Bringing Intracellular Mutations to the Surface: Computer-Designed Antibody Targets KRAS G12D

Most antibodies can recognize only targets on the cell surface and are often powerless against oncogenic proteins hidden inside cells. A research team has now taken a different approach: instead of sending antibodies directly into cells, it designed them to recognize the molecular complex formed when cells “display” a KRAS G12D mutant fragment on their surface, thereby creating an opening to attack an otherwise difficult-to-reach cancer mutation.

This type of “T-cell receptor-like antibody” does not target the full KRAS protein. Instead, it targets a mutant peptide composed of 10 amino acids together with the human leukocyte antigen HLA-C*08:02 that presents it. Cells break down internal proteins into short fragments, which are then transported to the surface by major histocompatibility complex class I molecules; the antibody therefore reads, from outside the cell, a molecular calling card originating from inside it.

The study first used computer simulations to generate human antibody variable regions capable of fitting the target complex and performed limited sequence design on the complementarity-determining regions directly involved in recognition. The team then assembled these variants into a yeast surface display library and, through multiple rounds of experimental screening and affinity optimization, obtained a candidate antibody capable of recognizing the KRAS G12D peptide–HLA-C*08:02 complex with high specificity. The point of this process was not for an algorithm to directly deliver a drug candidate, but to use computation to narrow the search space, followed by successive rounds of experimental elimination.

In selectivity testing, phage display analysis found no unintended reactions between the candidate antibody and the tested human protein fragments, while computational assessment also predicted low immunogenicity. However, these findings cover only the databases and models used and do not mean that all cross-reactivity in the human body has been ruled out.

Structural data provided another layer of validation. The 9WJ6 structure deposited in the Protein Data Bank contains the KRAS G12D peptide, an HLA class I antigen, β2-microglobulin, and the Fab light and heavy chains of the candidate antibody C8K10D5-1. The structure was determined by X-ray diffraction at a resolution of 4.54 angstroms. These data support that the molecules do form a complex, but the resolution is relatively limited, so interpretations of detailed atomic interactions still require caution.

The researchers further engineered the antibody into a chimeric antigen receptor and a bispecific T-cell engager. In cell experiments, both formats selectively killed cells carrying the target complex. This indicates that the same recognition element can be coupled to different immune killing mechanisms, but the evidence remains preclinical and has not yet demonstrated sufficient efficacy or a sufficient safety margin in animals or patients.

Background

This work also highlights the practical role of AI and computational design in drug development: models can accelerate the generation of candidate sequences, but biological validation remains the real hurdle. The therapy is currently restricted to HLA-C*08:02 and may be applicable only to patients who carry the matching HLA type and whose tumors can consistently present sufficient quantities of the mutant peptide. Tumor antigen heterogeneity, immune escape, cross-reactivity with normal tissues, and product safety are all hurdles that must be overcome before moving toward clinical development.