INTRODUCTION
Fusion genes are major drivers of leukemogenesis in acute leukemias, and as recurrent molecular aberrations, they serve as essential diagnostic and prognostic biomarkers [1, 2]. The recently updated 2022 WHO classification and the International Consensus Classification (ICC) have broadened the spectrum of diagnostically significant gene fusions, incorporating diverse rearrangements such as various KMT2A and NUP98 fusions into subtype classifications [3–5]. Notably, in AML, not only well-established fusions such as PML::RARA, RUNX1::RUNX1T1, and CBFB::MYH11 but also KMT2A or NUP98 fusions now permit an AML diagnosis even with blast counts <20%, highlighting their clinical significance [3, 6]. This expanded diagnostic framework emphasizes the need for accurate and comprehensive fusion detection to guide risk stratification and targeted therapies.
Conventional diagnostic methods, including karyotyping, FISH, and reverse transcription PCR (RT-PCR), have been the mainstay for detecting fusion genes in hematologic malignancies [6–9]. However, these methods cannot detect novel fusions or those with uncommon breakpoints and often require prior knowledge of fusion partners or specific chromosomal regions [10, 11]. To overcome these limitations, next-generation sequencing (NGS) technologies, particularly RNA sequencing (RNA-seq), are being increasingly adopted in clinical diagnostics [10–13].
RNA-seq-based fusion detection comprises two primary approaches: targeted RNA-seq and whole RNA-seq. Targeted RNA-seq focuses on a predefined panel of genes commonly associated with leukemias, enabling high sequencing depth and enhanced sensitivity for low-abundance fusions [12, 13]. Its streamlined workflow, relatively low sequencing cost per sample, and short turnaround time make it feasible for routine clinical use in many institutions [14–16]. In contrast, whole RNA-seq profiles the entire transcriptome, enabling comprehensive detection of rare or novel fusions without prior knowledge of fusion partners [1, 10, 17]. Although whole RNA-seq is more resource-intensive and computationally demanding, its ability to uncover previously uncharacterized fusions is critical in light of the updated classifications that emphasize diverse rearrangements [18].
Given the evolving diagnostic landscape and the need for comprehensive fusion profiling, we compared the fusion detection performance of whole RNA-seq with that of conventional diagnostics in patients with acute leukemia to assess its clinical utility and feasibility in routine diagnostics.
MATERIALS AND METHODS
Patients
In this retrospective study, we included 101 patients with newly diagnosed or relapsed acute leukemia who underwent bone marrow (BM) evaluation along with cytogenetic and molecular characterization at Samsung Medical Center, Seoul, Korea, between September 2022 and September 2023 as part of their standard clinical care. Specifically, patients with sufficient BM aspirates for RNA-seq were included in the study. The patients were selected to ensure representation of both positive and negative fusion gene detection cases, as identified using conventional diagnostic methods. All patients were diagnosed according to the 2016 WHO classification of hematolymphoid tumors [19] and were reclassified in this study according to the 2022 WHO classification or ICC [3–5]. BM aspirates were collected at the time of initial diagnosis or relapse. Laboratory results, including complete blood count, BM examination, karyotyping, FISH, multiplex RT-PCR, and targeted NGS, as well as clinical data, such as treatment, disease course, and prognosis, were obtained from electronic medical records. Conventional diagnostics for fusion detection included karyotyping, FISH, and multiplex RT-PCR. This study was approved by the Institutional Review Board of Samsung Medical Center (IRB No. 2024-12-079), which waived the need for informed consent because of the use of de-identified data and minimal risk to participants.
Conventional diagnostics
Karyotyping was performed on heparinized BM aspirates using a standard G-banding technique following short-term culturing without mitogen. FISH was performed on the BM aspirates using various probes (Supplemental Data Table S1), following the manufacturers’ instructions, as previously described [6]. KMT2A FISH was routinely performed, and selective FISH analysis was conducted when specific chromosomal abnormalities were suspected, based on cytogenetic findings. Multiplex RT-PCR was performed using the HemaVision kit (DNA Technology, Aarhus, Denmark) according to the manufacturer’s instructions.
