The genetic architecture of the human skeletal form

Abstract

The human skeletal form underlies our ability to walk on two legs, but unlike standing height, the genetic basis of limb lengths and skeletal proportions is less well understood. Here we applied a deep learning model to 31,221 whole body dual-energy X-ray absorptiometry (DXA) images from the UK Biobank (UKB) to extract 23 different image-derived phenotypes (IDPs) that include all long bone lengths as well as hip and shoulder width, which we analyzed while controlling for height. All skeletal proportions are highly heritable (~40-50%), and genome-wide association studies (GWAS) of these traits identified 179 independent loci, of which 102 loci were not associated with height. These loci are enriched in genes regulating skeletal development as well as associated with rare human skeletal diseases and abnormal mouse skeletal phenotypes. Genetic correlation and genomic structural equation modeling indicated that limb proportions exhibited strong genetic sharing but were genetically independent of width and torso proportions. Phenotypic and polygenic risk score analyses identified specific associations between osteoarthritis (OA) of the hip and knee, the leading causes of adult disability in the United States, and skeletal proportions of the corresponding regions. We also found genomic evidence of evolutionary change in arm-to-leg and hip-width proportions in humans consistent with striking anatomical changes in these skeletal proportions in the hominin fossil record. In contrast to cardiovascular, auto-immune, metabolic, and other categories of traits, loci associated with these skeletal proportions are significantly enriched in human accelerated regions (HARs), and regulatory elements of genes differentially expressed through development between humans and the great apes. Taken together, our work validates the use of deep learning models on DXA images to identify novel and specific genetic variants affecting the human skeletal form and ties a major evolutionary facet of human anatomical change to pathogenesis.

Deep learning based phenotyping of medical images improves power for gene discovery of complex disease

“…We first restricted the dataset to individuals of white British ancestry, applied standard variant and sample QC and analyzed 12.1 million common bi-allelic SNPs with minor allele frequency > 0.1% 1 ( Methods : Genetic QC). Next, as the bulk imaging data from the UKB comprised of DXA images that reflect scans of different body parts, we used a deep learning approach 15 to subset the imaging dataset to only AP view knee scans. We then removed individuals that had outlier image resolutions or poor quality DXA scans, and padded images to a standard size for processing (see Methods : Image segmentation, phenotype measurement and quality control).…”

“…DXA images in DICOM format were first organized by anatomy following the manifest files located in each directory output by the imaging machine. DXA scans were subject to further quality control following the methods described in Kun et al, 2022 15 . Following initial data cleaning, AP view knee DXA scans were converted from DICOM to JPG format using the pydicom library 35 .…”

“…Taking advantage of these developments in computer vision, recent genetic studies have successfully applied these methods to generate image derived phenotypes (IDPs) of distribution of body fat, heart structure, liver fat percentage, and brain morphology, and have linked these novel traits with genome-wide significant loci [10][11][12][13][14] . While some recent studies on musculoskeletal disease employ these novel phenotyping approaches [15][16][17] , neither these nor the studies on other traits have specifically investigated how generating quantitative IDPs that underlie binary disease status could be used to improve power for gene discovery at biobank scale.…”

The genetic architecture and evolutionary consequences of the human pelvic form

“…Each individual in the UKBiobank had a DXA image folder containing up to 8 different body parts. In order to check the labels of these body parts that were defined using their file name, we built a convolutional neural network (CNN) to sort the images by body part through the use of a multiclass classification model using a previously published protocol ( 36 ). After sorting and removal of images, we were left with 42,228 full skeleton X-rays ( Table S3 ).…”

“…S8), suggesting that an additional round of training was useful to reduce the variation in manual annotation to a minimum. The left panel displays the scatter plot of the left-to-right arm ratio from two imaging visits using HRNet, sourced from (36). The right panel shows the scatter plot of the same ratio from two imaging visits but using our optimized HRNet model.…”