Deep learning identifies pathological abnormalities predictive of graft loss in kidney transplant biopsies

Yi, Z.; Salem, F.; Menon, M. C.; Keung, K.; Xi, C.; Hultin, S.; Rasheed, M. R. H. A.; Li, L.; Su, F.; Sun, Z.; Wei, C.; Huang, W.; Fredericks, S.; Lin, Q.; Banu, K.; Wong, G.; Rogers, N. M.; Farouk, S.; Cravedi, P.; Shingde, M.; Smith, R. N.; Rosales, I. A.; O'Connell, P. J.; Colvin, R. B.; Murphy, B.; Zhang, W.

doi:10.1101/2021.04.18.440166

Home/Archive/bioRxiv

bioRxiv · 10.1101/2021.04.18.440166

Deep learning identifies pathological abnormalities predictive of graft loss in kidney transplant biopsies

View original source ↗Export metadata ↓

Abstract

BackgroundInterstitial fibrosis, tubular atrophy, and inflammation are major contributors to renal allograft failure. Here we seek an objective, quantitative pathological assessment of these lesions to improve predictive utility. MethodsWe constructed a deep-learning-based pipeline recognizing normal vs. abnormal kidney tissue compartments and mononuclear leukocyte (MNL) infiltrates from Periodic acid-Schiff (PAS) stained slides of transplant biopsies (training: n=60, testing: n=33) that quantified pathological lesions specific for interstitium, tubules and MNL infiltration. The pipeline was applied to 789 whole slide images (WSI) from baseline (n=478, pre-implantation) and 12-month post-transplant (n=311) protocol biopsies in two independent cohorts (GoCAR: 404 patients, AUSCAD: 212 patients) of transplant recipients to correlate composite lesion features with graft loss. ResultsOur model accurately recognized kidney tissue compartments and MNLs. The digital features significantly correlated with Banff scores, but were more sensitive to subtle pathological changes below the thresholds in Banff scores. The Interstitial and Tubular Abnormality Score (ITAS) in baseline samples was highly predictive of 1-year graft loss (p=2.8e-05), while a Composite Damage Score (CDS) in 12-month post-transplant protocol biopsies predicted later graft loss (p=7.3e-05). ITAS and CDS outperformed Banff scores or clinical predictors with superior graft loss prediction accuracy. High/intermediate risk groups stratified by ITAS or CDS also demonstrated significantly higher incidence of eGFR decline and subsequent graft damage. ConclusionsThis deep-learning approach accurately detected and quantified pathological lesions from baseline or post-transplant biopsies, and demonstrated superior ability for prediction of posttransplant graft loss with potential application as a prevention, risk stratification or monitoring tool.

Source connections

Published article ↗

Explore related subjects

pathology Preprint

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Yi, Z., Salem, F., Menon, M. C., Keung, K., Xi, C., Hultin, S., Rasheed, M. R. H. A., Li, L., Su, F., Sun, Z., Wei, C., Huang, W., Fredericks, S., Lin, Q., Banu, K., Wong, G., Rogers, N. M., Farouk, S., Cravedi, P., Shingde, M., Smith, R. N., Rosales, I. A., O'Connell, P. J., Colvin, R. B., Murphy, B., Zhang, W.. 2021-04-19. Deep learning identifies pathological abnormalities predictive of graft loss in kidney transplant biopsies. https://doi.org/10.1101/2021.04.18.440166

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

Midlife nonlinearities across scale in Biondi body amyloid formation in the human choroid plexus

Human choroid plexus epithelial cells (CPECs), which produce cerebrospinal fluid and form the blood-cerebrospinal fluid barrier, accumulate discrete intracellular amyloid inclusions called Biondi bodies (BBs). Using systematic manual and machine learning approaches on the choroid plexus glomus, we determined the percentages of CPECs containing thioflavin S-stained BBs as well as BB load and spatial distributions in whole slide images from 77 cases across the human lifespan. On average, BB prevalence was near zero until age 40, rose steeply in the 5th and 6th decades, and exceeded 50% after age 70. As prevalence plateaued, BB load continued to rise with age. Accelerated BB accumulation was seen in two cases of diabetic ketoacidosis, including a 16-year-old, while an 85-year-old with the relatively rare APOE2/E2 genotype had fewer BBs than expected. BB prevalence in the choroid plexus glomus was not associated with Alzheimer's disease, but was lower in females than males. Notably, spatial clustering of BBs was universal across cases while nearest-neighbor analyses suggested cell-to-cell, prion-like BB spread that can account for the nonlinearities seen at tissue and population scales. The abrupt, nonlinear BB accumulation during midlife coincides with or precedes other common aging-associated neuropathologies of the human brain, consistent with broad BB and choroid plexus contributions to brain aging and disease.

pathology↗

bioRxiv2026

Elevated precontractile calcium set point is sustained by limited PMCA1 export in dystrophic muscle

Duchenne muscular dystrophy is caused by loss of dystrophin and marked by elevated muscle calcium, usually attributed to contraction-induced membrane damage. Whether calcium rises before muscle can contract, and what sustains it, is unresolved. In Caenorhabditis elegans embryos and patient-derived human myoblasts, dystrophic muscle carries an elevated resting calcium set point before contraction is possible. Embryonic calcium is elevated 400 minutes after fertilization, before movement is detectable or the neuromuscular junction forms. Silencing the plasma membrane calcium pump gene mca-1 raises calcium in wild-type but not dystrophic animals, whereas silencing its paralogue raises it in both: the limitation is specific to that export route. Dystrophic human myoblasts have less pump; pump inhibition or silencing increases calcium preferentially in healthy cells, and raising pump expression lowers calcium most in dystrophic cells. Constrained calcium export is an early, continuing determinant of the dystrophic set point.

pathology↗

bioRxiv2026

Systematic comparison of gliflozins identifies a selective anti-calcifying effect of canagliflozin beyond canonical SGLT2 inhibition

Background: Sodium-glucose cotransporter 2 inhibitors (SGLT2i) provide important cardiorenal benefits in chronic kidney disease (CKD), but their effects on vascular calcification remain conflicting. Whether potential anti-calcifying effects are shared across gliflozins or involve mechanisms beyond canonical SGLT2 inhibition is unknown. Methods: We systematically compared canagliflozin, dapagliflozin and empagliflozin in a high-phosphate diet-induced mouse model of vascular calcification and complementary ex vivo and in vitro models, including rodent and human vascular tissues and vascular smooth muscle cells (VSMCs). Aortic transcriptional response was investigated by RNA sequencing. VSMC calcification, proliferation, migration, viability, glucose uptake and bioenergetics were investigated using complementary cellular assays, including OsteoSense staining, Annexin V/propidium iodide staining, radiolabeled 2-deoxyglucose uptake and Seahorse ATP rate analysis. SGLT2/1 expression was examined by RT-qPCR, Western blotting and immunofluorescence. We also evaluated sotagliflozin, a dual SGLT1/2 inhibitor, and mizagliflozin, a selective SGLT1 inhibitor. Results: Among the three gliflozins, only canagliflozin significantly reduced aortic calcification in vivo. Its anti-calcifying effect was also observed in rodent aortic rings, human vascular explants, cultured VSMCs and aortic rings from diabetic db/db mice. Aortic RNA-seq revealed that canagliflozin partially reversed HPD-induced transcriptional alterations, including changes in pro-calcific signaling and cellular metabolism, while some endothelial and inflammatory responses were shared among gliflozins. In mouse VSMCs, canagliflozin reduced calcification across all tested concentrations, dose-dependently decreased osteogenic marker expression, and reduced proliferation and migration. Increased apoptotic and damaged cell fractions were observed at higher concentrations, but not at lower concentrations. Canagliflozin also reduced glucose uptake and mitochondrial ATP production, without significantly affecting glycolytic ATP production. Neither dapagliflozin nor empagliflozin reproduced these cellular and metabolic effects at any of the concentrations tested. SGLT2 and SGLT1 expression was low or undetectable in vascular tissues and cells, and neither sotagliflozin nor mizagliflozin reproduced the anti-calcifying effect of canagliflozin in vivo or in vitro. Conclusions: Canagliflozin exhibits a selective anti-calcifying profile among the gliflozins tested, associated with changes in VSMC responses and cellular metabolism. These findings suggest that its vascular protective effects may involve mechanisms beyond canonical SGLT2 inhibition.

pathology↗