RNA-seq and fusion gene detection
RNA was extracted from BM aspirates using the QIAamp RNA Blood Mini Kit (Qiagen, Hilden, Germany). RNA-seq libraries were prepared from 300 ng of total RNA per sample using the Illumina Stranded mRNA Prep kit (Illumina, San Diego, CA, USA) with poly(A) enrichment and sequenced on the Illumina NextSeq 550Dx or NovaSeq 6000 platform (Illumina), generating 2× 150-bp paired-end reads. The median sequencing depth was 90 million reads per sample [interquartile range (IQR), 63–261 million reads; range, 14–1,021 million reads], and the median output was 12 Gbp per sample (IQR, 9–38 Gbp; range, 2–148 Gbp). Sequence quality was assessed using FastQC (v0.11.8) [20]. Adapters were trimmed, and low-quality reads removed, with Trimmomatic (v0.39) [21]. Reads were aligned to the GRCh37 reference genome using STAR (v2.7.7a) [22] with default parameters. Fusions were detected using Arriba (v2.4.0) [23], retaining high-confidence in-frame fusions and high-confidence fusions involving MECOM, immunoglobulin genes (IGH, IGK, and IGL), and T-cell receptor genes (TRA, TRB, TRD, and TRG) as true fusions. Fusion events were manually validated using the Integrative Genomics Viewer (v2.16.0) [24] by inspecting split reads and discordant read pairs at predicted breakpoints to confirm their presence and exclude potential artifacts. When residual samples were available, novel or rare fusions identified using RNA-seq were validated using RT-PCR with fusion-specific primers, followed by Sanger sequencing.
Statistical analysis
The fusion gene detection performance of RNA-seq was compared with that of conventional diagnostics by calculating concordance, sensitivity, and specificity. Categorical variables were compared using Pearson’s chi-squared or Fisher’s exact test, as appropriate. Cases with missing or invalid data were excluded from the analyses. All statistical analyses were performed using IBM SPSS Statistics, version 27 (IBM, Armonk, NY, USA). Statistical significance was set to P<0.05.
RESULTS
Patient characteristics
RNA-seq was performed on samples from 101 patients with acute leukemia, whose demographic and clinical characteristics are summarized in Table 1. The study population included 54, 37, 8, and 2 patients with acute myeloid leukemia (AML), B-cell acute lymphoblastic leukemia (B-ALL), T-cell acute lymphoblastic leukemia (T-ALL), and mixed-phenotype acute leukemia (MPAL), respectively. The median age was 48 yrs (range, 0–87 yrs), and 28.7% of patients were <18 yrs, including 45.9% of patients with B-ALL and 62.5% with T-ALL. Among the patients, 70.3% had newly diagnosed de novo acute leukemia, 12.9% had relapsed disease, and 16.8% had secondary AML arising from pre-existing myelodysplastic neoplasms or following cytotoxic therapy for other malignancies. The median blast percentage in BM aspirates was 70% (range, 8–99%); notably, six patients with <20% blasts in BM aspirates had >20% blasts on BM biopsies. All patients underwent cytogenetic evaluation, including karyotyping and KMT2A FISH, with multiplex RT-PCR performed in all except four patients. Conventional diagnostics detected fusion genes in 46.5% of patients, with detection rates of 40.7%, 62.2%, 25%, and 0% for AML, B-ALL, T-ALL, and MPAL, respectively. One patient with B-ALL harbored the rare IGH::CEBPA fusion, and one patient with T-ALL carried dual fusions (STIL::TAL1 and TRA/D::?).
