Effects of tDCS on Spontaneous Spike Activity in a Healthy Ambulatory Rat Model

Abstract

Background:

The neurophysiological effects of transcranial direct current stimulation (tDCS) are typically described with respect to changes in cortical excitability, defined by using transcranial magnetic stimulation pulses to determine changes in motor evoked potentials. However, how individual cortical neurons change firing patterns under the influence of tDCS is largely unknown. While the relatively weak currents produced in the brain by tDCS may not be adequate to directly depolarize neuronal membranes, ongoing neuronal activity, combined with subthreshold changes in membrane polarization might be sufficient to alter the threshold for neural firing.

Objectives:

The purpose of this study was to determine the effects of tDCS on neurophysiological activity in motor cortex of freely moving, healthy rats.

Methods:

In nine healthy, ambulatory rats, each studied under six different stimulation conditions varying in current intensity (maximum current density = 39.8 A/m² at 0.4 mA) and polarity (anodal or cathodal), neural activity was analyzed in response to 20 minutes of tDCS applied through bone screws insulated from the overlying scalp.

Results:

After analysis of 480 multi-unit channels that satisfied a rigid set of neurophysiological criteria, we found no systematic effect of tDCS stimulation condition on firing rate or firing pattern. Restricting the analysis to the most responsive units, subtle, but statistically significant changes occurred only in the highest intensity anodal condition.

Conclusions:

These results confirm that at current densities typically used in human or animal tDCS studies, observed effects of tDCS are likely to occur via mechanisms other than direct neuronal depolarization.

Introduction

Transcranial direct current stimulation (tDCS) is a non-invasive, safe and inexpensive neuromodulatory technique that has seen widespread use in research and clinical practice for treatment of pain, Parkinson’s disease, epilepsy, psychiatric disorders and stroke [1]. Direct currents are thought to modulate neuronal activity in a polarity-dependent manner, either by increasing or by decreasing the excitability of neurons at the locus of stimulation [2]. However, its mechanisms of action are still largely unresolved and this limitation in our understanding could be a significant contributor to contradictory results described in the literature regarding its efficacy. Understanding detailed neurophysiological effects of tDCS, such as the alteration of neuronal depolarization in vivo, will improve its application and interpretation in clinical populations.

Neurophysiological effects of tDCS generally have been described with respect to cortical excitability. In human and animal studies, excitability is typically described with reference to the amplitude of motor evoked potentials (MEPs) using test pulses with transcranial magnetic stimulation (TMS) in humans [3] or tES in experimental animals [4]. However, using MEPs to infer changes in synaptic efficacy and membrane properties of specific neuronal elements is decidedly indirect. The amplitudes of MEPs are a function of several synaptic connections of descending pathways as well as the action of local horizontal connections on corticospinal neurons, thus representing a global measure of corticospinal excitability. The use of electroencephalographic data (delta and sub-delta frequencies) [5], functional magnetic resonance imaging (fMRI) [6] or cerebral blood flow [7] to infer changes in excitability are even less direct. Direct evidence comes from hippocampal slice preparations where long-term potentiation and long-term depression have been shown to play a role in alteration of synaptic efficacy with tDCS [8]. However, these results provide nonspecific evidence, since DC fields are applied directly to the fluid bath surrounding the slice preparation.

The purpose of this study was to ascertain the effects of tDCS on multi-unit firing rates and patterns, and local field potential (LFP) activity in freely moving, healthy rats. Classic studies in the mid-20^th century showed that polarizing currents applied directly to the cortical surface in anesthetized animals altered spontaneous unit discharge [9–11]. Changes in firing rates and timing have also been shown in hippocampal slice preparations [12]. While the relatively weak currents produced by tDCS applied to the scalp or skull may not be adequate to directly depolarize neuronal membranes, ongoing neuronal activity of ambulatory animals, combined with subthreshold changes in membrane polarization might be sufficient to alter the threshold for neural firing, perhaps resulting in subtle, time-dependent changes in firing of individual neural units. To test this, we analyzed neural activity in response to 20 minutes of anodal or cathodal tDCS with three different stimulation intensity levels in healthy, ambulatory rats.

Methods

Animals

Nine male, 12 week-old Long-Evans rats (Charles River Laboratories, Willmington, MA) were used in this study. Rats were housed in groups of two per cage prior to the surgical procedure, and individually housed after surgery, with free access to food and water. Animal procedures were approved by the University of Kansas Medical Center Institutional Animal Care and Use Committee (IACUC) and complied with the Guide for the Care and Use of Laboratory Animals. Effort was made to minimize the number and degree of discomfort of animals in this study. Of the nine enrolled rats, only rats for which recordings with stable activity were obtained in at least half of the intensity and polarity combinations were continued on the study. This resulted in exclusion of four of the nine rats (in total, four recording sessions were excluded based on this criterion). All subsequent data analyses were conducted on the 32 total recordings obtained from the remaining five rats.

