AI is helping find the immune cells that recognise cancer

In a Nature study published on 16 September, 21 of the 25 highest-ranked immune receptors from a six-patient validation cohort reacted to tumour cells in laboratory tests. That 84% describes a selected shortlist, not treatment success.[1]

The shortlist came from PreGame, developed at Toronto’s Princess Margaret Cancer Centre. Its quarry is a specialised immune cell called a gamma-delta T cell. Finding the rare members that naturally recognise cancer has been a bottleneck for researchers hoping to turn such cells into therapies. A tumour contains bystanders as well as potential attackers; location alone does not tell scientists which is which.[2]

Two views of the same cell

One enabling technique is CITE-seq, introduced in 2017. It reads gene activity alongside selected proteins on a cell’s surface. Antibodies carrying DNA tags bind to those proteins; sequencing connects the antibody tags and RNA molecules to the same individual cell. Researchers can therefore compare what a cell is making with what it displays. It is a way to preserve biological distinctions that could disappear if thousands of cells were blended into one average.[3]

PreGame’s development used bone marrow samples from 22 people with multiple myeloma, divided into discovery and validation cohorts. A random forest learned gene, protein and cell-cycle signatures from cells whose receptors had already been tested for tumour reactivity. The computer’s labels were grounded in those experiments.[1]

That sequence changes the job of the algorithm. It helps decide which receptor deserves the next experiment. The laboratory still has to establish what the receptor recognises. A useful shortlist can accelerate discovery without requiring the model to explain every detail of the immune response.

Recognition with a brake

One receptor, TCR65, recognised HLA-C, a marker also found on healthy cells. Further experiments showed how an inhibitory receptor, KIR3DL1, could suppress activation when a second marker, Bw4, was present. This suggests a biological gate: recognise a target, but withhold attack when a reassuring second signal remains. Some tested receptors also reacted to healthy cells; clinical safety is unproven.[1]

Co-first author Michael St. Paul sees potential for gamma-delta receptors to work across different patients and cancer types because their recognition differs from conventional T-cell approaches. That is the therapeutic ambition behind the search. The team has filed a provisional patent covering PreGame and associated markers.[2]

The compelling possibility is that a better search method can uncover useful immune biology along the way. Selecting a receptor and understanding its restraint are connected discoveries: the second may determine whether the first can become a treatment.

Report an error or suggest a correction