A Three-Gene Assay for Monitoring Immune Quiescence in Kidney Transplantation

Abstract

Organ transplant recipients face life-long immunosuppression and consequently are at high risk of comorbidities. Occasionally, kidney transplant recipients develop a state of targeted immune quiescence (operational tolerance) against an HLA-mismatched graft, allowing them to withdraw all immunosuppression and retain stable graft function while resuming immune responses to third-party antigens. Methods to better understand and monitor this state of alloimmune quiescence by transcriptional profiling may reveal a gene signature that identifies patients for whom immunosuppression could be titrated to reduce patient and graft morbidities. Therefore, we investigated 571 unique peripheral blood samples from 348 HLA-mismatched renal transplant recipients and 101 nontransplant controls in a four-stage study including microarray, quantitative PCR, and flow cytometry analyses. We report a refined and highly validated (area under the curve, 0.95; 95% confidence interval, 0.92 to 0.97) peripheral blood three-gene assay (KLF6, BNC2, CYP1B1) to detect the state of operational tolerance by quantitative PCR. The frequency of predicted alloimmune quiescence in stable renal transplant patients receiving long-term immunosuppression (n=150) was 7.3% by the three-gene assay. Targeted cell sorting of peripheral blood from operationally tolerant patients showed a significant shift in the ratio of circulating monocyte-derived dendritic cells with significantly different expression of the genes constituting the three-gene assay. Our results suggest that incorporation of patient screening by specific cellular and gene expression assays may support the safety of drug minimization trials and protocols.

Our current limited ability to assess varying immune adaptive states to the allograft in different recipients results in the use of standard protocol-driven maintenance doses of immunosuppression in all patients. As a result, patients experience drug-specific toxicities, mainly cardiovascular morbidity, infections, diabetes, cancer, and nephrotoxicity.^1–4 Many patients, however, reveal stable graft function off immunosuppression without developing significant detrimental immune reactions or immune deficits.^5,6 This suggests that operational transplant tolerance is likely a transient state of alloimmune quiescence that can develop under the umbrella of maintenance immunosuppression.^7,8 These patients are conventionally called operationally tolerant (TOL) and provide a unique repertoire for study and development of monitoring methods that help to differentiate transplant recipients receiving immunosuppression with differing immune thresholds and thus help identify patients who may safely minimize their immunosuppression. Transcriptional studies in peripheral blood by our group and others have identified gene signatures for TOL after kidney^7–9 and liver^10,11 transplantation. But these studies are limited by insufficient cross-validations in independent cohorts, and, importantly, the frequency of a TOL signature is poorly defined in stable transplant recipients receiving immunosuppression. Therefore, the goals for the present study were to provide a highly cross-validated TOL gene signature in blood as a potential measure of immune quiescence to eventually guide safe reduction of immunosuppression; to evaluate the frequency of this signature in patients receiving different immunosuppressive regimens; and to further elucidate the underlying TOL biology by identifying potentially protolerogenic cell subsets in blood.

RESULTS

We investigated 571 unique peripheral blood samples collected from 348 renal transplant recipients and 101 nontransplant controls, in four stages, by microarray, quantitative PCR (qPCR), and FACS (Figure 1). Patient demographic characteristics for new microarray analysis (stage 1A) and qPCR validation, training, and prediction (stages 2 and 3) are list in Tables 1 and 2; patient demographic characteristics for TOL cell-specific analyses (stage 4, C and D) can be found in Table 3. Additional patient gene expression data used in this study were downloaded from the public domain^7,8,12 and used for the microarray cross-validations (stage 1, B and C) and for TOL biology analysis (stage 4, A and B).

Figure 1.

Study design. Four-stage study design: New microarray discovery (n=31) (A) and cross-platform microarray validations (B) (I [n=29] and II [n=58]) (stage 1) in peripheral blood to refine the present gene signature for TOL⁷ to a 21-gene signature; qPCR validation in 59 independent peripheral blood samples (stage 2); qPCR modeling and prediction in 220 peripheral blood samples for developing and training a three-gene assay in 70 samples (stage 3A) and for prediction of the prevalence of TOL under the umbrella of immunosuppression in 150 samples (stage 3B); and TOL biology analysis (stage 4) to identify TOL-specific cell types with enrichment of the 21 TOL genes by FACS and gene expression analysis. A total of 571 human blood and tissue samples across transplant centers in the United States and Mexico were investigated.

Table 1.

