Discovery

Protein charge emerges as key constraint for AI-designed cancer minibinders

A study from researchers at the University of Bonn demonstrates that the isoelectric point (pI) of AI-designed protein minibinders — not just their binding...

Protein charge emerges as key constraint for AI-designed cancer minibinders

Researchers at the University of Bonn have identified the isoelectric point (pI) of AI-designed protein minibinders as a key determinant of whether they function effectively when incorporated into chimeric antigen receptor (CAR) T cells, according to a study published August 20 in Nature Communications. The finding suggests that optimizing binding affinity alone may be insufficient when using generative protein design to build next-generation CARs.

The researchers developed minibinders against three B7-H family proteins expressed on cancer cells — PD-L1 (B7-H1), CD276 (B7-H3), and VTCN1 (B7-H4) — using a computational pipeline combining RFdiffusion for protein backbone generation, ProteinMPNN for sequence design, and AlphaFold2 and Chai-1 for structural evaluation. Candidates were subsequently screened experimentally and shown to recognize endogenous targets on human cancer cells.

The key finding emerged when selected minibinders were incorporated into second-generation CAR constructs. Despite retaining target-binding activity, some produced poor CAR surface expression and reduced tumor-cell killing selectivity. The researchers traced the effect to the pI of the minibinder domain, with highly basic binders showing reduced and heterogeneous presentation on the T-cell surface.

The authors propose that positively charged minibinders undergo nonspecific electrostatic interactions with negatively charged components of the cell surface, interfering with trafficking and membrane presentation. This means that the portions of an AI-designed binder outside its target-binding interface can materially affect the performance of the resulting cell therapy.

To test whether the property could be engineered, the researchers generated approximately 6,000 variants of HM9, a CD276 minibinder with a pI of 9.4, while preserving residues involved in target binding. Computational and experimental screening identified a pI range of approximately 6.5–8.5 as optimal. CAR-T cells incorporating variants within that range showed improved surface expression and greater selectivity for CD276-positive tumor cells across three independent human donors.

The AllSci BriefSystematic R&D and deal news. Daily.

The findings introduce an additional optimization parameter for developers using AI-generated minibinders as alternatives to conventional antibody-derived CAR recognition domains. A candidate with strong predicted structure and high target affinity may still perform poorly as part of a CAR if its broader physicochemical properties interfere with receptor expression or trafficking.

The result could be relevant to emerging commercial efforts to use generative protein design in cell therapy. Generate:Biomedicines has developed GB-5267, an AI-designed MUC16-targeting CAR-T therapy that the company has said is expected to enter Phase I development in 2026. Academic groups have also used RFdiffusion and ProteinMPNN to generate synthetic binders for CAR-T applications, including peptide-MHC targets.

The Bonn constructs remain preclinical. Testing was performed in cancer cell lines and primary T cells from healthy donors, with no animal efficacy studies or clinical development program reported. Further work will be required to determine whether the pI relationship generalizes across different minibinder architectures, CAR designs, and target classes.


Access the AllSci platform to explore the science behind the news.


Spot something wrong? Report an issue with this article