q-bio0309024/net.tex
1: \documentstyle[namedreferences,epsfig]{kluwer}
2: 
3: \setlength{\textwidth}{6.5 in} 
4: \setlength{\oddsidemargin}{0.5 in}
5: \setlength{\evensidemargin}{0.5 in} 
6: \setlength{\topmargin}{-0.3 in}
7: \setlength{\textheight}{9.0 in} 
8: \setlength{\headheight}{0.25 in}
9: \setlength{\headsep}{0.2 in} 
10: \setlength{\footheight}{0. in}
11: \renewcommand{\baselinestretch}{1.1} 
12: 
13: \newcommand{\et}{{\em et al.}}
14: \newcommand{\ul}{\underline}
15: 
16: %\runningtitle{Synchronization and oscillatory dynamics in
17: %heterogeneous neurons}    
18: 
19: \begin{opening}
20: 
21: \title{Synchronization and oscillatory dynamics in heterogeneous,
22: mutually inhibited neurons} 
23: 
24: 
25: % You can split the title and subtitle by putting 
26: % two backslashes at the appropriate place.            
27: 
28: \author{John A. \surname{White}}
29: \institute{Department of Biomedical Engineering,
30: Boston University, Boston, MA 02215}
31: \author{Carson C. \surname{Chow}}
32: \author{Jason \surname{Ritt}}
33: \author{Cristina \surname{Soto-Trevi\~no}}
34: \author{Nancy \surname{Kopell}}
35: \institute{Department of Mathematics,
36: Boston University, Boston, MA 02215}
37: 
38: % If there are more authors at one institute, you should first
39: % use \author{...} for each author followed by \institute{...}.
40: 
41: 
42: \date{\today}
43: % Gives you the possibility to enter a date. If you leave the
44: % space between the curly braces empty, no date is printed.
45: 
46: \end{opening}
47: 
48: 
49: \begin{document}  
50: 
51: 
52: \maketitle
53: 
54: \begin{abstract}
55: We study some mechanisms responsible for synchronous 
56: oscillations and loss of synchrony at physiologically relevant
57: frequencies (10-200 Hz)   
58: in a network of heterogeneous inhibitory neurons.
59: We focus on the factors that determine the level of synchrony and
60: frequency of the network response, as well as the effects of 
61: mild heterogeneity on network dynamics.  With
62: mild heterogeneity, synchrony is never perfect and is relatively
63: fragile. 
64: In addition, the effects of inhibition are more complex in 
65: mildly heterogeneous networks than in homogeneous ones.
66: In the former, synchrony is broken in two distinct 
67: ways, depending on the ratio of the synaptic decay time to the 
68: period of repetitive action potentials ($\tau_s/T$), where
69: $T$ can be determined either from the network or from a single,
70: self-inhibiting neuron.
71: With $\tau_s/T > 2$, corresponding to large applied current, small
72: synaptic strength or large synaptic decay time, the effects of inhibition 
73: are largely tonic and heterogeneous neurons spike relatively independently.
74: With $\tau_s/T < 1$, synchrony breaks when faster cells begin to
75: suppress their less excitable neighbors; cells that fire remain nearly
76: synchronous.  
77: We show numerically 
78: that the behavior of mildly heterogeneous networks can be related to
79: the behavior of single, self-inhibiting cells,
80: which can be studied analytically.
81: \end{abstract}
82: 
83: 
84: 
85: \section{Introduction}
86: Synchronous activity has been observed in many regions of the
87: brain and has been implicated as a correlate of
88: behavior and cognition~\cite{gray94,llinas93}.  In
89: the hippocampal formation, where such activity has been
90: studied most thoroughly, neurons
91: discharge in several behaviorally important synchronous
92: rhythms~\cite{buzsaki86}.
93: Among these patterns are the theta (4-12 Hz)
94: and gamma ($20-80$ Hz) rhythms, which appear as nested rhythms under
95: conditions of active exploration and paradoxical sleep, as well as
96: hippocampal sharp waves ($\sim 0.5$ Hz), which occur along with
97: embedded fast ripples ($\sim 200$ Hz) under conditions of rest and
98: slow wave sleep~\cite{bragin95,ylinen95}.  Here, we investigate some
99: mechanisms responsible for generating synchronous oscillations throughout
100: the physiologically relevant range of frequencies (10-200 Hz).
101: 
102: Two crucial results point to the importance of inhibitory
103: interneurons in generating synchronous rhythms in the hippocampal
104: formation.  First, it has been shown in intact animals that interneurons 
105: fire robustly and synchronously in both the theta-gamma state and
106: in the sharp wave-ripple state~\cite{bragin95,ylinen95}.  
107: Second, {\it in vitro} experiments
108: have demonstrated that a functional network containing interneurons 
109: alone can support synchronous gamma activity~\cite{whit95}.
110: These and other experimental results have spurred both 
111: analytic~\cite{ernst,hansel95,gerstner96b,vrees94}
112: and numerical~\cite{jeff96,traub96,traub96b,wang96,whit95}
113: studies of synchrony among neurons.
114: Among the principal conclusions of such studies are that
115: stable synchrony is supported by inhibition that is slow
116: compared with neuronal firing rates; and that firing rate 
117: decays linearly, eventually saturating, as a function of the decay 
118: time constant of inhibition ($\tau_s$). 
119: When the synaptic coupling is
120: extremely fast,  the coupling tends to push the
121: neurons towards
122: anti-synchrony~\cite{friesen94,perkel74,skinner94,vrees94,wang92}.
123: 
124: Synchronous oscillations generated {\it in vivo} are almost certainly 
125: the product of interactions among neurons with some (unknown) degree 
126: of heterogeneity in excitatory drive and intrinsic excitability.
127: Much of the earlier work in the area
128: has not explored the effects of heterogeneity in intrinsic spike
129: rates~\cite{ernst,gerstner96b,jeff96,traub96,traub96b,vrees94,whit95}.
130: Tsodyks \et (1993) considered a network of 
131: integrate-and-fire oscillators with heterogeneous external drive and
132: all-to-all
133: {\em excitatory} coupling. They found that for an infinite number of
134: oscillators, those with an external drive below a critical value would be
135: synchronized and those above the critical value would be asynchronous.
136: This co-existence around the critical value persisted in the limit of
137: vanishing heterogeneity. 
138: Golomb and Rinzel (1993) considered a heterogeneous network of
139: all-to-all coupled
140: inhibitory bursting neurons and found regimes of synchronous,
141: anti-synchronous and asynchronous behavior when the width of the
142: heterogeneity was changed. 
143: They considered a parameter regime that was synchronous for small
144: heterogeneity.  Wang and
145: Buzs\'aki~(1996) considered a 
146: hippocampal interneuron network with heterogeneity in the external
147: drive and network connectivity.  They found numerically that for a
148: physiologically plausible parameters, coherent activity is only
149: possible in the gamma range of frequencies.
150: 
151: Our purpose here is to understand more fully the implications of
152: small levels of heterogeneity 
153: for the degradation of synchrony in networks of inhibitory fast
154: spiking neurons 
155: and the mechanisms by which this degradation occurs.
156: To this end, we have begun a coordinated set of analytic and numerical
157: studies of the problem.  In this paper, we numerically analyze a
158: network of interneurons applicable to the CA1 region of the
159: hippocampus.  We consider slow inhibition and heterogeneity in the
160: external drive.  We find that small amounts of
161: heterogeneity in the external drive can greatly reduce coherence. 
