Associations Between Maternal Peak Bone Mass and Bone Mass in Prepubertal Male and Female Children


  • Presented in part at the 21st Annual Meeting of the American Society for Bone and Mineral Research, St Louis, Missouri, U.S.A., September 30-October 4, 1999.


The aim of this study was to estimate heritability of bone density in premenopausal women, prepubertal male, and prepubertal female child pairs. We studied 291 pairs (mothers, mean age, 33 years, range 22–45 years; children, mean age, 7.92 years, range 7.32-8.92 years). Bone density and body composition were assessed by dual-energy X-ray absorptiometry. Height and weight were measured in both mother and child. Body size-adjusted heritability estimates for areal bone density (g/cm2) were all statistically significant (femoral neck, 59%; lumbar spine, 38%; total body, 41%) and were consistently and significantly higher in mother-daughter pairs (n = 105) as compared with mother-son pairs (n = 186). Heritability estimates for bone mineral apparent density (BMAD; g/cm3) were marginally lower but remained statistically significant at all sites (femoral neck, 51%; lumbar spine, 32%; total body, 38%). Maternal osteopenia was associated with significant reductions in bone mass at all sites in the children (femoral neck, 0.75 SD and p < 0.0001; lumbar spine, 0.61 SD and p < 0.0001; total body, 0.43 SD and p = 0.012). Mother-child bone areal bone density correlation coefficients and prediction of low bone mass in the child were greater (but this did not reach statistical significance) if the corresponding anatomical site in the mother was used for prediction with the exception of the total body. These data confirm that heritability of bone mass extends to prepubertal children and is gender- and possibly site-specific as well as under separate genetic control to growth. Furthermore, the strength of the mother-child association is such that bone density screening of mothers would make it possible to identify most prepubertal children at higher risk of osteoporosis in later life.


FRACTURES caused by osteoporosis are a major public health problem.(1) Bone mineral density (BMD) is a major predictor of these fractures(2) and is, in turn, determined by both genetic and environmental factors. Family and twin studies have consistently indicated heritability estimates for bone mass in the order of 60–80%.(3–15) Concerns have been raised that the classic twin model may overestimate heritability because of gene-environment interaction,(3) and this is supported by family studies generally showing lower estimates than twin studies.(3–15) In addition, heritability estimates are higher in premenopausal daughters as compared with postmenopausal daughters(5,6) and appear to peak between the ages of 13 and 26 years(7,15) and decrease with increasing age,(8) suggesting that the major genetic effect is on the attainment of peak bone mass rather than bone loss in later life. Although this appears to be the case with bone density, it may be different for bone turnover.(16) The prepubertal period is a period in which environmental influences appear to have a much larger effect than in adulthood.(17–20) It would appear desirable to target interventions in this age group to those at highest genetic risk if they can be identified either directly or indirectly (by measurement of bone density in a parent). However, studies looking at associations between parents and children under the age of 15 years are sparse(21–23) and it is not clear if the association is mediated by bone mineralization in the prepubertal period or during the pubertal period or a combination of the two. Furthermore, studies have rarely considered the effect of body size on heritability estimates and the small sample sizes in most studies have made it difficult to examine whether heritability is either gender- or site-specific or a generalized phenomenon. Therefore, the aims of this study were to estimate heritability in premenopausal women, prepubertal male, and prepubertal female child pairs and to examine the effect of the maternal bone density site as well as the effect of adjustment for linear growth or bone size on these estimates.


In 1988, there were 6779 live births in Tasmania, Australia. Of these, 1380 were identified as being at high risk of Sudden Infant Death Syndrome by previously published criteria(24) and were invited to take part in a longitudinal study. In southern Tasmania (latitude 42°S), there were 735 births that met these criteria. Of these, 696 (95%) agreed to an in-hospital interview and 581 (80%) agreed to the 1-month follow-up.

