genome-wide-significant in two larger
meta-analyses of European and Asian populations from
the Genetic Factors for Osteoporosis (GEFOS) consor –
tium25,26 (which is discussed in more detail below). A dif –
ferent multi-ethnic GW AS, this time involving Chinese
women as the discovery cohort with replication in six
independent populations of European and Asian descent,
identified a novel locus with marginal genome-wide Table 1 | GWASs for BMD and osteoporotic fracture
Study group (year) Total
maximum
sample sizeTotal number of GWAS-
identified loci achieving
genome-wide significance*Number of new GWAS-
identified loci achieving
genome-wide significanceRefs
Richards et al. (2008) 8,557 2 2 20
Styrkarsdottir et al. (2008) 13,786 4 3 21
Styrkarsdottir et al. (2008) 15,375 7 2 22
Timpson et al. (2009) 5,275 1 1 23
Xiong et al. (2009) 9,109 1 1 24
Rivadeneira et al. (2009) 19,195 20 13 25
Guo et al. (2010) 11,568 1 1 35
Kung et al. (2010) 18,898 1 1 81
Hsu et al. (2010) 11,290 4 0 30
Koller et al. (2010) 2,193 0 0 82
Duncan et al. (2011) 21,798 2 2 29
Estrada et al. (2012) 83,894 56 32 26
Zheng et al. (2012) 5,672 1 0 31
Medina-Gomez et al. (2012) 13,712 1 0 32
*Genome-wide significance is defined as P < 5×10−8. Studies include only genome-wide association studies (GWASs) using SNP
data from >200,000 SNPs and including a replication sample or a meta-analysis of several cohorts. BMD, bone mineral density.REVIEWS
NATURE REVIEWS | GENETICS VOLUME 13 | AUGUST 2012 | 577
© 2012 Macmillan Publishers Limited. All rights reserved

significance ( P = 5.3 × 10−8), and these results were rep –
licated in the larger multi-ethnic GEFOS -2 analysis26.
Thus, multi-ethnic studies have been helpful in iden –
tifying risk variants that are shared across populations.
However, further work could be done to leverage the dif –
ferential allelic distributions across different ethnicities to
identify further loci, as has been described recently27,28.
The use of extreme phenotypes and expression data.
Using an alternative design from a case–control study
of extremely high versus extremely low hip BMD, two
novel loci (namely, UDP -N-acetyl-alpha -d-galactosa
mine:polypeptide N-acetylgalactosaminyltransferase 3 
(GALNT3 ) and R -spondin 3 ( RSPO3 )) were identified to
be genome-wide-significant for hip BMD29. These loci
were also genome-wide-significant in subsequent meta-
analysis, including in data from REFS  25,26. Although
it ranks among the smallest discovery cohorts used in
GW ASs for BMD, this study showed that smaller dis –
covery cohorts using individuals from the extremes of
the phenotypic continuum could be used to generate
findings that were replicated in the rest of the pheno –
typic distribution, thereby providing a paradigm for
cost-efficient GW ASs of continuous traits in which
the cost of selection and screening is cheaper than geno –
typing. Integrating functional data can also help to pri –
oritize candidates. Using expression data from humans
and animal models in addition to GW AS signals, the
Framingham cohort was able to prioritize SNPs aris –
ing from GW ASs at tumour necrosis factor receptor
superfamily, member 11b ( TNFRSF11B ; commonly
known as OPG ), wntless homologue ( WLS ; also known
as GPR177 ) and the transcription factor SOX6 (REF.  30).
Although these signals were identified in previous
GW ASs, their prioritization method outlined an efficient
way of selecting SNPs for further validation.
Other BMD sites. Recently, the WNT16 –FAM3C locus
was identified to be associated with forearm BMD, corti –
cal bone thickness, osteoporotic fracture risk and bone
strength in a mouse knockout of Wnt16 (REF.  31). These
findings strongly suggest an important role for WNT16
in cortical bone strength, as the BMD site (namely, the
forearm) is rich in cortical bone, and SNPs at this locus
strongly influenced forearm fracture risk in humans
(odds ratio = 1.33, P = 7.3 × 10−9). This same locus was
identified to be associated with total-body and skull
BMD in children and adults32.
Large-scale meta-analyses. Combining data from five
genome-wide association studies involving 19,195 sub –
jects of European descent, the first large-scale meta-anal –
ysis for BMD was undertaken. This GEFOS consortium
identified 13 novel regions that were genome-wide-
significant for BMD25. This effort therefore more than
doubled the number of loci to be associated with BMD
at a genome-wide-significant level.
Further insights into the allelic architecture and
genetic determinants of BMD arose from the second
GEFOS meta-analysis, which involved 32,961 individu –
als in the discovery phase and was replicated in 50,933 Table 2 | Loci associated with BMD at genome-wide-significant levels*
Locus Nearest gene or
candidateBest BMD
P valueFracture
odds ratioFracture
P valueRefs
1p31.3 WLS (also known
as GPR177 )2.6 × 10−1325,29,30
1p36 ZBTB40 7.4 × 10−571.07 3.6 × 10−622,25,26,29
1p36.12 WNT4 9.6 × 10−111.09 1.4 × 10−722,25,26,29
1q24.3 DNM3 8.5 × 10−1526
2p16 SPTBN1 2.3 × 10−181.06 2.6 × 10−825,26
2p21 PKDCC 1.3 × 10−926
2q13 ANAPC1 1.5 × 10−926
2q14.2 INSIG2 1.2 × 10−1026
2q24 GALNT3 3.9 × 10−3026,29
3p22 CTNNB1 4.4 × 10−251.06 2.9 × 10−725,26
3q13.2 KIAA2018 4.1 × 10−1026
3q25.31 LEKR1 4.5 × 10−1226
4p16.3 IDUA 5.2 × 10−1526
4q21.1 MEPE , SPP1 and
IBSP1.2 × 10−271.06 1.7 × 10−828,29
5q14 MEF2C 4.5 × 10−6125,26,29
5q31 ALDH7A1 6.4 × 10−62.25 2.1 × 10−935
6p21.1 SUPT3H and
RUNX25.6 × 10−1126
6p22.3 CDKAL1 and
SOX42.7 × 10−1326
6q22 RSPO3 8.1 × 10−1226,29
6q25 C6ORF97 and
ESR14.0 × 10−3521,22,25,26
7p14.1 STARD3NL 3.8 × 10−381.05 7.2 × 10−525,26
7q21.3 FLJ42280 and
SHFM19.4 × 10−1225,26
7q21.3 SLC25A13 8.1 × 10−481.08 5.9 × 10−1126
7q31.31 WNT16 and
FAM3C3.2 × 10−511.06 2.7 × 10−726,31,32
7q36.1 ABCF2 7.3 × 10−926
8q13.3 XKR9 and LACTB2 1.9 × 10−826
8q24 OPG 3.2 × 10−3920,21,25,26
9q34.11 FUBP3 3.4 × 10−221.05 3.5 × 10−526
10p11.23 MPP7 2.4 × 10−1626
10q21.1 MBL2 and DKK1 1.6 × 10−121.1 9.0 × 10−926
10q22.3 KCNMA1 5 × 10−1926
10q24.2 CPN1 9 × 10−1026
11p12 LRP4 , ARHGAP1
and F25.1 × 10−1825,26
11p14.1 DCDC5 2.2 × 10−1125,26
11p14.1 LIN7C and DCDC5 4.9 × 10−81.05 3.3 × 10−526
11p15 SOX6 1.1 × 10−3225,26,30
11q13.2 LRP5 2.1 × 10−261.09 1.4 × 10−820,25,26
12p11.22 KLHDC5 and
PTHLH1.9 × 10−1226
12p13.33 ERC1 and WNT5B 5.6 × 10−1226REVIEWS
578 | AUGUST 2012 | VOLUME 13 www.nature.com/reviews/genetics
© 2012 Macmillan Publishers Limited. All rights reserved

independent subjects26. All subjects were of European
or East Asian ancestry. This study was made possi –
ble through coordination of the analysis of discovery
cohorts and centralization and standardization of DNA,
genotyping and phenotyping for all follow -up cohorts.
This study identified 32 novel loci for BMD. It also was
the first study to interrogate the X chromosome sys –
tematically in a large-scale study, identifying one locus
on the X chromosome to be genome-wide-significant.
Although such large sample sizes can introduce hetero –
geneity, very few significant SNPs showed heterogeneity
of effect across 51 cohorts.
Genome-wide significance. Several of the above GW ASs
have reported ‘significant’ findings arising from SNPs
that failed to achieve a P value of less than 5 × 10−8. We
feel that this rigorous threshold for statistical significance
should be used in GW ASs, as it accounts for the number
of statistically independent common variants present
in HapMap33. SNPs that fail to meet this threshold, but
that are declared as significant, have subsequently been
shown to be false positives in larger meta-analyses; such
was the case for the major histocompatibility complex
(MHC) locus, which was reported to be associated with
BMD21, but did not achieve a P value <5 × 10−8 in either
GEFOS meta-analysis25,26.
Summary of GWASs for BMD. GW ASs have clarified
why linkage studies were underpowered for identifying
regions that are associated with BMD; the effect sizes of
these common variants are small. Remarkably, however,
despite the large number of loci from GW ASs, only 5.8%
of the variance in femoral neck BMD has been explained
by genome-wide-significant SNPs, even in the GEFOS -2
meta-analysis26. Importantly, the effect sizes of novel loci
have been decreasing as sample size increases (average
allelic effect per risk allele was −0.09 standard devia –
tions in early studies20,21 and decreased to −0.04 standard
deviations for novel alleles in GEFOS -2).
Thus, GW ASs have begun to describe the allelic
architecture of osteoporosis. The fact that small effects
from common variants are replicable across large
studies and that no locus contributed to a substantial
amount of the variance in this trait suggests that osteo –
porosis may either have an infinitesimal allelic archi –
tecture, wherein a large number of alleles across the
allele frequency spectrum have a small effect on risk,
or that rare variants contribute substantially to the phe –
notype34. Undoubtedly, larger, better-powered GW ASs
will identify more novel loci, but it seems quite likely
that the variance explained by common genome-wide-
significant SNPs is likely to remain the minority herit –
ability. Nonetheless, the overall goals of the genetics of
osteoporosis are not to explain its heritability, but rather
to contribute to clinical care.
GWASs for fracture
BMD is a risk factor for fracture, which is highly clini –
cally relevant when deciding which patients to treat with
primary prevention for fracture. Although BMD has
been helpful as a tool for understanding the pathways Table 2 (cont.) | Loci associated with BMD at genome-wide-significant levels*
Locus Nearest gene
or candidateBest BMD
P valueFracture
odds ratioFracture
P valueRefs
12q13 SP7 3.0 × 10−2022,23,25,26
12q13.12 DHH 1.2 × 10−1526
12q23.3 C12ORF23 9.6 × 10−1026
13q14 RANKL 2.0 × 10−2121,22,25,26
14q32 MARK3 5.2 × 10−161.09 0.0038 22,25,26
14q32.12 RPS6KA5 2 × 10−151.05 7.2 × 10−526
16p13.11 NTAN1 1.7 × 10−1026
16p13.3 AXIN1 1 × 10−1626
16p13.3 C16ORF38 and
CLCN71.5 × 10−1626
16q12.1 CYLD 1.9 × 10−2226
16q23 ADAMTS18 2.1 × 10−824
16q24 FOXL1 and
FOXC21.0 × 10−1425,26
17p13.3 SMG6 9.8 × 10−1926
17q12 CRHR1 1.4 × 10−825
17q21 SOST 2.0 × 10−111.07 6.9 × 10−622,26
17q21 HDAC5 1.7 × 10−825
17q24.3 SOX9 1.9 × 10−1126
18p11.21 C18ORF19 and
FAM210A4.9 × 10−81.08 8.8 × 10−1326
18q21.33 RANK 1.6 × 10−1722,25,26
19q13.11 GPATCH1 6.6 × 10−1126
20p12 JAG1 3.1 × 10−191.42 0.009 26,81
Xp22.31 FAM9B and
KAL11.2 × 10−826
*Genome-wide significance is defined as P < 5 × 10−8. Note that some loci are repeated in
subsequent studies to show their relationship with fracture. ABCF2 , ATP-binding cassette,
sub-family F; ADAMTS18 , ADAM metallopeptidase with thrombospondin type 1 motif, 18;
ALDH7A1 , aldehyde dehydrogenase 7 family, member A1; ANAPC1 , anaphase promoting
complex subunit 1; ARHGAP1 , rho GTPase activating protein 1; BMD, bone mineral density;
C12ORF23 , chromosome 12 open reading frame 23; CDKAL1 , CDK5 regulatory subunit
associated protein 1 -like 1; CLCN7 , chloride channel, voltage-sensitive 7; CPN1 ,
carboxypeptidase N, polypeptide 1; CRHR1 , corticotropin releasing hormone receptor 1;
CTNNB1 , catenin (cadherin-associated protein), beta 1; CYLD , cylindromatosis (turban
tumour syndrome); DCDC5 , doublecortin domain containing 5; DHH , desert hedgehog;
DKK1 , dickkopf 1; DNM3 , dynamin 3; ERC1 , ELKS/RAB6 -interacting/CAST family member 1;
ESR1 , oestrogen receptor 1; F2, coagulation factor II (thrombin); FAM210A , family with
sequence similarity 210, member A; FAM3C , family with sequence similarity 3, member C;
FAM9B , family with sequence similarity 9, member B; FOXC2 , forkhead box C2; FUBP3 , far
upstream element (FUSE) binding protein 3; GALNT3 , UDP -N-acetyl-alpha -D-galactosamine:
polypeptide N-acetylgalactosaminyltransferase 3; GPATCH1 , G patch domain containing 1;
HDAC5 , histone deacetylase 5; IBSP , integrin-binding sialoprotein; IDUA , alpha -L-
iduronidase; INSIG2 , insulin-induced gene 2; JAG1 , jagged 1; KAL1 , Kallmann syndrome 1
sequence; KCNMA1 , potassium large conductance calcium-activated channel, subfamily M,
alpha member 1; KLHDC5 , kelch domain containing 5; LACTB2 , lactamase, beta 2; LEKR1 ,
leucine, glutamate and lysine rich 1; LIN7C , lin‑7 homologue C; LRP4 , low-density lipoprotein
receptor-related protein 4; MARK3 , MAP/microtubule affinity-regulating kinase 3; MBL2 ,
mannose-binding lectin (protein C) 2, soluble; OPG , also known as TNFRSF11B ; MEF2C , myocyte
enhancer factor 2C; MEPE , matrix extracellular phosphoglycoprotein; MPP7 , membrane
protein, palmitoylated 7; NTAN1 , N-terminal asparagine amidase; PKDCC , protein kinase
domain containing, cytoplasmic homologue; PTHLH , parathyroid hormone-like hormone;
RANK , also known as TNFRSF11A ; RANKL , also known as TNFSF11 ; RPS6KA5 , ribosomal
protein S6 kinase, 90kDa, polypeptide 5; RSPO3 , R-spondin 3; RUNX2 , runt-related
transcription factor 2; SHFM1 , split hand/foot malformation (ectrodactyly) type 1; SLC25A13 ,
solute carrier family 25; SMG6 , smg‑6 homologue, nonsense mediated mRNA decay factor;
SOST , sclerostin; SOX4 , sex-determining region Y (SRY) box 4; SPP1 , secreted
phosphoprotein 1; SPTBN1 , spectrin, beta, non-erythrocytic 1; STARD3NL , STARD3
N-terminal like; SUPT3 , suppressor of Ty 3; WLS , wntless; WNT4 , wingless-type MMTV
integration site family, member 4; XKR9 , Kell blood group complex subunit-related family,
member 9; ZBT40 , zinc finger and BTB domain containing 40.REVIEWS
NATURE REVIEWS | GENETICS VOLUME 13 | AUGUST 2012 | 579
© 2012 Macmillan Publishers Limited. All rights reserved

and potential drug targets involved in fracture risk, it
is fractures themselves that are the most clinically rel –
evant endpoint, as they inflict considerable morbidity
and mortality.
Fracture has a considerably lower heritability than
BMD and decreases with age, as reviewed above.
Paradoxically, the age range at which most fractures
occur is the age range at which the disease is least herita –
ble. This is particularly challenging for population-based
studies that tend to accrue fracture cases disproportion –
ately from this low heritability age range. Specific study
designs that overcome this limitation by focusing on
younger individuals suffering fractures are likely to be
required to investigate the contribution of genetic vari –
ants to fracture risk fully, and they have already identi –
fied new genome-wide-significant hits for fracture, even
with small sample sizes32.
The lack of GW AS results from large-scale studies
testifies to this problem. We are aware of only one
published GW AS that used fracture as the discovery
phenotype35. One SNP in aldehyde dehydrogenase 7
family, member A1 (ALDH7A1 ) was associated with
hip fracture at a genome-wide-significant level, with a
high odds ratio of 2.25 ( P = 2.1 × 10−9) (TABLE 1 ), and
was genome-wide-suggestive for hip BMD in Chinese
and European subjects. This locus did not associate
with BMD in either GEFOS effort or in other GW ASs
performed.
Another path towards identifying the genetic deter –
minants of fracture is first to identify loci that are asso –
ciated with BMD through GW ASs and then to test their
association with fracture. Although this paradigm is
limited, as most individuals who suffer osteoporotic
fractures do not have a low BMD7,8, it can yield loci for
a common disease that is otherwise difficult to study.
