Introduction
In the field of precision medicine, identifying the presence of a protein is often insufficient for understanding disease pathology. Proteins undergo post-translational modifications (PTMs), such as phosphorylation, which act as chemical switches to alter function, stability, and interaction partners. These specific variants are known as proteoforms. While vast amounts of PTM data exist in scientific literature, standard databases frequently lack the granularity required to distinguish between unmodified proteins and specific proteoforms. This study utilizes the Protein Ontology (PRO) to provide a formal framework for capturing these nuances, enabling more sophisticated analysis of cancer mutations and the validity of animal models.
Methods and Framework
The researchers developed a pipeline integrating automated text mining with expert curation. Tools like eFIP and RLIMS-P were used to extract phospho-dependent protein-protein interactions (PPIs) from millions of PubMed abstracts and full-text articles. These findings were then mapped to PRO terms, which define entities across multiple levels of granularity—from protein families to specific modified isoforms.
To visualize these relationships, the team constructed a multi-relation network using Cytoscape. This network connects PTM enzymes, modified proteoforms, protein complexes, and associated diseases, providing a structured view of molecular mechanisms.
Key Insights
The Importance of Proteoform Specificity
The study demonstrates that different phosphorylation sites on a single protein can lead to divergent biological outcomes. For instance, phosphorylation of YAP1 at Tyr-357 is associated with cancer progression, whereas phosphorylation at Ser-127 is linked to Alzheimer’s disease. By treating these as distinct PRO entities, researchers can better model disease-specific pathways.
Limitations of Non-Primate Models
A critical finding involved the DNMT1 protein, which is implicated in glioma. Although the protein sequence is highly conserved across species, sequence alignment revealed that the regulatory phosphorylation site (Ser-127) is present in primates but absent in common model organisms like mice, rats, and dogs. This discrepancy suggests that mouse models may not accurately replicate the regulatory mechanisms of certain human diseases.
Functional Clustering in Cancer
Analysis of beta-catenin (CTNNB1) mutation patterns in the COSMIC database revealed that cancers do not mutate randomly. Instead, they form two distinct functional clusters based on their N-terminal mutation signatures. Cluster 1 (e.g., skin and pancreatic cancers) typically involves mutations that stabilize the protein by preventing degradation, while Cluster 2 (e.g., large intestine and prostate cancers) may alter cell adhesion or nuclear signaling. Hierarchical clustering of mutation data revealed that Cluster 1 showed high mutation rates at sites Ser-33 and Ser-37, whereas Cluster 2 was dominated by mutations at Ser-45.
Conclusion
The integration of PRO into disease analysis bridges the gap between raw literature and actionable biological insights. While manual curation remains a bottleneck, the framework allows for more precise modeling of PTM-driven diseases. Future work will focus on further integrating PRO with semantic web resources to enable automated computational reasoning across complex interaction networks.
Summary updated