Normalization Choice Impacts Molecular Rejection Diagnosis in Kidney Transplants

Refining Molecular Diagnostics in Allograft Surveillance

The long-term survival of kidney allografts remains contingent upon the precise detection and management of immune-mediated rejection. While traditional histology serves as the diagnostic cornerstone, the integration of molecular profiling has become essential for characterizing the complex immune responses that drive graft injury [1, 2]. These molecular insights are particularly critical given the intricate signaling pathways, such as the Janus kinase/signal transducer and activator of transcription (JAK/STAT) system, that mediate both innate and adaptive alloimmunity [3]. However, the clinical utility of these high-throughput genetic tools depends entirely on the stability and reproducibility of the underlying data [4, 1]. As transplant medicine moves toward standardized gene-expression panels, ensuring that analytical methods do not introduce diagnostic artifacts is paramount for informed therapeutic decision-making [5, 6]. A new multicenter study now evaluates how different data processing techniques influence the accuracy of these molecular diagnoses in real-world clinical cohorts.

A Multicenter Evaluation of Analytical Pipelines

The Banff 2022 classification now formally endorses intragraft gene-expression profiling for the diagnosis of rejection, specifically utilizing the Banff Human Organ Transplant (B-HOT) consensus gene panel. This panel provides a standardized set of genetic markers designed to identify immune activity within the graft, offering a molecular layer of evidence to supplement traditional microscopy. To assess the clinical reliability of this tool, researchers conducted a comprehensive evaluation of ten different normalization methods. These methods are statistical techniques used to remove technical noise from biological data, ensuring that the variations observed in gene expression reflect true pathology rather than laboratory artifacts or batch effects. The study utilized a large dataset of 868 kidney allograft biopsies collected from nine different centers across Europe and North America, providing a robust, multicenter foundation for the analysis. All 868 biopsies underwent traditional Banff-grading by pathologists and were simultaneously profiled using the nCounter platform, a digital color-coded barcode technology used to count individual mRNA molecules without the need for amplification. To ensure the findings were reproducible, the researchers structured the study population into three distinct groups: a derivation cohort of 441 biopsies, an internal validation cohort of 186 biopsies, and an external validation cohort of 241 biopsies. By testing ten different analytical pipelines across these cohorts, the study aimed to determine how the choice of data normalization impacts the diagnostic accuracy for both antibody-mediated rejection and T cell-mediated rejection. This rigorous design allows clinicians to understand which statistical approaches maintain the highest fidelity when translating raw genetic counts into a definitive clinical diagnosis.

Stability and Concordance Across Platforms

To determine the clinical utility of the various normalization pipelines, the researchers first assessed their downstream impact on gene count stability, which measures how consistently a method removes technical variance across different samples. Beyond internal consistency, the study evaluated the methods for differential expression and cross-platform concordance with RNA-sequencing data. This process involved comparing the results from the nCounter platform to RNA-sequencing (a more comprehensive method of measuring the entire transcriptome) to ensure that the gene expression signatures identified were biologically authentic and not platform-specific. The researchers found that most normalization methods successfully improved count stability and demonstrated high concordance with RNA-sequencing for overall gene expression, suggesting that the majority of these analytical tools can reliably capture the underlying molecular state of the kidney allograft. However, the choice of statistical pipeline significantly influenced the ability to detect specific immune signals. While most methods produced robust differential expression signatures that matched those detected by RNA-sequencing, two specific approaches, RUVSeq and RCRNorm, identified fewer differentially expressed genes. These are genes that show statistically significant changes in expression levels between rejection and non-rejection groups, serving as the primary markers for diagnosis. Furthermore, these two methods showed lower concordance with RNA-sequencing compared to the other pipelines tested. For the practicing clinician, these findings indicate that while the Banff Human Organ Transplant panel is a powerful diagnostic tool, the use of overly complex or poorly suited normalization algorithms like RUVSeq or RCRNorm may inadvertently mask critical molecular markers of injury, potentially leading to a less sensitive diagnosis of the underlying rejection process.