This approach has worked well for the identification
of loci that are associated with fracture. Specifically,
targeting highly heritable fractures, such as peri-men –
opausal wrist fractures, even small numbers of cases
and controls have generated genome-wide-significant
findings32.
The largest study for assessing the impact of BMD
SNPs on fracture was the GEFOS -2 effort, which
involved 31,016 cases and 102,444 controls from 50
independent studies26. The fracture definition used was
any type of fracture — thus it was designed to be inclu –
sive but was imprecise for fragility fractures — with the
intention of increasing statistical power. Although this
may produce results that are not specific to a skeletal
site or mechanism, it did demonstrate improved power
when compared to subclassification of fractures into
vertebral or non-vertebral fractures. Fourteen loci were
associated with any type of fracture, accounting for the
number of loci tested, of which six reached a genome-
wide-significant level ( P < 5 × 10−8). Interestingly, the
RANK–RANKL–OPG pathway, which is important
to BMD genetics, did not appear to have an influence
on risk of fracture. However, therapeutic targeting of
the RANKL–RANKL–OPG pathway clearly influences
fracture risk36, highlighting that false negatives do occur
in GW ASs.Summary of GWASs for fracture. In summary, the
genetics of fracture risk remains poorly understood,
and much progress is likely to be made through the
dissection of fracture risk, independently of BMD.
Future large-scale studies are required to address this
issue using fractures from the most heritable age range.
It is currently unclear which fracture phenotype will
bear the most fruit and at which site, but preliminary
evidence from GEFOS -2 suggests that an inclusive
phenotype definition, permitting inclusion of differ –
ent sites and different levels of fracture validation, with
an intent to increase sample size, may yield the most
power. However, given the age-dependent heritability
of fracture, it is likely that the most heritable fractures
will provide the greatest insights into the genetics of
this common and costly disease.
Insights into pathophysiology
The first promise of osteoporosis genetics was to
highlight proteins and pathways that are crucial
to its pathophysiology, thereby increasing our under –
standing of this condition and providing potential
treatments37. As it is still too early to understand the
function of novel proteins identified by GW ASs, we
will discuss the sequestering of the identified loci into
general pathways for proteins of better-known func –
tion, which were generally described before GW ASs.
In general, the non-novel pathways highlighted by
GW ASs include WNT, RANK–RANKL–OPG and
endochondral ossification. This Review is not meant
to describe these pathways and their interactions in
detail, and readers are referred elsewhere for these
purposes38–43. The importance of these pathways in
osteoporosis pathophysiology has been highlighted
both through loss-of-function mutations and gain-
of-function mutations can of course act in oppo –
site directions. Although other pathways certainly
influence bone physiology, we will restrict our focus
to these three pathways as they were most clearly
highlighted by GW ASs. TABLE 3 lists genes that are
present in other pathways, which have no previous
association with bone physiology. Below we outline
the gene expression, rodent knockouts and Mendelian
diseases that have improved our understanding of
the central importance of these pathways in bone
physiology.
Although below we have highlighted the relevance
of these pathways to bone biology, there is an impor –
tant body of work demonstrating the relevance of the
endocrine system as well as the central role of
the immune system in the development of osteopo –
rosis, among other tissues. This is of particular rel –
evance, as tissues such as lymphocytes and adipose
cells are more easily accessible than bone tissues that
are normally procured through bone biopsy. Thus,
although bone cells are of clear importance in under –
standing the relationship of identified alleles with
cis-expression quantitative trait loci ( cis-eQTLs), it is
important to consider whether other cell types may
be of more physiologic relevance for the transcript
that is under study.REVIEWS
580 | AUGUST 2012 | VOLUME 13 www.nature.com/reviews/genetics
© 2012 Macmillan Publishers Limited. All rights reserved

Osteoporosis-pseudoglioma
syndrome
An autosomal recessive
disorder conferring juvenile
osteoporosis and
juvenile-onset blindness that
is caused by mutations in
lipoprotein-receptor-related
protein 5 ( LRP5 ).WNT. The WNT signalling pathway is crucial to bone
biology, as it is important for embryogenesis and limb
development. WNT factors exert their mechanism by
binding to cell surface receptors, activation of which
leads to accumulation of β-catenin and its subse –
quent translocation to the nucleus, where it acts as a
transcriptional co -activator38. GW ASs for osteoporo –
sis have identified the following likely WNT-related
genes: catenin (cadherin-associated protein), beta 1
(CTNNB1 ), sclerostin (SOST ), low-density lipopro –
tein receptor-related protein 4 ( LRP4 ), LRP5 , GPR177 ,
wingless-type MMTV integration site family, mem –
ber 4 ( WNT4 ), WNT5B , WNT16 , dickkopf 1 ( DKK1 ),
secreted Frizzled-related protein 4 ( SFRP4 ), R-spondin 3
(RSPO3 ), jagged 1 ( JAG1 ), myocyte enhancer factor 2C
(MEF2C ) and AXIN1 . FIGURE 1 shows the probable
mechanism of the WNT pathway and displays that most
of the key proteins of the canonical WNT signalling
pathway were identified through GW ASs.
The first unequivocal demonstration of the impor –
tance of the WNT pathway in human bone physiology
was through a human genetics study that identified
inactivating mutations of LRP5 as causing the recessive
osteoporosis-pseudoglioma syndrome , which pre –
sents with low bone mass and fractures44. Although high-bone-mass syndromes have been associated with
activating mutations of LRP5 (REF.  45), inactivating
mutations of sclerostin ( SOST ) also present with high-
bone-mass syndromes46, and MEF2C potentiates SOST47
expression, thus demonstrating the relevance of both
gain- and loss-of-function mutations in understanding
osteoporosis aetiology. Of the loci in the WNT pathway
identified by GW ASs, transcript levels of DKK1 and
SOST in trans -iliac bone biopsies are shown to correlate
with BMD at the lumbar spine and femoral neck in direc –
tions that are consistent with their presumed function26.
At locus 1p36.12, cis-variant rs6426749[G] correlated
with reduced WNT4 expression in fibroblasts, osteo –
blasts and adipose tissue, and at 11p11.2, rs7932354[C]
correlates with increased LRP4 cis- expression in adipose
tissue26. Other members of the WNT pathway identi –
fied by GW ASs have monogenic skeletal phenotypes
in humans and/or mice. TABLE 3 catalogues the human
and mouse knockout data for GW AS-identified genes.
The WNT pathway also directly interacts with other
important bone pathways. For example, jagged 1 (JAG1)
is not only a WNT and β-catenin target but is also an
important component of the NOTCH signalling path –
way48. Mutations in JAG1 — a NOTCH ligand — that
lead to haploinsufficiency cause Alagille’s syndrome49, Table 3 | Evidence for the role of GWAS-identified genes in generating skeletal phenotypes
Pathway Gene Human monogenic skeletal syndrome Mouse knockout
WNT SOST Sclerosteosis MGI:1921749
LRP5 Osteoporosis-pseudoglioma syndrome,
osteopetrosis autosomal dominant 1MGI:1278315
WLS (GPR177 ) MGI:1915401
CTNNB1 MGI:88276
RSPO3 MGI:1920030
DKK1 MGI1329040
LRP4 Cenani–Lenz syndactyly syndrome MGI:2442252
AXIN1 MGI:1096327
WNT3 Tetra-amelia, autosomal recessive
JAG1 Alagille’s syndrome
RANK–RANKL–OPG RANKL Osteopetrosis, autosomal recessive 2 MGI:1100089
OPG Paget’s disease MGI:109587
RANK Paget’s disease MGI:1314891
Endochondral
ossificationPTHLH (encodes PTHRP) Brachydactyly, type E2 MGI:97800
RUNX2 MGI:99829
SP7 Osteogenesis imperfecta, type XII MGI:2153568
IBSP (encodes BNSP2) MGI:96389
SPP1 (encodes
osteopontin)MGI:98389
SOX6 MGI:98368
SOX9 Acampomelic campomelic dysplasia MGI:98371
This table is modified from REF.  26. This list is not meant to be exhaustive of all genome-wide association study (GWAS)-identified
genes but rather is intended to highlight certain genes clustering in the pathways discussed. BNSP2, bone sialoprotein 2; DKK1 ,
dickkopf 1; JAG1 , jagged 1; LRP4 , low-density lipoprotein receptor-related protein 4; OPG , also known as TNFRSF11B ; PTHLH ,
parathyroid hormone-like hormone; PTHRP , parathyroid hormone-related protein; RANK , also known as TNFRSF11A ; RANKL , also
known as TNFSF11 ; RSPO3 , R-spondin 3; RUNX2 , runt-related transcription factor 2; SOST , sclerostin; TNFRSF11 , tumour necrosis
factor receptor superfamily, member 11; WLS , wntless; WNT3 , wingless-type MMTV integration site family, member 3. REVIEWS
NATURE REVIEWS | GENETICS VOLUME 13 | AUGUST 2012 | 581
© 2012 Macmillan Publishers Limited. All rights reserved

Nature Reviews | GeneticsDKK1SOSTSFRP4
*‡ *‡
‡‡
‡ RSPO3Frizzled
WNT
‡AXIN1*‡*‡
Frizzled
LRP5
LRP6WNT
*‡WNT factors identified
by GWASs: WNT3 , WNT4 ,
WNT5B , WNT16 ,
involved in WNT
secretion WLS (GPR177 )
Cell membrane
β-catenin‡β-catenin‡
APCAXIN1
PP
Transcriptional co-activation
of genes including JAG1 Nucleus‡β-cateninN
NCPPa Inhibited state b Activated state
LRP5
LRP6
C
N CN CGSK3 β
and the NOTCH pathway directly influences members
of the endochondral ossification pathway (such as runt-
related transcription factor 2 (RUNX2) and osterix (also
known as SP7))50, as described in more detail below.
