Personalizing neuromodulation for chronic pain: A connectivity-guided randomized trial

In a randomized controlled trial of 90 patients with chronic pain, connectivity-guided target selection based on global cortical connectivity failed to improve rTMS outcomes compared to classic M1 stimulation, though lower local M1 connectivity was identified as a potential biomarker predicting better response to the classic approach.

The Gist

The Big Picture: Trying to Fix a Broken Radio

Imagine your brain is a massive, complex radio station. When you have chronic pain, it's like the station is broadcasting a loud, annoying static noise that never turns off.

Doctors have a tool called rTMS (repetitive transcranial magnetic stimulation). Think of this as a "signal booster" or a "tuning fork" that doctors can hold against your head to try to clear up the static and restore the music.

For years, doctors have known that hitting a specific spot on the radio (the Primary Motor Cortex, or M1) usually helps about half the people. But for the other half, it doesn't work at all. It's a bit of a "guess-and-check" game.

The Big Question: Can we stop guessing? Can we look at a person's brain before we start, find the exact spot that is most "broken" or "quiet," and tune that specific spot to fix the pain?

The Experiment: The "Connectivity Map"

The researchers in this study wanted to test a new, high-tech strategy. They believed that every person's brain has a unique "map" of how different parts talk to each other (connectivity).

They hypothesized that the best place to fix the radio wasn't always the same spot for everyone. Instead, they thought:

• The "Low-Connectivity" Theory: If a part of the brain is very quiet and disconnected from the rest, maybe that's the "weak link" we need to boost.
• The "High-Connectivity" Theory: Conversely, maybe the part that is too loud and connected is the problem, and we need to calm it down.

How they did it:

1. They took 90 patients with chronic pain.
2. Before any treatment, they used a special scanner (TMS-EEG) to zap four different spots on the brain and measure how much each spot "talked" to the rest of the brain.
3. They split the patients into three groups:
   • Group A (The Low-Connectivity Group): Got treated on the spot that was the quietest and most disconnected.
   • Group B (The High-Connectivity Group): Got treated on the spot that was the loudest and most connected.
   • Group C (The Classic Group): Got treated on the standard "M1" spot, ignoring the map entirely (the usual way doctors do it).

The Results: The Map Didn't Help

After 8 weeks of treatment, the researchers checked the results.

The Bad News:
The fancy "Connectivity Map" didn't work.

• Group A (Low-Connectivity target) did not feel better than Group C (Classic target).
• Group B (High-Connectivity target) did not feel better either.
• In fact, all three groups improved by about the same amount (roughly 30–40% pain reduction).

It turns out that just finding the "quietest" or "loudest" spot on the map didn't predict who would get better. The "personalized" approach didn't beat the standard "one-size-fits-all" approach.

The Good News (The Hidden Treasure):
While the big map failed, the researchers found a tiny, hidden clue when they looked closer at the data.

They noticed something interesting specifically in the Classic Group (the people who just got the standard treatment on the M1 spot):

• If a patient's M1 spot was locally quiet (meaning the tiny neighborhood right around the stimulation site wasn't talking much to itself), they tended to get much better.
• If that same spot was already loud and chatty, the treatment didn't help as much.

The Analogy:
Think of the brain like a crowded room.

• The Failed Strategy: The researchers tried to find the person in the room who was either the loudest or the quietest and shouted at them to fix the noise. It didn't work.
• The Hidden Clue: They realized that if the specific corner of the room where they were shouting was already a bit empty and quiet, the shout echoed nicely and fixed the problem. But if that corner was already packed with people shouting, adding more noise didn't help.

What Does This Mean for the Future?

1. Personalization is hard: We can't just look at a global "map" of brain connections and say, "Okay, treat this specific spot." It's more complicated than that.
2. Local matters more: The study suggests that for the standard treatment to work, the specific little neighborhood of the brain being treated needs to be in a certain "state" (a bit quiet/desynchronized) before we start.
3. Next Steps: Future doctors might be able to use a simple test to check if your "M1 neighborhood" is ready for treatment. If it is, they treat you. If not, maybe they try a different approach.

In a nutshell: The researchers tried to use a GPS to find the perfect spot to fix chronic pain, but the GPS led them to the same place everyone else goes. However, they discovered that the condition of the road at that specific spot matters more than the location itself.

Technical Summary

1. Problem Statement

Chronic pain affects a vast number of individuals, yet current therapeutic management relies heavily on a "trial-and-error" approach due to a lack of methods to individualize treatments. Repetitive transcranial magnetic stimulation (rTMS) is a promising non-invasive intervention, particularly when targeting the primary motor cortex (M1), but response rates remain modest (35–50%).

• The Gap: There is substantial inter-individual variability in pain processing and cortical network dynamics. Current protocols typically apply a "one-size-fits-all" target (usually M1) or select alternative targets (DLPFC, ACC, PSI) without considering the specific neurophysiological state of the patient's brain.
• The Hypothesis: The authors proposed that pre-therapy cortical connectivity, measured via TMS-evoked EEG (TMS-EEG), could serve as a biomarker to guide target selection. Specifically, they hypothesized that stimulating the cortical site with the lowest pre-therapy global connectivity would yield superior analgesic effects compared to stimulating the site with the highest connectivity or the standard M1 target, based on principles of homeostatic plasticity (i.e., regions with lower connectivity have greater capacity for modulation).

