Skip to main content
PLOS ONE logoLink to PLOS ONE
. 2013 Nov 20;8(11):e81402. doi: 10.1371/journal.pone.0081402

High-Frequency Stimulation of Excitable Cells and Networks

Seth H Weinberg 1,*
Editor: Matjaz Perc2
PMCID: PMC3835437  PMID: 24278435

Abstract

High-frequency (HF) stimulation has been shown to block conduction in excitable cells including neurons and cardiac myocytes. However, the precise mechanisms underlying conduction block are unclear. Using a multi-scale method, the influence of HF stimulation is investigated in the simplified FitzhHugh-Nagumo and biophysically-detailed Hodgkin-Huxley models. In both models, HF stimulation alters the amplitude and frequency of repetitive firing in response to a constant applied current and increases the threshold to evoke a single action potential in response to a brief applied current pulse. Further, the excitable cells cannot evoke a single action potential or fire repetitively above critical values for the HF stimulation amplitude. Analytical expressions for the critical values and thresholds are determined in the FitzHugh-Nagumo model. In the Hodgkin-Huxley model, it is shown that HF stimulation alters the dynamics of ionic current gating, shifting the steady-state activation, inactivation, and time constant curves, suggesting several possible mechanisms for conduction block. Finally, we demonstrate that HF stimulation of a network of neurons reduces the electrical activity firing rate, increases network synchronization, and for a sufficiently large HF stimulation, leads to complete electrical quiescence. In this study, we demonstrate a novel approach to investigate HF stimulation in biophysically-detailed ionic models of excitable cells, demonstrate possible mechanisms for HF stimulation conduction block in neurons, and provide insight into the influence of HF stimulation on neural networks.

Introduction

Electrical signaling is fundamental to the physiological function of excitable cells such as neurons and cardiac myocytes. Irregular electrical patterns in the brain and heart can lead to life-threatening conditions including epileptic seizures and ventricular fibrillation. External stimulation can terminate these irregular rhythms [1], [2]; however large strength stimuli are often associated with detrimental effects such as pain [3] and impaired cardiac function following defibrillation [4].

In the 1960s, it was shown that kilohertz-range high frequency (HF) sinusoidal stimulation could reversibly block conduction in neurons [5]. The use of 1–40 kHz HF-induced neural conduction block has recently been exploited in clinical studies for diagnostic and therapeutic purposes, improving bladder function [6], [7] and mitigating pain associated with peripheral nerve activity [8][10]. Despite the clinical usage of HF stimulation treatment, the mechanisms underlying therapeutic success in these physiological and pathological settings are unclear. Simulation studies in neurons have suggested two mechanisms: reduced sodium channel availability due to transmembrane potential depolarization and persistent activation of potassium channels [8][14]. However, the relative significance of the two mechanisms varies with the properties of the neuron, as well as the specific species and model. Further, in simulation studies, the transmembrane potential, ionic currents, and channel gating variables oscillate on the fast time scale of the HF stimulus, varying throughout the HF stimulation period, such that distinguishing the precise influence of the HF stimulus is difficult.

Alternatively, one can apply a multi-scale method, separating the fast time scale dynamics—due to the HF stimulus—and the slow dynamics of the excitable cell, and derive an averaged model, which accounts for the HF stimulus but does not contain a high-frequency term [15]. Using this type of approach, several studies have analyzed the influence of a HF stimulus in the simple FitzHugh-Nagumo (FHN) model [16]. Cubero and colleagues demonstrated that the model cell cannot repetitively fire when the HF stimulus amplitude-frequency ratio is above a critical value [17]. Ratas and Pyragas showed that this ratio also influenced conduction speed in a nerve axon and above a critical value led to conduction block [18], [19]. The FHN model is minimalistic, reproducing many important aspects of cellular excitability [20], and ideal for analysis, as the model only contains two variables, permitting the use of standard nonlinear dynamics techniques such as phase-plane analysis. However, in general biophysically-detailed models of excitable cells are more complex than represented by the simple two-dimensional FHN model.

In this study, we first illustrate the multi-scale method to derive the averaged FHN (AFHN) model equations and use phase-plane analysis to determine critical HF stimulus thresholds above which the model cannot exhibit repetitive firing or elicit a single action potential. We then extend this approach to simulate the dynamics of the classical Hodgkin-Huxley (HH) neuron model [21] and illustrate similarities and differences between the AFHN and averaged Hodgkin-Huxley (AHH) model. Further, we demonstrate that HF stimulation alters the ionic current activation and inactivation dynamics, illustrating possible mechanisms for conduction block in a single neuron. Finally, we simulate HF stimulation in a network of neurons and demonstrate that HF stimulation alters network synchrony and, above a critical stimulation strength, terminates persistent network activity, suggesting implications for clinical therapy.

Results

Averaged Fitzhugh-Nagumo model

We begin by deriving and analyzing a simplified model of spiking which accounts for the influence of HF stimulation. We consider the FitzHugh-Nagumo (FHN) model [16] with the addition of a constant current Inline graphic and a time-varying HF stimulus

graphic file with name pone.0081402.e002.jpg (1a)
graphic file with name pone.0081402.e003.jpg (1b)

where Inline graphic is the frequency of the HF stimulation, Inline graphic is the HF amplitude-frequency ratio,

graphic file with name pone.0081402.e006.jpg

and the dot indicates differentiation with respect to time Inline graphic. The HF stimulation term is defined in terms of Inline graphic and Inline graphic with the foresight that the influence of the HF stimulation depends on Inline graphic, not specifically the amplitude Inline graphic. The FHN model is a simplified model that reproduces many important properties and dynamics of excitable cells. The simplicity of the model permits a geometric illustration—through phase plane analysis—of many important biophysical phenomena such as repetitive spiking and depolarization block. In the model, Inline graphic is the dimensionless transmembrane potential, and Inline graphic represents the degree of refractoriness. Throughout the paper, we fix Inline graphic, Inline graphic, and Inline graphic, such that in the absence of any external stimuli, the neuron is excitable.

If the period of the fast HF stimulus is much smaller than all characteristic times of the FHN model, according to the method of averaging [15], an approximation to the slow system can be obtained by averaging over the period of the HF stimulus. As shown in the Methods, the variables of the averaged Fitzhugh-Nagumo model (AFHN), Inline graphic and Inline graphic, are governed by the following system of equations:

graphic file with name pone.0081402.e019.jpg (2a)
graphic file with name pone.0081402.e020.jpg (2b)

where

graphic file with name pone.0081402.e021.jpg
graphic file with name pone.0081402.e022.jpg

and

graphic file with name pone.0081402.e023.jpg

The AFHN model is very similar to the FHN model, with the only difference being the modification of the cubic function Inline graphic that influences the dynamics of Inline graphic. In the absence of HF stimulation, i.e., Inline graphic, the two models are identical. In the following sections, we investigate how the HF stimulus parameter Inline graphic influences the properties of repetitive action potential firing. In Figure 1, we plot Inline graphic from simulations of the FHN model (black) for various values of the HF stimulation frequency Inline graphic and compare with Inline graphic from a simulation of the AFHN neuron model (red), for fixed values for Inline graphic and Inline graphic. In general, as Inline graphic increases, Inline graphic from AFHN model becomes a better approximation of the average value of Inline graphic from the FHN model simulation, validating our formulation.

Figure 1. Validation of the AFHN model.

Figure 1

Simulated Inline graphic traces from the FHN (black) and AFHN (red) models for varying HF stimulus frequency Inline graphic. Parameters: Radial frequency Inline graphic is identified in each panel, Inline graphic, Inline graphic.

Repetitive firing in the AFHN model. In the parameter region considered in this study, the cell is excitable, that is, in the absence of an external stimulus, the cell is at rest, and the addition of a stimulus can induce a single or multiple action potentials. In this study, we will consider two types of applied current stimuli: a constant applied current and a brief applied current pulse, in addition to the HF stimulation.

We first consider the case of a constant applied current Inline graphic. In Figure 2A, we plot Inline graphic for a constant current Inline graphic and various values of Inline graphic. For no HF stimulus (Inline graphic), the cell fires repetitively. Increasing the amplitude of the HF stimulation parameter Inline graphic decreases the action potential amplitude and increases the firing frequency. Consistent with previous studies [17], increasing Inline graphic further results in cessation of repetitive firing, following a single action potential at the stimulus onset. Conditions for cessation of firing are derived as follows.

Figure 2. Repetitive firing in the AFHN model.

