A double-blind replication attempt of offline 5Hz-rTUS-induced corticospinal excitability Authors: Po-Yu Fong, Benjamin R Kop, Carys Evans, Vidya Gani Wijaya, Yongling Lin, Drew Cappotto, Jenny SA Lee, Anna Latorre, Joy Song, Bradley Treeby, Eleanor Martin, John Rothwell, Lennart Verhagen, Sven Bestmann Published: 2025 ABSTRACT AbstractTranscranial ultrasound stimulation (TUS) is a promising new form of non-invasive neuromodulation. As a nascent technique, replication of its effects on brain function is important. Of particular interest is offline 5 Hz repetitive TUS (5 Hz-rTUS), originally reported by Zeng et al. (2022) to elicit lasting corticospinal excitability increases, with large effect sizes. Here, we conducted a pre-registered (https://osf.io/p5n4q) replication of this protocol that benefited from three additional features: double-blind application of TUS, neuronavigation for consistent TMS positioning, and individualised 3D acoustic simulations to assess M1 target exposure to TUS. Changes in resting motor thresholds (rMT), motor-evoked potential (MEP) amplitude, short-interval intracortical inhibition (SICI), and intracortical facilitation (ICF) in response to TUS (5 Hz-rTUS vs. sham) were measured in the right first dorsal interosseous (FDI), abductor digiti minimi (ADM), and abductor pollicis brevis (APB) muscles. Transducer location was determined by the TMS-hotspot for motor representations of the right FDI, as in the original work. No significant effects of 5 Hz-TUS (vs. sham) were observed. Post-hoc simulations showed considerable variability of the acoustic focus, which was outside the anatomical M1-hand area in 67% of participants—in line with the known poor correspondence of TMS-hotspot location and M1-hand area. Our results indicate that the effect sizes of the neuromodulatory effects of 5 Hz-rTUS on M1 may be more variable than previously appreciated. We suggest that double-blinding, neuronavigated TMS, individualised acoustic simulations for TUS targeting and pre-registration will aid reproducibility across studies.Keywords: transcranial ultrasound stimulation (TUS), neuromodulation, corticospinal excitability, neuronavigation Introduction Transcranial ultrasonic stimulation (TUS) is a relatively novel technique for non-invasive neuromodulation in humans (Legon et al., 2014). As a nascent approach for neuromodulation, replication of the effects of TUS is a critical ingredient in progressing toward a mature technology with clinical utility. Here, we focus on 5 Hz-rTUS (also referred to as theta-burst TUS; tbTUS), an offline TUS protocol that has been reported to elicit strong facilitation of corticospinal excitability (CSE), with large effects sizes, that outlasted sonication by up to 30 min (Zeng et al., 2022). The same research group has subsequently replicated these excitatory effects in both healthy and clinical populations (Ding et al., 2024; Grippe et al., 2024; Samuel et al., 2022, 2023; Shamli Oghli et al., 2023; Xia et al., 2024; Zeng et al., 2024). However, the opposite neuromodulatory effects of 5 Hz-rTUS were recently reported by an independent research group, instead showing inhibition (not excitation) of CSE lasting up to 30 min following sonication (Bao et al., 2024). In contrast with targeting M1 based on the TMS-hotspot location and only at a fixed depth of ~30mm (Zeng et al., 2022), Bao and colleagues personalised TUS application with acoustic simulations to ensure precise targeting within M1 in each participant. The inhibitory effects were observed when sonicating either the lip of precentral gyrus or deeper sections of M1. Simulations also provided individual estimates of the acoustic intensity in the brain, as opposed to using measurements in free-water which cannot account for individual variance in absorption and refraction of ultrasound through the skull (Aubry et al., 2022; Chen et al., 2023). Given the contrasting outcomes of 5 Hz-rTUS reported by separate research groups, and a dearth of independent replication, at this early stage of offline human TUS application it is essential to obtain better estimates of the likely effect sizes of 5 Hz-rTUS and factors that may influence these. This necessity is further underscored by the intended application of these protocols in clinical populations (Grippe et al., 2024). In the present study, we therefore conducted an independent replication of the methodology initially reported by Zeng et al. (2022). We specifically focused on replicating the effect of 5 Hz-rTUS on CSE, short-interval intracortical inhibition (SICI), and intracortical facilitation (ICF). We employed the same targeting procedures, stimulation protocol, and outcome measures as Zeng et al. (2022). However, to guard against biases and additional sources of variance, we incorporated double-blinding, neuronavigation of TMS, more repetitions per condition, quantification of MEPs from adjacent hand muscles, post-hoc acoustic simulations of TUS, and pre-registered the study. Materials and Methods Unless stated otherwise, all elements of Experiment 2 by Zeng et al. (2022) were adhered to. Design Participants attended two sessions (1 week apart) with either 5 Hz-rTUS or sham-TUS, in counterbalanced order across participants (Fig. 1A-C). In addition to Zeng et al. (2022), we used a double-blind procedure for TUS application to reduce potential experimenter bias. MEP measures (MEP amplitude; SICI; ICF) were recorded at baseline, and 5, 30, and 60 min after 5 Hz-rTUS or sham (T5, T30, T60; Fig. 1D). Each session was conducted by two experimenters—one operating the TMS and TUS, and the other assisting with neuronavigation and stimulation equipment control. Experimental design and methodology. (A) TMS-elicited motor-evoked potentials (MEPs) were measured from the primary muscle of interest (FDI) and two adjacent hand muscles. (B) The 5 Hz-rTUS protocol. PTD: pulse train duration; PD: pulse duration; PRI: pulse repetition interval; DC: duty cycle; Isppa_fw: free-water spatial-peak pulse-average intensity; Ispta_fw: free-water temporal-average spatial peak. (C) TUS positioning (Zeng et al., 2022): TMS coil position over the motor hotspot was marked on the scalp (left), with the centre of the coil (middle) used to localise the TUS transducer on the scalp (right). (D) Experimental design and procedure. Participants We studied 15 healthy right-handed individuals (age: 31.3 ± 12; five males; eight Asian, six Caucasian, one African) after written informed consent, all without neurological or psychiatric diseases, no contraindications to brain stimulation, and no medications known to affect brain excitability. The study was approved by the UCL Research Ethics Committee (14233/003) and conducted in accordance with the Declaration of Helsinki. We adhered to Zeng and colleague’s original sample size, which was above the recommended sample size according to power analyses for repeated-measures ANOVA. Based on the effect size in Zeng’s original study (η2 = 0.602), G*Power v.3.1 (Faul et al., 2007) yielded a recommended sample size of six. This analysis was based on the significant main effect of TUS condition (sham-TUS, 5 Hz-rTUS, rTUS) on normalised MEP amplitudes (F(2,28) = 21.19, p < 0.001) from Experiment 2 of the Zeng study (Zeng et al., 2022). Other parameters included α of 0.05, 0.8 of Power, 2 groups, 4 measurements, 0.5 of correlation among repetitive measures, and 1 for non-spheric correction ε. MRI All participants had existing T1-weighted MRI scans obtained in one of UCL’s neuroimaging facilities, and all consented to re-use of these images for this study. While imaging sequence details varied for this reason, all images were acquired with 1 mm isotropic resolution, covering the whole head. Note that for the simulation conducted in this study, we focus on the location of the acoustic focus, not the actual intensity values, which will not have been systematically influenced by small idiosyncrasies in the MRI sequences across individuals. Transcranial magnetic stimulation TMS was delivered with a Magstim 2002 Monophasic stimulator (single-pulse) and a Magstim Bistim2 system consisting of two 2002 monophasic stimulators (paired-pulse) connected to a D70 alpha F8 coil with an internal diameter of 70 mm (Magstim Co. Ltd, Whitland, Wales). Participants were seated in a comfortable chair with their hands resting on a pillow on their lap. The TMS coil was placed over the left primary motor cortex (M1), tangentially to the scalp, with the handle of the coil pointing backwards at a 45-degree to the midline to induce an approximate posterior-anterior current. The TMS-hotspot was defined as the area of the scalp where the largest and most stable MEPs were observed in the right first dorsal interosseus (FDI) muscle (Rothwell et al., 1999) (Fig. 1A). Apart from the setup in the original work (Zeng et al., 2022), neuronavigation was used to ensure precise and consistent placement of TMS both before and after 5 Hz-rTUS/sham (Brainsight, Rogue Research, USA). We additionally recorded MEPs from adjacent abductor pollicis brevis (APB) and abductor digiti minimi (ADM) muscles, to capture any potential effects of the ultrasound beam on other intrinsic hand muscles (Fig. 1A). Our TMS procedures followed the original study (Zeng et al., 2022). Briefly, after identifying the TMS hand motor hotspot, we determined the baseline rMT and the stimulator intensity (SI) required to elicit a ~1mV MEP (SI1mV) (Groppa et al., 2012; Rothwell et al., 1999). This baseline SI1mV was the intensity for all subsequent single-pulse MEP blocks to assess changes in CSE. For the paired-pulse TMS block capturing intracortical excitability, the intensity of the conditioning stimulus was set to 80% rMT for SICI (2 ms ISI) and ICF (10 ms ISI), and the test stimulus (TS) was set to the SI1mV measured in the beginning of the same block. Paired-pulse trials were administered in pseudorandomised order. In contrast to Zeng and co-workers’ work (Zeng et al., 2022), we acquired 25 trials for each measurement—originally 20 and 10 trials for each single- and paired-pulse metric, respectively—using an ISI of 5 s and a 10% jitter. All TMS measures were acquired at baseline, T5, T30, and T60 (Fig. 1D). Each TMS block lasted approximately 9 min. Surface electromyography (EMG) signals were simultaneously recorded for target FDI, APB, and ADM with pairs of surface electrodes in a belly-tendon montage (WhiteSensor 40713, AmbuR, Denmark; Fig. 1A). Signals were amplified with a gain of 1000, bandpass filtered (5 Hz–2000 Hz) by a Digitimer D360 amplifier (Digitimer Ltd, Welwyn Garden City, Hers, UK), and digitised at 5000 Hz by a Power 1401 data acquisition interface and Signal software version 7.01 (Cambridge Electronic Design Ltd., Cambridge, UK). Transcranial ultrasound stimulation Transcranial ultrasound stimulation (TUS) was delivered using the NeuroFUS system (manufacturer: Sonic Concepts Inc., Bothell, WA, USA; supplier/support: Brainbox Ltd., Cardiff, UK) via a four-element 500 kHz annular array piezoelectric transducer with a 64 mm radius of curvature and aperture diameter (CTX-500-025, Sonic Concepts Inc., Bothell, WA, USA). A four-channel radiofrequency amplifier (Transducer Power Output system; TPO) powered the transducer via a four-channel electrical impedance matching network utilizing a rectangular pulse shape. The transducer was coupled with a 10 mm gel pad (Aquaflex, Parker, Laboratories, NJ, USA). Ultrasound gel (Aquasonic 100, Parker Laboratories, NJ, USA) was centrifuged to remove visible bubbles and applied between the gel pad and the transducer. We defined our sonication depth as 33 mm in accordance with the acoustic field peak reported previously (Grippe et al., 2024; Zeng et al., 2022). To reach 33 mm, the sonication depth on the TPO was set at 43.5 mm to account for the additional distance added by ultrasound gel (0.5 mm) and gel pad (10 mm). Table 1 and Supplementary Information 1 (Fig. S1) show the comparison of the acoustic intensities and profiles between Zeng et al. (2022) and the current study. The applied focal depth in the present study and the original study (Zeng et al., 2022) matches the scalp-to-cortex distance of ~30mm for the M1 omega formation. Table 1.The details of the intensities in the present study and the original experiment by Zeng et al.StudyTransducerPowerIsppa_fwIsppa_tcSimulated Isppa_tc Zeng et al. (2022)H246†20 W*9.04 W/cm22.26** W/cm2-Present studyCTX500-025†3.71 W*10.0 W/cm22.50** W/cm21.20 ± 0.43 W/cm2†Transducers from sonic concepts (Bothell, WA, USA).