A Baylor College of Medicine-led team has developed a strategy that combines the analysis of thousands of proteins with artificial intelligence (AI) to accelerate the discovery of small molecules called 'molecular glues' to treat disease. Their approach, published in Nature Communications , has uncovered a new class of molecular glues that can potentially neutralize harmful proteins that have been linked to blood cancers and autoimmune diseases. The work also shows how AI-based structural modeling can guide chemists early for compound optimization, well before experiments reveal how a compound works.
"Many scientists are increasingly exploring a new way to treat disease: instead of blocking harmful proteins, they aim at eliminating them entirely. One promising approach uses molecular glues, which act like matchmakers inside cells," said senior and co-corresponding author Dr. Jin Wang , director of the Center for NextGen Therapeutics and Michael E. DeBakey, M.D. endowed Professor in Pharmacology and in the Verna and Marrs McLean Department of Biochemistry and Molecular Pharmacology at Baylor. Wang also is a member of Baylor's Dan L Duncan Comprehensive Cancer Center. "These compounds bring a target protein to the cell's natural protein-disposal machinery, which destroys the target. In this study, our team discovered and optimized a new class of molecular glues that selectively remove a protein called VAV1, an important regulator of immune cell function that has been linked to blood cancers and autoimmune diseases."
VAV1 is found mainly in immune cells, where it helps transmit signals that activate T cells and other components of the immune system. However, abnormal VAV1 activity can contribute to diseases such as T-cell lymphomas and chronic inflammatory disorders. While traditional drugs typically inhibit only one function of a protein, targeted protein disposal eliminates the entire protein, potentially providing a more complete therapeutic effect.
Searching for molecular glues for VAV1
"To find compounds capable of degrading VAV1, we screened a library of molecules using high-throughput proteomics, a technology that can assess thousands of proteins simultaneously. This unbiased analysis revealed a series of compounds, including NGT-201-12, that caused VAV1 levels to drop while affecting relatively few other proteins," said first and co-corresponding author Dr. Hanfeng Lin , postdoc in the Wang lab . "Follow-up experiments confirmed that the compounds worked through the cell's natural protein-recycling system and specifically relied on the protein cereblon (CRBN), a key component of the protein-degradation pathway."
GluePlex: a computational workflow that can predict molecular structures
A major challenge in molecular glue research is understanding how these molecules recruit their targets. To address this, the researchers combined artificial intelligence, protein-structure prediction tools and physics-based modeling to develop a computational workflow called GluePlex. This platform predicted how VAV1, CRBN and the molecular glue come together to form a three-part complex. "The model identified a specific region of VAV1, known as the SH3-2 domain, as being essential for degradation. Experimental tests confirmed the prediction and pinpointed the exact spot the glue uses: a small surface loop on VAV1 that acts as a degradation signal, or 'degron.' This loop is different from degradation signals commonly associated with cereblon-targeting molecular glues," Lin said. A separate research team has since reported a similar conclusion about VAV1 using different experiment-based methods.
What stands out in the Baylor study is that the computational workflow arrived at this hard-to-anticipate contact point on its own – the prediction was made without reference to any experimental structure of the complex and was subsequently confirmed in the lab.
The result adds to a growing understanding of how molecular glues recognize proteins. Until recently, most known cereblon-dependent molecular glues were thought to require a specific structural feature known as a G-loop. Together with the independent report, this study indicates that cereblon also can recognize a different structure, broadening the range of proteins that may be targeted for degradation in the future.
Improving compounds through medicinal chemistry
After identifying the molecular mechanism, the team improved the compounds through medicinal chemistry. They introduced chlorine atoms into the molecules, which reduced molecular flexibility and increased degradation efficiency. One of the resulting compounds, NGT-201-18, showed substantially improved potency and resulted in a stronger protein complex required for degradation.
Importantly, the researchers tested NGT-201-18 in primary human T cells. The compound successfully reduced VAV1 levels and suppressed T-cell activation, demonstrating that the degrader could alter immune-cell signaling in a biologically relevant setting. Although more research is needed before treatment using these compounds could be considered for patients, these results support the idea that VAV1 degradation could become a useful strategy for treating autoimmune and inflammatory diseases.
The study also highlights the importance of comprehensive proteomic profiling. While VAV1 was the primary target, the researchers discovered that some compounds also degraded another protein, LIMD1. This finding underscores the need to evaluate both intended and unintended protein targets during drug development.
"Overall, this work introduces a series of VAV1-targeting molecular glues and, just as importantly, shows how artificial intelligence, structural modeling and proteomics can work together at the earliest stage of a project, when no experimental structure is yet available, to speed up drug discovery," Wang said. "These advances may help pave the way for new treatments for immune-related diseases and cancers that currently lack effective targeted therapies."
Other contributors to this work include co-first authors Xin Yu, Haiyang Zheng, as well as Ran Cheng, Min Zhang, Xiaoli Qi, Yen-Yu Yang, Shengmin Zhou, Rui Qi, Ly Le, Andrea Bortolato, Semen Yesylevskyy , Alan Nafiiev and Xing Che. The authors are affiliated with Baylor College of Medicine, Thermo Fisher Scientific, YDS Pharmatech, Inc., SandboxAQ or Receptor.AI Inc.
The research was supported in part by NIH grant R01-AI198974, the Michael E. DeBakey, M.D., Professorship in Pharmacology, seed funding for Center for NextGen Therapeutics and by CPRIT fellowships (RP210027 and RP210043).
