Dynamic Modulation of Neural Bandwidth
This page describes how the common input model can be used
to infer the bandwidth of a model neurone under large scale synaptic input.
Definition and measurement of neural bandwidth
Common input model configuration: Two identical model motoneurones,
with extensive dendritic tree (12 dendrites, 248200mm2),
and a total synaptic input of 32,000 EPSPs/second to each cell. A percentage
of this synaptic input is shared between the two cells, this is the subset
of the synaptic inputs for which the bandwidth is to be estimated.
- Neural bandwidth is the range of frequencies (in the Fourier sense)
over which the firing characteristics of the pre-synaptic inputs under consideration
can modulate the output discharges of the two neurones.
- A Correlation analysis of the output discharges can be used to indirectly
infer the neural bandwidth for the subset of the total synaptic input which
is common to both cells.
Bandwidth for Uncorrelated Synaptic Input
|Correlation analysis of the output discharge of the two cell model with
50 % common synaptic inputs. Each cell has 1000 inputs firing randomly at
32 spikes/sec, they share 16,000 EPSPs/sec.
(a) The two cells demonstrate a tendency for synchronous discharges
- due to the common input. (b) The range of frequencies involved is up to
Thus the bandwidth for this configuration, which reflects the ability
of 50% of the total synaptic input to modulate the output discharge, is 50
The strength of correlation between the two outputs (and thus the neural
bandwidth) varies systematically with the percentage of common input. With
10% common input there is no significant coherence between the outputs (bandwidth
is zero). With 80% common input significant coherence is present to frequencies
in excess of 200 Hz (bandwidth exceeds 200 Hz); the peak coherence is around
- Despite the fact that the two cells share 50% of their total synaptic
input, the output discharges are only weakly correlated at frequencies up
to 50 Hz. The coherence indicates that the output discharge of one cell can
predict at most 8% of the variability in the output discharge of the second
Bandwidth for Correlated Synaptic Input
- The bandwidth for 80% of the total synaptic inputs is in excess of
- With 80% shared synaptic input (25,600 EPSPs/sec) the output discharge
of one cell can predict at most 16% of the variability in the output discharge
of the second cell.
Altering the correlation structure, without altering the firing rates, or
numbers of inputs, has a dramatic impact on neural bandwidth.
||Coherence between the output discharges of the two cell model with 10
% common synaptic inputs (3200 EPSPs/sec), provided by:
(a) One hundred broad band inputs (32 spikes/sec; c.o.v = 1.0) which are
weakly correlated over a broad range of frequencies.
(b) Two sets of periodically firing weakly correlated inputs (160 inputs at
10 spikes/sec; 64 inputs at 25 spikes/sec) which are correlated at their mean
Inputs are distributed uniformly over the dendritic tree.
Dynamic Modulation of Neural Bandwidth.
- The correlation between the outputs accurately reflects the correlation
structure in the common inputs - The bandwidth of the neurones is determined
by the correlation structure present in the pre-synaptic inputs. In other
words the bandwidth of the neurone is determined by the firing characteristics
of the pre-synaptic inputs.
- A small change in the temporal correlation structure of a sub set
of the total synaptic input, without any alteration in mean firing rate,
can alter the bandwidth of the cells.
- The strength of correlation amongst the input spike train required
to achieve this is sufficiently weak, that a correlation analysis of pairs
of sample inputs may fail to reveal the presence of any correlation. The
peak coherence between sample inputs is typically around 0.01.
For further details see the following article
Halliday, D.M. (2000). Temporal correlation of large scale synaptic input
is a major determinant of neuronal bandwidth, Neural Computation,
12, 693-707. PDF is available in the publications page.
Abstract is available here.
If you require further details please contact David Halliday.
Next The Role of Correlated Synaptic
Activity in Neural Integration
Last Modified 09 July 2002