TMS for depression only works half the time: the bottleneck is we don't know how it changes the brain
TMS is in the clinic but still a black box; unclear mechanisms keep efficacy at 50%. Using intracranial electrodes in epilepsy patients to get ground truth from neurons, the next step is to refine EEG biomarkers, turning treatment from parameter guessing into closed-loop optimization.
The video won't play here. Listen to the audio instead:
The argument · timestamps estimated from transcript position
Efficacy is capped by unclear mechanisms, with only about 50% of patients showing significant improvement
Even with the most advanced protocols (at one-month follow-up), TMS response rate is about 50%, meaning roughly half of patients see symptoms reduced by about half. This number is striking because standard treatment lasts 6 weeks, and so-called accelerated protocols only bring response back to the same level. The problem is not that pulses fail to penetrate, but that we don't know how it changes the brain at synaptic and network levels, making it impossible to predict who will benefit or to adjust dose or target based on intermediate states.
— Corey KellerThe 80% remission rate is statistically inflated and not durable
Corey directly clarified the 80% figure from Nolan Williams' SAINT protocol: it comes from meeting the criteria for no longer meeting depression criteria at any assessment time point during the five-day course, not the endpoint remission rate at day five. What matters is the one-month follow-up, where treatment-resistant patients still show only about 50% response. The key word is 'not durable' — if TMS is to become a treatment that changes life trajectories, it must make synaptic plasticity changes last beyond a month.
— Corey KellerUsing intracranial electrodes in epilepsy patients to measure TMS's real effects
The core move of Corey's lab is to place microelectrodes in deep brain regions while epilepsy patients undergo preoperative intracranial monitoring, and simultaneously deliver single-pulse TMS from outside the skull. This yields ground truth at the neuronal level in humans: which neurons are activated or inhibited in the tens to hundreds of milliseconds after stimulation. He also used this to correct the view that TMS only reaches superficial layers, showing it can spread rapidly through networks across the whole brain.
— Nicholas WeilerSuppressing emotional brain regions may be an indirect pathway via the thalamus
Key finding: after a single TMS pulse to the dorsolateral prefrontal cortex, the first neurons activated are not the expected limbic regions (like sgACC), but inhibitory neurons in deep thalamo-striatal circuits; about 300 ms later, excitatory neurons in limbic regions are suppressed. That is, theoretically suppressing sgACC via DLPFC may not be a direct connection but an indirect pathway relayed through the thalamus. Corey stresses this is not yet causal — the sample is small, patients have epilepsy, and only single pulses were tested; further validation is needed.
— Corey KellerTranslating deep signals into scalp EEG via a 'transfer function'
They also place EEG on epilepsy patients' heads, so they can capture the correspondence between deep neuronal activity and scalp signals at the same moment — what Corey calls a transfer function. The long-term goal is to read from scalp EEG, without opening the skull, whether a specific deep network has been activated by TMS and whether plasticity has been induced. If this works, EEG could be upgraded from a research tool to a real-time, single-person biomarker in the clinic, rather than relying on group averages of 30 people.
— Corey KellerRobotic closed-loop noise reduction is necessary for single-subject measurement
TMS-evoked EEG signals are swamped by artifacts: the loud click of capacitor discharge, scalp discomfort, and auditory and somatosensory inputs all produce neural responses. To separate 'true cortical excitability' from noise, they built an automated feedback system: a robotic arm adjusts coil angle, position, and intensity in real time to maximize target signal and minimize artifacts, making single-session, single-person cortical excitability measurement stable within 5 minutes. This paves the way for future personalized treatment where each patient is measured once.
— Corey KellerTreating patients when their brain is not in an optimal state wastes TMS
In the clinic, patients often daydream, watch Netflix, or fall asleep, but basic research has long known that neurons must be in an activated state during stimulation for maximal plasticity. Keller's team is testing three approaches: different types of music (Jessica Ross found that applying TMS just before a beat enhances responses), applying TMS during specific sleep stages (sleep is the strongest plasticity window), and cognitive task loading — all aimed at putting the brain in the state most receptive to being shaped.
— Corey KellerThe future is a 'stimulation parameter trial run' before formal treatment
Their vision for five years out: first do an fMRI to localize the target, then spend a 'stimulation screening day' rapidly testing multiple parameters (different brain states, music, pulse patterns, intensities, etc.), measuring the direction of cortical excitability changes in real time to find the most effective protocol for that patient, then start the course; adjustments are made mid-course. This applies not only to TMS but could also evaluate any intervention's effect on brain plasticity, such as drugs, ultrasound, or ECT — essentially turning 'parameter guessing' into 'data-driven parameter tuning'.
— Corey KellerIn their own words · checked verbatim
Without that mechanism, I think there's a lot that we're missing out in terms of the clinical efficacy. TMS, even for the newest treatments a month after the treatment, only has about a 50% clinical response. That is 50% of our patients show about a 50% reduction in their depressive symptoms.
Corey Keller4:33
Meaning when you assess every single day at one point in their treatment course, they did not meet criteria for depression. So it was very effective, but not necessarily at the end of treatment.
Corey Keller7:11
This idea that TMS is only superficial. Yes, TMS only gets to that superficial region, but then it propagates throughout the brain very quickly. And even within 10 milliseconds, the thalamus, a very deep region becomes activated. Single neurons fire.
Corey Keller25:05
There's just a lot of parameters that we really haven't dove into, mostly because we haven't had the tools to measure it in such an acute way. Most people do the full TMS treatment and take a picture before and after using fMRI. And it's hard to understand that black box of each single pulse, each single dose and how they all interact. Once we get down to that level, we can do some pretty interesting neuroscience.
Corey Keller40:29
Figures
| TMS clinical response rate at one month | about 50% (about half of patients see symptoms reduced by at least half) | 4:33 |
| SAINT five-day accelerated protocol's apparent 80% statistical measure | any assessment time point during the course, not necessarily the endpoint | 7:11 |
| Number of pulses in one standard theta burst stimulation | 600 pulses in 3 minutes | 14:15 |
| Time window for cortical signal appearance | biphasic deflection 20–50 ms after stimulation | 31:27 |
| Time when deep brain regions are first activated | thalamus begins firing within 10 ms | 25:05 |
| Time when limbic regions are activated | about 300 ms later | 26:27 |
| Time to optimize single-site cortical excitability measurement | within about 5 minutes | 34:54 |
Glossary
- TMS / transcranial magnetic stimulation
- A non-invasive stimulation technique that uses a coil to generate a time-varying magnetic field, which induces electric currents in the brain through the skull to activate neurons.
- theta burst stimulation
- A protocol that delivers 600 pulses in 3 minutes with a specific timing pattern, thought to induce synaptic plasticity.
- cortical excitability
- The strength of a given cortical region's response to a fixed stimulus; often measured with TMS-EEG early evoked potentials.
- limbic system
- A set of brain regions involved in emotion, memory, and motivation, including the amygdala, hippocampus, and sgACC.
- subgenual cingulate (sgACC)
- A brain region located deep beneath the prefrontal cortex; its overactivity is thought to be closely linked to major depression.
How to listen
Researchers in psychiatry, neuromodulation, or EEG/fMRI translation, and clinicians interested in moving brain stimulation from empirical tuning to data-driven approaches.
A historical review section in the middle (about 10:14–13:16) has lower information density and can be skipped.