RANK–RANKL–OPG. Bone is a constantly remodel –
ling organ that is dependent on a cyclical resorption and
reforming process involving osteoclasts and osteoblasts
coordinated by the RANK–RANKL–OPG pathway.
Osteoblasts and osteocytes secrete a soluble factor called
RANKL (also known as TNFSF11)51. RANKL then
interacts with the RANK (also known as TNFRSF11A)
receptor, which is present on the osteoclast precursor
cell, leading to the migration, differentiation and fusion
of osteoclastic lineage cells52 (FIG. 2 ). However, to modu –
late this process intricately, osteoblasts and osteocytes
also produce a decoy receptor for RANKL called
OPG (also known as TNFRSF11B), which prevents its
binding to RANK, thereby halting the bone resorption cycle53. GW ASs for BMD have identified all three of
these important proteins in bone biology.
Murine knockout of OPG leads to severe osteopo –
rosis54. Furthermore, recent conditional deletion of
RANKL in mice showed that osteocytes are the inte –
gral source of RANKL, and its absence led to a twofold
increase in femoral cancellous bone volume51, which
is the expected direction of effect, as RANKL leads to
increased bone resorption (FIG. 2 ).
At locus 8q24.12, the expression of OPG in lympho –
cytes was positively correlated with BMD at the lum –
bar spine20. Likewise, allelic expression indicated that
SNPs in this region were associated with OPG expres –
sion in lymphocytes but not in primary osteoblasts.
Furthermore, for carriers of the top-risk allele near
OPG , expression of this protein was reduced by half in
lymphoblast cell lines. Although lymphoblast cell lines
are not bone tissue, lymphocytes are the main source
of OPG expression in bone marrow55. Similarly, leptin Figure 1 | Simplified depiction of members of the canonical WNT signalling pathway identified through
genome-wide association studies for bone mineral density. Proteins identified through genome-wide
association studies (GWASs) are indicated in bold font and with a bold outline. Inhibitory proteins of the WNT pathway
are in red, and activators of the WNT pathway are in green. a | The main role of the WNT signalling pathway is to control
the stability and subsequent abundance of β-catenin, the role of which is to activate gene transcription in the nucleus.
In the absence of WNT factors, β-catenin is phosphorylated by glycogen synthase kinase 3 β (GSK3 β), preventing the
translocation of β-catenin to the nucleus. Sclerostin (SOST), dickkopf 1 (DKK1) and secreted Frizzled-related protein
(SFRP4) act by inhibiting the interaction between Frizzled family members, low-density lipoprotein receptor-related
protein 5 (LRP5) and WNT. b | WNT proteins bind to the G-protein-coupled receptor Frizzled and LRP5 to form a
complex that ultimately leads to the recruitment of AXIN1 to the LRP5 co-receptor. R-spondin 3 (RSPO3) acts to disrupt
DKK1 association to LRP6 (REF.  78). This inhibits the degradation of the AXIN1– β-catenin complex and promotes the
translocation of β-catenin to the nucleus. Jagged 1 ( JAG1 ) has been reported to be a target of β-catenin transcriptional
control and is also an important component of the NOTCH signalling pathway. *Indicates the relevance of the gene to
human monogenic skeletal disease. ‡Indicates genes with evidence arising from mouse knockouts. The figure is
adapted, with permission, from REF. 38 © (2009) Macmillan Publishers Ltd. All rights reserved.REVIEWS
582 | AUGUST 2012 | VOLUME 13 www.nature.com/reviews/genetics
© 2012 Macmillan Publishers Limited. All rights reserved

Nature Reviews | Genetics*‡
*‡
Osteoblast
OsteocyteOPGMacrophage
Bone formation Bone resorptionRANKLRANKPre-fusion
osteoclastMultinucleated osteoclast
Activated osteoclast
Osteopetrosis
A syndrome of high bone mass
caused by an imbalance in the
constant remodelling of bone,
which favours bone formation,
or mutations leading to
increased bone formation,
such as activating mutations
in low-density lipoprotein
receptor-related protein 5
(LRP5 ).
Paget’s disease
Focalized bone lesions
characterized by enhanced
bone remodelling and resultant
overgrowth of bone, leading to
an increased risk of fracture.
Cleidocranial dysplasia
An autosomal dominant
condition characterized by
defective differentiation of
osteoblasts, resulting in
impaired bone formation, short
stature and abnormal teeth
owing to mutations of
runt-related transcription
factor 2 ( RUNX2 ), which
encodes core-binding factor
alpha 1.is produced only in adipocytes, and its administration
to patients with congenital leptin deficiency increases
bone mineral mass considerably while simultaneously
reducing body weight56. Thus, we are reminded that the
effect of altered expression in tissues apart from bone
must be considered.
Osteopetrosis is a syndrome of failed osteoclastic
bone resorption and leads to a high bone mass through
a disruption of the bone-remodelling cycle. Inactivating
mutations in both RANK , RANKL cause osteopetro –
sis57,58. Interestingly, mutations in TNFRSF11B cause
juvenile Paget’s disease59, and a recent GW AS for adult
Paget’s disease identified variants in TNFRSF11A60.
Endochondral ossification. Mature bone is formed
through the ossification of the cartilaginous skeleton
(reviewed elsewhere61). The parathyroid hormone-
related peptide (PTHRP), a ligand that binds the
parathyroid hormone 1 receptor encoded by parathy –
roid-hormone-like hormone ( PTHLH ) is important
for the development of the cartilage growth plate62, and
the transcription factor SOX6 is required for the estab –
lishment of the cartilage growth plate, allowing for the
development of endochondral bone63. The transcription
factor SOX9 regulates expression of collagen, type II,
alpha 1 (COL2A1 ), which encodes the major cartilage
matrix protein. All of these genes — which are essential
for cartilage development, except COL2A1 — were iden –
tified through BMD GW ASs (FIG. 3 ). After cartilage has
formed, it is then ossified through the deposition of min –
eral. Intriguingly, disruption of a single transcription fac –
tor — namely, RUNX2 — is able to prevent ossification of the cartilaginous skeleton completely40. Osterix acts
downstream of RUNX2 and is required for full differ –
entiation of osteoblasts64. Integrin-binding sialoprotein
(IBSP , which encodes bone sialoprotein 2 (BNSP2))
and secreted phosphoprotein 1 ( SPP1 , which encodes
osteopontin (OPN)) both bind strongly to calcium and
hydroxyapatite and may have a role in the adherence of
osteoclasts to the bone surface. GW ASs also identified
all of the above proteins (FIG. 3 ).
Disruption of RUNX2 results in the human monoge –
netic syndrome of cleidocranial dysplasia , which is char –
acterized by delayed skeletal development and absent
clavicles65, whereas murine knockouts die at birth owing
to a softened ribcage that is unable to support respira –
tion. In addition, osterix-null mice do not form bone64.
At the GW AS-identified locus 4q22.1 (TABLE 1 ), cis-
SNP rs6532023[G] was correlated with reduced SPP1
expression in adipose tissue26. These data suggest that
identified GW AS loci are involved in endochondral
ossification.
