Functional Magnetic Resonance Imaging–Guided Personalization of Transcranial Magnetic Stimulation Treatment for Depression

Antidepressant outcomes to repetitive transcranial magnetic stimulation (rTMS) are better when stimulation is serendipitously delivered to sites of the dorsolateral prefrontal cortex (DLPFC) showing negative (anticorrelated) functional connectivity with the subgenual cingulate cortex (SGC).^1-3 This suggests treatment response might be improved via prospective connectivity-guided targeting. However, DLPFC connectivity varies considerably between individuals.⁴ A pertinent question is whether treatment response could be improved via a single one-site-fits-all DLPFC target, representing the group average optimal site of SGC functional connectivity, or, alternatively, whether target site personalization is necessary.

We addressed this question using recently developed methodology enabling functional magnetic resonance imaging (fMRI)–guided personalized coordinates to be computed with millimeter precision.^5,6 Specifically, in a sample of individuals with major depressive disorder who previously received rTMS treatment, we tested whether proximity between the clinically applied and (1) fMRI-personalized or (2) fixed group average fMRI-guided DLPFC targets were associated with treatment response. We hypothesized that closer proximity to personalized targets would be associated with improved response.

Individuals with major depressive disorder underwent resting-state fMRI prior to and following rTMS, as part of a clinical trial (ACTRN12610001071011) from January 2011 to September 2015. Participants provided written consent and the protocol was approved by the Alfred Hospital, Monash University, and Swinburne University research ethics boards. Treatment comprised 3 weeks of daily (5 days per week, Monday through Friday) 10-Hz rTMS targeted to the left DLPFC (F3 beam method). Clinically applied stimulation sites were recorded for each individual and mapped to Montreal Neurological Space coordinates. Randomized clinical trial design and fMRI preprocessing are detailed elsewhere.³ Concatenated pretreatment and posttreatment fMRI scans (each 6 minutes and 40 seconds; total of 13 minutes and 20 seconds) were used to retrospectively compute personalized targets. Concatenation was justified based on the absence of significant pre-post differences in SGC-DLPFC connectivity (familywise error–corrected P>.05; FSL randomise). Connectivity was computed between the SGC and each DLPFC vertex. Vertices most anticorrelated with the SGC were spatially clustered, and the center of the largest cluster was defined as the personalized coordinate (Figure, A). Seed map methodology was applied to increase signal-to-noise ratio.⁶ The Euclidean distance between clinically applied and personalized targets was correlated with the percentage improvement in Montgomery-Asberg Depression Rating Scale score (3-week time point) (Figure, B). A group average target was defined as the DLPFC site of maximal anticorrelation with the SGC (Montreal Neurological Space coordinates: −41, 43, 27) using a normative connectivity map representing consensus across 2000 twenty-eight–minute resting-state scans from 1000 participants of the Human Connectome Project. Analysis began May 2019 and ended March 2020.

Of 26 individuals with major depressive disorder, 15 (57.7%) were male, and the mean (SD) age was 44 (14) years. fMRI-personalized targets varied substantially across the spatial extent of the DLPFC (Figure, C). The median distance between personalized and actual targets was 30 mm. Closer proximity between the clinically applied and personalized targets was associated with improved treatment response (R = −0.60; P < .001; Figure, D). Importantly, this association remained significant after controlling for proximity between the clinically applied and connectivity-based group average DLPFC target (partial R = −0.54; P = .002). The association was not significant when personalized targets were substituted with established group average stimulation targets (Table).

Stronger anticorrelation between the SGC and clinically applied targets was also associated with improved outcome (R = −0.57; P = .001). This association remained significant after controlling for connectivity derived from a normative connectivity map between the SGC and clinically applied targets (partial R = −0.43; P = .02).

Clinical response to rTMS was significantly better when patients were serendipitously treated closer in proximity to personalized connectivity-guided targets. Critically, therapeutic outcome was unrelated to proximity to nonpersonalized group average stimulation targets. Thus, one-site-fits-all group average targets may insufficiently account for interindividual variation in network architecture. Conversely, fMRI acquisition and target site personalization may improve rTMS clinical efficacy. Optimal targets could alternatively be generically stimulated using less spatially specific methods (eg, deep rTMS), but their neurobiological and clinical efficacy might be compromised by concurrent stimulation of regions of positive SGC functional connectivity. Limitations include the retrospective evaluation and moderate sample size. Future prospective randomized clinical trials are warranted to assess the clinical potential of fMRI-guided personalized rTMS.

Corresponding Author: Robin F. H. Cash, PhD, Melbourne Neuropsychiatry Centre, The University of Melbourne, Victoria, Australia (robin.cash@unimelb.edu.au).

Accepted for Publication: September 25, 2020.

Published Online: November 25, 2020. doi:10.1001/jamapsychiatry.2020.3794

Author Contributions: Dr Cash had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Concept and design: All authors.

Acquisition, analysis, or interpretation of data: All authors.

Drafting of the manuscript: Cash, Zalesky.

Critical revision of the manuscript for important intellectual content: All authors.

Statistical analysis: Cash, Cocchi, Lv, Zalesky.

Obtained funding: Zalesky.

Administrative, technical, or material support: Cash, Cocchi, Lv.

Supervision: Cocchi, Fitzgerald, Zalesky.

Conflict of Interest Disclosures: Dr Fitzgerald reports a patent for a type of transcranial direct current stimulation device (10,112,056 B2) issued; has received equipment for research from MagVenture A/S, Nextsim, Neuronetics, Brainsway, Cervel Neurotech, and Medtronic; funding for research from Neuronetics; and is a founder of TMS Clinics Australia. Dr Zalesky reports grants from National Health and Medical Research Council during the conduct of the study. No other disclosures were reported.

Funding/Support: Dr Cash is funded by the Australian Research Council (grant DE200101708) and Brain & Behaviour Research Foundation. Dr Cocchi is supported by the Australian National Health and Medical Research Council (grant APP1138711). Dr Lv was supported by the Australian National Health and Medical Research Council (grant APP1142801). Dr Zalesky was supported by the Australian National Health and Medical Research Council Senior Research Fellowship B (ID: 1136649).

Role of the Funder/Sponsor: No funders had any role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.

Meeting Presentation: Part of this paper was presented in poster format at the 9th Annual Scientific Meeting of Biological Psychiatry Australia; October 28, 2019; Melbourne, Australia; and 26th Annual Meeting of the Organization for Human Brain Mapping; June 27-30, 2020; virtual.

Additional Contributions: We thank all participants, nurses and staff involved, including Richard Thomson, PhD (Monash University), for study design; Jerome Maller, PhD (Monash University), for study design; Kate Hoy, PhD (Monash University), for randomized clinical trial design; Bernadette Fitzgibbon, PhD, (Monash University), for randomized clinical trial design; Rodney Anderson, BSc(Hons) (Monash University), for data collection; Caley Sullivan, BA/BSc(Hons) (Monash University), for data collection; Melanie Emonson, PhD (Monash University), for data collection; David Elliot, BSc(Hons) (Monash University), for clinical treatment; and Susan McQueen (Monash University), for clinical treatment.