Using causal network mapping to clarify pre-clinical brain stimulation results

We read with interest Teferi et al.’s [1] double-blinded, sham-controlled, cross-over trial in which intermittent theta-burst stimulation (iTBS) to individualized right dorsolateral prefrontal cortex targets unexpectedly increased laboratory anxiety [1]. This was particularly surprising given that prior results showed the same effect using continuous TBS (cTBS) [2], which is believed to affect brain circuits in the opposite way. We hypothesized that incidental intersection with our group’s causal anxiosomatic circuit could clarify these findings.

We calculated TMS site connectivity to the anxiosomatic circuit by modeling sites as concentric spheres of decaying intensity, generating sphere-site seed-maps using a normative connectome [3], and correlating seed-maps with the anxiosomatic circuit. This method was used to derive the anxiosomatic circuit [4] and has been validated by our group [5]. We used linear models to test relationships between anxiosomatic circuit connectivity and changes in anxiety.

As hypothesized, iTBS sites with more positive connectivity to the anxiosomatic circuit reduced anxiety potentiated startle (APS) more than sham (t = −3.59, permuted p = 0.003) (Fig. 1a). This relationship was detected even though most sites were negatively connected to the circuit (mean r = −0.21, sd = 0.29), limiting variance that could be explained in anxiety. Effects were frequency-specific (t = 2.32, permuted p = 0.03), with more positive connectivity predicting decreased APS following iTBS and increased APS following cTBS. However, cTBS did not differ from sham (Fig. 1b). Of note, sham sites with more negative circuit connectivity were associated with greater APS reductions in both studies. Results were unchanged when controlling for condition order, sex, age, baseline anxiety, and baseline working memory – or when rank transforming APS values (Supplementary Materials).

Fig. 1. TMS site connectivity to the anxiosomatic circuit shapes anxiety potentiated startle.

The x-axis depicts TMS site connectivity to the anxiosomatic circuit (Fisher z corrected r-values); the y-axis depicts anxiety potentiated startle (APS) changes from baseline. a iTBS condition (active/sham) interacted with anxiosomatic circuit connectivity to predict APS changes. b cTBS condition (active/sham) did not interact with anxiosomatic circuit connectivity to predict APS changes. Active stimulation frequency (iTBS/cTBS) interacted with anxiosomatic circuit connectivity to predict APS changes.

There are a few limitations. First, TMS pain is a potential confounder. However, APS was measured 24-hours after TMS, making this less likely [6]. Second, insufficient fMRI data were collected to generate reliable individual connectivity results, so a normative connectome was used as in prior work [4, 5]. Future studies using individual connectivity are necessary to validate the present results. Third, targets were not random but were localized to working memory peaks, which were generally negatively connected to the anxiosomatic circuit. More negative correlations were associated with APS decreases with sham, suggesting that stimulation sites were incidentally chosen at a working memory circuit that predicted anxiety learning in the absence of stimulation. This relationship was reversed by iTBS but not enhanced by cTBS, perhaps because it was in the same direction, potentially limiting additional cTBS effects. Fourth, it is possible that TMS effects spread to positive and negative portions of the anxiosomatic network and that this was unaccounted for by our TMS site seeding approach. Future studies should validate electric-field modeling for seeding TMS sites to better account for network bleed-over.

Our results suggest that Teferi et al.’s unexpected findings were driven partly by incidental intersection of iTBS sites with our causal anxiosomatic circuit. Future preclinical studies could prospectively target positive and negative circuit targets with different stimulation frequencies during different brain states to refine anxiosomatic protocols prior to clinical trials.

Author contributions

CRediT author statement according to: https://www.elsevier.com/authors/policies-and-guidelines/credit-author-statement. Conceptualization: RDW, SHS. Methodology: RDW, SHS. Data collection: NLB, Formal analysis: RDW. Writing - Original Draft: RDW, SHS. Writing - Review & Editing: DK, NLB.

Funding

SHS is primarily supported by public funds from the National Institute of Mental Health, but also has foundation funds from The Brain & Behavior Research Foundation and the Baszucki Family Foundation, and investigator-initiated grants from BrainsWay Ltd and Neuronetics Inc. SHS owns intellectual property involving the use of brain connectivity to target TMS, is a scientific consultant for Magnus Medical, investigator-initiated research funding from Neuronetics and Brainsway, has received speaking fees from Brainsway and Otsuka (for PsychU.org), and is a shareholder in Brainsway (publicly traded) and Magnus Medical (not publicly traded). None of these entities was directly involved in the present work. NB received 2 NARSAD Young Investigator Grants from the Brain & Behavior Research Foundation (NLB: 2018, 2021) and a K01 award K01MH121777 (NLB) that supported the studies this Commentary focuses on.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41386-025-02153-9.