Overall fusion gene detection using RNA-seq
RNA-seq helped identify 52 fusion genes in 51 of 101 patients (50.5%), including 24 unique fusion genes (Table 2). The fusion detection rate varied by leukemia subtype: 40.7% (22/54) in AML, 70.3% (26/37) in B-ALL, 37.5% (3/8) in T-ALL, and 50.0% (1/2) in MPAL. RNA-seq also helped characterize three fusion genes, RPN1::MECOM, KMT2A::THAP12, and RUNX1::PRPF19, whose partner genes were not identified using conventional diagnostics; among these, KMT2A::THAP12 and RUNX1::PRPF19 were novel (Fig. 1). Additionally, RNA-seq exclusively helped detect 12 fusion genes, including novel (FUS::ZNF362 and MLLT10:: UBE2L6), rarely reported (ETV6::NCOA2, HNRNPH1::ERG, and RUNX1::USP42) (Supplemental Data Fig. S1), and well-known (IGH::CRLF2, P2RY8::CRLF2, PAX5::JAK2, PAX5::NOL4L, and PICALM::MLLT10) fusion genes. Notably, one patient with B-ALL harbored dual fusion genes, BCR::ABL1 and P2RY8::CRLF2. Integrating conventional diagnostics and RNA-seq results helped detect at least one fusion gene in 56.4% (57/101) of all patients: 44.4% (24/54) in AML, 75.7% (28/37) in B-ALL, 50.0% (4/8) in T-ALL, and 50.0% (1/2) in MPAL.
Fusion gene detection performance of RNA-seq compared with that of conventional diagnostic methods
We compared the fusion detection performance of RNA-seq with that of conventional diagnostics and analyzed 104 fusion gene detection results (positive and negative) across 101 patients (Supplemental Data Table S2). Three patients (one each with AML, B-ALL, and T-ALL) were identified as harboring two fusion genes each using conventional diagnostics and/or RNA-seq, contributing to the total of 104 fusion gene detection results. The overall concordance rate between RNA-seq and conventional diagnostics was 80.8%. Of the 48 fusion genes detected using conventional diagnostics, RNA-seq helped identify 40 and missed eight (83.3% sensitivity) (Supplemental Data Table S3). The undetected fusions, including three BCR::ABL1, one ETV6:: RUNX1, one KMT2A::MLLT10, and three involving MECOM (N=1) and TRA/D (N=2), were attributed to factors such as low blast proportion (<20%), subclonal presence, inadequate BM aspirate quality, complex rearrangements, or enhancer hijacking mechanisms. Conversely, among 56 fusion-negative results based on conventional diagnostics, RNA-seq helped identify 12 fusion genes (21.4%, reflecting 1 – specificity). By leukemia subtype, RNA-seq helped detect fusions in 9.1% (3/33) of AML, 40.0% (6/15) of B-ALL, and 33.3% (2/6) of T-ALL fusion-negative cases based on conventional diagnostics, with B-ALL showing a significantly higher rate than that of AML (P=0.010).
Clinical relevance of fusion genes identified using RNA-seq
To investigate the clinical significance of the fusion genes identified using RNA-seq, we analyzed the laboratory and clinical features of 15 patients: three with fusion genes whose partner genes were identified using RNA-seq, and 12 with fusion genes detected only with RNA-seq (Table 3). Notably, RNA-seq enabled reclassification of five patients with B-ALL, not otherwise specified (NOS), into B-ALL with BCR::ABL1-like feature (N=3), B-ALL with ZNF362 rearrangement (N=1), or B-ALL with PAX5 alteration (N=1).
The following cases illustrate the clinical courses and prognostic implications of patients with novel and rare fusion genes identified using RNA-seq.
Novel fusion genes
1) KMT2A::THAP12 in AML, prior cytotoxic therapy (case No. 19)
A 35-yr-old man with AML, previously treated for diffuse large B-cell lymphoma, presented with acute myelomonocytic leukemia. FISH revealed a KMT2A break-apart signal despite a normal karyotype. RNA-seq helped identify a novel KMT2A::THAP12 fusion, likely resulting from a cryptic inversion, inv(11)(q23.3q13.5), based on the chromosomal positions of THAP12 (11q13.5, 5′ telomere-oriented) and KMT2A (11q23.3, 5′ centromere-oriented). Although induction therapy with idarubicin and cytarabine failed, re-induction with cladribine, cytarabine, granulocyte colony-stimulating factor, and mitoxantrone helped achieve complete remission (CR). However, the patient relapsed and died 7 months post diagnosis.
