CLINICAL RESEARCH www.jasn.org
A Peripheral Blood Gene Expression Signature to Diagnose Subclinical Acute Rejection
Weijia Zhang,1 Zhengzi Yi,1 Karen L. Keung,2 Huimin Shang,3 Chengguo Wei,1 Paolo Cravedi,1 Zeguo Sun,1 Caixia Xi,1 Christopher Woytovich,1 Samira Farouk,1 Weiqing Huang,1 Khadija Banu,1 Lorenzo Gallon,4 Ciara N. Magee,5 Nader Najafian,5 Milagros Samaniego,6 Arjang Djamali ,7 Stephen I. Alexander,2 Ivy A. Rosales,8 Rex Neal Smith,8 Jenny Xiang,3 Evelyne Lerut,9 Dirk Kuypers,10,11 Maarten Naesens ,10,11 Philip J. O’Connell,2 Robert Colvin,8 Madhav C. Menon,1 and Barbara Murphy1
Due to the number of contributing authors, the affiliations are listed at the end of this article.
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 indi- viduals 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 subse- quent 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 trans- plant 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.
JASN 30: 1481–1494, 2019. doi: https://doi.org/10.1681/ASN.2018111098 CLINICAL RESEARCH Kidney transplantation is the therapy of choice for ESRD. Although short-term allograft outcomes Received November 8, 2018. Accepted May 1, 2019. including clinical acute rejection episodes (i.e., occurring in the presence of graft dysfunction) Published online ahead of print. Publication date available at www.jasn.org. have declined over past decades, proportionate im- provement in long-term allograft survival remains Correspondence: Dr. Barbara Murphy, Division of Nephrology, Department of Medicine, Icahn School of Medicine at Mount 1–3 unrealized. The rates of clinically detected epi- Sinai, One Gustave L Levy Place, Box 1243, New York, NY 10029. sodes of acute rejection within the first year in the Email: [email protected] modern tacrolimus/mycophenolate era are ,10% Copyright © 2019 by the American Society of Nephrology
JASN 30: 1481–1494, 2019 ISSN : 1046-6673/3008-1481 1481 CLINICAL RESEARCH www.jasn.org among adult kidney recipients in the United States.2 The effect Significance Statement of these episodes on long-term graft survival is variable, de- pending on the severity of episode, time from transplantation, Biomarkers for noninvasive diagnosis of subclinical acute rejection and effective treatment of these episodes with return of creat- are needed to enable risk-stratification and tailoring of immuno- inine to prerejection baseline.4 suppression for kidney transplant recipients. Using RNA sequencing fl analyses of whole blood collected from a cohort of transplant re- Graft in ammation, however, may also occur initially in the cipients at the time of surveillance biopsy, the authors identified a absence of graft function decline. The wide prevalence of sub- transcriptional signature on the basis of a set of 17 genes that ac- clinical rejection, i.e., lymphocytic tubulitis and interstitial in- curately detects ongoing subclinical rejection. After extensive val- flammation in early surveillance biopsy specimens, and its idation, they developed a sequencing-based targeted expression effect on long-term allograft histology, function, and survival assay on the basis of this gene set that was able to identify subclinical rejection at 3 months post-transplant and increased risk of graft loss 5–10 has been appreciated only recently. The effect of milder in an independent cohort of 110 patients. This assay represents a subclinical inflammation, i.e., suspicious or borderline le- potentially useful tool to monitor kidney transplant recipients and sions, on allograft outcomes is even less characterized.11 optimize immunosuppressive therapy, although larger studies are In current clinical practice, the diagnosis of either clinical or needed to validate the assay’s clinical utility. subclinical allograft rejection requires a biopsy, a procedure burdened by clinical risks and costs. To obviate these issues, readings .1000 were taken as positive. The surveillance biopsy fi prior studies have tested noninvasive pro ling of urinary pro- specimens in the patients in the Belgian cohort were taken at the 12,13 14–16 teins and blood transcriptomic signatures, but results same time points (3, 6, 12, and 24 months) as for the GoCAR have been inconsistent so far. An assay that could be used in cohort. Highly sensitized patients requiring desensitization were clinical practice for the purpose of accurately diagnosing sub- excluded from the GoCAR study and not present in the Belgium clinical rejection offers the potential to identify and treat un- cohort. We used United Network for Organ Sharing and The derlying subclinical rejection without the need for a biopsy. Australia and New Zealand Dialysis and Transplant Registry da- Herein, we examined the incidence of subclinical rejection tabases to determine long-term outcomes for the GoCAR and borderline lesions over time in the Genomics of Chronic enrollees. Allograft Rejection study (GoCAR), a prospective, multicenter center study in which kidney transplant recipients underwent Genomic Experiments and Data Analysis serial surveillance biopsies. We determined the effect of sub- The details regarding genomic experiments (RNA sequencing, clinical graft inflammation on allograft function and survival, microarray, and targeted RNA expression [TREx] assay) are and developed a clinically applicable assay that detects sub- provided in the Supplemental Material and the data analysis clinical acute rejection by measuring the transcriptome in pe- workflow is depicted in Supplemental Figure 1. Briefly, mRNA ripheral blood. 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 sig- METHODS natures associated with ACR-3. After read quality control, mapping, and normalization steps on the raw sequencing Patients reads, the expression data were compared between ACR-3 The study included participants of the GoCAR study and kidney and non-ACR at 3 months (NACR-3), with both induction transplant recipients prospectively followed-up at the University therapy and deceased donor as confounders, to identify dif- Hospitals Leuven, Leuven, Belgium. The GoCAR study is a pro- ferentially expressed genes (DEGs) with a P value ,0.05 using spective,multicenterstudy(UnitedStatesandAustralia)aimedat unpaired LIMMA test,20 a linear model to assess differential investigating the genetics and genomics associated with the de- gene expression in the context of multiple variables. The DEGs velopment of allograft rejection or injury in kidney transplant were then subjected to enrichment analysis for canonical recipients. Patients underwent surveillance biopsies pretrans- pathway, gene ontology, and immune cell types to identify plant (before implantation), and at 3, 6, 12, and 24 months after classes or groups of genes that were associated with ACR. transplant. Patients were followed up for at least 5 years or until Next, a more focused gene set associated with ACR-3 was death. The details of patient enrollment criteria and study design identified from the DEGs using a randomization approach have been previously described.17,18 All biopsy specimens were described previously.18. An optimal gene set with the highest reported for Banff component scores by a central core laboratory AUC (area under the receiver operating characteristic curve) at Massachusetts General Hospital. The diagnosis of acute cel- score for diagnosis of ACR-3 was then determined by fitting a lular rejection (ACR) at 3 months (ACR-3) was made by apply- penalized logistic regression model on the expression data of ing Banff 2013 Classification19 on all clinically indicated and the focus gene set following a 5000-iterations methodology18 surveillance biopsy samples and included borderline subclinical (Supplemental Material). The AUC for the final gene set was rejection. Donor-specific anti-HLA antibodies (DSAs) were crossvalidated using a leave-one-out crossvalidation method measured before transplant and clinically indicated after trans- to avoid potential over-fitting. The gene set was then validated plant by Luminex (ThermoFisher). Mean fluorescence intensity for the diagnosis of ACR-3 using the microarray data from the
1482 JASN JASN 30: 1481–1494, 2019 www.jasn.org CLINICAL RESEARCH