*Power levels required to reach an equivalent Isppa differ for different transducers. Higher power does not suggest that ‘more stimulation’ was applied.**Transcranial Isppa (Isppa_tc) is estimated by Zeng et al. (2022) by applying 75% attenuation from free-water Isppa (Isppa_fw). The details of the intensities in the present study and the original experiment by Zeng et al. Transducers from sonic concepts (Bothell, WA, USA). *Power levels required to reach an equivalent Isppa differ for different transducers. Higher power does not suggest that ‘more stimulation’ was applied. Transcranial Isppa (Isppa_tc) is estimated by Zeng et al. (2022) by applying 75% attenuation from free-water Isppa (Isppa_fw). The 5 Hz-rTUS protocol was an 80-s train of 20-ms ultrasound pulses repeated every 200 ms (PRF 5 Hz; 10% duty cycle), for a total of 400 pulses (Fig. 1B). The spatial-peak-pulse average intensity in free water (Isppa_fw) was set to 10 W/cm2 (Table 1), similar to the original work. Applying the same 75% skull attenuation relative to free-water measurements, as was done in the original study, our estimated transcranial Isppa_tc (2.5 W/cm2) was consistent with the original study (2.26 W/cm2) (Zeng et al., 2022). However, skull attenuation varies under different settings (Attali et al., 2023; Hosseini et al., 2023; Mueller et al., 2017). Therefore, it is important to run individualised acoustic simulations to better estimate transcranial intensity. Based on individualised 3D acoustic simulations, our actual estimated transcranial Isppa_tc was 1.20 W/cm2 ± 0.43. The mechanical index was approximately 0.26 ± 0.05, and maximum temperature rise was about 0.22 °C. The TUS transducer location was determined by the TMS-hotspot, as previously (Ding et al., 2024; Samuel et al., 2022, 2023; Shamli Oghli et al., 2023; Xia et al., 2024; Zeng et al., 2022). The contours of the TMS coil were marked on the scalp using a chinagraph pencil, and the centre of the TMS coil was measured (Fig. 1C). We additionally recorded the TUS transducer location using neuronavigation for post hoc acoustic simulations. We also added a double-blind procedure, where both experimenter and participant were blind to which TUS condition (5 Hz-rTUS vs. sham) was administered, to minimise potential experimenter bias. To enable effective double-blinding, an independent researcher who was not present during testing designed a MATLAB script to randomly assign the TUS condition order across participants. At the beginning of the experimental session, one experimenter entered the participant code and session number into the MATLAB script, which determined whether real or sham TUS was administered. TUS parameters were automatically input into the TPO, and the TPO display was turned away from the participant and experimenters during TUS application. Hair preparation with centrifuged ultrasound gel was initially performed after threshold and hotspot estimation, and finalised after the baseline TMS block to minimise preparation time between baseline TMS and TUS application. During both 5 Hz-rTUS and sham, Gaussian white noise was played through bone-conductive headphones (Sportz3, AfterShokz, New York, USA) to maximise blinding of each condition (Braun et al., 2020). Participants selected the maximum volume for the white noise that they found acceptable. The Gaussian white noise was programmed with MATLAB using the randn() function to generate an auditory stimulus with a 44100 Hz sampling rate and 82.3 s of sound duration. Data preprocessing and analysis Raw EMG data were exported from Signal (version 7.01; Cambridge Electronic Design, UK) to MATLAB (version 9.7.0; R2019b). Peak-to-peak MEP amplitudes for each muscle (FDI, APB, ADM) were calculated using a custom script. Data and code to reproduce the results are provided here: https://doi.org/10.17605/OSF.IO/S5AG6. Within each block, trials identified as significant outliers using Grubbs’ test (1.61%) and trials with pre-contraction (1.78%) were excluded. Precontraction was defined as the root mean square (RMS) of EMG activity in the 100 ms prior to TMS exceeding the block’s average by two standard deviations. All trials with an RMS exceeding a liberal threshold of 0.045 were manually inspected (n = 10427/36000; 29%). Trials where noise in the EMG signal prevented sufficient quantification of MEP amplitude were excluded (0.55% across all muscles, 0% for FDI specifically). Paired t-tests were conducted to assess baseline differences between 5 Hz-rTUS and sham for rMT, SI1mV, MEP amplitude, SICI, and ICF. The MEP amplitudes for paired-pulse measures were expressed as a ratio to the mean TS. To assess main effects and interactions with maximal statistical power, linear mixed models (LMMs) with a maximal random effects structure included full random intercepts and slopes for each independent variable, and interactions were fitted using the lme4 package in R (Barr et al., 2013; Bates et al., 2015) with the bobyqa optimiser from the NLopt library, which was initially identified as the best-performing optimiser using the allFit function. Statistical significance was set at a two-tailed α = 0.05 and computed with t-tests using the Satterthwaite approximation