Hippocampal rhythm modulation via frequency-tuned trigeminal nerve stimulation

Abstract

Trigeminal nerve stimulation (TNS) is a non-invasive method for modulating brain activity through peripheral afferent pathways. Although pulse-based stimulation has been commonly used, the impact of sinusoidal alternating current stimulation of the trigeminal nerve (TN-ACS) on hippocampal network activity remains unclear. This study examined the effects of TN-ACS delivered at 8 Hz (theta) and 40 Hz (gamma) on neuronal activity in the rat hippocampus. We applied 3 min of TN-ACS at different frequencies in rats and recorded neural activity from the hippocampus using multi-channel electrophysiology. We analyzed spike-phase entrainment in CA1, spike rate, power spectral changes in local field potentials (LFP, focusing on off-frequency stimulation effects), and spike-field coherence between CA3 LFPs and CA1 spikes. Our results showed that TN-ACS modulated hippocampal dynamics in a frequency-specific manner. 40 Hz stimulation enhanced spike-phase locking in CA1, while 8 Hz did not. At the single neuron level, theta-band coherence between CA3 LFPs and CA1 spikes showed a wide range of effects. At the group level, theta-band coherence decreased during 8 Hz stimulation and increased during 40 Hz stimulation. Spike rates remained stable across all conditions. To our knowledge, this is the first study to demonstrate that TN-ACS can entrain hippocampal spike timing and modulate interregional coherence without affecting overall firing rates. These novel findings indicate that TN-ACS can selectively shape deep brain rhythms, highlighting 40 Hz TN-ACS as a promising intervention for targeting memory-related processes.

Introduction

Trigeminal nerve stimulation (TNS) is a non-invasive neuromodulation technique that delivers electrical current through patch electrodes placed on the forehead, cheek, or jaw to target specific branches of the trigeminal nerve [1]. Initially developed as an alternative therapy for migraine [2, 3] and epilepsy [4], TNS has gained increasing attention over the past decade for its potential in treating neuropsychiatric disorders, including depression, attention-deficit/hyperactivity disorder [5], post-traumatic stress disorder [6], and disorders of consciousness [7, 8]. Despite promising clinical outcomes, the mechanisms by which TNS exerts effects on higher-order cognitive networks remain poorly understood. Current well-accepted hypotheses suggest that TNS engages ascending brainstem pathways via afferent projections from the trigeminal nerve to the nucleus tractus solitarius and trigeminal sensory nuclei [9, 10]. TNS is thought to modulate several brainstem relay stations, such as the locus coeruleus (LC), raphe nuclei, and the ascending reticular activating system (ARAS), which play key roles in neurotransmitter release, particularly norepinephrine, dopamine, and serotonin [11–13]. We previously demonstrated in vivo that trigeminal nerve direct current stimulation (TN-DCS) in rodents causes a sustained increase in hippocampal spike activity [14]. This bottom-up modulatory pathway appears to rely on the activation of trigeminal nuclei [15] in the brainstem and appears to be primarily mediated by the LC and raphe nuclei [16], since hippocampal responses to TNS were abolished when LC and raphe nuclei activity was pharmacologically suppressed [14, 16, 17].

While DC stimulation is a promising approach, alternating current stimulation (ACS) introduces frequency-specific effects that could offer more precise targeting of intrinsic neural oscillations [18]. 40 Hz tACS has been widely used in memory and Alzheimer’s research [19, 20], where it has improved memory performance. One proposed mechanism is that external 40 Hz stimulation can entrain neural activity and synchronize intrinsic oscillations by enhancing phase alignment within gamma oscillations in brain networks involved in memory encoding and retrieval [21]. Similarly, rhythmic stimulation at 40 Hz using visual flickers [22, 23] or auditory tones [24] has also been shown to modulate endogenous brain rhythms such as gamma and theta oscillations within the hippocampus. These rhythms are closely linked to memory [25–27], each contributing in distinct but complementary ways to the encoding, storage, and retrieval of information during learning and memory tasks [28, 29]. Nevertheless, the effects of theta-frequency tACS on hippocampal activity and memory remain debated. Some studies [30, 31] have reported improvements in working memory and enhanced connectivity between the hippocampus and cortex, while others have found impairments in episodic memory or disruptions in working memory performance [32–34].

The CA3–CA1 circuit is a core component of the hippocampal trisynaptic pathway and plays a central role in memory encoding and retrieval [25]. Communication between these regions is strongly modulated by theta oscillations, providing a temporal framework for coordinating synaptic input and spike output across subregions [25, 26]. In particular, theta-phase coherence between CA3 local field potentials (LFPs) and CA1 neuronal spiking could serve as a useful metric for assessing communication from CA3 to CA1 during spatial and episodic memory processing [28, 35]. Recently, Paulson et al. [23] reported that daily intervention with 40 Hz audio-visual stimulation could facilitate coherence between hippocampal CA3 and CA1 during navigation, increasing prospective coding of future positions in rats with Alzheimer’s disease. Increased CA3-CA1 coherence reflects enhanced functional connectivity and temporal alignment [29], allowing CA1 neurons to respond more reliably to upstream signals from CA3.

The present study investigated the effects of TN-ACS at frequencies of 8 Hz (theta) and 40 Hz (gamma) on hippocampal oscillations. It was hypothesized that TN-ACS could influence hippocampal neural activity and CA3-CA1 coherence in a frequency-dependent manner. By stimulating the trigeminal nerve, the research aimed to understand how external peripheral ACS can impact the intrinsic oscillatory properties of hippocampal neurons and circuits. This study aims to characterize frequency-specific neuronal modulation within hippocampal rhythms associated with memory as a mechanistic step toward understanding how peripheral nerve stimulation may engage cognition-related circuits, without directly examining behavioral consequences.

