Three pleiotropic loci associated with bone mineral density and lean body mass

Abstract

Both bone mineral density (BMD) and lean body mass (LBM) are important physiological measures with strong genetic determination. Besides, BMD and LBM might have common genetic factors. Aiming to identify pleiotropic genomic loci underlying BMD and LBM, we performed bivariate genome-wide association study meta-analyses of femoral neck bone mineral density and LBM at arms and legs, and replicated in the large-scale UK Biobank cohort sample. Combining the results from discovery meta-analysis and replication sample, we identified three genomic loci at the genome-wide significance level (p < 5.0 × 10⁻⁸): 2p23.2 (lead SNP rs4477866, discovery p = 3.47 × 10⁻⁸, replication p = 1.03 × 10⁻⁴), 16q12.2 (rs1421085, discovery p = 2.04 × 10⁻⁹, replication p = 6.47 × 10⁻¹⁴) and 18q21.32 (rs11152213, discovery p = 3.47 × 10⁻⁸, replication p = 6.69 × 10⁻⁶). Our findings not only provide useful insights into lean mass and bone mass development, but also enhance our understanding of the potential genetic correlation between BMD and LBM.

Introduction

Osteoporosis is characterized by low bone mass and microarchitectural deterioration of bone tissue. As a common and aging-related bone metabolic disease, it predisposes people to fragility fractures (Cummings and Melton 2002). The femoral neck, located near the top of the femur bone, is especially susceptible to osteoporotic fracture because it is the weakest part of the femur. Femoral neck bone mineral density (FNK-BMD) is a strong predictor of hip fracture susceptibility in elderly (Rivadeneira et al. 2007). Low BMD is the most direct feature of osteoporosis. BMD is highly heritable with heritability to be as high as 85% (Ralston and de Crombrugghe 2006). In recent years, hundreds of genes/loci affecting BMD have been identified by genome-wide association studies (GWASs) and their meta-analyses (Estrada et al. 2012; Zheng et al. 2015; Kemp et al. 2017; Medina-Gomez et al. 2018; Pei et al. 2018; Morris et al. 2019; Pei et al. 2019b).

Sarcopenia is defined as a progressive, and generalized loss of skeletal muscle mass, strength, and function (Fielding et al. 2011). It is known to reduce the quality of life of the elderly and has become a concern in both developing and developed countries. According to Khosla et al., the age- and sex-adjusted prevalence of sarcopenia varied from 6 to 15% among subjects 65 years of age or over (Melton et al. 2000). Similar to the bone, muscle tissue deteriorates with advanced age. LBM has an important genetic component with a high heritability over 50% (Arden and Spector 1997; Nguyen et al. 1998). Researches on sarcopenia have been widely recognized for more than a decade. Previous studies have identified dozens of genomic loci associated with LBM (Liu et al. 2009a; Sun et al. 2011; Hai et al. 2012a, b; Guo et al. 2013; Zillikens et al. 2017; Pei et al. 2019a).

There are many similarities between osteoporosis and sarcopenia, including demographics, high prevalence, and huge socioeconomic costs (Edwards et al. 2015). Both are age-related decrements in mass and quality of bone and muscle, respectively (Sirola and Kroger 2011). The causes of both are multifactorial, such as hormones, nutrition, etc. (Rolland et al. 2008).

Twin studies have calculated the additive genetic correlation of BMD and LBM to range from 30 to 45% (Bogl et al. 2011). Park et al. suggested that LBM has a positively correlated genetic correlation with BMD (Park et al. 2012). Medina-Gomez et al. identified 8 pleiotropic loci that contribute to BMD and LBM in 10,414 children (Medina-Gomez et al. 2017). However, most loci with pleiotropic effects to both traits remain unknown.

This study aimed to identify pleiotropic SNPs/genes that contribute to FNK-BMD and LBM at arms and legs by bivariate genome-wide association meta-analyses. Our findings may help to improve understanding of the underlying genetic structure and pathophysiological mechanisms of osteoporosis and sarcopenia.

Materials and methods

Participants

The discovery stage includes six GWAS samples from different research and/or clinical centers. The study was approved by the local institutional review board of all agencies. All participants signed an informed consent before participating in the study.

