Automated detection of microstructural features that have unique protein markers and are prognostic for chronic kidney disease

The primary microstructural attributes seen in “normal” kidneys are nephrosclerosis (arteriosclerosis, global glomerulosclerosis, and interstitial fibrosis/tubular atrophy), nephron number, and nephron size. However manual measures of these microstructures are impractical. Further, understanding their pathophysiology may lead to new interventions for kidney disease. Automated morphometry with deep learning (DL) networks may be able to rapidly measure nephrosclerosis, nephron number, and nephron size. Novel morphometry of structures impacted by glomerular hyperfiltration (podocyte and parietal epithelial cell (PEC) density, Bowman’s space, and the diameter of proximal and distal tubule) and of microvasculature are needed to better understand the pathophysiology of early disease. Proteomic analysis of specific microstructures that differ between kidneys that do versus do not develop CKD outcomes has the potential to identify prognostic or even pathogenic proteins for early kidney disease. The multi-discipline multi-site Aging Kidney Anatomy study has unique resources for the study of microstructure in “normal” kidneys. This includes data and specimens on living kidney donors including needle core biopsies at donation with digitized whole slide images (WSI) and long-term CKD outcomes in the donor and recipient. This also includes data and specimens on patients who had a radical nephrectomy for tumor including digitized WSI of kidney wedge sections and annual eGFR testing for CKD outcomes during follow-up. Aim 1 will determine if automated morphometry of nephrosclerosis, nephron number, and nephron size predicts CKD outcomes to test the hypothesis that DL tools allows for efficient quantification of these clinically relevant microstructural attributes. This aim will use both previously developed and new DL networks, develop models to predict CKD outcomes from automated morphometry, and compare prediction of CKD outcomes between automated and manual morphometry. Aim 2 will characterize novel microstructural attributes that associate with kidney function, CKD risk factors, and CKD outcomes to test the hypothesis that encoded in the kidney tissue are unexplored structural attributes that reflect the glomerular hyperfiltration and interstitial microvascular status that are prognostic for CKD. This aim will automatically quantify podocytes, PECs, peritubular capillaries (PTC), Bowman’s space (volume), and proximal and distal tubules (diameter) on WSI using previously developed and newly developed DL tools and associate these structures with kidney function, CKD risk factors, and CKD outcomes. Aim 3 will discover protein markers linked to the microstructural attributes that are prognostic for CKD outcomes to test the hypothesis that differentially expressed proteins contained within kidney microstructures predict CKD outcomes. This aim will use laser capture microdissection, mass spectroscopy-based proteomics (both discovery and targeted validation approaches), and immunohistochemistry to identify proteins on kidney biopsy sections that predict CKD outcomes and to determine their association with microstructural attributes.