Abstract
Noninvasive biomarkers are needed to assess immune risk and ultimately guide therapeutic decision-making following kidney transplantation. A requisite step toward these goals is validation of markers that diagnose and/or predict relevant transplant endpoints. The Clinical Trials in Organ Transplantation-01 protocol is a multicenter observational study of biomarkers in 280 adult and pediatric first kidney transplant recipients. We compared and validated urinary mRNAs and proteins as biomarkers to diagnose biopsy-proven acute rejection (AR) and stratify patients into groups based on risk for developing AR or progressive renal dysfunction. Among markers tested for diagnosing AR, urinary CXCL9 mRNA (odds ratio [OR] 2.77, positive predictive value [PPV] 61.5%, negative predictive value [NPV] 83%) and CXCL9 protein (OR 3.40, PPV 67.6%, NPV 92%) were the most robust. Low urinary CXCL9 protein in 6-month posttransplant urines obtained from stable allograft recipients classified individuals least likely to develop future AR or a decrement in estimated glomerular filtration rate between 6 and 24 months (92.5–99.3% NPV). Our results support using urinary CXCL9 for clinical decision-making following kidney transplantation. In the context of acute dysfunction, low values can rule out infectious/immunological causes of injury. Absent urinary CXCL9 at 6 months posttransplant defines a subgroup at low risk for incipient immune injury.
Keywords: Acute rejection, biomarker, chemokines, kidney allograft, kidney graft function
Introduction
Noninvasive biomarkers capable of diagnosing acute transplant rejection and predicting long-term outcomes are needed to improve clinical decision-making following kidney transplantation. This goal requires identification and multicenter validation of reliable biomarkers. Results from single-center studies suggest that several markers may be diagnostically useful for acute kidney allograft rejection or fibrosis (1–15). However, multicenter validations are lacking. There is essentially no published evidence that any biomarker measured during times of clinical quiescence can predict long-term transplant outcome.
To address these unmet needs, we designed the Clinical Trials in Organ Transplantation Protocol-01 (CTOT-01) trial as a prospective, multicenter observational study to validate the diagnostic and predictive utility of a panel of candidate noninvasive biomarkers for transplant outcomes in primary kidney allograft recipients. The findings reported herein indicate that among assays tested, urinary levels of the chemokine CXCL9 are significantly higher in patients with ≥Banff 1A acute rejection (AR) demonstrated in for-cause biopsies (compared to those without rejection), and that the elevations in CXCL9 are commonly detectable up to 30 days before the graft dysfunction is clinically recognized. Perhaps more importantly, absent urinary CXCL9 during acute graft dysfunction rules out AR with a negative predictive value (NPV) of >92%. Low levels of CXCL9 in urine samples obtained from clinically stable allograft recipients 6 months after transplantation also uniquely identify a subgroup of patients at low risk for developing late AR and progressive allograft dysfunction over the ensuing 18 months. Together these findings provide the needed foundation for future work aimed at improving kidney transplant outcomes through biomarker-guided decision-making.
Methods
Study design and oversight
This prospective multicenter observational trial had a target accrual of 300 subjects (280 enrolled) to reach an evaluable group of 255 adult and pediatric recipients of living or deceased first kidney transplants. The CTOT-01 protocol development team was led by Drs. Heeger and Hricik. Clinical and/ or laboratory data were contributed by Drs. Birk, Fairchild, Formica, Gebel, Gibson, Goebel, Heeger, Hricik, Newell, Warshaw, Nickerson, Poggio, Rush and Shihab. Medical safety oversight was provided by Nancy Bridges. Data analysis was the responsibility of David Ikle (with the CTOT-01 team). Data were collected by the investigators and coordinators at each site. All authors vouch for data accuracy and completeness. Each site participated under the auspices of its Institutional Review Board. An independent, NIAID-appointed Data Safety Monitoring Board was responsible for periodic safety review.
Subjects
Adult and pediatric candidates for kidney transplantation having a negative crossmatch were eligible for enrollment. Plans for multi-organ transplantation and/or clinically significant liver disease were exclusion criteria.
Endpoints
The objectives were to determine the relationships between single candidate biomarkers, or combinations thereof, obtained during the first 6 months after kidney transplantation with a composite clinical endpoint (clinically evident or subclinical cellular AR of Banff grade IA or higher (16), increase in Banff chronic sum score ≥2, increase in interstitial fibrosis ≥ 15% (16), graft loss, or death at 6 months after transplant) and/or a change in renal function (>30% decrease in estimated glomerular filtration rate, eGFR) between months 6 and 24 after transplant. A diagnosis of antibody-mediated rejection (AMR) was defined by histological criteria and/or C4d staining with or without a donor specific antibody (DSA) as defined in the Banff criteria (17).
Interventions
Immunosuppression was not standardized in this observational study; doses and levels of immunosuppressive drugs were defined and maintained within therapeutic ranges as per local practice. Calcineurin inhibitor (CNI) levels were not collected or analyzed as part of the study.
Allograft protocol biopsies were obtained at implantation and 6 months after transplantation. “For-cause” biopsies were obtained at the discretion of the subject’s physician.
Blood and urine samples were obtained prior to transplantation, on Day 3, Weeks 1–4 and Months 2–6, 12 and 24 after transplant. Serum creatinine measurements were done at 3, 6, 12 and 24 months. Additional study visits occurred whenever a clinically indicated biopsy was scheduled. Blood and urine samples were collected prior to biopsies or associated treatments. Protocol biopsies were read centrally (by I.W.G., University of Manitoba). Clinically indicated biopsies were read locally for clinical management; tissue from the same biopsy was submitted to and read by the central pathology laboratory. We defined biopsy-proven AR as Banff grade ≥1A and, for some analyses, Banff grade suspicious (16). AR episodes were treated according to local practice.
Surveillance studies for viral infections including BK polyoma virus and cytomegalovirus (CMV) were performed according to local practice at each participating site. Bacterial (e.g. urinary) and viral (e.g. BK polyoma virus, CMV) infections were routinely tested for in patients with acute renal allograft dysfunction as per local standard of care.
Laboratory studies
Urine samples for gene expression, proteomics and chemokines were centrifuged at 2000g for 30 min at 4°C within 4 h of collection. The sediment was washed and stored at −80°C in RNAlater (Life Technologies, Grand Island, NY). Supernatants were divided into aliquots and frozen.
Gene expression profiling
Gene expression profiling by quantitative real-time polymerase chain reaction (qRT-PCR) was performed at the Cleveland Clinic after consultative assistance/training from M. Suthanthiran (5,18). RNA was isolated from urine pellets using the Micro to Midi RNA Purification System (Life Technologies). RNA was reverse transcribed with random hexamers using the TaqMan Reverse Transcription Reagents and normalized to a concentration of 1 µg/100 µL RNA equivalents. Urine cDNA was preamplified for 10 cycles in a 10 µL reaction containing 30 ng cDNA, 1.5 µM of forward and reverse primers for all target genes except 18S, dNTPs, buffer and AmpliTaq Gold DNA Polymerase. Following preamplification, the cDNAs were diluted 10-fold with TE buffer and 2.5 µL of this preamplified cDNA was used for qPCR analysis. TaqMan primers and probes were designed using Applied Biosystems, a division of Life Technologies, Primer Express version 3.0 in either the Suthanthiran or Fairchild lab. Assays for CCL5, IL8 and CXCL10 were custom-designed assays on demand from Applied Biosystems; CCR5 was an inventoried TaqMan assay. All primer sequences (and assay number for CCR5) are shown in Table S1. PCR reactions were performed in duplicate on an Applied Biosystems 7500 Fast System using the fast protocol (95°C for 20s; 40 cycles of 95°C for 3 s, 60°C for 30 s). Each reaction contained 2.5 µL of cDNA, 10 µL of TaqMan Fast Universal PCR master mix, 300 nM primers and 200nM probe in 20 µL. Quantities were calculated from a calibration curve using the mouse BAK amplicon as standard (19).
Urine protein profiling with SELDI-TOF-MS
Urine samples were thawed on ice and vortexed. Analysis was performed as described previously (10,20). Briefly, 5µL of urine was applied to normal phase chips (ProteinChip NP20; Ciphergen, Freemont, CA) and incubated for 30 min in a humidity chamber. Spots were then washed three times with 5µL of HPLC-grade water and air-dried for 10min. One microliter of 35% a-cyano-4-hydroxycinnamic acid (CHCA; Ciphergen) was applied to each spot and air-dried. Chips were read with a surface-enhanced laser desorption ionization time-of-flight mass spectrometric (SELDI-TOF-MS) instrument (ProteinChip Reader II; Ciphergen) in the positive ion mode with the following settings: laser intensity 215; detector sensitivity 6; detector voltage 1700V; 240 shots were collected per sample. Calibration was done externally with a mixture of four proteins with masses ranging from 2 to 16kDa. For comparison, spectra were normalized by total ion current and transformed into gel-view. Urine proteomic profiles were reported as having evidence of cleaved β2-microglobulin peptides present or absent or intact β2-microglobulin present to absent.
Urine was examined for the presence or absence of intact or cleaved β2-microglobulin, which has been associated with tubular injury (2) and AR (1), respectively. In AR, the cleaved but not the intact form of β2-microglobulin was associated with acute clinical rejection (p = 0.05).
Urine ELISA for CXCL9 and CXCL10
Frozen aliquots of urine supernatants were diluted (1:1) in 0.05% Tween-20/0.4% bovine serum albumin in phosphate buffered saline, pH 7.2–7.4, and tested by ELISA for CXCL9 and CXCL10 (R&D Systems, Minneapolis, MN) as per manufacturer’s instructions.
Serum creatinine and eGFR
Serum creatinine levels were assayed centrally (Cleveland Clinic) using an IDMS-traceable analyzer (Hitachi Module D/P, Indianapolis, IN). Missing central laboratory sample results (46/404, 11.3%) were imputed from local creatinine values (Pearson’s correlation = 0.92 between local and central values). A serum creatinine value of 7 mg/dL was imputed for graft loss. eGFR was calculated at 6 and 24 months posttransplant by the modified four-variable MDRD equation (21) for adults and by the modified Schwartz equation (22) for subjects younger than 18 years old.
Statistical methods
Data are summarized using descriptive statistics for categorical (counts/ percentages) and continuous (mean and standard deviations) variables. mRNA and chemokine levels were log10-transformed prior to analysis. Univariate analyses were performed using chi-squared or Fisher exact test for categorical variables and t-tests or analysis of variance (ANOVA) for continuous variables.
Following the general approach of Harrell et al. (23) to avoid overfitting of the models, we sought to reduce the number of candidate markers prior to estimating their association with the outcome. We estimated pairwise Spearman correlations among the nine mRNA markers without reference to any relationships they might have with either of the responses of interest. We made use of all the repeated measures from all the urine samples collected from the study participants over the first 6 months posttransplant. We computed the correlations in two ways: (a) by visit over the 6 months posttransplant and (b) pooled over the visits in the 6 months posttransplant. We focused on those pairwise correlations that indicated strong association (i.e. |ρ| > 0.60) and which remained strongly correlated over the course of the study. Because such correlated markers carry redundant information, a single marker can be selected from a given intercorrelated set based on biological and other considerations without much loss of information. Examination of the correlation matrix in Table S2 indicates that the correlations between granzyme B and CCR5, CCL5 and perforin are all greater than 0.60, suggesting that granzyme B can adequately represent the information contained in the other three markers.
Having reduced the number of candidate markers to a set of markers that carried primarily unique, independent information, we entered them into a multivariable logistic regression model in which the status of a binary endpoint was the binary outcome, for either the AR or the GFR endpoints. This model was fit separately to 500 bootstrap samples of the original study data set with replacement. Each model was reduced by backward elimination to yield 500 reduced models. Any marker that occurred in more than 60% of the reduced models was selected for the final diagnostic model (24). Optimism-adjusted measures of discrimination (area under the curve [AUC]) and calibration (Cox’s slope and intercept and a calibration plot) were obtained from the bootstrap fits (25–28). This approach of model development with internal bootstrap validation has been shown to provide models that have good generalizability (23,26,29). We used this approach to develop separate diagnostic models for AR as a function of urinary mRNA and urinary proteins. Finally, for a final model based on both mRNA and proteins, only three markers were included: granzyme B, CXCL9 and CXCL10. This final model was validated in 500 bootstrap samples and performance measures and calibration plots were created for that model.
Results
Patient characteristics
We enrolled 282 subjects, 280 of whom underwent transplantation (Table 1, Figure 1), including 40 children (14%), 192 recipients of living donor allografts (69%) and 80 African Americans (29%). Pretransplant peak panel reactive antibody in the cohort was low (12.9±25%), and all patients had negative flow cytometry crossmatches at transplantation.
Table 1.
Summary of baseline characteristics for all transplanted subjects
| Characteristics | All subjects (N=280) |
|---|---|
| Donor age | |
| Mean (SD) | 39.8 (13.02) |
| Donor gender | |
| Male | 134 (47.9) |
| Female | 146 (52.1) |
| Donor source | |
| Deceased | 88 (31.4) |
| Living | 192 (68.6) |
| Recipient age (year) | |
| Mean (SD) | 43.5 (17.57) |
| Recipient gender | |
| Male | 168 (60.0) |
| Female | 112 (40.0) |
| Recipient race | |
| Black or African American | 80 (28.6) |
| Other race | 199 (71.1) |
| Unknown or not reported | 1 (0.4) |
| Patient age group | |
| Adult | 240 (85.7) |
| Pediatric | 40 (14.3) |
| Peak PRA | |
| Mean (SD) | 12.9 (24.58) |
| Peak PRA (categorical) | |
| ≤60 | 257 (91.8) |
| >60 | 23 (8.2) |
| Number of HLA mismatches | |
| Mean (SD) | 3.5 (1.81) |
| Induction therapya | |
| Yes | 233 (83.2) |
| No | 47 (16.8) |
| Thymoglobulin | 125 (44.6) |
| Anti-IL-2 | 116 (41.4) |
| Maintenance therapy | |
| CNI | 230 (82.1) |
| Steroid | 210 (75.0) |
| Antimetabolite | 238 (85.0) |
| Sirolimus | 4 (1.4) |
| Other | 3 (1.1) |
| CNI, steroids, antimetabolite | 184 (65.7) |
Eight subjects received both thymoglobulin and anti-IL-2.
CNI, calcineurin inhibitor; PRA, panel reactive antibody.
Figure 1.
Consort diagram illustrating the outcome of patients throughout the course of the study including numbers and results of biopsies performed and numbers of patients who reached the 24-month endpoint.
Immunosuppression was heterogeneous (Table 1): 83% received induction antibody therapy, and discharge medications included a combination of a CNI, a mycophenolic acid derivative and steroids in 66%. Delayed graft function occurred in 9/192 (4.7%) living donor recipients and 11/88 (12.5%) deceased donor recipients. De novo DSAs developed in 11 patients. For-cause biopsies, performed principally to assess patients with acute kidney dysfunction, were obtained on 160 occasions (99 patients). One hundred and fifty of the 160 biopsies yielded adequate tissue for central readings. We categorized the biopsy findings as AR Banff grade ≥1A, Banff grade “suspicious,” infection or other diagnoses (Table 2). Among biopsies showing AR, six had histologic findings suggesting AMR. Of these, four showed mixed antibody and cellular rejection and two showed pure antibody-mediated changes. Comparison of clinical characteristics associated with AR (Table 3) showed that only HLA mismatches, de novo DSA and recipient African American race were associated with higher rates of rejection.
Table 2.
Diagnoses at time of for-cause biopsies
| Rejection | No rejection | |||
|---|---|---|---|---|
| Diagnosis n |
AR Banff >1A 37 |
AR Banff suspicious 32 |
Infection 7a |
Other 74 |
| Biopsy findings for infection and other categories | n (%) | |||
| Normal or nonspecific changes | 44 (54.3) | |||
| Acute tubular necrosis | 8 (9.9) | |||
| Focal and segmental glomerulosclerosis | 5 (6.2) | |||
| Chronic interstitial fibrosis, tubular atrophy and/or glomerulosclerosis | 19 (23.5) | |||
| Glomerulonephritis | 1 (1.2) | |||
| Pyelonephritis | 1 (1.2) | |||
| 3 (3.7) | ||||
AR, acute rejection.
Five bacterial urinary tract infections, one BK polyoma virus, one genital herpes, zero cytomegalovirus.
Table 3.
Clinical patient characteristics stratified by acute rejection and eGFR endpoints
| Characteristics | Acute rejection,1 yes (N=53) |
Acute rejection,1 no (N=212) |
p | GFR endpoint,2 yes (N=25) |
GFR endpoint,2 no (N=177) |
p |
|---|---|---|---|---|---|---|
| Donor age (years) | ||||||
| Mean (SD) | 39.5 (12.86) | 39.6 (12.83) | 0.957 | 38.2 (13.39) | 40.8 (12.83) | 0.338 |
| Donor gender | ||||||
| Male | 25 (47.2) | 100 (47.2) | >0.999 | 7 (28.0) | 81 (45.8) | 0.131 |
| Female | 28 (52.8) | 112 (52.8) | 18 (72.0) | 96 (54.2) | ||
| Donor type | ||||||
| Deceased | 21 (39.6) | 63 (29.7) | 0.188 | 11 (44.0) | 47 (26.6) | 0.097 |
| Living | 32 (60.4) | 149 (70.3) | 14 (56.0) | 130 (73.4) | ||
| Recipient age (years) | ||||||
| Mean (SD) | 41.3 (18.41) | 44.3 (17.37) | 0.260 | 35.2 (18.22) | 44.2 (17.52) | 0.017 |
| Recipient gender | ||||||
| Male | 34 (64.2) | 124 (58.5) | 0.532 | 10 (40.0) | 108 (61.0) | 0.053 |
| Female | 19 (35.8) | 88 (41.5) | 15 (60.0) | 69 (39.0) | ||
| Recipient race | ||||||
| Black or African American | 24 (45.3) | 51 (24.1) | 0.004 | 9 (36.0) | 44 (24.9) | 0.237 |
| Other race | 29 (54.7) | 160 (75.5) | 16 (64.0) | 132 (74.6) | ||
| Unknown or not reported | 0 | 1 (0.5) | NA | 0 | 1 (0.6) | NA |
| Recipient age group | ||||||
| Adult | 42 (79.2) | 184 (86.8) | 0.193 | 18 (72.0) | 152 (85.9) | 0.084 |
| Pediatric | 11 (20.8) | 28 (13.2) | 7 (28.0) | 25 (14.1) | ||
| Peak PRA | ||||||
| Mean (SD) | 14.0 (28.61) | 12.5 (23.71) | 0.701 | 11.8 (27.45) | 11.1 (22.84) | 0.887 |
| Number of HLA mismatches | ||||||
| Mean (SD) | 4.1 (1.49) | 3.3 (1.86) | 0.002 | 3.6 (1.55) | 3.5 (1.76) | 0.679 |
| Induction therapy | ||||||
| Yes | 47 (88.7) | 173 (81.6) | 0.306 | 22 (88.0) | 141 (79.7) | 0.424 |
| No | 6 (11.3) | 39 (18.4) | 3 (12.0) | 36 (20.3) | ||
| De novo DSA | ||||||
| Yes | 6 (11.3) | 5 (2.4) | 0.010 | 3 (12.0) | 5 (2.8) | 0.062 |
| No | 47 (88.7) | 207 (97.6) | 22 (88.0) | 172 (97.2) |
DSA, donor specific antibody; PRA, panel reactive antibody.
Biopsy-proven acute rejection between 0 and 24 months.
Greater than 30% decrease in eGFR between 6 and 24 months.
Bolded values are statistically significant at p< 0.05.
Urinary biomarkers to diagnose AR
To validate and compare urinary biomarkers to diagnose AR, we obtained urine at the time of for-cause biopsies performed for acute graft dysfunction. We performed qRT-PCR on urine sediment RNA and quantified the expression of cDNA for multiple genes previously reported to be elevated during AR: CCR1, CCR5, CXCR3, CCL5 (RANTES), CXCL9 (monokine induced by interferon gamma), CXCL10 (IP-10), IL-8, perforin and granzyme B (2,5,7,8,18). Of 2770 urine samples received by the central laboratory, 2095 (76%) yielded RNA of sufficient quality for analysis, underscoring the technical challenges of isolating urinary mRNA.
Using a correlation analysis (Table S2) to assess potential redundancy among urinary mRNA markers, we observed strong correlations among several of the markers, indicating interdependence. We eliminated highly correlated mRNAs with similar biological functions from subsequent analyses (e.g. perforin was eliminated as it strongly correlated with granzyme B and both are produced by cytotoxic T cells). We determined the strength of the relationships between each remaining mRNA and diagnoses in patients for whom urine samples for mRNAs were available. Analyses revealed higher quantities of mRNAs encoding granzyme B and CXCL9 in those with Banff ≥1A rejection (including AMR) compared to those without rejection (Figure 2A). CXCL9 mRNA values were significantly higher in patients with AR than in those with infection. Samples obtained from patients with “suspicious” grade rejection contained intermediate quantities of each of the mRNAs. Quantities of CCR1, CXCL10 and IL-8 did not differ among the groups. To validate that arbitrary removal of redundant markers did not alter the conclusions, we repeated the analyses replacing, for example, granzyme B with perforin and found that perforin behaved similarly to granzyme B (Table S3).
Figure 2. Urinary mRNAs as biomarkers for biopsy-proven acute rejection (AR) within the first 6 months posttransplant.
(A) Box and whisker plots depicting the median (dash), 25 and 75 percentiles (top and bottom of box)with whiskers extending to 1 and 100 percentile for each gene in patients with infection (Inf), Banff grade suspicious rejection (Susp) Banff grade ≥1A AR or other/no rejection/infection (Other). Results are absolute copy number normalized to the 18S RNA. p-Values are from an ANOVA with Tukey–Kramer multiple comparison corrections. (B) Receiver operator characteristics curves for granzyme B mRNA, CXCL9 mRNA and combined granzyme B + CXCL9 mRNA as diagnostic for Banff grade ≥1A AR using optimal thresholds of granzyme B mRNA = 1.017 × 10−6 and CXCL9 mRNA= 1.925 × 10−6 normalized units (copy number gene/copy number of 18S mRNA). Also see Tables 2, S4 and S5.
We performed logistic regression modeling to clarify the diagnostic utility of urinary granzyme B, CXCL9 and CXCL10 mRNAs for AR, validating results using bootstrap methods (Table 4, Figure 2B). This revealed that quantities of urinary CXCL9 and granzyme B mRNA are each strongly associated with AR but positive predictive values (PPVs) were modest at 61–65%. Including both in the model did not add diagnostic value (Table S4). We repeated the analysis after eliminating biopsies associated with infection and observed no significant change in the diagnostic value of the mRNA markers (Table S5).
Table 4.
Logistic regression and bootstrap validation of urinary markers for diagnosing Banff ≥1A acute rejection1
| Parameter estimates and tests |
ROC-based discrimination measures |
Positive/negative predictive value |
||||||
|---|---|---|---|---|---|---|---|---|
| Model predictors | OR (95% CI) | p-Value | AUC | Sensitivity | Specificity | PPV | NPV | |
| Univariate models | ||||||||
| Granzyme B mRNA | 2.26 (1.30, 3.92) | 0.0039 | 0.730 | 70.8 | 81.6 | 65.4 | 85.1 | |
| CXCL9 mRNA | 2.77 (1.59, 4.80) | 0.0003 | 0.788 | 66.7 | 79.6 | 61.5 | 83.0 | |
| CXCL9 protein | 3.40 (2.12, 5.47) | <0.0001 | 0.856 | 85.2 | 80.7 | 67.6 | 92.0 | |
| CXCL10 protein | 3.25 (1.89, 5.57) | <0.0001 | 0.768 | 74.1 | 86.0 | 71.4 | 87.5 | |
| Multivariate models | ||||||||
| CXCL9 protein | 3.06 (1.82, 5.16) | <0.0001 | 0.864 | 87.0 | 82.2 | 71.4 | 92.5 | |
| Granzyme B mRNA | 1.57 (0.83, 2.99) | 0.1696 | ||||||
| CXCL9 protein | 2.88 (1.68, 4.93) | 0.0001 | 0.885 | 82.6 | 86.7 | 76.0 | 90.7 | |
| CXCL9 mRNA | 1.95 (1.01, 3.73) | 0.0454 | ||||||
| CXCL9 protein | 2.89 (1.68, 4.96) | 0.0001 | 0.869 | 87.0 | 84.4 | 74.1 | 92.7 | |
| CXCL9 mRNA | 2.04 (0.83, 5.00) | 0.1205 | ||||||
| Granzyme B mRNA | 0.94 (0.38, 2.32) | 0.8842 | ||||||
AUC, area under the curve; PPV, positive predictive value; NPV, negative predictive value; OR, odds ratio; ROC, receiver operator characteristics. See Supplemental Statistical Methods for details.
Additional models with and without “suspicious” rejection are shown in Tables S2 and S3.
We also determined the strength of the relationships between CXCL9 and CXCL10 proteins in 2760 urinary supernatants (99% yielded interpretable measurements) and the diagnoses in patients for whom urine samples for the urinary proteins were available (Figure 3A). Mean values for CXCL10 protein were similar in patients with AR and infection, but values for CXCL9 protein were higher in patients with AR versus those with infection or other diagnoses. Logistical regression modeling and bootstrap validation (Table 4) showed that CXCL9 protein diagnosed rejection with an AUC = 0.856, PPV = 67.6% and NPV = 92.0%. Adding CXCL10 protein to the model did not enhance diagnostic utility (Table S4). Eliminating infectious etiologies improved the PPV to 73.3% for AR (Table S5).
Figure 3. Urinary CXCL9 and CXCL10 protein as biomarkers for biopsy-proven acute rejection (AR) and an incipient decrement in GFR.
(A) Box and whisker plots depicting the median (dash), 25 and 75 percentiles (top and bottom of box) with whiskers extending to 0 and 100 percentile for urinary CXCL9 and CXCL10 protein in patients infection (Inf), Banff grade suspicious rejection (Susp), Banff grade ≥1AAR or other/no rejection/infection (Other). p-Values are from an ANOVA with Tukey–Kramer multiple comparison corrections. (B) Receiver operator characteristics curves CXCL9 protein ± granzyme B RNA or CXCL9 mRNA as diagnostic for Banff grade ≥1AAR (optimal threshold for CXCL9 protein is 35pg/mL). See Tables 2, S4 and S5 for additional details. (C) Dot plot depicting individual urinary CXCL9 protein values obtained within 30 days prior to a for-cause biopsy, at the time of the biopsy or within 30 days after the for-cause biopsy in patients biopsied for suspected rejection, stratified by presence (red) or absence (blue) of Banff grade ≥1A AR on the biopsy. Dashed line depicts the mean for each set. *p < 0.0001 versus values for AR (red) at that time point. (D) Dot plot correlating 6-month urinary CXCL9 obtained during clinical quiescence with subsequent diagnosis of AR between 6 and 24 months. (E) Dot plot correlating individual values of 6-month urinary CXCL9 with individual “i” and “t” scores from simultaneous surveillance biopsies. p-Value of <0.0001 refers to an ANOVA comparing the means across Banff scores for each panel.
When we modeled combinations of urinary mRNA and proteins (Table 4, Figure 3B), we found that the combination of CXCL9 protein plus CXCL9 mRNA provided the best results. The usefulness of CXCL9 protein for ruling out rejection (NPV 92.5%) was not impacted by recipient age, HLA mismatch, race or the presence of de novo DSAs (not shown).
To test whether elevated urinary CXCL9 predates the onset of graft dysfunction that precipitated the for-cause biopsies, we analyzed urinary CXCL9 protein in samples obtained prior to the diagnostic biopsies (Figure 3C). We observed elevated CXCL9 concentrations in those with histologically diagnosed AR up to 30 days prior to clinical recognition of graft dysfunction. Urinary CXCL9 values fell within 30 days after treatment (Figure 3C), strengthening the relationship between urinary CXCL9 and graft inflammation.
We examined whether the urinary markers correlated with histological evidence of early progressive graft injury as determined by (a) an increase in Banff chronic sum score ≥2 or (b) an increase in interstitial fibrosis of ≥15%, between the implantation and 6-month surveillance biopsies (n = 156 patients with evaluable biopsies at both time points). These analyses revealed no significant associations among these histological endpoints and any of the markers tested at any time point (alone or in combination, data not shown). Nor were there significant relationships between the absence of any/all of the markers and stability of graft histology (not shown). Together with the previous data, the results suggest that measurements of urinary CXCL9 strongly indicate the presence or absence of active inflammation in the graft.
Previous single-center studies indicated that SELDI-TOF-MS detection of intact and cleaved β2-microglobulin in urine could distinguish AR from other diagnoses (10,11). Our analyses correlating urinary proteomic profiles with the presence of AR in all (for-cause plus surveillance) biopsies trended toward, but did not reach, statistical significance (p = 0.09, n = 188, data not shown).
Six-month urinary CXCL9 stratifies patients into low-versus high-risk subsets for future allograft injury
We tested whether urinary biomarkers measured during clinical quiescence differentiate individuals at risk for future AR. These analyses uncovered a relationship between 6-month urinary CXCL9 and subsequent AR, 6–24 months posttransplant (Figure 3D). Statistical analysis showed odds ratio (OR) = 4.695, sensitivity 83%, specificity 84%, p=0.005, AUC=0.88, NPV=99.3%, PPV 14.7%, independent of age, race, gender, previous AR or DSA (not shown), further supporting the clinical value of the test. None of the patients with low values developed rejection.
We also examined whether 6-month urinary CXCL9 stratified patients into those at high versus low risk for developing late graft dysfunction. We prespecified a clinically significant, >30% decrease in eGFR between 6 and 24 months posttransplant as the endpoint. Twenty-five of 202 evaluable patients (12%) met the endpoint (Table 3). Whereas none of the urinary mRNAs were informative (not shown), our analyses revealed a significant relationship between concentrations of urinary CXCL9 protein obtained at 6 months posttransplant and the GFR endpoint (Table 5). We found that absence of CXCL9 identified those destined to maintain stable renal function with high accuracy, independent of donor type, recipient age or gender, DSA at or before 6 months or 6-month eGFR (Table 5).
Table 5.
Logistic regression and bootstrap validation of 6-month urinary CXCL9 as a correlate of reaching the GFR endpoint (>30% decrease in eGFR between 6 and 24 months)
| n (endpoint+/ endpoint−) |
Parameter estimates and tests |
ROC-based discrimination measures |
Positive/negative predictive value |
|||||
|---|---|---|---|---|---|---|---|---|
| Model predictors | OR (95% CI) | p-Value | AUC | Sensitivity | Specificity | PPV | NPV | |
| 23/163 | CXCL9 protein | 1.929 (1.300, 2.861) | 0.0011 | 0.682 | 56.5 | 76.1 | 25.0 | 92.5 |
| 23/163 | CXCL9 protein GFR |
1.956 (1.314, 2.913) 1.006 (0.985, 1.027) |
0.0010 0.5758 |
0.672 | 60.9 | 75.5 | 25.9 | 93.2 |
| 23/163 | CXCL9 protein DSA |
1.933 (1.286, 2.907) 1.033 (0.245, 4.358) |
0.0015 0.9650 |
0.687 | 60.9 | 74.2 | 25.0 | 93.1 |
| 23/163 | CXCL9 protein Donor type (deceased) |
1.97 (1.32, 2.96) 2.75 (1.08, 6.96) |
0.0010 0.0335 |
0.746 | 87.0 | 57.1 | 22.2 | 96.9 |
| 11/115 | CXCL9 protein Recipient age |
1.94 (1.30, 2.90) 0.97 (0.95, 0.997) |
0.0011 0.0299 |
0.711 | 56.5 | 69.3 | 20.6 | 91.9 |
| 10/112 | CXCL9 protein Recipient gender (female) |
1.97 (1.32, 2.95) 2.29 (0.91, 5.75) |
0.0009 0.0779 |
0.721 | 82.6 | 46.6 | 17.9 | 95.0 |
AUC, area under the curve; DSA, donor specific antibody; NPV, negative predictive value; OR, odds ratio; PPV, positive predictive value; ROC, receiver operator characteristics.
We obtained 170 six-month protocol biopsies with adequate tissue from the 202 patients who completed the study. Subclinical AR Banff grade ≥1A was observed in eight patients. When we correlated urinary CXCL9 protein concentrations with individual inflammation (“i”) and tubulitis (“t”) components of the acute Banff scores (Figure 3E), we observed a direct relationship between the scores and urine CXCL9 protein. This relationship persisted for “t” scores when patients with AR on the biopsy were eliminated from the analysis (Figure S1).
Discussion
This unique multicenter study of 280 kidney transplant recipients indicates that urine biomarkers have utility for ruling out AR in kidney transplant recipients with acute graft dysfunction and for identifying patients likely to maintain stable allograft function without rejection episodes. Extending and independently validating previous work by others (2,5,18), we showed that measurements of urinary cell-derived granzyme B and CXCL9 mRNA each have utility for diagnosing AR in patients with graft dysfunction, but with PPVs of 61–67% positive urinary markers are unlikely to overcome the need for a diagnostic biopsy. In contrast, the high NPVs (>92%) suggest that it may be possible to avoid diagnostic biopsies in patients with impaired kidney function but low urine cell granzyme B/CXCL9 mRNA concentrations, because AR is highly unlikely. The bootstrap validations showed no diagnostic benefit of combining granzyme B and CXCL9 mRNAs over CXCL9 mRNA alone, consistent with the concept that both are produced by inflammatory infiltrates that mediate cellular AR.
In addition to demonstrating that low urinary CXCL9 protein could be used to rule out AR, our new data suggest that measurements of urinary CXCL9 protein are better than the urinary mRNAs tested for this purpose. The apparent superiority of CXCL9 protein is likely due to technical limitations of isolating high-quality mRNA from urine samples. This, along with the familiarity of clinical laboratories with ELISAs, favors using CXCL9 protein over mRNA as a noninvasive test to diagnose AR.
Combined measurements of CXCL9 protein and CXCL9 mRNA provided the best PPV (71.4%) and NPV (92.5%) for diagnosing (or ruling out) AR, suggesting that more sensitive (potentially earlier) detection of mRNA elevations enhances diagnostic utility. The kinetic observations showing that urinary CXCL9 protein is elevated ~30 days prior to clinical detection of AR (Figure 3C) indicate that CXCL9 can detect intragraft inflammation before overt renal dysfunction occurs, possibly permitting earlier therapy. We also observed that urinary CXCL9 protein levels decrease after treatment for rejection (Figure 3C), suggesting that serial measurements may be useful to identify adequate resolution of AR; additional studies are needed to confirm this.
Consistent with the concept that urinary CXCL9 detects subclinical injury, we observed that its presence in 6-month posttransplant urine samples (in the absence of clinically discernible graft dysfunction) correlated directly with biopsy evidence of tubulitis (Figure 3E), and its absence stratified patients into low-risk subgroups for future AR. Infections, especially those that directly involve the urinary tract or renal parenchyma (e.g. BK polyoma virus nephropathy or bacterial urinary tract infection), can elevate urinary CXCL9 (30). Although the prevalence of infection coincident with acute graft dysfunction in our cohort was low, our data show that patients with AR have higher levels of urinary CXCL9 than those with infection. In clinical practice, BK viral infection and urinary tract infections can be readily diagnosed, and should be ruled out in any patient with graft dysfunction. Low urinary CXCL9 protein in patients with renal dysfunction strongly correlates with absence of AR or infection; confirmation of this relationship in a prospective study might preclude the need for a biopsy in many patients.
Another novel result is that low 6-month urinary CXCL9 protein identifies patients without subclinical allograft injury and who are most likely to maintain stable kidney function. The data support the need for testing the utility of urinary CXCL9 protein measurements as a safety net to guide changes in immunosuppression.
Overall, the results from CTOT-01 indicate that urinary CXCL9 protein is an excellent marker for excluding acute kidney allograft rejection and for stratifying patients in low-versus high-risk subsets for incipient allograft injury. Low urinary CXCL9 indicates low acute immunological risk and is a valuable biomarker to identify patients destined to display stable long-term allograft function.
Supplementary Material
Acknowledgments
The study was supported by National Institutes of Health U01 grant AI63594-06 awarded to P. Heeger. The CTOT-01 consortium members thank the following personnel for the support of the work: Cincinnati Children’s Hospital, Cincinnati, OH: Barbara Logan; Cleveland Clinic, Cleveland, OH: Jennifer Czerr and Leslie Iosue; Emory University Medical Center, Atlanta, GA: Brandi Johnson, Margret Kamel and Amy S. Newell; Icahn School of Medicine at Mount Sinai, New York, NY: Dean Firkus, Brandy Haydel, Neha Karajgikhar, Sherif Mikhail, Katya Ostrow, Yasir Qureshi, Jason Rothfeld, Jennifer Smar, Paulina Trzcinka, Rosie Wickham, Tina Yao and Praeophayom Wauhop; University Hospitals Case Medical Center, Cleveland, OH: Victoria Rodriguez, Maureen Tessman and Tracey Lee; University of Manitoba, Winnipeg Manitoba, Canada: Jennifer Bestland, Iga Dembinski, Shirley Frederickson, Susan McMurrich and Myrna Ross.
Abbreviations
- AMR
antibody-mediated rejection
- AR
acute rejection
- AUC
area under the curve
- CMV
cytomegalovirus
- CNI
calcineurin inhibitor
- CTOT
clinical trials in organ transplantation
- DSA
donor specific antigen
- eGFR
estimated glomerular filtration rate
- NPV
negative predictive value
- PPV
positive predictive value
- PRA
panel reactive antibody
- qRT-PCR
quantitative real-time polymerase chain reaction
- ROC
receiver operator characteristics
- SELDI-TOF-MS
surface-enhanced laser desorption ionization time-offlight mass spectrometric
Footnotes
Disclosure
The authors of this manuscript have no conflicts of interest to disclose as described by the American Journal of Transplantation.
Supporting Information
Additional Supporting Information may be found in the online version of this article at the publisher’s web site.
References
- 1.Hauser IA, Spiegler S, Kiss E, et al. Prediction of acute renal allograft rejection by urinary monokine induced by IFN-gamma (MIG) J Am Soc Nephrol. 2005;16:1849–1858. doi: 10.1681/ASN.2004100836. [DOI] [PubMed] [Google Scholar]
- 2.Ho J, Wiebe C, Gibson IW, Rush DN, Nickerson PW. Immune monitoring of kidney allografts. Am J Kidney Dis. 2012;60:629–640. doi: 10.1053/j.ajkd.2012.01.028. [DOI] [PubMed] [Google Scholar]
- 3.Hu H, Aizenstein BD, Puchalski A, Burmania JA, Hamawy MM, Knechtle SJ. Elevation of CXCR3-binding chemokines in urine indicates acute renal-allograft dysfunction. Am J Transplant. 2004;4:432–437. doi: 10.1111/j.1600-6143.2004.00354.x. [DOI] [PubMed] [Google Scholar]
- 4.Lazzeri E, Rotondi M, Mazzinghi B, et al. High CXCL10 expression in rejected kidneys and predictive role of pretransplant serum CXCL10 for acute rejection and chronic allograft nephropathy. Transplantation. 2005;79:1215–1220. doi: 10.1097/01.tp.0000160759.85080.2e. [DOI] [PubMed] [Google Scholar]
- 5.Li B, Hartono C, Ding R, et al. Noninvasive diagnosis of renal-allograft rejection by measurement of messenger RNA for perforin and granzyme B in urine. N Engl J Med. 2001;344:947–954. doi: 10.1056/NEJM200103293441301. [DOI] [PubMed] [Google Scholar]
- 6.Matz M, Beyer J, Wunsch D, et al. Early post-transplant urinary IP-10 expression after kidney transplantation is predictive of short-and long-term graft function. Kidney Int. 2006;69:1683–1690. doi: 10.1038/sj.ki.5000343. [DOI] [PubMed] [Google Scholar]
- 7.Panzer U, Reinking RR, Steinmetz OM, et al. CXCR3 and CCR5 positive T-cell recruitment in acute human renal allograft rejection. Transplantation. 2004;78:1341–1350. doi: 10.1097/01.tp.0000140483.59664.64. [DOI] [PubMed] [Google Scholar]
- 8.Qin S, Rottman JB, Myers P, et al. The chemokine receptors CXCR3 and CCR5 mark subsets of T cells associated with certain inflammatory reactions. J Clin Invest. 1998;101:746–754. doi: 10.1172/JCI1422. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 9.Schaub S, Nickerson P, Rush D, et al. Urinary CXCL9 and CXCL10 levels correlate with the extent of subclinical tubulitis. Am J Transplant. 2009;9:1347–1353. doi: 10.1111/j.1600-6143.2009.02645.x. [DOI] [PubMed] [Google Scholar]
- 10.Schaub S, Rush D, Wilkins J, et al. Proteomic-based detection of urine proteins associated with acute renal allograft rejection. J Am Soc Nephrol. 2004;15:219–227. doi: 10.1097/01.asn.0000101031.52826.be. [DOI] [PubMed] [Google Scholar]
- 11.Schaub S, Wilkins JA, Antonovici M, et al. Proteomic-based identification of cleaved urinary beta2-microglobulin as a potential marker for acute tubular injury in renal allografts. Am J Transplant. 2005;5:729–738. doi: 10.1111/j.1600-6143.2005.00766.x. [DOI] [PubMed] [Google Scholar]
- 12.Tatapudi RR, Muthukumar T, Dadhania D, et al. Noninvasive detection of renal allograft inflammation by measurements of mRNA for IP-10 and CXCR3 in urine. Kidney Int. 2004;65:2390–2397. doi: 10.1111/j.1523-1755.2004.00663.x. [DOI] [PubMed] [Google Scholar]
- 13.Wang T, Dai H, Wan N, Moore Y, Dai Z. The role for monocyte chemoattractant protein-1 in the generation and function of memory CD8+ T cells. J Immunol. 2008;180:2886–2893. doi: 10.4049/jimmunol.180.5.2886. [DOI] [PubMed] [Google Scholar]
- 14.Hirt-Minkowski P, Amico P, Ho J, et al. Detection of clinical and subclinical tubulo-interstitial inflammation by the urinary CXCL10 chemokine in a real-life setting. Am J Transplant. 2012;12:1811–1823. doi: 10.1111/j.1600-6143.2012.03999.x. [DOI] [PubMed] [Google Scholar]
- 15.Ho J, Rush DN, Karpinski M, et al. Validation of urinary CXCL10 as a marker of borderline, subclinical, and clinical tubulitis. Transplantation. 2011;92:878–882. doi: 10.1097/TP.0b013e31822d4de1. [DOI] [PubMed] [Google Scholar]
- 16.Solez K, Colvin RB, Racusen LC, et al. Banff 07 classification of renal allograft pathology: Updates and future directions. Am J Transplant. 2008;8:753–760. doi: 10.1111/j.1600-6143.2008.02159.x. [DOI] [PubMed] [Google Scholar]
- 17.Mengel M, Sis B, Haas M, et al. Banff 2011 Meeting Report: New concepts in antibody-mediated rejection. Am J Transplant. 2012;12:563–570. doi: 10.1111/j.1600-6143.2011.03926.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 18.Muthukumar T, Dadhania D, Ding R, et al. Messenger RNA for FOXP3 in the urine of renal-allograft recipients. N Engl J Med. 2005;353:2342–2351. doi: 10.1056/NEJMoa051907. [DOI] [PubMed] [Google Scholar]
- 19.Kawai T, Cosimi AB, Spitzer TR, et al. HLA-mismatched renal transplantation without maintenance immunosuppression. N Engl J Med. 2008;358:353–361. doi: 10.1056/NEJMoa071074. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 20.Ho J, Lucy M, Krokhin O, et al. Mass spectrometry-based proteomic analysis of urine in acute kidney injury following cardiopulmonary bypass: A nested case-control study. Am J Kidney Dis. 2009;53:584–595. doi: 10.1053/j.ajkd.2008.10.037. [DOI] [PubMed] [Google Scholar]
- 21.Levey AS, Coresh J, Greene T, et al. Using standardized serum creatinine values in the modification of diet in renal disease study equation for estimating glomerular filtration rate. Ann Intern Med. 2006;145:247–254. doi: 10.7326/0003-4819-145-4-200608150-00004. [DOI] [PubMed] [Google Scholar]
- 22.Schwartz GJ, Munoz A, Schneider MF, et al. New equations to estimate GFR in children with CKD. J Am Soc Nephrol. 2009;20:629–637. doi: 10.1681/ASN.2008030287. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 23.Harrell FE, Jr., Lee KL, Mark DB. Multivariable prognostic models: Issues in developing models, evaluating assumptions and adequacy, and measuring and reducing errors. Stat Med. 1996;15:361–387. doi: 10.1002/(SICI)1097-0258(19960229)15:4<361::AID-SIM168>3.0.CO;2-4. [DOI] [PubMed] [Google Scholar]
- 24.Austin PC, Tu JV. Bootstrap methods for developing predictive models. Am Statistician. 2004;58:131–137. [Google Scholar]
- 25.Cox DR. Two further applications of a model for binary regression. Biometrika. 1958;45:562–565. [Google Scholar]
- 26.Steyerberg EW. A practical approach to development, validation, and updating. New York: Springer; 2009. Clinical prediction models. [Google Scholar]
- 27.Steyerberg EW, Vickers AJ, Cook NR, et al. Assessing the performance of prediction models: A framework for traditional and novel measures. Epidemiology. 2010;21:128–138. doi: 10.1097/EDE.0b013e3181c30fb2. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Tripepi G, Jager KJ, Dekker FW, Zoccali C. Statistical methods for the assessment of prognostic biomarkers (part II): Calibration and re-classification. Nephrol Dial Transplant. 2010;25:1402–1405. doi: 10.1093/ndt/gfq046. [DOI] [PubMed] [Google Scholar]
- 29.Efron B, Tibshirani RJ. An introduction to the bootstrap. Monographs on statistics and applied probability. New York: Chapman and Hall/CRC Press, LLC; 1993. [Google Scholar]
- 30.Jackson JA, Kim EJ, Begley B, et al. Urinary chemokines CXCL9 and CXCL10 are noninvasive markers of renal allograft rejection and BK viral infection. Am J Transplant. 2011;11:2228–2234. doi: 10.1111/j.1600-6143.2011.03680.x. [DOI] [PMC free article] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.