Transcranial ultrasound stimulation modulates neuronal membrane potentials across broad timescales in the awake mammalian brain

Abstract

Background:

Transcranial ultrasound stimulation (TUS) offers noninvasive neuromodulation with high spatiotemporal precision, but its cellular-level effects in the awake brain remain poorly understood.

Objective:

We investigated how low-intensity TUS modulates membrane voltage dynamics in individual cortical neurons in awake mice.

Methods:

Using the genetically encoded voltage indicator SomArchon, we performed high-speed kilohertz voltage imaging in awake head-fixed mice. TUS was delivered with a 0.35 MHz transducer at 10 or 40 Hz pulse repetition frequency, at intensities below the estimated threshold for auditory brainstem activation. We analyzed changes in membrane potentials (Vm), spiking, and coordination across simultaneously recorded neurons.

Results:

TUS evoked rapid (<10 ms) Vm depolarizations in 42.8 % of neurons, while only increasing spiking in 20.5 % of neurons, highlighting a direct effect of TUS on modulating synaptic inputs. Many neurons were entrained at both PRFs (20.8 % at 10 Hz; 12.7 % at 40 Hz) with Vm exhibiting significant phase-locking to individual TUS pulses. Vm entrainment was accompanied by increased temporal coordination across neurons and reset network synchrony. Furthermore, TUS-evoked cellular responses adapted over time, often transitioning from membrane depolarization to hyperpolarization upon repeated exposures, demonstrating prominent response depression.

Conclusion:

By resolving single-neuron responses in the awake mammalian brain, our results demonstrate that TUS directly activates individual cortical neurons with latencies often shorter than 10 ms. TUS pulsed at physiologically relevant frequencies of 10 and 40 Hz robustly entrains neural dynamics, alters network coordination and evokes neuronal plasticity. These results highlight the therapeutic potential of designing TUS pulsing patterns to target desired neural dynamics and plasticity features.

1. Introduction

Low-intensity, low-frequency transcranial ultrasound stimulation (TUS) is an emerging noninvasive neuromodulation technique capable of targeting deep brain regions with high spatiotemporal precision. Ultrasound stimulation elicits neural responses in both the central [1–5] and peripheral nervous systems [6–10]. While the precise physiological mechanisms of low-intensity TUS remain incompletely understood, several biological and physical factors have been proposed to contribute to TUS-induced changes in neural activity. Acoustic radiation force is considered a predominant factor [11,12], generating submicron membrane displacements that activate mechanosensitive ion channels [13–18]. Other proposed mechanisms include intramembrane cavitation [19–21] and synaptic vesicle release [22].

In clinical studies, TUS is often delivered at pulse repetition frequencies (PRFs) of 5–100 Hz [23–28], primarily to minimize thermal effects and ensure patient safety. However, stimulation frequency is a key factor influencing the therapeutic efficacy of neuromodulation techniques. For example, deep brain stimulation, commonly used to manage Parkinson’s disease and epilepsy, is typically delivered at high frequencies of 130–300 Hz, as lower frequencies yield inconsistent results [29,30]. In contrast, transcranial magnetic stimulation, primarily used for depression, employs lower frequencies <50 Hz [31] or bursts of 30–50 Hz pulses at ~5 Hz (theta-burst protocols) [32–34]. Although the underlying mechanisms remain unclear, lower frequencies (<100 Hz) are thought to promote neural entrainment and better engage intrinsic neural dynamics, while higher frequencies (>100 Hz) may disrupt pathological activity patterns without entrainment [35,36]. Supporting this, our recent voltage imaging study showed that 40 Hz deep brain stimulation robustly entrained neurons in awake mice, whereas 140 Hz induced a robust informational lesion without prominent entrainment [37].

TUS serves as an external input to neurons, with its effects shaped by the intrinsic properties of the targeted neurons [38]. TUS-evoked responses can be amplified by voltage-gated ion channels [22,39]; notably, TUS can also elevate cytosolic calcium via Ca²⁺ channels in brain slices, independent of Na⁺ channel activation [40]. These findings underscore the importance of voltage-gated ion channel kinetics in shaping membrane voltage dynamics, which constrain action potentials to a few hundred hertz [41–43]. Although TUS utilizes fundamental frequencies around 0.5–2 MHz to improve skull penetration, pulsing TUS at physiologically relevant timescales differentially modulates calcium dynamics [15,40] and spiking activity [44,45] in individual neurons. For instance, 40 Hz TUS enhances gamma-band oscillations, improves memory performance, and reduces amyloid pathology in Alzheimer’s disease pre-clinical models [46,47], mirroring effects observed with 40 Hz visual flickers [48]. Mechanistically, these frequency-specific effects likely reflect interactions between acoustic radiation force and intrinsic membrane properties, including mechanosensitive ion channel kinetics, membrane voltage time constants, and neuronal resonance. While excitatory and inhibitory neurons differ in PRF sensitivity, our recent work suggests that variability in TUS responses emerges more prominently at the single-neuron level than across canonical cell types [15].

Beyond immediate effects, TUS can induce sustained neuroplasticity. In humans, TUS reduces cortical GABA levels and enhances functional connectivity, consistent with lasting disinhibition [49]. TUS also increases motor cortical excitability for up to 30 min post-sonication, an effect abolished by blockade of Na⁺ and Ca²⁺ channels, N-methyl--D-aspartate (NMDA) receptors, and GABAergic signaling [50]. Mechanistic studies in rodents show that TUS-induced potentiation of motor responses requires Ca²⁺-activated bestrophin-1 channels, NMDA receptors, brain-derived neurotrophic factor, and tropomyosin receptor kinase B [51]. In hippocampal circuits, TUS directed at the medial perforant pathway induces long-term depression of field excitatory postsynaptic potentials, similar to that observed during low-frequency electrical stimulation [52]. Thus, TUS engages both rapid neuronal activity changes and slower plasticity-mediated processes to modulate circuit dynamics.

TUS consistently modulates brain activity across diverse experimental models, influencing behaviors [53,54], hemodynamic responses [55,56], population-level local field potentials [57,58], cytosolic calcium concentrations [15,59–61] and individual neuron spiking patterns [44,45]. However, these studies lack the temporal resolution to capture rapid dynamics evoked by TUS, due to either the intrinsic slowness of the signals, e.g., hemodynamics and cytosolic calcium, or the sparseness of the signals, e.g., spiking that requires temporal integration to detect rate changes. Thus far, the reported latencies of TUS-evoked neural responses range from ~50 to 300 ms [14,44,59], insufficient to confirm a direct neuromodulatory effect. Compounding this complexity is that TUS does not solely act through direct modulation of neurons. Astrocyte activation by ultrasound can modulate synaptic function via astrocytic glutamate release [62]. Moreover, TUS, when strong enough, can stimulate the auditory system, subsequently triggering cortical calcium influx and motor twitches hundreds of milliseconds later [63,64]. However, using ramped or lower-intensity waveforms can mitigate these auditory confounds [65,66]. Nonetheless, robust TUS effects persist in deafened mice [45,66,67], invertebrates [12,68,69], and cell cultures that lack auditory inputs [14,70,71].

To evaluate how TUS modulates membrane potentials (Vm), which captures synaptic inputs that dictate spiking outputs at the single-neuron level, we performed cellular voltage imaging in the cortex of awake mice using the genetically encoded voltage indicator SomArchon [72]. We delivered low-intensity TUS at PRFs of 10 Hz, an intrinsic rhythm broadly observed in the cortex [73], and 40 Hz, a frequency broadly associated with cortical fast spiking interneurons [74,75]. Stimulation was performed at an intensity estimated to be below the threshold for auditory brainstem activation [65]. We characterized the effects of TUS on Vm amplitude, entrainment, spiking, and plasticity.