162: In addition,  we find that coherence can be reduced in two qualitatively
163: different ways depending on the parameters -- either by a transition to
164: {\em asynchrony} where the cells fire independently of each other, or
165: through 
166: {\em suppression} where faster cells suppress slower cells. 
167: 
168: The reaction of a network to heterogeneity is shown in the paper
169: to be correlated with the dependence of firing frequency on the
170: time constant of synaptic decay.
171: We find in self-inhibiting cells or synchronous networks
172: that this dependence divides into two asymptotic regimes.  
173: In the first (the tonic-inhibition or {\em tonic} regime),
174: inhibition acts as if it were steady-state and only weakly affects discharge
175: frequency.  
176: In the second (the phasic-inhibition or {\em phasic} regime), 
177: time-varying
178: inhibition firmly controls discharge frequency.
179: There is a gradual crossover between these regimes.
180: The presence of
181: a neuron or network in the tonic or phasic regime can most easily
182: be determined  
183: by examining the ratio of the synaptic decay time constant
184: to discharge period ($\tau_s/T$).
185: (Discharge period $T$ can be obtained from the full network
186: or from a reduced model including only a single cell with 
187: self-inhibition.)
188: $\tau_s/T$ is large ($> 2$ for our parameters) and 
189: varies linearly with $\tau_s$ 
190: in the tonic regime.
191: $\tau_s/T$ is small ($< 1$) and only logarithmically 
192: dependent on $\tau_s$ in the phasic regime.  
193: However, if $\tau_s$ is {\em too} small ($<< 1$), the phasic regime
194: is departed and anti-synchrony is possible.
195: Networks of weakly heterogeneous (less than 5\%)  cells
196: generally exhibit   
197: asynchrony (defined here as the state of phase dispersion) 
198: in the tonic regime. 
199: In the phasic regime, cells generally exhibit a form of locking,
200: including synchrony, harmonic locking (locking at rational ratios),
201: and suppression.
202: These results can be demonstrated
203: analytically using a reduced model 
204: with mutual and self-inhibition~\cite{cc}.
205: 
206: We conclude 
207: that mild heterogeneity in inhibitory networks adds effects
208: that are not accounted for in previous analyses, but that are
209: tractable under our current framework.
210: In particular, we show that the prediction 
211: that slow inhibition leads to synchrony, 
212: made under assumptions of 
213: homogeneity~\cite{ernst,gerstner96b,vrees94},
214: must be modified in the presence of mild heterogeneity.
215: Thus, the new framework provides a context for understanding previous 
216: simulations~\cite{wang96}.
217: In particular, it explains the mechanisms 
218: underlying asynchrony (phase dispersion) with slow 
219: decay of inhibition.
220: These mechanisms differ from those underlying the loss of synchrony with
221: faster-decaying inhibition.  
222: 
223: \section{Methods}
224: 
225: \subsection{Numerical simulations}
226: 
227: Simulations were carried out using single-compartment neurons with
228: inhibitory synapses obeying first-order kinetics.
229: Membrane potential in each point neuron obeyed the current balance
230: equation
231: \begin{equation}
232: C\frac{dV_i}{dt}=I_i-I_{Na}-I_K-I_L-I_s,
233: \label{membrane}
234: \end{equation}
235: where $C=1 \mu$F/cm$^2$, $I_i$ is the applied current,
236: $I_{Na}=g_{Na} m_\infty^3 h (V_i - V_{Na})$ and $I_K=g_Kn^4(V_i-V_K)$ are
237: the Hodgkin-Huxley type spike generating currents, $I_L= g_L(V_i-V_L)$
238: is the leak current and $I_s = \sum_j^N (g_s/N) s_j (V_i-V_s)$ is the
239: synaptic current. 
240: The fixed parameters used were: $g_{Na}=30$~mS/cm$^2$, $g_K=20$~mS/cm$^2$, 
241: $g_L=0.1$~mS/cm$^2$,
242: $V_{Na}=45$~mV, $V_K=-75$~mV, $V_L=-60$~mV, $V_s=-75$~mV.
243: These parameters are within physiological ranges and give the
244: high spike rates typical of hippocampal interneurons.  
245: The phenomena described here seem largely independent of specific
246: neuronal parameters.
247: 
248: The activation variable $m$ was assumed fast and substituted with
249: its asymptotic value
250: $m_\infty(v)=(1 + \exp[-0.08(v+26)])^{-1}$.  The inactivation
251: variable $h$ obeys
252: \begin{equation}
253: \frac{dh}{dt}=\frac{h_\infty(v)-h}{\tau_h(v)},
254: \end{equation}
255: with $h_\infty(v)=(1 + \exp[0.13(v+38)])^{-1}$, 
256: $\tau_h(v)=0.6/(1 + \exp[-0.12(v+67)])$.  The variable $n$ obeys
257: \begin{equation}
258: \frac{dn}{dt}=\frac{n_\infty(v)-n}{\tau_n(v)},
259: \end{equation}
260: with $n_\infty(v)=(1 + \exp[-0.045(v+10)])^{-1}$, 
261: $\tau_n(v)=0.5+2.0/(1+\exp[0.045(v-50)]))$.
262: 
263: The gating variable $s_j$ for the synapse is assumed to obey first order
264: kinetics of the form
265: \begin{equation}
266: \frac{ds_j}{dt} = F(V_j)(1-s_j)-s_j/\tau_s,
267: \label{synapse}
268: \end{equation}
269: where $F(V_j)=1/(1+\exp[-V_j])$.
270: 
271: The ODEs were integrated using a fourth-order Runge-Kutta
272: method.  The free parameters were scanned across
273: the following ranges:  for applied current $I_i$, 0-10 $\mu$A/cm$^2$; 
274: for $g_s$, the maximal synaptic conductance per cell,
275: 0-2~mS/cm$^2$; for the synaptic decay time constant $\tau_s$,
276: 5-50~ms.
277: 
278: \subsection{Calculation of coherence}
279: 
280: As a measure of coherence between pairs of
281: neurons, we generated trains of square
282: pulses from the time domain responses of
283: each of the cells (Fig. \ref{coho_ex_fig}).
284: Each pulse, of height
285: unity, was centered at the time of
286: a spike peak (resolution = 0.1 ms); 
287: the width of the pulse was
288: 20\% of the mean firing period of the
289: faster cell in the pair (0.2~$T_1$ in Fig. \ref{coho_ex_fig}). 
290: We then took the
291: cross-correlation at zero time lag of these pulse trains.
292: This is equivalent to calculating the shared area of the 
293: unit-height pulses, as shown in Fig. \ref{coho_ex_fig}D.
294: We took coherence as the sum of these shared areas, 
295: divided by the square root of the product of the summed areas of each
296: individual pulse train.
297: For the example shown in Fig. \ref{coho_ex_fig}, our algorithm gives coherence
298: of 0.35.
299: 
300: Our approach differs from the algorithm used
301: by Wang and Buzs\'aki~(1996), in which 
302: trains of unit-height pulses are correlated for
303: a bin width equal to or greater than
304: the neuronal time scale. 
305: The difference between the two algorithms can be appreciated
306: by considering the contribution made to the coherence 
307: measure by two spikes (in two separate neurons)
308: occurring with time difference $t_d$.
309: The Wang and Buzs\'aki (1996) algorithm would see these as
310: perfectly coherent if the spikes are in the same time bin
311: and incoherent if they are not.
