1: %%%\documentclass[aps,pre,showpacs,floatfix,superscriptaddress]{revtex4}

2: \documentclass[aps,pre,twocolumn,floatfix,superscriptaddress,nofootinbib,showkeys]{revtex4}

3: %%%\documentclass[aps,prl,twocolumn]{article}

4: \usepackage{graphicx}

5: 

6: \newcommand {\be} {\begin{equation}}

7: \newcommand {\ee} {\end{equation}}

8: \newcommand {\Be}{\begin{eqnarray*}}

9: \newcommand {\Ee} {\end{eqnarray*}}

10: \newcommand {\bey} {\begin{eqnarray}}

11: \newcommand {\eey} {\end{eqnarray}}

12: \newcommand{\bit}{\begin{itemize}}

13: \newcommand{\eit}{\end{itemize}}

14: \newcommand{\bfl}{\begin{flusleft}}

15: \newcommand{\efl}{\end{flusleft}}

16: \newcommand{\bfr}{\begin{flushright}}

17: \newcommand{\bc}{\begin{center}}

18: \newcommand{\ec}{\end{center}}

19: \newcommand{\ben}{\begin{enumerate}}

20: \newcommand{\een}{\end{enumerate}}

21: \newcommand{\cl}{\centerline}

22: \newcommand{\ul}{\underline}

23: \newcommand{\nl}{\newline}

24: \renewcommand{\theparagraph}{\Alph{paragraph}}

25: \newcommand{\comment}[1]{}

26: 

27: \begin{document}

28: 

29: \title{Coherent response of the Hodgkin-Huxley neuron in the high-input regime}

30: 

31: \author{Alessandro Torcini}

32: \email{alessandro.torcini@isc.cnr.it}

33: \affiliation{Istituto dei Sistemi

34: Complessi  - CNR,

35: via Madonna del Piano 10, I-50019 Sesto Fiorentino, Italy}

36: \affiliation{INFN Sezione di Firenze,

37: via Sansone, 1 - I-50019 Sesto Fiorentino, Italy}

38: \author{Stefano Luccioli}

39: \affiliation{Istituto dei Sistemi

40: Complessi - CNR,

41: via Madonna del Piano 10, I-50019 Sesto Fiorentino, Italy}

42: \affiliation{INFN Sezione di Firenze,

43: via Sansone, 1 - I-50019 Sesto Fiorentino, Italy}

44: \author{Thomas Kreuz}

45: \affiliation{Istituto dei Sistemi Complessi - CNR,

46: via Madonna del Piano 10, I-50019 Sesto Fiorentino, Italy}

47: 

48: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

49: \begin{abstract}

50: We analyze the response of the Hodgkin-Huxley neuron to a large number of uncorrelated 

51: stochastic inhibitory and excitatory post-synaptic spike trains. 

52: In order to clarify the various mechanisms responsible for noise-induced 

53: spike triggering we examine the model in its silent regime.

54: We report the coexistence of two distinct coherence resonances:

55: the first one at low noise is due to the stimulation of 

56: {\it correlated} subthreshold oscillations; the second one at 

57: intermediate noise variances is instead related to the regularization 

58: of the emitted spike trains.

59: \end{abstract}

60: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

61: 

62: %%%\pacs{05.45.-a,87.10.+e,87.17.Aa,05.40.Ca}

63: % 87.19.La - Neuroscience

64: % 84.35.+i Neural networks

65: % 05.45.-a Nonlinear dynamics and nonlinear dynamical systems

66: % 05.45.Xt Synchronization; coupled oscillators

67: % 87.10.+e General theory and mathematical aspects (biophysics)

68: % 07.05.Mh Neural networks, fuzzy logic, artificial intelligence

69: % 89.75.-k Complex systems

70: % 87.17.Nn Electrophysiology of nerve cells

71: % 87.17.Aa Theory and modeling; computer simulation (cellular process)

72: %05.40.Ca        Noise

73: %05.45.Tp        Time series analysis

74: 

75: \keywords{Hodgkin-Huxley model, Coherence, Noise, Sub- and Suprathreshold oscillations}

76: 

77: \maketitle

78: 

79: \subsection{Introduction}

80: 

81: In the last decades a large number of studies have been

82: devoted to the characterization of the response of simple and more

83: elaborated neuronal models under the influence of a large variety

84: of stochastic inputs~\cite{tuckwell,koch,ger}.