The 696 subjects who agreed to the in-hospital interview were approached during 1996 to take part in a bone mass study. After 8 years, we were able, using school lists, to identify definitely 551 of these subjects (or 80%, which is in close agreement to the Australian Bureau of Statistics data on annual outward migration rates from Tasmania of 2.5%). Children and their mothers who provided informed consent to take part underwent an extensive protocol.(17,24) Ethical approval for this study was obtained from the University of Tasmania Ethics Committee (human experimentation).

Bone mass (at the right femoral neck, lumbar spine, and total body) and body composition in both mother and child were assessed using the technique of dual-energy X-ray absorptiometry on array setting (Hologic QDR2000 densitometer; Hologic, Inc., Waltham, MA, U.S.A.). Bone mass is reported as areal BMD and bone mineral apparent density (BMAD).(25) Lean mass is bone free unless otherwise stated. The longitudinal CV for our machine during 1996 using daily measurements of a spine phantom was 0.54%.

Height was measured using a stadiometer with the subject in bare feet. Weight was measured using bathroom scales that were calibrated daily using known weights. Although not directly relevant to this study, environmental factors of importance in children such as winter sunlight exposure, breast-feeding in early life, maternal smoking during pregnancy, dietary calcium intake, and sports participation were assessed by questionnaire as previously outlined.(17,24)


The t-tests (where appropriate) were used for comparison of means. Logistic regression was utilized for estimating the odds of low bone mass in the child. We used T scores from the BMD printout for the mother. In the absence of a reference distribution for 8-year-old children, we calculated Z scores in the child by subtracting the mean from the actual BMD value and dividing by the SD at that site. This, in effect, created a BMD distribution with a mean of 0 and an SD of 1 similar to the T distribution in the mothers. One cut point is reported (T < −1.0). This was chosen to reflect the World Health Organization (WHO) criteria for osteopenia because few women would be expected to meet the criteria for osteoporosis in this age group.(26) The logistic regression models only included BMD in the child as the dependent variable and BMD in the mother as the independent variable. The above statistics were carried out on SPSS version 8.0 for Windows (SPSS, Inc., Chicago, IL, U.S.A).

Mother-child resemblance for a variable trait was assessed for each sex separately by the Pearson's correlation coefficient. The test of hypothesis (whether the correlation coefficient was different from 0) was done by the Fisher's Z-transformation statistic.(27) Hypothesis testing for differences between product-moment correlation coefficients (r) was based on the unpaired t-test. However, because the sample correlation coefficient does not have a normal distribution, the test was calculated on Z scores, transformed according to the Fisher's Z-transformation, in which the variance of Z does not depend on the actual value of Z. By the established quantitative genetic theory,(28) the total phenotypic variance of a trait (Vp) was partitioned into two components, namely, genetic (Vg) and environmental (Ve) effects. The index of heritability, the proportion of the phenotypic variance explained by genetic factors, was estimated as H2 = Vg/Vp. Estimates of Vg and Ve were performed based on the phenotypic variance-covariance matrix and the kinship coefficient between mothers and children (in this case 0.5), by means of the Mx program(29) and were adjusted for height and weight. A test for difference in heritability estimates between mother-son and mother-daughter pairs was done by comparing two competing statistical models, namely, homogeneity and heterogeneity models. In the former, it is assumed that the same gene and the same features of family environment influence the trait in both males and females; thus, their parameters were constrained to be equal between mother-son and mother-daughter pairs; in the latter, it is assumed that there are different aspects of the parameters on female and male phenotypes. The Aikaike Information Criterion (AIC) was used in the comparison and in selecting a model. The AIC is a compromise measure of the goodness-of-fit in relation to the degrees of freedom and is calculated as c2 minus twice the number of degrees of freedom. A minimum value of AIC is preferable because it indicates the most parsimonious model, that is, the one with fewest parameters that still fits the data adequately. To test the hypothesis that a single set of genes is involved in the determination of bone density in different skeletal sites (i.e., pleiotropy), the contribution of genetic and environmental factors to the covariances of these traits was estimated in a bivariate genetic model as previously described.(30) In this model, it is possible to estimate the genetic (or environmental) correlation, which measures the extent to which two phenotypes share genetic (or environmental) effects. The bivariate genetic analysis was performed by means of the Mx system.(29) A value of p < 0.05 (two-tailed) or a 95% CI not including the null point was regarded as statistically significant.


