doi:10.1016/S0893-6080(01)00144-7
Copyright © 2002 Elsevier Science Ltd. All rights reserved.
Invited article
Synapses as dynamic memory buffers
References and further reading may be available for this article. To view references and further reading you must
purchase this article.
Wolfgang Maass
, 
, a and Henry Markram, b, 1
a Institute for Theoretical Computer Science, Technische Universität Graz, Inffeldgasse 166, A-8010 Graz, Austria
b Department of Neurobiology, The Weizmann Institute for Sciences, Rehovot, 76100, Israel
Received 27 September 2000;
accepted 16 August 2001.
Available online 30 November 2001.
Abstract
This article throws new light on the possible role of synapses in information transmission through theoretical analysis and computer simulations. We show that the internal dynamic state of a synapse may serve as a transient memory buffer that stores information about the most recent segment of the spike train that was previously sent to this synapse. This information is transmitted to the postsynaptic neuron through the amplitudes of the postsynaptic response for the next few spikes. In fact, we show that most of this information about the preceding spike train is already contained in the postsynaptic response for just two additional spikes. It is demonstrated that the postsynaptic neuron receives simultaneously information about the specific type of synapse which has transmitted these pulses. In view of recent findings by Gupta et al. [Science, 287 (2000) 273] that different types of synapses are characteristic for specific types of presynaptic neurons, the postsynaptic neuron receives in this way partial knowledge about the identity of the presynaptic neuron from which it has received information. Our simulations are based on recent data about the dynamics of GABAergic synapses. We show that the relatively large number of synaptic release sites that make up a GABAergic synaptic connection makes these connections suitable for such complex information transmission processes.
Author Keywords: Dynamic synapses; Spiking neurons; Memory
Fig. 1. Occupancy of dynamic states of a synapse (values of u and R, each divided into 100 equal sized bins) after different spike trains were sent to this synapse. N=250 Poisson spike trains, with an average firing rate of 30 spikes/s over a time interval of 1000 ms, were sent to a synapse with synaptic parameters U=0.16, F=376, D=45 (F1-type synapse in the terminology of Gupta et al. (2000).
Fig. 2. Protocol for using a dynamic synapse for storing and retrieving information. Information about a preceding spike train S (i.e. information about k and 
Δt,…,Δtk
) is stored in the dynamic state of a synapse at any later time point t0, and can subsequently be retrieved by measuring the amplitudes of postsynaptic responses for ‘test spikes’ at time points t0+Δs1 and t0+Δs1+Δs2.
Fig. 3. Mutual information that the amplitudes of postsynaptic responses for 1, 2, 3 subsequent ‘test spikes’ (results marked by +, *,×) contain about the identity of the preceding spike trains (in this case: uniform distribution over N=300 randomly chosen Poisson spike trains over 1000 ms, average firing rate 30 spikes/s), normalized by dividing through the entropy log N of the input distribution. Results shown are for the same F1-type synapse as in Fig. 1. Noise in the amplitudes of postsynaptic responses is modeled by assuming that the synaptic connection consists of 20 binary (release/no-release) stochastic contacts, each with current release probability unRn according to (2), (3), (4) and (5). The resolution (=number of bins) for amplitudes is plotted on the x-axis.
 |
Fig. 4. Mutual information about past spike train contained in the amplitudes of postsynaptic responses for two subsequent test spikes (Δs1=5, Δs2=50) is shown on the z-axis (normalized like in Fig. 3). More precisely, the past Poisson spike trains with an average firing rate of 40 spikes/s over 900 ms are divided into three segments of length 300 ms each, and mutual information with each of the three segments is measured separately (the numbers 1–3 on the x-axis denote the segment number). Ten Poisson spike trains over this segment are each combined with 1000 Poisson spike trains over the other two segments, which play here the role of additional ‘noise’. Apart from that the same model for noise in the amplitude of postsynaptic responses as in Fig. 3 is assumed, but for varying numbers of binary synaptic release sites (5, 10, 20, 30, 100; plotted on the y-axis). Results shown are for the same F1-type synapse as in Fig. 1. One can see that the information about the preceding spike train that the amplitudes of postsynaptic responses for subsequent spike contain is dominated by information about the most recent segment of the preceding spike train. Furthermore, this simulation shows that the amplitudes of postsynaptic responses contain a significant amount of information about the preceding spike train only for synapses with 20 or more synaptic release sites.
Fig. 5. Mutual information of the amplitudes of postsynaptic responses for two subsequent spikes about the identity of the preceding spike train (dark bars) and the identity of the synapse-type (light bars). Two hundred Poisson spike trains (mean 20 spikes/s, over 1000 ms) were sent to F1-, F2-, and F3-type synapses (in the terminology of Gupta et al., 2000); assumed parameters U=0.25, F=21, D=706 for F2-type and U=0.32, F=62, D=144 for F3-type synapses. Different levels of resolution in terms of the number of equal sized bins for amplitudes are plotted on the x-axis.
1 Tel.: +972-89343179; fax: +972-89316573.
Corresponding author. Tel.: +43-316-873-5822; fax: +43-316-873-5805; email: maass@igi.tu-graz.ac.at
Abstract
Purchase PDF (211 K)
E-mail Article
Add to my Quick Links
Cited By in Scopus (17)
Purchase PDF (196 K)
