Protein therapeutics have transformed treatment of diseases from cancer to immune disorders, yet many targets appear not only at diseased sites but also in healthy tissues, frequently causing dose-limiting toxicities. Conditionally active biologics that remain inert systemically and activate only at disease sites offer a compelling strategy to improve therapeutic indices. Most current approaches rely on masking moieties that transiently occlude ligand-binding interfaces until a cleavable linker responds to disease-specific stimuli such as protease activity or an external trigger like light. However, identifying suitable mask sequences has largely relied on labor-intensive display-based library screens that must be repeated for each new target. A computational approach capable of generating effective masks de novo would dramatically accelerate development of conditionally active therapeutics.
Abstract Figure
Researchers in the ChemSynBio Group, led by Benjamí Oller-Salvia at IQS, Universitat Ramon Llull, Barcelona, Spain, published in JACS, developed a workflow for de novo design of peptide masks that reversibly inactivate miniprotein binders. The team selected four previously reported miniproteins targeting therapeutically relevant receptors: EGFR domains I and III, FGFR2, and IL7Rα. These receptors show high expression on tumor cells but also appear in healthy tissues where off-target engagement causes adverse effects. Each miniprotein consists of a three-helix bundle with two helices mediating receptor interaction while the third provides structural scaffolding. The researchers hypothesized that a fourth helix linked to one terminus via a protease-cleavable sequence could sterically block the receptor-binding interface between the two binding helices.
The computational pipeline employed RoseTTAFold diffusion or RFjoint inpainting to generate peptide backbones complementary to the miniprotein surface, guided by hotspot residues at the central groove of the receptor-binding interface. ProteinMPNN then designed amino acid sequences for the 15–25 residue masks, and AlphaFold2 predicted structures for validation. In silico filtering prioritized candidates based on structure confidence scores exceeding 90, interchain predicted aligned error below 8, and coverage of 60–80% of receptor-interacting residues. The team selected five diverse candidates per miniprotein and incorporated matrix metalloproteinase cleavage sites between each mask and binder for tumor-selective activation. All 20 constructs expressed well in E. coli with yields of 10–200 mg per liter and showed no aggregation or crossmasking by size-exclusion chromatography. Binding assays on receptor-expressing cells revealed that all designs provided some degree of inhibition. Nine of twenty masks achieved greater than 100-fold affinity reduction, comparable to state-of-the-art masks identified through extensive yeast and bacterial display screening. The most effective design reduced EGFR binding by over three orders of magnitude. Treatment with MMP2 restored binding in 19 of 20 cases, with affinities reaching 50–100% of unmasked controls.
Detailed characterization of the lead EGFR antagonist mask revealed that micromolar affinity between the isolated mask peptide and miniprotein suffices for robust inactivation. Biolayer interferometry measured a dissociation constant of 5 μM with rapid off-rate kinetics that facilitate efficient mask displacement after cleavage without requiring buffer washes. The mask forms a compact four-helix bundle with the miniprotein through hydrophobic contacts and shape complementarity, directly contacting 13 of 21 receptor-interacting residues. Functional assays confirmed that the masked miniprotein failed to inhibit EGF-induced MAPK signaling while MMP2 treatment fully restored antagonist activity. To demonstrate stimulus versatility, the researchers installed a photocleavable linker through site-specific cysteine conjugation, creating a light-responsive variant that achieved comparable masking efficiency. UV irradiation at 365 nm cleaved greater than 90% of the mask within 45 minutes and fully restored cellular binding. This platform establishes a generalizable approach for designing conditionally active protein therapeutics responsive to both endogenous proteases and external triggers, bypassing the labor-intensive library screens traditionally required for mask discovery.
