The kSORT assay to detect renal transplant patients at high risk for acute rejection: results of the multicenter AART study

Abstract

Background: Development of noninvasive molecular assays to improve disease diagnosis and patient monitoring is a critical need. In renal transplantation, acute rejection (AR) increases the risk for chronic graft injury and failure. Noninvasive diagnostic assays to improve current late and nonspecific diagnosis of rejection are needed. We sought to develop a test using a simple blood gene expression assay to detect patients at high risk for AR.

Methods and findings: We developed a novel correlation-based algorithm by step-wise analysis of gene expression data in 558 blood samples from 436 renal transplant patients collected across eight transplant centers in the US, Mexico, and Spain between 5 February 2005 and 15 December 2012 in the Assessment of Acute Rejection in Renal Transplantation (AART) study. Gene expression was assessed by quantitative real-time PCR (QPCR) in one center. A 17-gene set--the Kidney Solid Organ Response Test (kSORT)--was selected in 143 samples for AR classification using discriminant analysis (area under the receiver operating characteristic curve [AUC] = 0.94; 95% CI 0.91-0.98), validated in 124 independent samples (AUC = 0.95; 95% CI 0.88-1.0) and evaluated for AR prediction in 191 serial samples, where it predicted AR up to 3 mo prior to detection by the current gold standard (biopsy). A novel reference-based algorithm (using 13 12-gene models) was developed in 100 independent samples to provide a numerical AR risk score, to classify patients as high risk versus low risk for AR. kSORT was able to detect AR in blood independent of age, time post-transplantation, and sample source without additional data normalization; AUC = 0.93 (95% CI 0.86-0.99). Further validation of kSORT is planned in prospective clinical observational and interventional trials.

Conclusions: The kSORT blood QPCR assay is a noninvasive tool to detect high risk of AR of renal transplants. Please see later in the article for the Editors' Summary.

Conflict of interest statement

MS is Chair of the SAB and Founder of Organ-I and Consultant for Immucor, Bristol Meyers Squibb, UCB, ISIS, Genentech; SR was a Consultant for Organ-I; TS and NS are Consultants for Organ-I, Immucor; FV has research grants with Astellas Pharma, Bristol Myers Squibb, Alexion, Pfizer, Novartis, Genentech.

Figures

Figure 1. AART study design in 436 unique adult/pediatric renal transplant patients from eight transplant centers.

558 PB samples from 436 adult and pediatric renal transplant patients collected from eight independent transplantation centers in the US, Spain, and Mexico were assessed. Emory, UCLA, UPMC, CPMC, UCSF, and Barcelona contributed adult samples, Mexico and Stanford, pediatric samples. For kSORT QPCR analysis, cross-sectional AART samples were divided into three cohorts: AART143, n = 143 adult samples from 135 patients for kSORT gene modeling; AART124, n = 124 independent adult (n = 59) and pediatric (n = 65) samples from 107 adult (n = 52) and pediatric (n = 55) patients (for independent kSORT validation); AART100, n = 100 adult (n = 78) and pediatric (n = 22) samples from 96 adult (n = 75, of which 21 patients were also included in AART143) and pediatric (n = 21) patients for final kSORT assay lock and clinical translation. For kSORT QPCR analyses, longitudinal samples were included in AART191: n = 191 adult (n = 94) and pediatric (n = 97) serial samples from 98 adult (n = 58) and pediatric (n = 40) patients for kSORT prediction. Detailed demographic patient information is shown in Table 1; detailed sample and patient flow is outlined in Table S2.

Figure 2. Training of a 17-gene model (kSORT) for acute rejection classification in 143 adult samples from real-life settings.

17 genes were used to classify 143 adult blood samples from four different sites into AR and No-AR by plsDA (threshold for AR prediction Θ was set at Θ = 50%). (A) AUC for AR in the training set was 0.94 (95% CI 0.91–0.98). (B) Predicted probabilities of AR were significantly higher in AR samples versus No-AR samples for each collection site and did not reach the threshold for AR prediction in the No-AR samples (predicted probability threshold Θ; Θ≥50%). Mean predicted probabilities for AR were 98.6%, 75.9%, 86.8%, and 77.2% for CPMC, Emory, UPMC, and UCLA, respectively. Mean predicted probabilities for No-AR were 13.7%, 19.4%, 11.3%, and 12.8% for CPMC, Emory, UPMC, and UCLA, respectively. Graphs show mean predicted AR probabilities in percent plus standard error of mean; p-values were calculated by two-sided Student's t test with Welch correction in case of unequal variances.

Figure 3. Validation of kSORT in 124 independent samples across different ages and settings.

(A) Independent validation of kSORT in 124 adult and pediatric AR and No-AR blood samples using the fixed plsDA model on the Fluidigm platform. 22 out of 23 AR samples were correctly classified as AR (red bars), and 100 out of 101 No-AR samples were correctly classified as No-AR (green bars). Shown are individual predicted AR probabilities (percent) grouped by patient age (adult, ≥20 y; pediatric, p<0.001) included in the AART124 group (shown are mean predicted AR probabilities in percent with standard error of mean; two-sided Student's t test with Welch correction was applied to calculate p-values). (C) ROC analyses demonstrated high sensitivity and specificity for AR classification by the 17 genes.

Figure 4. Evaluation of kSORT to predict acute rejection in 191 serially collected samples.

191 blood samples serially collected within 6 mo before (pre) or after (post) biopsy-confirmed AR were evaluated by kSORT. Frequencies of samples predicted as AR (red) or predicted as No-AR (green) were compared between sample collection time points (>3 mo prior to AR biopsy, n = 30; 0–3 mo prior to AR biopsy, n = 35; at AR biopsy, n = 74; 0–3 mo after AR biopsy, n = 31; >3 mo after AR biopsy, n = 21; and at No-AR/stable time points, n = 216). 62.86% of samples collected 0–3 mo (1.15±0.90 mo) prior to the AR biopsy had high probabilities for AR predicted by kSORT (96.36±0.08). High probabilities for AR persisted in 51.6% of samples collected 0–3 mo post-AR (94.60%±0.14); in comparison, 83.8% of the No-AR samples were always predicted as No-AR (8.20%±0.12). Mean AR scores were significantly different between pre-AR samples (0–3 mo) and No-AR/stable samples, as well as between AR samples and No-AR/stable samples.

Figure 5. Development of kSAS and the kSORT assay.

kSAS was developed to provide individual sample AR risk scores and AR risk categories. (A) Expression values of the 17-gene kSORT model in unknown samples were correlated to corresponding AR and No-AR reference values (centroids) by Pearson correlation. (B) For kSORT assay development, QPCR data from 100 samples were divided into training (n = 32) and independent validation sets (n = 68). (C) 13 12-gene models from the 17-gene kSORT model generated numerically aggregated AR risk scores for each sample and categorized them into three groups: high risk for AR (aggregated AR risk score ≥9), low risk for AR (aggregated AR risk score ≤−9), and indeterminate (aggregated AR risk score <9 and >−9) category.

Figure 6. Performance of the kSORT assay.

(A) kSORT score and classification category were calculated for each sample: the kSORT assay correctly classified 36 out of 39 AR samples as high risk for AR (red bars, 92.3%; risk score ≥9) and 43 out of 46 No-AR samples as low risk for AR (green bars, 93.5%, risk score ≤−9) across four different sample collection sites, and adult (≥20 y) versus pediatric (−9). (B) Mean aggregated kSORT scores (error bars give standard error of mean) were significantly higher in all true AR samples (10.2±7.2) than in all true No-AR samples (−8.4%±8.2) in the AART100 cohort by two-sided Student's t test. (C) ROC analysis demonstrated high sensitivity and specificity for the kSORT assay in the AART100 cohort.