Visual Abstract
Abstract
Background In kidney transplant recipients, surveillance biopsies can reveal, despite stable graft function, histologic features of acute rejection and borderline changes that are associated with undesirable graft outcomes. Noninvasive biomarkers of subclinical acute rejection are needed to avoid the risks and costs associated with repeated biopsies.
Methods We examined subclinical histologic and functional changes in kidney transplant recipients from the prospective Genomics of Chronic Allograft Rejection (GoCAR) study who underwent surveillance biopsies over 2 years, identifying those with subclinical or borderline acute cellular rejection (ACR) at 3 months (ACR-3) post-transplant. We performed RNA sequencing on whole blood collected from 88 individuals at the time of 3-month surveillance biopsy to identify transcripts associated with ACR-3, developed a novel sequencing-based targeted expression assay, and validated this gene signature in an independent cohort.
Results Study participants with ACR-3 had significantly higher risk than those without ACR-3 of subsequent clinical acute rejection at 12 and 24 months, faster decline in graft function, and decreased graft survival in adjusted Cox analysis. We identified a 17-gene signature in peripheral blood that accurately diagnosed ACR-3, and validated it using microarray expression profiles of blood samples from 65 transplant recipients in the GoCAR cohort and three public microarray datasets. In an independent cohort of 110 transplant recipients, tests of the targeted expression assay on the basis of the 17-gene set showed that it identified individuals at higher risk of ongoing acute rejection and future graft loss.
Conclusions Our targeted expression assay enabled noninvasive diagnosis of subclinical acute rejection and inflammation in the graft and may represent a useful tool to risk-stratify kidney transplant recipients.
Kidney transplantation is the therapy of choice for ESRD. Although short-term allograft outcomes including clinical acute rejection episodes (i.e., occurring in the presence of graft dysfunction) have declined over past decades, proportionate improvement in long-term allograft survival remains unrealized.1–3 The rates of clinically detected episodes of acute rejection within the first year in the modern tacrolimus/mycophenolate era are <10% among adult kidney recipients in the United States.2 The effect of these episodes on long-term graft survival is variable, depending on the severity of episode, time from transplantation, and effective treatment of these episodes with return of creatinine to prerejection baseline.4
Graft inflammation, however, may also occur initially in the absence of graft function decline. The wide prevalence of subclinical rejection, i.e., lymphocytic tubulitis and interstitial inflammation in early surveillance biopsy specimens, and its effect on long-term allograft histology, function, and survival has been appreciated only recently.5–10 The effect of milder subclinical inflammation, i.e., suspicious or borderline lesions, on allograft outcomes is even less characterized.11
In current clinical practice, the diagnosis of either clinical or subclinical allograft rejection requires a biopsy, a procedure burdened by clinical risks and costs. To obviate these issues, prior studies have tested noninvasive profiling of urinary proteins12,13 and blood transcriptomic signatures,14–16 but results have been inconsistent so far. An assay that could be used in clinical practice for the purpose of accurately diagnosing subclinical rejection offers the potential to identify and treat underlying subclinical rejection without the need for a biopsy.
Herein, we examined the incidence of subclinical rejection and borderline lesions over time in the Genomics of Chronic Allograft Rejection study (GoCAR), a prospective, multicenter center study in which kidney transplant recipients underwent serial surveillance biopsies. We determined the effect of subclinical graft inflammation on allograft function and survival, and developed a clinically applicable assay that detects subclinical acute rejection by measuring the transcriptome in peripheral blood.
Methods
Patients
The study included participants of the GoCAR study and kidney transplant recipients prospectively followed-up at the University Hospitals Leuven, Leuven, Belgium. The GoCAR study is a prospective, multicenter study (United States and Australia) aimed at investigating the genetics and genomics associated with the development of allograft rejection or injury in kidney transplant recipients. Patients underwent surveillance biopsies pretransplant (before implantation), and at 3, 6, 12, and 24 months after transplant. Patients were followed up for at least 5 years or until death. The details of patient enrollment criteria and study design have been previously described.17,18 All biopsy specimens were reported for Banff component scores by a central core laboratory at Massachusetts General Hospital. The diagnosis of acute cellular rejection (ACR) at 3 months (ACR-3) was made by applying Banff 2013 Classification19 on all clinically indicated and surveillance biopsy samples and included borderline subclinical rejection. Donor-specific anti-HLA antibodies (DSAs) were measured before transplant and clinically indicated after transplant by Luminex (ThermoFisher). Mean fluorescence intensity readings >1000 were taken as positive. The surveillance biopsy specimens in the patients in the Belgian cohort were taken at the same time points (3, 6, 12, and 24 months) as for the GoCAR cohort. Highly sensitized patients requiring desensitization were excluded from the GoCAR study and not present in the Belgium cohort. We used United Network for Organ Sharing and The Australia and New Zealand Dialysis and Transplant Registry databases to determine long-term outcomes for the GoCAR enrollees.
