Body Composition During Childhood and Adolescence: Relations to Bone Strength and Microstructure

Abstract

Context:

Numerous studies have examined the association of body composition with bone development in children and adolescents, but none have used micro-finite element (μFE) analysis of high-resolution peripheral quantitative computed tomography images to assess bone strength.

Objective:

This study sought to examine the relations of appendicular lean mass (ALM) and total body fat mass (TBFM) to bone strength (failure load) at the distal radius and tibia.

Design, Participants, and Setting:

This was a cross-sectional study of 198 healthy 8- to <15-year-old boys (n = 109) and girls (n = 89) performed in a Clinical Research Unit.

Results:

After adjusting for bone age, height, fracture history, ALM, and TBFM, multiple linear regression analyses in boys and girls, separately, showed robust positive associations between ALM and failure loads at both the distal radius (boys: β = 0.92, P < .001; girls: β = 0.66, P = .001) and tibia (boys: β = 0.96, P < .001; girls: β = 0.66, P < .001). By contrast, in both boys and girls the relationship between TBFM and failure load at the distal radius was virtually nonexistent (boys: β = −0.07; P = .284; girls: β = −0.03; P = .729). At the distal tibia, positive, albeit weak, associations were observed between TBFM and failure load in both boys (β = 0.09, P = .075) and girls (β = 0.17, P = .033).

Conclusions:

Our data highlight the importance of lean mass for optimizing bone strength during growth, and suggest that fat mass may have differential relations to bone strength at weight-bearing vs non-weight-bearing sites in children and adolescents. These observations suggest that the strength of the distal radius does not commensurately increase with excess gains in adiposity during growth, which may result in a mismatch between bone strength and the load experienced by the distal forearm during a fall. These findings may explain, in part, why obese children are over-represented among distal forearm fracture cases.

The pubertal years are recognized as the most opportune period to modify bone mineral density (BMD) and bone structure—traits that tend to track throughout life (1, 2). Consequently, suboptimal peak bone strength established early in life could have long-term negative ramifications for skeletal health and fracture risk. Moreover, associations between childhood obesity and distal forearm fractures (3, 4) implicate behavioral traits that could extend into adulthood. Thus, there is considerable interest in better understanding how modifiable factors, such as lean and fat mass, determine bone strength and microstructure during growth.

Mounting evidence connects lean mass to the achievement and maintenance of BMD and bone structure (5–9). Indeed, the highly integrated nature of the musculoskeletal system has been demonstrated in both children (5–7) and adults (8, 9). However, the relationship between fat and bone, particularly during childhood and adolescence, is more controversial (10). Mechanistically, these tissues are linked by multiple pathways (11, 12), and fat may modify the skeleton to accommodate increases in body weight with obesity. However, obese children and adolescents have proportionately greater muscle mass relative to height (13, 14); thus, any relation between fat and bone may be driven, in part, by skeletal modeling adaptations in response to muscle forces (15) and osteogenic factors that mediate crosstalk between muscle and bone cells (16).

Most studies of fat-bone relationships in children and adolescents have assessed bone mass and areal BMD using dual-energy x-ray absorptiometry (DXA), which can provide misleading values during growth (17). Recognizing this limitation, more recent studies have used standard peripheral quantitative computed tomography (CT) (pQCT) to assess weight-bearing volumetric BMD (vBMD) and bone macrostructure (18–25). Comparatively less attention has been given to fat-bone relations at non-weight-bearing skeletal sites, with findings frequently contradictory (18, 20, 24). With a resolution of approximately 400 μm, however, standard pQCT measurements lack the ability to accurately assess bone microstructure or evaluate bone strength (26).

Fortunately, better skeletal assessments are now feasible using high-resolution pQCT (HRpQCT). With a resolution of 82 μm, HRpQCT provides a noninvasive bone biopsy of the distal radius and tibia, thereby permitting accurate quantification of macro- and microstructure of cortical and trabecular bone compartments, separately. In addition, HRpQCT images can be used to generate voxel-based microfinite element (μFE) models highly predictive of biomechanically relevant bone strength (ie, failure load) at weight-bearing vs non-weight-bearing sites (27). To refine our knowledge of the relations between body composition and bone health, we conducted a cross-sectional study of healthy boys and girls (age range, 8 to < 15 years), using μFE analysis and HRpQCT at the distal radius and tibia to better characterize skeletal development by examining the relations of lean mass and fat mass to bone strength and microstructure.

Materials and Methods

Study subjects

After approval by the Mayo Clinic Institutional Review Board, study subjects were recruited from Olmsted County, Minnesota, as previously described (28). All guardians and subjects older than 12 years of age provided written informed consent; informed written assent was obtained from subjects younger than or equal to 12 years of age. Briefly, between October 2009 and April 2012, we recruited a total of 223 healthy boys and girls between the ages of 8 and 15 years (28). Because our questions focused primarily on growing children and adolescents, we excluded older (bone age ≥ 15 years [n = 25]) subjects from the current analysis. This resulted in 198 boys and girls of whom 105 subjects (58 boys and 47 girls) had sustained a distal forearm fracture (due to mild or moderate trauma) in the past year and 93 subjects (51 boys and 42 girls) had no fracture history. Because there are relatively few providers in Olmsted County, medical care is essentially self-contained within the community. Thus, the vast majority of all orthopedic care is provided by the Mayo Clinic, which has two large hospitals (Saint Marys and Rochester Methodist) that have maintained a common electronic medical record linkage system (29) that contains details of almost all of the medical care (both inpatient and outpatient) provided to Olmsted County residents. All subjects with a recent (<1 year) distal forearm fracture were recruited using this unique medical record linkage system as described previously (28). Fracture trauma levels were assigned using Landin's modified criteria (28) based on review of the clinical details in the original medical record and an interview with the guardian and child. We also recruited boys and girls from Olmsted Country with no fracture history and a similar age distribution (Figure 1). Exclusion criteria included medical conditions or medications known to affect skeletal structure or function. Reflecting the ethnic composition of Olmsted County, 97% of the sample was white.

Figure 1.

Proportion of A, boys; and B, girls in each bone age maturity category (I-III) stratified by group (nonfracture controls, all DFF patients). DFF = distal forearm fracture.

Study protocol

As described previously (28), height was measured to the nearest 0.1 cm using a customized stadiometer (Mayo Section of Engineering), and weight was obtained to the nearest 0.1 kg using an electronic scale (Model 5002, Tronic, White Plains, New York). Skeletal maturity (bone age) was determined from plain hand and wrist x-rays of the left wrist using the “gold standard” Tanner-Whitehouse III method (30). The primary study end-point was bone strength determined by μFE analysis; secondary end-points included HRpQCT-derived cortical and trabecular bone macro- and microstructural parameters at the distal radius and tibia. Owing to motion artifact, nine radius (five boys; four girls) and three tibia (one boy; two girls) scans were excluded. Body composition was assessed from whole-body DXA scans obtained on all subjects. All procedures were performed in the outpatient Clinical Research Unit at the Mayo Clinic (Rochester, Minnesota).

HRpQCT measurements

Details regarding the HRpQCT device and in vivo image processing and analysis protocols used in this cohort have been described previously (28). Briefly, the Xtreme-CT (Scanco Medical AG) was used to obtain high-resolution images of the distal radius and tibia. In subjects with a distal forearm fracture, the nonfractured distal radius (and in control subjects, a randomly assigned radius) was scanned. In all subjects, the nondominant distal tibia was scanned. For each scan, a reference line was placed at the proximal limit of the epiphyseal growth plate using a scout view image of the distal forearm/ankle. Subsequently, an automated program was used to obtain a 3-dimensional stack of 110 high-resolution (voxel size = 82 μm) CT slices (equivalent to 9.02 mm) starting 2 mm proximal to the reference line. Radiation exposure to subjects from HRpQCT is minimal, with a local absorbed dose of 0.065 Gy and total radiation exposure of < 0.01 mSv.

