Genetics of Bone Mass in Childhood and Adolescence: Effects of Sex and Maturation Interactions

Abstract

We aimed to determine if adult bone mineral density (BMD) susceptibility loci were associated with pediatric bone mass and density, and if sex and pubertal stage influenced any association. We analyzed prospective areal BMD (aBMD) and bone mineral content (BMC) data from the Bone Mineral Density in Childhood Study (n =603, European ancestry, 54% female). Linear mixed models were used to assess if 77 single-nucleotide polymorphisms (SNPs) near known adult BMD susceptibility loci interacted with sex and pubertal stage to influence the aBMD/BMC; adjusting for age, BMI, physical activity, and dietary calcium. The strongest main association was observed between an SNP near C7orf58 and distal radius aBMD. However, this association had a significant sex•SNP interaction, revealing a significant association only in females (b =−0.32, p =1.8 × 10⁻⁶). Furthermore, the C12orf23 locus had significant interactions with both sex and pubertal stage, revealing associations in females during Tanner stage I for total hip aBMD (b =0.24, p =0.001) and femoral neck aBMD (b =0.27, p =3.0 × 10⁻⁵). In contrast, the sex•SNP interactions for loci near LRP5 and WNT16 uncovered associations that were only in males for total body less head BMC (b =0.22, p =4.4 × 10⁻⁴) and distal radius aBMD (b =0.27, p =0.001), respectively. Furthermore, the LRP5 locus interacted with both sex and pubertal stage, demonstrating associations that were exclusively in males during Tanner V for total hip aBMD (b =0.29, p =0.003). In total, significant sex•SNP interactions were found at 15 loci; pubertal stage•SNP interactions at 23 loci and 19 loci interacted with both sex and pubertal stage. In conclusion, variants originally associated with adult BMD influence bone mass in children of European ancestry, highlighting the fact that many of these loci operate early in life. However, the direction and magnitude of associations for a large number of SNPs only became evident when accounting for sex and maturation.

Introduction

Optimizing peak bone mass (PBM) in early adulthood is one of the most important factors in preventing osteoporosis⁽¹⁾ and fracture later in life.^(2,3) Epidemiological studies suggest that a 10% increase in PBM at the population level would decrease the risk of fracture later in life by 50%.⁽⁴⁾ Lifestyle factors influence the accumulation and loss of bone over the life course.⁽⁵⁾ However, there is strong evidence for a genetic component in the predisposition of osteoporosis, with an estimated 60% to 80% of the variability in the risk explained by heritable factors.^(6–8) Twin studies also suggest that genetic predisposition determines up to 80% of PBM.⁽⁹⁾ Thus, understanding the genetic contribution to PBM is critical to developing effective intervention strategies for the prevention of osteoporosis and fractures.⁽¹⁰⁾

In 2008, OPG (TNFRSF11B) and LRP5 were the first adult bone mineral density (BMD) susceptibility loci identified in a genome-wide association study (GWAS).⁽¹¹⁾ To date, 56 adult BMD associated loci and 14 fracture risk–associated loci have been identified.⁽¹²⁾ However, the extent to which these loci impact bone accrual in early life prior to PBM versus affecting bone loss post-PBM is not known. A systematic investigation to assess whether adult BMD susceptibility loci are associated with bone density and content during childhood and adolescence is needed.

To investigate the potential role of adult bone loci on pediatric bone mass accrual, sex and pubertal stage need to be taken into account. During their lifetime women lose about 30% to 50% of PBM, whereas men lose 20% to 30% of PBM;⁽¹³⁾ the risk of osteoporosis is greater in women relative to men. Further, 26% of PBM is gained during the 2 years of peak bone accretion during adolescence⁽¹⁴⁾ and this pattern suggests that the regulation of bone accretion varies across maturational stages. Using a longitudinal study design and subjects of European ancestry, we aimed to determine if adult bone loci were associated with BMD or bone mineral content (BMC) during childhood, and if consideration of sex and pubertal stage provided new insights into such associations.

Subjects and Methods

Study sample

Participants from the Bone Mineral Density in Childhood Study (BMDCS) were invited to donate a blood or saliva sample at their final study visit. The BMDCS was a multicenter longitudinal study to establish norms for BMC and areal-BMD (aBMD) for children 5 to 20 years of age in the United States. Children were recruited from Children’s Hospital of Los Angeles (Los Angeles, CA), Cincinnati Children’s Hospital Medical Center (Cincinnati, OH), Creighton University (Omaha, NE), Children’s Hospital of Philadelphia (CHOP) (Philadelphia, PA), and Columbia University (New York, NY).^(15,16) Females aged 6 to 15 years and males aged 6 to 16 years were enrolled in 2002–2003 and were measured annually for 6 years (up to 7 visits). Additional older (age 19 years) and younger (age 5 years) subjects were enrolled in 2006–2007 and evaluated annually for 2 years (up to 3 visits) to extend the reference percentiles from ages 5 to 20 years.

Criteria for BMDCS entry were selected in order to enroll normally developing children with healthy bones. Key criteria included term birth (≥37 weeks’ gestation), birth weight >2.3 kg, no evidence of precocious or delayed puberty, and height, weight, or BMI within the 3rd to the 97th percentiles for age. Children were excluded for multiple fractures (more than two fractures if age <10 years or more than three fractures if age >10 years), current or previous medication use or medical condition known to affect bone health, and extended bed rest. Same sex siblings were excluded, but opposite sex siblings were not excluded from participation in BMDCS, because separate aBMD references curves were generated for males and females.

