Normalization of nCounter Gene Expression Data Alters Molecular Diagnostics in Kidney Transplantation

Area of Science:

• Transplant immunology
• Genomics
• Bioinformatics

Background:

• The Banff 2022 classification recommends intragraft gene-expression profiling for kidney transplant rejection diagnosis.
• Standardized analytical pipelines, particularly data normalization, are lacking for clinical implementation of gene expression profiling.
• The impact of normalization methods on diagnostic performance remains undetermined.

Purpose of the Study:

• To evaluate the impact of ten different normalization methods on kidney allograft biopsy gene expression data.
• To assess how normalization affects gene count stability, differential expression, and concordance with RNA-seq.
• To determine the influence of normalization on the diagnostic performance (discrimination and calibration) of predictive models for antibody-mediated rejection (AMR) and T cell-mediated rejection (TCMR).

Main Methods:

• Analysis of 868 kidney allograft biopsies from nine centers, profiled using the Banff Human Organ Transplant (B-HOT) consensus gene panel on nCounter.
• Evaluation of ten normalization methods across derivation, internal, and external validation cohorts.
• Assessment of downstream effects on gene count stability, differential expression, RNA-seq concordance, and diagnostic model performance for AMR and TCMR.

Main Results:

• Most normalization methods improved gene count stability and showed high concordance with RNA-seq data.
• nSolver-based approaches, nanostringr, NanoStringDiff, MetaNorm, and RCRNormFast demonstrated high diagnostic performance for both AMR and TCMR.
• RUVSeq and RCRNorm showed reduced performance, particularly for TCMR, with RCRNorm also exhibiting poor calibration.

Conclusions:

• The choice of normalization method significantly impacts gene expression profiles and diagnostic classifier performance in kidney allografts.
• Simpler normalization methods, including nSolver-based pipelines, consistently achieved robust discrimination for AMR and TCMR.
• Complex methods like RCRNorm and RUVSeq (for TCMR) diminished diagnostic accuracy, highlighting the importance of selecting appropriate normalization strategies for B-HOT-based molecular diagnostics.