Application of RNA-seq Derived Diagnostic Algorithms of T-cell Mediated Kidney Rejection (TCMR) to Publicly Available DNA Microarray-based Gene Expression Datasets. (July 2018)

• Record Type: Journal Article
• Authors: Wang, Zijie; Liu, Peng; Zeng, George; Randhawa, Parmjeet
• Is Part Of: Transplantation. Volume 102(2018)Supplement 7S-1
• Publication Date: 2018-07
• Subjects: Transplantation of organs, tissues, etc -- Periodicals; Transplantation immunology -- Periodicals; 617.95
• DOI: 10.1097/01.tp.0000542574.27498.a2
• Languages: English
• ISSNs: 0041-1337

Abstract

Abstract : Recent years have seen the development of DNA microarray based tools for the diagnosis of TCMR. This study seeks to develop a similar tool using RNA-seq, since this technology interrogates more genes, with better quantitation, and over a wider dynamic range, while allowing single base resolution. Moreover, we have adapted RNA-seq for use with formalin fixed tissue, which allows exactly the same biopsy core to be examined for both histologic and molecular changes. RNA-seq was performed on 10 renal transplant biopsies, 5 with TCMR and 5 with stable (STA) function, using the Invitrogen Pure Link FFPE tissue Total RNA Isolation method, the Ion Ampliseq Trancriptome Human Gene Expression Kit, and the IonTorrent sequencing platform. Machine learning tools were developed to distinguish between TCMR and STA samples using 252 DE genes identified by DSeq2 (False Discovery rate <0.05). Internal cross validation using five-fold leave-out-one-cross-validation demonstrated sensitivity and specificity of 100% with the Support Vector Machines algorithm, and 80% for linear discriminant analysis (LDA). Cross platform external validation was performed on two different sample sets containing a total of 703 biopsies analyzed by DNA microarray technology (GSE48581 INTERCOM 300 Study, and GSE36059 BFC403 study, with respectively 76 and 104 DE genes in common with the training set). The LDA based training algorithm correctly predicted TCMR in 58/67 biopsies in the validation dataset, if the discriminatory probability was set at 0.2 in accordance with prior clinically validated studies. TCMR was also identified in 74/105 biopsies designated as ABMR in GSE1, as well as 259/503 that were classified as Non-Rejection. These data illustrate that biopsies labeled simply as ABMR frequently have co-existent TCMR. The presence of a TCMR-like RNA-seq signature in biopsies labeled as Non-rejection highlights need to develop bioinformatics tools that will allow more precise molecular sub-classification of this category into disease that are mimics of acute rejection, including currently undiagnosed smoldering TCMR, infection associated interstitial nephritis, drug hypersensitivity reactions, and recurrent glomerulonephritis.