Three samples were from in-house studies and the other three were accessed through the NCBI Database of Genotypes and Phenotypes (dbGAP). The two in-house samples consisted of 866 (Omaha osteoporosis study, OOS) and 2205 (Kansas City osteoporosis study, KCOS) unrelated individuals, respectively, of European ancestry. The third in-house sample comprises 1539 (China osteoporosis study, COS) unrelated individuals of Chinese Han ancestry. The fourth sample was derived from the Framingham heart study (FHS), a longitudinal and prospective cohort comprising over 16,000 participants of European ancestry spanning three generations. A total of 6547 participants were qualified for analysis. Both the fifth and sixth samples are from the Women’s health initiative (WHI) observational study, a partial factorial randomized and longitudinal cohort with over 12,000 genotyped women aged 50–79 years, of African-American or Hispanic ancestry (1998). A total of 843 subjects of African-American ancestry (WHI-AA), 445 subjects of Hispanic ancestry (WHI-HIS).

Phenotype measurements and modeling

BMD, LBM, and fat body mass (FBM) were measured with dual-energy X-ray absorptiometry (DXA) scanners (either Lunar Corp., Madison, WI, USA; or Hologic Inc, Bedford, MA, USA) following the manufacturer protocols. LBM in the FHS sample was approximated by subtracting FBM from soft tissue mass at respective sites.

In each GWAS sample, covariates were screened among gender, age, age squared, height, height squared (in case of FNK-BMD) with the stepwise linear regression model implemented in the R function stepAIC. Raw FNK-BMD and LBM measurements were adjusted by significant covariates. To adjust for potential population stratification, the top five principal components derived from the genome-wide genotype data were included as covariates. The residuals were normalized by inverse quantiles of standard normal distribution.

We analyzed the combination of FNK-BMD with LBM at arms and legs, respectively. To correct for the effect of FBM on LBM, we also analyzed LBM after adjustment by FBM at corresponding site. Therefore, four combinations of bivariate analysis were performed.

Genotyping and quality control

All GWAS samples were genotyped by high-throughput SNP genotyping arrays (Affymetrix Inc., Santa Clara, CA, USA; or Illumina Inc., San Diego, CA, USA within individual samples) following the manufacturer’s protocols. Strict genotype quality control (QC) procedure was followed at both individual and SNP levels. At the individual level, genetic sex was inferred from genotype data on X-chromosome with PLINK (Purcell et al. 2007) and was compared with the self-reported sex. Individuals of inconsistent sex were removed. At the SNP level, SNPs that violate the Hardy–Weinberg equilibrium (p < 1.0 × 10⁻⁵) were removed. Population outliers were monitored by genotype-derived principle components, and if present, outliers were removed. In the FHS sample, SNPs with the Mendel error were set to a missing value.

Genotype imputation

GWAS samples were imputed by the 1000 genomes project phase 3 sequence variants (as of May 2013). Specifically, haplotype data from 240 individuals of European ancestry, 244 of East Asian ancestry, 319 of African ancestry and 170 of admixed American ancestry were downloaded from the project download site. Haplotypes of bi-allelic variants, including SNPs and bi-allelic insertions/deletions (INDELs), were extracted to form reference panels for imputation. As a QC procedure, variants with zero or one copy of minor alleles were removed.

Each GWAS sample was imputed by the respective reference panel with the closest ancestry. Before imputation, it is necessary to check for the consistency of allele strandedness between the test sample and the reference sample. The Chi-square test was used to examine the consistency. SNPs that failed the consistency test (p < 1.0 × 10⁻⁶) were transformed into the reverse strand in the test sample. SNPs that again failed the consistency test were removed. Imputation was performed with FISH (Zhang et al. 2014), a fast and accurate diploid genotype imputation algorithm.

Association analysis in individual samples

In each GWAS sample, both univariate and bivariate association analyses were performed assuming an additive model of inheritance. A univariate/bivariate mixed linear regression model was applied to account for genetic relatedness within each pedigree in the FHS sample. A univariate/bivariate linear regression model was used to examine the genetic association in other unrelated samples.