Analyzable bone density data were available on 330 children (mean age, 7.92 years; range, 7.32-8.92 years) and 278 mothers (mean age, 33 years; range, 22–45 years). There were 13 twin pairs with maternal data resulting in 291 mother-child pairs in total (186 mother-son and 105 mother-daughter). The response rate was 53% of those available in 1996 (60% for the children and 90% of the available mothers) or 42% of those in the original cohort. The only exclusion factor was refusal to participate and pregnancy in the mother. Anthropometric details are presented on these subjects in Table 1.

Table Table 1.. Characteristics of Study Sample
original image

Estimates of heritability of BMD varied between the different skeletal sites and gender (Table 2). Overall, the proportion of variance caused by additive genetic factors for height- and weight-adjusted BMD was 41, 59, and 38% for total body, femoral neck, and lumbar spine, respectively. Further adjustment for smoking, breast-feeding, sports participation, and winter sunlight exposure in the children led to little change in these estimates (43, 61, and 42% for total body, femoral neck, and lumbar spine, respectively). However, the heritability was consistently higher for mother-daughter pairs than for mother-son pairs. At the femoral neck, the adjusted estimates were 51% for mother-son pairs and 79% for mother-daughter pairs (p = 0.031). The corresponding estimates for lumbar spine BMD were 24% and 57% (p = 0.028) and for total body BMD were 35% and 54% (p = 0.26). After adjusting for bone size by using apparent BMD (g/cm3), the indices of heritability became marginally lower but remained statistically significant at all sites with similar gender differences. Genetic variances in body composition measures also were statistically significant. According to the model-fitting analysis, 40% and 60% of the variance of fat mass and lean mass, respectively, were ascribed to genetic factors. There was no appreciable variation between sexes with respect to these estimates (Table 2).

Table Table 2.. Estimates of Heritability for Mother-Son and Mother-Daughter Pairs Before and After Adjustment for Height and Weighta
original image

There was some evidence of site specificity with poorer correlation coefficients if sites other than the corresponding maternal anatomical site were used for prediction of that site with the exception of the total body in which both femoral neck and lumbar spine were better than the total body (Table 3). None of these differences reached statistical significance. Similarly, prediction of a Z score < −1.0 in the child was best predicted by maternal bone mass at that site as compared with other sites (Table 4). The models correctly classified 84–87% of children for a Z score < −1. Sensitivity and specificity were as follows: femoral neck (33% and 90%), lumbar spine (33% and 87%), and total body (21% and 88%). The absolute reduction in bone mass in the children of mothers with osteopenia was significant at all sites (Fig. 1) and varied from 0.43 to 0.75 SD (depending on site). This compares with the average 1.8 SD difference at all sites between women with and without osteopenia. The reduction in BMD was similar in mother-daughter pairs (femoral neck, 0.74 SD; lumbar spine, 0.72 SD; total body, 0.40 SD) and mother-son pairs (femoral neck, 0.72 SD; lumbar spine, 0.28 SD; total body, 0.37 SD) with the exception of the lumbar spine.

Table Table 3.. Correlation Between Maternal Bone Density Site and Child Bone Density Site: Evidence for Possible Site Specificity at the Lumbar Spine and Femoral Neck
original image
Table Table 4.. Prevalence and Odds of Low Bone Mass in Child (Defined as a Z-Score Less Than −1.0) if Mother Has Osteopenia (Defined as a T Score < −1.0)a
original image
Figure Figure 1.

Reduction of BMD in offspring of mothers with osteopenia at that site. There is a significant reduction in BMD at all sites. This compares with the average corresponding reduction in the mother of 1.8 SDs at all sites. Results are presented as mean ± SE.

