Modeling KRAS mutations in KMT2A::MLLT3 acute myeloid leukemia promotes leukemic progression and transcriptional rewiring

Acute myeloid leukemia (AML) is a biologically heterogeneous disease where the mutational profile of the leukemic cells is a key determinant of prognosis, including disease-free survival and overall survival (OS). Rearrangements of the KMT2A gene occur in about 10% of AML cases, with MLLT3 being the most common fusion partner [ 1 ]. Although KMT2A fusions can initiate leukemogenesis, most KMT2A -rearranged ( KMT2A -r) leukemias acquire additional mutations that provide a selective advantage to subclones, allowing them to expand and eventually dominate the leukemic population. Despite advances in next-generation sequencing that have identified these secondary alterations, their functional roles remain poorly understood. Among these events, activating mutations in the RAS pathway, most frequently involving NRAS , KRAS and PTPN11 , are detected in 10–25% of AML cases [ 2 , 3 ] and in up to 32% of KMT2A -r AML [ 4 ]. Their typically low variant allele frequency (VAF) suggests that they arise later during leukemogenesis as subclonal events [ 5 ]. Their clinical significance remains debated, with some studies linking KRAS pathogenic variants to lower remission rates and poorer survival [ 6 ]. Relapses remain common due to the persistence of chemotherapy-resistant leukemic stem cells (LSCs) [ 7 ]. To better understand the functional contribution of KRAS pathogenic variants in this context, we generated a human KMT2A::MLLT3 ( KM3 ) AML in vivo model to investigate their role in disease progression. We have previously shown that human CD34+ cells transduced with KM3 are sufficient to induce AML in vivo when injected into NSG-SGM3 immunodeficient mice [ 8 ]. To evaluate the impact of RAS activation, we introduced a KRAS G13D point mutation using CRISPR-Cas9 in KM3 -transduced human CD34+ cells before transplantation, while control mice received KM3 cells without the KRAS pathogenic variant (Fig. 1A ). KM3 + KRAS G13D mice developed leukemia with a significantly shorter latency, on average 2.16 ± 0.65-fold earlier than KM3 control mice across four independent experiments (Fig. 1B ). PCR and Sanger sequencing confirmed the presence of the KRAS pathogenic variant in leukemic cells, with VAFs ranging from 6.8% to 48.9%, comparable to those observed in patients [ 4 ]. The cell surface phenotype and morphology were similar between groups, expressing CD45, CD33, CD11b, HLA-DR and CD38, indicating that the shorter latency was not associated with phenotypic changes. All experiments were reproduced with the KRAS G12C pathogenic variant and showed similar results, except for the VAF, which was on average ~5%. KRAS pathogenic variants (G13D or G12C) slightly increase proliferation in KM3 cells, leading to a higher proportion of AML cells in the G2/M phase (Fig. 1C ). We next performed limiting dilution analysis (LDA) in secondary transplantation to quantify LSC frequency in KM3 control and KM3 + KRAS G13D leukemias. Secondary KM3 control mice remained healthy for >100 days with low engraftment (1–32%), whereas KM3 + KRAS G13D mice developed disease rapidly with high engraftment (Fig. 1D ). LSC frequency was estimated at >1 in 17 000 in KM3 + KRAS G13D AMLs versus 1 in 360 000 in KM3 controls ( p = 0.0069), indicating a > 20-fold increase in LSC frequency in our KMT2A -r AML model. PCR and sequencing confirmed the maintenance of the KRAS G13D pathogenic variant in secondary mice with VAFs between 16.8% and 68.8%. Fig. 1: KRAS pathogenic variants shorten KM3 AML latency, increase LSC frequency and induce gene expression changes. Full size image A Schematic representation of the in vivo experiments. B Representative Kaplan-Meier survival curves of mice KM3 control or KM3 + KRAS G13D injected with ~400 000 cells per mouse. C Cell-cycle analysis of KM3 control ( n = 6) and KM3 + KRAS -mutated ( n = 10) cells from bone marrow (BM) determined by flow cytometry. P value was determined by Student t test. * P < 0.05. D Kaplan-Meier survival curves of secondary mice ( KM3 control or KM3 + KRAS G13D) injected with 500,000, 50,000 