Lila said its AI learns from 950 thousand mRNA experiments and tested an instruction for immune T cells in primates
Lila said its AI learns from 950 thousand mRNA experiments and tested an instruction for immune T cells in primates
On August 28, Lila Sciences described its workflow: an AI model proposes mRNA variants, and automated laboratories synthesize them and test them in cells. In an American Society of Gene and Cell Therapy (ASGCT) conference abstract, the team described a preclinical test in which lipid particles deliver mRNA directly to immune T cells in the body, temporarily giving them a CAR that enables them to attack B cells.
In May, Lila appeared in the SciFM 26 announcement as one of the teams promising to repeat a cycle of model, laboratory, and new data. In its August disclosure, the company reported the scale of this cycle and presented a preclinical test of the approach.
mRNA is a temporary instruction that a cell uses to produce a protein. It contains a coding region that specifies the protein and two flanking regions: the 5'UTR affects the initiation of translation, while the 3'UTR affects how long the instruction remains stable in the cell. The number of possible combinations of these three regions is too large to test through a conventional series of experiments.
Lila reports that it physically synthesized and tested more than 950 thousand mRNA constructs in living cells. After each round, the results were returned to the model. The model proposed the next variants, and the laboratory synthesized and tested the best candidates again. For erythropoietin mRNA, which encodes a protein that stimulates red blood cell production, one construct produced as much protein on day 15 as the reference mRNA produced on day two. The company also reported greater stability for five additional constructs encoding other proteins.
Lila then tested whether this method could temporarily equip immune T cells with a CAR, an engineered receptor that recognizes a selected target. The mRNA encoding the CAR was packaged in lipid nanoparticles. An antibody against CD8, a protein found on the surface of a subset of T cells, directed the particles to those cells. The researchers first counted the T cells that expressed the CAR and then measured the number of B cells.
In mice with human immune cells, the CAR was detected after one day in approximately 30% of CD8 cells treated with the Lila variant and approximately 8% of those treated with the comparator. At a dose of 0.25 mg/kg, the Lila variant depleted B cells in the blood and spleen more strongly than the comparator at a dose of 1 mg/kg. In primates, a single 0.5 mg/kg dose of the Lila variant produced near-complete depletion of B cells in the blood through day seven. On day 14, B-cell levels remained at about half of baseline. With the comparator, B-cell levels had returned to baseline by that time.