Overcoming resistance to BRAF V600E inhibition in melanoma by deciphering and targeting personalized protein network alterations

Melanoma incidence is rising, with about half of cases harboring activating BRAF mutations (commonly V600E). Targeted therapies (e.g., BRAF and MEK inhibitors) have transformed care but most BRAF V600E patients relapse due to resistance. Genomic or single-biomarker–guided regimens may miss tumor-specific network alterations that evolve over disease course. The study proposes designing patient-specific therapies based on individualized altered signaling signatures derived from proteomic data using an information-theoretic, thermodynamic-like framework. This framework identifies tumor-specific groups of co-expressed proteins (‘unbalanced processes’) that constitute a patient-specific altered signaling signature (PaSS). The hypothesis is that concurrent targeting of all active unbalanced processes per tumor will more effectively collapse altered signaling flux, improve efficacy over standard BRAF-targeted combinations, and delay resistance.

Study design combined computational analysis of proteomic datasets with experimental validation in melanoma cell lines and mouse xenografts. Datasets: Proteomic data (RPPA) from TCGA PanCan32 were used: 353 skin cutaneous melanoma (SKCM) and 372 thyroid carcinoma (THCA) tumors profiled for 258 cancer-associated proteins (261 retained after filtering). A separate RPPA dataset included ~200 cell lines from 18 cancer types (including melanoma), each profiled for ~224 total and phospho-proteins. Additional analysis encompassed a broader set of 291 cell lines where 17 recurrent unbalanced processes were identified. Information-theoretic analysis: An information-theoretic, thermodynamic-inspired approach (referred to in the paper as SA/“Surprisal analysis”) decomposes protein expression deviations from a baseline steady state into contributions from multiple constraints (unbalanced processes). For each tumor k and protein i, the expression Xi(k) is represented as the balanced component minus the sum of deviations contributed by each process, parameterized by process amplitudes λj(k) and protein participation weights g (correlation structure within processes). Proteins with significant participation (g) cluster into unbalanced processes. Process amplitudes λ quantify the activity and direction (sign) of each process per tumor. Determining significant processes: Significance was assessed by how additional processes improved reconstruction/correlation of experimental data. Noise thresholds for λ were estimated from the most stable proteins’ standard deviations. Beyond n=17 processes, amplitudes did not exceed noise, thus 17 processes sufficed to describe the 725 tumors. Barcoding: For each patient, λ values for the 17 processes were thresholded and normalized to generate a barcode indicating active processes and direction (positive/negative amplitude). This barcode represents the patient-specific PaSS and guides drug target selection. Target prediction and combination design: For each active process in a tumor, central upregulated proteins were suggested as drug targets (from a curated map of BRAF-related signaling and process-specific protein maps). Personalized combinations were designed by selecting drugs to concurrently inhibit central nodes across all active processes; where appropriate, multi-kinase inhibitors were preferred if they matched process nodes and biomarker context. Experimental validation: - Cell lines: BRAFV600E melanoma lines A375 and G361 (and BRAFV600E A2058) were analyzed to predict cell line–specific PaSS and drug combinations. Treatments included clinically used agents (dabrafenib, trametinib), dasatinib, 2-Deoxy-D-glucose (2-DG), and controls (e.g., erlotinib/“eritinib”) at specified doses (typically up to 1 μM; 2-DG up to 2 mM; monotherapy and combinations). Viability assays (MTS/MTT and methylene blue) quantified survival; western blots measured pathway readouts (pS6, pERK, pAKT, pPKM2, PDGFRβ, pS6K, p53). Resistance development was assessed by twice-weekly treatments over 28 days. - In vivo: NSG mice bearing subcutaneous xenografts of A375, G361, or A2058 were treated up to 4 weeks (6×/week). Tumor growth was monitored; body weights were tracked (no significant weight loss reported). Dose example: trametinib 0.5–1 mg/kg; combinations matched computational predictions per line. Data sources and availability: TCGA/CPAT (tcpaportal.org) for patient and cell line RPPA; mutation data for SKCM (GDC).

- Across 725 tumors (353 SKCM, 372 THCA), 17 distinct unbalanced processes sufficiently reconstructed the proteomic data; each tumor typically exhibited 1–3 active processes. - Processes 1 and 2 differentiated THCA and SKCM: process 1 appeared almost exclusively in THCA, whereas process 2 characterized SKCM (331 SKCM tumors harbored process 2 versus 4 THCA tumors, where it had negative amplitude indicating opposite-direction deviations). - PaSS did not strictly correlate with BRAF mutation status: SKCM tumors with BRAFwt and BRAFV600E could share identical signatures. Example: 181 SKCM patients had a signature consisting only of process 2 (107 BRAFwt, 74 BRAFV600E); no THCA patients had this signature. - The cohort mapped to 138 distinct barcodes (patient-specific PaSS), highlighting extensive heterogeneity beyond conventional SKCM/THCA and BRAF status grouping. - Cell line–specific predictions and validations: • A375 (BRAFV600E): Predicted to require concurrent targeting of MEK/mTOR-S6 and PDGFRβ-linked process nodes. Trametinib + dasatinib (PaSS-based) achieved up to ~95% killing and broadly suppressed signaling (reduced pS6, pERK, pS6K, pPKM2; lowered PDGFRβ; increased p53). Clinically used dabrafenib + trametinib achieved ~85% killing but left PDGFRβ signaling and induced pAKT; dasatinib monotherapy inhibited PDGFRβ but not other active processes. • G361 (BRAFV600E): Predicted combination trametinib + 2-DG targeted MEK/mTOR and metabolic nodes (GAPDH/PKM2). Monotherapies: trametinib ~65% killing at 1 μM; 2-DG ~75% at 1 mM; dasatinib ~20% at 1 μM; erlotinib control ~10%. Dabrafenib + trametinib reached ~90% killing but induced pAKT and retained metabolic activity (high MTT). Trametinib + 2-DG nearly eradicated cells at 1 μM/2 mM and shut down pS6, pAKT, pS6K, and pERK. - Resistance assays (28 days, twice-weekly treatments): Standard dabrafenib + trametinib initially killed (~88–96%) but allowed regrowth by day 21–28 in A375 and G361. PaSS-based combinations (A375: trametinib + dasatinib; G361: trametinib + 2-DG) maintained or increased killing over time with no regrowth (A375 near-complete killing by day 28; G361 no regrowth). - In vivo xenografts: PaSS-based combinations significantly reduced tumor growth compared to monotherapies or clinically used combinations. A375: trametinib + dasatinib outperformed trametinib + dabrafenib and other regimens. G361: trametinib + 2-DG was superior. A2058 (BRAFV600E) exhibited a single active process; low-dose trametinib (0.5 mg/kg) was optimal, outperforming higher dose (1 mg/kg), consistent with predictions. - Overall, PaSS-guided combinations were more efficacious and selective than standard-of-care BRAF/MEK therapy, and better mitigated resistance development.

The study demonstrates that melanoma tumors, even when initially driven by BRAF V600E, evolve distinct altered signaling network states that are not captured by mutation status alone. By decoding patient-specific altered signaling signatures (PaSS) composed of 1–3 unbalanced processes, therapies can be rationally designed to concurrently inhibit the full altered signaling flux. This approach outperformed standard BRAF/MEK regimens in vitro and in vivo, and delayed or prevented resistance in prolonged treatment assays. The near mutual exclusivity of altered signatures between SKCM and THCA and the presence of shared signatures across BRAFwt and BRAFV600E patients underscore the limited utility of single genomic biomarkers for therapy selection. The results suggest that mapping tumors into a multi-dimensional process space (barcodes) enables precision combination therapy, potentially translating into improved durability and reduced likelihood of adaptive rewiring. The findings also highlight the need to monitor dynamic changes in PaSS during treatment and to consider integrating complementary modalities (e.g., immunotherapy) to manage heterogeneity and microenvironment-driven adaptations.

This work introduces and validates a PaSS-based framework for designing personalized drug combinations that target patient-specific unbalanced signaling processes in melanoma. Using proteomic data and an information-theoretic analysis, the authors identified 17 recurrent processes and 138 unique patient barcodes across 725 tumors. PaSS-guided combinations were highly selective, outperformed standard BRAF/MEK therapy in cell lines and mouse models, and delayed resistance. The study advances precision oncology by shifting from mutation-centric to network-state–centric therapy design. Future directions include serial biopsies to capture evolving subclonal processes, integration with immunotherapy, expansion to additional tumor types, and clinical trials to evaluate PaSS-guided regimens.

- In vivo, PaSS-based combinations did not fully flatten tumor growth in all cases, suggesting residual or emergent subpopulations. - Bulk proteomics may miss small subclones; evolving unbalanced processes during therapy (e.g., microenvironment/stromal interactions) may require serial sampling and adaptive regimen updates. - The approach depends on availability and quality of proteomic profiling and on accurate process-to-target mapping; not all predicted targets may have approved inhibitors or optimal dosing regimens. - Some reported dataset numbers and annotations show minor inconsistencies in the text, and broader clinical validation is needed to generalize findings.