Transcranial alternating current stimulation entrains single-neuron activity in the primate brain

Significance

Neurostimulation is a common therapy for a variety of neurological disorders, but the most effective stimulation approaches are often highly invasive, requiring electrodes to be implanted deep within brain structures like the hippocampus and basal ganglia. Here, we show that transcranial electrical stimulation (tES), a neuromodulatory technique that uses electrodes placed outside the scalp, can also affect patterns of neural activity in these areas. We find that tES can reliably control the timing, but not the rate, of spikes in individual neurons. Because changes in spike timing are thought to play a key role in many brain functions, the data shown here suggest that tES may be a valuable tool for clinical and fundamental studies of the human brain.

Abstract

Spike timing is thought to play a critical role in neural computation and communication. Methods for adjusting spike timing are therefore of great interest to researchers and clinicians alike. Transcranial electrical stimulation (tES) is a noninvasive technique that uses weak electric fields to manipulate brain activity. Early results have suggested that this technique can improve subjects’ behavioral performance on a wide range of tasks and ameliorate some clinical conditions. Nevertheless, considerable skepticism remains about its efficacy, especially because the electric fields reaching the brain during tES are small, whereas the likelihood of indirect effects is large. Our understanding of its effects in humans is largely based on extrapolations from simple model systems and indirect measures of neural activity. As a result, fundamental questions remain about whether and how tES can influence neuronal activity in the human brain. Here, we demonstrate that tES, as typically applied to humans, affects the firing patterns of individual neurons in alert nonhuman primates, which are the best available animal model for the human brain. Specifically, tES consistently influences the timing, but not the rate, of spiking activity within the targeted brain region. Such effects are frequency- and location-specific and can reach deep brain structures; control experiments show that they cannot be explained by sensory stimulation or other indirect influences. These data thus provide a strong mechanistic rationale for the use of tES in humans and will help guide the development of future tES applications.

When a neuron fires a volley of action potentials, information may be encoded in the precise timing of spikes as well as the overall firing rate. Moreover, spike timing may convey information that is not accessible from a rate code. In the hippocampus, the timing of spikes provides precise information about an individual’s position (1); similar mechanisms may control perceptual organization in visual cortex as well (2). At a larger scale, the fine temporal structure of neural signals is thought to provide a mechanism for flexibly routing information according to behavioral demands (3). As changes in the temporal structure of neural activity appear to underlie momentary lapses in performance (4) and serious pathological conditions (5), there has been considerable interest in understanding the factors that influence spike timing.

Recently, transcranial electrical stimulation (tES) has become a popular tool for clinicians and researchers seeking to noninvasively alter brain states. Electrodes are placed on a subject’s scalp, and a weak electric current is passed among them, generating an electric field that, in theory, passes through the skull and interacts with the electrical activity of the neurons beneath. Early results suggested that this simple technique could alter healthy subjects’ perceptual, cognitive, and emotional states and relieve a wide variety of disease symptoms in patients (6–8).

However, these results have been met with some skepticism. Subsequent meta-analyses have suggested that the behavioral effects of tES can be weak or highly variable (9–12). Furthermore, a clear neural mechanism by which tES can influence human behavior has yet to be identified. Previous studies have used in vitro (13, 14) or small animal preparations (15) that are difficult to translate to humans, as the experiments typically use highly invasive stimulation techniques and apply very strong electric fields. Extrapolation from these studies is further complicated by the fact that humans have thicker, bonier skulls and larger, gyrated brains, which attenuate and distort the fields produced during tES (16, 17). Taking these factors into consideration, Vöröslakos et al. (18) recently estimated that 75% of the current applied during tES never reaches the human brain as a result of shunting through the scalp; they concluded that the electric fields inside the skull are too feeble to affect neural activity. Based on these and similar studies, it has been suggested that the behavioral effects of tES are likely the results of nonspecific mechanisms, such as the placebo effect (19) or activation of peripheral nerves (20).

Ultimately, any intervention that seeks to modify brain function must affect single neuron activity. Furthermore, the nature of many clinical or research applications requires the ability to target specific brain structures. Both of these criteria can be tested most effectively in nonhuman primates, which, like humans, have thick skulls and large, gyrencephalic brains. To date, only a few studies have applied tES to nonhuman primates with noninvasive approaches that resemble those used in human experiments. These studies reported changes in secondary markers of neural activity, such as local field potentials, functional connectivity, and stimulus selectivity (21–23), rather than changes in the activity of individual neurons. Similarly, intracranial recordings from humans have generally lacked the ability to resolve single-neuron activity (24). Thus, it remains unclear whether and how tES affects neuronal activity in the primate brain.

