Distributed neural signatures of discomfort induced by transcranial magnetic stimulation

By analyzing an unprecedented dataset of 165 participants using concurrent TMS-fMRI, this study identifies distributed, cross-validated neural signatures of TMS-induced discomfort across multiple brain networks, revealing that these discomfort-related responses account for a significant portion of TMS-evoked activity and must be explicitly modeled to refine causal inference and therapeutic applications.

The Gist

Imagine you are trying to listen to a specific instrument in an orchestra, like the violin, to understand how it contributes to the music. But there's a problem: every time the violinist plays, they accidentally knock over a heavy cymbal, creating a loud, jarring crash.

If you just record the sound, you hear the violin and the crash. You might think the crash is part of the violin's music, or you might get so distracted by the noise that you miss the melody entirely.

This is exactly the problem scientists face with Transcranial Magnetic Stimulation (TMS).

The Setup: The "Magic" Hammer

TMS is a powerful tool used by doctors and researchers to zap specific parts of the brain with magnetic pulses. It's like a "remote control" for the brain, used to figure out what different brain areas do or to treat depression and anxiety.

However, there's a catch. When the magnetic hammer hits the brain, it also hits the scalp, the muscles, and the nerves under the skin. This causes a sensation that feels like a sharp tap, a throb, or even pain. It's uncomfortable.

For years, researchers knew this "tap" was annoying, but they didn't know how the brain was reacting to that annoyance. They assumed the brain activity they saw was purely the result of the magnetic pulse changing the brain's function. But what if a huge chunk of that activity was actually just the brain screaming, "Ouch, that hurt!"?

The Big Experiment: Mapping the "Ouch"

The authors of this paper decided to solve this mystery. They didn't just zap one spot; they zapped 11 different spots on the brains of 165 people (some healthy, some with depression or anxiety) while they were inside an MRI machine.

Think of this as testing 11 different keys on a piano to see which ones make the most noise when you hit them.

1. The Participants: They had two groups: "Healthy" people and "Symptomatic" people (those feeling down or anxious).
2. The Rating: After every single zap, the participants gave a "pain score" from 0 (no feeling) to 100 (worst pain imaginable).
3. The Scan: While they were being zapped, the MRI took a picture of their entire brain to see which lights turned on.

The Discovery: It's Not Just One Spot

Using a sophisticated computer algorithm (think of it as a super-smart detective that can find hidden patterns in a massive pile of clues), they mapped out exactly which parts of the brain light up when someone feels that "TMS discomfort."

Here is what they found, using simple analogies:

• The "Body Map" (Sensorimotor Network): Just like when you stub your toe, the part of your brain that feels touch and pain lit up. This makes sense; the scalp was being tapped.
• The "Alarm System" (Limbic & Attention Networks): The brain didn't just feel the pain; it got worried about it. Areas that handle fear, attention, and emotional evaluation lit up. It's like your brain's internal alarm system going off: "Hey! Something is wrong! Focus on this!"
• The "Default Mode" (The Daydreaming Network): Even the parts of the brain we use when we are just daydreaming got involved. The discomfort was so strong it pulled the brain out of its normal rhythm.

The Twist: Two Different Reactions

Here is the most fascinating part. The "Ouch" reaction looked different depending on who was getting zapped.

• Healthy Brains: They reacted like a standard security system. "I feel a tap, I check it out, and I move on." They used more of their motor areas (planning how to move away from the pain).
• Symptomatic Brains (Depression/Anxiety): Their reaction was more intense and different. They paid more attention to the pain and seemed to link it to their memories and emotions more strongly. It's like if a healthy person hears a car backfire and thinks, "Oh, a car," but a person with anxiety hears it and thinks, "Oh no, is that a threat? Is it happening again?"

Crucially, even though their brains reacted very differently, the participants rated the pain as being about the same. This means our "pain meter" (what we say out loud) doesn't always match the "pain engine" (what's actually happening inside our heads).

