The Banff Classification for Allograft Pathology includes the use of gene expression in the diagnosis of antibody-mediated rejection (AMR) of kidney transplants, but a predictive set of genes for classifying biopsies with 'incomplete' phenotypes has not yet been studied. Here, we developed and assessed a gene score that, when applied to biopsies with features of AMR, would identify cases with a higher risk of allograft loss. To do this, RNA was extracted from a continuous retrospective cohort of 349 biopsies randomized 2:1 to include 220 biopsies in a discovery cohort and 129 biopsies in a validation cohort. The biopsies were divided into three groups: 31 that fulfilled the 2019 Banff Criteria for active AMR, 50 with histological features of AMR but not meeting the full criteria (Suspicious-AMR), and 269 with no features of active AMR (No-AMR). Gene expression analysis using the 770 gene Banff Human Organ Transplant NanoString panel was carried out with LASSO Regression performed to identify a parsimonious set of genes predictive of AMR. We identified a nine gene score that was highly predictive of active AMR (accuracy 0.92 in the validation cohort) and was strongly correlated with histological features of AMR. In biopsies suspicious for AMR, our gene score was strongly associated with risk of allograft loss and independently associated with allograft loss in multivariable analysis. Thus, we show that a gene expression signature in kidney allograft biopsy samples can help classify biopsies with incomplete AMR phenotypes into groups that correlate strongly with histological features and outcomes.
Keywords: B-HOT; NanoString; pathology; rejection; transcriptomics; transplantation.
Copyright © 2023 International Society of Nephrology. Published by Elsevier Inc. All rights reserved.