Explainable AI helps decode the substrate specificity of γ-secretase enzyme

Using artificial intelligence, researchers show how γ-secretase recognizes substrates - an important advance for fundamental and translational research.

The γ-secretase enzyme is capable of cleaving more than 150 different membrane proteins. This includes the amyloid precursor protein, from which the deposits typical for Alzheimer's disease are formed, and the Notch1 protein, which plays an important role in cell communication and carcinogenesis. For a long time, however, it was unclear how γ-secretase recognizes its target proteins. Although many proteases identify substrates by means of characteristic amino acid sequences, γ-secretase does not.

An interdisciplinary team from LMU's Biomedical Center, the Technical University of Munich (TUM), and the German Center for Neurodegenerative Diseases (DZNE) has now managed to clarify details of the mechanism. The researchers have shown that substrates of the enzyme possess a complex physicochemical profile which is decisive for their recognition and cleavage.

New technique makes hidden features visible

The team developed a novel technique called Comparative Physicochemical Profiling (CPP), which allows them to compare the physicochemical properties of known substrates against reference proteins and identify characteristic patterns. In combination with explainable artificial intelligence (XAI), the team was also able to render visible the features that are characteristic for substrates of γ-secretase.

"The substrates of γ-secretase possess a specific physicochemical profile, which spans the entire transmembrane domain and adjacent sequence regions," explains Prof. Harald Steiner (LMU and DZNE), who led the study together with Dmitrij Frishman (TUM). Particularly close to the cleavage site of the substrates, the researchers discovered, substrates have the potential to develop an extended conformation as an alternative to their helical structure - a property that is supported by experimental data from γ-secretase enzyme-substrate complexes.

"We wanted to understand what actually defines a substrate, and not just generate a black-box prediction," adds lead author Dr. Stephan Breimann, who played a major part in developing the CPP method. "The use of explainable AI has given us precisely this transparency."

Perspectives for research and application

Using the CPP method, the researchers also managed to identify several previously unknown substrates of the enzyme, including proteins that play an important role in immune regulation and carcinogenesis.

The authors of the study are convinced their findings extend far beyond γ-secretase.

We see a new approach in here for also decoding the interplay of sequence, structure, and function in other proteases - or, say, in receptors."

Prof. Harald Steiner, LMU and DZNE

In the long term, the results could also contribute to the development of therapeutically relevant compounds such as small-molecule drugs, peptides, or antibodies with enhanced specificities.