The Big Takeaway: The "Noise" is Real

The researchers calculated that 12% to 25% of the brain activity seen during TMS was actually just the brain reacting to the discomfort, not the magnetic pulse itself.

Why does this matter?
Imagine you are trying to study how a specific lightbulb works, but every time you turn it on, a loud siren goes off. If you don't account for the siren, you might think the lightbulb causes the siren, or you might miss the lightbulb's true effect because the siren is so loud.

This paper tells us:

1. TMS isn't "clean": When we zap the brain, we are also zapping the pain system.
2. We need to subtract the "Ouch": To understand what TMS is really doing to treat depression or study the brain, scientists must mathematically remove the "discomfort signal" from their data.
3. One size doesn't fit all: The "discomfort signal" looks different in healthy people versus people with mental health conditions. We can't use the same correction for everyone.

The Bottom Line

This study is like a manual for cleaning up the signal. It tells scientists, "Hey, before you claim you found a new brain function, make sure you aren't just measuring how much the patient's scalp hurt." By acknowledging and measuring this discomfort, we can make TMS research more accurate, safer, and more effective for everyone.

Technical Summary

1. Problem Statement

Transcranial Magnetic Stimulation (TMS) is a critical tool for causal inference in neuroscience and a growing clinical therapy. However, TMS pulses induce scalp discomfort or pain by stimulating nociceptive fibers and muscles. This discomfort is a significant confound that can:

• Bias measured behavioral and neural outcomes.
• Increase dropout rates in research and clinical trials.
• Complicate the interpretation of TMS-evoked brain responses, as it is difficult to disentangle direct neuromodulatory effects from discomfort-driven neural activity.

Despite its ubiquity, the whole-brain neural signatures of TMS-induced discomfort remain poorly characterized. Previous studies were limited by:

• Small sample sizes and limited stimulation sites (often only 1–2 sites).
• Lack of comparison between healthy and clinical populations (e.g., depression/anxiety).
• Reliance on univariate analyses that fail to capture distributed, multivariate brain-behavior relationships.

2. Methodology

Study Design and Participants:

• Dataset: A large-scale concurrent TMS-fMRI study involving 165 participants (80 healthy, 85 with elevated depression/anxiety symptoms based on PHQ-9/GAD-7 scores).
• Protocol: Participants underwent up to 11 TMS-fMRI runs, each targeting a different cortical site (11 total targets: L/R Frontal Pole, L/R aMFG, L/R pMFG, R IFJ, R FEF, R preSMA, R M1, R IPL).
• Total Data: 1,535 TMS-fMRI runs.
• Stimulation: Single-pulse TMS delivered at 120% of resting motor threshold (or 100% machine output if no motor response) inside the MRI scanner.
• Behavioral Measure: Immediate post-run ratings of discomfort using the Subjective Units of Distress Scale (SUDS) (0–100).

Data Processing:

• Preprocessing: Standard fMRI preprocessing (motion correction, normalization to MNI space, smoothing) using fMRIPrep.
• Feature Extraction: TMS-evoked brain responses (beta weights) were extracted from 246 Regions of Interest (ROIs) defined by the Brainnetome atlas.
• Quality Control: Runs with excessive motion (>4mm RMS) were excluded.

Computational Framework:
The authors employed a Principal Component Analysis combined with Sparse Canonical Correlation Analysis (PCA-sCCA) framework to model the relationship between high-dimensional brain responses and SUDS ratings.

• PCA: Reduced the dimensionality of the 246-ROI brain data to capture dominant variance (retaining $k$ principal components).
• sCCA: Identified a linear weighted combination of PCs that maximally correlated with SUDS ratings, imposing sparsity ($\lambda$) to improve interpretability and generalizability.
• Validation: Model performance was optimized using 10 rounds of 10-fold cross-validation. Hyperparameters ($k$ and $\lambda$) were tuned separately for healthy and symptomatic groups.
• Comparison: PCA-sCCA was validated against alternatives (sCCA alone, sPLS, PCA-sPLS), demonstrating superior generalizability and stability.