Written informed consent was obtained from the study participants 18 years and older. For participants less than 18 years of age, consent was obtained from the parent or guardian and assent was obtained from participants. The protocol was approved by the Institutional Review Board of each Clinical Center.

Genotyping

We performed high-throughput genomewide SNP genotyping, using the Illumina Infinium II OMNI Express plus Exome BeadChip technology (Illumina, San Diego, CA, USA), at CHOP’s Center for Applied Genomics, as described.⁽¹⁷⁾ The SNPs analyzed survived the filtering of the genome wide dataset for SNPs with call rates <95%, minor allele frequency <1%, missing rate per person >2%, and Hardy-Weinberg equilibrium p < 1 × 10⁻⁵. We used previously reported SNPs or the best one or two surrogate SNPs available based on the CEU HapMap. In our study, we assessed 77 SNPs in or near known adult bone mass loci that were investigated using additive models.^{(11,12,18,19)}

Bone phenotypes

DXA scans were obtained using Hologic, Inc. (Bedford, MA, USA) bone densitometers (QDR4500A, QDR4500W, Delphi A, and Discovery models). Scans were obtained following manufacturer guidelines for patient positioning at the lumbar spine, proximal femur, forearm, and total body by trained research technicians using standardized protocols. Scans were centrally analyzed by the DXA Core Laboratory (University of California, San Francisco) using Hologic software version 12.3 for baseline scans and Apex 2.1 for follow-up scan analysis using the “compare” feature. aBMD values for the spine, total hip, femoral neck, and distal 1/3 radius, and BMC of the total body less head (TBLH) were adjusted based on the cross-calibration of DXA devices and longitudinal calibration stability using anthropomorphic spine and hip phantoms, and the Hologic whole-body phantom. BMC (rather than aBMD) of the TBLH was used because it is the preferred measure of bone status for the total body when adjusted for body size.^(20,21) aBMD/BMC Z-scores were calculated using the BMDCS reference values to account for the known increases and sex differences in aBMD/BMC during growth and development.⁽¹⁶⁾ aBMD/BMC Z-scores were adjusted for height-for-age Z-scores as described to minimize potential confounding by skeletal size.⁽²²⁾

Additional measures

Weight was measured on a digital scale and height was measured using a stadiometer.^(15,16) Height, weight, and BMI Z-scores were calculated using the CDC 2000 growth charts.⁽²³⁾ Race and ethnicity were self-identified by each participant using the National Institutes of Health and the U.S. Bureau of the Census classifications. Population ancestry was also confirmed using the participant’s genetic information using principal components analysis. Only participants of European ancestry by both criteria were included in our study.

Pubertal stage was assigned based on a physical examination by an experienced physician or nurse skilled in pediatric endocrinology. The participants were categorized as prepubertal (Tanner I), pubertal (Tanner II–IV), or postpubertal (Tanner V). Pubertal stage categorization in the females was based on breast development and Tanner criteria,^(24,25) and in the males was based on testicular volume measured by a Prader ochidometer.⁽²⁴⁾

Dietary calcium intake was assessed using a semiquantitative food frequency questionnaire (FFQ) (Block Dietary Data Systems, Berkeley, CA, USA). The FFQ consisted of 45 food and beverage items; the reported frequency and amount of intake in the last week was used to estimate calcium intake (mg/day) using an automated computer analysis program.

Physical activity (hours/day) was estimated using an expanded version of the questionnaire originally validated by Slemenda and colleagues.⁽²⁶⁾ Over 40 different physical and sedentary activities were queried. Responses were tabulated to estimate hours of physical activity per week.

Statistical methods

The number of visits per subject varied from one to seven in our longitudinal study (64% had 7 visits). We treated the subjects as a random sample from a larger population to which we wished to draw inferences. Using mixed-effect linear regression we modeled the between-subject variability as a random effect (random intercept term at the subject level), accounting for correlations arising from repeated-measures taken from each subject. Mixed-effects models allow use of all data under assumption of missing at random. We fitted random intercept models using the method of maximum likelihood (ML) estimation and employed the Huber-White approach to construct robust standard errors that give valid inferences for large sized samples. Briefly, letting “i” denote the ith subject and “j” denote the jth visit on that subject, for a given aBMD/BMC Z-score, we specified the following regression model examining the relationship of aBMD/BMC Z-score with time-invariant variables (ie, variables that do not change from visit-to-visit: SNP and sex) as well as with time-dependent variables (ie, variables that do change from visit-to-visit: age, puberty stage, BMI Z-score, physical activity, and dietary calcium):

$$\mathit{aBMD}/{\mathit{BMC}}_{ij}={\beta }_0+{\beta }_1{\mathit{Age}}_{ij}+{\beta }_2\mathit{BMIZ}{\mathit{score}}_{ij}+{\beta }_3\mathit{Physical}{\text{Activity}}_{ij}+{\beta }_4\mathit{Dietary}{\mathit{calcium}}_{ij}+{\beta }_5{\mathit{Tanner}}_{ij}+{\beta }_6{\mathit{Sex}}_i+{\beta }_7{\mathit{SNP}}_i+u_i+e_{ij}$$

with $e_{ij}~i.i.d.N(0,{\sigma }_e^2)$ and independently $u_i~i.i.d.N(0,{\sigma }_j^{\mathit{sub}})$

