Utility of peripheral blood cell-free DNA in detecting minimal residual disease in patients receiving CAR T-cell therapy for B-acute lymphoblastic leukemia (B-ALL)

Abstract Introduction: Current modalities for minimal residual disease (MRD) detection in B-ALL include flow cytometry and next-generation sequencing (NGS). While highly sensitive and specific, they require the presence of intact leukemia cells. Cell-free DNA (cfDNA) offers the potential advantage of subclinical disease detection in the absence of circulating blasts, but its use in B-ALL has been understudied. As an exploratory objective on “Pilot Prospective Study for PET-CT Imaging in Participants with Relapsed/Refractory Acute Leukemias” (NCT05969002), the use of cfDNA as an adjunctive marker for disease monitoring was evaluated in patients with relapsed/refractory (RR) B-ALL proceeding to Chimeric Antigen Receptor T-cells (CAR-T). Methods: Plasma samples were prospectively collected on day 0 (pre-infusion), +14, and +28 post-CAR-T infusion in patients with RR B-ALL enrolled on this study. cfDNA was isolated using the QIAmp Circulating Nucleic Acid kit (Qiagen). Libraries were constructed using the KAPA HyperPrep kit (Roche) and xGen UDI-UMI Adapters (IDT) and uniformly generated with 8 PCR cycles. Shallow whole-genome NGS was performed to a mean depth of 6x on a NovaSeq S4 platform (lllumina). cfDNA was assessed for genomic instability (copy number alteration (CNA) derived tumor fraction (TF) using ichorCNA) and fragmentomics (cfDNA fragment end motifs, non-negative matrix factorization (NMF) of fragment sizes, NMF end motifs (F-profiles), genome wide short/long fragment binned ratios (binwise)). Statistical analysis was performed in R, calculating features' area under the curve (AUC) from receiver operating characteristic curves with Youden's index. Dominant B-ALL VDJ sequences were identified by the Adaptive clonoSEQ® platform, based on NGS-MRD analysis from each patient. cfDNA libraries were queried using NCBI's Basic Local Alignment Search Tool (BLAST) for dominant VDJ sequences. Personalized droplet digital PCR (ddPCR) assays were designed based on a priori knowledge of each patient's VDJ clonotype(s). VDJ clonotype quantifications of cfDNA by ddPCR were normalized to housekeeping gene MRTFB. Active leukemia was defined as detectable disease in the peripheral blood (PB) or bone marrow (BM) via flow cytometry or NGS, or as extramedullary disease seen on PET scan. Results: Across 13 patients enrolled, the median age was 21.4 years (range: 5.1-38.1) and 92% were male. Libraries were successfully generated for 8 patients across 1-3 timepoints (20 total samples), all of whom had active leukemia in at least 1 timepoint. These were compared to 21 healthy control libraries for fragmentomic analysis. End-motif analysis modestly differentiated samples with active leukemia from healthy (F-profile 2 AUC 0.77, F-profile 4 AUC 0.75, F-profile 5 AUC 0.74). This improved when F-profiles 4 and 5 were combined into a polynomial support vector machine (AUC 0.77). NMF (AUC 0.95) and binwise (AUC 0.94) distinguished well between B-ALL and healthy states. Fragmentomic analyses and TF were combined into a meta-classifier, boosting performance with accurate predictions in 38/41 samples (100% specificity, 77% sensitivity). Due to lack of a dominant VDJ sequence, 3 samples from 1 patient were excluded from further testing. VDJ sequences were detected in PB cfDNA WGS libraries using BLAST for 2/10 samples (2/7 patients) with active leukemia. One of the 2 positive samples had negative PB flow cytometry and NGS, negative BM flow cytometry, and only 4-7 clones/million on BM NGS. No sequences were detected using BLAST for the 7 samples from 4 patients without active leukemia. Detection using BLAST was not consistent, likely due to depth of coverage. To improve sensitivity, we transitioned to ddPCR testing of PB cfDNA. ddPCR was positive in 9/10 samples (6/7 patients) with active leukemia. Of the 10 samples, only 1/10 had positive PB flow cytometry, and 7/9 had positive PB NGS (1 PB sample was not tested for NGS). The false negative sample was negative in PB flow, BM flow, and PB NGS, with <1 clone detected on BM NGS. ddPCR was negative in 6/7 samples (4/4 patients) without active leukemia. The false positive sample was negative on PB flow and NGS, but BM status was unknown. Conclusions: cfDNA shows promise as a non-invasive adjunctive assay for disease detection in B-ALL, even in the absence of circulating blasts. Further validation of our ddPCR testing is required to determine whether it can be used merely for detection or for quantitation of disease.