of degrees of freedom. Given the right-skewed nature of MEP amplitudes, trial-level square root corrected MEP amplitudes were used for LMMs. The time course of MEP amplitudes, SICI, and ICF was tested separately with models including TUS Condition (5 Hz-rTUS/sham) and Timepoint (Baseline/T5/T30/T60) as factors. TUS-induced changes in corticospinal excitability were further tested for MEP amplitudes expressed as a ratio to the baseline mean (i.e., baseline corrected), with TUS Condition (5 Hz-rTUS/sham) and Timepoint (T5/T30/T60) as factors. In addition to LMMs, the previously implemented statistical procedures of the original study (Zeng et al., 2022) were replicated using two-way rm-ANOVA on raw data without square root correction, with factors TUS Condition (5 Hz-rTUS/sham) and Timepoint (Baseline/T5/T30/T60; see Supplementary Results). Acoustic simulations To determine the actual anatomical location targeted by TUS, we conducted simulations of acoustic wave propagation for each individual, using k-Plan software, a user interface for the pseudospectral time-domain solver k-Wave (Treeby & Cox, 2010). To generate compatible skull images, a toolbox with pre-trained deep learning convolutional neural networks was used to convert T1-weighted MRI scans to pseudo-CT images (Yaakub et al., 2023). Next, the four-element CTX500 transducer location was imported from the neuronavigation TUS trajectory captured during the real 5 Hz-rTUS session (exported as 3D coordinates in Brainsight coordinate space). The simulation was run using six grid points per wavelength for a single pulse duration to obtain the steady-state pressure field. Using a custom MATLAB script, the k-Plan pressure field and grid settings were extracted using k-plan-matlab-tools (https://github.com/ucl-bug/k-plan-matlab-tools). We resampled the simulation grid to have the same spacing as the anatomical scans (i.e., T1w and pseudo-CT). Pulse average intensity was calculated from the simulated acoustic pressure amplitude p, using the plane wave approximation I = p2/2ρc, where the density ρ was 1000 kg/m3 and the sound speed c was 1500 m/s. We were primarily interested in determining the intersection between the acoustic focus and the anatomical location of the M1 hand area within the vicinity of the precentral gyrus. For this, we first segmented structural MRI scans into different tissue types, using SPM12 (https://www.fil.ion.ucl.ac.uk/spm/software/spm12/). Grey matter, white matter, and cerebrospinal fluid masks were merged to generate a binary brain mask per subject. In native space, we identified the omega formation in the pre-central gyrus at a depth of ~30mm, and the lip of the precentral gyrus at a depth of ~18mm (Bao et al., 2024; Osada et al., 2022). We extracted the location and value of the Isppa for each individual and calculated the Euclidean distance between the Isppa location and these two anatomical landmarks (Fig. 4E). We defined the acoustic focus as the volume where the intensity values are equal to or higher than half of the intensity maximum in the brain, equivalent to the full-width half-maximum intensity. Supplementary Table S2 reports the simulation of free water sonication and thermal simulation results, in line with recent reporting guidelines (Martin et al., 2024). K-Plan was also used to run a thermal simulation for one representative participant for a full 80 s PTD with no cooling time. Targeting precision and offline excitatory effects of 5 Hz-rTUS. (A) Distribution of ultrasound full-width half-maximum simulated acoustic fields across participants, represented in standard space. The anatomical region-of-interest (15 mm M1 sphere) is highlighted in yellow. (B) Dark grey: The percentage of the acoustic focus (i.e., the full-width half-maximum intensity volume) that fell within the M1 ROI volume. This percentage exceeded 20% in only 5/15 participants. Light grey: Maximum ultrasound intensities that fell within the M1 ROI. (C) TMS motor hotspot-based targeting leads to an anteromedial shift of the Isppa. The x-axis depicts the lateral-medial distance, and the y-axis depicts the posterior-anterior distance in millimeters. The grey shaded area represents the distribution of distances, with the mean denoted. (D) Distribution (density) of Euclidean distances in millimeters between the omega formation in the precentral gyrus and the Isppa in native space. (E) The lip/crown of the pre-central gyrus is at a scalp-to-cortex distance of ~18 mm, while the omega formation we targeted in this study is at a depth of ~30 mm. (F) There is no significant inferior shift of the acoustic focus (blue) in relation to the targeted omega formation (black line). However, there is a significant inferior shift relative to the lip/crown of the gyrus (purple). Distances distributions (density) are in millimeters and represent the relative inferior shift to the targeted omega formation (black line). (G) Even when M1 was more accurately targeted (>20% focal overlap with M1 ROI), there were no significant effects of sonication on corticospinal excitability. Data represent MEP amplitude as a ratio to baseline MEP amplitude at each time point. Bars depict means, and error bars depict standard error. (H) No significant association between targeting accuracy (i.e., percentage overlap with the M1 ROI) and MEP amplitude changes (i.e., MEP amplitude as a ratio to baseline MEP amplitude). Points represent individual participants, lines represent linear fits, and clouds represent the 95% CI. For group-level plots, the T1-weighted MRI scans and intensity maps were resampled in MNI standard space at a 1 mm isometric resolution. The binary acoustic focus maps in standard space were summed and overlaid on a standard brain with MRIcroGL. To quantify targeting accuracy through spatial overlap between the acoustic focus and the target, we required a volumetric definition of the M1 target. Therefore, we