Under Propofol, the Hippocampus Still Sorts Nouns From Verbs and Leans Toward the Next Word

What does an anesthetized brain do with the voices in the operating room? I had always filed that under “nothing.” You count backward, the propofol pours in, and the next voice you hear is a nurse in recovery asking how you feel. So I wasn’t braced for a paper showing that one part of an anesthetized brain keeps listening the whole time, and learns from what it hears.

A Baylor College of Medicine team led by neurosurgeon Sameer Sheth recorded single neurons in the hippocampus, the brain’s memory hub, in seven people under propofol general anesthesia. Those cells learned to flag an out-of-place tone. According to Baylor’s announcement, they produced distinct activity patterns for nouns, verbs and adjectives in a spoken story, and their firing carried information that could be used to predict words the patients hadn’t heard yet. The work appears in Nature as “Plasticity and language in the anaesthetized human hippocampus”, with neurosurgery resident Kalman “Vigi” Katlowitz as first author. “This kind of predictive coding is something we associate with being awake and attentive,” said Benjamin Hayden, a Baylor neurosurgery professor on the team, “yet it’s happening here in an unconscious state.”

How you get a microphone onto a neuron

You can’t run this experiment on volunteers, because scalp electrodes can’t pick out individual cells that deep. Instead the team worked with patients whose drug-resistant epilepsy was bad enough to call for anterior temporal lobectomies, operations that remove the tissue where the seizures start, so the hippocampus they recorded from was slated for removal anyway. With propofol holding sedation steady, the surgeons placed Neuropixels probes, high-density devices that can listen to hundreds of neurons at once, and captured 651 hippocampal neurons. That’s a lot of single cells from very few people. There were seven brains, each shaped by years of seizures, so none of them is a typical hippocampus.

The first test was the classic oddball setup: patients heard a repeating tone with occasional unexpected pitches. The neurons fired differently for the odd ones out, and over about 10 minutes they got better at telling the two apart. Nobody in that room was awake to practice, and the circuit improved anyway.

Wait, why would the memory center care that a beep is off-key? My first instinct was that it shouldn’t, and I had it backward, because catching a broken pattern is close to the hippocampus’s day job. To spot an oddball, the circuit has to keep a running sketch of what usually comes next and check each incoming sound against it. Propofol had taken away the patients’ responses and their recall, and that comparison kept going. The team trained a recurrent neural network on the same discrimination, and the model showed the same pattern as the human recordings. The researchers take that as a sign that local wiring can manage this kind of learning without conscious attention driving it. That’s a reasonable reading, but a model that reproduces a pattern shows a computation that could explain it, not what the living tissue actually did.

Then they played it a podcast

Four of the patients heard stories from The Moth Radio Hour. Some neurons separated nouns from everything else. Across the population, “dog” and “cat” landed closer together than “dog” and “pen,” which means the firing was tracking meaning and not only sound. Nautilus reports that the cells responded to differences in word frequency, part of speech or category about 86 percent of the time, and says that rate is similar to what’s seen in awake patients.

The prediction result is the one people will fight over. “The brain appears to anticipate what comes next in a story, even without conscious awareness,” Sheth said in the Baylor release. That’s the inference I’d push on hardest. A sentence’s early words constrain its later ones, so a cell still reacting to what it just heard can look as if it’s reaching ahead. Earth.com notes that whether the signal reflects true prediction or simply how words in context color a cell’s response is still unsettled. Either way, this hippocampus was doing a lot more than sitting dark.

Neither experiment can tell you what, if anything, the patients experienced. In this study “unconscious” is a clinical label: unresponsive, and with no memory afterward. Writing in Frontiers in Science last year, neuroscientist Lucina Uddin noted that “despite decades of investigation, there is still a lack of consensus on answers to many key questions” in consciousness research, and that experience can only be reliably studied in people who can report it. An anesthetized patient can’t. The Baylor team says plainly that its results don’t show the patients understood the stories, and that one anesthetic and one brain region tell you nothing yet about sleep, coma or other drugs.

Music, kind words and a blank tape

When anesthesia reassures you about awareness, it’s really reassuring you about memory. The UK and Ireland’s NAP5 audit counted reported accidental awareness in roughly 1 in 19,600 anaesthetics, and the cases clustered heavily around neuromuscular blockade: about 1 in 8,200 with the paralyzing drugs, against 1 in 135,900 without them. Those are patients who woke up and remembered. The Baylor recordings describe something that audit was never built to count, a brain taking in and organizing sound that it will never recall.

A randomized trial has already treated that sound as part of the care. At five German hospitals, researchers analyzed 385 of 400 recruited patients having one- to three-hour operations under general anaesthesia. Half wore earphones playing background music and positive, hypnotherapy-based suggestions in 20-minute blocks with 10-minute pauses, and the other half heard a blank tape. In the 24 hours after surgery, 63 percent of the suggestion group needed opioids, compared with 80 percent of controls, so one patient was spared opioids for every 6 treated. Median use was 4.0 mg of morphine equivalents against 5.3 mg, pain scores were about 25 percent lower on average, and no adverse events were reported.

The trial’s authors called the approach safe, feasible, inexpensive and non-drug. They also added a warning of their own: a considerable number of patients perceived sounds during surgery, so surgeons and anaesthetists should be careful about background noise and conversations. A tape of music and kind words has no sales force behind it.

The Baylor team never tested suggestions, and the German trial never recorded a neuron. Nobody has shown that hippocampal word-tracking is how the tape worked, and the trial tested music and suggestions together, not a playlist alone, and certainly not what idle chatter does. Read side by side, though, the two studies point the same direction. Under propofol, sound still reaches a brain that sorts it, and in a controlled trial, what reached patients’ ears changed how much pain medicine they needed afterward.

If I’m ever wheeled into an operating room, I’m asking for earphones with music and something kind on loop, and I’m asking the team to save their commentary on my scans for when I can answer back.

Sources

1. Nature: Katlowitz et al., Plasticity and language in the anaesthetized human hippocampus (2026)
2. ScienceDaily: Mysterious brain activity under anesthesia challenges what we know about consciousness (Baylor College of Medicine release)
3. BMJ: Nowak et al., therapeutic suggestions during general anaesthesia (2020)
4. British Journal of Anaesthesia: NAP5 on accidental awareness during general anaesthesia, summary of main findings (2014)
5. Nautilus: Your brain can learn things when you’re unconscious
6. Earth.com: Unconscious brains can still process language and predict words
7. The Debrief: The unconscious brain may be more capable than scientists realized
8. Frontiers in Science: Uddin, What we (can) know about consciousness (2025)