GoCAR study (n=65; 26 overlapping with the RNAseq co- were lost before 12- or 24-month biopsies within each group. hort, Supplemental Material) and three public blood micro- Univariate comparisons of continuous variables were done array datasets for clinical acute rejection (GSE14346,21 using unpaired t test (Mann–Whitney test for corresponding GSE15296,22 and GSE5008419). GSE14346 is an expression nonparametric analysis). Normality of sample distributions was dataset of 75 patients with 38 clinical acute rejections.21 confirmed using Kolmogorov–Smirnov and Shapiro–Wilk tests. GSE15296 is an expression dataset for 75 patients receiving Kaplan–Meier survival curves were calculated with graft loss kidney transplant with 51 clinical acute rejections including 12 (all-cause and death-censored) as outcome. Cox proportional borderline cases.22 GSE5008 is an expression dataset of 42 hazard models were used for multivariable survival associa- patients receiving kidney transplant including 28 cases with tions, including donor and recipient demographics on the ba- DSAs and clinical acute rejection. All three datasets were gen- sis of a priori–determined clinical importance for outcome of erated on the Affymetrix platform.19 interest (i.e., graft survival). Only patients with 3-month bi- The novel sequencing-based targeted expression (TREx) opsy samples were included in the Kaplan–Meier or Cox re- analysis technology (Illumina) was used to develop a molecular gression survival analyses; hence, all cases had survived past 3 assay for the 17-gene set and 12 house-keeping genes with the months. Time was determined from the day of transplant to potential for application in the diagnosis of acute rejection in the graft loss. All statistical analyses for demographic and clinical practice (experimental details in the Supplemental clinical data were completed using SPSS Statistics V23 (IBM Material). The assay first evaluated sensitivity and reproduc- Analytics) and GraphPad Prism V6 (La Jolla, CA). Statistical ibility with universal reference RNA, brain reference RNA, and significance was considered with two-tailed P,0.05. 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 RESULTS quality to be used for the TREx training set. A separate 64 samples from GoCAR patients combined with 46 patient sam- Study Cohorts ples from the Belgian cohort, with similar demographic and One hundred and ninety-one patients from the GoCAR study clinical characteristics in both cohorts (n=110, Supplemental that had peripheral blood RNA concurrent with 3-month kid- Table 1), composed the TREx independent testing set. The ney biopsy were included in this study (Figure 1).18 From this penalized logistic regression model was built on the expression group, 127 patients were randomly assigned to be used for values of 17 genes from the training dataset and tertile cutoffs identification of a peripheral blood gene signature associated on the basis of the probability scores were defined to stratify with subclinical acute rejection using RNA sequencing (n=88) patient risk of ACR into three groups: high, intermediate, and or microarray (n=65). Twenty-six patients underwent both low. The model from the training set was then applied to the RNAseq and microarray analyses, allowing correlation of independent testing set to compute the probability risk score gene expression between the two technologies (Supplemental and positive predictive value (PPV) and negative predictive Figure 2, Supplemental Material). Hence, 127 patients were value (NPV) on the basis of the tertile cutoffs in the second used as the training set for the development of a sequence- set of patients. based targeted expression assay (TREx) which was then vali- The later clinical end points of acute rejection after 3 dated on an independent cohort of 110 patients (64 GoCAR months, fibrosis (CADI score), and the risk of graft loss patients and 46 patients from the biobank of the University were evaluated in the high-, intermediate-, and low-risk Hospitals Leuven, Belgium) (Figure 1, Supplemental Table 1). groups, given the association between ACR-3 and these end The demographic and clinical characteristics of the sub- points in GoCAR. clinical ACR-3 and NACR-3 groups show similar graft function The RNA sequencing and microarray datawere deposited in at 3-month surveillance biopsy (Table 1). The only significant the National Center for Biotechnology Information Gene Ex- differences between ACR-3 and NACR-3 were donor age pression Omnibus database (GSE120398). (P=0.01), induction therapy (P=0.01), and m3 creatinine (P=0.04) (Table 1); however, these factors combined were un- Statistical Analyses able to accurately diagnose acute rejection (AUC=0.720 and Descriptive statistics (means and SD) were used to summarize crossvalidated AUC [cAUC]=0.672, Supplemental Figure 3A). the baseline characteristics of the ACR-3 and NACR-3 cohorts, Subclinical ACR is associated with later allograft fibrosis, and were compared using the chi-squared test and Fisher’s function decline, and loss. To study the natural history of allo- exact test. The diagnosis with selected significant demo- grafts with subclinical cellular rejection or borderline changes, graphic or clinical factors was estimated with logistic regres- we examined subclinical longitudinal histologic and functional sion and the AUC was calculated. Centrally reported Banff/ changes in the ACR-3 and NACR-3 GoCAR cohorts with 3-, CADI component scores were used for histologic compari- 12-, and 24-month surveillance biopsies. We excluded two pa- sons. For composite scores utilized in analysis (i.e.,CADI, tients with BK nephropathy on 3-month biopsy from outcome Ci+Ct), we imputed CADI scores=8 and Ci+Ct=6 (highest analysis. The ACR-3 group had significantly higher CADIscores scores in biopsy specimens at 12 months) for allografts that at 3-, 12-, and 24-month surveillance biopsies (CADI-3, CADI-12,
JASN 30: 1481–1494, 2019 Noninvasive Gene Expression Biomarker for Subclinical Rejection 1483 CLINICAL RESEARCH www.jasn.org
GoCAR Cohort (N=191)
Gene set Transcriptomic Analysis Discovery (N=127) (N=127)
Discovery Cohort Validation Cohort (RNAseq N=88) (Microarray N=65 *)
Bad QC for TREx (N=14)
TREx Assay Development GoCAR BELGIAN (N=223) Independent Independent Cohort (N=64) Cohort (N=46)
TREx TREx Training Set Testing Set (N=113) (N=110)
*26 patients overlapped with RNAseq cohort
Figure 1. GoCAR (n5191) and Belgian (n546) 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 ex- pression (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). and CADI-24) compared with NACR-3 (Figure 2, A and B). When ACR-3 diagnosis did not correlate with increased CADI at 12 or analyzed separately, patients with borderline ACR (BACR-3) 24 months. This suggests that clinically meaningful subclini- alone also had increased CADI scores compared with NACR-3 cal ACR was missed by local reporting (Supplemental Table 3). at 3-, 12-, and 24-month surveillancebiopsies(Figure2, AandB). The incidence of de novo DSAs during 24-month follow-up Interestingly, patients with increased i or i+t scores at 3-month was similar in ACR-3 and NACR-3 (Table 1). Eleven patients biopsies were also more likely to have higher i and t scores on developed acute antibody-mediated rejection (ABMR), all of biopsy at 12 and 24 months, suggesting a persistent inflammatory which were in the first 3 months, eight of them in the ACR-3 phenotype in these patients (Table 2). Consistent with this, the group, and three in NACR-3 (ACR-3 versus NACR-3, OR, presence of ACR-3 was associated with a significantly higher risk 8.29; 95% confidence interval 1.89 to 50.55; Fisher’s of acute rejection at 12 (odds ratio [OR], 7.09; 95% confidence P,0.01). Only one of these cases of ABMR was seen on the interval, 2.82 to 17.81; P,0.01) and 24 months (OR, 3.96; 95% 3-month surveillance biopsy specimen (i.e., subclinical confidence interval, 1.20 to 13.02; P=0.02) compared with ABMR), whereas other cases occurred before the 3-month bi- NACR-3 (Figure 2, C and D). In biopsy specimens with ACR-3, opsy. Microvascular inflammation scores (g+ptc) and C4d chronic injury was already increased at 3 months compared with staining were increased in ACR-3 biopsy specimens (Table NACR-3 (Figure 2A). However, in multivariable regression anal- 2). When the 11 ABMR cases were excluded, microvascular ysis, ACR-3 was associated with significantly higher CADI scores inflammation scores were still higher in ACR-3 versus NACR- at 12 and 24 months even after adjustment for 3-month CADI 3 in the remaining 178 patients. Furthermore, the ACR-3 and Ci+Ct scores (Supplemental Table 2, Table 2). group, excluding ABMR cases, had higher CADI-12 and As part of GoCAR, all biopsy specimens were read by a CADI-24 scores and an increased risk of ACR-12 and ACR- central core with three experienced transplant pathologists. 24 when compared with NACR-3 (Supplemental Table 4). Forty-seven percent of the biopsy specimens in the GoCAR These data suggest that ACR-3 had increased histologic decline cohort were also read by the local pathologist, enabling com- independent of preceding ABMR episodes. Because four of 37 parison of scoring. Forty-eight percent of the BACR-3 cases ACR-3 biopsy specimens were C4d positive, meeting probable identifiedbythecorelaboratorywereclassified as NACR ABMR diagnosis despite absent DSA, and because g/ptc scores locally. In contrast to the core ACR-3 diagnosis, the local in the absence of DSA could result from TCMR lesions,23 we
1484 JASN JASN 30: 1481–1494, 2019 www.jasn.org CLINICAL RESEARCH
