Paper Details

ABSTRACT

Introduction: Pediatric B-cell acute lymphoblastic leukemia (B-ALL) comprises genetically diverse molecular subtypes with distinct developmental programs and clinical outcomes. While molecular subtype is a major determinant of prognosis, patients within the same subtype frequently experience different clinical outcomes, suggesting that additional functional cell-state features contribute to relapse risk. Our recent study (PMID: 41576347) demonstrated that signaling and metabolic remodeling are associated with relapse risk in pediatric B-ALL. Whether these relapse-associated functional cell states are conserved across genetically distinct subtypes is unknown. We investigated whether diverse B-ALL molecular subtypes converge onto shared relapse-associated metabolic and signaling cell states at diagnosis.

Methods: Diagnostic bone marrow samples from 137 children with de novo B-ALL treated on Children's Oncology Group or St. Jude Children’s Research Hospital protocols, spanning multiple molecular subtypes and comprehensively annotated with clinical variables, were analyzed. Leukemia cells were profiled using two complementary single-cell mass cytometry (CyTOF) antibody panels with shared lineage markers. Together, the panels measured 76 unique proteins spanning B-cell development, intracellular signaling, transcription factors, metabolism, and apoptosis. To characterize metabolic and signaling cell states, cells were clustered independently within each CyTOF panel based on common lineage markers, together with either metabolic or signaling proteins. Patient-level glycolysis, oxidative phosphorylation (OxPhos), and signaling scores were calculated as frequency-weighted averages of metabolic or signaling cluster centroids. Metabolic heterogeneity was quantified using Shannon entropy and cluster dominance. Additionally, 48 samples were profiled using CITE-seq with lineage markers shared with the CyTOF panels, enabling future integration of the two datasets using our multimodal fusion framework, CellFuse for complementary transcriptomic and proteomic analysis.

Results: We observed molecular subtype-specific phenotypic partitioning, similar with previously described transcriptomic partitioning, with Hyperdiploid and PAX5alt leukemias forming distinct lineage states. Hyperdiploid ALL (n=19) occupied a CD19/TdT/PAX5-high, Ki67-high PreBI-like state, while PAX5alt ALL (n=11) localized to a distinct CD20/IgMi-high, Ki67-low pre-B-cell state PreBI-like state. Ph+ (n=6) and Ph-like (n=18) leukemias were distributed across broader, more immature ProBI and ProBII compartments, including CD34-high populations. Lineage-matched comparison of signaling and metabolic clusters identified relapse-associated cell states with coordinated signaling and metabolic programs. The relapse-enriched cluster showed high pS6, pERK, and pCREB in the context of a glycolytically active state, characterized by elevated GLUT1 and LDHA. Using conditional Density Resampled Estimate of Mutual Information (DREMI) to quantify dependency between protein pairs, we observed stronger pS6-pERK, pS6-pCREB, pERK-pCREB, and pS6-p4EBP1 coupling in relapse cells, indicating coordinated mTOR/MAPK/CREB signaling. In contrast, the continuous remission-enriched cluster showed a similar signaling profile to the relapse-enriched cluster but lower Ki67 and glycolytic marker expression, suggesting a less proliferative and less glycolytically active state.

Metabolic organization was largely independent of molecular subtype, with metabolic clusters comprising cells from multiple molecular subtypes, yet iAMP21, MEF2D and KMT2A-rearranged ALL demonstrated subtype specific metabolic states. Patients who would relapse exhibited higher metabolic heterogeneity (Shannon entropy, median 1.70 vs 1.30, P = 0.010) and lower cluster dominance (median 48% vs 60%, P = 0.031), suggesting leukemia cells are distributed across multiple metabolic states rather than concentrated within a single dominant metabolic program.

Conclusions: Pediatric B-ALL leukemias exhibit molecular subtype-specific proteomic partitioning, with outcome-associated signaling states coupled to distinct metabolic programs. These states are shared across molecular subtypes, suggesting convergent features. These findings demonstrate that relapse risk is associated with state-specific layer on subtype-specific proteomic architecture. Integration of CyTOF and CITE-seq data using CellFuse is ongoing.