Meta-analysis

Summary association statistics from individual GWAS samples were combined to perform meta-analysis. As a QC procedure, only well-imputed (r² > 0.3 in at least 2 samples) and common or less common (minor allele frequency, MAF > 0.01 in the European population) SNPs were included into analysis.

Both univariate and bivariate meta-analyses were performed under the fixed-effects model. Briefly, for a particular SNP, let β_i = (β_1i, β_2i) be the vector of regression coefficient for the two traits in the ith study (i = 1, …, n, n = 6), and let $V_i=\left[\begin{array}{ll}{v}_{11}\hfill & v_{12}\hfill \\ v_{21}\hfill & v_{22}\hfill \end{array}\right]$ be the corresponding symmetric variance–covariance matrix for the two regression coefficients. Both β_i and V_i are obtained from individual study analysis. Define the following data structure

$$B={\left({\beta }_1^{\prime },{\beta }_2^{\prime },\dots ,{\beta }_n^{\prime }\right)}_{2n\times 1}^{\prime },$$

$$X={\left[\begin{array}{cc}1& 0\\ 0& 1\\ \dots & \\ 1& 0\\ 0& 1\end{array}\right]}_{2n\times 2},V={\left[\begin{array}{cccc}{V}_1& & & \\ & V_2& & 0\\ & & \cdots & \\ & 0& & V_n\end{array}\right]}_{2n\times 2n},$$

where B is the vector of regression coefficient, X is the design matrix, and V is the variance–covariance matrix for all studies, respectively.

The generalized least-squared estimator $\widehat{\beta }$ of overall regression coefficients is given by

$$\widehat{\beta }={\left(X^{\prime }{V}^{-1}X\right)}^{-1}{X}^{\prime }{V}^{-1}B,$$

which has a normal distribution with mean β and covariance matrix Σ given by

$$\Sigma ={\left(X^{\prime }{V}^{-1}X\right)}^{-1}.$$

Under the null hypothesis of no association to either phenotype, that is, β = 0 (for both traits), the score statistic

$$T_{12}={\widehat{\beta }}^{\prime }{\Sigma }^{-1}\widehat{\beta }$$

will asymptotically follow a Chi-squared distribution with two degrees of freedom.

The two univariate test statistics are constructed similarly. Specifically,

$$T_1=\frac{{\widehat{\beta }}_1^2}{{\Sigma }_{11}},T_2=\frac{{\widehat{\beta }}_2^2}{{\Sigma }_{22}},$$

where ${\widehat{\beta }}_1$ and ${\widehat{\beta }}_2$ are two elements in $\widehat{\beta }$, and Σ₁₁ and Σ₂₂ are two diagonal elements in Σ. Under the null hypothesis of no univariate association, that is, β₁=0 or β₂=0, T₁ or T₂ will follow a Chi-squared distribution with 1 degree of freedom.

The above meta-analysis model was implemented in an in-house java program BiMeta.jar, which is available upon request to the corresponding authors.

Replication in the UKB sample

Significant findings from the discovery stage were replicated in the independent UKB sample. In brief, the UKB sample is a large prospective cohort study of ~ 500,000 participants from the United Kingdom, aged between 40 and 69 at recruitment. Ethics approval for the UKB study was obtained from the North West Centre for Research Ethics Committee (11/NW/0382), and informed consent was obtained from all participants. This study used the data requested under the UKB application number 41542, which was covered by the general ethical approval for the UKB study. Genome-wide genotypes for all subjects were available at 784,256 genotyped autosome markers and were imputed into UK10K haplotype, 1000 Genomes project phase 3 and Haplotype Reference Consortium (HRC) reference panels. All the included participants are those who self-reported as white (data field 21000). Participants who had a self-reported gender inconsistent with the genetic gender, who were genotyped but not imputed or who withdraw their consents were removed. A set of unrelated participants were then sampled for subsequent analysis with KING (Manichaikul et al. 2010).

