Multiscale RNA editing analysis reveals cell type–specific regulatory programs across disease states in acute myeloid leukemia

Abstract Background Acute myeloid leukemia (AML) is characterized by marked cellular heterogeneity and immune dysregulation. Adenosine-to-inosine (A-to-I) RNA editing, primarily catalyzed by ADAR and ADARB1, represents an important post-transcriptional regulatory mechanism, yet its condition- and cell type–specific landscape in AML remains poorly defined, particularly at single-cell resolution. Methods We analyzed publicly available single-cell RNA sequencing data from healthy donors (HL), newly diagnosed AML (ND), remission (RM), and persistent disease (PO), integrating single-cell and pseudo-bulk analyses in a multiscale framework. RNA editing sites were identified using a stringent discovery pipeline and quantified at both pseudo-bulk and cell type–resolved levels. Differential RNA editing was assessed using regression-based read-count models, primarily beta-binomial regression with subject-specific random effects when applicable. Pairwise contrasts between clinical conditions were evaluated using delta-method inference, with statistical significance defined by false discovery rate and a minimum effect-size threshold. Selected editing sites were examined in independent human AML cohorts for validation and clinical association. Results We identified 2,875 recurrent A-to-I RNA editing sites enriched in intronic and 3′ untranslated regions and linked to immune and inflammatory pathways. At the pseudo-bulk level, 150 sites were differentially edited across clinical states, and global RNA editing varied by condition, showing an overall negative association with ADAR and ADARB1 expression with context-dependent exceptions. Cell type–resolved analyses identified 148 differentially edited sites with strong lineage specificity. In ND, leukemia-associated cell states consistently exhibited lower editing than lineage-matched healthy counterparts. T cells consistently harbored differential editing signals across all condition contrasts, while progenitor-like cells showed the strongest RM-versus-ND differences despite minimal changes in global editing. Notable editing events were observed in GBP4, SPN, TNFSF10, EMB , and FKBP5 . Several candidate sites were validated in independent AML cohorts and were associated with clinical features. Conclusions This multiscale analysis reveals that RNA editing in AML is condition- and cell type–specific and is not fully captured by bulk transcriptomic measures. Site-specific, lineage-restricted RNA editing represents a distinct regulatory layer that reflects disease state and cellular context, highlighting its potential relevance for understanding AML biology and informing future biomarker development.