Low-intensity transcranial focused ultrasound amygdala neuromodulation: a double-blind sham-controlled target engagement study and unblinded single-arm clinical trial

Abstract

Mood, anxiety, and trauma-related disorders (MATRDs) are highly prevalent and comorbid. A sizable number of patients do not respond to first-line treatments. Non-invasive neuromodulation is a second-line treatment approach, but current methods rely on cortical targets to indirectly modulate subcortical structures, e.g., the amygdala, implicated in MATRDs. Low-intensity transcranial focused ultrasound (tFUS) is a non-invasive technique for direct subcortical neuromodulation, but its safety, feasibility, and promise as a potential treatment is largely unknown. In a target engagement study, magnetic resonance imaging (MRI)-guided tFUS to the left amygdala was administered during functional MRI (tFUS/fMRI) to test for acute modulation of blood oxygenation level dependent (BOLD) signal in a double-blind, within-subject, sham-controlled design in patients with MATRDs (N = 29) and healthy comparison subjects (N = 23). In an unblinded treatment trial, the same patients then underwent 3-week daily (15 sessions) MRI-guided repetitive tFUS (rtFUS) to the left amygdala to examine safety, feasibility, symptom change, and change in amygdala reactivity to emotional faces. Active vs. sham tFUS/fMRI reduced, on average, left amygdala BOLD signal and produced patient-related differences in hippocampal and insular responses. rtFUS was well-tolerated with no serious adverse events. There were significant reductions on the primary outcome (Mood and Anxiety Symptom Questionnaire General Distress subscale; p = 0.001, Cohen’s d = 0.77), secondary outcomes (Cohen’s d of 0.43–1.50), and amygdala activation to emotional stimuli. Findings provide initial evidence of tFUS capability to modulate amygdala function, rtFUS safety and feasibility in MATRDs, and motivate double-blind randomized controlled trials to examine efficacy.

ClinicalTrials.gov registration: NCT05228964

Introduction

Mood, anxiety, and trauma-related disorders (MATRD), i.e., major depressive disorder (MDD), bipolar disorder, anxiety disorders, and posttraumatic stress disorder (PTSD), are highly prevalent with an enormous public health burden [1, 2]. All are characterized by intense/prolonged negative affect triggered by emotionally provocative stimuli [3, 4]. Exaggerated emotional reactivity is associated with hyperactivity of the amygdala [5], a subcortical structure underlying detection of salient environmental stimuli [6] and engagement of adaptive physiological, behavioral, and affective responses [7, 8]. Consistent with a role in salience detection and emotional valence [9–12], the amygdala contributes to numerous MATRD-relevant processes [13]. Amygdala hyperactivity in MATRDs is consistent with transdiagnostic models of mental illness, i.e. the Research Domain Criteria [14], which identify the amygdala as a key mediator of all negative valence subdomains [15]. The amygdala therefore remains a promising but largely unexploited therapeutic target.

First-line MATRD interventions have been shown to alter amygdala function, including antidepressants [16] and psychotherapy [17–24]. However, non-invasive neuromodulation approaches, such as repetitive transcranial magnetic stimulation (rTMS) [25, 26], necessitate modulation of cortical activity to indirectly modulate subcortical function [27], which may depend upon integrity of cortical-subcortical structural connections [28]. As an rTMS therapeutic mechanism may be indirect modulation of subcortical limbic pathways via TMS-accessible cortical targets [29], the capacity to non-invasively and directly modulate subcortical regions without reliance upon cortical targets could offer an important therapeutic complement to existing approaches.

Low-intensity transcranial low-intensity focused ultrasound (tFUS), the focused administration of low-intensity, high-frequency sound waves to the brain, is a novel method of reversibly augmenting brain function [30]. One advantage is its ability to directly and non-invasively modulate deep brain structures without reliance on cortical connections [31]. The ellipsis (focus) can also be delivered with spatial precision ranging from millimeters to centimeters in width and with adjustable focal depth [32]. Several forms of tFUS are in clinical use or are currently under investigation [30, 33].

Candidate mechanisms for tFUS neuromodulation include thermal effects, i.e., rise in temperature, and mechanical effects, likely mediated through mechanosensitive ion channels and/or changes in membrane capacitance [33, 34]. Target-engagement experiments combining tFUS and fMRI have demonstrated changes in blood oxygenation level-dependent (BOLD) signal and perfusion in striatum [35–37], amygdala [38], and entorhinal cortex [39]. Recent work in healthy adults shows that active vs. sham tFUS attenuates amygdala activation to subsequent threat induction [38], and tFUS can also effectively modulate function of the somatosensory [40–42] and visual cortices [43]. Effects of a single sonication on neurobiological function have been demonstrated in macaques to last over an hour [44, 45]. tFUS also induces persistent changes in markers of neuroplasticity, a potential mechanism for sustained effects [46–48]. The accumulated safety profile of tFUS in humans is favorable [49]. Reviews and retrospective reports of tFUS studies reported no serious adverse events (AEs), with 4–11% of participants experiencing mild-to-moderate severity AEs [50, 51] that onset quickly and resolved 1 week to 1 month later.

Few studies have examined therapeutic tFUS neuromodulation [52–54], and tFUS modulation of amygdala across a range of MATRDs has not yet been undertaken. Given findings for amygdala hyperactivity across MATRDs [5], an intuitive strategy is to repeatedly inhibit the amygdala with tFUS. This might induce persisting neuroplastic changes to facilitate restructuring of brain function and symptom alleviation. A recent pilot study administered putative inhibitory tFUS to the right amygdala in individuals with treatment-resistant generalized anxiety disorder (GAD) once weekly for 8 weeks and observed a significant reduction in anxiety [55]. However, it is unclear whether results would generalize to a heterogeneous group of MATRDs and whether daily dosing may exert superior therapeutic effects. This study also lacked brain function assessment to provide confirmatory evidence for target engagement and post-intervention changes in amygdala function.

Here, we report findings from a double-blind, sham-controlled, within-subject tFUS/fMRI target-engagement study in patients with MATRDs and healthy controls (HCs) followed by a single-arm unblinded pilot clinical trial of daily repetitive tFUS (rtFUS) targeted to the left amygdala in MATRD patients. The target engagement study was designed to confirm a modulatory effect of left amygdala tFUS and examine potential case-control differences in tFUS-evoked responses. We used task-based fMRI to probe amygdala function before and after rtFUS to test the hypothesis that rtFUS would promote sustained post-intervention attenuations of amygdala activity. Left amygdala was selected based on evidence for a more prominent left-sided abnormality [5]. tFUS was targeted using structural MRI-guided placement against the left temporal bone (thinnest portion of lateral skull) [56] of a single-element transducer designed for human subcortical neuromodulation [57]. Patients and HCs first underwent two double-blind administrations of tFUS/fMRI (active and sham order counterbalanced) separated by about a week. Patients then commenced with once-daily rtFUS treatment delivered 5 days a week (Mon-Fri) for 3 consecutive weeks using the same protocol administered during tFUS/fMRI, previously demonstrated to attenuate left amygdala fMRI activation to fear provocation [38] and employed in the GAD treatment study [55]. We hypothesized that active vs. sham tFUS/fMRI would reduce left amygdala BOLD signal and that rtFUS in patients would be safe, show high rates of treatment completion, produce significant improvements on our primary clinical outcome, and attenuate task-based amygdala fMRI activation to naturalistic probes (emotional faces) following treatment.

Methods

Ethics approval and consent to participate

Procedures were reviewed and approved by The University of Texas at Austin Institutional Review Board (IRB00000130, OHRP Federalwide Assurance #00002030). The study was registered prior to data collection on ClinicalTrials.gov (NCT05228964), which began in Dec. 2021 and ended in Dec. 2023. All methods were performed in accordance with the relevant guidelines and regulations. Informed consent was obtained from all participants. When publishing identifiable images from human research participants, the authors have obtained written informed consent for publication of the images.

Participants

Adults ages 18–65 were recruited by flyer, word-of-mouth, and referral. Inclusion criteria for MATRD participants were English-speaking; willingness and ability to safely undergo study procedures; a Mood and Anxiety Symptom Questionnaire (MASQ) [58] General Distress (MASQ-GD) subscale score of ≥19 at screening (which provides good sensitivity/specificity in distinguishing psychiatric outpatients from a control group [59]); and meeting diagnostic criteria (assessed by Structured Clinical Interview for DSM-5; SCID-5 [60]) for MDD, bipolar disorder, GAD, social anxiety disorder (SAD), panic disorder (PD), and/or PTSD. Healthy comparison (HC) inclusion criteria differed regarding MASQ-GD score inclusion (<19) and psychiatric diagnoses, for which they could not currently nor ever have met criteria for a psychiatric disorder. See Supplementary Table 1 for exclusion criteria. Target completer sample size was selected at N = 20 or more in the MATRD group to afford power of 0.8 to detect a moderate effect size (d = 0.7) for pre/post change on symptom measures, with N = 20 or more matched HCs recruited to afford 0.8 power to detect group-related differences of a moderate effect size.

Procedures

Screening

Potential participants completed online screening and reported demographics, mental health diagnoses, current/past treatment history, and current mental health symptoms using the 30-item MASQ [58], the Patient Health Questionniare-9 (PHQ-9) [61], and the Generalized Anxiety Disorder-7 (GAD-7) [62].

Consent visit and clinical assessment

Individuals underwent informed consent and were then administered the SCID-5 [60] by trained study personnel. Individuals also completed the Vocabulary and Matrix Reasoning subtests from the Wechsler Abbreviated Scale of Intelligence 2nd Edition [63] to provide an estimate of IQ.

tFUS targeting visit

In the first MRI visit, tFUS was targeted to the left amygdala using structural MRI-guided placement (3D MPRAGE, sagittal, TR/TE = 2400/2.18 ms, slice thickness = 0.8 mm, flip angle = 8 deg, FOV = 256 mm, matrix size = 200 × 308 × 320, voxel size = 0.8 × 0.8 × 0.8 mm, duration = 6:38) with subsequent use of neuronavigation (ANT Neuro visor2; eemgagine GmbH; Berlin, Germany) to replicate placement for later visits. Proper targeting was agreed upon by consensus of both the lead scanner operator and assisting operator, both trained by GAF to visually distinguish the left amygdala using the following anatomical landmarks: (a) first, by locating in the inferior/posterior direction the separation from the anterior hippocampus by the medial protrusion of cerebrospinal fluid from the left inferior lateral ventricle; (b) then, scrolling forward in the anterior direction, the amygdala borders the entorhinal/perirhinal temporopolar cortex, distinguished by the thickness of the gray matter sheath relative to that of inferior/lateral portions of the medial temporal lobe in the coronal view; (c) in the coronal view, the amygdala will border laterally the temporal white matter core, medially it will border the subarachnoid cerebrospinal fluid, and superiorly it will border the ventral basal forebrain and cerebrospinal fluid in the choroidal fissure. See Supplementary Table 3 and Supplementary Fig. 1 for details of the transducer targeting procedure and procedure used to visually landmark the left amygdala.

Task-based fMRI and in-scanner sonication visits

Participants returned on a separate day to undergo task-based fMRI before and after a double-blind, sham-controlled (visually/tactilely identical tFUS-blocking and tFUS-transmitting transducer pads) delivery of the tFUS protocol inside the MRI scanner during collection of T2*-weighted images sensitive to blood oxygenation level-dependent (BOLD) response (TR/TE = 1390/30 ms, slice thickness = 2.5 mm, anterior-to-posterior phase encoding direction, 54 slices, flip angle = 52 deg, FOV = 215 mm, matrix size = 86 × 86 × 54, voxel size = 2.5 × 2.5 × 2.5 mm, duration = 8:00, SMS factor = 3). T2*-weighted images with identical parameters but opposite phase encoding direction were acquired prior to the task run for susceptibility distortion correction. Pads were labeled with letters “A” and “B”, and both investigators and participants were blinded to which was active and which was sham. Participants were counterbalanced for pad administration order and for task administration order of two task versions administered prior to and after sonication (different ordering of blocks). Sessions were separated by approximately 1 week (minimum 3 days), an arbitrarily chosen time delay to minimize possibility of carryover effects based upon available data [44, 45]. Pre- and post-tFUS task versions were held constant within participant across pads. Transducer was placed initially using neuronavigation with anatomical scout images used to verify adequacy of initial placement or to adjust until optimal placement was reached. The pre-treatment fMRI measure was the first run of the emotional face matching task acquired on the first MRI visit, i.e., prior to any active/sham stimulation.

fMRI task amygdala probe

We employed a widely utilized emotional face processing task [22, 64] to test effects of tFUS on amygdala function (Supplementary Fig. 2). Faces from the NimStim set [65] were matched within trial for gender and ethnicity and balanced across trials for gender. The task incorporated both shape and neutral face comparator conditions to allow for isolation of emotion-specific and face-processing effects. Emotional expressions were uniform on every trial and within blocks. On each 4-sec trial, individuals viewed a trio of faces (1 on top, 2 on bottom) for 3 s (followed by a 1 s fixation cross between trials) and were told to match the face at the top to one of the two faces on the bottom through button box key press. Two pictures on each trial were always the same (person and expression), while the third was an individual of matched gender and expression but different identity. The shape processing comparator condition used circles/ovals to induce assessment/matching processes in absence of facial stimuli and emotion. Faces were presented in 2 blocks of each condition (anger, fear, happy, neutral, shapes), pseudo-randomly ordered, with 10 trials in each block. Task was presented using PsychoPy version 3.0.1 [66].

tFUS administration

In tFUS/fMRI and rtFUS treatment sessions, participants received the same tFUS stimulation protocol with the Brainsonix Pulsar 1002 [57], which has demonstrated engagement of amygdala and other subcortical structures in previously-published studies [35, 38, 39, 55]. Participants received tFUS with the following parameters: 10 Hz pulse repetition frequency (PRF), 5% duty cycle (DC), 5 ms pulse width, derated spatial peak pulse average intensity of 14.4 watts/cm², derated spatial peak temporal average intensity of 719.91 milliwatts/cm² (720 milliwatts/cm² is the current FDA safety limit), and derated instantaneous peak pressure of 0.64 megapascals over 10 min in 10 sets of 30 s on/off blocks. Parameters were informed by: (a) animal work suggesting lower DCs (percentage of the time from the start of one tone burst duration to the start of the next tone burst duration, i.e. one cycle of tFUS sonication in a pulsed design, during which ultrasound is delivered) showed inhibitory effects whereas higher DCs (e.g., 50%) showed excitatory effects [67]; and (b) unpublished pilot data. AEs and blinding integrity/confidence was assessed via written questionnaire after each session.

rtFUS administration sessions

On the Monday following completion of both tFUS/fMRI sessions, patients with MATRDs initiated rtFUS treatment. Transducer placement was replicated using neuronavigation. Participants rested with eyes open in a seated position in a quiet room (to minimize variability in behavioral context). Participants returned once daily, Mon-Fri, for 3 weeks to repeat the stimulation procedure. MASQ-GD was collected at the start of rtFUS sessions 1, 3, 5, 7, 9, 11, 13 and 15.

Post-treatment measures

After completing treatment, participants repeated symptom measures collected at study outset. Participants also underwent a post-treatment MRI scan to assess changes in brain function during the emotional face matching task. Scan occurred, on average, within 3–7 days of cessation of treatment.

Target engagement outcome measures

Primary outcome for tFUS/fMRI was change in fMRI BOLD signal during tFUS delivery (10 × 30-sec long blocks of sonication separated by 30-sec rest blocks), modeled as a single boxcar regressor convolved with the hemodynamic response function (HRF). Blinding was assessed querying each participant’s guess for active or sham after the session as well as their confidence rating from 0 (Not at All Confident) to 10 (Extremely Confident).

Pilot rtFUS clinical trial outcome measures

Safety

Primary outcomes were rates and severity of AEs, elicited through verbal query after sonication and at the start of each study visit. If new-onset symptoms or unresolved symptoms were endorsed, participants completed a standardized questionnaire to gather further information on the type/nature of the AE and its severity.

Acceptability

Primary outcome was rate of treatment completion, which was defined as the proportion of individuals completing all FUS open-label treatment sessions of those individuals receiving any FUS open-label treatment sessions.

Clinical outcomes

Primary clinical outcome was change in MASQ-GD from baseline to after the rtFUS clinical trial [58]. We also examined change in MASQ-GD from baseline to the start of the rtFUS clinical trial as an exploratory measure of potential effects from the single in-scanner active administration. The MASQ-GD was designed and validated as a dimensional measure of negative affect symptoms in mood and anxiety diagnoses [68]. MASQ-GD was selected for its transdiagnostic relevance and its theoretical and empirical relationship to negative affect [69], which is elevated across MATRDs [70] and empirically related to amygdala function [71]. See Supplementary Table 2 for list of secondary outcomes.

