Several anaesthetics that work in opposite directions at the molecular level push six species toward exactly the same shape.
Someone pushes a syringe before surgery and asks you to count to ten. You get to three, and then — nothing. Those hours weren't sleep, and they weren't sleep you later forgot. They simply never happened for you. Hundreds of thousands of people go through this every day, and what exactly that syringe switches off is still an open question.
What makes it harder: the common anaesthetics work in nearly opposite ways down at the molecules. Some strengthen the brain's brake, others cut its throttle. Different targets, different proteins — yet people go under looking exactly the same. Which sharpens the question: if there's no common ground in pharmacology, what level is it hiding on? A study published in September 2026 rephrased the question and got an answer out of six species. The answer is not "less activity".
Consciousness research has an old habit: pick a theory first, then go hunting for evidence that supports it. This study inverts the order. No theory assumed, no definition of consciousness required — just one question: which mathematical properties of the brain's activity curve change between awake and anaesthetised?
"Dynamics" here doesn't mean what the brain is computing. It means the shape of the wiggly line itself: how jittery, how fast, how regular, how much the current value has to do with the value a moment ago. It's a bit like ignoring what someone is saying and attending only to their rhythm, pauses and swings in volume. Content thrown away, form kept.
The approach is deliberately brute-force. Using a toolbox called hctsa, the researchers computed more than 6,000 statistical features for every activity time series — useful and useless alike — and then asked which ones moved the same way across every species and every anaesthetic. The dataset spans not just species but scales: functional MRI in human, macaque, marmoset and mouse (whole brain regions), and single-neuron imaging in larval zebrafish and in the nematode C. elegans. A few hundred neurons at one end, a whole human brain at the other.
Of those 6,000-plus features, the ones that kept showing up were dominated by a single family: autocorrelation. The name is clunky, the idea isn't. Take a stretch of activity and compare it with how it looked a moment earlier. Very similar means this patch of brain is still carrying its recent state forward. Not similar means it has already moved on. How long it carries things forward has a name: the intrinsic timescale.
Under anaesthesia, autocorrelation decays markedly faster — intrinsic timescales shorten. And here's the part that's easy to get backwards: anaesthesia does not simply turn the volume down. Neurons still fire, signals still fluctuate, and the amount of activity isn't necessarily lower. What's lost is continuity: the brain is still moving, it just remembers less and less of what it was doing one step ago.
Artificial networks have a family called recurrent networks (used for sequences — speech, text, anything that arrives one piece at a time). Each unit keeps a fraction of what it just received and mixes it into the present. Keep more, and that unit's effective time constant is longer, so it can carry something mentioned at the start of a sentence all the way to the end. Shrink that coefficient across the board and the network doesn't break or stop — it simply loses the ability to handle long-range dependencies. On these statistics, the anaesthetised brain looks like someone turned that one knob down.
Forgetting faster locally produces a second consequence at the whole-brain level: inter-regional synchrony drops. The two are really one thing. Think of a choir — to stay with everyone else you first have to hold your own pitch. If each singer jumps to a random note twice a second, sitting together changes nothing. Shorten local timescales and cross-regional synchrony has to follow.
Put together: anaesthesia isolates local activity in space and in time — cut off from its own past, and cut off from everyone else.
What's interesting is that this result takes nobody's side. The two camps that have been arguing for thirty years — Global Workspace, where consciousness depends on brain-wide broadcast, and Integrated Information Theory, where it depends on the whole being irreducible — predict the same direction here. Broadcast cut and whole fragmented are two descriptions of one phenomenon. So this isn't a referee. It's closer to a hard constraint every theory has to satisfy: if your account lets a brain with a one-moment memory and no cross-talk still be conscious, you owe us an explanation for why all six species go down right here.
The change isn't uniform across cortex: some areas shift hard, others barely move. The researchers compared that map of "where it shifts most" against a completely independent one — the transcriptional atlas of how strongly genes for excitatory and inhibitory neurotransmission are expressed across cortex. The two maps covary, and the covariation holds in human, macaque and mouse cortex alike.
In plain terms: how vulnerable a patch of cortex is to anaesthesia depends on how its throttle and brake are wired. (neurotransmitter systems inhibitory interneurons)
But a beautiful correlation is still only a correlation. The sharpest cut comes from the macaque: put an electrode into the centromedian thalamus — the group of thalamic nuclei that doesn't relay any particular sense but governs global arousal (thalamus) — switch on the current, and the brain's dynamics move back the other way while the animal recovers behavioural responsiveness. Drugs push it to one end, current pulls it back; one axis, both directions traversed. That step turns correlation into causation, and turns a set of statistics into a handle you can actually grab.
The "now" turns out to have thickness — and that thickness can be thinned continuously by a drug. Several fields that never talk to each other meet on this: