ORIGINAL ARTICLE

JBMR

Epigenome-wide Association of DNA Methylation in Whole Blood With Bone Mineral Density
John A Morris,1,2Ã Pei-Chien Tsai,3Ã Roby Joehanes,4,5Ã Jie Zheng,6Ã Katerina Trajanoska,7Ã Mette Soerensen,8,9Ã Vincenzo Forgetta,2 Juan Edgar Castillo-Fernandez,3 Morten Frost,10 Tim D Spector,3 Kaare Christensen,8,9,11 Lene Christiansen,8 Fernando Rivadeneira,7 Jonathan H Tobias,6,12 David M Evans,6,13 Douglas P Kiel,4,5 Yi-Hsiang Hsu,4,5,14ÃÃ J Brent Richards,1,2,3,15ÃÃ and Jordana T Bell3ÃÃ
1Department of Human Genetics, McGill University, Montreal, Canada 2Lady Davis Institute for Medical Research, Department of Medicine, Jewish General Hospital, McGill University, Montreal, Canada 3Department of Twin Research and Genetic Epidemiology, King’s College London, London, UK 4Institute for Aging Research, Hebrew SeniorLife, Boston, MA, USA 5Department of Medicine, Beth Israel Deaconess Medical Center and Harvard Medical School, Boston, MA, USA 6MRC Integrative Epidemiology Unit, University of Bristol, Bristol, UK 7Departments of Internal Medicine and Epidemiology, Erasmus Medical Center, Rotterdam, Netherlands 8The Danish Twin Registry and The Danish Aging Research Center, Epidemiology, Biostatistics, and Biodemography, Institute of Public Health, University of Southern Denmark, Odense, Denmark 9Department of Clinical Genetics, Odense University Hospital, Odense, Denmark 10Endocrine Research Unit, KMEB, University of Southern Denmark, Odense, Denmark 11Department of Clinical Biochemistry and Pharmacology, Odense University Hospital, Odense, Denmark 12Musculoskeletal Research Unit, University of Bristol, Bristol, UK 13The University of Queensland Diamantina Institute, Translational Research Institute, Princess Alexandra Hospital, Brisbane, Australia 14Broad Institute of MIT and Harvard, Cambridge, MA, USA 15Departments of Epidemiology and Biostatistics and Medicine, McGill University, Montreal, Canada
ABSTRACT
Genetic and environmental determinants of skeletal phenotypes such as bone mineral density (BMD) may converge through the epigenome, providing a tool to better understand osteoporosis pathophysiology. Because the epigenetics of BMD have been largely unexplored in humans, we performed an epigenome-wide association study (EWAS) of BMD. We undertook a large-scale BMD EWAS using the Inﬁnium HumanMethylation450 array to measure site-speciﬁc DNA methylation in up to 5515 European-descent individuals (NDiscovery ¼ 4614, NValidation ¼ 901). We associated methylation at multiple cytosine-phosphate-guanine (CpG) sites with dual-energy Xray absorptiometry (DXA)-derived femoral neck and lumbar spine BMD. We performed sex-combined and stratiﬁed analyses, controlling for age, weight, smoking status, estimated white blood cell proportions, and random effects for relatedness and batch effects. A 5% false-discovery rate was used to identify CpGs associated with BMD. We identiﬁed one CpG site, cg23196985, signiﬁcantly associated with femoral neck BMD in 3232 females (p ¼ 7.9 Â 10À11) and 4614 females and males (p ¼ 3.0 Â 10À8). cg23196985 was not associated with femoral neck BMD in an additional sample of 474 females (p ¼ 0.64) and 901 males and females (p ¼ 0.60). Lack of strong consistent association signal indicates that among the tested probes, no large-effect epigenetic changes in whole blood associated with BMD, suggesting future epigenomic studies of musculoskeletal traits measure DNA methylation in a different tissue with extended genome coverage. © 2017 The Authors. Journal of Bone and Mineral Research Published by Wiley Periodicals Inc.
KEY WORDS: BONE MINERAL DENSITY; EPIGENETICS; DNA METHYLATION; EPIGENOME-WIDE ASSOCIATION STUDY (EWAS); GENETIC
EPIDEMIOLOGY