2. Materials and methods

2.1. Animal preparation

All procedures were approved by the Boston University Institutional Animal Care and Use Committee (IACUC) and Biosafety Committee. Mice were group-housed prior to surgery and individually housed postoperatively with environmental enrichment, such as igloos or running wheels. The animal facility maintained a 12-h light/dark cycle at approximately 70 °F and 50 % humidity. The study included a total of 20 adult mice, aged 8–16 weeks at experiment onset, including 8C57BL/6 (Jackson Laboratory #000664, 3 female, 5 male), 5 NDNF-IRES-Cre transgenic mice (Jackson Laboratory #030757, 2 female, 3 male), 5 PV-tdTomato transgenic mice (1 male, 4 female) obtained by crossing B6.Cg-Gt(ROSA)26Sortm14(CAG-tdTomato)Hze/J (Jackson Laboratory, #007914) with B6.129P2-Pvalb^tm1(cre)Arbr/J (Jackson Laboratory, #017320) mice, and 2 PV-Cre (B6.129P2-Pvalbtm1(cre)Arbr/J, Jackson Laboratory #017320, 2 male).

Surgical preparations were as previous described [37,59]. Briefly, mice were anesthetized with 1–3 % isoflurane, and a 3 mm craniotomy was made over either the motor cortex (16 mice; AP: +1.75 mm, ML: 1.75 mm) or visual cortex (4 mice; AP: −2.8 mm, ML: 2.5 mm). AAV vectors were infused through a 36-gauge stainless steel needle (World Precision Instruments, NF36BL-2) connected to a 10 μL NANOFIL microsyringe (World Precision Instruments) and controlled by an UltraMicroPump3-4 microinjector (World Precision Instruments). Injections were made at 2–6 sites per craniotomy, terminating 180–250 μm below the dura. For NDNF- Cre mice, a total of 300 nL of AAV9--Syn-FLEX-SomArchon-GFP (titer: 1.28 × 10¹² GC/mL) was infused at 50 nL/min. C57BL/6 mice and the PV-Cre mouse received a total 400 nL of AAV8-CaMKII-SomArchon-GFP (titer: 3.2 × 10¹² GC/mL), and PV-tdT mice and sham mice received 400 nL of AAV9-Synapsin-SomArchon-GFP (titer: 5.42 × 10¹² GC/mL). After each injection, the needle was left in place for 5–10 min to aid viral diffusion. A #0 glass coverslip (3 mm outer diameter, Warner Instruments, 64–0726) was placed over the craniotomy and sealed to the skull using either surgical silicone adhesive (Kwik-Sil, World Precision Instruments) or ultraviolet-curable dental cement (Tetric EvoFlow, Ivoclar). Exposed skull regions were reinforced with Metabond Quick Adhesive Cement (Parkell, S380), and a custom aluminum headbar was affixed with dental cement (Stoelting, 5145).

All mice received 72 h of postoperative analgesia via a single pre-operative intramuscular injection of sustained-release buprenorphine (0.03 mg/kg; Reckitt Benckiser Healthcare). Mice were allowed to recover for 3–4 weeks before imaging.

2.2. High speed voltage imaging

Mice were habituated to head fixation on the imaging platform for five consecutive days prior to voltage recordings. During each session, neurons expressing SomArchon were identified by detecting GFP fluorescence, which is co-expressed with SomArchon. Voltage imaging was conducted either using a custom widefield fluorescence microscope as described previously [37,72], or, most of the time, a targeted illumination confocal microscope (TICO) as described previously [76]. Briefly, the widefield microscope was equipped with a Hamamatsu ORCA Fusion Digital sCMOS camera (Hamamatsu Photonics K.K., C14440-20UP), a 40x NA = 0.8 water immersion objective (Nikon, CFI APO NIR), a 140-mW 637 nm laser (Coherent Obis 637-140X) coupled with a reverse 2x beam expander (ThorLabs Inc., GBE02-E) to achieve an illumination with a diameter of 30–40 μm, and a mechanical shutter (Newport corp., model 76995) controlled laser timing via a NI DAQ board (USB-6259). Imaging frames were captured using HCImage Live (Hamamatsu Photonics K.K., U11158) at ~828 Hz (16 bits, 2 × 2 binning), and data were stored as DCAM image files (DCIMG). The TICO microscope was equipped with a 637 nm diode laser (Ushio America Inc., Red-HP-63X) for SomArchon excitation, a 488 nm laser (Lasertack, GmbH277 PD-01376) for GFP excitation, an sCMOS camera (Teledyne Photometrix, Model: Kinetix), DMD module (Vialux, V-7000 VIS) for targeted illumination, and an 16X objective (Nikon Corp., 16 × /0.8NA LWD). Image frames were captured using Teledyne Photometrics PVCAM software (Teledyne Photometrics, PVCAM).

2.3. Transcranial ultrasound stimulation

A planar ultrasound transducer with a center frequency of 350 kHz (Ultran GS350-D13) was positioned beneath the chin of awake, head-fixed mice, aligned under the microscope objective. Ultrasound gel (Aquasonic Clear^® Ultrasound Gel 03–08_BX, Parker Laboratories, Fairfield, NJ) was applied between the transducer and the chin to ensure efficient acoustic coupling and propagation to the brain. Ultrasound was delivered in 1-s bursts per trial, pulsed at either 10 or 40 Hz with a 20 % duty cycle. Pulse trains were triggered via MATLAB (MathWorks, Inc.) through a USB-6259 NI-DAQ multifunction I/O system (National Instruments). TTL triggers from the NI-DAQ and the imaging camera were simultaneously recorded using the Open Ephys platform (http://open-ephys.org) for precise offline alignment of stimulation and imaging frames. Sham stimulation controls were performed by coupling the transducer, with ultrasound gel, to the ventral abdomen of awake, head-fixed mice.

2.4. Acoustic field characterization and simulation

Free-field acoustic pressure was measured in degassed, deionized water using a calibrated 0.2 mm needle hydrophone (Precision Acoustics, UK). The hydrophone was mounted in a custom-built 3D motorized scanning tank constructed according to an open-source design [77]. Pressure waveforms were acquired with a digital oscilloscope (TDS2024C, Tektronix) and processed in MATLAB to extract peak acoustic pressures across a 3D grid centered on the transducer’s focus. Numerical simulations were performed using the k-Wave MATLAB toolbox to model ultrasound propagation through a C57BL/6 mouse skull [78]. The simulation domain incorporated anatomically realistic geometry and acoustic properties of the skull, brain, and surrounding soft tissues, including density, sound speed, and attenuation coefficients derived from the literature, as previously detailed [15,59]. To model realistic acoustic heterogeneity, a static air-filled cavity was incorporated along the nasal cavity and through the pharyngeal airway. The air cavity was modeled as a rectangular prism ~1 mm² in cross-section, with a gradual dorsoventral taper extending from the start to the end of the pharyngeal airway. The ultrasound source was modeled as a 13 mm diameter circular array. Peak intracranial pressures and spatial peak pulse average intensities (I_SPPA) were computed directly from k-Wave outputs at each voxel in the simulated brain volume. Spatial peak temporal average intensity (I_SPTA) was computed by multiplying I_SPPA by the pulse duty cycle (20 %).

