Summary
A bioRxiv preprint describes a sequence-design model that encodes preferred protonation states to create pH-sensitive PD-L1 binders. Yeast-display screens recovered hundreds of candidate designs with transition pH values from 4.0 to 5.8.
A bioRxiv preprint introduces Proton-PottsMPNN, a sequence-design model that represents alternative protonation states directly when designing protein binders. The researchers used it to generate 8,407 de novo candidate binders to PD-L1, then screened the library on yeast across different pH values. The screens recovered hundreds of pH-dependent designs, with reported transition pH values between 4.0 and 5.8.
The approach targets a design problem: a protein’s binding behaviour can change as pH changes, because pH affects the charge of certain amino-acid side chains. The preprint reports designs that bind PD-L1 under acidic conditions—a different operating direction from earlier computational approaches described by the authors, which aimed for binding at physiological pH followed by release in acidic conditions.
Encoding protonation in the design
Histidine, aspartate and glutamate can change protonation state as acidity changes. Protonation alters their electrical charge and can affect interactions within a protein and at its binding surface. In the authors’ account, previous design models did not explicitly represent those states, so designers could introduce titratable residues but had to screen to find out whether the resulting binder responded to pH as intended.
Proton-PottsMPNN treats protonated and deprotonated forms of histidine, aspartate and glutamate as distinct sequence tokens. The model can therefore design the surrounding amino-acid environment to favour a specified state over its alternative. The preprint says it jointly designed sequence–structure compatibility and protonation-state preference under one energy function.
Yeast-display screening of the designs
The researchers synthesized the full library of 8,407 candidates and displayed them on yeast. They screened for PD-L1 binding across a series of pH values using fluorescence-activated cell sorting (FACS) and magnetic-activated cell sorting (MACS). FACS sorts cells using fluorescence measurements; MACS separates them magnetically. The screens recovered 237 and 288 unique pH-dependent designs, respectively.
Reported transition pH values ranged from 4.0 to 5.8. The switches involved protonated aspartate and glutamate as well as histidine. The authors also report that recovered designs showed a stronger model preference for their specified protonation state than designs made from the same structural backbones that were not recovered. That preference was strongest among designs that needed the most acidic conditions to bind.
The results demonstrate a computational route to specifying protonation-state preferences during binder design, followed by experimental selection of candidate PD-L1 binders. The evidence reported is from yeast-display screening; it does not establish performance in animals or people, or therapeutic effectiveness. The work is a preprint and has not yet undergone peer review.