This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. Received in original form January 30, 2017; revised form March 31, 2017; accepted April 5, 2017. Accepted manuscript online April 10, 2017. Address correspondence to: Jordana T Bell, DPhil, St Thomas’ Hospital Campus 3rd & 4th Floor South Wing Block D Westminster Bridge Road London SE1 7EH. E-mail: jordana.bell@kcl.ac.uk; J Brent Richards, MD, 3755 Ch de la Co^te-Sainte-Catherine, Montreal, QC H3T 1E2. E-mail: brent.richards@mcgill.ca; Yi-Hsiang Hsu, PhD, 1200 Centre St, Boston, MA 02131, USA. E-mail: yihsianghsu@hsl.harvard.edu ÃJAM, P-CT, RJ, JZ, KT, and MS contributed equally to this work. ÃÃY-HH, JBR, and JTB contributed equally to this work. Additional Supporting Information may be found in the online version of this article.
Journal of Bone and Mineral Research, Vol. 32, No. 8, August 2017, pp 1644–1650 DOI: 10.1002/jbmr.3148 © 2017 The Authors. Journal of Bone and Mineral Research Published by Wiley Periodicals Inc.
1644

Introduction
Osteoporosis is primarily an aging-related disease characterized by compromised bone strength that increases the risk of fracture. Because of population aging worldwide, the incidence of osteoporosis is increasing, exceeding $17 billion per year in direct care costs within the United States(1) and costing upward of s37 billion per year in the EU-27 member states.(2) Identifying the causes of osteoporosis will improve the understanding of its pathology, leading to better or more efﬁcient treatments of this common and costly disease. Low bone mineral density (BMD) is one of the major risk factors for fracture and is largely used in clinical prediction tools for fracture and gauging response to treatment. Genome-wide association studies (GWAS) of BMD directly assessed at the femoral neck (FN) and lumbar spine (LS), the two most commonly measured sites for quantifying BMD and diagnosing osteoporosis, have been instrumental in identifying novel genetic loci inﬂuencing osteoporosis disease risk.(3,4) However, epigenetic variation in the genome, which can be inﬂuenced by both genetic and environmental factors,(5,6) may also inﬂuence BMD, yet the epigenetic inﬂuences on BMD have largely been unexplored.
One of the most stable epigenetic processes is DNA methylation, or the addition of a CH3 methyl group to cytosine, typically in the context of cytosine paired sequentially to a guanine nucleotide, separated by a phosphate group (CpG). DNA methylation is known to play a role in gene expression and cell differentiation,(7,8) and differential DNA methylation has been linked to multiple human complex traits and disease phenotypes.(5,9–12) Studies performed using bone samples have identiﬁed epigenetic alterations that inﬂuence bone cell function.(13,14) We studied epigenetic variation in whole blood, as a proxy for difﬁcult-to-acquire samples such as bone, in relation to BMD because epigenetic markers are often stable across multiple tissues, and immune cells within blood are known to inﬂuence bone homeostasis.(15) Furthermore, osteoclasts are derived from the monocyte-macrophage lineage found in whole blood.(16) Although epigenetic proﬁling has been performed previously in bone samples from osteoporotic and osteoarthritic patients(17) and an epigenome-wide association study (EWAS) of BMD has been performed in mice,(18) EWAS of BMD have not been reported in humans with validation of signiﬁcant ﬁndings.
We, therefore, undertook a large-scale BMD EWAS, assessing the association of up to 473,882 CpGs quantiﬁed in whole blood with BMD measured in up to 4614 individuals across ﬁve cohorts from Europe and North America. To our knowledge, this study is the largest EWAS of a musculoskeletal trait performed to date. We used an additional 901 individuals as a validation cohort to increase the reliability of our results.
Materials and Methods
Individual cohorts
We performed our EWAS in cohorts composed of Europeandescent individuals. Cohorts used for the discovery analysis were the TwinsUK Registry (TUK), Framingham Study Offspring Cohort (FOS), Avon Longitudinal Study of Parents and Children (ALSPAC; further information on the ALSPAC cohorts and ARIES project is included in Supplemental Table S1), Rotterdam Study (RS), and the Danish Twin Registry (DTR). The cohort used for validation of signiﬁcant ﬁndings was the Framingham Study