Study design

Whereas the current intensity level used in humans is 0.5–2.5 mA [1], intensity was scaled to the size of the electrode to produce an equivalent current density [2]. Each rat received anodal and cathodal stimulation at three different current intensities (0, 0.2 and 0.4 mA) according to the safety limits recommended by Liebetanz et al [13], for a total of six conditions. The purpose of this design was to increase neuronal activity, either directly through anodal stimulation on the side ipsilateral to the microelectrode array, or through the reduction of inhibitory contralateral cortico-cortical connections by cathodal stimulation on the side contralateral to the electrode array. Electrodes were placed in such a way that differences in polarity could be observed at the location of the recording electrode. For each animal, the stimulation conditions were block randomized using a random number generator. Two stimulation sessions were performed for each of the six conditions per rat, utilizing a minimum wash-out period of 48 hours between sessions. To blind the study, the randomization, conduct of the stimulation sessions, and spike detection analyses were done by different individuals with only a randomly assigned number linking animal and stimulation condition. Once all neurophysiological analyses were complete, number assignment of animal and condition was unblinded.

Surgery

Rats were induced into an anesthetic plane through inhalation of isoflurane followed by a combination of intraperitoneal bolus injection of ketamine (90 mg/kg); and intramuscular injection of xylazine (5 mg/kg). Anesthetic depth was monitored in 15-minute intervals by the absence of both ocular and pinch reflexes. Maintenance boluses of ketamine (10–100 mg/kg/h) were injected intramuscularly or intraperitoneally, as determined by the presence of a positive pinch reflex, throughout the procedure. The head was shaved and subsequently mounted in a stereotaxic frame. Ocular ointment was applied to both eyes and a rectal temperature probe attached to a homeothermic blanket was inserted to monitor and maintain body temperature throughout the procedure. The incision site was cleaned with ethanol and iodine wipes, lidocaine cream was applied, and subcutaneous injections of bupivicaine (0.5 cc) were made along the midline of the scalp and upper neck prior to incision. A midline incision of the scalp was made and the skull was exposed. A laminectomy was performed by making an incision in the membrane of the cisterna magna to reduce brain swelling. For tDCS stimulation, two small burr holes were made bilaterally (stereotaxic coordinates relative to bregma, +1mm AP, +3 ML) corresponding to the forelimb area of primary motor cortex (M1) and an additional burr hole was made in the interparietal bone (+3mm AP, 0 ML from lambda) for the electrode return path. Care was taken to not traverse completely through the skull. Three 00–90 stainless steel screws (part number 9178A410; McMaster-Carr, Elmhurst, IL) were implanted into the burr holes and connected to a miniature connector through silver wire tied to the screws. An additional three burr holes were made in the frontal, parietal, and interparietal bones and implanted with 00–80 stainless steel screws for anchoring the tDCS connector and the microelectrode array (MEA). A second screw in the interparietal bone was used as the animal ground for the neurophysiological recording system.

The surface area of the skull screw in contact with bone was estimated by importing the McMaster-Carr CAD drawings (part number 9178A410; screw diameter = 1.176 mm) and selecting the corresponding surfaces that were exposed to bone, using Fusion 360 (Autodesk, Inc., San Rafael, CA). Assuming a depth of screw penetration into the skull of 0.75 mm, the electrode surface area based was calculated to be approximately 4.421 mm². Since the maximum current intensity was 0.4 mA, the estimated maximum current density was 39.8 A/m², well below current densities known to cause damage in rat cortex [13].

To implant the MEA, a craniectomy spanning premotor cortex and the rostral aspect of M1 was made. Dura overlying the cortex was resected, and a 16-shank MEA (Tucker-Davis Technologies, Alachua, FL) was inserted to a depth of 1600μm from the cortical surface using an electronic micropositioner to target layer V cortical neurons. Kwik-cast polymer (WPI, Sarasota, FL) was used to embed the electrode shanks and protect the cortical surface. Dental acrylic was then applied to permanently mount the screws, tDCS connector, and MEA. Following the surgical procedure, the incision was sutured, treated with topical lidocaine cream and triple antibiotic and analgesics were administered over the next three days as the animal recovered from the procedure. Following completion of the study, all rats were humanely euthanized by an overdose of sodium pentobarbital (Beuthanasia-D, 390mg).

Neurophysiological recording sessions

Rats were lightly sedated using isofluorane and connected to a tethered headstage capable of amplifying and digitizing signals acquired from the probe implanted in cortex (Intan Technologies, Los Angeles, CA). A separate lead, capable of delivering constant-current stimulation, was secured to an adapter affixed to the caudal end of the skull. Recordings were initiated after recovery of the animal from anesthesia. Custom slip-ring commutators (Moflon MC573; Intan M4430 or C3430) were designed to minimize noise due to animal movements during the recording. Neural signals were sampled at 20 kHz (Intan RHD2000), or 30 kHz (Intan RHS2000), and stored for offline filtering and analysis.

Recording sessions were initiated once the animal had awoken from anesthesia and was moving normally around the testing chamber. Each recording session lasted at least an hour, with some recordings of up to two-hours. After an initial 20-minute period, stimulation or sham stimulation was initiated for an ensuing 20-minute period. Recordings in which the rat removed its headstage or otherwise disrupted the first 60-minute period were discarded. All subsequent analyses focus on epochs (Pre, Stim, and Post) that were isolated within the three discrete experimental phases, as outlined in Figure 1A. These epochs were chosen to give a consistent, direct comparison across the different conditions, while reducing movement artifact while exploring (Pre) and avoiding slight timing differences related to onset or offset of stimulation. Stimulation was delivered using a constant-current stimulus isolator (Bak Electronics BSI-2). Current was manually ramped to the final stimulation intensity over the course of 20 seconds. For anodal stimulation, the screw ipsilateral to the recording electrode was used as the tDCS electrode. For cathodal stimulation, the contralateral screw was used. In both conditions, the second pole was the occipital bone screw (Figure 1B). The onset of tDCS was synchronized to the neural data by coupling the stimulator output with an input on the acquisition hardware. Though tethered to the commutator, the animals were free to move about the recording chamber within a space of 30.5 cm × 30.5 cm. The RHS2000 system was used to measure the tDCS voltage offset using the additional low-gain DC-coupled amplifier on the headstage (Intan M4016) to verify timing and level of stimulation (Supplementary Figure S1).

Figure 1. Methods.

A: Recording timeline. Recording sessions consisted of 15-minutes of baseline pre-stimulation activity, a 20-minute tDCS-stimulation period and a 60-minute period of post-stimulation activity. B: Electrode orientation. For neural recordings, a 16-shank microwire array was embedded within the left premotor cortex and the rostral aspect of left M1. Current was delivered through either the screw ipsilateral to the recording electrode (anodal stimulation) or the contralateral screw (cathodal stimulation). In both conditions, the occipital bone screw was used as neutral pole (counter electrode).

Finite Element Analysis (FEA)

Finite element models were solved for all stimulation conditions. The magnitude of the current density and the projection of the current density onto a vector oriented radially (normal to the surface of the brain) were calculated within the cortex for each model. Figure 2 shows overall current density magnitude and the projection of the overall current density onto a radially oriented vector. The overall current density vector is derived from the radial current and an orthogonal horizontal current. The rat brain is lissencephalic, therefore the radial and horizontal components can be interpreted as the components of the electric field that would depolarize neurons projecting from the cortex to deeper structures, such as pyramidal cells in layer 5, and components that would depolarize horizontal corticocortical projections and interneurons respectively. The model output for anodal stimulation at an intensity of 0.2 mA within coronal cross sections was aligned to the microelectrode array, anode screw, and current return screw. For the 0.2 mA stimulation intensities, cortical currents with a current density magnitude of 19.9 A/m² and a radial current density of 17.5 A/m² oriented inward were observed below the anode screw and a current density magnitude of 1.85 A/m² at the location of the microelectrode array. At the higher 0.4mA stimulation level, current densities were increased with peak cortical current densities of 39.8 A/m² for both anodal and cathodal stimulation. Additional methodological details regarding the finite element modeling can be find in the Supplementary Materials.

Figure 2. Finite Element Model Results.

Finite element models were used to simulate the current distribution for associated with each stimulation condition. Here model results are shown for anodal stimulation with a stimulus intensity of 0.2 mA. The magnitude and radial component of the current density is shown in coronal slices at the locations indicated in the schematic on the left. a) At the location of the microelectrode array anterior to the anode screw, current density magnitudes with a maximum of 1.85 A/m² were observed. As the radial component of the current density is approximately 20% of the overall magnitude (0.36 A/m²), the current density at the microelectrode array was primarily horizontal. b) Below the anode, current density magnitudes of 17.0 A/m² were observed with a peak negative radial current density of −13.3 A/m² indicating current flowing radially inwards. c) Below the return screw, current density magnitudes of 17.6 A/m² were observed with a peak positive radial current density of 10.1 A/m² indicating current flowing radially outwards.

Neural signal analyses

Neurophysiological data were analyzed offline using Matlab (R2017a); analyses specific to this study are available in a public repository at https://github.com/m053m716/TDCS. Due to the sensitivity of the neurophysiological recordings and the relatively unconstrained ambulatory behavior during the experimental sessions, it was necessary to apply a secondary screening step to the data prior to subsequent endpoint extraction to exclude epochs in which the signal was excessively noisy. Such non-experimental noise sources are typically evident by visual inspection of the streaming data on the acquisition computer, and often correlate with such behaviors as the rat beginning to chew on the tether cables or bumping its headstage against the side of the recording chamber. Although behavior was not formally documented during the recording sessions, no events were noted that would suggest systematic effects of the stimulation protocol on overt motor behaviors. Therefore, in lieu of removing potential noise epochs by screening directly based on the animal’s behavior, we used a “mask” method in which the root-mean-square (RMS) power of the full-bandwidth neurophysiological signal (after application of a 60-Hz notch filter) was computed in non-overlapping windows that matched the timescale of any time-series analyses. Windows with an average RMS across all channels exceeding 150-μV were considered as periods of non-neurophysiological noise and were excluded from subsequent analyses. This secondary exclusion step caused the removal of two recording sessions due to the total exclusion of data during the Pre or Stim epochs, resulting in a final dataset of 30 total recordings derived from five rats. Additional details on the masking method and a table (Table S2) detailing the total number of included time-samples from each recording based on this exclusion process can be found in the Supplementary Materials.

We extracted two processed endpoints from the masked neural time-series: 1) the times of multi-unit spike peaks in the bandpass filtered LFP, and 2) the average frequency content of the decimated, low-pass filtered LFP. To detect multi-unit activity, the full bandwidth signal was bandpass filtered using the Matlab `designfilt` function to create an infinite impulse response (IIR) filter with passband lower and upper frequencies of 300-Hz and 3,000-Hz respectively. Spikes in the filtered signal were detected using a combination of thresholds designed to robustly screen for unit activity as described previously. [14] Briefly, candidate spike samples were identified if the smoothed nonlinear energy operator (SNEO) met or exceeded 4.5 times the median deviation of the SNEO time-series (Figure 3). The Matlab `findpeaks` algorithm was then used to identify peaks in the SNEO-valid reduced time-series, with the ‘MinPeakAmplitude’ parameter set to 4.5 times the median absolute deviation of the bandpass filtered time-series. [15, 16] Spike times of multi-unit activity were used for all subsequent spike analyses on a per-channel basis. Multi-unit peak times were used to compute the firing rate (FR) in 30-second bins, as well as the revised coefficient of local variation (LvR) [17] which was computed on a per-epoch basis.

Figure 3. Example signals and spike detection.

A: Example extracellular field potentials. Trials consisted of a 15-minute pre-stimulation baseline period, a 20-minute stimulation (or sham) period, and 60-minutes of post-stimulation recording. For data analyses, segments contained within each period were defined as Pre (10-minutes), Stim (15-minutes), and Post (10-minutes), respectively. B: Filtered recording data. Highlighted sub-section from A after unit filter and re-referencing. Multi-unit spikes (blue asterisks) are detected in the filtered signal if both the time-series amplitude (black) meets the voltage threshold (dashed-blue), and the corresponding smoothed nonlinear energy operator (SNEO; dotted-red) meets the corresponding threshold (dashed red).

To estimate the frequency content of the LFP, the full-bandwidth signal was down-sampled using the Matlab `decimatè function, which includes a Chebyshev anti-aliasing lowpass filter, so that regardless of whether the data was acquired at 20- or 30-kHz, all subsequent steps were applied to a 1-kHz signal. Because we focused on LFP frequency content as a proxy for spatial processes occurring on a slower timescale than multi-unit spiking, all frequency analyses used the average time-series signal from across all recording probes, which were embedded in a planar configuration so that they were located within the same level. To make use of the “mask” corresponding to the time-series RMS power, non-overlapping 30-second windows were used to compute the power spectrum of the LFP during non-noise samples using the Matlab `spectrogram` function, such that there were 20 logarithmically-spaced sampled frequencies within each of the following bands: Delta (2-Hz to 4-Hz), Theta (4-Hz to 8-Hz), Alpha (8-Hz to 12-Hz), Beta (12-Hz to 30-Hz), Low-Gamma (30-Hz to 50-Hz), and High-Gamma (70-Hz to 105-Hz). These power values were used for any subsequent LFP frequency-content statistics. [18]

Statistical Analyses

All endpoints were evaluated over epochs corresponding to the pre-stimulation baseline (Pre), during stimulation or sham-stimulation delivery (Stim), and after stimulation (Post). Statistical analyses were performed in Matlab (R2019b) or JMP (version 11.0, SAS Institute, Inc.). FR was always square root transformed prior to any model fit. To summarize spiking statistics, we computed the changes in LvR (ΔLvR) and changes in median FR (Δμ) using the channel matched Stim and Pre epoch values. These values were then fit using a one-way ANOVA with the fixed effect of treatment (Sham, 0.4-mA Anodal, or 0.4-mA Cathodal) and the random effect of animal to account for intrinsic differences specific to each rat. This reduced set of levels, which includes only the most-extreme stimulation conditions and groups together all rats who received 0.0-mA stimulation in one sham group, was selected to reduce the total number of comparisons in the model. For visualizing FR time-series (Figure 4), we computed percentage changes in FR (ΔFR_i,j,t) for the t-th sample observed on the i-th channel of recording j were using the channel-matched median μ_i,j value, for which each t was contained within the Pre epoch, according to equation (1).