Statistical analyses

Preprocessing, individual-level analysis, and quality control of fMRI

fMRI data was preprocessed using FSL [72]. Affine transformation of functional to structural images was added to non-linear normalization of each participant’s T1 image to the Montreal Neurological Institute (MNI) 152-person 1 mm³ T1 template using FNIRT from FSL 5.0 [73]. Functional images were re-aligned to the middle volume of the run using rigid-body motion correction (FSL’s mcflirt), and FSL’s topup program was used to construct a susceptibility-induced off-resonance field [74] which was then used to apply a correction to the data from the task functional run. Motion and susceptibility distortion-corrected functional data from each participant was then normalized to MNI atlas space and resampled to a 2 × 2 × 2 mm voxel size. Functional data was smoothed with an isotropic 6 mm full-width half max Gaussian kernel to account for individual anatomical variability. For quality control, participants were set to be excluded a priori if they had a root mean square absolute movement >3 mm across the mean of the squared maximum displacements in each of the 6 translational and rotational motion parameters. However, all participants met this criterion (max root mean square of absolute displacement = 2.91 mm).

AFNI [75] was used for individual-level analysis. Individual time series images were scaled to percent signal changes. For tFUS/fMRI, a box car regressor modeling onset and duration of the condition of interest (tFUS on) was convolved with the HRF. For task-based fMRI, box car regressors modeling onset and duration of blocks of interest (anger, fear, happy, neutral, and shapes) were convolved with the HRF. Additional regressors of no interest for both scans included 6 motion regressors (translations and rotations in x, y, and z dimensions) and second-order Legendre polynomials to account for scanner intensity drift over time. Volumes with motion exceeding 0.3 of the Euclidean norm of motion parameter derivatives (and preceding volumes) were censored from analysis. The program 3dDeconvolve was used to set up the deconvolution matrix, which was then executed by 3dREMLfit to estimate effects using generalized least squares time series fit with restricted maximum likelihood (REML) estimation of temporal autocorrelation.

tFUS/fMRI generalized psychophysiological interaction analysis (gPPI)

To better characterize tFUS effects, we examined tFUS-dependent changes in connectivity using a generalized psychophysiological interaction (gPPI) analysis [76] in AFNI. From the left amygdala area demonstrating a main effect of active vs. sham tFUS on activation, individual time series of BOLD signal changes was extracted from fully preprocessed fMRI time series data. Timeseries were detrended for scanner drift, and the canonical HRF was removed from each timeseries to estimate the neuronal timeseries. The interaction of these neuronal timeseries with the boxcar regressor for tFUS delivery was then calculated to yield a PPI effect. The PPI timeseries was then convolved with the canonical HRF to model the ideal timeseries for a context-dependent connectivity effect with left amygdala activity for active vs. sham tFUS. This interaction effect, the detrended left amygdala timeseries, the HRF-convolved boxcar regressor for tFUS delivery, and additional regressors of no interest, i.e., 6 motion regressors (translations and rotations in x, y, and z dimensions) and second-order Legendre polynomials to account for scanner intensity drift over time, were entered into a deconvolution analysis (executed as above). The beta coefficient for the interaction is the primary effect of interest, which indexes magnitude of active vs. sham tFUS modulation of regional relationships with left amygdala activity.

Group-level analyses

Blinding/Confidence assessment for tFUS/fMRI

Confidence ratings were analyzed as a continuous variable from 0 to 10 using a linear mixed effects (LME) model in IBM SPSS version 28 [77]. The effects of active or sham, session 1 or session 2 of tFUS/fMRI, and patient group were examined as fixed effects with all main effects and higher-order interactions, subject as a random factor, variance components for random effects, REML estimation, and the Satterthwaite approximation of degrees of freedom (DOF). Correct or incorrect guesses were analyzed as a dichotomous outcome using chi-square tests to sensitively assess distributions of correct/incorrect guesses across all participants and sessions, within each session (session 1 or 2), within each group (HC and MATRD patients), and within each group and session (session 1 and 2 for HC and MATRD patients).

tFUS/fMRI activation

For group analyses, individual parameter estimates for the effect of tFUS delivery were carried forward to a LME analysis in the AFNI program 3dLME [78]. At each voxel, a random and fixed intercept was modeled with fixed effects of session (active or sham), patient group (MATRD or HC), and the group x session interaction. A priori effects of interest were the signed t-contrast for the effect of session (active vs. sham tFUS across all participants) and the session (active vs. sham) x group (MATRD vs. HC) interaction. The t-value statistical map for each effect was then subsequently converted to Z scores and subjected to probabilistic threshold free cluster enhancement (pTFCE) [79] within a dilated bilateral amygdala and anterior hippocampus mask (bilateral amygdala from FSL’s subcortical segmentation program FIRST [80] and bilateral MNI152 hippocampus regions anterior to y = −21 as in prior work [81]), with anterior hippocampus included to detect potential off-target effects on most proximal nearby structure. A whole-brain analysis was also conducted with pTFCE correction for multiple comparisons.

tFUS/fMRI context-dependent amygdala connectivity

Individual beta weights for the PPI effect (tFUS delivery x left amygdala timeseries) were carried forward to a group LME analysis in the AFNI program 3dLME [78] and analyzed as above. The t-statistic map for each effect was converted to Z scores and subjected to probabilistic threshold free cluster enhancement (pTFCE) [79] within a limbic and prefrontal mask encompassing bilateral amygdala (from FSL’s subcortical segmentation program FIRST [80]), anterior hippocampus (bilateral MNI152 hippocampus regions anterior to y = −21 [81]), striatum (caudate and putamen from the California Institute of Technology Reinforcement Learning atlas [82]), insula (defined by Automatic Anatomical Labeling (AAL) atlas [83]), anterior and mid-cingulate cortex (anterior cingulate, mid-cingulate, and olfactory cortex sites of the AAL atlas with y > 0, −14 < x < 14, and −12 < z < 44), and bilateral lateral/dorsolateral prefrontal cortex (bilateral inferior frontal, middle frontal, and superior frontal gyri from the AAL atlas constrained by z > −4, 16 < x < 60 for right hemisphere, −60 < x < −16 for left hemisphere, and y > −10). A whole-brain analysis was also conducted with pTFCE correction for multiple comparisons.

Pre- to post-rtFUS change in task-based fMRI

Individual activation magnitudes for each task condition were carried forward to a LME analysis in the AFNI program 3dLME [78]. At each voxel, a random and fixed intercept was modeled along with fixed effects of time, task condition, and the time x task condition interaction. A priori effects of interest were the main effect of time and the time x condition interaction. The F-value statistical map for each effect was converted to Z scores and subjected to probabilistic threshold free cluster enhancement [79] (pTFCE) within a bilateral amygdala mask from FSL’s subcortical segmentation program FIRST [80] (to test hypotheses regarding rtFUS-induced changes in amygdala activity). A whole brain analysis was also conducted with pTFCE correction for multiple comparisons. All reported fMRI findings exceed the pTFCE corrected threshold of p < 0.05. Post-hoc LMEs recapitulating voxelwise models were run in IBM SPSS 28.0 [77] to characterize directionality of time-related changes and/or decompose time x condition interaction effects.

Clinical outcomes and safety/feasibility

To assess changes on clinical outcomes, we employed longitudinal LME models in IBM SPSS version 28 [77]. All models employed a random and fixed intercept and fixed effect of time using variance components for random effects, REML estimation, and Satterthwaite approximation of DOF. Intent-to-treat (ITT) analyses encompassed all participants receiving at least one in-scanner sonication and all participants with available post-treatment data. Sensitivity analyses were conducted in individuals meeting diagnostic criteria for diagnoses represented in at least half of the completing sample (N = 10 or more). Rates of rtFUS treatment completion were assessed as the percent of the sample initiating the 15-day treatment course who also completed all 15 rTFUS sessions. Rates of AEs were calculated as the number of participants reporting the AE in each condition, and severity was rated by the participant on a 0–10 scale with 0 being “not noticeable at all” and 10 being “extremely noticeable.” McNemar’s chi-square tests in IBM SPSS 28.0 [77] were used to assess differences in frequency of AEs by active or sham.

Results

Completion rates

Of 83 individuals assessed, 52 (29 patients, 23 HC) were eligible and enrolled (Table 1 and Supplementary Table 4). Of 52, 47 completed both tFUS/fMRI scans (25 patients, 22 HC), and 24 patients initiated daily rtFUS. Of 24, 3 discontinued prior to completing treatment, leaving 21 individuals receiving all sessions of rtFUS (~88% of sample initiating; Supplementary Fig. 3). A wide spectrum of diagnoses was represented, including MDD (N = 16), bipolar I disorder (N = 4), bipolar II disorder (N = 6), GAD (N = 16), PTSD (N = 10), SAD (N = 7), and PD (N = 2).

