Prognostic and therapeutic implications of BRAF mutations in acute myeloid leukemia Authors: Darren Lee, Yazan Abu-Shihab, Kate Plas, Deedra Nicolet, Krzysztof Mrózek, Mark J. Routbort, Keyur P. Patel, Christopher J. Walker, Jill Buss, Andrew R. Stiff, Andrea Laganson, Courtney D. DiNardo, Naval G. Daver, Tapan M. Kadia, Farhad Ravandi, Andrew J. Carroll, Jonathan E. Kolitz, Bayard L. Powell, William G. Blum, Maria R. Baer, Guido Marcucci, Geoffrey L. Uy, Wendy Stock, Richard M. Stone, L. Jeffrey Medeiros, John C. Byrd, James S. Blachly, Robert L. Bowman, Jeffrey W. Tyner, Sanam Loghavi, Ann-Kathrin Eisfeld, Linde A. Miles Published: 2026 ABSTRACT Mutations in the RAS/MAPK signaling pathway are recurrent in acute myeloid leukemia (AML), primarily involving NRAS and KRAS. In contrast, mutations in the gene encoding an effector protein, BRAF, occur at relatively lower frequencies in AML and are associated with poor outcomes. To date, no comprehensive analysis has assessed the clinical and molecular characteristics of BRAF-mutated AML. In this study, we report the identification of canonical and non-canonical BRAF mutations in ~1% of 5779 consecutive clinically and molecularly fully annotated AML patients treated at two major United States Cancer Centers (50/5779 AML patients: 21 newly diagnosed AML; 9 relapsed/refractory; 20 newly diagnosed secondary AML). We performed single-cell multiomic analysis on a subset of AML samples. BRAF mutations were enriched in myelodysplasia-related AML (AML-MR), and most mutations were located outside the V600 hotspot. Single-cell multiomic profiling delineated BRAF mutation class-specific patterns of co-mutations, clonality, and immunophenotypes. Notably, BRAF mutations and other signaling co-mutation(s) could be found in the same cell, a finding that significantly diverges from prior studies of RAS-mutant AML. In this cohort, BRAF-mutant AML patients had poor overall survival with currently available treatments, including venetoclax-based regimens. Drug sensitivity data suggest possible avenues for targeted treatment of BRAF-mutated AML. INTRODUCTION Mutations in the RAS/MAPK pathway genes, including NRAS, KRAS, and PTPN11, occur in >20% of patients with acute myeloid leukemia (AML), both de novo and secondary [1, 2]. Typically, these mutations are subclonal, suggesting they are later, transforming events in disease development [3]. Mutations in BRAF, a gene encoding a protein kinase also in the RAS/MAPK signaling pathway, are well-known driver mutations in other cancer types [4, 5], such as melanoma [6], papillary thyroid cancer [4], colorectal cancer [4] and hairy cell leukemia [7]. BRAF mutations are also recurrent mutational events in AML, albeit at a much lower frequency [3, 8]. While RAS-mutant AML patients may respond to standard cytotoxic chemotherapy, depending on the molecular context, BRAF mutations in AML have been divergently associated with dismal outcomes [3]. However, since many routinely used genetic testing panels have limited BRAF gene coverage, BRAF mutations might be overlooked and, their actual frequency, especially in relapsed/refractory disease, is likely underestimated. Importantly, in view of the prognostic impact of RAS pathway mutations on non-intensive treatment regimens, there is an urgent need to better understand the impact of BRAF mutations in AML, including their clinical, molecular, and cellular context, as well as their associations with response to intensive and non-intensive treatment regimens. The most common BRAF mutation in cancer, known as p.V600E, substitutes the valine residue at codon 600 with glutamic acid [8]. Some hematologic cancers, such as hairy cell leukemia, have recurrent BRAF V600 mutations in ~100% of cases [7]. Notably, alternative non-canonical BRAF mutations exist in other cancer types and have been sorted separately from the V600E hotspot into two classes based on kinase activity and oncogenic RAS-dependence [5]. In total, three classes of BRAF mutations exist – class I (i.e. V600E) and class II mutations generate constitutively active kinases, but the class I mutations are unique in that these kinases can function as monomers [8-10]. In contrast, class II mutant BRAF kinases still require dimerization for activity. Conversely, class III mutations are kinase-dead mutants with little to no BRAF kinase activity [11, 12]. However, class III mutant BRAF proteins are still able to amplify RAS pathway signals, but require an active or oncogenic RAS to transmit activation of the pathway. Little is currently known about the impact of the various classes of BRAF mutations on AML disease phenotype and therapeutic consequences, which may be attributed to the rarity of these mutations compared with other recurrent AML-associated mutations. However, the actual frequency of BRAF mutations is unknown, because most routinely used targeted next-generation sequencing panels cover only the V600 BRAF mutation. In this study, we identified a cohort of BRAF-mutant AML patients and examined their leukemic molecular landscape, clonal architecture, and response to therapy. We found that BRAF mutations were highly enriched in myelodysplasia-related AML (AML-MR) and that most of them were non-V600E, non-canonical BRAF mutations, with each class of BRAF mutation showing a unique co-mutation and immunophenotype pattern. Moreover, we found that BRAF-mutant AML patients had dismal outcomes, and their disease showed resistance to venetoclax similar to RAS-mutated AML. However, through characterization of published ex vivo drug treatment data, we also identified unique drug sensitivities for BRAF-mutant AML that are not shared by other RAS-pathway mutant AML. METHODS Patient cohort Diagnosis according to World Health Organization (WHO) classification [13] and the International Consensus Classification (ICC) [14] criteria and disease status assignment are provided in Supplementary Table 1. Mononuclear cells were isolated via Ficoll gradient centrifugation from obtained bone marrow aspirates or peripheral blood specimens. Single-cell suspensions of samples were viably frozen and stored in liquid nitrogen. Bulk DNA molecular profiling All patients and samples included in the BRAF-mutant clinical cohort and control cohorts underwent bulk DNA molecular profiling. Samples in the BeatAML cohort obtained from OHSU were sequenced with a combination of whole-exome sequencing, targeted sequencing panels, and clinical genetic platforms (Sequenom and GeneTrails [2, 15]). Analysis of samples collected at MDACC used targeted PCR-based NGS panels covering 81 recurrently mutated genes [16] or 28 recurrently mutated genes [17] in myeloid disorders. The Alliance samples were sequenced using an NGS panel covering 80 cancer and/or leukemia-associated genes, as described previously [18]. OSU patient sequencing was obtained from the patients’ clinical records using the OSU CLIA-approved panels. Whole-exome sequencing Whole-exome hybrid capture and NGS sequencing were performed on a subset of BRAF-mutant AML samples at OSU. The capture included chromosomal tiling probes (250-fold coverage) and was performed on leukemic and matched germline tissue for each patient for the detection of single nucleotide variants and small insertions/deletions, as described previously [19]. Reagents Mission Bio Tapestri-related reagents were included in the Myeloid DNA +Protein V2 sequencing kit except: Cell Staining Buffer, TotalSeqD CD135 antibody (313325), and TotalSeqD Human Heme Oncology Cocktail v1 (399906) were purchased from Biolegend. The Myeloid amplicon panel is a previously described panel [20]. Single-cell multi-omic sequencing AML patient samples were thawed, washed with FACS buffer (PBS supplemented with 2% fetal bovine serum), filtered to obtain single-cell suspensions, and quantified using a CellDrop (Denovix). Cells (1 × 106 viable cells) were then incubated with Tapestri blocking buffer and TruStainFcX (Biolegend) on ice for 15 min followed by a 30 min incubation with the TotalSeqD Antibody Cocktail supplemented with 2 μL of TotalSeqD CD135 on ice. Stained cells were then further processed as previously described in commercially available protocols [21-23]. Sequencing of pooled libraries were performed by the CCHMC DNA Genomic Sequencing shared facility. Single-cell DNA sequencing analysis Sequencing reads were processed as follows: reads were trimmed, aligned (hg19 human genome), assigned barcodes, and genotypes called