To address these issues, we recorded the activity of single neurons from the brains of alert nonhuman primates while applying tES methods identical to those used in humans. Our results confirm that the resulting electric field actually reaching the brain is weak, and that it does not change the firing rates of single neurons. However, we find that tES reliably alters single-neuron spike timing, even in deep brain structures distant from the sites of stimulation on the scalp. These effects are not consistent with activation via nonspecific or indirect pathways, but suggest a direct effect on neurons near the targeted area. These data show that tES can reliably modulate neural activity and provide important constraints for the future design of research and therapeutic interventions.

Results

Our experiments focus on transcranial alternating current stimulation (tACS), a form of tES that uses alternating current to generate oscillating electric fields. Previous work has suggested that tACS might exert two broad classes of influences on neural activity: changes in firing rate (25, 26) and changes in spike timing (27, 28). To investigate these possible effects, we applied tACS through the scalps of two macaque monkeys while recording single-unit activity from the hippocampus, a popular target in rodent (15, 29) and slice studies (30), as well as a frequent target for invasive stimulation in humans (31–35). We also recorded from neurons in the nearby basal ganglia (BG), which is similarly of interest as a therapeutic target for brain stimulation (36, 37).

tACS Produces Weak Electric Fields in the Primate Brain.

We first sought to identify scalp locations for the tACS electrodes that delivered field strengths similar to those observed in human experiments. To this end, we built individualized models of each animal’s head from high-resolution MRIs and intraoperative records of implant locations [see Datta et al. (38) and SI Appendix, SI Materials and Methods], which were then solved to find electrode montages that generated appropriate electric field strengths around the recording sites. We considered montages containing as many as eight electrode sites, with a total current budget of ±2 mA (4 mA peak to peak) across the entire montage, as is typical in human tES experiments. The resulting field strengths were estimated from the model to be 0.29 V/m in monkey F and 0.26 V/m in monkey N.

Consistent with these predictions, we measured peak field strengths of 0.28 V/m in monkey F and 0.35 V/m in monkey N. The mean ± SE of the field strengths in the two monkeys were 0.23 ± 0.01 and 0.19 ± 0.02 V/m, respectively. These are comparable to, and, in fact, slightly lower than, values reported in human studies, which can reach as high as 0.8 V/m (16, 17), albeit at sites closer to the brain’s surface.

tACS Alters Spike Timing in Deep Brain Structures.

Next, we used these stimulation montages to investigate how tACS affects single-unit activity. The animals performed a simple visual fixation task while we recorded single-unit activity (Methods) and applied tACS. Two kinds of stimulation were applied in randomly interleaved 5-min blocks. During active tACS blocks, the stimulation consisted of a ±2-mA sine wave oscillating at 5, 10, 20, or 40 Hz. During sham blocks, we applied only the onset and offset of the active tACS waveform (Methods). This latter approach mimics the sensations evoked by tACS so well that human subjects are generally unable to distinguish between them (39–41), and thus serves as a useful control for nonspecific sensory effects of stimulation. Blocks were separated by a 5-min interstimulus interval, and 6–10 blocks were collected per session.

We recorded from a total of 197 cells: 136 from the BG and 61 from HC. The BG neurons came from monkey F (89 cells; 65%) and monkey N (47 cells; 35%), whereas all 61 hippocampal neurons were recorded from monkey N. These recording locations were confirmed by aligning postoperative CT scans, with the electrodes inserted, to the preoperative MRI scans (SI Appendix, Fig. S1).

Data from an example BG neuron are shown in Fig. 1A. The raster plot on the left shows the neuron’s spiking activity during 10 1-s segments taken from a block of 20-Hz active tACS (Fig. 1A, orange) and sham stimulation (Fig. 1A, blue). There was no obvious difference in this neuron’s firing rate between stimulation conditions in these example intervals, nor, as shown in column 2, across the entire experimental session (P = 0.37, mixed-effects model). At the same time, it is clear that the timing of the spikes became more regular during tACS, with the neuron preferentially firing at the beginning of each tACS cycle (Fig. 1A, black trace at top), when the phase was around zero degrees (Fig. 1A, gray lines).

Fig. 1.