Table 1.

Sample demographic and clinical characteristics.

Measure/Characteristic	HC Sample: Mean (SD) or N	MATRD Sample: Mean (SD) or N	MATRD vs. HC Statistic (pval)
N	23	29	--
Age	22.59 (4.47)	24.14 (7.72)	t = 0.84 (0.41)
Sex Assigned at Birth	N = 13 female N = 10 male	N = 22 female N = 7 male	${\mathcal{X}}^2$ = 2.18 (0.14)
Gender Identity	N = 13 female N = 10 male	N = 22 female N = 7 male	${\mathcal{X}}^2$ = 2.18 (0.14)
Ethnicity	N = 8 Hispanic N = 15 non-Hispanic	N = 13 Hispanic N = 16 non-Hispanic	${\mathcal{X}}^2$ = 0.54 (0.46)
Race	N = 15 White/Caucasian N = 1 Black or African American N = 7 Asian N = 1 Native American or Alaskan Native	N = 25 White/Caucasian N = 1 Black or African American N = 3 Asian N = 2 Native American or Alaskan Native	${\mathcal{X}}^2$ = 3.60 (0.31)
Work/Career	N = 3 Full-time employment N = 7 Part-time employment N = 0 Recently unemployed N = 1 Unemployed for 6+ months N = 16 Full-time student N = 1 Part-time student	N = 9 Full-time employment N = 7 Part-time employment N = 1 Recently unemployed N = 1 Unemployed for 6+ months N = 16 Full-time student N = 3 Part-time student	${\mathcal{X}}^2$ = 3.15 (0.53)
WASI Full Scale IQ	119.62 (17.05)	119.71 (17.48)	t = 0.02 (0.99)
Years of Education	15.91 (2.02)	15.43 (2.51)	t = −0.73 (0.47)
Diagnoses	N/A	N = 16 MDD N = 4 BD I N = 6 BD II N = 4 AUD (mild) N = 2 PD N = 7 SAD N = 16 GAD N = 10 PTSD	--
Psychiatric Meds	N/A	N = 18 Yes, N = 11 No N = 4 Fluoxetine N = 2 Escitalopram N = 4 Bupropion N = 2 Venlafaxine N = 2 Desvenlafaxine N = 1 Sertraline N = 1 Duloxetine N = 3 Lamotrigine N = 3 Lithium N = 1 Quetiapine N = 1 Clonazepam N = 1 Trazodone N = 1 Lurasidone N = 1 Gabapentin N = 1 Lisdexamfetamine	--
Current Therapy	N/A	N = 11 Yes, N = 18 No	--
MASQ-GD	14.27 (2.57)	26.79 (6.41)	t = 9.56 (<0.001)
MASQ-AD	27.35 (4.60)	38.45 (8.48)	t = 4.45 (<0.001)
MASQ-AA	11.95 (2.19)	18.41 (6.16)	t = 5.23 (<0.001)
PHQ-9	2.55 (3.31)	11.00 (5.29)	t = 6.99 (<0.001)
GAD-7	2.27 (3.95)	11.31 (4.91)	t = 7.07 (<0.001)
SHAPS	0.86 (2.52)	3.00 (3.60)	t = 2.45 (0.018)
STAI Trait	33.70 (10.74)	53.50 (11.91)	t = 5.91 (<0.001)
STAI State	30.45 (10.86)	48.29 (12.64)	t = 5.10 (<0.001)
PCL-5	5.67 (7.13)	30.14 (16.83)	t = 6.91 (<0.001)
WHO-QOL Global	8.43 (1.86)	6.68 (1.85)	t = −3.27 (0.002)
WHO-QOL Physical	16.90 (2.98)	13.53 (2.47)	t = −4.32 (<0.001)
WHO-QOL Psych	15.24 (3.48)	10.48 (2.57)	t = −5.51 (<0.001)
WHO-QOL Social	15.62 (3.30)	12.57 (3.76)	t = −2.96 (0.005)
WHO-QOL Envir	15.98 (3.81)	14.80 (2.62)	t = −1.28 (0.208)
BDI-II	2.76 (5.49)	20.57 (13.05)	t = 6.99 (<0.001)
PANAS PA	32.29 (9.76)	24.07 (7.97)	t = −3.24 (0.002)
PANAS NA	13.24 (5.47)	25.89 (8.15)	t = 6.50 (<0.001)
BAI	3.50 (4.47)	21.07 (11.73)	t = 7.23 (<0.001)
ASI-3	6.20 (6.96)	24.86 (14.41)	t = 5.95 (<0.001)
PSQI Global Sum	4.29 (3.11)	9.31 (3.37)	t = 4.91 (<0.001)
TEPS AP	4.34 (1.13)	3.75 (1.08)	t = −1.86 (0.069)
TEPS CP	4.66 (1.15)	4.23 (0.89)	t = −1.47 (0.147)
S-UPPS-P Total	37.05 (7.17)	41.70 (8.18)	t = 2.08 (0.043)
CD-RISC	28.38 (8.97)	22.00 (6.80)	t = −2.80 (0.007)
DDQR TW NofDrinks	1.48 (2.44)	2.71 (4.58)	t = 1.12 (0.267)
DDQR TW NofHrs	1.62 (2.84)	2.75 (4.17)	t = 1.07 (0.290)
DDQR HW NofDrinks	2.71 (3.51)	5.25 (8.17)	t = 1.47 (0.149)
DDQR HW NofHours	2.67 (3.48)	4.89 (7.32)	t = 1.41 (0.166)
AUDIT	1.67 (1.62)	3.11 (3.86)	t = 1.78 (0.084)

AA anxious arousal, AD anhedonic depression, AP Anticipatory Pleasure, ASI-3 Anxiety Sensitivity Index 3, AUDIT Alcohol Use Disorder Identification Test, BAI Beck Anxiety Inventory, BDI-II Beck Depression Inventory II, CD-RISC Connor Davidson Resilience Scale, CP Consummatory Pleasure, DDQR Daily Drinking Questionnaire Revised, Envir Environmental Health, GAD-7 Generalized Anxiety Disorder 7, GD General Distress, HW Heaviest Week, MASQ Mood and Anxiety Symptom Questionnaire, NA Negative Affect, NofDrinks Number of Drinks, NofHrs Number of Hours, PA Positive Affect, PANAS Positive and Negative Affect Scale, PCL-5 PTSD Checklist for DSM-5, PHQ-9 Patient Health Questionnaire 9, PSQI Pittsburgh Sleep Quality Inventory, Psych Psychological Health, QIDS Quick Inventory of Depressive Symptomatology Self Report, Social Social Relationships, SHAPS Snaith Hamilton Pleasure Scale, STAI Spielberger State-Trait Anxiety Inventory, S-UPPS-P Short Version of the Urgency-Premeditation-Perseverance-Sensation Seeking-Positive Urgency Impulsive Behavior Scale, TEPS Temporal Experiences of Pleasure Scale, TW Typical Week, WHO-QOL World Health Organization Quality of Life Brief Version.

tFUS/fMRI outcomes

All reported fMRI findings display voxelwise pTFCE-corrected p’s < 0.05 within either ROI-constrained or whole brain analysis.

Assessment of blinding

Distributions of real/fake guesses by active/sham condition did not significantly differ across groups and sessions (χ² = 0.03, p = 0.87), within session (χ²’s < 0.66, p’s > 0.41), within group (χ²’s < 0.62, p’s > 0.43), or within group and session (χ²’s < 1.77, p’s > 0.18; Supplementary Table 5), suggesting that blinding was maintained. Analyses of confidence data revealed a main effect of session (F_1,43 = 2.34, p = 0.024), with participants more confident in their guesses in the second session. There was also a significant interaction of Group x Active or Sham stimulation (F_1,62 = 2.08, p = 0.042), with MATRD participants equally confident in their guesses in active (mean = 4.73) and sham conditions (mean = 4.19) while HC participants reported less confidence in their guesses for active (mean = 3.09) vs. sham (mean = 4.18) conditions.

Active vs. Sham tFUS on fMRI activation

ROI-constrained analyses revealed a significant attenuation of left amygdala and left anterior hippocampal BOLD signal as a main effect of active vs. sham tFUS (Fig. 1). In the MATRD group, 15 of 25 showed a BOLD signal reduction (60%), i.e. negative parameter estimates, for active tFUS and within-subject contrasts of active vs. sham tFUS in the left amygdala region showing a main effect for active vs. sham tFUS in voxelwise analyses. In the HC group, 16 of 22 showed a BOLD signal reduction (73%) for active tFUS and 14 of 22 showed a BOLD signal reduction (64%) for the within-subject contrast of active vs. sham tFUS in this left amygdala region. The binary distributions of activation or deactivation did not differ between MATRDs and HCs for either active tFUS (χ² = 0.84, p = 0.36) or within-subject contrasts of active vs. sham tFUS (χ² = 0.07, p = 0.80). Visual examination of the distribution (Fig. 1) for active tFUS BOLD signal changes in the left amygdala reveals 5 participants with robust deactivation on the tail (% signal changes > −0.20). In post-hoc exploratory analyses, we examined these 5 participants for unique clinical/demographic characteristics. Of these 5, 3 were in the MATRD group and 2 were in the HC group. Post-hoc testing for age, years of education, continuous outcome measures (using t-tests), gender, and diagnoses (using logistic regressions) revealed these 3 patients vs. the 22 others displayed significantly higher scores on the MASQ-GD (equal variances not assumed, Levene’s Test F = 7.832, p = 0.010; t₂₂ = 6.33, p < 0.001) and the PHQ-9 (t₂₄ = 2.43, p = 0.023). The MASQ-GD difference survived Bonferroni correction for the 32 continuous metrics examined (posterior threshold of p < 0.0016). The non-parametric correlation between active tFUS inhibition in the MATRD group in this left amygdala region and MASQ-GD score was also significant (ρ₂₆ = −0.52, p = 0.006). Likelihood of diagnosis presence or absence did not differ in these 3 patients vs. the others (logistic regression p’s > 0.13). The 2 HCs did not significantly differ from the other HCs on any examined metric (all p’s > 0.13), nor were MASQ-GD scores associated with active tFUS BOLD signal changes in HCs (ρ₂₁ = 0.25, p = 0.28). ROI-constrained analyses additionally revealed a condition x group interaction in the right anterior hippocampus such that MATRD patients did not show a differential active vs. sham effect (t₂₅ = 0.51, p = 0.62) while HCs showed reduced activation in this region for active vs. sham tFUS (t₄₂ = −2.22, p = 0.03) (Fig. 1). Whole-brain analyses (Fig. 1) revealed additional main effects of active vs. sham tFUS (attenuation of BOLD signal in the left hypothalamic region and the left ventral anterior insula/putamen) and a condition x group interaction in the right middle insula, where patients show within-group activation for active vs. sham (t₂₅ = 2.38, p = 0.03) and HCs do not (t₄₂ = −1.44, p = 0.16). See Supplementary Table 6 for details.

Fig. 1. Evoked effects from Active vs. Sham tFUS/fMRI and differences by Patient Group.

Figure depicts amygdala and hippocampal areas displaying probabilistic threshold-free cluster enhancement corrected significant (p < 0.05) effects from the region of interest-constrained analysis (left half) and the whole-brain exploratory analysis (right half) of voxelwise linear mixed effects models for the main effect of active vs. sham transcranial focused ultrasound (tFUS)(top half) and the interaction effect of mood, anxiety, and trauma-related disorder (MATRD) patients vs. healthy comparison (HC) group differences for active vs. sham tFUS (bottom half). Areas of significant difference in evoked effects of active vs. sham tFUS and its interaction with patient group are depicted as colorized and signed based upon the contrast of active vs. sham tFUS mean activation values at each voxel (main effect) and the MATRD vs. HC differences for active vs. sham tFUS (interaction effect) at each voxel. Therefore, cool (light to dark blue) colors indicate regions of significantly decreased activation for active vs. sham tFUS (top half) while warm (light to deep red) colors indicate regions of significantly larger activation/less deactivation for MATRD patients vs. HC participants for evoked effects of active vs. sham tFUS (bottom half). Suprathreshold voxels are outlined in black with a linear fade applied to subthreshold voxels as a function of distance from significance threshold. Results are displayed on the MNI152 ICBM 2009c non-linear asymmetric average brain. Split stringed violin plots depict observed average fMRI % signal changes for active and sham tFUS for each individual averaged over suprathreshold voxels. White dots indicate observed mean values for each participant averaged over suprathreshold voxels for sham- (left/red distribution) and active (right/blue distribution), with individual participant observations connected by black lines. Widest lines directly adjacent to white dots indicate the mean value at that time point, with connecting lines above/below indicating the within-subject 95% confidence interval for the mean based upon the Cousineau–Morey method. Asterisks(*) with black connecting lines indicate comparisons that are statistically significant.

Active vs. Sham tFUS on left amygdala connectivity

The gPPI analysis with seed region defined by the active vs. sham effect on left amygdala activation revealed in the ROI-constrained analysis a main effect of active vs. sham (increased connectivity with right posterior dorsolateral prefrontal cortex, i.e., right superior frontal gyrus in Brodmann Area 6; Supplementary Fig. 4) and a condition x group interaction in which MATRD patients show for active vs. sham tFUS increased connectivity with right dorsal mid-insula (t₅₀ = 3.72, p < 0.001) while HC participants show decreased connectivity (t₅₀ = −3.26, p = 0.002) (Supplementary Fig. 4). Whole brain analyses also identified both effects (Supplementary Table 6).

Change on primary symptom outcome following tFUS/fMRI

MASQ-GD was assessed again at the start of the first rtFUS treatment session. This facilitated an exploratory analysis to examine potential MASQ-GD changes following a single active sonication (and a single sham sonication), both delivered in the MRI scanner. An ITT LME model revealed a significant attenuation (F_1,24 = 15.18, p < 0.001) in MASQ-GD scores from 26.79 ± 6.41 at baseline to 23.48 ± 6.98 at the start of the rtFUS treatment course (Fig. 2).

Fig. 2. Changes on primary outcome from before to after repetitive transcranial focused ultrasound.

A depicts a split stringed violin plot of observed Mood and Anxiety Symptom Questionnaire General Distress (MASQ-GD) subscale scores pre- and post-repetitive transcranial focused ultrasound (rtFUS). White dots indicate observed values at pre- (left/red distribution) and post-rtFUS (right/blue distribution), with individual participant observations connected by black lines. Widest lines directly adjacent to white dots indicate the mean value at that time point, with connecting lines above/below indicating the within-subject 95% confidence interval for the mean based upon the Cousineau–Morey method. B depicts the individual predicted values from the linear mixed effects model analysis for pre- to post-treatment change. Gray dots indicate individual predicted values at pre- and post-rtFUS. Black line connects points at the mean of each distribution at each time point. Error bars indicate +/− 1 standard error at each time point. C depicts the observed means and standard errors of MASQ-GD scores collected at baseline, before the first rtFUS session and prior to every odd-numbered rtFUS session, and at post-treatment. Error bars indicate +/− 1 standard error at each time point. The period on this graph during which individuals underwent active/sham tFUS/fMRI is noted. Asterisk (*) with black connecting lines indicates a statistically significant (p < 0.05) difference in the MASQ-GD comparison between indicated time points.

Safety

There were 24 AEs reported across 29 participants, with ~92% (22 of 24) occurring in-scanner (Supplementary Table 7). There was an even split of AEs across active and sham (11 in each) with no significant differences in frequency by condition (χ² = 0.00, p = 1.00). There were 2 additional AEs reported in the treatment portion (headache and irritability). Most common AEs across conditions were decreased concentration, tingling in limbs/hands, headache, and dizziness/light-headedness (2 of each). Severities ranged from 1–7, with mean severity in the active in-scanner sonication being the highest at ~5. Nearly all AEs resolved during the study visit, except for two cases of mild headache that abated prior to next study visit and one case of mild increased irritability during treatment that abated within a few days. About 50% of participants reported some AE (14 of 29) across all study procedures, with only 4 participants reporting an AE only during verum FUS. There were no serious AEs nor AEs requiring medical intervention.

rtFUS pilot trial outcomes

Primary symptom outcome

The ITT LME model revealed a significant attenuation (F_1,25 = 12.89, p = 0.001) of MASQ-GD scores from 26.79 ± 6.41 at baseline to 21.10 ± 7.02 at post-treatment (Fig. 2 and Table 2). The effect size was a Cohen’s d of 0.77. We then explored additional metrics of reliable change and clinically significant change (CSC). A reliable change index (RCI) was calculated [84] based upon a test/re-test reliability estimate of 0.80 for the MASQ-GD [85] and the standard deviation in MATRD scores at baseline (6.41) which yielded a SD in measurement variance of ~4 points. We used the threshold of >1.96 SDs of measurement variance as a criterion for change greater than that expected due to measurement variance alone. 10 of 20 MATRD participants with post-treatment data demonstrated a reduction that was larger than this threshold. CSC [84] was defined as post-treatment MASQ-GD scores in the MATRD group falling at or below midway between the means in the MATRD group and that of the HCs (the recommendation when distributions overlap between patient and normative samples) [59]. We observed that 11 of 20 MATRD participants fell at or below this midway point at post-treatment.

Table 2.

Effects of repetitive transcranial focused ultrasound on primary and secondary clinical outcomes.

Measure	Baseline	PostTx	DOF		ITT		Cohen’s d
			Num	Denom	F	p
Primary Outcome
MASQ-GD	26.79 ± 6.41	21.10 ± 7.02	1	24.67	12.89	0.001	0.77
Secondary Outcomes
MASQ-AD	38.45 ± 8.48	33.60 ± 8.00	1	22.29	8.31	0.009	0.54
MASQ-AA	18.41 ± 6.16	15.00 ± 3.74	1	26.29	6.46	0.017	0.62
PHQ-9*	11.00 ± 5.29	6.55 ± 4.75	1	19.65	16.85	<0.001	0.87
GAD-7*	11.31 ± 4.91	6.80 ± 5.35	1	23.49	21.14	<0.001	1.02
SHAPS	3.00 ± 3.60	1.14 ± 2.82	1	24.70	7.86	0.010	0.55
STAI Trait	53.50 ± 11.91	48.24 ± 11.30	1	25.00	5.26	0.031	0.36
STAI State	48.29 ± 12.64	39.67 ± 11.19	1	23.40	10.43	0.004	0.45
PCL-5*	30.14 ± 16.83	18.14 ± 13.50	1	23.62	19.84	<0.001	1.51
QIDS*	10.46 ± 4.57	6.90 ± 4.05	1	24.89	14.49	<0.001	0.77
WHO-QOL Global	6.68 ± 1.85	7.43 ± 1.54	1	23.95	4.28	0.049	−0.43
WHO-QOL Physical	13.53 ± 2.47	14.67 ± 2.54	1	25.11	4.11	0.053	−0.39
WHO-QOL Psych	10.48 ± 2.57	12.13 ± 3.08	1	24.86	8.15	0.009	−0.61
WHO-QOL Social	12.57 ± 3.76	12.76 ± 4.27	1	22.98	0.02	0.893	0.06
WHO-QOL Envir	14.80 ± 2.62	15.31 ± 2.75	1	23.41	1.74	0.200	−0.39
BDI-II	20.57 ± 13.05	11.95 ± 9.60	1	19.76	13.47	0.002	0.79
PANAS PA	24.07 ± 7.97	26.90 ± 8.10	1	25.20	1.94	0.180	−0.21
PANAS NA*	25.89 ± 8.15	20.05 ± 7.70	1	21.98	13.38	0.001	0.73
BAI*	21.07 ± 11.73	11.30 ± 9.71	1	20.53	14.78	<0.001	0.71
ASI-3	24.86 ± 14.41	17.57 ± 15.85	1	22.83	7.76	0.011	0.56
PSQI Global Sum	9.31 ± 3.37	6.27 ± 3.77	1	19.08	13.26	0.002	0.85
TEPS AP	3.75 ± 1.08	4.23 ± 1.11	1	21.13	9.04	0.007	−0.61
TEPS CP	4.23 ± 0.89	4.69 ± 0.80	1	21.76	7.46	0.012	−0.51
S-UPPS-P Total	41.71 ± 8.18	41.40 ± 8.81	1	19.95	0.12	0.730	0.11
CDRISC	22.00 ± 6.80	25.95 ± 5.09	1	20.55	5.32	0.032	−0.37
DDQR TW NofDrinks	2.71 ± 4.58	2.85 ± 4.55	1	19.86	0.47	0.501	0.18
DDQR TW NofHrs	2.75 ± 4.17	3.75 ± 5.52	1	22.38	0.84	0.369	−0.17
DDQR HW NofDrinks	5.25 ± 8.17	5.55 ± 7.14	1	20.40	0.31	0.582	0.13
DDQR HW NofHours	4.89 ± 7.32	5.50 ± 7.34	1	20.98	0.03	0.868	0.05
AUDIT	3.11 ± 3.86	3.00 ± 2.83	1	21.44	0.41	0.529	0.16

Baseline and PostTx columns depict the measure mean + the standard deviation.

AA anxious arousal, AD anhedonic depression, AP Anticipatory Pleasure, ASI-3 Anxiety Sensitivity Index 3, AUDIT Alcohol Use Disorder Identification Test, BAI Beck Anxiety Inventory, BDI-II Beck Depression Inventory II, CD-RISC Connor Davidson Resilience Scale, CP Consummatory Pleasure, DDQR Daily Drinking Questionnaire Revised, Denom denominator, DOF degrees of freedom, Envir Environmental Health, GAD-7 Generalized Anxiety Disorder 7, GD General Distress, HW Heaviest Week, ITT intent to treat, MASQ Mood and Anxiety Symptom Questionnaire, NA Negative Affect, NofDrinks Number of Drinks, NofHrs Number of Hours, Num numerator, PA Positive Affect, PANAS Positive and Negative Affect Scale, PCL-5 PTSD Checklist for DSM-5, PHQ-9 Patient Health Questionnaire 9, PostTx post-treatment, PSQI Pittsburgh Sleep Quality Inventory, Psych Psychological Health, QIDS Quick Inventory of Depressive Symptomatology Self Report, Social Social Relationships, SHAPS Snaith Hamilton Pleasure Scale, STAI Spielberger State-Trait Anxiety Inventory, S-UPPS-P Short Version of the Urgency-Premeditation-Perseverance-Sensation Seeking-Positive Urgency Impulsive Behavior Scale, TEPS Temporal Experiences of Pleasure Scale, TW Typical Week, WHO-QOL World Health Organization Quality of Life Brief Version.

*measure survives Bonferroni correction (p < 0.0017).

Secondary symptom outcomes

Most secondary outcomes demonstrated pre- to post-treatment improvements at p < 0.05 uncorrected (Supplementary Fig. 5 and Table 2), including measures of anhedonia/pleasure; symptoms of depression, anxiety, and PTSD; negative affect; sleep difficulties; and certain quality of life facets. Of these, effect sizes were moderate (0.43) to very large (1.50). Employing a Bonferroni correction for 29 secondary outcomes (p < 0.0017), measures of depression, anxiety, PTSD symptoms, and negative affect continue to surpass this threshold (Table 2).

Sensitivity analyses

To examine if changes on diagnosis-specific secondary outcomes were consistent in individuals meeting diagnostic criteria, we conducted sensitivity analyses in sub-samples restricted by inclusion diagnosis (Supplementary Table 8) for MDD (N = 16), GAD (N = 16), and PTSD (N = 10). ITT analyses demonstrated significant attenuations on depressive symptoms in those with MDD (F’s > 7.37, p’s < 0.023), on anxiety symptoms in those with GAD (F’s > 11.74, p’s < 0.009), and on PTSD symptoms in those with PTSD (F_1,9 = 15.31, p = 0.004).

Changes in task-based amygdala activation

All reported fMRI findings display voxelwise pTFCE-corrected p’s < 0.05 within either ROI-constrained or whole brain analysis. There was a significant main effect of time in left and right amygdalae (Fig. 3 and Supplementary Table 9), driven by decreased activation from pre- to post-treatment across all task conditions (right amygdala parameter estimate = −0.135, t₂₃₀ = −4.30, p < 0.001; left amygdala parameter estimate = −0.118, t₂₃₀ = −4.09, p < 0.001). There was a significant time x condition effect in the left amygdala bordering the active vs. sham tFUS/fMRI main effect (Fig. 4 and Supplementary Table 9). Post-hoc analysis revealed attenuation of activation to angry faces (parameter estimate for time x anger (vs. shapes) = −0.25, t₂₀₆ = −2.65, p = 0.009) with no other fixed parameter estimates for time x emotion (vs. shapes) effects reaching statistical significance (all p’s > 0.08). It is also notable the anger condition evoked the largest level of baseline task activation in this region (parameter estimate for anger (vs. shapes) = 0.27, t₂₀₆ = 4.43, p < 0.001) followed by happy faces (parameter estimate for happy (vs. shapes) = 0.13, t₂₀₆ = 2.05, p = 0.042), with fearful and neutral (vs. shapes) showing no significant baseline activation (p’s > 0.32). Additional regions demonstrated a significant main effect of time (Supplementary Fig. 6) in the whole-brain analysis, but no additional areas displayed significant time x condition interaction effects.

