Clinical implications of RAS mutations in AML: prognostic significance is based upon the involved gene and mutation complexity

Key Points

• •Complex mutations in KRAS predict poor outcomes in AML; complex NRAS and KRAS mutations confer an inferior prognosis in KMT2A-r AML.
• •Noncomplex KRAS mutations confer poor prognosis in the KMT2A-r AML cohort but are not associated with outcome in patients with non–KMT2A-r AML.

Visual Abstract

Abstract

Acute myeloid leukemia (AML) is a heterogeneous disease with complex mutational profiles that lead to variable clinical outcomes. NRAS and KRAS are among the most frequently mutated genes in AML, but their clinical impact has not been well characterized. In this cohort of >2000 children and young adults with AML, we evaluated the role of mutations in RAS genes and mutation complexity in outcome determination. Given enrichment in KMT2A-rearranged (KMT2A-r) AML, we specifically studied the significance of RAS mutations in KMT2A-r AML. Using variant calls from next-generation sequencing platforms, we identified RAS mutations in 35.1% (N = 669; NRAS, n = 518; KRAS, n = 216). We demonstrated that NRAS mutations were not associated with outcome in AML or in KMT2A-r AML. In contrast, KRAS mutations demonstrated inferior outcomes in AML, with enrichment of prevalence and enhancement of prognostic implications in KMT2A-r AML, including non–high-risk KMT2A fusions. Additionally, we describe a complex RAS (Comp-RAS) mutation cohort characterized by 2 distinct RAS mutations or high variant allele frequency RAS mutations that collectively account for 13.5% (n = 90) of the patients with RAS mutations. Patients with complex KRAS mutations and those with Comp-RAS mutations in the KMT2A-r cohort had a distinctly adverse outcome, and data demonstrate that Comp-RAS status drives adverse outcomes for those with KRAS mutations in the whole AML cohort.

Introduction

Acute myeloid leukemia (AML) is a highly heterogeneous disease at the cytomolecular level.¹ To a large degree, the genomic lesions define the variable clinical outcomes achieved in patients, with some subsets of patients harboring highly refractory disease.^2,3 Mutations of the RAS pathway, and subsequent aberrant RAS signaling, are prevalent across the age spectrum.^4,5 Combined, NRAS and KRAS gene mutations are among the most common somatic alterations in AML, present in 15% to 25% of all patients and 30% of pediatric patients.^2,3,6,7 Although the prevalence and significance of these mutations are increasingly recognized, there have been variable results in their utility as a prognostic biomarker or as a successful therapeutic target in AML.5, 6, 7, 8, 9, 10, 11 The prognostic and therapeutic relevance of RAS mutations within this group has not been well characterized in a large pediatric AML cohort. Emerging evidence suggests that RAS mutations are enriched or acquired at relapse following targeted therapies, including FLT3 inhibitors,^12,13 IDH1/2 inhibitors,^14,15 JAK2 inhibitors,^16,17 and the venetoclax-azacitidine combination,^18,19 implicating MAPK pathway activation as a mechanism of therapeutic resistance.

KMT2A-rearranged (KMT2A-r) AML is a subtype of particular interest, given the frequency of 20% to 25% in childhood and young adult AML.²⁰ The prognosis of patients with KMT2A-r AML varies based on the fusion partner.21, 22, 23, 24 The heterogeneous nature of the KMT2A-r cohort, with >90 recognized fusion partners, means determining the prognostic significance of each alteration is challenging.²⁵ Within small cohorts, KRAS mutations are enriched and associated with poor outcome in KMT2A-r AML; however, these genes have not been well characterized in a large pediatric AML cohort.⁸

We hypothesized that the prognostic impact of RAS pathway mutations in AML, including KMT2A-r AML, requires comprehensive evaluation. Correspondingly, we analyzed extracted DNA and RNA from 2018 AML diagnostic specimens from those enrolled in Children’s Oncology Group studies. Here, we demonstrate that KRAS mutations are associated with adverse outcomes, especially within the KMT2A-r cohort, whereas NRAS mutations lack prognostic significance. We also establish that increased activation of RAS, either because of biallelic RAS mutations, predominance of variant allele or associated loss of heterozygosity (LOH), is a significant determinant of disease outcome and confers inferior prognosis.