Methods

Animals and surgical procedures

This study used 5 adult male Sprague-Dawley rats (weighing 250–400 g) obtained from Charles River Laboratories. All procedures were carried out in accordance with protocols approved by the KU Leuven Animal Ethics Committee (protocol P072/2020). Anesthesia was induced via intraperitoneal injection of urethane (10 mg/mL; Sigma-Aldrich, USA). Once anesthetized, each animal was positioned in a stereotaxic apparatus (Narishige SR-6) for surgery. After making a midline incision to expose the skull, a small craniotomy was performed over the right dorsal hippocampus. The burr hole was made at coordinates −3.45 mm anterior–posterior and 2.4 mm medial–lateral relative to Bregma.

Electrical stimulation setup

TN-ACS was delivered using a metal electrode (0.8 cm×0.15 cm) coated with conductive gel (Signa Gel, Parker Labs, New Jersey). This electrode was positioned on the lower jaw to target the marginal mandibular branch of the trigeminal nerve. A return electrode of the same size, also coated with gel, was attached to the tail. The jaw electrode was connected to the positive terminal and the tail electrode to the negative terminal of an isolated current source (AM 2200 analog stimulus isolator, A-M Systems, Sequim, WA), enabling delivery of sinusoidal alternating current. Stimulation waveforms were controlled by an analog voltage signal generated via a data acquisition system (100 kHz, NI USB-6216, National Instruments, Austin, TX) using custom MATLAB scripts. Each experimental condition consisted of a 9 min recording block, including 3 min of baseline (pre-stimulation), 3 min of stimulation, and 3 min post-stimulation (referred to as the pre, during, and post time epochs). TN-ACS was applied at either 40 Hz or 8 Hz with current amplitudes of 0.25, 0.5, 1.0 mA. To control potential time-on-task effects between different tests, additional 3 min recordings were performed every 15 min after each stimulation block until multi-unit spike activity returned to pre-stimulation levels.

Electrophysiological recording setup

Neural signals were recorded using a single-column, 32-channel silicon probe with a vertical span of 1550 μm (E32 + R-50-S1M-L20 NT, Atlas Neuro, Leuven, Belgium). Signals were amplified (×192), bandpass filtered between 0.1 Hz and 7.9 kHz, and digitized at 30 kHz with 16-bit resolution using an RHD 32 Intan headstage (Intan Technologies, Los Angeles, CA) connected to an Open Ephys acquisition system (www.openephys.org). The probe was inserted into the hippocampus via the craniotomy, reaching depths from approximately 1950 ± 150 μm (uppermost contact) to 3500 ± 150 μm (deepest contact) relative to the dura. Electrode sites were assigned to putative hippocampal subregions (CA1, dentate gyrus, CA3) based on their depth, single-unit firing characteristics, and anatomical reference to the Paxinos and Watson rat brain atlas [36]. In an earlier study we also cross validated this with histology [14].

Neural data processing and analysis

Extracting single units

Based on the raw data recorded from the multi-channel probe, spike sorting was carried out using SpykingCircus [37], followed by manual curation in Phy (https://github.com/cortex-lab/phy) to ensure the selection of well-isolated single units. From these curated units, we extracted spike times and computed spike rates separately for the pre-, during-, and post-stimulation time epochs.

Spike-train thinning

To standardize data resolution, both spike-train and LFP recordings were down-sampled to 1000 Hz. Since metrics such as LFP-spike coherence are sensitive to firing rate, we applied a spike-train thinning procedure to minimize rate-related bias [38]. Specifically, for each unit, we randomly removed spikes from the higher-rate conditions until the spike rates across the pre-, during-, and post-stimulation periods matched the lowest observed rate among them.

Power spectrum density analysis

To evaluate changes in oscillatory power across stimulation conditions, we computed the power spectral density (PSD) [39] of LFP signals using custom MATLAB scripts referred to open-source processing principles (http://www.fieldtriptoolbox.org). LFP recordings were first low-pass filtered at 300 Hz (2nd-order Butterworth) and down-sampled from 30–1 kHz to reduce computation load. PSD was estimated via short-time Fourier transform using 1 s Hanning windows with 50% overlap and an FFT length of 1000 samples.

Stimulation onset and offset were identified from digital trigger channels, allowing precise segmentation into pre-, during-, and post-stimulation time epochs. Each 3 min phase was further subdivided into three equal 1 min segments, resulting in nine total epochs per session. Frequency bands overlapping with known stimulation artifacts (±2 Hz around 8 or 40 Hz) were detected and excluded from analysis. PSD was calculated for each segment, and average band power was extracted for canonical frequency ranges: delta (1–4 Hz), theta (4–6 Hz), beta (18–22 Hz; 26–30 Hz), slow gamma (30–60 Hz, excluding harmonic contamination). Band power was averaged across time and normalized to µV²/Hz. For each frequency band, we computed mean power and 95% confidence intervals across segments and the time epochs.

Spike-field coherence analysis

For the spike-field coherence (SFC) analysis we calculated coherence in the theta (4–6 Hz) and slow gamma (42–46, 50–54, 58–60) bands between LFP from putative CA3 and single-unit spike from putative CA1, for each stimulation time epoch using the following formula,

$$\mathit{SFC}\left(f\right)=|\frac{S_{\mathit{xy}}\left(f\right)}{\sqrt{S_{\mathit{xx}}\left(f\right)S_{\mathit{yy}}\left(f\right)}}|$$

where S_xx was the power spectral density of the LFP on the electrode in question, S_yy was the power spectral density of the spike-train, and S_xy was the cross-power spectral density between the spike-train and the LFP.

