Heritability of prevalent vertebral fracture and volumetric bone mineral density and geometry at the lumbar spine in three generations of the framingham study



Genetic factors likely contribute to the risk for vertebral fractures; however, there are few studies on the genetic contributions to vertebral fracture (VFrx), vertebral volumetric bone mineral density (vBMD), and geometry. Also, the heritability (h2) for VFrx and its genetic correlation with phenotypes contributing to VFrx risk have not been established. This study aims to estimate the h2 of vertebral fracture, vBMD, and cross-sectional area (CSA) derived from quantitative computed tomography (QCT) scans and to estimate the extent to which they share common genetic association in adults of European ancestry from three generations of Framingham Heart Study (FHS) families. Members of the FHS families were assessed for VFrx by lateral radiographs or QCT lateral scout views at 13 vertebral levels (T4 to L4) using Genant's semiquantitative (SQ) scale (grades 0 to 3). Vertebral fracture was defined as having at least 25% reduction in height of any vertebra. We also analyzed QCT scans at the L3 level for integral (In.BMD) and trabecular (Tb.BMD) vBMD and CSA. Heritability estimates were calculated, and bivariate genetic correlation analysis was performed, adjusting for various covariates. For VFrx, we analyzed 4099 individuals (148 VFrx cases) including 2082 women and 2017 men from three generations. Estimates of crude and multivariable-adjusted h2 were 0.43 to 0.69 (p < 1.1 × 10−2). A total of 3333 individuals including 1737 men and 1596 women from two generations had VFrx status and QCT-derived vBMD and CSA information. Estimates of crude and multivariable-adjusted h2 for vBMD and CSA ranged from 0.27 to 0.51. In a bivariate analysis, there was a moderate genetic correlation between VFrx and multivariable-adjusted In.BMD (−0.22) and Tb.BMD (−0.29). Our study suggests vertebral fracture, vertebral vBMD, and CSA in adults of European ancestry are heritable, underscoring the importance of further work to identify the specific variants underlying genetic susceptibility to vertebral fracture, bone density, and geometry. © 2012 American Society for Bone and Mineral Research.


Vertebral fractures (VFrx) comprise almost half of the 1.5 million new osteoporotic fractures annually in the United States.1 Prevalent vertebral fracture increases the risk of future fracture at the spine and other skeletal sites2 and increases the risk of death in older adults.3 Studies have shown that both volumetric bone density (vBMD, g/cm3) and vertebral cross-sectional area (CSA) are important contributors to bone strength and VFrx risk,4 and low quantitative computed tomography (QCT)–assessed vBMD has been associated with VFrx.5

Whereas dual-energy X-ray absorptiometry (DXA)–measured areal BMD (aBMD, g/cm2) is widely used clinically to assess fracture risk and is considered the “gold standard” for the diagnosis of osteoporosis, the conventional 2D DXA does not provide volumetric bone density and can be influenced by artifacts such as aortic calcification and degenerative changes in the spine. In contrast, QCT provides 3D measures of vBMD that offers the opportunity to examine specific compartments such as trabecular bone and to better estimate cross-sectional geometry.6, 7

Although VFrx, vBMD, and CSA are affected by many environmental factors, genetic factors also have been implicated. However, to our knowledge, there have been no previously published estimate of heritability for VFrx. The objective of this study was to estimate the heritability of VFrx, vBMD, and CSA and to explore the extent to which VFrx and volumetric bone measurements share common genetic components.

Materials and Methods

Study samples

Participants in this study include members of the original, second-generation, and third-generation cohorts in the Framingham Heart Study (FHS). The analysis of prevalent vertebral fracture includes 684 members of the original cohort who obtained lateral spine radiographs in years 1992 to 19938 and 3448 members of the second- and third-generation cohorts who underwent CT examinations including lateral scout views in years 2002 to 2005.9 The vBMD and CSA measurements were derived from the same QCT images in members of the second- and third-generation cohorts. All participants signed informed consent before enrollment and clinical examinations, and the study was approved by the Institutional Review Boards at Boston University Medical Center and Hebrew Rehabilitation Center.

Semiquantitative (SQ) measurements for vertebral fracture

Prevalent vertebral fracture was assessed from T4 to L4 using the semiquantitative grading system.10 In the original cohort, lateral radiographs8 were used, and in the second- and third-generation cohorts, scout images9 were independently evaluated by trained radiologists. The SQ readings at each vertebral level were graded as 0 (no fracture), 1 (mild fracture: 20% to 25% reduction in any vertebral height), 2 (moderate fracture: 25% to 40% reduction), and 3 (severe fracture: >40% reduction). In this study, prevalent fracture was defined as SQ grade ≥2 at any level, although we also repeated the analyses considering grade 1 and higher vertebral fractures. We also repeated the vertebral fracture heritability analyses in the subset of individuals who had both vertebral fracture assessments and vBMD measures.

Volumetric bone density and vertebral cross-sectional area

QCT images of the lumbar spine obtained for second- and third-generation cohorts were analyzed using previously published software.11 The integral vBMD (In.BMD, g/cm3) was measured for the entire vertebral body including both cortical and trabecular compartments. Trabecular vBMD (Tb.BMD, g/cm3) was determined within an elliptical region inserted into the anterior vertebral body. CSA (cm2) was calculated from a central 10-mm-thick slice perpendicular to the axis of the vertebral body. Note that although lumbar levels L2 to L4 were targeted for analysis on the QCT image, in many instances the L2 level was not included because of technical issues. Also, we excluded images because of poor quality or severe scoliosis. Lumbar level 3 (L3) was present in the highest proportion of participants (around 98%) such that for analyses of vBMD and CSA we only present results for L3.

Statistical analysis

Descriptive statistics were generated using SAS version 9.1 (SAS, Inc., Cary, NC, USA). We used SOLAR statistical software to estimate heritability (h2).12 We calculated h2 estimates for VFrx, vBMD, and CSA, after adjustment for generation (model 1), for age, age2, and sex (model 2), and additionally for height (model 3). Age2 was included in the model based on the observation of a nonlinear relationship between prevalence of VFrx and age in our data. Using model 2 covariates, we also estimated h2 of VFrx additionally adjusting for either CSA, In.BMD, or Tb.BMD (models 4a to 4c). Bivariate analyses were also conducted in SOLAR to examine the genetic correlation (RhoG) among QCT-derived measurements of vBMDs, CSA, and VFrx, adjusting for generation, age, age2, and sex.


Sample characteristics

The characteristics of the sample are presented in Table 1. We analyzed data from 4099 individuals including 2082 women and 2017 men for VFrx analysis, whereas there were only 3333 individuals including 1737 men and 1596 women available for vBMD analysis. Among 148 VFrx cases with SQ grade ≥2, 48 were present in men and 100 in women, with ages ranging from 34 to 96 years. The original cohort participants (aged 72 ∼ 96 years) had the highest proportion of VFrx cases, 13.5% versus less than 3% in generation 2 (aged 40 ∼ 87 years) and less than 1% in generation 3 (aged 31 ∼ 72 years).

Table 1. Characteristics of Participants From Three Generations of the Framingham Heart Study
 Original cohortSecond generationThird generation
  • a

    Individuals with information on VFrx, vBMD, or CSA.

  • b

    Individuals with VFrx information.

Total sample size, na68414012047
Age, mean (SD), years78.5 (4.6)63.9 (9.0)45.0 (6.2)
Male, n (%)245 (35.82)661 (47.18)1128 (55.11)
Body mass index, mean (SD), kg/m226.88 (4.60)28.50 (5.37)27.51 (5.33)
Height, mean (SD), g/cm263.23 (3.85)65.88 (3.78)67.58 (3.69)
Vertebral fracture (SQ grade ≥1) ncase/ncontrolb203/481256/1136206/1817
Vertebral fracture (SQ grade ≥2) ncase/ncontrolb92/59241/135115/2008
L3 In.BMD, mean (SD), g/cm3N/A0.17 (0.04)0.20 (0.03)
L3 Tb.BMD, mean (SD), g/cm3N/A0.12 (0.04)0.16 (0.04)
L3 CSA, mean (SD), cm2N/A11.02 (1.86)11.33 (1.58)

Heritability estimates and bivariate correlation analysis

Both volumetric BMD measures (In.BMD, Tb.BMD) and CSA showed significant and strong h2 in crude analyses and in models adjusting for covariates including age, age2, sex, and with/without height. The h2 in covariate-adjusted models consistently remained around 0.41 (SE = 0.04) and 0.43 (SE = 0.04) for Tb.BMD and In.BMD, respectively. Depending on covariates included, h2 for CSA ranged from 0.27 to 0.43 (Table 2). The h2 for VFrx with grade ≥2 was 0.43 to 0.53, similar to the values observed for vBMD in all models. The h2 for VFrx increased (approximately 0.69) after additional adjustment for either vBMD or CSA measurement. Using only individuals who were included in the vBMD analyses, the heritability was similar (approximately 70%) as observed using the larger sample. When we repeated the analysis using only VFrx with grade ≥1, the heritability was much lower (19% to 27%) (intrareader agreement for grade ≥1, kappa = 0.56 ∼ 0.59, was not as good as agreement for grade ≥2, kappa = 0.68 ∼ 0.72, and the agreement varied for grade = 1 versus grade = 0 with kappa = 0.39 ∼ 0.67 depending on the reader). Bivariate analyses showed a negative genetic correlation between In.BMD and VFrx (RhoG = −0.22, p = 0.047) and a negative genetic correlation between Tb.BMD and VFrx (RhoG = 0.29, p = 0.015).

Table 2. Heritability of Vertebral Fracture and Vertebral Volumetric Bone Mineral Density and Cross-sectional Area
Model: covariates adjustedh2SEp Value
  • a

    Crude model: adjusting for generation information (original, second generation, third generation).

  • b

    A total of 33 individuals (excluded) had no VFrx information.

  • c

    A total of 115 individuals (excluded) in the second and third generations had no vBMD information.

  • d

    VFrx is defined as SQ grade ≥1.

  • e

    VFrx is defined as SQ grade ≥2.

Vertebral fractured (VFrx is defined as SQ grade ≥1) (n = 4099b)
 Model 1: crudea0.250.071.56 × 10−4
 Model 2: age + age2 + sex + generation0.200.081.96 × 10−3
 Model 3: age + age2 + sex + height + generation0.190.073.86 × 10−3
 Model 4a: age + age2 + sex + L3_CSA + generation0.270.097.12 × 10−4
 Model 4b: age + age2 + sex + L3_In.BMD + generation0.260.091.20 × 10−3
 Model 4c: age + age2 + sex + L3_Tb.BMD + generation0.270.087.98 × 10−4
Vertebral fracturee (VFrx is defined as SQ grade ≥2) (n = 4099b)
 Model 1: crudea0.490.174.32 × 10−3
 Model 2: age + age2 + sex + generation0.430.191.02 × 10−2
 Model 3: age + age2 + sex + height + generation0.530.193.06 × 10−3
 Model 4a: age + age2 + sex + L3_CSA + generation0.690.265.07 × 10−3
 Model 4b: age + age2 + sex + L3_In.BMD + generation0.680.301.10 × 10−2
 Model 4c: age + age2 + sex + L3_Tb.BMD + generation0.670.298.87 × 10−3
L3 In.BMD (n = 3312c)
 Model 1: crudea0.510.047.06 × 10−44
 Model 2: age + age2 + sex + generation0.430.043.92 × 10−33
 Model 3: age + age2 + sex + height + generation0.430.042.09 × 10−32
L3 Tb.BMD (n = 3333c)
 Model 1: crudea0.500.041.23 × 10−40
 Model 2: age + age2 + sex + generation0.410.041.59 × 10−29
 Model 3: age + age2 + sex + height + generation0.410.041.79 × 10−28
L3 CSA (n = 3312c)
 Model 1: crudea0.270.044.37 × 10−15
 Model 2: age + age2 + sex + generation0.430.041.88 × 10−36
 Model 3: age + age2 + sex + height + generation0.370.041.04 × 10−30
Bivariate analysis (model 3: age + age2 + sex + height + generation)e
 In.BMD and vFrxTb.BMD and vFrxCSA and vFrx
 Phenotypic correlation RhoP−0.26−0.19−0.07
 Genetic correlation RhoG (SE)−0.22 (0.16)−0.29 (0.16)−0.16 (0.05)
 p for test RhoG = 04.67 × 10−21.50 × 10−22.96 × 10−1
 p for test RhoG = 1 or −19.06 × 10−39.51 × 10−34.99 × 10−3
 Environmental correlation RhoE (SE)−0.29 (0.15)−0.11 (0.14)0.01 (0.05)
 p for test RhoE = 06.55 × 10−33.40 × 10−19.40 × 10−1


We found in this sample of European ancestry that genetic factors contribute significantly to VFrx, lumbar vertebral vBMD, and CSA. Risk factors for VFrx, including integral vBMD and trabecular vBMD of the L3 vertebra and L3 CSA, demonstrated moderate to high heritability (27% to 51%), whereas VFrx, defined as at least 25% height reduction in any vertebral level, was highly heritable (>43%). Spinal BMD was genetically correlated with VFrx, suggesting that there are common genetic variants affecting VFrx and its risk factors. Although we observed a higher heritability when examining only vertebral fractures of grades 2 and higher, the number of such vertebral fractures was relatively small (n = 148), therefore the results should be confirmed by family studies with larger numbers of VFrx.

Studying the heritability of VFrx, spinal vBMD, and CSA is important because no one has yet investigated the genetic variants associated with VFrx. These results provide a strong justification to pursue the genetic contributors to these morbid and mortal fractures. This study highlights three major findings using spine images from the large family-based cohort, the Framingham Study. First, VFrx are heritable with estimates of h2 from 43% to 53%, depending on the adjustment model. This heritability estimate is comparable to other types of osteoporotic fractures,13, 14 although lower than that for hip fractures occurring earlier in life (65%),15 which is similar to the VFrx heritability estimate (around 68%) with adjustment for vBMD or CSA. Second, QCT-derived measurements of volumetric bone density and geometry are moderately to highly heritable (Tb.BMD ∼41%, In.BMD ∼43%, and CSA ∼37%). Similarly, DXA-derived BMD and bone geometry are under strong genetic control with 50% to 70% heritability. Importantly, adjusting CSA for age, sex, and body size noticeably increased h2 of CSA, implying that genetic factors may operate independently of effects on body size. We also observed that the h2 of CSA remained the same (37.5%) if models included weight adjustment (the result is not shown).

Third, there were moderate genetic correlations between VFrx and multivariable-adjusted L3 In.BMD (RhoG = −0.22, p = 0.047) and Tb.BMD (RhoG = −0.29, p = 0.015). Although these genetic correlations are higher than both phenotypic and environmental correlations (RhoE), they are modest in magnitude. Genetic contributions to a risk factor, vBMD and CSA, may differ from the genetic contribution to the ultimate disease phenotype, VFrx in this study. This is especially true for a “proxy phenotype,” such as DXA-derived aBMD, and a complex event, such as VFrx. Similarly, Deng and colleagues16 estimated that less than 1% of additive genetic variance is shared between aBMD and fractures at the hip, which strengthens the argument that there are other (nonskeletal) factors that contribute to the risk of fractures. Our estimates based on single vertebral level L3 are higher than those previously reported (with genetic correlation 0.26 for In.BMD and 0.07 for CSA), but the common genetic contribution to VFrx shared with the vBMD or CSA is still modest. Neither vBMD nor CSA alone can individually serve as a perfect surrogate of the skeleton's ability to withstand the forces that produce fractures. Similarly, they are not perfect proxies for genetic studies of VFrx either.

There are several caveats worth noting. We conducted this analysis using up to 4099 available samples of European ancestry in the Framingham Study. Results may not be generalizable to other ethnic groups. Moreover, the prevalence of osteoporotic fractures varies across different ethnicities. Research has shown the highest fracture rate in women of European ancestry, whereas the mortality rate resulting from fracture is higher in African Americans.17 Studying prevalent VFrx is another limitation, and future studies should examine incident VFrx. The bivariate analyses were conducted using SOLAR, which treats the phenotype as quantitative in the correlation. This may lead to inflated type I error for our VFrx measurement, which was dichotomous.

In conclusion, our study suggests that prevalent vertebral fracture, vertebral vBMD, and CSA in white adults are heritable and some of the heritability is shared between the traits, underscoring the importance of further work to identify the specific variants underlying genetic susceptibility to vertebral fracture and its risk factors.


All authors state that they have no conflicts of interest.


This work was supported by NIH R01AR053986, R01AR/AG041398, T32 AG023480, K01 AR053118, and by the National Heart, Lung, and Blood Institute (NHLBI), Framingham Heart Study (NIH/NHLBI Contract N01-HC-25195).

Authors' roles: Study design and data collection: DK, MLB, LAC, EJS, and DPK. Data analysis: CTL and YZ. Data interpretation: CTL, DK, MLB, LAC, and DPK. Drafting manuscript: CTL, DK, and DPK. Revising manuscript content: CTL, DK, YZ, YHH, HKG, KEB, TFL, EJS, SD, MLB, LAC, and DPK. Approving final version of manuscript: CTL, DK, YZ, YHH, HKG, KEB, TFL, EJS, SD, MLB, LAC, and DPK. CTL, DK, and DPK take responsibility for the integrity of the data analysis.