Methods

Patients and samples

Samples were obtained from 2018 children and young adults (aged 0-29 years) enrolled in clinical trials Children's Cancer Group-2961 (ClinicalTrials.gov identifier: NCT00002798; n = 89),²⁶ AAML03P1 (ClinicalTrials.gov identifier: NCT00070174; n = 87),²⁷ AAML0531 (ClinicalTrials.gov identifier: NCT00372593; n = 734),²⁸ and AAML1031 (ClinicalTrials.gov identifier: NCT00372593; n = 1108)²⁹ with written, informed consent collected from patients and their legal guardians in accordance with the Declaration of Helsinki. Each protocol was approved by the National Cancer Institute’s central institutional review board and the local institutional review board for each participating institution. Clinical data were available for all 2018 patients, with 1907 of those patients also having accompanying survival data, and analyses were performed for that cohort with complete data.

Mutation calls and variant allele frequency

Mutation calls were concatenated from a combination of whole-genome, targeted capture, transcriptome, and fusion capture sequencing. Mutation calls with a number of mutant reads ≥3 or a variant allele frequency (VAF) of ≥5% were kept. For each patient, if there were multiple same-gene mutation calls made for either NRAS or KRAS, mutations that did not meet the initial threshold were also kept, representing potentially late-acquired mutations. To effectively capture a population in which aberrant RAS pathway signaling was dominant, mutations with the same-gene single mutations in RAS genes with VAF ≥60% were added to the double mutant group to form the complex RAS (Comp-RAS) group. Noncanonical RAS variants were classified according to available clinical and functional evidence. Variant pathogenicity was curated using ClinVar, supplemented by published literature and hotspot annotation. Pathogenic, likely pathogenic, and variants of unknown significance noncanonical RAS variants were included for analysis; canonical hot spot mutations were analyzed separately.

Survival analysis and statistical analyses

All figures and statistics were generated in R (version 4.3.2) and plotted with ggplot2 (version 3.4.2). Associated P values were calculated using the log-rank test and then adjusted using the Benjamini-Hochberg procedure to control for the false discovery rate arising from multiple testing. After adjustment, a P value of < .05 was considered significant. The 5-year event-free survival (EFS), overall survival (OS), and relapse rate (RR) were calculated using the Kaplan-Meier method. Cumulative incidence of relapse was calculated using relapse as the primary event and by removing all competing events: induction failure, death, and death without remission. Kaplan-Meier survival curves and Cox proportional hazards ratios were generated using the survival (version 3.5-5) and survminer (version 0.4.9) R packages. Oncoprints were generated using ComplexHeatmap (version 2.16.0). The comparator group for survival analysis, in those with KRAS or NRAS mutations, was the cohort without the selected mutation.

Read calls and allelic imbalance

Same-gene G12/G13 double mutants were assessed for read coverage using short-read binary alignment mapping (BAM) files visualized in Interactive Genome Viewer. Other codon combinations were too infrequent or spaced too distantly to analyze for frequency. Missense mutations were grouped, and same-read mutations were counted manually. Allelic imbalance plots were generated using copy-number data from whole-genome sequencing. Single-nucleotide polymorphisms were plotted on top of copy-number data to screen samples for copy neutral (CN)-LOH.

Fusion detection

BAM files for fusion and mutation calls were created by extracting and purifying total RNA from diagnostic peripheral blood or bone marrow using the QIAcube automated system with AllPrep DNA/RNA/miRNA Universal Kits (Qiagen, Valencia, CA). Libraries were generated for 75-bp strand-specific paired-end sequencing using the ribodepletion version 2.0 protocol by the British Columbia Genome Sciences Center (Vancouver, BC, Canada). Libraries were sequenced using the Illumina HiSeq 2000/2500 and aligned to the hg19 (GRCh37-lite) reference genome with Burrows-Wheeler aligner version 0.5.7, applying default parameters with the addition of the “-s” option. Duplicate reads were labeled with Picard Tools. Patient fusions were identified through karyotyping or a combination of fusion-detecting algorithms: STAR-fusion version 1.8.1, TransAbyss version 1.4.10, and CICERO version 0.1.8. STAR-fusion was run with default parameters with a premade GRCh37 resource library with Gencode version 19 annotations (https://data.broadinstitute.org/Trinity/CTAT_RESOURCE_LIB/). The TransAbyss software was run with the GRCh37-lite reference genome and the following parameters: fusion breakpoint reads ≥1, flanking pairs, and spanning reads ≥2 counts. CICERO was run with default parameters with GRCh37-lite. Fusions detected computationally were verified with Fusion Inspector (version 1.8.1; Broad Institute, Cambridge, MA).