Generation 3 cohort (Gen3), a cohort including family members of FOS (Table 1; Supplemental Table S1).
Both, or one of, FN and LS BMD were measured in each cohort by dual-energy X-ray absorptiometry (DXA) (Supplemental Table S2). All cohorts, except the ALSPAC and DTR, followed the same methods for extracting DNA from whole-blood tissue and quantifying DNA methylation. Whole-blood tissue DNA was extracted using the DNeasy Blood & Tissue Kit (Qiagen, Inc., Valencia, CA, USA), followed by bisulﬁte conversion of 750 ng DNA using the EZ DNA Methylation Kit (Zymo Research, Irvine, CA, USA) following manufacturer instructions. The ALSPAC cohort and the DTR cohort performed DNA extraction and conversion as described previously.(19,20) DNA methylation across the genome was quantiﬁed using the Inﬁnium HumanMethylation450 BeadChip (Illumina, San Diego, CA, USA), assaying up to 482,421 CpGs throughout the human genome. Image intensities were extracted using GenomeStudio Methylation Module (v1.8) software. Cohort-speciﬁc criteria were applied in further quality control and normalization of probe intensities (Supplemental Table S3).
Statistical analyses
For discovery analyses, each cohort followed a prespeciﬁed analysis plan. FN and LS BMD residuals were calculated by ﬁtting a linear regression model, adjusting for age, weight, and sex. For sex-speciﬁc analyses, the term for sex was removed from the model. In addition, the DTR adjusted for birth weight discordance as historically DTR samples were selected to address birth weight discordance in twins (Supplemental Table S1). To address the issue of cell heterogeneity in whole blood tissue, each cohort calculated the estimated white blood cell proportions of B cells, T cells (CD4þ and CD8þ), granulocytes, natural killer cells, and monocytes using the Houseman and colleagues method for quadratic projections.(21) DNA methylation for each probe was transformed to a standard normal distribution using quantile normalization. The association between DNA methylation and BMD was then calculated by ﬁtting a linear mixed effects model for normalized DNA methylation, including BMD residuals, smoking (measured as smokers, non-smokers, or former smokers), age, weight, sex, and estimated white blood cell proportions as ﬁxed effects, and terms for family structure and batch effects as random effects, where relevant. We used BMD residuals to reduce problems due to collinearity between BMD, weight, and age when ﬁtting our linear mixed-effects model. Association testing was performed in male, female, and combined samples. Each cohort was assessed for epigenome-wide statistical inﬂation by calculating the genomic inﬂation factor lambda (l) and generating a quantile-quantile plot (QQ-plot). Lambda can be calculated to estimate the deviation of a distribution from a null expected distribution, whereas QQ plots can be used to visualize the deviation of a distribution from a null expected distribution.
Fixed-effects meta-analyses were performed using METAL(22) for FN and LS sex-combined and sex-stratiﬁed analyses. We used the I2 statistic to quantify the variability in association effect estimates due to statistical heterogeneity, excluding probes with heterogeneous I2 statistics (pHet < 0.05). Statistical signiﬁcance, when considering the multiple testing burden, was determined by calculating Benjamini-Hochberg (BH) adjusted p values for each meta-analysis. Probes with signiﬁcant BHadjusted p values (pBH < 0.05) would, therefore, be signiﬁcant at a 5% false-discovery rate (FDR). Summary statistics for the FN

Journal of Bone and Mineral Research

ASSOCIATION OF DNA METHYLATION IN WHOLE BLOOD WITH BMD 1645

Table 1. Sample Sizes of Discovery and Replication Cohorts

Sample size

FN BMD

LS BMD

Phase

Cohort

Pooled Females Males Pooled Females Males

Discovery Validation

Avon Longitudinal Study of Parents and Children (ALSPAC) Danish Twin Registry (DTR) Framingham Study Offspring Cohort (FOS) Rotterdam Study (RS) TwinsUK (TUK) Discovery total FOS Gen3 Discovery þ Validation total

