Cross-Cohort Analytical Validation of Mass-Spectrometry Proteomic and Phosphoproteomic Signalling Signatures in Molecularly Stratified Cancer Cohorts

Background: The functional implications of cancer genomes can be measured using mass-spectrometry proteomics, but for clinical translation, the ability to measure the same signaling across independent disease cohorts needs to be demonstrated.  Methods: Harmonized CPTAC breast cancer data (n=122) were used for discovery, with ovarian cancer (n=83) and lung adenocarcinoma (n=110) for external validation. After cross-cohort filtering, 7,053 proteins and 3,822 phosphosites were analyzed. Three models (elastic-net regression) were used to predict the activity of MAPK/RTK, PI3K-AKT-mTOR, and cell-cycle activity, which were assessed by correlation, RMSE, R², bootstrap confidence intervals, and AUC.   Results: Pooled external correlations of MAPK/RTK, PI3K-AKT-mTOR, and cell cycle were 0.603, 0.561, and 0.782, respectively. External AUCs ranged from 0.810 to 0.956. However, integration was not always superior to the best single modality. Bootstrap-adjusted Rand index for the two stable states identified by the clustering was 0.974: signaling-low (n=78) and co-activated (n=114). LAMC2, GRB10-S104, RRM2, and CEP152-S1461 were some of the key contributors.  Conclusions: Computational validity and development of targeted assays were supported by the validation of locked mass-spectrometry signatures, which signalled states across cancers, particularly those involving cell-cycle activity, with prospective clinical applications pending.