Figure 2

(A) Simulated Inline graphic traces for Inline graphic and different values for Inline graphic. The dashed lines indicate Inline graphic. (B) Phase-plane portrait for variable Inline graphic and Inline graphic. In each panel, the Inline graphic-nullcline (green) is shown for 3 values of Inline graphic. The Inline graphic-nullcline (blue) is independent of Inline graphic and Inline graphic. (C) Inline graphic-Inline graphic parameter space, denoting regions of rest, repetitive firing, and block. The limit cycle lower and upper limits (Inline graphic, Eq. 6 ) and rheobase (Inline graphic, Eq. 9 ) as functions of Inline graphic. (D) Frequency and amplitude of action potentials, as functions of Inline graphic and Inline graphic.

For the parameters chosen, the AFHN model has a single steady-state Inline graphic, which satisfies the implicit expression

graphic file with name pone.0081402.e067.jpg (3a)
graphic file with name pone.0081402.e068.jpg (3b)

and shown in Figure 3. As Inline graphic increases, the resting potential Inline graphic becomes more depolarized and approaches 0 for large Inline graphic. The degree of refractoriness also increases as Inline graphic increases, such that Inline graphic approaches Inline graphic for large Inline graphic.

Figure 3. Steady-state of the AFHN model.

Figure 3

The steady-state transmembrane potential Inline graphic and degree of refractoriness Inline graphic are shown as functions of the HF stimulation parameter Inline graphic. Critical values of Inline graphic for repetitive firing Inline graphic and for evoking a single action potential following a brief applied current pulse Inline graphic are identified. See text for description of critical values.

Using standard techniques from linear stability analysis [22], the stability of the steady-state Inline graphic can be determined by linearizing around Inline graphic, and evaluating the matrix of partial derivatives, the Jacobian Inline graphic, at the steady-state,

graphic file with name pone.0081402.e085.jpg (4)

When the steady-state becomes unstable, specifically the real part of the eigenvalues of Inline graphic, Inline graphic, a stable limit cycle emerges, which can be interpreted biophysically as repetitive action potential firing. The critical parameter value at which the limit cycle emerges is known as a Hopf bifurcation. Many previous studies have shown that in the FHN model (i.e., Inline graphic), as the applied current Inline graphic increases, there are two critical values for Inline graphic, Inline graphic and Inline graphic, which correspond to the onset and offset of the stable limit cycle, respectively [23][25]. Below Inline graphic, the steady-state is stable corresponding to the cell at rest, between Inline graphic and Inline graphic the steady-state is unstable and the cell repetitively fires, and above Inline graphic, the steady-state is stable again and the cell is in depolarization block [23].

In Figure 2B, we plot the nullclines of the AFHN model for several values of Inline graphic and Inline graphic. The Inline graphic-nullcline (green)—given by the set of all points Inline graphic such that Inline graphic—is a cubic function of Inline graphic, while the Inline graphic-nullcine (blue)—similarly defined as the set of all points Inline graphic such that Inline graphic—is linear, and the nullclines intersection denotes the location of the steady-state. For a given value of Inline graphic, increasing Inline graphic shifts the Inline graphic-nullcline upwards, while the Inline graphic-nullcline is independent of both Inline graphic and Inline graphic.

If Inline graphic is such that the steady-state is located on the middle branch of the Inline graphic-nullcline, and if Inline graphic is sufficiently slow compared to Inline graphic, that is, Inline graphic, then it can be shown that the steady-state is unstable, and a stable limit cycle exists [23]. From a geometric illustration, we can anticipate a critical value of Inline graphic, Inline graphic, above which a stable limit cycle and repetitive firing cannot exist, consistent with Figure 2A (bottom panel). As Inline graphic increases, the slope of the middle branch of the Inline graphic-nullcline decreases, and the “knees” of the nullcline move towards the steady-state Inline graphic. When the slope at Inline graphic equals 0, the middle branch of the nullcline no longer exists and, therefore regardless of Inline graphic, a stable limit cycle also does not exist. Using the slope of the Inline graphic-nullcline alone as a criterion for the critical value of Inline graphic, Inline graphic.

From linear stability analysis, we can more precisely determine the necessary condition for a limit cycle, Inline graphic, such that Inline graphic is given by

graphic file with name pone.0081402.e129.jpg (5)

For all values of Inline graphic the steady-state is always stable, regardless of Inline graphic, as previously shown by [17]. Further, for Inline graphic, the critical stimulus upper and lower limits, Inline graphic and Inline graphic, respectively, are given by

graphic file with name pone.0081402.e135.jpg (6)

The Inline graphic curves separate the regions of rest, repetitive firing, and depolarization block in the Inline graphic-Inline graphic parameter space and coalesce when Inline graphic at a double Hopf bifurcation (Figure 2C). For the parameters used in this study, Inline graphic.

In the regime for repetitive firing, we derive an approximation for the action potential frequency and amplitude in the AFHN model (see Methods). For a given value of Inline graphic, the frequency first increases then decreases as Inline graphic increases (Figure 2D), while the amplitude is constant, consistent with a relaxation oscillator. Increasing Inline graphic increases frequency and decreases the action potential amplitude, consistent with Figure 2A.

Excitability in the AFHN model. We next consider the excitability of the AFHN model following a brief applied current, in the presence of HF stimulation, by determining the strength-duration curve, the relationship between the duration Inline graphic of an applied current pulse and the minimum amplitude Inline graphic such that an action potential fires [26].

With the system initial at rest, i.e., Inline graphic, we make the assumption that an action potential is fired when Inline graphic reaches some threshold Inline graphic. Although it has been shown that the FHN model does not strictly exhibit all-or-none threshold behavior [23], when Inline graphic is sufficiently slow compared with Inline graphic, the middle root of the Inline graphic-nullcline is a reasonable approximation for an action potential threshold, which we show increases as Inline graphic increases (see Methods for details and references on firing threshold, Eq. 44, Figure 4C). This threshold-like behavior is illustrated in Figure 4. We plot Inline graphic as a function of time following brief Inline graphic current pulses for Inline graphic and 1. For both values of Inline graphic, an action potential is elicited if Inline graphic during the brief pulse, while if Inline graphic during the current pulse, Inline graphic returns to rest Inline graphic. Increasing Inline graphic increases both the Inline graphic threshold for evoking an action potential, Inline graphic, and the stimulus threshold Inline graphic necessary to elicit an action potential (Figure 4C).

Figure 4. Excitability in the AFHN model.

Figure 4

(A) Simulated Inline graphic traces during brief Inline graphic stimuli pulses of amplitude Inline graphic for Inline graphic, 1, and 1.5. In simulations that Inline graphic exceeds the threshold Inline graphic, an action potential is elicited. Inset shows an expanded time course. (B) Strength-duration curve ( Eq. 8 ) for several values of Inline graphic. (C) Rheobase (Inline graphic, Eq. 9 ) and chronaxie (Inline graphic, Eq. 10 ) as functions of Inline graphic.

In the Methods section, we show a critical value for Inline graphic, Inline graphic, exists, which for all values of Inline graphic,

graphic file with name pone.0081402.e178.jpg (7)

the AFHN model cannot be excited by a brief applied current, where Inline graphic is defined in Eq. 27. Using the parameters used in this study, Inline graphic. For Inline graphic, regardless of the magnitude of the stimulus pulse Inline graphic, Inline graphic relaxes back towards the steady-state value Inline graphic following the applied current pulse, without a large amplitude excursion typical of an action potential (Figure 4A, right panel).

For Inline graphic, and the strength-duration curve is approximated by

graphic file with name pone.0081402.e186.jpg (8)

where the prime indicates differentiation with respect to Inline graphic, such that

graphic file with name pone.0081402.e188.jpg

We plot the strength-duration curves in Figure 4B for several values of Inline graphic. For all values of Inline graphic, Inline graphic decreases linearly with Inline graphic when presented on a logarithmic scale and approaches a constant value for long d, a relationship typical of excitable cells. For a given stimulus duration d, the strength required to elicit an action potential Inline graphic increases as Inline graphic increases. Two important values are typically determined from the strength-duration curves: rheobase (Inline graphic, defined as Inline graphic for an infinite duration pulse, and chronaxie (Inline graphic), defined as the pulse duration having a threshold that is twice the rheobase. From Eq. 8, Inline graphic and Inline graphic are given by

graphic file with name pone.0081402.e200.jpg (9)

and

graphic file with name pone.0081402.e201.jpg (10)

respectively. We plot Inline graphic and Inline graphic as a function of Inline graphic in Figure 4C. Both Inline graphic and Inline graphic are fairly constant for small Inline graphic. Inline graphic increases and Inline graphic decreases, as Inline graphic further increases towards Inline graphic. We also plot Inline graphic in Figure 2C for comparison with Inline graphic, and note that for all values of Inline graphic, Inline graphic, that is, a smaller Inline graphic is required to elicit a single action potential than to elicit repetitive spiking, as expected. We note that the derivation of Eq. 8 assumes the stimulus Inline graphic is brief—that is, Eq. 8 is strictly valid for small d—therefore, Inline graphic should not be interpreted as a critical Inline graphic above which no action potentials can be elicited by longer duration stimuli. Indeed, Inline graphic, and therefore, the cell can repetitively fire during long duration stimuli for Inline graphic, and a single action potential can be elicited by long duration stimuli for Inline graphic. Further, since rheobase is defined as a stimulus threshold for infinite d, Eqs. 9 and 10 should be interpreted as approximations derived from Eq. 8, which nonetheless provide qualitative relationships between the strength-duration curve parameters Inline graphic and Inline graphic and HF stimulation parameter Inline graphic that can be compared with a biophysically-detailed model, as discussed in the next section.