Cortical area (mm²), cortical thickness (mm), endocortical circumference (mm), periosteal circumference (mm), cortical vBMD (mg/cm³), cortical porosity (%), trabecular area (mm²), trabecular bone volume fraction (BV/TV), trabecular number (1/mm), trabecular thickness (mm), and trabecular separation (mm) at the distal radius and tibia were derived from HRpQCT images as previously described (28). Short-term precision of the HRpQCT device in our laboratory has been reported previously (31).

μFE analysis

To evaluate bone strength (ie, failure load [Newtons]) at the distal radius and tibia, μFE models were created directly from the HRpQCT images using the manufacturer's software (μFE element analysis solver v.1.15, Scanco Medical AG) as described previously (28). Failure load was derived by scaling the resulting load from a test simulating 1% compression, such that 2% of all elements had an effective strain > 7000 microstrain (27).

DXA measurements

Whole-body DXA scans were performed using a Lunar densitometer (Lunar Prodigy System; GE Healthcare, Madison, Wisconsin). Appendicular lean mass (ALM, kg) was derived as the lean mass of the arms and legs (kg); total body fat mass (TBFM, kg) was derived as the fat mass of the whole body.

Statistical analyses

Differences in means between boys and girls were assessed using two-sample t tests or Wilcoxon rank-sum tests as appropriate. χ² tests were used to compare frequency differences between groups. Multivariable linear regression analysis was used to test the relationships of ALM and TBFM with the primary study end-point (bone strength) in boys and girls, separately, after adjusting for skeletal maturity (bone age), height, and fracture history. Similar models were then fit for the other study end points. Model residuals were used to check normality assumptions, nonconstant variance, and influential points. Multicollinearity was assessed using the variance inflation factor. Standardized regression coefficients were used to summarize the results. All testing was performed at a significance level of P < .05 (two tailed). Analyses were performed using SAS 9.3 and R 3.0.2.

Results

Descriptive characteristics are summarized in Table 1 for all subjects combined and, separately, according to sex. In this sample of healthy boys and girls, chronological age, height, weight, and body mass index (BMI) did not differ between the sexes. By contrast, the boys had significantly higher ALM and lower TBFM compared with the girls. Based on US National Center for Health Statistics/Centers for Disease Control and Prevention percentiles for BMI (BMI, kg/m²) in children and adolescents (32), 9.6% of the sample (12 boys; 7 girls) was obese (BMI > 95th percentile).

Table 1.

Descriptive Characteristics of all Study Subjects, Both Combined and Stratified by Sex

Characteristic	All	Boys	Girls	P
N	198	109	89
Mild/Moderate trauma DFF, n/n	54/51	29/29	25/22	.947
Bone age, y	11.6 ± 1.7	11.8 ± 1.6	11.4 ± 1.8	.045
Chronological age, y	11.5 ± 1.7	11.6 ± 1.7	11.4 ± 1.7	.514
Height, cm	151 ± 12	151 ± 13	150 ± 11	.371
Weight, kg	44.5 ± 12.2	45.6 ± 13.5	43.3 ± 10.4	.191
BMI, kg/m2	19.3 ± 3.5	19.5 ± 3.7	19.1 ± 3.2	.376
ALM, kg	13.7 ± 3.7	14.5 ± 4.1	12.8 ± 3.0	.002
TBFM, kga	11.4 ± 7.0	10.8 ± 7.9	12.0 ± 5.8	.237

Values are presented as mean ± sd unless otherwise noted.

P-values are for sex differences using the two-sample t test or χ² test as appropriate.

Statistics in bold are significant at P < .05.

^aRank-sum test P = 0.008 [boys, median = 9.0 kg; girls, median = 11.0 kg].

Table 2 shows the full regression models for boys and girls, separately, with failure load as the dependent variable and bone age, height, fracture history, ALM, and TBFM as independent variables. The results showed robust positive associations between ALM and bone strength (ie, failure load) at both the radius and tibia. As evident in the table, associations between ALM and failure loads were strong for both the radius (boys: β = 0.92, P < .001; girls: β = 0.66, P < .001) and tibia (boys: β = 0.96, P < .001; girls: β = 0.66, P < .001). By contrast, in both boys and girls, the relationship between TBFM and failure load at the radius was virtually nonexistent (boys: β = −0.07; P = .284; girls: β = −0.03; P = .729). At the tibia, positive, albeit weak, associations were observed between TBFM and failure load in both boys (β = 0.09, P = .075) and girls (β = 0.17, P = .033).

Table 2.

Multiple Linear Regression Models for Failure Loads at the Distal Radius and Tibia in Boys and Girls, Separately

Independent Variable	Boys				Girls
	Radius Adj. R2 = 0.64		Tibia Adj. R2 = 0.78		Radius Adj. R2 = 0.53		Tibia Adj. R2 = 0.63
	β	P	β	P	β	P	β	P
Bone Age	0.20	.090	0.01	.920	0.28	.058	0.06	.657
Height	−0.29	.077	−0.15	.246	−0.20	.248	0.01	.974
Mild Trauma DFF	−0.06	.341	−0.16	.002	−0.19	.018	−0.08	.251
Moderate Trauma DFF	0.11	.078	−0.07	.146	−0.01	.885	−0.02	.740
ALM	0.92	< .001	0.96	< .001	0.66	< .001	0.66	< .001
TBFM	−0.07	.284	0.09	.075	−0.03	.729	0.17	.033

Standardized β coefficients and P values are presented.

Adjusted R² values are presented for each full regression model.

In boys, regression analyses with bone age, height, fracture history, ALM, and TBFM as independent variables (Table 3) showed strong positive associations (β = 0.50–0.77, all P < .05) between ALM and greater radial cortical bone dimensions (ie, cortical area, cortical thickness, and endocortical/periosteal circumferences) in particular. Further, whereas as a positive association was observed between ALM in boys and radial cortical porosity (β = 0.86, P < .001), ALM in boys was negatively associated with radial cortical vBMD (β = −0.39, P = .121). Similar associations were observed between ALM in boys and cortical bone parameters at the tibia (β = 0.23–1.20, all P < .05), except ALM in boys was positively associated with tibial cortical vBMD (β = 0.77, P = .002). Further, ALM in boys was associated with better trabecular bone microstructure at both the radius and tibia (ie, greater trabecular area [β = 0.68 and 0.31, respectively, both P < .01], BV/TV [β = 0.80 and 0.88, respectively, both P < .01] and number [β = 0.82 and 0.57, respectively, both P < .01]; lower trabecular separation [β = −0.70 and −0.55, both P < .01]). However, associations between ALM in boys and trabecular thickness at both the radius and tibia were comparatively lower (β = 0.43 and 0.41, respectively, both P = .102).

Table 3.

Independent Associations of Appendicular Lean Mass and Total Body Fat Mass with Cortical and Trabecular Bone Parameters at the Distal Radius and Tibia in Boys

Dependent Variable	ALM, kg		TBFM, kg		Adjusted R2
	β	P	β	P
Radius
Cortical area, mm2	0.65	.010	0.05	.641	0.16
Cortical thickness, mm	0.50	.042	0.00	.994	0.18
Endocortical circumference, mm	0.62	< .001	−0.05	.432	0.71
Periosteal circumference, mm	0.77	< .001	−0.06	.249	0.74
Cortical vBMD, mg/cm3	−0.39	.121	0.25	.015	0.16
Cortical porosity, %	0.86	< .001	−0.20	.026	0.31
Trabecular area, mm2	0.68	< .001	−0.06	.324	0.70
Trabecular BV/TV	0.80	.001	0.08	.408	0.27
Trabecular number, 1/mm	0.82	< .001	0.31	< .001	0.42
Trabecular thickness, mm	0.43	.102	−0.24	.026	0.08
Trabecular separation, mm	−0.70	.002	−0.22	.014	0.32
Tibia
Cortical area, mm2	1.20	< .001	−0.05	.619	0.29
Cortical thickness, mm	1.00	< .001	−0.12	.183	0.29
Endocortical circumference, mm	0.23	.043	0.16	.001	0.80
Periosteal circumference, mm	0.30	.006	0.16	.001	0.82
Cortical vBMD, mg/cm3	0.77	.002	0.03	.769	0.11
Cortical porosity, %	0.69	.003	−0.01	.890	0.25
Trabecular area, mm2	0.31	.008	0.16	.001	0.81
Trabecular BV/TV	0.88	< .001	0.10	.251	0.36
Trabecular number, 1/mm	0.57	.002	0.34	< .001	0.50
Trabecular thickness, mm	0.41	.102	−0.26	.014	0.06
Trabecular separation, mm	−0.55	.006	−0.28	.001	0.43

Standardized β coefficients and P values are presented for ALM and TBFM.

Adjusted R² values are presented for each full regression model.

Model covariates: bone age, height, mild trauma DFF, moderate trauma DFF, ALM, and TBFM.

In the same regression models, associations between TBFM in boys and bone parameters at both the radius and tibia were relatively weak (Table 3), although some did reach statistical significance. For example, TBFM in boys was positively associated with radial cortical vBMD (β = 0.25, P = .015) and trabecular number (β = 0.31, P < .001), and negatively associated with cortical porosity (β = −0.20, P = .026), trabecular thickness (β = −0.24, P = .026) and trabecular separation (β = −0.22, P = .014). At the tibia, TBFM in boys was positively associated with endocortical/periosteal circumferences (both β = 0.16, P = .001), trabecular area (β = 0.16, P = .001) and trabecular number (β = 0.34, P < .001), and negatively associated with trabecular thickness (β = −0.26, P = .014) and trabecular separation (β = −0.28, P = .001). All other associations between TBFM and cortical and trabecular bone parameters at the radius and tibia were not significant (Table 3).

Regression analyses in girls using the same set of independent variables (bone age, height, fracture history, ALM, and TBFM) resulted in similar associations between ALM and most cortical and trabecular bone parameters (Table 4), although the magnitude of these associations tended to be lesser than those observed in boys. Nonetheless, in particular, ALM in girls was clearly associated with greater radial cortical area (β = 0.66, P = .003) and endocortical/periosteal circumferences (β = 0.47 and 0.55, respectively, both P < .01). Similarly, ALM in girls was strongly associated with larger endocortical and periosteal circumferences (β = 0.56 and 0.59, respectively, both P < .001) at the tibia. In addition, ALM in girls was associated with better trabecular bone microstructure at both the radius and tibia (ie, greater trabecular area, BV/TV and number [β = 0.28–0.73]; lower trabecular separation [β = −0.43 and −0.65]), although not all of these associations were statistically significant. Furthermore, ALM in girls was not associated with trabecular thickness at either the radius (β = −0.03, P = .907) or tibia (β = −0.05, P = .833).

Table 4.

Independent Associations of Appendicular Lean Mass and Total Body Fat Mass with Cortical and Trabecular Bone Parameters at the Distal Radius and Tibia in Girls

Dependent Variable	ALM (kg)		TBFM (kg)		Adjusted R2
	β	P	β	P
Radius
Cortical area, mm2	0.66	.003	0.01	.896	0.37
Cortical thickness, mm	0.44	.080	0.01	.950	0.16
Endocortical circumference, mm	0.47	.003	−0.10	.198	0.67
Periosteal circumference, mm	0.55	< .001	−0.12	.082	0.75
Cortical vBMD, mg/cm3	0.47	.064	0.10	.424	0.11
Cortical porosity, %	0.29	.286	−0.10	.474	0.00
Trabecular area, mm2	0.52	.001	−0.11	.152	0.66
Trabecular BV/TV	0.28	.283	0.15	.249	0.03
Trabecular number, 1/mm	0.52	.044	0.28	.030	0.11
Trabecular thickness, mm	−0.03	.907	−0.02	.885	0.01
Trabecular separation, mm	−0.43	.096	−0.27	.038	0.08
Tibia
Cortical area, mm2	0.26	.207	0.10	.331	0.37
Cortical thickness, mm	0.15	.529	−0.01	.937	0.13
Endocortical circumference, mm	0.56	< .001	0.00	.954	0.67
Periosteal circumference, mm	0.59	< .001	0.01	.849	0.71
Cortical vBMD, mg/cm3	0.12	.621	0.03	.795	0.07
Cortical porosity, %	0.37	.153	0.02	.859	0.05
Trabecular area, mm2	0.61	< .001	−0.01	.945	0.67
Trabecular BV/TV	0.59	.013	0.27	.022	0.20
Trabecular number, 1/mm	0.73	.003	0.25	.036	0.19
Trabecular thickness, mm	−0.05	.833	0.09	.486	0.07
Trabecular separation, mm	−0.65	.007	−0.29	.015	0.19

Standardized β coefficients and P values are presented for ALM and TBFM.

Adjusted R² values are presented for each full regression model.

Model covariates: bone age, height, mild trauma DFF, moderate trauma DFF, ALM, and TBFM.

In the same regression models, TBFM in girls was not significantly (all P > .05) associated with any cortical bone parameters at either the radius or tibia (Table 4). However, significant associations were observed between TBFM in girls and both trabecular number (β = 0.28, P = .030) and trabecular separation (β = −0.27, P = .038) at the radius. Similarly, TBFM in girls was positively associated with trabecular BV/TV (β = 0.27, P = .022) and trabecular number (β = 0.25, P = .036), and negatively associated with trabecular separation (β = −0.29, P = .015) at the tibia. Finally, as observed for ALM, TBFM in girls was not associated with trabecular thickness at either the radius (β = −0.02, P = .885) or tibia (β = 0.09, P = .436).

Discussion

Our study expands on previous work by applying recent innovations in skeletal imaging and image analysis to show, for the first time, associations of lean mass and fat mass with μFE-derived bone strength and HRpQCT-derived bone microstructure at the distal radius and tibia in children and adolescents. Our data clearly suggest that lean mass is an important determinant of bone strength and microstructure during growth. Further, our results suggest that fat mass may have differential associations with bone strength at non-weight-bearing (ie, radius) vs weight-bearing (ie, tibia) sites. Indeed, in analyses adjusted for bone age, height, fracture history, and ALM in boys and girls, separately, the relationships between fat mass and radial bone strength were virtually nonexistent, whereas positive, albeit weak, associations were observed between fat mass and tibial bone strength.

It has long been recognized that adaptation of bone morphology is dominated by mechanical loads placed on the skeleton (33). Examples of such loads include forces generated from skeletal muscle contractions (15). Although lean mass is only a surrogate for mechanical loading, we found strong positive associations between ALM (lean mass of the limbs) and bone strength in both boys and girls. This was due to skeletal adaptations that included periosteal expansion, a thicker cortex, and more optimal trabecular bone microstructure. Notably, our group has shown similar associations between ALM and HRpQCT-derived bone parameters in adult men and women (8). However, despite the close relationships between lean mass and bone parameters across the life span, the precise mechanisms responsible for coupling and synchronization of muscle-bone interactions still need to be defined more clearly (16). In addition, we found a positive relationship between lean mass and cortical porosity, which is perhaps not surprising given that cortical porosity increases during peak linear growth (34)—a time in childhood also characterized by rapid gains in muscle mass.

Contrary to the relatively strong relations between lean mass and bone parameters, our data show weaker fat-bone associations during growth. A strong point of our study was the assessments of bone strength and microstructure at both non-weight-bearing and weight-bearing sites using μFE analysis and HRpQCT, which allowed us to identify that the relations between fat mass and bone parameters in children and adolescents seem to be site-specific. Further, because the relationship between fat mass and bone may also be driven, in part, by association with greater lean mass (13, 14), our findings highlight the importance of considering lean mass as a confounder when examining fat–bone relationships in youth. Indeed, the associations between fat mass and bone parameters are dependent upon whether lean mass is included as a covariate in the analysis. This observation may help explain why Vandewalle et al (25) found that obese boys have larger bone size (assessed by standard pQCT) at the radius compared with nonobese boys in analyses that did not adjust for lean mass, whereas in analyses adjusted for lean mass, Wey et al (24) found reduced radial bone size (assessed by standard pQCT) in children with higher vs lower fat mass. Unfortunately, differences in imaging technology and skeletal sites measured limit our ability to directly compare our results with previous studies.

Although the mechanisms connecting fat and bone remain incompletely understood, available evidence suggests that adipose tissue may influence bone remodeling in several ways. For example, excess adiposity can result in the increased production and secretion of factors (eg, cytokines, estrogens, adiponectin, resistin) that could act locally or systemically to mediate positive and/or negative effects on bone (11, 12). Further, obesity results in increased production of adipokines (eg, leptin) that feedback to the central nervous system and potentially alter sympathetic outputs to bone tissue (35). Further, recent genetic evidence in Caucasian populations from North America and Europe strongly suggests that higher obesity leads to lower circulating 25(OH)D levels (36). Because vitamin D deficiency is a risk factor for a number of adverse health outcomes, including fractures, 25(OH)D levels should be closely monitored and treated in obese children and adolescents. In addition, excess adiposity could have a mechanical loading effect on the skeleton at weight-bearing sites (11), which may explain, in part, the differential associations between fat mass and bone parameters at the radius vs the tibia. Clearly, further delineation of specific pathways that mediate fat-bone crosstalk is needed to better understand the variable effects of fat on skeletal integrity at different sites, and ultimately, the clinical implications of obesity on future fracture risk. Although obesity is an established risk factor for childhood forearm fractures (3, 4), it also seems to be protective against fracture in adulthood (10). Thus, it cannot be assumed that relationships between fat and bone are unchanged over the life span.

There are several limitations to our study. First, because our findings are cross-sectional, it is difficult to infer causality between body composition and skeletal parameters. Nonetheless, prospective studies in children provide at least some evidence that these relationships track longitudinally during development (24, 37). A second concern is that our assessments of body composition were limited to whole-body DXA scans. We recognize that the relationships between body composition and bone may vary depending on the lean mass region (eg, ALM vs total body lean mass) and fat depot (eg, sc vs visceral vs marrow) assessed. Nevertheless, a significant strength our study was the adjustment for potential confounders during growth including skeletal maturity and height, important considerations in studies of children and adolescents in whom a wide range of body sizes exist in the same cohort. A third concern is that our study did not include functional measures of muscle strength. Although muscle mass, size, and strength are highly correlated across the life span (38–40), the relations between muscle functional parameters and bone strength during growth warrant additional study. A fourth potential concern is that our sample included a limited number of nonwhite subjects, which may affect the generalizability of these findings to other racial and ethnic groups.

Another important issue in studies of children and adolescents is the need to control for potential confounders (ie, heterogeneity in skeletal maturity, height, and fracture history); thus, we adjusted all analyses for these covariates. Further, based on the biological rationale described earlier, we additionally adjusted for TBFM and ALM. We realize that inappropriate treatment of highly intercorrelated variables as being independent can cause statistical problems (11). Recognizing this concern, we were careful to perform formal diagnostic statistics to test for the influence of multicollinearity. Although there was some indication of multicollinearity issues when including both height and ALM in the model at the same time, the coefficient and SE estimates for all of the variables were very similar if height was removed from the models. Therefore, we do not believe that multicollinearity is a major issue.

Our study also had a number of significant strengths including the use of HRpQCT and μFE analysis, as well as the inclusion of both boys and girls. Interestingly, our analyses stratified by sex revealed that the associations between lean mass, but not fat mass, and bone parameters were stronger in boys vs girls. Conceivably, this is the result of the significantly greater lean mass observed in boys compared with girls.

In conclusion, our findings highlight the importance of lean mass for optimizing bone strength during growth, and suggest that the highly integrated nature of the musculoskeletal system is established early in life. Furthermore, fat mass seems to have differential associations with bone strength at weight-bearing vs non-weight-bearing skeletal sites in both boys and girls. At the distal tibia, fat mass has a positive, albeit weak, association with bone strength, whereas there is a virtually nonexistent association between fat mass and bone strength at the distal radius after adjustment for bone age, height, fracture history, and lean mass. These observations suggest that the strength of the distal radius does not commensurately increase with excess gains in adiposity during growth, which may result in a mismatch between bone strength and the load experienced by the distal forearm during a fall. These findings may explain, in part, why obese children are over-represented among distal forearm fracture cases (3, 4).