first delineated M1 with a 15 mm radius spherical region of interest (ROI) that covered both the lip of the precentral gyrus and the omega formation (Supplementary Fig. S2). This ROI was chosen because stimulation of both these areas, and likely intermediate regions as well, can elicit effects of TUS on CSE (Bao et al., 2024). We then calculated the percentage of the acoustic focus that fell within the ROI, as well as the peak ultrasound intensity within the ROI. When no clear omega formation or w-shape could be identified, we set the seed of calculation at the convergence of these two semi-circular shapes of the cortex. Results None of the participants reported discomfort or adverse effects following participation. Participants could not hear 5 Hz-rTUS over the mask, nor could they distinguish between the 5 Hz-rTUS and sham sessions. No significant difference in baseline physiological measures At baseline, no differences were observed for MEP amplitude (t(14) = 0.837, p = 0.417, d = 0.216, Sham: mean (M) = 1.075, standard error (SE) = 0.098; TUS: M = 1.005, SE = 0.091), similar to prior work. There were also no significant differences in baseline paired-pulse measurements for SICI (t(14) = -0.619, p = 0.546, d = -0.160, Sham: M = 0.445, SE = 0.052; TUS: M = 0.473, SE = 0.070) or ICF (t(14) = -0.724, p = 0.481, d = -0.187, Sham: M = 1.303, SE = 0.156; TUS: M = 1.379, SE = 0.091; Supplementary Fig. S3). Moreover, there were no differences between baseline 5 Hz-rTUS and sham conditions for rMT (t-tests; t(14) = 0.811, p = 0.431, d = 0.209, Sham: M = 52.8% MSO, SE = 3.3; TUS: M = 51.9% MSO, SE = 3.4) or SI1mV (t(14) = 1.353, p = 0.198, d = 0.349, Sham: M = 74.0% MSO, SE = 5.4; TUS: M = 72.5% MSO, SE = 5.1; Supplementary Fig. S3), similar to prior work. No significant effect of 5 Hz-rTUS on corticospinal or intracortical excitability To examine the time course of corticospinal excitability changes in the primary muscle of interest (FDI), a linear mixed model was fitted predicting square root corrected MEP amplitude by Condition (5 Hz-rTUS/sham), Timepoint (Baseline/T5/T30/T60), and their interaction. No significant effects were observed in the main effect of Timepoint (F(3,14) = 1.757, p = 0.201, ηp² = 0.273) or Condition (F(1,14) = 1.855, p = 0.195, ηp² = 0.117; Sham: M = 1.060, SE = 0.036; TUS: M = 0.992, SE = 0.030), nor the interaction Timepoint*Condition (F(3,14) = 0.236, p = 0.87, ηp² = 0.048). When expressing MEP amplitude as a ratio to baseline, we similarly found no evidence of neuromodulation (main effect of Timepoint: F(2,14) = 1.712, p = 0.216, ηp² = 0.195; main effect Condition: F(1,14) = 0.687, p = 0.421, ηp² = 0.047, Sham: M = 1.097, SE = 0.038; TUS: M = 1.053, SE = 0.035; interaction of Timepoint*Condition: F(2,14) = 0.013, p = 0.988, ηp² = 0.002; Fig. 2). No significant effect of 5 Hz-rTUS on MEP amplitude. (A) There was no significant effect of 5 Hz-rTUS on single-pulse TMS amplitude between sham and TUS at any time point. MEP amplitudes are expressed as a ratio to baseline for each time point. Points and error bars represent group mean ± standard error. (B) The two plots depict the data at participant-level. Data are the ratio of normalised amplitudes of TS at each time point. As shown in Figure 3, there were also no significant effect of 5 Hz-rTUS for either SICI (Timepoint: F(3,14) = 0.648, p = 0.597, ηp² = 0.12; Condition: F(1, 14) = 0.187, p = 0.672, ηp² = 0.013, Sham: M = 0.651, SE = 0.023; TUS: M = 0.642, SE = 0.024; Timepoint*Condition: F(3,18) = 0.479, p = 0.701, ηp² = 0.075) or ICF (Timepoint: F(3,15) = 0.8, p = 0.513, ηp² = 0.141; Condition: F(1,14) = 2.773, p = 0.118, ηp² = 0.165, Sham: M = 1.103, SE = 0.028; TUS: M = 1.160, SE = 0.020; Timepoint*Condition: F(3,21) = 0.099, p = 0.96, ηp² = 0.014). RM-ANOVAs conducted on raw MEP amplitudes, following the same analyses as Zeng and colleagues, also did not reveal any significant effects (Zeng et al., 2022) (see Supplementary Information 2). For the adjacent APB and ADM muscles, there was also no evidence for offline excitatory effects of sonication (Supplementary Information 3-4, Supplementary Fig. S5-8). Collectively, we did not observe any reliable effect of 5 Hz-rTUS on CSE or intracortical excitability. No significant effect of 5 Hz-rTUS on SICI or ICF. (A) For both SICI and ICF, there are no significant differences between sham and TUS at any time point. Data are the ratio of paired-pulse MEP amplitude to baseline MEP amplitude for each timepoint. Bars and errors bars represent the mean and standard error, respectively. (B) The two plots depict the data at participant-level. Data are the ratio of SICI and ICF at each time point. No significant effect of 5 Hz-rTUS on rMT or the stimulator intensity required to elicit 1mV amplitude MEPs There was no significant effect of sonication on resting motor thresholds (Timepoint: F(3,42) = 2.711, p = 0.057, ηp² = 0.162; Condition: F(1,14) = 0.546, p = 0.472, ηp² = 0.038; Timepoint*Condition: F(3,42) = 0.319, p = 0.812, ηp² = 0.022) nor SI1mV (Timepoint: F(1.6,22) = 0.68, p = 0.485, ηp² = 0.046; Condition: F(1,14) = 0.303, p = 0.591, ηp² = 0.021; Timepoint*Condition: F(2.1,29) = 1.737, p = 0.192, ηp² = 0.11; Supplementary Fig. S9). Variable ultrasound targeting of M1 based on TMS hotspot location It is possible that ultrasound targeting of M1 based on the scalp location of the TMS hotspot is variable, and thereby may reduce the consistency and overall efficacy of 5 Hz-rTUS. To address this, we conducted post-hoc simulations of the sonication target based on individual head models and assessed the degree of targeting variance in our population. First, we observed that previous reports may have overestimated acoustic transmission, where the reported transcranial Isppa (2.26-2.93 W/cm2) was estimated by uniformly applying 75% attenuation from free-water Isppa (9.04-11.72 W/cm2; Table 1) (Samuel et al., 2022; Shamli Oghli et al., 