Atomic-level mapping of amorphous diabetes drug reveals hydrogen bonds as key to stability

Scientists at EPFL and AstraZeneca have developed a method to map the atomic-level structure of amorphous drugs, demonstrated on a GLP-1 receptor agonist candidate for diabetes and obesity treatment. Their work appears in the Journal of the American Chemical Society.

One of the big hurdles in drug development is solubility. Many promising drug molecules just don’t dissolve well enough in the body, making them tough to absorb—especially in pill form. To get around this, scientists often turn to “amorphous” forms of drugs. Unlike crystalline solids, where atoms line up in an orderly grid, amorphous materials are a molecular jumble. This disorder can boost solubility dramatically, but it comes at a cost: instability. Over time, amorphous drugs can reorganize into crystals, losing their effectiveness.

To stop this from happening, scientists need to understand what keeps amorphous drugs stable. But that’s tricky. These materials are so disordered that traditional techniques like X-ray crystallography don’t work. Until very recently, no one had ever mapped a full 3D atomic structure of a pharmaceutical amorphous solid.

Now, researchers led by Professor Lyndon Emsley at EPFL and Staffan Schantz at AstraZeneca have developed a powerful new method to map the atomic-level structure of amorphous drugs used in diabetes. The researchers used a method they developed in 2023 that combines NMR crystallography with machine learning and molecular dynamics simulations to produce the first experimentally validated 3D atomic-level ensemble structure of an amorphous pharmaceutical solid.

The team used the method to study one of AstraZeneca’s experimental GLP-1 receptor agonists, a type of drug used to treat type-2 diabetes and obesity. These drugs are usually injectable, but unlocking ways to make stable, effective oral versions could make a big difference for patients worldwide.

The team first measured how atomic nuclei in the drug respond to magnetic fields using solid-state nuclear magnetic resonance (NMR). These measurements gave them chemical shift distributions for 17 carbon and 16 hydrogen atoms in the molecule.

Then they simulated over nine million possible molecular environments using supercomputers and applied a machine learning tool called ShiftML2 to predict the corresponding chemical shifts. By matching predictions to experiments, they selected the most likely structures.

Hydrogen bonds are the key

By combining experimental data with advanced simulations, the team could pinpoint which molecular conformations and interactions actually existed within the amorphous structure. They found that the drug molecules tend to form hydrogen bonds either with each other or with surrounding water molecules. These bonds act like molecular “anchors,” reducing the tendency of the molecules to rearrange themselves into a crystalline, and less soluble, structure, especially when water molecules are involved.

The analysis even showed which ring structures in the molecule tended to adopt specific orientations to help prevent crystallization. For example, certain parts of the molecule, like the benzodioxole and benzimidazole rings, were more likely to twist at particular angles, specifically around −150° and −60° for the benzodioxole ring relative to the piperazine ring.

These “preferred” twists help lock the molecule into a stable, non-crystalline form, making it harder for the material to reorganize into a crystalline structure—a process that would significantly reduce the drug’s solubility and, by extension, its effectiveness in the body.

A breakthrough method

Crucially, the study also demonstrated a major methodological advancement. Instead of relying solely on computer predictions about which molecular structures should be stable—an approach that can sometimes miss important real-world behavior—the researchers validated their models against actual experimental data from NMR measurements.

This meant they could confirm, with high confidence, the exact atomic-level arrangements present in the amorphous drug. By filtering their computational models by comparison to real experimental observations, they produced an unprecedented and reliable 3D map of an amorphous pharmaceutical solid.

“This constitutes an unprecedented level of structural detail, and the insights into the stabilization mechanisms of the amorphous form of this GLP-1R agonist that it provides, represent a big step forward in understanding amorphous drug forms,” says Emsley.

For GLP-1R agonists, widely used in the treatment of type 2 diabetes and obesity, this could eventually lead to effective oral alternatives to injections, significantly improving patient comfort and treatment adherence.