A computationally designed version of a human T-cell receptor improved receptor expression, reduced unwanted chain pairing, and made engineered T cells better at recognizing and killing tumor cells.

T-cell therapies can turn the immune system into a powerful tool against cancer. T cells are immune cells that help identify and destroy infected or abnormal cells. They recognize potential targets using proteins on their surface called T-cell receptors (TCRs). Cells display small fragments of proteins on their surface, and TCRs recognize specific fragments presented by other surface molecules. When a T cell recognizes its target, it receives signals that activate it to respond. Cancer cells can evade this process, so one cancer treatment involves genetically modifying a patient’s T cells to recognize a specific cancer-related target.

But engineering these receptors comes with a problem: a TCR is made from two chains that must pair correctly for the receptor to function. When an engineered TCR is introduced into a patient’s T cells, its chains can pair incorrectly with the cell’s original TCR chains, reducing the amount of functional receptor on the cell surface and potentially creating unwanted specificities.

A new study used AI to redesign the less visible part of the TCR. The resulting receptor, called structurally enhanced TCR (SET), improved T-cell function and increased the ability of engineered T cells to control tumors in mice.

A TCR contains variable regions that recognize the target and constant regions that help the two TCR chains fit together and connect with the rest of the T-cell signaling machinery. For an engineered T cell to work, both chains have to assemble correctly before the completed receptor reaches the surface. However, the engineered chains must also compete with the cell’s natural TCR chains, which can further reduce the number of functional engineered receptors.

The TCR’s constant regions do not determine which cancer target the receptor recognizes. Two computational protein-design algorithms searched for small changes that could make the receptor more stable and encourage the chains to pair with each other.

The first round focused mainly on the outer surface of the constant regions, while the second focused on the point where the two chains meet.

The final SET receptor contained 18 changes to the human TCR constant regions. Some of these changes altered the way the two chains fit together. For example, one change recreated stabilizing interactions between the two chains, while another repositioned an amino acid to form a new interaction with the first chain.

More Receptors Reach the Cell Surface

Despite competing with the cell’s natural TCR chains, the redesigned TCR produced more functional receptor on the surface of human T cells. In primary human T cells, SET produced about twice as much target-specific receptor staining as the original TCR.

The increase in receptor expression translated into stronger T-cell activity. When SET-equipped T cells were exposed to melanoma cells carrying their target, they released more signaling proteins that indicate T-cell activation. SET also improved activity in other T cells, even though these cells lacked the molecule that normally helps stabilize interactions between a TCR and its target.

SET was also tested with TCRs designed to recognize four unrelated targets, including other cancer antigens and a viral protein. In each case, the redesigned constant regions increased the release of signaling proteins compared with the original receptor, suggesting that changes can be transferred to different TCRs without changing their targets.

Turning Better Receptors into Better Cancer Killing

One key question was whether the redesigned receptor could actually improve tumor control. In laboratory experiments, SET-equipped T cells killed more melanoma cells than T cells carrying the original receptor. After 48 hours, about 30% of the tumor cells remained viable when exposed to SET T cells, compared with about 78% with the original TCR.

In mice, the original receptor had little effect on tumor growth. SET-expressing T cells, however, significantly delayed tumor growth and extended survival.

After 83 days, tumors in the SET group were 35% smaller on average than those in the groups receiving the original or earlier redesigned receptors. By day 127, all of the mice treated with SET T cells were still alive, compared with fewer than half of the mice in the control groups.

The study demonstrates a different way to improve engineered T-cell therapies. Instead of repeatedly modifying the portion of a TCR that recognizes a cancer target, computational design can improve the receptor structure.

The approach is also relatively compact, as these improvements were achieved through a limited number of computational design rounds. Because the SET changes can be added to TCRs with different target specificities, the same engineering strategy could potentially be used across multiple therapies.

Additional testing will be needed before SET-modified TCRs can be evaluated as treatments in people. But it points to a broader possibility for computational protein design: AI may not need to invent an entirely new therapeutic molecule to make a difference.