Heritability of BMD of Femoral Neck and Lumbar Spine: A Multivariate Twin Study of Finnish Men


  • The authors state that they have no conflicts of interest.


Of the 80% variation in BMD among male twins that is caused by genetics, part was explained by genetic influences on lifting strength and lean body mass/height. Lifting strength was significant in both the femoral and spine BMD and body weight only for lumbar BMD.

Introduction: The dominant role of heritability in BMD has been shown in twin studies among women. However, the mechanisms of genetic influences are poorly understood. BMD is associated with lean body mass and muscle strength, which both have a genetic component, but the relative effects of muscle strength and lean body mass/height on the total genetic and environmental variations influencing BMD of men are unclear.

Materials and Methods: Measurements of BMD from a DXA scanner on a representative sample of 147 monozygotic and 153 dizygotic male twin pairs (age, 35–70 yr) were related to a variety of anthropometric and behavioral covariates and interview data. Data were analyzed with univariate modeling of genetic characteristics, bivariate modeling of covariates that were significant in univariate models, and multivariate modeling of the simultaneous effects of significant covariates from the bivariate models.

Results: Heritability influences were estimated to account for 75% of the variance in femoral BMD and 83% in lumbar BMD. Univariate and bivariate modeling showed that, of the factors studied, only lifting force and lean body mass/height had statistically significant influences. Of the total genetic variation in femoral BMD, lifting force explained 9%, and lean body mass/height 18%; the proportions for lumbar BMD were 9% and 11%, respectively. Of the total environmental variation, the correlation with isokinetic lifting force explained 9% for femoral BMD and 10% for lumbar BMD. The genetic correlations between lifting force and femoral and lumbar BMD were ∼0.3, as were the environmental correlations of isokinetic lifting force and femoral and lumbar BMD and of lean body mass/height and femoral BMD. The environmental correlation of lean body mass/height and femoral BMD was not significant.

Conclusions: Lifting force had effects on both femoral and lumbar BMD. Body weight was important, but only for lumbar BMD. Muscle strength may have the best potential for modification among behavioral factors to increase both femoral and lumbar BMD.


The major role of heredity in BMD was reported in a classic twin study in 1987,(1) which has since been confirmed in several studies.(2–12) However, most studies are based on sample sizes smaller than 100 twin pairs, and mostly in women; the two classic twin studies among men are based on <50 pairs.(13,14) Estimates of the relative roles of total heredity vary: intergenerational studies estimate that genetic factors account for 50–70% of the variance in BMD, whereas twin study estimates reach 80–90%.(13,14) Body weight, and in particular lean mass, has been the most significant single anthropometric or environmental parameter affecting BMD.(4,7,15–19) However, heritability estimates for lean body mass have been 56% and 60% for leg strength; adjusting for these muscle-related variables decreased the total additive genetic variance of BMD at the femoral neck by close to 20% and at the lumbar spine by 7%.(2) It can be expected that differences in adjustment for other traits may explain some of the differences in the heritability estimates.

Although the importance of specific identified genes associated with BMD varies, their known influence accounts for only a small fraction of the heritability variation in BMD; for example, the population-attributable risk for the combined vitamin D receptor gene risk genotypes is only 4%.(20,21) However, genes can influence BMD through several different mechanisms, so that many of the behavioral and constitutional determinants that affect BMD—such as body and muscle mass, physical activity, and some dietary habits that aggregate in families—may be genetically influenced.(2,22–26) However, attempts to assess the role of environmental exposures and behavioral factors in studies of the genetics of BMD are rare and challenging, due, in part, to the inherent inaccuracy of assessing most exposures over the lifespan, which is typically the time period of possible influence and interest.

Our aim was to estimate the magnitude of the influences of total heredity, and of anthropometric, lifting performance measures, and several behavioral factors in BMD among a representative sample of adult Finnish men. The specific goal was to estimate the relative effects of muscle strength and lean body weight on the total genetic and environmental influences on femoral neck and lumbar spine BMD.



The male twin pairs for this study were selected from the population-based Finnish Twin Cohort (with 13,888 pairs of known zygosity) based on relevant prior information available from surveys in 1975 and 1981, which had elicited response rates of 89% and 84%, respectively.(27) Selection of 116 monozygotic (MZ) and 116 dizygotic (DZ) pairs was based on significant discordance in exercise, occupational sitting, lifting (71 pairs together), or driving (45 pairs) and 31 MZ and 37 DZ randomly selected pairs (mean age, 49 yr; range, 35–70 yr). Zygosity was confirmed with DNA analyses. The twins were studied at a single clinical center, where interviews and measurements were carried out.