2) RUNX1::PRPF19 in relapsed AML with mutated NPM1 (case No. 141)
A 67-yr-old woman with relapsed AML, harboring an NPM1 mutation, initially presented with a normal karyotype and negative multiplex RT-PCR, with NPM1 status untested. After induction, consolidation, and allogeneic hematopoietic stem cell transplantation (HSCT), she relapsed 5 months post-transplant, showing 26% myeloblasts and trilineage dysplasia. Targeted NGS helped confirm an NPM1 mutation (W288fs, c.860_863dup, variant allele frequency 21.4%). Cytogenetics revealed clonal evolution: 46,XY,t(8;21)(q12;q22)[9]/46,XY,t(11;21)(q12;q22)[5]/46,XX[6]. FISH for RUNX1::RUNX1T1 indicated a jumping translocation with RUNX1 rearrangement, showing three RUNX1 signals. RNA-seq helped identify RUNX1::PLAG1 [t(8;21)(q12;q22), out-of-frame] and a novel RUNX1::PRPF19 fusion [t(11;21)(q12;q22)]. The patient achieved CR with mitoxantrone, etoposide, and cytarabine (MEC) but died 4 months post-relapse.
3) MLLT10::UBE2L6 in relapsed AML with KMT2A::MLLT10 (case No. 111)
A 64-yr-old man with AML harboring KMT2A::MLLT10 relapsed 2 months post-chemotherapy, presenting with acute monocytic leukemia and a complex karyotype: 50,XY,+4,+8,der(10)t(10;11)(p12;q23)inv(11)(q23q13),der(11)t(10;11)(p12;q13),+16, +20[16]/46,XY[4]. The der(10)t(10;11)(p12;q23)inv(11)(q23q13) yielded KMT2A::MLLT10, whereas der(11)t(10;11)(p12;q13) produced a novel MLLT10::UBE2L6 fusion. The patient achieved CR with MEC induction but relapsed 1 month later and, despite allogeneic HSCT, died 2 months post-transplant.
4) FUS::ZNF362 in B-ALL (case No. 58)
A 29-yr-old man with B-ALL presented with a normal karyotype. He achieved CR following induction chemotherapy and underwent allogeneic HSCT but relapsed 4 months post-transplant. Re-induction with blinatumomab was attempted; however, the patient died 9 months after relapse.
Rare fusion genes
RUNX1::USP42 was identified in a 7-yr-old boy with AML with myelodysplasia-related changes (case No. 52), presenting with acute myelomonocytic leukemia. Cytogenetic analysis revealed 91<4n>,XXYY,del(5)(q15q33)×2,–17[18]/46,XY[2]; dysplasia was not evident. HNRNPH1::ERG was detected in a 36-yr-old woman with AML with maturation (case No. 168), whereas ETV6::NCOA2, corresponding to t(8;12)(q13;p13), was identified in a 6-yr-old girl with MPAL, T/myeloid (case No. 119). All three patients achieved CR following induction therapy and remained stable post-allogeneic HSCT.
