AUTHORS
Roger Mulet-Lazaro, PhD, Remco Hoogenboezem, MSc2, Jolinda Konijnenburg3, Aniko Sijs-Szabo4, Stanley Van Herk5, Berna Beverloo, PhD6, Vincent H.J. Van Der Velden, PhD7, Peter JM Valk, PhD8, Anita W. Rijneveld, MD, PhD9, Mathijs Arnoud Sanders, MSc, PhD10
AFFILIATIONS
1 Department of Hematology, Erasmus MC Cancer Institute, Erasmus University Medical Center, Rotterdam, Netherlands
2 Department of Hematology, Erasmus MC Cancer Institute, Erasmus University Medical Center, Rotterdam, Netherlands
3 Department of Hematology, Erasmus MC Cancer Institute, Erasmus University Medical Center, Rotterdam, Netherlands
4 Department of Hematology, Erasmus MC Cancer Institute, Erasmus University Medical Center, Rotterdam, Netherlands, Department of Hematology, Leiden University Medical Center, Leiden, Netherlands
5 Department of Hematology, Erasmus MC Cancer Institute, Erasmus University Medical Center, Rotterdam, Netherlands
6 Department of Clinical Genetics, Erasmus University Medical Center, Rotterdam, Netherlands
7 Department of Immunology, Erasmus University Medical Center, Rotterdam, Netherlands
8 Department of Hematology, Erasmus MC Cancer Institute, Erasmus University Medical Center, Rotterdam, Netherlands
9 Department of Hematology, Erasmus MC Cancer Institute, Erasmus University Medical Center, Rotterdam, Netherlands
10 Department of Hematology, Erasmus MC Cancer Institute, Erasmus University Medical Center, Rotterdam, Netherlands
ABSTRACT
Introduction
Acute leukemias are aggressive blood cancers categorized into acute myeloid leukemia (AML), acute lymphoblastic leukemia (ALL) and acute leukemia of ambiguous lineage (ALAL), with subtypes defined by the International Consensus Classification (ICC) and the World Health Organization (WHO). RNA sequencing (RNA-seq) has emerged as a powerful diagnostic tool in leukemia, enabling simultaneous characterization of gene expression profiles (GEP) and genetic lesions. However, its analysis and interpretation are challenging. Computational pipelines such as HAMLET (Arindrarto W, 2021) or IntegrateALL (Wolgast N, 2026) facilitate this effort but are restricted to hotspot mutations and a single disease.
To overcome these limitations, we developed DRACULA (Diagnostic RNA-seq Analysis and Classification Utility for Leukemia Assessment), a Nextflow pipeline that detects a broad spectrum of clinically relevant genetic alterations and performs GEP-driven classification across acute leukemias. Here, we benchmark DRACULA on RNA-seq data of adult AML (n=145), B-ALL (n=223) and T-ALL (n=85) to investigate its potential for diagnostics and research.
Methods
DRACULA implements state-of-the-art tools to detect genetic lesions, including small variants, copy number alterations (CNAs) and fusion genes. Besides, it features the newly developed Breakpoint and Aberrant Transcript Scanner (BATS), an algorithm that extracts unannotated splice junctions to identify aberrant splicing and structural variants. The pipeline also includes machine learning models for GEP-driven classification: E-CAL determines lineage, while ALLCatchR, TALLSorts, and AMLmapR identify B-ALL, T-ALL, and AML subtypes, respectively.
DRACULA results were evaluated via orthogonal validation techniques. Small variants were assessed using DNA sequencing (DNA-seq), namely whole exome sequencing (WES) in AML (n=145) and targeted sequencing in B-ALL (n=85) and T-ALL (n=30). Fusion transcripts were compared to results from karyotyping (n=408) and RT-PCR, whereas CNAs were validated against DNA-seq and karyotyping.
Results
In AML, DRACULA detected 96% of WES-reported mutations with VAF > 0.05. Although frameshift and nonsense mutations comprised 13/20 missed variants, their high prevalence among detected variants (182/512) indicated limited nonsense-mediated decay. In ALL, 91% of the 322 mutations detected by capture sequencing were directly identified, but missed splicing mutations could be inferred by BATS, raising the recovery rate to 94%.
Fusion genes were identified in 41/145 AML cases, including 28/29 (96.5%) cytogenetically visible translocations. In ALL, DRACULA found fusions in 235/308 patients, accounting for 135/138 (98%) of events identified by cytogenetics or RT-PCR. Undetected events were subclonal or involved enhancer hijacking.
Of the 30 arm-level CNAs reported by WES in AML, 29 (96%) were identified by DRACULA. In ALL, the pipeline captured 92% of the 206 CNAs found using DNA-seq, and 84% of those derived from cytogenetics. Remarkably, DNA-seq showed higher concordance with DRACULA than with conventional karyotyping.
Based on DRACULA results, patients were assigned to leukemia subtypes per ICC guidelines. Mutational profiling correctly assigned 143/145 (98.6%) AML cases, with discrepancies restricted to MECOM-rearranged patients. AMLmapR predictions agreed for 87% of cases. In B-ALL, mutational data classified 222/223 (99.5%) patients; the single unclassified case was identified as ETV6::RUNX1-like by ALLCatchR. This model agreed with the mutation-based classification in 94.6% of the cases, struggling only to distinguish between BCR::ABL and BCR::ABL-like.
Given the prevalence of noncoding alterations in T-ALL, mutational analysis stratified only 22/54 (40.7%) non-early thymocyte progenitor (non-ETP) patients. However, incorporating individual marker expression resolved 23 additional cases, adding up to 83.3%. TALLSorts confirmed this assignment and identified 32 of the remaining cases as ETP-ALL and ETP-like, raising the overall T-ALL classification rate to 91% (77/85).
Discussion
In summary, analysis of RNA-seq data with DRACULA detects most driver genetic lesions in acute leukemia, extending beyond known hotspots. Although here tested only in AML and ALL, DRACULA also performs well in ALAL. Integration with the GEP-driven classifiers included with the pipeline enables ICC-based classification of > 95% of cases, underlining its potential as a diagnostic tool.
DISCLOSURES
Roger Mulet-Lazaro, PhD: No relevant conflicts of interest to declare.