$$\Delta F{R}_{i,j,t}=\frac{100*\left(F{R}_{i,j,t}-{\mu }_{i,j}\right)}{\text{max}\left(F{R}_{i,j,t},{\mu }_{i,j}\right)}$$ Eq. 1

The upper- and lower-5^th percentile of changes computed in such a way were then used to define thresholds that identified units with the most-substantial average changes in firing rate between the Stim and Pre epochs. Once labeled in this way, chi-squares tests of homogeneity were used to compare the proportion of labeled units in each category (Decrease, No Change, or Increase) against the expected proportion for each category (5%, 90%, and 5%, respectively). Because LvR values are computed as a single value for an epoch, as opposed to as a time-series, those values were used directly for visualization (Figure 5).

Figure 4. Multi-unit firing rate is not influenced by tDCS in awake, ambulatory rats.

We tested the hypothesis that anodal (left) or cathodal (right) tDCS influences the intrinsic excitability of cortical spiking units, which would probabilistically manifest as an average increase or decrease in firing rate (FR) during and possibly after the Stim epoch. Panels are arranged from top-to-bottom in order of increasing current magnitude. Each panel contains the mean (dark line) and ± 1 standard deviation (lighter shading) of the percent-difference in firing rates between the epoch/treatment condition of the given panel, and the temporally-matched combination for all 0.0-mA (sham) controls. Black brackets above the plot denote a statistically significant difference at the level of α = 0.001. Epochs are discontinuous to indicate the temporally excluded segments occurring around the stimulation “ramp” periods, during which time a current is applied, but not at a constant level.

Figure 5. Temporal patterns of multi-unit spiking may be influenced by tDCS in awake, ambulatory rats.

We tested the hypothesis that anodal (left) or cathodal (right) tDCS may exert a subtle effect on temporal variability of spiking activity, which could be measured by the revised coefficient of local variation (LvR), an index that measures the relative uniformity or burstiness of a given unit. Panels are arranged from top-to-bottom in order of increasing current magnitude. Each panel contains the smoothed distribution of LvR values for the Pre, Stim, and Post epochs. Each data point corresponds to the multi-unit activity for a single recording channel. Bold inset text (N) gives the total number of channels used to generate the smoothed density estimate. Corresponding labels for μ indicate the mean LvR value for samples during that epoch in each panel. One-way ANOVA comparing Sham (all data points in row A), 0.4-mA anodal (C, left) and 0.4-mA cathodal (C, right) indicates that cathodal stimulation may cause a decrease of LvR.

For all LFP data, we fit median log-transformed power values from each of the six frequency bands under consideration (Delta, Theta, Alpha, Beta, Low-Gamma, and High-Gamma) using standard least-squares (SLS) regression in JMP estimated by the restricted maximum likelihood (REML) method. The LFP model used the fixed effects of Epoch, Polarity, Intensity, Frequency Band, and their factorial interactions, as well as the random effect of Animal. In this model, the fixed effect of Frequency Band by itself was excluded, due to the known expected differences in average power as an effect of frequency (in general, low frequencies have higher power than high frequencies in the temporal fluctuations of the extracellular electrical field). For ANOVA and the LFP multi-variate regression model, F-ratios were computed to estimate the significance of whole model fixed effects using a single-tailed test on the F-Distribution. To test the significance of model parameter estimates, T-ratios were computed in order to make post-hoc comparisons, with significance estimates based on two-tailed tests of the T-Distribution.

For data points with many comparisons (time samples in Figure 4, or frequency bins in Figure 6), a conservative procedure was used to indicate individual comparison points that were significantly different from matched points in the aggregate sham distribution (See Supplementary Materials/ Multiple Comparisons Significance Tests Methods). This allowed us to test whether there were, for example, transient periods during the Stim or Post epochs in which an effect on FR may have manifested.

Figure 6. Average LFP spectral power.

To assess the influence of tDCS on population activity that may be reflected at a slower timescale than effects related to multi-unit activity, the frequency content of the extracellular LFP was analyzed. Each panel shows the difference in magnitude of log-transformed RMS power between the frequency content of the corresponding epoch under either anodal (left) or cathodal (right) and sham stimulation. Black brackets indicate frequency bins with a difference between 0.4-mA (maximum treatment) and 0.0-mA (sham) stimulation that was statistically significant at the level of α = 0.05, but not α = 0.01.