Spike entrainment

To visualize how spike timing was modulated by the phase of the sine-wave stimulation, we analyzed neural entrainment across different stimulation time epochs. For each epoch, spike times were aligned to stimulus triggers and separated into three blocks: 3 min pre-stimulation (baseline), 3 min stimulation, and 3 min post-stimulation time epoch. Within each block, we calculated cycle histograms using 30 bins per cycle to visualize spike timing relative to the stimulus phase at either 40 or 8 Hz. Spikes were assigned to bins based on the phase of the sine wave at which they occurred, and the histograms were normalized to reflect spike rate per phase bin. To compute similar cycle histograms, we used the same stimulus frequency as a reference, even though there was no stimulation during the pre-stimulation and post-stimulation time epochs. This approach enabled us to effectively model how spikes would align to a hypothetical sine wave of the same frequency without active stimulation. To quantify spike-phase alignment, we calculated the Phase-Locking Value (PLV), defined as:

$$\mathrm{PLV}=|\sum _b{R}_b{e}^{{i\theta }_b}|$$

where θ_b is the center phase of bin b$i$, and R_b is the probability of a spike occurring in that bin. A PLV of 1 indicates perfect phase locking (all spikes occur within a single-phase bin), while a PLV of 0 reflects uniform spike distribution across all phases. This metric provides a robust quantification of neural entrainment, defined here as the synchronization of spike timing to the phase of an external oscillatory input [40]. To evaluate frequency-dependent effects, we computed cycle histograms and PLVs for all single units under each time epoch and used a bar plot for visualization.

Inter-trial phase coherence

To quantify trial-to-trial consistency of spike timing, we computed the inter‑trial phase coherence (ITPC) for each neuron, condition, and epoch. For this analysis we defined one trial as one time-window of data lasting 10 s, and progressively moved this time-window through the full three-minute dataset to create different trials for each condition (pre, during and post). Then using the same phase estimation procedure described above for the cycle histograms, for each trial we computed a single phase angle:

$$\mathrm{Phase}=\mathrm{angle}|\sum _b{R}_b{e}^{{i\theta }_b}|$$

ITPC was calculated as the magnitude of the mean complex vector across all trials for that condition and epoch:

$$\mathrm{ITPC}=|\left(\mathrm{mean}\right(\mathrm{Phase}\left)\right)|$$

Values range from 0 (random phase across trials) to 1 (perfect phase alignment). ITPC is insensitive to firing rate because it depends only on phase angles.

Rayleigh test–based quantification of neuronal entrainment

As an additional entrainment metric, we assessed per-neuron phase-locking significance using a Rayleigh test of circular non-uniformity. Spike phases were defined relative to the stimulation cycle by expressing spike times modulo the TN-ACS period and converting to radians (0–2π). For each neuron, Rayleigh tests were performed separately for pre-, during-, and post-stimulation epochs using the binned phase histograms, and resulting p-values were FDR-corrected across neurons within each epoch. Neurons were classified as significantly entrained when the Rayleigh test survived correction (q < 0.05). To summarize amplitude dependence, we modeled the fraction of entrained neurons during stimulation as a function of amplitude using a binomial generalized linear model (logit link).

Statistics

The effect of TN-ACS frequency (8 and 40 Hz), amplitude (0.25, 0.5 and 1 mA) and time epoch on: 1) neuronal entrainment (PLV); 2) CA3-LFP and CA1-spike coherence; and 3) spike rate; were analyzed using linear mixed effects models with neuron number and rat as random effects and frequency, amplitude and time epoch as fixed-effects, (MATLAB, fitlme.m). Model syntax using the Wilkinson notation was as such:

1. PLV ~ frequency + amplitude + time epoch + amplitude*time epoch + (1|Neuron) + (1|Rat)),

2. SFC~ frequency + amplitude + time epoch + amplitude*time epoch + (1|Neuron) + (1|Rat)),

3. Spike Rate~ frequency + amplitude + time epoch + amplitude*time epoch + (1|Neuron) + (1|Rat)).

To assess the effects of TN-ACS on LFP power across hippocampal regions, we included “subregion”, which contains both pyramidal cell layer and dendritic layer, to test whether stimulation affected LFP power differently across hippocampal subregions, which may reflect their distinct anatomical and functional roles. CA3 is known to generate stronger recurrent activity and intrinsic oscillations compared to CA1, and prior evidence has reported regional differences in baseline LFP power and responsiveness to modulation. Therefore, we adapted our linear mixed-effects model to include “subregion” (CA1 or CA3) as a fixed effect and excluded “amplitude.” This decision was based on two considerations: 1) Exploratory analyses using the full model indicated that stimulation amplitude had only modest effects on LFP power compared to other variables, and 1 mA consistently led to the most robust effect across animals and time epochs. 2) Including both amplitude and subregion in the same model introduced multicollinearity and complicated interpretation due to increased model complexity. The effect of TN-ACS frequency, subregion, and time epoch on PSD was assessed using a linear mixed model (MATLAB, fitlme.m). Model syntax using Wilkinson notation:

PSD~frequency+subregion+time epoch+time epoch*frequency+(1|ratNumber).

For the various LFP frequency ranges in PSD analysis, we used the same model. Because the study’s a priori target intensity was 1 mA, we omitted amplitude from the PSD LME to maintain parsimony; including amplitude increased multicollinearity (max VIF ≈ 9.8; GVIF^(1/(2·df)) ≈ 2.2–2.9 for amplitude terms) without changing inferences for band, epoch, or subregion (see supplementary table 12).

All linear mixed model fits were verified by checking residual normality. Full model details are in supplementary tables. For all linear mixed models, ANOVA was performed on the model output to test the significance of the fixed-effects and interactions. ANOVA F-statistics and p-values for fixed-effects and interactions are reported in Results. If a significant main effect was found, post-hoc testing was conducted using a simpler LME model based (with time epoch as fixed effect) linear contrasts test. For analysis of neuron entrainment, CA3-LFP and CA1-spike coherence and spike rate three comparisons (pre-during, during-post and pre-post) were performed. For analysis of PSD, because subregion was included in the model, six comparisons (CA1pre-dur, CA1dur-post, CA1pre-post, CA3pre-dur, CA3dur-post, CA3pre-post) were performed. All reported p-values are Bonferroni corrected for multiple comparisons. Alpha of all tests was set to 0.05 and effect sizes are reported as Cohen’s d.