2023; Xia et al., 2024; Zeng et al., 2022). Here, we find a mean±sd transcranial Isppa of 1.2 ± 0.4 W/cm2, corresponding to a ~12% transmission rate, in line with empirically observed and theoretical estimations of percentage intensity transmission at f = 500 kHz (Bao et al., 2024; Chen et al., 2023). Critically, our simulations reveal substantial variability in the location of the acoustic focus across participants (Fig. 4A, B; Supplementary Fig. S4). The acoustic focus overlapped with the M1 ROI in only 7 out of 15 participants. Only 33% of participants had more than 20% of the acoustic focus volume in the M1 ROI. In those 33% of subjects, the maximum intensity within the ROI was 1.0 ± 0.3 W/cm2 (Fig. 4B). In native space, the Euclidean distance from the target omega formation seed voxel to the peak intensity voxel (Isppa) was 21.1 ± 9.5 mm (Fig. 4D). Along each 3D axis, we performed one-sample t-tests and found a significant anteromedial shift of the foci (Fig. 4C; anterior: t(14) = 2.86, p = 0.013; medial: t(14) = 7.36, p = 3.548). The degree of anterior shift (10.5 ± 14.2 mm) and the distribution of Euclidean distances correspond with the well-known anterior shift of the TMS motor hotspot relative to the anatomical hand motor area (Ahdab et al., 2010, 2016; Diekhoff et al., 2011; Kim et al., 2021; Niskanen et al., 2010). We furthermore observed a significant inferior shift relative to the lip of the gyrus, which has a scalp-to-cortex distance of ~15 mm (t(14) = 14.17, p = 1.079). However, there was no inferior shift relative to the targeted omega formation at a depth of ~30 mm (t(14) = 1.63, p = 0.126; Fig. 4E,F). Taken together, our results demonstrate that TMS motor hotspot-based targeting method for rTUS introduces considerable dispersion around the intended target M1 location. While such misalignment between the hotspot scalp location and underlying anatomy is known, it quantifies the degree of bias and variability in directing sonication reliably to the same brain target with this procedure. Lack of excitatory offline 5 Hz-rTUS effects when anatomical targeting is accurate 5 Hz-rTUS might have been effective in participants in whom sonication was reliably directed to the presumed anatomical target region. We, therefore, identified those participants where the acoustic focus overlapped with the M1 ROI by >20%. However, in none of the 5 participants meeting this criterion did we observe an excitatory effect of 5 Hz-rTUS on single-pulse MEP amplitude (Fig. 4G). Given the previously reported effect sizes (Zeng et al., 2022), one may expect these participants to show an excitatory effect; this was not observed here. Finally, it is conceivable that there is a systematic relationship between the targeting accuracy and the sonication effect. Using linear models, we found no significant relationship between the percentage acoustic focus overlap and the ratio of post-TUS MEPs to baseline MEPs for any timepoint (Fig. 4H; T5: b = -0.002, t(14) = -1.483, p = 0.162; T30: b = -4.194, t(14) = -0.175, p = 0.864; T60: b = 6.948, t(14) = 0.342, p = 0.738). In sum, even when sonication is directed to M1 as intended, the excitatory effects of 5 Hz-rTUS did not replicate. Discussion In this pre-registered replication study, we did not observe a significant effect of 5 Hz-rTUS on corticospinal or intracortical excitability. This result suggests effects sizes of 5 Hz-rTUS can be more variable than suggested by previous work, and we highlight potential factors influencing these discrepancies. Neuromodulation with TUS holds promise for clinical interventions, due to its spatial precision and potential for targeting deeper brain structures. Replication is an essential process for scientific rigor and has been earmarked by the International Transcranial Ultrasonic Stimulation Safety and Standards (ITRUSST) as crucial for accelerating TUS toward an effective neuromodulation approach. There is, indeed, growing attention to the rigorous experimental control required to demonstrate replicability in the field of focused ultrasound (Martin et al., 2024; Murphy et al., 2024; Nandi et al., 2024). This need is further highlighted by examples from the fields of electrical and magnetic stimulation, where initial reports of novel stimulation protocols have often been followed by more nuanced appraisals of their efficacy (Jannati et al., 2017; Wiethoff et al., 2014). In the present study, we sought to replicate recently published neuromodulatory effects of offline 5 Hz-rTUS directed to M1 (Zeng et al., 2022). Our results did not reveal a significant effect of 5 Hz-rTUS on corticospinal or intracortical excitability, or any of the previously reported effect sizes. This contrasts with the 14 out of 15 participants showing enhanced corticospinal excitability in response to 5 Hz-rTUS in the original study (Zeng et al., 2022), and the similarly large effect sizes of the same protocol in subsequent studies by the same group (Ding et al., 2024; Grippe et al., 2024; Samuel et al., 2022, 2023; Shamli Oghli et al., 2023; Xia et al., 2024; Zeng et al., 2022, 2024). However, we also note some moderate effect sizes in these comparisons in the current study. These moderate effect sizes without statistical significance may be attributed to the small sample sizes, with inherent variability (Gelman & Stern, 2006; Lakens, 2013). Variable effects of 5 Hz-rTUS The absence of significant effects held for the targeted FDI and the adjacent APB and ADM muscles. Here, we initially assessed our results using linear mixed models, which have greater statistical power than RM-ANOVAs. However, our null results were the same when employing the statistical procedures as in the original study (Zeng et al., 2022). Additional qualitative assessment of individual data indicated that in only 2 out of 15 participants of the current study were the observed MEP changes consistent with the originally reported