Chronic Lymphocytic Leukemia (CLL), the most common adult leukemia, poses diagnostic and prognostic challenges. Current prognostic markers, such as immunoglobulin heavy variable mutation status (IGHV), cytogenetic aberrations, and ZAP70/CD38 expression, help stratify risk, but their relevance in the modern targeted therapy era is debated. CLL treatment has evolved from cytotoxic therapies to targeted agents inhibiting B-cell receptor signaling (BTK, PI3K inhibitors) or apoptosis regulation (BCL2 inhibitors). The decision to initiate treatment and select therapy is complex, relying on symptoms, patient factors, and prognostic markers. This study hypothesized that proteomic analysis, reflecting the integrated consequence of genetic, epigenetic, and environmental influences, could provide crucial information to improve prognostication, therapy selection, and identify novel therapeutic targets. Existing methods for quantifying these influences are insufficient; therefore, a proteomic approach was employed.

The introduction section reviews the history of CLL treatment, highlighting the shift from cytotoxic therapies to targeted therapies. It discusses the limitations of current prognostic markers in the context of modern treatment and the need for a more comprehensive approach to patient risk stratification and therapy selection. The existing literature supports the idea that proteomics could provide a more complete picture of the disease's complexity and inform better treatment strategies. However, previous proteomic studies of CLL were limited by small sample sizes, preventing robust classification and prognostication.

Reverse phase protein array (RPPA) was used to analyze protein expression in 871 patient samples (795 CLL, others MSBL). Samples included both fresh and frozen blood and bone marrow specimens collected over several years. The RPPA platform probed for 384 validated antibodies targeting both total and post-translationally modified proteins. Rigorous data normalization and quality control steps were performed to mitigate biases from sample collection, processing, and treatment status. Data were analyzed at three levels: individual proteins, functionally related protein groups (PFGs), and system-wide signatures (SGs) using various computational methods including principal component analysis (PCA), k-means clustering, hierarchical clustering, and random forest. Statistical analyses (Kaplan-Meier, Cox hazard, ANOVA, Tukey HSD, Chi-square, Fisher exact, Kruskal-Wallis) were used to assess relationships between protein expression patterns and clinical outcomes (overall survival (OS), time to first treatment (TTF), time to second treatment (TST)). Multivariate analyses were conducted to evaluate the independent prognostic value of proteomic signatures alongside established clinical and molecular markers. A Random Forest approach identified a 30-protein classifier set for prospective patient classification into signature groups. The Leukemia Protein Atlas website was used to share the data and analyses.

The study identified six recurrent proteomic signatures (SGs A-F) that were strongly prognostic for OS, TTF, and TST, surpassing the predictive power of traditional markers (IGHV, cytogenetic abnormalities, Rai/Binet staging). SG membership was an independent predictor of outcome in multivariate analyses. A novel SG (SG-A) was identified, encompassing 5% of CLL cases, characterized by hairy cell leukemia-like proteomics but poor response to all therapies. This group was termed Hybrid Cell Proliferative CLL (HCPLC). Analysis of protein functional groups (PFGs) revealed that a high percentage were prognostic, highlighting the importance of a systems-level approach. The study identified numerous individually prognostic proteins, including both known and novel markers. Specific PFGs associated with treatment response were identified, showing differential responses to various therapeutic classes (BTK inhibitors, chemotherapy, antibody therapy). A 30-protein classifier set was developed, achieving 77% accuracy for SG-A and 87.5% for SG-C, crucial for clinical application. Analysis of differentially expressed proteins within each SG suggested potential therapeutic targets for each signature group. The potential use of ANXA1, TFRC, and SMAD2:p245 as predictors of time to first treatment (TTFT), particularly in early-stage patients, was highlighted.

This study demonstrates the considerable prognostic and therapeutic value of proteomic profiling in CLL. The proteomic classification system provided more precise prognostic information than traditional methods, which is crucial for tailoring treatment strategies. The identification of novel signature groups, particularly SG-A (HCPLC), and specific differentially expressed proteins within each signature group suggests novel therapeutic targets and potential biomarkers for monitoring treatment response. The development of a 30-protein classifier set moves the approach closer to clinical translation and implementation. The systems biology approach, by considering the interplay of multiple proteins and pathways, captures the complexity of CLL pathogenesis, surpassing the information obtained from single markers. The findings have implications for refining risk stratification, optimizing treatment selection, and accelerating the development of novel targeted therapies for CLL.

This large-scale proteomic study of CLL provides a novel classification system significantly improving prognostication and therapy selection. Six signature groups were identified, each with distinct clinical characteristics and therapeutic implications. The identification of a unique, therapy-resistant subgroup (HCPLC) and several potential therapeutic targets opens new avenues for CLL research. Future studies should focus on validating these findings in larger, independent cohorts and translating the 30-protein classifier set into a clinically useful diagnostic tool. Further investigation into the biological mechanisms driving the observed proteomic signatures is also warranted.

The study acknowledges limitations, including potential bias in the selection of proteins analyzed, the relatively small number of patients treated with modern therapies, and the absence of experimental validation of the proposed therapeutic targets. The accuracy of the 30-protein classifier, while promising, could be further improved. The diverse range of MSBLs included might have influenced the results and warrants further investigation of CLL only in future studies.