Abstract
Transcranial magnetic stimulation (TMS) is a noninvasive brain stimulation method that is rapidly growing in popularity for studying causal brain–behavior relationships. However, its dose-dependent centrally induced neural mechanisms and peripherally induced sensory costimulation effects remain debated. Understanding how TMS stimulation parameters affect brain responses is vital for the rational design of TMS protocols. Studying these mechanisms in humans is challenging because of the limited spatiotemporal resolution of available noninvasive neuroimaging methods. Here, we leverage invasive recordings of local field potentials in a male and a female nonhuman primate (rhesus macaque) to study TMS mesoscale responses. We demonstrate that early TMS-evoked potentials show a sigmoidal dose–response curve with stimulation intensity. We further show that stimulation responses are spatially specific. We use several control conditions to dissociate centrally induced neural responses from auditory and somatosensory coactivation. These results provide crucial evidence regarding TMS neural effects at the brain circuit level. Our findings are highly relevant for interpreting human TMS studies and biomarker developments for TMS target engagement in clinical applications.
SIGNIFICANCE STATEMENT Transcranial magnetic stimulation (TMS) is a widely used noninvasive brain stimulation method to stimulate the human brain. To advance its utility for clinical applications, a clear understanding of its underlying physiological mechanisms is crucial. Here, we perform invasive electrophysiological recordings in the nonhuman primate brain during TMS, achieving a spatiotemporal precision not available in human EEG experiments. We find that evoked potentials are dose dependent and spatially specific, and can be separated from peripheral stimulation effects. This means that TMS-evoked responses can indicate a direct physiological stimulation response. Our work has important implications for the interpretation of human TMS-EEG recordings and biomarker development.
Introduction
Modulating local neural activity has great potential to restore affected brain regions in neurologic and psychiatric disorders. Noninvasive neuromodulation is a cost-effective tool for large-scale use in patients with a wide range of symptoms and symptom severity. The current gold standard for noninvasive neuromodulation in clinical and basic research is transcranial magnetic stimulation (TMS). TMS can induce action potentials in neurons and, through its repeated application, brain plasticity (Huang et al., 2005; Hoogendam et al., 2010). TMS-based therapies received Food and Drug Administration approval to treat major depressive disorder, obsessive-compulsive disorder, and nicotine addiction (Brunoni et al., 2017; Lefaucheur et al., 2020). Despite its increasing use in research and clinical applications, the physiological effects of TMS are not fully understood. This gap of knowledge has hampered rational dose selection and optimization of TMS parameters. In particular, a clear physiological marker of TMS target engagement would be crucial to advance clinical applications in brain regions other than the motor cortex, such as the dorsolateral prefrontal cortex (DLPFC).
In human research, TMS physiological effects can be measured noninvasively through motor-evoked potentials (MEPs; Brasil-Neto et al., 1992; Numssen et al., 2021; Wischnewski et al., 2022), functional magnetic resonance imaging (MRI; Nahas et al., 2001a; Denslow et al., 2005; Oathes et al., 2021), and electroencephalography (EEG; Ilmoniemi et al., 1997; Massimini et al., 2005; Bonato et al., 2006; Daskalakis et al., 2008; Thut et al., 2011; Rogasch et al., 2015). Combined TMS-EEG is a promising method to study neural population responses to TMS with high temporal accuracy. A primary way of investigating transient stimulation effects on EEG is by measuring TMS-evoked potentials (TEPs). For example, causal brain dynamics during TMS perturbation in motor, visual, and prefrontal brain regions (Ilmoniemi et al., 1997; Bonato et al., 2006) and modulation of brain state during behavior (Massimini et al., 2005; Thut et al., 2011) have been studied through TMS-EEG. It has also been used to quantify cortical inhibition paradigms that are resulting from TMS perturbation (Daskalakis et al., 2008; Rogasch et al., 2015). However, despite its promise for studying the neural responses of TMS, the interpretation of human TMS-EEG neurophysiological results is challenging.
One challenge for the interpretation of TMS-EEG is a poor spatial localization of results because of volume conduction, meaning that the location of TEP sources is unclear. Second, multisensory side effects during TMS include auditory and somatosensory coactivation (by TMS click sound and muscle and peripheral nerve costimulation), which also affect brain responses. This results in a challenge to disentangle stimulation effects induced in the brain because of the TMS electric field and peripheral sensory stimulation effects (Ilmoniemi and Kičić, 2010; Conde et al., 2019). For consistency, we define the following: (1) centrally induced TMS effects as the effects originating in the CNS [for a given brain area, such effects can be because of the (i) local electric field inducing suprathreshold depolarization of neurons or (ii) circuit-mediated activation propagated from another brain area where the TMS electric fields induced suprathreshold depolarization]; and (2) peripherally induced stimulation effects originate in the peripheral nervous system in the form of auditory or somatosensory activation and later propagate to the CNS. To systematically investigate TMS physiological mechanisms, several control conditions carefully accounting for various confounds are required (Biabani et al., 2019; Rocchi et al., 2021). Nonhuman primate (NHP) models can overcome the main limitations of human TMS-EEG. Using invasive electrophysiological recordings in NHPs offers a unique opportunity to study the physiological effects of TMS with high spatiotemporal precision and control for somatosensory effects. Further, compared with smaller animal models, the size and cortical folding patterns of the NHP brain will result in comparable TMS-induced electric fields as in the human brain, making them an ideal translational model (Alekseichuk et al., 2019).
Previous work using TMS in NHPs has studied the effect of single-pulse TMS on single-unit activity (Mueller et al., 2014; Romero et al., 2019). These findings have shown that TMS (compared with sham stimulation) causes neural activity within milliseconds, and this effect is observed in various neuron types (Mueller et al., 2014). Furthermore, these effects are specific to the targeted area and affect behavioral responses (Romero et al., 2019). While these studies have resulted in important insights into TMS mechanisms, they were limited to the microscale neurophysiological level and a single brain location. However, it is well understood that TMS effects are widespread, involving several brain networks (Fox et al., 2012; Ozdemir et al., 2020). Furthermore, it is not straightforward to translate the findings from cellular response to EEG. On the other hand, invasive stereo-EEG recordings can measure mesoscale, neurocircuit activity across the brain while preserving excellent spatiotemporal precision. In addition, one can directly translate findings in local field potentials (LFPs) to the macroscale of human EEG.