The fixed portion of the model included age, BMI Z-score, physical activity, dietary calcium, Tanner stage, sex, and SNP, providing an overall regression line representing the population average. The random effect (u_i) serves to shift this regression line up or down according to each subject. The xtmixed procedure in Stata 12.0 (StataCorp LP, College Station, TX, USA) was used to perform the statistical analysis. Each SNP was tested individually as an additive trait (ie, coded 0, 1, and 2) to assess the association between each SNP and each bone outcome adjusting for the covariates. We then included in the fixed portion of the model a series of interaction terms. Three-way interaction effects of SNP, sex, and puberty stage were used to determine whether any association varied as a function of both sex and maturation stage. Two-way interaction effects of SNP by sex were used to determine whether any association between a SNP and bone Z-scores varied as a function of sex; and two-way interaction effects of SNP by puberty stage were used to determine whether any association varied as a function of maturation stage. For the models that included three-way interactions, the lower order interactions were also included. The overall interaction p values were extracted using a “contrast” statement in Stata after fitting the models and the sex and/or puberty stage specific beta coefficients and p values were extracted using a “margins” statement from the same fitted model. If we observed an interaction term with a p <0.05, we extracted the sex and/or Tanner stage-specific SNP beta coefficients, standard errors, and p values from the model. The strata-specific SNP associations were considered significant at p <0.05, because all loci being tested are known to associate with adult bone density.^(27,28) However, to address the issue of multiple testing, we applied the Benjamini and Hochberg false discovery rate (FDR) method to derive corrected p values by accounting for the number of independent loci and the number of skeletal sites investigated for each interaction analysis.⁽²⁹⁾ Finally, we performed a series of sensitivity analyses to assess the confidence of our finding by: ⁽¹⁾ using restricted maximum likelihood (REML) in place of ML; ⁽²⁾ removing age as a covariate; and ⁽³⁾ restricting our analyses to those with more than three study visits.

Results

The final European ancestry cohort with complete data included 603 individuals at baseline (Supporting Table 1). A total of 20 SNPs yielded at least nominal significant evidence of association with aBMD/BMC at one or more skeletal sites (p <0.05) (Supporting Table 2). The two loci most strongly associated with bone mass were C7orf58 and LIN7C (Table 1). The rs13245690 effect allele (C7orf58) was associated with lower distal radius aBMD (b =−0.20, p =1.8 × 10⁻⁴) whereas the rs7104230 effect allele (LIN7C) was associated with lower femoral neck aBMD (b =−0.14, p = 0.005), TBLH BMC (b = −0.08, p =0.033), and total hip aBMD (b =−0.14, p =0.010).

Table 1.

Association Between Known Adult Bone Loci and Pediatric Bone Mass (full results in Supporting Table 2)

Chr	SNP	Minor	Major	MAF	Nearest gene	Skeletal site	Beta ± SEa	pa
7	rs13245690	G*	A	0.37	C7orf58	Rad aBMD	−0.20 ±0.05	1.8 × 10−4
11	rs7104230	T*	C	0.47	LIN7C	FN aBMD	−0.14 ±0.05	0.005
						TH aBMD	−0.14 ±0.05	0.010
						TBLH BMC	−0.08 ±0.04	0.033

Chr =chromosome; SNP =single-nucleotide polymorphism; MAF =minor allele frequency; SE =standard error; Rad =distal radius; aBMD =areal bone mineral density; FN =femoral neck; TH =total hip; TBLH =total body less head; BMC =bone mineral content.

^aBeta coefficients, SEs, and p values were derived from linear mixed models that were adjusted for age (years), sex, Tanner stage, body mass index (Z-score), physical activity (hours/week), and dietary calcium (g/day). An additive model was used and so the beta coefficients are interpreted as the difference in aBMD/BMC Z-score per effect allele (as indicated by * in the minor or major column).

We then tested whether any sex•SNP interactions were associated with aBMD/BMC, and observed 15 such interactions (p <0.05, Supporting Tables 3 and 4). Eleven of these SNPs yielded sex-specific associations (Table 2). The previous association with C7orf58 and lower distal radius aBMD was driven by females (b =−0.32, p =1.8 × 10⁻⁶), and the same variant was associated with lower TBLH-BMC in females (b =−0.11, p =0.034). In contrast, C7orf58 was associated with higher spine aBMD (b =0.19, p =0.034) and higher total hip aBMD (b =0.17, p =0.033) in males (Table 2). Other loci associated with bone outcomes in females included TXNDC3 (femoral neck aBMD: b =0.18, p =0.022; TBLH BMD: b =0.14, p =0.030), MPP7 (distal radius aBMD: b =−0.24, p =0.001), and C12orf23 (TBLH BMC: b =0.19, p =1.3 × 10⁻⁴; spine aBMD: b =0.21, p =0.002). Other loci associated with bone outcomes in males included LRP5 (TBLH BMC: b =0.22, p =4.4 × 10⁻⁴), KLHDC5/PTHLH (radius aBMD: b =0.29, p =0.002; spine aBMD: b =0.20, p =0.048; and TBLH BMC: b =0.14, p =0.046), WNT16 (radius aBMD: b =0.27, p =0.001), DNM3 (femoral neck aBMD: b =0.21, p =0.013), ESR1 (distal radius aBMD: b =−0.21, p = 0.027), and RANKL/AKAP11 (TBLH BMD: b =−0.15, p =0.008) (Table 2). Correcting for multiple testing, the following loci remained statistically significantly associated with aBMD/BMC in males or females (Table 2): C7orf58 (female, distal radius), WNT16 (male, distal radius), MPP7 (female, distal radius), LRP5 (male, TBLH BMC), and C12orf23 (female, TBLH BMC).