312: The answer depends on where the bin edges fall, with 
313: probability of a coherence ``hit'' falling to zero when
314: the bin width is less than $t_d$.  In their algorithm
315: coherence is a function of the bin width, and averaging
316: across the population of cells ameliorates effects due to
317: the placement of bin edges.
318: In our algorithm, the two spikes make a
319: contribution to coherence that is continuously distributed
320: between 0 ($t_d >$ 20\% of firing period) and 1 ($t_d = 0$).
321: Although both algorithms give results that depend on 
322: the percentage of the firing period considered significant,
323: our measure allows us to examine coherence in small networks
324: with less discretization error.
325: This change is important here specifically because we
326: analyze small networks that phase-lock with a short but measurable
327: phase difference. 
328: 
329: We mapped coherence vs. $I_i$, $g_s$, and $\tau_s$ for networks 
330: of 2, 10, and 100 cells with all-to-all inhibitory coupling.
331: In networks with $N = 2$, coherence is plotted in the maps.
332: In larger networks, the plots show the average of the coherence
333: measure taken for all pairs of neurons.
334: 
335: \section{Results}
336: 
337: \subsection{Single self-inhibited neuron}
338: 
339: We first consider the firing characteristics of a single
340: self-inhibited neuron or, equivalently, a network of identical,
341: synchronized, mutually inhibitory neurons.
342: These simulations 
343: validate predictions from analytic work on simpler models~\cite{cc} 
344: and determine the ranges of
345: the phasic and tonic regimes in parameter space.
346: Firing frequency of the single neuron was tracked over the
347: parameter space of $I_i$, $g_s$, and $\tau_s$.  
348: Figure \ref{1cell_fig}A shows sample time-domain traces
349: for three values of $I_i$ (0.4, 1.6 and 9.0 $\mu$A/cm$^2$).
350: Like mammalian interneurons, the modeled 
351: system of differential equations produces
352: action potentials at rates up to 250 Hz.
353: Figure \ref{1cell_fig}B shows discharge frequency as a
354: function of $I_i$, for several values of 
355: $g_s$.  
356: For large values of $g_s$ (lower traces), this curve is roughly linear.
357: For smaller values (upper traces), discharge frequency rises along a
358: somewhat parabolic trajectory.
359: For negative values of $I_i$, the self-inhibited neuron can fire at
360: arbitrarily low frequencies (data not shown), indicative of a 
361: saddle-node bifurcation and synchrony through slow inhibition~\cite{bard96}.
362: In Fig.~\ref{1cell_fig}C
363: we show discharge frequency versus $\tau_s$ for several values of $I_i$,
364: with $g_s$ fixed.
365: The dependence of the frequency on $\tau_s$ for the lower two traces
366: is similar to what was
367: observed in the full network and {\it in vitro} by Whittington {\it et
368: al.}~(1995).  
369: The phasic and tonic regimes are clearly illustrated in
370: Fig.~\ref{1cell_fig}D, 
371: in which the ratio $\tau_s/T$ is plotted versus $\tau_s$ for
372: various values of $I_i$.  For large $I_i$ (top traces),
373: $\tau_s/T$ is large and linearly related to $\tau_s$,
374: indicative of the tonic regime.
375: In contrast, for small $I_i$ (bottom trace),
376: $\tau_s/T$ is small and depends 
377: only weakly on $\tau_s$, indicative of the phasic regime. 
378: For our model and level of heterogeneity, 
379: parameter sets that give $\tau_s/T < 1$ are in the
380: phasic regime; sets that give $\tau_s/T > 2$ are in the tonic regime.
381: 
382: Presence in either the phasic or tonic regime is dependent
383: on parameters other than $I_i$.
384: Generally, the tonic regime
385: is characterized by strong applied current and a relatively weak
386: synapse so 
387: that the firing period is much faster than the synaptic decay time.
388: The phasic regime occurs when either the applied current is weak and/or the
389: synapse is strong so that the firing period is locked to the decay
390: time. 
391: 
392: 
393: \subsection{Two cell network}
394: 
395: We simulated networks of two 
396: mutually inhibitory cells with self-inhibition.  
397: We include
398: self-inhibition because it better mimics the behavior of a large
399: network.
400: In these and all other network simulations, mutual and 
401: self-inhibition are of equal weight.
402: In networks of two
403: interneurons with identical properties but different initial
404: conditions, the cells
405: quickly synchronize (phase-lock with zero phase difference) over the
406: entire examined range of $I_i$, $g_s$, and $\tau_s$ (data not shown).
407: Slow-firing cells tend to synchronize more quickly than
408: fast-firing cells, but the exact delay before synchronization
409: depends on initial conditions and was not examined systematically.
410: Anti-synchrony is not stable in the parameter regime we considered,
411: but could be with very small values of $\tau_s$.
412: 
413: When the input $I_i$ to each neuron is made mildly
414: heterogeneous (intrinsic spike rates $<$ 5\% different), 
415: a more complex picture emerges. 
416: Under the conditions of mild heterogeneity modeled here,
417: but not necessarily under conditions of greater heterogeneity~\cite{golomb93},
418: the behavior of the two-cell network falls into one of four
419: qualitative states, as exemplified by the traces of
420: membrane potential and inhibitory conductance vs. time in 
421: Fig.~\ref{domains_fig}.
422: For small $g_{s}$, large $I_i$, and large $\tau_s$ -- conditions
423: associated with the tonic regime -- the phasic component of synaptic
424: inhibition received by each cell is small (Fig. \ref{domains_fig}A).
425: The neurons influence each other's firing frequencies, but
426: firing times are independent.
427: We refer to this phase-dispersed state as the {\it asynchronous
428: state}.
429: As the phasic component of inhibition is increased, the phasic
430: regime is approached.
431: Within the phasic regime lie three qualitative states.
432: For appropriate choices of the level of inhibition, 
433: the two-cell network enters a phase-locked state with a non-zero
434: phase difference (Fig.~\ref{domains_fig}B).  We will continue to use the term
435: synchrony to refer to this near-synchronous regime. 
436: For this model, 
437: heterogeneity of some sort (in this case, heterogeneity of 
438: intrinsic firing frequencies) is a necessary and sufficient
439: condition for near, as opposed to pure, synchrony~\cite{cc}.
440: The size of the phase difference depends on the parameters chosen.
441: With further increases in the level of inhibition,
442: the faster cell begins to suppress its slower partner, leading
443: to what we term {\it harmonic locking} (Fig.~\ref{domains_fig}C).
444: In this example, cells
445: fire in a 4:3 ratio, and exert temporally complex effects
446: on each other during the course of one cycle ($\sim$ 50 ms).
447: Finally, with enough inhibition, the faster
448: neuron inhibits its slower counterpart totally, in what we term
449: {\it suppression} (Fig.~\ref{domains_fig}D).
450: In suppression, the sub-threshold dynamics of
451: membrane potential in the suppressed cell 
452: are exactly phase locked to those of the faster cell. 
453: This exact relationship holds because our simulations do not
454: include a synaptic delay term.
455: 
456: Without self-inhibition, this harmonic-locking regime is very small
457: and not seen in the analogous parameter space (data not shown). 
458: Our heuristic explanation for this difference is as follows.
459: Without self-inhibition, once the slower neuron is suppressed, 
460: the instantaneous preferred frequencies of the two cells diverge.
461: The faster cell is uninhibited and, by firing faster, adds more
462: inhibition to the slower cell, making it more difficult for the slower
463: cell to escape.
464: With self-inhibition, each of the cells in the two-cell network receives
465: an identical synaptic signal, effectively making the two cells more
466: homogeneous.