Subjects were excluded from analysis if they suffered from conditions or medications known to affect bone metabolism; these included any kidney or liver disorders (n = 4 men) or a history of the following conditions or medications in the prior year: thyroid or parathyroid disorders (n = 1); anti-epileptic medication (n = 5); cortisone or steroid use, arthritis, or hypertension therapy (n = 23); any history of bed rest of >1 mo (n = 34); history of cancer (n = 8). After taking into account exclusions and missing data, 407 subjects (in 98 MZ and 107 DZ pairs) were left for the analysis of femoral BMD and 409 (in 98 MZ and 108 DZ pairs) for the analysis of lumbar BMD.

To assess the representativeness of our sample, we compared the MZ pairs that participated in the study with all MZ male pairs from the 1981 Finnish Twin Cohort survey, and we compared our sample of DZ pairs to our sample of MZ pairs. No significant differences were observed between our MZ sample and all male MZ pairs in the entire cohort for the following characteristic: level of leisure-time physical activity, level of education, outdoor versus indoor work, shift work, work monotony, or life satisfaction. Also, no significant differences were observed for a history of work-incapacitating neck, shoulder, or back pain, or of sciatica, nor in means of smoking history, occupational category, or social class. However, our subjects were more likely to be working and to have slightly more physically demanding jobs than the subjects in the Finnish Twin Cohort, which is probably because of their selection partly on these characteristics.(28) Given the representativeness of the selected MZ pairs is known and the DZ pairs were chosen using the same criteria as the MZ pairs, the selected DZ pairs are likely representative of the larger population as well.

Study protocols were reviewed and approved by the Ethical Committee of the Department of Public Health at the University of Helsinki and the University of Alberta. All subjects received written information about the study procedures and provided informed consent before participating.

Measurements of BMD

BMD was measured with the same DXA absorptiometry instrument (Lunar DPX, Madison, WI, USA; with the same X-ray tube) at the femoral neck and L1–L4 lumbar vertebrae. The CVs at the department where the subjects were examined for measurements were 0.9% for the spine and 1.5% for the femoral neck. National, ethnic mean values, and SDs for the BMD of the femoral neck in 20- to 29-yr-old men are 1.06 and 0.14 g/cm2; for lumbar spine BMD, they are 1.23 and 0.15 g/cm2.(29) The WHO criteria for white white women were applied to that national database to create limits for normal BMD, osteopenia (BMD between 1 and 2.5 SD below the mean peak sex-matched BMD), and osteoporosis (BMD < mean −2.5 SD).(30) Using these definitions, 2% of subjects had osteoporotic femoral necks (BMD < 0.72 g/cm2) and 31% of subjects had osteopenia (0.72–0.92 g/cm2). For the lumbar spine, 2% of subjects were osteoporotic (<0.87 g/cm2) and 30% had osteopenia (0.87–1.08 g/cm2). Nineteen percent had osteoporosis or osteopenia in both the spine and the hip and 25% at only one site.

Estimates of lifetime exposures to suspected determinants based on interview

Exposure data were obtained from an extensive, structured interview that reviewed the subject's work history and common leisure activities (Table 1). A series of health-related questions identified subjects with health conditions that had led to medication or inactivity. Because, on average, 74% of Finnish dietary calcium comes from dairy products,(31) calcium intake was calculated from present dairy product consumption as reported on structured questionnaires, including information about possible changes in consumption. None of the subjects was taking calcium supplements or vitamin D at the time of interview. Cigarette smoking was calculated in pack-years; at baseline, 34% were current smokers. Alcohol and coffee use were summarized as grams per month and cups per day, respectively, based on most recent consumption. The reliability and coverage issues related to alcohol and coffee consumption and smoking have been shown to be acceptable.(32,33) However, smoking and alcohol consumption were not associated with BMD of femoral neck and lumbar spine in an earlier study based on MZ twin pairs,(34) and, therefore, were not included in this analysis.