2.5. SomArchon fluorescence trace preprocessing

SomArchon fluorescence traces were analyzed in MATLAB (MathWorks, R2021a). Image frames were first motion-corrected on a per-trial basis, following procedures described previously [37]. Briefly, each trial video was boxcar smoothed over 21 frames and sharpened using a Gaussian filter (filter size = 50; filter sigma = 1 and 25, respectively). The processed videos were then input into the NormCorre algorithm to estimate X and Y frame shifts, using a bin width of 10, maximum shift of 50 pixels, and no upsampling [79]. Each trial was individually motion-corrected. After within-trial correction, the mean motion-corrected images from all trials were aligned across trials using NormCorre. The resulting X and Y shifts from the cross-trial alignment were applied to each frame of the respective trials to generate fully motion-corrected videos.

Following motion correction, regions of interest (ROIs) corresponding to individual neurons were manually selected using the drawpolygon tool in MATLAB. For each ROI, the SomArchon fluorescence trace was computed by averaging the pixel intensities within the selected region. Traces were then interpolated to 1000 Hz using piecewise cubic Hermite interpolation (pchip) to facilitate alignment with ultrasound stimulation timestamps. To remove slow baseline fluctuations, traces were detrended by subtracting a boxcar-smoothed version of the signal using a sliding 1-s window.

Detrended traces were visually inspected, and those containing obvious motion artifacts or lacking discernible spiking activity were excluded from further analysis. The remaining detrended traces, referred to as membrane potential (Vm) traces, were used in all subsequent analyses.

2.6. Spike identification

Spikes were detected in the Vm traces using a custom algorithm as detailed previously [80]. To estimate spontaneous Vm fluctuations, traces were first high-pass filtered using a 10 ms moving median filter, and then smoothed using a 201 ms moving average to generate a smoothed trace (ST). An asymmetric separation of sub- and supra-threshold fluctuations was applied. The lower trace ($T_L$) was computed by replacing all Vm values greater than ST with ST values, as defined in Equation (1):

$$T_L=\{\begin{array}{l}Vm,Vm<ST\\ ST,Vm\ge ST\end{array}$$ Equation 1

To estimate periods with spiking, we computed the upper trace ($T_U$) by replacing all values of the Vm trace below the ST with the ST using Equation (2):

$$T_U=\{\begin{array}{l}Vm,Vm\ge ST\\ ST,Vm<ST\end{array}$$ Equation 2

Next, we computed the rate changes in $T_L$ and $T_U$ by calculating their derivatives ($d{T}_L$ and $d{T}_U$) respectively. We then identify putative spikes as the timepoints where $d{T}_U$ was greater than mean ($\mu$) $d{T}_U$ signal plus 6x the standard deviation ($\sigma$) of $d{T}_L$ using Equation (3):

$$Putativespikes=d{T}_U>\mu \left(d{T}_U\right)+6\times \sigma \left(d{T}_L\right).$$ Equation 3

Putative spikes were then evaluated using the following for three criteria. First, putative spikes within the first three and last three datapoints were excluded. Second, the Vm value of the putative spike was greater than the previous point. Third, the datapoint values after the spike drop sharply (i.e., have a negative derivative). That is, $d{T}_U$ values within the three datapoints after the identified putative spikes were more negative than 5x $\sigma \left(d{T}_L\right)$ below the mean. After excluding spikes not meeting the criteria, remaining putative spikes were deemed as final spikes for further analyses.

2.7. Classification of spike-modulated and Vm-modulated neurons

To determine whether TUS evoked significant changes in a neuron’s Vm, we analyzed two distinct time windows following TUS onset: a transient period (0–250 ms) and a delayed period (250 ms–1 second). During the transient period, the mean Vm within the first 250 ms after TUS onset was calculated for each trial. The mean Vm was then compared to the corresponding pre-stimulation mean Vm of the same trial (1-s pre-TUS onset) using a Wilcoxon signed-rank test (α = 0.05). Neurons with transient mean Vm significantly different than the pre-stimulation mean Vm were further classified as transiently activated or transiently suppressed. For the delayed period (250 ms–1 second), the window was divided into three consecutive 250 ms bins: 250–500 ms, 500–750 ms, and 750–1000 ms. For each bin, the mean Vm was calculated and compared to the pre-stimulation mean Vm for each trial using a Wilcoxon signed-rank test (α = 0.05). A neuron was classified as having delayed activation if at least two consecutive bins showed significant increases in mean Vm compared to the pre-stimulation mean Vm. To avoid overlap between transient and sustained classifications, neurons classified as transiently modulated within the first 250 ms were excluded from the sustained modulation category. Neurons with no significant changes in either time window were classified as non-modulated.

To assess TUS-induced changes in spiking activity, we followed a similar approach. Due to the inherently high variability of spike rates during brief 1-s windows, suppression of spike rate during either the transient or the sustained period was not analyzed.

To validate the statistical method used to detect TUS-induced modulation of Vm and spiking activity, we calculated the false positive rate by measuring the percent of false positive neurons when repeating the statistical tests at twenty randomized pseudo-stimulation onsets within the pre-stimulation period. False positive rate for all modulation tests was below 5 %, confirming that our statistical approach is robust with a false positive rate below the expected 5 % threshold (α = 0.05).

2.8. TUS-evoked response latency and temporal profiles

Vm latency was analyze across neurons that were classified as Vm transiently activated. Vm traces were first z-scored to the mean of the pre-stimulation period. A Wilcoxon rank-sum test (p < 0.05) was then used to compare the instantaneous z-scored Vm for each millisecond time bin across all trials to the mean of the pre-stimulation period (zeros). The first occurrence of a significant point was recorded as Vm latency for a given neuron.

To characterize the temporal dynamics of TUS-evoked responses, we computed the peak time and full-width at half-maximum (FWHM) from trial-averaged Vm and spike rate traces. For each neuron, we identified the time of peak response within a 500 ms stimulation window and defined peak time as the time from TUS onset to this maximum. FWHM was computed as the duration over which the response remained above half the peak amplitude. This was restricted to neurons with significant transient activation.

2.9. Analysis of evoked Vm responses across repeated TUS trials

To quantify the change in evoked Vm responses across repeated trials, we z-scored the Vm for each trial relative to the 1-s pre-stimulation period. For each neuron and trial, the response was computed as the mean z-scored Vm within a specific analysis window during stimulation: either the transient period (0–250 ms) or the delayed period (250–1000 ms). The rate of change across trials was then estimated as the slope of the linear regression of these trial-wise mean responses. To determine whether a neuron exhibited a significant change, we applied a threshold based on the slope’s p-value from the linear fit (α < 0.05).

2.10. Vm phase-locking analysis

To evaluate TUS-induced entrainment of Vm, we calculated the phase-locking value (PLV), which measures the consistency of Vm phase relative to stimulation pulse onsets. The PLV was computed with equation (4):

$$PLV(f)=\left|\left(\frac{1}{N}\right)\right|\sum _N{e}^{i\theta (f,n)}$$ Equation 4

where f is the Vm frequency and N is the total number of stimulation pulses. The phase $\varphi$ of the Vm at the pulse time was obtained from the Hilbert Transform of the narrowband (±5 % of f) filtered Vm. The average PLV value of each neuron was calculated as the mean across all pulses during all trials. To build the pre-stimulation shuffled distribution for statistical comparison, we randomly shifted the TUS onsets, maintaining pulse repetition frequency, and calculated the mean phase locking value for each iteration. This process was iterated 1500x. If a neuron had a PLV >95th percentile of the shuffled distribution, the neuron was considered significantly entrained by TUS.

2.11. Vm and spike cross-correlation between simultaneously recorded neuron pairs

Pairwise cross-correlation of Vm was computed for each neuron pair within the same field of view (FOV) during the pre-stimulation, stimulation, and post-stimulation periods, then averaged to get one value per FOV. The Vm signal of each neuron was pre-stimulation z-scored for each trial using the mean and standard deviation from the prestimulation. Neuron pairwise cross-correlation values were calculated for each trial and then averaged across trials using the MATLAB xcorr function, with zero time lag. Friedman’s test with Wilcoxon signed-rank post-hoc testing was used to assess the significance of changes in correlation coeffects across the pre-stimulation, stimulation, post-stimulation time windows.

Spike cross-correlation was computed using peri-stimulation time histograms (PSTHs) derived from the spike rasters. Spike events were binned at 50 ms intervals (corresponding to 20 Hz effective resolution) and converted into spike rates by dividing by the bin width. PSTHs were trial-averaged and z-scored based on the pre-stimulation period. Pairwise cross-correlation of PSTHs between neuron pairs was calculated at zero lag using the MATLAB xcorr function, following the same procedure as for Vm. Correlation values were averaged across trials, and changes in PSTH cross-correlation across the pre-stimulation, stimulation, and post-stimulation periods were assessed using Friedman’s test with Wilcoxon signed-rank post-hoc test.

2.12. Auditory brainstem response (ABR) estimation

To estimate potential auditory activation, we implemented the computational ABR prediction model described by Choi et al. [65] Briefly, the squared ultrasound pressure waveform was decomposed with short-time Fourier transforms, differentiated over time, and weighted by the inverse of mouse auditory thresholds across frequency bands. The resulting signal was convolved with the ABR impulse response to yield predicted ABR amplitude as a function of stimulation time.

2.13. Statistical analysis

All statistical analyses were performed in MATLAB 2021a using standard non-parametric tests unless otherwise specified. Because data distributions were non-normal, only non-parametric tests (e.g., Wilcoxon signed-rank, rank-sum, Friedman) were used throughout. Statistical test details, including sample sizes and p-values, are reported in the text or summarized in Suppl. Table 3. For circular data, Rayleigh’s test was used. A significance level of α = 0.05 was applied to all tests.

3. Results

3.1. Temporally resolved membrane potential analysis of TUS-evoked responses in individual cortical neurons in awake mice

To characterize rapid, real-time neuronal responses evoked by low-frequency and low-intensity TUS, we performed high-speed cellular voltage imaging in awake, head-fixed mice (Fig. 1a). Briefly, the genetically encoded voltage sensor SomArchon [72] was virally expressed in various cortical neuron subtypes (Fig. 1d, Methods). A glass coverslip was then placed over the pia to provide optical access. Imaging was conducted using a custom targeted-illumination confocal microscope, at a frame rate of 828 Hz [81], allowing simultaneous recording of multiple neurons while avoiding TUS-induced artifacts common to metal electrodes. Each recording trial was 3 s in duration, comprising a 1-s pre-stimulation baseline with the transducer off, 1-s of ultrasound stimulation at either 10 or 40 Hz PRF with a 20 % duty cycle, and a 1-s post-stimulation with the transducer off (Fig. 1c; Supplemental Table 1). Each neuron was typically recorded over 20 trials (range: 10–40), with an inter-trial interval of ~2–10 s. SomArchon fluorescence was processed offline to obtain Vm and spike timing (Methods, Fig. 1b–d).

Fig. 1. Voltage imaging analysis of TUS evoked responses in individual cortical neurons in awake mice.

(a) Experimental setup: awake mice expressing SomArchon in the cortical neurons were imaged using a custom, high-speed, confocal microscope, equipped with a 637 nm laser, a digital micromirror device (DMD) for patterned excitation, a high-speed sCMOS camera, and a 16 × objective. A 6 mm planar transducer delivered TUS from under the chin via ultrasound gel. (b) TUS protocol: 350 kHz ultrasound, 1 s duration, 20 % duty cycle, delivered at 10 or 40 Hz PRF. (c) Example field of view with SomArchon-GFP neurons; Vm and spikes from circled neuron shown in (d) (scale bar: 50 μm). (d) Vm traces of example neurons (circled in (d)) during three trials of 10 Hz TUS. Red ticks: spikes; black traces: Vm; gray ticks: pulse onsets; shaded gray: 1 s TUS. (e) Free-field acoustic pressure map in water (left, peak: 401 kPa) and simulated in situ acoustic pressure map in mouse brain using k-Wave (right, peak: 209 kPa), with glass imaging window shown in blue.

TUS at 0.35 MHz was delivered using a planar transducer positioned beneath the mouse’s chin and coupled with ultrasound gel (Fig. 1a), with the resulting ultrasound radiation volume (~6.5 × 6.5 × 20 mm at −6 dB) closely matching the dimensions of the mouse brain (~13 × 14 × 8 mm). This experimental configuration has been shown to deliver TUS to cortical structures in awake mice [15,59], and to provide unobstructed optical access of the objective lens to the cranial window for high resolution cellular imaging to quantify the real-time effect of TUS on individual neurons.

To estimate the ultrasound intensity at the cortical recording site, we first measured the free-field acoustic waveform of the transducer in degassed water using a needle hydrophone. The peak pressure was 401 kPa with a focal region of 20 mm height and 6 mm width (Fig. 1e, left). Using this empirically measured pressure as input, we performed k-Wave simulations to model intracranial pressure and intensity distributions. Simulations predicted a maxium intensity of 0.18 MPa at the intracranial bone and 0.11 MPa at the imaging site, the latter corresponding to a maximum in situ I_SPPA of 0.129 W/cm² (Fig. 1e, right; Supplemental Table 2; Methods). This I_SPPA value at the imaging site is consistent with the low-intensity range reported in prior neuromodulation studies [44,51,52,82], where effective stimulation pressures and I_SPPA values ranged ~88–438 kPa and ~0.079–6.4 W/cm², respectively.

As TUS is known to activate auditory pathways [63,64], we further estimated the auditory brainstem responses (ABRs) under our experimental conditions using the model developed by Choi et al. [65]. At 0.127 MPa (RMS, 0.18 MPa peak), the intensity at the intracranial bone, and 20 % duty cycle, the predicted ABR for 10 Hz PRF remained subthreshold throughout the entire 1-s-long stimulation duration (Supplemental Fig. 1). For 40 Hz PRF, while ABR was below the auditory activation threshold within ~250 ms of stimulation onset, it gradually increased above the threshold during the remaining 250 ms to 1 s stimulation period.

3.2. TUS induced heterogeneous Vm responses in individual neurons with many exhibiting rapid Vm depolarization within 10 ms of TUS onset, unaccompanied by increased spiking

TUS repeatedly evoked robust Vm changes across many neurons at both PRFs, with heterogeneous temporal profiles (Fig. 2). Many neurons exhibited transient depolarization within ~250 ms of TUS onset (Fig. 2a–b,e–f), when TUS-evoked ABR was estimated to be under the auditory detection threshold for both 10 Hz and 40 Hz PRFs (Supplemental Fig. 1). However, Vm depolarization was typically not accompanied by spike rate increase (Fig. 2a–e), although a small fraction exhibited increased spiking (Fig. 2b–f). Some neurons displayed delayed Vm depolarization, marked by gradual Vm shifts persisting throughout stimulation (Fig. 2c–g), while a smaller group showed consistent hyperpolarization (Fig. 2d–h). Most strikingly, many neurons’ Vm fluctuations were entrained by TUS at either 10 or 40 Hz PRFs, faithfully following pulse timing (Fig. 2g). The diverse Vm responses evoked by TUS, including depolarization, hyperpolarization, and entrainment, are largely in agreement with that observed during conventional intracranial electrical stimulation [37] and our previous finding that TUS-evoked calcium signals are largely independent of cell type and PRF [15].

