Bonn researchers used AI to design short proteins for CAR-T, a therapy that directs T cells against tumors; cell experiments linked receptor function to protein charge
Bonn researchers used AI to design short proteins for CAR-T, a therapy that directs T cells against tumors; cell experiments linked receptor function to protein charge
On August 20, researchers from the University of Bonn and University Hospital Bonn published a study on short proteins designed with AI for CAR-T cell therapy. In cell experiments, the authors linked receptor function to properties of the protein surface outside the target recognition site.
CAR-T is a cell therapy in which an immune T cell is equipped with an artificial receptor. Its external domain recognizes a protein on a tumor cell, while its internal domain transmits a signal that triggers an attack. The external domain can be a mini-binder, a short artificial protein designed for a specific target.
In Biohub experiments with designed mini-proteins, the laboratory tested the first requirement: whether a candidate binds its target. The Bonn study addresses the next question: whether such a protein remains functional when used as the external domain of a CAR.
The team generated thousands of candidates for three proteins on the surface of tumor cells, PD-L1, CD276, and VTCN1; one thousand candidates entered the first cell-based screen. Among cells carrying the mini-binders, 2.4% produced a positive PD-L1 binding signal; the proportions for CD276 and VTCN1 were below 1%. The results varied substantially by target even among related proteins, so laboratory testing became part of the design process.
The authors incorporated several CD276 mini-binders into CARs. They compared T cells exposed to CD276-expressing A375 melanoma cells, the same cell line with CD276 knocked out, and cells in which CD276 expression had been restored. This comparison allows the T-cell response to be attributed to the presence of CD276.
A CAR containing one of the mini-proteins, HM9, showed poor surface expression on T cells. The authors preserved the amino acids at the site where HM9 contacts CD276 and used an algorithm to explore substitutions at the remaining positions. The library expanded to 5,992 variants. Variants with a pI of 6.5–8.5 were more likely to bind CD276 and less likely to have poor representation on the cell surface. The pI describes the protein’s charge balance: it is the pH at which the protein has, on average, no net charge.
The selected variants were incorporated into CAR T cells obtained from three donors. When cocultured with A375 cells, T cells carrying these variants produced more IFN-γ and TNF-α than cells carrying the original HM9 when they encountered CD276-expressing cells. These signaling proteins are produced during T-cell activation. Cells lacking the target induced almost no response. Over the next 24 hours, the variants also killed CD276-expressing A375 cells more effectively than the A375 version without CD276. The experiment defined a specific goal for the next design round: preserve the recognition site, optimize the properties of the remaining mini-protein surface, and test receptor function in a living T cell.