Example neurons demonstrating the range of responses to tACS. Each row contains data from a single neuron during 20-Hz (A), 40-Hz (B), and 10-Hz tACS (C and D), along with the corresponding sham conditions. The schematic electrode indicates that the neurons in C and D were recorded simultaneously on adjacent channels. Neurons in A and B were recorded during separate sessions. Column 1 contains raster plots from 10 1-s segments of the data, showing the tACS waveform (black), spikes occurring during tACS (orange), and spikes occurring during sham trials (blue). Vertical gray lines denote each neuron’s preferred phase during tACS; these are omitted in D because the neuron had no preferred phase. Column 2 shows firing rates for each condition estimated by binning the cell’s responses into 30-s windows. Average firing rates across conditions were then compared via a mixed-effects model test, the results of which are shown in the upper right. Column 3: spike density plots show the relative probability of a spike at each phase of the tACS waveform (orange) or the corresponding LFP frequency band during sham trials (blue). To quantify the strength of the phase locking, PLVs are shown beneath each plot, as is the result of a randomization test comparing the PPC values between conditions. Asterisks indicate statistical significance (*P < 0.05 and **P < 0.01). Column 4: left traces show the average waveform during tACS (thick orange line) superimposed over a band depicting the mean ± SD of the waveform during sham stimulation (blue band). Traces on the right show the average waveform during sham stimulation (thick blue line) superimposed over a band indicating the mean ± SD of the waveform during tACS (orange band). SI Appendix, Fig. S1 shows recording sites, and SI Appendix, Fig. S2 shows additional waveforms.

To quantify this entrainment, we computed the phase-locking value (PLV), which provides a measure of the reliability of spike times relative to an oscillation, in this case the waveform of the tACS stimulation. A PLV of zero means that spike times are uniformly distributed across the tACS cycle, whereas a value of one means that spikes appear exclusively at a single phase. For the neuron shown in Fig. 1A, the PLVs were 0.003 during sham stimulation and 0.73 during tACS; a randomization test reveals that this increase was statistically significant (P = 0.004).

Fig. 1 B–D shows three other example neurons, also recorded from the BG. As in Fig. 1A, none of the neurons showed a significant change in firing rate between the sham and active stimulation conditions (P = 0.23, P = 0.75, and P = 0.72, respectively, via a mixed-effects model). However, the cells in Fig. 1 B and C exhibited modest entrainment to the tACS, whereas the neuron in Fig. 1D appears to be unaffected. Note that the neurons in Fig. 1 C and D were recorded simultaneously on neighboring channels 150 µm apart, suggesting that the effects of tACS can be heterogeneous, even very locally within the stimulated area.

To ensure that our recordings were stable throughout the active and sham stimulation blocks, we plotted spike waveforms separately for each condition. These are shown for the example neurons in the rightmost column of Fig. 1. The thick orange lines show the units’ average waveform during active tACS application, superimposed on error bars derived from the sham data (Fig. 1, Right, light blue band). The average sham waveform is also shown (Fig. 1, Right, thick blue line) superimposed against the error bars from the tACS condition (Fig. 1, Right, orange band). Spike waveforms appear very similar across conditions. To quantify this similarity for the population, we calculated the median correlation between neurons’ average waveforms during tACS and sham blocks. The overall correlation was very high (r_s = 0.993), indicating that spike waveforms were stable across conditions. As shown in SI Appendix, Fig. S2, the waveforms were also extremely stable across tACS phases, indicating that the entrainment is not a result of a loss of signal during part of the stimulus cycle.

Fig. 2 summarizes firing rate and entrainment effects across our data set for all stimulation frequencies. Firing rates were very similar in the sham and active stimulation conditions (Fig. 2A). At the single-cell level, only 14 of 197 cells (7%) showed individually significant changes in firing rate (mixed-effects models, P < 0.05). These were evenly distributed across the frequencies tested (n = 4, n = 4, n = 2, and n = 4 cells, respectively), and, even though decreases in firing rate were slightly more common (n = 9 cells) than increases (n = 5 cells), there was no statistical evidence of a preferred sign for these changes [χ(1)² = 0.58; P = 0.45]. As cells were individually assessed at the P = 0.05 level, we would expect to see between 4 and 17 modulated cells by chance alone, so we interpret these firing rate changes as type II errors rather than an effect of tACS. Across the population of neurons recorded in each condition (SI Appendix, Fig. S3), the median changes in firing rate were not significantly different at any frequency (P > 0.05, Wilcoxon signed-rank tests). Thus, there was no consistent evidence for an effect of tACS on firing rate for individual cells or for the population of cells.

Fig. 2.

Population distributions for firing rates and PLVs. (A) Distributions of average firing rate during 5-Hz, 10-Hz, 20-Hz, and 40-Hz tACS. Each point compares a single neuron’s firing rate during sham stimulation (x-axis) and active tACS (y-axis). Rates were distributed along the equality line, indicating that firing rates did not change across conditions. The percentage of cells exhibiting significant changes for active tACS vs. sham is indicated on the top (increases) or bottom (decreases) corners of each graph, respectively. Red text indicates that proportions are larger than would be expected by chance (i.e., outside the 95% binomial CI). (B) PLVs presented in the same style as A. Unlike the firing rates, PLV values tend to be above the diagonal line, indicating a strengthening of entrainment during active tACS. SI Appendix, Fig. S3 shows population effect sizes.

In contrast, we found considerable evidence for the entrainment of single-unit activity by tACS (Fig. 2B). At the single-cell level, we used a randomization test to determine whether PLV values significantly changed between the stimulation and sham conditions (Methods). Overall, 44%, or 86 of 197 neurons, showed individually significant PLV changes at the P = 0.05 level. Of these, the vast majority of cells (77 of 86, or 90%) were more strongly entrained during tACS [χ(1)² = 30; P < 0.001]. The distribution of modulated cells did not differ across the tested frequencies (Fisher’s exact test, P = 0.25).

To examine these effects at the population level, we computed the median change in PLV (i.e., tACS − sham) for each cell and each frequency of stimulation. Overall, changes in PLV were significantly larger than zero for all frequencies (5 Hz, Z = 3.7, P < 0.0004; 10 Hz, Z = 3.2, P < 0.003; 20 Hz, Z = 4.6, P < 0.00004; 40 Hz, Z = 3.0, P = 0.0052, right-tailed Wilcoxon signed-rank tests; SI Appendix, Fig. S3). However, there was a statistical effect of stimulation frequency on the change in PLV [Kruskal–Wallis test, χ²(3) = 13.7, P = 0.033], driven primarily by the relative weakness of 40-Hz tACS compared with the remaining frequencies, particularly 5 Hz (P = 0.06, post hoc, Dunn–Šidák corrected) and 20 Hz (P = 0.0092). The numerically largest effect of tACS was found for 20-Hz tACS in HC, in which the median PLV reached 0.11, but this effect was not significantly larger (P = 0.12, corrected Wilcoxon rank-sum test) than the same conditions in BG. Overall, the effects of tACS on firing rate and entrainment were similar between the two areas, as shown in Fig. 2 and detailed in SI Appendix, Table S1.

The tACS field is generated by a sinusoidal current and therefore reaches its maximum strength at 90°. If the external field were very strong, or the only input to the cell, neurons should be precisely phase-locked to this peak, as reported by Ozen et al. (15). By comparison, the relatively weak fields in our experiments did not completely override neurons’ intrinsic phase preferences, as the preferred phases across the population remained scattered across the stimulus cycle (σ²_circular = 0.89, P = 0.38). However, consistent with the expected biophysics, the phase preferences of the neurons that individually showed significant entrainment clustered around 90° (P = 0.03, Rayleigh’s V test with Θ₀ = 90°), where the field’s polarizing influence was strongest. This clustering did not occur for the subset of neurons with significant entrainment during sham stimulation (P = 0.90, Rayleigh’s V test with Θ₀ = 90°). Although neurons with phase preferences near the peak (90 ± 45°) were entrained more strongly (∆PLV = 0.052) than those with other phase preferences (∆PLV = 0.034), this difference was not statistically significant (P = 0.12, Z = 1.56, Wilcoxon rank-sum tests).

At an even finer temporal scale, we observed an effect of tACS on the prevalence of bursting activity in single neurons. Bursts are neuronal responses in which two or more spikes are elicited in quick succession, and they are thought to reflect intrinsic membrane currents (42) that can, in principle, be modulated by tACS (43). To test for potential effects of tACS on bursting, we calculated a bursting index by dividing the average number of spikes in the 3–5-ms bins of the spike autocorrelogram by the average number of spikes in the 100–200-ms bins. Across the entire data set, bursting decreased from a median of 0.35 [interquartile range (IQR), 0.11–1.06] during sham stimulation to 0.30 (IQR, 0.12–0.89) during tACS; this effect was at the threshold of statistical significance (P = 0.0529, Wilcoxon signed-rank test). However, the low burst index values indicate that many cells do not burst in either condition, so we restricted our analysis to the 71 cells that exhibited bursting (i.e., burst index > 1). For these cells, the median burst index was 2.99 (IQR, 1.65–4.49) during sham stimulation but significantly decreased during tACS to 2.57 (IQR, 1.35–4.02; P = 0.023, Wilcoxon signed-rank test).

Cells may have differing susceptibility to external electric fields, as a consequence of a number of biophysical factors. One such factor is cellular morphology, which has been shown in slice recordings to be predictive of spiking responses to electrical fields (26). Although we cannot directly recover morphology from extracellular recordings, it is often linked to cell type (44). Accordingly, we divided our data set into putative interneurons and putative pyramidal cells according to their action potential width, using a threshold of 200 µs (44). However, we found no significant difference in phase-locking during tACS between the neuron clusters (P = 0.77, Z = 0.29).This may be because the HC and BG contain neurons at all orientations in close proximity (45); their orientation relative to the electric field (13, 26) may mask any cell-type–specific effects. Furthermore, the GABAergic neurons in the deep brain are more morphologically similar to projection neurons than their cortical counterparts (46–48).

A second factor is overall excitability, which might be expected to determine a cell’s sensitivity to electrical fields. However, there was also no relationship between neuronal excitability (as measured by firing rate) and changes in PLV in response to tACS (r_s = 0.06, P = 0.32). At the same time, we did find a modest association between a neuron’s PLV during sham stimulation and the subsequent change in PLV during tACS (r_s = 0.18, P = 0.0094), which suggests that some cells have a propensity to become coupled to external oscillations (49).

The Effects of tACS Are Specific to the Targeted Location and Frequency.

Merely changing spiking activity is necessary but not sufficient for tACS to be useful, as many proposed applications rely on the ability to target specific brain regions, often in specific frequency bands. We therefore examined the specificity of the observed spike timing effect in space and in frequency.

We first tested the frequency specificity of the tACS effects by extending our phase-locking analysis to all frequencies between 1 and 100 Hz (in ±1-Hz bins). Fig. 3A shows the median PLVs for 5-, 10-, 20-, and 40-Hz stimulation, with the blue lines corresponding to sham stimulation and the orange lines to tACS. In each case, entrainment effects, indicated by the horizontal line segment in each panel, were found only at frequencies near those of the stimulation frequency (e.g., 10 Hz for 10-Hz tACS; P < 0.01, Wilcoxon signed-rank tests); no significant changes were observed at any other frequency.

Fig. 3.

(A) Neurons in the targeted area entrain only to the stimulation frequency. To test for possible cross-frequency effects, we calculated phase-locking spectra showing entrainment to frequencies between 1 and 100 Hz (1-Hz steps, ±1-Hz bins; SI Appendix) during sham trials (blue) and active tACS (orange). Solid lines and shaded error bars correspond to medians and IQRs (i.e., 25th and 75th percentiles). Stimulation was applied at 5 Hz (Top), 10 Hz, 20 Hz, and 40 Hz (Bottom) to the HC/BG target. In each condition, entrainment of HC/BG neurons only increased (horizontal black lines indicate differences significant at the P < 0.01 level) at or near the tACS frequency (black triangles). No other significant changes were observed. (B) Neurons outside the targeted area are not entrained at all. To test the spatial specificity of tACS, the same analysis was performed by using simultaneously recorded neurons from area TEO. No consistent entrainment was found for these cells in any condition or at any frequency. As TEO is a visual area, this result also corroborates the hypothesis that spike entrainment was caused by tACS rather than retinal stimulation. SI Appendix, Fig. S4 shows single-cell effects within TEO.

Next, we examined the spatial specificity of tACS effects. In these experiments, we applied the same stimulation montage as before while also recording from the left inferotemporal cortex (area TEO), a superficial visual area that is ∼20 mm away from the targeted areas. Fig. 3B shows that there was no consistent entrainment in TEO at any frequency during tACS. Data from individual TEO neurons are shown in SI Appendix, Fig. S4. Of the 21 TEO cells recorded, only three (14%) showed individually significant changes in entrainment as a result of tACS, which is significantly different (χ² = 6.2, P = 0.01) from the distribution observed in HC and BG. Furthermore, two of those three cells showed significantly decreased entrainment during 10-Hz tACS, which is again different from the data from the targeted regions, where 90% of cells showed increased entrainment during tACS. Unlike in the targeted regions, stimulation did not affect the PLV of TEO neurons (P = 0.877, Z = −0.15, Wilcoxon signed-rank test), and the median change in TEO neurons’ PLV was 0, significantly smaller than that recorded in the targeted regions (P = 0.0042, Z = −2.86). This demonstrates that neurons far from the targeted areas are not affected by tACS.