Results

Frequency-dependent effects of TN-ACS on CA1 spike entrainment

We first examined how TN-ACS at 8 and 40 Hz affected spike timing in CA1 neurons. As shown in Fig. 1A, B, single-neuron examples demonstrated how spike timing across the sine wave cycle changed during stimulation with 8 Hz TN-ACS at 0.25 and 1.0 mA, respectively (0.5 mA example not shown). To provide a baseline for comparison, we applied the same sine cycle to the pre-time epoch, as described in Methods. We observed no change in PLV between the pre- and the during-time epochs in this example for 8 Hz TN-ACS (Fig. 1A), even when the amplitude was increased to 1 mA. Figure 1I, J show examples of single neurons recorded during 40 Hz TN-ACS at 0.25 and 1.0 mA, respectively. Here, we observed a clear increase in phase-specific spike timing during stimulation, compared to a more uniform distribution during the pre-stimulation time epoch for both 0.25 and 1 mA.

Fig. 1. Effects of 8 Hz and 40 Hz TN-ACS on spike-phase entrainment in hippocampal CA1 neurons.

A, B Representative single-neuron examples showing spike-phase distributions during pre-stimulation (blue) and during-stimulation (red) time epochs at 8 Hz TN-ACS, for amplitudes of 0.25 mA (A) and 1.0 mA (B). Top: cycle histograms (30 bins per cycle); Bottom: polar plots of spike count across the sine wave cycle. C–E Group-level phase-locking values (PLVs) for 8 Hz TN-ACS at 0.25 mA (n = 48), 0.5 mA (n = 28), and 1.0 mA (n = 44), showing no significant differences across pre-, during-, and post-stimulation time epochs. H-J Group-level PLVs for 40 Hz TN-ACS at 0.25 mA (n = 38), 0.5 mA (n = 31), and 1.0 mA (n = 28), showing significantly increased entrainment during stimulation compared to pre-stimulation. F, G Single-neuron examples for 40 Hz TN-ACS at 0.25 mA (F) and 1.0 mA (G), demonstrating enhanced phase-specific spike timing during stimulation. Bar plots show mean ± 95% CI. Asterisks denote significant linear-contrast comparisons after Bonferroni correction (*p < 0.05; *p < 0.01).

At the group level, ANOVA of the linear mixed effect model (see methods and supplementary table 1) showed that PLV was significantly affected by stimulation frequency (F(1, 644) = 3.892, p = 0.048) and time epoch interacted with amplitude (F(2, 644) = 5.03, p = 0.006), but not by amplitude (F(1, 644) = 0.456, p = 0.456). Group-level results for 8 Hz TN-ACS are shown in Fig. 1C–E. For the 8 Hz TN-ACS, post-hoc tests showed no significant differences between pre- and during-time epochs (0.25 mA, t(141) = −0.643, p = 1; 0.5 mA, t(81) = 1.200, p = 0.701; 1.0 mA, t(129) = 1.302, p = 0.586) nor between pre- and post-time epochs at any of the tested amplitudes (0.25 mA, t(141) = 0.044, p = 1; 0.5 mA, t(81) = 1.395, p = 0.5; 1.0 mA, t(129) = −1.156, p = 0.749). However, 40 Hz TN-ACS did increase spike entrainment in hippocampal CA1 neurons. As shown in Fig. 1H-J, the PLV during 40 Hz TN-ACS was higher than in pre-stimulation across different amplitudes (0.25 mA, t (111) = 3.252, p = 0.005; 0.5 mA, t (90) = 3.076, p = 0.008; 1.0 mA, t (81) = 2.438, p = 0.049). Also, the post-stimulation PLV was not significantly different from the pre-stimulation PLV for any of the tested amplitudes (0.25 mA, t(111) = −0.259, p = 1; 0.5 mA, t(90) = 0.120, p = 1; 1.0 mA, t(81) = −0.796, p = 1), indicating that entrainment effects disappeared when stimulation was stopped. We observed similar effects of TN-ACS on spike entrainment for CA3 neurons. Results are shown in the Supplementary (see Supplementary Fig. S1 and table 2).

Inter-trial phase coherence (ITPC) was quantified for each unit during the pre-stimulation (pre), stimulation (dur), and post-stimulation (post) epochs and plotted separately for TN-ACS frequency and amplitude conditions. The results from the ITPC analysis are shown in Fig. 2. Consistent with the PLV results, ITPC significantly increased during 40 Hz TN‑ACS at 1 mA (t (81) = 3.450, p = 0.00268), indicating enhanced trial‑to‑trial precision of spike timing. In contrast, 8 Hz TN‑ACS did not increase ITPC. No stimulation frequency produced post‑stimulation increases in ITPC, suggesting that entrainment effects were present only during ongoing stimulation (LME model and statistical details see supplementary table 3). These findings strengthen the interpretation that 40 Hz TN‑ACS entrains CA1 spike timing.

Fig. 2. Effects of 8 Hz and 40 Hz TN-ACS on inter-trial phase coherence (ITPC) in hippocampal CA1 neurons.

A-C Group-level ITPC values across pre-stimulation (pre), during-stimulation (dur), and post-stimulation (post) epochs for 8 Hz TN-ACS at 0.25 mA (A), 0.5 mA (B), and 1.0 mA (C). D–F Group-level ITPC values for 40 Hz TN-ACS at 0.25 mA (D), 0.5 mA (E), and 1.0 mA (F). Dots indicate individual neurons; bars show mean ± 95%CI. Statistical testing was performed using a linear mixed-effects model with time condition as a fixed effect and random intercepts for neuron and rat, followed by planned linear contrasts between epochs. Asterisks denote significant linear-contrast comparisons after Bonferroni correction (*p < 0.05; **p < 0.01).

