Skip to main content
PLOS Computational Biology logoLink to PLOS Computational Biology
. 2012 Apr 12;8(4):e1002478. doi: 10.1371/journal.pcbi.1002478

Impact of Adaptation Currents on Synchronization of Coupled Exponential Integrate-and-Fire Neurons

Josef Ladenbauer 1,2,*, Moritz Augustin 1, LieJune Shiau 3, Klaus Obermayer 1,2
Editor: Olaf Sporns4
PMCID: PMC3325187  PMID: 22511861

Abstract

The ability of spiking neurons to synchronize their activity in a network depends on the response behavior of these neurons as quantified by the phase response curve (PRC) and on coupling properties. The PRC characterizes the effects of transient inputs on spike timing and can be measured experimentally. Here we use the adaptive exponential integrate-and-fire (aEIF) neuron model to determine how subthreshold and spike-triggered slow adaptation currents shape the PRC. Based on that, we predict how synchrony and phase locked states of coupled neurons change in presence of synaptic delays and unequal coupling strengths. We find that increased subthreshold adaptation currents cause a transition of the PRC from only phase advances to phase advances and delays in response to excitatory perturbations. Increased spike-triggered adaptation currents on the other hand predominantly skew the PRC to the right. Both adaptation induced changes of the PRC are modulated by spike frequency, being more prominent at lower frequencies. Applying phase reduction theory, we show that subthreshold adaptation stabilizes synchrony for pairs of coupled excitatory neurons, while spike-triggered adaptation causes locking with a small phase difference, as long as synaptic heterogeneities are negligible. For inhibitory pairs synchrony is stable and robust against conduction delays, and adaptation can mediate bistability of in-phase and anti-phase locking. We further demonstrate that stable synchrony and bistable in/anti-phase locking of pairs carry over to synchronization and clustering of larger networks. The effects of adaptation in aEIF neurons on PRCs and network dynamics qualitatively reflect those of biophysical adaptation currents in detailed Hodgkin-Huxley-based neurons, which underscores the utility of the aEIF model for investigating the dynamical behavior of networks. Our results suggest neuronal spike frequency adaptation as a mechanism synchronizing low frequency oscillations in local excitatory networks, but indicate that inhibition rather than excitation generates coherent rhythms at higher frequencies.

Author Summary

Synchronization of neuronal spiking in the brain is related to cognitive functions, such as perception, attention, and memory. It is therefore important to determine which properties of neurons influence their collective behavior in a network and to understand how. A prominent feature of many cortical neurons is spike frequency adaptation, which is caused by slow transmembrane currents. We investigated how these adaptation currents affect the synchronization tendency of coupled model neurons. Using the efficient adaptive exponential integrate-and-fire (aEIF) model and a biophysically detailed neuron model for validation, we found that increased adaptation currents promote synchronization of coupled excitatory neurons at lower spike frequencies, as long as the conduction delays between the neurons are negligible. Inhibitory neurons on the other hand synchronize in presence of conduction delays, with or without adaptation currents. Our results emphasize the utility of the aEIF model for computational studies of neuronal network dynamics. We conclude that adaptation currents provide a mechanism to generate low frequency oscillations in local populations of excitatory neurons, while faster rhythms seem to be caused by inhibition rather than excitation.

Introduction

Synchronized oscillating neural activity has been shown to be involved in a variety of cognitive functions [1], [2] such as multisensory integration [3], [4], conscious perception [5],[6], selective attention [7], and memory [9], [10], as well as in pathological states including Parkinson's disease [11], schizophrenia [12], and epilepsy [13]. These observations have led to a great interest in understanding the mechanisms of neuronal synchronization, how synchronous oscillations are initiated, maintained, and destabilized.

The phase response curve (PRC) provides a powerful tool to study neuronal synchronization [14]. The PRC is an experimentally obtainable measure that characterizes the effects of transient inputs to a periodically spiking neuron on the timing of its subsequent spike. PRC based techniques have been applied widely to analyze rhythms of neuronal populations and have yielded valuable insights into, for example, motor pattern generation [15], the hippocampal theta rhythm [16], and memory retrieval [10]. The shape of the PRC is strongly affected by ionic currents that mediate spike frequency adaptation (SFA) [17], [18], a prominent feature of neuronal dynamics shown by a decrease in instantaneous spike rate during a sustained current injection [19][21]. These adaptation currents modify the PRC in distinct ways, depending on whether they operate near rest or during the spike [18]. Using biophysical neuron models, it has been shown that a low threshold outward current, such as the muscarinic voltage-dependent Inline graphic-current (Inline graphic), can produce a type II PRC, characterized by phase advances and delays in response to excitatory stimuli, in contrast to only phase advances, defining a type I PRC. A high threshold outward current on the other hand, such as the Inline graphic-dependent afterhyperpolarization Inline graphic-current (Inline graphic), flattens the PRC at early phases and skews its peak towards the end of the period [18], [22], [23]. Both changes of the PRC indicate an increased propensity for synchronization of coupled excitatory cells [22], and can be controlled selectively through cholinergic neuromodulation. In particular, Inline graphic and Inline graphic are reduced by acetylcholine with different sensitivities, which modifies the PRC shape [23][25].

In recent years substantial efforts have been exerted to develop single neuron models of reduced complexity that can reproduce a large repertoire of observed neuronal behavior, while being computationally less demanding and, more importantly, easier to understand and analyze than detailed biophysical models. Two-dimensional variants of the leaky integrate-and-fire neuron model have been proposed which take into consideration an adaptation mechanism that is spike triggered [26] or subthreshold, capturing resonance properties [27], as well as an improved description of spike initiation by an exponential term [28]. A popular example is the adaptive exponential leaky integrate-and-fire (aEIF) model by Brette and Gerstner [29], [30]. The aEIF model is similar to the two-variable model of Izhikevich [31], such that both models include a sub-threshold as well as a spike-triggered adaptation component in one adaptation current. The advantages of the aEIF model, as opposed to the Izhikevich model, are the exponential description of spike initiation instead of a quadratic nonlinearity, and more importantly, that its parameters are of physiological relevance. Despite their simplicity, these two models (aEIF and Izhikevich) can capture a broad range of neuronal dynamics [32][34] which renders them appropriate for application in large-scale network models [35], [36]. Furthermore, the aEIF model has been successfully fit to Hodgkin-Huxley-type neurons as well as to recordings from cortical neurons [29], [37], [38]. Since lately, this model is also implemented in neuromorphic hardware systems [39].

Because of subthreshold and spike-triggered contributions to the adaptation current, the aEIF model exhibits a rich dynamical structure [33], and can be tuned to reproduce the behavior of all major classes of neurons, as defined electrophysiologically in vitro [34]. Here, we use the aEIF model to study the influence of adaptation on network dynamics, particularly synchronization and phase locking, taking into account conduction delays and unequal synaptic strengths. First, we show how both subthreshold and spike-triggered adaptation affect the PRC as a function of spike frequency. Then, we apply phase reduction theory, assuming weak coupling, to explain how the changes in phase response behavior determine phase locking of neuronal pairs, considering conduction delays and heterogeneous synaptic strengths. We next present numerical simulations of networks which support the findings from our analysis of phase locking in neuronal pairs, and show their robustness against heterogeneities. Finally, to validate the biophysical implication of the adaptation parameters in the aEIF model, we relate and compare the results using this model to the effects of Inline graphic and Inline graphic on synchronization in Hodgkin-Huxley-type conductance based neurons. Thereby, we demonstrate that the basic description of an adaptation current in the low-dimensional aEIF model suffices to capture the characteristic changes of PRCs, and consequently the effects on phase locking and network behavior, mediated by biophysical adaptation currents in a complex neuron model. The aEIF model thus represents a useful and efficient tool to examine the dynamical behavior of neuronal networks.

Methods

aEIF neuron model

The aEIF model consists of two differential equations and a reset condition,

graphic file with name pcbi.1002478.e010.jpg (1)
graphic file with name pcbi.1002478.e011.jpg (2)
graphic file with name pcbi.1002478.e012.jpg (3)

The first equation (1) is the membrane equation, where the capacitive current through the membrane with capacitance Inline graphic equals the sum of ionic currents, the adaptation current Inline graphic, and the input current Inline graphic. The ionic currents are given by an ohmic leak current, determined by the leak conductance Inline graphic and the leak reversal potential Inline graphic, and a Inline graphic-current which is responsible for the generation of spikes. The Inline graphic-current is approximated by the exponential term, where Inline graphic is the threshold slope factor and Inline graphic is the threshold potential, assuming that the activation of Inline graphic-channels is instantaneous and neglecting their inactivation [28]. The membrane time constant is Inline graphic. When Inline graphic drives the membrane potential Inline graphic beyond Inline graphic, the exponential term actuates a positive feedback and leads to a spike, which is said to occur at the time when Inline graphic diverges towards infinity. In practice, integration of the model equations is stopped when Inline graphic reaches a finite “cutoff” value Inline graphic, and Inline graphic is reset to Inline graphic (3). Equation (2) governs the dynamics of Inline graphic, with the adaptation time constant Inline graphic. Inline graphic quantifies a conductance that mediates subthreshold adaptation. Spike-triggered adaptation is included through the increment Inline graphic (3).

The dynamics of the model relevant to our study is outlined as follows. When the input current Inline graphic to the neuron at rest is slowly increased, at some critical current the resting state is destabilized which leads to repetitive spiking for large regions in parameter space [34]. This onset of spiking corresponds to a saddle-node (SN) bifurcation if Inline graphic, and a subcritical Andronov-Hopf (AH) bifurcation if Inline graphic at current values Inline graphic and Inline graphic respectively which can be calculated explicitly [33]. In the former case a stable fixed point (the neuronal resting state) and an unstable fixed point (the saddle) merge and disappear, in the latter case the stable fixed point becomes unstable before merging with the saddle. In the limiting case Inline graphic, both bifurcations (SN and AH) meet and the system undergoes a Bogdanov-Takens (BT) bifurcation. The sets of points with Inline graphic and Inline graphic are called Inline graphic-nullcline and Inline graphic-nullcline, respectively. It is obvious that all fixed points in the two-dimensional state space can be identified as intersections of these two nullclines. Spiking can occur at a constant input current lower than Inline graphic or Inline graphic depending on whether the sequence of reset points lies exterior to the basin of attraction of the stable fixed point. This means, the system just below the bifurcation current can be bistable; periodic spiking and constant membrane potential are possible at the same input current. Thus, periodic spiking trajectories do not necessarily emerge from a SN or AH bifurcation. We determined the lowest input current that produces repetitive spiking (the rheobase current, Inline graphic) numerically by delivering long-lasting rectangular current pulses to the model neurons at rest. Note that in general Inline graphic depends on Inline graphic, such that in case of bistability, Inline graphic can be reduced by decreasing Inline graphic [33].

We selected realistic values for the model parameters (Inline graphic, Inline graphic, Inline graphic, Inline graphic, Inline graphic, Inline graphic, Inline graphic) and varied the adaptation parameters within reasonable ranges (Inline graphic, Inline graphic). All model parametrizations in this study lead to periodic spiking for sufficiently large Inline graphic, possibly including transient adaptation. Parameter regions which lead to bursting and irregular spiking [34] are not considered in this study. Inline graphic was set to Inline graphic, since from this value, even without an input current, Inline graphic would rise to a typical peak value of the action potential (Inline graphic) within less than Inline graphic while Inline graphic essentially does not change due to its large time constant. Only in Fig. 1A–C we used Inline graphic to demonstrate the steep increase of Inline graphic past Inline graphic.