Fig. 2. TUS-evoked Vm and spiking responses across example neurons.

(a–d, red) Example neurons showing TUS-evoked changes in Vm and spike rate during 10 Hz TUS and (e-h, blue) during 40 Hz TUS. For each neuron: (i) Vm traces across trials, aligned to TUS onset; vertical ticks indicate TUS pulse onsets. (ii) Corresponding spike raster plot across trials. (iii) Trial-averaged Vm (colored traces, mean ± 95 % CI), and spike rate (black histogram, 50 ms bins). Horizontal lines show the average Vm (colored) and spike rate (black) during pre-stimulation, the second before TUS onset.

To determine whether TUS-evoked significant transient responses in each neuron, we computed the mean Vm amplitude in the first 250 ms of each stimulation trial and the corresponding pre-stimulation period immediately before stimulation onset, and compared across trials using the Wilcoxon signed-rank test (α = 0.05). Neurons were classified as transiently modulated if Vm significantly exceeded pre-stimulation levels (Methods, Fig. 3a and b). Of the 161 recorded neurons across 15 mice, 38.0 % (61/161) were significantly transiently modulated, with 29.2 % (47/161) activated (depolarized) and 8.7 % (14/161) suppressed (hyperpolarized) (Table 1, Fig. 3c–f). The fraction of modulated neurons showed no noticeable difference between 10 Hz and 40 Hz PRFs. In contrast, only 5.7 % and 3.6 % of neurons were modulated during the delayed period (250 ms to 1 s after onset) at 10 Hz and 40 Hz PRFs, respectively, and all were activated. Although 40 Hz PRF may engage auditory pathways more strongly than 10 Hz PRF during the delayed window (Supplemental Fig. 1), this did not translate into significant evoked responses in the motor cortex neurons recorded. Finally, most Vm-modulated neurons (88 %, 61/69) responded during the transient window, indicating that rapid Vm changes dominated TUS responses.

Fig. 3. TUS-evoked transient and delayed Vm and spiking changes across individual neurons at 10 and 40 Hz PRFs.

(a–b) Heatmaps of z-scored Vm (left) and spike rate (right) aligned to TUS onset at 10 Hz (a, N = 105) and 40 Hz (b, N = 55). Neurons are grouped by response type (1: transient activation, 2: transient suppression, 3: delayed activation, 4: no modulation); spike rate heatmaps are sorted the same as Vm but independently classified. (c–e) Mean z-scored Vm for neurons with (c) transient activation, (d) transient suppression, or (e) delayed activation (10 Hz, red: N = 31, 9, 6; 40 Hz, blue: N = 16, 5, 2). (f) Violin plots of normalized Vm during the corresponding periods for different populations of Vm-modulated neurons. *, p < 0.05, **, p < 0.01, ***, p < 0.001 Wilcoxon signed-rank test (see Suppl. Table 3). (g–h) Mean z-scored spike rate for neurons with (g) transient or (h) delayed activation (10 Hz, red: N = 7, 9; 40 Hz, blue: N = 8, 9). (i) Violin plots of normalized spike rate during the corresponding periods for different populations of spike-modulated neurons. *, p < 0.05, **, p < 0.01, ***, p < 0.001 Wilcoxon signed-rank test (see Suppl. Table 3).

Table 1.

Vm and spike modulation across PRFs and conditions

PRF	Metric	Total Neuron #	Transient (−) # (%)	Transient (+) # (%)	Total Transient # (%)	Delayed # (%)	Total Modulated # (%)
Ultrasound
10	Vm	106	9 (8.5 %)	31 (29.2 %)	40 (37.7 %)	6 (5.7 %)	46(43.4 %)
40	Vm	55	5 (9.1 %)	16 (29.1 %)	21 (38.2 %)	2 (3.6 %)	23(41.8 %)
Total	Vm	161	14 (8.7 %)	47 (29.2 %)	61 (37.9 %)	8 (5.0 %)	69 (42.9 %)
10	Spike	106	0 (0 %)	7 (6.6 %)	7 (6.6 %)	9 (8.5 %)	16 (15.1 %)
40	Spike	55	0 (0 %)	8 (14.5 %)	8 (14.5 %)	9 (16.4 %)	17 (30.1 %)
Total	Spike	161	0 (0 %)	15 (9.9 %)	15 (9.3 %)	18 (11.2 %)	33 (20.5 %)
Sham Ultrasound
10	Vm	39	0 (0 %)	0 (0 %)	0 (0 %)	0 (0 %)	0 (0 %)
40	Vm	21	0 (0 %)	0 (0 %)	0 (0 %)	0 (0 %)	0 (0 %)
Total	Vm	60	0 (0 %)	0 (0 %)	0 (0 %)	0 (0 %)	0 (0 %)
10	Spike	39	0 (0 %)	0 (0 %)	0 (0 %)	0 (0 %)	0 (0 %)
40	Spike	21	0 (0 %)	0 (0 %)	0 (0 %)	2 (5.1 %)	2 (5.1 %)
Total	Spike	60	0 (0 %)	0 (0 %)	0 (0 %)	2 (3.3 %)	2 (3.3 %)

As Vm depolarization often precedes spiking, we next examined spike rate changes during transient and delayed periods (Methods). Overall, 20.5 % of neurons showed increased spike rates, in agreement with a prior primate study [83], while none showed significant decreases (Fig. 3g–i, Table 1). Among spike-modulated neurons, 90.9 % (30/33) also exhibited significant transient Vm depolarization. However, only 45.5 % (15/33) of spike-modulated neurons showed transient increases in spiking, in contrast to 88.4 % (61/69) of Vm-modulated neurons showing transient Vm changes. Further evaluation revealed that just 23.4 % of transient Vm-activated neurons were spike-modulated, suggesting that many Vm responses occur independently of spiking.

As a control, we performed off-target sham ultrasound to the abdomen of the mouse using ultrasound gel. Sham ultrasound induced minimal changes in Vm, with no Vm-modulated neurons (N = 60 neurons in 5 mice, Table 1). A small fraction of neurons (3.3 %, 2/60, Table 1) exhibited increased spiking, within the expected 5 % false positive rate.