Differential gene expression analysis

Differential expression analysis was performed with DESeq2 (version 1.40.2). Complex KRAS and NRAS were combined and contrasted with other KRAS and NRAS. Low-count genes were filtered by retaining genes with ≥10 counts in a number of samples equal to the smallest sample size (n = 82; transcriptomic data were missing for 6 patients with Comp-RAS). Log-fold change shrinkage was performed on the differential expression results using the Bayes shrinkage estimator from the apeglm package (version 1.22.1). P values were adjusted using the Benjamini-Hochberg correction. Differentially expressed genes with adjusted P values < .05 were kept. Heat maps were generated using ComplexHeatmap (version 2.16.0). Z scores were generated from mRNA expression values in transcripts-per-million, and samples and genes were clustered with k-means clustering. Gene set enrichment analysis of gene ontology for biological process was performed using the results from the differential expression analysis and the clusterProfiler package (version 4.8.2). The resulting pathways were adjusted for significance using a Benjamini-Hochberg correction.

Results

Among 1907 patients with AML, 669 (35.1%) harbored NRAS or KRAS mutations. NRAS mutations (n = 518 [27.2%]) were more common than KRAS mutations (n = 216 [11.3%]); 65 patients (9.7%) had both. Both KRAS and NRAS mutations clustered at codons G12 (NRAS, n = 212; KRAS, n = 95), G13 (NRAS, n = 130; KRAS, n = 63), and Q61 (NRAS, n = 213; KRAS, n = 23; Figure 1). Noncanonical mutations were seen more commonly in KRAS (n = 46 [21.3%]) than NRAS (n = 16 [3.1%]; P < .001).

Figure 1.

AML RAS mutation frequency based upon gene, involved codon, mutation type, and complexity. (A) Lollipop plot of NRAS mutations based upon location (G12, G13, Q61, and noncanonical), complexity (single vs double), and mutation type in the AML cohort. (B) Lollipop plot of KRAS mutations based upon location (G12, G13, Q61, and noncanonical), complexity (single vs double), and mutation type in the AML cohort. GDI, guanine nucleotide dissociation inhibitor; GEF, guanine nucleotide exchange factor; NRA, nuclear receptor associated; UTR, untranslated region.

Patient characteristics for this group are summarized in supplemental Table 1. When examining for co-occurring cytogenetic alterations in this cohort, of note, KMT2A-r were enriched in KRAS mutant AML when compared with the whole cohort, a feature not seen in either the NRAS or dual-mutant cohort. KRAS mutant samples were also more likely to have co-occurring MLLT10 rearrangements and less likely to have NUP98::NSD1 fusions and CEBPA, WT1, and NPM1 mutations. When compared with the remaining cohort, CBFB::MYH11 fusions were enriched in NRAS, KRAS, and dual-mutation cohorts. FLT3-ITD mutations were depleted in NRAS, KRAS, and the dual mutation groups. When assessing treatment and treatment response, those with KRAS mutations had a higher rate of death during induction 1 (EOI1) compared with those without KRAS mutations, whereas those with NRAS mutations were less likely to be minimal residual disease (MRD) positive at EOI1. Stem cell transplant (SCT) in first remission was less common in those with NRAS, KRAS, and dual mutations compared with the remaining cohort.

Contrasting clinical implications of NRAS vs KRAS mutations in pediatric AML

KRAS mutations are associated with adverse outcomes in AML

Among the overall cohort, patients with NRAS mutations were not associated with adverse outcomes, with an EFS at 5 years from study entry of 46% (95% confidence interval [CI], 42-51) vs 45% (95% CI, 42-47; P = .25) for those without NRAS mutations and an improved OS of 67% (95% CI, 63-71) vs 61% (95% CI, 58-64; P = .02; Figure 2A). In contrast, patients with KRAS mutations had inferior outcomes, with an EFS at 5 years from study entry of 39% (95% CI, 33-46) vs 46% (95% CI, 44-48; P = .02) for those without KRAS mutations and an OS of 57% (95% CI, 50-64) vs 63% (95% CI, 61-66; P = .05; Figure 2B). Those with KRAS mutations also demonstrated a notably higher RR of 54% (95% CI, 47-61) when compared with the non-KRAS cohort RR of 46% (95% CI, 43-48; P = .009; supplemental Figure 1).

Figure 2.

Outcomes in the whole AML cohort based on RAS pathway mutation status. Kaplan-Meier estimates for the probability of EFS (left) and OS (right) stratified based on the presence (or absence) of an RAS pathway mutation (with a VAF or VAF threshold of ≥5%). The analysis was performed in the following cohorts: (A) NRAS mutation in the whole AML cohort, (B) KRAS mutation in the whole AML cohort.

