Protein expression signatures reveal age-specific risk patterns in B-ALL

Abstract Introduction Current B-cell acute lymphoblastic leukemia (B-ALL) risk stratification relies on clinical and genetic features, but often doesn't accurately predict response or relapse. We hypothesized that Proteomic profiling may provide a superior classification approach, but large-scale proteomic studies comparing pediatric (PED) and adult (AD) B-ALL are limited. Methods Leukemia blasts from pre-treatment patient bone marrow (BM) or peripheral blood (PB) samples from 273 PED and 390 AD B-ALL cases were isolated to >80% purity and protein lysates prepared. A custom reverse-phase protein array (RPPA) was printed and probed with 433 antibodies. CD34+ cells from normal, non-G-CSF-stimulated BM were used as controls, and patient protein expression was normalized against these. Protein expression signatures (PES) were analyzed using hierarchical clustering, Pearson correlation, and classification and regression tree (CART) analysis. The proteins were divided into 65 Protein Functional Groups (PFGs) based on known shared pathways or biological functions. Hierarchical clustering within PFGs was performed using the ProgenyClust algorithm, and hierarchical “metagalaxy” clustering was performed using PFG cluster membership to identify constellations of related PFGs and recurrent PES. Survival analysis using Kaplan-Meier curves, Cox proportional hazards models, and multivariate analysis was performed to assess overall survival (OS), remission duration (RD), and prognostic significance of individual proteins, PFG clusters, and metagalaxy-derived signatures. PED patients were treated on COG protocols, including AALL1731 and AALL1732, and AD patients with a variety of intensive chemotherapy regimens, most based on the HCVAD Rituximab/Inotuzumab/Blinatumumab backbone. Results Univariate survival analysis of individual proteins revealed 70 proteins significantly associated with clinical outcomes in PED patients, 91 in AD, with 26 shared in both. We first divided the proteins into 65 PFGs and then determined the optimal number of clusters within each PFG. PED and AD cases were proportionally represented in all 65 PFGs. We then clustered the patients based on PFG cluster membership, identifying 13 PES, based on 14 distinct constellations of correlated PFG clusters. The PES fell into three patterns, with 4 being heavily or exclusively dominated by PED, 5 by AD, and 4 mixed. We next searched for associations between PES membership and known clinical features, such as Ph+, Ph-like, CRLF2, or KMT2A status, but found none (Fisher's exact tests, all p > 0.05). However, PES showed a significant association with traditional risk group stratification in both groups. Among PED, most signatures demonstrate excellent outcomes (good-risk: 5-year OS 89.1%), but signatures 6, 9, and 10 show inferior survival (high-risk: 5-year OS 74.5%). Among AD, signatures stratified patients into relatively better-risk (5-year OS 54.7%) and worse-risk groups (5-year OS 38.6%). Protein expression signatures demonstrated prognostic significance in univariate analysis in both cohorts (PED HR=2, p=0.02, AD HR=1.6, p=0.007). In multivariate analysis, statistical significance was maintained in both the AD (HR=1.53, 95% CI: 1.09-2.15, p=0.013) and PED (HR=1.98, 95% CI: 0.999-3.92, p=0.050) populations. Most children in PED-predominant groups (signatures 5, 11, 12) achieved excellent outcomes, with only signature 10 representing an aggressive PED-specific subtype. However, PED whose protein expression resembled AD disease patterns, faced significantly higher risks, with poor outcomes more than doubling when children clustered with AD-predominant or mixed signatures. These high-risk patients predominantly fell into signatures 6 and 9, where AD-like molecular patterns consistently predicted treatment failure in children. This reveals two mechanisms of poor prognosis in pediatric B-ALL: intrinsically aggressive PED biology (signature 10) versus AD-type disease occurring in children (signatures 6 and 9). Conclusion We identified recurrent protein expression signatures in B-ALL with independent prognostic value beyond traditional risk factors in both PED and AD groups. Pediatric patients show two pathways to poor outcomes: aggressive pediatric-specific biology (signature 10) and adult-type disease in children (signatures 6 and 9), supporting the hypothesis that “kids who look like adults” harbor more aggressive disease biology.