|Home | About | Journals | Submit | Contact Us | Français|
We have previously validated the use of dual-energy X-ray absorptiometry (DXA) for measuring body composition of mice using the GE-Lunar PIXImus and software version 1.42 . Since that report, newer versions of the software have been released. The purpose of the present study was to compare results from our original study with results analyzed using two newer versions of software (versions 1.44 and 1.45). Body composition data (lean tissue mass [LTM], fat mass [FM], bone mineral content [BMC], and bone mineral density [BMD]) were obtained from DXA scans of twenty-five, anesthetized male C57Bl/6J mice (6-11 weeks old; 19 to 29g). Relative to version 1.42, versions 1.44 and 1.45 significantly (P<0.001) overestimated LTM and BMD and underestimated FM and BMC. However, compared to carcass analysis, versions 1.44 and 1.45 significantly overestimated both FM and LTM and underestimated BMC. Results from 1.44 and 1.45 were highly correlated with carcass values for all body composition parameters. Prediction equations were developed for the two new software versions. Applying the prediction equation from 1.42, to the data obtained from 1.44 and 1.45 resulted in FM and LTM that were worse than if no equation was used. However, using their own developed equations resulted in data that were not significantly different than that from carcass analysis. These data suggest that software-specific equations are necessary for comparing DXA-derived data to that of chemical analysis.
Although small animals are used extensively in the field of obesity research, the techniques for analyzing their body composition are limited. Dual-energy X-ray absorptiometry (DXA) has gained popularity as a research tool for determining in vivo body composition in humans, providing measures of fat mass (FM), lean tissue mass (LTM), bone mineral content (BMC), and bone mineral density (BMD). There has also been a growth in the use of DXA for animal osteoporosis studies, allowing longitudinal measurements to be made [2;3]. We have previously validated the use of DXA for measuring the body composition of mice using the GE-Lunar PIXImus and software version 1.42 . The DXA-derived values were closely correlated with the chemical values although DXA did overestimate fat mass and underestimate lean mass. However, prediction equations were developed allowing the data to be ‘normalized’ to that obtained using chemical extraction techniques. DXA has now become an established method for conducting longitudinal studies of body composition, and bone mineral density in small animals. Since our previous validation, two subsequent software versions 1.44 and 1.45 have been developed. According to the manufacturer, these newer versions were modified only in the analysis portion of the software, rather than the acquisition portion. Therefore the purpose of this study was to compare results from our original study with results analyzed using the two newer versions of software (versions 1.44 and 1.45), to determine whether different prediction equations are needed for the new software versions.
The animals and data for the carcass analysis and DXA software version 1.42 are from . The chemical extraction and DXA procedures are described in more detail in the previous publication. Briefly, twenty-five male C57BL/6J mice (6-11 weeks old; 19 to 29g) were anesthetized and scanned using a peripheral densitometer (GE-Lunar PIXImus). Due to the size limitation of the imaging area, the head was excluded from the DXA analysis using the region of interest tool. In addition, the heads were removed from the carcasses prior to body composition analysis. Thus, all data presented for body composition exclude the head. Water content was determined by opening the carcasses, and drying until constant weight at 60°C. The remaining dried carcass was ground and placed in a Soxhlet apparatus with petroleum ether for 8 hours to extract the fat, and leaving fat-free dry mass (FFDM). Ash content was determined by burning at 600°C for 8 hours . Lean mass was calculated as the fat-free dry mass, minus ash content plus the water content. Bone ash was calculated from total body ash using the correction factor determined in Nagy and Clair .
The original DXA scans of the mice were reanalyzed in the two newer software versions – 1.44 and 1.45. This resulted in three different estimates of fat mass (FM), lean tissue mass (LTM), bone mineral content (BMC), and bone mineral density (BMD) from the three software versions. DXA scans from 1.42 were analyzed by one person and all 1.44 and 1.45 scans were analyzed by another person. To determine the effect of having two people analyze the scans, 20 scans were analyzed separately by two different people to assess the reproducibility of the analysis. The average difference in the body composition parameters were: BMD 0.0001 g/cm2, BMC 0.0025g, LTM 0.09g, FM 0.035g.
Data were analyzed by paired t-tests and regression techniques. Data were analyzed using SAS (version 7.0; SAS Institute, Cary, NC) with significance levels set at 0.05.
Software version 1.42 gave significantly different results than both 1.44 and 1.45 for all body compartments (Table 1). Results for BMC and FM were significantly higher with 1.42, than either 1.44 or 1.45 (P<0.001). LTM and BMD were significantly lower with 1.42 than with either 1.44 or 1.45 (P<0.001). Compared to carcass values, both newer versions still significantly overestimated FM (P<0.001), however the extent of the overestimation was much reduced (v1.44, 0.19 ± 0.05g; v1.45, 0.21 ± 0.05g). Whereas v1.42 underestimated LTM (compared to carcass), both newer versions significantly overestimated LTM (P<0.001, v1.44, 1.34 ± 0.06g; v1.45, 1.31 ± 0.06g). Bone ash was significantly underestimated in both 1.44 (−0.039 ± 0.002g) and 1.45 (−0.038 ± 0.002g) (P<0.001). None of the body compartments were significantly different between versions 1.44 and 1.45 (P>0.05).
Results for FM from all three software versions were significantly related to carcass fat (P<0.001, r2>0.77). None of the slopes were significantly different than one (P>0.10), however the intercept of the relationship between version 1.42 and carcass fat was significantly different than zero (P<0.05). All three software versions significantly overestimated FM (P<0.001), however the extent of the overestimation was much greater for 1.42 (2.19 ± 0.06g) than for either 1.44 (0.19 ± 0.05g), or 1.45 (0.21 ± 0.05g).
The relationship between DXA LTM and chemical LTM was highly significant with all three software versions (P<0.001, r2>0.98). None of the relationships had intercepts that were significantly different than zero (P>0.05), however, the slope of the relationship between DXA LTM using 1.44 and chemical LTM was significantly less than one (P=0.03). Whereas 1.42 had significantly underestimated LTM (−0.58 ± 0.05g P<0.001), versions 1.44 and 1.45 significantly overestimated LTM (1.34 ± 0.06g and 1.31 ± 0.06g respectively, P<0.001).
DXA BMC was significantly related to chemical bone ash using all three software versions (P<0.001, r2>0.96). However, the intercepts using all three versions were significantly different than zero (P<0.01), and the slope of the relationship using version 1.42 was significantly less than one (P=0.003). Whereas 1.42 slightly overestimated bone ash (0.005 ± 0.002g, P=0.065), versions 1.44 and 1.45 both significantly underestimated bone ash (−0.039± 0.002g and −0.038 ± 0.002g respectively P<0.001).
New prediction equations were developed based on the above relationships for versions 1.44 and 1.45 and are presented in Table 2 (for v1.45 only). Backward elimination regression was used entering DXA values of FM, LTM, and BMC into all the models. Chemical FM was best predicted by DXA FM for both 1.44 (r2=0.80, P<0.001, RMSE=0.26g) and 1.45 (r2=0.79, P<0.001, RMSE=0.27g). Chemical LTM was best predicted by a model containing both DXA LTM and FM for both 1.44 (r2=0.99, P<0.001, RMSE=0.23g) and 1.45 (r2=0.99, P<0.001, RMSE=0.24g). Chemical bone ash was best predicted by the model including DXA BMC for 1.44 (r2=0.97, P<0.001, RMSE=0.009g) and 1.45 (r2=0.96, P<0.001, RMSE=0.01g).
These new equations for versions 1.44 and 1.45 are different from those developed for 1.42 . To assess the effect of these different equations, predicted FM and LTM were calculated for data analyzed in version 1.45 using both the 1.42 equations and the 1.45 equations and the results compared to carcass FM and LTM. This was repeated for 1.44, using 1.42 and 1.44 equations, but only the results for v1.45 are presented as they are the same. Predicted FM using the software specific equations was not significantly different than the carcass FM (Figure 1). However, using the equations developed for 1.42 resulted in significantly lower FM than either carcass or software-specific predicted FM (P<0.001). In addition, the predicted FM using 1.42 was actually less than 0. In the comparison of carcass LTM to the predicted LTM using software specific equations, there was no significant difference (P>0.05) (Figure 2). However, using the 1.42 equation to predict LTM analyzed in version 1.45 resulted in significantly higher LTM (P<0.001). Predicted bone ash using either 1.44 or 1.45 was not significantly different than carcass bone ash (P>0.05) (Figure 3).
This study examined whether two newer versions of software for the PIXImus differed in the analysis of existing scans, and whether prediction equations developed for software version 1.42  are valid for the newer software versions. Although the results from the two newer software versions were significantly correlated with carcass values, they were significantly different than the results from version1.42 for all body compartments. The results from the two newer versions were not significantly different from each other. New, software-specific, prediction equations were developed and when applied to the data, there was no longer a significant difference compared to carcass analysis.
It has been shown previously that DXA overestimates FM when compared with carcass analysis [1;4], and the two newer software versions were no different in this. However, the extent of the overestimation was much reduced in the newer versions: from 2.19g , to 0.20g with the new software. This is a much more accurate result than that obtained with v.1.42, or that obtained by Brommage  of 3.3g overestimated.
Although the new software is more accurate in measuring FM, the same is not true for LTM. Whereas v1.42 underestimated the amount of LTM , both 1.44 and 1.45 significantly overestimated the amount of lean mass. The magnitude of this discrepancy is also greater in the new software (almost 2g) compared with v1.42 (−0.6g). Interestingly, the magnitude of the improvement in the measurement of FM was almost identical to the magnitude of the new error in LTM.
Bone mineral content was underestimated using the newer software versions and a prediction equation had to be developed. In the previous software version, DXA and chemical estimates of BMC were not significantly different, and so an equation was not needed.
Although the original analysis was performed by someone different than the analyses of 1.44 and 1.45, it is unlikely that this explained the differences in the results. The average differences in the DXA data (as presented in the methods) as a result of having 2 people analyze the data, are too small to account for the differences between the software versions. For example, there was a 2g difference in fat mass between v1.42 and v1.44/v1.45, whereas the error associated with having different people analyze the data was only 0.035g for fat mass.
Although chemical and DXA estimates of body composition were significantly different with the newer software versions, there was no relationship between the discrepancy and the chemical values for any of the body compartments (Figures (Figures11--3).3). For example, the discrepancy between the DXA and chemical estimate of fat mass was not related to how fat the mice were.
Due to these differences between v1.42 and the newer versions, it was not possible to apply the prediction equations from v1.42 to the newer software. Correcting data obtained using v1.44 or 1.45 with the equation from v1.42, resulted in a negative value for fat mass and an overestimation of lean mass that was no better than before applying the equation. New, software-specific prediction equations were developed from v1.44 and v1.45, which when applied to the data result in numbers that are not significantly different than carcass values. These data suggest that software-specific equations are necessary for comparing DXA-derived data to that from chemical analysis. Therefore, it is important to know what software version is being used so the appropriate prediction equations can be applied.
Supported by NIH DK56336, and AR46031.