Natural Phenotypes of Fat, Lean, and Bone and Their Association to Metabolic Markers

Abstract

Introduction Our group have developed a method to deconvolute whole body DXA scans into three images, each consisting uniquely of fat, lean, and bone masses. These images are being explored as more sensitive markers of risk for a variety of diseases. We asked if images of fat, lean, and bone distribution can be used to define natural phenotypical groupings and if these groupings are related to common risk factors for metabolic disease and bone fragility. Methods Participants were from the Shape Up! Adults cohort study. Each had the following: a whole body DXA scans (Hologic Horizon/A), strength assessments, and blood metabolic biomarkers. Each DXA scan was deconvolved using proprietary methods to create fat, lean, and bone mass images accurate on the pixel level. Like image types were spatially registered using 105 fiducial points. Statistical appearance and shape modeling was then performed. The sex-specific population variances for appearance (DXA) was captured as principal components (PC) resulting in 6 PC models: PCFATmen, PCFATwomen, PCLEANmen, PCLEANwomen, PCBONEmen and PCBONEwomen. K-means cluster algorithms were used to determine the clusters of PC variables that minimized the variable variances and to define a natural number of clusters. Simple t-tests were used to determine the variables that best described the uniqueness of each cluster in terms of metabolic markers, demographics, body composition, and bone density measures. Results Presently, 176 participants (72 men) were available with a mean age of 41 for men and 46 for women. For women, it took 38 PCs to describe 95% of the fat variance, 38 PCs to describe 95% of the lean variance, 39 PCs to describe 95% of the bone variance. For men, it took 32, 35, and 35 PCs to describe 95% of the fat, lean, and bone variances respectively. As example, using the fat images, there were nine phenotypes for women and five for men defined by the fat images where each had unique metabolic profiles. Several phenotypes were found to have a high risk of metabolic syndrome. See Figure 2 for women and Figure 3 for men. Conclusions Fat, lean, and bone images created from a single DXA scan can be used to classify patients into meaningful phenotypes that relate to metabolic status. These phenotypes are unique for men and women.