Identification of drug targets
Another promise of osteoporosis genetics has been the
delivery of clinically relevant drug targets that can be
manipulated pharmaceutically to prevent osteoporotic
fractures. Osteoporotic fractures represent a reasonable
disease class for preventive intervention as they occur
later in life, are common and incur a substantial financial
burden. Furthermore, BMD and clinical risk factors are
able to identify groups of individuals who are at high
risk for fracture, thus justifying preventive therapy for
individuals at risk3.Figure 2 | Simplified depictions of members of the RANK–RANKL–OPG signalling pathway identified through
genome-wide association studies for bone mineral density. Proteins identified as genome-wide-significant
through genome-wide association studies (GWASs) are indicated in bold font and with a bold outline. RANK is encoded
by tumour necrosis factor receptor superfamily, member 11a ( TNFRSF11A ), its ligand RANKL is encoded by TNFSF11 , and
the decoy receptor OPG is encoded by TNFRSF11B. To generate activated osteoclasts, RANKL is secreted by osteoblasts
and osteocytes in bone, and these bind to its natural receptor, RANK, on the surface of pre-fusion osteoclasts. To
fine-balance this activation system, osteoblasts and osteocytes also secrete OPG, which is a natural decoy receptor for
RANKL and prevents binding of RANKL to RANK. *Indicates the relevance of the gene to human monogenic skeletal
disease. ‡Indicates genes with evidence arising from mouse knockouts. The figure is adapted, with permission, from
REF. 79 © (2008) Health Plexus Ltd.REVIEWS
NATURE REVIEWS | GENETICS VOLUME 13 | AUGUST 2012 | 583
© 2012 Macmillan Publishers Limited. All rights reserved

Nature Reviews | GeneticsPeriosteal
bone collarPrimary
centre of
ossification
Cartilage
Bone
Bone marrow (including
blood vessels)PTHLH
SOX6
SOX9
COL2A1*‡*‡‡*‡ RUNX2
SP7
BNSP2
OPN‡‡*‡‡OssificationCartilage
formation
Because the development of a novel therapeu –
tic requires over 10 years in a drug pipeline, it is too
early to judge whether genetic studies for osteoporo –
sis have identified loci for which manipulation will
influence patient care. However, a simplistic way to
address this question theoretically is to ask whether
GW ASs for osteoporosis have been able to identify targets for drugs that have been shown to be effica –
cious in phase III trials or that are in preclinical tri –
als. If GW ASs are then able to identify known drug
targets, it remains possible that among the novel loci
identified, there will be potential targets for the phar –
maceutical industry.
The current drugs that are available for the treat –
ment of osteoporosis, and their most probable targets,
are listed in TABLE 4 . All of these drug targets were iden –
tified in the pre-GW AS era and therefore provide an
opportunity to test whether GW ASs can identify clini –
cally validated drug targets. Of the eight drugs classes
currently used to treat osteoporosis, or showing prom –
ise in drug development (such as DKK1 inhibitors66,
sclerostin inhibitors67 and cathepsin K inhibitors68),
the targets of five of these drugs were directly identi –
fied through GW ASs for BMD. The pathway of a sixth
drug class, parathyroid hormone analogues, was high –
lighted through the identification of the PTHLH. This
shows that among the GW AS loci, there is a ~267 -fold
enrichment for validated drug targets in humans
(P = 2.5 × 10−15), assuming there are 20,000 genes in the
genome. Thus, GW ASs for BMD have been able to iden –
tify most clinically relevant drug targets for osteoporosis
through an agnostic scan of the genome. It therefore
is possible that there are useful drug targets among
the dozens of loci that have not yet been thoroughly
investigated.
Some observers have suggested that GW AS findings
will be unlikely to be helpful in the design of therapies or
in the understanding of common disease, partly because
of the small effect sizes that have been attributed to the
common variants under study69. However, these com –
ments miss the point. What is of interest is not the effect
of any single common variant that has survived selec –
tion pressure to become common, but rather the demon –
stration that even a marginal modification of a protein’s
function through a change in one allele can influence a
disease phenotype. The findings from GW ASs of osteo –
porosis suggest that the identification of such proteins
using tools as blunt as common variants can identify
clinically relevant drug targets.
Osteoporosis and fracture prediction
The third general objective of osteoporosis genet –
ics has been to identify a set of variants that can be
measured to allow identification of groups at high risk
for future fracture. This is of particular relevance to
osteoporosis as safe interventions exist for osteoporo –
sis, and such preventive therapy years before disease
onset could decrease the population health burden of
this disease.
Because many genome-wide-significant alleles
have been identified, a weighted allele score can be
implemented, which simply counts the number of
deleterious alleles per person, weighting each allele
by the effect size that is attributed to this allele in an
independent population cohort. Individuals with more
deleterious alleles would therefore have a higher risk
score. Indeed, using 15 genome-wide-significant SNPs
for lumbar spine BMD in the GEFOS -1 GW AS, this Figure 3 | Simplified depiction of members of the
endochondral ossification pathway identified through
genome-wide association studies for bone mineral
density. Proteins identified as genome-wide-significant
through genome-wide association studies (GWASs) are
indicated in bold font. Bone is generated in the
developing skeleton by first forming cartilage, which is
then ossified, starting at the primary centre of ossification
and moving outwards to the peripheral bone. Arrows
indicate genes that are involved in the promotion of a
process. Genes involved in the formation of cartilage are
indicated in the left panel. Parathyroid-hormone-related
protein (PTHRP), which is encoded by the gene PTHLH ,
binds to the PTHRP receptor to promote development of
the cartilage growth plate. The transcription factor SOX6
is involved in the establishment of cartilage growth plate,
allowing for the development of endochondral bone.
The transcription factor SOX9 regulates collagen, type II,
alpha 1 ( COL2A1 ) expression, the product of which is a
structural protein that is the main component of
cartilage. Genes involved in ossification are depicted in
the right-hand panel. RUNX2 is a transcription factor that
is a regulator of ossification of cartilaginous skeleton,
SP7 is a transcription factor that permits differentiation
of osteoblasts, and BNSP2 is a major noncollagenous
structural protein. Osteopontin (OPN) is a secreted
protein that permits the attachment of osteoclasts to
mineralized bone. *Indicates the relevance of the gene
to human monogenic skeletal disease. ‡Indicates genes
with evidence arising from mouse knockouts. The figure
is adapted, with permission, from REF. 80 © (2001) Society
for Endocrinology.REVIEWS
584 | AUGUST 2012 | VOLUME 13 www.nature.com/reviews/genetics
© 2012 Macmillan Publishers Limited. All rights reserved

consortium was able to show a difference between the
highest risk group and lowest risk group of approxi –
mately 0.7 standard deviations25. Increasing the num –
ber of SNPs by including those identified in the larger
GEFOS -2 effort to 63 autosomal SNPs increased this
gradation of effect to approximately 0.86 standard
deviations26. Comparing these two findings sug –
gests that a substantial increase in the number of risk
alleles (of decreasing effect size) does not dramatically
improve the ability of a weighted allelic risk score to
partition individuals.
More clinically relevant, however, was the dem –
onstration of the effect of this allelic risk score on the
receiver operating characteristic (ROC) curve (BOX 1 ) for
fracture risk. A considerable body of work has already
gone into understanding the relevance of clinical risk
factors (such as age, sex and weight) and BMD in
risk stratification for fracture. Using only well-validated
clinical risk factors, irrespective of genotype, the area
under the ROC curve for osteoporotic hip fracture is
0.83 (REF.  70).
However, the area under the ROC curve for frac –
ture risk using the allelic risk score without any clinical
information was marginally better than chance alone
and had a value of 0.57 (95% confidence interval: 0.55–
0.59). Similarly, the allelic risk score for a diagnosis of
osteoporosis (BMD T score ≤ −2.5) had an area under
the curve of 0.59 (95% confidence interval: 0.56–0.61),
which was again worse than a risk score including
just age and weight (0.75 (95% confidence interval:
0.73–0.77)) and improved only marginally when add –
ing the allelic risk score to age and weight (0.76 (95%
confidence interval: 0.74–0.78)) (BOX 1 ).
Some observers have suggested that to predict risk of
disease, 150 genes with odds ratios of 1.5 or 250 genes
with odds ratios of 1.25 will be needed71. Recent data that
show the combined effect of many common variants on
the explanation of trait variance may improve our ability
to prognosticate osteoporosis72. However, the published
data from the field of osteoporosis suggest that simple
clinical risk factors, such as age, weight and height,
outperform an allelic risk score that is comprised of
susceptibility alleles.These findings may be improved in the future by
identifying a set of alleles that influence risk of fracture
independently of BMD or by identifying less common
variants that have a large effect on the risk of fracture
and/or BMD. However, the highly polygenic allelic
architecture of BMD and the low variance explained
of this trait suggest, at present, that the reliable predic –
tion of individuals at risk for fracture or osteoporosis
using genetic information is not feasible.
Conclusions, perspectives and future studies
In the past 4 years since the first GW ASs for osteopo –
rosis were published, great advances have been made,
with the identification of 62 genome-wide-significant
loci for this common and costly disease. Many of the
identified proteins have clear and relevant mechanisms
of action for osteoporosis pathophysiology. Further,
the genes identified have mirrored many of the cur –
rent therapeutic drug targets, suggesting that among
the novel genes identified, there are opportunities for
new therapeutic approaches. Despite these successes,
the use of these identified variants for diagnosis and
risk prediction has been disappointing.