or 5000 leukemic cells for limiting dilution experiment assay (LDA). ELDA software was used to calculate LSC frequencies. E OS of KM3 AML in the Harmony cohort of patients ≤ 60 years. F Principal Component Analysis (PCA). Principal components were calculated and the first two ranked by proportion of variation explained are shown (PC1 and PC2). CD34+ cells are represented in blue, KM3 control cells in purple (WT), KM3 + KRAS G13D in green and KM3 + KRAS G12C in red. G Volcano plot showing log2 fold-change (x-axis) and their −log10 FDR values (y-axis) for the DEGs detected between KM3 control vs. KM3 + KRAS pathogenic variants (G13D + G12C) cells. AML acute myeloid leukemia, BM bone marrow, DEGs differentially expressed genes, ELDA extreme limiting dilution analysis, KM3 KMT2A::MLLT3 ; LDA, limiting dilution analysis, LSC leukemic stem cell, OS overall survival, PCA principal component analysis, WT wild type. Taken together, KRAS G13D accelerates disease onset without altering cellular phenotype and markedly increases LSC frequency in secondary recipients. KRAS VAFs closely mirror those observed in patients and remain stable upon serial transplantation, supporting the clinical relevance of our model. Consistent with these findings, analysis of the HARMONY KMT2A::MLLT3 adult AML cohort showed that KRAS pathogenic variants ( n = 22) were associated with significantly reduced OS (median OS: 9.6 vs. 38.1 months in KRAS -mutated vs. wild-type (WT) patients; p = 0.006 in patients younger than 60 years (Fig. 1E )), a trend that remained evident across the entire cohort regardless of KMT2A fusion partner (data not shown). RNA sequencing (RNA-seq) was conducted on leukemic cells isolated from mice to assess transcriptional changes induced by KRAS pathogenic variants (G13D and G12C). Principal component analysis (PCA) showed clear separation between healthy cord blood CD34+ cells (blue, CD34 + ), KM3 control cells (purple, WT), and KM3 + KRAS- mutated cells (green, KRAS G13D and red, KRAS G12C) (Fig. 1F ). Notably, KRAS -mutated leukemias clustered tightly together, indicating a high degree of transcriptional homogeneity within the groups. Differential expression analysis (baseMean > 200, adjusted p < 0.0001) identified 842 upregulated and 790 downregulated genes in KRAS variant AMLs compared with KM3 controls (Fig. 1G ; Supplementary Table 1 ). Gene expression profiles were highly similar between KRAS G13D and G12C leukemias. Among the most upregulated genes were the hematopoietic transcription factors GATA1 , GATA2 , and TAL1 , as well as IKZF2 , AEBP1 , ALOX5 , and ALOX15 , genes previously linked to leukemic stem cell maintenance or inflammatory signaling [ 9 , 10 , 11 , 12 ]. Gene set enrichment analysis revealed enrichment of TNF-α signaling via NF-κB and dendritic cell maturation pathways in KRAS variant AMLs. We next performed single-cell RNA-seq (scRNA-seq) on KM3 control and KM3 + KRAS G13D AMLs (primary and secondary). KM3 control leukemias displayed relatively homogeneous transcriptional profiles, whereas most of KRAS variant samples showed the emergence of a distinct population annotated as conventional dendritic cell type 1 (cDC1)–like cells (Fig. 2A ). This suggests that KRAS G13D not only preserves the existing leukemic architecture but can also drive the emergence of a novel subpopulation with potential functional significance. Fig. 2: Single-cell analysis and cell type composition of KM3 control and KM3 + KRAS G13D cells. Full size image A Representative UMAP of scRNA-seq from KM3 control (left), primary (middle) and secondary (right) KM3 + KRAS G13D whole BM mice samples and stacked bar plot showing the composition of ANN-predicted cell types. B Representative UMAP showing ALOX15 and ALOX5 expression in KM3 control and KM3 + KRAS G13D (primary and secondary) cells from whole BM mice samples. Color scales represent normalized expression levels for each gene, with higher expression indicated by warmer colors. C Gene expression levels of ALOX5 and ALOX15 in KM3 control ( n = 16) and KM3 + KRAS -mutated ( n = 10) samples. BM bone marrow, CD34 + CD34-positive cells, KM3 KMT2A::MLLT3 ; scRNA-seq, single-cell RNA sequencing, UMAP Uniform Manifold Approximation and Projection. Mapping bulk RNA-seq differentially expressed genes at single-cell resolution revealed that several genes, including IKZF2 , DAPK2 , and ALOX15 , were specifically expressed in this cDC1-like population. RNA-seq analysis revealed that ALOX15 was exclusively expressed in KM3 + KRAS -mutated samples, whereas ALOX5 was consistently detected in both control and KRAS -mutated conditions (Fig. 2C ). UMAP visualization further showed that ALOX5 - and ALOX15 -expressing cells represent distinct cellular populations. Notably, ALOX15 expression, but not ALOX5 , was restricted to the cDC1-like population emerging in KM3 + KRAS G13D samples (Fig. 2B ). Together, these results indicate that KRAS G13D reshapes the leukemic transcriptional landscape, promoting the emergence of a cDC1-like population. The paucity and genetic heterogeneity of acute leukemias complicate their study and contribute to discrepancies in prognostic impact. In KMT2A -rearranged AML, RAS pathway pathogenic variants are generally associated with poor outcomes, although results vary depending on cohort characteristics (age, fusion partners, co-mutations). Beyond prognosis, the biological consequences of these pathogenic variants remained poorly understood. To answer these important questions, we added a KRAS pathogenic variant to our well characterized in vivo KM3 AML model. Given the tight regulation of the KRAS signaling pathway, the mutation was introduced via CRISPR-Cas9 genome editing rather than by viral transduction. This strategy ensured that no additional copies of the gene were inserted, thereby preserving physiological expression levels. We observed a more aggressive disease characterized by reduced latency, increased proliferation, and a higher frequency of leukemic stem cells (LSCs), without major changes in immunophenotype. To our knowledge, this is the first experimental evidence directly linking KRAS pathogenic variants to increased LSC frequency in human leukemia. LSCs are known for their self-renewal capacity, quiescence, and resistance to therapy, making them key drivers of disease progression and relapse. While the mechanisms underlying their expansion remain unclear, our results suggest that KRAS pathogenic variants promote a stem-like program in a context-dependent manner. Indeed, the impact of KRAS pathogenic variants likely varies depending on the cellular and genetic background, as illustrated by their variable effects across AML subtypes and in RASopathies. Transcriptomic analyses revealed upregulation of hematopoietic regulators including GATA1 , GATA2 , and TAL1 , consistent with transcriptional programs associated with stem cell maintenance. We also observed upregulation of ALOX15 , a gene implicated in LSC survival in other hematological malignancies. Single-cell RNA-seq further revealed the emergence of a cDC1-like population in most KRAS -mutated samples, absent in controls. Notably, ALOX15 expression was restricted to KM3 KRAS -mutated samples and was specifically enriched within this cDC1-like population, suggesting a potential link between this pathway and the emergence or function of this distinct cellular subset. Although the role of this population remains unclear, similar expansions of differentiated cell types have been described in other AML contexts, and targeting such differentiated subpopulations could represent a potential therapeutic approach. Overall, our study establishes a biologically and clinically relevant model of KRAS -mutated KM3 AML and highlights KRAS as a driver of LSC expansion and disease aggressiveness. Importantly, our findings are supported by clinical data showing reduced overall survival in KM3 patients harboring KRAS pathogenic variants, consistent with observations in broader KMT2A -r AML cohorts. While KRAS inhibitors have demonstrated efficacy in solid tumors [ 13 , 14 ], their applicability to non-G12C pathogenic variants remains limited, underscoring the need for alternative strategies. Targeting KRAS-driven pathways or associated cellular states may therefore represent a promising avenue to improve outcomes in this AML subgroup.