Demographic information for TOL patients and SI patients, and varying kidney function used for novel discovery (stage 1), qPCR validation (stage 2), and TOL modeling (stage 3) (n=121 unique patients)

Variable	TOL (n=43)a	SI (n=78)a
Recipients
Male patients (%)	68.4%	74.0%
Mean age±SD (yr)	28±20	15±13
Race (%)
White	78	56
Hispanic	0	11
Asian	22	0
African American	0	22
Other	0	11
Post-transplant time (mo)
Mean	216.8	47.6
Median±SD	195.7±139.2	23.5±71.7
Minimum/maximum.	11.4/460	0.36/300
Induction therapy	NA	Daclizumab/antithymocyte globulin
Maintenance therapy	–	CNI+steroids/MMF, with or without AZA
Serum creatinine (mg/dl)	0.95±0.2	2.92±2.9
Donors
LRD donor source (%)	0.32	0.67
Mean HLA mismatch (x/6) ±SD	0.75±1.5	2.92±2.9
Male donors (%)	0.5	0.42
Mean age±SD (yr)	39.8±16.6	42.86±10.84

NA, not applicable; CNI, calcineurin inhibitor (cyclosporine, tacrolimus); MMF, mycophenolate mofetil; AZA, azathioprine; LRD, living-related donor; x/6, number of HLA mismatches out of a total of 6.

^aUnique patients used in novel microarray discovery; qPCR validation and modeling.

Table 2.

Patient demographic information for the SI patient group (n=150) used for independent prediction (stage 3B) (n=150 unique patients)

Variable	Data
Recipients
Male patients (%)	63.3
Mean age±SD (yr)	33.3±19.2
Post-transplant time (mo)
Mean	25.5
Median±SD	29.0±11.4
Minimum/maximum	0.3/63.2
Induction therapy (%)
Daclizumab	25.3
Alemtuzumab	21.3
Basiliximab	53.3
Maintenance therapy(%)
CNI	60.7
Cyclosporine	14
Tacrolimus	19.86
Belatacept	34.7
Steroid-Free	39.3
Serum creatinine (mg/dl)	1.27±0.34
GFR (ml/min. per 1.73 m2)	60.95±22.8
Donor
LRD donor source (%)	59.3
HLA mismatch	2.93±1.74
Male donor (%)	48.0
Donor age (yr)	37.7±13.1

Patients in the SI group were receiving maintenance immunosuppression and had stable clinical graft function.

Table 3.

Patient demographic information for the TOL and SI patient groups used for FACS analyses of TOL cells (n=11 unique patients)

Variable	TOL (n=5)a	SI (n=6)b
Recipients
Male patients (%)	40.00	100.00
Mean age±SD (yr)	24±7.87	11±6.86
Post-transplant time (mo)
Mean	221.8	54.4
Median±SD	219.1±56.08	16.7±64.06
Minimum/maximum	157.7/291.3	11.9/159.2
Induction therapy	NA	Daclizumab/antithymocyte globulin
Maintenance therapy(%)	–	CNI+steroids/MMF, with or without AZA
Serum creatinine (mg/dl)	0.95±0.2	1.4±0.73
Donors
LRD donor source (%)	80	66.70
HLA mismatch	NA	0.75±1.5
Male donor (%)	75	50
Donor age (yr)	27.5±8.06	NA

NA, not applicable.

^aOperational tolerant.

^bStable immunosuppression.

Stage 1: Cross-Platform Microarray Discovery and Cross-Validation

Stage 1A: 21-Gene Signature for Operational Tolerance

In the new microarray discovery set of 31 peripheral blood samples, 141 unique genes (153 Agilent probes) were significantly differentially expressed in TOL (statistical analysis of microarrays [SAM],¹³ false discovery rate [FDR], 5%) (Supplemental Table 1). Among these, a minimal set of 21 unique genes (34 Agilent probes) (Table 4) correctly classified TOL patients (n=16) from patients with chronic allograft injury (CAN) (n=10) and from healthy nontransplant individuals (HC) (n=5) (prediction analysis of microarrays [PAM]¹⁴) (Figure 2A) and provided excellent segregation of samples by unsupervised hierarchical clustering (Figure 2B).

Table 4.

Significant changes between CAN and SI versus TOL for 21 genes in 417 peripheral blood samples across 5 independent platforms.

