Distinct Genetic and Fragmentomic Signatures of Plasma Ctdna Synergize and Resolve Clinical Challenges in Patients with Large B-Cell Lymphoma

Introduction

Seizing the full potential of lymphoma circulating tumor DNA (ctDNA) for scientific discoveries and accurate precision medicine tools requires diminished dependence on tumor tissues, enhanced detection of rare variants and harnessing non-random ctDNA fragmentation patterns for non-invasive profiling. Establishment of such tools is pivotal to overcome clinical challenges set by pervasive heterogeneity and different treatment responses in the patients with large B-cell lymphoma (LBCL).

Methods

Using our previous data on targeted and whole-genome sequencing (WGS) levels of lymphoma ctDNA (Meriranta, 2022), we constructed a novel targeted sequencing assay with B-cell lymphoma-relevant loci and used duplex barcoding to improve analytical sensitivity. We optimized fragment pattern conservation by comparison of different library preparation strategies and established the requirements for exhaustive and reproducible duplex sequencing with error-rate down to 2-4/10^6. We examined replicate samples and in vitro and in silico dilution series of ctDNA spike-in in healthy donor cell-free DNA (cfDNA) to determine the limits of tumor-naïve variant calling, ctDNA quantification and minimal residual disease (MRD) detection. We applied this assay to >500 plasma, tumor and whole-blood DNA samples from primary LBCL patients treated in biomarker driven Nordic phase II trial (Leppä, 2022) and correlated the results with biological and clinical data (Fig A).

Results

Ultra-sensitive charting of pretreatment ctDNA revealed profound spatial heterogeneity in LBCL patients at diagnosis. Plasma profiles were formed by contributions from multiple cell subpopulations with divergent hypermutations and driver events, some exhibiting convergent trajectories diverged even at early B-cell stages preceding germinal center entrance. The cell populations represented in the diagnostic biopsies contributed an unpredictable proportion of the overall ctDNA pool and majority (63%) of the subclonal ctDNA mutations were not detected in the matched tissues resulting in conflicting genotypes between different materials. Dominant but not subclonal TP53 mutations in the ctDNA associated with poor survival also uncovering patients with no TP53 aberrations in their diagnostic biopsies.

On therapy, allele fractions of lymphoma mutations, regardless of their clonal status, declined with early responding patients being MRD negative after two treatment cycles with excellent survival. At the end-of-therapy, Duplex-MRD predicted progression with excellent specificity (98%) but with limited negative predictive value (90%) and sensitivity (47%). Fragmentomic analysis revealed that duplex-recovery was suboptimal for subnucleosomal cfDNA fragment lengths with the highest mutational content. Remarkably, harnessing cfDNA fragments with only single-strand recovery (i.e. non-duplex) for combined ‘Dual-MRD’ enhanced ctDNA detection, translated to better distinction of early responders and resolved most false-positive end-of-therapy FDG-PET scans (Fig A).

Besides enriched for shorter fragment sizes, additional non-random structural features of ctDNA molecules extended to their distinct repertoire of fragment nucleotide end motifs (Fig B). Notably, mutant fragments carried less physiological DNAse1L3 nuclease end signatures than their reference counterparts or healthy donor cfDNA (Fig B). Distinct fragmentomic elements of ctDNA molecules contained unexploited utility for non-invasive disease phenotyping and detection that we spanned to shallow (~10x) WGS breadth in independent validation series of lymphoma and healthy donor samples.

Conclusion

By harnessing synergy between duplex technology, single-strand recovery and distinct fragmentomic features, we discovered novel characteristics in lymphoma ctDNA that correlate with disease biology and improve non-invasive disease profiling to overcome limitations of tumor biopsies and clinical estimates.