In summary, increasing the HF stimulation parameter Inline graphic increases the thresholds for both repetitive firing and a single action potential, Inline graphic and Inline graphic, respectively. We derive expressions for critical values of Inline graphic and determine the influence of HF stimulation on the resting potential, firing frequency and amplitude, action potential threshold, rheobase, and chronaxie. These theoretical relationships provide references that can be compared to results from a more realistic neuron model described in the next section.

Averaged Hodgkin-Huxley model

We next derive and analyze the influence of HF stimulation on a biophysically-detailed model of the neuron, utilizing the techniques described in the previous section. We consider the classical space-clamped Hodgkin-Huxley (HH) neuron model of the giant squid axon [21], with the addition of an applied current Inline graphic and HF stimulus, given by the following system of equations:

graphic file with name pone.0081402.e231.jpg (11a)
graphic file with name pone.0081402.e232.jpg (11b)
graphic file with name pone.0081402.e233.jpg (11c)
graphic file with name pone.0081402.e234.jpg (11d)

where HF stimulus parameters Inline graphic and Inline graphic are defined as before. In the HH neuron model, Inline graphic represents the transmembrane voltage Inline graphic relative to the resting potential Inline graphic, Inline graphic the sodium activation gating variable, Inline graphic the sodium inactivation gating variable, and Inline graphic the potassium activation variable. Current conductances, reversal potentials, and gating variable dynamics are described in the Methods.

Assuming that the period of the fast HF stimulation is much shorter than the characteristic times of the dynamics of Inline graphic and the gating variables, as in the previous section, we approximate the dynamics of the slow variables by averaging over the period of the HF stimulus. The variables of the averaged Hodgkin-Huxley (AHH) model, Inline graphic, Inline graphic, Inline graphic, and Inline graphic, are governed by the following system of equations:

graphic file with name pone.0081402.e248.jpg (12a)
graphic file with name pone.0081402.e249.jpg (12b)
graphic file with name pone.0081402.e250.jpg (12c)
graphic file with name pone.0081402.e251.jpg (12d)

where

graphic file with name pone.0081402.e252.jpg (12e)
graphic file with name pone.0081402.e253.jpg (12f)
graphic file with name pone.0081402.e254.jpg (12g)

and

graphic file with name pone.0081402.e255.jpg (12h)

for Inline graphic. Because of the simplicity of the FHN model, we could derive analytical expressions for the dynamics of the AFHN model variables. In the HH model, the expressions for the Inline graphic and Inline graphic terms that govern the dynamics of the gating variables are complex, and as such, it is not possible to derive analytical expressions for Eqs. 12e and 12f without using approximations for the exponential function. Therefore, Eqs. 12e and 12f are computed by numerical integration for particular values of Inline graphic and Inline graphic.

As with the FHN model, we plot Inline graphic from simulations of the HH model (black) for various values of the HF stimulation frequency and compare with Inline graphic from a simulation of the AHH neuron model (red), for fixed values for Inline graphic and Inline graphic (Figure 5). Below an HF stimulus frequency Inline graphic of 5 kHz, there is significant disagreement between the averaged and original model. As Inline graphic increases, Inline graphic from the AHH model becomes a better approximation of the average value of Inline graphic from the HH model simulation, validating the use of the averaging method.

Figure 5. Validation of the AHH model.

Figure 5

Simulated Inline graphic traces from the HH (black) and AHH (red) models for varying HF stimulus frequency Inline graphic. Parameters: Inline graphic rad/s (where Inline graphic is the frequency identified in each panel), Inline graphic  =  30 Inline graphic, Inline graphic Inline graphic.

Repetitive firing in the AHH model. As in the previous section, we consider the influence of an applied current Inline graphic in the AHH model, in the presence of HF stimulation. In Figure 6A, we plot Inline graphic for different values of Inline graphic, such that the neuron is repetitively firing, i.e., Inline graphic. For sufficiently large Inline graphic, the neuron does not repetitively fire.

Figure 6. Repetitive firing in the AHH model.

Figure 6

(A) Simulated Inline graphic traces for Inline graphic Inline graphic and different values for Inline graphic. The dashed lines indicate Inline graphic mV. (B) Inline graphic-Inline graphic parameter space, denoting regions of rest, repetitive firing, and block. The limit cycle lower and upper limits (Inline graphic) and rheobase (Inline graphic) as functions of Inline graphic. (C) Action potential frequency and amplitude, as functions of Inline graphic and Inline graphic. Inline graphic in units of Inline graphic.

We plot the Inline graphic-Inline graphic parameter space for the AHH model in Figure 6B. The parameter space is qualitatively similar to the AFHN model, such that the range of Inline graphic for which the neuron repetitively fires becomes smaller as Inline graphic increases, and above a critical value of Inline graphic, Inline graphic, the neuron does not repetitively fire. In the HH model, it has been shown that action potential frequency increases and the action potential amplitude decreases for increasing Inline graphic [23], and we find that this is true for a given value of Inline graphic. For a given Inline graphic, as Inline graphic increases, in agreement with the AFHN model, action potential amplitude decreases. However, in contrast with the AFHN model, the frequency decreases as Inline graphic increases (Figure 6C).

Excitability in the AHH model. We next consider excitability in the AHH model following brief applied current pulses. Here, we consider both positive (cathodal) and negative (anodal) applied current stimuli. As with the FHN model, the HH model is known to not exhibit a strict all-or-none firing threshold. However, especially for brief (0.1 ms) pulses, the HH model demonstrates a threshold-like response. In Figure 7A, we plot Inline graphic for different values of Inline graphic and Inline graphic. Consistent with the AFHN model, the Inline graphic threshold for evoking an action potential, Inline graphic, increases for increasing Inline graphic (left, middle panels). Further, above a critical value of Inline graphic, Inline graphic, an action potential cannot be evoked, regardless of Inline graphic (right panel). Although Inline graphic reaches levels near 0 mV, these responses should not be considered action potentials, as the depolarization of Inline graphic does not arise as a consequence of the regenerative activation of inward currents but rather solely as a perturbation due to the large applied stimulus. Specifically, above Inline graphic, regardless of stimulus amplitude Inline graphic, Inline graphic is maximally depolarized as the end of the stimulus pulse and does not become further depolarized following the pulse.

Figure 7. Excitability in the AHH model.

Figure 7

(A) Inline graphic traces following brief (0.1 ms) cathodal and (B) anodal stimulus pulses, for different values of Inline graphic. Threshold Inline graphic indicated in each panel. (C) Cathodal and anodal strength-duration curves for different values of Inline graphic. (D) Cathodal and anodal threshold Inline graphic (for 0.1 ms stimuli), rheobase, and chronaxie, as functions of Inline graphic. Current pulse amplitudes in (A): 64-66 (left); 633-635 (middle); 600, 800, 1000 (right); in (B): 198-200 (left); 397-399 (middle); 400, 600, 800 (right); in Inline graphic.

We also consider the influence of HF stimulation on excitability following anodal break stimulation, also known as post-inhibitory rebound. In the classical HH model (i.e., Inline graphic), a negative (anodal) applied current pulse Inline graphic hyperpolarizes the steady-state resting transmembrane potential Inline graphic (Figure 7B), which permits sodium inactivation recovery, i.e., Inline graphic moves closer to 1. Following the pulse offset (break), Inline graphic returns towards the more depolarized initial resting potential, and due to the slower sodium inactivation kinetics, Inline graphic rebound can be sufficiently large to evoke an action potential. As with cathodal stimulation, the threshold for stimulation, Inline graphic (determined as the magnitude of the hyperpolarization necessary for a post-inhibitary rebound), increases for increasing Inline graphic (left, middle panels), and above a critical value of Inline graphic, Inline graphic an action potential cannot be evoked (right panel). Inline graphic is larger for anodal stimulation (Figure 7D top panel, red), compared with cathodal stimulation (black), and the difference increases as Inline graphic increases, meaning a relatively larger anodal stimulation is necessary to evoke an action potential. Consistent with this finding, we find that Inline graphic Inline graphic.