Table 1. Demographic characteristics of ACR-3 and NACR-3 Table 1. Continued group patients in GoCAR cohort Characteristics ACR-3 (n=46)b NACR-3 (n=145) P Value Characteristics ACR-3 (n=46)b NACR-3 (n=145) P Value CNI .0.99 Recipient age 46.52613.22 49.42613.66 0.2 N 1 (2.17) 4 (2.76) Recipient sex 0.07 Y 45 (97.83) 141 (97.24) Female 21 (45.65) 44 (30.34) 6 Male 25 (54.5) 101 (69.66) Data are mean SD or number (%). P value, comparison of ACR-3 with NACR- 3 (unpaired t test or nonparametric test), chi-squared or Fisher’s exact test. Recipient race 0.09 DM, diabetes mellitus; HTN, hypertension; PKD, polycystic kidney disease; White 30 (65.22) 90 (62.07) NDSA, non donor specific anti-HLA-antibody; DSA, donor specificantibody; Black 7 (15.22) 26 (17.93) CNI, calcineurin inhibitors. a Hispanic 1 (2.17) 13 (8.97) Only deceased-donor allografts included in analysis. bACR-3 includes 35 patients with BACR and 11 patients with $1a cellular Asian 7 (15.22) 7 (4.83) rejection. Patients with BK nephropathy at 3-mo biopsy were included in Other 1 (2.17) 9 (6.21) NACR-3. Kidney disease 0.12 DM 11 (23.91) 54 (37.24) GN 10 (21.74) 34 (23.45) examined ACR-3 biopsy specimens with/without C4d. C4d- HTN 6 (13.04) 25 (17.24) negative ACR-3 biopsy specimens had significantly higher PKD 7 (15.22) 10 (6.9) g+ptc score than NACR-3 (Supplemental Figure 3B). These bi- Other 12 (26.09) 22 (15.17) opsy specimens likely represent microvascular inflammation as- Previous renal transplant 0.48 sociated with TCMR. These cases also had higher CADI-12 and N 41 (89.13) 122 (84.14) CADI-24 scores and increased risk of ACR-12 and ACR-24 when Y 5 (10.87) 23 (15.86) compared with NACR-3 (Supplemental Table 5). These data Donor age 46.67617.08 39.46615.75 0.01 Donor sex 0.87 suggest that ACR-3 alone is associated with adverse outcomes Female 22 (47.83) 72 (49.66) independent of antibody-mediated injury. Male 24 (52.17) 73 (50.34) We compared the changes of eGFR from 3 (or 6) months Donor race 0.76 to 12 (or 24) months between ACR-3 and NACR-3 groups White 38 (82.61) 117 (80.69) (mean D eGFRs). ACR-3 recipients had greater declines in Black 4 (8.7) 9 (6.21) mean eGFR by 12 and 24 months compared with NACR-3 Hispanic 2 (4.35) 11 (7.59) (Supplemental Figure 3C). Adjusted Cox models showed Asian 2 (4.35) 4 (2.76) that, over a median (interquartile range) follow-up of 1713 Other 0 (0) 4 (2.76) (165–2793) days, ACR-3 was associated with an increased haz- Donor status 0.73 ard of death-censored and all-cause allograft loss versus Living 20 (43.48) 58 (40) NACR-3 (Figure 2E, Supplemental Table 6 [death-censored Deceased 26 (56.52) 87 (60) 3-mo serum creatininea 1.7661.28 1.3460.37 0.04 graft loss], Supplemental Figure 3D [all-cause graft loss]). Cold ischemia time (h)a 9.4567.93 9.86680.76 Delayed graft function 0.67 Peripheral Blood Transcriptomic Signatures Are N 36 (78.26) 118 (81.38) Associated with Subclinical Rejection Y 10 (21.74) 27 (18.62) We next evaluated transcriptomic signatures associated with ACR- Anti-HLA antibodies 3inperipheralbloodtakenfrom88patients(22ACR-3and66 Class-I 0.74 NACR-3)at the time of thebiopsy.Datawereanalyzed according to NDSA 19 (82.61) 22 (75.86) the outline depicted in Supplemental Figure 1. Comparison of DSA 4 (17.39) 7 (24.14) gene expression by LIMMA test20 on normalized data adjusted for Class-II 1 confounders associated with ACR in the GoCAR cohort (induc- NDSA 12 (85.71) 19 (86.36) tion therapy and deceased donor) identified 1115 DEGs (609 up- DSA 2 (14.29) 3 (13.64) P, De novo DSA 0.28 and 506 downregulated) associated with ACR-3 ( 0.05; Figure N 7 (77.78) 23 (92) 3A). Gene ontology enrichment analysis revealed that upregulated Y 2 (22.22) 2 (8) genes were involved in transcriptional regulation and cell cycle, Induction therapy ,0.01 whereas downregulated genes were involved in transport and cy- Lymphocyte-depleting 26 (56.52) 44 (30.34) toskeleton organization processes (Figure 3B). Canonical and in- Nondepleting 10 (21.74) 55 (37.93) genuity pathway analysis showed that these dysregulated genes None 10 (21.74) 46 (31.72) were involved in multiple pathways, including those related to 3-mo maintenance immunosuppression extracellular matrix, cell cycle, TGF-b signaling, B cell receptor Steroid 0.3 signaling, integrin signaling, Jak/Stat signaling, and leukocyte ex- N 1 (2.17) 11 (7.59) travasation signaling (Supplemental Figure 4A). The immune re- Y 45 (97.83) 134 (92.41) sponse genes were enriched in the most downregulated genes
JASN 30: 1481–1494, 2019 Noninvasive Gene Expression Biomarker for Subclinical Rejection 1485 CLINICAL RESEARCH www.jasn.org
AB 6 4 ** *** ** ***** * ACR-3 3 4 BACR-3 NACR-3 2 Ci+Ct 2 CADI-scores 1
0 0 3-months 12-months 24-months 3-months 12-months 24-months
C D ACR-12 ACR-24 150 NACR-12 150 NACR-24 *** * 100 100
50 50 Percentage (n) Percentage (n)
0 0 NACR-3 ACR-3 NACR-3 ACR-3
E 1.0 NACR-3, N=145
0.9
ACR-3, N=46 0.8 Cum Survival 0.7
p < 0.01 0.6
.0 500.0 1000.0 1500.0 2000.0 2500.0 Follow-up duration (days)
Figure 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. ranked by fold-change (Supplemental Figure 4B). These findings pattern was replicated in publicly available blood expression are in keeping with the known immunologic processes involved in datasets of patients with acute rejection (Figure 3D). acute rejection. Immune cell type enrichment analysis using immune cell A Peripheral Blood Gene Signature Accurately profiles in the ImmGene database24 showed that, in addition to Diagnoses Subclinical Cellular Rejection macrophage and natural killer cells, pro-B or pre-T cell genes Of 1115 DEGs for subclinical acute rejection, we identified an were significantly associated with ACR-3 (Figure 3C), and this optimal gene set for the diagnosis of ACR-3 using a combination
1486 JASN JASN 30: 1481–1494, 2019 www.jasn.org CLINICAL RESEARCH
b of logistic regression and permutation-based approaches, as pre- 18 n 0.01 0.01 0.01 0.01 0.01 viously described. The discovery set ( =88) was randomly di- Value , , , , ,
P vided 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- =16) 0.9 0.6 0.01 1.1 0.8 0.38 0.3 0.39 0.7 0.5 0.72 1.1 0.01 2.1 2.6 n 6 6 6 6 6 6 6 6 6 6 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). Diagnosis of ACR-3 using the 17-gene set was internally
24-mo validated on our microarray cohort (n=65; AUC/cAUC=1.000/ =56) ACR-3 ( n 0.7 0.81 0.7 0.94 0.3 0.63 0.4 0.27 0.7 0.07 0.6 0.94 0.7 0.20 0.7 1.56 1.9 2.76 2.7 4.33 0.802, respectively; Figure 4B, Supplemental Table 8). In ad- 6 6 6 6 6 6 6 6 6 6 dition, it accurately diagnosed ACR in three public microarray 6 (13.3) 0 (0.0) 0.09 3 (6.6%)4 (8.8%) 0 (0.0) 0 (0.0) 0.34 0.29 datasets that used blood taken at the time of clinically indi- NACR-3 ( cated biopsies at varying times post-transplant (Figure 4, C–E; b e e e GSE14346: AUC/cAUC=0.959/0.818; AUC=1.000 GSE15296: AUC/cAUC=0.980/0.832; GSE50084: AUC/cAUC=1.000/ 0.01 0.01 0.01 0.41 0.01 0.27 0.01 0.50 0.01 1.45 0.01 2.26 Value , , , , , , ,
P 0.979). These data show that the differential expression of our 17-gene set is identifiable in peripheral blood during sub- clinical or clinical ACR. =30) 0.8 0.03 0.41 0.7 0.04 0.57 0.5 0.30 0.09 0.9 0.03 0.14 0.8 0.41 0.23 0.5 0.8 1.1 1.9 2.8 brosis; Ct, tubular atrophy; Cv, vascular intimal sclerosis; mm, mesangial n fi 6 6 6 6 6 6 6 6 6 6 Development of TREx Assay for Diagnosis of ACR 0.20 Using sequencing-based TREx analysis technology, we developed a molecular assay to measure expression of the 17 12-mo genes on whole blood RNA that demonstrated high sensitiv- =90) ACR-3 ( SD. Ci, interstitial n 0.8 0.77 0.5 0.7 0.90 0.6 0.58 0.5 0.23 0.9 0.73 0.5 0.47 1.2 0.93 1.5 2.17 2.1 3.65 6 ity and reproducibility when evaluated with human universal 6 6 6 6 6 6 6 6 6 6 reference, brain reference RNA, and RNA from our clinical 5 (6.4%)4 (5.2%) 8 (32.0%) 2 (8.0%) 0.63 7 (10.0%) 8 (32.0%) samples (Supplemental Material, Supplemental Figures 1, 5, NACR-3 ( and 6, A–E). a e e e Next, we performed the TREx assay on 113 of the original 0.01 0.41 0.01 0.01 0.01 0.01 0.28 0.01 0.12 0.01 0.43 0.01 0.16 0.01 1.89 127 samples from the transcriptomic analysis to build the sta- Value , , , , , , , , , P tistic diagnostic model on the basis of the expression values
c of the 17 genes (TREx training set; Figure 1, Supplemental Figure 1). Fourteen samples were excluded because of poor =10) 1.0 0.7 0.01 0.67 0.7 0.15 0.27 1.1 0.0 0.47 0.13 0.7 1.7 0.5 1.7 0.03 1.16 2.4 n 6 6 6 6 6 6 6 6 6 6 QC. Expression of the 17 genes by TREx validated our original findings using RNAseq and microarray, clearly differentiat- ing 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 =36) ACR-3 ( n (high, intermediate, and low risk) with NPV=0.98 and 0.8 0.60 0.6 0.60 0.3 0.30 0.7 1.80 0.8 0.00 0.7 0.30 0.9 3.20 0.4 1.66 1.3 1.22 2.0 4.20 6 6 6 6 6 6 6 6 6 6 PPV=0.79 (Figure 5C). 3-mo ammation; t, tubulitis; CADI, Chronic Allograft Damage Index; ptc, peritubular capillaritis. 5 (14.2%) 2 (20.0%) Using the same analytic model derived on the training fl set, the 17-gene TREx assay was validated in the totally inde- pendent cohort of 110 subjects (64 independent GoCAR s test. ’ subjects+46 subjects from the Belgian cohort, Figure 1, Sup-