Table 2.

SNP-Sex Interactions and Pediatric Bone Mass (full results in Supporting Table 4)

Chr	SNP	Minor	Major	MAF	Nearest gene	Skeletal site	p interaction	Males		Females
								Beta ±SEa	pa	Beta ± SEa	pa
1	rs2586392	C*	T	0.27	DNM3	FN aBMD	0.029	0.21 ±0.08	0.013	−0.04 ±0.08	0.595
6	rs7751941	A	G*	0.23	ESR1	Rad aBMD	0.015	−0.21 ±0.10	0.027	0.10 ±0.08	0.244
7	rs13245690	G*	A	0.37	C7orf58	Rad aBMD	0.011	−0.05 ±0.08	0.521	−0.32 ±0.07	1.8 × 10−6
7	rs13245690	G*	A	0.37	C7orf58	Sp aBMD	0.006	0.19 ±0.09	0.034	−0.13 ±0.08	0.081
7	rs13245690	G*	A	0.37	C7orf58	TBLH BMC	0.009	0.10 ±0.06	0.105	−0.11 ±0.05	0.034
7	rs13245690	G*	A	0.37	C7orf58	TH aBMD	0.013	0.17 ±0.08	0.033	−0.10 ±0.07	0.174
7	rs10276139	C*	T	0.16	TXNDC3	FN aBMD	0.024	−0.10 ±0.10	0.303	0.18 ±0.08	0.022
7	rs10276139	C*	T	0.16	TXNDC3	Rad aBMD	0.026	−0.19 ±0.08	0.020	0.08 ±0.09	0.371
7	rs10276139	C*	T	0.16	TXNDC3	TBLH BMC	0.014	−0.10 ±0.08	0.168	0.14 ±0.06	0.030
7	rs3779381	G*	A	0.26	WNT16	Rad aBMD	0.026	0.27 ±0.08	0.001	0.02 ±0.07	0.837
10	rs4568902	G*	A	0.22	MPP7	Rad aBMD	0.004	0.09 ±0.09	0.300	−0.24 ±0.08	0.001
11	rs16921914	A	G*	0.29	DCDC5	FN aBMD	0.010	0.18 ±0.08	0.018	−0.09 ±0.07	0.214
11	rs273592	C*	T	0.33	DCDC5	FN aBMD	0.033	−0.07 ±0.07	0.347	0.14 ±0.07	0.034
11	rs16921914	A	G*	0.29	DCDC5	TH aBMD	0.005	0.21 ±0.08	0.010	−0.11 ±0.08	0.176
11	rs273592	C*	T	0.33	DCDC5	TH aBMD	0.020	−0.10 ±0.07	0.184	0.13 ±0.07	0.047
11	rs3781586	A	C*	0.14	LRP5	TBLH BMC	0.004	0.22 ±0.06	4.4 × 10−4	−0.05 ±0.07	0.471
12	rs1053051	T	C*	0.49	C12orf23	Rad aBMD	0.031	−0.16 ±0.08	0.049	0.07 ±0.07	0.310
12	rs1053051	T	C*	0.49	C12orf23	Sp aBMD	0.001	−0.13 ±0.08	0.088	0.21 ±0.07	0.002
12	rs1053051	T	C*	0.49	C12orf23	TBLH BMC	0.001	−0.07 ±0.06	0.221	0.19 ±0.05	1.3 × 10−4
12	rs7304170	T	C*	0.18	KLHDC5/PTHLH	Rad aBMD	0.008	0.29 ±0.09	0.002	−0.03 ±0.08	0.668
12	rs7304170	T	C*	0.18	KLHDC5/PTHLH	Sp aBMD	0.013	0.20 ±0.10	0.048	−0.14 ±0.09	0.135
12	rs7304170	T	C*	0.18	KLHDC5/PTHLH	TBLH BMC	0.015	0.14 ±0.07	0.046	−0.10 ±0.07	0.149
13	rs9594738	T	C*	0.47	RANKL/AKAP11	TBLH BMC	0.017	−0.15 ±0.06	0.008	0.03 ±0.05	0.545

SNP =single-nucleotide polymorphism; Chrn=chromosome; MAF =minor allele frequency; SE =standard error; FN =femoral neck; aBMD =areal bone mineral density; Rad =distal radius; Sp =spine; TBLH =total body less head; BMC =bone mineral content; TH =total hip.

^aBeta coefficients, SEs, and p values were derived from linear mixed models that were adjusted for age (years), pubertal stage, body mass index (Z-score), physical activity (hours/week), and dietary calcium (g/day). An additive model was used and so the beta coefficients are interpreted as the difference in aBMD/BMC Z-score per effect allele (as indicated by * in the minor or major column). Values of p in bold and italicized remain significant after correction for multiple testing using the Benjamini and Hochberg false discovery rate procedure.