Table Table 1.. BMD, Age, and Various Anthropometric and Behavioral Factors in Monozygotic (MZ) and Dizygotic (DZ) Twins
original image

Each subject was asked to review each job held and its associated tasks from the time he began working. The product of the most common weight lifted and the frequency formed a summary measure of lifting per day. Eventually, each job was placed in one of four categories based primarily on materials handling activities and associated positional loading (i.e., twisted and bent postures): 1 = sedentary work, 2–3 = progressive degrees of materials handling and positional loading, and 4 = very heavy physical loading. A limited evaluation of interview reliability was conducted of exposures at work. The intraclass correlation coefficients (ICCs) between initial phone interviews and those repeated 1 yr later were 0.74 for sitting time and 0.60 for mean total lifting per day. Subjects were also questioned about regularly performed exercise and other leisure-time physical activities during adolescence and adulthood. Test–retest reliability, using a 5-yr test–retest interval, of the summary variable of mean hours of exercise per week yielded acceptable reliability (ICC, 0.73).(35)

Isokinetic lifting

Isokinetic lifting tests were performed from a forward-bent, knees-straight position, and the subjects were asked to lift as rapidly and forcefully as possible and lifting force (Newtons) and work (Joules) were recorded.(36)

Body fat measurement

Bioelectric impedance was used to obtain percentage of body fat following the manufacturer's instructions (Spectrum II; RJL Systmes, Detroit, MI, USA). The CVs of repeat measurements were from 0.2% to 4.1%.(37) Lean body mass was obtained by subtracting fat from body weight.

Statistical analyses

To examine relationships of covariates with BMD, regression models using the twins as individuals were age-adjusted using transformed variables, but reported results are on untransformed variable scales. In the regression model, the standard assumption of independent observations does not hold with sampling based on twins clustered within twin pairs. Therefore, to obtain valid p values, the model accounted for clustering of twin pairs, and SE were adjusted appropriately.(38) Also, to test that the DZ and MZ pairs come from the same population, equality of means, and variances (Table 1) by zygosity were tested, also adjusted for clustering of twins. STATA 9.1 statistical software was used.(39)

In quantitative genetic modeling, we aim to partition sources of variance in BMD into genetic and environmental effects. Using data on twins reared together, it is possible to model four separate parameters: an additive genetic (A) variance component, effects caused by genetic dominance (D), and shared (C) and unique (E) environmental variance components. One can develop models based on the different combinations of these parameters (e.g., AE, ACE, ADE, and E), but effects caused by dominance and shared environmental effects cannot be simultaneously modeled with data on twins alone.(40) We used the principle of parsimony to support accepting a simple model (e.g., AE) until evidence in support of a more complex model (e.g., ACE) requires us to abandon it. The goodness-of-fit statistics (χ2 values) were used to assess the relative fit of the models; they were calculated from the difference between estimated −2 × log-likelihood values of a full model (i.e., ACE) and a corresponding hierarchically nested model (such as AE). This was done to compare models where different components of variance have been specified. Degrees of freedom of χ2-test were calculated from the differences of degrees of these models, and p values were computed to evaluate the relative fit. All estimation was done in Mx.(40)

Univariate modeling

The first step of univariate genetic modeling was to estimate univariate saturated models (i.e., models of phenotype means, twin covariances, and age and calcium intake regression) to adjust the variance. Also, MZ and DZ twin correlations were calculated by standardizing twin covariances. Genetic and environmental components of variance were estimated by standard univariate twin modeling and by following the principles described above.

Bivariate modeling

Bivariate genetic factor models were estimated to quantify the relationship between BMD covariates in terms of reported genetic effects (A) and unique environmental effects (E) of total variation (i.e., standardized variance component estimates), genetic correlations (ra) between the genetic effects on BMD and the genetic effect in the covariate, and the corresponding environmental correlations (re) with 95% CI. The proportion of total genetic/environmental variation in BMD explained by genetic/environmental variation in common with a covariate was also reported. All models, except the bivariate model for lumbar BMD and heavy leisure-time activities, were fitted to raw data using maximum likelihood estimation and including all observations of phenotypes (i.e., full information maximum likelihood). Models were adjusted for age and calcium by regressions on phenotype means. The bivariate model for categorized lumbar BMD (seven categories) and binary heavy leisure-time activities (yes/no) was fitted using raw ordinal maximum likelihood estimation, also adjusted for age and calcium. Finally, we extended the analysis to multivariate modeling to consider the simultaneous effects of most relevant covariates that were significant in bivariate models on BMD. The same principles were followed as in bivariate modeling.


No significant differences between the MZ and DZ twins were observed in the means and variances for BMD of lumbar spine and femoral neck (Table 1).

Associations of BMD with covariates in individuals

Age was associated with BMD of the femoral neck, but not of the L1–L4 lumbar vertebrae. In age-adjusted analyses, all basic anthropometric and isokinetic lifting parameters, except body fat percent, were associated with both BMD locations (Table 2). Calcium intake was associated with only femoral BMD. Among physical activity histories, years of endurance, and ball sports were associated with both BMD locations and heavy leisure-time activities with lumbar spine BMD. An increase of 100 N in isokinetic lifting force was associated with 0.017 and 0.023 g/cm2 increases in femoral and lumbar BMD, respectively; similarly, a 10-kg increase in lean body mass was associated with 0.06 g/cm2 increases in both BMD measures.

Table Table 2.. Univariate Regressions of Age, Behavioral Factors, Physical Activity Factors, and Anthropometric and Lifting Performance Measures on BMD of Femoral Neck (N = 407) and Lumbar Spine (L1–L4) (N = 409)
original image

Results from univariate quantitative genetic modeling

In lumbar and femoral BMD, the age- and calcium intake–adjusted pairwise ICCs for the MZ twins (rMZ = 0.81 and 0.76, respectively) were about twice as high as the correlations for the DZ twins (rDZ = 0.30 and 0.36, respectively), suggesting a genetic influence on BMD. The best-fitting univariate model for both femoral neck and lumbar BMD consisted of two latent factors: additive genetic and unique environmental factors (AE model; Table 3). The model including also genetic dominance factor (ADE) provided a good fit for the lumbar BMD, but it was not significantly better than the AE model. For femoral neck BMD, in the most parsimonious model, additive genetic effects accounted for 75% (95% CI = 66–82%) of the variance, with the remainder being caused by the unique environment. For the lumbar BMD, genetic effects accounted for 83% (95% CI = 76–88%) of the variance. Models omitting genetic effects fitted significantly worse (Table 3).

Table Table 3.. Comparison of the Goodness-of-Fit
original image

Results from bivariate models

Correlations between femoral neck and lumbar BMD in one twin with the covariate in the other twin were moderate: between BMD and height 0.09–0.11 and body mass measures 0.28–0.42, between BMD and isokinetic lifting performance 0.22–0.31 and between lumbar BMD and heavy leisure-time activities 0.25; for MZ pairs, they were higher or equal to values in DZ pairs, suggesting genetic effects in common between BMD and the covariate. Corresponding correlations for other covariates were lower.

To quantify the strength of these genetic and environmental covariations between the BMD and other studied constitutional factors we ran a series of bivariate genetic models. When tested against a full bivariate model, the shared environmental parameters (C) could be dropped from the models without a reduction in fit, and additive genetic environmental (AE) models could be used in all analyses. In these models, genetic correlations were statistically significant for femoral neck BMD and height, lean body mass/height, BMI, and isokinetic lifting force, as were correlations for unique (unshared between twins) environmental factors for isokinetic lifting capacities. Of total genetic variation, 18% in femoral neck BMD was explained by common genetic variation with lean body mass/height; similarly, body mass index explained 17%, height explained 2%, isokinetic lifting force explained 9%, and isokinetic lifting work explained 7%. The values for lumbar BMD were a few percentages lower, but the corresponding genetic correlations were statistically significant, except with height. Of the total environmental variation in femoral neck BMD, 9% was explained by unique environmental correlation with isokinetic lifting force; in lumbar BMD, most (21%) was explained by unique environmental correlation with body mass index.

Results from multivariate models

In genetic factor modeling of femoral neck BMD, lumbar BMD, lean body mass/height, and isokinetic lifting force, one genetic factor in common and three trait-specific genetic factors were found (Fig. 1). All genetic correlations were statistically significant, being highest between femoral and lumbar BMD phenotypes (0.72) and between isokinetic lifting force and lean body mass/height (0.61). The genetic correlations between either femoral or lumbar BMD and isokinetic lifting force or lean body mass/height were between 0.30 and 0.41. Environmental correlations between BMD phenotypes and isokinetic lifting force were significant (0.29–0.46), but lean body mass/height correlated only with lumbar BMD (0.37).

Figure Figure 1.

Most parsimonious multivariate AE-model* of femoral and lumbar BMD, isokinetic lifting force, and lean body mass/height. Path coefficients are presented in the figure. *χ2(4) = 6.958, p = 0.138, AIC = −1.042.


Genetic effects accounted for ∼80% of the variance in men's BMD of the femoral neck and lumbar spine; the rest was explained by environmental factors. Most of the suspected determinants could be left out without reducing the explanatory power of the models, leaving only lean body mass/height and isokinetic lifting force in the multivariate models. Of the total variation in femoral and lumbar BMD explained by environmental factors, lifting force explained one tenth in femoral and lumbar BMD and body mass explained one fifth in lumbar BMD, but practically nothing in femoral BMD. The genetic correlation between lifting force and lean body mass/height was high, but the environmental correlation was not statistically significant. Overall, these results indicate that lifting force accounted for an important portion of interindividual variation in BMD, in particular for the femoral neck.

Our bivariate results indicate how the phenotypic association of a determinant with BMD can be separated into both genetic and environmental components. Of the total genetic variation of femoral BMD (75% of all variation), the proportions explained by lifting force (9%) and lean body mass/height (18%) were close to earlier findings of an 18.6% decrease in total genetic variation of BMD after adjustment for lean body mass, grip, and leg extensor strength.(2) Our isokinetic lifting force explained 9% and lean body mass/height explained 11% of the genetic variation of lumbar BMD (83% of all variation) compared with 6.8% in an earlier study after adjustment for the muscle variables.(2) In addition to the fact that the measures are correlated, another explanation for why adjusting for lean body mass diluted the effects of muscle strength on BMD may be that strength measures are less accurate and stable than loading parameters because of body anthropometric factors.(2,7)

Environmental factors explain relatively small proportions of the variation in BMD (25% in femoral neck and 17% in lumbar spine), with modest contributions from lifting force parameters and lean body mass/height, particularly in the femur. This shops the complex genetic and environmental determinants of body composition and lifting force, which contribute to BMD. Those who are strong must have a history of heavy loading peaks affecting both the lumbar spine and hip joint. The finding that body mass had a smaller effect at the femur than at vertebrae may be explained by the fact that body mass loads the spine also during sitting when the femoral neck does not carry the body mass. In addition, body mass is typically responsible for higher and more sustained mechanical forces than those imposed by occasional objects lifted; however, the additional loading peaks caused by muscles seem to have clinical importance. Unfortunately, how to get a population to adopt sustained physical activity habits and associated good muscle strength is a significant challenge, perhaps partly because of important familial and genetic influences on exercise lifestyle and physical activity behavior.(24,41,42)

Although our twin study sample size is among the largest using DXA measurements, ideally it should be larger to have more statistical power to separate genetic and specific shared environmental effects. However, in adult twins, the role of shared (i.e., those common to family members) influences is likely to be minimal.(6,43) Another consideration is that lean body mass was based on the percent of body fat measured by bioelectric impedance, which is less commonly used than methods based on DXA; however, it has been shown to be a repeatable and valid measure.(19,37)

Although adjusting for height controls for the effects of skeletal size on bone's contribution to lean body mass to some degree, there is the possibility that other variations in bone size and BMD may still influence this measure. This, in turn, may have some effect, inflating the apparent influence of lean body mass on BMD; and this must be acknowledged as a limitation. On the other hand, controlling for lean body mass/height may overadjust for the effects of lifting strength on BMD because of the correlation of muscle mass to strength. Thus, the relation observed between isokinetic lifting strength and BMD may be an underestimate. In the total contribution (9%) of lifting strength on BMD the relative role genetics was around four times larger than that of environmental factors. One of the study strengths was that the subjects were a representative sample of Finnish men and that data on most commonly suspected determinants were available.(24) This allowed multivariable analyses to examine genetic and environmental pathways through which specific determinants, such as anthropometrics and strength, affect BMD. We are not aware of other multivariate classic twin studies of femoral and lumbar BMD.

The total heritability of femoral neck and lumbar spine BMD in men was high and within the range of earlier twin studies, mainly of women. A portion of the total genetic variation of BMD was explained by lifting force and lean body mass/height, suggesting that not all genetic influences on BMD act directly on the trait itself. These results indicate that body weight was important only for lumbar BMD, whereas lifting force accounted for an important portion of both femoral and lumbar BMD and may have more practical importance for modification than other lifestyle factors, in particular for femoral neck. In other words, although body mass and lifting force are associated, they have different loading exposure times, mechanisms, and effects on the bone. Enhancing optimal muscle strength and endurance properties may have best preventive potential among behavioral factors.


This research was supported by National Institutes of Health Grant AR 40857; The Work Environment Fund, Finland; The Academy of Finland, Grants 38332 and 42044; The Alberta Heritage Foundation for Medical Research, Canada; the European Union funded EURODISC-project (QLK6-CT-2002-02582); and the Canada Research Chairs Program. The Finnish Twin Cohort study is part of the Academy of Finland's Centre of Excellence for Complex Disease Genetics.