=143) Borderline ( plemental Figure 1; clinical epidemiologic data in Supplemen- 0.5 0.66 0.5 0.72 0.7 0.07 0.1 1.35 0.5 0.24 0.4 0.37 0.3 1.52 0.3 0.25 0.8 1.36 1.5 2.53 n 6 6 6 6 6 6 6 6 6 6 tal Material and Supplemental Table 1). The 17-gene TREx 2 (1.8%) 0.22 0.40 0.36 0.01 0.10 0.07 0.03 0.03 0.61 1.36 assay accurately diagnosed ACR-3 on the testing set with NPV=0.89 and PPV=0.73 using the tertile cutoffs (Figure 5D). NACR-3 ( value. P suspicious at 3-mo biopsy. 2 5 (4.5%) 8 (22.8%) 4 (40.0%) fi fi . $ test TREx Risk Pro le Strati es Risk for Late ACR and U Graft Loss CADI subscores in serial biopsy specimens Because ACR-3 correlated with graft loss in the GoCAR cohort, Wallis ANOVA and post-test Dunn 2 3 (2.7%) 5 (14.2%) 4 (40.0%) d – Whitney 0 or g+ptc 0 we determined whether the 17-gene TREx assay predicted sub- $ – . . sequent ACR and graft loss using all GoCAR patients included Banff Subscore Mann C4d by immunohistochemistry method. Chi-squared test comparing ACR-3 and NACR-3 groups. Kruskal ACR-3, Banff ACR n Table 2. At 12-mo and 24-mo biopsies, 83.3% and 81.2% of ACR-3 groupa were from BACR-3 group, respectively. Datab are mean c d e ci score matrix increase; g, glomerulosclerosis; i, interstitial in g+ptc C4d ct score cv score t score mm score g score I+t score i score Ci+Ct score CADI score C4d in this study ( =177). Patients in the high/intermediate group
JASN 30: 1481–1494, 2019 Noninvasive Gene Expression Biomarker for Subclinical Rejection 1487 CLINICAL RESEARCH www.jasn.org
A
1.0 ETAA1
PHTF2 AVIL ZMAT1
ETNK1 ZBTB1 USP32P1 FHDC1 ANKRD12 ZNF493 KIAA1033NFYB FAR1 STAG2 ZNF292 FLJ31306 CAPZA2 ASTN2 SENP7 0.5 LYSMD3 THAP5 CLK1 LINC00672 CCDC82
BBS10 CLK4 DNAJB14 SP3 PSMC6 PCMTD1
0.0
THBS1 SIL1 PARVB ITGA2B Log2Rat SPARC PLTP BCKDHA FAM127A GP9 TSC22D1 SND1 SPOCD1 BRE –0.5 CD14 S100A10 TUBB1 CLU CTSD PRF1 PTGDS F13A1 MLC1 GZMH LGR6 GPR56 Log2Rat CMKLR1 S1PR5 CLIC3 0.5 SPON2 –1.0 0.0
FGFBP2 –0.5 –1.0
23 –log10(pvalue.AR) B regulation of transcription vesicle–mediated transport transcription protein localization RNA splicing intracellular transport 10.0 10.0 cell cycle cytoskeleton organization 7.5 7.5 RNA splicing response to DNA damage stimulus 5.0 5.0 cellular response to stress 2.5 regulation of lymphocyte differentiation 2.5 protein amino acid lipidation 0.0 cell death 0.0 DNA repair Ras protein signal transduction meiosis I positive regulation of leukocyte activation
0.0 2.5 5.0 7.5 10.0 0.0 2.5 5.0 7.5 10.0 –log(p) –log(p) C D
-1.0 1.0 0.0 3.0
Macrophages GSE14346 GSE15296 GSE50084 Microarray RNAseq
Natural Killer pro-B cell preT cell Nature Killer cells Pro-B cells Macrophages Pre-T cells CD8 T cells Stem Cells CD4 T cells
B Cells Stromal cells Dendritic Stem Cells Monocytes NK T Cellsgd T Cells Pro-B Cells Cells NeutrophilsPre-T Cells CD4 T CellsCD8 T Cells Monocytes Macrophages Stromal Cells Nature KillerCD4CD8 Cell T Cells
Figure 3. 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 2log10 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 2log10 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
1488 JASN JASN 30: 1481–1494, 2019 www.jasn.org CLINICAL RESEARCH had higher risk of subsequent subclinical ACR at 12 months sequencing of peripheral blood to identify a 17-gene set that (31.9% versus 13.6%; P=0.04) and 24 months (53.8% versus strongly correlated with the presence of ACR-3. Using TREx 32.1%; P=0.08) or of clinical ACR (Banff 1A or greater) at any technology with MiSEQ sequencer, a targeted sequencing time post-transplant (20.6% versus 6.1%; P,0.01) than the platform approved for clinical application, we developed low-risk group (Supplemental Table 9). This was also signif- and validated the 17-gene assay, demonstrating that recipients icant when the intermediate group was considered alone, with stratified into high, intermediate, and low risk for ACR-3. subsequent ACR 1A or higher and ACR/BACR occurring more The current approach to immunosuppression in kidney frequently than the low-risk group (14 of 85 versus four of 70; transplantation is protocol-driven, with adjustments dic- OR, 3.15; P=0.04, and 39 of 85 versus 20 of 70; OR, 2.12; tated by changes in serum creatinine—a poorly sensitive P,0.01, respectively). ABMR episodes, nine of which oc- marker for graft damage. Surveillance biopsies could help curred before the 3-month biopsy, and de novo DSAs were define the intensity of the alloimmune response and guide not significantly different between risk groups (Supplemental immunosuppression accordingly, but they are costly, time- Table 9). consuming, and burdened by a risk of potential complica- The clinical outcomes of GoCAR and the Belgian cohort tions. Our assay enables the identification of underlying in- used for TREx assay are summarized in Supplemental Table 10. flammation, while serum creatinine is still within the normal Multivariable Cox models performed on the combined range, allowing surveillance of the graft status without the GoCAR/Belgian cohorts (n=223) demonstrated that patients need for a biopsy and before there is functional evidence in the high- or intermediate-risk groups had lower death- of injury. Interestingly, it identified ACR in four patients censored graft survival compared with the low-risk group with DGF in whom creatinine is not a functional marker (Figure 5E, Supplemental Table 11). Collectively, these data of rejection. The 17-gene set has the potential to be useful indicate that our TREx assay of the 17-gene set not only ac- to diagnose subclinical rejection at other time points, iden- curately differentiates those at high versus low risk of ACR-3, tifying clinical ACR in the publicly available data sets in but also gives their ongoing risk for rejection and graft loss. which biopsy specimens were taken at multiple times post- Despite the fact that 74.2% of the intermediate group were transplantation. Therefore, these data provide the back- NACR-3, we observed that the group overall had a significantly ground for future studies testing the hypothesis that serial higher risk of graft loss compared with the low-risk group measurements of our assay may guide maintenance immu- (Supplemental Table 11). Stratified analysis comparing the nosuppression management more accurately than our cur- intermediate-risk NACR-3 (I-NACR-3) with the low-risk rent approach using creatinine. NACR-3 (L-NACR-3) demonstrated that the I-NACR-3 and Interestingly, even in the absence of ACR-3 in the interme- the L-NACR-3 had similar Banff subscores at 3 months, with diate group, the TREx-based risk stratification also associates the exception of the mm-scores (Supplemental Table 12). with detection of inflammatory infiltrates in subsequent sur- However, I-NACR-3 developed significantly higher CADI veillance biopsy specimens, and graft function decline and loss. and Ci+Ct scores by 24 months (P=0.01 and ,0.01, respec- This may be explained by the fact that renal biopsy specimens tively; Supplemental Table 12) and had greater risk of graft loss represent 3-mm sections of the total core and may not than L-NACR-3 in adjusted Cox regression (Figure 5F, Sup- capture a focal phenomenon like allograft rejection.25 Besides plemental Table 13). Of note, graft survival in I-NACR-3 did limitations in overall representation, the clinical utility of graft not differ significantly from I-ACR-3 (data not shown). These biopsy is limited by the subjectivity of biopsy reporting. These data demonstrate that biopsy specimens with NACR classified findings emphasize the potential clinical utility of the 17-gene as intermediate are not simply misclassified, but rather have set to avoid pitfalls in the reporting of renal biopsy specimens. outcomes that are truly intermediate between the high- and Although baseline anti-HLA antibodies, including DSAs low-risk groups. and non-DSA, all of which were at low level, and de novo DSAs did not significantly associate with graft loss in our co- hort, ABMR episodes and microvascular inflammation scores DISCUSSION were all more common in ACR-3 versus NACR-3, implying the coexistence of endothelial injury with cellular inflamma- In this study, we used the extensively phenotyped GoCAR co- tion. Of the ten episodes of ABMR, nine were clinical ABMR hort to accurately identify ACR-3 and demonstrate its associ- and occurred before the 3-month biopsy. ation with progressive allograft damage and functional decline, The DEGs associated with ACR-3 were predominantly resulting in a higher risk for graft loss. We then used RNA from pathways not related to the immune response or
shows expression of DEGs that were significantly enriched for immune cell types in the ImmGene dataset. (D) The heatmap of enrichment P value (2log10 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.
JASN 30: 1481–1494, 2019 Noninvasive Gene Expression Biomarker for Subclinical Rejection 1489 CLINICAL RESEARCH www.jasn.org
AB 1.0 1.0
0.8 0.8
0.6 0.6
0.4 0.4 True Positive Rate True Positive Rate 0.2 0.2 original AUC = 0.980 original AUC = 1.000 cross validated AUC = 0.833 cross validated AUC = 0.802 0.0 0.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 False Positive Rate False Positive Rate
CDGSE14346 GSE15296 1.0 1.0
0.8 0.8
0.6 0.6
0.4 0.4 True Positive Rate True Positive Rate 0.2 0.2 original AUC = 0.959 original AUC = 0.980 cross validated AUC = 0.818 cross validated AUC = 0.832 0.0 0.0 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 False Positive Rate False Positive Rate
E GSE50084 1.0
0.8
0.6
0.4 True Positive Rate 0.2 original AUC = 1.000 cross validated AUC = 0.979 0.0 0.0 0.2 0.4 0.6 0.8 1.0 False Positive Rate
Figure 4. 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).
1490 JASN JASN 30: 1481–1494, 2019 www.jasn.org CLINICAL RESEARCH
Table 3. Seventeen-gene set for 3-month ACR diagnosis
Symbol RefSeq Name P Value log2 Ratio ZMAT1 NM_001011657 Zinc finger, matrin-type 1 0.01 0.744899 ETAA1 NM_019002 Ewing tumor-associated antigen 1 0.04 0.703351 ZNF493 NM_001076678 Zinc finger protein 493 0.002 0.663139 CCDC82 NM_024725 Coiled-coil domain containing 82 0.02 0.622008 NFYB NM_006166 Nuclear transcription factor Y, b 0.03 0.589093 SENP7 NM_001077203 SUMO1/sentrin specific peptidase 7 ,0.001 0.581319 CLK1 NM_001162407 CDC-like kinase 1 0.01 0.567156 SENP6 NM_001100409 SUMO1/sentrin specific peptidase 6 0.01 0.510019 C1GALT1C1 NM_001011551 C1GALT1-specific chaperone 1 0.01 0.292181 SPCS3 NM_021928 Signal peptidase complex subunit 3 homolog (S. cerevisiae) 0.03 0.221611 MAP1A NM_002373 Microtubule-associated protein 1A 0.01 20.27124 EFTUD2 NM_001142605 Elongation factor Tu GTP binding domain containing 2 0.001 20.33174 AP1M1 NM_001130524 Adaptor-related protein complex 1, mu 1 subunit ,0.001 20.33596 ANXA5 NM_001154 Annexin A5 ,0.001 20.36446 TSC22D1 NM_001243797 TSC22 domain family, member 1 0.01 20.44781 F13A1 NM_000129 Coagulation factor XIII, A1 polypeptide 0.02 20.5544 TUBB1 NM_030773 Tubulin, b 1 class VI 0.03 20.55441 lymphocyte activation, but rather cell repair, metabolism, and biopsy specimens we have been able to define a unique signa- stress response pathways (Figure 3B), in contrast to previously ture that effectively identifies even borderline subclinical re- reported expression of immune response–related genes in the jection, and predicts the risk for ongoing immune injury and graft biopsy specimens.18 This has previously been described graft loss. Second, validation on public cohorts demonstrated by others26 and could potentially reflect migration of immune the ability of the gene set to identify clinical ACR at multiple cells from the periphery to the allograft. However, immune time points. Third, we have technically validated our assay on cell analysis revealed enrichment for pro-B or pre-T cells or multiple platforms and developed an accurate and reproduc- stem cells in the GoCAR cohort and public validation datasets ible clinically applicable assay (TREx) using MiSEQ with clinical acute rejection, suggesting active B and T cell sequencer, a sequencing platform which is approved by the proliferation, consistent with an active immune response in Food and Drug Administration for clinical application. This the setting of ACR. is highly reproducible and provides absolute transcript Other studies have examined peripheral blood biomarkers measures. This contrasts significantly with other studies in to diagnose ACR.14,21,22,27,28 Initial data using donor-derived which the methods used will be hard to implement on a larger cell-free DNA demonstrated that levels correlated with clin- scale, such as, microarray15 or PCR.21 Lastly, our assay per- ically severe acute T cell–mediated rejection ($Banff Ib) and forms with a high degree of accuracy even in validation co- ABMR; however, leakage of donor-derived cell-free DNA horts with high NPVs and PPVs. This contrasts with a more may lack specificity for rejection and instead reflect graft in- recent paper using gene expression for diagnosis of ACR, in jury in general.29 Several studies have examined peripheral which a microarray assay with 57 genes reported by Friedwald gene expression for the diagnosis of ACR; however, there and colleagues,15 representing a technical challenge for clin- are several pertinent differences compared with the data pre- ical implementation, performed poorly, with PPVs in 51% sented here. First, we used an unbiased approach through and 47% in two validation cohorts, and NPVs in the range RNAseq profiling for gene selection, with both clinical and of 70%.15 Our 17-gene set did not overlap with their 57 technical validation on multiple platforms. Other groups have genes,15 which could be due to the following reasons: (1) taken a reductive approach, narrowing down the genes for Assay variation: we used RNA sequencing for biomarker dis- consideration on the basis of those that are related to immune covery, whereas microarray was applied in their study. RNA- cell expression, and have focused on clinical acute rejec- seq allows absolute quantification of transcripts. Microarray tion.14,21 Prior data have shown that immune response genes allows relative quantification and is restricted to predefined areelevatedintheallograftduringrejectionepisodes,but probe sets. (2) We only investigated 3-month blood profiles, share minimal overlap with simultaneous gene signatures in whereas they analyzed the expression profiles at various times periphery26; thus, by only focusing on immune response from 4 to 24 months; overall gene expression profiles in blood genes, the genes of highest expression and reflecting the great- were reported to change upon immune suppression admin- est changes in expression are excluded from the analysis. This istration post-transplant30. will also account for the lack of overlap between our gene set In summary, our study highlights the negative effect of andthegenesetinthestudybyRoedderet al. 14 By including untreated early subclinical inflammation on subsequent his- all genes and building the diagnostic model with protocol tologic and functional decline in kidney allografts. We
JASN 30: 1481–1494, 2019 Noninvasive Gene Expression Biomarker for Subclinical Rejection 1491 CLINICAL RESEARCH www.jasn.org
A C -1.0 1.0 ACR ZNF493 CIGALT1C1 0.8 Graft Loss SENP7 CLK1 ACR and Graft Loss SENP6 CCDC82 0.6 ZMAT1 PPV = 0.79 NFYB ETAA1 SPCS3 0.4 AP1M1
ANXA5 Probabilities F13A1 MAP1A 0.2 EFTUD2 TSC22D1 NPV = 0.98 TUBB1 0.0 ACR-3 NACR-3 0 20406080100 Samples
B D 1.0 0.8 ACR 0.8 Graft Loss 0.6 ACR and Graft Loss PPV = 0.73 0.6 0.4 0.4
Probabilities 0.2 0.2 True Positive Rate AUC = 0.830 NPV = 0.89 0.0 0.0 0.0 0.2 0.4 0.6 0.8 1.0 0 20406080100 False Positive Rate Samples
E F 1.00 1.00 Low Risk, N=89 Low Risk, N=67 0.95 0.95 High/ 0.90 Intermediate Intermediate Risk, N=134 0.90 Risk, N=63 0.85 0.85
Cum Survival 0.80 Cum Survival
0.80 0.75 p = 0.04 p = 0.03 0.70 0.75
.0 .0 500.0 500.0 1000.0 1500.0 2000.0 2500.0 1000.0 1500.0 2000.0 2500.0 Follow-up (days) Follow-up (days)
Figure 5. 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. found a peripheral blood 17-gene set utilizing a novel TREx with the ultimate goal of controlling subclinical intragraft assay that accurately diagnoses subclinical rejection, includ- inflammation and prolonging graft survival. However, these ing borderline lesions, and stratifies renal recipients accord- findings need further, larger studies to validate the clinical ing to those at high risk for histologic decline and allograft utility of this assay before it can be used as a substitute for loss. Our assay offers the potential to be used as an immune- renaltransplant biopsies for the determination of clinical and monitoring tool to guide the use of immunosuppression, subclinical rejection.
1492 JASN JASN 30: 1481–1494, 2019 www.jasn.org CLINICAL RESEARCH
ACKNOWLEDGMENTS Supplemental Methods. Supplemental Results. We thank the Genomics Resources Core Facility at Weill Cornell Supplemental Figure 1. Correlation between RNA sequencing and Medical Center for sequencing and TREx experiments and Scientific microarray data. Computing at the Icahn School of Medicine at Mount Sinai for Supplemental Figure 2. Association of demographic and patho- providing computational resources. logic characteristics with clinical outcomes. This project is a substudy of the GoCAR study supported by Supplemental Figure 3. Genomic data analysis workflow. National Institutes of Health grant 5U01AI070107-03. Supplemental Figure 4. Pathway and gene ontology enrichment Dr. Zhang, project leader, study design, analysis and interpretation analysis for DEGs associated with ACR-3. of genomic data, and drafting of manuscript. Miss. Yi, genomic data Supplemental Figure 5. The procedure of development of TREx analysis (major contribution) and interpretation. Dr. Keung, analysis assay for 17-gene set. and interpretation of clinical data. Dr. Shang, TREx experiments. Supplemental Figure 6. Development of TREx assay for 17-gene Dr. Wei, sample preparation and quality control. Dr. Cravedi, analysis set. and interpretation of clinical data and critical revision of draft. Supplemental tables. Mr. Sun, genomic data analysis. Miss. Xi, clinical data management/ Supplemental Table 1. Comparison of demographic statistics be- query. Mr. Woytovich, sample preparation. Dr. Farouk, analysis and tween GoCAR and Belgian dataset. interpretation of clinical data. Dr. Huang and Dr. Banu, data in- Supplemental Table 2. ACR-3 predicts CADI-12 and -24 in- terpretation and manuscript revision. Dr. Gallon, patient enrollment dependent of simultaneous chronic damage indices. and follow-up. Dr. Magee and Dr. Najafian, pathology reporting. Supplemental Table 3. Comparison of local and central biopsy Dr. Samaniego, patient enrollment and follow-up. Dr. Djamali, pa- reports at 3-month biopsy. tient enrollment and follow-up. Dr. Alexander, study design discus- Supplemental Table 4. Comparison of clinical characteristics be- sion. Dr. Rosales, pathology reporting. Dr. Smith, pathology re- tween ACR-3 and NACR-3 without AMBR. porting. Dr. Xiang, TREx and sequencing experiments. Dr. Lerut, Supplemental Table 5. Comparison of clinical outcomes post 3 sample management and preparation of Belgian cohort. Dr. Kuypers, months between C4d-negative ACR-3 and NACR-3 groups. clinical data management of Belgian cohort. Dr. Naesens, principle Supplemental Table 6. ACR-3 independently predicts long-term investigator of Belgian cohort study, and critical review of manu- allograft survival. script. Dr. O’Connell, interpretation of clinical data, discussion of Supplemental Table 7. The list of 240 focus genes set. study design, and critical review of manuscript. Dr. Colvin, pathology Supplemental Table 8. Demographiccharacteristics of RNAseq and reporting. Dr. Menon, analysis and interpretation of clinical data, and microarray cohorts in GoCAR cohort. critical review of drafting of manuscript. Dr. Murphy, principle in- Supplemental Table 9. Frequency of anytime rejection episodes in vestigator, study conception and design, and drafting of manuscript. TREx risk groups. Supplemental Table 10. Summary of clinical events of TREx co- horts post kidney transplant. DISCLOSURES Supplemental Table 11. TREx risk group status affects allograft survival. Dr. Zhang reports personal fees from RenalytixAI, outside the submitted Supplemental Table 12. Comparison of Banff scores between in- work; in addition, Dr. Zhang has a patent ‘Method for identifying kidney termediate- and low-risk NACR-3 groups. ’ ‘ allograft recipients at risk for chronic injury pending, a patent Methods for Supplemental Table 13. High-/intermediate-risk NACR-3 affects diagnosing risk of renal allograft fibrosis and rejection (miRNA)’ pending, a allograft survival. patent ‘Method for diagnosing subclinical acute rejection by RNA sequencing analysis of a predictive gene set’ pending, and a patent ‘Pretransplant predic- tion of post-transplant acute rejection’ pending. Dr. Kuypers reports grants and personal fees from Astellas, outside the submitted work. Dr. Murphy REFERENCES reports personal fees from RenalytixAI, outside the submitted work; in addi- tion, Dr. Murphy has a patent ‘Method for identifying kidney allograft recip- ients at risk for chronic injury’ pending, a patent ‘Methods for diagnosing risk 1. Hart A, Smith JM, Skeans MA, Gustafson SK, Stewart DE, Cherikh WS, of renal allograft fibrosis and rejection (miRNA)’ pending, a patent ‘Method et al.: Kidney. Am J Transplant 16[Suppl 2]: 11–46, 2016 for diagnosing subclinical acute rejection by RNA sequencing analysis of a 2. Hart A, Smith JM, Skeans MA, Gustafson SK, Wilk AR, Robinson A, et al.: predictive gene set’ pending, and a patent ‘Pretransplant prediction of post- OPTN/SRTR 2016 annual data report: Kidney. Am J Transplant 18 transplant acute rejection’ pending. Dr. Menon acknowledges funding [Suppl 1]: 18–113, 2018 support from American Heart Association grant 15SDG25870018. 3. Hariharan S, Johnson CP, Bresnahan BA, Taranto SE, McIntosh MJ, Stablein D: Improved graft survival after renal transplantation in the United States, 1988 to 1996. N Engl J Med 342: 605–612, 2000 SUPPLEMENTAL MATERIALS 4. Pirsch JD, Ploeg RJ, Gange S, D’Alessandro AM, Knechtle SJ, Sollinger HW, et al.: Determinants of graft survival after renal transplantation. Transplantation 61: 1581–1586, 1996 This article contains the following supplemental material 5. Shishido S, Asanuma H, Nakai H, Mori Y, Satoh H, Kamimaki I, et al.: The online at http://jasn.asnjournals.org/lookup/suppl/doi:10.1681/ impact of repeated subclinical acute rejection on the progression of ASN.2018111098/-/DCSupplemental. chronic allograft nephropathy. J Am Soc Nephrol 14: 1046–1052, 2003
JASN 30: 1481–1494, 2019 Noninvasive Gene Expression Biomarker for Subclinical Rejection 1493 CLINICAL RESEARCH www.jasn.org
6. Nankivell BJ, Borrows RJ, Fung CL, O’Connell PJ, Allen RD, Chapman 19. Haas M, Sis B, Racusen LC, Solez K, Glotz D, Colvin RB, et al.; Banff JR: Natural history, risk factors, and impact of subclinical rejection in meeting report writing committee: Banff 2013 meeting report: In- kidney transplantation. Transplantation 78: 242–249, 2004 clusion of c4d-negative antibody-mediated rejection and antibody- 7. Kurtkoti J, Sakhuja V, Sud K, Minz M, Nada R, Kohli HS, et al.: The utility associated arterial lesions. Am J Transplant 14: 272–283, 2014 of 1- and 3-month protocol biopsies on renal allograft function: A ran- 20. Ritchie ME, Phipson B, Wu D, Hu Y, Law CW, Shi W, et al.: Limma domized controlled study. Am J Transplant 8: 317–323, 2008 powers differential expression analyses for RNA-sequencing and mi- 8. Szederkényi E, Iványi B, Morvay Z, Szenohradszki P, Borda B, Marofka F, croarray studies. Nucleic Acids Res 43: e47, 2015 et al.: Treatment of subclinical injuries detected by protocol biopsy 21. Li L, Khatri P, Sigdel TK, Tran T, Ying L, Vitalone MJ, et al.: A peripheral improves the long-term kidney allograft function: A single center pro- blood diagnostic test for acute rejection in renal transplantation. Am J spective randomized clinical trial. Transplant Proc 43: 1239–1243, 2011 Transplant 12: 2710–2718, 2012 9. El Ters M, Grande JP, Keddis MT, Rodrigo E, Chopra B, Dean PG, et al.: 22. Kurian SM, Williams AN, Gelbart T, Campbell D, Mondala TS, Head SR, Kidney allograft survival after acute rejection, the value of follow-up et al.: Molecular classifiers for acute kidney transplant rejection in pe- biopsies. Am J Transplant 13: 2334–2341, 2013 ripheral blood by whole genome gene expression profiling. Am J 10. Moreso F, Carrera M, Goma M, Hueso M, Sellares J, Martorell J, et al.: Transplant 14: 1164–1172, 2014 Early subclinical rejection as a risk factor for late chronic humoral re- 23. Gibson IW, Gwinner W, Bröcker V, Sis B, Riopel J, Roberts IS, et al.: jection. Transplantation 93: 41–46, 2012 Peritubular capillaritis in renal allografts: Prevalence, scoring system, 11. Mehta R, Bhusal S, Randhawa P, Sood P, Cherukuri A, Wu C, et al.: reproducibility and clinicopathological correlates. Am J Transplant 8: Short-term adverse effects of early subclinical allograft inflammation in 819–825, 2008 kidney transplant recipients with a rapid steroid withdrawal protocol. 24. Heng TS, Painter MW; Immunological Genome Project Consortium: Am J Transplant 18: 1710–1717, 2018 The immunological genome project: Networks of gene expression in 12. Hricik DE, Nickerson P, Formica RN, Poggio ED, Rush D, Newell KA, immune cells. Nat Immunol 9: 1091–1094, 2008 et al.; CTOT-01 consortium: Multicenter validation of urinary CXCL9 25. Furness PN, Philpott CM, Chorbadjian MT, Nicholson ML, Bosmans JL, as a risk-stratifying biomarker for kidney transplant injury. Am J Trans- Corthouts BL, et al.: Protocol biopsy of the stable renal transplant: A plant 13: 2634–2644, 2013 multicenter study of methods and complication rates. Transplantation 13. Suthanthiran M, Schwartz JE, Ding R, Abecassis M, Dadhania D, 76: 969–973, 2003 Samstein B, et al.; Clinical Trials in Organ Transplantation 04 (CTOT-04) 26. Flechner SM, Kurian SM, Head SR, Sharp SM, Whisenant TC, Zhang J, Study Investigators: Urinary-cell mRNA profile and acute cellular re- et al.: Kidney transplant rejection and tissue injury by gene profiling of jection in kidney allografts. N Engl J Med 369: 20–31, 2013 biopsies and peripheral blood lymphocytes. Am J Transplant 4: 1475– 14. Roedder S, Sigdel T, Salomonis N, Hsieh S, Dai H, Bestard O, et al.: The 1489, 2004 kSORT assay to detect renal transplant patients at high risk for acute 27. Naesens M, Salvatierra O, Benfield M, Ettenger RB, Dharnidharka V, rejection: Results of the multicenter AART study. PLoS Med 11: Harmon W, et al.; SNS01-NIH-CCTPT Multicenter Trial: Subclinical in- e1001759, 2014 flammation and chronic renal allograft injury in a randomized trial on 15. Friedewald JJ, Kurian SM, Heilman RL, Whisenant TC, Poggio ED, steroid avoidance in pediatric kidney transplantation. Am J Transplant Marsh C, et al.: Clinical Trials in Organ Transplantation 08 (CTOT-08): 12: 2730–2743, 2012 Development and clinical validity of a novel blood-based molecular 28. Li L, Khush K, Hsieh SC, Ying L, Luikart H, Sigdel T, et al.: Identification biomarker for subclinical acute rejection following kidney transplant. of common blood gene signatures for the diagnosis of renal and car- Am J Transplant 19: 98–109, 2019 diac acute allograft rejection. PLoS One 8: e82153, 2013 16. Naesens M: The special relativity of non-invasive biomarkers for acute 29. Bloom RD, Bromberg JS, Poggio ED, Bunnapradist S, Langone AJ, rejection. Am J Transplant 19: 5–8, 2019 Sood P, et al.; Circulating Donor-Derived Cell-Free DNA in Blood for 17. Menon MC, Chuang PY, Li Z, Wei C, Zhang W, Luan Y, et al.: Intronic Diagnosing Active Rejection in Kidney Transplant Recipients (DART) locus determines SHROOM3 expression and potentiates renal allograft Study Investigators: Cell-free DNA and active rejection in kidney allo- fibrosis. J Clin Invest 125: 208–221, 2015 grafts. J Am Soc Nephrol 28: 2221–2232, 2017 18. O’Connell PJ, Zhang W, Menon MC, Yi Z, Schröppel B, Gallon L, et al.: 30. Dorr C, Wu B, Guan W, Muthusamy A, Sanghavi K, Schladt DP, et al.: Biopsy transcriptome expression profiling to identify kidney transplants Differentially expressed gene transcripts using RNA sequencing from at risk of chronic injury: A multicentre, prospective study. Lancet 388: the blood of immunosuppressed kidney allograft recipients. PLoS One 983–993, 2016 10: e0125045, 2015