We then tested whether any SNPs had significant interactions with pubertal stage to influence aBMD/BMC, and observed 23 such interactions (p <0.05, Supporting Tables 3 and 5). Nine of these loci yielded maturation-specific associations (Table 3). The following loci were particularly associated with bone mass at one or more skeletal sites before Tanner stage V was reached: WNT4, MHC, and TNFRSF11B (Table 3). For example, the rs6993813 (TNFRSF11B) effect allele was associated with lower spine aBMD (b =−0.16, p =0.002) and lower TBLH BMC (b =−0.12, p =0.002) during Tanner stage I. The following loci were particularly associated with bone mass once Tanner stage V was reached: SPTNB1, LIN7C, SOX9, and GPATCH1 (Table 3). For example, the rs7104230 (LIN7C) effect allele was associated with lower total hip aBMD (b =−0.18, p =0.002) during Tanner stage V. Correcting for multiple testing, the following loci remained statistically significantly associated with aBMD/BMC in specific Tanner stages (Table 3): TNFRSF11B (Tanner I and II–IV, spine; and Tanner I, TBLH BMC) and GPATCH1 (Tanner V, TBLH BMC).

Table 3.

SNP-Maturation Interactions and Pediatric Bone Mass (full results in Supporting Table 5)

Chr	SNP	Minor	Major	MAF	Nearest gene	Skeletal site	p interaction	Tanner I		Tanner II–IV		Tanner V
								Beta ± SEa	pa	Beta ± SEa	pa	Beta ± SEa	pa
1	rs17130546	G*	A	0.08	WLS	FN aBMD	0.015	−0.21 ± 0.10	0.042	−0.26 ± 0.10	0.010	−0.08 ± 0.11	0.443
1	rs12042083	A	G*	0.20	WNT4	Sp aBMD	0.022	0.15 ± 0.07	0.043	0.03 ± 0.07	0.613	0.07 ± 0.08	0.367
2	rs11898505	A	G*	0.39	SPTBN1	Sp aBMD	0.033	−0.04 ± 0.06	0.501	−0.06 ± 0.05	0.306	−0.17 ± 0.07	0.017
2	rs6752877	G*	T	0.40	SPTBN1	Sp aBMD	0.016	0.07 ± 0.06	0.249	0.08 ± 0.05	0.124	0.21 ± 0.07	0.003
6	rs3130340	C*	T	0.21	MHC	Rad aBMD	0.010	0.15 ± 0.07	0.033	0.03 ± 0.07	0.632	−0.07 ± 0.08	0.371
8	rs13277230	T	C*	0.46	TNFRSF11B	Rad aBMD	0.001	−0.15 ± 0.06	0.015	−0.14 ± 0.05	0.010	0.03 ± 0.06	0.595
8	rs6993813	T	C*	0.48	TNFRSF11B	Rad aBMD	0.010	−0.17 ± 0.06	0.007	−0.17 ± 0.05	0.002	−0.04 ± 0.06	0.579
8	rs6993813	T	C*	0.48	TNFRSF11B	Sp aBMD	0.017	−0.16 ± 0.05	0.002	−0.16 ± 0.05	0.001	−0.05 ± 0.06	0.411
8	rs13277230	T	C*	0.46	TNFRSF11B	TBLH BMC	0.004	−0.08 ± 0.04	0.028	−0.05 ± 0.04	0.199	0.04 ± 0.04	0.326
8	rs6993813	T	C*	0.48	TNFRSF11B	TBLH BMC	0.025	−0.12 ± 0.04	0.002	−0.06 ± 0.04	0.106	−0.01 ± 0.04	0.853
11	rs7104230	T*	C	0.47	LIN7C	TH aBMD	0.008	−0.12 ± 0.06	0.057	−0.08 ± 0.06	0.152	−0.18 ± 0.06	0.002
17	rs12937692	A	G*	0.22	C17orf53/HDAC5	Rad aBMD	0.040	−0.05 ± 0.07	0.494	−0.14 ± 0.06	0.034	−0.03 ± 0.08	0.736
17	rs7217932	A	G*	0.48	SOX9	TBLH BMC	0.036	0.02 ± 0.04	0.602	0.03 ± 0.04	0.450	0.10 ± 0.05	0.031
19	rs2287679	C*	T	0.28	GPATCH1	TBLH BMC	0.041	0.03 ± 0.05	0.551	0.08 ± 0.04	0.048	0.14 ± 0.05	0.004

SNP =single-nucleotide polymorphism; Chr =chromosome; MAF =minor allele frequency; SE =standard error; FN =femoral neck; aBMD =areal bone mineral density; Sp =spine; Rad =distal radius; TBLH =total body less head; BMC =bone mineral content; TH =total hip.

^aBeta coefficients, SEs, and p values were derived from linear mixed models that were adjusted for age (years), sex, body mass index (Z-score), physical activity (hours/week), and dietary calcium (g/day). An additive model was used and so the beta coefficients are interpreted as the difference in aBMD/BMC Z-score per effect allele (as indicated by * in the minor or major column). Values of p in bold and italicized remain significant after correction for multiple testing using the Benjamini and Hochberg false discovery rate procedure.