We further tested for spatial specificity by recording from the same brain regions (BG and HC) as before while changing the configuration of the stimulation montage. In these experiments, the electrode montage was mirrored across the midline, so that the strongest stimulation would no longer be expected to reach the brain regions near the recording electrodes. Fig. 4A compares the entrainment produced by the original optimized montages (Fig. 4A, Left; n = 18 neurons) and mirrored montages (Fig. 4A, Right; n = 20 neurons), along with the corresponding sham controls. As before, the optimal montage entrained neurons significantly more effectively (P = 0.026, Wilcoxon signed-rank test) than the sham control. However, entrainment was unchanged at all frequencies (P > 0.05) when tACS and sham stimulation were applied through the mirrored montage. Similar results were obtained with 20-Hz stimulation (optimal, n = 12; mirrored, n = 8): the optimal montage produced a significant increase in entrainment around 20 Hz (P = 0.005, Wilcoxon signed-rank test), but tACS and sham stimulation applied through the mirrored montages were again equally ineffective (P > 0.05) at all frequencies.

Fig. 4.

Single-neuron entrainment is specific to the stimulation montage. (A) Phase-locking spectra, as in Fig. 3, for a montage optimized to stimulate the recording site (Left; see also Inset) and a mirror-image version applied over the contralateral hemisphere (Right). Stimulation was applied at 10 Hz (Top) and 20 Hz (Bottom). Data from sham (blue) and active tACS (orange) are shown for each montage, with solid lines and shaded error bars indicating the medians and IQRs (i.e., 25th and 75th percentiles), respectively. As in Fig. 3, stimulating via the optimal montage (Fp1 and O1) leads to significantly increased entrainment, indicated by the horizontal black bars, which is specific to the tACS frequency (black arrow). When the mirrored montage (Fp2 and O2) was used instead, no significant changes in entrainment were observed. (B) tACS produced significantly larger entrainment effects when applied through the optimal montage. To determine whether the differences in entrainment seen in A are themselves significant, we calculated the change in phase locking (tACS minus sham) produced by the optimal montage (red) vs. the mirrored montage (green). The data for 10-Hz (Left) and 20-Hz (Right) tACS are plotted in the same format as A. This analysis confirms that tACS is spatially specific (see also Fig. 3) and acts as control against peripheral nerve stimulation (Results).

Fig. 4B compares the effectiveness of the two montages by subtracting the data from the stimulation and sham conditions. Here the red line indicates the change in PLV produced by the original montage (i.e., the difference between the orange and blue traces in Fig. 4, Left), whereas the green line indicates the changes produced by the mirrored montage. The original montages are significantly more effective (P = 0.01 and P = 0.004, Wilcoxon rank-sum tests) than the mirrored controls, demonstrating again the spatial specificity of tACS stimulation.

Entrainment Is Unlikely to Be Caused by Peripheral or Retinal Stimulation.

The data in Fig. 4 also serve as an important control for phosphenes, a common side effect of tACS. During tACS, some of the current applied to the head is shunted along the scalp and into the eyes, directly stimulating cells in the retina (50–52). This activity then propagates throughout posterior and temporal visual areas (53). Neurons in or downstream of these areas could, in principle, become entrained to this input rather than the electric field itself, and a similar technique has been shown to improve HC-dependent memory in humans (54, 55).

However, the mirrored montage experiments suggest that this pathway does not produce the entrainment effects shown in Figs. 1–3. The anterior–posterior positions of the stimulating electrodes determine the amount of shunting through the retina and therefore the strength of phosphene production (50). Montages matched in terms of anterior–posterior position should thus produce equivalent amounts of retinal stimulation. Furthermore, shunted current affects projections to both hemispheres via the temporal and nasal parts of the retina, even if applied unilaterally, as human observers report tACS-induced phosphenes in both hemifields (56). If our entrainment results were caused by visual input, we would therefore expect the optimized and mirrored montages to entrain neurons equally well. However, as Fig. 4 shows, we observed significantly less (P < 0.01)—and, in fact, zero—entrainment when the mirrored montages were applied. Additionally, a retinal influence would likely be strongest in visual areas (53), but recordings from TEO, a midlevel visual area, showed no entrainment when tACS was targeted at our sites in the deep brain (Fig. 3B and SI Appendix, Fig. S4).

Finally, perceptual data from humans receiving tACS under similar conditions also suggest that phosphenes and other peripheral effects are unlikely to produce the entrainment we see here. Phosphenes are a common side effect of stimulation at 10–20 Hz, but are absent or difficult to evoke at other frequencies (50, 56–58). However, we observed strong, statistically significant entrainment at 5 and 40 Hz, which is again inconsistent with a retinal origin.

Along with phosphenes, the onset of tACS can produce tactile sensations as a result of activation of somatosensory afferents in the scalp. These percepts allow human subjects to effectively discriminate between tACS and no-stimulation conditions (59). However, subjects rapidly adapt to them, and a properly designed sham stimulus can effectively blind human subjects to the stimulation condition without passing appreciable current into the brain (39–41). It is difficult to interrogate a monkey’s subjective experience, but because of the close anatomical and perceptual similarities between monkeys and humans, their perception of tACS is likely comparable. We thus compared our active stimulation against a sham rather than a no-stimulation condition (SI Appendix, SI Materials and Methods). Furthermore, we did not analyze the first 30 seconds of each block, when these onset effects are likely to be strongest.

Taken together, the design of our experiments, the spatial and frequency specificity of the responses, and the correspondence with known biophysical mechanisms all suggest that entrainment is unlikely to be the result of phosphenes, tactile sensations, or other indirect effects.

Discussion

There has been considerable uncertainty about the potential effects of tES on the primate brain. Published results suggest a range of attitudes, from cautious optimism that it can effectively modulate brain and behavior (12) to skepticism that enough current enters the brain to have any effect at all (18). These assessments, however, are based on extrapolations, often from less realistic model systems, or indirect measures of neural activity. Here, we have presented data from nonhuman primates, a highly realistic model of the human brain, demonstrating that tACS reliably entrains the spiking activity of single neurons. This entrainment is limited to the frequency of stimulation and the vicinity of the targeted brain region.

Data from transcranial stimulation experiments are often difficult to interpret because of the numerous possible artifacts and confounds. Human experiments have purported to show that tACS entrains EEG oscillations, but there has been some skepticism regarding these results. In an EEG experiment, the stimulation artifact (e.g., 10-Hz tACS) and putative neural effect (change in 10-Hz oscillations) overlap precisely. Focusing on spiking activity, which has very different spectral content from the tACS waveform, allowed us to avoid this contamination. We were also able to exclude a potential influence of retinal stimulation, which has clouded the interpretation of previous work (50), by comparing two stimulation montages that would be expected to produce similar effects on the retina. These produced very different results on brain activity (Fig. 4), suggesting that our effects are not a result of retinal activation. Other sources of noise or indirect effects (e.g., tactile percepts, changes in arousal) would also have relatively poor spatial specificity. However, we found that the influence of tACS was heterogeneous within a small volume of tissue (Fig. 1 C and D) and yet confined to the targeted region (Figs. 3B and 4).

In our previous work in nonhuman primates, we showed that tES using direct, rather than alternating, current alters synchrony between multiunit activity in one brain area and LFPs in another. However, we were unable to detect changes in single-unit spike rate or spike timing (21). Contemporary work by Kar et al. (23) also found that 10-Hz tACS altered neuronal adaptation to visual stimuli, but the authors were unable to measure single neuron responses during stimulation as a result of technical limitations. Although these results imply that tES can somehow affect the brain, they did not demonstrate that it ultimately changes spiking activity, which is widely regarded to be the primary means of coding and communication within the nervous system.

There has been some debate over the strength of the electric field reaching the brain and the field strengths necessary to produce neuronal effects. In our experiments, we modeled and measured field strengths of 0.2–0.3 V/m in the target area; these values are similar to—and, in some cases, slightly weaker than—those produced in human experiments (16, 17). The results from in vitro and small animal experiments at similar field strengths are mixed. In slices, Anastassiou et al. (60) found increased entrainment at 1 Hz, but not at 8 or 30 Hz, using a field of ∼0.75 V/m. Reato et al. (30), however, also found robust entrainment effects in slices exposed to 0.2-V/m fields, as did Chan and Nicholson (61) at 0.1–0.2 V/m. In a rodent model, Ozen et al. (15) argued that fields below 1 V/m had little effect, but some of their experiments with weaker fields also produced entrainment in as many as 33% of neurons (see figure 4 in ref. 15), similar to what we report here. These previous results can be reconciled with ours if one considers that we collected nearly an order of magnitude more spikes per neuron (∼15,000), which provided the statistical power needed to detect the sometimes-subtle entrainment effects shown in Fig. 2.

Our results also differ from those of Lafon et al. (24), who reported an unsuccessful attempt to entrain sleep spindles by applying low-frequency tACS in human subjects. However, the electric fields used in this study were quite weak (<<0.05 V/m) and, unlike single unit activity, sleep spindles are complex nonlocal phenomena generated by interactions between multiple brain regions (62). As such, stronger fields and novel, multifocal stimulation may be needed to entrain them. Consistent with this idea, Ketz et al. (63) recently described a closed-loop tACS approach during sleep that improves long-term memory consolidation in a target detection task by modulating endogenous slow-wave oscillations.

If the effects of tACS require 15,000 spikes to be detectable, are they too weak to influence behavior? Our data do not directly address this point, but the entrainment changes reported here, particularly during 20-Hz stimulation, are comparable to those linked with behavioral changes. Specifically, increases of ∼0.1 PLV have been associated with changes in sensory representations and working memory (64, 65), perceptual organization (2), behavioral state (66), learning and reward expectancy (67), and the retrieval of memories (68). Changes in synchronous activity appear to underlie momentary lapses in performance (4) and serious pathological conditions (5), so even small changes in spike timing, when distributed across a network of neurons, may exert a powerful influence on behavior. Moreover, optogenetic tools have allowed researchers to causally manipulate spike timing in laboratory animals with observable effects on behavior (69). Our results suggest that tACS may provide a way to noninvasively manipulate rhythmic activity in humans as well.

We focused on the HC and BG because they are appealing targets for translational research. Deep brain stimulation of the BG is already a routine treatment for several conditions, and there is some evidence for a role of invasive stimulation of the HC in memory enhancement (31). From a technical standpoint, the HC may be a particularly challenging test bed for neurostimulation because it has strong endogenous oscillations that generate their own electric fields (70). However, HC and BG do have the advantage of bordering a large ventricle, which allows for stimulation by current passing directly through the brain as well as current shunted through the highly conductive cerebrospinal fluid (71). It may be more difficult to produce fields as strong as these in other deep structures, especially with conventional tES techniques. Nevertheless, fields as high as 0.8 V/m, nearly four times stronger than the fields used here, have been produced by tES targeting the cortical surface (16, 17), so it should be possible to entrain neurons in those areas at least as well. As shown in Fig. 4, the montage must be chosen carefully to do so.

Grossman et al. (29) recently demonstrated that a novel form of tES using temporally interfering waveforms could be used to stimulate deep structures in the mouse brain. Although this and other advanced techniques may produce stronger and more focal stimulation, our data demonstrate that similar effects can be achieved by using conventional two-electrode tACS. This is true even in the primate, in which the HC is ∼30 mm below the skull surface, rather than ∼3 mm as it is in the mouse. The human tES literature has largely focused on cortical targets, but our results suggest that stimulation of deep structures may be surprisingly easy to achieve in humans as well.

One limitation of this work is that we could record single-unit activity from only a small subset of possible brain areas. Because the current passes through many areas as it travels between the stimulating electrodes, it remains possible that stimulation actually affected some other, unobserved area, which subsequently entrained neurons near our recording sites. We think many of the controls detailed here earlier mitigate against this possibility: entrainment effects were not found in brain regions near the target site, for which the identical argument could be advanced. Moreover, we observed that tACS altered bursting in single neurons. Although bursting can arise from network dynamics (72) or very specific sensory input (73), it is typically thought to reflect local, cell-intrinsic properties (42, 74), especially when the overall firing rate does not change, as we report here. Because subthreshold depolarizations are known to inhibit bursting (42, 75, 76), these results are thus consistent with a local effect whereby tACS causes small, subthreshold changes in neurons’ membrane potentials. Ultimately, a full picture of the effects of tACS on brain activity will require simultaneous measurement of activity in many different brain regions with millisecond temporal resolution.

In conclusion, our data indicate that tACS entrains the activity of individual neurons in the primate brain. It does so in a spatially localized and frequency-specific manner that suggests that the electric fields produced during tACS directly affect neurons within the targeted area. Our data therefore provide a mechanistic rationale for applying tACS to human subjects and, more generally, suggest that other forms of tES may have direct neural effects as well.

Materials and Methods

Data for these experiments were collected from two adult male rhesus monkeys (Macaca mulatta) by using techniques very similar to those used in our previous study (21). All procedures described here were approved by the Animal Care Committee of the Montreal Neurological Institute and conformed to the guidelines of the Canadian Council on Animal Care. Full details of our task design, stimulation, neurophysiological recordings, and analysis are provided in the SI Appendix.

Note Added in Proof.

Following the submission of our manuscript, Asamoah et al. (77) published a report suggesting that tACS affects cortical neurons via transcutaneous stimulation of peripheral nerves. Our data—especially those presented in Fig. 4—and the data in ref. 21 are difficult to reconcile with that hypothesis. However, it will be important in future work to identify a potential contribution of peripheral nerves, using methods comparable to those used here.