Molecular Signature in the Peripheral Blood for Sub-Clinical Acute Kidney Rejection.

Abstract

Background: Sub-clinical acute rejection (SCAR) is defined as histologic rejection with normal serum creatinine and is associated with worse long term graft survival. Protocol kidney biopsies (bx) are done at some centers to detect SCAR. Methods: We identified 23 whole blood samples from recipients with SCAR and profiled these using Affymetrix HG-U133 Peg microarrays. We compared these to 21 pts with clinical acute rejection (AR), and 25 with normal bx and a normal serum creatinine (TX). SCAR bx included 16 “Borderline” and 7 Banff 1A (Banff 07). 3-way ANOVA analysis was performed between Groups 1-3. Results: The arrays yielded over 6,000 differentially expressed probesets at p<0.001. Even setting a False Discovery Rate (FDR) cut-off of <10% gave us over 2,500 significantly differentially expressed probesets. For the diagnostic signature, we used only the top 200 differentially expressed probesets as ranked by p values. These 200 top probesets have FDR values of <0.05%. We used 3 different predictive algorithms: Nearest Centroid (NC), Support Vector Machines (SVM) and Diagonal Linear Discriminant Analysis (DLDA) to build predictive models. NC, SVM and DLDA picked classifier sets of 188, 192 and 200 probesets as the best classifiers, respectively. Table 1 shows the performance of the 3-way NC classifier using both one-level cross validation as well as Optimism Corrected Bootstrapping (1000 data sets) as described by Harrell. We also tested a 2-way prediction of SCAR vs. TX to further validate that a phenotype as potentially subtle clinically as SCAR can be truly distinguished from TX. At a p-value <0.001, there were 33 probesets whose expression signals highly differentiated SCAR and TX. When these 33 probesets were used in NC to predict SCAR and TX creating a 2-way classifier, the predictive accuracies with a one-level cross-validation was 96% and with the Harrell 1000 test optimism correction it was 94%. Table 1 shows NC algorithm values.Table: No Caption available.Conclusion: Peripheral blood gene expression profiling can distinguish SCAR, TX and AR and promises a minimally invasive method for monitoring KTx recipients. This signature is now being further validated. DISCLOSURE:Friedewald, J.: Grant/Research Support, Pfizer, Stockholder, TGI. Kurian, S.: Stockholder, TGI. Levitsky, J.: Stockholder, TGI. Abecassis, M.: Stockholder, TGI. Salomon, D.: Stockholder, TGI.