Strength-duration curves for cathodal and anodal stimulation in the AHH model are shown in Figure 7C. Consistent with the AFHN model, for a given duration, the necessary cathodal applied current strength Inline graphic increases as Inline graphic increases. Further, rheobase Inline graphic increases and chronaxie Inline graphic decreases as Inline graphic increases, as in the AFHN model (Figure 7D, black traces, middle and bottom panels). As Inline graphic approaches Inline graphic, the strength-duration curve becomes flatter, consistent with a decreasing chronaxie, and illustrating that for large Inline graphic the magnitude of the applied pulse, and not the duration, determine whether an action potential is evoked. For a given Inline graphic, anodal rheobase is slightly larger compared with cathodal rheobase (Figure 7C, D). As Inline graphic approaches Inline graphic, in contrast with cathodal strength-duration curves, the anodal curves become steeper, such that chronaxie increases as Inline graphic increases (Figure 7D, bottom panel), illustrating that short duration anodal pulses become relatively less effective for evoking post-inhibitory rebound action potentials.

Dynamics of the AHH model. For the AFHN model, we demonstrate that Inline graphic and Inline graphic can be approximated via theoretical analysis of the two-dimensional dynamical system, based primarily on analysis of the influence of Inline graphic on the phase plane. Various approaches have been used to simplify the HH model to a FHN-like two-dimensional system, often assuming fast sodium activation and a linear relationship between gating variables Inline graphic and Inline graphic for a given Inline graphic [20]. However, we found that a similar phase plane analysis using this type of reduction of the AHH model was only moderately successful at reproducing AHH dynamics, likely due to the complex relationship between the gating variables dynamics over a wide range of Inline graphic and Inline graphic (not shown).

In the AFHN model, the HF stimulation parameter Inline graphic influences the dynamics of Inline graphic through the cubic function Inline graphic. In contrast, in the AHH model the dynamics of Inline graphic are altered indirectly through the influence of Inline graphic on the gating variables. In Figure 8A, we plot the steady-state activation, inactivation, and time constant curves as functions of Inline graphic for different values of Inline graphic. As Inline graphic increases, the sodium activation Inline graphic and inactivation Inline graphic steady-state curves are shifted to the right, the potassium activation Inline graphic steady-state curve is shifted to the left, and all three curves are less steep (Figure 8A). The time constants Inline graphic, Inline graphic, and Inline graphic all decrease as Inline graphic increases.

Figure 8. Steady-state of the AHH model.

Figure 8

(A) Steady-state gating variables Inline graphic, Inline graphic, Inline graphic and time constants Inline graphic, Inline graphic, Inline graphic as functions of Inline graphic in the AHH neuron model for different Inline graphic values. (B) Steady-state values for the transmembrane potential Inline graphic and the gating variables (left), and gating variable time constants at Inline graphic (right), as functions of Inline graphic. Vertical dashed lines indicate Inline graphic, Inline graphic, and Inline graphic (see text for description). In the top panels, the horizontal dashed line indicates Inline graphic mV for Inline graphic. Time constants in units of ms, and Inline graphic in units of Inline graphic.

Shifts in the activation, inactivation, and time constant curves alter the AHH system steady-state (Figure 8B). As Inline graphic increases, the steady-state transmembrane potential Inline graphic becomes more hyperpolarized, reaching a minimum of Inline graphic mV hyperpolarized below the baseline resting potential Inline graphic mV. As Inline graphic increases further, Inline graphic is gradual depolarized, approaching a maximum value of Inline graphic mV depolarized above Inline graphic. The steady-state sodium activation gate Inline graphic decreases and approaches 0 as Inline graphic increases. Despite Inline graphic becoming more hyperpolarized for small Inline graphic, the steady-state sodium inactivation gate Inline graphic also decreases, and then increases and approaches 1 for large Inline graphic. The steady-state potassium activation gate Inline graphic is also complex, first increasing then decreasing and approaching 0 as Inline graphic increases.

Mechanisms of conduction block

The influence of the HF stimulus parameter Inline graphic on the dynamics of the gating variables provides significant insight into the mechanism of conduction block in neurons and the various thresholds for repetitive firing and excitability (Inline graphic, Inline graphic, and Inline graphic) (Figure 8). For small Inline graphic (the critical value of Inline graphic for repetitive firing), the neuron can repetitively fire for some range of the applied current Inline graphic. As Inline graphic increases, the resting sodium channel activation gate Inline graphic and inactivation gate Inline graphic decrease and the resting potassium channel activation gate Inline graphic increases, which all drive the neuron towards being less prone to firing.

As Inline graphic increases for small Inline graphic (the critical value of Inline graphic for anodal stimulation), the time constant for sodium channel inactivation Inline graphic also decreases, that is, following a negative applied current pulse, Inline graphic will return to its resting value in a shorter amount of time. Combined with a more hyperpolarized resting transmembrane potential Inline graphic and decreasing Inline graphic, the threshold for anodal excitation Inline graphic increases dramatically as Inline graphic increases (Figure 7). Inline graphic when Inline graphic is at a minimum and Inline graphic has decreased by approximately a factor of 2.

As Inline graphic increases for Inline graphic (the critical value of Inline graphic for cathodal stimulation), the sodium activation curve Inline graphic is right shifted, and combined with a more hyperpolarized Inline graphic and decreasing Inline graphic, results in an increasing threshold for cathodal simulation Inline graphic (Figure 7). Inline graphic when Inline graphic mV and Inline graphic are at a minimum, Inline graphic and Inline graphic are at a maximum and gating dynamics are fast, that is, the time constants for sodium activation Inline graphic and inactivation Inline graphic are near 0. Rapid and persistent activation of the potassium current opposes the sodium current and prevents sufficient depolarization to reach the threshold for evoking an action potential. As Inline graphic increases for large Inline graphic, Inline graphic and Inline graphic decrease, as Inline graphic gradually transitions from hyperpolarized to depolarized, relative to Inline graphic. For very large Inline graphic, all three time constants are essentially equal to 0, such that the gating variable kinetics can be defined by instantaneous functions of Inline graphic. In this regime, the four-dimensional AHH model (Eqs. 12a–12h) is reasonably approximated by a one-dimensional system, for which a large amplitude excursion typical of an action potential is no longer possible. Indeed, for a system with a single stable steady-state (as is the case for large Inline graphic), all perturbations from the steady-state are followed by a gradual relaxation back to rest, as observed for large Inline graphic in Figure 7A and B (right panels).

The AHH model suggests that there are different mechanisms of conduction block, depending on the strength of the HF stimulus. For small Inline graphic, repetitive firing ceases due to decreased sodium channel activation (decreased Inline graphic) and availability (decreased Inline graphic) and increased potassium channel activation (increased Inline graphic). For intermediate Inline graphic, gating variable dynamics are fast (i.e., the time constants approach 0), and therefore eliciting a single action potential via anodal and cathodal excitation fails due to rapid sodium current inactivation, in addition to decreased sodium activation and increased potassium activation. For large Inline graphic, sodium and potassium currents are persistently de-activated (i.e., Inline graphic and Inline graphic), preventing an action potential from being evoked, and Inline graphic is depolarized due to the influence of the leak current.

In summary, the influence of the HF stimulation parameter Inline graphic on the properties of action potential firing in the AHH model is similar to that demonstrated in the AFHN model, however with differences in the influence on the resting potential and firing frequency. Further, simulation and analysis of the biophysically-detailed model provides insight into the mechanisms of conduction block. We determined three critical values for Inline graphic, which above the neuron cannot repetitively fire (Inline graphic), and an action potential cannot be evoked by cathodal (Inline graphic) or anodal (Inline graphic) stimulation. Below the critical values, we demonstrated that the thresholds for evoking repetitive firing or a single action potential increase as Inline graphic increases.

HF stimulation of an AHH neuronal network

Finally, we consider the influence of HF stimulation on a random network of 100 neurons, each with, on average, 10 connections, coupled via both excitatory and inhibitory synapses (see Methods for description of synaptic currents and network architecture). Following a single initial applied current pulse, in the absence of HF stimulation (i.e., Inline graphic), a rastergram shows that most neurons in the network fire repetitively (Figure 9A), while a few neurons remain quiescent due to the absence of incoming excitatory synaptic connections (arrows). The pseudo-electroencephalogram (pEEG, Eq. 32 ) becomes disorganized after an initial time period of synchronization following electrical activity initiation.

Figure 9. Electrical activity in a network of AHH model neurons.

Figure 9

(A) Rastergram of action potentials and the pseudo-electroencephalogram (pEEG). Arrows indicated quiescent neurons. Parameters: Inline graphic, Inline graphic. (B) pEEG (left, Eq. 32 ), the corresponding power spectrum (middle, value indicates the average neuron firing rate), and synchrony measure Inline graphic (right, Eq. 33 ), as functions of Inline graphic. Inline graphic in units of Inline graphic.

