Designing Molecular Switches

Reflecting work in the Baker Lab

Published here October 6, 2026

De novo design of a macrocycle-induced dimerization system for cellular control

Stephanie Hanna, Patrick J. Salveson, Basile Wicky, Madison A. Kennedy, Derrick R. Hicks, Carolina Moller, Suna Cheng, Xinting Li, Mohamad Abedi, Brian Coventry, Meerit Y. Said, Asim K. Bera, Alex Kang, Barry L. Stoddard & David Baker

Nature Communications 2026, 17, 6683. https://doi.org/10.1038/s41467-026-71345-8

View Original Publication

Chemically induced dimerization, CID, systems are indispensable tools for switching cellular processes on and off with a small molecule. The workhorse of the field, the rapamycin–FKBP–FRB system, works well but brings an unwanted complication: rapamycin is an immunosuppressant that also engages endogenous proteins, producing off-target effects in vivo. Workarounds have relied on derivatizing rapamycin, transplanting plant-derived CID systems into mammalian cells, or using directed evolution and phage display to engineer binding partners for existing ligands. Each strategy inherits constraints from the natural starting material, and the universe of accessible ligand–protein pairs stays narrow. What the field needs is a route to CID systems designed from the ground up, where both the chemical inducer and its cognate protein are engineered together with no natural template to constrain them.

Researchers in the Baker Lab at the University of Washington, published in Nature Communications, reasoned that structured, membrane-permeable macrocycles with large exposed surface areas could serve as ideal de novo CID ligands, provided a matching protein could be designed to recognize them. The team generated libraries of backbone-cyclized peptides by identifying pairs of hydrogen-bonding loops whose terminal geometry matched, fusing them into closed rings in which all backbone amides were either engaged in transannular hydrogen bonds or N-methylated, satisfying every polar group and favoring membrane permeability. RosettaFastDesign then assigned sequences permitting both L and D amino acids, and crystal structures of four designs confirmed close agreement with computational models. One design, MC1, carries internal C2 symmetry, a property the team exploited to simplify the protein-design problem: a homodimeric binding protein was sufficient. Placing MC1 into the central cavities of de novo-designed helical repeat homodimers via RIFdock, followed by interface redesign and AlphaFold 2 filtering, yielded CID7, a homodimer that binds MC1 with a KD of 36 nM and whose co-crystal structure matches the design model with a Cα RMSD of 0.97 Å.

In mammalian cells, MC1 drives measurable reconstitution of a split reporter and reverses cleanly upon ligand washout, validating the system as a tunable switch. Because both the ligand and the protein were designed computationally with no natural scaffold, the framework is in principle extensible to a large family of orthogonal CID pairs, pointing toward simultaneous, independent control of multiple cellular pathways for synthetic biology and adoptive cell therapy. The full design methodology and experimental characterization await in the original publication.

Article Archive
Designing Molecular Switches

Figure 1. a | Five, six, or seven residue loops were constructed to mimic beta hairpins. Construction of macrocyclic peptide backbones is shown using a representative, five-residue starting loop. 6D transforms were calculated and binned. Pairs of loops were fused together if their termini geometry match, as indicated by green bin numbers. b | RosettaFastDesign was used to generate sequences for macrocyclic peptides, allowing both L and D hydrophobic residues. Design models of the MCs are shown in white with dashed black lines indicating hydrogen bonds. c | X-ray crystal structures, color, overlaid on the corresponding design models, white, for four designs. The all-atom RMSDs between the design models and crystal structures are depicted beneath the MCs. d | A rotamer interaction field was generated from the MC1 crystal structure, green; leucine side chains in the RIF process are shown for illustration, gray. RifDock was then used to place the peptide into the central cavity of homodimeric scaffold proteins. Protein peptide interactions were optimized with Rosetta fastDesign, retaining the protein-protein interfaces, and the designs were selected for experimental characterization based on AF2 and Rosetta metrics.