A Physiological Marker for Deep Brain Ultrasonic Neuromodulation

Abstract

Objectives:

Transcranial ultrasound neuromodulation (TUSN) is a non-invasive and spatially specific therapy that promises to deliver treatments tailored to the specific needs of individuals. To fulfill this promise, each treatment must be modified to adequately correct for variation across individual skulls and neural anatomy. This study examines the use of ultrasound-induced voltage potentials (measured with electroencephalography; EEG) to guide TUSN therapies.

Materials and Methods:

We measured EEG responses in two awake nonhuman primates during sonication of 12 targets surrounding two deep brain nuclei, the left and right lateral geniculate nucleus.

Results:

We report reliable ultrasound evoked potentials measured with EEG following the deep brain ultrasonic modulation in nonhuman primates. Robust responses are observed following just 10 repetitions of the ultrasonic stimuli. Moreover, these potentials are only evoked for specific deep brain targets. Further, a behavioral study in one subject shows a direct correspondence between the target with maximal EEG response and ultrasound based modulation of visual choice behavior. Thus, this study provides evidence for the feasibility of EEG-based guidance for ultrasound neuromodulation therapies.

Introduction

Through its capacity to perturb deep brain regions with millimeter and microsecond precision, transcranial ultrasound promises to deliver precise treatment of neurological disorders [1]. Recent studies have demonstrated ultrasound’s capacity to modulate visual circuits, mood networks, motivation, and disease of brain function in both nonhuman primate and human subjects [2–12]. These results have led to an explosion of interest in clinical trials using ultrasound to treat depression [13, 14], epilepsy [15], chronic pain [16], and other conditions [17].

Widespread clinical adoption of this approach would benefit from protocols that deliver repeatable intensity to the target anatomy. Variations in the human skull cause aberration and attenuation of the ultrasound signal in a manner that is unique in each patient. [18–20]. These variations in skull properties introduce at least a four-fold variation [21, 22]in the intensity delivered to the target across patients.

One solution to this problem is to measure the threshold at which ultrasound evokes changes in neural activity in each patient and then adjust the acoustic intensity relative to this threshold. Transcranial magnetic stimulation (TMS) treatments, for example, use a similar approach in which treatment intensities are calibrated by measuring the energy needed to elicit a response in the motor cortex [23]. Results in small animal models suggest that such an approach should be possible for transcranial ultrasound as well. In rodents, ultrasound can elicit motor evoked potentials and overt muscle contractions [24]. However, these effects have yet to be observed in larger animal studies which have instead relied on changes in behavior [3, 5–7] or changes in the amplitude of electroencephalography (EEG) potentials evoked by other modalities (e.g., TMS [9]) to infer the effects of ultrasound on neural tissue.

This study demonstrates ultrasound’s capacity to evoke a transient EEG response in the awake primate brain. The evoked potential is specific to the targeted neural anatomy and independent of the location of the transducer—the placement of which varied across the two subjects. This result lays the groundwork for the development of patient-specific calibration methods to address the variation in patient skulls that has so far precluded the delivery of a controlled and consistent acoustic intensity into the brain.

Materials and Methods

Animals

Two male macaque monkeys (macaca mulatta, subjects B, and H, ages 7 and 6 years and weight 15.0, and 10.8 kg respectively) participated in the study. All procedures were conducted as approved by the Institutional Animal Care and Use Committee of the University of Utah.

Measurement of Ultrasound Evoked Potentials

The stimulation of the deep brain regions was achieved using Remus (Figure 1a), a system for remote ultrasound delivery into the brain [2–4, 25]. In Remus, a 256 element phased array transducer is affixed to the head of an awake and head- fixed nonhuman primate (NHP) subject. Each NHP subject is implanted with four titanium pins that enable the placement of a custom 3D frame that simultaneously provides for head fixation and placement of the transducer array. The pins, attached by titanium screws to the skull, additionally provide a strong intracranial electrical connection. Head fixation is achieved by affixing the frame to a primate chair (Crist Instrument Company, Hagerstons, MD). Ultrasound is delivered through the intact skin, muscle, and skull. The transducer array has a center frequency of 650 kHz but was driven at 480 kHz for this study. There is evidence that lower frequencies are more effective at eliciting neural responses [24, 26–28]and 480 kHz was selected as the lowest frequency at which the array gives sufficient pressure output.

Figure 1: Experimental Design.

(A) Apparatus: Remus, a remote ultrasound system described in [2, 3] delivers ultrasound to deep brain regions in awake, behaving nonhuman primates. Implanted head holder pins serve as electrodes for EEG recordings.

(B) Deep Brain Targets: The target locations surrounding the left and right lateral geniculate nucleus (LGN; green). The targets are taken from a single axial plane and varied by 4 mm in the front/back and left/right dimensions. The B and H mark the approximate locations of the array’s geometric focus within the sonication plane in subject B and H respectively. Relative to the transducer, the estimated depth of the plane containing the LGN is 58 mm in subject B and 51 mm in subject H.

MR thermometry—performed only once—is used to validate targeting. Targeting of the lateral geniculate nucleus (LGN) has been demonstrated previously [2, 3, 25]). Briefly, the subject was placed in a sphinx position and a custom 3D printed stand was mounted to a head frame which was affixed to the subject’s skull via four titanium pins. MR thermometry was performed simultaneously with high energy ultrasound sonications (sufficient to increase the temperature at target by about 2 C). The MR imaging was acquired with a 3D gradient recalled segmented echo planar imaging pulse sequence; TR = 24 ms, TE = 11 ms, BW = 592 Hz/pixel, flip angle 12, echo train length = 5 with monopolar readout, FOV = 144 x 117 x 36mm, resolution 1.5 x 1.5 x 3.0 mm, acquisition time 4.6 s per dynamic image.

The steering settings that resulted in a temperature rise within each LGN were saved for use in future trials. It is important to note that neuromodulation pulses are much shorter and have much lower duty cycles, thus heating is not expected in those experiments (see Discussion).

Ultrasound evoked potentials were measured across six sessions in subject B and eight sessions in subject H. The total number of trials delivered to each animal was the same (60 sonications per target). Some of the sessions in subject H were shorter because he was less willing to sit for the full session duration. After head fixation, each session proceeded in a dark room to minimize the influence of visual stimuli on the EEG recordings. We measured the EEG response during sonication of 24 targets arranged in two grids with three rows and four columns each. The grids were centered on the left and right LGN (Figure 1b). During a session each target was sonicated 5-10 times using a randomization without replacement approach. Sonication of the left and right hemisphere was strictly interleaved. All steering was strictly programmatic–the position of the transducer was identical across targets and sessions.

In each session, the time between sonications was random with a mean of 17 seconds. The maximum delay was 50 seconds and the minimum delay was 14 seconds. The standard deviation was 4 seconds. The average session length in subject B (6 total sessions) was 70 minutes, ranging from a minimum of 66 minutes to a maximum of 73 minutes. Each target was sonicated 10 times per session. The average session length in subject H (8 total sessions) was 50 minutes, ranging from a minimum of 33 minutes to a maximum of 68 minutes. In 4 of the 8 sessions, each target was sonicated 10 times and in the other four sessions each target was sonicated 5 times. Thus, each target was sonicated a total of 60 times in both subjects. At the conclusion of each session subjects were returned to their home environment—no other data were acquired during these sessions. A juice reward was delivered 8 seconds after each sonication to keep the subject content throughout the procedure.

EEG

The EEG response was measured using alligator clips which were electrically connected to the titanium pins which attach the frame to the subject’s head (Figure 1). This method of recording EEG signals has been reported previously [3]. The titanium pins are connected to the skull via titanium screws and thus provide a strong electrical connection to the brain. The placement of the pins is primarily driven by stability requirements for head fixation but the approximate 10/20 EEG electrode positions are P3 and P4 for the rear pins and FP1 and FP2 for the front pins.

