Pruning Season
What's actually happening in an adolescent brain, with citations.
8 to 10 minAbout today
Today we're covering five things: synaptic pruning, the social pain network, dopamine as prediction error, why risk-taking is adaptive, and the glymphatic system. Each one is a real mechanism with a real lab behind it. If you don't know those terms yet, you will by the end of the hour.
Then we'll play an evidence game. You'll see a claim and decide whether to commit or unlock clues first. Clues cost points. So does being wrong. That trade-off is how clinicians actually think, and it's what separates real knowledge from pop-psychology confidence. Today is the actual research, with citations.
What's shared here stays here. Names, stories, details, all of it is protected. The only exception is safety.
You can pass at any point. No explanation needed. If something doesn't fit right now, say "pass" and we move on.
10 to 12 minBridge questions
Share your name, then go through these in order. You don't have to give details. Just notice what comes up.
15 to 18 minFive things your brain is doing right now
Each of these is a real mechanism with a real lab behind it. Before you open one, take a second and predict the answer. Prediction is not extra, it's the mechanism that makes the reveal land. That's literally one of the five things we'll cover.
The brain deletes roughly 40% of its excitatory synapses in the prefrontal cortex across adolescence and into the mid-twenties. This is synaptic pruning.
The rule is Hebbian: synapses that fire together strengthen through long-term potentiation, and quiet ones get tagged by microglia and eliminated. Use it or lose it, at the cellular level.
Sarah-Jayne Blakemore's lab at UCL established the prefrontal pruning timeline. The landing: the Spanish you practiced at eleven, the piano you quit at twelve, the social reflexes you rehearse every night on your phone, whatever you are currently doing is a vote on which circuits survive into your thirties.
Not metaphor. Actual cellular-level editing. You are choosing what to keep.
In 2003, Naomi Eisenberger, Matthew Lieberman, and Kipling Williams at UCLA put people in an fMRI scanner and had them play a virtual game of catch called Cyberball. Partway through, the other "players" (actually a script) stopped throwing to the participant.
Being excluded from a meaningless video game activated the dorsal anterior cingulate cortex and anterior insula, the same regions that fire when you physically hurt. Later work by Ethan Kross (2011) found the overlap extends into somatosensory cortex.
When being left out of the group chat feels like being kicked, that is not drama or melodrama. It is shared neuroanatomy. The alarm system evolved to run exclusion and injury through the same circuits because, for most of human history, being exiled from the group was a death sentence.
The signal is calibrated to a prehistoric stake. That doesn't make the stake real now. It makes the signal real now. There's a difference.
Wolfram Schultz's monkey studies, starting in the 1990s, changed what we thought dopamine was for. He found that dopamine neurons fire hardest for unexpected reward. They fall silent for predicted reward. They drop below baseline when a predicted reward fails to arrive.
This is reward prediction error. Dopamine is not the pleasure chemical. It is the "was that better or worse than I expected" chemical.
Why the first bite is the best one: prediction error collapses once your brain has calibrated. Why scrolling, slot machines, and loot boxes are engineered around variable-ratio schedules: unpredictable reward keeps the error signal firing. Why dopamine does not equal happiness: chasing the signal leaves you wanting more of what you just had while enjoying it less.
Addiction is a learning disorder of prediction, not a pleasure disorder. Pop-psychology "dopamine hits" misses the whole mechanism.
Laurence Steinberg (Temple) and B.J. Casey (formerly Cornell, now Yale) independently proposed the dual systems model: the socioemotional and reward system (ventral striatum, limbic) peaks in sensitivity around age 14 to 16. The cognitive control system (dorsolateral prefrontal cortex) matures on a slower track into the mid-twenties.
This imbalance is not a defect. Adrena Galvan's fMRI work shows the teen ventral striatum responds harder to reward than either children's or adults'. The argument is evolutionary: a brain that wouldn't leave the nest, try unfamiliar food, pursue a partner outside the family, or take the social risk of a new identity would be a dead end.
The imbalance is not a defect. The environment evolution calibrated for (small groups, low speed, no firearms, no algorithms) is not the one any adolescent is navigating.
Steinberg's driving-simulator studies also found that peer presence doubles risk-taking in teens, but not in adults. The ventral striatum fires harder in teens when peers are watching. This is not peer pressure in the moralized sense. It's a neurobiological shift in how reward gets weighted.
Two mechanisms. First, Mary Carskadon's sleep lab at Brown showed that puberty delays melatonin onset by roughly two hours. Adolescent biology runs on a timezone two hours west of the school bell. A 7 AM alarm at 16 is biologically equivalent to a 5 AM alarm at 30.
Second, in 2012, Maiken Nedergaard's lab at Rochester discovered the glymphatic system. During NREM sleep, cerebrospinal fluid flushes through the spaces between brain cells via aquaporin-4 channels on astrocyte endfeet, clearing metabolic waste including beta-amyloid. The system runs at roughly 10% efficiency while you're awake.
Translation: sleep is not optional downtime. It is active neural maintenance. Miss it, and the metabolic trash does not get taken out. You feel it the next day because there is literally more junk between your neurons.
The American Academy of Pediatrics recommends high schools start no earlier than 8:30 AM. Most do not. That is a policy failure, not a personal failure.
Neuroplasticity does not end at 25. Plasticity changes form across the lifespan, but the adolescent version is faster and more dramatic than anything that comes after. What you practice now consolidates harder than what you practice later. Which is either a warning or an invitation, depending on what you're practicing.
15 to 18 minHow clinicians actually think
Each round shows a claim. You can commit now, or unlock clues first. Every clue costs 1 point. Right answer with zero clues is worth 3. With three clues, it's worth 0. Wrong is wrong.
That trade-off is the whole game. Experts don't guess and don't demand certainty either. They pay for evidence in proportion to the stakes. This is clinical reasoning as a mechanic.
10 to 12 minOpen it up
Take these wherever the room goes. You do not have to answer every one.
If someone brings up something real, sit with it. These mechanisms explain things that have felt personal and confusing for years. That lands differently than abstract discussion.
5 minPick one
One experiment between now and next session. Not all of them. Pick the one that hooked your attention most, and commit to noticing specifically.
One sentence
Before you go, one takeaway and one specific plan. Specific means naming when, where, and what will be different. "I'll try to sleep more" is not a plan. "Thursday night I'm putting my phone in the kitchen at 10 and going to bed" is a plan.
Push for specificity. Vague plans do not reinforce. If a teen says "I'll pay more attention to my feelings," reflect back the vagueness and ask for a when, where, and what. That refusal to accept the first version is itself a form of respect.