467: The added homogeneity increases the size of the region in which
468: harmonic locking occurs at relatively small locking ratios.
469: 
470: In order to observe network behavior over a large parameter range,
471: we used the relatively simple measure of firing coherence (see Methods).
472: A given level of coherence does not uniquely determine the qualitative
473: behavior of the network (asynchronous, synchronous, harmonic,
474: or suppressed).
475: However, the structures of coherence maps
476: are stereotyped, and coherence maps can be correlated to the four
477: qualitative network states.
478: 
479: Figures~\ref{coho_fig}A-B show three dimensional plots of coherence in a 
480: two-cell network, plotted versus $\tau_s$
481: and $g_s$ for low ($I_1=1.6$ and $I_2=1.78 \mu$A/cm$^2$)
482: and high ($I_1=9$ and $I_2=9.9 \mu$A/cm$^2$) applied currents.  
483: (The gray scale, which does not relate to coherence, is discussed
484: below.) 
485: Even though the differences in intrinsic (uncoupled) firing
486: frequencies for 
487: the two cells are small ($<5\%$ in each case), 
488: coherence is high and smoothly varying,
489: corresponding to  synchrony,
490: only over a small region of parameter space. 
491: The extent of the synchronous region increases as $\tau_s$ decreases.  
492: Increasing the heterogeneity reduces the size of the synchronous region.  
493: For differences
494: greater than a few percent in the intrinsic (uncoupled) frequencies, the
495: synchronous region was dramatically reduced in size (data not shown).
496: 
497: For a given $\tau_s$, synchrony is
498: broken in two distinct ways if $g_s$ is either too small or too large.  
499: For large $I_i$, large $\tau_s$, and (especially) small $g_s$, 
500: the phasic coupling between the two cells is weak and they fire
501: asynchronously (i.e., with dispersed phase).  
502: In this state, which is particularly large on the left side of 
503: Fig.~\ref{coho_fig}B, 
504: coherence has a value of about 0.2, corresponding to the expected value
505: of our coherence measure  with ``memory'' equal to 
506: 20\% of the spiking period.
507: For large $g_s$, high levels of coherence are lost when the faster 
508: cell begins to suppress the slower cell, resulting in harmonic
509: spiking.
510: The particular pattern of harmonic spiking can change
511: dramatically with small changes in parameters, resulting in the
512: jagged coherence regions seen in Figs.~\ref{coho_fig}A-B.  Again, the harmonic
513: region is particularly noticeable with large $I_i$, as in Fig.~\ref{coho_fig}B.
514: 
515: Eventually, with large enough $g_s$, the full suppression 
516: state can take hold, and coherence plummets to give a very flat
517: region of coherence at a value of 0.
518: This state, favored by large $g_s$ and large $\tau_s$, 
519: occupies a large region on the right side of Fig.~\ref{coho_fig}A.
520: 
521: We argued in the discussion of Fig.~\ref{domains_fig} that the network's
522: presence in the asynchronous state is associated with the tonic
523: regime, and that the transition from asynchrony to 
524: locking is associated with the transition from the tonic
525: regime to the phasic regime.
526: To demonstrate this effect, 
527: we have gray-scale-coded the coherence maps
528: of Fig.~\ref{coho_fig} according to the value of $\tau_s/T$ obtained from
529: single, self-inhibited cells with the same values of
530: $\tau_s$ and total inhibition $g_s$ and $I_i$ taken as the average
531: of the range seen in the heterogeneous population.
532: 
533: The single-cell value of $\tau_s/T$ is useful as an indicator
534: of the qualitative state of all the cells in the network because
535: all the cells that are not suppressed fire at similar frequencies.
536: This result is demonstrated by 
537: Fig. \ref{compare_n}, which shows plots of $\tau_s/T$ for four conditions:
538: the N = 1 case (solid lines), the N = 2 case with differences
539: in intrinsic rates of around 4\% (dashed lines) and 2\%
540: (dashed-and-dotted lines); and the N = 10 case with maximal
541: heterogeneity of around 4\% (dotted lines).
542: In all cases with more than one cell, a pair of traces corresponding
543: to the fastest and slowest cells of the simulations are shown.
544: In all cases, the traces follow similar trajectories
545: until the slowest cell is suppressed 
546: (indicated by an abrupt end of the lower branchbefore
547: the rightmost point is reached).
548: This similarity in $\tau_s/T$ (and hence $T$) for all unsuppressed cells
549: is seen in both the phasic (Fig. \ref{compare_n}C) and tonic 
550: (Fig. \ref{compare_n}D, right side) regimes.
551: 
552: Returning to Fig.~\ref{coho_fig}, 
553: the value of $\tau_s/T$ as a predictor of transitions in qualitative
554: state and hence coherence implies that 
555: we should see transitions from asynchrony 
556: when $\tau_s/T$ drops below $\sim 2$.
557: As Figs.~\ref{coho_fig}A-B show, this approximate relationship does hold.
558: Furthermore, factors that change $\tau_s/T$ (e.g., changing $I_i$;
559: cf. Figs. \ref{domains_fig}A and \ref{domains_fig}B) 
560: have predictable effects on the extent of
561: the asynchronous state in ($\tau_s,g_s$)-space.
562: 
563: Figures \ref{coho_fig}C-D show similar results with less heterogeneity
564: ($I_1 = 1.64, I_2 = 1.74 \mu$A/cm$^2$ for panel C; 
565: $I_1 = 9.2, I_2 = 9.7 \mu$A/cm$^2$ for panel D;
566: these values approximate the mean $\pm$ one standard deviation
567: for uniform distributions with limits as in Figs.~\ref{coho_fig}A-B). 
568: In these cases, the same qualitative coherence map is evident, with
569: a somewhat larger region of coherence.
570: The qualitative coherence regions correspond to the same qualitative
571: states from Fig.~\ref{domains_fig}.
572: 
573: \subsection{Large networks}
574: 
575: We also simulated all-to-all connected networks of 10 and 100
576: heterogeneous 
577: inhibitory neurons and found qualitatively similar results.  
578: Figures~\ref{coho_fig}E-F
579: show the coherence plots over the same parameter space as 
580: Figs.~\ref{coho_fig}A-B
581: for a network of 10 heterogeneous cells. 
582: The level of inhibition per synapse, $g_s/N$, scales with
583: $N$ to keep the level of inhibition per postsynaptic cell,
584: $g_s$, constant. 
585: For the ten-cell case, applied current $I_i$ is uniformly distributed
586: through the same ranges as in panels A-B ([1.6,~1.78] for panel
587: E; [9.0~9.9] for panel F).
588: Again, there are four qualitative states: an
589: asynchronous state for small $g_s$, more prevalent with higher $I_i$;
590: a near synchronous state; a harmonic state; and a suppressed
591: state.
592: For the 10 cell network, the transition to suppression is smoother
593: than in the two-cell case.  
594: Cells fall out of the rhythm to suppression one at a time,
595: leading to a relatively smooth drop in coherence.
596: At the highest values of $g_s$, coherence has not yet dropped
597: to zero because some cells are still able to synchronize with
598: the fastest neuron of the network.
599: In the harmonic state, examination of time-domain traces 
600: (data not shown) reveals harmonic patterns, with a cluster of
601: cells in  synchrony while the slower cells drop in and
602: out of the population rhythm.
603: The coherent region for the ten-cell network is larger than in
604: Figs.~\ref{coho_fig}A-B.