Heel BMD (data field 3148) as evaluated by quantitative ultrasound speed of sound (SOS) and broadband ultrasound attenuation (BUA) was used for replicating FNK-BMD because of the large number of phenotyped subjects. Arm LBM was quantified as the sum of fat-free mass at arms (data fields 23121 and 23125). Leg LBM was quantified as the sum of fat-free mass at legs (data fields 23113 and 23117). Phenotype modeling for both FNK-BMD and LBM was similar to that in the discovery samples, with the exception that both phenotypes were mandatorily adjusted by the top ten principal components, to adjust for potential population structures. Association was again examined by the univariate/bivariate linear regression model.

Functional annotation

Functional annotation of the identified SNPs was performed using the bioinformatical software HaploReg (Ward and Kellis 2012). HaploReg provides functional information for non-coding SNPs with multiple functional categories, including conservation sites, DNase hypersensitivity sites (DHS), transcription factor binding sites (TFBS), promoter sites, enhancer sites, and others. We annotated lead SNPs and their neighbor SNPs with strong LD pattern (r² > 0.8). Meanwhile, 3DSNP (https://cbportal.org/3dsnp/) was used to identify the potential pathogenicity and function of the SNPs (Lu et al. 2017).

Result

The basic characteristics of the studied samples are summarized in Table 1. A total of 12,445 subjects from six samples are included in the FNK-BMD and LBM bivariate analyses. Principle components analysis (PCA) was applied to each sample and no population outliers were observed. The 1000 genomes sequencing project generated over 10 million qualified SNPs. After removing variants either of low frequency or of poor imputation accuracy, 12,061,510 variants are qualified for analysis.

Table 1.

The basic characteristics of study participants

Sample	Anc	Source	N	Female (%)	Age (years)	Height (cm)	FNK-BMD (g/cm2)	Arm LBM (kg)	Arm FBM (kg)	Leg LBM (kg)	Leg FBM (kg)
OOS	EUR	In-house	866	48.15	50.64(18.16)	170.98 (9.74)	0.81 (0.14)	6.47 (2.19)	3.04(1.34)	17.98 (4.54)	8.74 (3.53)
KCOS	EUR	In-house	2205	76.64	51.53(13.71)	166.23 (8.41)	0.79 (0.15)	5.64(1.90)	3.22(1.77)	16.81 (3.90)	9.04 (3.83)
COS	EAS	In-house	1538	50.58	34.71 (13.39)	164.31 (8.20)	0.81 (0.13)	4.92 (1.55)	1.76 (0.98)	14.91 (3.47)	4.82(1.91)
FHS	EUR	dbGAP	6547	44.60	56.00(13.76)	168.03 (9.88)	0.94 (0.16)	4.09 (0.77)	3.59 (2.02)	15.16(3.62)	8.41 (3.62)
WHI-AA	AFR	dbGAP	843	100.00	61.17(7.30)	162.75 (5.76)	0.82 (0.14)	–	–	13.00 (2.50)	14.15(5.07)
WHI-HIS	AMR	dbGAP	445	100.00	60.11 (7.51)	158.16(5.55)	0.73 (0.11)	–	–	10.79(1.96)	11.47 (3.85)

Data are presented as mean (sd)

OOS Omaha osteoporosis study, KCOS Kansas city osteoporosis study, COS Chinese osteoporosis study, FHS Framingham heart study, WHI-AA Women’s health initiative study African-American sub-sample, WHI-HIS Women’s health initiative study Hispanic sub-sample, Anc ancestor, EUR European population, EAS East Asian population, AFR African population, AMR Admixed American population. The sample size after quality control was reported. N Number, – raw phenotype was not available

Univariate and bivariate analysis of FNK‑BMD and LBM

A logarithmic quantile–quantile plot of meta-analysis test statistics for bivariate analyses showed a marked deviation in the tail of the distribution, implying the possible existence of true associations (Fig. 1). The Manhattan plot of the GWAS meta-analyses is displayed in Fig. 2.

Fig. 1.

QQ plot. Logarithmic quantile–quantile (QQ) plot of the discovery GWAS results. Results were plotted bivariate GWAS for FNK-BMD and arm LBM (red), FNK-BMD and arm LBM_adj (green), FNK-BMD and leg LBM (blue) and FNK-BMD and leg LBM_adj (yellow)

Fig. 2.

Manhattan plot. Manhattan plot of bivariate GWAS for FNK-BMD and arm LBM (red), FNK-BMD and arm LBM_adj (green), FNK-BMD and leg LBM (blue) and FNK-BMD and leg LBM_adj (yellow). The dotted line represents the genome-wide significance level (GWS, 5.0 × 10⁻⁸)

Univariate GWAS meta-analysis of FNK-BMD identified ten distinct loci at the genome-wide significance (GWS, 5.0 × 10⁻⁸) level: 1p31.3, 2q23.2, 2q36.3, 3q29, 5q14.3, 7q21.3, 8q21.3, 11p15.2, 11p15.5, and 17q12. Here, an independent locus was defined as a genomic region of 500 kb length from either side of the lead SNP.

Univariate GWAS meta-analysis of arm LBM identified two distinct loci 1q21.3 and 3q24. In the FBM-adjusted analysis, neither of them remained significant, while two additional loci 9q21.13 and 9p21.1 emerged to be significant. Univariate analysis of leg LBM identified eight distinct loci: 3q24, 5p22.3, 6p21.1, 6p25.2, 9q21.13, 11q21, 16q12.2, and 18q21.32. In the FBM-adjusted analysis, three of them, 3q24, 6p21.1 and 9q21.13, remained significant. Besides, six additional loci emerged to be significant: 3q21.3, 3q22.1, 3q27.1, 5q31.2, 10q22.3, and 17q21.31.

Together, univariate analyses identified 4 loci for arm LBM, 14 loci for leg LBM and 10 loci for FNK-BMD at the GWS level. Two of them (3q24 and 9q21.13) overlap, resulting in a total of 26 loci identified in the univariate analyses (Supplementary Table 1).

In the bivariate analysis of FNK-BMD and arm LBM, 5 of the above 26 loci remained significant at the GWS level. Among them, three loci are nominally significant (p < 0.05) in both univariate analyses: 2p23.2, 5q14.3, and 8q21.3. In the arm FBM-adjusted analysis, four of them are GWS significant, among which 6p21.1 and 8q21.3 are nominally significant in both univariate tests.

In the bivariate analysis of FNK-BMD and leg LBM, 9 of the above 26 loci remain significant at the GWS level. Among them, six loci are nominally significant in both univariate tests, including 5q14.3, 6q21.1, 8q21.3, 9q21.13, 16q12.2, and 18q21.32. In the leg FBM-adjusted analysis, eight of them remained significant at GWS level, among which three loci (5q14.3, 6p21.1, and 8q21.3) are nominally significant in both univariate tests.

In sum, the bivariate analyses identified seven distinct loci at the GWS level; all of them are nominally significant in both univariate analyses, implying a pleiotropic effect (Supplementary Table 2). The lead SNPs between univariate and bivariate analyses are identical or in strong LD with each other at these loci. Bivariate analyses also identified five loci (3q24, 5q22.3, 11q21, 3q22.1, and 17q21.3) that are only associated with LBM at the GWS level, but not with BMD even at the nominal level. The bivariate association signals at these five loci are likely to be driven by LBM alone. Likewise, two other loci (1p31.3 and 7q21.3) that are significant for BMD but not for LBM are likely to be driven by BMD alone.

Replication in the UKB sample

We replicated the above seven bivariately associated lead SNPs in the UKB sample. The lead SNPs rs61540635 at 2p23.2 and rs36088869 at 5q14.3 were not found in the imputed UKB genotypes and were replaced by the second lead SNPs rs4477866 (bivariate p = 3.47 × 10⁻⁸, LD r² = 0.61) and rs10037512 (bivariate p = 5.30 × 10⁻⁸, LD r² = 1.00), respectively. Of all the seven lead SNPs including rs4477866 and rs10037512, three are nominally significant in both univariate analyses and in bivariate analysis of the UKB sample, providing strong evidence of replication. These significant SNPs include rs4477866 (p_BMD = 2.00 × 10⁻¹³, p_{Arm LBM} = 1.20 × 10⁻⁵, bivariate p = 1.03 × 10⁻⁴), rs11152213 (p_BMD = 5.50 × 10⁻⁷, p_{Leg LBM} = 1.30 × 10⁻¹⁵⁰, bivariate p = 6.69 × 10⁻⁶) and rs1421085 (p_BMD = 6.40 × 10⁻¹⁵, p_{Leg LBM} = 1.20 × 10⁻³³⁸, bivariate p = 6.47 × 10⁻¹⁴). The main association results are listed in Table 2.