However, the clinical use of these findings is dem –
onstrated by the ability of GW ASs for BMD to high –
light loci that are validated drug targets in humans.
Through an agnostic scan of the genome, GW ASs
were able to identify the drug target for five of eight
validated drugs for osteoporosis and highlighted
the pathway of a sixth. This tremendous enrich –
ment (~250-fold) of the human genome for known
drug targets suggests that among the novel BMD loci
there are likely to be other clinically relevant drug
targets.
Almost all of these advances have been made by
first identifying variants that are associated with BMD.
However, paradoxically, much of the risk of the clini –
cally relevant endpoint — that is, fracture — is inde –
pendent of BMD. Therefore, well-powered consortia
GW ASs that focus on the heritable age range for frac –
ture will further help to address the use of genetics
in understanding the pathophysiology, therapeutic
options and prognosis of this burdensome disease.Table 4 | Drugs, drug targets and whether the locus encoding the target was identified through GWASs
Drug class Drug target Target locus identified through GWASs Refs
Denosumab RANKL RANKL 36
Sclerostin inhibitors Sclerostin (SOST) SOST 67
Selective oestrogen receptor
modulatorsOestrogen receptor ESR1 83
Parathyroid hormone
analoguesParathyroid hormone
receptorNot identified, but the pathway has been
highlighted through PTHLH (encodes PTHRP)84,85
Bisphosphonates Farnesyl pyrophosphate Not identified 86
Oestrogen Oestrogen receptor ESR1 87
Cathepsin K inhibitors Cathepsin K Not identified 68
DKK1 inhibitors DKK1 DKK1 66
DKK1, dickkopf 1; GWASs, genome-wide association studies; PTHLH , parathyroid hormone-like hormone; PTHRP , parathyroid
hormone-related protein.REVIEWS
NATURE REVIEWS | GENETICS VOLUME 13 | AUGUST 2012 | 585
© 2012 Macmillan Publishers Limited. All rights reserved

Sensitivity
1 − specificity0.4 0.6 1.0 0.2 0.8 01.0
0.8
0.6
0.4
0.2
0
Nature Reviews | GeneticsOsteoporosis: genetic score
Osteoporosis: age weight
On the basis of current evidence, it is tempting to
suggest that BMD has an infinitesimal allelic architec –
ture, such that there are many thousands of genetic
variants at hundreds to thousands of genes that all
impart a small influence on BMD. If this were the
case, then larger and therefore better powered GW ASs
would continue to uncover loci that are of relevance
to the three aims of osteoporosis genetics. These stud –
ies should be done but with the prior appreciation that
the alleles identified from such additional studies will
probably have smaller and smaller effects. The current
evidence from GW ASs, however, does not preclude
the important contribution of rare genetic variation to
osteoporosis.
Researchers should therefore turn their attention
away from common genetic variation and explore the
extent to which rare genetic variation underpins the her –
itability of osteoporosis. However, for such studies, large
sample sizes will also be required, given the reduced
power for single SNPs, even those of large effect73. Fine-
mapping studies for GW AS loci will also be helpful in
improving our understanding of the pathophysiology of
this disease, particularly when there are several genes
at a highlighted locus. Currently, large-scale sequenc –
ing studies that involve thousands of individuals are
underway in several centres, and these should address
this question shortly.Low-coverage whole-genome sequencing, such as
that progressing within the UK10K project, will pro –
vide direct insights into the role of rare variation in
non-coding regions. This may have particular rele –
vance to osteoporosis as many of the disease-associated
SNPs have resided outside coding sequences. However,
in the short term, there will be more whole-exome
sequence data generated, considering the substan –
tially lower cost of these studies. Whichever region
of the genome is interrogated for rare variation, it
is of paramount importance that these findings are
well replicated, given the modest statistical power of
most rare variant association methods73. This is
of particular concern for non-coding regions, in which
the unit of analysis is less circumscribed than in cod –
ing regions, leading to a more arbitrary set of multiple
tests. The ‘exome chip’ , which is designed to genotype
amino-acid-changing polymorphisms, may be of par –
ticular relevance as this study design obviates the need
for sequence alignment and calling and will provide
a standardized set of variants across studies. Finally,
extreme phenotypes may be helpful for rare variant
studies and have already been collected for fracture
and BMD29,74.
Regardless of which approach is taken for rare vari –
ants for osteoporosis, their results will be of particular
interest to the wider genomics community, given the
success of mapping this trait to regions of the genome,
its medical use and the availability of centralized DNA
collections for replication.
Epigenetics is another potential source of miss –
ing heritability in all areas of age-related diseases, and
osteoporosis is no exception75. This field has recently
expanded from rare diseases to common complex traits
with the advent of methylation arrays and immunopre –
cipitation methods, such as methylated DNA immuno –
precipitation sequencing (MeDIP-seq). For osteoporosis,
there is a paucity of studies mainly owing to difficult
access to bone tissue. However, there appears to be
some overlap between methylation changes across dif –
ferent tissues — with correlations of approximately 70%
between blood and buccal tissue76. Advances in epige –
netics may alter approaches towards drug therapy and
may provide insight into therapeutic responses earlier
than changes in bone density. How these advances will
alter care will depend, however, on the relative costs of
the drugs and prognostic tests used.
Other possible genetic avenues worth exploring
include the role of copy number variants, which despite
having been difficult to identify so far have shown
promise in neuro-developmental diseases and could still
have large effects in complex traits such as osteoporosis.
Finally, the use of more sophisticated associated endo-
phenotypes, such as high-throughput metabolomics and
proteomics, could dramatically increase power to detect
genes in GW ASs77.
Given the history of collaboration and coordination
of large-scale studies within the osteoporosis genetics
community, these future studies have a high likelihood
of dissecting the contribution of rare and epigenetic
variants for osteoporosis.Box 1 | Receiver operating characteristic curves for risk of osteoporosis
Clinicians who are faced
with a multitude of tests
to help discern whether
a patient is diseased
or at risk of disease are
often aided by
identifying a cutoff for
a diagnostic test that
minimizes both false
positives and false
negatives. A measure of
the ability of a screening
test to reduce both false
positives and false negatives
for binary outcomes, such
as fracture or diagnosis of
osteoporosis, is the receiver
operating characteristic
(ROC) curve. The ROC curve
compares the proportion of
truly diseased individuals correctly identified as diseased (that is, the sensitivity)
against the proportion of false positives (that is, one minus the specificity).
Therefore, an ideal test would have a threshold that maximizes sensitivity while
minimizing the proportion of false positives. Simply measuring the area under this
ROC curve can then provide a gauge of the clinical use of such a test. If the test were
no better than chance alone, then the area under the ROC curve would be 0.5, and a
near-perfect test would have an area under the ROC curve of 0.99. The figure shows
a ROC curve for risk of osteoporosis as derived from a weighted allelic risk score,
where weights were derived from 61 autosomal genome-wide-significant SNPs for
bone mineral density and applied to an exterior cohort. Note that the area under the
ROC curve is better when age and weight are used alone than the genetic risk score.
The figure is adapted, with permission, from REF. 26 © (2012) Macmillan Publishers
Ltd. All rights reserved.REVIEWS
586 | AUGUST 2012 | VOLUME 13 www.nature.com/reviews/genetics
© 2012 Macmillan Publishers Limited. All rights reserved

1. Kanis, J. A. Diagnosis of osteoporosis. Osteoporos. Int.
7 (Suppl. 3), S108–S116 (1997).
2. Burge, R. et al. Incidence and economic burden of
osteoporosis-related fractures in the United States,
2005–2025. J. Bone Miner. Res. 22, 465–475
(2007).
3. Kanis, J. A., Johnell, O., Oden, A., Johansson, H. &
McCloskey, E. FRAX and the assessment of fracture
probability in men and women from the UK.
Osteoporos. Int. 19, 385–397 (2008).
4. Slemenda, C. W. et al. The genetics of proximal
femur geometry, distribution of bone mass and bone
mineral density. Osteoporos. Int. 6, 178–182
(1996).
5. Smith, D. M., Nance, W. E., Kang, K. W., Christian, J. C.
& Johnston, C. C. Jr. Genetic factors in determining
bone mass. J. Clin. Invest. 52, 2800–2808 (1973).
6. Arden, N. K., Baker, J., Hogg, C., Baan, K. &
Spector, T . D. The heritability of bone mineral density,
ultrasound of the calcaneus and hip axis length: a
study of postmenopausal twins. J. Bone Miner. Res.
11, 530–534 (1996).
7. Marshall, D., Johnell, O. & Wedel, H. Meta-analysis of
how well measures of bone mineral density predict
occurrence of osteoporotic fractures. BMJ 312,
1254–1259 (1996).
8. Jergas, M. & Gluer, C. C. Assessment of fracture risk
by bone density measurements. Semin. Nucl. Med.
27, 261–275 (1997).
9. Andrew, T ., Antioniades, L., Scurrah, K. J.,
Macgregor, A. J. & Spector, T . D. Risk of wrist fracture
in women is heritable and is influenced by genes that
are largely independent of those influencing BMD.
J. Bone Miner. Res. 20, 67–74 (2005).
10. Michaelsson, K., Melhus, H., Ferm, H., Ahlbom, A. &
Pedersen, N. L. Genetic liability to fractures in the
elderly. Arch. Intern. Med. 165, 1825–1830 (2005).
11. Kaufman, J. M. et al. Genome-wide linkage screen of
bone mineral density (BMD) in European pedigrees
ascertained through a male relative with low BMD
values: evidence for quantitative trait loci on
17q21 -23, 11q12 -13, 13q12 -14, and 22q11. J. Clin.
Endocrinol. Metab. 93, 3755–3762 (2008).
12. Hsu, Y. H. et al. Variation in genes involved in the
RANKL/RANK/OPG bone remodeling pathway are
associated with bone mineral density at different skeletal
sites in men. Hum. Genet. 118, 568–577 (2006).
13. Peacock, M. et al. Bone mineral density variation in
men is influenced by sex-specific and non sex-specific
quantitative trait loci. Bone 45, 443–448 (2009).
14. Xiao, P . et al. Genomic regions identified for BMD
in a large sample including epistatic interactions
and gender-specific effects. J. Bone Miner. Res. 21,
1536–1544 (2006).
15. Ioannidis, J. P . et al. Meta-analysis of genome-wide
scans provides evidence for sex- and site-specific
regulation of bone mass. J. Bone Miner. Res. 22,
173–183 (2007).
16. Ralston, S. H. & Uitterlinden, A. G. Genetics of
osteoporosis. Endocr. Rev. 31, 629–662 (2010).
17. Richards, J. B. et al. Collaborative meta-analysis:
associations of 150 candidate genes with osteoporosis
and osteoporotic fracture. Ann. Intern. Med. 151,
528–537 (2009).
18. Ferrari, S. L. et al. Polymorphisms in the low-density
lipoprotein receptor-related protein 5 ( LRP5 ) gene are
associated with variation in vertebral bone mass,
vertebral bone size, and stature in whites. Am. J. Hum.
Genet. 74, 866–875 (2004).
19. van Meurs, J. B. et al. Large-scale analysis of
association between LRP5 and LRP6 variants and
osteoporosis. JAMA 299, 1277–1290 (2008).
20. Richards, J. B. et al. Bone mineral density, osteoporosis,
and osteoporotic fractures: a genome-wide association
study. Lancet 371, 1505–1512 (2008).
Together with reference 21, these papers were the
first GWASs for osteoporosis, and they made it
clear that there were no common genetic variants
of large effect for osteoporosis.
21. Styrkarsdottir, U. et al. Multiple genetic loci for bone
mineral density and fractures. N. Engl. J. Med. 358,
2355–2365 (2008).
22. Styrkarsdottir, U. et al. New sequence variants
associated with bone mineral density. Nature Genet.
41, 15–17 (2009).
23. Timpson, N. J. et al. Common variants in the region
around Osterix are associated with bone mineral
density and growth in childhood. Hum. Mol. Genet.
18, 1510–1517 (2009).
This is the only GWAS for BMD that has been
conducted in children.24. Xiong, D. H. et al. Genome-wide association and
follow -up replication studies identified ADAMTS18
and TGFBR3 as bone mass candidate genes in
different ethnic groups. Am. J. Hum. Genet. 84,
388–398 (2009).
25. Rivadeneira, F. et al. T wenty bone-mineral-density loci
identified by large-scale meta-analysis of genome-wide
association studies. Nature Genet. 41, 1199–1206
(2009).
This is the first large-scale consortium-style GWAS
for osteoporosis. This paper identified many
clinically relevant drug targets for osteoporosis and
substantially expanded the number of loci
associated with BMD.
26. Estrada, K. et al. Genome-wide meta-analysis
identifies 56 bone mineral density loci and reveals 14
loci associated with risk of fracture. Nature Genet. 44,
491–501 (2012).
This is the largest global effort to describe the
genetic determinants of BMD and fracture. Using
BMD loci, this paper was also the first to assess
the contribution of BMD GWAS loci to fracture in a
large sample size.
27. Morris, A. P . T ransethnic meta-analysis of genomewide
association studies. Genet. Epidemiol. 35, 809–822
(2011).
28. Dastani, Z. et al. Novel loci for adiponectin levels and
their influence on type 2 diabetes and metabolic
traits: a multi-ethnic meta-analysis of 45,891
individuals. PLoS Genet. 8, e1002607 (2012).
29. Duncan, E. L. et al. Genome-wide association study
using extreme truncate selection identifies novel
genes affecting bone mineral density and fracture risk.
PLoS Genet. 7, e1001372 (2011).
This is an important paper demonstrating the
relative use of selecting extremes of a phenotypic
continuum to identify common genetic variants.
30. Hsu, Y. H. et al. An integration of genome-wide
association study and gene expression profiling to
prioritize the discovery of novel susceptibility loci for
osteoporosis-related traits. PLoS Genet. 6, e1000977
(2010).
31. Zheng, H. et al. WNT16 influences bone mineral
density, cortical bone thickness, bone strength and
osteoporotic fracture risk. PLoS Genet. 5 Jul 2012
(doi:10.1371/journal.pgen.1002745).
Together with reference 30, this paper showed that
forearm and total body BMD had allelic
architectures that appeared to be less polygenic
than lumbar spine and femoral neck sites.
32. Medina-Gomez, C. et al. Meta-analysis of genome-
wide scans for total body BMD in children and adults
reveals allelic heterogeneity and age-specific effects at
the WNT16 locus. PLoS Genet. 5 Jul 2012
(doi:10.1371/journal.pgen.1002718).
33. Frazer, K. A. et al. A second generation human
haplotype map of over 3.1 million SNPs. Nature 449,
851–861 (2007).
34. Visscher, P . M., Hill, W. G. & Wray, N. R. Heritability in
the genomics era—concepts and misconceptions.
Nature Rev. Genet. 9, 255–266 (2008).
35. Guo, Y. et al. Genome-wide association study identifies
ALDH7A1 as a novel susceptibility gene for
osteoporosis. PLoS Genet. 6, e1000806 (2010).
36. Cummings, S. R. et al. Denosumab for prevention of
fractures in postmenopausal women with
osteoporosis. N. Engl. J. Med. 361, 756–765 (2009).
37. Duncan, E. L. & Brown, M. A. Clinical review 2: genetic
determinants of bone density and fracture risk—state
of the art and future directions. J. Clin. Endocrinol.
Metabolism 95, 2576–2587 (2010).
38. Angers, S. & Moon, R. T . Proximal events in
Wnt signal transduction. Nature Rev. Mol. Cell Biol.
10, 468–477 (2009).
39. Kearns, A. E., Khosla, S. & Kostenuik, P . J.
Receptor activator of nuclear factor κB ligand and
osteoprotegerin regulation of bone remodeling in
health and disease. Endocr. Rev. 29, 155–192 (2008).
40. Komori, T . Signaling networks in RUNX2 -dependent
bone development. J. Cell. Biochem. 112, 750–755
(2011).
41. Kubota, T ., Michigami, T . & Ozono, K. Wnt signaling in
bone metabolism. J. Bone Miner. Metabolism 27,
265–271 (2009).
42. Guo, J. et al. Suppression of Wnt signaling by Dkk1
attenuates PTH-mediated stromal cell response and
new bone formation. Cell. Metabolism 11, 161–171
(2010).
43. Akiyama, H. et al. Interactions between Sox9 and
β-catenin control chondrocyte differentiation. Genes
Dev. 18, 1072–1087 (2004).44. Gong, Y. et al. Osteoporosis-pseudoglioma syndrome,
a disorder affecting skeletal strength and vision,
is assigned to chromosome region 11q12 -13.
Am. J. Hum. Genet. 59, 146–151 (1996).
45. Van Wesenbeeck, L. et al. Six novel missense
mutations in the LDL receptor-related protein 5
(LRP5 ) gene in different conditions with an increased
bone density. Am. J. Hum. Genet. 72, 763–771
(2003).
46. Balemans, W. et al. Increased bone density in
sclerosteosis is due to the deficiency of a novel
secreted protein (SOST). Hum. Mol. Genet. 10,
537–543 (2001).
47. Leupin, O. et al. Control of the SOST bone enhancer
by PTH using MEF2 transcription factors. J. Bone
Miner. Res. 22, 1957–1967 (2007).
48. Estrach, S., Ambler, C. A., Lo Celso, C., Hozumi, K. &
Watt, F. M. Jagged 1 is a β-catenin target gene
required for ectopic hair follicle formation in adult
epidermis. Development 133, 4427–4438 (2006).