605: Applied currents (and hence intrinsic frequencies) of the two 
606: neurons in panels A-B are at the limits of the range of 
607: applied currents in the ten-cell network, making the 
608: effective level of heterogeneity smaller in the ten-cell case.
609: The close agreement between panels C-D and E-F supports this
610: contention.
611: 
612: We also performed a limited number of simulations of a 
613: 100-cell network with the same architecture, at parameter values
614: representing orthogonal slices through the 3-dimensional
615: coherence maps.
616: Results from these simulations are shown in Fig.~\ref{slices_fig},
617: along with slices from the coherence maps of Fig.~\ref{coho_fig}.
618: In Figs.~\ref{slices_fig}A-B, coherence is plotted vs. $g_s$ for a 
619: fixed value of $\tau_s$ = 15~ms and at two levels of applied
620: current.
621: In Figs.~\ref{slices_fig}C-D, coherence is plotted vs. $\tau_s$ for a fixed value
622: of $g_s$ = 0.5~mS/cm$^2$.
623: Results from the 100-cell (N~=~100) and 10-cell (N~=~10) 
624: cases are quite similar,
625: at both low (panels A, C) and high (panels B, D) levels of
626: applied current.
627: These results support the argument that the qualitative behavior
628: of the network does not change with N, and thus that predictions
629: based on single-cell analysis and simulations are applicable to
630: moderately heterogeneous networks of arbitrary size.
631: Results are shown for both levels of heterogeneity in 2-cell networks.
632: The dashed lines (N~=~2), which are slices through the coherence maps
633: of 
634: Fig.~\ref{coho_fig}A-B, have lower coherences that reflect the relatively large
635: amounts of heterogeneity in these cases.
636: The dashed-and-dotted lines (N~=~2*) show coherence values for
637: slices through Figs.~\ref{coho_fig}C-D, with closer intrinsic frequencies
638: chosen to approximate the standard deviations of the appropriate
639: uniform distributions.
640: These slices more nearly match the 10- and 100-cell cases.
641: 
642: Results from Figs. \ref{compare_n} and 6 also demonstrate the close relationship
643: between the ratio $\tau_s/T$ and coherence (as well as underlying
644: qualitative states).
645: Values of $\tau_s/T < 2$ from Fig. \ref{compare_n} 
646: are almost invariably associated
647: with one of the locked states.
648: Values of $\tau_s/T > 2$, on the other hand, give rise to the 
649: asynchronous state, associated in Fig.~\ref{slices_fig} with regions of flat
650: coherence at a value of 0.2 (e.g., the leftmost portion of 
651: Fig.~\ref{slices_fig}B
652: and the rightmost portion of Fig.~\ref{slices_fig}D).
653: 
654: \section{Discussion}
655: 
656: We show that the behavior of the firing frequency of a single
657: self-inhibited cell can give insight into the network
658: frequency and coherence.  
659: In particular, the ratio of the synaptic decay constant $\tau_s$
660: to the neuronal firing period $T$ has rough predictive value
661: in determining whether a mildly heterogeneous network is
662: synchronous or asynchronous.
663: This predictive value only holds with mild heterogeneity,
664: however; greater heterogeneity leads to a mixture of qualitative
665: states~\cite{golomb93} which invalidates our analyses.
666: 
667: We also emphasize the importance of even mild 
668: heterogeneity in affecting network dynamics.  
669: Previously, it had been argued   
670: that slowly decaying inhibition generally had a synchronizing
671: influence~\cite{gerstner96b,terman96,vrees94}.  
672: However, for mildly heterogeneous cells, 
673: the relation of the frequency (or period) 
674: to the synaptic decay time must also be
675: considered.  
676: For homogeneous cells, the synaptic coupling is only
677: required to align the phases in order to obtain synchrony.  
678: For mildly heterogeneous cells, 
679: the coupling must both align the phases and
680: entrain the frequencies.  
681: The latter is more difficult for the network to achieve.
682: It occurs only when the inhibition is strong enough so that firing
683: period is 
684: dominated by the decay time.  However, if the inhibition
685: is too strong then the slower cells will never fire.
686: Thus, there are
687: two ways to destroy full network synchrony.  
688: The first is through
689: effective de-coupling where the cells tend to fire asynchronously.
690: The second is through suppression, in which the neurons with
691: higher intrinsic rates fire in near-synchrony and
692: keep their slower counterparts from firing.  
693: Between synchrony and
694: suppression harmonic locking is also possible.  This occurs when
695: the suppression of the slower cell is temporary but lasts longer
696: than the period of the faster cell.
697: We should note that anti-synchrony, not seen in the
698: parameter regimes presented here,
699: can become stable with very fast 
700: synapses (i.e., $\tau_s/T 
701: << 1$)~\cite{friesen94,perkel74,skinner94,vrees94,wang92}.
702: 
703: For even mildly heterogeneous cells, synchrony in which all inhibitory
704: cells participate is possible only over a small
705: region of parameter space 
706: that decreases as the heterogeneity is increased.  The region where
707: synchrony occurs in a large network of known (mild) heterogeneity and 
708: connectivity can be approximated from a two-cell network.  
709: The frequency of firing and conditions allowing synchrony 
710: can be estimated analytically from a reduced model neuron with
711: self-inhibition~\cite{cc}
712: As in large networks~\cite{traub96}, the frequency in single cells 
713: depends on the applied
714: current, the synaptic strength and the synaptic decay time.  
715: In the synchronous region, the firing period depends linearly on the
716: decay time and logarithmically on the other parameters so that
717: frequency will depend directly on the decay rate.  However, 
718: the contribution from the logarithmic factor can be fairly
719: large and thus must be calculated explicitly.  This can be estimated
720: analytically from the reduced model~\cite{cc}, or from simple
721: simulations of a single, self-inhibiting cell (see Fig.~\ref{1cell_fig}).
722: 
723: The result that the value of $\tau_s/T$ from single-cell 
724: simulations has predictive value for the qualitative state and
725: coherence of a network of arbitrary size is intriguing and potentially
726: useful, because
727: it points the way to determining the qualitative and quantitative
728: behavior of a neuronal network based on simple behavior that
729: can be studied numerically or even analytically.
730: However, the predictive capabilities of this index should not
731: be overestimated.
732: A careful examination of Fig.~\ref{coho_fig} shows that 
733: the mapping between $\tau_s/T$ and asynchrony is not
734: precise.  The value of $\tau_s/T$ at which the transition will
735: occur is dependent on many factors, including the level of
736: heterogeneity and, in all likelihood, the level and form 
737: of connectivity in the network.
738: The value of $\tau_s/T$ alone is not sufficient to determine
739: the point of transition from synchrony to harmonic locking
740: and suppression, even in a model of known heterogeneity and
741: architecture.
742: Making this determination requires knowledge of $I_i$ and $g_s$
743: in addition to $\tau_s/T$~\cite{cc}.
744: 
745: Studies of the 2-cell network were successful
746: in elucidating the qualitative states of the larger circuit,
747: though the exact form of transitions 
748: from asynchrony to synchrony and synchrony to suppression is
749: different in detail for our simulations of the 2-cell and N-cell cases.
750: In general, the behavior of the 2-cell network matches that of the
751: N-cell circuit better in the asynchronous state, associated with
752: the tonic regime, than in the harmonic and suppression states,
753: associated with the phasic regime.
754: This result is expected from our theoretical framework since
755: the tonic regime is defined as the regime in which
756: only the tonic level of inhibition is important.  Since we
757: normalized the synaptic strength by N, the net amount of inhibition is
758: independent of the network size.
