Opening an Intracellular View for Antibodies: Computational Design Targets the KRAS G12D Cancer Mutation

Many cancer-driving proteins are hidden inside cells. Even when conventional antibodies can precisely recognize molecules, they have difficulty crossing the cell membrane to reach their targets. KAIST and protein design company TheraZyne took a different approach: rather than pursuing the target into the cell, they recognize the KRAS G12D mutant peptide that cancer cells transport to their surface, turning a previously hidden oncogenic signal into a target visible to immunotherapy.

These “T-cell receptor-like antibodies” target a complex formed by the KRAS G12D decapeptide and HLA-C*08:02. Cells continuously break down internal proteins and display some of the fragments on their surface through major histocompatibility complex class I. If an antibody can distinguish the mutant peptide without mistakenly recognizing fragments of normal proteins, it may be possible to convert an intracellular mutation into an attackable surface marker.

The team first used computational methods to construct human antibody variable regions capable of binding to the target complex. They then introduced sequence variations at a small number of positions in the complementarity-determining regions, created a yeast surface-display library, and performed repeated screening. In this process, the model does not directly produce a drug candidate. Instead, it first narrows the experimental search space, after which cellular and molecular screening eliminates candidates with insufficient binding affinity or selectivity. Following affinity maturation, the researchers obtained antibodies with dissociation constants in the nanomolar range.

Functional tests showed that, after these antibodies were engineered into chimeric antigen receptors or bispecific T-cell engagers, they could selectively kill cells carrying the target complex. Phage-display analysis found no off-target reactivity against the human peptides tested, while computational assessment predicted low immunogenicity. The Protein Data Bank also contains an X-ray structure of a candidate antibody Fab in complex with the KRAS peptide and HLA-C*08:02. The 4.54-angstrom resolution provides experimental evidence of binding, but is insufficient to replace higher-resolution, atomic-level validation.

Several hurdles remain before this can become a treatment. Current results come primarily from in vitro models. The research team said animal validation is underway, and there are not yet any human data on safety, efficacy, or durability. The absence of cross-reactivity in the phage database also cannot cover all tissues and protein environments in living organisms. The actual number of peptides displayed by tumors, heterogeneity among different cancer cells, and immune escape could all influence the efficacy and toxicity of CAR-T cells or bispecific molecules.

Eligibility is also determined by more than the KRAS mutation alone. The candidate antibodies depend on HLA-C*08:02 to present the peptide, so only patients who have both KRAS G12D and the matching HLA type may benefit. Patients with pancreatic cancer, colorectal cancer, or lung cancer would still need to undergo genetic and HLA testing for selection. TheraZyne lists the program as TZ-Ab101, and the paper also discloses that the company has filed a provisional patent application for the related antibodies. At this stage, the more firmly established result is that computational design and limited experimental evolution can jointly generate functional TCR-like antibodies—not that a cancer therapy ready for clinical use has been established.