Protein allostery (the phenomenon in which a signal received by one region of a protein triggers a response in a distant functional region) constitutes the fundamental mechanism of real-time information processing and energy transfer at the cellular level. The vast majority of protein networks in living organisms function through protein switches based on this principle, regulating cellular responses by detecting signals ranging from small molecules to proteins. Therefore, the design of artificial allosteric protein systems with tunable input and output parameters is among the most fundamental goals of protein engineering and synthetic biology. Artificial fluorescent protein switches and enzyme-based biosensors have revolutionized cell and neurobiology research by enabling real-time monitoring of intracellular concentrations of calcium, neurotransmitters, and other small molecules. However, current approaches have largely relied on the presence of natural receptor sites that exhibit global conformational changes (shape changes affecting the entire protein structure) upon ligand (small molecule binding to the protein structure) binding. The limited number of regions possessing such properties constitutes one of the primary obstacles in the development of artificial biosensors; it also restricts sensor design for new classes of ligands.
Guo, Smutok, Lee and colleagues aimed to test whether minimal ligand-binding domains designed de novo (from scratch) using machine learning algorithms, which do not exhibit global conformational changes, could function as efficient receptors in single-component allosteric protein switches. It was established that the structural constraints imposed by naturally evolved receptor domains created a significant bottleneck in next-generation biosensor design. Accordingly, the study was designed to test the hypothesis that allosteric regulation could be possible through conformational entropy (structural dynamic variability) and to develop synthetic receptor domains that can be integrated with both natural and fully synthetic reporter enzymes. Furthermore, the study aimed to investigate the extensibility of this platform to different ligand classes, such as small molecules, peptides, and proteins, and to validate the functionality of the developed systems in applied contexts.
In the study, small molecule and peptide binding domains designed de novo via machine learning methods were inserted into the TEM-1 β-lactamase enzyme using circular permutation (the rearrangement of the N- and C-termini of a protein), thereby creating chimeric (combined) protein switches. Structures responding to steroid hormones, peptides, and proteins were identified by screening focused libraries for various ligand classes. The functional mechanism of the developed biosensors was examined in detail using complementary biophysical methods such as circular dichroism spectroscopy, ¹⁹F-nuclear magnetic resonance (¹⁹F-NMR), and hydrogen/deuterium exchange mass spectrometry (HDX-MS). The autonomy and portability of the receptor domains were tested by creating combinations with alternative reporter enzymes such as NanoLuciferase and PQQ-glucose dehydrogenase. Additionally, fully synthetic switches were produced using the AI-designed LuxSit Pro luciferase protein; the functionality of the developed systems was validated in both bacterial culture experiments and on a graphene nanosheet-based electrochemical bioelectrode platform.
The findings demonstrated that ligand-binding domains designed via machine learning function as effective receptors in single-component allosteric protein switches, despite not exhibiting global conformational changes. Colorimetric, bioluminescent, and electrochemical biosensors were produced for steroid hormones such as cortisol and 17α-hydroxyprogesterone, and it was shown that these systems could be compiled as intramolecular YES and AND logic gates. Biophysical analyses suggest that ligand binding does not lead to protein refolding but instead increases the catalytic activity of the reporter domain by reducing the conformational entropy of the system. It was observed that circular permutation reduced ligand binding affinity to a certain extent; however, it was shown that this effect could be partially compensated for by the addition of auxiliary binding domains. The electrochemical bioelectrode developed within the scope of the study enabled the measurement of steroid hormones at nanomolar levels and was determined to be reusable. These findings indicate that binding domains produced via machine learning can form a modular and extensible platform for synthetic biology applications, providing a conceptual framework for further research in diagnostic systems, analytical biotechnology, and bioorthogonal signal processing in engineered organisms.
Translated by: Hazel BİRGÜL
Editor: Elinsu AK
Reference: Guo, Z., Smutok, O., Lee, G. R., Cui, Z., Qianzhu, H., Kish, M., Ergun yva, C., Wu, K., Mutschler, R., Jackson, C. J., Fiorito, M. M., Warden, A. C., Smith, O. B., Quijano-Rubio, A., Huber, T., Phillips, J. J., Otting, G., Katz, E., Baker, D., & Alexandrov, K. (2026). Artificial allosteric protein switches with machine-learning-designed receptors. Nature Biotechnology. https://doi.org/10.1038/s41587-026-03081-9
-Bioinfocodes Scientific News Service-
News articles prepared by our team members, reviewing and compiling scientific research
published in journals with and impact factor greater than 20 (click here for the list)