After observing rapid TUS-evoked Vm responses in individual neurons, we further quantified response latency and temporal profiles in neurons with transient Vm activation (Methods). Across 47 neurons, TUS-evoked depolarization latencies ranged from 5 to 53 ms (10.85 ± 1.59 ms, mean ± SEM), with 34 neurons (72.3 %) having a latency under 10 ms (Fig. 4). Interestingly, latencies were modestly shorter during 10 Hz PRF (range: 5–37 ms, mean ± SEM: 9.58 ± 1.31 ms, N = 31) compared to during 40 Hz (range: 5–53 ms, mean ± SEM: 13.31 ± 3.94 ms, N = 16). This difference may reflect the higher initial energy delivered during the first 20 ms of stimulation onset, as 10 Hz TUS has a 20 ms pulse, but 40 Hz has only a 5 ms pulse. Further quantification of the overall activation profiles revealed slight difference between PRFs, with 10 Hz TUS-evoked responses peaking at 154.6 ± 13.3 ms with a full-width at half-maximum (FWHM) of 274.0 ± 23.8 ms and 40 Hz TUS-evoked responses peaking at 117.1 ± 11.7 ms with a FWHM of 199.8 ± 31.9 ms. FWHM values were significantly broader at 10 Hz than 40 Hz (Mann–Whitney U test, one-tailed, p = 0.024), indicating more sustained depolarization at 10 Hz PRFs. The fact that many neurons exhibited short-latency responses under 10 ms confirms that TUS directly modulates neuronal dynamics at the cellular level, rather than through indirect somatosensory or auditory activation, or systematic heating.

Fig. 4. Rapid onset of TUS-evoked depolarizations.

(a) Voltage traces across 18 trials for a single neuron. Gray line indicates TUS onset. Colored boxes denote traces zoomed in shown in (bi–iii). (bi–iii) Zoomed-in Vm traces from (a) across different trials (green, blue, red) aligned to TUS onset (gray line), with many trials showing rapid Vm depolarization within 9 ms (yellow line). (c) Normalized Vm of the example neuron shown in (a). Shown are mean (solid black) ± SEM (gray shading, N = 18 trials). (d) Histogram of Vm response latencies across transiently Vm activated neurons.

3.3. TUS-evoked responses systematically change across repeated stimulation trials with more neurons showing response depotentiation

TUS can induce long-lasting changes in neural activity, with effects persisting minutes to hours beyond stimulation offset [84–86]. To assess whether TUS-evoked responses systematically changed across repeated trials, we z-scored the Vm of each trial relative to the 1 s pre-stimulation baseline. Since responses were primarily restricted to the transient period within the first 250 ms after stimulation onset (Table 1), our analysis focused on this window.

Many neurons exhibited progressive reductions in TUS-evoked transient Vm responses across trials, often shifting from depolarization to hyperpolarization (Fig. 5a–c), while a smaller subset exhibited progressive increases (Fig. 5b–d). Overall, 19 of 161 neurons (11.8 %) showed significant trial-by-trial changes in the transient window (p < 0.05, Methods), with 16 progressively decreases and 3 increases. In contrast, during the delayed window (250–1000 ms post-onset) evoked Vm responses showed minimal changes, only 7 neurons (4.3 %) showed significant changes across trials, with 6 exhibiting increased Vm and 1 decreased Vm. At the population level, trial-to-trial changes during the transient window were significantly different from zero (Wilcoxon signed-rank test, p = 5.3e-04, Fig. 5e), whereas responses during the delayed period were not (p = 0.77, Fig. 5f), suggesting a temporally specific effect of evoked response adaptation. Thus, TUS-evoked cellular responses are shaped by plasticity mechanisms and dominated by depotentiation.

Fig. 5. Evoked Vm responses across repeated TUS trials.

(a–b) Heatmaps of baseline-normalized Vm across trials for two example neurons showing progressive reduction (a) or increases (b) of evoked transient responses across trials. Each row represents one trial. Colored bars indicate analysis windows: red = transient period (0–250 ms), blue = delayed period (250–1000 ms). (c–d) Trial-by-trial evoked Vm response amplitude during the transient (ci, di) and delayed (cii, dii) periods for the neurons shown in (a) and (b), respectively. The darker lines are linear fits with the corresponding R² and p-values indicated. (e) Distributions of the linear fit slope values (rate of change across trials) of the transient Vm across all recorded neurons (N = 161). (f) Same as (e), but for the delayed Vm.

3.4. TUS entrained cortical neuron Vm at both 10 and 40 Hz PRFs

Upon observing rapid Vm changes following TUS onset, we further evaluated the temporal precision of these responses relative to individual TUS pulses by assessing Vm entrainment (Fig. 6a–c). Specifically, we computed the phase-locking value (PLV) between TUS pulse onsets and Vm, and compared the observed PLVs to a shuffled null distribution (α = 0.05, see Methods). We identified 22/106 (20.8 %) significantly entrained neurons at 10 Hz and 7/55 (12.7 %) at 40 Hz PRF (Fig. 6a–l). Sham control recordings showed chance-level entrainment at 5 % (3/ 60), the expected false-positive rate. Thus, TUS delivered at physiologically relevant 10 and 40 Hz PRFs entrained many cortical neurons, consistent with prior findings in anesthetized rats that tonically firing Purkinje cells exhibited more temporally regular spiking in response to 50 and 100 Hz PRF TUS [82]. The temporal precision of the evoked somatic Vm responses by each individual TUS pulse further supports a direct effect of TUS on individual neurons.

Fig. 6. 10 and 40 Hz TUS strongly entrains Vm and enhances inter-trial phase coherence.

(a, c) Vm traces from example 10 Hz− (a) and 40 Hz-entrained (c) neurons. (b, d) Polar plots of phase angles at pulse onset for the neurons in (a) and (c). (e, g) Normalized Vm traces across all 10 Hz− (e, N = 22) and 40 Hz-entrained (g, N = 7) neurons. (f, h) Polar histograms of phase angles at pulse onset across all 10 Hz− (f, N = 22) and 40 Hz-entrained (h, N = 7) neurons. (i, k) Heatmaps of Vm responses for all entrained neurons at 10 Hz (i) and 40 Hz (k), sorted by response amplitude. (j, l) Normalized pulse-triggered Vm across all entrained neurons at 10 Hz (j) and 40 Hz (l). (m, p) Inter-trial coherence (ITC) across frequencies aligned to stimulation onset for all entrained neurons at 10 Hz (m) and 40 Hz (p). (n, q) ITC amplitude at 10 Hz (n) and 40 Hz (q). (o, r) Violin plots comparing ITC across pre-, during (stimulation), and post-TUS periods for 10 Hz (o) and 40 Hz (r) entrained neurons. *, p < 0.05, Wilcoxon signed-rank test (Suppl. Table 3).

Vm fluctuations in entrained neurons were tightly time-locked to stimulation pulses at both PRFs. For example, one 10 Hz-entrained neuron showed peak depolarization at 9 ms after each pulse onset, corresponding to a preferred phase angle of 2° (Fig. 6a and b), while a 40 Hz-entrained neuron showed a depolarization peak at 3 ms post-pulse onset, corresponding to a preferred phase angle of 303° (Fig. 6c and d). Across all 10 Hz entrained neurons, preferred phase angles clustered around 352 ± 62.6° (mean ± S.D., 22 neurons, N = 220 pulses, Fig. 6e and f), significantly deviated from uniformity (Rayleigh test; p = 1.28e-14, r = 0.40, N = 220 pulses). In contrast, 40 Hz-entrained neurons exhibited a broader distribution centered at 201 ± 68.1° (mean ± S.D., 7 neurons, N = 280 pulses, Fig. 6g and h), nonetheless significantly deviated from uniformity (Rayleigh test; p = 5.22e-9, r = 0.29, N = 280 pulses). Across all entrained neurons, pulse-triggered population Vm peaked around 0 ms of 10 Hz post-pulse onset, and around 2 ms of 40 Hz post-pulse onset (Fig. 6i–l).