Figure 1. Influence of adaptation on spiking behavior and Inline graphic-Inline graphic curves of aEIF neurons.

Figure 1

A–C: Membrane potential Inline graphic and adaptation current Inline graphic of aEIF neurons without adaptation (A), with subthreshold adaptation (B) and with spike-triggered adaptation (C), in response to step currents Inline graphic. To demonstrate the steep increase of Inline graphic past Inline graphic, Inline graphic was set to Inline graphic. Note that the neuron in C has not reached its steady state frequency by the end of the rectangular current pulse. D,E: Inline graphic-Inline graphic relationships for Inline graphic, Inline graphic (black – blue, D) and Inline graphic, Inline graphic (black – red, E). All other model parameters used for this figure are provided in the Methods section.

Traub neuron model

In order to compare the effects of adaptation in the aEIF model with those of Inline graphic and Inline graphic in a biophysically detailed model and with previously published results [18], we used a variant of the conductance based neuron model described by Traub et al. [41]. The current-balance equation of this model is given by

graphic file with name pcbi.1002478.e089.jpg (4)

where the ionic currents consist of a leak current Inline graphic, a Inline graphic-current Inline graphic, a delayed rectifying Inline graphic-current Inline graphic, a high-threshold Inline graphic-current Inline graphic with Inline graphic, and the slow Inline graphic-currents Inline graphic, and Inline graphic. The gating variables Inline graphic, Inline graphic and Inline graphic satisfy first-order kinetics

graphic file with name pcbi.1002478.e104.jpg (5)
graphic file with name pcbi.1002478.e105.jpg (6)
graphic file with name pcbi.1002478.e106.jpg (7)

with Inline graphic and Inline graphic, Inline graphic and Inline graphic, Inline graphic and Inline graphic. The fraction Inline graphic of open Inline graphic-channels is governed by

graphic file with name pcbi.1002478.e115.jpg (8)

where Inline graphic, Inline graphic, and the intracellular Inline graphic concentration Inline graphic is described by

graphic file with name pcbi.1002478.e120.jpg (9)

Units are mV for the membrane potential and ms for time. Note that the state space of the Traub model eqs. (4)–(9) is six-dimensional.

The dynamics of interest is described below. Starting from a resting state, as Inline graphic is increased, the model goes to repetitive spiking. Depending on the level of Inline graphic, this (rest-spiking) transition occurs through a SN bifurcation for low values of Inline graphic or a subcritical AH bifurcation for high values of Inline graphic, at input currents Inline graphic and Inline graphic, respectively. The SN bifurcation gives rise to a branch of stable periodic solutions (limit cycles) with arbitrarily low frequency. Larger values of Inline graphic cause the stable fixed point to lose its stability by an AH bifurcation (at Inline graphic). In this case, a branch of unstable periodic orbits emerges, which collides with a branch of stable limit cycles with finite frequency in a fold limit cycle bifurcation at current Inline graphic. The branch of stable periodic spiking trajectories extends for currents larger than Inline graphic and Inline graphic. This means that in the AH bifurcation regime, the model exhibits hysteresis. That is, for an input current between Inline graphic and Inline graphic a stable equilibrium point and a stable limit cycle coexist. On the contrary, Inline graphic does not affect the bifurcation of the equilibria, since it is essentially nonexistent at rest.

We used parameter values as in [22]. Assuming a cell surface area of Inline graphic, the membrane capacitance was Inline graphic, the conductances (in Inline graphic) were Inline graphic, Inline graphic, Inline graphic, Inline graphic, Inline graphic, Inline graphic, and the reversal potentials (in mV) were Inline graphic, Inline graphic, Inline graphic, Inline graphic; Inline graphic Inline graphic and Inline graphic.

Network simulations

We considered networks of Inline graphic coupled neurons with identical properties using both models (aEIF and Traub), driven to repetitive spiking with period Inline graphic,

graphic file with name pcbi.1002478.e153.jpg (10)

where the vector Inline graphic consists of the state variables of neuron Inline graphic (Inline graphic for the aEIF model, or Inline graphic for the Traub model), Inline graphic governs the dynamics of the uncoupled neuron (according to either neuron model) and the coupling function Inline graphic contains the synaptic current Inline graphic (received by postsynaptic neuron Inline graphic from presynaptic neuron Inline graphic) in the first component and all other components are zero. Inline graphic was modeled using a bi-exponential description of the synaptic conductance,

graphic file with name pcbi.1002478.e164.jpg (11)
graphic file with name pcbi.1002478.e165.jpg (12)

where Inline graphic denotes the peak conductance, Inline graphic the fraction of open ion channels, Inline graphic the conduction delay which includes axonal as well as dendritic contributions, and Inline graphic the synaptic reversal potential. Inline graphic is a normalization factor which was chosen such that the peak of Inline graphic equals one. The spike times Inline graphic of neuron Inline graphic (at the soma) correspond to the times at which the membrane potential reaches Inline graphic (in the aEIF model) or the peak of the action potential (in the Traub model). Inline graphic and Inline graphic are the rise and decay time constants, respectively. For excitatory synapses the parameters were chosen to model an AMPA-mediated current (Inline graphic, Inline graphic, Inline graphic), the parameters for inhbitory synapses we set to describe a Inline graphic-mediated current (Inline graphic, Inline graphic, Inline graphic).

We simulated the aEIF and Traub neuron networks, respectively, taking Inline graphic, homogeneous all-to-all connectivity without self-feedback (Inline graphic), and neglecting conduction delays (Inline graphic). We further introduced heterogeneities of several degrees w.r.t. synaptic strengths and conduction delays to the computationally less demanding aEIF network. Specifically, Inline graphic (Inline graphic) and Inline graphic were sampled from a uniform distribution over various value ranges. The neurons were weakly coupled, in the sense that the total synaptic input received by a neuron from all other neurons in the network (assuming they spike synchronously) resulted in a maximal change of ISI (Inline graphic) of less than 5%, which was determined by simulations. As initial conditions we used points of the spiking trajectory at times that were uniformly sampled from the interval Inline graphic, i.e. the initial states were asynchronous. Simulation time was Inline graphic for each configuration of the aEIF networks and Inline graphic for the Traub neuron networks. All network simulations were done with BRIAN 1.3 [42] applying the second-order Runge-Kutta integration method with a time step of Inline graphic for coupled pairs and Inline graphic for larger networks.

We measured the degree of spike synchronization in the simulated networks using averaged pairwise cross-correlations between the neurons [43],

graphic file with name pcbi.1002478.e196.jpg (13)

where Inline graphic if neuron Inline graphic spikes in time interval Inline graphic, otherwise Inline graphic, for Inline graphic. Inline graphic indicates the average over all neuronal pairs (Inline graphic) in the network. Calculation period Inline graphic was Inline graphic and time bin Inline graphic was Inline graphic. Inline graphic assumes a value of Inline graphic for asynchronous spiking and approaches Inline graphic for perfect synchronization.

In order to quantify the degree of phase locking of neurons in the network we applied the mean phase coherence measure Inline graphic [44], [45] defined by

graphic file with name pcbi.1002478.e212.jpg (14)

where Inline graphic is the phase difference between neurons Inline graphic and Inline graphic at the time of the Inline graphic spike Inline graphic of neuron Inline graphic, Inline graphic. Inline graphic is the largest spike time of neuron Inline graphic that precedes Inline graphic, Inline graphic is the smallest spike time of neuron Inline graphic that succeeds Inline graphic. Inline graphic is the number of spikes of neuron Inline graphic in the calculation period Inline graphic. Inline graphic and Inline graphic denotes the average over all pairs Inline graphic. Inline graphic means no neuronal pair phase locks, Inline graphic indicates complete phase locking. Inline graphic was calculated using for Inline graphic the last Inline graphic (aEIF networks) or Inline graphic (Traub networks) of each simulation.

PRC calculation

The PRC can be obtained (experimentally or in simulations) by delivering small perturbations to the membrane potential of a neuron oscillating with period Inline graphic at different phases Inline graphic and calculating the change of the period. The PRC is then expressed as a function of phase as Inline graphic, where Inline graphic is the period of the neuron perturbed at Inline graphic. Positive (negative) values of Inline graphic represent phase advances (delays). An alternative technique of determining the PRC is to solve the linearized adjoint equation [22],

graphic file with name pcbi.1002478.e244.jpg (15)

subject to the normalization condition Inline graphic (see Text S1). Inline graphic, Inline graphic are as described above (cf. eq. (10)) and Inline graphic is the Jacobian matrix of Inline graphic. Inline graphic denotes the asymptotically stable Inline graphic-periodic spiking trajectory as a solution of the system

graphic file with name pcbi.1002478.e252.jpg (16)

of differential equations and a reset condition in case of the aEIF model. Eq. (16) together with the reset condition describe the dynamics of an uncoupled neuron. Inline graphic is an attractor of this dynamical system and nearby trajectories will converge to it. To obtain Inline graphic, we integrated the neuron model equations for a given set of parameters and adjusted the input current Inline graphic, such that the period was Inline graphic. Analysis was restricted to the regular spiking regime (cf. [34] for the aEIF model). Parameter regions where bursting and chaotic spiking occurs were avoided.

For Traub model trajectories, the peak of the action potential is identified with phase Inline graphic, for aEIF trajectories Inline graphic corresponds to the point of reset. The first component Inline graphic of the normalized Inline graphic-periodic solution Inline graphic of eq. (15) represents the PRC, also called infinitesimal PRC, which characterizes the response of the oscillator to a vanishingly small perturbation (cf. Text S1). For continuous limit cycles Inline graphic, as produced by the Traub model, Inline graphic can be obtained by solving eq. (15) backward in time over several periods with arbitrary initial conditions. Since Inline graphic is asymptotically stable, the Inline graphic-periodic solution of the adjoint system, eq. (15), is unstable. Thus, backward integration damps out the transients and we arrive at the periodic solution of eq. (15) [48][50]. In case of the aEIF model with an asymptotically stable Inline graphic-periodic solution Inline graphic, that involves a discontinuity in both variables Inline graphic, Inline graphic at integer multiples of Inline graphic, we treated the adjoint equations as a boundary value problem [18]. Specifically, we solved the adjoint system

graphic file with name pcbi.1002478.e271.jpg (17)
graphic file with name pcbi.1002478.e272.jpg (18)

subject to the conditions

graphic file with name pcbi.1002478.e273.jpg (19)
graphic file with name pcbi.1002478.e274.jpg (20)

where Inline graphic denote the two components of Inline graphic, and Inline graphic is the left-sided limit. Eq. (19) is the normalization condition. Eq. (20) is the continuity condition, which ensures Inline graphic-periodicity of the solution (see Text S1, derivation based on [51][53]). From the fact, that the end points of Inline graphic-periodic aEIF trajectories differ, i.e. Inline graphic, Inline graphic and Inline graphic, it follows that Inline graphic, which in turn leads to Inline graphic. Perturbations of the same strength, which are applied to Inline graphic just before and after the spike, have therefore a different effect on the phase, leading to a discontinuity in the PRC.

The PRCs presented in this study were calculated using the adjoint method. For validation purposes, we also simulated a number of PRCs by directly applying small perturbations to the membrane potential Inline graphic of the oscillating neuron at different phases and measuring the change in phase after many cycles – to ensure, that the perturbed trajectory had returned to the attractor Inline graphic. The results are in good agreement with the results of the adjoint method.

Phase reduction

In the limit of weak synaptic interaction, which guarantees that a perturbed spiking trajectory remains close to the attracting (unperturbed) trajectory Inline graphic, we can reduce the network model (10) to a lower dimensional network model where neuron Inline graphic is described by its phase Inline graphic [48][50], [54], [55] as follows.

graphic file with name pcbi.1002478.e291.jpg (21)
graphic file with name pcbi.1002478.e292.jpg (22)

where Inline graphic is the PRC of neuron Inline graphic and Inline graphic the first component (membrane potential) of the spiking trajectory Inline graphic (see previous section and Text S1). Inline graphic is the Inline graphic-periodic averaged interaction function calculated using Inline graphic with conduction delay Inline graphic (11). Note that Inline graphic simply causes a shift in the interaction function: Inline graphic. Inline graphic only depends on the difference of the phases (in the argument) which is a useful property when analyzing the stability of phase locked states of coupled neuronal pairs. In this case (without self-feedback as already assumed) the phase difference Inline graphic evolves according to the scalar differential equation

graphic file with name pcbi.1002478.e305.jpg (23)

whose stable fixed points are given by the zero crossings Inline graphic of Inline graphic for which Inline graphic and Inline graphic. If Inline graphic is differentiable at Inline graphic, these left and right sided limits are equal and represent the slope. Note however that Inline graphic is continuous, but not necessarily differentiable due to the discontinuity of the PRC of an aEIF neuron. Therefore, the limits might not be equal in this case. The case where Inline graphic is discontinuous at Inline graphic, which can be caused by Inline graphic-pulse coupling, i.e. Inline graphic is replaced by a Inline graphic-function, is addressed in the Results section. We calculated these stable fixed points, which correspond to stable phase locked states, for pairs of identical cells coupled with equal or heterogeneous synaptic strengths and symmetric conduction delays, Inline graphic, using PRCs derived from the aEIF and Traub neuron models, driven to Inline graphic periodic spiking. Periodic spiking trajectories of both models and PRCs of Traub neurons were computed using variable order multistep integration methods, for PRCs of aEIF neurons a fifth-order collocation method was used to solve eqs. (17)–(20). These integration methods are implemented in MATLAB (2010a, The MathWorks). Bifurcation currents of the Traub model were calculated using MATCONT [56], [57].

Results

PRC characteristics of aEIF neurons

We first examine the effects of the adaptation components Inline graphic and Inline graphic, respectively, on spiking behavior of aEIF neurons at rest in response to (suprathreshold) current pulses (Fig. 1A–C). Without adaptation (Inline graphic) the model produces tonic spiking (Fig. 1A). Increasing Inline graphic or Inline graphic leads to SFA as shown by a gradual increase of the inter spike intervals (ISI) until a steady-state spike frequency Inline graphic is reached. Adaptation current Inline graphic builds up and saturates slowly when only conductance Inline graphic is considered (Fig. 1B) in comparison to spike-triggered increments Inline graphic (Fig. 1C). Fig. 1D,E depicts the relationship between Inline graphic and the injected current Inline graphic for various fixed values of Inline graphic and Inline graphic. Increased subthreshold adaptation causes the minimum spike frequency to jump from zero to a positive value, producing a discontinuous Inline graphic-Inline graphic curve (Fig. 1D). A continuous (discontinuous) Inline graphic-Inline graphic curve indicates class I (II) membrane excitability which is typical for a SN (AH) bifurcation at the onset of spiking respectively. An increase of Inline graphic causes this bifurcation to switch from SN to AH, thereby changing the membrane excitability from class I to II, shown by the Inline graphic-Inline graphic curves. An increase of Inline graphic on the other hand does not produce a discontinuity in the Inline graphic-Inline graphic curve, i.e. the membrane excitability remains class I (Fig. 1E). Furthermore, increasing Inline graphic shifts the Inline graphic-Inline graphic curve to larger current values without affecting its slope, while an increase of Inline graphic decreases the slope of the Inline graphic-Inline graphic curve in a divisive manner. When Inline graphic is large, the neuron is desensitized in the sense that spike frequency is much less affected by changes in the driving input.

In Fig. 2A,B we show how Inline graphic and Inline graphic differentially affect the shape of the PRC of an aEIF neuron driven to periodic spiking. The PRCs calculated using the adjoint method (solid curves) match well with those obtained from simulations (circles). While non-adapting neurons have monophasic (type I) PRCs, which indicate only advancing effects of excitatory perturbations, increased levels of Inline graphic produce biphasic (type II) PRCs with larger magnitudes, which predict a delaying effect of excitatory perturbations received early in the oscillation cycle. An increase of Inline graphic on the other hand flattens the PRC at early phases, shifts its peak towards the end of the period and reduces its magnitude. The type of the PRC however remains unchanged (type I). Indeed, if Inline graphic the PRC must be type I, since in this case the component Inline graphic of the solution of the adjoint system, eqs. (17)–(20), can be written as Inline graphic, where Inline graphic is given by the right-hand side of eq. (17). Thus, Inline graphic cannot switch sign.

Figure 2. Effects of adaptation on PRCs of aEIF neurons.

Figure 2

A,B: PRCs associated with adaptation parameters as in Fig. 1D,E. Solid curves are PRCs calculated with the adjoint method and scaled by 0.1 mV, circles denote PRC points that were obtained from numerical simulations of eqs. (1)–(3), using 0.1 mV perturbations at various phases Inline graphic (see Methods and Text S1). The input currents Inline graphic were chosen to ensure 40 Hz spiking. Note that the discontinuity of the PRCs at Inline graphic is caused by the reset of the spiking trajectories. C–F Top: PRCs for adaptation parameters as indicated and Inline graphic (C), Inline graphic (D), Inline graphic (E), Inline graphic (F). C–F Bottom: Vector field, Inline graphic- and Inline graphic-nullclines, and periodic spiking trajectory in the respective state space. The reset point (solid square) of the trajectory corresponds to the phase Inline graphic. A solid arrow marks the location along the trajectory where the PRC (shown above) has its maximum. Dashed arrows in D, F mark the trajectory points that correspond to the zero crossings of the PRCs. Trajectory points change slowly in regions where the vector field magnitudes are small. The dashed blue curve in D denotes the boundary of the domain of attraction of the fixed point, which is located at the intersection of the nullclines. Note that differences in the vector fields and Inline graphic-nullclines between C and E as well as D and F, are due to the changes in Inline graphic.

To provide an intuitive explanation for the effects of adaptation on the PRC, we show the vector fields, Inline graphic- and Inline graphic-nullclines, and periodic spiking trajectories of four aEIF neurons (Fig. 2C–F). One neuron does not have an adaptation current (Inline graphic), two neurons possess only one adaptation mechanism (Inline graphic, Inline graphic and Inline graphic, Inline graphic, respectively) and for one both adaptation parameters are increased (Inline graphic, Inline graphic). An excitatory perturbation to the non-adapting neuron at any point of its trajectory, i.e. at any phase, shifts this point closer to Inline graphic along the trajectory, which means the phase is shifted closer to Inline graphic, hence the advancing effect (Fig. 2C). The phase advance is strongest if the perturbing input is received at the position along the trajectory around which the vector field has the smallest magnitude, i.e. where the trajectory is “slowest”. In case of subthreshold adaptation (Fig. 2D), the adapted periodic spiking trajectory starts at a certain level of Inline graphic which decreases during the early part of the oscillation cycle and increases again during the late part, after the trajectory has passed the Inline graphic-nullcline. A small transient excitatory input at an early phase pushes the respective point of the trajectory to the right (along the Inline graphic-axis) causing the perturbed trajectory to pass through a region above the unperturbed trajectory, somewhat closer to the fixed point around which the vector field is almost null. Consequently, the neuron is slowed down and the subsequent spike delayed. An excitatory perturbation received at a later phase (to the right of the dashed arrow) causes phase advances, since the perturbed trajectory either remains nearly unchanged, however with a shorter path to the end of the cycle, compared to the unperturbed trajectory, or it passes below the unperturbed one where the magnitude of the vector field (pointing to the right) is larger. Note that for the parametrization in Fig. 2D, both, the resting state as well as the spiking trajectory are stable. In this case, a strong depolarizing input at an early phase can push the corresponding trajectory point into the domain of attraction of the fixed point, encircled by the dashed line in the figure, which would cause the resulting trajectory to spiral towards the fixed point and the neuron would stop spiking. On the other hand, increasing Inline graphic would shrink the domain of attraction of the fixed point and at Inline graphic, it would be destabilized by a subcritical AH bifurcation. When Inline graphic and Inline graphic, we obtain a type I PRC (Fig. 2E), as explained above. The advancing effect of an excitatory perturbation is strongest late in the oscillation cycle, indicated by the red arrow, where the perturbation pushes a trajectory point from a “slow” towards a “fast” region closer to the end of the cycle, as shown by the vector field. When Inline graphic as well as Inline graphic are increased, the PRC exhibits both adaptation mediated features (type II and skewness), see Fig. 2F. A push to the right along the corresponding trajectory experienced early in the cycle brings the perturbed trajectory closer to the fixed point and causes a delayed next spike. Such an effect persists even if the fixed point has disappeared due to a larger input current. In this case, the region where the fixed point used to be prior to the bifurcation, known as “ghost” of the fixed point, the vector field is still very small. This means that type II PRCs can exist for larger input currents Inline graphic. Note that differences of the vector fields and the shift of the nullclines relative to each other in Fig. 2C,D as well as Fig. 2E,F are due to different input current values (as an increase of Inline graphic moves the Inline graphic-nullcline upwards). The maximal phase advances, indicated by solid arrows in Fig. 2A,B, are close to the threshold potential Inline graphic (where the Inline graphic-nullcline has its minimum) in all four cases.

We next investigate how the changes in PRCs caused by either adaptation component are affected by the spike frequency. Bifurcation currents, rheobase currents and corresponding frequencies, in dependence of Inline graphic and Inline graphic, as well as regions in parameter space where PRCs are type I and II, are displayed in Fig. 3A–D. Fig. 3E,F shows how individual PRCs are modulated by spike frequency (input current). Both PRC characteristics, caused by Inline graphic and Inline graphic, respectively, are more pronounced at low frequencies. Increasing Inline graphic changes a type II PRC to type I and shifts its peak towards an earlier phase. The input current which separates type I and type II PRC regions (in parameter space) increases with both, Inline graphic and Inline graphic (Fig. 3A,B). That is, an increase of Inline graphic can also turn a type I into a type II PRC, by bringing the spiking trajectory closer to the fixed point or its “ghost”. This is however only possible if the system is in the AH bifurcation regime (Inline graphic) or close to it. Spike-triggered adaptation thereby considerably influences the range of input currents for which the PRCs are type II. The spike frequency according to the input current, at which a type II PRC turns into type I increases substantially with increasing Inline graphic, but only slighly with an increase of Inline graphic (Fig. 3C,D). The latter can be recognized by the similarity of the respective (green) curves in the subfigures C and D. Type II PRCs thus only exist in the lower frequency band whose width increases with increasing subthreshold adaptation.

Figure 3. Bifurcation currents of the aEIF model and dependence of PRC characteristics on the input current.

Figure 3

A,B: Rheobase current (solid black), SN and AH bifurcation currents Inline graphic, Inline graphic (dashed grey, dashed black) respectively, as well as input current (green) which separates type I (blue) and type II (yellow) PRC regions, as a function of Inline graphic, for Inline graphic (A) and Inline graphic (B). At Inline graphic a BT bifurcation occurs at Inline graphic (where the SN and the AH bifurcations meet) marked by the red dot. The region around Inline graphic is displayed in a zoomed view. If Inline graphic the system undergoes a SN bifurcation at Inline graphic, if Inline graphic an AH bifurcation occurs at Inline graphic. C,D: Spike frequencies Inline graphic corresponding to the input currents in A and B. Note that the region in Inline graphic-Inline graphic space where the PRCs are type II is very shallow in A compared to B, the corresponding regions in Inline graphic-Inline graphic space shown in C and D however are rather similar. This is due to the steep (flat) Inline graphic-Inline graphic relationship for Inline graphic (Inline graphic) respectively (see Fig. 1D,E). E,F: PRCs with locations in Inline graphic-Inline graphic space as indicated, scaled to the same period Inline graphic.

Phase locking of coupled aEIF pairs

In this section, we examine how the changes in phase response properties due to adaptation affects phase locking of coupled pairs of periodically spiking aEIF neurons. Specifically, we first analyze how the shape of the PRC determines the fixed points of eq. (23) and their stability, and then show how the modifications of the PRC mediated by the adaptation components Inline graphic and Inline graphic change those fixed points. Finally, we investigate the effects of conduction delays and heterogeneous coupling strengths on phase locking in dependence of adaptation.

Relation between phase locking and the PRC

In case of identical cell pairs and symmetric synaptic strengths, Inline graphic, the interaction functions in eq. (23) are identical, Inline graphic, where Inline graphic is the conduction delay. Inline graphic then becomes an odd, Inline graphic-periodic function, which has roots at Inline graphic and Inline graphic. Thus, the in-phase and anti-phase locked states always exist. The stability of these two states can be “read off” the PRC even without having to calculate Inline graphic, as is explained below. Let Inline graphic in the following. The fixed point Inline graphic of eq. (23) is stable if Inline graphic and Inline graphic. Note that the left and right sided limits are not equal if Inline graphic is not differentiable at Inline graphic, due to the discontinuity of the PRC of an aEIF neuron.

First, consider a synaptic current with infinitely fast rise and decay. In this case we use a positive (or negative) Inline graphic-function in eq. (21) instead of Inline graphic to describe the transient excitatory (or inhibitory) pulse. Inline graphic is then given by

graphic file with name pcbi.1002478.e450.jpg (24)

that is, Inline graphic becomes the PRC, mirrored at Inline graphic, rightwards shifted by the delay Inline graphic and scaled by Inline graphic. The sign of the slope of Inline graphic is thus given by the negative (positive) sign of the PRC slope at Inline graphic, Inline graphic, for excitatory (inhibitory) synapses respectively. For the aEIF model, the case Inline graphic requires a distinction, because Inline graphic and Inline graphic are discontinuous at Inline graphic. Let Inline graphic be the distance between Inline graphic and the closest root of Inline graphic. Since Inline graphic is odd and Inline graphic-periodic, Inline graphic implies stability of Inline graphic, in the sense that Inline graphic increases on the interval Inline graphic and decreases over Inline graphic. Thus, Inline graphic can be considered an attractor. Inline graphic is equivalent to Inline graphic which in turn is equivalent to Inline graphic for excitatory coupling and Inline graphic for inhibitory coupling. Hence, it is the discontinuity of the PRC which determines the stability of Inline graphic in this case.

A synaptic current with finite rise and decay times causes an additional rightwards shift and a smoothing of the interaction function. The stability of the fixed point Inline graphic is then determined by the slope of the PRC and its discontinuity on the interval Inline graphic, where Inline graphic is on the order of the synaptic timescale. If the PRC slope is negative on this interval and its discontinuity (if occurring in the interval) is also negative, i.e. Inline graphic, then Inline graphic is stable for excitatory coupling and unstable for inhibitory coupling. In Fig. 4A we show the effect of the synaptic timescale, i.e. Inline graphic and Inline graphic, on the interaction function for a given PRC. Fig. 4B,C illustrates how the stability of the synchronous state of a neuronal pair is given by the slope of the PRC, for three different delays. The slope of the PRC is positive at Inline graphic, Inline graphic and negative at Inline graphic and remains positive (negative) until Inline graphic has decayed to a small value. Therefore, synchrony is unstable for delays Inline graphic, Inline graphic and stable for Inline graphic, indicated by the slope of Inline graphic at Inline graphic, which is negative for the first two and positive for the third delay.

Figure 4. Relationship between the PRC and the interaction function.

Figure 4

A: PRC of an aEIF neuron (top) spiking at Inline graphic and interaction functions Inline graphic (bottom) obtained for synaptic conductances with three different sets of synaptic time constants: Inline graphic, Inline graphic (blue), Inline graphic, Inline graphic (green); Inline graphic, Inline graphic (magenta), and Inline graphic. The synaptic current Inline graphic associated with each pair of time constants (center) illustrates the three synaptic timescales relative to the period Inline graphic. Note that Inline graphic shown here is received by the neuron at the beginning of its ISI. B: PRC (solid black) of an aEIF neuron spiking at Inline graphic and excitatory synaptic currents Inline graphic with Inline graphic, Inline graphic (dashed blue) received at three different phases. Assuming the input comes from a second, synchronous neuron, these phases represent three different conduction delays Inline graphic, Inline graphic, and Inline graphic. Note that synaptic input received at an earlier phase causes a larger peak of Inline graphic, due to the smaller value Inline graphic of the membrane potential which leads to a larger difference Inline graphic to the synapse's reversal potential Inline graphic. C: Interaction functions Inline graphic for pairs of neurons with the PRC shown in B, coupled by excitatory synapses with Inline graphic, Inline graphic, and delays Inline graphic and Inline graphic. The values of Inline graphic at Inline graphic are highlighted by blue circles. The slopes of Inline graphic, in terms of both left and right sided limits Inline graphic and Inline graphic, indicate whether the synchronous states are stable or unstable (see main text).

Effects of adaptation on phase locking of coupled aEIF pairs

First, consider pairs of identical aEIF neurons with the PRCs shown in Fig. 2A,B, symmetrically coupled through instantaneous synapses (Inline graphic and Inline graphic) and without conduction delays (Inline graphic). When the coupling is excitatory, the in-phase locked state (synchrony) is unstable in case of type I PRCs, since they have a positive “jump” at Inline graphic, i.e. Inline graphic. Synchrony is stable for pairs with type II PRCs however, as Inline graphic. The anti-phase locked state on the other hand is unstable because of the positive PRC slopes at Inline graphic. In case of inhibitory coupling, synchrony is stable for type I pairs and the anti-phase locked state is stable for all pairs. This means, bistability of in-phase and anti-phase locking occurs for inhibitory neurons with type I PRCs.

Next, we consider pairs that are coupled through synaptic currents Inline graphic with finite rise and decay times, as described in the Methods section. In Fig. 5 we show how the stable (and unstable) phase locked states of pairs of neurons with symmetric excitatory (A, B) and inhibitory (C, D) synaptic interactions and without conduction delays change, when the PRCs are modified by the adaptation components Inline graphic and Inline graphic. For excitatory pairs, stable fixed points shift towards synchrony, when Inline graphic or Inline graphic is increased. The phase differences become vanishinly small, when the PRCs switch from type I to type II due to subthreshold adaptation. Perfect synchrony is stabilized, where the PRC slopes at Inline graphic for small Inline graphic become negative, due to even larger values of Inline graphic (not shown) or lower spike frequency (see Fig. 3C–F). Neurons that have type I PRCs with a pronounced skew, as caused by spike-triggered adaptation, lock almost but not completely in-phase, if the adaptation is sufficiently strong. Inhibitory pairs on the other hand show stable synchrony independent of PRC type and skewness. Larger values of Inline graphic or Inline graphic lead to additional stabilization of the anti-phase locked state. That is, strong adaptation in inhibitory pairs mediates bistability of in-phase and anti-phase locking. All phase locking predictions from the phase reduction approach are in good agreement with the results of numerically simulated coupled aEIF pairs.

Figure 5. Effects of adaptation on phase locked states of coupled aEIF pairs.

Figure 5

Stable (solid black) and unstable (dashed grey) phase locked states of pairs of aEIF neurons spiking at Inline graphic with identical PRCs as a function of adaptation parameters. These phase locked states were obtained by evaluating the interaction function. Circles denote the steady-state phase differences by numerically simulating pairs of aEIF neurons according to eqs. (1)–(3). To detect bistability, the simulations were run multiple times and the pairs initialized either near in-phase or anti-phase with values of the periodic spiking trajectory. In A and B the neurons are coupled through excitatory, in C and D through inhibitory synapses, as indicated by the diagrams on the left. Synaptic conductances are equal (Inline graphic) and conduction delays are not considered here (Inline graphic). Synaptic time constants were Inline graphic, Inline graphic for excitatory and Inline graphic, Inline graphic for inhibitory connections. In A and C, Inline graphic varies from 0 to 0.1 Inline graphic with Inline graphic, whereas in B and D, Inline graphic while Inline graphic varies from 0 to 0.2 nA. All other model parameters are given in the Methods section. The corresponding changes in PRCs are indicated in the top row.

Phase locking of aEIF pairs coupled with delays

We next investigate how phase locked states of excitatory and inhibitory pairs are affected by synaptic currents that involve conduction delays, considering the PRC of a neuron without adaptation, and two PRCs that represent adaptation induced by either Inline graphic or Inline graphic. Neurons symmetrically coupled through excitatory synapses with a conduction delay do not synchronize irrespective of whether adaptation is present or not (Fig. 6A–C). Instead, stable states shift towards anti-phase locking with increasing mutual delays. Inhibitory pairs on the other hand synchronize for all conduction delays (Fig. 6D–F), but the anti-phase locked states of coupled inhibitory neurons with type II PRCs or skewed type I PRCs are destabilized by the delays. The bistable region is larger in case of spike-triggered adaptation compared to subthreshold adaptation (Fig. 6E,F). Again, all stable phase locked states obtained using phase reduction are verified by numerical simulations. Fig. 7 illustrates the phenomenon that synchronous spiking of excitatory pairs is destabilized by the delay, while synchrony remains stable for inhibitory pairs. Consider two neurons oscillating with a small phase difference Inline graphic (neuron 1 slightly ahead of neuron 2). Then, a synaptic input received by neuron 2 at a delay Inline graphic after neuron 1 has spiked, arrives at an earlier phase (Inline graphic) compared to the phase at which neuron 1 receives its input (Inline graphic). Consequently, if the synapses are excitatory and the PRCs type I, the leader neuron 1 advances its next spike by a larger amount than the follower neuron 2 (Fig. 7A). In case of excitatory neurons and type II PRCs, depending on Inline graphic and Inline graphic, the phase of neuron 1 is advanced by a larger amount or delayed by a smaller amount than the phase of neuron 2, the latter of which is shown by the changed spike times in Fig. 7B. It is also possible that the phase of the leader neuron is advanced while that of the follower neuron is delayed. Hence, for either PRC type, Inline graphic increases due to delayed excitatory coupling, that is, synchrony is destabilized. For inhibitory synapses and type I PRCs, the leader neuron 1 delays its subsequent spike by a larger amount than the follower neuron 2 (Fig. 7C). In case of type II PRCs, neuron 1 experiences a weaker phase advance or stronger phase delay than neuron 1, or else the phase of neuron 1 is delayed while that of neuron 1 is advanced, depending on Inline graphic and Inline graphic (Fig. 7D). Thus, delayed inhibitory coupling causes Inline graphic to decrease towards zero for either PRC type, that is, synchrony is stabilized.

Figure 6. Phase locking of coupled aEIF pairs with conduction delays.

Figure 6

Stable (solid black) and unstable (dashed grey) phase locked states of aEIF pairs without adaptation, Inline graphic (A and D), and with adaptation, Inline graphic, Inline graphic (B and E), Inline graphic, Inline graphic (C and F), as a function of the conduction delay Inline graphic. The neurons are coupled through excitatory (A–C) or inhibitory synapses (D–F) with equal conductances (Inline graphic). Synaptic time constants are as in Fig. 5. Circles denote steady-state phase differences of numerically simulated pairs of aEIF neurons. The corresponding PRCs are shown in the top row. Inline graphic was 25 ms.

Figure 7. Effects of conduction delays on the stability of synchrony in coupled pairs.

Figure 7

Spike times (solid bars) of two neurons oscillating with a small phase difference Inline graphic and coupled through excitatory (A and B) or inhibitory synapses (C and D) with a symmetric conduction delay Inline graphic. The PRCs of the neurons that make up each pair are displayed below. In A and C the neurons have type I PRCs, in B and D the PRCs are type II. The time (phase) at which each neuron receives a synaptic current is shown along the spike trace. Phase advances or delays, considering the time of input arrival and the shape of the PRC, are indicated by advanced or delayed subsequent spike times. Dashed bars indicate spike times without synaptic inputs. The consequent changes in Inline graphic are highlighted.

Phase locking of aEIF pairs coupled with delays and unequal synaptic strengths

In the following we analyze phase locking of neuronal pairs with unequal synaptic peak conductances Inline graphic. Due to the linearity of the integral in eq. (21) we can substitute Inline graphic in eq. (23), which yields

graphic file with name pcbi.1002478.e581.jpg (25)

By setting eq. (25) to zero, we obtain the condition eq. (26) for the existence of phase locked states,

graphic file with name pcbi.1002478.e582.jpg (26)

Phase locked states therefore only exist if the ratio of conductances Inline graphic is not larger than the maximum of the periodic function Inline graphic. This upper bound primarily depends on the type of the PRCs and the synaptic time constants. In case of type I PRCs, Inline graphic is limited because the minimum of Inline graphic is positive. Inline graphic is either positive (for excitatory synapses) or negative (for inhibitory synapses) for all Inline graphic. Inline graphic is small for slow synapses, since the slower the synaptic rise and decay times, the larger Inline graphic, see Fig. 4A. For a type II PRC on the other hand, this minimum is zero (unless the negative lobe of the PRC is small and the synapse slow), from which follows that Inline graphic. The effects of heterogeneous synaptic strengths on phase locking of neuronal pairs without adaptation, as well as either adaptation parameter increased, are shown in Fig. 8. For excitatory pairs coupled without a conduction delay it is illustrated, how the right hand side of eq. (25) changes when the coupling strengths are varied (A–C). In addition, stable phase locked states of excitatory and inhibitory pairs coupled through synapses with various mutual conduction delays (Inline graphic, Inline graphic, or Inline graphic) are displayed as a function of Inline graphic (D–I). When the ratio of conductances Inline graphic is increased, the zero crossings of Inline graphic given by eq. (25), i.e. phase locked states, disappear for neurons with type I PRCs (through a SN bifurcation). Inline graphic then continuously increases (or decreases) (mod Inline graphic) as shown by the dashed curves (without roots) in Fig. 8A,C and indicated by the arrows in Fig. 8D,F,G,I. This means, the spike frequency of one neuron becomes faster than that of the other neuron. Neurons with type II PRCs on the other hand have stable phase locked states even for diverging coupling strengths. Bistability of two phase locked states can occur for a ratio Inline graphic close to one (equal coupling strengths), depending on the PRC and the delay. Synchronization of excitatory-inhibitory pairs is not considered in this paper. It should be noted however, that if both neurons have type I PRCs, phase locking is not possible, irrespective of the ratio of coupling strengths. In this case, one interaction function is strictly positive and the other strictly negative and thus, the condition (26) for fixed points of eq. (25) cannot be fulfilled.

Figure 8. Phase locking of aEIF pairs coupled with delays and heterogeneous synaptic strengths.

Figure 8

A–C: Change of phase difference Inline graphic given by equation (25), as a function of Inline graphic for pairs of excitatory aEIF neurons coupled with different ratios of synaptic conductances Inline graphic (Inline graphic). Zero crossings with a negative slope indicate stable phase locking and are marked by black dots. Adaptation parameters of the neurons and PRCs are shown in the top row. D–I: Stable phase locked states of excitatory (D–F) and inhibitory (G–I) pairs as a function of the synaptic conductance ratio, for three different conduction delays Inline graphic, Inline graphic and Inline graphic (black, brown, green). Unstable states are not shown for improved clarity. Dashed lines denote equal synaptic strengths, grey arrows indicate a continuous increase or decrease of Inline graphic (mod Inline graphic) for ratios Inline graphic at which phase locked states do not exist (see main text).

Synchronization and clustering in aEIF networks

In order to examine how the behavior of pairs of coupled phase neurons relates to networks of spiking neurons, we performed numerical simulations of networks of oscillating aEIF neurons without adaptation and with either a subthreshold or a spike-triggered adaptation current, respectively, and analyzed the network activity. The neurons were all either excitatory or inhibitory and weakly coupled. Fig. 9 shows the degree of synchronization Inline graphic (A, C) and the degree of phase locking Inline graphic (B) for these networks considering equal as well as heterogeneous conduction delays and synaptic conductances. An increase of either adaptation parameter (Inline graphic or Inline graphic) leads to increased Inline graphic in networks of excitatory neurons with short delays. It can be recognized however, that Inline graphic increases to larger values and this high degree of synchrony seems to be more robust against heterogeneous synaptic strengths, when the neurons are equipped with a subthreshold adaptation current (Fig. 9A,C). These effects correspond well to those of the adaptation components Inline graphic and Inline graphic on synchronization of pairs, presented in the previous section. Parameter regimes (w.r.t. Inline graphic and Inline graphic) that cause stable in-phase or near in-phase locking of pairs, such as subthreshold adaptation in case of short delays or spike-triggered adaptation for short delays and coupling strength ratios close to one (Fig. 6A–C and Fig. 8D–F), lead to synchronization, indicated by large Inline graphic values, in the respective networks. Networks of non-adapting excitatory neurons remain asynchronous as shown by the low Inline graphic values. For equal synaptic strengths, these networks settle into splay states where the neurons are pairwise phase locked, with uniformly distributed phases (Fig. 9B,D). When the delays are large enough and the synaptic strengths equal, splay states also occur in networks of neurons with large Inline graphic, indicated by low Inline graphic and high Inline graphic values in Fig. 9A,B. As far as inhibitory networks are concerned, non-adapting neurons synchronize, without delays or with random delays of up to 10 ms. Furthermore, synchrony in these networks is largely robust against heterogeneities in the coupling strengths (Fig. 9A). Networks of inhibitory neurons with subthreshold adaptation only show synchronization and pairwise locking for larger delays (i.e. Inline graphic random in Inline graphic or larger). Spike-triggered adaptation promotes clustering of the network into two clusters, where the neurons within a cluster are in synchrony, as long as the delays are small. These cluster states seem to be most robust against heterogeneous synaptic strengths when the delays are small but not zero. For larger delays, inhibitory neurons of all three types (with or without adaptation) synchronize, in a robust way against unequal synaptic strengths. The behaviors of inhibitory networks are consistent with the phase locked states found in pairs of inhibitory neurons (Fig. 6D–F). Particularly, stable synchronization of pairs with larger conduction delays and the bistability of in-phase and anti-phase locking of pairs with spike-triggered adaptation for smaller delays, nicely carry over to networks. In the former case, synchrony of pairs relates to network synchrony, in the latter case, bistability of in-phase and anti-phase locking of individual pairs can explain the observed two cluster states. Note that bistability of in-phase and anti-phase locking is also shown for inhibitory pairs with subthreshold adaptation and Inline graphic. In this case however, the slope of Inline graphic at Inline graphic is almost zero (not shown), which might explain why the corresponding networks do not develop two-cluster states. The behavior of all simulated networks does not critically depend on the number of neurons in the network, as we obtain qualitatively similar results for network sizes changed to Inline graphic and Inline graphic (not shown). The numerical simulations demonstrate that stable phase locked states of neural pairs can be used to predict the behavior of larger networks.

Figure 9. Impact of adaptation on the behavior of aEIF networks.

Figure 9

Degree of network synchronization Inline graphic (A) and phase locking Inline graphic (B) of Inline graphic aEIF neurons without adaptation, Inline graphic (black frame) and either adaptation component, respectively, Inline graphic, Inline graphic (blue frame), Inline graphic, Inline graphic (red frame), driven to 40 Hz spiking, all-to-all coupled without self-feedback, for various conduction delays and synaptic conductances. Inline graphic and Inline graphic are random (uniformly distributed) in the indicated intervals. Specifically, Inline graphic, Inline graphic and Inline graphic, Inline graphic, with units in parenthesis. The PRCs of the three neuron types described above are shown in the top row. C: Time course of Inline graphic for networks without delays and equal synaptic strengths, as indicated by the symbols in A. Each Inline graphic and Inline graphic value represents an average over three simulation runs. D: Raster plots for neuron and network parameters as indicated by the symbols in B, where the neurons in the columns are sorted according to their last spike time.

Synchronization properties of Traub neurons with adaptation currents Inline graphic, Inline graphic

To understand the biophysical relevance of the subthreshold and spike-triggered adaptation parameters, Inline graphic and Inline graphic, in the aEIF model, we compare them with the adaptation currents Inline graphic and Inline graphic in a variant of the Hodgkin-Huxley type Traub model neuron. Specifically, in this section we investigate the effects of the low- and high-threshold currents Inline graphic and Inline graphic, respectively, on spiking behavior, Inline graphic-Inline graphic curves and PRCs of single neurons, and on synchronization of pairs and networks, using the Traub model, and compare the results with those of the previous two sections. It should be stressed, that the aEIF model was not fit to the Traub model in this study. Therefore, the comparison of how adaptation currents affect SFA, PRCs and synchronization in both models, are rather qualitative than quantitative.