Here, we record TMS-evoked responses in the NHP brains with implanted depth electrodes along a whole hemisphere. Our experimental setup allows for answering previously unanswered key questions of TMS dose response and spatial specificity. First, we study the effects of TMS intensity and coil location on TEPs. Second, we measure TMS multisensory coactivation (both auditory and somatosensory effects) through exhaustive control experiments. Our results demonstrate clear TMS-evoked brain responses, that are separable from peripheral effects, with increasing doses corresponding to larger evoked responses. Additionally, we show spatial specificity of these physiological responses. Our findings are crucial for the interpretation of human TMS-EEG studies, rational dose control, and the development of robust markers of target engagement based on TMS-evoked potentials.
Materials and Methods
Subjects and surgical procedures.
All procedures were approved by the Animal Care and Use Committee of the Nathan S. Kline Institute for Psychiatric Research. We used two adult nonhuman primates (Macaca mulatta) in this study (Monkey W: female, 5 kg; Monkey H: male, 10 kg). The monkeys were implanted with an MRI-compatible polyetheretherketone headpost positioned over the occipitoparietal region. Three multicontact stereo EEG (sEEG) depth electrodes (spacing, 5 mm; Ad-Tech) were permanently implanted through a skull incision over the left occipital cortex. Recording electrodes were oriented along the posterior–anterior axis with medial prefrontal cortex, frontal eye field (FEF), auditory cortex (AUD), and temporal cortex (TEM) as the end points. In Monkey W, an additional electrode targeted the geniculate complex of the thalamus.
Magnetic resonance imaging and electrode localization.
For both monkeys, we acquired the T1 and T2 spin echo sequences using a Siemens TIM Trio 3 T scanner with the following parameters: pixel dimensions = 0.5 × 0.5 × 0.5 mm; flip angle = 80°; TR = 2600 ms; TE = 3.55 ms; and TI = 900 ms. Exact electrode positions were identified on a postimplantation MR image, registered to the preimplantation image, and referenced to the common stereotaxic atlas (Saleem and Logothetis, 2012; Tables 1, 2, for details). MRI registration was done using the FSL FLIRT package (Jenkinson and Smith, 2001; Jenkinson et al., 2002; Greve and Fischl, 2009). The electrode positions with respect to the brain are shown in Figure 1a.
Monkey W—the anatomic regions associated with each contact of the sEEG electrodes
Monkey H—the anatomic regions associated with each contact of the sEEG electrodes
Overview of electrode implantation and experimental setup. a, Contact locations of invasive depth electrodes and TMS coil locations in the reconstructed head and brain surfaces of each monkey. The electrode shafts are directed at frontal eye fields (blue) and auditory (green) and temporal (red; superior temporal in Monkey W and inferior temporal in Monkey H) regions of the brain. TMS is delivered to 5 locations (numbered) in Monkey W and 6 locations in Monkey H. b, Active TMS is delivered at 5 intensities (10, 25, 50, 70, and 90% of the MSO) to both monkeys, with an additional power mode setting of 125% MSO in Monkey H. For auditory control, TMS click is delivered at 70% MSO by turning the coil by 90° away from the head, to mimic the auditory response. For somatosensory control, the scalp is stimulated with 6 mA electric pulse while delivering TMS click at 70% MSO. This control was performed at location 2 in Monkey H. To control the effects of auditory coactivation, active TMS was delivered at 70% MSO with and without auditory masking at locations 1 and 5 in Monkey H.
The brain morphology of the two monkeys was assessed by an expert in nonhuman primate anatomy. Monkey W has typical gross morphology with normal sulcal variability throughout the brain. Monkey H shows a variant of normal anatomy with multiple microgyri in the parietal and occipital regions as well as enlarged ventricles in the posterior regions. The frontal, temporal, and deep brain areas are within the typical range of gross anatomy.
Transcranial magnetic stimulation.
We performed the TMS experiments on the two nonhuman primate subjects in sphinx position. Monkeys were anesthetized with Dexdomitor 0.015 mg/kg, i.m.; ketamine 6 mg/kg, i.m.; and atropine 0.045 mg/kg, i.m., followed by 1–2% isoflurane. In all experimental sessions, Biphasic single-pulse TMS was delivered using a magnetic stimulator (model MagPro X100, MagVenture) with a butterfly coil (model MC-B35, MagVenture). The coil was positioned at locations designated in Figure 1a above the prefrontal cortex (Monkey W and H positions 1 and 2), premotor cortex (Monkey H, position 3), temporal areas (Monkey W, position 3; Monkey H, position 4), and auditory cortex (Monkey W, position 4; Monkey H, position 5). Stimulation location contralateral to the implanted sEEG electrodes, above the right temporal cortex (Monkey W, position 5; Monkey H, position 6) was used for both monkeys as a control condition. For frontal stimulation locations (Monkey W, positions 1–3; Monkey H, positions 1–4), the coil orientation was maintained parallel to anterior–posterior line. Because of the intrusion of the headpost on the lateral regions, for lateral stimulation locations (Monkey W, positions 4 and 5; Monkey H, positions 5 and 6), the coil was oriented ∼45° with respect to the anterior–posterior line. We ensured that the connectors and the wires to the electrodes were directed away from the coil and the stimulator, and the amplifier was substantially away from the monkey head during stimulation. For each location, we delivered five different TMS intensities in both NHPs: 10%, 25%, 50%, 70%, and 90% of maximum stimulator output (MSO). In addition, in Monkey H we applied 90% of maximum in power mode, equivalent to ∼125% standard MSO. These intensities are linearly related to the current input (coil dI/dt) as follows: 10% MSO = 13 A/μs, 25% MSO = 37 A/μs, 50% MSO = 76 A/μs, 70% MSO = 107 A/μs, 90% MSO = 142 A/μs, and 125% MSO = 201 A/μs. Each stimulation block (condition × location × intensity) consisted of ∼80 TMS pulses given every 3–5 s.
We implemented the following experimental conditions (Fig. 1b): (1) active TMS, (2) passive TMS (produces auditory click associated with TMS), (3) passive TMS with somatosensory control (adds focal electric pulse to the scalp), and (4) active TMS with auditory masking (continuous auditory white noise-masking auditory click associated with TMS).