To further confirm entrainment, we measured trial-to-trial phase consistency through inter-trial coherence (ITC) (Fig. 6m–p, Methods). ITC values were significantly different across pre-, during, and post-stimulation windows at both PRFs (Fig. 6n–q; Friedman tests; statistical results in Suppl. Table 3). ITC at both stimulation frequencies increased during stimulation relative to pre-stimulation, indicating phase-aligned responses across trials (Wilcoxon sign-rank tests; statistical results in Suppl. Table 3). At 10 Hz, ITC persisted significantly into the post-stimulation period, whereas at 40 Hz, ITC returned to prestimulation levels (Fig. 6o–r). Together, these results demonstrate that TUS evoked precisely timed Vm responses at both 10 and 40 Hz PRFs, with 10 Hz PRF producing more consistent responses across individual pulses, highlighting the potential for frequency-specific pulsing protocols to modulate network dynamics through phase-dependent mechanisms.

3.5. TUS resets network dynamics

As our imaging technique allows for simultaneous imaging from multiple neurons at near kilohertz within a large field of view (FOV) [81] (Fig. 7a), we further examined how TUS influences network dynamics by evaluating coordination between neurons. Across FOVs, we recorded between 3 and 25 neurons (8.5 ± 6.52 neurons per FOV, mean ± S.D., N = 10 FOVs). The exact number of neurons per FOV was determined by SomArchon positive neurons and their anatomical locations. In many recordings, we noted that Vm and spiking became more coordinated immediately after TUS offset (Fig. 7b). To quantify this, we computed cross-correlations of Vm and spiking activity across simultaneously recorded neuron pairs (Methods). Vm-Vm correlations decreased significantly during stimulation compared to baseline, and then increased significantly post-stimulation than during TUS (Fig. 7c, Friedman test, followed by post-hoc Wilcoxon signed-rank test; see Suppl. Table 3), suggesting that TUS desynchronizes the network, and TUS offset triggers a reset in network dynamics. Similarly, spike-spike correlation was significantly higher post-stimulation than during TUS (Fig. 7d, Friedman test, followed by post-hoc Wilcoxon signed-rank test; see Suppl. Table 3). This rebound synchrony may contribute to the plasticity effects observed (Fig. 5) and could be leveraged in therapeutic applications to modulate network states more effectively.

Fig. 7. TUS modulates Vm and spiking synchronization between simultaneously recorded neurons.

(a) SomArchon fluorescence image of a field of view (FOV) containing 25 simultaneously recorded neurons during 10 Hz TUS (scale bar: 50 μm). (b) SomArchon fluorescence from a FOV containing 4 simultaneously recorded neurons during 40 Hz TUS (scale bar: 50 μm). (c) Example Vm traces and spike times across two trials of the neurons shown in (b). Red ticks: spikes; black: Vm; gray ticks: pulse onsets; shaded gray: 1 s TUS (scale bar: 100 ms). (d) Violin plot of Vm zero-lag cross-correlations across pre-, during (stim), and post-TUS periods (N = 10 FOVs, Wilcoxon signed-rank test; see Suppl. Table 3). (e) Violin plot of zero-lag spike cross-correlations (N = 10 FOVs, Wilcoxon signed-rank test; see Suppl. Table 3).

4. Discussion

Despite the clinical promise of TUS, its direct effects on individual neurons in the awake mammalian brain remains unknown. We employed high-speed cellular voltage imaging in awake mice to assess single-neuron Vm responses to TUS delivered at 10 and 40 Hz PRFs, free of TUS-mediated electromechanical artifacts often associated with electrodes [87]. We found that TUS robustly modulates individual neurons, producing diverse Vm changes, including depolarization, hyperpolarization, increased spiking, and entrainment. While 42.8 % of recorded neurons showed significant Vm modulation, predominantly within 250 ms of stimulation onset, only 20.5 % showed increased spiking, typically emerging after 250 ms, consistent with a gradual ramping depolarization evoked by TUS. Notably, many activated neurons responded with a latency shorter than 10 ms, suggesting direct neuronal effects. Reflecting the observed short-latency responses, many neurons were entrained by TUS at both frequencies (20.8 % at 10 Hz; 12.7 % at 40 Hz), with Vm fluctuations precisely time-locked to each pulse, especially during 10 Hz TUS, highlighting TUS as a temporally precise and biophysically selective neuromodulatory tool. Additionally, we discovered that the evoked Vm response magnitude systematically changed in many neurons, with most declining across repeated TUS trials, highlighting a plasticity mechanism during repeated TUS. In agreement with the plasticity effects, TUS reset network synchrony upon stimulation offset. These results demonstrate that pulsed TUS at physiologically relevant PRFs can directly drive temporally precise Vm changes in individual neurons and engage cellular plasticity processes supporting prolonged network modulation.

We observed rapid Vm changes with latencies shorter than 10 ms from TUS onset in many neurons. Although transient onset and offset transitions of strong TUS could create broad frequency activation of the auditory brainstem within ~6 ms [66], it requires ~15–20 ms for cochlear activation to propagate to auditory cortex [88], which is much longer than the latencies observed in our recordings. Furthermore, prior studies that observed ultrasound-evoked auditory responses used high PRFs of 1–1.5 kHz [63,64], which are within the mouse hearing range (~1–100 kHz [89,90]). We used a 0.35 MHz carrier frequency, well above auditory frequency thresholds, and low PRFs (10 or 40 Hz) to minimize cochlear activation. Furthermore, at 0.127 MPa RMS (0.18 MPa peak), the modeled ABR response remained subthreshold at 10 Hz, and at 40 Hz it did not exceed threshold until after ~250 ms of stimulation. Thus, the majority of neurons we observed to respond within the first 250 ms (88 %, 61/69) were modulated when auditory activation was predicted to be absent. Additionally, TUS-evoked auditory pathway activation takes several hundred milliseconds to result in significant changes in the visual cortex [64], and motor cortex neurons are minimally modulated by auditory stimuli without significant motor implications, further arguing against an auditory origin. Astrocytes have been shown to be activated by TUS [91], but astrocyte-mediated neuronal responses generally occur on slower time scales of hundreds of milliseconds [92], and cannot explain the short latency activation observed here. Finally, we used the abdominal-directed TUS as a sham control to account for systemic and auditory confounds, such as transducer-generated sound or somatic sensation, independent of brain-targeted sonication. Off-target sham ultrasound evoked no significant changes in Vm and spiking, further ruling out non-specific systemic and auditory confounds. Thus, our results demonstrate a direct effect of TUS on neurons, which cannot be explained by auditory or indirect mechanisms.

TUS evoked both excitatory and inhibitory Vm responses across cortical neurons, especially during the transient period within 250 ms of stimulation onset. Thus, restricting each TUS burst to 250 ms may optimize neural modulation while minimizing the overall acoustic energy delivery into the brain. Under the tested PRFs (10 and 40 Hz), responsive neuron proportions and diversity were similar, suggesting that cortical neurons exhibit broadly distributed sensitivity to these stimulation parameters. Prior work suggests that PRFs can differentially modulate cell types, for instance, excitatory neurons increased firing with higher PRFs (30–4500 Hz), while inhibitory neurons remained largely unresponsive, even when duty cycle was held constant, suggesting intrinsic differences in temporal sensitivity [45]. Similarly, another study showed selective activation of PV + interneurons and suppression of CaMKII+ excitatory neurons in the hippocampus during 900 Hz continuous-wave stimulation, implicating both waveform shape and frequency in driving cell-type-specific divergence [61]. In the present study, 10 and 40 Hz PRF produced heterogeneous responses across cell types with no clear bias in evoked responses among CaMKIIα+ excitatory neurons, NDNF+ inhibitory interneurons, or a pan-neuronal Syn-expressing population (Suppl. Table 4). Thus, slower PRFs at 10 and 40 Hz broadly recruit cortical neurons with overlapping activation thresholds.