The EEG signal was recorded on the two rear pins (alligator clips connected to the front pins served as ground). The impedance between the channels and ground was consistently below 1 kΩ. We recorded the EEG signals using a 128-channel recording system (RHS2000, Intan Technologies, Los Angeles, CA). EEG processing was performed by custom, in house MATLAB scripts. Each signal was low-pass filtered at 7.6 kHz and sampled at 20 kHz. The recorded signal showed strong interference at 60 Hz and at the second and third harmonics of 60 Hz, thus notch filters at 60, 120, and 180 Hz with a 4 Hz bandwidth were applied to the raw data before averaging. Trials in which the filtered EEG signal exceeded 500 μV were excluded since such high signals are not likely to be physiological. The temporal EEG response is then the EEG signal recorded in the one second after sonication, averaged across all sonications not excluded by the 500 μV threshold.

Behavior

In a separate experiment, we tested whether the strength of the ultrasound evoked potential detected in EEG could guide ultrasound neuromodulation of awake behavior. In subject B, we selected four targets - the two targets in each hemisphere with a maximal (active sonication) and minimal (control) EEG response. We then measured whether the strength of the ultrasound evoked potential predicted the behavioral effects of ultrasound neuromodulation delivered to each target.

Behavioral modulation was measured with a visual discrimination task that has been described in detail elsewhere [2, 5]. Briefly, the subject fixates on a central target. After a random delay, targets appear in the right/left visual hemifield, separated by a random delay between -90 and 90 ms (negative delays mean the right target appeared first while positive delays mean the left target appeared first). The subject is rewarded if they look at the target that appeared first within a 1.5 second period.

One sonication is delivered in each behavioral session (behavioral sessions were performed separately from EEG sessions and did not begin until after all EEG sessions were complete). We measured behavioral changes in 16 total sessions, four sessions per anatomical target (targets with maximal and minimal responses in both the left and right hemispheres). The sonication follows a baseline period of 375 trials. The task continues without interruption during and after the sonication and the animal is then allowed to work until they are satiated (water restriction is used to aide training of the subjects). The subject’s behavior is quantified by fitting a sigmoid to their choice behavior and identifying the delay at which the subject has equal preference for the left/right target. Changes in behavior are then quantified by the subject’s preference for the left target at the baseline delay. Behavioral trials were performed in a prospective study after EEG measurements. The location of the sonication was guided by the EEG result.

Ultrasound Parameters

The ultrasound parameters for the ultrasound evoked potential experiments and the behavioral experiments are given in Table 1. The parameters designed to elicit an ultrasound evoked potential (UEP) were selected to match a prior study in which changes in EEG rhythms were observed in anesthetized subjects [4]. Compared to existing studies of ultrasound neuromodulation in humans, the UEP parameters have a shorter pulse duration but higher time average, spatial peak intensity (I_SPTA) across that 100 ms duration (15 W/cm²) [1]. The I_SPTA across the entire session (averaged over the full 33-70 minute session time) was quite low (90 mW/cm²).

Table 1:

Sonication Parameters for the ultrasound evoked potential and behavioral experiments. Left to right columns: Spatial peak pulse average intensity (I_SPPA; W/cm²), duty cycle (DC; %), pressure (P; MPa), pulse repetition frequency (PRF; Hz), pulse repetition interval (PRI; ms), pulse duration (PD; ms), center frequency (F; MHz), mechanical index (MI), sonication duration (SD; ms).

Experiment	I SPPA	P	DC	PRF	PRI	PD	F	MI	SD
UEP	31	1.0	50	200	5	2.5	0.48	1.4	100
Behavior	7.7	0.5	14.4	4.8	208	30	0.48	0.72	30,000

In contrast to the parameters selected to elicit ultrasound evoked potentials, the parameters used in the behavioral study were designed to cause sustained changes in awake behavior. Thus, we selected sonication parameters for the behavioral experiments from a prior study showing changes in visual choice behavior during sonication of the LGN [3]. Note that the total energy delivered by these sonications is two orders of magnitude less than the energy used during thermometry. Thus, no significant heating is expected (see Discussion).