DISCUSSION
Our study demonstrated that RNA-seq outperforms conventional diagnostics in fusion gene detection. RNA-seq detected fusions in 50.5% of all patients, with leukemia subtype-specific rates of 40.7%, 70.3%, 37.5%, and 50% in AML, B-ALL, T-ALL, and MPAL, respectively. Integrating RNA-seq with conventional diagnostics improved fusion detection by approximately 10% compared to fusion detection achieved via conventional diagnostics alone. RNA-seq enables the precise identification of fusion transcripts, particularly for cryptic or rare rearrangements unresolved with cytogenetic analyses [1, 11, 25], contributing to the enhanced detection rates observed in our study. Notably, among fusion-negative results with conventional diagnostics, RNA-seq helped identify fusion genes in 40% of B-ALL and 33% of T-ALL cases, significantly exceeding the 9.1% detection rate observed for AML. These findings reflect the detection of recurrent cryptic rearrangements in B-ALL, including Philadelphia-like ALL-associated fusions (e.g., IGH::CRLF2, P2RY8::CRLF2, PAX5::JAK2), PAX5 alterations (e.g., PAX5::NOL4L), and B-ALL with ZNF384-rearranged-like features (e.g., FUS::ZNF362), as well as PICALM:: MLLT10 in T-ALL. This highlights the greater fusion-related genetic complexity of ALL and the superior diagnostic yield of RNA-seq in this context.
The overall concordance rate between RNA-seq and conventional diagnostics was 80.8%, with a fusion detection sensitivity of 83.3%, comparable across AML (86.4%) and B-ALL (87.0%), which was consistent with findings in previous studies. A large-scale study in 806 AML patients reported RNA-seq-based detection of 89.9% of true fusion events identified using routine diagnostics, surpassing karyotyping (87.7%) and molecular diagnostic (77.5%) detection rates, and identified 26 recurrent fusion events undetected via conventional methods [11]. In pediatric AML, panel-based RNA-seq achieved 83% concordance for risk-relevant fusions, improved detection of cryptic fusions (e.g., NUP98::NSD1, KMT2A::MLLT10), enhanced risk stratification in 10.4% of cases, and increased measurable residual disease-monitorable cases from 44.4% to 75.5% [26]. Similarly, a study in 126 pediatric patients with ALL reported 86% concordance for recurrent rearrangements [27].
However, the sensitivity of RNA-seq can be limited by low mapping efficiency, inadequate sequencing coverage, or low transcript expression, which led to missed detections of CBFB or KMT2A rearrangements in previous studies [11, 27]. Suboptimal samples with low blast proportions also pose challenges, as reported in pediatric ALL, where molecular subtype classification was hindered [28]. We could not detect common fusions such as BCR::ABL1 and ETV6::RUNX1, likely because of low blast proportions, presence of subclonal populations, or poor aspirate quality, which reduced the fusion burden. Additionally, RNA-seq has limitations in detecting rearrangements that do not produce functional fusion transcripts, particularly those involving enhancer hijacking mechanisms, such as MECOM, IGH, and TRA/D rearrangements, the effects of which depend on the gene expression levels [25, 29]. In our study, we identified one RPN1::MECOM and one IGH::CEBPA fusion but missed one MECOM and two TRA/D rearrangements, likely because of these constraints. These findings suggest that although RNA-seq is a powerful tool, it may require complementary conventional methods to enhance clinical utility. Ensuring high-quality specimens with sufficient blast proportions is essential, as peripheral blood dilution can significantly reduce detection sensitivity. In cases with a low blast burden, detection methods such as karyotyping, FISH, or targeted RT-PCR may be effective alternatives. For fusions involving enhancer hijacking, bioinformatics pipelines incorporating expression-based screening (e.g., detecting aberrant gene overexpression) may help identify candidate events, which can then be validated using orthogonal methods such as karyotyping or FISH. Nevertheless, clinical adoption of RNA-seq remains limited because of challenges such as long turnaround times, need for high-throughput infrastructure, difficulties in assay standardization, and complex data analysis pipelines. Addressing these barriers is crucial for the integration of RNA-seq into routine clinical workflows.
In this study, four novel fusions in acute leukemias were identified, each associated with poor prognostic implications, along with three rare fusions associated with relatively favorable outcomes. KMT2A rearrangements, involving over 100 partner genes, are well-documented in acute leukemias [30], and our identification of the novel KMT2A::THAP12 fusion in AML expands this repertoire. THAP12, located at 11q13.5, encodes a zinc finger protein that interacts with ZFP574 to regulate cell-cycle progression and hematopoiesis, contributing to B-cell leukemogenesis by promoting leukemic cell proliferation [31]. Although AML-associated data are limited, the KMT2A::THAP12 fusion likely enhances oncogenic potential in AML, with rapid relapse in our patient suggesting an aggressive phenotype.