759: Thus, we take this result as additional evidence that our
760: hypothesized mechanisms of loss of coherence are correct.
761: 
762: Our numerical results are similar to those of 
763: Wang and Buzs\'aki~(1996), but our explanations differ considerably.
764: In heterogeneous networks, they also saw a decline in coherence with
765: both low and high firing rates.  
766: They attributed the decline in synchrony for low rates to two factors.
767: First, they point out that cells are more sensitive at low firing rates 
768: than at higher rates to changes in applied
769: current, a source of heterogeneity in both studies.
770: This point is correct, but in our work we controlled for this
771: factor, using smaller percent differences in small currents than
772: in large currents to achieve similar percent differences in
773: intrinsic firing rates, and we still saw a drop-off in coherence
774: at low rates.
775: Second, Wang and Buzs\'aki (1996) cite what they call a 
776: ``dynamical'' effect, in which inhibition is fast enough to destabilize
777: the synchronous state.
778: Previous work~\cite{vrees94,wang92} shows that the outcome of 
779: such dynamical effects for
780: {\em homogeneous} networks is anti-synchrony.
781: In our parameter regime, the loss of coherence in 
782: {\em heterogeneous} networks at low firing rates 
783: (i.e., with $\tau_s/T$ small) is associated with the phasic regime
784: and is due to suppression of firing in slower cells.
785: Wang and Buzs\'aki~(1996) make the phenomenological argument
786: that the loss of synchrony
787: at high firing rates is related to a need for greater density of
788: synaptic connectivity. 
789: We considered all-to-all connectivity and found that
790: loss of coherence 
791: associated with high firing rates (tonic regime)
792: is caused by a loss of too much of the phasic component of inhibition.
793: Furthermore, we argue that one can approximate the parameters for which 
794: this loss of coherence occurs by analyzing the single, self-inhibitory
795: cell. 
796: It should be possible to generalize these results and arguments to
797: the case with less than all-to-all coupling.
798: 
799: It has been suggested that the selection of the network frequency
800: {\it in vivo} is determined by the tonic
801: excitation and the parameters regulating the synaptic
802: coupling~\cite{traub96}.  Our results support this hypothesis.
803: However, we have demonstrated that with heterogeneous cells,
804: synchrony may not be possible at all frequencies.  
805: In particular, a network of this kind seems unlikely to support
806: synchronous firing at 200 Hz, a frequency that seems too fast
807: to be synchronized by GABA$_A$ receptors with $\tau_s \sim 15$ ms
808: (and $\tau_s/T \sim 3$).
809: Our framework implies that this result, which has been seen in
810: simulations 
811: before~\cite{wang96}, holds in general for heterogeneous cells in the 
812: tonic regime.
813: 
814: Our results emphasize the difficulty of generating synchronous 
815: oscillations in
816: interneuronal networks over a large range of frequencies,
817: such as in the transition from the gamma/theta mode to the
818: sharp wave/fast ripples mode.
819: At gamma frequencies, the factor $\tau_s/T$ should be less than 1
820: with typical values of $\tau_s$.
821: Thus, full synchrony at gamma frequencies is possible but requires
822: careful regulation of the system to prevent suppressive effects.
823: The question of whether or not the suppression we see is incompatible 
824: with physiological data cannot be answered, because it is extremely
825: difficult to estimate the number of interneurons participating in the
826: rhythm.
827: We believe that this issue can be explored, and our model tested, by
828: examining the power of the gamma field potential in a brain slice as
829: $\tau_s$ is modified by pentobarbital.
830: Our model predicts that the power in this signal should decrease as 
831: $\tau_s$ rises
832: and suppression becomes more evident.
833: A negative result in these experiments would indicate that our model is
834: missing a fundamental element.
835: One such element is intrinsic or synaptic noise, which can act to release
836: neurons from suppression (White, unpublished observations).
837: 
838: The more difficult goal for our model to achieve is that of
839: firing synchronously at ripple (200 Hz) frequencies, as has been
840: reported in the behaving animal~\cite{ylinen95}.
841: One or more of several conceivable explanations may underlie 
842: this apparent robustness in hippocampal function at high frequencies.  
843: First, it is possible, but unlikely, that heterogeneity in 
844: the intrinsic firing frequencies of interneurons is very low
845: ($ < 4$\%).
846: Second, the operant value of $\tau_s$ may be lower than we
847: believe; a value of 5 ms would conceivably allow synchrony
848: at 200 Hz with levels of heterogeneity of around 5\%.
849: Third, each interneuron may fire not at 200 Hz, but rather
850: at a lower frequency of, say, 100 Hz, during sharp waves.
851: Under this explanation, the 200-Hz ripple would be generated
852: by clusters of two or more populations of neurons spiking
853: independently.
854: Finally, some factor(s) not considered here may enhance synchrony
855: at high frequencies.
856: Gap junction-mediated electrical coupling among interneurons, 
857: for which some evidence exists in the hippocampal region CA1~\cite{katsu},
858: is perhaps the most likely such factor~\cite{traub95}.
859: 
860: 
861: \acknowledgements
862: We thank M. Camperi for assistance in writing code, 
863: and S. Epstein, O. Jensen, C. Linster, and F. Nadim for helpful
864: discussions.  
865: B. Ermentrout, J. Rinzel, and R. Traub provided valuable feedback
866: on earlier versions of the manuscript.
867: This work was supported by grants from 
868: the National Science Foundation (DMS-9631755 to C.C., N.K. and J.W.), 
869: the National Institutes of Health (MH47150 to N.K.; 
870: 1R29NS34425 to J.W.), 
871: and The Whitaker Foundation (to J.W.)
872: 
873: 
874: \begin{thebibliography}{}
875: 
876: \bibitem[\protect\citeauthoryear{Bose \et}{1997}]{bose} Bose
877: A, Terman D, and Kopell N. (1997) Manuscript in 
878: preparation. 
879: 
880: \bibitem[\protect\citeauthoryear{Bragin \et }{1995}]{bragin95} 
881: Bragin A, Jand\'o G, N\'adasdy Z, Hetke J, Wise K,
882: and Buzs\'aki G. (1995) Gamma (40-100 Hz) oscillation in the hippocampus of
883: the behaving rat. {\it J. Neuroscience} 15:47-60.
884: 
885: \bibitem[\protect\citeauthoryear{Buzs\'aki}{1986}]{buzsaki86}
886: Buzs\'aki G. (1986) Hippocampal sharp waves: their origin 
887: and significance. {\it Brain Res.} 398:242-252.
888: 
889: \bibitem[\protect\citeauthoryear{Chow \et}{1997}]{cc} Chow CC, White
890: JA, Ritt J, and Kopell N. (1997) Manuscript in preparation.
891: 
892: \bibitem[\protect\citeauthoryear{Ermentrout \et}{1996}]{bard96} 
893: Ermentrout B. (1996) Type I membranes, phase resetting curves, and
894: synchrony. {\it Neural Comp.} 8:979-1001.
895: 
896: \bibitem[\protect\citeauthoryear{Ernst \et}{1995}]{ernst} Ernst U,
897: Pawelzik K, and Geisel T. (1995) Synchronization 
898: induced by temporal delays in pulse-coupled oscillators. {\it
899: Phys. Rev. Lett.} 74:1570-1573.   
900: 
901: \bibitem[\protect\citeauthoryear{Friesen}{1994}]{friesen94}
902: Friesen W. (1994) Reciprocal inhibition, a mechanism underlying 
903: oscillatory animal movements. {\it Neurosci Behavior} 18:547-553.
