Omics Biomarkers for Prediction of Rejection—A Thermometer for Alloimmune Activation

The good short- and long-term outcomes of liver transplantation are largely attributed to the ability to control alloimmune response with effective immunosuppression. The standard practice of management of immunosuppression therapy (IST) is protocolized, aimed at predetermine drug levels, with minimal adjustments over the lifetime of the recipients. While successful, this approach is associated with the consequences of high burden of immunosuppression, leading to an increased incidence of infection and malignancies and significant comorbidities that are related to immunosuppressive drug-related side effects, including progressive kidney injury and cardiometabolic complications. There is a need to improve IST management aiming at minimizing and personalizing immunosuppression. An article published by Levitsky et al¹ in this issue proposes a gene expression signature that assesses alloimmune activation by predicting rejection and, therefore, may be used for adjustment of immunosuppression.

Recent studies described outcomes of minimization, or complete withdrawal, of IST after liver transplantation in large prospective trials that involved staged, clinically guided IST withdrawal in adult and pediatric recipients.^2-6 In these trials, “clinically guided” is translated to empiric, gradual reduction of IST dose over the course of the study following liver function tests. Recipients were recruited for withdrawal at a median of 10.5 y (adult) and 8.5 y (pediatric) posttransplant and achieved a remarkable 40%–60% of operational tolerance. Early withdrawal starting 1 y after transplantation demonstrated that only 13% of the recipients developed operational tolerance. However, many recipients tolerated significant minimization of calcineurin inhibitors-based IST.⁴ In the nontolerant recipients, clinically guided minimization of immunosuppression invariably resulted in acute rejection (ACR) and was not very informative for the personalized needs of the individual recipient. Ideally, minimization and personalizing IST should be guided by biomarkers, which are informative of the status of alloimmune activity and allograft injury. These biomarkers should identify recipients who can tolerate minimization and detect early alloimmune injury to the allograft before development of clinical liver allograft injury. Preferably, they should guide management of IST irrespective of the specific drug combination that is used for the individual recipient.

Previous studies have tested the potentials of noninvasive genomic markers, including blood mRNA, serum microRNA (miRNA) profiles, and cell-free DNA (cfDNA) for the diagnosis and prediction of ACR. The mRNA assays are performed using blood cells, hypothesizing that allograft dysfunction is reflected with changes in the arrangement or state of activation of cells of the immune system. In contrast, miRNA and cfDNA are assayed in the plasma or serum, suggesting that immune-mediated injury is associated with a leak of molecular byproducts to the periphery. Noninvasive diagnosis of rejection studies the differences in the expression of these biomarkers at time of allograft dysfunction and when biopsy discriminates rejection from no rejection event. In contrast, prediction of rejection requiring the collection of blood samples at predetermined intervals before the development of allograft dysfunction and assessing the differential regulation of these biomarkers before the development of allograft dysfunction.

A recent prospective randomized Immune Tolerance Network clinical trial examined the upregulation of serum miRNA in predicting and diagnosing allograft rejection when immunosuppression is minimized. The study was designed to assess the safety of minimizing immunosuppression while searching for molecular profiles that are diagnostic and predictive of ACR.⁷ In this study, the final parsimonious ACR diagnostic test included 2 miRNAs, hsa-miR-483-3p and hsa-miR-885-5p, that differentiated ACR from non-ACR (area under the curve was 89.5% [95% confidence interval, 82%-96%], with 83.8% sensitivity, 87.1% specificity, 72% positive predictive value, and 93% negative predictive value). Interactive pathways analyses identify these miRNAs’ biological mechanisms underpinning ACR. Moreover, the trajectory of our ACR diagnostic test demonstrates that molecular events initiating pathways of ACR can be observed 2–3 wk before clinical allograft injury and that intrapatient trajectory of ACR-associated miRNAs shows more significant, larger fold changes in miRNAs ACR profile. Measurements of hepatocyte-specific cfDNA methylation markers demonstrate elevation before and at the time of clinical rejection, validating the sensitivity of this biomarker for the detection of early allograft injury.⁸

The approach suggested by Levitsky et al advocates for blood-based mRNA profiles for the diagnosis and prediction of rejection. The study includes 2 cohorts; the majority of the recipients are patients from Northwestern Department of Surgery, Division of Transplantation, University (NU) who underwent for-cause liver biopsy or were otherwise stable. A smaller cohort was from the National Institutes of Health Clinical Trials in Organ Transplantation-14 (CTOT-14) prospective study in which recipients had long-term follow-up starting from the transplant event. The patients had blood withdrawal at predetermined intervals and at the time of for-cause liver biopsy, allowing to establish a predictive gene expression signature f Department of Surgery, Division of Transplantation, or ACR. In a previous publication, the authors used the NU samples as the primary training set for biomarker discovery and the CTOT-14 samples as the validation set, generating a 36-probe classifier that distinguishes ACR from non-ACR.⁹ The study demonstrated probability score line slopes that were positive in preceding ACR and negative for nonrejecting stable recipients and those with non-ACR allograft dysfunction, up to 100 d from a biopsy event. In the current publication, the authors used the same patient population; however, NU and CTOT-14 recipients were merged and randomly split into training and testing sets. This modeling generated a 59-probe classifier that was associated with some improvement of the previous predictive signature. Canonical pathway analysis demonstrated that nearly half of the significant pathways were associated with immune responses and liver-related function, including allograft rejection signaling. The ability to define peripheral blood mRNA predictive rejection profile is remarkable but must be interpreted with caution. Rejection episodes were observed in a small number of recipients (n = 14) for whom multiple prerejection samples were available, allowing to generate predictive signature. The fact that 2 different classifiers are reported for the same patient population demonstrates the potential for unintentional errors when training and testing sets are resampled. Moreover, the arrays and software are often changed, and complex multiprobe signature may be impacted by the changes.

Noninvasive omics diagnostic biomarkers are presenting powerful diagnostic technology in the transplant setting. Once validated, they should be tested in prospective randomized clinical trial, aiming to determine whether they are able to function as a “thermometer” of alloimmune activation, providing the opportunities for personalized immunosuppression.