Results

tDCS treatment does not overtly affect spiking activity in intact, ambulatory rats

A total of 480 multi-unit channels were used in the analyses. When fluctuations in spike rate are considered on a finer time-scale within each epoch, there may be transient fluctuations resulting in significant deviations from the Pre epoch median multi-unit spike rate (Figure 4, black brackets) that are indicative of the nonstationary nature of spiking units. However, the appearance of these deviations even during the Pre epochs indicates that analyses of transient changes in spike rate on this timescale are susceptible to false positive detection due to the intrinsic variability of spiking activity, even after applying statistical corrections. By contrast, the ΔFR time-series in the 0.2-mA and 0.4-mA current intensity conditions for each polarity (Figure 4B,C) did not diverge significantly from aggregate sham stimulation at matched time-points (Also see Supplementary Figure S6 for a longer post-stimulation time period). The one-way ANOVA (R² = 0.1634) that summarizes these differences by comparing the FR data from the two most-intense stimulation conditions to the aggregate sham FR fails to reject the null hypothesis that either anodal or cathodal stimulation is significantly different from sham stimulation (DF = 2, F-Ratio = 0.4171, p = 0.66). Although one-way ANOVA of ΔLvR for the same groupings indicates a significant effect of treatment group (DF = 2, F-Ratio = 5.8866, p = 0.0031), the overall model fit was poor (R² = 0.0479), indicating that this effect may be driven by a subset of outlier units. Indeed, as shown in Figure 5, there may be a tendency for stimulation to cause a reduction in LvR during or after Cathodal stimulation. Because of the reliance of time-series statistics on the assumption of statistical independence for multiple simultaneously-observed spiking units, we performed an exploratory analysis of pairwise multi-unit cross-covariance (See Supplemental Materials/ Cross-Covariance Principal Components Analysis Methods). In general, the top-3 time-series principal components, which explained the majority of the pairwise covariance structure (Figure S3), remained close to zero (Figure S4), indicating the suitability of using multiple simultaneously-recorded channels of multi-unit activity for these analyses, although in individual recordings there were occasionally transient (1-minute or less) epochs of substantially correlated or anti-correlated activity across recording units during each of the three epochs (Figure S4). These occurrences were too sparse to statistically characterize.

To determine if a lack of FR changes was due to some vagary in the distribution of units that changed firing patterns, we analyzed the proportion of units changing firing rates in response to tDCS which might be obscured when all units are included in analysis. We categorized the bottom (Decrease) and top (Increase) 5^th percentiles of changes in multiunit FR between Pre and Stim epochs to test against the null hypothesis that the distribution of such units would be evenly distributed across treatment conditions. The responsive units, as defined, were not distributed homogeneously throughout the groups (Table 1; Pearson’s X² = 56.0660; p < 0.0001), suggesting that there is a significant difference in the proportion of responsive units that depends upon stimulation polarity and intensity. However, post-hoc comparison indicates this effect is primarily driven by a disproportionate number of units that had a significantly-reduced rate in the 0.2-mA Anodal condition (14 of 24; p < 0.0001), as well as a disproportionately large number of units with significant rate increases in the 0.0-mA Cathodal condition (9 of 24; p = 0.00119), suggesting this effect may relate more to the transient non-stationarities in FR that are typical during ambulatory recordings, as opposed to the treatment itself (See Supplementary Materials/ Table S3. Contingency table of responsive units).

Table 1. Contingency table of responsive units as determined using rate thresholding.

Each column indicates a different treatment. Within each cell, the top value is the total number of units for a given treatment meeting that condition, the middle value is the corresponding treatment-wise proportion, and the bottom row is the likelihood of that proportion being observed by chance, given the average proportion of responsive units to units that showed no response. Units were considered responsive if the change in spike rate between the Pre and Stim epochs was either in the lowest 5^th percentile (a decrease) or the upper 5^th percentile (an increase). Dark grey shading indicates a significant increase in FR compared to what would be expected given the average treatment-wise proportions.

Local Field Potentials (LFP) during tDCS

To assess whether slow fluctuations in LFP may be modified by tDCS, we computed the average frequency spectrum power in 30-second non-overlapping bins for the cross-channel average LFP signal (Figure 6). No power changes within any frequency bands, including low frequency oscillation corresponding to entrainment of local neural activity [19, 20]and in the high-gamma frequency band corresponding to localized changes in neuronal spiking [21, 22].