In the passive TMS condition, we turned the TMS coil away from the head to solely mimic the TMS auditory response because of the click sound of the coil without direct brain stimulation. For auditory/somatosensory costimulation, we used a coil (Cool-B65 A/P, MagVenture) that concurrently provides electric current stimulation to the scalp. To study whether the current stimulation alone elicits peripheral effects, we used the sham coil in combination with the current stimulation over stimulation location 1. We delivered an auditory click at 70% MSO along with 6 mA electrical stimulation on the scalp using pairs of disk electrodes of 10 mm diameter (3 M) placed across the region of interest. This sham stimulation approach follows (Conde et al., 2019). Given that the cutaneous perception of electrical stimulation because of active TMS and somatosensory control cannot be assessed subjectively through feedback from nonhuman primates, we visually assessed and confirmed equivalent scalp and shoulder twitches during active and sham conditions.
For auditory masking, we created a masking acoustic noise using the TMS Adaptable Auditory Control tool (Russo et al., 2022). The sound pressure level of the masking noise was set to 80 dB for the above stimulation intensities. Sound pressure level was measured at the ear canal using a sound meter (Brüel & Kjær). The masking noise was presented via two speakers placed 50 mm from the head on each side. The masking was applied concurrently with active TMS at stimulation locations 1 and 5 with the intensity of 70% MSO.
Local field potentials.
We recorded the LFPs using 32-channel ActiveTwo amplifier (Cortech) at a sampling rate of 40 kHz. For both monkeys, contacts in the occipital cortex (proximal to the headpost) were chosen as the reference and the ground (9th and 11th contacts from the front in the FEF electrode, respectively). To prevent the electric current induction in the electrode contacts because of the magnetic field, we made sure that the recording setup, including the electrode contacts and the electrical cables, were free of loops. Furthermore, high-input resistance of the amplifier (1 GΩ) would minimize induced currents to negligible levels, even if they occur.
Data preprocessing was done in MATLAB (MathWorks) using the FieldTrip toolbox (Oostenveld et al., 2011). The data acquired from the amplifier were preprocessed by separating data into epochs time locked to TMS delivery (t = 0). The epoch length was 4 s (2 s before and 2 s after TMS delivery). The epochs were then detrended and demeaned. In the next step, we removed the TMS pulse artifact (−0.5 to 0.4 ms) and the TMS-induced muscle artifact (0.4 to 25 ms in both monkeys) by padding the time windows. The time windows to be excluded were determined by the visual inspection of the LFPs and concurrent electromyographic activity recorded from scalp (Fig. 2a,b). We interpolated the excluded time window using the cubic Hermite interpolating polynomial (pchip) with before and after data segments of 200 ms. We filtered the resulting LFPs using a bandstop Butterworth filter of order 2 with cutoff frequencies at 57 and 63 Hz. Noisy contacts and trials were excluded through visual inspection; 11 ± 6 (mean ± SD) trials were excluded per experimental block. We performed independent component analysis to manually remove residual artifacts (one component on average per experimental block; maximum of four components). The resulting time series data were then resampled to 1 kHz and bandpass filtered from 1 to 50 Hz using a fourth-order Butterworth filter, followed by baseline correction (Fig. 2c). We processed baseline data by splitting the traces into 4 s epochs with 50% overlap. We followed the same preprocessing steps for baseline data except for artifact removal and interpolation. For subsequent analysis, we excluded the sEEG contacts located in white matter (WM) or outside the brain.
Preprocessing of the LFP data. a, Top, A fragment of the raw sEEG signal with TMS pulses marked by arrows. Bottom, A representative trial depicting the presence of TMS pulse, muscle artifact, and neural signal in the sEEG recording. The range of amplitude has been truncated to clearly show the magnitude of the muscle artifact. b, Electromyographic (EMG) recordings obtained by needle electrodes inserted into left frontalis and left posterior temporalis muscles in Monkey W. The general area of the muscle is shaded on the reconstructed head model (blue, left frontalis; orange, left posterior temporalis). Top, Time-locked EMG signals resulting from stimulating frontal brain region. Frontal muscle, which is closest to the stimulation site, shows higher EMG activity compared with posterior muscle. Bottom, Time-locked EEG signals resulting from stimulating posterior brain region. In both cases, the muscle activity does not exceed the 15 ms time window that is shown in dashed vertical lines. Stimulation locations are shown by X. Time-locked EEG signals are calculated by interpolating the TMS pulse duration and calculating the average across trials. c, The detailed pipeline used for preprocessing sEEG signals.
Electric field modeling.
We modeled the electric fields induced by TMS using SimNIBS 3 (Thielscher et al., 2015). We extracted the gray matter (GM) and white matter surfaces using a modified Human Connectome Project pipeline for nonhuman primates (Fair et al., 2020). The surfaces for scalp, skull, and CSF were created by manual segmentation. Once the complete head mesh was created, we ran simulations for the TMS coil positions used experimentally. We assigned fixed conductivities to each layer, as follows: 0.126 S/m (white matter), 0.275 S/m (gray matter), 1.654 S/m (CSF), 0.01 S/m (skull), and 0.465 S/m (scalp; Opitz et al., 2011). The electric fields were calculated in a quasistatic regime (Thielscher et al., 2011). Through electric field simulations, we sought to visualize the affected brain region and the extent of activation. We computed the robust maximum (99.95th percentile) of the electric field magnitude to remove numerical inaccuracies. To calculate the spatial similarity of electric field distributions on gray matter surface, we first obtained the binary mask by thresholding the electric field magnitude at 50% of the robust maximum. Then, we computed the Sørensen–Dice similarity coefficients on pairs of gray matter surfaces.
Data analysis.
We performed a three-step analysis on the electrophysiological data. First, we analyzed the dose dependency of identified TEPs. Second, we investigated the effect of coil-to-contact distance on the TEP response. Finally, we evaluated the region-specific differences in LFPs and TEP components in the presence of multisensory coactivation.