KRAS mutations are enriched in KMT2A-r AML and associated with poor outcomes

We subsequently evaluated the prevalence and impact on outcome of the distinct RAS pathway mutations within KMT2A-r AML. Within this cohort, 80 (19.6%) patients had NRAS mutations, 68 (16.6%) patients had KRAS mutations, and 19 (4.6%) had both, with an aggregate incidence of 40.8% (n = 167) for all RAS mutations. There was a similar prevalence of NRAS mutations in patients with KMT2A-r compared with those without (24.2% vs 28.0%; P = .15). In contrast, KRAS mutations were enriched in patients with KMT2A-r AML compared with those without (21.3% vs 8.6%; P < .001). NRAS mutations were not significantly associated with EFS or OS as patients with and without NRAS mutations had similar EFS of 30% (95% CI, 22-41) vs 35% (95% CI, 30-40; P = .44) and OS of 53% (95% CI, 44-65) vs 52% (95% CI, 47-58; P = .89; supplemental Figure 2). However, the presence of KRAS mutations was associated with poor outcome in the KMT2A-r group, with demonstrably worse EFS of 22% (95% CI, 15-33) vs 37% (95% CI, 32-42; P < .001) and OS of 43% (95% CI, 33-55) vs 55% (95% CI, 50-61; P = .01) compared with those without KRAS mutations (Figure 3A). Those with KRAS mutations in the KMT2A-r cohort demonstrated a RR of 73% (95% CI, 60-82), which was higher compared with the non-KRAS KMT2A-r cohort (58% [95% CI, 51-63]; P = .007).

Figure 3.

Outcomes in KMT2A-r AML based on RAS mutation status. Kaplan-Meier estimates for the probability of EFS and OS stratified based on the presence (or absence) of a RAS pathway mutation (with a VAF or VAF threshold of ≥5%). The analysis was performed in the following cohorts: (A) KRAS mutation in KMT2A-r AML cohort, (B) KRAS mutation in SR KMT2A-r AML cohort.

To mitigate possible bias driven by high-risk fusion partners, we excluded patients with KMT2A-r AML with known high-risk fusions (KMT2A::AFF1, KMT2A::MLLT4, KMT2A::ABI1, KMT2A::MLLT10, and KMT2A::MLLT1) and reanalyzed the impact of NRAS and KRAS mutations in the standard risk KMT2A-r AML (SR KMT2A-r) cohort. Analysis of NRAS mutations (n = 48 [22.6%]) in the SR KMT2A-r cohort demonstrated no significant difference in EFS or OS (supplemental Figure 2). Among the SR KMT2A-r cohort, 41 patients (19.3%) had KRAS mutations, and this cohort demonstrated inferior outcomes compared with those without KRAS mutations with an EFS of 25% (95% CI, 15-43) vs 46% (95% CI, 39-53; P = .03), an OS of 47% (95% CI, 34-66) vs 64% (95% CI, 58-72; P = .012), and an RR of 70% (95% CI, 49-82) vs 49% (95% CI, 41-52; P = .007; Figure 3B).

Comp-RAS mutations determine outcome in AML

Biallelic and high VAF RAS mutations are a distinct entity in AML

Given the number of patients in this cohort with dual RAS mutations, we studied the prognostic implications of dual mutations in AML, as double mutations would lead to unopposed activation of the affected RAS gene without the mitigating effects of the normal allele. Double mutations in NRAS or KRAS were seen in 77 patients (12% of all RAS mutations), of whom 57 patients had double NRAS mutations, 18 patients had double KRAS mutations, and 2 patients had double mutations in both genes (Figure 4A). Short-read BAM data were available for a subset (n = 8 [9%]) of these patients. Interrogation of the sequence data demonstrated that the 2 mutations occurred in different reads in 99% of reads analyzed, suggesting the likelihood of biallelic RAS mutations (supplemental Figure 3).

Figure 4.

AML Comp-RAS cohort. (A) Incidence of double RAS mutations based on isoform (KRAS or NRAS) in the whole AML cohort. (B) Visual representation of RAS VAF distribution with >60% used to predict dominance of aberrant RAS signaling in single mutant RAS samples.