Genomic Experiments and Data Analysis
The details regarding genomic experiments (RNA sequencing, microarray, and targeted RNA expression [TREx] assay) are provided in the Supplemental Material and the data analysis workflow is depicted in Supplemental Figure 1. Briefly, mRNA sequencing (Illumina HiSeq4000 sequencer) was performed on 88 samples obtained at 3 months post-transplant in the GoCAR cohort as discovery set for identification of gene signatures associated with ACR-3. After read quality control, mapping, and normalization steps on the raw sequencing reads, the expression data were compared between ACR-3 and non-ACR at 3 months (NACR-3), with both induction therapy and deceased donor as confounders, to identify differentially expressed genes (DEGs) with a P value <0.05 using unpaired LIMMA test,20 a linear model to assess differential gene expression in the context of multiple variables. The DEGs were then subjected to enrichment analysis for canonical pathway, gene ontology, and immune cell types to identify classes or groups of genes that were associated with ACR.
Next, a more focused gene set associated with ACR-3 was identified from the DEGs using a randomization approach described previously.18. An optimal gene set with the highest AUC (area under the receiver operating characteristic curve) score for diagnosis of ACR-3 was then determined by fitting a penalized logistic regression model on the expression data of the focus gene set following a 5000-iterations methodology18 (Supplemental Material). The AUC for the final gene set was crossvalidated using a leave-one-out crossvalidation method to avoid potential over-fitting. The gene set was then validated for the diagnosis of ACR-3 using the microarray data from the GoCAR study (n=65; 26 overlapping with the RNAseq cohort, Supplemental Material) and three public blood microarray datasets for clinical acute rejection (GSE14346,21 GSE15296,22 and GSE5008419). GSE14346 is an expression dataset of 75 patients with 38 clinical acute rejections.21 GSE15296 is an expression dataset for 75 patients receiving kidney transplant with 51 clinical acute rejections including 12 borderline cases.22 GSE5008 is an expression dataset of 42 patients receiving kidney transplant including 28 cases with DSAs and clinical acute rejection. All three datasets were generated on the Affymetrix platform.19
The novel sequencing-based targeted expression (TREx) analysis technology (Illumina) was used to develop a molecular assay for the 17-gene set and 12 house-keeping genes with the potential for application in the diagnosis of acute rejection in clinical practice (experimental details in the Supplemental Material). The assay first evaluated sensitivity and reproducibility with universal reference RNA, brain reference RNA, and RNA extracted from peripheral blood samples from GoCAR. Of the original 127 GoCAR samples used in the RNAseq and microarray cohorts, only 113 had RNA that was of sufficient quality to be used for the TREx training set. A separate 64 samples from GoCAR patients combined with 46 patient samples from the Belgian cohort, with similar demographic and clinical characteristics in both cohorts (n=110, Supplemental Table 1), composed the TREx independent testing set. The penalized logistic regression model was built on the expression values of 17 genes from the training dataset and tertile cutoffs on the basis of the probability scores were defined to stratify patient risk of ACR into three groups: high, intermediate, and low. The model from the training set was then applied to the independent testing set to compute the probability risk score and positive predictive value (PPV) and negative predictive value (NPV) on the basis of the tertile cutoffs in the second set of patients.
The later clinical end points of acute rejection after 3 months, fibrosis (CADI score), and the risk of graft loss were evaluated in the high-, intermediate-, and low-risk groups, given the association between ACR-3 and these end points in GoCAR.
The RNA sequencing and microarray data were deposited in the National Center for Biotechnology Information Gene Expression Omnibus database (GSE120398).
Statistical Analyses
Descriptive statistics (means and SD) were used to summarize the baseline characteristics of the ACR-3 and NACR-3 cohorts, and were compared using the chi-squared test and Fisher’s exact test. The diagnosis with selected significant demographic or clinical factors was estimated with logistic regression and the AUC was calculated. Centrally reported Banff/CADI component scores were used for histologic comparisons. For composite scores utilized in analysis (i.e., CADI, Ci+Ct), we imputed CADI scores=8 and Ci+Ct=6 (highest scores in biopsy specimens at 12 months) for allografts that were lost before 12- or 24-month biopsies within each group. Univariate comparisons of continuous variables were done using unpaired t test (Mann–Whitney test for corresponding nonparametric analysis). Normality of sample distributions was confirmed using Kolmogorov–Smirnov and Shapiro–Wilk tests. Kaplan–Meier survival curves were calculated with graft loss (all-cause and death-censored) as outcome. Cox proportional hazard models were used for multivariable survival associations, including donor and recipient demographics on the basis of a priori–determined clinical importance for outcome of interest (i.e., graft survival). Only patients with 3-month biopsy samples were included in the Kaplan–Meier or Cox regression survival analyses; hence, all cases had survived past 3 months. Time was determined from the day of transplant to the graft loss. All statistical analyses for demographic and clinical data were completed using SPSS Statistics V23 (IBM Analytics) and GraphPad Prism V6 (La Jolla, CA). Statistical significance was considered with two-tailed P<0.05.
Results
Study Cohorts
One hundred and ninety-one patients from the GoCAR study that had peripheral blood RNA concurrent with 3-month kidney biopsy were included in this study (Figure 1).18 From this group, 127 patients were randomly assigned to be used for identification of a peripheral blood gene signature associated with subclinical acute rejection using RNA sequencing (n=88) or microarray (n=65). Twenty-six patients underwent both RNAseq and microarray analyses, allowing correlation of gene expression between the two technologies (Supplemental Figure 2, Supplemental Material). Hence, 127 patients were used as the training set for the development of a sequence-based targeted expression assay (TREx) which was then validated on an independent cohort of 110 patients (64 GoCAR patients and 46 patients from the biobank of the University Hospitals Leuven, Belgium) (Figure 1, Supplemental Table 1).
GoCAR (n=191) and Belgian (n=46) cohorts were used in this study. Of 191 patients, 129 were randomly selected for transcriptomic analysis using RNA sequencing (n=88, discovery set) and microarray (n=65, validation) for identification of a peripheral blood gene signature to diagnose subclinical acute rejection. Of note, 26 patients were overlapped between the RNAseq and the microarray cohorts for correlation analysis of gene expression between the two technologies. The sequencing-based targeted expression (TREx) assay was developed on the gene set identified from transcriptomic analysis. In TREx assay, 113 of 127 patients from the transcriptomic analysis cohort were used for the training set to build the penalized logistic regression statistical model which was validated on an independent testing cohort of 110 patients (64 GoCAR patients and 46 patients from the Belgian cohort).
The demographic and clinical characteristics of the subclinical ACR-3 and NACR-3 groups show similar graft function at 3-month surveillance biopsy (Table 1). The only significant differences between ACR-3 and NACR-3 were donor age (P=0.01), induction therapy (P=0.01), and m3 creatinine (P=0.04) (Table 1); however, these factors combined were unable to accurately diagnose acute rejection (AUC=0.720 and crossvalidated AUC [cAUC]=0.672, Supplemental Figure 3A).
Demographic characteristics of ACR-3 and NACR-3 group patients in GoCAR cohort
Subclinical ACR is associated with later allograft fibrosis, function decline, and loss. To study the natural history of allografts with subclinical cellular rejection or borderline changes, we examined subclinical longitudinal histologic and functional changes in the ACR-3 and NACR-3 GoCAR cohorts with 3-, 12-, and 24-month surveillance biopsies. We excluded two patients with BK nephropathy on 3-month biopsy from outcome analysis. The ACR-3 group had significantly higher CADI scores at 3-, 12-, and 24-month surveillance biopsies (CADI-3, CADI-12, and CADI-24) compared with NACR-3 (Figure 2, A and B). When analyzed separately, patients with borderline ACR (BACR-3) alone also had increased CADI scores compared with NACR-3 at 3-, 12-, and 24-month surveillance biopsies (Figure 2, A and B). Interestingly, patients with increased i or i+t scores at 3-month biopsies were also more likely to have higher i and t scores on biopsy at 12 and 24 months, suggesting a persistent inflammatory phenotype in these patients (Table 2). Consistent with this, the presence of ACR-3 was associated with a significantly higher risk of acute rejection at 12 (odds ratio [OR], 7.09; 95% confidence interval, 2.82 to 17.81; P<0.01) and 24 months (OR, 3.96; 95% confidence interval, 1.20 to 13.02; P=0.02) compared with NACR-3 (Figure 2, C and D). In biopsy specimens with ACR-3, chronic injury was already increased at 3 months compared with NACR-3 (Figure 2A). However, in multivariable regression analysis, ACR-3 was associated with significantly higher CADI scores at 12 and 24 months even after adjustment for 3-month CADI and Ci+Ct scores (Supplemental Table 2, Table 2).
ACR-3/BACR-3 is associated with adverse outcomes compared with NACR-3. Line graphs compare CADI (A and B) Ci+Ct scores between ACR-3 (bold red line), BACR-3 (BACR at 3 months, dotted red line), and NACR-3 (no-ACR at 3 months, green line) on serial 3-, 12-, and 24-month surveillance biopsy specimens (line through median, whiskers=EM). Bar graphs compare ACR prevalence on (C) 12- and (D) 24-month biopsy specimens in ACR-3 and NACR-3 groups. These increases in ACR and CADI at 12 or 24 months are subclinical observations. (E) Kaplan–Meier curves compare adjusted death-censored survival of ACR-3 (green) and NACR-3 (blue) groups in the GoCAR cohort (see Supplemental Table 1B). *P<0.05; ***P<0.001.
CADI subscores in serial biopsy specimens
As part of GoCAR, all biopsy specimens were read by a central core with three experienced transplant pathologists. Forty-seven percent of the biopsy specimens in the GoCAR cohort were also read by the local pathologist, enabling comparison of scoring. Forty-eight percent of the BACR-3 cases identified by the core laboratory were classified as NACR locally. In contrast to the core ACR-3 diagnosis, the local ACR-3 diagnosis did not correlate with increased CADI at 12 or 24 months. This suggests that clinically meaningful subclinical ACR was missed by local reporting (Supplemental Table 3).
The incidence of de novo DSAs during 24-month follow-up was similar in ACR-3 and NACR-3 (Table 1). Eleven patients developed acute antibody-mediated rejection (ABMR), all of which were in the first 3 months, eight of them in the ACR-3 group, and three in NACR-3 (ACR-3 versus NACR-3, OR, 8.29; 95% confidence interval 1.89 to 50.55; Fisher’s P<0.01). Only one of these cases of ABMR was seen on the 3-month surveillance biopsy specimen (i.e., subclinical ABMR), whereas other cases occurred before the 3-month biopsy. Microvascular inflammation scores (g+ptc) and C4d staining were increased in ACR-3 biopsy specimens (Table 2). When the 11 ABMR cases were excluded, microvascular inflammation scores were still higher in ACR-3 versus NACR-3 in the remaining 178 patients. Furthermore, the ACR-3 group, excluding ABMR cases, had higher CADI-12 and CADI-24 scores and an increased risk of ACR-12 and ACR-24 when compared with NACR-3 (Supplemental Table 4). These data suggest that ACR-3 had increased histologic decline independent of preceding ABMR episodes. Because four of 37 ACR-3 biopsy specimens were C4d positive, meeting probable ABMR diagnosis despite absent DSA, and because g/ptc scores in the absence of DSA could result from TCMR lesions,23 we examined ACR-3 biopsy specimens with/without C4d. C4d-negative ACR-3 biopsy specimens had significantly higher g+ptc score than NACR-3 (Supplemental Figure 3B). These biopsy specimens likely represent microvascular inflammation associated with TCMR. These cases also had higher CADI-12 and CADI-24 scores and increased risk of ACR-12 and ACR-24 when compared with NACR-3 (Supplemental Table 5). These data suggest that ACR-3 alone is associated with adverse outcomes independent of antibody-mediated injury.
We compared the changes of eGFR from 3 (or 6) months to 12 (or 24) months between ACR-3 and NACR-3 groups (mean Δ eGFRs). ACR-3 recipients had greater declines in mean eGFR by 12 and 24 months compared with NACR-3 (Supplemental Figure 3C). Adjusted Cox models showed that, over a median (interquartile range) follow-up of 1713 (165–2793) days, ACR-3 was associated with an increased hazard of death-censored and all-cause allograft loss versus NACR-3 (Figure 2E, Supplemental Table 6 [death-censored graft loss], Supplemental Figure 3D [all-cause graft loss]).
Peripheral Blood Transcriptomic Signatures Are Associated with Subclinical Rejection
We next evaluated transcriptomic signatures associated with ACR-3 in peripheral blood taken from 88 patients (22 ACR-3 and 66 NACR-3) at the time of the biopsy. Data were analyzed according to the outline depicted in Supplemental Figure 1. Comparison of gene expression by LIMMA test20 on normalized data adjusted for confounders associated with ACR in the GoCAR cohort (induction therapy and deceased donor) identified 1115 DEGs (609 up- and 506 downregulated) associated with ACR-3 (P<0.05; Figure 3A). Gene ontology enrichment analysis revealed that upregulated genes were involved in transcriptional regulation and cell cycle, whereas downregulated genes were involved in transport and cytoskeleton organization processes (Figure 3B). Canonical and ingenuity pathway analysis showed that these dysregulated genes were involved in multiple pathways, including those related to extracellular matrix, cell cycle, TGF-β signaling, B cell receptor signaling, integrin signaling, Jak/Stat signaling, and leukocyte extravasation signaling (Supplemental Figure 4A). The immune response genes were enriched in the most downregulated genes ranked by fold-change (Supplemental Figure 4B). These findings are in keeping with the known immunologic processes involved in acute rejection.
Whole blood transcriptomic signatures of the patients at 3 months post-transplant are associated with subclinical acute rejection (ACR-3). (A) The volcano plot of DEGs between the recipients who developed or did not develop ACR-3. The x axis depicts the log2 ratio of gene expression and the y axis depicts the −log10 of LIMMA P test. The top up- or downregulated genes are labeled with boxes. (B) The bar chart of significant gene ontology function groups by enrichment analysis on DEGs. The bars represent −log10 P value of enrichment significance of gene pathways by Fisher exact test; the lengths of red and green bars represent the percentages of up- and downregulated genes, respectively. (C) Immune cell enrichment analysis of DEGs associated with ACR-3. The heatmap shows expression of DEGs that were significantly enriched for immune cell types in the ImmGene dataset. (D) The heatmap of enrichment P value (−log10 P) of immune cell–specific signatures in DEGs between ACR and non-ACR in GoCAR RNAseq, microarray, and three public datasets (GSE14346, GSE15296, and GSE50084). AR, acute rejection; gd, gama delta; NK, natural killer.
Immune cell type enrichment analysis using immune cell profiles in the ImmGene database24 showed that, in addition to macrophage and natural killer cells, pro-B or pre-T cell genes were significantly associated with ACR-3 (Figure 3C), and this pattern was replicated in publicly available blood expression datasets of patients with acute rejection (Figure 3D).
A Peripheral Blood Gene Signature Accurately Diagnoses Subclinical Cellular Rejection
Of 1115 DEGs for subclinical acute rejection, we identified an optimal gene set for the diagnosis of ACR-3 using a combination of logistic regression and permutation-based approaches, as previously described.18 The discovery set (n=88) was randomly divided into two equal groups and LIMMA analysis was performed with 100 iterations to identify 240 transcripts (significant on ≥2 iterations) (see Methods section; Supplemental Table 7). A 17-gene set was then determined as an optimal set for the diagnosis of ACR-3 (AUC=0.980) with a leave-one-out crossvalidation (cAUC=0.833) (Figure 4A, Table 3).
The 17-gene set for diagnosis of ACR-3 was identified from GoCAR and validated in internal and external datasets. (A) The receiver operating characteristic (ROC) curve for diagnosis of ACR-3 with 17-gene set in GoCAR RNAseq discovery set (n=88; AUC=0.980, shown by black curve; leave-one-out cAUC=0.833, shown by blue curve). (B) The ROC curve for diagnosis of ACR-3 with 17-gene set in GoCAR microarray validation set (n=65, AUC/cAUC=1.000/0.802). (C) The ROC curve for diagnosis of ACR with 17-gene set in a public dataset (GSE14346: AUC/cAUC=0.959/0.818). (D) The ROC curve for diagnosis of ACR with 17-gene set in a public dataset (GSE15962: AUC/cAUC=0.988/0.832). (E) The ROC curve for diagnosis of ACR with 17-gene set in a public dataset set (GSE50084: AUC/cAUC=1.000/0.979).
Seventeen-gene set for 3-month ACR diagnosis
Diagnosis of ACR-3 using the 17-gene set was internally validated on our microarray cohort (n=65; AUC/cAUC=1.000/0.802, respectively; Figure 4B, Supplemental Table 8). In addition, it accurately diagnosed ACR in three public microarray datasets that used blood taken at the time of clinically indicated biopsies at varying times post-transplant (Figure 4, C–E; GSE14346: AUC/cAUC=0.959/0.818; AUC=1.000 GSE15296: AUC/cAUC=0.980/0.832; GSE50084: AUC/cAUC=1.000/0.979). These data show that the differential expression of our 17-gene set is identifiable in peripheral blood during subclinical or clinical ACR.
Development of TREx Assay for Diagnosis of ACR
Using sequencing-based TREx analysis technology, we developed a molecular assay to measure expression of the 17 genes on whole blood RNA that demonstrated high sensitivity and reproducibility when evaluated with human universal reference, brain reference RNA, and RNA from our clinical samples (Supplemental Material, Supplemental Figures 1, 5, and 6, A–E).
Next, we performed the TREx assay on 113 of the original 127 samples from the transcriptomic analysis to build the statistic diagnostic model on the basis of the expression values of the 17 genes (TREx training set; Figure 1, Supplemental Figure 1). Fourteen samples were excluded because of poor QC. Expression of the 17 genes by TREx validated our original findings using RNAseq and microarray, clearly differentiating between ACR-3 and NACR-3 (Figure 5A), with an AUC of 0.830 (Figure 5B). Tertile probability cutoffs (0.146 and 0.463) were defined to stratify the patients into three groups (high, intermediate, and low risk) with NPV=0.98 and PPV=0.79 (Figure 5C).
ACR-3 diagnosis with 17-gene set was validated by TREx assay. (A) The heatmap of expression of 17-gene set in TREx training set (n=113); ACR-3 or NACR-3 cases ordered by risk scores are on the left or right of the vertical yellow line, respectively. The up- or downregulated genes in ACR-3 are above or below the horizontal yellow line, respectively. (B) The ROC curve for diagnosis of ACR-3 with 17-gene set in the training set (n=113, AUC=0.830). (C) The dot plot of the probability risk scores for the patients in the training set (n=113, PPV=0.79, NPV=0.98 at tertile cutoffs). (D) The dot plot of the probability risk scores for the patients in the testing set (n=110, PPV=0.73, NPV=0.89 at tertile cutoffs defined from the training set). (E) Kaplan–Meier curve of graft loss for the kidney transplant recipients stratified into two groups (high/intermediate and low probability risks) in TREx (n=223). (F) The Kaplan–Meier curve of graft loss with the kidney transplant recipients without ACR (NACR-3) stratified by intermediate or low probability risks in TREx cohort. Cum, cumulative.
Using the same analytic model derived on the training set, the 17-gene TREx assay was validated in the totally independent cohort of 110 subjects (64 independent GoCAR subjects+46 subjects from the Belgian cohort, Figure 1, Supplemental Figure 1; clinical epidemiologic data in Supplemental Material and Supplemental Table 1). The 17-gene TREx assay accurately diagnosed ACR-3 on the testing set with NPV=0.89 and PPV=0.73 using the tertile cutoffs (Figure 5D).
TREx Risk Profile Stratifies Risk for Late ACR and Graft Loss
Because ACR-3 correlated with graft loss in the GoCAR cohort, we determined whether the 17-gene TREx assay predicted subsequent ACR and graft loss using all GoCAR patients included in this study (n=177). Patients in the high/intermediate group had higher risk of subsequent subclinical ACR at 12 months (31.9% versus 13.6%; P=0.04) and 24 months (53.8% versus 32.1%; P=0.08) or of clinical ACR (Banff 1A or greater) at any time post-transplant (20.6% versus 6.1%; P<0.01) than the low-risk group (Supplemental Table 9). This was also significant when the intermediate group was considered alone, with subsequent ACR 1A or higher and ACR/BACR occurring more frequently than the low-risk group (14 of 85 versus four of 70; OR, 3.15; P=0.04, and 39 of 85 versus 20 of 70; OR, 2.12; P<0.01, respectively). ABMR episodes, nine of which occurred before the 3-month biopsy, and de novo DSAs were not significantly different between risk groups (Supplemental Table 9).
The clinical outcomes of GoCAR and the Belgian cohort used for TREx assay are summarized in Supplemental Table 10. Multivariable Cox models performed on the combined GoCAR/Belgian cohorts (n=223) demonstrated that patients in the high- or intermediate-risk groups had lower death-censored graft survival compared with the low-risk group (Figure 5E, Supplemental Table 11). Collectively, these data indicate that our TREx assay of the 17-gene set not only accurately differentiates those at high versus low risk of ACR-3, but also gives their ongoing risk for rejection and graft loss.
Despite the fact that 74.2% of the intermediate group were NACR-3, we observed that the group overall had a significantly higher risk of graft loss compared with the low-risk group (Supplemental Table 11). Stratified analysis comparing the intermediate-risk NACR-3 (I-NACR-3) with the low-risk NACR-3 (L-NACR-3) demonstrated that the I-NACR-3 and the L-NACR-3 had similar Banff subscores at 3 months, with the exception of the mm-scores (Supplemental Table 12). However, I-NACR-3 developed significantly higher CADI and Ci+Ct scores by 24 months (P=0.01 and <0.01, respectively; Supplemental Table 12) and had greater risk of graft loss than L-NACR-3 in adjusted Cox regression (Figure 5F, Supplemental Table 13). Of note, graft survival in I-NACR-3 did not differ significantly from I-ACR-3 (data not shown). These data demonstrate that biopsy specimens with NACR classified as intermediate are not simply misclassified, but rather have outcomes that are truly intermediate between the high- and low-risk groups.
Discussion
In this study, we used the extensively phenotyped GoCAR cohort to accurately identify ACR-3 and demonstrate its association with progressive allograft damage and functional decline, resulting in a higher risk for graft loss. We then used RNA sequencing of peripheral blood to identify a 17-gene set that strongly correlated with the presence of ACR-3. Using TREx technology with MiSEQ sequencer, a targeted sequencing platform approved for clinical application, we developed and validated the 17-gene assay, demonstrating that recipients stratified into high, intermediate, and low risk for ACR-3.
The current approach to immunosuppression in kidney transplantation is protocol-driven, with adjustments dictated by changes in serum creatinine—a poorly sensitive marker for graft damage. Surveillance biopsies could help define the intensity of the alloimmune response and guide immunosuppression accordingly, but they are costly, time-consuming, and burdened by a risk of potential complications. Our assay enables the identification of underlying inflammation, while serum creatinine is still within the normal range, allowing surveillance of the graft status without the need for a biopsy and before there is functional evidence of injury. Interestingly, it identified ACR in four patients with DGF in whom creatinine is not a functional marker of rejection. The 17-gene set has the potential to be useful to diagnose subclinical rejection at other time points, identifying clinical ACR in the publicly available data sets in which biopsy specimens were taken at multiple times post-transplantation. Therefore, these data provide the background for future studies testing the hypothesis that serial measurements of our assay may guide maintenance immunosuppression management more accurately than our current approach using creatinine.
Interestingly, even in the absence of ACR-3 in the intermediate group, the TREx-based risk stratification also associates with detection of inflammatory infiltrates in subsequent surveillance biopsy specimens, and graft function decline and loss. This may be explained by the fact that renal biopsy specimens represent 3-µm sections of the total core and may not capture a focal phenomenon like allograft rejection.25 Besides limitations in overall representation, the clinical utility of graft biopsy is limited by the subjectivity of biopsy reporting. These findings emphasize the potential clinical utility of the 17-gene set to avoid pitfalls in the reporting of renal biopsy specimens.
Although baseline anti-HLA antibodies, including DSAs and non-DSA, all of which were at low level, and de novo DSAs did not significantly associate with graft loss in our cohort, ABMR episodes and microvascular inflammation scores were all more common in ACR-3 versus NACR-3, implying the coexistence of endothelial injury with cellular inflammation. Of the ten episodes of ABMR, nine were clinical ABMR and occurred before the 3-month biopsy.
The DEGs associated with ACR-3 were predominantly from pathways not related to the immune response or lymphocyte activation, but rather cell repair, metabolism, and stress response pathways (Figure 3B), in contrast to previously reported expression of immune response–related genes in the graft biopsy specimens.18 This has previously been described by others26 and could potentially reflect migration of immune cells from the periphery to the allograft. However, immune cell analysis revealed enrichment for pro-B or pre-T cells or stem cells in the GoCAR cohort and public validation datasets with clinical acute rejection, suggesting active B and T cell proliferation, consistent with an active immune response in the setting of ACR.
Other studies have examined peripheral blood biomarkers to diagnose ACR.14,21,22,27,28 Initial data using donor-derived cell-free DNA demonstrated that levels correlated with clinically severe acute T cell–mediated rejection (≥Banff Ib) and ABMR; however, leakage of donor-derived cell-free DNA may lack specificity for rejection and instead reflect graft injury in general.29 Several studies have examined peripheral gene expression for the diagnosis of ACR; however, there are several pertinent differences compared with the data presented here. First, we used an unbiased approach through RNAseq profiling for gene selection, with both clinical and technical validation on multiple platforms. Other groups have taken a reductive approach, narrowing down the genes for consideration on the basis of those that are related to immune cell expression, and have focused on clinical acute rejection.14,21 Prior data have shown that immune response genes are elevated in the allograft during rejection episodes, but share minimal overlap with simultaneous gene signatures in periphery26; thus, by only focusing on immune response genes, the genes of highest expression and reflecting the greatest changes in expression are excluded from the analysis. This will also account for the lack of overlap between our gene set and the gene set in the study by Roedder et al. 14 By including all genes and building the diagnostic model with protocol biopsy specimens we have been able to define a unique signature that effectively identifies even borderline subclinical rejection, and predicts the risk for ongoing immune injury and graft loss. Second, validation on public cohorts demonstrated the ability of the gene set to identify clinical ACR at multiple time points. Third, we have technically validated our assay on multiple platforms and developed an accurate and reproducible clinically applicable assay (TREx) using MiSEQ sequencer, a sequencing platform which is approved by the Food and Drug Administration for clinical application. This is highly reproducible and provides absolute transcript measures. This contrasts significantly with other studies in which the methods used will be hard to implement on a larger scale, such as, microarray15 or PCR.21 Lastly, our assay performs with a high degree of accuracy even in validation cohorts with high NPVs and PPVs. This contrasts with a more recent paper using gene expression for diagnosis of ACR, in which a microarray assay with 57 genes reported by Friedwald and colleagues,15 representing a technical challenge for clinical implementation, performed poorly, with PPVs in 51% and 47% in two validation cohorts, and NPVs in the range of 70%.15 Our 17-gene set did not overlap with their 57 genes,15 which could be due to the following reasons: (1) Assay variation: we used RNA sequencing for biomarker discovery, whereas microarray was applied in their study. RNAseq allows absolute quantification of transcripts. Microarray allows relative quantification and is restricted to predefined probe sets. (2) We only investigated 3-month blood profiles, whereas they analyzed the expression profiles at various times from 4 to 24 months; overall gene expression profiles in blood were reported to change upon immune suppression administration post-transplant30.
In summary, our study highlights the negative effect of untreated early subclinical inflammation on subsequent histologic and functional decline in kidney allografts. We found a peripheral blood 17-gene set utilizing a novel TREx assay that accurately diagnoses subclinical rejection, including borderline lesions, and stratifies renal recipients according to those at high risk for histologic decline and allograft loss. Our assay offers the potential to be used as an immune-monitoring tool to guide the use of immunosuppression, with the ultimate goal of controlling subclinical intragraft inflammation and prolonging graft survival. However, these findings need further, larger studies to validate the clinical utility of this assay before it can be used as a substitute for renal transplant biopsies for the determination of clinical and subclinical rejection.
Disclosures
Dr. Zhang reports personal fees from RenalytixAI, outside the submitted work; in addition, Dr. Zhang has a patent ‘Method for identifying kidney allograft recipients at risk for chronic injury’ pending, a patent ‘Methods for diagnosing risk of renal allograft fibrosis and rejection (miRNA)’ pending, a patent ‘Method for diagnosing subclinical acute rejection by RNA sequencing analysis of a predictive gene set’ pending, and a patent ‘Pretransplant prediction of post-transplant acute rejection’ pending. Dr. Kuypers reports grants and personal fees from Astellas, outside the submitted work. Dr. Murphy reports personal fees from RenalytixAI, outside the submitted work; in addition, Dr. Murphy has a patent ‘Method for identifying kidney allograft recipients at risk for chronic injury’ pending, a patent ‘Methods for diagnosing risk of renal allograft fibrosis and rejection (miRNA)’ pending, a patent ‘Method for diagnosing subclinical acute rejection by RNA sequencing analysis of a predictive gene set’ pending, and a patent ‘Pretransplant prediction of post-transplant acute rejection’ pending. Dr. Menon acknowledges funding support from American Heart Association grant 15SDG25870018.
Supplemental Materials
This article contains the following supplemental material online at http://jasn.asnjournals.org/lookup/suppl/doi:10.1681/ASN.2018111098/-/DCSupplemental.
Supplemental Figure 1. Correlation between RNA sequencing and microarray data.
Supplemental Figure 2. Association of demographic and pathologic characteristics with clinical outcomes.
Supplemental Figure 3. Genomic data analysis workflow.
Supplemental Figure 4. Pathway and gene ontology enrichment analysis for DEGs associated with ACR-3.
Supplemental Figure 5. The procedure of development of TREx assay for 17-gene set.
Supplemental Figure 6. Development of TREx assay for 17-gene set.
Supplemental Table 1. Comparison of demographic statistics between GoCAR and Belgian dataset.
Supplemental Table 2. ACR-3 predicts CADI-12 and -24 independent of simultaneous chronic damage indices.
Supplemental Table 3. Comparison of local and central biopsy reports at 3-month biopsy.
Supplemental Table 4. Comparison of clinical characteristics between ACR-3 and NACR-3 without AMBR.
Supplemental Table 5. Comparison of clinical outcomes post 3 months between C4d-negative ACR-3 and NACR-3 groups.
Supplemental Table 6. ACR-3 independently predicts long-term allograft survival.
Supplemental Table 7. The list of 240 focus genes set.
Supplemental Table 8. Demographic characteristics of RNAseq and microarray cohorts in GoCAR cohort.
Supplemental Table 9. Frequency of anytime rejection episodes in TREx risk groups.
Supplemental Table 10. Summary of clinical events of TREx cohorts post kidney transplant.
Supplemental Table 11. TREx risk group status affects allograft survival.
Supplemental Table 12. Comparison of Banff scores between intermediate- and low-risk NACR-3 groups.
Supplemental Table 13. High-/intermediate-risk NACR-3 affects allograft survival.
Acknowledgments
We thank the Genomics Resources Core Facility at Weill Cornell Medical Center for sequencing and TREx experiments and Scientific Computing at the Icahn School of Medicine at Mount Sinai for providing computational resources.
This project is a substudy of the GoCAR study supported by National Institutes of Health grant 5U01AI070107-03.
Dr. Zhang, project leader, study design, analysis and interpretation of genomic data, and drafting of manuscript. Miss. Yi, genomic data analysis (major contribution) and interpretation. Dr. Keung, analysis and interpretation of clinical data. Dr. Shang, TREx experiments. Dr. Wei, sample preparation and quality control. Dr. Cravedi, analysis and interpretation of clinical data and critical revision of draft. Mr. Sun, genomic data analysis. Miss. Xi, clinical data management/query. Mr. Woytovich, sample preparation. Dr. Farouk, analysis and interpretation of clinical data. Dr. Huang and Dr. Banu, data interpretation and manuscript revision. Dr. Gallon, patient enrollment and follow-up. Dr. Magee and Dr. Najafian, pathology reporting. Dr. Samaniego, patient enrollment and follow-up. Dr. Djamali, patient enrollment and follow-up. Dr. Alexander, study design discussion. Dr. Rosales, pathology reporting. Dr. Smith, pathology reporting. Dr. Xiang, TREx and sequencing experiments. Dr. Lerut, sample management and preparation of Belgian cohort. Dr. Kuypers, clinical data management of Belgian cohort. Dr. Naesens, principle investigator of Belgian cohort study, and critical review of manuscript. Dr. O’Connell, interpretation of clinical data, discussion of study design, and critical review of manuscript. Dr. Colvin, pathology reporting. Dr. Menon, analysis and interpretation of clinical data, and critical review of drafting of manuscript. Dr. Murphy, principle investigator, study conception and design, and drafting of manuscript.
Footnotes
Published online ahead of print. Publication date available at www.jasn.org.
- Copyright © 2019 by the American Society of Nephrology
References
In this issue