Following the data preprocessing, we estimated the significance of time-locked LFPs or TEPs. The average time-locked activity was calculated for each experimental block. For each stimulation location, we performed a nonparametric cluster-based permutation test (Oostenveld et al., 2011; p < 0.05) to investigate whether a significant main effect of stimulation intensity is present following TMS (25–400 ms after stimulation pulse) relative to the baseline. The F-statistic was calculated for each contact/time pair and clustered according to their temporal adjacency. We used the sum of the F values within a cluster as a cluster-level statistic and compared it against a permutation distribution of 1000 permutations to determine the p value. The highest F value per cluster is reported in Results. The contacts in which the effect was present during at least one stimulation intensity were considered to be “responsive” contacts.
Identified significant TEPs were named according to the time point of the maximum deflection (e.g., TEP with the maximum negative deflection at 50 ms is N50). We extracted the N50 component as the minimum amplitude of the LFP in the time window of 40–70 ms from TMS onset, per each trial. To analyze the dose dependency of the TEP components, we extracted N50 for all intensities, from all responsive contacts, and from each stimulation location. TEP amplitudes were normalized to the maximum at each stimulation location. The responses from both monkeys, responsive contacts, and stimulation locations were pooled and fit to a four-parameter logistic function, as follows:
Next, we studied the effect of stimulation location on the TEP response through a distance–response relationship. We calculated the Euclidian distances from the TMS coil center projected on the scalp to each electrode contact at each stimulation location. Here, we excluded the contacts without TEP response relative to the baseline (N50 amplitude is within the 99th percentile of the mean LFP amplitude). We assessed the relationship between the distance and response by fitting a linear (y = a * x + b) and an exponential decay function (y = a + b/x3). The goodness-of-fit was estimated using adjusted R2.
Furthermore, we studied the LFPs for the different control conditions. To investigate the peripheral auditory effects, we compared active TMS with and without auditory masking. First, we calculated the averaged total response by combining the trials from all responsive contacts in both tested stimulation locations (locations 1 and 5). To narrow down the auditory and nonauditory responses, we grouped the response of auditory and nonauditory contacts separately and calculated the average response for each condition (i.e., masked and nonmasked).
To test whether there is a statistically significant effect of masking on the TEP responses, we extracted the TEP components from all the trials, and performed Wilcoxon rank-sum test (p < 0.05) on the total, auditory, and nonauditory TEP responses.
Finally, we studied the effects of auditory click and peripheral somatosensory effects by comparing active TMS, passive TMS, and passive TMS plus electrical stimulation. Similar to the auditory masking analysis, we calculated the total averaged response from all responsive contacts. Then we grouped the auditory and nonauditory contacts and calculated the average response. To study the effect of auditory click and electrical stimulation on TEP responses, we extracted TEP responses from all contacts and performed the Kruskal–Wallis test (p < 0.05) on total, auditory, and nonauditory TEP responses. We performed follow-up multiple comparisons to evaluate the significant differences across the conditions.
Results
TMS-induced neural effects show a dose dependency
To study the stimulation intensity (dose)-dependent effects of TMS on time-locked LFPs, we first identified time windows that showed a significantly different LFP amplitude after stimulation pulse relative to the baseline for at least one stimulation intensity. This was done at five and six separate stimulation locations for two monkeys (named W and H, respectively). The responses were tested using a nonparametric cluster-based permutation test. The TMS-evoked LFPs from both monkeys show high-amplitude activity at ∼25–100 ms after stimulation pulse followed by a rebound. In both monkeys, this pattern of activity is conspicuous at the stronger stimulation intensities (≥50% of MSO).
In Monkey W, on stimulation of location 1, three frontal contacts showed significant responses (29–102 ms, Fmax = 7.93–11.89, p = 0.003–0.037). Stimulation of location 2 showed significance in two frontal contacts (25–99 ms, Fmax = 6.45–9.53, p = 0.005–0.046). Stimulation of locations 3 and 4 resulted in significant responses across four frontal contacts (25–99 ms, Fmax = 12.36–24.16, p = 9.9 × 10−4 to 0.0017) and five frontal contacts (25–133 ms, Fmax = 13.27–49.52, p = 9.9 × 10−4 to 0.003), respectively. The contacts mentioned above were located in the corpus striatum and anterior cingulate cortex (Table 1, details). Stimulation location 4 also led to significant responses in the two contacts within the auditory belt (25–91 ms, Fmax = 19.93–23.92, p = 9.9 × 10−4 to 0.004) and a cluster of four contacts within the superior temporal plane (25–111 ms, Fmax = 9.35–21.65, p = 9.9 × 10−4 to 0.014). The responses from stimulation location 4 is shown in Figure 3b (top). Figure 4a and Table 3 contain responses from all stimulation locations and summary statistics. In Monkey H, the main stimulation locations 1, 2, 4, and 5 showed early latency responses in the contacts that are in the orbitofrontal cortex, corpus striatum, and anterior cingulate [i.e., three contacts in location 1 (25–93 ms, Fmax = 6.63–12.95, p = 0.0015–0.05), three contacts in location 2 (27–93 ms, Fmax = 7.39–23.31, p = 9.9 × 10−4 to 0.03), three contacts in location 4 (25–130 ms, Fmax = 8.08–18.25, p = 9.9 × 10−4 to 0.042), and five contacts in location 5 (25–133 ms, Fmax = 12.17–44.03, p = 9.9 × 10−4)]. For all main stimulation locations, the contacts in the auditory cortex saw early latency responses (25–145 ms, Fmax = 6.64–22.61, p = 9.9 × 10−4 to 0.045). Locations 1, 3, 4, and 5 also showed significant responses in the contacts in the ventral insular region (25–161 ms, Fmax = 4.40–18.85, p = 9.9 × 10−4 to 0.045). Furthermore, all main stimulation locations resulted in significant responses in the superior temporal and the middle temporal regions, while the responses because of stimulating location 5 resulted in longer time windows (locations 1, 2, 3, and 4: 25–117 ms, Fmax = 5.46–11.79, p = 0.003–0.041; location 5: 33–174 ms, Fmax = 10.03–31.64, p = 9.9 × 10−4). The responses from stimulation location 5 are shown in Figure 3b (bottom). Figure 4b and Table 4 contain responses from all stimulation locations and summary statistics. The control stimulation location (location 5) in Monkey W did not elicit a significant response in the LFPs. However, in Monkey H, control stimulation location (location 6) resulted in significant early latency responses in the frontal contacts. The LFPs resulting from stimulating the control location compared with its ipsilateral counterpart is shown in Figure 5, a and b, for Monkey W and H, respectively.