715 267 2207 650 775 4614 901 5515

715 132 1254 356 775 3232 448 3680

0 135 953 294
0 1382 453 1835

0 260 2203 633 770 3866
0 3866

0 132 1259 346 770 2507
0 2507

0 128 953 287
0 1368
0 1368

FN ¼ femoral neck; BMD ¼ bone mineral density; LS ¼ lumbar spine.

and LS sex-combined and sex-stratiﬁed meta-analyses are available for download (www.gefos.org).
Associated probes were assessed for their twin-based heritability using normalized methylation beta values estimated from 330 female MZ twin pairs and 34 female DZ twin pairs from the TUK cohort, adjusted for age, body mass index, smoking status, alcohol consumption, predicted whole blood cell-type counts, DNA methylation plate, and position on the plate. Heritability was estimated by ﬁtting the classical ACE model in OpenMX.(23) The observed variance in the adjusted beta values was partitioned into additive genetic (A), common environmental (C), and unique environmental (E) factors. Heritability was deﬁned as A/(AþCþE).
Associated probes were assessed for the inﬂuence of single nucleotide polymorphisms (SNPs) that overlapped the probe body by mapping these probes to dbSNP 146.(24) This was assessed by adding a term for the dosage of each SNP to the discovery linear mixed-effects models in cohorts with genotype data to observe if the association between DNA methylation and BMD was inﬂuenced by the genetic polymorphism at the probe.
Gen3 samples were assessed using the same methods as in the analysis of FOS samples to perform validation analyses of signiﬁcantly associated probes. Probes were deemed robustly associated with BMD if they met a validation p value of less than 0.05. These samples are not completely independent from the FOS samples because the Framingham Study is a family-based study with several cohorts, and, therefore, there is underlying family structure.
To assess the power of our study, we performed 5000 permutations on the 775 TUK samples with FN BMD measurements. FN was randomly sampled based on the twin and family structure before ﬁtting linear mixed-effects models, and the power was deﬁned as the number of permutations with p values greater than the observed p value for the TUK samples (p ¼ 1.14 Â 10À5; Supplemental Table S8).
Results
Meta-analysis
Meta-analyses of discovery cohorts identiﬁed CpG site cg23196985 associated at a 5% FDR for FN sex-combined (b ¼ 0.66, SE ¼ 0.19, p ¼ 2.99 Â 10À8, pBH ¼ 1.30 Â 10À2) and FN female (b ¼ 0.95, SE ¼ 0.15, p ¼ 7.86 Â 10À11, pBH ¼ 3.41 Â 10À5; Fig. 1, Fig. 2, Table 2) analyses. CpG site cg23196985 maps to the

5’ untranslated region of the liver carboxylase 1 gene (CES1), which is expressed in the liver and whole blood(25) yet with no
currently reported associations with BMD by GWAS in the same
chromosomal region (16q12.2) and with the nearest BMD-
associated SNP mapping approximately 4 mega base pairs upstream at the SALL1 and CYLD locus.(26) The calculated lambdas and QQ plots for the meta-analyses of FN female and sex-combined analyses revealed no statistical inﬂation of the association p values (lfemale ¼ 1.02, lsex-combined ¼ 0.97; Supplemental Table S9). We observed no signiﬁcantly associated CpG sites with LS BMD in sex-combined or sex-stratiﬁed analyses (Supplemental Figures).
We tested for the inﬂuence of SNPs underlying cg23196985 in females from the FOS, RS, and ALSPAC cohorts, as the strength of
the association was stronger in females than in the sex-
combined analysis. Because some of the samples in our cohorts included twins, we ﬁrst estimated the evidence for heritability of DNA methylation levels at cg23196985 in the TUK cohort. We
observed evidence for additive genetic effects with a heritability
estimate of 0.69 at cg23196985 and therefore pursued further
analyses exploring the association between DNA methylation
levels at this CpG site and BMD conditional on SNP genotypes.
All twins were homozygous for the reference allele at rs144950224, a SNP that maps directly to the probe’s target CpG site. Four SNPs mapped to the cg23196985 50 base-pair
probe sequence, and these included rs144950224, rs12149371,
rs12149373, and rs3815583. SNP rs144950224 was found to be
rare within our cohorts, with a minor allele frequency (MAF) of
approximately 0.5% in FOS samples, 0.1% in RS samples, and no
carriers in ALSPAC samples. We observed no notable change in
association p values upon conditioning with each of the four SNPs (Supplemental Table S10).
In the validation sample, cg23196985 was not associated with FN in Gen3 female (p ¼ 0.64) and sex-combined (p ¼ 0.60) analyses. However, after meta-analyzing Gen3 validation data
with discovery results, the probe remained strongly associated in female-only analyses (b ¼ 0.86, SE ¼ 0.14, p ¼ 3.7 Â 10À10; n ¼ 4345) but not in sex-combined analysis (p ¼ 0.68, n ¼ 5301).
For our power calculation, we found all permutations were less signiﬁcant than our observed p value for the 775 TUK samples at p ¼ 1.14 Â 10À5 (permuted p value range 0.99 to 2.46 Â 10À5); this suggested we had 100% power to detect the observed effect size (b ¼ 1.20, SE ¼ 0.27) between bone density measurements and methylation at cg23196985 in CES1.

1646 MORRIS ET AL.

Journal of Bone and Mineral Research

Fig. 1. Quantile-quantile plots (QQ plots) of the distribution of observed Àlog10 association p values against the expected null distribution, for discovery meta-analyses of FN BMD in (A) females-only and (B) sexcombined analyses. Genomic inﬂation lambda scores are given in each QQ plot to quantify statistical inﬂation of p values. No evidence for inﬂation was observed in the QQ plots or as calculated by lambda scores.
Individual cohorts
Individual cohort analyses identiﬁed seven probes that were signiﬁcantly associated in two cohorts with sex-combined or sex-stratiﬁed analyses (Supplemental Table S4), but there were no other cohort-speciﬁc signiﬁcant associations. The DTR LS female analysis identiﬁed four signiﬁcantly associated probes, two of which map to genes, cg04081651 (MAP3K8), cg09832237, cg14793931 (ZFR2), and cg24029028 (Supplemental Table S5). The DTR LS male analysis identiﬁed one signiﬁcantly associated probe, cg23214071 (HLA-DQB1) (Supplemental Table S6). The TUK LS female analysis identiﬁed two signiﬁcantly associated probes, cg24117468 (P4HA2) and cg02526790 (TG)

(Supplemental Table S7). The calculated lambda and QQ plot for the DTR LS female analysis revealed statistical inﬂation of the association p values (l ¼ 1.46), but the lambda and QQ plots for the remainder of the cohorts showed no large inﬂation or deﬂation of association p values (Supplemental Table S9).
Discussion
In the ﬁrst large-scale assessment of the contribution of epigenetic changes in whole blood to BMD, we did not identify methylation changes reliably associated with this clinically relevant trait. CpG site cg23196985 was found in the discovery meta-analysis to be strongly associated with FN BMD in females only and in analyses combining males and females, but upon validation in an extended sample that included related individuals, the association was attenuated in the female analysis and completely absent in sex-combined analyses.
These ﬁndings provide important insights into the ﬁeld of epigenetics. The ﬁrst is that by using a precisely measured trait, BMD, which is highly heritable with estimates from 50% to 85%(27) and for which genetic determinants have been identiﬁed through GWAS,(3,4,27) there do not appear to be associations between methylation changes and BMD. Although whole blood methylation changes may not be the ideal tissue within which to test epigenetic inﬂuences on bone, this conveniently accessible tissue has many links to bone biology, including the fact that osteoclasts and monocyte/macrophages originate from the same precursors.(15,16) The extent to which methylation changes are shared between bone and whole blood is not well known. However, evidence shows that a signiﬁcant proportion of methylation variation genome-wide can be conserved across tissues.(28) Additional explanations for our mostly null ﬁndings include the possibility that DNA methylation changes may not have a large inﬂuence on BMD.
Notwithstanding the general lack of consistent associations with BMD across the genome, we did generate evidence for suggestive association of cg23196985 with FN BMD in females. However, we caution that these ﬁndings require further replication. Because we are unaware of any available replication data to test this hypothesis, these ﬁndings will require replication in future studies.
There is limited evidence for the effects of DNA methylation on bone. A methylation proﬁling study that compared the differences between bone samples of 27 osteoporotic and 23 osteoarthritic patients was undertaken on an earlier DNA methylation platform, the HumanMethylation27 BeadChip (assessing approximately 27,000 CpGs in the genome), and was able to identify bone genes following pathway analyses of more than 200 differentially methylated CpGs; however, to date, these results lack replication.(17) Another study failed to demonstrate speciﬁc effects of DNA methylation, assessed by sequencing methods, on RANKL in the bones of patients with osteoporotic fractures.(29) The evidence from previous studies and our own suggests that if strong effects of DNA methylation on bone biology are to be identiﬁed, they may not be detectable with current analytical approaches.
A strength of our study was the sample size and a conservative estimate of statistical power to identify epigenetic effects on BMD that account for 0.8% of its variance. The large sample size also allowed us to classify several cohort-level associations that were likely to be false positives. For example, TUK female analyses identiﬁed two probes signiﬁcantly associated with LS

Journal of Bone and Mineral Research

ASSOCIATION OF DNA METHYLATION IN WHOLE BLOOD WITH BMD 1647

Fig. 2. Manhattan plots of Àlog10 association p values for discovery meta-analyses of FN BMD in (A) females-only and (B) sex-combined analyses.

BMD, but this association was not observed in any other cohort, suggesting the associations were false positives (Supplemental Tables S4 and S5).
One of the key limitations of cohort-based epigenetic studies is the lack of cell-sorted data for analysis. As discussed by Birney and colleagues,(30) optimal planning at the outset of a study is ideal; however, such coordination is difﬁcult to implement in

large cohorts and so bioinformatics methods must be applied post hoc to adjust for suboptimal study designs. We adjusted for cell heterogeneity within whole blood, and therefore the signal we tested for association with BMD would be ubiquitous within whole blood. Such ubiquitous signals within whole blood may only be detectable for extremely strong environmental modiﬁers of DNA methylation, such as cigarette smoking.(5,6)

Table 2. Meta-Analysis Results for Association of cg23196985 with FN BMD in Both Female and Sex-Pooled Analyses in Discovery phase, with Replication and Combined Discovery and Replication Analyses

Discovery

Validation

Combined

b SE

p

pBH pHet b

SE p

b SE

p

pHet

Female

0.95 0.15 7.9E-11 3.4E-05 0.81 0.19 0.39 0.64 0.86 0.14 3.8E-10

0.43

Sex pooled 0.66 0.12 3.0E-08 1.3E-02 0.1 –0.01 0.02 0.6 0.01 0.02 0.68 3.0E-07

FN ¼ femoral neck; BMD ¼ bone mineral density; b ¼ effect size; SE ¼ standard error; pBH ¼ Benjamini-Hochberg adjusted p value; pHet ¼ heterogeneity p value.