using GATK by the Mission Bio Tapestri v2 pipeline. Further analysis of processed H5 files was performed using the scDNA package (https://github.com/bowmanr/scDNA, v1.01) in R. In the scDNA package, H5 files were input and variants of interest were identified based on clinical sequencing identified variants. All variants included were manually investigated in IGV. We selected non-synonymous, exonic variants that were genotyped in >50% of assayed cells and had a computed VAF > 1%. The variant list was further refined to exclude variants that were 1) confirmed SNPs, 2) recurrently mutated broadly across cohorts at a fixed VAF, and/or 3) represented exclusively in low-quality reads or clipped reads. The ‘tapestri_h5_to_sce’ function from the scDNA package was used after variant selection to create a SingleCellExperiment class object using the default cutoffs of allele frequency (AF) variance >25, depth (DP) > 10, and genotype quality (GQ) > 30. The AF variance refers to the maximum deviation from 50% by which a heterozygous call should be masked as inaccurate. We only retained variants that passed all three filters in over 80% of cells and only cells that passed all three filters were included in the final analysis and termed “Complete” cells, indicating they received a reliable genotype for all genes of interest. Clones were identified and statistically summarized following variant identification using the ‘enumerate_clones’ and ‘compute_clone_statistics’ functions, respectively. Single cell DNA+Protein (scDNA+Protein) sequencing analysis Protein matrices were extracted from the H5 files using the scDNA package after genotyping and clone enumeration above. The SingleCellExperiment object was converted to a Seurat object (v5.1) with metadata containing genotyping information. All complete cells identified above were bound to a single protein matrix for cohort level protein analysis across all samples. Protein data were normalized across cells using DSB, scaled across all samples, and analyzed by PCA. Samples were integrated with Harmony, clustered (SLM) and then subsequently visualized by Uniform Manifold Approximation and Projection (UMAP) [24]. Cell type calls were performed by manual interpretation of protein expression. Data were visualized using Seurat, ggplot2, gridExtra, ggpubr, magick, patchwork, raster and scCustomize packages (https://samuel-marsh.github.io/scCustomize/). Beat AML ex vivo drug screen data Ex vivo functional drug screens were performed as described previously [2, 15] on freshly isolated mononuclear cells as part of the BeatAML cohort. Data in Fig. 3 and Supplementary Fig. 3 were generated as part of the BeatAML cohort and downloaded from their online browser (http://vizome.org/aml2). Data points were replotted using GraphPad Prism v10. Statistical analysis Wilcoxon rank tests were used to assess significant differences in scDNA +Protein protein expression (Fig. 2, Supplementary Fig. S1). Survival data were analyzed by logrank (Mantel–Cox) tests (Fig. 3, Supplementary Fig. S2). Comparisons of area under the curve (AUC) metrics from the BeatAML ex vivo drug data were analyzed by Kruskal–Wallis tests (Fig. 3D, E, Supplementary Fig. S3) or Mann–Whitney tests (Fig. 3H, I) when appropriate. Plotting and graphical representations Oncoprint (Fig. 1A) was generated using R oncoPrint package. Clonographs and UMAPs were generated in R. Graphs unless otherwise noted were generated using GraphPad Prism v10. UMAP data were plotted using the R ggplot2 package (RRID:SCR_014601). Other data processing was performed in R utilizing packages including: tidyr, dplyr, RColorbrewer, pals, and cowplot. Code availability & data sharing statement All scripts are available at https://github.com/bowmanr/scDNA and processed data files are available at https://github.com/plask3189/scProtein_Analysis for DNA+Protein analyses. Raw data files are available upon request from the authors. Data files from BRAF WT AML samples used for comparisons in Fig. 2 are available on dbGAP (phs004401.v1p1). RESULTS Clinical analysis shows enrichment of BRAF mutations in AML-MR We identified 50 patients with BRAF mutations (1% of all patients with AML treated at our centers during the study period) including 