Monkey W—detailed statistics of cluster-based permutation tests to identify main effects of intensity
Monkey H—detailed statistics of cluster-based permutation tests to identify main effects of intensity
TMS-induced early evoked potential N50. a, Aggregated magnitude of early evoked potential N50 in all responsive contacts (significant for at least one stimulation intensity) in both monkeys. The N50 amplitude is normalized per stimulation location per monkey. The curve shows the sigmoidal (logistic) dose–response function with the upper and lower confidence bounds (shaded). The mean squared error of the fit is 0.018. b, Left, The 3D rendering of the brain with the recording contacts (responsive contacts, dark orange) in both monkeys for a representative stimulation location. Right, Normalized LFPs (overlayed responsive contacts). Responsive contacts were determined by a nonparametric cluster-based permutation test, *p < 0.05.
LFPs from responsive contacts for each stimulation location. a, Time-locked averaged signal from responsive contacts for each main stimulation location (1–4) in Monkey W. b, Time-locked averaged signals from responsive contacts for each main stimulation location (1–5) in Monkey H. All the signals are normalized to the maximum amplitude of each contact. The responsive contacts are shaded in orange on the 3D rendering of the brain. The time window during which a significant effect of stimulation intensity on the LFPs is observed, is shaded in gray for each stimulation location.
Comparison of local field potentials between left and right hemispheres. a, The butterfly plot of the local field potentials from all channels when stimulating location 4 (ipsilateral to recording hemisphere) and location 5 (contralateral control location) in Monkey W. In Monkey W, the overall activity is lower when stimulating the control location compared with stimulation location 4. b, Same plots for Monkey H (location 5, ipsilateral to recording hemisphere; location 6, contralateral control location). In Monkey H, the control location induces activity in the recording hemisphere, albeit less than that of location 5.
In the early latency time windows, the LFPs showed a maximum deflection ∼50 ms after TMS delivery. This N50 component demonstrated a pronounced increase in its magnitude with increasing stimulation intensity in the lateral-most stimulation location (locations 3 and 4 in Monkey W, locations 4 and 5 in Monkey H) in both monkeys. In locations 1 and 2 in Monkey W and locations 1, 2, and 3 in Monkey H, the N50 component showed a robust response for higher stimulation intensities (90% MSO and 125% MSO). We extracted the N50 component from all responsive contacts across all three electrodes from both monkeys to investigate the dose-dependent behavior. The number of total responsive contacts was 20 (3 + 2 + 4 + 11) and 57 (11 + 10 + 8 + 12 + 16) of 108 and 135 in Monkeys W and H, correspondingly. The N50 components were normalized to the maximum N50 value for each stimulation location. To estimate the overall dose dependency of N50, we fit a four-parameter sigmoid (logistic) curve with TMS dose as the independent variable and the normalized N50 deflection as the dependent variable (Fig. 3a). The mean squared error of the fit was 0.018.
Effect of coil location and response–distance relationship
We performed computational simulations to evaluate the spatial extent of the TMS-induced electric field in the brain under different coil positions (Fig. 6a). As shown in Figure 6b, in Monkey W, a pairwise Sørensen-Dice similarity coefficient showed anticipated moderate overlap between the main stimulation locations 2–4 (DICE score, 0.19–0.56). In Monkey H, the frontomedial stimulation locations 1, 2, and 3 showed high spatial similarities (DICE score, 0.54–0.75), while the lateral locations 4 and 5 showed moderate similarity between themselves (DICE score, 0.36). The control stimulation location in Monkey W showed no impact on the recorded hemisphere (DICE scores with the main stimulation locations, <0.01). In Monkey H, the control stimulation location showed little-to-no impact on recorded hemisphere (DICE scores with the stimulation locations 1, 2, 3, and 4 was <0.15) except for stimulation location 3 (DICE score, 0.29). The sagittal plane of the models is given in Figure 7. The recording electrodes were predominantly within the volume of >50% of the maximum electric field magnitude, which we define here as an area of stimulation.
TMS-induced electric fields on cortical surfaces. a, The simulations of the TMS-induced electric field strength on the brain surface are shown in Monkeys W and H for each coil location. The electric field strength is color coded in relative units with respect to the maximum value. In Monkey W, locations 1 (anterior medial) and 2–4 (lateral) constitute main experimental conditions, and location 5 is the contralateral control condition. In Monkey H, locations 1–3 (medial) and 4 and 5 (lateral) are main experimental conditions, and contralateral location 6 is the control. b, The spatial similarity between TMS-induced electric fields as quantified by Sørensen-Dice similarity coefficient. In Monkey W (top), main stimulation locations demonstrated moderate spatial similarity between pairs (DICE score, 0.19–0.56), while the control stimulation location showed no impact on the recorded hemisphere. In Monkey H (bottom), stimulation locations 1, 2, and 3 showed high spatial similarities (DICE score, 0.54–0.75), while the lateral locations 4 and 5 showed moderate similarity between themselves (DICE score, 0.36). The control stimulation location showed little impact on the recorded hemisphere except for stimulation location 3 (DICE score, 0.29).
Profile of the penetration of electric fields at electrode locations. The simulation results show the electric field magnitude captured by the electrode sites. The cross sections along the depth electrodes are shown in dotted red lines, and the white dots show the contacts of the electrodes. The maximum is the 75th percentile of the electric field magnitude. We observe that significant electric fields reach the frontal regions in almost all stimulation locations, which would potentially contribute to TEPs. However, electric fields rarely penetrate into the lateral regions, and the TEP responses from these regions could be a result of activation because of neural circuits.
We next explored the relationship between stimulation location and the N50 response. We focus on the high-intensity (90% MSO) condition as it elicits a robust response across most contacts for both monkeys. We combined the stimulation locations 1, 2, and 3 in Monkey W and locations 1, 2, 3, and 4 in Monkey H for this analysis that were comparable in coil orientation. The lateral-most locations (4 in Monkey W and 5 in Monkey H) were excluded given the different coil orientation. Thus, N = 33 (15 + 7+11) and N = 49 (14 + 11 + 11 + 13) contacts in both monkeys from the said stimulation locations were included in this analysis. To assess the relationship between the N50 response and the coil-to-contact distance, we fit exponential decay and linear models. A negative linear relationship between the N50 response and the coil-to-contact distance in Monkeys W (R2 adjusted = 0.51) and H (R2 adjusted = 0.23) is significant but less accurate than a decaying relationship. The exponential decay fit is much better for Monkey W (R2 adjusted = 0.74) and similar to the linear fit in Monkey H (R2 adjusted = 0.18). The resulting exponential decay fit is shown in Figure 8.
Relationship between N50 response and the coil-to-contact distance. Scatter plots show the amplitude of the N50 response against the coil-to-contact distance for each stimulation location. Stimulation locations 1–3 from Monkey W and locations 1–4 from Monkey H are included in this analysis. The relationship between the N50 amplitude and the coil-to-contact distance in Monkeys W (R2 adjusted = 0.74) and H (R2 adjusted = 0.18) show an exponential decay.
TMS induces central and peripheral stimulation effects in the auditory cortex
We performed three control experiments to study the effects of peripheral auditory and somatosensory costimulation. First, we compared auditory and nonauditory regions during active TMS and active TMS with a masking noise, which suppresses the TMS-associated click sound. For this control, we stimulated locations 1 and 5 in Monkey H and extracted LFPs from all contacts that were responsive for both locations. We computed the average LFP responses across all contacts and separately for contacts in auditory and nonauditory brain areas for masked and unmasked conditions separately. The average nonauditory response did not show a visible change between the masked and unmasked conditions. However, the average response and average auditory response show a reduction in amplitude in the early time window in the masked condition compared with the unmasked condition (Fig. 9a). We further investigated this by extracting the N50 components from the LFPs. The components were normalized to the maximum at each stimulation location. We performed a Wilcoxon rank-sum test to compare the N50 responses between masked and unmasked conditions for both regions. We found that nonauditory regions are immune to masking (z = 1.09, p = 0.27), whereas auditory regions show a significantly lower N50 response during masking (z = 5.36, p = 8.1 × 10−8). The average response also shows a decreased response when masked (z = 3.47, p = 5.1 × 10−4). The results suggest that TMS creates centrally induced effects on neuronal populations in the stimulation area, while also inducing a separate response in the auditory cortex because of stimulation click sound.
Centrally induced TMS neural effects are separable from peripheral auditory and somatosensory activation. Total average response, average nonauditory response, and average auditory response are calculated from the responsive contacts. The responsive contacts are shaded in blue (nonauditory) and green (auditory). The TMS coil locations are indicated by X above the cortical surface. The N50 responses are extracted on a trial basis for comparison across controls. a, Auditory masking results in diminished N50 response localized to the auditory region when active TMS is delivered. The Wilcoxon rank-sum test indicates a significantly lower N50 response in the average response (p = 5.1 × 10−4). When the responses are clustered into auditory and nonauditory regions, the same test indicates that the N50 response in the auditory region shows a significantly lower response (p = 8.1 × 10−8) in the masked condition compared with unmasked when delivering active TMS. The test returns no significant difference for the N50 response between masked and unmasked conditions in the nonauditory regions. b, Auditory click and auditory click plus electrical stimulation do not induce N50 responses comparable to active TMS. Kruskal–Wallis tests for the N50 component for total response, nonauditory response, and auditory response showed a significant difference across the three conditions (total: χ2 = 114.68, p = 1.25 × 10−25; auditory: χ2 = 30.56, p = 2.31 × 10−7; nonauditory: χ2 = 89.15, p = 4.38 × 10−20). Follow-up multiple-comparisons tests revealed that the total N50 response was significantly different for the active TMS versus click (p = 9.56 × 10−9) and active TMS versus click with electric stimulation (ES; p = 9.56 × 10−9). The auditory N50 response was also significantly different for the active TMS versus click (p = 5.29 × 10−5) and the active TMS versus click with ES (p = 7.26 × 10−7). A similar observation was made on the nonauditory N50 response where the active TMS versus click (p = 9.74 × 10−10) and active TMS versus click with ES (p = 9.56 × 10−10) were significantly different. n.s., Not significant. *p < 0.05.
Auditory controls show a localized response in the auditory cortex
We compared the effects of auditory click control condition and auditory click with electric scalp stimulation to active TMS (Fig. 9b). As in the previous section, we considered the total, auditory, and nonauditory responses across conditions. The average LFPs from all three groups showed a pronounced N50 response in the active TMS condition (Fig. 9b). Click only and click with electric stimulation did not show a pronounced N50 response in the total and nonauditory groups. However, both showed a moderate N50 response in the auditory group that was less than that of the active TMS condition. We performed the Kruskal–Wallis test on the extracted N50 responses to test for a significant main effect of stimulation condition (active TMS, click only, and click plus electric stimulation). For all three groups, we observed significance (total: χ2 = 114.68, p = 1.25 × 10−25; nonauditory: χ2 = 89.15, p = 4.38 × 10−20; auditory: χ2 = 30.56, p = 2.31 × 10−7) for at least one stimulation condition. Follow-up multiple-comparisons tests indicated that active TMS evokes a significantly higher N50 response than both control conditions across all groups. Comparison of total response showed significance for the active TMS versus click (p = 9.56 × 10−9) and active TMS versus click with electric stimulation (p = 9.56 × 10−9). Similarly, response in the auditory cluster showed significance for the active TMS versus click (p = 5.29 × 10−5) and active TMS versus click with electric stimulation (p = 7.26 × 10−7). The same pairs of conditions were significant in the nonauditory cluster: active TMS versus click (p = 9.74 × 10−10) and active TMS versus click with electric stimulation (p = 9.56 × 10−10). This result confirms that the sham protocol used to mimic auditory click and scalp stimulation does not induce TMS effects in nonauditory regions. Furthermore, scalp stimulation did not induce any significant effect on total or nonauditory response, confirming that somatosensory coactivation did not affect the recorded brain regions. In the auditory regions, the auditory click results in N50 response for both control conditions, which is significantly less than that in the active TMS condition. This corroborates the observation made in the auditory-masking condition that effects on auditory regions include centrally induced TMS effects as well as peripherally induced effects because of auditory activation.