Single RAS mutations with high VAF may also reflect loss of the normal allele through copy number variation or CN-LOH. Therefore, we used VAF as a surrogate for allelic imbalance. We identified 14 (2%) patients with single RAS mutations (n = 12 NRAS, n = 2 KRAS) with a VAF of >60% (Figure 4B). Remission MRD data were available in 4 (29%) of these patients, and all 4 of these patients demonstrated no evidence of RAS mutations in their remission sample, thus indicating these observed high VAF RAS mutations were not germ line in origin in these patients. Whole-genome sequencing data were available in 6 patient samples, and in 2 of these samples with NRAS mutations with high VAF (VAF of 82% and 75%), we demonstrated CN-LOH in chromosome 1p, demonstrating the underlying mechanism of haplo-insufficiency (supplemental Figure 3).

These 2 groups, biallelic RAS mutations and RAS mutations with VAF >60%, were combined to form a “Comp-RAS” group, given a hypothesized shared mechanism whereby dominance of the mutated RAS gene leads to subsequent hyperactivation of the downstream effector pathway. After accounting for group overlap, 69 patients met criteria to be included in the NRAS mutant Comp-RAS cohort (Comp-NRAS) and 21 patients in the KRAS mutant Comp-RAS (Comp-KRAS) cohort. Supplemental Table 2 describes the clinical and molecular characteristics of the Comp-RAS cohorts. Comparison of the Comp-RAS cohorts with non–Comp-RAS mutant AML demonstrated broadly similar baseline clinical characteristics, molecular comutation patterns, and early treatment response profiles.

When examining outcomes for these groups, the Comp-NRAS cohort demonstrated 5-year EFS of 37% (95% CI, 27-50) compared with 45% (95% CI, 42-47; P = .29) in the non-NRAS mutant AML cohort, and 47% (95% CI, 43-53; P = .16) compared with the non–Comp-NRAS mutant AML cohort (Figure 5A). The Comp-KRAS cohort demonstrated 5-year EFS of only 15% (95% CI, 5-43), significantly worse than the EFS for both patients who were non-KRAS mutant, at 45% (95% CI, 44-48; P = .006) and those who were non–Comp-KRAS mutant at 41% (95% CI, 35-49; P = 0.039; Figure 5B). Similarly, the Comp-KRAS cohort experienced an increased RR of 83% (95% CI, 53-94), which compared unfavorably to both the non–Comp-KRAS cohort at 49% (95% CI, 40-49; P = .0001) and the non-KRAS mutant cohort at 46% (95% CI, 43-49; P = .0001). Given the impact of Comp-RAS status on outcome in the AML cohort, we sought to determine the role mutation complexity plays in the adverse outcomes seen in the overall KRAS mutant cohort of AML. After removing the Comp-RAS cohort, there were no significant differences in EFS (P = .39), OS (P = .44), or RR (P = .48) from study entry (Figure 5B) when non–Comp-KRAS mutant AML was compared with non-KRAS mutant AML.

Figure 5.

Outcomes in the whole AML cohort based on the presence of Comp-RAS mutation status. Kaplan-Meier estimates for the probability of EFS (left) and OS (right) stratified based on the presence (or absence) of a Comp-RAS pathway mutation (defined as dual RAS gene mutation or VAF >60%). The analysis was performed in the following cohorts: (A) Comp-NRAS mutation in the whole AML cohort, (B) Comp-KRAS mutation in the whole AML cohort.

The Comp-RAS group demonstrates inferior outcomes in KMT2A-r AML

We next interrogated the prognostic significance of Comp-RAS within the KMT2A-r AML cohort, where it occurred in 25 patients (13%; n = 9 KRAS, n = 15 NRAS, n = 1 both). In the KMT2A-r cohort, the Comp-NRAS group demonstrated 5-year EFS of 13% (95% CI, 3-46), inferior to both the non-NRAS mutant (35% [95% CI, 30-40]; P = .09) and non–Comp-NRAS KMT2A-r (34% [95% CI, 24-46]; P = .09) cohorts (Figure 6A). The Comp-KRAS cohort demonstrated 5-year EFS of 10% (95% CI, 2-64), which was inferior to both the non–Comp-KRAS mutant KMT2A-r cohort of 23% (95% CI, 16-35; P = .95) as well as the non-KRAS cohort (37% [95% CI, 32-42]; P = .12; Figure 6B). Given the strong signal seen in the Comp-KRAS cohort among the KMT2A-r cytogenetic subgroup, we once again analyzed the patients in the non–Comp-KRAS cohort to determine whether Comp-KRAS status was driving inferior outcomes. In this instance, the patients in the non–Comp-KRAS cohort continued to have an inferior 5-year EFS compared with the non-KRAS cohort (P = .008) when the patients in the Comp-KRAS cohort were excluded (Figure 6B). Additionally, we examined the role of Comp-RAS status in the previously described SR-KMT2A-r cohort, which included only 15 patients (n = 5 KRAS, n = 10 NRAS). Within this cohort, patients in the Comp-NRAS cohort demonstrated EFS of 20% (95% CI, 3-48), which was inferior to patients in both the non-Comp NRAS cohort (40% [95 CI, 28-55]; P = .28) and the non-NRAS cohort (44% [95% CI, 37-51]; P = .18). Patients in the Comp-KRAS cohort within the SR-KMT2A-r subgroup demonstrated an EFS of 20% (95% CI, 1-59), which compared unfavorably to both the non–Comp-KRAS cohort (26% [95% CI, 13-41]; P = .85) and the non-KRAS cohort (46% [95% CI, 39-53]; P = .18; supplemental Figure 4).