904: 
905: \bibitem[\protect\citeauthoryear{Gerstner \et}{1996}]{gerstner96b}
906: Gerstner W, van Hemmen JL, and Cowen J. (1996) What 
907: matters in neuronal locking? {\it Neural Computation} 8:1653-1676.
908: 
909: \bibitem[\protect\citeauthoryear{Golomb and Rinzel}{1993}]{golomb93}
910: Golomb D and Rinzel J. (1993) Dynamics of globally coupled inhibitory
911: neurons with heterogeneity. Phys. Rev. E {\bf 48}, 4810-4814.
912: 
913: \bibitem[\protect\citeauthoryear{Gray}{1994}]{gray94} Gray CM. (1994)
914: Synchronous oscillations in neuronal 
915: systems: mechanisms and functions. {\it J. Comp. Neuro.}  1:11-38.
916: 
917: \bibitem[\protect\citeauthoryear{Hansel \et}{1995}]{hansel95} Hansel
918: D, Mato G, and Meunier C. (1995) Synchrony in 
919: excitatory neural networks. {\it Neural Comp.} 7:307-337.
920: 
921: \bibitem[\protect\citeauthoryear{Jeffreys \et}{1996}]{jeff96} Jeffreys
922: JGR, Traub RD, and Whittington MA. (1996) 
923: Neuronal networks for induced `40 Hz' rhythms. {\it Trends Neurosci.}
924: 19:202-207. 
925: 
926: \bibitem[\protect\citeauthoryear{Katsumaru \et}{1988}]{katsu}
927: Katsumaru H, Kosaka T, Heizman CW, and Hama K. (1988) 
928: Gap-junctions on GABAergic neurons containing the calcium-binding
929: protein parvalbumin in the rat hippocampus (CA1 regions).
930: {\it Exp. Brain Res.} 72:363-370.
931: 
932: \bibitem[\protect\citeauthoryear{Llin\'as and Ribary}{1993}]{llinas93}
933: Llin\'as R and Ribary U. (1993) Coherent 40-Hz oscillation 
934: characterizes dream state in humans. {\it Proc. Natl. Acad. Sci. USA}
935: 90:2078-2081.
936: 
937: \bibitem[\protect\citeauthoryear{Perkel and Mulloney}{1974}]{perkel74}
938: Perkel D and Mulloney B. (1974) Motor patterns in reciprocally 
939: inhibitory neurons exhibiting postinhibitory rebound. 
940: {\it Science} 185:181-183.
941: 
942: \bibitem[\protect\citeauthoryear{Pinsky and Rinzel}{1994}]{pinsky}
943: Pinsky P and Rinzel J. (1994) Intrinsic and network 
944: rhythmogenesis in a reduced Traub model for CA3
945: neurons. {\it J. Comp. Neurosci.} 1:39-60.
946: 
947: \bibitem[\protect\citeauthoryear{Skinner \et}{1994}]{skinner94}
948: Skinner F, Kopell N, and Marder E. (1994) Mechanisms for oscillations 
949: and frequency control in networks of mutually inhibitory relaxation 
950: oscillators.  {\it J. Comp Neurosci.} 1:69-87.
951: 
952: \bibitem[\protect\citeauthoryear{Terman \et}{1996}]{terman96} Terman
953: D, Bose A, and Kopell N. (1996) 
954: Functional reorganization in thalamocortical networks: transition between
955: spindling and delta sleep rhythms. {\it Proc. Natl. Acad. Sci. USA}
956: 93:15417-15422.
957: 
958: \bibitem[\protect\citeauthoryear{Traub}{1995}]{traub95} Traub
959: RD. (1995) Model of synchronized population bursts 
960: in electrically coupled interneurons containing active dendritic
961: conductances. {\it J. Comp. Neurosci.} 2:283-289.
962: 
963: \bibitem[\protect\citeauthoryear{Traub \et}{1996a}]{traub96} Traub RD,
964: Whittington M, Colling S, Buzs\'aki G, 
965: and Jefferys J. (1996a) 
966: Analysis of gamma rhythms in the rat hippocampus in vitro
967: and in vivo. {\it J. Physiol.} 493:471-484.
968: 
969: \bibitem[\protect\citeauthoryear{Traub \et}{1996b}]{traub96b} Traub
970: RD, Jefferys JGR, and Whittington MA. (1996b) 
971: Simulation of gamma rhythms in networks of
972: interneurons and pyramidal cells. {\it J. Comp. Neurosci.} In press.
973: 
974: \bibitem[\protect\citeauthoryear{Tsodyks \et}{1993}]{tsodyks93}
975: Tsodyks M, Mitkov I, and Sompolinsky, H. (1993) Pattern of synchrony
976: in inhomogeneous networks of oscillators with pulse interactions.
977: {\it Phys. Rev. Lett.} 71:1280-1283.
978: 
979: \bibitem[\protect\citeauthoryear{Wang and Buzs\'aki}{1996}]{wang96}
980: Wang X-J and Buzs\'aki G. (1996) Gamma oscillation by 
981: synaptic inhibition in an interneuronal network model.
982: {\it J. Neurosci.} 16:6402-6413.
983: 
984: \bibitem[\protect\citeauthoryear{Wang and Rinzel}{1992}]{wang92}
985: Wang X-J and Rinzel J. (1992) Alternating and synchronous rhythms in
986: reciprocally inhibitory model neurons. {\it Neural Comp.} 4:84-97. 
987: 
988: \bibitem[\protect\citeauthoryear{Whittington \et}{1995}]{whit95}
989: Whittington MA, Traub RD, and Jefferys JGR. (1995) 
990: Synchronized oscillations in interneuron networks
991: driven by metabotropic glutamate receptor activation. 
992: {\it Nature} 373:612-615.
993: 
994: \bibitem[\protect\citeauthoryear{van Vreeswijk \et}{1994}]{vrees94}
995: van Vreeswijk C, Abbott L, and 
996: Ermentrout GB. (1994) When inhibition not excitation synchronizes neural
997: firing. {\it J. Comp. Neuroscience} 1:313-321.
998: 
999: \bibitem[\protect\citeauthoryear{Ylinen \et}{1995}]{ylinen95} Ylinen
1000: A, Bragin A, N\'adasdy Z, Jand\'o G, Szab\'o I, 
1001: Sik A, and Buzs\'aki G. (1995) Sharp wave-associated high-frequency
1002: oscillation (200 Hz) in the intact hippocampus: network and
1003: intracellular mechanisms. {\it J. Neurosci.} 15:30-46.
1004: 
1005: \end{thebibliography}
1006: 
1007: \begin{figure}[p]
1008: \centerline{\epsfig{figure=fig1.eps, height=6.in,bbllx=12pt,bblly=12pt,bburx=599pt,bbury=780pt, clip=}}
1009: \caption{An example of the coherence measure used in this work.
1010: Panels A and B show idealized periodic spike traces with periods
1011: $T_1$ and $T_2~>~T_1$.  
1012: Panel C shows the pulse trains compared in the algorithm.
1013: The solid line corresponds to Trace 1 and the dotted line to Trace 2.
1014: Each pulse has unit height, width =~$0.2~T_1$, and is centered at 
1015: the appropriate spike peak.