We plot the pEEG and the corresponding power spectrum for the same network as Inline graphic increases (Figure 9B). In general, the average neuron firing rate tends to decrease as Inline graphic increase, although though this trend is not strictly monotonic with Inline graphic (Figure 9B, middle panels). In general, decreased firing rate is associated with increased network synchronization (Figure 9B, right panels), illustrated by the narrowing of the dominant peak in the power spectrum and increase in the synchrony measure Inline graphic ( Eq. 33 ). When Inline graphic increases further above a critical value of Inline graphic, Inline graphic, network electrical activity ceases immediately following the initial initiation (Figure 9B, bottom panel, Inline graphic for this example).

The critical value Inline graphic for complete cessation of network electrical activity was highly dependent on the specific architecture of the network and the relative proportion of excitatory and inhibitory synaptic connections (see Methods). We determined Inline graphic as functions of Inline graphic, the probability that each synaptic connection was excitatory, and the specific network architecture. For each value of Inline graphic, Inline graphic different networks were randomly constructed with the same average connection properties but different connections. The mean value for Inline graphic, plus/minus one and two standard errors of the mean (SEM, standard deviation divided by Inline graphic) are shown as functions of Inline graphic (Figure 10). For all network architectures, Inline graphic for all networks with Inline graphic or Inline graphic, that is, the network does not exhibit persistent activity, even in the absence of HF stimulation. In this specific type of network, persistent network activity requires both excitatory and inhibitory synaptic connections, and indeed that a majority of synapses are inhibitory. For Inline graphic, Inline graphic is an inverted U-shaped function of Inline graphic, with a maximum near Inline graphic. In all networks, the values of Inline graphic are less than the single cell critical values for repetitive spiking Inline graphic, and cathodal and anodal stimulation, Inline graphic and Inline graphic, respectively.

Figure 10. HF stimulation of a network of AHH model neurons.

Figure 10

The critical value Inline graphic for repetitive activity in a AHH model neural network, as a function of the probability of excitatory synaptic connections Inline graphic. The mean Inline graphic (thick black) over Inline graphic simulations, Inline graphic 1 (solid black) and 2 (thin black) SEM (standard error of the mean, standard deviation of Inline graphic) are shown. Single cell critical values of Inline graphic (Inline graphic and Inline graphic) are identified (see text for description). Inline graphic in units of Inline graphic.

Discussion

Summary of main findings

In summary, we find that HF stimulation alters the dynamics of excitable cells and networks, resulting in conduction block and electrical quiescence. In a simplified excitable cell model, we identify analytical expressions for HF stimulation strength critical values for repetitive firing and evoking action potentials. In a biophysically-detailed neuronal model, we demonstrate that HF stimulation alters the dynamics of ionic current gating, leading to reduced cellular excitability and conduction block. HF stimulation of a neural network reduces the overall network activity and increases network synchronization, leading to network quiescence for a sufficiently large HF stimulus.

Relation to prior work

Previous studies have investigated the mechanisms underlying conduction block in neurons [8][14]. In these studies, two primary mechanisms for conduction block are posed: persistent potassium current activation of potassium opposing sodium current and preventing the neuron transmembrane potential from reaching a threshold; and reduced sodium channel availability due to a baseline depolarization of the transmembrane potential. It is noted in several studies that these mechanisms are not mutually exclusive and indeed both likely play a role, depending on the species and specific cell type. Our findings are generally consistent with these previous studies, in that we find persistent potassium current activation and reduced sodium channel availability (or de-inactivation). However, a multi-scale method approach permits us to identify how changes in ionic current gating result in different critical thresholds. Further, we are able identify the influence of HF stimulation on the gating variable kinetics, i.e. the gating variable steady-state activation, inactivation and time constant curves, as a significant and novel factor regulating conduction block.

The method of averaging has been previously used to investigate the influence of HF stimulation in the FHN model [17][19]. Cubero and colleagues previously demonstrated that HF stimulation can lead to cessation of repetitive firing above a critical threshold (a finding reproduced in the present study) [17]. Ratas and Pyragas determined conditions for which HF stimulation results in slowed and failed propagation. [18], [19]. Here, we extend the approach of these prior studies to demonstrate the influence of HF stimulation on the threshold for evoking a single action potential and additionally to investigate HF stimulation in a biophysically-detailed neuronal single-cell model and network. The FHN model is ideal for mechanistic studies, as the two-dimensional model enables phase-plane analysis and often permits analytical expressions for critical values. We demonstrate that many aspects of HF stimulation of FHN model neurons are similarly reproduced in the HH model, specifically qualitatively similar Inline graphic-Inline graphic parameter space for repetitive firing; influence of HF stimulation on the action potential threshold; and the existence of critical HF stimulation amplitudes above which neurons cannot repetitively firing or trigger a single action potential.

However, some important properties of HF stimulation of the AHH model are not qualitatively reproduced by the AFHN model, which paints a simpler picture for the mechanism of conduction block. In the AFHN model, the resting potential is gradually depolarized and the gating variable Inline graphic gradually increases as Inline graphic increases, suggesting that conduction block occurs to do the collective influence of transmembrane potential depolarization and a larger degree of refractoriness. In the AHH model, the resting potential is hyperpolarized for small Inline graphic and depolarized as Inline graphic increases. Additionally, the gating variable steady-state values and time constants are altered in a complex manner, such that conduction block is due to both altered refractoriness and the time-dependent dynamics of the refractory variables. The AFHN model also not does reproduce the influence of HF stimulation on firing frequency. In the AFHN model, the frequency increases as Inline graphic increases. However, in the AHH model, frequency decreases, likely due to the reduced sodium channel availability, i.e. the decrease in Inline graphic, that occurs as Inline graphic increases for small Inline graphic. Understanding how HF stimulation influences firing frequency is significant and necessary for optimizing HF stimulation therapy, as frequency plays a significant role in neural computing [27].

Physiological significance of findings

The term, high frequency stimulation, is often used in different clinical settings with different meanings. HF stimulation frequencies range several orders of magnitude from 100 Hz—typical of studies of deep brain stimulation to treat movement disorders such as Parkinson's disease and other neurological disorders such as epilepsy [28], [29]—to 40 kHz—including clinical applications such as pain mitigation and improved bladder voiding [8][10]. An inherent assumption in the derivation of the averaged excitable cell models is that the time scale of the HF stimulus is significantly shorter than the time scale of cellular dynamics. We show that the averaged and original HH model begin to agree when the HF stimulation frequency is near 5 kHz (Figure 5), consistent with Inline graphic ms (Inline graphic kHz) time-scale for sodium channel activation, and there is greater agreement as the HF frequency increases. This suggests that the method of averaging approach may not be strictly appropriate to the investigation of deep brain stimulation using lower frequency HF in the 100–200 Hz range but highly relevant to the study of peripheral nerve stimulation and clinical applications typically utilizing kilohertz-range HF stimulation. Recently, Weinberg and colleagues demonstrated that HF stimulation in the 100–200 Hz range could block electrical conduction in cardiac tissue, a novel approach to terminate arrhythmias [30], [31]. Since the time scale of cardiac dynamics is generally slower than neuronal dynamics, future work is necessary to determine the validity of the averaging method for investigation of the influence of HF stimulation in cardiac tissue.

In this study, we found that HF stimulation could prevent persistent network electrical activity at lower HF stimulation amplitudes necessary for single cell quiescence, which suggests network activity may cease due to HF stimulation sufficiently reducing excitability in a subset of neurons that prevents re-excitation throughout the network. We only considered network activity that persists (or ceases) following a single initial applied current at the simulation onset. The network response to a constant, repetitive, or random (Poissonian) applied current, in addition to the HF stimulation, may be significantly different. We additionally only consider sparsely connected random network architectures. In this study, we found that the specific network architecture was highly important in determining the response to HF stimulation, and thus it is reasonable to speculate that the network response in highly connected and/or directed neural networks could be different from our findings. Several studies of models including more detailed network architecture and specific cell types have suggested mechanisms underlying deep brain stimulation treatment using HF stimulation in the 100–200 Hz range. Rubin and Terman demonstrated that HF stimulation of the subthalamic nucleus can regularize globus pallidus firing and eliminate pathological thalamic rhythmicity [32]. Using a systems theoretic approach, Agarwal and Sarma demonstrate that HF deep brain stimulation improves reliability of thalamic relay [33]. As noted above, the averaging method may not be strictly appropriate in this frequency range. However, investigation of more physiological neural network architectures and cell types may suggest alternative deep brain stimulation therapies within a higher frequency (kilohertz) regime or provide insight into the role of specific network components with different responses to HF stimulation, as in the aforementioned studies.