2. Methodology

Study Design:

• Type: Randomized, double-blind, controlled, three-arm parallel-group trial.
• Duration: 8 weeks (12 sessions total: 5 induction sessions over 1 week, followed by 7 maintenance sessions).
• Participants: 90 patients with chronic pain (pain >3 months, NRS 3–9). Diagnoses included neuropathic, nociceptive, and nociplastic pain.
• Groups:
  1. Low-Connectivity Group: rTMS applied to the cortical target (M1, DLPFC, ACC, or PSI) with the lowest pre-therapy global connectivity.
  2. High-Connectivity Group: rTMS applied to the target with the highest pre-therapy global connectivity.
  3. Classic-M1 Group: rTMS applied to M1 regardless of connectivity measures (standard of care).

Neurophysiological Assessment (TMS-EEG):

• Targets: Four cortical regions were mapped: Primary Motor Cortex (M1), Dorsolateral Prefrontal Cortex (DLPFC), Anterior Cingulate Cortex (ACC), and Posterosuperior Insula (PSI).
• Connectivity Metric: Global connectivity was quantified using a distance-weighted, debiased weighted Phase Lag Index (wPLI). This metric was derived from TMS-evoked potentials (TEPs) recorded from 63 EEG channels. It captured the magnitude and spatial extent of TMS-induced oscillatory phase locking (8–35 Hz) across the cortex.
• Target Selection: For the Low/High groups, the target was dynamically assigned based on which of the four sites exhibited the extreme global connectivity value for that specific patient.

Intervention:

• Protocol: 10-Hz rTMS, 30 trains of 10 seconds (3000 pulses total per session).
• Intensity: Adjusted per target (90% or 110% of resting motor threshold).
• Blinding: Double-blind (patients and assessors); blinding efficacy was tested at the end of the trial.

Outcomes:

• Primary: Proportion of patients achieving ≥30% reduction in Visual Analog Scale (VAS) pain intensity at Week 8.
• Secondary: ≥50% pain reduction, pain interference, sleep, fatigue, mood (HADS), quality of life (EQ-5D), and Patient/Operator Global Impression of Change (PGIC/OGIC).
• Exploratory: Correlation between local connectivity (within the stimulation cluster) and treatment response.

3. Key Results

Primary and Secondary Outcomes:

• No Significant Difference: There were no statistically significant differences between the three groups for the primary outcome (≥30% pain reduction) or any secondary outcomes (p > 0.05).
  • Responder Rates (≥30% reduction): Low-Connectivity (48%), High-Connectivity (50%), Classic-M1 (37%).
  • Responder Rates (≥50% reduction): Low-Connectivity (41%), High-Connectivity (37%), Classic-M1 (33%).
• Safety: rTMS was well-tolerated across all groups with no serious adverse events. Side effects (headache, dizziness) were mild and transient.
• Blinding: Blinding was largely successful; only a small minority of participants or therapists correctly guessed the group allocation.

Exploratory Analyses (The Critical Finding):

• Global vs. Local Connectivity: While the global connectivity index failed to predict group-level outcomes, local connectivity within the stimulated region showed a significant association with treatment response, but only in the Classic-M1 group.
• Correlation: In the Classic-M1 group, lower pre-therapy local connectivity (within the M1 EEG cluster) was significantly associated with greater pain reduction (r = 0.50, p = 0.005).
• Interaction: This relationship was not observed in the Low- or High-Connectivity groups. A regression model confirmed a significant interaction between group and local connectivity (p = 0.038), suggesting that the predictive value of local connectivity is specific to the M1 target or the standard protocol.
• Specificity: The connectivity index was not driven by local cortical excitability (Local Mean Field Power), confirming it measured functional synchronization.

4. Key Contributions

1. First Large-Scale Connectivity-Guided Trial: This is the first randomized controlled trial (RCT) to prospectively use TMS-EEG-derived global connectivity to stratify patients and guide rTMS target selection in chronic pain.
2. Refutation of Global Connectivity Hypothesis: The study provides robust evidence that a "global" connectivity metric (measuring network-wide propagation) is insufficient for optimizing rTMS target selection across different cortical regions.
3. Identification of a Local Biomarker: The study discovered that local phase-based connectivity within the M1 region is a potential biomarker for treatment response. Specifically, patients with lower local synchronization in M1 prior to treatment are more likely to respond to M1-rTMS.
4. Trans-Etiological Efficacy: The study demonstrated that rTMS efficacy (approx. 35–50% response rate) is consistent across diverse pain mechanisms (neuropathic, nociceptive, nociplastic), suggesting that cortical network properties may be more predictive of response than the specific pain etiology.

5. Significance and Future Directions

• Clinical Implication: The "connectivity-guided" strategy of selecting the target with the lowest global connectivity does not improve outcomes over standard M1 stimulation. Clinicians should not currently adopt this specific global metric for target selection.
• Precision Medicine Shift: The findings suggest that precision medicine in neuromodulation may need to shift from global network selection to local network characterization. The data supports the hypothesis that M1-rTMS is most effective in patients with a "desynchronized" local M1 network state, which may indicate a higher capacity for plasticity.
• Future Research: The authors recommend future prospective trials that stratify patients based on local M1 connectivity rather than global connectivity. This could significantly improve responder rates for M1-rTMS, which remains the gold standard for chronic pain neuromodulation.
• Limitations: The study lacked a sham control (though responder rates matched historical active controls), and the global connectivity metric may have inherent biases across different anatomical targets (e.g., DLPFC vs. PSI).

Conclusion:
While the specific strategy of using global connectivity to select the stimulation target failed to outperform standard care, the trial successfully identified local M1 connectivity as a promising biomarker for predicting individual response to M1-rTMS. This moves the field closer to truly personalized neuromodulation by identifying who responds to M1 stimulation based on their specific neurophysiological state.