3. Key Contributions

1. First Brain-Wide Mapping: Provided the first comprehensive, whole-brain characterization of neural signatures associated with TMS-induced discomfort across 11 cortical targets.
2. Multivariate Approach: Established a robust PCA-sCCA framework capable of capturing distributed, joint patterns of brain activity linked to subjective discomfort, overcoming limitations of univariate ROI-based analyses.
3. Population Specificity: Demonstrated that neural signatures of discomfort are group-specific, revealing distinct patterns in healthy individuals versus those with elevated affective symptoms.
4. Quantification of Confounds: Quantified the proportion of TMS-evoked brain responses that are attributable to discomfort, providing a metric for researchers to control for this confound.

4. Key Results

Behavioral Findings:

• Site Variability: Discomfort varied significantly by stimulation site. Frontal regions (especially Frontal Pole) induced the highest discomfort, while posterior regions (M1, IPL) induced the lowest.
• Group Variability: Despite differences in neural processing, no significant difference in subjective SUDS ratings was found between healthy and symptomatic groups.
• Individual Variability: Substantial inter-individual variability existed within each site (e.g., at the Frontal Pole, ratings ranged from <20 to >65).

Neural Signatures (PCA-sCCA Results):

• Distributed Networks: Discomfort-related activity engaged distributed networks including sensorimotor, attentional, limbic, and default mode networks.
• Key ROIs:
  • Shared: Somatosensory areas (Postcentral Gyrus), Caudate Nucleus.
  • Healthy Group: Greater engagement of motor/premotor areas (PreCG, PMv), visual sensory regions, and medial frontal/limbic regions (mPFC, dACC).
  • Symptomatic Group: Greater engagement of somatosensory-associative regions (SMG), attentional networks (Anterior Insula), and memory-associated regions (Parahippocampal Gyrus, aSTG).
• Group Specificity: Cross-group removal analyses confirmed that the identified signatures are largely non-interchangeable; removing the healthy group's signature did not eliminate the brain-discomfort correlation in the symptomatic group, and vice versa.

Quantification of Discomfort Influence:

• Healthy Group: Discomfort-related activity accounted for ~12.3% of TMS-evoked responses in top ROIs.
• Symptomatic Group: Discomfort-related activity accounted for ~25.2% of TMS-evoked responses in top ROIs.
• Site Dependence: Prefrontal stimulation sites (e.g., Frontal Pole) showed the highest proportion of discomfort-related responses (up to ~33% in symptomatic groups), while motor sites showed the lowest.

5. Significance and Implications

• Refining Causal Inference: The study reveals that a substantial portion of TMS-evoked brain activity (especially in prefrontal targets) may reflect discomfort processing rather than specific neuromodulatory effects. This challenges previous interpretations of TMS-fMRI studies where ACC or insula activation was attributed solely to network modulation.
• Clinical and Research Recommendations:
  • Researchers must routinely assess and report TMS-induced discomfort (e.g., SUDS).
  • Analytic models should explicitly control for discomfort, either by treating ratings as covariates or by regressing out the identified discomfort-related neural signatures.
  • Special caution is needed when interpreting TMS effects in prefrontal regions and in clinical populations (depression/anxiety), where discomfort-related neural processing is more pronounced.
• Future Directions: The identified signatures provide a benchmark for future studies to disentangle therapeutic neuromodulation from pain/discomfort artifacts, improving the translational validity of TMS therapies.

In conclusion, this paper establishes that TMS-induced discomfort is a complex, distributed neural phenomenon that varies by stimulation site and population. Ignoring these signatures risks misinterpreting the mechanisms of TMS, necessitating a paradigm shift toward explicitly modeling and controlling for discomfort in TMS research.