In this study, we investigate a simple network with a random architecture and consider the influence of HF stimulation as function of the relative fraction of excitatory synaptic connections. We illustrate a general approach to study HF stimulation in a large neural network which does not require simulation of the HF stimulation term and thus does not require a prohibitively small simulation time step. Here, we consider HF stimulation in the context of only a few network parameters. However, neural networks can exhibit rich and complex dynamics, and much work has demonstrated that the local network architecture can have significant influence on global behavior, e.g., the small-world phenomenon [34][36], and network architecture will likely significantly influence our findings. Additional work is necessary to understand the influence of HF stimulation in the context of networks of varying degrees of connectivity and structure.

The critical values for repetitive firing and evoking action potentials are defined in terms of the HF stimulus frequency-to-amplitude ratio; that is, as the HF stimulus frequency increases, so must the HF stimulus amplitude for the same response. In a therapeutic device, it is ideal to minimize the amplitude of an applied stimulus, to minimize power consumption and mitigate safety issues for both the patient and device. Thus, determining the optimal frequency regime to minimize the amplitude for optimal HF stimulation is an important and practical issue. Future work will consider these complications, which must also include analysis of HF stimulation at frequencies approaching the same time scales as cellular dynamics.

Limitations

In most clinical settings of interest, HF stimulation is applied in the form of an external electrical field stimulus. In this study, we do not account for the influence of an external electrical field nor account for the spatial extent of the nerve axon. Such levels of details are significant for studies of local neural conduction block [8][14], as spatial gradients in the extracellular space create virtual electrodes resulting in non-uniform HF stimulation through the nerve axon [37]. It has been shown in multi-compartment models that somatic and axonal firing can become decoupled during 100–200 Hz HF stimulation, such that somatic quiescence does not necessarily preclude activation in neuronal processes [38]. Simulation studies in one-dimensional nerve axons have shown that as the amplitude of a kilohertz-range HF stimulation is increased, the system can transition from regimes of conduction block to rapid firing several times, such that a strict conduction block threshold is not clearly defined [39]. As such, non-uniform stimulation could lead to conduction block in one region of a neuron and rapid activation in another. Further studies of kilohertz-range HF stimulation in more spatially-detailed neuronal models are necessary to investigate these complex issues.

The HH model of the giant squid nerve axon is a classical model of an excitable cell, highly studied and well-characterized, and thus it was a reasonable biophysically-detailed ionic model to characterize the influence of HF stimulation using the method of averaging approach. However, several more detailed neuronal models relevant to mammalian physiology have been described, incorporating more detailed and multiple sodium, potassium, and calcium currents [40][42]. Indeed, the interaction between voltage-gated calcium channels and calcium-mediated synaptic transmission may be important for understanding the influence of HF stimulation on network activity and designing an optimal therapy and warrants further study.

Finally, the simulation results presented here are deterministic and do not account for stochastic fluctuations inherent in neuronal signaling at both the cellular and subcellular levels. Indeed, studies have shown that noise-induced firing can be enhanced by HF stimulation for sufficiently large noise levels, termed vibrational resonance [17], closely related to the well-known phenomenon of stochastic resonance [43]. Further work is necessary to investigate the influence of stochastic fluctuations on spiking in biophysically-detailed averaged neuronal models.

Methods

Derivation of AFHN model

We derive the AFHN model equations, following the approach in [19]. See [18], [19] for a more details. For HF stimuli with large frequencies Inline graphic, the period of HF oscillations is much less than the characteristic time scales of the FHN neuron. Therefore, we seek to eliminate the HF stimulus term Inline graphic from Eq. 1a and obtain an autonomous system which approximates the original system on the time scale of the FHN neuron. First, we change the variables in Eqs. 1a and 1b, substituting

graphic file with name pone.0081402.e532.jpg (13a)
graphic file with name pone.0081402.e533.jpg (13b)

and derive the following equations for Inline graphic and Inline graphic:

graphic file with name pone.0081402.e536.jpg (14a)
graphic file with name pone.0081402.e537.jpg

Mathematically, we are interested in the limit Inline graphic, for a fixed Inline graphic. By rescaling time Inline graphic, we can transform the system to

graphic file with name pone.0081402.e541.jpg (15a)
graphic file with name pone.0081402.e542.jpg (15b)

The variables Inline graphic and Inline graphic vary slowly relative to the periodic function Inline graphic, due to the small parameter Inline graphic. According to the method of averaging [15], an approximate solution to the system can be obtained by averaging over the fast periodic function, and the averaged variables Inline graphic and Inline graphic satisfy the following ODEs:

graphic file with name pone.0081402.e549.jpg (16a)
graphic file with name pone.0081402.e550.jpg (16b)

After calculating the integrals and returning to the original time scale, the averaged system is as given above in Eqs. 2a and 2b. Importantly, we note that the assumption implicit in stating Eq. 13a, specifically that the original system voltage Inline graphic can be expressed as the sum of slow varying Inline graphic and high frequency term Inline graphic, underlies an important conclusion of the method of averaging theorem [44]: an equilibrium point in the averaged system (e.g., rest or depolarization block) corresponds to a periodic solution in the original system (due to the additional HF term superimposed on top of the averaged system equilibrium). Similarly, a periodic orbit in the averaged system (e.g., repetitive firing) corresponds to a more complex oscillation or tori, observed in the Inline graphic traces of the FHN and AFHN models in Figure 1.

Firing frequency and amplitude in the AFHN model

Following a similar approach as described in [24], expressions for the firing frequency and action potential amplitude in the AFHN model are derived as follows. Recall from Eqs. 2a and 2b that the Inline graphic-nullcline Inline graphic is cubic in shape. Therefore, over the finite range of values for Inline graphic, there are three solutions of the equation Inline graphic, which we can denote by Inline graphic, Inline graphic, and Inline graphic (see Figure 11). The minimal value of Inline graphic for which Inline graphic exists is Inline graphic, the maximal value of Inline graphic for which Inline graphic exists is Inline graphic, both of which are functions of Inline graphic and Inline graphic. Inline graphic, the left and right branches, are termed the stable branches of the Inline graphic-nullcline, and Inline graphic, the middle branch, is termed the unstable branch, because, in the limit that Inline graphic is much faster than Inline graphic, a steady-state located on the (un)stable branch is (un)stable.

Figure 11. Period of a stable limit cycle in the AFHN model.

Figure 11

For Inline graphic, the period Inline graphic of the stable limit cycle is approximately the sum of the time required to traverse the two stable branches of the Inline graphic-nullcline, Inline graphic, denoted by the points Inline graphic and Inline graphic, respectively. As Inline graphic increases, the amplitude and period of the stable limit cycle, indicated by the second set of points labeled Inline graphic, both decrease. See text for description of other variables in figure. Parameters: Inline graphic.

The locations of (Inline graphic and Inline graphic are given by the local minimum and maximum, respectively, of the Inline graphic-nullcline, given by

graphic file with name pone.0081402.e587.jpg

and

graphic file with name pone.0081402.e588.jpg

Because Inline graphic is fast compared to Inline graphic, Inline graphic rapidly moves between stable branches of the Inline graphic-nullcline, Inline graphic. We can approximate the period of an oscillation Inline graphic by the time required to travel along the two stable branches Inline graphic

graphic file with name pone.0081402.e596.jpg (17)

where points Inline graphic-Inline graphic are indiciated in Figure 11. Along the stable branches, the dynamics of Inline graphic are determined by

graphic file with name pone.0081402.e600.jpg (18)

and, therefore, Eq. 17 is equivalently given by

graphic file with name pone.0081402.e601.jpg (19)

Note that both terms of the integral are positive, since Inline graphic and Inline graphic over the range of values Inline graphic. The frequency of oscillations is given by Inline graphic.

From Figure 11, we observe that the amplitude of the limit cycle—and thus, of the action potential—is given by Inline graphic, the difference between the values of Inline graphic at points Inline graphic and Inline graphic, respectively. Inline graphic is the non-repeated root solution of

graphic file with name pone.0081402.e611.jpg (20)

and Inline graphic is similarly the non-repeated root solution of

graphic file with name pone.0081402.e613.jpg (21)

Strength-duration curve in the AFHN model

An approximation for the strength-duration for the AFHN model is derived as follows. Near the steady-state Inline graphic, the dynamics of Inline graphic can be approximated by

graphic file with name pone.0081402.e616.jpg (22)

where the dot indicates differentiation with respect to time and the prime indicates differentiation with respect to Inline graphic, such that

graphic file with name pone.0081402.e618.jpg

and Inline graphic. Solving Eq. 22 with the initial condition Inline graphic,

graphic file with name pone.0081402.e621.jpg (23)

We set Inline graphic, the Inline graphic threshold for eliciting an action potential, and after rearranging, we arrive at the strength-duration curve relationship in Eq. 8.

To evaluate Eq. 8, we must determine the dependence of the threshold Inline graphic on Inline graphic. Many studies have discussed the absence of a well-defined threshold in the FHN model [45][47]. Izhikevich notes that canard trajectories following the repelling slow manifold provide the best approximation to the excitability threshold [45]. Recent work has show that this manifold is well approximated by inflection sets (regions of flow lines with zero curvature in the phase plane) [46]. For simplicity, we approximate the threshold Inline graphic by the middle solution of

graphic file with name pone.0081402.e627.jpg (24)

which, as shown in Figure 12 (left panel), does reasonably well-approximate the action potential threshold. We are interested in the threshold Inline graphic from a non-stimulated state, i.e., Inline graphic. Therefore, Eq. 41 can be written as

graphic file with name pone.0081402.e630.jpg (25a)
graphic file with name pone.0081402.e631.jpg (25b)

Figure 12. Sub- and super-threshold brief stimuli in the AFHN model.

Figure 12

(Left) For Inline graphic, Inline graphic indicates the threshold for evoking an action potential. Two trajectories starting near Inline graphic are identified by arrows: (1, left arrow) when the initial condition Inline graphic, Inline graphic returns quickly to the resting potential Inline graphic; (2, right arrow) when Inline graphic, an action potential is evoked: the system follows a counterclockwise trajectory, quickly approaching the Inline graphic-nullcline, following the right stable branch until reaching the right knee, quickly reaching the left stable branch, and then returning to rest. (Right) When Inline graphic, Inline graphic, the critical value above which an action potential cannot be evoked by a brief perturbation from the steady-state.

where Eq. 25b is implied, since Inline graphic by construction is a solution of Eq. 25a. Matching coefficients, Inline graphic and Inline graphic, and using the quadratic formula, Inline graphic is given by

graphic file with name pone.0081402.e646.jpg (26)

where the dependence of Inline graphic on Inline graphic is due to the dependence of Inline graphic and Inline graphic on Inline graphic. Further, from the definition of Inline graphic, if the terms under the radical equal 0, Inline graphic is equal to a critical value, which we will call Inline graphic—above which the threshold is ill-defined, that is, the system cannot be excited by a brief perturbation from the steady-state. Using Eq. 3b, Inline graphic is the real solution of the cubic equation

graphic file with name pone.0081402.e656.jpg (27)

Note that Inline graphic does not depend on Inline graphic, only the system parameters. Using Eqs. 25a and 27, for all values of Inline graphic, if

graphic file with name pone.0081402.e660.jpg

and therefore,

graphic file with name pone.0081402.e661.jpg (28)

the system cannot be excited by a brief stimulus pulse, illustrated in Figure 12 (right panel).

Hodgkin-Huxley model equations

The equations governing the dynamics of the gating variables Inline graphic and Inline graphic are given by

graphic file with name pone.0081402.e664.jpg
graphic file with name pone.0081402.e665.jpg
graphic file with name pone.0081402.e666.jpg

where Inline graphic is the shifted transmembrane potential, in which the resting potential Inline graphic has been subtracted. The standard parameters for the HH model are given in Table 1.

Table 1. Hodgkin-Huxley model current parameters.

Parameter Definition Units Value
Inline graphic maximum NaInline graphic current conductance mS/cmInline graphic 120
Inline graphic maximum KInline graphic current conductance mS/cmInline graphic 36
Inline graphic maximum leak current conductance mS/cmInline graphic 0.3
Inline graphic NaInline graphic current reversal potential mV 115
Inline graphic KInline graphic current reversal potential mV –12
Inline graphic leak current reversal potential mV 10.6
Inline graphic resting potential mV –80
Inline graphic capacitance Inline graphic 1

Parameters for ionic currents in the HH model.

Neuronal network model and architecture

A network of Inline graphic AHH neurons are simulated by adding a synaptic current to the AHH model, such that the Inline graphic dynamics of the Inline graphic-th neuron are governed by the following equation:

graphic file with name pone.0081402.e688.jpg (29)

where

graphic file with name pone.0081402.e689.jpg

and Inline graphic and Inline graphic are the set of presynaptic neurons with connections to neuron Inline graphic, with excitatory and inhibitory, respectively, synapses. Inline graphic is the averaged gating variable for the postsynaptic conductance, and assumed to be an instantaneous, sigmoidal function of the presynaptic cell potential with a threshold Inline graphic [48], that is

graphic file with name pone.0081402.e695.jpg (30)

where

graphic file with name pone.0081402.e696.jpg (31)

Parameters are in Table 2. The specific synaptic connections between neurons are determined randomly, as follows. The number of presynaptic connections to the Inline graphic-th neuron is drawn from a Gaussian distribution with mean Inline graphic and standard deviation Inline graphic, rounded to the nearest whole number. The presynaptic neuron indices Inline graphic are chosen at random. The type of each synapse, excitatory or inhibitory, is determined at random, such that the probability of an excitatory synapse is Inline graphic. Electrical activity is evoked in the neural network by applying a 200-Inline graphic, 0.1-ms applied current in 50 randomly selected neurons at time Inline graphic.

Table 2. Synaptic current parameters.

Parameter Definition Units Value
Inline graphic maximum synaptic conductance mS/cmInline graphic 0.3
Inline graphic excitatory synapse reversal potential mV 80
Inline graphic inhibitory synapse reversal potential mV –12
Inline graphic presynaptic cell potential threshold mV 50
Inline graphic threshold parameter mV 2

Parameters for excitatory and inhibitory synaptic currents in a network of AHH model neurons.

The collective activity of the neural network can be represented by the pseudo-electroencephalogram (pEEG) [49], given by Inline graphic,

graphic file with name pone.0081402.e711.jpg (32)

the transmembrane potential averaged over all neurons. The frequency-domain representation of the pEEG is computed by the Fast Fourier Transform.

The synchrony of the electrical activity in the network is given by the synchrony measure Inline graphic [50],

graphic file with name pone.0081402.e713.jpg (33)

where the variance of the time fluctuations of the average transmembrane potential, Inline graphic,

graphic file with name pone.0081402.e715.jpg

the variance of the time fluctuations of the individual transmembrane potentials Inline graphic,

graphic file with name pone.0081402.e717.jpg

and Inline graphic denotes time-averaging over the duration of the simulation Inline graphic. Note that if Inline graphic are identical for all Inline graphic, then Inline graphic.

Numerical simulations

All numerical simulations were performed in MATLAB. For simulations of the AHH model, the modified gating variable rate functions (Eqs. 12e–12f) were pre-calculated for a given value of Inline graphic for Inline graphic mV (Inline graphic mV), and values were linearly interpolated from look-up tables during simulations.

Acknowledgments

We thank the anonymous reviewers for their comments and suggestions that have greatly improved the discussion of the study results.

Funding Statement

The work was supported by the Biomathematics Initiative and the Dean of the College of Arts and Sciences at The College of William and Mary. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

References

  • 1. Anderson WS, Kudela P, Weinberg S, Bergey GK, Franaszczuk PJ (2009) Phase-dependent stimulation effects on bursting activity in a neural network cortical simulation. Epilepsy Research 84: 42–55. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 2. Zipes DP, Fischer J, King RM, deB NA, Jolly WW (1975) Termination of ventricular fibrillation in dogs by depolarizing a critical amount of myocardium. The American Journal of Cardiology 36: 37–44. [DOI] [PubMed] [Google Scholar]
  • 3. Nitsche MA, Cohen LG, Wassermann EM, Priori A, Lang N, et al. (2008) Transcranial direct current stimulation: State of the art 2008. Brain Stimulation 1: 206–223. [DOI] [PubMed] [Google Scholar]
  • 4. Tokano T, Bach D, Chang J, Davis J, Souza JJ, et al. (1998) Effect of ventricular shock strength on cardiac hemodynamics. Journal of Cardiovascular Electrophysiology 9: 791–797. [DOI] [PubMed] [Google Scholar]
  • 5. Tanner JA (1962) Reversible blocking of nerve conduction by alternating-current excitation. Nature 195: 712–713. [DOI] [PubMed] [Google Scholar]
  • 6. Boger A, Bhadra N, Gustafson KJ (2008) Bladder voiding by combined high frequency electrical pudendal nerve block and sacral root stimulation. Neurourology and Urodynamics 27: 435–439. [DOI] [PubMed] [Google Scholar]
  • 7. Boger AS, Bhadra N, Gustafson KJ (2012) High frequency sacral root nerve block allows bladder voiding. Neurourology and Urodynamics 31: 677–682. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8. Bhadra N, Kilgore KL (2004) High-frequency nerve conduction block. Conference proceedings: Annual International Conference of the IEEE Engineering in Medicine and Biology Society IEEE Engineering in Medicine and Biology Society Conference 7: 4729–4732. [DOI] [PubMed] [Google Scholar]
  • 9. Bhadra N, Lahowetz EA, Foldes ST, Kilgore KL (2007) Simulation of high-frequency sinusoidal electrical block of mammalian myelinated axons. Journal of Computational Neuroscience 22: 313–326. [DOI] [PubMed] [Google Scholar]
  • 10. Ackermann DM, Foldes EL, Bhadra N, Kilgore KL (2010) Conduction block of peripheral nerve using high-frequency alternating currents delivered through an intrafascicular electrode. Muscle & Nerve 41: 117–119. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 11. Kilgore KL, Bhadra N (2004) Nerve conduction block utilising high-frequency alternating current. Medical & Biological Engineering & Computing 42: 394–406. [DOI] [PubMed] [Google Scholar]
  • 12. Zhang X, Roppolo JR, de Groat WC, Tai C Mechanism of Nerve Conduction Block Induced by High-Frequency Biphasic Electrical Currents. IEEE Transactions on Biomedical Engineering 53: 2445–2454. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13. Liu H, Roppolo JR, de Groat WC, Tai C (2009) The role of slow potassium current in nerve conduction block induced by high-frequency biphasic electrical current. IEEE Transactions on Biomedical Engineering 56: 137–146. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14. Tai C, de Groat WC, Roppolo JR (2005) Simulation of nerve block by high-frequency sinusoidal electrical current based on the Hodgkin-Huxley model. IEEE transactions on neural systems and rehabilitation engineering : a publication of the IEEE Engineering in Medicine and Biology Society 13: 415–422. [DOI] [PubMed] [Google Scholar]
  • 15.Sanders J, Verhulst F, Murdock JA (2007) Averaging methods in nonlinear dynamical systems. Berlin: Springer.
  • 16. Fitzhugh R (1961) Impulses and Physiological States in Theoretical Models of Nerve Membrane. Biophysical Journal 1: 445–466. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 17. Cubero D, Baltanás JP, Casado-Pascual J (2006) High-frequency effects in the FitzHugh-Nagumo neuron model. Physical Review E 73: 061102. [DOI] [PubMed] [Google Scholar]
  • 18. Ratas I, Pyragas K (2011) Effect of high-frequency stimulation on nerve pulse propagation in the FitzHugh–Nagumo model. Nonlinear Dynamics 67: 2899–2908. [DOI] [PubMed] [Google Scholar]
  • 19. Ratas I, Pyragas K (2012) Pulse propagation and failure in the discrete FitzHugh-Nagumo model subject to high-frequency stimulation. Physical Review E 86: 046211. [DOI] [PubMed] [Google Scholar]
  • 20. Rinzel J (1985) Excitation dynamics: insights from simplified membrane models. Federation Proceedings 44: 2944–2946. [PubMed] [Google Scholar]
  • 21.Hodgkin AL, Huxley AF (1952) A quantitative description of membrane current and its application to conduction and excitation in nerve. The Journal of Physiology. [DOI] [PMC free article] [PubMed]
  • 22.Strogatz SH (2008) Nonlinear Dynamics and Chaos. With Applications to Physics, Biology, Chemistry, and Engineering. Westview Press.
  • 23.Koch C (2004) Biophysics of Computation. Information Processing in Single Neurons. Oxford University Press.
  • 24.Keener JP, Sneyd J (2009) Mathematical Physiology. I: Cellular Physiology. Springer.
  • 25.Fall CP, Marland ES, Wagner JM, Tyson JJ (2002) Computational Cell Biology. Springer.
  • 26.Plonsey R, Barr RC (2007) Bioelectricity. A Quantitative Approach. Springer.
  • 27. Brown EN, Kass RE, Mitra PP (2004) Multiple neural spike train data analysis: state-of-the-art and future challenges. Nature Neuroscience 7: 456–461. [DOI] [PubMed] [Google Scholar]
  • 28. Benabid AL, Koudsie A, Benazzouz A, Vercueil L, Fraix V, et al. (2001) Deep brain stimulation of the corpus luysi (subthalamic nucleus) and other targets in Parkinson's disease. Extension to new indications such as dystonia and epilepsy. Journal of Neurology 248 Suppl 3III37–47. [DOI] [PubMed] [Google Scholar]
  • 29. Loddenkemper T, Pan A, Neme S, Baker KB, Rezai AR, et al. (2001) Deep brain stimulation in epilepsy. Journal of Clinical Neurophysiology 18: 514–532. [DOI] [PubMed] [Google Scholar]
  • 30. Tandri H, Weinberg SH, Chang KC, Zhu R, Trayanova NA, et al. (2011) Reversible Cardiac Conduction Block and Defibrillation with High-Frequency Electric Field. Science Translational Medicine 3: 102ra96–102ra96. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 31. Weinberg SH, Chang KC, Zhu R, Tandri H, Berger RD, et al. (2013) Defibrillation success with high frequency electric fields is related to degree and location of conduction block. Heart Rhythm 10: 740–748. [DOI] [PubMed] [Google Scholar]
  • 32. Rubin JE, Terman D (2004) High frequency stimulation of the subthalamic nucleus eliminates pathological thalamic rhythmicity in a computational model. Journal of Computational Neuroscience 16: 211–235. [DOI] [PubMed] [Google Scholar]
  • 33.Agarwal R, Sarma SV (2012) Performance Limitations of Relay Neurons. PLoS Computational Biology : e1002626. [DOI] [PMC free article] [PubMed]
  • 34. Watts DJ, Strogatz SH (1998) Collective dynamics of `small-world'networks. Nature 393: 440–442. [DOI] [PubMed] [Google Scholar]
  • 35. Gao Z, Hu B, Hu G (2001) Stochastic resonance of small-world networks. Physical Review E 65: 016209. [DOI] [PubMed] [Google Scholar]
  • 36. Vasalou C, Herzog ED, Henson MA (2009) Small-world network models of intercellular coupling predict enhanced synchronization in the suprachiasmatic nucleus. Journal of Biological Rhythms 24: 243–254. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 37. Knisley SB, Hill BC, Ideker RE (1994) Virtual electrode effects in myocardial fibers. Biophysical Journal 66: 719–728. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38. McIntyre CC, Grill WM, Sherman DL, Thakor NV (2004) Cellular effects of deep brain stimulation: model-based analysis of activation and inhibition. Journal of Neurophysiology 91: 1457–1469. [DOI] [PubMed] [Google Scholar]
  • 39. Tai C, de Groat WC, Roppolo JR (2005) Simulation analysis of conduction block in unmyelinated axons induced by high-frequency biphasic electrical currents. IEEE Transactions on Biomedical Engineering 52: 1323–1332. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 40. Schwarz JR, Reid G, Bostock H (1995) Action potentials and membrane currents in the human node of Ranvier. Pugers Archiv : European Journal of Physiology 430: 283–292. [DOI] [PubMed] [Google Scholar]
  • 41. Wang XJ (1998) Calcium coding and adaptive temporal computation in cortical pyramidal neurons. Journal of Neurophysiology 79: 1549–1566. [DOI] [PubMed] [Google Scholar]
  • 42. McIntyre CC, Richardson AG, Grill WM (2002) Modeling the excitability of mammalian nerve fibers: inuence of afterpotentials on the recovery cycle. Journal of Neurophysiology 87: 995–1006. [DOI] [PubMed] [Google Scholar]
  • 43. Longtin A (1997) Autonomous stochastic resonance in bursting neurons. Physical Review E 55: 868–876. [Google Scholar]
  • 44.Guckenheimer J, Holmes P (1983) Nonlinear Oscillations, Dynamical Systems and Bifurcations of Vector Fields. Springer.
  • 45.Izhikevich EM (2007) Dynamical Systems in Neuroscience: The Geometry of Excitability and Bursting. Cambridge, MA: MIT Press.
  • 46. Desroches M, Krupa M, Rodrigues S (2013) Inection, canards and excitability threshold in neuronal models. Journal of Mathematical Biology 67: 989–1017. [DOI] [PubMed] [Google Scholar]
  • 47. Mitry J, McCarthy M, Kopell N, Wechselberger M (2013) Excitable neurons, firing threshold manifolds and canards. Journal of Mathematical Neuroscience 3: 12. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48. Wang XJ, Rinzel J (1992) Alternating and Synchronous Rhythms in Reciprocally Inhibitory Model Neurons. Neural Computation 4: 84–97. [Google Scholar]
  • 49. Weinberg S, Iravanian S, Tung L (2008) Representation of collective electrical behavior of cardiac cell sheets. Biophysical Journal 95: 1138–1150. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 50. Golomb D, Rinzel J (1993) Dynamics of globally coupled inhibitory neurons with heterogeneity. Physical Review E 48: 4810–4814. [DOI] [PubMed] [Google Scholar]

Articles from PLoS ONE are provided here courtesy of PLOS

RESOURCES