In-situ pressure was estimated by derating the free field intensity by a factor of 21%. The derivation of this estimate has been described previously [2, 3]. Briefly, MR thermometry data in two NHP subjects (subject B and a subject not included in this study) was combined with a simplified version of Pennes’ Bioheat equation [29] to estimate the pressure at target. The average de-rating factor across the two subjects was 21%.

The half power beamwidth (HPBW) of the transducer when steered to the approximate location of the LGN (11, 3, and 15 mm from the natural focus in the left/right, anterior/posterior, and superior/inferior dimensions respectively) was measured in water using a hydrophone. The resulting HPBW measured in free field was 1.5, 5, and 5 mm in the left/right, anterior/posterior, and superior/inferior dimensions respectively.

Results

We found that brief pulses of transcranial focused ultrasound applied to deep brain targets of NHPs elicited target-specific evoked EEG potentials. The targeting was achieved electronically; the transducer was fixed to the head always in the same location. Figure 2 shows the ultrasound-evoked potentials for each of the 24 targets (Figure 1). Each axis provides the average response to the 60 stimuli after removing trials in which the EEG voltage exceed 500 μV . The resulting number of averaged trials is shown in the bottom left of each plot. In both subjects, ultrasound elicited the strongest responses in the target located 4 mm medial and 4 mm posterior to the assumed location of the LGN (Figure 2). Relative to the onset of the ultrasound, the latencies of the negative peak ranged from 139 to 180 ms in subject B and 86 to 115 ms in subject H. This spatially confined response was consistent across subjects, in spite of differences in the placement of the ultrasonic transducer (Figure 1b). This result demonstrates the capacity of ultrasound to directly perturb neural activity in deep brain targets.

Figure 2: Deep Brain Region-Specific Ultrasound Evoked Potentials.

The EEG responses (mean ± s.e.m) as a function of time at the 24 target locations shown in Figure 1. The green rectangle shows the 100 ms sonication. Each plot is referenced to the stimulus onset. Results for Subject B are shown on top and results for Subject H are shown on bottom. The numbers in the bottom left of each axis give the number of averages used to create each plot (after removing traces in which the EEG exceeded 500 μV ). Time points at which a two tailed T-test reveals a significant deviation from the baseline measurement (average value between -500 and 0 ms) are marked with a black line at the top (p < 0.05) and bottom (p < 0.001) of each plot. Notably, the strongest ultrasound evoked potentials are observed at the same anatomical locations in both subjects.

We hypothesized that the target that evoked the strongest response would also lead to strongest effects on visual choice behavior [3]. In other words, we tested whether this ultrasound evoked response could be used to guide targeting to achieve behavioral effects. Indeed, Figure 3 shows that the evoked-potential-optimized targets lead to notable and significant (time window 17-22 minutes following sonication, n=4 left and n=4 right, two-tailed t-test, p < 0.05) effects on visual choice behavior (Figure 3a). No significant effect was observed for control targets (n=4 left, n=4 right; Figure 3b).

Modulation of Visual Choice Behavior.

Choice preference (mean ± s.e.m.) in subject B as a function of time, aligned to stimulus onset, in response to stimulation of the targets that resulted in maximal EEG response (left; see blue and red star) and minimal response (right; blue and red star). Each curve comprises n = 4 sessions.

Discussion

We demonstrated a measurable and transient physiological response to ultrasound stimulus of deep brain anatomy. The response is specific to the targeted anatomy and is engaged independently of the location of the transducer array. The latency of the evoked potential suggests that it takes some time for the ultrasound stimulation to elicit a response. This agrees with a prior study in mice which found latencies up to 200 ms when measuring evoked potentials using electromyography [27]. Behavioral data in one subject validates that the strength of the physiological response correlates with modulation of awake behavior. Indeed, we have observed that stimulation of those locations with maximal response led to substantial effects on visual choice behavior. Stimulation of locations without a physiological response did not affect choice behavior. These results provide evidence of the feasibility of developing treatment guidance algorithms that rely on physiological feedback to provide robust, repeatable ultrasound therapies. Such novel approaches could provide insight into both spatial targeting and intensity calibration for TUSN therapies.