TOL versus CAN	Gene ID	q Value (%)			P Values
		Agilent	Lymphochip	Affymetrix	SAB	Fluidigm
BNC2	54796	0.0000	Not expressed	0.0000	0.0002	0.0000
CYP1B1	1545	0.0000	Not expressed	0.0000	0.0001	0.0000
KLF6	1316	1.4100	3.5000	3.2468	0.0000	0.0096
IGFL2	147920	4.9800	Not expressed	0.0000	0.0059	0.0000
CCL4	6351	0.0000	Not expressed	0.0000	0.0000	0.04
SHCBP1	79801	0.0000	Not expressed	NS	0.0001	0.03
SPC25	57405	0.0000	0.8500	NS	0.014	0.07
UHRF1	29128	1.4100	0.0000	NS	0.0099	0.0024
NXF3	56000	3.7000	Not expressed	5.9524	0.0000	Not expressed
IGHA2	3494	0.0000	Not expressed	Not expressed	0.02	0.47
TNFRSF17	608	0.0000	0.0000	0.0000	0.0009	0.22
IGJ	3512	0.0000	0.8500	0.0000	0.17	0.71
IGHG1a	3500	0.0000	Not expressed	NS	0.068	0.27
IGHG4a	3503	0.0000	Not expressed	NS	0.068	0.27
IGH@	3492	0.0000	Not expressed	0.7666	0.47	Not expressed
FAM110C	642273	0.0000	Not expressed	0.0000	0.0000	0.67
VN1R2	317701	1.4100	Not expressed	NS	0.02	0.36
CLVS1	157807	1.4100	Not expressed	NS	0.05	Not expressed
GDEP	118425	3.7000	Not expressed	NS	0.0005	Not expressed
C1QC	714	3.7000	Not expressed	NS	0.0000	0.64
PRAMEF3	401940	4.9800	Not expressed	NS	0.02	Not expressed
TFDP3	51270	3.7000	Not expressed	NS	0.30	0.66

Significance of the 21 genes between TOL and CAN were calculated by SAM¹³ for microarray data (Agilent, Lymphochip, Affymetrix) and between TOL and SI by two-sided t test for qPCR data (SAB, Fluidigm). Any q value ≤5% and P value ≤0.05 were considered to represent a statistically significant difference. NS, nonsignificant by q≤5%.

^aIGHG1/IGHG4 represented by same qPCR primer/probe set.

Figure 2.

Refining the TOL signature to 21 unique genes for correct prediction (A) and segregation of TOL, CAN, and HC samples (B): 21 genes were identified from a set of 141 significantly differentially expressed genes and were correctly classified in 16 TOL, 10 CAN, and 5 HC samples by nearest shrunken centroid (PAM FDR, 5%) from our new Agilent microarray discovery set (A) and were correctly segregated the same patients by phenotype by unsupervised hierarchical clustering (B). Part A shows predicted probabilities for TOL by PAM using 21 genes. The threshold for TOL prediction was set at predicted probability score >50%.

Stage 1B: Discrimination of TOL Patients in Two Public Microarray Datasets

Homologues of the 21 genes from the Agilent arrays were evaluated for their ability to reclassify independent TOL blood samples analyzed on two different microarray platforms from a 34 blood sample set of TOL, CAN, and stable immunosuppression (SI) patients on the cDNA Lymphochip,⁷ and from a separate 58 blood sample set of TOL, SI, and HC patients on the Affymetrix HG U133 plus 2.0 gene chip (GSE22229⁸). Given the 4-fold smaller representation of genes on the Lymphochip versus the Agilent platform, re-annotation to the most recent National Center for Biotechnology Information gene identifiers and mapping across different platforms using Array Information Library Universal Navigator (http://ailun.stanford.edu)¹⁵ found five overlapping genes on the cDNA Lymphochip (IGJ, TNFRSF17, SPC25, KLF6, and UHRF1), which provided sample classification similar to that reported in the published study⁷ (Supplemental Figure 1A). These genes were also significantly differentially expressed in TOL (FDR, 5%) (Table 4). In the Newell dataset,⁸ all 21 genes were present and provided a high rate of accurate sample segregation by phenotype (Supplemental Figure 1B) with correct class assignment of 15 of the 19 TOL samples and of 34 of the 41 SI and HC samples. Ten of the 21 genes in the Newell dataset were also significantly differentially expressed in TOL (SAM, FDR 5%) (Table 4).

Stage 2: qPCR Validation of TOL-Specific Genes

Twenty-one TOL Genes Discriminate an Independent Set of 31 TOL Patients by qPCR

Standard qPCR (SABiosciences Superarray) was done in 59 independent peripheral blood samples (31 TOL, 28 SI) (Table 1) for the 21 genes (plus 18S). The qPCR data allowed clear segregation of the TOL and SI phenotypes by unsupervised principal component analysis (PCA, 70.3%) (Figure 3A) and by hierarchical clustering (Figure 3B); qPCR also validated the differential expression of 17 of the 21 genes in TOL (two-sided t test, P<0.05) (Table 4 and Figure 3B).

Figure 3.

Validation of 21 TOL genes in 59 independent patients (28 SI, 31 TOL) from multiple centers in the United States by qPCR. This assay validated significance of identified TOL genes in 31 TOL and 28 SI patients. Unsupervised principal component analysis showed 70.3% segregation between SI and TOL using 18 of 21 genes with sufficient expression levels (A) and correctly clustered samples by phenotype (26 of 28 SI patients and 26 of 31 TOL patients clustered correctly) by unsupervised hierarchical clustering (B); significant differential expression between TOL and SI (B) (P values were calculated by two-sided t test).

Stage 3: qPCR Modeling of a Three-Gene TOL Assay

Stage 3A: Selection of KLF6, BNC2, and CYP1B1 for a Minimal Three-Gene Assay for TOL

High-throughput microfluidic qPCR (Biomark HD, Fluidigm, CA) for the 21 genes in a second independent sample set (n=70, 17 TOL, 53 SI) (Table 1) resulted in a quality control filtering of five genes and seven samples. The best-performing and minimal gene-set to detect TOL was a set of three genes (KLF6, BNC2, and CYP1B1), which correctly classified the TOL samples by penalized logistic regression with 84.6% sensitivity, 90.2% specificity, and an area under the curve of 0.95 (95% confidence interval, 0.97 to 0.92) (Figure 4A). Penalized logistic regression provided accurate estimates for the regression coefficients and a numeric probability estimate for each patient (rather than a simple categorical class estimate), calculated as a percentage predicted probability of TOL. The cutoff for the predicted probability for a sample to be classified as TOL (θ) was θ=0.25, which had the best sensitivity and specificity by maximizing the correct rate. These three genes also separated TOL from SI by unsupervised clustering (Figure 4B) and had significantly different (two-sided t test) expression levels (P<0.05) in TOL (Figure 4A and Table 4).

Figure 4.

Development of a peripheral blood three-gene TOL assay. Independent qPCR for the 21 genes was performed in 65 peripheral blood samples. A three-gene model (KLF6, BNC2, and CYP1B1) predicted TOL with an area under the curve (AUC) of 0.95 (95% confidence interval [95% CI], 0.97 to 0.92), with 84.6% sensitivity and 90.2% specificity; the threshold for TOL prediction was set at θ=25% (A); the same genes segregated TOL and SI samples by unsupervised principal component analysis (B) and were significantly increased in TOL samples (P<0.05, two-sided t test with Welch correction) (C). Shown are individual gene expression fold changes with mean and SEM calculated against a universal RNA using the ddCt method.⁴³ ROC, receiver-operating characteristic curve.

Stage 3B: Assessing the Frequency of TOL Prediction in 150 Stable Patients Receiving Standard Immunosuppression

In a purely observational analysis, we applied the three-gene assay (KLF6, BNC2, and CYP1B1) to peripheral blood samples from 150 stable renal transplant recipients (mean serum creatinine±SD, 1.27±0.34 mg/dl; mean GFR, 60.95±23.81ml/min per 1.73 m²; no detectable donor-specific antibody [DSA]) receiving long-term maintenance immunosuppression with a minimal 3-year clinical follow-up (Table 2). Immunosuppression for induction and maintenance in these patients differed: anti-CD52 (alemtuzumab) plus calcineurin inhibitor (CNI) (n=32); anti-CD25 (daclizumab) plus CNI (n=38); anti-CD25 (basiliximab) plus CNI (n=21); and anti-CD25 (basiliximab) plus belatacept (n=59). CNI consisted of tacrolimus (n=73) or cyclosporine (n=21); belatacept recipients were a subset of patients enrolled in BENEFIT (Belatacept Evaluation of Nephroprotection and Efficacy as First-line Immunosuppression Trial).¹⁶ To reduce false-positive rates for predicting whether a patient has a TOL phenotype, the three-gene assay specificity was increased to 98.4% to maximize assay safety by increasing the threshold θ for TOL prediction from 0.25 to 0.6. As a result, 11 patients (7.33%) were predicted as being TOL (Supplemental Figure 2, A and B). Of these, 5 patients were receiving belatacept (8.5%) compared with 2 patients receiving CNI (3.4%) after the same induction (CD25; n=118) (Supplemental Figure 2A). Receiving CNI for maintenance (n=91), 4 patients with alemtuzumab induction were predicted as TOL (n=4 of 32; 12.5%) compared with 2 patients with anti-CD25 induction and CNI maintenance (n=2 of 59; 3.4%) (Supplemental Figure 2B). According to the small numbers of patients in each drug treatment subgroup, none of the differences in frequencies were statistically significant by two-sided chi-squared test (threshold of significance P≤0.05). Additional clinical parameters, such as time since transplantation, donor age, donor source, HLA mismatch, and cause of ESRD, did not affect the frequency of predicted TOL phenotype. Three patients with a predicted TOL phenotype had repeat blood samples within 1 year of the original sample, and the predicted probabilities for the TOL phenotype by the three-gene assay were consistently elevated in all 3 patients. On clinical follow-up of all 11 patients with predicted TOL phenotypes, none experienced a rejection episode or had detectable DSAs within 12 months after sampling; despite a transient drift in serum creatinine in 2 of the 11 patients, all continued to have stable graft function at 3-year follow-up (average serum creatinine, 1.31 mg/dl; mean GFR, 59.89 ml/min per 1.73 m²).

Stage 4: Biologic Analysis of TOL Genes

Stage 4A: Peripheral Blood Cell Subset Analysis

To identify the peripheral blood cell types that were specifically contributing to the differential expression of the 21 TOL gene set, we first analyzed their expression profiles in 158 microarrays from 79 normal human cells and tissues (BioGPS, GSE1133). Hypergeometric enrichment analysis of the 21 TOL genes in this dataset identified a signature suggestive of maximal enrichment in dendritic cells (n=7 of 21 genes; P=0.013), with additional gene enrichment in B lymphocytes (B cells; n=7 of 21 genes; P=0.047) and NK cells (n=6 of 21 genes; P=0.042) (Table 5). Gene expression >3-fold higher in a given cell/tissue compared with the median expression in all samples was considered significant enrichment. KLF6 and CYP1B1 were also highly enriched in dendritic cells and myeloid cells.

Table 5.

Hypergeometric enrichment analysis of 21 genes in peripheral blood cells from healthy samples

Related Cell	Common Targets with ≥3-Fold of Median Intensity	Mapped Upregulated Probes in TOL (n)	Hypergeometric P Value
Dendritic cells	3922	7	0.01
B cells	5028	7	0.05
CD56+ NK cells	3802	6	0.04
CD8+ T cells	3278	3	0.26
CD4+ T cells	3157	3	0.25
Monocytes	3241	3	0.26
CD34+ cells	3898	3	0.27
CD33+ cells	11,218	4	0.21

Frequency and hypergeometric P values for common targets between BioGPS from HU133A platform (GSE1133) and identified 18 upregulated targets specific to tolerance from Agilent platform and P values of enrichment across related blood cell types are shown.

Stage 4B: Inferred Biologic Function of the TOL Genes

Downstream analyses of the 21 TOL genes for their biologic function by Ingenuity Pathway Analysis (Ingenuity, Redwood City, CA) revealed that 13 of the 21 genes were involved in an apoptosis network with a central signaling role for TNF, IL6, and IL4 (Supplemental Figure 3A), and canonical pathway analyses identified the complement system (P=0.03) and the B cell activating factor signaling pathway (P=0.04) associated with these genes (Supplemental Figure 3B). BNC2, which encodes for a DNA-/metal-binding protein, has previously been associated with peripheral gene expression profiling of operationally tolerant liver transplant recipients¹⁰; KLF6 belongs to the Krueppel-like family of transcription factors, which participate in diverse aspects of leukocyte growth, development, differentiation, and activation of cells of myeloid lineages.¹⁷ Finally, CYP1B1 belonged to the cytochrome P450 family of monooxygenases that catalyzes reactions involved in drug metabolism and synthesis of cholesterol, steroids, and lipids.^18,19

Stage 4C: FACS Quantitative Analysis of Selected Peripheral Blood Cell Subsets

On the basis of the significant TOL gene enrichment in dendritic, myeloid, B, and NK cells, FACS analysis was conducted on these selected cell populations in 5 TOL patients, 6 SI patients, and 5 HCs (Table 3). Although the total numbers of T cells were significantly lower in TOL patients than in HCs (P<0.001) (Figure 5A), the difference between T cells in TOL and SI patients did not reach significance (Figure 5A). When T cells were sorted for CD4 staining, CD4⁻/CD3⁺ T-cells were significantly lower in TOL than in both HCs (P<0.01) and SI patients (P=0.05); this difference was not seen for CD4⁺ T-cells (Figure 5A). As suggested by the hypergeometric gene enrichment data for the 21 TOL-specific genes, CD14hi CD16⁻ CD11C^- monocytes were significantly enriched in the TOL patients compared with the SI patients (P=0.03) and HCs (P=0.0005). In addition, CD11C⁺, CD304⁻, CD14^low dendritic cells were significantly increased in TOL patients compared with both SI patients (P=0.05) and HCs (P=0.028) (Figure 5B, Supplemental Table 2). In examining NK cells, CD14⁻, CD16⁻, and CD56^bright NK cells were increased in TOL patients compared with HCs (P=0.0084); there were no significant changes in NK cells between TOL and SI patients, although NK cells trended toward lower numbers (P=0.13) in TOL compared with SI. Interestingly, the number of B cells that stained positive for CD19 did not show consistent elevation in TOL patients, as suggested in other publications^7,8; however, because of limited sample volumes we could not perform substaining for immature and transitional B cells.

Figure 5.

Variations in peripheral blood cell subsets in operational tolerance. Quantitative analyses of peripheral blood cells in TOL show decreased CD8 T cells (A) and increased cell populations of myeloid lineage (B). Dendritic cells, monocytes, NK cells, and B cells were quantified in PBMCs from 5 TOL patients, 6 SI patients, and 5 HCs by FACS. TOL patients showed significantly reduced numbers of T cells, which additionally stained negative for CD4 compared with both HCs and SI patients (A). Dendritic cells that additionally stained low for CD14 (CD14^lo DC) and monocytes that stained high for CD14 (CD14^hi) were significantly enriched in TOL compared with HC and SI samples (B, upper panel left and right; ratio to CD4⁻ T cells). In contrast, NK cells that stained bright for CD56 (CD56^Bright) were significantly enriched in TOL samples only compared with HC samples and showed slightly decreased numbers in TOL compared with SI samples (lower panel left). Graphs in A show mean cell counts per million live cells; graphs in B show mean cell counts per million live cells calculated as ratio to CD4⁻ T cells. Significance was calculated by two-sided t test. *P≤0.05; **P<0.01; ***P<0.001.

Stage 4D: Expression of the BNC2, KLF6, and CYP1B1 TOL Genes in CD11c⁺ Dendritic Cells in TOL Patients

To investigate whether the three-gene TOL signature identified in whole blood from TOL patients originated from the CD11c⁺ cells enriched in TOL, we isolated CD11c⁺ cells from 5 TOL and 13 non-TOL patients and analyzed the expression of the TOL genes (BNC2, KLF6, and CYP1B1). All three genes showed significant differential expression in CD11c⁺ cells isolated from the TOL patients (BNC2, P=0.04; CYP1B1, P=0.01; KLF6, P=0.05) compared with non-TOL patients.

Discussion

Transcriptional signals in peripheral blood and tissue have been found to track with the state of clinical operational transplant tolerance^{7–9,11,20,21} but lack cross-validation and additional analyses in stable transplant patients receiving different immunosuppression regimens. The present study highlights a peripheral three-gene assay that detected operational TOL in HLA-mismatched, clinically stable renal transplant recipients off immunosuppression with high sensitivity and specificity, highly confirmed by independent cross-validations. When applied to HLA-mismatched stable renal allograft recipients receiving long-term maintenance immunosuppression, this assay also discerned a clinical homeostatic state of low alloimmune risk in 7.3% of patients receiving immunosuppression. Additionally, TOL patients showed significant enrichment of myeloid-derived cells in peripheral blood with significantly differential expression of BNC2, KLF6, and CYP1B1 constituting the three-gene assay of TOL.

Different TOL-specific gene panels have been reported in different studies of both liver and kidney transplant tolerance.^7,8,21,22 Interestingly, BNC2 and CYP1B1 in our three-gene assay have been independently linked to TOL in liver and kidney transplantation by us and others,^7,8,10 and CYP1B1 has additionally been reported as drug target to influence antitumor immune responses.^23,24 Different studies have also reported different cell subtypes to play a role in operational tolerance, particularly B cells and NK cells,^8,21 with emerging evidence for a role of the antigen presenting dendritic cells.^25,26 An immunosuppressive, protolerogenic role for the myeloid-derived dendritic cell subpopulation^27,28 and other cells of myeloid lineage^7,8 has been recently shown, and in the ONE-Study different monocyte-derived regulatory cell populations are being tested for their cell therapy potential²⁹ in renal transplant recipients. While we also noted enrichment of the 21 TOL genes in B cells and NK cells by microarray analyses, in addition to the highest enrichment of the genes seen in dendritic cells, only cells of myeloid origin (myeloid-derived dendritic cells, monocytes) showed actual increased presence in blood of TOL patients by downstream FACS analysis with additional differential expression of the three genes in these cells.

Most patients receiving long-term maintenance immunosuppression with stable graft function will require continuation of their long-term maintenance immunosuppression, as the frequencies for the presence of a TOL phenotype predicted by gene signatures in these patients are low: approximately 7% of patients in this United States and Mexican multicenter study defined by our three-gene assay; 8% of patients reported in our previous multicenter study from the United States, Canada, and Europe⁷; and 3.5% of patients in a single-center European cohort.³⁰ Given these small patient numbers it is not possible to definitively assess whether a protolerogenic cell composition and a protolerogenic milieu can be achieved by specific immunosuppression options. However, the trend toward more patients receiving a belatacept/CNI-free regimen seen in the present study is thought provoking and may correlate the excellent long-term graft function in this cohort. The association of anti-HLA DSAs and persistence of the predicted TOL probability is unclear. Low levels of DSA have been observed in TOL-predicted patients in some studies,^7,31 but DSA were not detected in any of the TOL-predicted patients in this study.

In conclusion, this study provides a highly validated, peripheral blood, three-gene assay to detect a TOL patient phenotype and infers mechanisms into this state of operational allospecific tolerance. The three-gene assay offers a potential means to monitor for donor-specific hyporesponsiveness and graft accommodation in all immunosuppressed transplant recipients, segregating patients who may be on a larger burden of long-term immunosuppression than is “immunologically” necessary for customized immunosuppression management.

CONCISE METHODS

All study methods are described briefly below and fully detailed in the Supplemental Methods.

Design

The study was performed in four distinct stages (Figure 1). Stage 1 consisted of new microarray TOL gene signature discovery (A) and cross-platform validation (B) (Figure 1) in 118 samples from 101 unique renal transplant recipients (44 TOL, 21 CAN, 36 SI) and 17 HCs. New Agilent 4×44-k whole-genome arrays were performed in 31 samples for identification of the 21 gene set (GEO: GSE45218) (A); cross-microarray platform validations I and II (B) were performed using publicly available microarray data from 87 samples (validation I, n=29, cDNA Lymphochip; validation II, n=58, Affymetrix HG U133 Plus2.0, GSE22229). Next, qPCR validation (stage 2) for the 21 TOL genes was performed in 59 independent samples by standard qPCR. Stage 3 comprised qPCR TOL gene-assay modeling and prediction in 220 independent samples by microfluidic high-throughput qPCR. We trained a minimal TOL gene panel in 70 samples (A) (17 TOL, 53 SI) and we applied the final three-gene model to 150 unique peripheral blood samples from patients receiving long-term immunosuppression to test for the frequency of a potentially tolerant phenotype as defined by the three-gene model (B). For TOL biology analysis (stage 4), we investigated whether the TOL gene signature corresponded to specific blood cell types. Initially, enrichment of the 21 TOL genes was investigated in whole-genome expression data from different blood lymphocytes (GSE1133) in normal patients (A). Predicted TOL gene-enriched cell subsets were then pursued by FACS in blood from TOL patients, HCs, and SI patients (C) and by gene expression analysis in the most informative cell type from the FACS analysis (D). Patient demographic characteristics can be found in Tables 1–3.

Patients and Samples

We collected a total of 326 peripheral blood samples from 282 unique renal allograft recipients and 10 nontransplant HCs. Renal allograft recipients belonged to three distinct clinical phenotypes: (1) TOL, defined as long-term stable graft function without any immunosuppressive drug for >2 years and no history of rejection off immunosuppression (n=69); (2) long-term SI, patients with varying graft function in the absence of acute rejection as defined by the Banff classification receiving double or triple maintenance immunosuppression and with or without history of rejection (n=97); within the SI group, we included 10 CAN samples from patients with chronic allograft histologic injury (biopsy specimens confirmed by Banff classification and Chronic Allograft Damage Index^32,33); and (3) patients with stable graft function and absence of rejection who were receiving SI (n=150): one immunosuppressant (minimal immunosuppression; n=9) or conventional triple immunosuppression (n=141). There were differences in immunosuppression induction: daclizumab, n=38³⁴; antithymocyte globulin, n=166^35,36; alemtuzumab, n=32^37–39; and basiliximab, n=80.^16,40 Immunosuppression maintenance also varied: steroid-free (n=54) or steroid-based ( n=262); CNI free (belatacept, n=59^16,40) or CNI based (tacrolimus or cyclosporine, n=257). The study adhered to the Declarations of Helsinki and Istanbul and was approved by the institutional review boards of California Pacific Medical Center (San Francisco, CA), Stanford University (Stanford, CA), the University of Pittsburgh (Pittsburgh, PA), and the Instituto Nacional de Ciencias Medicas y Nutricion (Mexico City, Mexico), with written informed consent obtained from all participants. De-identified samples were also provided for this study by Bristol-Myers Squibb from a subset of patients receiving belatacept and cyclosporine from the BENEFIT study⁴⁰ as part of an investigator-initiated grant.

Sample Collection, RNA Extraction, and Cell Isolation

Peripheral blood samples were collected in PAXgene Blood RNA Tubes (Qiagen, Hilden, Germany) and in EDTA tubes for gene expression analysis and for isolation of PBMCs for FACS analysis in TOL and SI samples and in leukoreduction system chambers for isolation of PBMCs for FACS analysis in HC samples. Total RNA was extracted from whole blood using the PAXgene Blood RNA Kit or the RNeasy kit (both from Qiagen), RNA concentrations were measured (NanoDrop Technologies, Wilmington, DE), and RNA integrity was assessed (Agilent 2100 Bioanalyzer). Only samples with an RNA integrity number >7 were accepted for further processing. FACS analysis was done using the FACSAria II flow cytometer.

Microarray Preparation and Hybridization

Standard published protocols⁴¹ were used for hybridization of samples onto Agilent Whole Human Genome 4×44-k 60-mer oligonucleotide arrays (G4112F, Agilent Technologies, Santa Clara, CA), using 150ng of total RNA as template/sample. The arrays were scanned on an Agilent scanner and further processed using Agilent Feature Extraction Software (Agilent Technologies).

Microarray Gene Expression Data Processing and Analysis

Using a cutoff for absolute value of log₂ red channel/green channel >0.5 for at least one array Agilent array data were processed and normalized using LOWESS in Gene Spring GX7.3 (Agilent Technologies). SAM and PAM programs with two- and three-class comparison analyses and nested loop cross-validation of PAM with a minimum error rate were used to identify the minimal gene set differentiating TOL from CAN and HCs with an FDR<5%. Agilent microarray raw data have been deposited in GEO (GSE45218). Ingenuity Pathway Analysis was used to assess biologic functions and examine canonical pathways for the genes significant in TOL. BioGPS (http://biogps.org/downloads/, GSE1133) was used to assess cell-specific target genes based on their relative expression as examined in 79 different human cells and tissues. Genes >3 times higher expressed in a given cell type compared with the median expression in all other cell types were considered specific for a specific cell type and tested for significant enrichment by hypergeometric enrichment analysis.⁴²

Probe sets on Agilent, Lymphochip, and Affymetrix Gene chip platforms were reannotated using Array Information Library Universal Navigator¹⁵ and mapped to Gene Entrez IDs before interrogation of the 21 genes on the public domain Affymetrix and on the Lymphochip cDNA array data, which were originally discovered from the Agilent arrays in this study.

qPCR

For TOL qPCR validation and minimal gene set selection, a customized qPCR platform (RT2 QPCR System) was used with standard PCR technology (Superarray, SABiosciences, Qiagen) in which 10 ng total RNA transcribed into cDNA was analyzed in duplicate, each in 25-µl reaction volumes for the 21 genes according to the manufacturer’s instructions. The qPCR primers were custom designed from publicly available mRNA sequences for the 21 TOL genes using Primer 3.0 (http://frodo.wi.mit.edu) and synthesized by SABiosciences.

For subsequent qPCR modeling and prediction studies, we chose the Fluidigm microfluidic high-throughput qPCR platform (Fluidigm Inc., South San Francisco, CA) for sensitive and simultaneous analysis of 96 samples across TOL genes in duplicates with a 14×10³-fold reduction in template required per reaction cycle compared with our previous standard qPCR, due to the use of a target specific amplification step (18 cycles) in this assay. Experiments were performed according to the manufacturer’s protocols and described in the Supplemental Methods.

PCR Data Processing and Statistical-Analyses

Raw Ct values were imported into Excel 2007 (Microsoft, Redmond, WA) for quality control and calculation of relative expression values against 18S and universal human reference RNA (Stratagene; Agilent Technologies) using the delta Ct (ddCt) method as described.⁴³ Genes that were expressed in <80% of samples and samples with expression <90% of genes were excluded from further analyses, resulting in 16 genes and 65 samples (14 TOL, 51 SI). DdCt values were analyzed in Partek Genomic Suite, version 6 (Partek Inc., St. Louis, MO) and GraphPad Prism for data visualization and unsupervised clustering. Development of a mathematical model for detecting and predicting TOL and clinical confounder analyses were performed in the latest versions of R (R 2.14.2); Bioconductor packages were used for further normalization of data, feature selection, and classifier development (Supplemental Methods). F-statistic P values by Fisher exact test or chi-square test were applied for the nested models with and without one of the following factors to test whether any of these were significantly associated with the TOL prediction score: recipient age, categorized as pediatric and adolescent (age ≤18 years) and into adults (age >18 years), time since transplantation, immunosuppression induction, and immunosuppression maintenance protocol.

Isolation of TOL Cell Types by Flow Cytometry for Quantitative Analyses

FACS and quantitative analysis of cell subtypes were performed in PBMCs from 6 SI patients, 5 TOL patients, and 5 HCs; PBMCs were thawed in 15-ml conical tubes containing 12 ml of IMDM medium (GIBCO, Invitrogen) supplemented with 10% human AB serum, 100 U/ml penicillin, 100 μg/ml streptomycin, 2 mM l-glutamine, sodium pyruvate, nonessential amino acids, and 50 μM 2-ME. PBMCs were washed two times in PBS containing 2 mM EDTA and 2% human AB serum before staining with fluorescently labeled monoclonal antibodies against CD3, CD4, CD11c, CD19, CD56, HLA-DR (BD Biosciences, San Jose, CA), CD14, CD16 (BioLegend, San Diego, CA), and CD304 (Miltenyi Biotech, Cologne, Germany) along with propidium iodide (Invitrogen). Samples were analyzed and isolated on a BD FACSAria II flow cytometer.

Disclosures

None.