Longitudinal genomic characterization of pediatric acute myeloid leukemia at diagnosis and relapse.

10020 Background: Relapse remains the leading cause of treatment failure and mortality in pediatric acute myeloid leukemia (pAML). In AML, relapse is common and is increasingly recognized as a consequence of therapy-driven clonal evolution rather than mere persistence of the diagnostic clone. This study aimed to delineate mutational trajectories from diagnosis to relapse and identify relapse-associated genomic alterations of potential clinical relevance. Methods: This prospective study was conducted from July 2021 to December 2025. Bone marrow or peripheral blood samples were collected from children ≤18 years at diagnosis and relapse. Genomic DNA from paired samples was sequenced exomically to identify somatic mutations. Longitudinal comparative analyses assessed clonal persistence, emergence of relapse-specific mutations, and pathway-level enrichment. Results: A total of 100 children with AML were enrolled. Paired longitudinal analysis (n=18) identified 133 mutated genes. Seventy-nine (59.4%) were shared between diagnosis and relapse, forming a conserved truncal genomic backbone. Fifty genes (37.6%) were exclusive to diagnosis, while only four (3.0%) were exclusive to relapse, suggesting strong therapeutic selection pressure rather than widespread genomic diversification. At diagnosis, missense variants predominated (88.4%), followed by in-frame insertions/deletions (6.5%), nonsense (2.4%), and frameshift alterations (2.5%). Relapse samples showed further enrichment of missense variants (90.5%) and a relative depletion of truncating events, consistent with selection for functionally adaptive mutations. Truncal alterations were consistently observed in epigenetic regulators TET2 and KMT2C, supporting their role as early leukemogenic events. In contrast, relapse samples showed enrichment of signaling and transcriptional regulators, including FLT3 and PTPN11, along with relapse-exclusive emergence of WNT pathway components (FZD7, LEF1) and the PI3K regulator PIK3R1. Functional enrichment analysis revealed that diagnostic samples were primarily enriched for DNA damage response, p53 signaling, apoptosis, and cell-cycle checkpoint pathways (p<0.05). At relapse, these core programs persisted alongside marked enrichment of oncogenic signaling pathways (WNT, PI3K–AKT, FLT3) and metabolic processes, including glycolysis and ATP generation. Reactome analysis highlighted convergence on FLT3- and AKT-driven signaling, TP53-mediated DNA repair, and escape from apoptosis. Conclusions: Longitudinal genomic profiling demonstrates that pAML relapse is driven by the persistence of early truncal mutations, coupled with the selective expansion of signaling- and metabolism-adapted subclones. Enrichment of WNT, PI3K–AKT, and FLT3-associated pathways at relapse reveals potential targets for relapse-directed precision strategies.