Diagnostic Accuracy for AMR and TCMR

The clinical utility of molecular profiling depends on the ability of statistical models to accurately differentiate between rejection and non-rejection states. To determine this, the researchers assessed each normalization method for the discrimination and calibration of predictive models for both antibody-mediated rejection (AMR) and T cell-mediated rejection (TCMR). Discrimination refers to the ability of the test to correctly sort patients into those with and without the condition, while calibration measures how closely the predicted probabilities of the model match the actual observed frequency of rejection. In the overall validation cohort of 427 kidney allograft biopsies, diagnostic performance remained consistently high across several standardized pipelines, including nSolver-based approaches, nanostringr, NanoStringDiff, MetaNorm, and RCRNormFast. For these high-performing methods, the Area Under the Receiver Operating Characteristic curve (AUROC), a metric where 1.0 represents a perfect diagnostic test and 0.5 represents random chance, ranged from 0.88 to 0.91 for antibody-mediated rejection. The Area Under the Precision-Recall Curve (AUPRC), which is particularly useful for evaluating diagnostic accuracy in datasets where the condition may be less common, ranged from 0.86 to 0.89 for AMR. Performance was similarly robust for T cell-mediated rejection, with an AUROC between 0.90 and 0.92 and an AUPRC between 0.78 and 0.83. These figures suggest that most standard normalization techniques provide the high level of diagnostic precision required for clinical decision-making in transplant medicine. However, the study identified a significant decline in diagnostic performance when using more complex or less compatible normalization algorithms. The RCRNorm method, in particular, failed to provide clinically useful data, yielding an AMR AUROC of 0.55 and an AUPRC of 0.41. Its performance for T cell-mediated rejection was equally poor, with an AUROC of 0.53 and an AUPRC of 0.18, essentially rendering the molecular panel no better than a coin flip for diagnosis. Furthermore, the RUVSeq method showed reduced performance specifically for TCMR, with an AUROC between 0.84 and 0.85 and an AUPRC between 0.64 and 0.65. For the practicing clinician, these results emphasize that while molecular diagnostics are highly effective, the choice of data processing is not merely a technicality; using suboptimal methods like RCRNorm or RUVSeq can lead to a substantial loss of diagnostic accuracy, potentially resulting in the misclassification of active rejection.

Clinical Reliability and Calibration

For a clinician to rely on a molecular diagnostic tool, the numerical risk score provided must accurately reflect the actual clinical state. This relationship is known as calibration, which is the agreement between predicted probabilities and observed outcomes. In this study, the researchers found that calibration was satisfactory for most methods tested, suggesting that the majority of available analytical pipelines provide reliable probability estimates for both antibody-mediated and T cell-mediated rejection. When calibration is high, a clinician can trust that a 70 percent probability of rejection generated by the molecular panel truly corresponds to a 70 percent frequency of rejection in similar patient populations. However, the choice of normalization method significantly influenced the reliability of these risk estimates. The researchers observed that calibration was poor for RCRNorm and for T cell-mediated rejection models after RUVSeq normalization. In these instances, the predicted probability of rejection generated by the model did not align with the actual frequency of rejection observed in the biopsy samples. Such discrepancies could lead to clinical misinterpretation, where a high-risk score might not correspond to a high clinical probability of graft rejection, potentially complicating treatment decisions and leading to inappropriate immunosuppressive adjustments. Ultimately, the study demonstrates that the complexity of the statistical pipeline does not necessarily translate to better clinical utility. The findings indicated that simpler normalization approaches consistently showed higher diagnostic accuracy than complex methods like RCRNorm and RUVSeq for molecular diagnostics based on the Banff Human Organ Transplant (B-HOT) consensus gene panel. For the practicing nephrologist, these results suggest that standardized, straightforward data processing methods are currently the most dependable choice for integrating molecular rejection signatures into routine clinical practice, ensuring that the diagnostic output remains both precise and clinically actionable.

References

1. Mengel M, Loupy A, Haas M, et al. Banff 2019 Meeting Report: Molecular diagnostics in solid organ transplantation–Consensus for the Banff Human Organ Transplant (B-HOT) gene panel and open source multicenter validation. American Journal of Transplantation. 2020. doi:10.1111/ajt.16059

2. Sun L, Su Y, Jiao A, Wang X, Zhang B. T cells in health and disease. Signal Transduction and Targeted Therapy. 2023. doi:10.1038/s41392-023-01471-y

3. Hu X, Li J, Fu M, Zhao X, Wang W. The JAK/STAT signaling pathway: from bench to clinic. Signal Transduction and Targeted Therapy. 2021. doi:10.1038/s41392-021-00791-1

4. Kumar MA, Baba SK, Sadida HQ, et al. Extracellular vesicles as tools and targets in therapy for diseases. Signal Transduction and Targeted Therapy. 2024. doi:10.1038/s41392-024-01735-1

5. Mack CL, Adams D, Assis DN, et al. Diagnosis and Management of Autoimmune Hepatitis in Adults and Children: 2019 Practice Guidance and Guidelines From the American Association for the Study of Liver Diseases. Hepatology. 2019. doi:10.1002/hep.31065

6. Kotton CN, Kamar N, Wojciechowski D, et al. The Second International Consensus Guidelines on the Management of BK Polyomavirus in Kidney Transplantation. Transplantation. 2024. doi:10.1097/tp.0000000000004976