Identification of a peripheral blood transcriptional biomarker panel associated with operational renal allograft tolerance

Abstract

Long-term allograft survival generally requires lifelong immunosuppression (IS). Rarely, recipients display spontaneous "operational tolerance" with stable graft function in the absence of IS. The lack of biological markers of this phenomenon precludes identification of potentially tolerant patients in which IS could be tapered and hinders the development of new tolerance-inducing strategies. The objective of this study was to identify minimally invasive blood biomarkers for operational tolerance and use these biomarkers to determine the frequency of this state in immunosuppressed patients with stable graft function. Blood gene expression profiles from 75 renal-transplant patient cohorts (operational tolerance/acute and chronic rejection/stable graft function on IS) and 16 healthy individuals were analyzed. A subset of samples was used for microarray analysis where three-class comparison of the different groups of patients identified a "tolerant footprint" of 49 genes. These biomarkers were applied for prediction of operational tolerance by microarray and real-time PCR in independent test groups. Thirty-three of 49 genes correctly segregated tolerance and chronic rejection phenotypes with 99% and 86% specificity. The signature is shared with 1 of 12 and 5 of 10 stable patients on triple IS and low-dose steroid monotherapy, respectively. The gene signature suggests a pattern of reduced costimulatory signaling, immune quiescence, apoptosis, and memory T cell responses. This study identifies in the blood of kidney recipients a set of genes associated with operational tolerance that may have utility as a minimally invasive monitoring tool for guiding IS titration. Further validation of this tool for safe IS minimization in prospective clinical trials is warranted.

Conflict of interest statement

The authors declare no conflict of interest.

Figures

Fig. 1.

Identification of “tolerance genes” in training- and test-set patient samples. (A) Tolerance prediction by two-class comparison of TOL (T) and CR (C). A cross-validated comparison of a training set of 5 TOL (green bars), 11 CR patients (blue bars), and 6 additional TOL-Test (TT) patients show a >80% fit to phenotype for most samples across this selected gene set. (B) Tolerance prediction by two-class comparison of TOL and N. A cross-validated comparison of a training set of five TOL (green bars) and eight N (yellow bars) blood samples. (C) Tolerance prediction by three-class comparison of TOL, CR, and N performs best for tolerance. A cross-validated comparison of a training set of 5 TOL, 11 CR, and 8 N blood samples by using three-class prediction identifies a subset of 49 known unique genes (59 transcripts). In the training set, class prediction scores for the CR (blue bars) and N (yellow bars) groups are weaker in three-class over two-class analysis. For the tolerance phenotype (green bars), consistently weak class prediction is seen for T5 by three-class analysis, as also seen by two-class analysis, but >90% tolerance prediction scores come up for most of the TOL and TOL-Test samples. Given the strength of this gene set for tolerance prediction, MIS and STA (S1–12) patients were also typed by PAM for their best fit to TOL, N, or CR class. Patients MIS1, MIS3, MIS4, MIS6, MIS8, and STA9 had very strong expression signatures that match operational tolerance. (D Right) Venn diagram of statistically significant genes (q < 0.05%) by SAM for all patient groups, relative to chronic rejection, using two-class comparison with CR (n = 11) as the reference group. More genes distinguish CR from TOL (n = 893) and AR (n = 982) than either MIS (n = 297) or N (n = 249). (D Left) Relative expression of the top 100 transcripts discriminating AR, CR, N, and TOL patients by PAM four-class analysis. The majority of these transcripts are expressed in highest abundance in TOL and lowest abundance in CR, suggesting that these two patient groups are the most distinct. A dendrogram of the relatedness of all patient samples across 11 K genes is shown in SI Fig. 3.