AI can spot an unusual cell. SPARCS lets researchers retrieve it.

The useful cells were a tiny minority. In one experiment, researchers examined 29.4 million cultured human cells and selected 2,071 candidates whose images suggested that an immune-signalling protein had failed to reach its expected destination. They then used a laser to cut out the selected nuclei and recovered genetic information from the collected material. An unusual picture had become something they could investigate.[1]

That connection is the central achievement of SPARCS, reported by Niklas Schmacke, Sophia Mädler and colleagues in Cell online on 6 October. The platform helped identify a regulator of STING, a protein involved in innate immunity, and trace its effect to the acidity of an intracellular compartment. The AI guided the search; gene editing, chemical perturbations and molecular measurements established the biological explanation.[1][3]

A microscope image with a recoverable address

A genetic screen asks what changes when particular genes are disabled. In a pooled CRISPR screen, many differently edited cells grow together, each carrying a short guide-RNA sequence that identifies its perturbation. Researchers need a way to find the cells with an interesting response and recover those genetic labels. A simple fluorescence measurement can report how brightly a cell glows. Microscopy can also reveal where the glow sits, whether it forms puncta and how structures are arranged.[1]

SPARCS preserves that spatial information. Cells grow on a thin membrane suitable for laser microdissection and are photographed with a confocal microscope. Image analysis separates individual cells; trained classifiers recognise the patterns relevant to the experiment. Software then translates selected image regions into cutting paths, allowing a focused ultraviolet laser to excise material at those coordinates.[1]

Across screens involving 70 million cells, the team investigated autophagy and STING. For autophagy, the relevant pattern was the appearance of fluorescent puncta marking autophagosomes, the structures cells use to package material for recycling. For STING, the classifier learned different locations inside the cell. These questions depend on the arrangement of a signal, so retaining the image changes what the screen can ask.[1][2]

The collection stage has its own practical limits. The reported system cut about 1,000 nuclei an hour and recovered approximately 80% of excised membrane regions in its collection test. For genetic readout, the researchers sequenced pools of at least 500 excised cells. The current workflow therefore connects an image-defined group to enriched gene perturbations; it does not assign a separate genetic label to every individual image.[1]

Following an immune signal through the cell

STING made a revealing test because its location is part of its function. Under the study’s trafficking-dependent stimulus, the protein moves from the endoplasmic reticulum, an internal membrane network, through the Golgi apparatus and onward. The screen looked for cells in which that journey stalled. One enriched gene was GPR89B, which, along with its close relative GPR89A, encodes the Golgi channel GPHR.[1]

The Golgi normally maintains an acidic interior. GPHR helps sustain that environment by allowing chloride ions to move across its membrane. When the researchers disabled both GPR89 genes in cultured cells, the Golgi became less acidic and STING remained in the endoplasmic reticulum after treatment with the activating compound diABZI. Restoring GPR89B enabled STING to travel to the trans-Golgi network again.[1]

Two further interventions sharpened the explanation. A different compound, C53, which activates STING directly in the endoplasmic reticulum, still worked largely normally in the knockout cells. And monensin, a chemical that makes the Golgi less acidic, reproduced the delay in STING signalling. Together with direct pH measurements and tests of downstream signalling, these results connect the defect to the compartment’s acidity and the protein’s route through the cell.[1]

Physical retrieval also made a second kind of measurement possible. Mass spectrometry quantified roughly 5,000 proteins from samples of 50 selected cells. Comparing groups with different image patterns exposed molecular differences that microscopy alone could not reveal. This is why collecting the material matters: a shape or location can become the starting point for another experiment.[1]

SPARCS makes the image an entry point into mechanism. The decisive finding was that a protein’s journey depended on the chemical conditions inside an organelle. The platform supplied a route from millions of pictures to the small set of cells that made that dependence visible.[1]

Report an error or suggest a correction