Discussion
We investigated how single-pulse TMS affects mesoscale level neuronal activity in nonhuman primates depending on TMS intensity and coil position. Further, we used several control conditions to dissociate the target effects in the brain from peripheral auditory and somatosensory side effects. We found that early evoked potentials, occurring ∼50 ms after TMS pulse, characterize a centrally induced neural response to stimulation that could be a result of local suprathreshold activation in neuronal populations or connectivity to cortical brain regions more proximal to TMS. These early local field potentials demonstrate a characteristic sigmoidal dose–response dependency and inversely relate to the distance from the brain area to the stimulation coil. This centrally induced brain response is well dissociable from acoustic costimulation, particularly evident in the auditory cortex, and somatosensory costimulation, which does not induce specific early evoked components. These findings provide the basis for concurrent TMS-EEG studies in humans and put forward early evoked potentials as a clinically relevant target engagement biomarker.
TMS-evoked potentials and dose response
We found that TMS stimulation results in TEPs characterized by high amplitude and short latency. In both monkeys, the early activity expressed itself as a positive peak at ∼50 ms. In Monkey W, we saw a later rebound component at the highest stimulation intensity. Comparable observation has previously been reported in TMS-EEG studies in humans in unconscious or anesthetized states (Massimini et al., 2005; Ferrarelli et al., 2010; Sarasso et al., 2015). During such states, the TEPs are localized to the stimulation region, which results in less complex brain activity. However, in attentional states, the long-range brain connectivity results in complex TEPs with multiple peaks. The early response we observed here was robust in several brain areas across multiple TMS coil locations. While frontomedial stimulation locations are only effective at high doses (e.g., 90% MSO), lateral stimulation locations showed a characteristic, sigmoidal dose response dependency with increasing TMS intensity. In particular, we found that the early LFP component (N50) captures a mesoscale brain response not reported in the nonhuman primate literature so far. Nevertheless, the effects of single-pulse TMS on the visual cortex of anesthetized cats have been studied with electrophysiological recordings and optical imaging (Moliadze et al., 2003; Kozyrev et al., 2014). While these studies provide important insights into complex TEP patterns, research in nonhuman primates enables straightforward translation to humans (Alekseichuk et al., 2019). Two groups have studied the effects of TMS on the microscale neuronal level in nonhuman primates. One group examined single-unit activity during stimulation in different types of neurons (Mueller et al., 2014). Example neurons showed bursts of activity at short latencies (<100 ms), especially at ∼50 ms. The other group showed that the majority of single units demonstrated increased short-latency (10–40 ms after TMS delivery) spiking activity in parietal neurons during task-related activity, while some neurons demonstrated an excitation–inhibition–excitation pattern (Romero et al., 2019). While the summation of single neural firing should result in the population-level response, our study directly shows the aggregate LFP activity in neural circuits and its potentiation ∼50 ms after TMS delivery. Hence, the N50 component captures direct physiological effect of TMS.
To ensure that the TEPs are not a consequence of currents induced by inductive loops in the recording setup, we ensured that the contacts of the electrodes and the electrical cables were free of loops. Furthermore, the high-input impedance of the amplifier results in negligible currents compared with the biological currents that give ride to the LFPs. An interesting question is whether the TEPs observed here are because of direct suprathreshold neural stimulation at the recording sites or are propagated from more superficial brain areas via neural circuits, or both. The TMS-induced electric fields have maximum strength at the neocortex; however, a substantial part of the electric field in the small NHP brain also reaches deeper areas. In addition, the absolute field values necessary to induce suprathreshold neural activation in the brain are unknown to date. Thus, we focus our analysis on the relative values of electric field magnitude. The cross sections of the electric field magnitudes at each electrode for each stimulation location are shown in Figure 7. Electric field modeling cannot directly distinguish between the above-mentioned possibilities, which remain an exciting avenue for future research.
We also observed that the lateral-most stimulation locations in both monkeys (location 4 in Monkey W, location 5 in Monkey H) showed activation in the distal contacts compared with the proximal ones. We speculate that this could be because of a couple of reasons: (1) the coil orientation used for these two locations were different from the rest because of the headpost placement, which could have resulted in activation of deeper, frontal regions; and (2) the lack of activation in the proximal contacts could be because of the placement of some of these contacts in the WM (or WM and GM interface; Tables 1, 2). We did not observe the same LFPs in these contacts, and they were excluded from the analysis. The brain of Monkey H presented a variant brain anatomy, particularly in the parietal and occipital neocortical regions. Given that we operated outside these areas and normal daily behavior demonstrated by this NHP, we consider the recorded TEPs in Monkey H as normal. The similarity of TEP components observed in both monkeys further validates our designation.
Dose selection is a vital ingredient in TMS application, whether in research or therapy. Rational dose selection enables achieving desired treatment outcomes (Herrmann and Ebmeier, 2006) and eliciting desired activity in the brain when probing brain networks. Coupling the stimulation dose with a biomarker reflects that neural activity centrally induced by TMS is a key prerequisite for developing optimized dose control. In human TMS studies, targeting the motor cortex, the effect of TMS stimulation intensity has mostly been studied using MEPs (Darling et al., 2006). Given that MEPs are an indirect readout of cortical reactivity in the form of a muscle response, the effects of TMS intensity can be quantified easily. However, in brain regions pertaining to cognitive processes, for example, prefrontal or parietal cortex, such dose–response relationships have not been established with easily accessible readouts. In such cases, TEPs can serve as important biomarkers that explain the effects of stimulation intensity and connect them with behavioral responses. In previous human work, the intensity dependency of TEPs in human PFC was demonstrated using TMS-EEG (Kähkönen et al., 2005). However, this study was limited to a single brain region and did not disentangle centrally induced TMS effects from multisensory coactivation.
Spatial dependency
We used electric field simulations to study the effect of coil position on the field distribution. The Sørensen-Dice similarity coefficient analysis indicated similarity in electric field distributions between stimulation locations that are on the hemisphere ipsilateral to recording electrodes. The smaller size of the monkey brain and the range of coil movement, which was ≤3 cm between adjacent stimulation locations, lead to overlapping field distributions. Control location 5 in Monkey H scarcely overlapped with distributions from other locations. However, in Monkey H, two medial stimulation locations had overlapping distributions with the control location. Furthermore, the electric field distribution when stimulation location 6 in the monkey head also showed electric fields reaching the medial electrode in the left hemisphere. This can explain the response we observed in LFPs when stimulating location 6. As the monkey brain is considerably smaller than the human brain, the percentage of the brain volume activated by TMS is higher in monkeys than in humans (Alekseichuk et al., 2019). Thus, the spatial specificity of TMS is likely reduced to what would be expected in humans.
Investigating the spatial dependency of TEP responses, we observed a relationship between the N50 amplitude and the distance from the TMS coil to the recording contact. An inverse linear model and an exponential decay model can explain the seen relationship. Yet, the decay model captures this effect better, especially in Monkey W. This observation agrees with the nature of the magnetic field, which is rapidly decaying in magnitude with increasing distance from its source.
The effect of coil-to-cortex distance on TMS effects on DLPFC (Nahas et al., 2001b) and motor cortex (McConnell et al., 2001; Stokes et al., 2007) has been studied previously across subjects. While Romero et al. (2019) showed that TMS elicits spiking activity with a very high spatial specificity, we observed more widespread neural effects on the brain. The key differences in this study and our study potentially led to different observations: (1) their stimulation paradigm is limited to one brain region and, hence, is limited in explaining effects arising from stimulating other regions; (2) they record from a single brain region and therefore cannot capture the induced effects that are propagated to connected regions; and (3) the difference of the scale of recording (i.e., single units vs neural circuits).
Multisensory coactivation
We incorporated several control conditions to disentangle centrally induced neural effects from peripheral sensory effects. We conclude that auditory masking only affects the response in auditory regions, confirming that a fraction of the response because of TMS is in fact auditory evoked in nature. Similarly, the difference of responses in active TMS and passive TMS (auditory click) conditions confirms the above observation. We did not observe a visible response in nonauditory regions during passive TMS. Electrical stimulation of the scalp (coupled with passive TMS) to control for somatosensory activation did not produce a significant response in the nonauditory regions. Since the depth electrodes did not have contacts in the somatosensory cortex, we could not determine whether a response localized to that region was present during electrical stimulation. The evoked potentials we observed during passive TMS and electrical stimulation did not elicit long-latency responses. In fact, the somatosensory evoked potentials in anesthetized monkey preparations have shown to be short latency in nature and lasting up to 25 ms (Allison et al., 1991). Therefore, it should not impact our findings in the responses appearing after 25 ms. Research into different anesthetics in humans shows the normal modulation of middle latency auditory-evoked potentials at 10–100 ms after stimulus delivery (Goto et al., 2001). These modulatory effects are also shown for somatosensory sensation (Sparks et al., 1975). Hence, our analysis is confined to the TEP components that show modulatory effects for the given conditions. Present results have implications in TMS-EEG studies. Interpreting TEP components and assessing their origins is challenging because of the limitations of EEG recordings (Conde et al., 2019). A recent study has attempted to disentangle peripheral activations from centrally induced TMS effects using similar controls (Rocchi et al., 2021). The study has compared the TEPs across conditions in the vicinity of a single stimulation site. Despite similar observations being made on the modulation of early evoked activity, direct recordings from auditory regions in our study gave us valuable insights about centrally and peripherally induced TMS effects on auditory and nonauditory regions.
Nonhuman primate–human translation
The main TEP component we observed in this study can be compared with some of the human TMS-EEG studies. Several human TMS-EEG studies have reported N100 response resulting from stimulation in the vicinity of M1 (Kičić et al., 2008; Bikmullina et al., 2009). The N100 component has also been observed in TEPs resulting from prefrontal cortex stimulation (Kähkönen et al., 2005; Rogasch et al., 2015). This component has been implicated as a neurophysiological marker of cortical inhibition through MEP measurements (Nikulin et al., 2003; Kičić et al., 2008). The disparity in latency could be attributed to the differences in neural dynamics, gross functional anatomy, and cellular physiology between humans and monkeys (Paller et al., 1992). However, there is also evidence for TMS-evoked responses in humans that are commonly characterized by early potentials such as cortical N40 and occur on the same timescale as we observed here (Casula et al., 2014; Kerwin et al., 2018). Importantly, the distinction between recording modalities also contributes to differences in latency. Furthermore, using anesthetized preparations of nonhuman primates also contributes to such differences. In human literature, several TMS-EEG studies reveal complex TEP waveforms during attentional states. These include peaks at 15, 30, 45, 60, and 100 ms during motor cortex stimulation (Ilmoniemi and Kičić, 2010; Rogasch and Fitzgerald, 2013) and peaks at 27, 40, 60, 100, and 180 ms following prefrontal stimulation (Kähkönen et al., 2004; Lioumis et al., 2009). However, in sleep or unconscious states, the TEP components are less complex, more localized, and higher in amplitude (Massimini et al., 2005; Ferrarelli et al., 2010; Sarasso et al., 2015). We observe response patterns in two anesthetized monkeys equivalent to the sleep or unconscious findings in humans. Furthermore, the human data on subcortical ERPs is very scarce, and it is not clear whether the subcortical ERP we observed here is more or less complex than its human counterpart. While TMS-induced effects are state dependent, anesthetized models still provide important mechanistic understanding about neural effects induced by TMS. Previous studies have used anesthetized monkey models to explain the activity of visual cortex when presented with visual stimuli (Logothetis et al., 1999) and the effects of repetitive TMS using functional imaging (Hayashi et al., 2004; Ohnishi et al., 2004). Nevertheless, future work is needed to establish N50 as a biomarker in humans, and the present work lays significant advances toward this goal.
Conclusion
Understanding the relationship between TMS dose and physiological responses is crucial for optimizing TMS parameters to achieve desired clinical outcomes, elicit network activity, and induce plasticity. Furthermore, understanding the origins of evoked components in noninvasive EEG and the correct interpretation of TEPs in the presence of multisensory coactivation improve their utility as biomarkers for target engagement. Our study provides important findings regarding the dose-dependent physiological effects of TMS, the role of stimulation location and the distance between the target brain region and the coil position, and the impact of multisensory coactivation on TEPs evoked potentials because of TMS. Together, present results provide a mechanistic basis for human TMS-EEG studies and TMS biomarkers for clinical therapy.
Footnotes
The research presented here was supported by the National Institutes of Health Grants R01-NS-109498, RF1-MH-117428, and K99-MH-128454, and the University of Minnesota MnDRIVE Initiative.
The authors declare no competing financial interests.
- Correspondence should be addressed to Alexander Opitz at aopitz{at}umn.edu