Figure 6.

Outcomes of the KMT2A-r AML cohort based on the presence of Comp-RAS mutation status. Kaplan-Meier estimates for the probability of EFS (left) and OS (right) stratified based on the presence (or absence) of a Comp-RAS pathway mutation (defined as dual RAS gene mutation or VAF >60%). The analysis was performed in the following cohorts: (A) Comp-NRAS mutation in KMT2A-r AML cohort, (B) Comp-KRAS mutation in KMT2A-r AML cohort.

Comp-RAS is an independent poor prognostic marker with a distinct transcriptomic profile

Given the potential for enriched molecular alterations to affect outcomes, we examined co-occurring alterations within both the non–Comp-RAS mutant and Comp-RAS cohorts. Oncoprint of co-occurring mutations showed few co-occurring mutations and no enrichment of alterations in Comp-RAS samples (χ² P > .05; Figure 7A). Data from our patient characteristics table also noted a strong enrichment of KMT2A-r AML in those with KRAS mutations. This raises the possibility that poorer outcomes observed in Comp-RAS AML and KRAS mutant KMT2A-r AML may be driven by an overrepresentation of KMT2A rearrangements. To determine whether observed RAS mutations have independent prognostic significance, we performed multivariable analysis (MVA) for EFS, including all relevant high-risk covariates (including cytogenetic risk groups, recurrent molecular lesions, MRD status, and relevant clinical variables). Hazard ratios (HRs) for EFS, from multivariable Cox-proportional hazards, demonstrated that a Comp-KRAS determination was an independent poor prognostic indicator of EFS for those with AML in the whole cohort (HR, 1.8 [95% CI, 1.08-3.09]; P = .02; Figure 7B). There was also a numerical suggestion that Comp-NRAS was an independent poor prognostic indicator of EFS in the whole cohort, though this result was not statistically significant (HR, 1.4 [95% CI, 0.99-2.00]; P = .05). A MVA for EFS among the KMT2A-r subgroup also identified non–Comp-KRAS as independently associated with inferior EFS (HR, 1.65 [95% CI, 1.15-2.38]; P = .007) whereas Comp-RAS status was also independently associated with inferior EFS in this cohort (HR, 1.98 [95% CI, 1.18-3.35]; P = .01; Figure 7C). In this KMT2A-r MVA, the Comp-RAS cohort was combined, given the small sample size and relatively equal distribution (Comp-NRAS = 16; Comp-KRAS = 10).

Figure 7.

Comp-RAS is independently associated with inferior outcomes in AML. (A) Oncoprint of co-occurring mutations in the whole AML cohort shows few co-occurring mutants and no enrichment of mutations or fusions in Comp-RAS samples (χ²P > .05). (B) Multivariate CoxPH identifies Comp-KRAS as being independently associated with inferior outcomes in the whole cohort. (C) Multivariate CoxPH identifies Comp-RAS and non–Comp-KRAS as being independently associated with poor prognosis in KMT2A-r AML. CBFB-MYH11, core-binding factor subunit beta-myosin heavy chain 11; CEBPA, CCAAT/enhancer-binding protein alpha; CoxPH, Cox proportional hazards; CR EOI1, complete remission at end of induction 1; GATA2, GATA-binding protein 2; NPM1, nucleophosmin 1; PTPN11, protein tyrosine phosphatase non-receptor type 11; RUNX1-RUNX1T1, RUNX family transcription factor 1-RUNX1 partner transcriptional co-repressor 1; SCT CR, stem cell transplantation in complete remission.