Further, we implemented a per-neuron Rayleigh test of circular non-uniformity to quantify statistically significant phase locking (entrainment) during TN-ACS. Specifically, for each unit we tested whether the distribution of spike phases relative to the stimulation cycle deviated from uniformity (Fig. 3A, C, Rayleigh test), and we controlled for multiple comparisons across neurons within each epoch using FDR correction. This analysis confirms and complements the PLV magnitude results by providing a clear estimate of “how many neurons were entrained.” At 8 Hz, significant entrainment during stimulation was rare across amplitudes (0/48 at 0.25 mA; 0/28 at 0.5 mA; 2/44 at 1 mA), and there was no evidence for an amplitude-dependent increase in entrained fraction (binomial GLM p_amp = 1). In contrast, at 40 Hz, the fraction of entrained neurons increased with amplitude (5/38 at 0.25 mA; 12/31 at 0.5 mA; 16/28 at 1 mA), yielding a significant dose–response relationship (binomial GLM p_amp = 4.88×10⁻⁴). As expected when phases are defined with respect to the stimulation-cycle reference, entrainment was not detected outside the stimulation epoch (Pre/Post).

Fig. 3. Rayleigh-test quantification of single-unit entrainment across amplitude and stimulation frequency.

A For 8 Hz TN-ACS, bar plots summarize the fraction of neurons exhibiting significant phase locking (Rayleigh test; FDR-corrected across neurons within each epoch) during the pre-, during-, and post-stimulation epochs at 0.25, 0.5, and 1.0 mA (left to right). Numbers above bars indicate n significant / n valid neurons. B Dose–response analysis for 8 Hz showing the proportion of Rayleigh-significant neurons during stimulation as a function of amplitude; the line indicates a binomial generalized linear model (logit link), with p_ₐmp testing for a monotonic amplitude effect. C, D Same analyses for 40 Hz TN-ACS. Entrainment was rare across amplitudes at 8 Hz (p_ₐmp = 1), whereas at 40 Hz the fraction of entrained neurons increased with amplitude (p_ₐmp = 4.88 × 10⁻⁴).

TN-ACS caused effects on CA3-LFP_theta and CA1-spike coherence in a frequency-dependent manner

Given the established importance of theta and gamma oscillations in hippocampus-mediated learning and memory, and the well-characterized CA3-to-CA1 projections within the trisynaptic pathway, we used CA3 LFP–CA1 spike coherence as a cross-subregional phase-alignment metric. Specifically, CA3 LFP provided an oscillatory reference signal, and coherence quantified how consistently CA1 spikes aligned to that rhythm across time, without implying causal directionality (Fig. 4A).

Fig. 4. TN-ACS Caused Effects on the CA3-LFP_theta and CA1-Spike Coherence.

A Schematic diagram illustrates the measurement of spike-field coherence between CA3 local field potentials (LFP) and CA1 neuronal spiking. B Example coherence spectrum under 8 Hz TN-ACS at 1 mA, showing a decrease in low-frequency (0.5–10 Hz) coherence during stimulation. C Time-frequency (coherence) heatmap showing the suppression of coherence in the delta and theta bands during 8 Hz stimulation. D Group-level analysis of coherence changes (Δcoherence = Dur - Pre) at theta and slow gamma frequency bands under 1 mA 8 Hz TN-ACS. Theta-band coherence significantly decreased during stimulation compared to the baseline. E and F Individual coherence spectrum and time-frequency heatmap under 40 Hz TN-ACS at 1 mA, showing an opposite pattern to the one observed in 8 Hz TN-ACS at 1 mA. G Group-level coherence changes for 40 Hz TN-ACS: Theta-band coherence significantly increased during stimulation, with no changes observed in the slow gamma band. Data are presented as mean ± 95% CI. **p < 0.01, ***p < 0.001.

An example from the 1 mA 8 Hz TN-ACS data (Fig. 4B) showed a decrease in delta and theta band (0.5–10 Hz) coherence during stimulation, which returned to baseline levels after stimulation ended. This decrease was further visualized in the time-frequency (coherence) heatmap (Fig. 4C), where suppression of coherence was most pronounced in the theta and delta ranges during stimulation and returned in the post-stimulation time epoch. In contrast, 40 Hz TN-ACS at the same amplitude caused the opposite pattern. As shown in Fig. 4E, coherence in the low-frequency range increased during stimulation, with the heatmap (Fig. 4F) indicating a focal rise around 4–5 Hz (theta band).

What stands out across both frequency conditions is the variability at the single-unit level. As shown in Fig. 4D, G, some units exhibited increased coherence while others showed a decrease. Despite the variability in responses observed in individual neurons, a consistent group level effect was observed which depended on stimulation frequency: ANOVA of the linear mixed-effects model on theta band coherence confirmed significant main effects of stimulation frequency (F(1, 2909) = 145.34, p < 0.001), time epoch (F (2, 2909) = 42.07, p < 0.001), and stimulation amplitude (F (1, 2909) = 21.47, p < 0.001), as well as a significant interaction between time epoch and amplitude (F(1, 2909) = 55.16, p < 0.001). Linear contrast post-hoc testing showed that 8 Hz TN-ACS at 1 mA (n = 118, Fig. 4D) led to a significant reduction in theta-band CA3-LFP_theta and CA1-spike coherence during stimulation (t(1578.0) = −3.063, p = 0.007, Cohen’s d = −0.136) compared to pre-stimulation, while 40 Hz TN-ACS at 1 mA (n = 90, Fig. 4G) caused a significant increase in theta-band coherence during stimulation (t(1332.0) = 3.657, p < 0.001, Cohen’s d = 0.208) compared to pre-stimulation. Thus, a significant net decrease in theta-band coherence during 8 Hz TN-ACS and a significant net increase during 40 Hz TN-ACS were observed.