85: One of the main reasons justifying the interest

86: of neuroscientists for this subject resides in the observation

87: that {\it in vivo} neocortical neurons are subjected to a constant

88: bombardment of excitatory and inhibitory post-synaptic potentials

89: (EPSPs and IPSPs), somehow resembling a background noise \cite{destexhe}.

90: 

91:  Among the many proposed biophysical models the Hodgkin-Huxley model~\cite{hh} can still be considered

92: as a reasonably valid framework for exploring the main features of neural 

93: dynamics~\cite{hh_devil}. In order to understand the origin of the

94: variability observed in the distribution of spikes emitted by cortical neurons~\cite{shadlen_newsome_1998}

95: the response of the Hodgkin-Huxley (HH) model has recently been

96: studied under the influence of additive noise~\cite{tiesinga_2000,aguera03,rudolph_2003}

97: and subjected to trains of post-synaptic potentials~\cite{brown_feng_1999,hasegawa,tiesinga_2000,rudolph_2003}.

98: 

99: One of the most interesting phenomena observed experimentally

100: and numerically for excitable neuronal systems driven by noise is {\it coherence resonance} (CR)~\cite{gang,pikovsky_kurths_1997}:

101: it refers to the regularization of noise-induced oscillations occuring for

102: finite (not zero) noise intensity in the absence of any external injected signal

103: (for a comprehensive review see \cite{lindner_ojalvo_2004}).

104: Evidences of CR have been reported experimentally

105: for the cat's spinal and cortical neural ensembles \cite{manjarrez}

106: and theoretically for various neuronal models~\cite{lindner_ojalvo_2004}.

107: In particular, CR has been observed for the HH model~\cite{lee_neiman_1998,yu_wang_2001},

108: but these results mainly refer to additive continuous noise.

109: 

110: Our aim is to perform a detailed analysis of the response of the HH model

111: in the silent regime when it is subjected to many (hundreds or thousands) 

112: stochastic trains of post-synaptic potentials (PSPs) per emitted

113: spike (i.e. the system is in the so-called {\it high-input regime}~\cite{shadlen_newsome_1998,destexhe}).

114: 

115: The different mechanisms responsible for noise-induced neuronal firing are analyzed in

116: terms of statistical indicators (interspike-interval distributions and

117: their first moments) as well as of dynamical indicators (autocorrelation functions).

118: The analysis of the correlation time reveals a CR associated to the stimulations of 

119: coherent subthreshold oscillations around the rest state. This new type of CR coexists with 

120: {\it standard} CR related to the regularization of spike sequences~\cite{lindner_ojalvo_2004}.

121: 

122: 

123: \subsection{Model and methods}

124: \label{model}

125: 

126: The HH model describes the dynamical

127: evolution of the membrane potential $V(t)$. It can be written as

128: \bey

129: C\frac{dV}{dt}&=&g_{Na}m^{3}h(E_{Na}-V)+g_{K}n^{4}(E_{K}-V)+

130: \label{hh}

131: \\

132: &+&g_{L}(E_{L}-V) + I(t)\quad,

133: \nonumber

134: \eey

135: where $I(t)$ represents the synaptic current and the evolution of the

136: gating variables $X=m,n,h$ is ruled by three ordinary

137: differential equations of the form

138: $ dX/dt=\alpha_X(V)(1-X)-\beta_X(V)X $.

139: The parameters entering in Eq. (\ref{hh}) 

140: and the expressions of the nonlinear functions $\alpha_X(V)$ and

141: $\beta_X(V)$ are reported in \cite{noi}.

142: 

143: In this study we consider the response of the single HH model to $N_e$ (resp. $N_i$) uncorrelated 

144: trains of EPSPs (resp. IPSPs) with $N_e (N_i)$  ranging from 10 to 10,000.

145: By following Refs.~\cite{brown_feng_1999,feng_zhang_2001},

146: each PSP is schematized as an instantaneous

147: variation of the membrane potential by a positive (resp. negative)

148: amount $\Delta V$ for excitatory (resp. inhibitory) synapses.

149: The amplitude of each voltage kick is assumed to be 0.5 mV, i.e.

150: reasonably small ($\approx 7$ \%) with respect to the distance

151: between the "threshold" for spike initiation for rapid EPSPs

152: and the resting potential ($\approx 6 - 7 $ mV) estimated for the HH model~\cite{noble&stein,tuckwell}.

153: Moreover, amplitudes $\approx 0.5$ mV are comparable with average

154: EPSPs experimentally measured for pyramidal neurons in the visual cortex of rats~\cite{koch}.