effects at T5 and T30. These results suggest the effects of 5 Hz-rTUS protocol across different cohorts are likely more nuanced and less robust than originally appreciated. This conclusion aligns with a recent replication from another group showing the opposite, that is, inhibitory rather than excitatory effects lasting approximately 30 min after offline 5 Hz-rTUS (Bao et al., 2024). How the same TUS protocol directed to the same neural structure with the same overall approach can lead to completely opposite neuromodulatory effects is currently unclear. One major difference was that Bao et al. (2024) targeted TUS based on structural landmarks with a priori simulations at two focal depths—the lip/crown of the pre-central gyrus and one targeting the deeper, omega-shaped formation of the pre-central gyrus. Inhibitory effects of 5 Hz-rTUS were observed at both stimulation depths. While one cannot rule out that relatively subtle differences in applied sonication intensity between the two studies explain the opposite effects on excitability, a complete reversal of the neuromodulatory effects with such subtle intensity variation raises concerns about the clinical utility of such protocols, and certainly provides a mandate for further independent replication. Regardless of the mechanistic explanation for how 5 Hz-rTUS at very low intensities might elicit such strong neuromodulatory effects, our findings here and those by Bao et al. (2024) suggest that across different cohorts, the effects of low-intensity 5 Hz-rTUS are more variable and potentially even orthogonal to the ones initially reported. Variability in targeting with TMS motor hotspot-based transducer placement One factor that will determine the effects of TUS is the precision and consistency of its targeting. To target the primary motor cortex in combined TUS-TMS experiments, the TMS motor hotspot is commonly used as a heuristic to determine the TUS transducer location on the scalp that sonicates M1 (Fomenko et al., 2020; Kop et al., 2024; Legon et al., 2018; Xia et al., 2021; Zeng et al., 2022). However, this method may lead to poor target exposure to TUS. First, when using TMS over M1, the location and thresholds of the neural activation depend on several factors, including cortical folding, tissue type, and induced current direction (Jung et al., 2010; Lefaucheur & Picht, 2016). Even when TMS evokes large and stable MEPs, the axial projection from the coil centre does not necessarily correspond to the site of relevant excitation in the cortex. Moreover, the location of the maximum electric field (E-field) in the brain can deviate up to 1.4 cm from the centre of a figure-of-eight coil (Gomez-Tames et al., 2018; Gomez et al., 2021). Future work assessing the relationship between the sonication beam, and characteristics of the TMS-induced electric field, including the E-field amplitude and direction of the induced current in cortex could assess whether E-field estimates provide suitable markers for targeting. However, the direct axial projection from the geometrical coil centre (marked on the scalp) seems unlikely to be a reliable guide for targeting with TUS. Secondly, considering that the width of the sonication beam is considerably smaller than the spatial specificity of TMS (Fomenko et al., 2020; Legon et al., 2018), even small deviations of the TUS transducer in the axial plane will lead to mis-targeting. Compared to TMS, where an altogether much larger region of the brain is targeted, this will increase the probability of directing sonication to different cortical targets across individuals. Indeed, our 3D individualised post-hoc acoustic simulations demonstrate such inter-individual variability in the location of the acoustic focus relative to M1, where only 5/15 of participants showed a substantial degree of overlap between sonication and the M1 target. Scalp-based transducer placement can thus introduce inter-individual variation in the specific targeted pre- or postcentral elements. Such variation aligns with the known complexity of mapping the TMS-hotspot on the scalp to a specific anatomical target, for example, a specific section of the omega-shaped hand knob (Ahdab et al., 2010, 2016; Diekhoff et al., 2011; Kim et al., 2021; Niskanen et al., 2010). Interestingly, the only study that used personalised targeting to eliminate this variability reported inhibitory effects of 5 Hz-rTUS (Bao et al., 2024). Neuronavigation and double-blinding may explain failure to replicate While differences in transducer model may come to mind as a possible explanation for the different findings (Supplementary Fig. S1), we would argue that the collapse from neuromodulatory effects in 14/15 participants (Zeng et al., 2022) to non-significance cannot be explained solely by a difference in transducers. The axial intensity profiles between our four-element transducer and the two-element transducer used previously (Zeng et al., 2022) are similar, apart from a near-field peak at 12 mm for the original two-element transducer that is too superficial to reach the brain (Supplementary Fig. S1). If the neuromodulatory effects of 5 Hz-rTUS were, indeed, entirely dependent on these very small variations in the pressure fields, it would require extraordinary precision in targeting the same neural structures consistently in every subject. This seems unlikely. The key differences between the original work (Zeng et al., 2022) and the present study were the inclusion of TMS neuronavigation and double-blinding. Using only markings on the scalp, precise TMS repositioning is challenging, particularly in terms of orientation. Without neuronavigation, slight trial-by-trial shifts in TMS position can bias MEP amplitudes (Lefaucheur & Picht, 2016; Ruohonen & Karhu, 2010). Further, the approach is prone to circular reasoning if the MEPs—the dependent variable—are themselves used to determine repositioning of the TMS coil. In