We then tested whether any SNP had significant interactions with both sex and pubertal stage to influence bone outcomes, and observed 19 such interactions (Supporting Tables 3 and 6). Twelve of these loci yielded sex-and-maturation-specific associations: GALNT3, PKDCC, CTNNB1, ABCF2, TXNDC3, TNFRSF11B, CPN1, MBL2, LRP5, C12orf23, RANKL/AKAP11, and MARK3 (Table 4). For example, specific to females during Tanner stage I the rs12185748 (GALNT3) effect allele was associated with lower spine aBMD (b =−0.26, p =0.001), total hip aBMD (b =−0.16, p =0.020), and TBLH BMC (b =−0.14, p =0.015). Also specific to females, the rs1053051 (C12orf23) effect allele was associated with higher total hip aBMD and femoral neck aBMD and the strongest associations were observed during Tanner stage I (total hip: b =0.24, p =0.001; femoral neck: b =0.27, p =3.0 × 10⁻⁵). Specific to males once Tanner stage V was reached, the rs9594738 (RANKL/AKAP11) effect allele was associated with lower distal radius aBMD (b =−0.29, p =0.001) and the rs3781586 (LRP5) effect allele was associated with higher total hip aBMD (b =0.29, p =0.003). The C12orf23 locus interaction with sex and pubertal stage for femoral neck aBMD is shown in Fig. 1 to illustrate a SNP interaction with both sex and maturation. Further, Supporting Table 7 compares the direction of the SNP associations we observed with those reported in adult GWAS. Correcting for multiple testing, the following loci remained statistically significantly associated with aBMD/BMC in males or females at specific Tanner stages (Table 4): GALNT3 (female/Tanner I, spine) and C12orf23 (female/Tanner I, femoral neck).

Table 4.

SNP-Sex-Maturation Interactions and Pediatric Bone Mass (full results in Supporting Table 6)

Chr	SNP	Minor	Major	MAF	Nearest gene	Skeletal site	p interaction	Sex	Tanner I		Tanner II–IV		Tanner V
									Beta ± SEa	pa	Beta ± SEa	pa	Beta ± SEa	pa
2	rs12185748	T	C*	0.49	GALNT3	Sp aBMD	5.4×10−5	Male	0.02 ±0.08	0.792	−0.11 ±0.08	0.180	−0.12 ± 0.10	0.206
								Female	−0.26 ±0.07	0.001	−0.08 ±0.07	0.286	−0.03 (0.08)	0.683
2	rs12185748	T	C*	0.49	GALNT3	TBLH BMC	0.049	Male	−0.05 ±0.06	0.332	−0.10 ±0.06	0.089	−0.10 (0.07)	0.163
								Female	−0.14 ±0.06	0.015	−0.06 ±0.05	0.257	−0.02 (0.06)	0.762
2	rs12185748	T	C*	0.49	GALNT3	TH aBMD	0.003	Male	−0.11 ±0.08	0.185	−0.14 ±0.08	0.077	−0.13 (0.09)	0.127
								Female	−0.16 ±0.07	0.020	−0.02 ±0.07	0.713	0.00 (0.07)	0.976
2	rs10205005	C*	T	0.24	PKDCC	Rad aBMD	0.001	Male	0.02 ±0.09	0.832	−0.04 ±0.09	0.639	−0.23 (0.11)	0.034
								Female	−0.00 ±0.11	0.965	0.03 ±0.09	0.754	0.23 (0.10)	0.019
3	rs430727	T	C*	0.44	CTNNB1	Sp aBMD	0.016	Male	0.07 ±0.09	0.437	0.20 ±0.08	0.017	0.24 (0.10)	0.012
								Female	0.11 ±0.08	0.185	0.04 ±0.07	0.566	0.00 (0.08)	0.978
7	rs6967282	G*	A	0.23	ABCF2	TBLH BMC	0.014	Male	−0.09 ±0.07	0.210	0.01 ±0.07	0.911	−0.08 (0.09)	0.375
								Female	0.13 ±0.07	0.045	0.02 ±0.06	0.778	−0.03 (0.07)	0.653
7	rs10276139	C*	T	0.16	TXNDC3	FN aBMD	0.047	Male	−0.05 ±0.10	0.587	−0.07 ±0.10	0.483	−0.16 (0.11)	0.166
								Female	0.07 ±0.09	0.464	0.18 ±0.09	0.036	0.28 (0.10)	0.004
7	rs10276139	C*	T	0.16	TXNDC3	TH aBMD	0.021	Male	−0.12 ±0.10	0.227	−0.09 ±0.11	0.385	−0.21 (0.12)	0.067
								Female	0.09 ±0.09	0.323	0.13 ±0.09	0.142	0.24 (0.10)	0.012
8	rs6993813	T	C*	0.48	TNFRSF11B	Sp aBMD	0.050	Male	−0.20 ± 0.08	0.009	−0.13 ±0.07	0.069	0.06 (0.09)	0.473
								Female	−0.14 ±0.08	0.073	−0.19 ±0.07	0.005	−0.14 (0.08)	0.079
10	rs11599750	T	C*	0.40	CPN1	FN aBMD	0.015	Male	−0.19 ±0.08	0.017	−0.05 ±0.08	0.562	0.03 (0.09)	0.761
								Female	−0.10 ±0.07	0.174	−0.13 ±0.08	0.096	−0.16 (0.08)	0.044
10	rs7898709	G*	T	0.13	MBL2	FN aBMD	0.040	Male	0.00 ±0.10	0.971	0.11 ±0.11	0.327	−0.02 (0.12)	0.899
								Female	−0.10 ±0.11	0.368	−0.22 ±0.11	0.040	−0.15 (0.12)	0.194
11	rs3781586	A	C*	0.14	LRP5	TH aBMD	0.043	Male	0.12 ±0.11	0.262	0.18 ±0.11	0.098	0.29 (0.10)	0.003
								Female	0.03 ±0.12	0.806	−0.06 ±0.12	0.643	−0.12 (0.13)	0.343
12	rs1053051	T	C*	0.49	C12orf23	FN aBMD	0.002	Male	−0.04 ±0.08	0.587	0.10 ±0.08	0.213	0.06 (0.08)	0.505
								Female	0.27 ±0.07	3.0 × 10−5	0.17 ±0.07	0.014	0.17 (0.08)	0.022
12	rs1053051	T	C*	0.49	C12orf23	TH aBMD	0.002	Male	−0.03 ±0.09	0.736	0.09 ±0.08	0.285	0.06 (0.09)	0.459
								Female	0.24 ±0.07	0.001	0.15 ±0.07	0.033	0.16 (0.07)	0.034
13	rs9594738	T	C*	0.47	RANKL/AKAP11	Rad aBMD	0.004	Male	0.05 ±0.09	0.550	−0.11 ±0.08	0.183	−0.29 (0.09)	0.001
								Female	0.07 ±0.09	0.409	0.12 ±0.07	0.101	0.14 (0.08)	0.086
14	rs2010281	A	G*	0.35	MARK3	Rad aBMD	0.015	Male	−0.03 ±0.11	0.788	0.00 ±0.09	0.996	0.28 (0.11)	0.009
								Female	−0.02 ±0.09	0.785	−0.01 ±0.07	0.886	0.01 (0.07)	0.876
14	rs2273703	G*	C	0.36	MARK3	Rad aBMD	0.015	Male	0.03 ±0.11	0.788	−0.00 ±0.09	0.996	−0.28 (0.11)	0.009
								Female	0.02 ±0.09	0.826	0.01 ±0.07	0.872	−0.01 (0.07)	0.879
14	rs2010281	A	G*	0.35	MARK3	TH aBMD	0.010	Male	−0.09 ±0.09	0.322	−0.17 ±0.08	0.036	0.02 (0.09)	0.799
								Female	−0.06 ±0.07	0.447	0.01 ±0.07	0.842	0.04 (0.07)	0.574
14	rs2273703	G*	C	0.36	MARK3	TH aBMD	0.025	Male	0.09 ±0.09	0.322	0.17 ±0.08	0.036	−0.02 (0.09)	0.799
								Female	0.03 ±0.07	0.680	−0.02 ±0.07	0.820	−0.05 (0.07)	0.526

