# Lesson 4: How Psychedelics Change the Brain: Networks, Plasticity and Learning

*Companion-podcast transcript • Sarah & Kiffer*  
*Office Hours episode to listen to after working through the lesson*

---

**Sarah:** In a study published in 2026, a team in Copenhagen used positron emission tomography, or PET, to scan fifteen healthy volunteers before and one week after a single dose of psilocybin. They were looking in living human brains for the extra synapses that mouse and pig studies keep reporting.

**Kiffer:** They measured SV2A, a protein in synaptic vesicles that serves as a marker of synapse density. Overall, they found no statistically significant increase, and the study was stopped for futility at a planned interim analysis.

**Sarah:** And yet the same paper can be read as a blow to the plasticity story or as support for it, because the volunteers dosed in a therapy-like room showed larger increases than the ones dosed inside the scanner.

**Kiffer:** Which reading holds up depends on the numbers, and Sarah and I don't land in quite the same place on them.

**Sarah:** Welcome back to Office Hours. I'm Sarah.

**Kiffer:** And I'm Kiffer. What a brain measurement can and can't tell us runs all the way through Lesson four, How Psychedelics Change the Brain: Networks, Plasticity and Learning, so let's start with the networks.

**Sarah:** Question one. Does psilocybin switch the default mode network off?

**Kiffer:** Phrases like that turn up in headlines, and the studies describe something more specific. Start with the direction of change. In 1997, a PET study of ten volunteers found that frontal glucose metabolism rose after psilocybin. In 2012, a study of two groups of fifteen found only decreases in activity, largest in hub regions, including parts of the default mode network. Then in 2017, Lewis and colleagues scanned fifty-eight volunteers and found both at once. Relative blood flow rose in right frontal regions, while absolute flow fell across most of the brain.

**Sarah:** Both at once sounds impossible until you do the arithmetic, so let's try it with made-up numbers. Suppose that after psilocybin, blood flow across the whole brain falls by ten percent, and flow in one frontal region falls by four percent. Expressed relative to the whole brain, does that frontal region go up or down, and by roughly how much? Take a few seconds, and pause the audio if you'd like more time.

*(Pause)*

**Kiffer:** Relative flow is the region divided by the whole brain. Call both of them one before the drug. Afterward the region is at 0.96 and the whole brain is at 0.90, and 0.96 divided by 0.90 is about 1.07. So the region has lost blood flow in absolute terms and gained about seven percent in relative terms. Both numbers are correct. They answer different questions.

**Sarah:** So a headline that says frontal activity rises on psilocybin and one that says psilocybin quiets the brain could be describing the same scans.

**Kiffer:** They could. A direction can also flip with the analysis. Tagliazucchi and colleagues reported that LSD increased the global connectivity of association cortex. Preller and colleagues removed the global signal, the average signal across the whole brain, and found decreases, and when they kept it in, they reproduced the increases. Before I believe up or down, I want to know what was measured and how it was processed.

**Sarah:** What about the words switch off?

**Kiffer:** The most detailed description we have is desynchronization. In the precision mapping study, seven people were scanned about eighteen times each. Psilocybin changed functional connectivity more than three times as much as methylphenidate, the active control matched for arousal, and the change was largest in the default mode network. Regions stopped rising and falling together in their usual pattern. So disrupted coordination fits the data, and switched off overstates it.

**Sarah:** Question two. How much can a correlation from a small imaging study tell us?

**Kiffer:** It can tell us less than most headlines suggest. In the first modern LSD imaging study, reduced default mode network integrity correlated 0.49 with ego-dissolution in fifteen people, and its interval runs from about minus 0.03 to 0.80. Let's push that further with a hypothetical. Suppose a study of seven people, the size of the precision mapping study, reports a correlation of 0.8 between desynchronization and the intensity of the experience. Using Fisher's z, what's the ninety-five percent confidence interval? Take a few seconds.

*(Pause)*

**Sarah:** Fisher's z first. It's half the natural log of 1.8 divided by 0.2, which is half the log of nine, about 1.10. The standard error is one divided by the square root of seven minus three, so one divided by two, which is 0.5. Then 1.96 times 0.5 is 0.98. So the limits are 1.10 minus 0.98 and 1.10 plus 0.98, and the interval runs from about 0.12 to 2.08.

**Kiffer:** Before you go on, what's the largest value a correlation can take?

**Sarah:** It's one, so 2.08 is impossible. I stopped on the z scale and never converted back.

**Kiffer:** That's the slip, and it's easy to miss for a reason. Near zero, the z value and the correlation are almost the same number, so the lower limit looks believable whether you convert it or not. Only the upper limit gives the mistake away.

**Sarah:** Okay, let me convert. For each limit it's e to the power of two z, minus one, divided by the same thing plus one. The lower limit stays at about 0.12. For the upper one, two times 2.08 is about 4.16, e to the power of 4.16 is about 64, and 63 divided by 65 is about 0.97. So a correlation of 0.8 in seven people is compatible with anything from about 0.12 to 0.97.

**Kiffer:** That range runs from a weak relationship to a nearly perfect one. It fits the warning from an analysis of about fifty thousand people by Marek and colleagues, that brain measures relate to differences between people more weakly than small studies suggest, and that studies of typical size inflate those correlations. The warning applies most to correlations across people. Scanning each of seven people about eighteen times makes each person's comparison of psilocybin with methylphenidate well measured. A correlation across those people still rests on seven data points.

**Sarah:** There's a version of this trap that catches careful readers too. By 2022, the psychedelic resting-state literature had forty-two articles from only seventeen unique studies, and two datasets underlie more than half of it.

**Kiffer:** So here's a quick one. Suppose a hypothetical review collects three papers that each analysed the same fifteen LSD volunteers, adds up the participants, and treats a correlation of 0.49 as if it came from forty-five people. What happens to the interval?

**Sarah:** The standard error becomes one over the square root of forty-two, about 0.15. Fisher's z for 0.49 is about 0.54, so the limits on the z scale are about 0.23 and 0.84, which convert back to about 0.23 and 0.69. The interval no longer includes zero, so it looks like a confirmed relationship.

**Kiffer:** And nothing new was learned, because the same fifteen people were counted three times. Repeated analyses of one sample can look like replication. My first question about any imaging finding is whether an independent group has reproduced it in new people.

**Sarah:** Question three. Do psychedelics grow new connections in the human brain? The question reaches a lot of people in Canada. In the 2025 Canadian Substance Use Survey, published in June 2026, about seven percent of people aged fifteen and older, roughly 2.6 million, reported using a psychedelic in the past year, and psilocybin was the most common.

**Kiffer:** In cells and animals, several research groups have reported growth. Cultured rat neurons grew more branches after DMT, psilocin and LSD. In mice, one injection of psilocybin raised spine density in the frontal cortex by about seven percent after a day and twelve percent after a week. In pigs, a synapse marker was six to nine percent higher a week after a dose. The thorny part is how people read the mouse experiment with ketanserin.

**Sarah:** I'll admit I read it wrong the first time. Ketanserin blocks the serotonin 2A receptor, it abolished the head twitch, and the spines still changed. So I concluded that the receptor behind the experience isn't needed for plasticity, which would mean the experience isn't needed either.

**Kiffer:** That's a common reading, and it goes two steps too far. First, the authors noted that their ketanserin dose blocks only about thirty percent of serotonin 2A receptors in rodents, so the remaining receptors could still be doing the work. Second, a head twitch is a behavioural marker of hallucinogenic potential in mice, and nobody can ask a mouse what it experienced. What the study shows is narrower. With ketanserin, spine heads still widened and new spines still formed, while the rise in spine density became smaller and was no longer statistically significant.

**Sarah:** And another group found a result that points the other way. In mice bred without the receptor, Vargas and colleagues saw no spine growth after a related psychedelic.

**Kiffer:** Right, while Moliner and colleagues reported that the plasticity effects of LSD and psilocin in mice didn't require the receptor. So whether it's necessary is still open. And that brings us to people, and back to Copenhagen.

**Sarah:** Let me set out the design. It was a pre-registered, open-label study with a single arm and fifteen healthy volunteers. SV2A was measured with PET in the frontal cortex and hippocampus at baseline and one week after psilocybin. The team planned the sample so that fifteen people would give eighty percent power to detect a five percent increase, which is smaller than what the pigs showed.