1648 MORRIS ET AL.

Journal of Bone and Mineral Research

Targeted EWAS of speciﬁc cell types within whole blood with clear roles in bone biology, such as monocytes due to their role in osteoclastogenesis, may be more fruitful. Furthermore, longitudinal studies can identify disease-risk biomarkers and provide mechanistic insights; however, these are generally underpowered when studying BMD because of the relatively small changes in BMD that occur over time. As we have shown that large-scale whole blood EWAS of BMD does not identify disease-risk biomarkers for osteoporosis risk, a well-powered longitudinal study with a wide range of time points and bone samples may be more informative.
In the largest EWAS meta-analysis to date of BMD, we observed a probe near CES1 to be associated with FN BMD in the discovery sample of up to 4826 individuals but not with the same phenotype in a related validation sample of 901 individuals. In conclusion, these ﬁndings suggest that there are no large effects of methylation changes on BMD in whole blood in the epigenome, which are common and well captured by the Inﬁnium HumanMethylation450 BeadChip.
Disclosures
All authors state that they have no conﬂicts of interest.
Acknowledgments
JAM, VF, and JBR were supported by the RMGA, FRQS, and CIHR. PCT, JEC-F, TDS, and JTB were supported by the European Commission and the UK Economic and Social Research Council (ES/N000404/1). The Framingham Study was funded by NIH contract N01-HC-25195. DPK was supported by the NIH (R01AR41398). JZ, JHT, and DME were supported by the Medical Research Council Integrative Epidemiology Unit. KT was supported by Erasmus Mundus Western Balkans. MS, MF, KC, and LC were supported by the European Union’s Seventh Framework Programme (FP7/2007-2011) under grant agreement 259679 and the Danish National Program for Research Infrastructure 2007 (09-063256). The DTR was supported by the Velux Foundation. JZ, JHT, and DME were supported by the Medical Research Council Integrative Epidemiology Unit.
We are extremely grateful to all the families who took part in this study, the midwives for their help in recruiting them, and the whole ALSPAC team, which includes interviewers, computer and laboratory technicians, clerical workers, research scientists, volunteers, managers, receptionists and nurses. The UK Medical Research Council and Wellcome (grant 102215/2/13/2) and the University of Bristol provide core support for ALSPAC. Methylation data in the ALSPAC cohort were generated as part of the UK BBSRC-funded (BB/I025751/1 and BB/I025263/1) Accessible Resource for Integrated Epigenomic Studies (ARIES, http://www. ariesepigenomics.org.uk).
Authors’ roles: JAM, P-CT, Y-HH, JBR, and JTB conceived of the study design. JAM, P-CT, Y-HH, RJ, JZ, KT, MS, VF, and JEC-F performed the analyses. P-CT, MF, KC, and LC participated in acquisition of data. All authors have revised the manuscript and given their ﬁnal approval.
References
1. Burge R, Dawson-Hughes B, Solomon DH, Wong JB, King A, Tosteson A. Incidence and economic burden of osteoporosis-related fractures in the United States, 2005-2025. J Bone Miner Res. 2007;22(3): 465–75.