21 (42%) with newly diagnosed AML, 9 (18%) with relapsed/refractory AML, and 20 (40%) with newly diagnosed secondary AML, defined as having a documented antecedent chronic myeloid neoplasm or therapy-related AML. BRAF-mutant AML patients had a median age of 67 years (range, 19–84 years), and 54% of patients were male (Supplementary Table 1). The median variant allele frequency for all BRAF mutants was 15% (range, 1–83%). The median bone marrow blast percentage was 43% (range, 3–93%), and the median absolute monocyte count was 0.52/μL (range, 0–94.56). BRAF mutations were enriched in patients with AML-MR based on both the World Health Organization (WHO 5th) [13] (n = 34/50, 68%) and the International Consensus Classification (ICC) [14] (n = 26/50, 52%) classifications, including 11 (22%) patients who had a history of a previous MDS/MPN. Molecular profiling of BRAF-mutant AML patients The most frequent co-mutations in BRAF-mutant AML patients were in TET2 (45%) and ASXL1 (30%) (Fig. 1A). Oncogenic RAS pathway mutations were also common, with NRAS detected in 30% and KRAS in 23% of patients. In fact, 50% of BRAF-mutant patients (n = 25/50) harbored at least one other oncogenic RAS/MAPK pathway mutation, suggesting RAS pathway addiction and that the RAS pathway in general may be a convergent mechanism of leukemogenesis for this AML subtype. Surprisingly, the majority (37/50, 73%) of BRAF-mutant cases were non-V600E variants (Fig. 1B, C). We found that 35 and 22% of patients harbored class II (n = 18) and class III (n = 11) mutations, respectively, with G469 (class II) and D594 (class III) mutations most common. This is in contrast to prior findings in solid tumors and lymphoid malignancies, where the V600E mutation represented the majority of BRAF alterations [4]. As class II and class III mutations have different dependencies on RAS-pathway activation, we next examined co-mutations across the BRAF mutation classes identified in our cohort (Fig. 1D). We found that indeed patients with class III D594 mutations were significantly enriched with NRAS co-mutations (P = 0.003). In contrast, class I V600 and class II G469 mutations seldom co-occurred with other RAS pathway mutations. Additionally, while BRAF V600 mutations were mutually exclusive with ASXL1 co-mutations (P = 0.005), class II G469 mutations were significantly enriched in ASXL1 co-mutated cases (P = 0.0003). The BRAF class-specific co-mutational patterns with common myeloid malignancy mutations are suggestive of unique pathogenesis compared to other AML subtypes. Whole-exome sequencing uncovers recurrent RAS-pathway addiction in BRAF-mutant AML We performed whole-exome sequencing (WES) in a subset of patients (n = 5) to identify mutations outside of the targeted sequencing panels. BRAF-mutant AML harbored a median mutational burden of 22 non-synonymous, coding mutations (range 13–32). Interestingly, patients on average harbored 3 RAS/MAPK pathway mutations (range, 2–4; Supplementary Table 2). In two patients, WES was also performed at relapse. In one patient, the BRAF mutation was stable at relapse (Fig. 1E). In the second patient, the BRAF mutation was lost at relapse, but the patient instead acquired an IQGAP3 mutation (Fig. 1F). IQGAP3 regulates cell proliferation by modulating the RAS/ERK signaling pathway, acting downstream of Rho family GTPases and interacting with ERK to promote cell cycle progression [25]. These findings support a convergent reliance on the MAPK/RAS pathway in this leukemia subset. Single-cell molecular profiling uncovers unique co-mutation and immunophenotypic patterns To resolve the clonal architecture of BRAF-mutant AML, we performed simultaneous single-cell DNA (scDNA) + cell surface protein analysis on 7 samples using the Mission Bio Tapestri platform [21, 26]. We found that BRAF mutations could be identified either in the dominant, largest non-wild-type clone (n = 4) or in a sub-clone (n = 3) (Fig. 2A). BRAF mutations co-occurred on a single-cell level with oncogenic NRAS and KRAS mutations but were mutually exclusive with a PTPN11 mutation in one sample. This observation is significantly divergent from the clonal co-mutation patterns we previously observed in AML patients harboring signaling