1016: Panel D shows the shared area of the two pulse trains in graphical and
1017: numerical form.}
1018: \label{coho_ex_fig}
1019: \end{figure}
1020: 
1021: \begin{figure}[p]
1022: \centerline{\epsfig{figure=fig2.eps, height=6.in,bbllx=12pt,bblly=12pt,bburx=599pt,bbury=780pt, clip=}}
1023: \caption{Behavior of the single, self-inhibited neuron.
1024: {\it A.}~~Time-domain responses of the self-inhibited neuron
1025: ($g_s$ = 0.25 mS/cm$^2$, $\tau_s$ = 10 ms) for three values
1026: if $I_i$ (from bottom to top: 0.4, 1.6, and 9.0 $\mu$A/cm$^2$).
1027: Horizontal scale bar: 20 ms.  Vertical scale bar: 50 mV.
1028: {\it B.}~~Neuronal discharge frequency vs. applied current $I_i$
1029: for several values of $g_s$ (from top to bottom: 0.05, 0.45,
1030: 0.85, 1.65 mS/cm$^2$).  $\tau_s$ = 10 ms.
1031: {\it C.}~~Firing frequency vs. $\tau_s$.
1032: From bottom to top, $(g_s, I_i)$ = (0.45,2.0) (solid line), 
1033: (0.45,4.0) (dashed line), 
1034: (0.2,6.0) (dotted line), and (0.05,10.0) (dashed-and-dotted line).  
1035: Conductances have units of mS/cm$^2$.  
1036: Currents have units of $\mu$A/cm$^2$.
1037: {\it D.}~~The ratio of the synaptic decay time constant ($\tau_s$)
1038: to the neuronal discharge period ($T$), plotted vs. $\tau_s$.
1039: Different line types represent the same values
1040: of $g_s$ and $I_i$, in the same order, as in panel {\it C}.
1041: }
1042: \label{1cell_fig}
1043: \end{figure}
1044: 
1045: \begin{figure}[p]
1046: \centerline{\epsfig{figure=fig3.eps,height=6.5in,bbllx=12pt,bblly=12pt,bburx=599pt,bbury=780pt, clip=}}
1047: \caption{Plots of membrane potential ($V_m$) vs. time for two
1048: heterogeneous neurons
1049: at four different points in ($I_i,g_s,\tau_s$)-space.
1050: In all cases, the solid (dotted) line is the more (less) 
1051: excitable cell.
1052: Also plotted in each panel is $g_s(t)$, the time-varying synaptic conductance
1053: (in mS/cm$^2$) received by each of the two cells.
1054: {\it A.}  Asynchrony with $I_1$ = 9.0, $I_2$ = 9.9 $\mu$A/cm$^2$;
1055: $g_s$ = 0.25 mS/cm$^2$; $\tau_s$ = 10 ms.
1056: {\it B.}  Near-synchrony with $I_1$ = 1.6, $I_2$ = 1.78 $\mu$A/cm$^2$;
1057: $g_s$ = 0.25 mS/cm$^2$; $\tau_s$ = 10 ms.
1058: {\it C.}  Harmonic locking with $I_1$ = 9.0, $I_2$ = 9.9 $\mu$A/cm$^2$;
1059: $g_s$ = 0.5 mS/cm$^2$; $\tau_s$ = 10 ms.
1060: {\it D.}  Suppression with $I_1$ = 1.6, $I_2$ = 1.78 $\mu$A/cm$^2$;
1061: $g_s$ = 0.5 mS/cm$^2$; $\tau_s$ = 10 ms.
1062: }
1063: \label{domains_fig}
1064: \end{figure}
1065: 
1066: \begin{figure}[p]
1067: \centerline{\epsfig{figure=fig4.eps, height=7in,bbllx=12pt,bblly=12pt,bburx=599pt,bbury=780pt, clip=}}
1068: \caption{Coherence maps in ($\tau_s,g_s$)-space.
1069: Top row: Coherence vs. $\tau_s$ and $g_s$ for the two-cell
1070: network, with $I_1$ = 1.6 and $I_2$ = 1.78 $\mu$A/cm$^2$ ({\it A}),
1071: $I_1$ = 9.0 and $I_2$ = 9.9 $\mu$A/cm$^2$ ({\it B}).
1072: Middle row: Coherence vs. $\tau_s$ and $g_s$ for the two-cell
1073: network, with $I_1$ = 1.64 and $I_2$ = 1.74 $\mu$A/cm$^2$ ({\it C}),
1074: $I_1$ = 9.2 and $I_2$ = 9.7 $\mu$A/cm$^2$ ({\it D}).
1075: Bottom row: Coherence maps for ten cells with $I_i$ uniformly
1076: distributed in the intervals [1.6,~1.78] ({\it E}),
1077: [9.0,~9.9] ({\it F}).
1078: In all maps, the gray scale gives the ratio $\tau_s/T$ (see scale bar).
1079: }
1080: \label{coho_fig}
1081: \end{figure}
1082: 
1083: \begin{figure}[p]
1084: \centerline{\epsfig{figure=fig5.eps,height=6.5in,bbllx=12pt,bblly=12pt,bburx=599pt,bbury=780pt, clip=}}
1085: \caption{Plots of $\tau_s/T$ vs. $g_s$ ({\it A-B}) and $\tau_s$ ({\it C-D})
1086: for networks of size N = 1, 2, and 10. 
1087: For the N = 2 (dashed lines) and N = 10 (dotted lines) cases, 
1088: values of $I_i$ were evenly distributed between the inclusive limits shown,
1089: as in Figs.~\ref{coho_fig}A-B and \ref{coho_fig}E-F,
1090: giving about 4\% maximum heterogeneity in intrinsic firing rates.
1091: For the N = 2* case (dashed-and-dotted lines), $I_1$ and $I_2$
1092: were set to the same values as in Fig.~\ref{coho_fig}C-D,
1093: giving about 2\% heterogeneity.
1094: For N = 1, $I_i$ is the center point of the interval.
1095: For all cases with N $> 1$, two traces are shown, representing values
1096: from the fastest and slowest neurons from the simulations.
1097: Suppression of the slowest cell is represented by early terminations
1098: of the curves.
1099: }
1100: \label{compare_n}
1101: \end{figure}
1102: 
1103: \begin{figure}[p]
1104: \centerline{\epsfig{figure=fig6.eps, height=7in,bbllx=12pt,bblly=12pt,bburx=599pt,bbury=780pt, clip=}}
1105: \caption{Two-dimensional slices through coherence maps.
1106: Left column: coherence vs. $g_s$ at $\tau_s$ = 15 ms, 
1107: for $I_i$ uniformly distributed 
1108: in the ranges [1.6,~1.78] ({\it A}) and [9.0,~9.9] ({\it B}).
1109: Shown are coherence for 100 cells (solid line), 10 cells (dotted line;
1110: data from Fig.~\ref{coho_fig}E-F),
1111: 2 cells at the limits of the distribution of $I_i$ (dashed line;
1112: data from Fig.~\ref{coho_fig}A-B),
1113: and 2  cells at intermediate values of $I_i$ (dashed-and-dotted line;
1114: data from Fig.~\ref{coho_fig}C-D). 
1115: Right column: coherence vs. $\tau_s$ at $g_s$ = 0.5 mS/cm$^2$,
1116: for low ({\it C}) and high ({\it D}) values of $I_i$ (specific
1117: values as in {\it A} and {\it B}, respectively).  
1118: Line types have the same meaning as in {\it A-B}.  
1119: }
1120: \label{slices_fig}
1121: \end{figure}
1122: 
1123: \end{document}
1124: 
1125: 
1126: