Secondary Logo

Investigating Blood-Based, Cell-Specific Biomarkers of Acute Cardiac Allograft Rejection

Shannon, Casey P.1; Kim, JiYoung1; Chen, Virginia1; Hollander, Zsuzsanna1; Lam, Karen1; Wilson-McManus, Janet1; Assadian, Sara1; Balshaw, Robert1,2; Tebbutt, Scott J.1,3,8; McMaster, Robert4; Keown, Paul5; Ng, Raymond T.1,6,8; McManus, Bruce M.1,7,8

doi: 10.1097/01.tp.0000520331.07917.b7
118.3
Free

1PROOF Centre of Excellence, UBC, Vancouver, BC, Canada; 2Statistics, UBC, Vancouver, BC, Canada; 3Medicine, Division of Respiratory Medicine, UBC, Vancouver, BC, Canada; 4Medicine, Division of Nephrology, UBC, Vancouver, BC, Canada; 5Computer Science, UBC, Vancouver, BC, Canada; 6Pathology and Laboratory Medicine, UBC, Vancouver, BC, Canada; 7Department of Transplantation and Immunology, Vancouver Coastal Health Research Institute, Vancouver, BC, Canada; 8Centre for Heart Lung Innovation, UBC, Vancouver, BC, Canada.

Biomarkers in Transplantation.

Introduction: Cardiac transplantation is the main intervention for patients with end-stage heart failure. Despite great improvements in maintenance immunosuppressive therapies, acute allograft rejection remains a clinical problem. Timely detection of moderate rejection allows for treatment to be modified, preventing organ damage, graft failure and patient death. Regular monitoring for rejection is thus crucial. The endomyocardial biopsy (EMB) is the current standard for monitoring the allograft, but the procedure is highly invasive and costly. Consequently, there is great interest in developing blood-based biomarker tests to monitor for allograft rejection. Blood is a complex tissue, however, and accounting for its dynamic cellular heterogeneity is likely key to identifying robust biomarkers. We investigated whether integrating cellular composition estimates can lead to better performing biomarkers in the context of acute cardiac allograft rejection.

Methods: Genome-wide transcript abundance was assayed from PAXgene peripheral whole blood samples obtained from patients undergoing biopsy-confirmed acute allograft rejection (≥2R; n = 29) or not (n = 198), using Affymetrix Human Gene 1.1 ST microarrays. The cellular composition of the blood samples was inferred from their gene expression profiles. Elastic net classifier panels were then identified using the whole blood gene expression, either unadjusted for cellular composition, adjusted for cellular composition for all cell types, or all but one cell type. Out-of-sample performance of the various models was estimated using stratified 5-fold cross-validation.

Results: Best cross-validation performance (AUC = 0.75) was achieved by an NK cell-specific classifier panel. Both cell-adjusted and cell-specific classifiers outperformed those identified using unadjusted whole blood gene expression data.

Conclusion: Cell-adjusted/specific biomarker panels outperformed those derived from unadjusted whole blood gene expression. Best performance was achieved by an NK cell-specific biomarker panel. The method presented here is a promising way to incorporate cellular composition in the context of biomarker discovery from gene expression in tissue admixtures.

Figure

Figure

Copyright © 2017 Wolters Kluwer Health, Inc. All rights reserved.