mutations (RAS, FLT3), where we very rarely observed more than one signaling mutation in a clone [21]. We next investigated immunophenotype patterns in BRAF-mutant AML samples. We found that BRAF mutations were present in cells spanning different lineage states, highly dependent on the type of BRAF mutation (Fig. 2B). Similarly, we found that other RAS pathway co-mutations in BRAF-mutant samples spanned different lineage states (Fig. 2C). Specifically, we found BRAF G469 mutant clones to be significantly enriched for a stem/progenitor phenotype with CD34 expression (P < 2 × 10−16) (Fig. 2D, E). KRAS-mutant cells were also significantly enriched in expressing CD34 in BRAF-mutant samples (P < 2 × 10−16). Conversely, BRAF V600E and BRAF D594 mutant clones were enriched in monocytic cells expressing CD16 and/or CD11b (P <2×10−16 for all) (Fig. 2F, G, Supplementary Fig. S1A). NRAS and PTPN11 mutant clones were also found to have increased CD11b and CD16 expression, as previously identified [21, 26, 27] (Fig. 2E-G, Supplementary Fig. S1A, B). Compared to BRAF WT clones, all BRAF-mutant clones had higher CD123 expression (Supplementary Fig. S1C, D). BRAF V600E mutant clones, specifically, displayed higher expression of FLT3 compared to class II and class III mutant clones (Supplementary Fig. S1E, F). These findings suggest that BRAF mutations from different classes may lead to highly divergent cell states and immunophenotypes of the resulting AMLs. We then compared our BRAF-mutant AML samples to a separate subset of BRAF WT/RAS-mutant AML samples previously analyzed by scDNA+Protein [23], which harbor similar AML co-mutations found in a subset of our BRAF-mutant AML patients, including NRAS mutations (Fig. 2H). We found that BRAF G469 mutant cells had overall higher expression of CD34 compared with NRAS-mutant cells from both sample cohorts (P < 2 × 10−16; Fig. 2I-K, Supplementary Fig. S1G). Interestingly, even though BRAF V600 and BRAF D594 mutant clones expressed CD11b and CD14, the NRAS-mutant cells from BRAF WT/RAS-mutant samples expressed significantly higher levels of the monocyte markers (NRAS vs BRAF V600, P < 2 × 10−16; NRAS vs BRAF D594 P = 0.039) (Fig. 2L-N, Supplementary Fig. S1G-J). BRAF V600 mutant clones had expression of CD123 comparable to that of RAS mutant clones from BRAF WT/RAS-mutant AML samples (Supplementary Fig. S1G, K-M). BRAF mutations are associated with poor overall survival regardless of treatment regimen The overall response rate (CR, CRi, or CRh) was 46% and the median OS was 5.67 months (range, 0.13–90.7 months) for the entire cohort (Supplementary Table 3, Supplementary Fig. S2A). Newly diagnosed (ND) BRAF-mutant patients were treated with a variety of low-intensity (LI, n = 7) and high-intensity (HI, n = 14) treatment regimens with or without venetoclax [28] (Fig. 3A). ND BRAF-mutant AML patients had poor overall survival (OS) regardless of therapeutic regimen (P = 0.2752; LI+/−VEN: 15.09 months, 95% Cl 0.13-non-estimable (NE) vs HI+/−VEN: 13.90 months, 95% Cl 2.37-NE). BRAF-mutated patients with de novo AML treated with HI regimens had OS similar to that of 2022 European LeukemiaNet [29] (ELN) Intermediate-risk patients (Supplementary Fig. S2B). BRAF variant allele burden did not have an impact on OS in newly diagnosed patients (VAF ≥15%: 5.02 months, 95% CI: 0.43–22.68 vs VAF < 15%: 18.47 months, 95% CI: 1.02 to NE; P = 0.132; Fig. 3B). Prior exposure to venetoclax among R/R AML did not impact OS (Supplementary Fig. S2C, P = 0.884). We next reviewed the BeatAML cohort [2, 15] and identified three ND patients with BRAF mutations. Their OS was very short, with all three patients deceased at 4 months post-study enrollment. Interestingly, patients with BRAF-mutant AML from the BeatAML cohort had significantly higher percentages of monocytes (P < 0.0018) in the peripheral blood compared to NRAS/KRAS-mutant patients (Fig. 3C). Ex vivo drug sensitivity data suggest BRAF-specific vulnerabilities To further evaluate treatment options in BRAF-mutant AML, we looked at relative drug efficacies using the samples in the BeatAML cohort [2, 15]. Five BRAF-mutant AML samples were tested, including 3 harboring V600E mutations and 2 harboring class II G469 mutations (Fig. 3D). BRAF-mutant AML samples were significantly more resistant to 8 therapeutic agents compared with BRAF-WT AML samples. These drugs included Erlotinib, an EGFR inhibitor, and two multi-kinase inhibitors, KW-2449 and Lenvatinib. Interestingly, we also found that BRAF-mutant AML samples showed increased resistance to venetoclax compared to RASWT patients, providing a rationale for the poor responses to venetoclax-based regimens seen in our patients (P = 0.049; Fig. 3A, E, Supplementary Fig. S3A). NRAS/KRAS-mutant AMLs have been previously shown to be resistant to venetoclax-based regimens and showed resistance levels similar to BRAF-mutant AML [27, 30-32]. Conversely, these samples were found to have significantly increased sensitivity to 8 therapeutic agents compared to BRAFWT AML, with the highest sensitivity to flavopiridol, a cyclin-dependent kinase (CDK) inhibitor [33], which had previously shown exciting efficacy data in specific AML subtypes, and had been more recently suggested as a strategy to overcome venetoclax resistance [34]. Not surprisingly, additional efficacies were observed upon treatment with vemurafenib, a mutant BRAF V600E inhibitor [35], trametinib, a MEK inhibitor [36], and rapamycin, an mTOR complex inhibitor [37], and also with 17-AAG, an HSP90 inhibitor [38]. We compared the drug efficacies of BRAF-mutant AML to those identified in NRAS/KRAS-mutant AML samples. RAS- and BRAF-mutant samples all shared sensitivities to Trametinib, Selumetinib, and CI-1040, compared to other AML genotypes (Fig. 3F, Supplementary Fig. S3B). NRAS- and BRAF-mutant AML shared sensitivity to flavopiridol, but KRAS-mutant AMLs were not significantly sensitive (Fig. 3G). Conversely, NRAS-mutant AML was sensitive to panobinostat; however, the responses of KRAS- and BRAF-mutant AML did not reach statistical significance (Supplementary Fig. S3C). BRAF-mutant AML showed unique sensitivity to the HSP90 inhibitor, 17-AAG and the broad CDK inhibitor, AT7519, compared to BRAF WT AMLs and trended towards significance with RAS-mutant AMLs (Fig. 3H, I, Supplementary Fig. S3D). These findings suggest that BRAF-mutant AML may have unique drug sensitivities that extend beyond BRAF/RAF pathway inhibition, with specific promise of both HSP90 and broad CDK inhibition that should be studied further as alternative therapeutic options for patients. DISCUSSION Oncogenic mutations in the RAS/MAPK cascade occur in almost all types of human cancer. Alterations in the BRAF gene occur in ~15% of all human cancers [4]. Interestingly, while RAS pathway mutations are frequent in AML occurring in ~20% of cases, mutations in the gene encoding the downstream effector kinase BRAF occur much less frequently [1, 3]. This is in contrast to many solid malignancies where BRAF mutations represent significant cancer drivers, and to the B-cell derived hairy cell leukemia, where BRAF V600 mutations occur in almost 100% of cases. However, almost 75% of the BRAF mutations in our patient cohort were non-canonical mutations that were not located at the V600 hotspot. Since BRAF V600E mutation makes up 70–80% of all BRAF gene alterations in human cancer, our findings suggest that AML is unique in being enriched for non-hotspot BRAF mutations. BRAF mutant proteins have been classified based on the mutation location and the intrinsic kinase activity of the mutated protein. Class I (V600) and class II (i.e. G469 and others) mutants are hyperactive and RAS-independent [8, 9], whereas class III (i.e. D594 and others) mutants are inactive or kinase-dead and require an activated or oncogenic RAS protein [11, 12]. Indeed, BRAF D594 mutant AML patients in our cohort frequently harbored NRAS co-mutations, and our single-cell data from a BRAF D594 mutant patient showed that the BRAF and NRAS mutations co-occurred in the same clone, indicating the potential dependence of BRAF D594 mutations on activated RAS in AML as well. Interestingly, we observed divergent co-mutation patterns between V600 and G469 mutations, particularly with ASXL1 co-mutations, suggesting the BRAF mutation classes may divergently synergize with recurrent AML mutations. Similarly, we also found unique immunophenotype patterns between the different BRAF mutant clones, with BRAF G469 mutant clones enriched in CD34 expression and BRAF D594 mutant clones having higher expression of CD11b and CD16. Additionally, our BRAF-mutant AML samples showed higher levels of CD123 expression, which warrants testing the efficacy of recent CD123-targeted therapies [39]. Not only do these findings suggest that the BRAF mutations may be prevalent with certain co-mutations based on synergistic interactions, but also that BRAF mutant classes are enriched in distinct leukemic cell compartments. Although limited by the number of samples available, the results of our multiomic analysis underscore the divergent phenotypes among AML samples with different BRAF mutations and suggest that further studies are needed to determine how the BRAF mutations interact with recurrent AML drivers and affect the phenotype of the leukemia. While RAS mutations in other cancers typically bestow poor patient outcomes, the prognostic relevance of RAS mutations remains less clear in AML. NRAS-mutant AML patients typically show superior responses to standard chemotherapy compared with patients with other signaling mutations [3], but RAS/MAPK pathway mutations are also enriched in AML patients who are either refractory to or who develop resistance to targeted and novel therapies. BRAF mutations in AML have been associated with poor prognoses [40-42]; however, many studies lacked sufficient numbers of BRAF-mutated cases for analysis. Although our study still includes a relatively small number of patients, which precludes us from performing a multivariable outcome analysis, it also shows that BRAF-mutated AML patients have poor outcomes of BRAF-mutated AML patients, regardless of the frontline regimen, which is consistent with the literature [3, 40-42]. These findings align with the recurrent acquired resistance/relapse mechanism for the venetoclax/azacitidine combination treatment, which lies in the expansion of a myelomonocytic blast population enriched for NRAS/MAPK mutations [27, 31]. This is further supported by our drug sensitivity data on BRAF-mutant AML obtained from the BeatAML cohort, which suggests that cells harboring BRAF mutations are more resistant to venetoclax; however, it is still unknown if they utilize similar mechanisms as N/KRAS-mutant cells. These findings suggest that BRAF mutations, like other RAS/MAPK mutations, appear to play a significant role in therapy resistance and may serve as a convergent mechanism for therapy escape in AML. Studies in other cancer subtypes have uncovered that each class of BRAF mutant proteins has distinct biochemical and signaling properties, which may lead to different dependencies on interacting proteins and alternative pathways. For example, these studies suggest that class III BRAF mutants are sensitive to RAS-pathway inhibition [12], whereas the class II BRAF mutants may be sensitive to MAPK-pathway inhibitors outside of RAS-targeted compounds [43]. Data from the BeatAML study uncovered multiple therapeutic agents that may hold promise in BRAF-mutant AML patients, including HSP90 and CDK inhibition. However, the majority of the 5 samples harbored the V600E mutation, and no samples tested contained a class III mutation. Therefore, further examination of these agents in class II and III mutant AML samples is required to fully characterize their efficacy as potential avenues for BRAF-mutant AML treatment. In conclusion, our study demonstrates that while rare, BRAF mutations are recurrent in AML, and are typically non-canonical. They exhibit distinct, class-specific co-mutation and immunophenotypic patterns, and are associated with poor survival outcomes, even in newly diagnosed patients. Ex vivo drug sensitivity studies identified potential therapeutic agents for BRAF-mutant AML. These findings highlight the unique nature of BRAF-mutated AML, underscoring the critical need for improved understanding of this genotype to enhance survival outcomes, and the inclusion of expanded BRAF gene coverage into routine testing panels for newly diagnosed and relapsed/refractory AML patients. Supplementary Material Supplementary information The online version contains supplementary material available at https://doi.org/10.1038/s41375-026-02996-1.