We next attempted to disentangle the deleterious effect of KMT2A rearrangements with the observed effect of RAS mutations, focusing on the Comp-KRAS cohort as the only subgroup associated with prognostic significance independent of KMT2A-r status. In the non–KMT2A-r cohort, Comp-KRAS status still conferred inferior EFS (5-year EFS, 21% [95% CI, 6-71]) when compared with the non–Comp-KRAS cohort (5-year EFS, 53% [95% CI, 44-62]; P = .03) and the non-KRAS cohort (5-year EFS, 48% [95% CI, 46-51]; P = .09). Additionally, in the non–KMT2A-r cohort, Comp-KRAS status was also associated with high rates of relapse (RR, 76% [95% CI, 22-93]) when compared with both the non–Comp-KRAS cohort (RR, 40% [95% CI, 29-49]; P = .007) and the non-KRAS cohort (5-year cumulative incidence of relapse, 42% [95% CI, 39-45]; P = .009; supplemental Figure 5).

We also examined whether SCT or exposure to gemtuzumab ozogamicin (GO) modified the adverse prognostic impact of Comp-RAS mutations in the overall cohort or KRAS mutations within the KMT2A-r subgroup. Analyses were limited by small sample size, particularly within the Comp-KRAS cohort. In subgroup analyses stratified by SCT and GO exposure, the inferior outcomes associated with these RAS categories were maintained. Inferior EFS was noted in the Comp-KRAS group who did not receive GO (P = .03). There was no evidence that any intervention mitigated the adverse prognostic effects seen in these groups (supplemental Figures 6 and 7).

Finally, given the significant difference in outcomes observed between the Comp-RAS and non–Comp-RAS cohorts, we sought to determine whether differences in transcriptional profiles could provide insight into the underlying biology driving these disparities. To this end, we performed a differential gene expression analysis comparing diagnostic RNA-sequencing data from patients in the Comp-RAS and non–Comp-RAS groups. This analysis revealed distinct transcriptional patterns, with gene set enrichment analysis demonstrating suppression of immune-related pathways, most notably those involved in adaptive immune responses, and concomitant upregulation of pathways regulating transmembrane ion transport in the Comp-RAS cohort (supplemental Figure 8).

Discussion

AML is a heterogeneous disease, and this work demonstrates the importance of considering mutation complexity, including among distinct cytogenetic subgroups, when analyzing RAS pathway mutations. Here, we clarify prognostic implications of NRAS vs KRAS mutations in AML, establishing a lack of association of NRAS mutations overall with outcome. Crucially, we demonstrate that KRAS mutations are enriched in KMT2A-r AML and are associated with inferior outcomes, particularly for patients with Comp-KRAS. We show that this is independent of risk-stratifying KMT2A lesions and that the prognostic impact is seen in the SR KMT2A-r cohort. This work critically demonstrates that KRAS can help identify a cohort of high-risk KMT2A-r cases that are considered SR by the current classification schema. This aligns with recently published data in both adult groups and smaller pediatric studies identifying KRAS as an adverse prognostic factor in those with KMT2A-r.^8,30,31 The biologic explanation for gene and subgroup-specific outcomes in leukemia remains incomplete; however, our findings support the functional differences in distinct RAS genes and proteins. Among all RAS proteins, KRAS is recognized as an oncogenic mutation that influences a cell’s neoplastic transformation and progression, with independent mechanisms of oncogenic signaling when compared with NRAS.^32,33 Further studies are required to elucidate the exact biological mechanism underpinning these associations.

Our data refine the prognostic role of RAS mutation complexity across AML subgroups. In the overall cohort, adverse outcomes were driven primarily by Comp-KRAS, whereas Comp-NRAS showed only a nonsignificant trend toward inferior survival, likely reflecting limited power. In KMT2A-r AML, Comp-RAS status emerged as an independent adverse factor on MVA. Importantly, overlap between Comp-NRAS and Comp-KRAS was minimal (n = 1), indicating that outcomes in the Comp-NRAS subgroup were not driven by co-occurring Comp-KRAS. Together, these findings support gene- and context-specific effects of RAS mutation complexity and demonstrate that non-Comp-KRAS remains independently adverse in KMT2A-r AML.

