In silico evolution of autoinhibitory domains for a PD-L1 antagonist using deep learning models.

There has been considerable progress in the development of computational methods for designing protein-protein interactions, but engineering high-affinity binders without extensive screening and maturation remains challenging. Here, we test a protein design pipeline that uses iterative rounds of deep learning (DL)-based structure prediction (AlphaFold2) and sequence optimization (ProteinMPNN) to design autoinhibitory domains (AiDs) for a PD-L1 antagonist. With the goal of creating an anticancer agent that is inactive until reaching the tumor environment, we sought to create autoinhibited (or masked) forms of the PD-L1 antagonist that can be unmasked by tumor-enriched proteases. Twenty-three de novo designed AiDs, varying in length and topology, were fused to the antagonist with a protease-sensitive linker, and binding to PD-L1 was measured with and without protease treatment. Nine of the fusion proteins demonstrated conditional binding to PD-L1, and the top-performing AiDs were selected for further characterization as single-domain proteins. Without any experimental affinity maturation, four of the AiDs bind to the PD-L1 antagonist with equilibrium dissociation constants (KDs) below 150 nM, with the lowest KD equal to 0.9 nM. Our study demonstrates that DL-based protein modeling can be used to rapidly generate high-affinity protein binders.

Design of a protease-activated PD-L1 inhibitor.

Immune checkpoint inhibitors that bind to the cell surface receptor PD-L1 are effective anti-cancer agents but suffer from immune-related adverse events as PD-L1 is expressed on both healthy and cancer cells. To mitigate toxicity, researchers are testing prodrugs that have low affinity for checkpoint targets until activated with proteases enriched in the tumor microenvironment. Here, we engineer a prodrug form of a PD-L1 inhibitor. The inhibitor is a soluble PD-1 mimetic that was previously engineered to have high affinity for PD-L1. In the basal state, the binding surface of the PD-1 mimetic is masked by fusing it to a soluble variant of its natural ligand, PD-L1. Proteolytic cleavage of the linker that connects the mask to the inhibitor activates the molecule. To optimize the mask so that it effectively blocks binding to PD-L1 but releases upon cleavage, we tested a set of mutants with varied affinity for the inhibitor. The top-performing mask reduces the affinity of the prodrug for PD-L1 120-fold, and binding is nearly fully recovered upon cleavage. In a cell-based assay measuring inhibition of the PD-1:PD-L1 interaction on the surface of cells, the IC50s of the masked inhibitors were up to 40-fold higher than their protease-treated counterparts. The changes in activity we observe upon protease treatment are comparable to systems currently tested in the clinic and provide evidence that natural binding partners are an excellent starting point for creating a prodrug.

In silico evolution of protein binders with deep learning models for structure prediction and sequence design.

There has been considerable progress in the development of computational methods for designing protein-protein interactions, but engineering high-affinity binders without extensive screening and maturation remains challenging. Here, we test a protein design pipeline that uses iterative rounds of deep learning (DL)-based structure prediction (AlphaFold2) and sequence optimization (ProteinMPNN) to design autoinhibitory domains (AiDs) for a PD-L1 antagonist. Inspired by recent advances in therapeutic design, we sought to create autoinhibited (or masked) forms of the antagonist that can be conditionally activated by proteases. Twenty-three de novo designed AiDs, varying in length and topology, were fused to the antagonist with a protease sensitive linker, and binding to PD-L1 was tested with and without protease treatment. Nine of the fusion proteins demonstrated conditional binding to PD-L1 and the top performing AiDs were selected for further characterization as single domain proteins. Without any experimental affinity maturation, four of the AiDs bind to the PD-L1 antagonist with equilibrium dissociation constants (KDs) below 150 nM, with the lowest KD equal to 0.9 nM. Our study demonstrates that DL-based protein modeling can be used to rapidly generate high affinity protein binders.

Engineering Improved Photoswitches for the Control of Nucleocytoplasmic Distribution.

Optogenetic techniques use light-responsive proteins to study dynamic processes in living cells and organisms. These techniques typically rely on repurposed naturally occurring light-sensitive proteins to control subcellular localization and activity. We previously engineered two optogenetic systems, the light activated nuclear shuttle (LANS) and the light-inducible nuclear exporter (LINX), by embedding nuclear import or export sequence motifs into the C-terminal helix of the light-responsive LOV2 domain of Avena sativa phototropin 1, thus enabling light-dependent trafficking of a target protein into and out of the nucleus. While LANS and LINX are effective tools, we posited that mutations within the LOV2 hinge-loop, which connects the core PAS domain and the C-terminal helix, would further improve the functionality of these switches. Here, we identify hinge-loop mutations that favorably shift the dynamic range (the ratio of the on- to off-target subcellular accumulation) of the LANS and LINX photoswitches. We demonstrate the utility of these new optogenetic tools to control gene transcription and epigenetic modifications, thereby expanding the optogenetic "tool kit" for the research community.