Monitoring of CNS involvement in AML using ultrahigh-sensitivity next-generation sequencing of cell-free DNA Free

Evaluation and monitoring of central nervous system (CNS) involvement in AML patients with standard of care methods (cytomorphology, flow cytometry) is limited, as they require sufficient cell numbers in the cerebrospinal fluid (CSF). To date, no method is available for the analysis of measurable residual disease (MRD) in the CNS compartment of AML patients. Here, we evaluate ultrahigh-sensitivity next-generation sequencing (NGS) of cell-free DNA (cfDNA) from CSF and plasma for MRD analysis of CNS involvement in AML.

CSF, plasma and mononuclear cells (MCs) were collected from peripheral blood (PB) and bone marrow (BM) of 14 patients with confirmed CNS involvement, including 12 patients with AML and 2 patients with blastic plasmacytoid dendritic cell neoplasm (BPDCN), who were treated at our institution between 2021 - 2025. Samples were also obtained from 5 patients (AML n=4; mixed-phenotype acute leukemia n=1) evaluated for CNS involvement but tested negative. MCs from BM/PB and CSF were sequenced using a 46-gene panel at the time of CNS involvement. Amplicon-based ultrahigh-sensitivity NGS-MRD (limit of quantification 0.01%) on cfDNA from matched CSF and plasma samples was performed using the known mutations from diagnosis as previously described at all available follow-up timepoints before and after CNS involvement (Thol et al., Blood 2018; Hupe et al., Hemasphere 2025).

In total, 18 CNS episodes from 14 patients were analyzed: 7 cases of isolated CNS relapse and 11 cases of combined CNS and systemic involvement. At the time of CNS involvement, the most frequent mutations identified in CSF-derived MCs were DNMT3A (n=7), FLT3-ITD (n=6) and NPM1 (n=6), with a median amount of 3 mutations per patient (range 1-6).

First, the utility of cfDNA from PB plasma was assessed to detect CNS-derived mutations in patients with isolated CNS relapse. Of the 15 mutations identified in CSF MCs, only 1 FLT3-ITD mutation (6.6%) was detectable in PB plasma with a markedly lower variant allele frequency (VAF) (53.6% in CSF MCs vs. 0.04% in Plasma cfDNA).

Conversely, in the 5 patients without CNS involvement, 8 out of 9 mutations (89%) identified in PB plasma cfDNA at the time of CSF evaluation were also detectable in cfDNA from CSF. Notably, median VAF of mutations in CSF cfDNA were significantly lower compared to PB plasma cfDNA (1.4% vs 9%; p<0.01), most likely reflecting spillover of cfDNA from PB into the CSF.

In contrast, patients with combined BM and CNS involvement showed a different pattern. All mutations (15 of 15) were detectable in both compartments, with similar median VAFs in cfDNA from CSF and Plasma (42% vs. 38%; p=0.44).

To evaluate the potential utility of cfDNA from CSF for MRD assessment, the initial treatment response was analyzed and compared to cytomorphology, the current diagnostic standard. Ultrahigh-sensitivity NGS of cfDNA from CSF revealed persistent CNS disease for a significantly longer duration after CNS infiltration (median: 10 weeks; range 4-13 weeks) despite ongoing treatment. In comparison, cytomorphology showed earlier clearance of blasts with a median time to acellular CSF of only 2 weeks (range: 1–4 weeks; p < 0.01).

Two patients in our cohort had extended follow-up and experienced three and four CNS relapses, respectively. In three events (two isolated and one combined CNS involvement), paired CSF and PB plasma samples were available at a timepoint within four months prior to clinical relapse. In all three cases, MRD was detectable in CSF cfDNA prior to clinical manifestation - at 4, 14, and 16 weeks - while PB cfDNA remained MRD-negative.

These findings demonstrate that cfDNA-based ultrahigh-sensitivity NGS from CSF enables detection of residual CNS disease in the absence of detectable cells by cytomorphology. Plasma cfDNA does not adequately reflect CNS involvement and CSF analysis remains essential whenever CNS disease is suspected. To monitor disease dynamics, cfDNA from plasma and CSF must be analysed separately to distinguish true CNS involvement from PB-derived mutation spillover. Prospective studies are needed to evaluate the diagnostic and prognostic value of cfDNA-based MRD analysis from CSF. Ultrahigh-sensitivity NGS of CSF-cfDNA represents a promising tool for sensitive evaluation and longitudinal monitoring of CNS involvement and response to CNS-directed therapies.