Computational design of potent and selective BAK and BAX binders

There are numerous binders of the pro-survival BCL2 family proteins such as BCL2, MCL1, and BCL-XL, but development of potent and selective binders of their pro-apoptotic counterparts BAK and BAX has remained a major unsolved challenge. We use computational protein design to generate 13 kDa binders of BAK and BAX with 400 pM and 3 nM affinity, orders of magnitude higher than any existing native or designed binder, and with greater than 100-fold specificity against pro-survival BCL2 family members. The crystal structure of the BAKᐧɑBAK2 complex is very close to the computational design model, with the binder making specific interactions extending out from the canonical BH3-binding groove. Liposome- and cell-based analyses reveal that ɑBAK2 inhibits membrane permeabilization when in excess of BAK, but activates BAK when BAK is in excess. Structural analyses indicate that binding of ɑBAK2 results in partial unfolding and exposure of BAK’s BH3 domain. Similar to ɑBAK2, ɑBAX2 activates BAX at low concentrations and does not activate BAX at high concentrations. This work provides valuable insight into design of small molecule or protein inhibitors of BAK and BAX; inhibition requires high affinity binding as well as a saturating concentration of binder at the site of action. Our designs are the first binders with the high specificity required for efficient modulation of apoptosis via direct interaction with BAK and BAX and they provide highly selective molecular probes for addressing outstanding cell biological questions about cell death. In the related work, proteins were computationally designed to bind BAK or BAX. Site-directed saturation mutagenesis libraries were generated based on the sequences of sub-optimal minibinders, transformed into yeast for surface display, and selected for high-affinity binding to each target, optionally in the presence of unlabeled competitors (other BCL2 homologs). Next generation sequencing analysis of the naive and selected pools was used to determine the fitness of each mutation for affinity and specificity.