Beyond cell-type effects, accumulating evidence suggests that PRF alone can critically shape neural recruitment. In ex vivo slices, increasing PRF from 300 Hz to 1500 Hz markedly enhanced calcium response rates [40]. In humans, low-frequency TUS at 10–100 Hz produced sustained inhibition of motor-evoked potentials, while higher PRFs (i.e.,1000 Hz) failed to elicit lasting effects, underscoring the nuanced influence of PRF on neuromodulatory outcomes [93]. Our findings extend this principle to the cellular level: although both 10 and 40 Hz stimulation reliably evoked rapid depolarizations, 40 Hz pulses produced slightly delayed response onset and earlier peak depolarization with a narrower temporal profile (i.e., reduced FWHM). These differences suggest that despite delivering less energy in the initial milliseconds, the higher temporal precision of 40 Hz stimulation may more efficiently trigger membrane responses through rapid pulse onsets and recovery intervals that minimize desensitization. In contrast, the longer 10 Hz pulse may yield more prolonged membrane engagement. This suggests that PRF influences both the likelihood of activation and the temporal dynamics of Vm responses.

Across repeated trials, TUS-evoked neuronal responses within 250 ms of stimulation onset progressively shifted, with most neurons exhibiting response decay, but some showing augmentation, highlighting cellular adaptation. However, evoked responses during the delayed period, 250–1000 ms after stimulation onset, showed minimal changes over repeated TUS, consistent with the evoked Vm changes being primarily restricted to the first 250 ms of stimulation. This Vm shift reflects cellular and circuit plasticity mechanisms that regulate excitability during repeated stimulation. TUS-evoked responses could engage intrinsic cellular plasticity mechanisms across time scales, involving ion channel desensitization and protein turnover, and circuit plasticity mechanisms via well-established short-term and long-term synaptic changes. For example, activity-dependent recruitment of hyperpolarizing potassium conductances is upregulated by sustained depolarization or intracellular calcium [43,94]. TUS-mediated increases in intracellular calcium could thus contribute to the observed response depotentiation via recruitment of potassium channels. Another example is mechanosensitive ion channel Piezo1, which exhibits rapid inactivation under sustained or repetitive mechanical force [95]. Piezo1 may enter desensitized states that reduce subsequent responsiveness to TUS. Alternatively, repeated TUS may enhance inhibitory synaptic inputs via facilitation of GABAergic transmission or increased interneuron recruitment [96]. Synaptic depression at excitatory terminals, due to vesicle depletion or presynaptic calcium channel inactivation, may also contribute [97]. Finally, the polarity shift and response decay align with homeostatic plasticity mechanisms such as synaptic scaling or firing rate adaptation, which could act to stabilize activity and prevent overexcitation [98,99].

In this study, we examined the effect of 1 s long TUS, as the photostability of SomArchon limits the total recording duration of a given neuron. With the continued progress on voltage sensor development, more photostable high-performance sensors (e.g., Electra-Off [100]) would permit the analysis of the effect of prolonged TUS exposure. This is particularly advantageous for revealing TUS-mediated large network changes and sustained cellular plasticity effects. The ultrasound beam (~6.5 mm diameter, ~20 mm length at −6 dB) used is expected to sonicate a large portion of the mouse brain due to its small size (~13 mm × 14 mm × 8 mm). Even with improved focusing, diffraction limits focal zones to millimeter–centimeter scales [101], and low frequencies, skull heterogeneity, and skull geometry further degrade spatial targeting precision [3,102]. These physical constraints make it unavoidable that a sizable brain volume is simultaneously modulated in both preclinical and clinical studies. However, the short latency response within 10 ms of stimulation onset cannot be explained by indirect network responses from interconnected brain regions. Furthermore, we observed 42.8 % of neurons showed membrane potential changes, and 20.5 % increased firing. This is comparable to that reported previously in nonhuman primates, where focused ultrasound to the premotor cortex increased firing in ~20 % of neurons [83]. Moreover, 20.8 % of neurons at 10 Hz and 12.7 % at 40 Hz exhibited significant phase-locking of Vm to each TUS pulse, further ruling out the involvement of indirect pathways. However, it is possible that direct pre-synaptic terminal activation by TUS could contribute to the observed somatic membrane voltage changes that we measured.

Beyond single-neuron Vm amplitude changes, our findings revealed that TUS can entrain cortical neurons to externally imposed frequencies. Significant phase-locking was observed in many neurons at both 10 and 40 Hz PRFs, with clear population-level clustering of preferred phase angles despite cell-to-cell variability. These results indicate that TUS can reliably bias the timing of membrane dynamics, aligning individual neurons to the temporal structure of stimulation. The significant phase clustering suggests that TUS exerts a reliable mechanical drive across neurons, synchronizing Vm oscillations through stereotyped membrane displacements at each pulse. Prior mesoscopic studies have shown TUS-induced increases in LFP power at stimulation frequencies, including 40 Hz, which has been linked to enhanced memory and amyloid clearance in Alzheimer’s models [46,47]. Here, we extend these findings by resolving entrainment at the single-cell level in the cortex of awake mice. A previous study in anesthetized rats showed that 50 Hz and 100 Hz TUS altered interspike intervals in tonically active Purkinje cells [103], though without assessing entrainment. Using cellular voltage imaging in awake mice, our study provides the first in vivo characterization of temporally resolved Vm entrainment to ultrasound stimulation.

Stimulation induced neuronal entrainment critically depends on cell types, stimulation frequencies, and brain regions [37,104,105]. Recent studies demonstrated that rhythmic intracranial electrical stimulation at lower frequencies (e.g. 40 Hz) produced strong and consistent entrainment of fast-spiking interneurons across multiple cortical areas, whereas higher frequency stimulation (e.g., at 140 Hz) showed restricted efficacy in entraining fast-spiking interneurons only in some areas [104]. Similarly, hippocampal pyramidal neurons were reliably entrained by 40 Hz electrical stimulation, but not 140 Hz stimulation [37]. Additionally, 10 Hz, but not 140 Hz, sensory stimulation entrained population local field potentials in sensorimotor striatum [105]. Future studies systematically exploring the effect of TUS pulse repetition frequencies on different cell types across brain regions will provide mechanistic insights on how intrinsic membrane biophysical properties and circuit-level functional connectivity features shape TUS-induced responses.

EEG studies have reported increased signal complexity and entropy during TUS, reflecting decorrelated activity in broader cortical networks [106]. During TUS, we observed a significant reduction in cortical network synchrony. Furthermore, we detected a statistically significant increase in synchronization between simultaneously recorded neurons for both Vm and spiking activity upon TUS offset. This network reset after TUS offset could contribute to the plasticity effects observed during repeated stimulation. The rebound increase in synchrony is likely due to the reduction during stimulation, which was only statistically significant for Vm, but not spiking. Future studies with larger neuronal populations and longer recording durations may uncover subtle but functionally relevant spiking changes.

Appendix A. Supplementary data

Supplementary data to this article can be found online at https://doi.org/10.1016/j.brs.2025.09.004.

Code availability

Codes used for data analysis are available on Github repository: https://github.com/HanLabBU/TUS-Brain-Stimulation-2025. Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.