155: This amounts to exciting the neuron (\ref{hh}) with an impulsive current

156: \begin{equation}

157: I(t)=Q\Big[\sum_{k=1}^{N_{e}}\sum_{l} \delta(t-t_{k}^{l})

158: -\sum_{m=1}^{N_{i}}\sum_{n} \delta(t-t_{m}^{n})\Big]

159: \label{I_imp}

160: \end{equation}

161: where $t_{k}^{l}$ (resp. $t_{m}^{n}$) are the arrival times of the

162: EPSPs (resp. IPSPs) and $Q=C \Delta V$ is the charge

163: associated to each kick. 

164: In order to mimic the inputs received by cortical

165: neurons, for each afferent (excitatory and inhibitory) synapse the time interval distribution between PSP

166: inputs is chosen to be Poissonian with an average frequency $\nu_0 = 100$ Hz~\cite{shadlen_newsome_1998}

167: \footnote{It should be stressed that the model here analyzed is driven by

168: impulsive post-synaptic currents and it should not be confused

169: with conductance-driven models~\cite{tiesinga_2000,rudolph_2003}.}.

170: 

171: The stochastic input can be characterized in terms of the average and the variance of the net

172: spike count within a temporal window $\Delta T$

173: \begin{equation}

174: N(\Delta T) = \sum_{k=1}^{N_e} n^E_k(\Delta T) - \sum_{m=1}^{N_i} n^I_m(\Delta T)

175: \label{sp_c}

176: \end{equation}

177: where $n^E_k(\Delta T)$ (resp. $n^I_m(\Delta T)$) represents the

178: number of afferent EPSPs (resp. IPSPs) received from neuron $k$ (resp. $m$)

179: in the interval $\Delta T$. In the high input regime, by assuming uncorrelated

180: input spike trains, $N(\Delta T)$ follows a Gaussian distribution (cf. \cite{noi}),

181: with average and variance derivable within the shot noise formalism as

182: \begin{equation}

183: < N(\Delta T) >=(N_e-N_i) \Delta T \nu_0 

184: \end{equation}

185: and

186: \begin{equation}

187: Var[N(\Delta T)]=(N_e+N_i) \Delta T \nu_0 = \sigma^2 \Delta T \nu_0.

188: \end{equation}

189: The parameter $\sigma=\sqrt{(N_e+N_i)}$ measures the standard deviation

190: of the stochastic input process.

191: 

192: The average current stimulating the neuron is given by

193: \begin{equation}

194: \bar I= \frac{C \Delta V <N(\Delta T)>}{\Delta T} = C \Delta V

195: \nu_0 (N_e - N_i).

196: \label{I_ave}

197: \end{equation}

198: $\bar I$ represents the bifurcation parameter ruling the dynamics of the deterministic 

199: HH model; for $\bar I < I_{SN} \simeq 6.27 \mu$A/cm$^2$ the model is in the {\it silent regime},

200: i.e., in the absence of noise the dynamics is attracted towards a stable fixed point 

201: and the neuron does not fire. However, since the fixed point is a focus

202: the relaxation towards the rest state occurs via damped 

203: oscillations ({\it subthreshold oscillations}) 

204: \cite{rinzel_miller_1980}. Periodic firing is observed for the HH model only above 

205: $I_{SN}$ and it is associated to frequencies in the range $50-80$ Hz.

206: 

207: We have verified that in the high input regime the response of the

208: neuron (for fixed $\nu_0$) is determined once fixed $\bar I$ and

209: $\sigma$, therefore it does not depend separately on $N_e$  and $N_i$,

210: but only on their difference $(N_e - N_i) \propto \bar I$ 

211: and sum $(N_e + N_i) \propto \sigma^2$~\cite{noi}. 

212: 

213: In order to characterize the output of the neuron and

214: to examine the coherence effects in the response we

215: have employed the following indicators:

216: the distribution of the output interspike intervals

217: $P_{ISI}(t)$ and its first moments; 

218: the coefficient of variation of the output interpike intervals distribution ($P_{ISI}(t)$) defined as

219: $R={S_{ISI}}/{A_{ISI}} $, where $A_{ISI}$ is the average 

220: and $S_{ISI}$ the standard deviation of the $P_{ISI}(t)$ (for a perfectly periodic response we have $R=0$ and $R=1$ for Poissonian output;

221: the normalized autocorrelation function $C(t)$ for the membrane potential and the

222: correlation time~\cite{pikovsky_kurths_1997} defined as

223: \begin{equation}

224: \tau_c= \int_0^\infty C^2(t) \enskip dt

225: \quad .

