Urinary Biomarkers and Kidney Transplant Rejection: Fine-Tuning the Radar



This editorial addresses the promise of urinary biomarkers in refining the diagnosis of acute rejection and their potential clinical utility. See article by Hricik et al on page 2634.

Acute rejection in the transplanted kidney is diagnosed in the clinic using a combination of renal dysfunction, renal biopsy and alloantibody levels [1]. Despite its invasive nature and high interobserver variability, the renal biopsy remains the gold standard for the diagnosis of acute rejection [2]. Because of these limitations of the renal biopsy, the goal of discovering accurate and reproducible biomarkers in the urine and blood among kidney transplant recipients has been an important line of research [3-5].

In this issue of AJT, Hricik et al [6] report on the findings of a multicenter observational study conducted under the aegis of the NIH-sponsored Collaborative Trials in Organ Transplantation study consortium (CTOT). In the CTOT-01 study, Hricik et al discovered that low urinary CXCL9 (a T cell chemoattractant induced by interferon gamma) at 6 months posttransplantation was associated with a lower probability of development of acute rejection or a decline in renal allograft function over subsequent follow-up to 24 months in the 280 kidney transplant recipients studied [6]. The authors concluded that low urinary CXCL9 protein could be used as a biomarker to identify transplant recipients at low risk for future immunologic events. The authors are to be applauded on the conduct and reporting of their multicenter prospective inquiry. However, the study raises several questions that bear reflection.

The CTOT-01 study examined, serially, urinary mRNA of several candidate cytokine and effector molecules as well as proteins pertaining to, in the context of a 6-month per protocol allograft biopsy, biopsies performed for cause and clinical follow-up.

Using this approach, the positive predictive value (PPV) of urinary CXCL9 protein for acute rejection was modest at 67% whereas the negative predictive value (NPV) was much higher at 92%. The performance of urinary CXCL9 mRNA in the prediction of acute rejection was inferior (PPV, 61.5%; NPV, 83%). This discrepancy in urinary CXLC9 protein and mRNA was attributed to the difficulties in using urine as a matrix for mRNA samples. Further, the authors propose that low urinary CXCL9 at Month 6 posttransplantation would allow the identification of subjects at low risk for acute rejection or decline in GFR between 6 and 24 months posttransplantation (NPV, 92.5–99.3%). The CTOT-01 study population was at low immunologic risk; the mean panel reactive antibody of the subjects was 12.9%. Most recipients received living donor transplants, most received antibody induction and all flow crossmatches at transplant were negative. Notably, as reported by Hricik et al [6], in the CTOT-1 population, rejection (grade suspicious or higher) was reported in 25/170 (13.8%) of 6-month protocol biopsies and in 43/79 (54.43%) of for-cause biopsies by Month 6.

These findings are an important contribution to the development of biomarkers in transplantation and certainly are a welcome addition to all clinicians as a piece in the puzzle of clinical decision making. However, the study also teaches us another important lesson. In the modern era of transplantation, 1-year acute rejection rates in low-immunologic risk patients should approach 10%. In this setting, if one were to consider using a biomarker as a substitute for the invasive renal biopsy, what is desirable is a test not only with a high NPV but also with a high PPV. The ideal test would help detect disease early in its clinical course without an excess of false positives. In this case, the marker must have high sensitivity and specificity. As the baseline event rate (acute rejection) is expected to be low with modern immunosuppression in a low-risk population, a high negative predictive value for rejection while indisputably of value does not answer what one may consider the even more difficult question of who will develop disease (rejection/injury).

Clinical decisions are in large part based on conditional probabilities given the known pretest probability (prevalence) and the ability to distinguish when a test is either a true positive or true negative. Given this one must invoke Bayes theorem, simply stated mathematically as, PPV given Prevalence = Sensitivity × Prevalence/Sensitivity × Prevalence +(1 − Prevalence)(1 − specificity) [7].

Assuming an overall occurrence of rejection of 10% in a population, if the sensitivity and specificity relevant to CXCL9 protein reported by Hricik et al [6] were to be substituted in the equation above, the PPV of CXCL9 protein for acute rejection would be about 36%.

Even applying these findings of CTOT-01 to populations at higher immunologic risk is not a straightforward issue as parameter estimates obtained from a low-risk population may not perform well in a population different from that used to derive the biomarker that is limited external validity [8].

Receiver operating characteristic curves (ROC curves) are used to compare the performance of diagnostic tests as an extrapolation from initial development in the military [9]. In the context of ROC curves, the area under the ROC curve (AUC ROC) would approach 1.0 for the ideal test and 0.5 for when the test is equal to or no better than a coin flip. In CTOT-01, the AUC ROC for the prediction of acute rejection averaged 0.86. Although the authors do not provide 95% confidence intervals for the point estimate of the AUC ROC curve, this would reasonably be expected to lie between 0.6 (somewhat better than a coin flip) and 0.95 (excellent test). Interestingly, adding granzyme B protein measurements to the analysis did not add to the performance of the test in a significant manner. This latter problem is likely related to the overlap in biologic pathways in acute rejection that form the basis for these biomarkers that, in turn, manifested as collinearity in the analyses.

Despite overlap in the levels of urinary CXCL9 protein across the range spanning transition normal allograft histology and acute rejection, a dose dependency underlying relationship was discernible, higher urinary CXCL9 levels being associated with higher Banff lesion sum scores, a measure of inflammatory cell burden in the allograft [6]. Interestingly, urinary CXCL9 protein declined with clinical resolution of rejection [6]. However, we need to consider these findings in the context of several limitations: (1) immunosuppressive drug concentrations at the time of biopsy or treatment of rejection were not collected or analyzed; (2) monitoring for CMV and polyoma virus were conducted per center protocol but not included in the analyses and (3) the large number of biomarkers and the relatively low event rate could inflate statistical significance given multiple comparisons.

The findings of CTOT-01 represent an important advance in our understanding of the relationship between biomarkers and clinically significant events in renal transplantation. Another salient contribution of this study is in demonstrating the difficulties in discovering the ideal biomarker in transplantation. So, how should we view the findings of CTOT-01 in the clinic?

  1. Can urinary CXCL9 be used to decrease the number of patients who need renal biopsies?

    The findings of CTOT-01 lead us to believe that this is a very real possibility. As a low CXCL9 decreases the conditional probability of acute rejection; it could help guide the clinician in better selecting patients with graft dysfunction for a biopsy. However, the validity of this inference needs verification in more extensive studies.

  2. Is CXCL9 an adequate clinical tool to screen for rejection or injury?

    As the authors candidly report, we are still far away from this elusive goal. In an era of low acute rejection rates, this task is a daunting one. In order for any biomarker to achieve this end, both excellent NPV and PPV are needed for events with clinically relevant proximity to the event. Such a test would need to be sensitive and specific and highlights the need for continuing this line of research.

  3. Could the absence of urinary CXCL9 protein help define a subset of patients whose immunosuppression can be reduced further as they could be considered to be at very low risk for rejection?

    The study design of CTOT-1 per se does not allow us to answer this question as this will necessarily require a study design that is tailored to answer that specific question.

    Such a study would include the need to define the dose–response and concentration effect relationships between immunosuppression type and intensity and urinary CXC9 protein and rejection rates in prospective randomized studies.

    These issues further highlight the difficulties in discovering the ideal biomarker in transplantation and maximizing its clinical utility.

CTOT-01 is a laudable first step toward biomarker-guided clinical decision making in the transplant clinic. The results of CTOT-01 are an excellent contribution that also serve to inform us of the complexity of biomarker discovery and point to the need for further study of relevant and reproducible biomarkers and defining their utility in clinical transplantation.


The authors of this manuscript have no relevant conflicts of interest as described by the American Journal of Transplantation.