Rapid diagnosis of acute leukemia with integrated epigenetic and genetic profiling Free

Introduction Acute leukemia (AL) diagnosis is a complex process that integrates multiple modalities, including histology, flow cytometry, cytogenetics, and targeted sequencing. In prior work, we have developed MARLIN, a machine learning classifier that generates accurate AL diagnoses solely on the basis of DNA methylation patterns (Steinicke*, Benfatto*, et al., Nature Genetics, in press). When coupled with Oxford Nanopore Technologies (ONT) long-read sequencing, MARLIN generates confident real-time predictions for most AL cases in under two hours from sample receipt. However, methylation-based classifications alone do not provide complete information as some alterations, such as FLT3 internal tandem duplications (ITD), do not show distinct methylation profiles. Therefore, we sought to establish an integrated approach for AL classification in the clinic that uses ONT long-read sequencing to capture DNA methylation and genetics within a single, rapid assay.

Methods To incorporate genetics with methylation-based classifications (MARLIN), we applied adaptive sampling functionality on the ONT PromethION sequencing platform (hereafter, ONT-AS) to perform real-time enrichment of 274 genes and other regions of interest (ROI) that are recurrently altered in AL and other blood cancers. On each flow cell, we devoted 90% of sequencing pores to ONT-AS to obtain high coverage over ROIs, and used the remaining 10% of pores for conventional (non-adaptive) sequencing to obtain genome-wide data suitable for methylation-based rapid classifications (MARLIN) and detection of copy-number changes. Using this approach, we sequenced a retrospective cohort of 10 AL patient bone marrow samples (median age: 61.5 years, range: 23-87 years) representing diverse AL subtypes (n=7 AML, n=2 B-ALL, n=1 T-ALL) and leukemic drivers. Data analysis was performed using Dorado basecaller, minimap2 aligner, Clair3 for SNV/indel calling, qDNAseq for copy-number analysis, and Sniffles for structural variant/fusion detection, in addition to manual review in igv.

Results In 10/10 cases, methylation data generated from pores devoted to conventional (non-adaptive) sequencing yielded confident (>0.8) predictions for AL lineage and molecular subtype within the first two hours of sequencing. These classifications recapitulated (7/10 cases) or refined (3/10 cases) the final WHO/ICC diagnosis for each case, consistent with our prior validation of MARLIN for rapid AL diagnosis. In parallel, genomic data generated from pores devoted to ONT-AS showed strong enrichment over targeted ROIs. Across our 10 cases, an average of 68.0 gigabases (range 36.5-131.7) of output were generated over 72 hours of ONT-AS with a median coverage of 133x (range 75.0-192.4x) over ROIs, including key genes such as NPM1 (105x), TP53 (113x), and FLT3 (163x). We detected 27/31 SNVs/indels reported by routine clinical testing; these included class-defining variants per WHO/ICC criteria (for example, Patient 1: TP53 p.H187D; Patient 2: NPM1 p.W288Cfs*12) and variants that predict response to targeted therapies (for example, Patient 2: FLT3-ITD; Patient 3: IDH1 p.R132C). We also identified split-reads corresponding to the four reported rearrangements/fusions in our cases, including inv(16), KMT2A::MLLT3, PAX5::ETV6, and MEF2D::BCL9L. Finally, copy-number analysis was consistent with changes reported by routine clinical testing, including class-defining alterations in MDS-associated AML (for example, Patient 1: del(5q)/-7/del(12p)/del(17p)).

Conclusions We establish proof-of-concept for rapid diagnosis of AL through integrated detection of epigenetic and genetic information with ONT sequencing. Within a single sequencing run, this approach can produce accurate epigenetic classifications in just 1-2 hours from the time of sample receipt while also achieving high (>100x) coverage over a comprehensive panel of genetic variants after 48-72 hours. Moreover, this approach requires minimal library preparation and no additional technical steps for target enrichment or amplification. We envision that this approach will streamline the initial diagnostic evaluation, risk stratification, and treatment selection for patients presenting with AL and provide complementary information to existing testing.