Table 2.

Main results of the identified lead SNPs in bivariate GWAS meta-analyses

LBM Site	Marker	Chr	Position	Locus	A1/A0	EAF	Discovery					Replication
							BMD		LBM		Bivariate	BMD		LBM/LBMadj		Bivariate
							Beta (SE)	P	Beta (SE)	P	P	Beta (SE)	P	Beta (SE)	P	P
Arm, unadjusted	rs61540635	2	29,047,358	2p23.2	W/M	0.54	−0.037 (0.015)	1.13 × 10−6	− 0.060 (0.015)	6.33 × 10−5	1.27 × 10−8	–	–	–	–	–
	rs4477866*	2	29,009,089	2p23.2	C/A	0.43	0.077 (0.014)	3.80 × 10−8	0.045 (0.015)	2.70 × 10−3	3.47 × 10−8	0.018(0.003)	2.00 × 10−13	0.009 (0.002)	1.20 × 10−5	1.03 × 10−4
Leg, unadjusted	rs 1421085	16	53,800,954	16q 12.2	T/C	0.59	− 0.042 (0.014)	2.70 × 10−3	− 0.083 (0.014)	3.06 × 10−9	2.04 × 10−9	− 0.019 (0.003)	6.40 × 10−15	− 0.078 (0.002)	1.20 × 10−338	6.47 × 10−14
	rsl 1152213	18	57,852,948	18q21.32	C/A	0.25	0.039 (0.015)	9.32 × 10−3	0.086 (0.015)	9.85 × 10−9	3.47 × 10−8	0.014(0.003)	5.50 × 10−7	0.060 (0.002)	1.30 × 10−150	6.69 × 10−6

Chr chromosome, A1 effect allele. Allele frequencies (EAF) are reported for the A1 allele. Beta regression coefficient, SE the standard deviation of beta, P the p value calculated in meta-analysis

^*Represents the proxy SNP

Functional annotation

We annotated the three successfully replicated lead SNPs and their neighbor SNPs (LD r² > 0.8) with HaploReg. rs4477866 is an intron variant in PPP1CB (protein phosphatase 1 catalytic subunit beta) gene. It has 27 variants with a strong LD structure (r² > 0.8), including a missense SNP rs1128416 and another SNP rs3190 located in the 3′-UTR of PPP1CB gene. In the GTEx project eQTL datasets (v6), polymorphisms at this variant are strongly associated with TRMT61B (tRNA methyltransferase 61B) gene expression in skeletal muscle (p = 4.31 × 10⁻²³). rs11152213 is located in the 3′-end (186 kb) of MC4R (melanocortin 4 receptor) gene. It is a DNAse hypersensitivity site (DHS). This SNP has been reported to be associated with obesity (p = 3 × 10⁻²²) (Berndt et al. 2013), and MC4R is an obesity candidate gene. Besides, rs6567160 (p = 9.35 × 10⁻¹³) in MC4R has been reported to be associated with LBM (Pei et al. 2014), and rs11152213 is in strong LD (r² = 0.96) with rs6567160. At last, rs1421085 is located in an intron of FTO (fat mass and obesity-associated protein) gene, which is a well-established obesity candidate gene. In the Roadmap epigenomic study, it was predicted to have enhancer activity in skeletal myoblasts cells, as implied by 25-state model and H3K4me1 histone marks. It was also predicted to have enhancer and promoter activity in osteoblast primary cells, as implied by core 15-state model, 25-state model, H3K4me1, and H3K27ac histone marks.

Through the 3D SNP annotation, rs61540635 may interact with TRMT61B, SNORD53, SNORD92, and WDR43 genes. rs11152213 may interact with MC4R gene. rs1421085 may interact with BBS2, CRNDE, IRX5, and MT4 genes.