SNP =single-nucleotide polymorphism; Chr =chromosome; MAF =minor allele frequency; SE =standard error; Sp =spine; aBMD =areal bone mineral density; TBLH =total body less head; BMC = bone mineral content; TH =total hip; Rad =distal radius; FN =femoral neck.

^aBeta coefficients, SEs, and p values were derived from linear mixed models that were adjusted for age (years), body mass index (Z-score), physical activity (hours/week), and dietary calcium (g/day). An additive model was used and so the beta coefficients are interpreted as the difference in aBMD/BMC Z-score per effect allele (as indicated by * in the minor or major column). Values of p in bold and italicized remain significant after correction for multiple testing using the Benjamini and Hochberg false discovery rate procedure.

Fig. 1.

Example of an SNP-sex-maturation interaction at the femoral neck. SNP rs1053051 interacted with sex and maturation (p interaction = 0.002) to influence FN aBMD. The rs1053051 effect allele was associated with FN aBMD in the females and most strongly during Tanner stage I (b =0.27, p =3.0 × 10⁻⁵). FN =femoral neck; aBMD =areal bone mineral density.

Our findings did not change when we used REML instead of ML (data not shown). Similarly, our findings did not change when we removed age as a covariate or when we restricted our analysis to those with more than three study visits (data not shown).

Discussion

C7orf58 and LIN7C were the loci most strongly associated with aBMD/BMC in our pediatric sample, before any interactions were considered. As such, the initial interpretation was that only a small subset of bone-related loci uncovered in adult GWA studies operated in childhood. However, these two loci and several other bone density loci first identified in adults were subsequently found to have statistically significant interactions with sex and/or pubertal stage to influence pediatric bone mass. Our interaction analyses revealed associations, or stronger associations, that would have otherwise gone undetected. Indeed, the C7orf58 locus was specifically associated with higher spine and total hip aBMD in males and with lower distal radius aBMD and TBLH BMC in females, whereas the LIN7C locus was most strongly associated with lower total hip aBMD during late puberty. Other notable loci that had significant sex and/or pubertal stage interactions included genes involved in Wnt signaling (eg, LRP5 and WNT16), RANK-OPG-RANKL signaling, and phosphate regulation (GALNT3); and genes with unknown function with respect to bone development (eg, C12orf23). This is the first longitudinal study to elucidate the role of adult bone density loci on aBMD/BMC in childhood while also incorporating sex and pubertal stage interactions. The findings support our hypothesis that genetic determinants of bone accretion are not constant during the development of the growing skeleton and vary by sex.

