Protease activity in plasma extracellular vesicles enables diffuse large B-cell lymphoma detection, relapse- and therapy monitoring

Technical Summary: Protease Activity in Plasma Extracellular Vesicles for DLBCL Detection and Monitoring

Problem Statement
Diffuse large B-cell lymphoma (DLBCL) remains a significant clinical challenge, with a substantial proportion of patients experiencing refractory disease or relapse despite advances in therapy, including CAR-T cell transfer. Current risk stratification tools, such as the International Prognostic Index (IPI), have shown diminishing reliability in the rituximab era, and regular imaging lacks the sensitivity to detect molecular recurrence early. While circulating tumor DNA (ctDNA) offers a promising liquid biopsy approach, its clinical utility is often constrained by prolonged turnaround times, methodological complexity, and limited sensitivity for early relapse detection. Furthermore, existing extracellular vesicle (EV) biomarker strategies, primarily relying on EV-derived RNA, lack standardized isolation methods and robust readout assays. There is a critical need for a simple, sensitive, and cost-effective biomarker system capable of detecting DLBCL, stratifying risk, and monitoring minimal residual disease (MRD) and therapy response.

Methodology
The study employed a multi-modal approach centered on the analysis of plasma extracellular vesicles (pEVs) isolated from patient blood samples.

• Sample Cohort: Peripheral blood was collected from 61 DLBCL patients, 12 Hodgkin's lymphoma patients, 30 indolent B-cell lymphoma patients, and 50 age-matched healthy controls (HC). Samples were collected at diagnosis and longitudinally during/after rituximab-containing immunochemotherapy.
• Isolation: pEVs were purified using dual-mode chromatography (DMC), combining size-exclusion chromatography (SEC) to remove free proteins and cation exchange (CEX) chromatography to deplete lipoproteins. Isolation was validated via electron microscopy, Western blot (CD63), and interferometric nanoparticle tracking analysis (iNTA) for size and concentration.
• Protease Activity Assay (PAA): A functional assay was developed to measure pEV-associated protease activity. Thirty-three Förster resonance energy transfer (FRET) peptides were screened; 10 peptides exhibiting consistent linear activity patterns were selected. The cumulative protease activity (pEV-PA) was quantified by measuring the slope of fluorescence increase over time.
• Proteomic Profiling: To validate functional findings, pEV content from 61 DLBCL patients and 22 HC was analyzed using Olink® proteomics, targeting 3,000 proteins.
• Data Analysis: Machine learning models (including neural networks, stochastic gradient boosting, and k-nearest neighbors) were trained on training/test splits (80/20) to classify samples based on protease features. Differential expression analysis (limma) and dimensionality reduction (t-SNE, PCA) were applied to proteomic data.

Key Contributions and Results

• Differentiation of DLBCL from Controls: The pEV-PA assay successfully distinguished DLBCL patients from healthy controls. The cumulative activity of 10 proteases yielded an Area Under the Curve (AUC) of 0.91, with 87% sensitivity and 100% specificity. Machine learning classifiers achieved AUCs ranging from 0.91 to 0.97 in test datasets. Proteomic profiling corroborated these findings, achieving perfect group discrimination (AUC = 1.0) with clear linear separation in t-SNE and PCA plots.
• Identification of Specific Biomarkers: Among the proteases analyzed, ADAM10, ADAM17, Aggrecanase, MMP2, MMP3, MMP8, MMP9, and MMP10 were identified as key features. Proteomic analysis highlighted upregulated factors such as Secreted Phosphoprotein 1 (SPP1) and Endothelial Cell Specific Molecule 1 (ESM1), which are known to modulate the tumor microenvironment.
• Prediction of Treatment Response: Baseline pEV-PA levels correlated with treatment outcomes. Non-responders (progression/relapse within 52 months) exhibited significantly higher cumulative pEV-PA at diagnosis compared to responders. Specifically, elevated activities of ADAM17, Aggrecanase, MMP8, MMP10, and MMP13 were associated with non-response. Machine learning models applied to proteomic data achieved AU-ROC scores of 0.67–0.84 in test sets for distinguishing responders from non-responders.
• Longitudinal Monitoring: The study demonstrated that pEV-PA dynamics reflect disease activity. Responders showed a marked reduction in cumulative PA following therapy, whereas non-responders exhibited increases or stable high levels. High baseline activities of ADAM17, CD10, MMP10, and Aggrecanase were associated with significantly shorter progression-free survival (PFS), outperforming the prognostic accuracy of an IPI score >3.
• Specificity of pEV Enrichment: The study noted that PA detection in plasma without prior pEV isolation yielded mostly negative results, confirming that pEV enrichment is necessary for discriminating protease signatures.

Significance and Claims
The authors claim that pEV-associated protease activity serves as a sensitive, indirect measure of tumor burden and dynamic disease activity in DLBCL. The study posits that pEVs are not solely shed by malignant cells but also by host cells (e.g., hepatocytes, PBMCs) responding to the cancer cell secretome (CCS), which includes immunostimulatory components like transposons and endogenous retroviruses. This mechanism explains the correlation between elevated PA and inferior outcomes.

The paper asserts that the PAA offers a distinct advantage over existing methods due to its simplicity, sensitivity, and cost-effectiveness. By capturing activity-based biomarkers rather than mere protein abundance, the assay amplifies signals and may provide higher specificity than freely circulating proteases. The authors conclude that pEV-PA is a promising tool for risk stratification, early relapse prediction, and monitoring minimal residual disease, potentially enabling more precise tailoring of first-line therapies and earlier initiation of advanced therapies like CAR-T cells. However, the authors remain modest regarding clinical implementation, acknowledging limitations such as the relatively small cohort size, the absence of controls with relevant comorbidities, and the need for larger prospective studies to establish definitive diagnostic cut-off values.