TL;DR
This review explores the cutting-edge intersection of chemical biology and deep learning to target protein post-translational modifications (PTMs) like phosphorylation and SUMOylation. It highlights how new computational tools and innovative small molecules (including molecular glues and PROTACs) are transforming these dynamic "biological switches" into actionable targets for cancer therapy.
What Problem Does This Paper Address?
Proteins don't just exist in a static state; they are constantly modified by "tags" called PTMs (e.g., adding a phosphate or a methyl group) that act as on/off switches for cellular activity. While we have identified over 500,000 phosphorylation sites in the human proteome, we only understand the function of fewer than 10,000 of them.
This "functional annotation gap" is a massive bottleneck. Without knowing which PTMs actually drive disease, we cannot effectively design drugs to stop them. Furthermore, traditional drugs often target the enzymes that add these tags (the "writers"), which can lead to off-target effects because one enzyme might modify hundreds of different proteins. The challenge is to move toward targeting specific modified sites or using more sophisticated chemical tools to modulate these pathways.
Methods
The paper summarizes a multi-pronged approach to PTM research:
- Deep Learning for Functional Prediction: The authors highlight the development of models like FuncPhos-SEQ (sequence-based) and FuncPhos-STR (structure-based, utilizing AlphaFold data) to predict which PTM sites are actually functional.
- Chemical Proteomics: Techniques like Activity-Based Protein Profiling (ABPP) and proximity labeling are used to "fish" for specific PTM-enzyme interactions in living cells.
- Targeted Protein Degradation (TPD): The review discusses the use of PROTACs (Proteolysis Targeting Chimeras) and Molecular Glues to selectively degrade proteins involved in PTM signaling, rather than just inhibiting their activity.
- High-Throughput Screening: Utilizing dual-luciferase reporter systems and high-content screening to find "lead" compounds that interfere with specific PTM axes (e.g., the CBX4-HIF-1α SUMOylation pathway).
Datasets
The computational models discussed (FuncPhos series) were trained and validated using several major biological databases:
- PSP (PhosphoSitePlus): A comprehensive resource for known PTM sites.
- EPSD, iPTMnet, and PTMD: Databases containing experimentally verified functional information for protein modifications.
- AlphaFold DB: Used to provide structural context for PTM sites in the FuncPhos-STR model.
Key Insights and Ideas
- PTMs as Dynamic Nodes: PTMs are not just markers; they are the "software" of the cell. The review emphasizes that PTM "crosstalk"—where one modification influences another—is a critical regulatory layer in cancer metastasis and immune evasion.
- Bridging the Gap with AI: Deep learning is successfully identifying functional PTM sites that were previously ignored. For example, the FuncPhos model correctly identified that phosphorylation at specific sites on the enzyme NADK is a key trigger for its activation.
- Moving Beyond Kinase Inhibitors: While kinase inhibitors are common, the field is shifting toward "undruggable" PTMs like SUMOylation and acetylation by using covalent probes and molecular glues that stabilize specific protein-protein interactions.
- Metabolic-PTM Link: The paper highlights a fascinating link where metabolic enzymes (like HADH) can influence PTMs (like SUMOylation of CBX4), suggesting that targeting metabolism could be a "backdoor" to controlling PTM-driven cancers.
Key Results
- STK19 Inhibition: Researchers identified STK19 as a critical "writer" for NRAS-mutant melanoma. A specific inhibitor, ZT-12-037-01, showed significant anti-tumor activity in animal models.
- Targeting Autophagy: The development of DC-LC3in-C42, a highly potent small molecule (IC50 = 7.6 nM), targets the acetylation site of the LC3 protein to inhibit autophagy, providing a potential new strategy for cancer treatment.
- Molecular Glues for CDK12: The review notes the discovery of molecular glues (e.g., LL-K12-4) that stabilize the complex between CDK12 and the DDB1 ubiquitin ligase, leading to the selective degradation of Cyclin K.
Limitations and Future Work
- Selectivity: Many PTM-modifying enzymes have broad substrate ranges. Designing a drug that stops an enzyme from modifying one specific protein without affecting its other vital functions remains a major hurdle.
- Drug Resistance: Similar to traditional kinase inhibitors, tumors often develop secondary mutations or bypass pathways to evade PTM-targeted drugs.
- Complexity of Crosstalk: The "PTM code" is incredibly complex. Future work needs to focus on how multiple different types of modifications (e.g., a protein being both phosphorylated and ubiquitinated) work together to drive disease.
- Clinical Translation: While many "苗头" (lead) compounds have been found, moving these from the lab to clinical trials requires overcoming significant pharmacokinetic and toxicity challenges.
Summary updated