combination with unblinded researchers, the lack of neuronavigation and double-blinding risks introduction of unconscious confirmation bias. The changes to the original work we introduced sought to minimise such bias. It would seem prudent to suggest that inclusion of neuronavigation for TMS positioning and TUS targeting, ideally based on a priori acoustic simulations (Murphy et al., 2025), and double-blinding should become standard ingredients of future TUS-TMS investigations (Bao et al., 2024; Darmani et al., 2025; Murphy et al., 2025; Oh et al., 2024; Riis et al., 2024; Zhang et al., 2022). The applied stimulation intensity, or depth of target was not obviously different in our study, but participant demographics, or preparation of transducer-skull coupling may additionally contribute to the variability across studies. Where no detail was provided in the original study by Zeng et al. (2022), we followed the procedures outlined by ITRUSST (Murphy et al., 2025), and going forward, adherence to such guidelines will allow for a better comparison across studies. The present study sought to independently replicate the original experiment (Zeng et al., 2022), but some limitations remain. Firstly, the actual stimulation intensity used in the original experiment cannot be determined with certainty. As Table 1 shows, the Isppa_fw applied by Zeng et al. is not specified. Instead, an estimate of 2.26 W/cm2 for the transcranial Isppa (Isppa_tc) after 75% skull attenuation was reported. The use of 75% skull attenuation is explicitly described in later publications (Grippe et al., 2024; Shamli Oghli et al., 2023; Zeng et al., 2024), and Zeng et al. (2022) refer readers to their acoustic beam properties as the basis for deriving in-situ values. On this basis, the 2.26 W/cm² Isppa_tc reported in Zeng et al. (2022) corresponds to a free-water intensity of approximately 9.04 W/cm² (i.e., 2.26 divided by 0.25). In the present study, we applied a similar Isppa_fw of 10 W/cm2. A subsequent study by Zeng et al. (2024) reports 20 W/cm2 ‘acoustic intensity’ to also reach 2.26 W/cm2 Isppa_tc. However, when applying a 75% attenuation factor, 20 W/cm2 would correspond more closely to 5 W/cm2 Isppa_tc. One possible explanation is a reporting error, in which acoustic power (20 W) was expressed as acoustic intensity (20 W/cm2). Therefore, while some uncertainty remains, the available evidence supports the accuracy of the present study’s estimated stimulation intensity. Ambiguities in reporting Isppa_fw and Isppa_tc across studies highlight the challenges of reproducibility and the importance of standardised reporting guidelines (Martin et al., 2024). Nevertheless, the assumption of a fixed 75% skull attenuation for all participants is a limitation of any TUS study (Zeng et al., 2022). The skull absorption for sonication varies considerably across individuals (Murphy et al., 2025). However, as our primary goal was to replicate the previous work as closely as possible, we assumed the same skull attenuation value (75%). We note that our estimated transcranial Isppa derived from individual post-hoc acoustic simulations differs substantially from the estimate obtained by applying a fixed attenuation value and varies across individuals. Going forward, personalised assessment of skull absorption to sonication should be considered, as indeed advocated for by the ITRUSST guidelines (Murphy et al., 2025). Novel approaches to neuromodulation require realistic assessment of their efficacy, to develop methodological standards, direct further development, and ultimately help accelerate clinical use. A key component for this is independent replication. Our replication results here reappraise the effect sizes of excitatory neuromodulation of 5 Hz-rTMS to M1 previously reported but also suggest avenues toward consistent and reproducible evaluation of novel TUS protocols. Supplementary Material Acknowledgements We thank Paul Hammond for their help and assistance. Data and Code Availability Data and code to reproduce the results reported in this study are available at: https://doi.org/10.17605/OSF.IO/S5AG6 Author Contributions Po-Yu Fong: investigation, conceptualisation, methodology, data curation, writing—original draft, writing—review & editing, Resources, and Supervision (the investigation team). Benjamin Kop: conceptualisation, methodology, software, formal analysis, data curation, writing—original draft, writing—review & editing, and visualisation. Carys Evans: conceptualisation, methodology, writing—original draft, writing—review & editing, and visualisation. Vidya Gani Wijaya: investigation. Yongling Lin: investigation. Drew Cappotto: investigation. Jenny S. A. Lee: investigation. Anna Latorre: investigation. Joy Song: investigation. Bradley Treeby: Resources, writing—review & editing. Eleanor Martin: Resources, writing—review & editing. John Rothwell: Funding acquisition, Resources, and Supervision. Lennart Verhagen: Methodology, Writing—review & editing, Funding acquisition, Conceptualisation, and Supervision. Sven Bestmann: Supervision, conceptualisation, methodology, writing—original draft, and writing—review & editing. Funding This work was funded by Medical Research Council UK (Grant Reference No. MR/P006671/1 to J.C.R), Nederlandse Organisatie voor Wetenschappelijk Onderzoek (Grant Reference No. VIDI 18919 to L.V.), European Innovation Council (Grant Reference No. Pathfinder 101071008 to C.E. and L.V.), and Dunhill Medical Trust (Grant Reference No. RPGF1810\93 to C.E. and S.B.). Declaration of Competing Interest No conflicts of interest. Supplementary Materials Supplementary material for this article is available with the online version here: https://doi.org/10.1162/IMAG.a.1046 References Associated Data Supplementary Materials Data Availability Statement Data and code to reproduce the results reported in this study are available at: https://doi.org/10.17605/OSF.IO/S5AG6