Additionally, we applied the same analysis to the gamma band coherence: ANOVA of linear mixed effect model showed that slow gamma band coherence value was not affected by time epoch (F (2, 2909) = 2.14, p = 0.118). Similarly, no significant interaction was detected between time epoch and amplitude (F (2, 2909) = 0.32, p = 0.726). This meant that neither 8 Hz nor 40 Hz TN-ACS affected slow gamma coherence (Fig. 4D, G). Together, these findings suggest that TN-ACS modulates theta-band spike-field coherence across hippocampal subregions in a frequency-dependent manner, with opposite effects on CA3-LFP_theta and CA1-spike coupling for 8 Hz and 40 Hz.

TN-ACS has minimal effect on hippocampal neuronal spike rate

The results above show that 8 and 40 Hz TN-ACS caused distinct effects on CA1 spike timing and CA3 LFP–CA1 spike coherence. To further understand whether these changes are accompanied by alterations in overall firing rate, we analyzed the effects of TN-ACS on CA1 spike rate. Figure 5 shows how 8 and 40 Hz TN-ACS affected the spontaneous spike activity of putative CA1 across different stimulation amplitudes. ANOVA of the linear mixed effect model (see methods and supplementary table 7) showed that the spike rate was not affected by time epoch (pre, during, post; F(2, 1502) = 1.46, p = 0.234), nor was there a significant interaction between time epoch and amplitude (F(2, 1502) = 2.05, p = 0.129). This means spike rate did not significantly vary across time epoch (pre, dur, post), nor did it exhibit a significant interaction between time epoch and amplitude. ANOVA of the linear mixed effect model showed that spike rate was significantly influenced by stimulation frequency (F (1, 1502) = 10.45, p = 0.001) and stimulation amplitude (F (1, 1502) = 5.98, p = 0.015). These effects may reflect frequency-specific modulation of background activity, or shifts in background excitability. However, because these changes did not interact with the stimulation time course—and because spike rates remained largely consistent before, during, and after stimulation—they are unlikely to account for the entrainment or coherence effects observed in earlier analyses. These results suggest that TN-ACS at both 8 and 40 Hz produced no effects on firing rates, indicating that the changes in spike timing observed above are unlikely to be driven by general changes in spontaneous excitability (e.g., spike rate).

Fig. 5. Effects of 8 Hz and 40 Hz TN-ACS on spike rates in hippocampal CA1 neurons.

A Spike rate changes in CA1 neurons during 8 Hz TN-ACS at different stimulation amplitudes, 0.25 mA, 0.5 mA, and 1.0 mA (from left to right). B Spike rate changes in CA1 neurons during 40 Hz TN-ACS at different stimulation amplitudes, 0.25 mA, 0.5 mA, and 1.0 mA (from left to right). Each bar represents the average spike rate (spikes/s) during pre-stimulation (pre), stimulation (dur), and post-stimulation (post) time epochs. Data are presented as mean ± 95% CI.

TN-ACS modulates frequency-specific oscillations in hippocampal CA1 and CA3 regions

Based on the effects of TN-ACS on spike entrainment, we expected to see the largest effects on the LFP at the stimulation frequencies (i.e., 8 or 40 Hz). However, TN-ACS causes a large artifact in the LFP exactly at the stimulation frequency. Since it was not possible to disentangle this artifact from the neural response, we expanded our PSD analysis to a broad range of clean, artifact-free bands to capture potential off-frequency effects of TN-ACS. This approach allowed us to assess whether TN-ACS induced broader spectral shifts in hippocampal activity, either as downstream consequences of entrainment or through neuromodulatory pathways that affect oscillations beyond stimulation frequency. Although it should be pointed out that a limitation of this approach was that we probably missed the largest effects on LFP at the stimulation frequency. For 8 Hz stimulation, artifacts appeared at 8, 16, 24 Hz, and higher harmonics (Supplementary Fig. S3A), while for 40 Hz stimulation, artifacts were observed at 40, 80, 120 Hz, and higher harmonics (Supplementary Fig. S3B). To avoid contamination from stimulation artifacts, we excluded frequency bands within ±2 Hz around the stimulation frequency and its harmonics. Thus, subsequent band-specific analyses for both 8 and 40 Hz stimulation focused on clean frequency ranges: Delta (2–4 Hz), Theta (4–6 Hz), Beta (18–22, 26–30 Hz), Slow Gamma (34–38, 42–46, 50–54, 58–60 Hz). Because the removal of artifact-contaminated bands fragmented the fast gamma range during 8 Hz TN-ACS, we did not analyze fast gamma activity.

Figure 6A shows the dynamic changes in LFP power within the theta band during 8 Hz TN-ACS at 1 mA, with time periods corresponding to 1–3 min (pre-stimulation), 4–6 min (during stimulation), and 7–9 min (post-stimulation). Further, we visualized the same data with the PSD values averaged across each 3 min block. Figure 6B shows the dynamic changes in LFP power and averaged PSD value at each 3 min block in LFP power within the slow gamma band during 8 Hz TN-ACS at 1 mA. Figure 6C and D show the same data for 40 Hz TN-ACS at 1 mA.

Fig. 6. Effects of 8 Hz and 40 Hz TN-ACS on LFP power (theta and slow gamma band) across hippocampal CA1 and CA3 regions.

A Dynamic changes in LFP power within theta band during 8 Hz TN-ACS at 1 mA (left panel), shaded areas indicate mean ± 95% CI across animals. Group-level comparisons of theta band power for 8 Hz TN-ACS (right panel). Time epochs correspond to pre-stimulation (1–3 min), during-stimulation (4–6 min), and post-stimulation (7–9 min). Power spectral density (PSD) values were averaged over each 3 min. B Dynamic changes in LFP power within slow gamma band during 8 Hz TN-ACS at 1 mA (left panel) and group-level comparisons of slow gamma band power for 8 Hz TN-ACS (right panel). Slow gamma power significantly decreased during stimulation and returned to baseline post-stimulation. C Dynamic changes in LFP power within theta band during 40 Hz TN-ACS at 1 mA (left panel) and group-level comparisons of theta power for 40 Hz TN-ACS (right panel). D Dynamic changes in LFP power within slow gamma band during 40 Hz TN-ACS at 1 mA (left panel) and group-level comparisons of slow gamma band power for 40 Hz TN-ACS (right panel). Model based linear contrast was used as post-hoc testing, *p < 0.05; **p < 0.01; and ***p < 0.001.

To analyze the effects of 8 and 40 Hz TN-ACS on the LFP within each of the different LFP frequency bands, we separately applied a linear mixed effects model to the data from each LFP band. The results from each of the 4 LFP bands are shown in the supplementary table 8. In summary, we found that: 1) PSD was significantly affected by stimulation frequency (8 and 40 Hz) for all LFP bands. 2) PSD was significantly affected by time epoch for all bands. 3) There was a significant interaction between time epoch and frequency for all bands except theta and beta. 4) PSD was significantly affected by hippocampal subregions (CA1 and CA3) for all bands except delta.

We then did linear contrasts post-hoc testing on the data from each LFP band (6 comparisons: CA1pre-dur, CA1dur-post, CA1pre-post, CA3pre-dur, CA3dur-post, CA3pre-post). Comparisons that were significant after Bonferroni correction (n = 6) are marked with an asterisk in Fig. 6. Full results are shown in the Supplementary Table 11. In summary, 8 Hz TN-ACS significantly decreased theta power in CA1 (Pre-Dur: t (132.0) =−3.124, p = 0.048, Cohen’s d = −0.591) but not in CA3 (Pre-Dur: t (102.0) =−1.318, p = 1, Cohen’s d = −0.253). It is note that LFP power of CA3 within slow gamma band was significantly decreased (Pre-Dur: t(102.0) = −3.787, p = 0.0015, Cohen’s d = −0.683) during 8 Hz TN-ACS, and returned to the baseline level after stimulation was turned off (Dur-Post: p <0.001；Pre-Post: p = 1). However, there are no significant changes within slow gamma band were observed in CA1. In contrast, 40 Hz TN-ACS did not significantly change theta band power in either CA1 (Pre-Dur: p = 0.210) or CA3 (Pre-Dur: p = 1). Different from 8 Hz TN-ACS, 40 Hz TN-ACS also did not affect slow gamma band power in CA1 (Pre-Dur: p = 1) or CA3 (Pre-Dur: p = 1).

To characterize the dynamic effects of these two stimulation frequencies across additional bands, we also visualized the result for delta band and beta band in Supplementary Fig. S5.

Discussion

In this study, we examined the effects of TN-ACS at two physiologically relevant frequencies—8 Hz (theta) and 40 Hz (gamma)—on hippocampal network dynamics in vivo. We showed for the first time that TN-ACS can entrain spike timing in hippocampal CA1 neurons in a frequency-specific manner, with 40 Hz stimulation significantly increasing phase-locking of spike activity to the stimulation waveform, while 8 Hz stimulation showed no such effect. Here, the PLV provided a measure of the temporal alignment between neuronal firing and the external oscillatory input [40]. These results extend prior work on tACS, where gamma-frequency cortical stimulation was shown to modulate perception and memory by aligning neuronal spiking to an external rhythm [21, 41]. The possible mechanism for the stronger entrainment at 40 Hz, even at low amplitudes, is that gamma-frequency stimulation more effectively recruits fast-spiking interneuron networks that drive precise spike timing in CA1 pyramidal neurons, whereas slower theta-frequency input may be less effective in directly modulating spike-phase alignment at the cellular level [42–44]. Unlike transcranial methods, where external oscillatory electric fields are generated within cortical to subcortical regions [45]. TN-ACS recruits afferent brainstem circuits, via a bottom-up pathway, likely involving the LC and raphe nuclei, key relay stations of arousal and synaptic gain for neurotransmitter release. Our group’s earlier research showed that TN-DCS caused an LC-dependent increase in hippocampal spiking, with the trigeminal nerve providing afferent input rather than volume conduction [14, 16]. The current results add to this by demonstrating that rhythmic TN-ACS can structure spike timing without increasing spike rates, supporting the idea that subcortical pathways can mediate frequency-selective entrainment of deep brain circuits [46]. This is supported by modeling and animal slice work showing that ACS can modulate firing probability in a rhythmic, oscillatory manner without altering average firing rates over longer time scales [47–49].

The spike-entrainment analyses provide a direct description of how an externally imposed rhythm (TN-ACS) shapes the timing of CA1 spiking relative to the stimulation cycle. To complement this, we investigated whether TN-ACS also alters cross-subregional phase alignment between CA3 population activity and CA1 spike timing by computing CA3 LFP–CA1 spike coherence. Importantly, we interpret this metric as a descriptive readout of spike timing relative to an oscillatory reference signal recorded in CA3, rather than as evidence for reverse anatomical communication, because spike–field coupling can reflect multiple physiological relationships and does not by itself establish causal directionality [50]. Within this framework, we found that TN-ACS modulates this coherence selectively in the theta band, with no significant effects in slow gamma ranges: 8 Hz TN-ACS reduced theta-band coherence, while 40 Hz TN-ACS increased it. However, we noted some large variability in the results (see Fig. 4D, G). This may reflect heterogeneity in cellular composition and spike–field coupling properties across pyramidal cells and interneurons, as well as differences in how individual units engage with ongoing network rhythms [51]. Theta-phase organization is frequently implicated in coordinating activity across hippocampal subregions, whereas gamma-band interactions can be more state- and circuit-dependent and may reflect distinct local and pathway-specific processes [52]. The suppression of theta band coherence at 8 Hz TN-ACS may reflect interference or competition with endogenous theta rhythms [32, 53, 54], while the increased coherence in the theta band observed during 40 Hz TN-ACS may reflect an indirect effect of gamma-frequency stimulation on slower network dynamics, potentially via cross-frequency interactions and changes in spike timing precision [55]. The increase in theta-band coherence during 40 Hz TN-ACS might initially seem paradoxical, given that spike entrainment occurred at a gamma frequency. However, we suggest that these effects are not mutually exclusive. Increased entrainment at 40 Hz may enhance the precision and timing regularity of neuronal firing in CA1, thereby facilitating more stable phase relationships with upstream theta-modulated input from CA3 [27, 43, 47]. In other words, by enhancing gamma oscillatory activity that is temporally nested within theta cycles in CA3, 40 Hz TN-ACS may indirectly strengthen the coherence in the theta oscillation in CA3 and CA1 spike timing [28]. This subregional interaction between entrainment and coherence underscores the possibility that TN-ACS may affect hippocampal processing not only by imposing an external rhythm but also by modulating how neurons synchronize with endogenous inputs across timescale, while acknowledging that mechanistic directionality cannot be inferred from spike–LFP coherence alone [29].

An additional point concerns the absence of a post‑stimulation “entrainment echo,” which has been reported in some neural entrainment studies [56, 57] but not others [58, 59]. Whether rhythmic stimulation produces persistent phase alignment after offset is highly state‑dependent and varies with stimulation pathway and circuit architecture. Many demonstrations of entrainment echoes involve awake cortical networks receiving strong or direct rhythmic input [56, 57, 59, 60], conditions that favor reverberatory activity after the external drive ends. In contrast, our experiments were performed under urethane anesthesia, where externally imposed timing can dissipate once the input is removed [61, 62]. Peripheral rhythmic stimulation such as TN‑ACS also reaches the hippocampus through multi‑synaptic brainstem pathways [14, 16, 17, 63], which further reduces the likelihood of self‑sustaining oscillatory continuation after stimulation offset. Consistent with this, our ITPC analysis confirms frequency‑specific entrainment at 40 Hz during stimulation but no residual alignment afterward. Thus, the lack of an entrainment echo in our data is compatible with the known physiological constraints of the anesthetized hippocampal network [62] and does not contradict the presence of robust online entrainment during stimulation [58].

While these findings provide important new insights, several limitations must be acknowledged. First, we did not directly record LFP or spiking activity from brainstem nuclei such as the locus coeruleus or raphe, so the involvement of specific bottom-up pathways in TN-ACS-induced modulation remains inferred from prior TN-DCS studies [14–16] rather than being empirically confirmed in this work. Importantly, we cannot rule out the additional possibility that TN-ACS may first activate the trigeminal-to-somatosensory pathway, followed by top-down projections from somatosensory cortex to the hippocampus [64]. Such a combined bottom-up and top-down route may also contribute to the observed effects and warrants further investigation. Second, only two stimulation frequencies were tested in this study; future studies using a broader frequency range would better characterize oscillatory properties and optimal stimulation parameters. Finally, although we observed robust physiological changes in hippocampal activity, we did not measure behavior, and thus the cognitive relevance of these effects remains unknown. Neural modulation of memory-related rhythms does not necessarily translate into improved memory performance. Future studies should therefore pair TN-ACS with hippocampal-dependent tasks and test whether the observed entrainment/coherence changes predict behavioral outcomes.

Conclusion

This study provides the first evidence that TN-ACS can entrain spike timing in hippocampal CA1 neurons and modulate cross-subregional coherence in a frequency-dependent manner. Our findings showed that 40 Hz TN-ACS enhanced phase-locking and increased theta-band coherence between CA3 LFPs and CA1 spikes, while 8 Hz stimulation had the opposite effect, reducing coherence. These effects occurred without major changes in overall firing rates, suggesting that TN-ACS acted primarily through the temporal coordination of neural activity. From a translational perspective, these findings highlight 40 Hz TN-ACS as a promising candidate for non-invasive modulation of memory-related hippocampal dynamics, in line with prior reports linking gamma-frequency stimulation to cognitive enhancement. In contrast, the disruptive effects of 8 Hz TN-ACS on coherence and oscillatory power suggest it may interfere with intrinsic theta rhythms critical for memory processing—consistent with mixed outcomes observed in clinical studies using theta-frequency tACS. Together, these results underscore the importance of stimulation frequency in determining TN-ACS outcomes and provide a foundation for developing frequency-specific protocols to modulate deep brain function through trigeminal nerve stimulation.

Author contributions

Liyi Chen: Writing – original draft, Visualization, Validation, Software, Methodology, Investigation, Data curation, Conceptualization. Alireza Majdi: Review & editing, Visualization, Methodology. Boateng Asamoah: Review & editing, Methodology. Myles Mc Laughlin: Writing – Review & editing, Visualization, Validation, Supervision, Project administration, Methodology, Funding acquisition, Data curation, Conceptualization.

Funding

This work was supported by FWO Funding G0B4520N and NIH Funding 1R01MH123508-01. Liyi Chen was funded by China Scholarship Council at CSC202006380043.

Data availability

Any datasets and codes that are a part of this study are available from the corresponding author upon reasonable request.

Competing interests

All authors confirm that there are no conflicts of interest that could have influenced the outcome of this work.

Supplementary information

The online version contains supplementary material available at 10.1038/s41398-026-04086-3.