PRC characteristics of Traub neurons

Without adaptation, Inline graphic (hence Inline graphic), the model exhibits tonic spiking in response to a rectangular current pulse (Fig. 10A). When either adaptation current is present, that is the conductance Inline graphic or Inline graphic is increased to Inline graphic, the membrane voltage trace reveals SFA. Note that Inline graphic causes stronger differences in subsequent ISIs after stimulus onset, when comparing the Inline graphic-traces of neurons with either adaptation conductance set to Inline graphic. The Inline graphic-Inline graphic curves in Fig. 10B indicate that the presence of Inline graphic predominantly has a subtractive effect on the neuron's Inline graphic-Inline graphic curve and gives rise to class II excitability. The presence of Inline graphic on the other hand flattens the Inline graphic-Inline graphic curve, in other words its effect is divisive. Furthermore, an increase of Inline graphic changes a type I PRC to type II, whereas increased Inline graphic reduces its amplitude at early phases and skews its peak to the right (Fig. 10C). Evidently, the effects of Inline graphic and Inline graphic on SFA, Inline graphic-Inline graphic curves and PRCs of Traub neurons are consistent with the effects of the adaptation parameters Inline graphic and Inline graphic in aEIF neurons (Figs. 1, 2).

Figure 10. Effects of adaptation on spiking dynamics, Inline graphic-Inline graphic curves, PRCs and bifurcation currents of Traub model neurons.

Figure 10

A: Membrane potential Inline graphic of Traub model neurons without adaptation, Inline graphic (black), Inline graphic-mediated, Inline graphic (blue) and Inline graphic-mediated adaptation, Inline graphic (red), in response to step currents Inline graphic, B: the corresponding Inline graphic-Inline graphic curves, and C: the corresponding PRCs. Solid lines in C denote the PRCs, calculated with the adjoint method and scaled by 0.2 mV. Open circles denote the results of numerical simulations of eqs. (4)–(9) with 0.2 mV perturbations at various phases. D,E: Rheobase current Inline graphic (solid black), Inline graphic (dashed grey) and Inline graphic (dashed black), as a function of Inline graphic, for Inline graphic (left) and Inline graphic (right). Inline graphic and Inline graphic converge at Inline graphic marked by the red dot. The input current indicated by the green curve separates type I and type II PRC regions (blue and yellow, respectively). F,G: Spike frequencies Inline graphic according to the input currents Inline graphic in D and E. H,I: PRCs for parametrizations as indicated in F and G (with Inline graphic corresponding to Inline graphic), scaled to the same period Inline graphic. All other model parameters are provided in the Methods section.

We further show how the PRC characteristics caused by the adaptation currents depend on the injected current Inline graphic, hence the spike frequency Inline graphic, and the bifurcation type of the rest-spiking transition (Fig. 10D–I). An increase of Inline graphic reduces the effects of Inline graphic and Inline graphic on the PRC. That means, at higher frequencies Inline graphic, larger levels of Inline graphic and Inline graphic are required to obtain type II and skewed PRCs, respectively. This frequency dependence of adaptation current-mediated changes of the PRC is similar in both neuron models (Figs. 3, 10D–I). Note, that in the Traub model a rather low value of Inline graphic (Inline graphic) is sufficient to guarantee a type II PRC for spike frequencies of up to Inline graphic (Fig. 10F,G), compared to the aEIF model, where a much larger value of Inline graphic (Inline graphic) would be necessary (Fig. 3C,D). As far as the bifurcation structures of both models are concerned, an increase of the low-threshold adaptation parameters Inline graphic and Inline graphic has a comparable effect in the Traub and the aEIF models, respectively, changing the transition from rest to spiking from a SN via a BT to an AH bifurcation. The exact conductance values at which this change, i.e. the BT bifurcation, occurs, differ (Inline graphic for the Traub model and Inline graphic for the aEIF model).

Synchronization of coupled Traub neurons

We show the effects of the adaptation currents Inline graphic and Inline graphic on phase locked states of pairs of Traub neurons symmetrically coupled without conduction delays in Fig. 11A–D. Excitatory pairs of neurons without adaptation phase lock with a small phase difference. Low levels of Inline graphic are sufficient to stabilize in-phase locking, by turning the PRC from type I to II (Fig. 11A), while an increase of Inline graphic reduces the locked phase difference to almost but not exactly zero, that is, near in-phase locking, by skewing the PRC (Fig. 11B). Inhibitory synaptic coupling produces bistability of in-phase (synchrony) and anti-phase locking (anti-synchrony) for pairs of neurons without adaptation or either adaptation current increased (Fig. 11C,D). Note that the domain of attraction of the anti-synchronous state grows with increasing Inline graphic or Inline graphic, while that of the synchronous state shrinks. In contrast to the aEIF model, this bistability also occurs for neurons without an adaptation current (compare Figs. 5C,D, 11C,D).

Figure 11. Influence of adaptation on synchronization properties of Traub model neurons.

Figure 11

A–D: Stable (solid black) and unstable (dashed grey) phase locked states of coupled pairs of Traub neurons with identical PRCs, as a function of conductances Inline graphic and Inline graphic, respectively. Corresponding changes in PRCs are displayed in the top row. The neurons are coupled through excitatory or inhibitory synapses as indicated by the diagrams on the left, with equal synaptic strengths, Inline graphic and Inline graphic. E: Network synchronization Inline graphic over time, of Inline graphic coupled excitatory (solid) and inhibitory (dashed) Traub neurons without, Inline graphic (black) or with adaptation, Inline graphic, Inline graphic (blue) and Inline graphic, Inline graphic (red), driven to 40 Hz spiking. The neurons are all-to-all coupled with equal synaptic conductances, Inline graphic (black and blue), Inline graphic (red), but without self-feedback, Inline graphic, and conduction delays, Inline graphic. F: Raster plots showing the spike times during the last 200 ms for the three excitatory networks and the network of inhibitory neurons without adaptation (bottom). The neurons in the columns are sorted according to their last spike time.

The effects of Inline graphic or Inline graphic on synchronization of networks of Traub neurons coupled without conduction delays and equal synaptic strengths, are shown in Fig. 11E,F. In correspondence with the effects on pairs, Inline graphic and Inline graphic promote synchronization of excitatory networks, shown by the course of network synchronization measure Inline graphic over time (Fig. 11E). The mean values of phase locking measure Inline graphic are 0.26 for nonadapting neurons and 0.98 for networks where either adaptation current is increased. An increased adaptation current Inline graphic leads to larger Inline graphic values, compared to an increase of Inline graphic, which is similar to the aEIF networks where increased Inline graphic causes larger Inline graphic values than an increase of Inline graphic (compare Figs. 9C, 11E). In contrast to networks of excitatory aEIF neurons without adaptation, which develop splay states, Inline graphic values of nonadapting excitatory Traub neuron networks increase to about 0.5, while low Inline graphic values indicate poor phase locking, hence splay states do not occur (Fig. 11F). Networks of inhibitory neurons organize into clusters, indicated by Inline graphic values that converge to 0.5 (Fig. 11E) and large Inline graphic values (0.96 without adaptation, 0.94 for either Inline graphic or Inline graphic increased). Particularly, clustering into two clusters was revealed by the raster plots, see Fig. 11F. These two-cluster states of networks can be explained by the bistability of synchrony and anti-synchrony of individual pairs. Clustering emerges for all three types of Traub neurons, with and without adaptation, as opposed to networks of inhibitory aEIF neurons, where cluster states only occur in case of spike-triggered adaptation (Fig. 9). Considering the collective behavior of coupled excitatory neurons, the synchronizing effects of Inline graphic and Inline graphic in the Traub model are comparable to those of the adaptation components Inline graphic and Inline graphic in the aEIF model.

Discussion

In this work we studied the role of adaptation in the aEIF model as an endogenous neuronal mechanism that controls network dynamics. We described the effects of subthreshold and spike-triggered adaptation currents on the PRC in dependence of spike frequency. To provide insight into the synchronization tendencies of coupled neurons, we applied a common phase reduction technique and used the PRC to describe neuronal interaction [48], [55]. For pairs of coupled oscillating neurons we analyzed synchrony and phase locking under consideration of conduction delays and heterogeneous synaptic strengths. We then performed numerical simulations of aEIF networks to examine whether the predicted behavior of coupled pairs relates to the activity of larger networks. Finally, to express the biophysical relevance of the elementary subthreshold and spike-triggered adaptation mechanisms in the aEIF model, we compared their effects with those of the adaptation currents Inline graphic and Inline graphic in the high-dimensional Traub neuron model, on single neuron as well as network behavior.

Conductance Inline graphic, which mostly determines the amount of adaptation current in absence of spikes, that is, subthreshold, qualitatively changes the rest-spiking transition of an aEIF neuron, from a SN to an AH via a BT bifurcation as Inline graphic increases. Thereby the neuron's excitability, as defined by the Inline graphic-Inline graphic curve, and its PRC, are turned from class I to class II, and type I to type II, respectively. A similar effect of a slow outward current that acts in the subthreshold regime on the PRC has recently been shown for a two-dimensional quadratic non-leaky integrate-and-fire (QIF) model derived from a normal form of a dynamical model that undergoes a BT bifurcation [18], [48]. The relation between the PRC and the bifurcation types has further been emphasized by Brown et al. [47] who analytically determined PRCs for bifurcation normal forms and found type I and II PRC characteristics for the SN and AH bifurcations, respectively. A spike-triggered increment Inline graphic of adaptation current does not affect the bifurcation structure of the aEIF model and leaves the excitability class unchanged. When Inline graphic is small such that the model is in the SN bifurcation regime, an increase of Inline graphic cannot change the PRC type. In the AH bifurcation regime, Inline graphic substantially affects the range of input current for which the PRC is type II but causes only a small change in the corresponding frequency range. Furthermore, spike-triggered adaptation strongly influences the skew of the PRC, shifting its peak towards the end of the ISI for larger values of Inline graphic. Such a right-skewed PRC implies that the neuron is most sensitive to synaptic inputs that are received just before it spikes. Similar effects of spike-triggered negative feedback with slow decay on the skew of the PRC have been reported for an extended QIF model [18], [22], [48], [58].

PRCs determine synchronization properties of coupled oscillating neurons. When the synapses are fast compared to the oscillation period, the stability of the in-phase and anti-phase locked states (which always exist for pairs of identical neurons) can be “read off” the PRC for any mutual conduction delay, as we have demonstrated. A similar stability criterion that depends on the slopes of the PRCs at the phases at which the inputs are received has recently been derived for pairs of pulse-coupled oscillators [59]. Under the assumption of pulsatile coupling, the effect of a synaptic input is required to dissipate before the next input is received. In principle, the synaptic current can be strong, but it must be brief such that the perturbed trajectory returns to the limit cycle before the next perturbation occurs [14].

We have shown that, as long as synaptic delays are negligible and synaptic strengths equal, excitatory pairs synchronize if their PRCs are type II, as caused by Inline graphic, and lock almost in-phase if their PRCs are type I with a strong skew, as mediated by Inline graphic. Inhibitory pairs synchronize in presence of conduction delays and show bistability of in-phase and anti-phase locking for small delays, particularly in case of skewed PRCs. Conduction delays and synaptic time constants can affect the stability of synchrony in a similar way, by producing a lateral shift of the interaction function Inline graphic, as shown in Fig. 4. Note however, that the synaptic timescale has an additional effect on the shape of Inline graphic, smoothing it for slower synaptic rise and decay times. We have further demonstrated that heterogeneity in synaptic strengths desynchronizes excitatory and inhibitory pairs and leads to phase locking with a small phase difference in case of type II PRCs and small delays. While neurons with type II PRCs have stable phase locked states even for large differences in synaptic strengths, pairs of coupled neurons with type I PRCs are only guaranteed to phase lock when the synaptic strengths are equal. Similar effects of heterogeneous synaptic conductances have recently been observed in a computational study of weakly coupled Wang-Buszaki and Hodgkin-Huxley neurons (with class I and II excitability, respectively) [60].

The activity of larger aEIF networks, simulated numerically, is consistent with the predictions of the behavior of pairs. In fact, knowledge on phase locking of coupled pairs helps to explain the observed network states. Both adaptation mediated PRC characteristics, i.e. a negative lobe or a pronounced right skew, favor synchronization in networks of excitatory neurons, in agreement with previous findings [17], [22], [61]. This phenomenon only occurs when the conduction delays are negligible. It has been shown previously that synchrony in networks of excitatory oscillators becomes unstable when considering coupling with delays [62], [63]. We have demonstrated that increased conduction delays promote asynchrony in excitatory networks, with or without adaptation currents. Inhibitory neurons on the other hand are able to synchronize spiking in larger networks for a range of conduction delays. This provides support to the hypothesis that inhibitory networks play an essential role in generating coherent brain rhythms, as has been proposed earlier [43], [64], [2] for review. Inhibition rather than excitation has been found to generate neuronal synchrony particularly in case of slow synaptic rise and decay [40], [61], [65], and in the presence of conduction delays as has recently been shown experimentally [66]. In regimes that lead to bistability of in-phase and anti-phase locking according to our analysis of pairs, the simulated networks break up into two clusters of synchronized neurons. Recently it has been shown that a stable two cluster state of pulse coupled neural oscillators can exist even when synchrony of individual pairs is unstable [67]. Such cluster states have been invoked to explain population rhythms measured in vitro, where the involved neurons spike at about half of the population frequency [68].

Spike frequency has been shown to affect the skewness of PRCs, using type I integrate-and-fire neurons with adaptation [58], and to modulate the negative lobe in type II PRCs of conductance based model neurons [45]. Using the aEIF model we have demonstrated that the spike frequency strongly attenuates the effect of either adaptation mechanism on the PRC. At high frequency, unphysiologically large adaptation parameter values are necessary to produce a negative lobe or a significant right-skew in the PRC. This means, for a given degree of adaptation in excitatory neurons, synchronization is possible at frequencies up to a certain value. The stronger the adaptation, the larger this upper frequency limit. It has been previously suggested that the degree of adaptation can determine a preferred frequency range for synchronization of excitatory neurons, based on the observation (in vitro and in silico) that the neurons tend to spike in phase with injected currents oscillating at certain frequencies [69]. This preferred oscillation frequency increases with increasing degree of SFA. According to our results, at low frequencies synchronization of local circuits through excitatory synapses is possible, provided that the neurons are adapting and delays are short. At higher frequencies, adaptation much less affects the synchronization tendency of excitatory neurons and inhibition may play the dominant role in generating coherent rhythms [43], [64].

The adaptation currents Inline graphic and Inline graphic have previously been found to influence the phase response characteristics of the biophysical Traub neuron model, turning a type I PRC to type II (through Inline graphic) and modulating its skew (through Inline graphic) [18], [22]. We have shown that these changes of the PRC are reflected in the aEIF model by its two adaptation parameters and that in both models (aEIF and Traub) these changes are modulated by the spike frequency. As a consequence, the adaptation induced effects on synchronization of pairs and networks of oscillating neurons are qualitatively similar in both models. Quantitative differences with respect to these effects may well be reduced by fitting the aEIF model parameters to Traub neuron features.

Our analysis of phase locked states is based on the assumption that synaptic interactions are weak. Experimental work lending support to this assumption has been reviewed in [14], [50]. Particularly for stellate cells of the entorhinal cortex, synaptic coupling has been found to be weak [70]. Another assumption in this study is that the neurons spike with the same frequency. Considering a pair of neurons spiking at different frequencies, equation (23) needs to be augmented by a scalar Inline graphic, which accounts for the constant frequency mismatch between the two neurons [71]: Inline graphic. In this case, the condition for the existence of phase locked states is Inline graphic. Due to the assumption of weak synaptic strengths however, Inline graphic must be small, which means that the above condition can only be met if Inline graphic is small. In other words, in the limit of weak coupling phase locking is only possible if the spike frequencies are identical or differ only slightly. The phase reduction technique considered here, and PRCs in general, are of limited applicability for studying network dynamics in a regime where individual neurons spike at different frequencies, or even irregularly. How adaptation currents affect network synchronization and rhythm in such a regime nevertheless remains an interesting question to be addressed in the future.

Supporting Information

Text S1

Supplementary Methods. A) Calculation of the PRC using the adjoint method. B) Phase reduction.

(PDF)

Footnotes

The authors have declared that no competing interests exist.

This work was supported by the DFG Collaborative Research Center SFB910 (JL,MA,KO) and the NSF grant DMS-0908528 (LS). The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

References

  • 1.Singer W. Neuronal synchrony: A versatile code for the definition of relations? Neuron. 1999;24:49–65. doi: 10.1016/s0896-6273(00)80821-1. [DOI] [PubMed] [Google Scholar]
  • 2.Wang XJ. Neurophysiological and computational principles of cortical rhythms in cognition. Physiol Rev. 2010;90:1195–1268. doi: 10.1152/physrev.00035.2008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Roelfsema P, Engel A, Konig P. Visuomotor integration is associated with zero time-lag synchronization among cortical areas. Nature. 1997;385:157–161. doi: 10.1038/385157a0. [DOI] [PubMed] [Google Scholar]
  • 4.Ghazanfar A, Chandrasekaran C, Logothetis N. Interactions between the superior temporal sulcus and auditory cortex mediate dynamic face/voice integration in rhesus monkeys. J Neurosci. 2008;28:4457–4469. doi: 10.1523/JNEUROSCI.0541-08.2008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Melloni L, Molina C, Pena M, Torres D, Singer W, et al. Synchronization of neural activity across cortical areas correlates with conscious perception. J Neurosci. 2007;27:2858–2865. doi: 10.1523/JNEUROSCI.4623-06.2007. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Hipp J, Engel A, Siegel M. Oscillatory synchronization in large-scale cortical networks predicts perception. Neuron. 2011;69:387–396. doi: 10.1016/j.neuron.2010.12.027. [DOI] [PubMed] [Google Scholar]
  • 7.Fries P, Reynolds J, Rorie A, Desimone R. Modulation of oscillatory neuronal synchronization by selective visual attention. Science. 2001;291:1560–1563. doi: 10.1126/science.1055465. [DOI] [PubMed] [Google Scholar]
  • 8.Doesburg SM, Roggeveen AB, Kitajo K, Ward LM. Large-scale gamma-band phase synchronization and selective attention. Cereb Cortex. 2008;18:386–396. doi: 10.1093/cercor/bhm073. [DOI] [PubMed] [Google Scholar]
  • 9.Herrmann C, Munk M, Engel A. Cognitive functions of gamma-band activity: memory match and utilization. Trends Cogn Sci. 2004;8:347–355. doi: 10.1016/j.tics.2004.06.006. [DOI] [PubMed] [Google Scholar]
  • 10.Lengyel M, Kwag J, Paulsen O, Dayan P. Matching storage and recall: hippocampal spike timing-dependent plasticity and phase response curves. Nat Neurosci. 2005;8:1677–1683. doi: 10.1038/nn1561. [DOI] [PubMed] [Google Scholar]
  • 11.Hammond C, Bergman H, Brown P. Pathological synchronization in Parkinson's disease: networks, models and treatments. Trends Neurosci. 2007;30:357–364. doi: 10.1016/j.tins.2007.05.004. [DOI] [PubMed] [Google Scholar]
  • 12.Uhlhaas PJ, Singer W. Abnormal neural oscillations and synchrony in schizophrenia. Nat Rev Neurosci. 2010;11:100–113. doi: 10.1038/nrn2774. [DOI] [PubMed] [Google Scholar]
  • 13.Zijlmans M, Jacobs J, Zelmann R, Dubeau F, Grotman J. High-frequency oscillations mirror disease activity in patients with epilepsy. Neurology. 2009;72:979–986. doi: 10.1212/01.wnl.0000344402.20334.81. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Smeal R, Ermentrout GB, White J. Phase-response curves and synchronized neural networks. Phil Trans R Soc B. 2010;365:2407–2422. doi: 10.1098/rstb.2009.0292. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Ermentrout GB, Kopell N. Learning of phase-lags in coupled neural oscillators. Neural Comput. 1994;6:225–241. [Google Scholar]
  • 16.Netoff T, Banks M, Dorval A, Acker C, Haas J, et al. Synchronization in hybrid neuronal networks of the hippocampal formation. J Neurophysiol. 2005;93:1197–1208. doi: 10.1152/jn.00982.2004. [DOI] [PubMed] [Google Scholar]
  • 17.Crook SM, Ermentrout GB, Bower JM. Spike frequency adaptation affects the synchronization properties of networks of cortical oscillators. Neural Comput. 1998;10:837–854. doi: 10.1162/089976698300017511. [DOI] [PubMed] [Google Scholar]
  • 18.Ermentrout GB, Il BB, Netoff T. Phase response curves to measure ion channel effects on neurons. In: Schultheiss N, Prinz A, Butera R, editors. PRCs in neuroscience: theory, experiment and analysis. New York: Springer; 2012. pp. 207–236. [Google Scholar]
  • 19.McCormick D, Connors BW, Lighthall J, Prince D. Comparative electrophysiology of pyramidal and sparsely spiny stellate neurons of the neocortex. J Neurophysiol. 1985;54:782–806. doi: 10.1152/jn.1985.54.4.782. [DOI] [PubMed] [Google Scholar]
  • 20.Connors BW, Gutnick MJ. Intrinsic firing patterns of diverse neocortical neurons. Trends Neurosci. 1990;13:99–104. doi: 10.1016/0166-2236(90)90185-d. [DOI] [PubMed] [Google Scholar]
  • 21.La Camera G, Rauch A, Thurbon D, Lüscher HR, Senn W, et al. Multiple time scales of temporal response in pyramidal and fast spiking cortical neurons. J Neurophysiol. 2006;96:3448–3464. doi: 10.1152/jn.00453.2006. [DOI] [PubMed] [Google Scholar]
  • 22.Ermentrout GB, Pascal M, Gutkin BS. The effects of spike frequency adaptation and negative feedback on the synchronization of neural oscillators. Neural Comput. 2001;13:1285–1310. doi: 10.1162/08997660152002861. [DOI] [PubMed] [Google Scholar]
  • 23.Stiefel KM, Gutkin BS, Sejnowski TJ. The effects of cholinergic neuromodulation on neuronal phase-response curves of modeled cortical neurons. J Comput Neurosci. 2009;26:289–301. doi: 10.1007/s10827-008-0111-9. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Madison D, Lancaster B, Nicoll R. Voltage clamp analysis of cholinergic action in the hippocampus. J Neurosci. 1987;7:733–741. doi: 10.1523/JNEUROSCI.07-03-00733.1987. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Stiefel KM, Gutkin BS, Sejnowski TJ. Cholinergic neuromodulation changes phase response curve shape and type in cortical pyramidal neurons. PloS One. 2008;3:e3947. doi: 10.1371/journal.pone.0003947. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Treves A. Mean-field analysis of neuronal spike dynamics. Network. 1993;4:259–284. [Google Scholar]
  • 27.Richardson MJE, Brunel N, Hakim V. From subthreshold to firing-rate resonance. J Neurophysiol. 2003;89:2538–2554. doi: 10.1152/jn.00955.2002. [DOI] [PubMed] [Google Scholar]
  • 28.Fourcaud-Trocmé N, Hansel D, van Vreeswijk C, Brunel N. How spike generation mechanisms determine the neuronal response to uctuating inputs. J Neurosci. 2003;23:11628–11640. doi: 10.1523/JNEUROSCI.23-37-11628.2003. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 29.Brette R, Gerstner W. Adaptive exponential integrate-and-fire model as an effective description of neuronal activity. J Neurophysiol. 2005;94:3637–3642. doi: 10.1152/jn.00686.2005. [DOI] [PubMed] [Google Scholar]
  • 30.Gerstner W, Brette R. Adaptive exponential integrate-and-fire model. J Scholarpedia. 2009;4:8427. doi: 10.1152/jn.00686.2005. URL http://www.scholarpedia.org/article/Adaptive exponential integrate-and-fire model. [DOI] [PubMed] [Google Scholar]
  • 31.Izhikevich EM. Simple model of spiking neurons. IEEE Trans Neural Netw. 2003;14:1569–1572. doi: 10.1109/TNN.2003.820440. [DOI] [PubMed] [Google Scholar]
  • 32.Izhikevich EM. Which model to use for cortical spiking neurons? IEEE Trans Neural Netw. 2004;15:1063–1070. doi: 10.1109/TNN.2004.832719. [DOI] [PubMed] [Google Scholar]
  • 33.Touboul J, Brette R. Dynamics and bifurcations of the adaptive exponential integrate-and-fire model. Biol Cybern. 2008;99:319–334. doi: 10.1007/s00422-008-0267-4. [DOI] [PubMed] [Google Scholar]
  • 34.Naud R, Marcille N, Clopath C, Gerstner W. Firing patterns in the adaptive exponential integrate-and-fire model. Biol Cybern. 2008;99:335–347. doi: 10.1007/s00422-008-0264-7. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 35.Izhikevich EM, Edelman G. Large-scale model of mammalian thalamocortical systems. Proc Natl Acad Sci U S A. 2008;105:3593–3598. doi: 10.1073/pnas.0712231105. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 36.Destexhe A. Self-sustained asynchronous irregular states and up-down states in thalamic, cortical and thalamocortical networks of nonlinear integrate-and-fire neurons. J Comput Neurosci. 2009;27:493–506. doi: 10.1007/s10827-009-0164-4. [DOI] [PubMed] [Google Scholar]
  • 37.Clopath C, Jolivet R, Rauch A, Luscher H, Gerstner W. Predicting neuronal activity with simple models of the threshold type: Adaptive exponential integrate-and-fire model with two compartments. Neurocomputing. 2007;70:1668–1673. [Google Scholar]
  • 38.Jolivet R, Schürmann F, Berger TK, Naud R, Gerstner W, et al. The quantitative singleneuron modeling competition. Biol Cybern. 2008;99:417–426. doi: 10.1007/s00422-008-0261-x. [DOI] [PubMed] [Google Scholar]
  • 39.Brüderle D, Petrovici MA, Vogginger B, Ehrlich M, Pfeil T, et al. A comprehensive workow for general-purpose neural modeling with highly configurable neuromorphic hardware systems. Biol Cybern. 2011;104:263–296. doi: 10.1007/s00422-011-0435-9. [DOI] [PubMed] [Google Scholar]
  • 40.Jeong H, Gutkin BS. Synchrony of neuronal oscillations controlled by GABAergic reversal potentials. Neural Comput. 2007;19:706–729. doi: 10.1162/neco.2007.19.3.706. [DOI] [PubMed] [Google Scholar]
  • 41.Traub RD, Wong RK, Miles R, Michelson H. A model of a CA3 hippocampal pyramidal neuron incorporating voltage-clamp data on intrinsic conductances. J Neurophysiol. 1991;66:635–650. doi: 10.1152/jn.1991.66.2.635. [DOI] [PubMed] [Google Scholar]
  • 42.Goodman D, Brette R. Brian: a simulator for spiking neural networks in Python. Front Neuroinform. 2008;2:5. doi: 10.3389/neuro.11.005.2008. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 43.Wang XJ, Buzsáki G. Gamma oscillation by synaptic inhibition in a hippocampal interneuronal network model. J Neurophysiol. 1996;16:6402–6413. doi: 10.1523/JNEUROSCI.16-20-06402.1996. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.Mormann F. Mean phase coherence as a measure for phase synchronization and its application to the EEG of epilepsy patients. Physica D. 2000;144:358–369. [Google Scholar]
  • 45.Fink CG, Booth V, Zochowski M. Cellularly-driven differences in network synchronization propensity are differentially modulated by firing frequency. PLoS Comput Biol. 2011;7:e1002062. doi: 10.1371/journal.pcbi.1002062. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 46.Ermentrout GB. Type I neurons, phase resettings curves and synchrony. Neural Comput. 1996;8:979–1001. doi: 10.1162/neco.1996.8.5.979. [DOI] [PubMed] [Google Scholar]
  • 47.Brown E, Moehlis J, Holmes P. On the phase reduction and response dynamics of neural oscillator populations. Neural Comput. 2004;16:673–715. doi: 10.1162/089976604322860668. [DOI] [PubMed] [Google Scholar]
  • 48.Ermentrout GB, Terman D. Mathematical foundations of neuroscience. New York: Springer; 2010. 422 [Google Scholar]
  • 49.Schwemmer MA, Lewis TJ. The theory of weakly coupled oscillators. In: Schultheiss N, Prinz A, Butera R, editors. PRCs in neuroscience: theory, experiment and analysis. New York: Springer; 2012. pp. 3–32. [Google Scholar]
  • 50.Hoppenstaedt F, Izhikevich EM. Weakly connected neural networks. New York: Springer; 1997. 400 [Google Scholar]
  • 51.Mueller PC. Calculation of Lyapunov exponents for dynamic systems with discontinuities. Chaos Soliton Fract. 1995;5:1671–1681. [Google Scholar]
  • 52.Samoilenko AM, Perestyuk NA. Impulsive differential equations. Singapore: World Scientific; 1995. 462 [Google Scholar]
  • 53.Akhmet M. Principles of discontinuous dynamical systems. New York: Springer; 2010. 180 [Google Scholar]
  • 54.Ermentrout GB, Kopell NJ. Multiple pulse interactions and averaging in systems of coupled neural oscillators. J Math Biol. 1991;29:195–217. [Google Scholar]
  • 55.Kuramoto Y. Chemical oscillations, waves and turbulence. New York: Springer; 1984. 164 [Google Scholar]
  • 56.Dhooge A, Govaerts W. MATCONT: A MATLAB package for numerical bifurcation analysis of ODEs. ACM T Math Software. 2003;29:141–164. [Google Scholar]
  • 57.Govaerts W, Kuznetsov YA, Sautois B. MATCONT. Scholarpedia. 2006;1:1375. URL http://www.scholarpedia.org/article/MATCONT. [Google Scholar]
  • 58.Gutkin BS, Ermentrout GB, Reyes A. Phase-response curves give the responses of neurons to transient inputs. J Neurophysiol. 2005;94:1623–1635. doi: 10.1152/jn.00359.2004. [DOI] [PubMed] [Google Scholar]
  • 59.Woodman MM, Canavier CC. Effects of conduction delays on the existence and stability of one to one phase locking between two pulse-coupled oscillators. J Comput Neurosci. 2011;31:401–418. doi: 10.1007/s10827-011-0315-2. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 60.Bradley PJ, Wiesenfeld K, Butera RJ. Effects of heterogeneity in synaptic conductance between weakly coupled identical neurons. J Comput Neurosci. 2011;30:455–469. doi: 10.1007/s10827-010-0270-3. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 61.Hansel D, Mato G, Meunier M. Synchronization in excitatory neural networks. Neural Comput. 1995;7:307–337. doi: 10.1162/neco.1995.7.2.307. [DOI] [PubMed] [Google Scholar]
  • 62.Ernst U, Pawelzik K, Geisel T. Synchronization induced by temporal delays in pulse-coupled oscillators. Phys Rev Lett. 1995;74:1570–1573. doi: 10.1103/PhysRevLett.74.1570. [DOI] [PubMed] [Google Scholar]
  • 63.Ermentrout GB, Ko TW. Delays and weakly coupled neuronal oscillators. Phil Trans Soc A. 2009;367:1097–1115. doi: 10.1098/rsta.2008.0259. [DOI] [PubMed] [Google Scholar]
  • 64.Bartos M, Vida I, Jonas P. Synaptic mechanisms of synchronized gamma oscillations in inhibitory interneuron networks. Nat Rev Neurosci. 2007;8:45–56. doi: 10.1038/nrn2044. [DOI] [PubMed] [Google Scholar]
  • 65.van Vreeswijk C, Abbott LF, Ermentrout GB. When inhibition not excitation synchronizes neural firing. J Comput Neurosci. 1994;1:313–321. doi: 10.1007/BF00961879. [DOI] [PubMed] [Google Scholar]
  • 66.Wang S, Chandrasekaran L, Fernandez FR, White JA, Canavier CC. Short conduction delays cause inhibition rather than excitation to favor synchrony in hybrid neuronal networks of the entorhinal cortex. PLoS Comput Biol. 2012;8:e1002306. doi: 10.1371/journal.pcbi.1002306. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 67.Chandrasekaran L, Achuthan S, Canavier CC. Stability of two cluster solutions in pulse coupled networks of neural oscillators. J Comput Neurosci. 2011;30:427–445. doi: 10.1007/s10827-010-0268-x. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 68.Pervouchine DD, Netoff TI, Rotstein HG, White JA, Cunningham MO, et al. Lowdimensional maps encoding dynamics in entorhinal cortex and hippocampus. Neural Comput. 2006;18:2617–2650. doi: 10.1162/neco.2006.18.11.2617. [DOI] [PubMed] [Google Scholar]
  • 69.Fuhrmann G, Markram H, Tsodyks M. Spike frequency adaptation and neocortical rhythms. J Neurophysiol. 2002;88:761–770. doi: 10.1152/jn.2002.88.2.761. [DOI] [PubMed] [Google Scholar]
  • 70.Netoff TI, Acker CD, Bettencourt JC, White JA. Beyond two-cell networks: experimental measurement of neuronal responses to multiple synaptic inputs. J Comput Neurosci. 2005;18:287–295. doi: 10.1007/s10827-005-0336-9. [DOI] [PubMed] [Google Scholar]
  • 71.Izhikevich EM. Dynamical systems in neuroscience. Cambridge: MIT Press; 2007. 497 [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

Text S1

Supplementary Methods. A) Calculation of the PRC using the adjoint method. B) Phase reduction.

(PDF)


Articles from PLoS Computational Biology are provided here courtesy of PLOS

RESOURCES