**Kiffer:** At the planned interim analysis, in twelve people with complete data, the mean change in frontal binding was 0.3 units, which the authors say corresponds to 2.3 percent, and the standard deviation of the change was 2.6 units. The estimated chance of reaching significance at full size was twenty-four percent if the planned effect were true and two percent if the observed effect held, both below their futility threshold of fifty percent, so they stopped. Here's the question. From that mean and standard deviation in twelve people, roughly what's the ninety-five percent interval for the average change, and does it rule out an increase the size seen in pigs? Take a few seconds.

*(Pause)*

**Sarah:** The standard error of a mean change is the standard deviation divided by the square root of the number of people. The square root of twelve is about 3.46, and 2.6 divided by 3.46 is about 0.75. Times 1.96, that's about 1.47. So the interval runs from 0.3 minus 1.47, about minus 1.17, to 0.3 plus 1.47, about 1.77.

**Kiffer:** Now turn that into percent. If 0.3 units is 2.3 percent, each unit is about 7.7 percent.

**Sarah:** So minus 1.17 is about minus nine percent, and 1.77 is about plus fourteen percent. The data fit a nine percent decrease, a fourteen percent increase or anything in between, and that range includes the six to nine percent seen in pigs.

**Kiffer:** Those are rough figures we've worked out ourselves from the reported mean and standard deviation. With only twelve people the multiplier should be a little larger than 1.96, so a more exact interval is slightly wider still. Either way, stopping for futility means the study was unlikely to reach significance at its planned size. On its own, it can't show that the true change was zero.

**Sarah:** Then there's the setting result. Ten of the fifteen were dosed in a calm room with curated music, and five inside the MRI scanner. The setting wasn't randomized. Participants one to five were dosed in the scanner and participants six to fifteen in the room. The room group showed increases of about five to six percent.

**Kiffer:** And the room group's own increases had confidence intervals that crossed zero, in a small subgroup, from an exploratory analysis. The authors call the setting differences preliminary and hypothesis-generating, and Lesson five takes up setting properly. One more design point cuts the other way. There was no placebo arm, and the authors note that two earlier studies in healthy people found SV2A about one to eight percent lower on a second scan, so drift between scans could have hidden part of an increase.

**Sarah:** Question four. When people say MDMA helps people with PTSD reprocess traumatic memories, is that extinction or reconsolidation, and how would anyone tell?

**Kiffer:** The two predict different things about the return of fear. Extinction builds a new safety memory that competes with the old one, so fear can come back with time, in a new context or after the feared event happens again. Reconsolidation would change the original memory, so fear should stay down under those same tests. Extinction is well established in people, and a critical review in 2018 described human reconsolidation as viable but hotly contested. None of the human MDMA laboratory studies from the lesson tested reconsolidation.

**Sarah:** And the extinction studies in people are mixed. In mice, MDMA before extinction training gave a lasting improvement that depended on BDNF in the amygdala. In healthy adults, Maples-Keller and colleagues randomized thirty-four people and found no group difference in retention on their main analysis. Vizeli and colleagues gave thirty healthy men MDMA and placebo and found smaller skin conductance responses during extinction and at recall, with no change in the startle reflex.

**Kiffer:** So the result depends on the measure, and the retention measure has its own problem. Lonsdorf and colleagues found sixteen versions of the extinction retention index in the literature, and across four independent datasets the correlations among them varied widely. Let's run one hypothetical participant through two versions, with made-up numbers. Her largest response to the CS plus, the cue paired with the shock, was 0.90 microsiemens during conditioning. On the first two recall trials, her responses to it were 0.45 and 0.39. What's her extinction retention index on the CS plus version? Take a few seconds.

*(Pause)*

**Sarah:** The mean recall response is 0.45 plus 0.39, divided by two, which is 0.42. Then 0.42 divided by 0.90 is about forty-seven percent of the conditioning maximum, and one hundred minus forty-seven gives about fifty-three percent. So she kept about half of her extinction learning.

**Kiffer:** Now a second version, which subtracts her response to the CS minus, a cue that was never paired with the shock. On the conditioning trial where her CS plus response peaked, her CS minus response was 0.30, so the difference was 0.60. On the two recall trials, her CS minus responses were 0.30 and 0.26, which average 0.28.

