Shallow whole-genome sequencing of cell-free DNA for B-cell lymphoma diagnosis and disease monitoring : a standardized approach with underappreciated potential

Background Genetic material from B-cell lymphoma (BCL) is typified by recurrent multiple somatic alterations. Given the large cell turnover and its physical relation to the cardiovascular system, the disease presents itself as an excellent candidate for cell-free DNA (cfDNA) based research through liquid biopsies (LBs). Recent studies mainly focused on targeted sequencing techniques for single nucleotide variation and translocation detection. However, these methods remain expensive, require targeted panels and are yet to be clinically standardized. In contrast, the much cheaper shallow whole-genome sequencing (sWGS) for copy number alteration (CNA) detection, is operative at hospitals that offer noninvasive prenatal testing. Despite these benefits, a thorough evaluation of this approach for BCL diagnosis and disease monitoring was, until now, lacking. Methods Between May 2016 and September 2019, 126 BCL patients (39 Hodgkin [HL] and 87 non-Hodgkin [nHL] lymphoma) have been recruited for this study. In total, 257 case samples were analyzed, including 126 LBs at baseline; 98 longitudinal follow-up samples from 32 patients; and 33 paired solid biopsies (SBs), obtained as formalin-fixed paraffin-embedded (FFPE) samples. HL patients were initially treated with ABVD (Adriamycin, Bleomycin, Vinblastin, Dacarbazin), whereas nHL patients received R-CHOP(-like) (Rituximab, Cyclophosphamide, Hydroxydaunorubicin, Oncovin, Prednisone) therapy. The cohort was ultimately extended with 60 negative LB controls. DNA was sequenced by sWGS (0.5x coverage), where resulting reads were mapped to both human and lymphoma-associated viral reference genomes (e.g. Epstein-Barr virus [EBV]). The copy number profile abnormality (CPA) score was developed to quantify the genome-wide deviation from the healthy diploid state. Likewise, the normalized viral read fraction (VRF) was defined to reflect the viral load. The 99th percentile of previous variables’ theoretical normal distributions, defined by the controls’ mean and standard deviation, was used as a 1% false discovery rate cutoff for abnormality calling. These limits were subsequently used to obtain tumor and virus detection sensitivity and specificity statistics. Results During tumor staging, 78.3% of HL and 73.5% of nHL LBs had detectable CNAs. Copy number profiles from solid-liquid pairs showed high concordance within nHL patients (r=0.815±0.043). The CPA score was higher in LBs for both HL (P=.005) and nHL (P=.026). In 39/225 (17.3%) LBs, an elevated EBV VRF was noted. All but two EBV positive samples also had an abnormal CPA score, which significantly associates the latter categorical variables (P=.005). All samples that tested positive during routine chromogenic in situ hybridization (CISH) for EBV-encoded RNA detection on FFPE tissue had abnormally elevated EBV levels in plasma, whereas only two of the 34 CISH negative samples were retrieved as ‘false positives’, with subtle yet elevated amounts of plasmatic EBV titers. Observations were consistent for refractory and relapsed patients: when detectable, CNA patterns were very similar across different longitudinal staging moments. Amongst the patients with follow-up, five were detected to be EBV positive (according to sWGS of cfDNA) at diagnosis. Of these, three went into complete remission and consequently no longer had abnormally elevated EBV levels; one refractory patient was recurrently noted with high EBV titers; and one relapsed patient was ultimately detected with an unusually high EBV VRF, after it earlier dropped to 0. Very similar conclusions could be made for the CPA score. Following an extensive concordance analysis, the CPA score was found to be positively and significantly associated to tumor fraction; Ann Arbor stage; International Prognostic Index; lactate dehydrogenase and beta-2 microglobulin concentration; and, metabolic tumor volume, derived from PET/CT scans. Finally, distinguishing HL from nHL, is characterized by an area under the curve of 0.968 during receiver operating characteristic analysis applied to an iterative out-of-bag random forest modeling procedure. Conclusions Both CNAs and viral fragments can be detected more frequently in LBs than SBs of BCL patients. CNA patterns remain more or less constant in refractory/relapsed patients across staging moments. There were no false positive LBs regarding tumor detection. Changes in copy number state fluently correlate to disease status. Finally, as a proof of principle, copy number profiles enable histological classification between, amongst others, HL and nHL