Control for Artifacts

Studies in rodents, which have small cranium, have shown that ultrasound can engage auditory or vestibular pathways [30, 31]. We controlled for such confounds by working with non-human primates, in which the ultrasound focus is well confined with respect to the brain dimensions, and by evaluating responses to multiple deep brain targets in each session. The energy delivered to both the optimal and the control targets was the same, yet there was a clear target dependence for both the neural and behavioral effects. Thus, the effects cannot be explained by generic potential artifacts.

Non-Significant Thermal Effects

Previously published thermometry data [2] additionally allow us to estimate the temperature increase resulting from each 100 ms sonication. In that experiment, 5 seconds of continuous ultrasound was delivered at an input voltage of 20 Vp. In the first 2.3 seconds (the time it took to acquire one thermometry image) the measured temperature rise was approximately 1.8 degrees C. The sonication in this study delivered 100 ms of 50% duty cycle ultrasound at a voltage of 18.1 Vp. Assuming a linear relationship between input power and temperature rise, we thus expect the sonications in this study to result in a temperature increase of approximately 0.03 degrees C. Thus, it is unlikely that temperature played a significant role in evoking the EEG response.

Limitations and Future Work

This study has several important limitations. First, the study shows a robust, measurable EEG response in only one deep brain region. It is not clear whether other deep brain regions will similarly respond to ultrasound stimulus. Indeed, the study shows that—at least for the parameters used in these experiments—some deep brain regions do not produce response measurable by EEG electrodes placed over the occipital lobes. Thus, clinical translation of physiological feedback to guide TUSN interventions will depend on whether targets relevant to neurological disease also produce a reliable response. Therefore, future studies should seek to replicate these findings in both NHP and human subjects and to determine other neural targets that produce an EEG response.

A second, related limitation, is uncertainty in the specific neural anatomy underlying the responses shown in Figure 2. The maximum response did not occur at the assumed location of the LGN and the cause of this is unclear. One possibility is that ultrasound interacts more strongly with the white matter tracts posterior to the LGN than with the LGN itself [32]. An additional possibility is systematic error in the targeting protocol used to identify the LGN—this could be the result of inaccuracies in the overall protocol or a result of targeting the wrong portion of the LGN which is larger than the HPBW of the transducer [3]. The EEG readout similarly lacks the spatial specificity required to definitively state that the measured evoked potential results from activity in the visual cortex. Thus, while changes in visual choice behavior show that the target has some effect on visual circuits, this study cannot conclusively state whether the measured evoked potential is a result of direct modulation of the visual system. Future studies could use denser electrode arrays or invasive methods to better identify the specific neuroanatomy involved in the measured response.

The estimated pressure delivered to the target region constitutes another limitation because it is not subject or target specific, it relies on assumptions about the thermal properties of the brain, and it does not account for the temporal and spatial averaging inherent in MR thermometry [3]. Thus, there remains considerable uncertainty in the estimated pressure at target.

Due to constraints on the NHP subject’s time, data demonstrating a correlation between ultrasound evoked potentials and ultrasound induced modulation of choice behavior was performed in only one subject. As a result, the behavioral data can only offer a proof of concept and do not offer definitive evidence of a link between the magnitude of the UEP and the efficacy of altering behavior by sonication of the targeted anatomy. Future work is required to confirm whether the magnitude of the UEP is directly correlated to behavioral changes resulting from TUSN.

Conclusion

This study shows that ultrasound modulation of deep brain targets can elicit robust evoked potentials. Ultrasound based modulation of behavior correlates with the magnitude of the evoked potential. A robust EEG response to ultrasound may provide a physiological marker to validate target engagement during neuromodulation procedures. Such validation would greatly enhance the efficacy of neuromodulation protocols that are plagued by uncertainty in the acoustic intensity delivered to the targeted tissue [19, 33].