2. Hernlund E, Svedbom A, Ivergard M, et al. Osteoporosis in the European Union: medical management, epidemiology and economic burden: a report prepared in collaboration with the International Osteoporosis Foundation (IOF) and the European Federation of Pharmaceutical Industry Associations (EFPIA). Arch Osteoporos. 2013;8(1–2):136.
3. Estrada K, Styrkarsdottir U, Evangelou E, et al. Genome-wide metaanalysis identiﬁes 56 bone mineral density loci and reveals 14 loci associated with risk of fracture. Nat Genet. 2012;44(5):491–501.
4. Zheng H-F, Forgetta V, Hsu Y-H, et al. Whole-genome sequencing identiﬁes EN1 as a determinant of bone density and fracture. Nature. 2015;526(7571):112–7.
5. Tsaprouni LG, Yang T-P, Bell J, et al. Cigarette smoking reduces DNA methylation levels at multiple genomic loci but the effect is partially reversible upon cessation. Epigenetics. 2014;9(10):1382–96.
6. Joehanes R, Just AC, Marioni RE, et al. Epigenetic signatures of cigarette smoking. Circ Cardiovasc Genet. 2016;CIRCGENETICS. 116.001506.
7. Robertson KD. DNA methylation and human disease. Nat Rev Genet. 2005;6(8):597–610.
8. Suzuki MM, Bird A. DNA methylation landscapes: provocative insights from epigenomics. Nat Rev Genet. 2008;9(6):465–76.
9. Grundberg E, Meduri E, Sandling JK, et al. Global analysis of DNA methylation variation in adipose tissue from twins reveals links to disease-associated variants in distal regulatory elements. Am J Hum Genet. 2013;93(5):876–90.
10. Yuan W, Xia Y, Bell CG, et al. An integrated epigenomic analysis for type 2 diabetes susceptibility loci in monozygotic twins. Nat Commun. 2014;5:5719.
11. Chambers JC, Loh M, Lehne B, et al. Epigenome-wide association of DNA methylation markers in peripheral blood from Indian Asians and Europeans with incident type 2 diabetes: a nested case-control study. Lancet Diabetes Endocrinol. 2015;3(7):526–34.
12. Tsai P-C, Van Dongen J, Tan Q, et al. DNA methylation changes in the IGF1R gene in birth weight discordant adult monozygotic twins. Twin Res Hum Genet. 2015;18(6):635–46.
13. Reppe S, Noer A, Grimholt RM, et al. Methylation of bone SOST, its mRNA, and serum sclerostin levels correlate strongly with fracture risk in postmenopausal women. J Bone Miner Res. 2015;30(2): 249–56.
14. Delgado-Calle J, San~udo C, Bolado A, et al. DNA methylation contributes to the regulation of sclerostin expression in human osteocytes. J Bone Miner Res. 2012;27(4):926–37.
15. Greenblatt MB, Shim J-H. Osteoimmunology: a brief introduction. Immune Netw. 2013;13(4):111–5.
16. Novack DV, Mbalaviele G. Osteoclasts-key players in skeletal health and disease. Microbiol Spectr. 2016;4(3).
17. Delgado-Calle J, Fernandez AF, Sainz J, et al. Genome-wide proﬁling of bone reveals differentially methylated regions in osteoporosis and osteoarthritis. Arthritis Rheum. 2013;65(1):197–205.
18. Orozco LD, Morselli M, Rubbi L, et al. Epigenome-wide association of liver methylation patterns and complex metabolic traits in mice. Cell Metab. 2015;21(6):905–17.
19. Christiansen L, Lenart A, Tan Q, et al. DNA methylation age is associated with mortality in a longitudinal Danish twin study. Aging Cell. 2016;15(1):149–54.
20. Relton CL, Gaunt T, McArdle W, et al. Data resource proﬁle: Accessible Resource for Integrated Epigenomic Studies (ARIES). Int J Epidemiol. 2015;44(4):1181–90.
21. Houseman EA, Accomando WP, Koestler DC, et al. DNA methylation arrays as surrogate measures of cell mixture distribution. BMC Bioinformatics. 2012;13(1):86.
22. Willer CJ, Li Y, Abecasis GR. METAL: fast and efﬁcient meta-analysis of genomewide association scans. Bioinformatics. 2010;26(17):2190–1.
23. Boker S, Neale M, Maes H, et al. OpenMx: an open source extended structural equation modeling framework. Psychometrika. 2011;76 (2):306–17.
24. Sherry ST, Ward MH, Kholodov M, et al. dbSNP: the NCBI database of genetic variation. Nucleic Acids Res. 2001;29(1):308–11.

Journal of Bone and Mineral Research

ASSOCIATION OF DNA METHYLATION IN WHOLE BLOOD WITH BMD 1649

25. GTEx Consortium. Human genomics. The Genotype-Tissue Expression (GTEx) pilot analysis: multitissue gene regulation in humans. Science. 2015;348(6235):648–60.
26. Welter D, MacArthur J, Morales J, et al. The NHGRI GWAS Catalog, a curated resource of SNP-trait associations. Nucleic Acids Res. 2014;42(Database issue):D1001–6.
27. Richards JB, Zheng H-F, Spector TD. Genetics of osteoporosis from genome-wide association studies: advances and challenges. Nat Rev Genet. 2012;13(8):576–88.

28. Dick KJ, Nelson CP, Tsaprouni L, et al. DNA methylation and bodymass index: a genome-wide analysis. Lancet. 2014;383(9933): 1990–8.
29. Delgado-Calle J, San~udo C, Fernandez AF, Garcıa-Renedo R, Fraga MF, Riancho JA. Role of DNA methylation in the regulation of the RANKL-OPG system in human bone. Epigenetics. 2012;7(1):83–91.
30. Birney E, Smith GD, Greally JM. Epigenome-wide association studies and the interpretation of disease -omics. PLoS Genet. 2016;12(6): e1006105.

1650 MORRIS ET AL.

Journal of Bone and Mineral Research