Our study has several key strengths. The bone phenotypes we modeled were derived from a highly standardized protocol that included centralized, expert analysis of DXA scans. To accurately account for the age and sex-related patterns of bone acquisition and the known effects of height status, we used aBMD/BMC Z-scores adjusted for height-for-age Z-scores as outcomes. Pediatric endocrinologists or nurse specialists performed pubertal evaluations. Our longitudinal models adjusted for growth, pubertal timing, physical activity, and diet to assure that any associations were not likely to be related to these known influences on pediatric bone acquisition. We performed a series of sensitivity analyses to assess the confidence of our findings. Importantly, our findings remain consistent when we: ⁽¹⁾ used restricted expected maximum likelihood (REML); ⁽²⁾ restricted our sample to the initial cohort with more extensive follow-up; and ⁽³⁾ removed age as a covariate as there was concern of overfitting. However, our study also has weaknesses. For some of our statistical significant interactions, no SNP associations were observed for specific sex and/or pubertal stage categories. We may have been able to observe associations with a larger sample size; alternatively, these loci may not play a major role in skeletal biology in childhood. We used DXA to estimate aBMD/BMC, and, in the future, replication of our findings using volumetric BMD and cortical and trabecular bone density outcomes will be important.

A cross-sectional study recently associated the 7q31.31 region, which includes C7orf58 and WNT16, with TBLH BMD in children aged 6 and 10 years.⁽³⁰⁾ We repeatedly measured aBMD/BMC bone phenotypes across all maturational stages and observed a sex interaction with C7orf58, documenting for the first time that this gene may function differently in males and females in influencing pediatric bone accrual. The function of C7orf58 is not known. In contrast, its neighboring gene, WNT16, encodes a ligand for the Wnt signaling pathway known to have a key role in bone homeostasis.⁽³¹⁾ We observed a sex interaction with WNT16 at the distal radius and also observed sex and/or pubertal stage interactions with other genes involved in the Wnt signaling pathway. In particular, LRP5 (a co-receptor in Wnt signaling) interacted with sex and was strongly associated with higher TBLH BMC and total hip aBMD in males in our study. LRP5 was one of the earliest adult bone density loci identified in the general population,^(11,32) and mutations in LRP5 have been identified in two rare forms of pediatric osteoporosis (juvenile primary osteoporosis and osteoporosis-pseudoglioma syndrome).^(33,34) However, this locus has not been extensively associated with pediatric bone mass in the general population.⁽³⁵⁾ The lack of LRP5 associations with pediatric bone mass in the general population may be masked in analyses that combine males and females. RANK-OPG-RANKL signaling is another well-established pathway involved in bone homeostasis,⁽³⁶⁾ and variants in this pathway have been associated with cortical volumetric BMD at the tibia in 15-year-old children.^(37–39) Although direct comparisons cannot be made with our aBMD/BMC phenotypes, it is interesting to note that in one of these studies a sex-RANKL interaction was tested and there was evidence that the association was stronger in males.⁽³⁷⁾ We observed sex and/or pubertal stage interactions with RANKL/AKAP11 and TNFRSF11B (OPG). The RANKL/AKAP11 locus was associated with lower TBLH BMC in the males and with lower distal radius aBMD in the males during late maturation, whereas the TNFRSF11B locus was most strongly associated with lower distal radius aBMD, spine aBMD, and TBLH BMC during early maturation.

No established mechanisms explain why any gene would affect pediatric bone mass or density differently by sex and pubertal stage. However, one possible explanation for the interactions involving Wnt signaling loci is related to the sensitivity of bone to mechanical loading. Osteocytes embedded within the bone matrix function as mechanosensors and Wnt signaling is important for osteocyte mechanosensation.⁽³¹⁾ Compared to wild-type mice, Lrp5 knockout mice experience lower gains in bone mass, whereas mice with the Lrp5 G171V high bone mass mutation experience greater gains in bone mass in response to mechanical loading.⁽⁴⁰⁾ In childhood, males have greater lean muscle mass and tend to gain more lean mass during maturation;⁽⁴¹⁾ males are also more likely to engage in higher-intensity physical activities and experience less decline in physical activity during maturation.^(42,43) Such differences translate to a male skeleton exposed to greater mechanical loading and may explain why we observed associations between LRP5, and possibly other Wnt signaling pathway genes, and higher bone mass in males, but not in females. We envision that biomedical scientists and clinicians will collaborate to lead translational research to further investigate sex and developmental specific strategies that could help to optimize skeletal health across the lifespan. The National Institutes of Health is developing policies to help balance the study of male and female animals to allow for the investigation of sex differences;⁽⁴⁴⁾ such policies will help establish mechanisms underlying the sex and maturation differences we observed.

In conclusion, our findings suggest that multiple variants originally associated with adult BMD do indeed influence bone mass in childhood, especially when sex and pubertal stage are taken into account. These sex- and maturation-specific effects were most notable for loci involved in Wnt signaling and RANK-OPG-RANKL signaling and for C7orf58 and C12orf23, which have unknown functions. These findings underscore the need to understand the mechanisms by which genetic determinants of bone accretion are regulated across puberty in males and females.