**Sarah:** So the recall difference is 0.42 minus 0.28, which is 0.14, and 0.14 divided by 0.60 is about twenty-three percent. One hundred minus twenty-three gives about seventy-seven percent. So the same person in the same session goes from keeping about half her extinction learning to keeping about three quarters.

**Kiffer:** The two versions count different things. Part of her recall response may have been general arousal, since she reacted to the safe cue as well. The first version counts that arousal as fear, and the second removes it. Each is defensible, so a study should name its index before collecting data and report the others as secondary. When I read that MDMA improved extinction retention, I ask which index was used and whether the result holds on another.

**Sarah:** And the clinical hypothesis came from authors with ties to the developer of MDMA-assisted therapy.

**Kiffer:** It did. Feduccia was affiliated with MAPS Public Benefit Corporation, and Mithoefer trained and supervised therapists for its phase 3 trial. A conflict of interest leaves a hypothesis exactly as testable as it was, and it's a good reason to look for tests by independent groups.

**Sarah:** Okay. Now the part where we disagree, which takes us back to the PET study. How much should it change the plasticity story?

**Kiffer:** Let me hear your case first, since you feel more strongly about it than I do.

**Sarah:** I think it should change it a lot. This was a direct human test of what the animal work predicted. It was pre-registered, it was planned to detect a five percent increase, which is smaller than the pigs showed, and it measured the same marker as the pig study. It came back flat overall and stopped for futility. Any consent form or clinic page that still says psilocybin works by growing new connections is now indefensible, in my view. I'd go further and say the field should stop leading with plasticity when it explains why these treatments might help.

**Kiffer:** I agree about the consent form, and I'd have said that before this study. Where I differ is on how far one small null result should move the science. The interval we worked out reaches about fourteen percent. There was no placebo arm, and second scans came out lower in two earlier studies. And it measured one marker in healthy volunteers, one week after a single dose. I read it as an imprecise test that makes a large increase unlikely and leaves a pig-sized one open.

**Sarah:** But that argument can rescue almost any hypothesis, since every small null can be called imprecise. I'm also wary of the setting subgroup, because an exploratory split of ten and five is where a struggling hypothesis looks for help. And the clinical questions don't wait on it. When the federal government announced nearly three million dollars through the Canadian Institutes of Health Research in June 2023 for three trials of psilocybin-assisted psychotherapy, those trials were built to answer whether people improve, and they can answer it with no mechanism story at all.

**Kiffer:** On the subgroup, I'm with you, and the authors call it preliminary themselves. On rescue, it's fair to ask what would change my mind. I'd want a study in patients with a comparator arm, large enough that its interval excludes a six percent increase, to come back flat. I'd accept that as a real failure of the idea in people. Until something like that exists, I'd keep plasticity as a working hypothesis for research and describe it everywhere else as an animal finding.

**Sarah:** I'm more skeptical than Kiffer, but we agree on the practical rule. In anything a patient reads, describe plasticity as an animal finding that hasn't been shown in people, and read a human null result the way you'd read a positive one, by its interval and its design.

**Kiffer:** Let's pull it together with three things to take away.

**Sarah:** First, ask what was measured before you accept a direction. Relative and absolute blood flow, or an analysis with and without the global signal, can point opposite ways in the same data, and desynchronization describes the default mode network better than switching off.

**Kiffer:** Second, read every small-sample number by its interval. A correlation of 0.8 in seven people fits values from 0.12 to 0.97, the same fifteen people counted three times add no information, and on our own rough calculation, a null result in twelve people is still compatible with an increase of about fourteen percent.

**Sarah:** And third, state the level of evidence along with the claim. Spine growth is a finding in rodents and the synapse marker rose in pigs, a small PET study in healthy people found no overall rise, and even a procedure shared across species, like fear extinction, gives answers that depend on which index you choose.

**Kiffer:** If you'd like more practice, rework the extinction retention example with the CS minus recall responses raised to 0.35 and 0.33, and work out which version of the index changes and by how much.

**Sarah:** Next time, it's Lesson five, The Psychedelic Experience: Mystical States, Set and Setting, Expectancy and Blinding, where the setting question from the PET study comes back, along with the debate over whether the experience itself matters.

**Kiffer:** Take care, everyone.

**Sarah:** See you in Lesson five.