Discussion

As a therapeutic tool, the effects of tDCS typically have been described with respect to improvements in clinical scales and outcomes. For example, in stroke survivors, Boggio et al. demonstrated a significant effect of tDCS vs. sham stimulation on the Jebsen-Taylor Hand function test [23]. This effect was significant with either anodal tDCS of the affected hemisphere or cathodal tDCS of the unaffected hemisphere. Others have shown positive effects of tDCS on reaction times and pinch force in chronic stroke patients [24]. Clinical effects of tDCS have now been examined in a vast array of clinical conditions, including stroke, traumatic brain injury, Parkinson’s disease, dystonia, multiple disorders [25]. sclerosis, epilepsy, pain syndromes, and neuropsychiatric Unfortunately, the cumulative results of these clinical studies lack consistency, either due to variability of testing procedures, populations, or individual response to tDCS. A recent meta-analysis of 26 tDCS randomized controlled stroke trials with 754 participants found evidence for a moderate effect of cathodal tDCS for Activities of Daily Living, but not for Upper Extremity Fugl-Meyer Motor scores [26]. While several stimulation parameters were independently examined, such as pad size, charge density, current density, etc., none were found to be significant modifiers. In addition to improvements in reporting stimulation conditions in future clinical trials, it is imperative that the mechanisms underlying the effects of tDCS be resolved to improve the scientific rigor of this promising and easily applied non-invasive treatment approach.

tDCS in ambulatory rats does not result in overall changes in neuronal firing rates

The maximum current density (39.8 A/m² at 0.4 mA) predicted from the FEA is theorized to induce changes in neuronal excitability but below the level that would cause damage to the cerebral cortex [13]. When compared to finite element models of clinical tDCS parameters [27–29], we would expect to have effective stimulation both beneath the stimulating screws and near the site of the electrode array. In the entire population of units examined, there were no significant changes in firing rate or pattern in any of the stimulation conditions. While we observed some alterations in the power of the LFP signal during the 0.4mA Cathodal condition, transient increases in power in various LFP bands are expected due to intrinsic non-stationarity of the signal in ambulatory rats. LFP fluctuations in the present study were unlikely to be directly related to changes in neuronal firing, since we did not observe consistent increases (or decreases) in power in the high-gamma frequency range (60–200 Hz) correlated with the period of stimulation or post-stimulation epochs [30].

Despite negative results for FR and LFP, we found that there were subtler changes in the variability of spiking patterns, measured by LvR. Specifically, cathodal stimulation appears to cause LvR to decrease. Low LvR values indicate a relatively uniform spike train, while high ones indicate burstiness; therefore, this shift may relate to a reduction in the relative burstiness of spike activity during the stimulation epoch, relative to preceding stimulation onset. A possible confounding factor for the analyses of LvR could be the short periods during which activity is substantially correlated or anti-correlated (Figure S4). For example, if a channel comprised of predominantly multi-unit activity arising from one or two units bursting in isolation began to detect the activity of other cells as more units with tonic firing profiles became active due to factors inducing an aggregate increase in correlated activity within the observed motor cortical region, it would result in an apparent decrease of LvR for that channel despite the possibility of no actual change in firing on the previously bursting detected units. Therefore, while we observed a trend of subtle reduction in LvR, it is not possible from to conclude from these data whether the change arises due to alterations in the activity profile within an observed spiking unit, or whether it is due to the emergence or suppression of previously undetected or detected units, respectively.

Limitations of present study

We attempted to produce identical stimulation conditions for each animal and confirmed that the bone screws used for the electrodes did not penetrate completely through the skull. Based on our estimates of maximum screw depth (0.75 mm) and materials properties, we estimated a maximum current density of 39.8 A/m² using an FEA. This resulted in a higher current density than is observed clinically, but well under levels that are known to cause cortical lesions[13]. Due to the required stability of the bone screws in ambulatory animals, we are confident that the depth of penetration could not have been < 0.5 mm. However, small differences in current density among animals due to slightly different screw depths may have contributed to some variability in our outcomes. Nevertheless, the current densities estimated both at the level of the tDCS and recording electrodes should have been sufficient to modulate activity without damaging the neural tissue.

We attempted to optimize the likelihood that changes in FR could be detected by placing the recording microelectrodes approximately in layer V in M1. M1 slice preparations have demonstrated that action potential thresholds in response to polarizing currents are lower in layer V/VI pyramidal soma compared to layer II/III pyramidal neurons and interneurons [31]. Also, the microelectrode recording array in M1 was placed as closely as possible to the ipsilateral anodal bone screw. While the FEA indicates that at this location there are likely radial currents in the opposite polarity of the anodal current delivered, we still expect increased firing rates due to the theorized likelihood of activation directly under the stimulating electrode and the subsequent horizontal activation at the level of the recording array, especially in behavioral conditions in which the animals can move with minimal constraint, but not performing a specific motor task. Spontaneous FR might be too low to detect decreased firing directly under a cathodal electrode. Since contralateral cathodal stimulation is thought to induce interhemispheric disinhibition, we reasoned that this arrangement would maximize our changes of detecting significant increases in spontaneous FR in the recorded neurons.

Further, the rats were freely ambulatory, but by design, were not performing any specific behavioral task. Indeed, several studies have shown that tDCS effects are dependent on the ongoing activity and these effects are more pronounced when such activity is associated with cortical engagement of the stimulated area [32, 33]. Temporal alignment of neurophysiological analyses to external stimuli or elements of a specific behavioral context may help illuminate subtle, subthreshold modulations in FR or LFP that are not coherent during ambulatory behavior. Additional studies would be needed to understand whether tDCS can modify neuronal firing when used in conjunction with a behavioral task.

Mechanisms underlying tDCS effects based on neuronal depolarization

Studies of the effect of polarizing currents on neuronal activity using direct measures of neuronal firing patterns in awake animals are relatively rare. In seminal experiments, Bindman et al. applied direct current (5–50 μA) to the exposed cortical surface of urethane-anesthetized rats and found that spontaneous FR of cortical neurons increased in response to surface-positive currents and decreased in response to surface-negative currents. Evoked potentials produced by peripheral stimulation were also altered in a similar polarity-dependent manner, demonstrating increased excitability [9]. However, it is difficult to translate these effects of stimulation applied directly to the cortical tissue and in anesthetized animals to currents applied to the scalp or skull in ambulatory animals. As demonstrated in slice preparations, the effects of electric fields on different neuronal compartments are complex, and whether an external stimulus elicits an action potential in a neuron depends on the integration of activity in its dendrites, soma, axon hillock, axons and presynaptic terminals [2]. Further, in hippocampal slice studies, the excitatory or inhibitory effects of DC stimulation are determined by the orientation of the axons in the electric field [34]. It is unclear how well in vitro models predict the complex interaction of cortical neurons in intact, behaving animals.

The weak effect of tDCS in the present study is likely due to several factors. First, to induce measurable effects on neuronal spiking, Vöröslakos et al. found that at least 1 mV/mm voltage gradient was necessary. However, such gradients are not achieved in typical human protocols. Based on cadaver experiments, currents would need to exceed 4–6 mA to achieve these gradients in brain tissue [35]. Typical protocols used in clinical studies with wet sponges use currents of 1–2 mA [36]. Moreover, the scalp, subcutaneous tissue and muscles function as an effective shunt, resulting in at least 50% reduction of applied current intensity. The resistance of the skull further reduces the current flow by another 10–25% depending on its thickness. Given this significant attenuation of physical current flow due to the material properties of physiological tissue alone, the likelihood of directly depolarizing neuronal populations is small. Behavioral and cognitive effects reported in previous tDCS studies are likely driven by effects other than changes in spontaneous neuronal firing.

Synaptic mechanisms of tDCS effects

Evidence is accumulating that supports various synaptic mechanisms of excitability changes not necessarily involving spontaneous neuronal firing changes. At the synaptic level, one of the effects induced by DC electric fields is the attraction of the acetylcholine receptors and the tropomyosin-receptor-kinase (TrK) families, leading to an increase in the probability of synaptic vesicle and neurotransmitter release [37]. More recently, Monai et al. demonstrated in transgenic mice that tDCS induced not only enhanced sensory evoked cortical potentials, but also synchronous, large-amplitude Ca²⁺ surges across the cortex with no changes in LFP that would be indicative of neuronal depolarizations [38]. Most interestingly, Ca²⁺ surges were restricted to astrocytic populations and did not occur in neurons. Finally, Ca²⁺ surges were dependent upon inositol triphosphate receptor type 2 (IP₃R2). Since Ca²⁺/ IP₃ signaling is thought to play a role in synaptic plasticity in the cortex and hippocampus [39], this mechanism might provide a partial explanation for changes in synaptic efficacy in the absence of spontaneous changes in neuronal FR. In addition to astrocytic changes, glial cells have also been reported to be sensitive to changes in response to electrical fields and may be serve as another intermediate pathway for inducing the neuroplastic changes necessary to explain the clinical efficacy of tDCS [40]

Differences between tDCS and other modes of non-invasive tES

It should be noted that the present results and interpretation are restricted to DC current fields. A different mode of non-invasive brain stimulation, transcranial alternating current stimulation (tACS), produces substantial effects on neuronal FR when applied to the scalp. Using sinusoidal electrical stimulation of either the scalp or the dura, neuronal firing can be entrained in widespread regions of the cortex [41]. However, based on the results reported here, changes in FR in direct response to tDCS without behavioral intervention are unlikely to occur at the stimulation intensities known to be safe and typically employed in human or animal studies.

Conclusion

In an ambulatory rat model, tDCS applied directly through the skull at safe current densities that have been shown to elicit behavioral effects in prior studies had limited impact on firing rates or patterns of extracellularly recorded units within layer V of M1. Furthermore, there were no changes observed in the LFP activity within the array indicative of neuronal firing related to the stimulation. These results indicate that tDCS delivered agnostic to behavior and at current intensities typically used in humans or animals, likely produces observed effects via mechanisms other than direct neuronal depolarization.

Highlights.

• In awake rats, tDCS had no effect on overall neuronal firing rate or pattern.

• The highest intensity anodal condition had a greater proportion of responsive units.

• tDCS does not result in widespread neuronal firing at typical current intensities.