Effects of suppressing 12 779 genes measured in primary human T-cells
Effects of suppressing 12 779 genes measured in primary human T-cells
On 28 August, Cell published a study of CD4+ T-cells, immune cells that coordinate the immune response. The team collected blood cells from four healthy donors, partially suppressed one gene at a time, and measured the response at rest and at 8 and 48 hours after restimulation.
A conventional single-cell atlas shows which genes are active at the same time. In Perturb-seq, researchers introduce a short guide RNA into a cell. The CRISPRi system uses this RNA to partially suppress a selected gene, after which the researchers measure which other genes become more or less active in response. This method therefore measures the effects of a specific intervention.
After quality control, the researchers retained 21,99 million of the 33,4 million cells in which the guide RNA could be assigned unambiguously to an intervention. For 11 527 suppressed genes, they assessed changes in the activity of other genes and identified 2 035 311 significant pairs linking a suppressed gene to an altered gene. Each pair was associated with a specific T-cell state.
The researchers compared effects across different guide RNAs, donors, and independent, separately conducted CRISPR experiments. They then tested nine known and newly identified regulators of two immune signaling proteins, IL10 and IL21, in separate experiments. These tests used two independent guides per target, repeated RNA measurements, and protein measurements. In all 12 independently tested relationships between the suppression of one gene and a change in another, the effect had the same direction as in the main experiment. Ten of these relationships were statistically significant.
Suppressing the same gene produced different effects in the three cell states. At 8 hours after stimulation, the authors observed early groups of genes that acted together, including groups involved in intracellular calcium signaling. At 48 hours, they observed late groups containing genes involved in cell division and genome maintenance.
This range of responses turns the “virtual cell” into a testable problem: given the cell state and intervention, a computational model must predict the measured response. Senior author Alex Marson described this requirement in a Gladstone Institutes news release:
“If cells are examined in only one state, AI will struggle to predict their behavior in the real world. A useful virtual cell requires systematic data on responses that depend on cell state.”
The researchers also applied the map to an age-related CD4+ T-cell signature, a set of changes in gene activity associated with age. They derived the signature from the OneK1K study, which included 782 participants, and validated it in another 199 participants. Perturbation effects measured in resting T-cells reproduced this signature more accurately than measurements from stimulated cells or a laboratory cell line. The model identified the mTORC1 signaling system, which is involved in growth and metabolism, and the enzymes that attach small SUMO molecules to proteins and modify their function after synthesis as regulators of different parts of the signature.
The authors released the experimental data and code. These resources allow researchers to test such a prediction against the measured effect of a selected perturbation in one of the three cell states.