Similarly, in AML, beyond the common RUNX1 translocations (RUNX1::RUNX1T1 and ETV6::RUNX1), we identified a novel RUNX1::PRPF19 fusion, along with the rare RUNX1::USP42 fusions, both suggesting an association with myelodysplasia-related changes. PRPF19, located at 11q12.2 and encoding a factor involved in mRNA pre-processing, has been reported as a KMT2A partner in AML and ALL [30, 32]. The RUNX1::PRPF19 fusion, detected as a jumping translocation in a patient with an NPM1 mutation, likely represents a secondary event, with the emergent dysplasia supporting its association with myelodysplasia-related changes. Conversely, RUNX1::USP42, typically resulting from a cryptic t(7;21)(p22;q22) translocation, is associated with adverse outcomes and frequently co-occurs with del(5q) in AML [32–35]. In our patient, del(5q) and monosomy 17 accompanied RUNX1::USP42, suggesting an association with AML with myelodysplasia-related changes, despite the absence of dysplasia.
Another novel fusion, MLLT10::UBE2L6, was identified alongside KMT2A::MLLT10 in AML. UBE2L6, at 11q12.1, encodes a ubiquitin-conjugating enzyme E2 L6, which promotes leukemic cell differentiation via ISGylation [36]. KMT2A::MLLT10 is associated with poor prognosis in AML and is often associated with rapid relapse (<1 yr) [37], as observed in our patient. The concurrent MLLT10::UBE2L6 fusion may exacerbate this poor prognosis, with rapid relapse in our patient suggesting a potential synergistic effect.
In B-ALL, we identified the novel FUS::ZNF362 fusion, adding to the previously reported ZNF362 rearrangements such as SMARCA2::ZNF362 and TAF15::ZNF362 [38]. ZNF362-rearranged B-ALL cases share a gene expression profile with ZNF384 rearrangements, leading to their classification as B-ALL with ZNF384(382) rearrangement in the ICC [5], characterized by prevalence in adolescents and young adults, high expression of myeloid-associated antigens CD13 and CD33, frequent CD10 negativity, and variable prognosis depending on the fusion partners [38–40]. The immunophenotype of our patient was consistent with this profile (CD10dim and CD33+). The rapid disease progression suggested a poor prognosis for FUS::ZNF362.
Among the rare fusions identified in our study, HNRNPH1:: ERG, similar to FUS::ERG, has been reported in pediatric and young adult AML and is associated with poor survival [41]. However, our patient with HNRNPH1::ERG underwent allogeneic HSCT and remained stable at follow-up at 16 months. Similarly, ETV6::NCOA2, predominantly reported in MPAL with t(8;12)(q13;p13), is associated with favorable outcomes in pediatric patients [42, 43]. Concordantly, our patient with ETV6::NCOA2 has remained stable following transplantation.
This study has several limitations. Sequencing depth varied across samples, likely because of differences in sample quality, library preparation efficiency, or sequencing performance. Although the median depth of 90 million reads provided robust coverage for most samples [44–46], some samples exhibited lower read depths, which may have reduced sensitivity for rare or low-abundance fusions. Additionally, we did not perform RNA expression profiling. Future studies incorporating such analysis may enhance the diagnostic and clinical impact of RNA-seq.
In conclusion, to the best of our knowledge, we present the first evaluation of whole RNA-seq–based fusion detection in patients with acute leukemia in Korea. RNA-seq demonstrated superior fusion detection compared with that of conventional diagnostics and enabled improved risk stratification. We identified novel fusions and clarified their clinical relevance, further highlighting the added value of RNA-seq. Our findings provide a foundation for optimizing RNA-seq strategies and integrating complementary testing to overcome technical and biological limitations, thereby enabling comprehensive diagnostic applications.