226: \label{tauc}

227: \end{equation}

228: 

229: CR is usually identified by a minimum (resp. a maximum) in the coefficient of variation 

230: $R$ (resp. in $\tau_c$) occurring at finite noise variance \cite{lindner_ojalvo_2004}.

231: 

232: \subsection{Results}

233: 

234: The HH neuron subjected to stochastic input exhibits 

235: a noise-induced (irregular) spiking behaviour with an average firing rate $\nu_{out}=1/A_{ISI}$.

236: For fixed $\bar I$ the firing rate increases with 

237: the standard deviation of the noise (cf. Fig.~\ref{ivar_ave.total}). 

238: 

239: The noise-induced firing activity becomes 

240: more and more pronounced by approaching the repetetive firing bifurcation 

241: $I_{SN}$ for the deterministic HH. The HH model is a type II neuron, therefore

242: the onset of oscillation at $I \simeq I_{SN}$ is associated with a finite frequency

243: ($\simeq 50$ Hz), however, in presence of noise, arbitrarily low spike rates can be observed

244: even in the silent regime.

245: 

246: \begin{figure}[h]

247: \begin{center}

248: \includegraphics*[width=6cm]{ivar_ave.total.eps}

249: \end{center}

250: \caption{Silent regime: average firing rate $\nu_{out}$ as 

251: a function of $\sigma$ for increasing values of $\bar I$ 

252: (from the bottom to the top $\bar I$=-1,0,1,2,3,4,5,6.15 $\mu$A/cm$^2$).}

253: \label{ivar_ave.total}

254: \end{figure}

255: 

256: \subsubsection{Response of the HH model in the low and high noise limit}

257: \label{response}

258: 

259: Let us now examine more in detail the origin of noise-induced spiking

260: in the low and high noise limits.

261: For low noise we observe the coexistence of

262: multi-peaks and an exponential tail in the $P_{ISI}(t)$ (cf. Fig.~\ref{multi}). 

263: As explained in the following the multi-peak structure

264: is related to the relaxation dynamics of the membrane potential $V(t)$

265: following a spike emission, while the exponential tail is associated with noise 

266: induced activation processes.

267: \begin{figure}[h]

268: \begin{center}

269: \includegraphics*[width=6cm]{prova}

270: \end{center}

271: \caption{$P_{ISI}(t)$ obtained for the stochastic input

272: (\ref{I_imp}) with $\sigma=4.6$ (continuous curve) and the

273: relaxation dynamics of the membrane potential $V(t)$ following a step 

274: current stimulation (dashed line). The

275: position of the output spike has been shifted to $t=0$ and

276: the action potential has been rescaled to better

277: reveal the relaxation oscillations. In the inset

278: the exponential tail due to the activation mechanism

279: is shown. The data refer to $\bar I =6.15 \mu$A/cm$^2$.

280: } \label{multi}

281: \end{figure}

282: 

283: To better understand the first mechanism we have studied the evolution of $V(t)$

284: following a step current of amplitude $\bar I$, i.e.,

285: \begin{equation}

286: I(t)=\left\{

287: \begin{array}{ll} 0  &

288: \mbox{if  $t \le 0$}     \\

289: \bar I

290: & \mbox{if $t>0$}

291: \end{array} \right.

292: \qquad .

293: \label{step}

294: \end{equation}

295: As shown in Fig.\ref{multi} (dashed line) $V(t)$ exhibits one spike followed by damped oscillations.

296: This suggests that the probability of eliciting a second spike is enhanced in

297: correspondence the maxima of the relaxation oscillations, where $V(t)$ is nearest to the firing threshold.

298: Moreover, it is possible to show that the period $T_{nl}$ of the first oscillation

299: has a clear nonlinear origin, while the period of successives oscillations $T_l$ can be explained by 

300: a linear stability analysis around the fixed point.

301: Indeed we have verified that the first peak ($T_{fp}$) observed in the distribution $P_{ISI}(t)$ 

302: (continuous curve in Fig.\ref{multi}) is related to $T_{nl}$,

303: while the successives peaks in the $P_{ISI}(t)$ correspond to the

304: linear oscillations of period $T_l$ \cite{noi}. 

305: 

306: The second firing mechanism responsible for the exponential tail in the $P_{ISI}(t)$ is due to the

307: competition between two effects, on one side the tendency of the dynamics to relax

308: towards its stable fixed point and on the other side noise fluctuations that instead lead

309: the system towards the firing threshold. 