Results of bivariate genetic analyses are shown in Table 5. There are two main clusters of correlations between BMD and measures of body composition. The genetic correlations among the three measures of BMD ranged between 0.76 (between lumbar spine and femoral neck) and >0.9 (total body vs. lumbar spine or femoral neck BMD). These genetic correlations were higher than the environmental correlations (which ranged from 0.19 to 0.48). The genetic correlation between fat mass and BMD (around 0.5) was lower than that between lean mass and BMD (from 0.6 to 0.7). However, the genetic correlation between lean and fat mass (0.46) was lower than the environmental correlation (0.69).

Table Table 5.. Genetic and Environmental Correlations Between Bone Density and Body Composition Measurements
original image

Results of the above analyses did not change to any notable extent if twins were omitted.


The results from this study confirm that the genetic associations for bone mass previously reported also apply to growing prepubertal children. In part, the associations were mediated by linear growth but remained significant after adjustment for body and/or bone size implying that bone mineralization is, to a substantial extent, under separate genetic control to growth. Furthermore, there was some evidence of site specificity in both heritability estimates and prediction of low bone mass in children whose mothers had low bone mass.

Indices of heritability estimated from this study were generally lower than those estimated from twin studies, but the difference may be caused by the statistical modeling of data. In the twin model, it is assumed that monozygotic and dizygotic twins share the same degree of environmental effects, despite the fact that their parents usually treat monozygotic twins more alike. If this effect is present then it will result in an overestimation of the genetic component and an underestimation of the environmental component in the twin statistical model.(3) Indeed, our estimates of heritability for mother-daughter pairs were at the lower limit of previous family and sibling studies, which have largely involved females.(6,9–12) Alternatively, the lower estimates in this study may be because of size adjustment although Table 2 would suggest that this is unlikely with the exception of the lumbar spine in mother-son pairs.

Although there have been a considerable number of studies describing bone density relationships in adults, there have been very few in growing children.(13,15,21–23) Such studies are important because they shed additional light on the timing of genetic effects and whether gene-environment interaction may be important in growing children. We found consistently and markedly higher heritability for bone density in mother-daughter pairs as compared with mother-son pairs. This is consistent with the results of McKay et al. who reported consistently higher correlation coefficients for height, weight, and unadjusted bone mass in mother-daughter pairs as compared with mother-son pairs in 12-year-old children at varying stages of maturity.(21) Of the other studies, one had only females,(15) one included fathers but not male children,(22) one had only mother-daughter pairs,(23) and the other had a very wide age range of children, which may bias estimates because of the wide variations in size as shown in this study in which a very high correlation of 0.86 between parents and offspring was reported.(13) Studies in adult parent-offspring pairs have been variable with some showing no gender variation,(7) no consistent trend,(12) or consistent differences with greater mother-daughter and father-son comparisons as compared with cross-gender comparisons.(14) This finding may suggest that there is gene-sex interaction in the determination of BMD in the population; that is, genes that influence BMD in males may not be the same as genes that influence BMD in females. In support of this, Orwoll et al.(31) recently reported that quantitative trait loci affecting bone mass in male rats were different from those that affect bone mass in female rats. Our results also may suggest that genomic imprinting is relevant in bone mineralization. Genomic imprinting refers to the epigenetic marking of a gene based on its parental origin resulting in monoallelic expression in the same sex offspring.(32) The mechanism of this is complex and not clear at present but does not follow Mendelian inheritance. It has not been implicated previously in osteoporosis but may be important given its well-recognized role in early growth and differentiation.(32) We did not include fathers in this study because of the high proportion of single mothers and greater difficulty in confirming paternity as compared with maternity. However, it would be desirable to include them and siblings in further studies to determine if the associations we have observed also apply to father-child pairs and gender concordant and discordant siblings. In addition, although it is desirable to measure bone density in the offspring at different stages of growth to compare directly genetic contributions to bone mass in the same population, a recent study found little difference in heritability over a 2-year period of observation,(23) suggesting that a much longer period of observation may be necessary to observe changes over time.

Site specificity rarely has been examined with regard to bone mass. A review concluded, based on the small studies available at the time, that there was no evidence to support this concept.(4) We report some evidence to support the concept of site specificity at the femoral neck and lumbar spine but not total body with regard to both heredity estimates and identification of low bone mass in the offspring of parents with low bone mass. Furthermore, the phenotypic correlation between lumbar spine and femoral neck BMD was 0.55 among children and 0.66 among mothers, suggesting that the acquisition of BMD during prepubertal and pubertal periods is not homogeneous and is dependent on skeletal site. Results of the present study also indicated that there was significant genetic determination of covariation of BMD at the lumbar spine and the femoral neck. The genetic correlation between lumbar spine and femoral neck BMD was 0.77, which is higher than 0.57 for the environmental correlation. This, together with previous findings,(33) suggests that part of the genetic variation in BMD at these two important sites is influenced by a set of common genes or that the underlying genes for the two traits are closely linked in the genome. Furthermore, the lower correlation (as compared with other BMD sites) also provides support for the hypothesis of an additional but different set of genes regulating bone mineralization at these two sites. Site specificity is of clinical importance because physical interventions may be more important for those at risk of low hip BMD whereas hormonal or pharmaceutical interventions may be more efficacious for those at risk of low spinal BMD. The genetic correlation between body composition and BMD is also noteworthy. Both lean mass and, to a lesser extent, fat mass correlated significantly with BMD. This may suggest that there may be common genes with pleiotropic effects on lean mass and BMD and fat mass and BMD or that the genes influencing these body compartments may be closely linked.

In this study, body size (especially height) was under strong genetic control (data not shown). The effect of this may be to magnify genetic associations with BMD by including both bone density and body size, which may be under separate control. However, heritability estimates remained significant at all sites after statistical adjustment for body size, suggesting different genetic mechanisms for body size and bone density. In addition, BMAD estimates were also significant but slightly lower than those for size-adjusted BMD, indicating that adjustment for body size substantially but incompletely adjusted for bone size.

Environmental factors, both dietary(18) and especially physical activity,(17,19,20) appear to have a larger effect on bone mass in the prepubertal age group as compared with other stages of life although it remains to be determined how much of this can be maintained into adult and later life. The strength of the association between maternal bone mass and child bone mass is sufficiently strong for women who are osteopenic (particularly given the very high specificity) to suggest that it is worthwhile targeting the offspring of these women with environmental modification aiming to offset the 0.43- to 0.75-SD reduction in bone mass caused by maternal genetic factors. Identification of these mothers is best done by bone density screening rather than clinical risk factors(34) and becomes more attractive when combined with the observation, in our cohort, that bone density feedback also can modify lifestyle behavior in the mothers.(34)

Last, this study has a number of potential strengths and weaknesses. The strengths include the large sample size and careful assessment of potential confounders of the mother-child associations. However, the participants in this study are not representative of Tasmanian women and children as a whole. As outlined in previous papers,(24,34) the selection process resulted in younger women who were more likely to smoke and less likely to breast-feed. This would suggest a lower bone mass in this population; however, the BMD distribution was in close agreement with the National Health and Nutrition Examination Survey (NHANES) population used for the Hologic reference database.(34) Furthermore, adjustment for these confounders did little to the mother-child associations we report, suggesting that these associations may be generalizable to other premenopausal mother-prepubertal child populations. The number of comparisons reported in this article suggests some caution is necessary in interpreting individual results and indicating the need to replicate these findings in other similar age populations.

In conclusion, these data confirm that heritability of bone mass extends to prepubertal children where it is both gender specific and under separate genetic control to linear growth. The apparent differences in site specificity merit further investigation in larger family studies. Furthermore, the strength of the mother-child association is such that bone density screening of mothers would make it possible to identify most prepubertal children at higher risk of osteoporosis in later life.


This work was supported by the National Health and Medical Research Council of Australia, Royal Hobart Hospital Acute Care Program, Tasmanian Dairy Authority, Lions Club of Australia, Coca-Cola Amatil, and Talays.