We hypothesize that the adverse outcomes observed in Comp-KRAS AML and Comp-RAS KMT2A-r AML reflect dominant oncogenic signaling driven by mutant RAS clones. An important aspect of our analysis was the identification of a subset of patients harboring high-VAF (>60%) RAS mutations, which we interpret as representing dominant leukemic clones or LOH. Although RAS mutations are frequently detected as subclonal events in AML and can be lost at relapse, emerging evidence suggests that oncogenic RAS signaling operates in a dose-dependent manner,34, 35, 36 with mutant allele burden influencing signaling output and cellular fitness. Experimental models have demonstrated that wild-type RAS alleles can partially restrain oncogenic signaling,^34,37 whereas increased mutant allele dosage enhances MAPK pathway activation and leukemogenic potential. Consistent with this concept, selection for increased mutant allele imbalance is widely observed across cancers,³⁸ supporting the notion that higher oncogenic allele burden may confer a competitive advantage. These observations provide a biologically plausible explanation for the adverse outcomes observed in patients with dominant RAS-mutant clones in our cohort.

To further explore potential biological explanations for this observation, we performed additional genomic and transcriptional analyses. Short-read BAM data, available in ∼10% of this cohort, suggest the mutations are biallelic, therefore leaving no wild-type allele and favoring aberrant RAS pathway signaling. Similarly, the presence of CN-LOH in the 2 samples with the highest VAF supports CN-LOH as another possible mechanism for hyperactive RAS signaling. Additionally, our transcriptional analyses further suggest that the adverse outcomes observed in these cohorts may, in part, be driven by altered immune and microenvironmental dynamics. The suppression of genes involved in adaptive immune responses in this cohort may reflect impaired immunosurveillance, potentially limiting clearance of residual leukemic cells after induction therapy. This is consistent with previous observations that diminished immune activation signatures in AML are associated with higher relapse risk and treatment resistance.^39,40 Concurrently, the upregulation of transmembrane ion transport pathways may influence leukemic cell survival through effects on membrane potential, intracellular ion homeostasis, and downstream signaling cascades. All these changes have been implicated in promoting proliferation, metabolic adaptation, and resistance to apoptosis in myeloid malignancies.41, 42, 43 Therefore, these coordinated transcriptional changes may create a permissive microenvironment for leukemic persistence, providing a potential mechanistic link between the molecular complexity of RAS mutations and the aggressive clinical phenotype observed in Comp-KRAS AML.

Our findings support the concept that leukemic fitness depends on the degree of RAS pathway activation. Together with previous evidence that RAS allelic imbalance enhances oncogenic fitness,^32,44,45 these data suggest that dominant RAS signaling may mark a biologically aggressive AML subset and raise the possibility that Comp-RAS leukemias could be especially susceptible to RAS-directed therapy.

These findings support the incorporation of RAS pathway alterations into upfront molecular risk stratification. In particular, KRAS-mutant KMT2A-r AML, including standard-risk fusion partners, may warrant higher-risk classification and consideration of intensified consolidation. Although our retrospective analysis did not demonstrate a statistically significant mitigation of risk with hematopoietic stem cell transplantation, there was a numerical suggestion of benefit among patients with KRAS-mutant KMT2A-r AML who underwent transplantation. This is consistent with emerging adult data suggesting that SCT may mitigate the adverse prognostic impact of KRAS mutations.³⁰ Prospective studies are needed to determine whether transplantation or other intensified approaches can meaningfully offset the adverse biology of KRAS-driven pediatric AML.

Although KRAS mutations were enriched among patients with KMT2A-r AML, SCT in first complete remission was less frequently performed in patients with NRAS, KRAS, or dual RAS mutations within this analysis. This likely reflects the fact that RAS mutation status was not used to guide SCT allocation during the study period. In addition, patients with RAS-mutant AML, particularly those with Comp-RAS alterations, experienced higher rates of early relapse and morbidity, which may have limited the opportunity to proceed to SCT in first remission despite high-risk cytogenetic features.

We demonstrate that KRAS mutations are enriched and associated with inferior outcomes in KMT2A-r AML while also delineating the critical role of KRAS mutation status to define a high-risk population within the broader and standard-risk KMT2A-r AML cohort. We also define the “Comp-RAS” cohort as those with double mutations, VAF >60%, or CN-LOH. The Comp-KRAS cohort demonstrates inferior outcomes when compared with both the non–comp-RAS mutant and non-KRAS/NRAS population. Thus, a comprehensive assessment of RAS mutations is required to determine treatment and prognostic implications. Additional analysis in prospective cohorts, along with more detailed mechanistic studies, is required to further evaluate the impact of RAS signaling within AML.

Conflict-of-interest disclosure: The authors declare no competing financial interests.