Discussion

In the present study, by performing bivariate genome-wide association meta-analyses for FNK-BMD and LBM at arms and legs, and replicating the significant variants in the UKB cohort sample, we identified three genomic loci 2p23.2, 16q12.2, and 18q21.32 that may play a pleiotropic effect to both BMD and LBM. Two loci 16q12.2 and 18q21.32 have been reported to be associated with BMD and LBM by previous GWASs, while 2p23.2 is a novel locus.

Functional annotations highlighted several candidate genes, such as MC4R, FTO, PPP1CB, and TRMT61B. Among them, the protein encoded by MC4R gene is a membrane-bound receptor and member of the melanocortin receptor family. We previously found that rs6567160, near MC4R, had a pleiotropic effect to both FBM and LBM, implying that MC4R may regulate the development of both FBM and LBM (Pei et al. 2014). Braun et al. observed an increased LBM and femoral BMD phenotypes with aging in the mc4r knocked-out mice (Braun et al. 2012). FTO gene is a well-established obesity candidate gene. One previous study found that the loss of fto in mice leads to postnatal growth retardation and a significant reduction in adipose tissue and LBM (Fischer et al. 2009). Another study showed that FTO genotype was associated with both FBM and LBM and the associations were attenuated but remained significant after adjusting for each other (Sonestedt et al. 2011). The protein encoded by PPP1CB gene is one of the three catalytic subunits of protein phosphatase 1 (PP1) and is an effective adipogenic activator that promotes 3T3-L1 adipogenesis (Cho et al. 2015). The isoform PPP1CB is muscle-specific and involved in glycogen metabolism (Printen et al. 1997), while glycogen is a major fuel source in contracting skeletal muscles. However, the mechanism by which PPP1CB functions warrants further inquiry.

Genome-wide association studies (GWASs) and their meta-analyses have successfully identified many novel genomic loci for common complex diseases/traits of public health importance. The current GWASs and their meta-analyses generally focus on analyses of individual traits. In practice, multiple correlated phenotypes of interest are usually collected from a single study population. Joint consideration of such traits can provide additional information compared to information contained in individual traits, and hence, can provide greater power to detect pleiotropic loci that are important to the pathogenesis of many correlated human diseases. Earlier studies on association tests of multiple correlated traits have been shown to improve the statistical power to evaluate the effects of pleiotropic loci that jointly influence complex traits (Liu et al. 2009b; Tan et al. 2015). However, there has been surprisingly little work done on GWAS meta-analysis of multiple correlated traits. We here performed bivariate GWAS meta-analyses analyzing simultaneously two correlated diseases of public health significance, osteoporosis and sarcopenia. Moreover, aiming to maintain homogeneity, we performed bivariate analyses using LBM at arms and legs respectively instead of appendicular LBM (the sum LBM of arms and legs). We found that the loci identified in two analyses were not identical, implying a possible heterogeneity across skeletal sites.

The bivariate model indeed tests the association with either trait instead of with both traits, which is a fundamental limitation of all such bivariate association methods (Zhang et al. 2010; Zhou and Stephens 2014; Zhu et al. 2015). In our analysis, we checked the two univariates p values in both the discovery and replication samples. In the discovery samples, the bivariate association was declared only when they were both nominally significant. In the replication sample, we also checked the two univariates p values. Therefore, we believe that the bivariate association declared in this way may represent a pleiotropic effect on two traits, not just one trait.

In our study, the estimated heel BMD was used to replicate the femoral neck BMD. Despite being two different skeletal sites, heel BMD and femoral neck BMD may have shared genetic background. In a previous study, a total of 64 BMD SNPs from 56 loci identified at femoral neck and spine (Estrada et al. 2012). Of these loci, 54 were successfully replicated in the UKB heel BMD GWAS analysis (Kemp et al. 2017). Therefore, heel BMD could partially replicate the genetic variants identified in other skeletal sites including femoral neck, though the two traits were not perfectly matched.

In conclusion, our bivariate GWAS meta-analysis suggested that 2p23.2, 16q12.2, and 18q21.32 may play a pleiotropic role in both bone mass and muscle mass metabolism. Our findings provide useful insights that further enhance our understanding of the genetic association between BMD and LBM and provide a rationale for subsequent functional studies of these implicated genes in the pathophysiology of diseases related to osteoporosis and sarcopenia.