RELIABLE ESTIMATION OF BLOOD-BASED TUMOUR MUTATIONAL BURDEN: ANALYTICAL VALIDATION AND IMPLICATIONS FOR CLINICAL IMPLEMENTATION

The clinical implementation of biomarkers in oncology is essential for tailoring treatment strategies to individual patients. Tumor Mutational Burden (TMB), defined as the number of somatic mutations per megabase of interrogated genomic sequence, has emerged as a pan-cancer biomarker for response prediction to immune checkpoint inhibitors. However, significant analytical and biological variability complicates its interpretation, particularly in non-standardized settings and across different sample types such as formalin-fixed paraffin-embedded (FFPE) tissue and cell-free DNA (cfDNA). While tissue-based TMB (tTMB) has been investigated extensively, its clinical utility remains limited by tissue availability, quality, and turnaround time. Blood-based TMB (bTMB) offers a minimally invasive alternative, yet its analytical validity and biological interpretability remain poorly standardized.
This thesis investigates the implementation and validation of a plasma-based comprehensive genomic profiling (CGP) approach for bTMB quantification using the Illumina TSO500 ctDNA assay, the focus being the technical and clinical evaluation of bTMB, using both contrived reference standards and cfDNA from healthy individuals. In addition, we confirmed robusteness of TMB estimation using TSO500 tissue assay through an FFPE cohort across diverse tumor types and compared results against independent bioinformatics pipelines.
Using engineered ctDNA controls with defined bTMB mutation profiles and input amounts, we examined the impact of key technical and biological parameters, cfDNA input quantity, library conversion efficiency, allele frequency or tumor content fraction, and sequencing depth, on variant detection and bTMB accuracy. Notably, high reproducibility of bTMB scores was observed at high input and library conversion rates, while sensitivity deteriorated rapidly under suboptimal conditions, especially for low allele frequency variants. In such cases, reduced representation of unique molecules impaired variant detection and led to artificial underestimation of bTMB.
To investigate biological background noise, we profiled cfDNA from healthy donors. Despite no clinical evidence of malignancy, we observed measurable bTMB scores in majority of samples. Detailed variant annotation and matched blood analysis revealed that these elevated bTMB signals were originating partially from rare white blood cells (WBC)-matched clones. In addition, pipeline-specific artifacts, particularly low specificity for small deletion calls, artificially inflated bTMB values. Comparative analysis of biological and technical replicates further revealed significant inconsistencies in low allele frequency variant detection, reflecting the stochastic nature of variant calling under low-input and borderline coverage conditions.
Our findings emphasize that bTMB is a highly input- and context-dependent metric. Accurate interpretation requires harmonized preprocessing, rigorous artifact suppression, and integration of matched blood controls. By addressing these critical factors, this work establishes a robust framework for bTMB implementation, paving the way for its reliable use in both clinical decision-making and translational cancer research.