BindCraft: one-shot design of functional protein binders

Key results

1. ShownAcross 12 targets spanning cell-surface receptors, allergens, a de novo beta barrel, a centriole protein and the nucleic-acid-binding grooves of two nucleases, testing 6-53 designs per target produced binders, mostly at nanomolar affinity and without library screening or affinity maturation.

   Binders found per designs purified: PD-1 13 of 53 (best apparent Kd below 1 nM, exact value not determined), PD-L1 7 of 9 (615 nM), IFNAR2 3 of 9 (260 nM), CD45 4 of 16 (14.7 nM), BBF-14 6 of 11 (20.9 nM), Der f7 4 of 10 (12.8 nM), Der f21 4 of 7 (793 nM), Bet v1 2 of 7 (120 nM); CrSAS-6 binder4 bound at 5.7 micromolar (monomer) and 4.2 micromolar (dimer). Stated range 10-100%, mean 46.3%. · Targeting cell surface receptors; Binder design beyond known and natural binding…; Blocking immunogenic epitopes of common…; Discussion

2. ShownCrystal structures of three designed complexes reproduced the predicted fold and binding mode, including a binder against a de novo beta barrel with no sequence homolog in the Protein Data Bank; cryoEM of the Cas9 complexes supported only docking-level validation.

   Backbone Calpha RMSD to the design model: 1.7 A for BBF-14/binder4 and 1.7 A for Der f7/binder2 (crystal forms at 2.2 A and 3.0 A resolution); 3.1 A for Der f21/binder10 (2.6 A resolution), attributed to a different rotamer of an interface tyrosine. Deposited as 9HAC, 9HAD, 9HAE, 9HAF. For apo SpCas9 the maps showed binder density at REC1 but no atomic model could be built. · Binder design beyond known and natural binding…; Blocking immunogenic epitopes of common…; Modulating the function of large multi-domain…; Data availability

3. ShownBinders aimed at nucleic-acid-binding sites worked as artificial inhibitors: all six anti-REC1 designs bound apo SpCas9, the designed binders lowered Cas9 editing in cells, and two of twelve anti-CbAgo designs slowed guide-directed DNA cleavage.

   SpCas9 binder3 and binder10 showed apparent affinity around 300 nM by SPR; co-transfected into HEK293T cells with Cas9 and an NSD2 sgRNA they cut editing more than the natural AcrIIC2, while AcrIIA2 and AcrIIA4 nearly abolished it. CbAgo kcat dropped from 0.004 /s to 5e-5 /s (80-fold) with binder2 and 9.8e-5 /s (40-fold) with binder3; binder2 bound at Kd 5 nM (BLI) and guide DNA displaced it. · Modulating the function of large multi-domain…; SpCas9 gene editing

4. ShownSplicing designed miniproteins into a capsid loop redirected AAV6 transduction toward cells carrying the targeted receptor, with candidates picked straight from the design pipeline in a cell assay rather than by biochemical prescreening.

   HER2- and PD-L1-targeting binders were inserted between residues 497 and 498 of VR-V in an AAV6-cmv-GFP capsid mutated to remove heparin and sialic acid binding. A supernatant-transfer screen gave one HER2-directed and four PD-L1-directed variants with enhanced specificity for HEK293 cells stably overexpressing the receptor; an antibody against the same receptor site blocked transduction by the PD-L1 variant. · AAV re-targeting for cell-type specific gene…

5. SuggestedOne designed binder against the birch allergen Bet v1 partially blocked patient IgE from binding the allergen, though less completely than a three-antibody cocktail.

   Blocking ELISA with sera from three birch-allergic patients: the REGN5713/5714/5715 cocktail blocked up to 90% of Bet v1 binding to IgE at low concentrations, while the single binder2 blocked up to 50% in two of the three donors. Binder2 bound Bet v1 at 120 nM (SPR), gave the expected 27.8 kDa complex by SEC-MALS, and competed with REGN5713 for its epitope. · Blocking immunogenic epitopes of common…; Birch allergen blocking assay