AFFILIATIONS
1Division of Nephrology, Department of Medicine, Icahn School of Medicine at Mount Sinai, New York, New York; 2Department of Medicine, Westmead Clinical School, The University of Sydney, Sydney, New South Wales, Australia; 3Department of Microbiology and Immunology, Cornell Medical Center, New York, New York; 4Department of Medicine-Nephrology, Northwestern University Feinberg School of Medicine, Chicago, Illinois; 5Department of Medicine, Brigham and Women’s Hospital, Boston, Massachusetts; 6Division of Nephrology, Department of Medicine, Henry Ford Hospital, Detroit, Michigan; 7Division of Nephrology, Department of Medicine, University of Wisconsin, Madison, Wisconsin; 8Department of Pathology, Massachusetts General Hospital, Harvard Medical School, Boston, Massachusetts; Departments of 9Pathology and 11Nephrology and Renal Transplantation, University Hospitals Leuven, Leuven, Belgium; and 10Department of Microbiology and Immunology, Katholieke Universiteit Leuven, Leuven, Belgium
1494 JASN JASN 30: 1481–1494, 2019 1 Supplementary Materials 2 Table of Contents...... 1 3 Supplementary methods ...... 2 4 Supplementary results...... 6 5 Supplementary figures...... 8 6 Figure S1: Correlation between RNA sequencing and microarray data...... 8 7 Figure S2: Association of demographic and pathological characteristics with clinical 8 outcomes …………………………………………………...... 9 9 Figure S3: Genomic data analysis workflow….….………...... 10 10 Figure S4: Pathway and Gene Ontology enrichment analysis for DEGs associated 11 with ACR-3………………………………………………………..…………….11 12 Figure S5: The procedure of development of TREx assay for 17-gene set………...... 12 13 Figure S6: Development of TREx assay for 17-gene set…...... 13 14 Supplementary tables...... 14 15 Table S1: Comparison of demographic statistics between GoCAR and Belgian 16 dataset...... 14 17 Table S2: ACR-3 predicts CADI-12 and 24 independent of simultaneous chronic damage 18 indices...………………………………………………………………….……...16 19 Table S3: Comparison of local and central biopsy reports at 3-month biopsy……..……….17 20 Table S4: Comparison of clinical characteristics between ACR-3 and NACR-3 without 21 AMBR ….……………………………………………………………………….18 22 Table S5: Comparison of clinical outcomes post 3 month between C4d negative ACR-3 and 23 NACR-3 groups……………………………………….…………………………19 24 Table S6: ACR-3 independently predicts long-term allograft survival……...……………..20 25 Table S7: The list of 240 focus gene set……………………………………...…………….21 26 Table S8: Demographic characteristics of RNAseq and Microarray cohorts in GoCAR 27 Cohort……………………………………………………………………………30 28 Table S9: Frequency of anytime Rejection episodes in TREx risk groups………………...31 29 Table S10: Summary of clinical events of TREx cohorts post kidney transplant…………...32 30 Table S11: TREX-risk group status impact allograft survival………………………………33 31 Table S12: Comparison of Banff scores between Intermediate- and Low-risk NACR-3 32 Groups………………………………………………………………………………..34 33 Table S13: High/Intermediate risk NACR-3 impacts allograft survival……………………..35 34 References...... 36 35
36 37 Supplementary Methods:
38 RNA sequencing
39 Total RNA was extracted from blood samples using Trizol and the RNA quality was
40 assessed by the Bioanalyzer 2100 (Agilent Technologies). The libraries were generated by
41 following manufactory protocol and were sequenced on Illumina HisSeq2000 sequencer: Briefly,
42 mRNA was firstly extracted from 2ug of total RNA using oligo-dT magnetic beads and
43 fragmented at high temperature. A cDNA library was then prepared from the fragmented mRNA
44 by reverse transcription, second strand synthesis and ligation of specific adapters. Next generation
45 sequencing was performed on Illumina Hiseq 4000 (Illumina Inc.) with single-ended 51 read
46 cycles. Image analysis and bases calling was conducted in real-time by the Illumina analysis
47 pipeline.
48 The raw RNAseq data was processed as follows: The clean reads with good quality were
49 firstly aligned to several human reference databases including hg19 human genome, exon,
50 splicing junction and contamination database including ribosome and mitochondria RNA
51 sequences using BWA 1 alignment algorithm. After filtering the reads mapped to contamination
52 database, the reads that are uniquely aligned to the exon and splicing-junction sites with a
53 maximal 2 mis-matches for each transcript were then counted as the expression level for
54 corresponding transcript and further subjected to quantile normalization cross samples after log2
55 transformation.
56 Microarray experiments
57 Microarray experiments using Affymetrix human Exon 1.0 ST geneChip were performed
58 on total RNA of blood samples following standard protocols provided by the manufacturer
59 (Affymetrix). Briefly, ENCORE amplification and labeling kit (NuGen, San Carlos, CA) was
60 applied to blood RNA samples starting with approximately 100 ng of total RNA to generate
61 biotin-labeled RNA fragments for hybridization to the chip. The chips were scanned by GeneChip
62 Scanner 7G ( Affymetrix Inc) 63 The raw intensity data of Exon geneChip experiments at gene level were extracted and
64 summarized with RMA algorithm 2 and data quality was assessed in Affymetrix Expression
65 Console (Affymetrix Inc). The Affymetrix control probesets or the probesets with low intensity
66 across all samples were excluded from downstream analysis.
67 Correlation of microarray and RNA sequencing data were investigated on 26 patients.
68 Top 10 percentile of genes with the most variable expression levels across the samples were
69 selected from RNA sequencing and microarray data for Pearson correlation analysis.
70 Design of sequencing-based targeted expression (TREx) assay
71 The TREx assay for diagnosis of acute cellular rejection using peripheral blood was
72 designed using the 17-gene set (Figure S5). Twelve house-keeping genes with following criteria
73 were included: 1) minimum variation gene expression across samples and 2) expression values at
74 the average level for all genes detected by RNA sequencing. Sixty-four PCR primers assays were
75 designed for the 17 genes, 12 housekeeping genes and the controls from Illumina Design Studio.
76 The assay kit was manufactured by Illumina Inc (Product No. 75629, Illumina Inc.). PCR was
77 performed on total RNA to amplify the groups of genes using the primer sets and to generate
78 sequencing libraries on the amplicons. After barcoding, the libraries were sequenced using
79 MiSEQ sequencer. The short sequencing data for each sample were analyzed using the
80 sequencing analysis pipeline after de-multiplexing mixed raw sequences.
81
82 Bioinformatics data analysis:
83 Identification of ACR-3 gene set using RNA sequencing
84 Data analysis workflow to identify a set of focus genes for the diagnosis of acute
85 rejection post-transplant and subsequently develop a TREx assay was depicted in Suppl. Figure
86 1.
87 Using RNA sequencing data of 88 patients we identified genes correlated with ACR-3,
88 based on unpaired differential LIMMA test 3 with p value < 0.05 by including clinical factors 89 (induction therapy and deceased donor) as confounders. Biological functional/pathways for the
90 DEGs were determined by enrichment analysis with fisher-exact test using the databases of Gene
91 Ontology (GO) and pathways (KEGG, Ingenuity, BIOCARTA, NABA, Panther, PID,
92 REACTOME, Wiki-pathway). The immune cell types correlated with ACR-3 were evaluated by
93 fisher-exact test of enrichment of immune cell specific genes amongst the DEGs. The immune
94 cell specific genes were identified from ImmGene databases as described previously 4.
95 We next chose a focused geneset that was specifically associated with ACR-3 from the
96 pre-selected ACR-3 genes using an approach of 100-times randomization analysis described
97 previously 4. Briefly, the whole cohort was randomly assigned to 2 groups of equal size (1:1 ratio)
98 and LIMMA testing was performed on each group to identify DEG associated with ACR-3, and
99 this process was repeated 100 times. Genes that occurred more than twice in the 100 iterations
100 with a LIMMA P<0.05 were considered as the focused geneset. An optimal gene set with the
101 highest AUC (area under the receiver operating characteristic (ROC) curve) for prediction of
102 ACR-3 was then determined after fitting penalized logistic regression model on expression data
103 of the focus geneset with 5000 time iterations4. The process started by randomly selecting 20
104 genes from the focus geneset. The expression values of the 20-gene group were fitted into the
105 penalized regression model in logistf R package for ACR-3 diagnosis. The penalized logistic
106 regression model used Firth’s bias reduction method to reduce the bias of maximum likelihood
107 estimates due to small sample size, which will resolve the issue of overfitting from standard
108 logistic regression method. The genes that were significantly associated with AC-3 were
109 identified from the regression model for the 20-gene group. These steps were repeated 5000 time
110 and statistically significant genes (P<0.05) were identified from each iteration. The occurrence of
111 significant genes from the 5000 iterations was computed. Finally, the top 40 genes ranked by the
112 occurrences were applied back to the penalized logistic regression model for ACR-3 diagnosis.
113 Statistically significant genes (P<0.05) in this model were considered as the final optimal geneset. 푒∑ 훽푋 114 The probability was calculated based on the formula 푃 = where 훽 are coefficients 1+푒∑ 훽푋
115 generated from the penalized logistic regression model using the final geneset and 푋 are
116 expression values ( normalized sequencing read count). The AUC was calculated to estimate the
117 overall accuracy for the diagnosis of ACR-3. The final gene set was cross-validated using a leave-
118 one-out cross-validation method to avoid over-fitting issue of self-training of the dataset. In leave
119 one crossvalidation for a dataset (n samples), one sample was left out and the model was built on
120 the rest of samples (n-1) and applied to the left-out one sample to generate the probability score.
121 This step was repeated n times until all samples were tested. AUC was calculated based on the
122 probability scores of all the samples. The gene set was validated for diagnosis of ACR-3 on
123 microarray data for 65 GOCAR patients and 3 public blood microarray datasets for clinical acute
124 rejection (GSE14346 5, GSE15296 6 and GSE50084 7)
125
126 127 Supplementary Result:
128 Correlation of RNA sequencing and microarray data
129 Due to different scale of gene expression measurement by RNA sequencing and
130 microarray technologies, we firstly subtracted the expression value of each gene by its median
131 value across 26 samples to generate relative expression values for RNAseq and microarray,
132 respectively. We then calculated the sample correlation between RNAseq and microarray based
133 on the relative expression values of top 10% of the genes with the most variable expression. The
134 average Pearson correlation was 0.83±0.06 (Figure S2). Our data indicated that RNA sequencing
135 and microarray had good correlation in detecting expression changes among the samples
136
137 Development of TREx assay
138 We developed a molecular assay of 17 gene set from blood RNA to diagnose acute
139 rejection using sequencing-base targeted RNA expression (TREx) analysis technology (Figure
140 S5). Prior to application of blood RNA from transplant patients, the TREx assay was tested with
141 universal human RNA (UHR) and brain RNA samples. The replicated experiments with test
142 samples showed high reproducibility (R=0.993 for UHR samples; Figure S6a), and the fold
143 change for brain RNA versus UHR samples showed concordance between standard RNA
144 sequencing and targeted RNA sequencing ( R=0.949; Figure S6b). The reproducibility of TREx
145 assay on blood RNA from transplant recipients was similar to universal reference samples
146 (R=0.998; in Figure S6c) and overall reproducibility by heatmap (median R =0.978
147 [0.840~0.998]; Figure S6d). The median correlation between standard RNA sequencing and
148 targeted RNAseq data of blood RNA from 87 transplant recipients is 0.87 [0.933-0.745] (Figure
149 S6f). These data on high quality reference samples and clinical blood RNA from kidney
150 transplant recipients indicated that the TREx assay we developed is a reliable assay for clinical
151 diagnosis.
152 153 Demographic characteristics of Belgian cohort
154 46 patients with 3-month allograft biopsies and simultaneous PAXgene RNA were
155 collected from University of Leuven hospital, Belgium. The clinical epidemiologic characteristics
156 of the Belgian cohort are compared with GoCAR cohort in Table S1. The Belgian cohort had
157 similar a ACR-3 rate and follow-up as GoCAR, but differed in donor/recipient demographics
158 given the Northern European ancestry of its population and immunosuppression protocols with no
159 lymphocyte depleting therapy and a statistically higher number of patients on a steroid
160 withdrawal maintenance regimen (Table S1).
161
162 Figure S1. Genomic data analysis workflow: The genomic data analysis workflow includes transcriptomic analysis for identification of ACR-3 diagnosis gene set and TREx assay validation. The transcriptomic analysis identified the transcriptomic signatures and pathways associated with ACR-3 in the RNAseq discovery set and further discovered a 17- gene set for ACR-3 diagnosis which was validated in GoCAR microarray and public dataset. TREx assay was developed for 17-gene set. 113 out of 127 patients in transcriptomic analysis cohort was used as training set to build a penalized logistic regression model which was validated on an independent ACR-3 testing cohort (N= 110 ).
m3-ACR Geneset Discovery Discovery set (m3, N=127)
RNAseq Set (N=88)
Expression Differential Analysis
Pathway/Function m3-ACR Analysis associated Genes Microarray Set (N=65) Logistic Regression (26 duplicated with RNAseq) Analysis Validation 17 gene set 3 Public data sets
TREx Assay Validation TREx Assay Panel (17 genes+12 Training set (N=113, GoCAR discovery set) Housekeepers) Building model
Technical Validation Logistic Model and (reference RNA) Tertile Cutoffs Model validation
Independent cohort (m3 ACR, N=110) Figure S2. The heatmap of Pearson correlation matrix of RNA sequencing and Microarray data on 26 patients.
P9
P8
P7
P6
P5
P4
P3
P26
P25
P24
P23 Correlation P22 (−0.8,0.3] P21 (0.3,0.6] P20 (0.6,0.8] P2 (0.8,1] P19
P18
P17 RNA sequencing RNA P16
P15
P14
P13
P12
P11
P10
P1
P1 P2 P3 P4 P5 P6 P7 P8 P9 P10 P11 P12 P13 P14 P15 P16 P17 P18 P19 P20 P21 P22 P23 P24 P25 P26
Microarray Figure 3S. Association of demographic and pathological characteristics with clinical outcomes: a) The Receiver operating charac- teris- tic (ROC) curve of association of donor age, induction types and 3m creatinine with ACR-3 ( AUC (area under the curve) = 0.720 (black curve) and cross-validated AUC=0.672 (blue curve)); c) The bar charts compares delta eGFR ( 12m-6m, 12m-3m, 24m-3m or 24m-6m) between ACR-3 and NACR-3 d) Kaplan Meier curves compare all cause survival of ACR-3 (green) and NACR-3 (blue) groups in the GoCAR cohort.
a) b) 1.0 0.8 e Rate 0.6 ositi v 0.4 ure P T
0.2 original AUC = 0.720 cross validated AUC = 0.672 0.0
0.0 0.2 0.4 0.6 0.8 1.0 False Positive Rate d)
c) d) e)
ns NACR-3, N=145 50 * *** **
R 0 F
G ACR-3, N=46 e a t l e
D -50
12m-to-3m 12m-to-6m 24m-to-3m 24m-to-6m -100 p < 0.01 -3 -3 -3 -3 -3 -3 -3 -3
ACR ACR ACR ACR NACR NACR NACR NACR
Follow-up duration (days) Figure S4: Pathway and Gene Ontology enrichment analysis on differentially expressed genes (DEGs): a) The barchart of enriched Canonical pathways for DEGs from multiple public pathway databases (upper panel) and Ingenuity (lower panel) . The bar represents –log10 p value of enrichment significance of gene pathways by Fisher-exact test; the lengths of red and green bars represent the percentage of up and down regulated genes, respectively; b) The barchart of enriched Gene Ontology functional terms for the top 100 upregulated (upper panel) or down-regulated (lower panel) genes.
a) b) Canonical Pathway Analysis
SPLICEOSOME(KEGG) PLATELET_ACTIVATION_SIGNALING(REACTOME) regulation of transcription 10 ECM_PATHWAY(BIOCARTA) transcription 8 TGF-BETA SIGNALING PATHWAY(WIKI) 6 REGULATION_OF_APOPTOSIS(REACTOME) change macromolecule catabolic process 4 Dn REGULATION_OF_ACTIN_CYTOSKELETON(KEGG) 2 Up EGF/EGFR_SIGNALING_PATHWAY(WIKI) positive regulation of biosynthetic process DNA_REPLICATION(REACTOME) 0.0 5.0 10.0 15.0 CELL_CYCLE_CHECKPOINTS(REACTOME) SIGNALING_BY_THE_B_CELL_RECEPTOR(REACTOME) MAPK_PATHWAY(BIOCARTA) CELL_CYCLE_MITOTIC(REACTOME) response to organic substance 0.0 2.5 5.0 7.5 wound healing Ingenuity Pathway Analysis lipid transport Integrin Signaling 8 ERK/MAPK Signaling response to wounding PTEN Signaling 6 cell death Caveolar-mediated Endocytosis Signaling 4
B Cell Receptor Signaling change vesicle−mediated transport 2 IGF-1 Signaling Dn Up VEGF Signaling cell adhesion Apoptosis Signaling response to oxidative stress Leukocyte Extravasation Signaling JAK/Stat Signaling immune response 0.0 2.5 5.0 7.5 0.0 5.0 10.0 15.0 -log10(P) −log(p)
Figure S5: The procedure of development of TREx assay for 17-gene set: 12 genes that had the least expression variation and similar expression range as for 17-gene set were selected as house-keeping genes. 64-assay kit was designed for 29 genes (17- gene set and 12 house-keeping genes) with at least 2 amplicons for each gene. The sequencing library were generated on 96-well plate and sequenced by MiSEQ sequencer. The sequence read count of 17-gene set was normalized using the house keeping genes as reference.
17 diagnosis 12 house-keeping genes genes
29 gene panel
Illumina Design Studio
64 assay primers
Assay Kit
RNA 96-well Plate extraction
Library generation Data Analysis and sequencing
Diagnosis Figure S6: Development of TREx assay for 17-gene set: a) The scatterplot of log2 count of duplicated experiments using RNA sample of the Universal Human Reference (UHR) (R=0.994); b)The scatterplot of log2 ratio of the count data of brain vs UHR RNA between duplicated experiments ( R=0.949); c) The scatterplot of log2 count of duplicated experiments using RNA sample of the blood of a kidney transplant (R=0.998); d) The heatmap of expression correlation of duplicated experiments for a group of transplant patients. e) The barchart of correlation of 17 gene expression from TREx assay and RNA sequencing (mean correlation coefficient=0.864, [0.745-0.933], IQR=0.049).
c) a) b)