Fig. 3. Condition-general changes in emotional face processing task-based functional magnetic resonance imaging amygdala activation.

Figure depicts mean individual activation values at pre- and post-rtFUS for left (A) and right (B) amygdalae averaged over suprathreshold voxels that were identified by the linear mixed effects model as demonstrating a probabilistic threshold-free cluster enhancement-corrected significant (p < 0.05) main effect of time and no time x task condition interaction. Top row of each panel displays split stringed violin plots of observed mean amygdala activation magnitudes pre- and post-repetitive transcranial focused ultrasound (rtFUS) for each individual averaged over suprathreshold voxels. White dots indicate observed mean values averaged over suprathreshold voxels at pre- (left/red distribution) and post-rtFUS (right/blue distribution), with individual participant observations connected by black lines. Values are separated by task condition (anger, fear, happy, neutral, and shapes). Widest lines directly adjacent to white dots indicate the mean value at that time point, with connecting lines above/below indicating the within-subject 95% confidence interval for the mean based upon the Cousineau–Morey method. Lower row of each panel depicts the individual predicted values from the linear mixed effects model analysis of each subject’s mean activation at pre- and post-rtFUS averaged over suprathreshold voxels, which was run post-hoc to characterize directionality of pre- to post-rtFUS changes. Gray dots indicate individual predicted values at pre- and post-rtFUS, separated on the x-axis by task condition. Black line connects points at the mean of each distribution at each time point. Error bars indicate +/− 1 standard error at each time point. Asterisk(*) with black connecting lines indicates the comparison that was statistically significant, i.e. mean overall change from pre- to post-rtFUS across all conditions. Brain pictures between panels depict the probabilistic threshold-free cluster enhancement-corrected main effects of time within the bilateral amygdalae on the MNI152 ICBM 2009c non-linear asymmetric average brain.

Fig. 4. Condition-specific changes in emotional face processing task-based functional magnetic resonance imaging amygdala activation.

Figure depicts mean individual pre- and post-rtFUS activation values averaged over suprathreshold voxels in the portion of the left amygdala identified by the linear mixed effects model as demonstrating a significant time x task condition interaction. Middle graph displays split stringed violin plots of observed mean left amygdala activation magnitudes pre- and post-repetitive transcranial focused ultrasound (rtFUS) for each individual averaged over suprathreshold voxels. White dots indicate observed mean activation values at pre- (left/red distribution) and post-rtFUS (right/blue distribution), with individual participant observations connected by black lines. Values are separated by task condition (anger, fear, happy, neutral, and shapes). Widest lines directly adjacent to white dots indicate the mean value at that time point, with connecting lines above/below indicating the within-subject 95% confidence interval for the mean based upon the Cousineau–Morey method. Lower graph depicts the individual predicted values from the linear mixed effects model analysis that was run post-hoc to decompose the time x condition effect. Gray dots indicate individual predicted values at pre- and post-rtFUS, separated on the x-axis by task condition. Black line connects points at the mean of each distribution at each time point. Error bars indicate +/− 1 standard error at each time point. Asterisk (*) with black connecting lines indicates the condition showing a significant change from pre- to post-rtFUS. Brain pictures on top row depict the probabilistic threshold-free cluster enhancement-corrected time x condition interaction effect within the left amygdala displayed on the MNI152 ICBM 2009c non-linear asymmetric average brain. Suprathreshold voxels are outlined in black with a linear fade applied to subthreshold voxels as a function of distance from significance threshold.

Assessing relationships between amygdala activation changes and MASQ-GD change

We conducted an exploratory analysis to examine if MASQ-GD symptom change trajectories were related to changes in amygdala activation using a LME model. For the main effect of time, we calculated average activation for each participant in each amygdala effect from the LME across task conditions and calculated individual changes scores (pre vs. post; positive values indicating greater amygdala attenuation). We then tested if left and right amygdala time-related effects demonstrated an interaction with changes in MASQ-GD scores from pre- to post-treatment. The time x activation change effect was significant for the right (F_1,29 = 8.36, p = 0.007) but not left amygdala (F_1,34 = 1.47, p = 0.234). Visualization of symptom change trajectories with a median split by activation change (greater or lesser attenuation) revealed that individuals showing greater pre- to post-treatment attenuation demonstrated greater pre- to post-treatment MASQ-GD reductions (Supplementary Fig 7). Pattern of association was consistent in the left amygdala but not statistically significant. In a separate LME analysis to rule out pre-treatment brain activation prediction of baseline MASQ-GD scores or symptom change trajectories as driving the aforementioned effect, we observed that neither the pre-treatment brain activation x time interaction effect (F_1,22 = 1.748, p = 0.200) nor the main effect of pre-treatment brain activation (F_1,41 = 0.021, p = 0.887) were significant, suggesting the effect indexes a relationship between change in right amygdala activation and change in MASQ-GD scores. We also undertook this analysis for the left amygdala time x condition effect driven by attenuated activation to angry faces. However, individual left amygdala attenuations to anger (F_1,33 = 0.004, p = 0.952) and anger vs. shapes (F_1,32 = 1.480, p = 0.233) did not interact with magnitude of MASQ-GD changes.

Discussion

Here, we examined tFUS/fMRI target engagement and daily rtFUS to the left amygdala in MATRD patients. This study produced four primary findings. First, active vs. sham tFUS to the left amygdala reduced, on average, BOLD signal and modulated activation and connectivity in interconnected limbic and prefrontal circuitry, with patient-related differences in evoked effects in the contralateral (right) hippocampus and insula. Second, rtFUS appears safe and feasible as an intervention approach. Third, rtFUS to the left amygdala using parameters, frequency, and duration/intensity employed here demonstrates promise as a potential transdiagnostic MATRD intervention. Fourth, we observed attenuations in amygdala activation to naturalistic emotional stimuli post-rtFUS, and exploratory analyses indicate greater right amygdala attenuations were associated with greater reductions on the primary outcome. Together, findings provide initial evidence that rtFUS may be a safe and promising method to directly modulate subcortical brain function for improved mental health.

Our double-blind, sham-controlled tFUS/fMRI findings provide solid evidence of target engagement. There was noticeable variability in magnitude of active tFUS effects on left amygdala BOLD signal, though ~66% of the sample showed some deactivation. This variability may stem from individual factors such as degree of tFUS displacement due to skull thickness/curvature/composition, fidelity of targeting, degree to which tFUS effects on BOLD signal adhered to the block design regressor (i.e. near-instantaneous on/off effects with little carryover), or (in patients) possibly individual variation in symptom severity. Exploratory analyses suggested that patients showing the greatest target deactivation had higher levels of symptoms on the primary outcome. This raises the possibility that, here, higher MASQ-GD scores might be indexing a pre-stimulation brain state characterized by greater resting amygdala activity, thereby providing the opportunity for a larger reduction. Additional characterization of baseline resting brain state with perfusion or metabolism measures may provide useful information in future studies to test this hypothesis. We note it is possible that variability in targeting accuracy/energy delivery may have induced off-target effects in the nearby left hippocampus, adjacent to the left amygdala target, that produced the observed active vs. sham tFUS hippocampal main effect. However, the amygdala and hippocampus are highly interconnected structurally [86] and functionally [87], which is also consistent with the notion that the hippocampal effect is a downstream consequence of left amygdala modulation. Additional causal modeling may be helpful to better support one or the other hypothesis, though a temporally and spatially precise method of neuronal recording (such as intracranial depth electrodes) in both structures (where time-lagged analyses could produce findings with a high degree of confidence) would be needed to definitively tease apart the possible contributions.

tFUS/fMRI administration also evoked significant patient-related differences in the right insula and hippocampus, contralateral to the targeted left amygdala. These findings thus highlight the potential utility of tfUS/fMRI as a causal probe of interregional subcortical connections (cf. TMS/fMRI for cortical connections [88]), which might be used to map dysfunctional subcortical circuitry to guide discovery and therapeutic development. Since we did not have a priori hypotheses for patient-related response abnormalities evoked with left amygdala tFUS, these findings should be considered preliminary and in need of replication. Future studies should continue to characterize the nature, directionality, and magnitude of tFUS-evoked patient abnormalities and how effects of these causal perturbations may change over time with rtFUS and other treatments. Of further note is the distinct contralateral segregation of patient-related abnormalities (all in the non-targeted right hemisphere), while active vs. sham tFUS main effects were primarily ipsilateral. Alterations in interhemispheric functional/structural connectivity (i.e. connectivity between the same structures in both hemispheres) have previously been reported in anxiety, depression, and PTSD [89–91]. The current data thus raise the possibility that contralaterality of abnormal left amygdala tFUS-evoked effects in MATRDs may be indicative of perturbed interhemispheric connections. As structural and functional connectivity have both been shown to moderate magnitude of TMS/fMRI effects on BOLD signal in interconnected structures [27, 28, 92], future studies should assess whether baseline functional/structural connectivity from the tFUS target to interconnected structures is associated with tFUS-evoked effects and whether these relationships are perturbed in MATRDs.