310: In this framework the dynamics of $V(t)$ can be described as the overdamped dynamics

311: of a particle in a potential well under the influence of thermal

312: fluctuations \cite{Kramers} and the average firing time, or activation time $T_a$, can be expressed 

313: in terms of the Kramers expression \cite{Kramers} for sufficiently small noise:

314: \begin{equation}

315: T_a \propto {\rm e}^{W/\sigma^2}

316: \quad ,

317: \label{rate}

318: \end{equation}

319: where $\sigma^2$ plays the role of

320: an effective temperature and $W$ of an energy barrier.

321: For $\sigma^2 < W$ , i.e., in the low noise limit, the dynamics can be characterized as an activation

322: process: $P_{ISI}(t)$ resembles a Poissonian distribution with $R \approx 1$.

323: For $\sigma^2 > W$, i.e., in the high noise limit, the multi-peak structure and the exponential tail 

324: disappear and $P_{ISI}(t)$ reduces to an inverse Gaussian corresponding to

325: the distribution of the first passage times for a diffusive process plus drift~\cite{tuckwell}.

326: 

327: As a further indicator we have estimated

328: the spike triggered average potential (STAP)~\cite{bialek}, that gives the average shape of $V(t)$ preceding the emission of a spike,

329: for sufficiently long ISIs and for small $\sigma$.

330: As shown in Fig.~\ref{stap}, $V(t)$ oscillates with period $\approx T_l$

331: before firing; therefore the emission of a spike (for long ISIs) is triggered by

332: the excitation of linear subthreshold oscillations around the fixed point.

333: Thus the HH neuron nearby the rest state acts as a sort of selective filter

334: since it responds (by emitting a spike) with higher probability when it is stimulated 

335: with a specific input frequency ($\approx 1/T_l=61-88$ Hz for $ 0 \le \bar I \le I_{SN}$).

336: This result agrees with the analysis of~\cite{yu_wang_2001}

337: where it has been shown that a silent HH neuron subjected

338: to a sinusoidal current optimally resonates when forced with a

339: frequency linearly correlated with $1/T_l$.

340: 

341: \begin{figure}[h]

342: \begin{center}

343: \includegraphics*[clip,width=6cm]{f7}

344: \end{center}

345: \caption{$\bar I= 5 \mu A/cm^2$ and $\sigma=5.7$. STAP preceding the emission of a spike.  At time $t=0$

346: the potential overcomes the spike detection threshold (fixed at $-5$ mV).

347: STAP has been calculated by averaging only over ISIs longer

348: than $150$ ms.

349: }

350: \label{stap}

351: \end{figure}

352: 

353: 

354: \subsubsection{Coherent response of the Hodgkin-Huxley neuron}

355: \label{tre2}

356: 

357: In the silent regime we have found the coexistence of two CRs: one is related to the regularization of

358: the emitted spike trains at intermediate noise levels and 

359: has previously been reported in~\cite{lee_neiman_1998,yu_wang_2001}; the

360: second one, observed for the first time, occurs at very low noise and is associated to the coherence 

361: of subthreshold oscillations.

362: 

363: The first CR can be explained by noticing that the dynamics of excitable systems

364: can be characterized in terms of two characteristic times~\cite{pikovsky_kurths_1997}: 

365: an activation time $T_a$, representing 

366: the time needed to excite the system, and an excursion time $T_e$, 

367: indicating the duration of the spike (i.e., the time spent in the excited state).

368: An ISI is given by the sum of these two times and therefore $R$ can be (formally) expressed as the 

369: sum of two parts depending separately on $T_a$ and $T_e$. 

370: These two contributions vary in an opposite way when the noise is increased and

371: the competition of these two tendencies leads to the maximal coherence (associated to a minimum in $R$) 

372: for finite noise~\cite{pikovsky_kurths_1997,lindner_ojalvo_2004}.

373: So there is an initial range of $\sigma$-values where by increasing the noise the spike emission is facilitated 

374: ($T_a$ reduces accordingly to Eq. (\ref{rate})) and this renders the response of the neuron more and more regular.

375: On the other side at high $\sigma$-values the response becomes again more irregular, because 

376: the noise induces firing even during the relative refractory period

377: thus modifying even the duration $T_e$ of the spike itself.

378: To summarize, the activation process 

379: responsible for firing is gradually

380: substituted by a diffusive process with drift and at the transition from one mechanism to the other

381: a regularization of the output signal is observed.

382: As shown in Fig.~\ref{ivar_tauc_cv}, for different values of $\bar I$, both the coefficient of variation $R$ and the signal correlation time $\tau_c$ 

383: were able to identify clearly the first kind of coherence in the silent regime.  

384: We can also see that $\tau_c$ increases and $R$ decreases when $\bar I \rightarrow I_{SN}$, i.e., the coherence of the 

385: emitted spike trains increases by approaching the repetetive firing regime.

386: \begin{figure}[h]

387: \begin{center}

388: %\includegraphics*[width=6cm]{ivar_tauc_cv_hor}

389: \includegraphics*[width=8cm]{ivar_tauc_cv_ver}

390: \end{center}

391: \caption{Coherence of the emitted spike trains for increasing values of $\bar I$ in the silent regime. (a) $\tau_c$ 

392: (from bottom to top $\bar I$=-1,0,1,2,3,4,5 $\mu$A/cm$^2$). (b) $R$ (from top to 

393: bottom $\bar I$=-1,0,1,2,3,4,5 $\mu$A/cm$^2$ ).}

394: \label{ivar_tauc_cv}

395: \end{figure}

396: 

397: The second kind of resonance could only be detected by $\tau_c$.

398: In fact this property is not related to spike emission (suprathreshold dynamics) but to the behaviour of the signal below the firing threshold.

399: In Fig.\ref{tauI4}a the behaviour of $\tau_c$ for $\bar I= 4 \mu A/cm^{2}$ is shown in a wider range of noise 

400: with respect to Fig.~\ref{ivar_tauc_cv}a. 

401: In this range $\tau_c$ exhibits indeed two maxima: the first and higher maximum 

402: at low noise ($\sigma \approx 3$) is related to the coherence of subthreshold oscillations, while the second one 

403: at intermediate noise ($\sigma \approx 33$) is due to the CR just discussed above.

404: 

405: \begin{figure}[h]

406: \begin{center}

407: \includegraphics*[width=4.2cm]{f13a}

408: \includegraphics*[width=4.2cm]{timeautocorr}

409: \end{center}

410: \caption{$\bar I= 4 \mu A/cm^{2}$: (a) correlation time $\tau_c$ 

411: (empty circles) as a function of the standard deviation of 

412: the input noise $\sigma$: (b) autocorrelation function

413: $C(t)$ for three increasing noise values $\sigma=2.9$ (continuous line), 4.5 (dotted line) and 9.7 (dashed line); 

414: the vertical dot-dashed lines indicates the period $T_l$ and the period $T_{fp}$ for $\sigma=9.7$.}

415: \label{tauI4}

416: \end{figure}

417:  

418:  Let us now gain deeper insight about the origin of these two maxima.

419:  For extremely low noise ($\sigma < 3$) the neuron rarely fires, i.e., $T_a \rightarrow \infty$; increasing the noise the subthreshold oscillations generated 

420:  by the input kicks are more and more correlated.

421:  For $\sigma > 3$ the occurrence of rare spikes tends to decorrelate the signal leading to a

422:  decrease of $\tau_c$. 

423:  With a further increase of the noise $V(t)$ becomes essentially a sequence of spikes and 

424:  $C(t)$ represents the correlation between them; a second peak appears in $\tau_c$ indicating the 

425: maximal coherence of the spike sequence.

426: 

427: To investigate the transition from one to the other behaviour let us analyze the autocorrelation function of $V(t)$

428: for increasing value of $\sigma$, as shown in Fig.~\ref{tauI4}b.

429: For $\sigma \approx 3$ the autocorrelation function $C(t)$ oscillates

430: with a period $\approx T_l$ (see Sec.\ref{response}), while at $\sigma \approx 9.7$ the maxima of

431: $C(t)$ correspond to multiples of $T_{fp}$, i.e., the first peak of $P_{ISI}(t)$.

432: In between these two values there is a transition where the course of $V(t)$, initially consisting of 

433: mere subthreshold oscillations, is dominated by the spikes.

434: The transition is observed for $\sigma \approx 4.5$, where in correspondence to the first oscillation  $C(t)$ reveals 

435: two maxima, one located at $t \approx T_l$ and another at $t \approx T_{fp}$.

436: 

437: 

438: 

439: \subsection{Conclusions}

440: 

441: We have presented an analysis of the response of the silent HH model in the {\it high-input regime}. In this regime the HH neuron 

442: displays a large variety of dynamical behaviors and its response is completely determined by the average and the variance 

443: of the stochastic input.  Our main result is the coexistence of two different coherence resonances:

444: the first one, at intermediate noise levels, associated with the regularization of the spikes emitted by the neuron;

445: the second one, at very low noise, related to the coherence of subthreshold oscillations around the fixed point.

446: The first one can be revealed using both the ISI coefficent of variation $R$ and the autocorrelation time 

447: of the signal $\tau_c$. The second type of resonance, observed for the first time, is not related to

448: spike emission and can thus only be detected by means of $\tau_c$.

449: 

450: Moreover, we have examined the behaviour of the HH neuron for low and high input noise variance. 

451: The richness of the HH dynamics is 

452: particularly pronounced for low input noise where we have found the coexistence of

453: a multimodal structure and an exponential tail in the $P_{ISI}(t)$.

454: The peaks in the $P_{ISI}(t)$ suggests that the system,

455: under the influence of stochastic inputs, can resonate when forced 

456: with specific frequencies: the first peak is associated to frequencies in the $\gamma$-range~\cite{sheperd}

457: (namely, from $40$ to $66$ Hz for $\bar I \in [0:9] \enskip \mu A/cm^2$),

458: and the second one to lower frequencies (namely, from $30$ to $37$ Hz

459: for the same interval of currents) \cite{noi}.

460: 

461: In the silent regime the responsiveness of the single neuron is enhanced by

462: stochastic stimulations, in particular the regularity of the emitted

463: spike trains is maximal in correspondence to an optimal

464: noise level. Moreover, we expect that a population of such neuronal elements will have the

465: capability to exhibit coherent and correlated activity over

466: different time scales (mainly in the $\gamma$ and $\beta$-ranges

467: ~\cite{sheperd}), a property that is believed to be important for

468: information encoding for cortical

469: processing~\cite{salinas_seinoski_nature}. Indeed it has been

470: found ~\cite{yu_wang_2001} that for sufficiently

471: strong synaptic coupling a globally coupled HH

472: network subjected to stochastic inputs reveals a collective synchronized rhythmic

473: firing in a range of $40-60$ Hz, induced via CR.

474: 

475: To proceed in the direction of more realistic situations the present analysis performed for 

476: a current-driven model should be extended by considering conductance-based synaptic 

477: inputs~\cite{tiesinga_2000,rudolph_2003}. It is surely worth to address this issue in forthcoming

478: studies, because the response of these two classes of models to noise fluctuations can sometimes be

479: even opposite, as shown in~\cite{tiesinga_2000} for the output firing rate.

480: 

481: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

482: %       References

483: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

484: 

485: \begin{thebibliography}{9}

486: 

487: \bibitem{aguera03} B. Ag{\"u}era y Arcas, A.L. Fairhall, and

488: W. Bialek, Computation in a single neuron: Hodgkin-Huxley revisited,

489: Neural Computation \textbf{15}, (2003) 1715-1749.

490: 

491: \bibitem{brown_feng_1999} D. Brown, J. Feng and S. Feerick,

492: Variability of Firing of Hodgkin-Huxley and FitzHugh-Nagumo Neurons with Stochastic Synaptic Input,

493: \prl \textbf{82} (1999) 4731-4734.

494: 

495: \bibitem{destexhe} A. Destexhe, M. Rudolph and D. Par\'{e}, 

496: The high-conductance state of neocortical neurons {\it in vivo},

497: Nature Rev. Neurosci.  \textbf{4}, (2003) 739-751.

498: 

499: \bibitem{ger} W. Gerstner and W. Kistler, {\em Spiking Neuron Models},

500: (Cambridge University Press, Cambridge, 2002).

501: 

502: \bibitem{hh} A.L. Hodgkin and A.F. Huxley, 

503: A qualitative description of membrane current and its application to

504: conduction and excitation in nerve, J. Physiol. (London) 117 (1952) 500-544.

505: 

506: \bibitem{koch} C. Koch, {\em Biophysics of computation},

507: (Oxford University Press, New York, 1999).

508: 

509: \bibitem{feng_zhang_2001} J. Feng and P. Zhang,

510: Behavior of integrate-and-fire and Hodgkin-Huxley models

511: with correlated inputs, Phys. Rev. E, \textbf{63} (2001) 051902.

512: 

513: \bibitem{gang} H. Gang, T. Ditzinger, C.Z. Ning, and H. Haken,

514: Stochastic resonance without external periodic force,

515: \prl \textbf{71} (1993) 807-810.

516: 

517: \bibitem{hasegawa} H. Hasegawa, 

518: Responses of a Hodgkin-Huxley neuron to various types of spike-train inputs,

519: \pre \textbf{61} (2000) 718-726.

520: 

521: \bibitem{lee_neiman_1998} S.G Lee, A. Neiman, and S. Kim,

522: Coherence resonance in a Hodgkin-Huxley neuron,

523: \pre \textbf{57} (1998) 3292-3297.

524: 

525: \bibitem{lindner_ojalvo_2004} B. Lindner, J. Garc\`ia Ojalvo,

526: A. Neiman, and L. Schimansky-Geier, Effects of noise in excitable systems, 

527: Physics Reports \textbf{392}, (2004) 321-424.

528:  

529: \bibitem{noi} S. Luccioli, T. Kreuz, and A. Torcini,

530: Dynamical response of the Hodgkin-Huxley model in the high-input regime,

531: Phys. Rev. E \textbf{73} (2006) 041902.

532: 

533: \bibitem{noble&stein} D. Noble and R.B. Stein,

534: The threshold conditions for initiation of action potentials by excitable cells,

535: J. Physiol. \textbf{187} (1966) 129-161.

536: 

537: \bibitem{Kramers} N.G. van Kampen, {\it Stochastic Processes in Physics and Chemistry},

538: Elsevier, North Holland,  (2001)

539: 

540: \bibitem{manjarrez} E. Manjarrez , J.G. Rojas-Piloni, I. M\'endez, L. Mart\'inez, D. V\'elez, D. V\'azquez and A. Flores,

541: Internal stochastic resonance in the coherence between spinal and cortical neuronal ensembles in the cat,

542: Neurosci. Lett. \textbf{326} (2002) 93-96.

543: 

544: \bibitem{hh_devil} C. Meunier and I. Segev, 

545: Playing the devil's advocate: is the Hodgkin-Huxley model useful?,

546: TRENDS Neurosci. \textbf{25} (2002) 558-63.

547: 

548: \bibitem{pikovsky_kurths_1997} A.S. Pikovsky and J. Kurths,

549: Coherence Resonance in a Noise-Driven Excitable System,

550: \prl \textbf{78} (1997) 775-778.

551: 

552: \bibitem{bialek} F. Rieke, D. Warland, Rob de Ruyter van Steveninck e W. Bialek, 

553: {\em Spikes: exploring the neural code}, 

554: MIT: Cambridge, Massachusetts (1996).

555: 

556: \bibitem{rinzel_miller_1980} J. Rinzel and R. Miller,

557: Numerical calculation of stable and unstable periodic solutions to the Hodgkin-Huxley equations,

558: Math. Biosci. \textbf{49} (1980) 27-59.

559: 

560: \bibitem{rudolph_2003} M. Rudolph and A. Destexhe,

561: The discharge variability of neocortical neurons during high-conductance state,

562: Neuroscience {\bf 119} (2003) 855-873

563: 

564: \bibitem{salinas_seinoski} E. Salinas and T.J. Sejnowski,

565: Impact of correlated synaptic input on output firing rate

566: and variability in simple neuronal models,

567: The Journal of Neuroscience {\bf 20} (2000) 6193-6209.

568: 

569: \bibitem{salinas_seinoski_nature} E. Salinas and T.J. Sejnowski,

570: Correlated neuronal activity and the flow of neural information,

571: Nature Rev. Neurosci. \textbf{2} (2001) 539-550.

572: 

573: \bibitem{shadlen_newsome_1998} M.N. Shadlen and W.T. Newsome,

574: The variable discharge of cortical neurons: implications

575: for connectivity, computation and information coding,

576: J.  Neurosci. \textbf{18}, (1998) 3870-3896.

577: 

578: \bibitem{sheperd} {\it The synaptic organization of the brain},

579: ed. G.M. Sheperd (Oxford University Press, New York, 2004)

580: 

581: \bibitem{tiesinga_2000} P.H.E. Tiesinga, J.V. Jos\`e, and T.J.

582: Sejnowski, 

583: Comparison of current-driven and conductance-driven neocortical model neurons

584: with Hodgkin-Huxley voltage-gated channels,

585: \pre \textbf{62}, (2000) 8413-8419

586: 

587: \bibitem{tuckwell} H.C Tuckwell, {\em Introduction to theoretical

588: neurobiology}, (Cambridge University Press, New York, 1988).

589: 

590: \bibitem{yu_wang_2001} Y. Yu, W. Wang, J. Wang, and F. Liu,

591: Resonance-enhanced signal detection and transduction in the Hodgkin-Huxley neuronal systems,

592: \pre \textbf{63}, (2001) 021907.

593: 

594: \end{thebibliography}

595: 

596: 

597: 

598: \end{document}

599: 

600: