Combining Blood Gene Expression and Cellfree DNA to Diagnose Subclinical Rejection in Kidney Transplant Recipients

Abstract

Background and objectives: Subclinical acute rejection is associated with poor outcomes in kidney transplant recipients. As an alternative to surveillance biopsies, noninvasive screening has been established with a blood gene expression profile. Donor-derived cellfree DNA (cfDNA) has been used to detect rejection in patients with allograft dysfunction but not tested extensively in stable patients. We hypothesized that we could complement noninvasive diagnostic performance for subclinical rejection by combining a donor-derived cfDNA and a gene expression profile assay.

Design, setting, participants, & measurements: We performed a post hoc analysis of simultaneous blood gene expression profile and donor-derived cfDNA assays in 428 samples paired with surveillance biopsies from 208 subjects enrolled in an observational clinical trial (Clinical Trials in Organ Transplantation-08). Assay results were analyzed as binary variables, and then, their continuous scores were combined using logistic regression. The performance of each assay alone and in combination was compared.

Results: For diagnosing subclinical rejection, the gene expression profile demonstrated a negative predictive value of 82%, a positive predictive value of 47%, a balanced accuracy of 64%, and an area under the receiver operating curve of 0.75. The donor-derived cfDNA assay showed similar negative predictive value (84%), positive predictive value (56%), balanced accuracy (68%), and area under the receiver operating curve (0.72). When both assays were negative, negative predictive value increased to 88%. When both assays were positive, positive predictive value increased to 81%. Combining assays using multivariable logistic regression, area under the receiver operating curve was 0.81, significantly higher than the gene expression profile (P<0.001) or donor-derived cfDNA alone (P=0.006). Notably, when cases were separated on the basis of rejection type, the gene expression profile was significantly better at detecting cellular rejection (area under the receiver operating curve, 0.80 versus 0.62; P=0.001), whereas the donor-derived cfDNA was significantly better at detecting antibody-mediated rejection (area under the receiver operating curve, 0.84 versus 0.71; P=0.003).

Conclusions: A combination of blood-based biomarkers can improve detection and provide less invasive monitoring for subclinical rejection. In this study, the gene expression profile detected more cellular rejection, whereas donor-derived cfDNA detected more antibody-mediated rejection.

Keywords: cellfree nucleic acid; diagnostic tests; gene expression; graft rejection; kidney transplantation; rejection; routine.

Graphical abstract

Figure 1.

CONSORT diagram illustrating the number of patients and the samples available for analysis on the basis of inclusion and exclusion criteria in addition to sample availability. CONSORT, Consolidated Standards of Reporting Trials; cfDNA, cellfree DNA; ci, interstitial fibrosis; ct, tubular atrophy; CTOT-08, Clinical Trials in Organ Transplantation-08.

Figure 2.

Differential performance of the gene expression profile and donor-derived cfDNA based on rejection type (acute cellular versus acute antibody-mediated rejection). (A) AUROC of the gene expression profile only for distinguishing acute cellular rejection versus no rejection. (B) AUROC of donor-derived cfDNA only for distinguishing acute cellular rejection versus no rejection. (C) AUROC of the gene expression profile only for antibody-mediated rejection versus no rejection. (D) AUROC of donor-derived cfDNA only for antibody-mediated rejection versus no rejection. AUROC, area under the receiver operating curve; dd-cfDNA, donor-derived cellfree DNA.

Figure 3.

Gene expression profile and donor-derived cellfree DNA preferentially detect different types of rejection. (A) All cases of subclinical rejection versus no rejection by biopsy and assay result. (B) All cases of subclinical rejection by rejection type and assay result. (A) Summary of the gene expression profile and donor-derived cfDNA performance at sample levels. Of 428 samples, the subclinical rejection group (n=103) consists of gene expression profile alone positive (n=23), donor-derived cfDNA alone positive (n=27), both gene expression profile and donor-derived cfDNA positive (n=21), and both gene expression profile and donor-derived cfDNA negative (n=32). Of the normal biopsies (n=325), both tests were negative for n=242, both were positive for n=5, gene expression profile alone was positive for n=45, and donor-derived cfDNA alone was positive for n=33. (B) Of 103 subclinical rejection cases, they are divided by histology phenotypes into antibody-mediated rejection alone (n=42), acute cellular rejection alone (n=38), and combined antibody-mediated rejection plus acute cellular rejection (n=23) cases, with breakdown of acute cellular rejection Banff grade as shown. The numbers demonstrate the true positives and false negatives with each assay. Although some overlap exists, the two assays tend to detect different types of rejection. 1A, Banff 1A acute cellular rejection; 1B, Banff 1B acute cellular rejection; 2A, Banff 2A acute cellular rejection; BL, borderline cellular rejection.

Figure 4.

Performance metrics of individual gene expression profile and donor-derived cfDNA assays compared with the logistic regression model with continuous variables for combined gene expression profile and donor-derived cfDNA to distinguish subclinical rejection versus no rejection. (A) AUROC of the gene expression profile only for subclinical rejection versus no rejection. (B) AUROC of the combined gene expression profile and donor-derived cfDNA performance on the CTOT-08 cohort (training set) by the multivariable logistic regression model using the continuous score output of both tests. (C) AUROC of donor-derived cfDNA only for subclinical rejection versus no rejection. (D) AUROC of an external validation with an independent cohort (n=105 samples) by the multivariable logistic regression model using the continuous score output of both tests. AUC, area under the curve.