The rtFUS pilot trial provides initial evidence of safety, feasibility, and possible utility of daily rtFUS as a transdiagnostic intervention. We observed a significant reduction on our primary outcome, a general measure of negative affect symptoms in MATRDs. The effect was statistically significant, though smaller in magnitude, following a single active and sham administration in the MRI scanner. Following the entire treatment course, the effect size was moderate-to-large for the primary outcome as well as for several secondary outcomes that survived multiple comparisons correction, including depression, anxiety, and PTSD symptom severity. Given the single-arm design, we cannot determine an unbiased effect size of rtFUS apart from placebo/expectancy effects. As a proof-of-concept study, these findings support conduction of future double-blind randomized controlled trials.

A prior single-arm study in treatment-resistant GAD [55] administered the same protocol once weekly and observed a significant reduction in anxiety symptoms with a large effect size. It remains unclear whether the daily sessions employed here were necessary to produce the observed magnitude of effects or if less frequent administrations would have produced comparable results. Given the absence of longitudinal follow-up in this and the prior study [55], it is also unclear whether dosing strategy may impact durability of therapeutic effects. Optimal dosing/duration for rtFUS remains unknown, and this needs to be addressed in future studies.

The tFUS safety profile observed here is favorable, consistent with a recent summary [50] and systematic review [51]. There were no serious AEs in this study nor in prior investigations. Most AEs occurred during in-scanner sonication, with only 2 AEs occurring in the treatment portion. This may suggest that AEs may be more likely during the first tFUS administration and diminish with repeated administrations (similar to AEs with rTMS [93]), or conversely/in addition, that there is an interaction between the MR scanner environment and tFUS administration that results in a higher likelihood of AEs. This could also be an artifact of a different form of AE solicitation in the tFUS/fMRI vs. rtFUS pilot trial portions. Following tFUS/fMRI sessions, initial queries for AEs were incorporated into a written measure that also assessed blinding guesses/confidence, while in the rtFUS pilot trial AEs were solicited by verbal query by study personnel (and followed up by the written questionnaire if endorsed). Since this was an initial pilot trial to examine broad overall safety for future studies, we did not use a more lengthy/time-consuming but rigorous standardized AE rating scale, as is typically employed in clinical trials, to reduce participant burden and maintain study engagement. Future studies with first tFUS administration not in the MR scanner and using a uniform/standardized method of solicitation throughout will inform this point.

Imaging findings also provide initial mechanistic evidence that rtFUS may induce sustained amygdala emotional reactivity attenuations (tempered by absence of a comparator group). There was an anger-specific attenuation effect in the left amygdala (within the basolateral division, by standardized atlases [94, 95]), which was directly adjacent to the tfUS/fMRI target engagement finding. This raises the possibility that, although the most prominent effects were in both the targeted (left) and non-targeted (right) contralateral structure irrespective of emotion type (perhaps reflecting a mix of tFUS-related and tFUS-unrelated effects), the more subtle emotion-specific rtFUS effect in the targeted region may more prominently index a tFUS-induced attenuation. Consistent with this, amygdala hyperactivity to angry faces has been demonstrated in MDD [96, 97], GAD [98], PTSD [99], and SAD [100–102]. However, we also note that the angry face condition evoked the largest level of baseline task activation in this region, consistent with the known engagement of the amygdala by angry faces [6, 103]. This raises the possibility that specificity of change is also a function of increased detection power facilitated by baseline anger-specific elevated activation. A sham-controlled trial is needed to differentiate these possibilities. Finally, we observed that reductions on the MASQ-GD were associated with activation attenuations in the right amygdala but not left amygdala target, again highlighting a possible contralaterality of effects in patients (cf. tFUS/fMRI-evoked differences). This was an exploratory analysis, and effects should be interpreted with caution. However, we postulate if tFUS/fMRI contralateral differences reflect perturbed interhemispheric connections in patients, this brain-symptom change relationship may indicate that those with these connections more intact (which may manifest as greater attenuations in right amygdala activation, perhaps induced through contralateral connections with the left amygdala target) may have benefitted most from this treatment modality, with right amygdala attenuation magnitudes possibly reflecting a mechanistic signature of this process. This remains strictly a hypothesis that may warrant further study.

Several limitations warrant consideration. First, we chose a non-traditional primary outcome measure of negative affect (MASQ-GD) that lacks established benchmarks for clinically significant/reliable change in patient populations. We provide initial estimates of these metrics, but these should be interpreted with caution. The MASQ-GD was chosen to maximize sensitivity to rtFUS effects in a transdiagnostic sample, but this also renders comparisons difficult to effects of established interventions often measured with diagnosis-specific outcome measures. Second, the sample size is relatively small for an intervention study. Third, distribution of MATRD diagnoses was not uniform, and we could not conduct diagnosis-specific sensitivity analyses across the entire spectrum of MATRDs. Fourth, the rtFUS portion lacked blinding and a sham condition that could differentiate the treatment effect from those related to expectancy and knowledge of treatment receipt. Thus, effect sizes observed here cannot be extrapolated to the true treatment effect size. Fifth, we employed self-report measures rather than clinician-administered outcome measures to accommodate the transdiagnostic nature of this study and to mitigate participant burden of multiple structured clinical interviews measuring different symptom outcomes. Although clinician and patient-reported outcomes have been found to substantially overlap [104] and to typically support identical conclusions [105], future studies would benefit from collecting both. Sixth, we did not use acoustic modeling tools [106] (not available at the time of study conduction) to guide our targeting approach, which incorporate information on individual head size/shape and skull thickness/curvature to provide individualized models of energy delivery. The employment and validation of such tools will likely enhance the precision and consistency of energy delivery to the desired brain structure, which may serve to enhance the modulatory potency of this approach.

Author contributions

GAF conceptualized the study and acquired funding. Data collection was supported by LKE, AD, RK, and GAF. Analyses were undertaken by GAF, MKD, and BRB. The initial manuscript was drafted by BRB and GAF. BRB, LKE, AD, RK, MKD, CBN, GAF reviewed and edited the manuscript and provided final approval of the manuscript.

Data availability

Deidentified data are available upon reasonable request from the corresponding author.

Competing interests

GAF has received funds for consulting from Synapse Bio AI and Alto Neuroscience and owns equity in Alto Neuroscience. MKD has received consulting funds from VCENNA. CBN has received funds for consulting from ANeuroTech (division Anima BV), Abbott Laboratories, Signant Health, Janssen Research and Development LLC, BioXcel Therapeutics, Silo Pharma, Engrail Therapeutics, Clexio Bioscience LTD, EcoR1 Capital LLC, EmbarkNeuro (formerly AncoraBio), Galen Mental Health LLC, Goodcap Pharmaceuticals Inc., ITI Inc., LUCY Scientific Discovery, Relmada Therapeutics Inc., Sage Therapeutics Inc., Senseye Inc., Precisement Health, Autobahn Therapeutics Inc., EMA Wellness, Heading Health LLC, Synapse Bio AI, Ninnion, Pasithea, and Skyland Trails. CBN owns stock/equity in Corcept Therapeutics Company, EMA Wellness, Precisement Health, Relmada Therapeutics Inc., Signant Health, Galen Mental Health, and Senseye Inc. CBN is on the steering/advisory board for ANeuroTech (division Anima BV), Signant Health, Laureate Institute for Brain Research (LIBR) Inc., Galen Mental Health LLC, Heading Health LLC, Pasithea Therapeutics Corp., Sage Therapeutics Inc., Senseye Inc., and Skyland Trails. All other authors report no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41380-025-03033-w.