49. Li, L. et al. Alagille syndrome is caused by mutations in
human Jagged1, which encodes a ligand for Notch1.
Nature Genet. 16, 243–251 (1997).
50. Engin, F. et al. Dimorphic effects of Notch signaling
in bone homeostasis. Nature Med. 14, 299–305
(2008).
51. Xiong, J. et al. Matrix-embedded cells control osteoclast
formation. Nature Med. 17, 1235–1241 (2011).
52. Fuller, K., Wong, B., Fox, S., Choi, Y. & Chambers, T . J.
TRANCE is necessary and sufficient for osteoblast-
mediated activation of bone resorption in osteoclasts.
J. Exp. Med. 188, 997–1001 (1998).
53. Simonet, W. S. et al. Osteoprotegerin: a novel
secreted protein involved in the regulation of bone
density. Cell 89, 309–319 (1997).
54. Mizuno, A. et al. Severe osteoporosis in mice lacking
osteoclastogenesis inhibitory factor/osteoprotegerin.
Biochem. Biophys. Res. Commun. 247, 610–615
(1998).
55. Li, Y. et al. B cells and T cells are critical for the
preservation of bone homeostasis and attainment of
peak bone mass in vivo. Blood 109, 3839–3848
(2007).
56. Farooqi, I. S. et al. Effects of recombinant leptin
therapy in a child with congenital leptin deficiency.
N. Engl. J. Med. 341, 879–884 (1999).
57. Sobacchi, C. et al. Osteoclast-poor human
osteopetrosis due to mutations in the gene encoding
RANKL. Nature Genet. 39, 960–962 (2007).
58. Guerrini, M. M. et al. Human osteoclast-poor
osteopetrosis with hypogammaglobulinemia due to
TNFRSF11A (RANK) mutations. Am. J. Hum. Genet.
83, 64–76 (2008).
59. Whyte, M. P . et al. Osteoprotegerin deficiency and
juvenile Paget’s disease. N. Engl. J. Med. 347,
175–184 (2002).
60. Albagha, O. M. et al. Genome-wide association study
identifies variants at CSF1, OPTN and TNFRSF11A as
genetic risk factors for Paget’s disease of bone. Nature
Genet. 42, 520–524 (2010).
61. Karsenty, G., Kronenberg, H. M. & Settembre, C.
Genetic control of bone formation. Annu. Rev. Cell Dev.
Biol. 25, 629–648 (2009).
62. Amizuka, N. et al. Haploinsufficiency of parathyroid
hormone-related peptide (PTHrP) results in abnormal
postnatal bone development. Dev. Biol. 175,
166–176 (1996).
63. Smits, P ., Dy, P ., Mitra, S. & Lefebvre, V. Sox5 and
Sox6 are needed to develop and maintain source,
columnar, and hypertrophic chondrocytes in the
cartilage growth plate. J. Cell Biol. 164, 747–758
(2004).
64. Nakashima, K. et al. The novel zinc finger-containing
transcription factor osterix is required for osteoblast
differentiation and bone formation. Cell 108, 17–29
(2002).
65. Lee, B. et al. Missense mutations abolishing DNA
binding of the osteoblast-specific transcription factor
OSF2/CBFA1 in cleidocranial dysplasia. Nature Genet.
16, 307–310 (1997).
66. Canalis, E. Update in new anabolic therapies for
osteoporosis. J. Clin. Endocrinol. Metab. 95,
1496–1504 (2010).
67. Lewiecki, E. M. Sclerostin: a novel target for
intervention in the treatment of osteoporosis.
Discovery Med. 12, 263–273 (2011).
68. Gauthier, J. Y. et al. The discovery of odanacatib
(MK-0822), a selective inhibitor of cathepsin
K. Bioorg. Med. Chem. Lett. 18, 923–928 (2008).
69. McClellan, J. & King, M. C. Genetic heterogeneity in
human disease. Cell 141, 210–217 (2010).REVIEWS
NATURE REVIEWS | GENETICS VOLUME 13 | AUGUST 2012 | 587
© 2012 Macmillan Publishers Limited. All rights reserved

70. Leslie, W. D. et al. Independent clinical validation of a
Canadian FRAX tool: fracture prediction and model
calibration. J. Bone Miner. Res. 25, 2350–2358
(2010).
71. Pepe, M. S., Gu, J. W. & Morris, D. E. The potential
of genes and other markers to inform about risk.
Cancer Epidemiol. Biomarkers Prev. 19, 655–665
(2010).
72. Yang, J. et al. Genome partitioning of genetic variation
for complex traits using common SNPs. Nature Genet.
43, 519–525 (2011).
73. Ladouceur, M., Dastani, Z., Aulchenko, Y. S.,
Greenwood, C. M. & Richards, J. B. The empirical
power of rare variant association methods: results
from Sanger sequencing in 1,998 individuals.
PLoS Genet. 8, e1002496 (2012).
74. Ladouceur, M., Leslie, W. D., Dastani, Z., Goltzman, D.
& Richards, J. B. An efficient paradigm for genetic
epidemiology cohort creation. PLoS ONE 5, e14045
(2010).
75. Bell, J. T . & Spector, T . D. A twin approach to unraveling
epigenetics. T rends Genet. 27, 116–125 (2011).
76. T alens, R. P . et al. Variation, patterns, and temporal
stability of DNA methylation: considerations for
epigenetic epidemiology. FASEB J. 24, 3135–3144
(2010).
77. Suhre, K. et al. Human metabolic individuality in
biomedical and pharmaceutical research. Nature 477,
54–60 (2011).
78. Kim, K. A. et al. R-spondin family members regulate
the Wnt pathway by a common mechanism. Mol. Biol.
Cell 19, 2588–2596 (2008).79. Josse, R. G. Bone biology and the role of RANK/
RANKL/OPG pathway. HealthPlexus [online], http://
www.healthplexus.net/article/bone-biology-and-role-
rankranklopg-pathway (2008).
80. Mackie, E. J., T atarczuch, L. & Mirams, M. The
skeleton: a multi-functional complex organ: the growth
plate chondrocyte and endochondral ossification.
J. Endocrinol. 211, 109–121 (2011).
81. Kung, A. W. et al. Association of JAG1 with bone
mineral density and osteoporotic fractures: a
genome-wide association study and follow -up
replication studies. Am. J. Hum. Genet. 86,
229–239 (2010).
82. Koller, D. L. et al. Genome-wide association study of
bone mineral density in premenopausal European-
American women and replication in African-American
women. J. Clin. Endocrinol. Metabolism 95,
1802–1809 (2010).
83. Ettinger, B. et al. Reduction of vertebral fracture risk in
postmenopausal women with osteoporosis treated with
raloxifene: results from a 3 -year randomized clinical
trial. Multiple Outcomes of Raloxifene Evaluation
(MORE) investigators. JAMA 282, 637–645 (1999).
84. Neer, R. M. et al. Effect of parathyroid hormone (1 -34)
on fractures and bone mineral density in
postmenopausal women with osteoporosis. N. Engl. J.
Med. 344, 1434–1441 (2001).
85. Greenspan, S. L. et al. Effect of recombinant human
parathyroid hormone (1 -84) on vertebral fracture and
bone mineral density in postmenopausal women with
osteoporosis: a randomized trial. Ann. Internal Med.
146, 326–339 (2007).86. Liberman, U. A. et al. Effect of oral alendronate on
bone mineral. Density and the incidence of fractures
in postmenopausal osteoporosis. New Engl. J. Med.
333, 1437–1444 (1995).
87. Rossouw, J. E. et al. Risks and benefits of estrogen plus
progestin in healthy postmenopausal women: principal
results from the Women’s Health Initiative randomized
controlled trial. JAMA 288, 321–333 (2002).
Acknowledgements
This work has been supported by the Canadian Institutes of
Health Research, the Lady Davis Institute for Medical
Research, Ministère de Développement économique,
de l’Innovation et de l’Exportation du Québec, the Arthritis
Research Campaign, the Wellcome T rust, Guy’s & St. Thomas’
NHS Foundation T rust and the King’s College London
Biomedical Centre. We would like to acknowledge the contri –
butions of F. Rivadeneira, C. Greenwood and C. Polychronakos
for their input on this Review.
Competing interests statement
The authors declare no competing financial interests.
FURTHER INFORMATION
J. Brent Richards’ homepage : http://www.mcgill.ca/genepi
GEFOS : http://www.gefos.org
TwinsUK : http://www.twinsuk.ac.uk
UK10K: http://www.uk10k.org
ALL LINKS ARE ACTIVE IN THE ONLINE PDFREVIEWS
588 | AUGUST 2012 | VOLUME 13 www.nature.com/reviews/genetics
© 2012 Macmillan Publishers Limited. All rights reserved