neural coding of behavior: evidence for precise timing of spikes ?

Mark Laubach laubach at biogfx.neuro.wfu.edu
Fri Dec 8 15:25:04 EST 1995


carp at neuro.med.umn.edu wrote:

>There was a poster from the Merzenich lab at this year's SFN meeting 
>(461.13) which demonstrated stimulus encoding by temporal correlation of 
>spikes.  Pairs of neurons were recorded from primary auditory cortex in 
>anaesthetized monkeys during tone pulses - often, changes in firing 
>rated would signal the onset and offset of the stimulus, but not the 
>entire stimulus duration.  Cross correlations, on the other hand, were 
>(at least in some examples) elevated during the entire stimulus 
>duration, sometimes with no corresponding change in firing rate 
>whatsoever.  Since the monkeys were anaesthetized one cannot 
>assign any "behavioral significance" to such a finding - nevertheless I 
>think this is a more compelling demonstration of temporal integration 
>than those seen in slice preparations.

>Adam Carpenter
>University of Minnesota

The abstract from the poster you mention did not include info on the
bin size used in the correlational analysis.  Do you know if this
correlation occurred with precision (i.e., 1-5 msec bins) or over a
longer time scale (i.e., 10s of msec)?

Also, a similar finding was previously reported by Vaadia et al.
(Nature, 373, 515).  One of their figs (#2) showed a neural
interaction between a pair of cells in the frontal cortex in the
absence of phasic activity by either cell.  I think that this
observation is also "a more compelling demonstration of temporal
integration than those seen in slice preparations",  In fact, these
data were collected in _behaving primates_.  However, the JPSTHs used
in this paper were constructed with large bins (10s of msec),
suggesting that these interactions were NOT based on precise spike
timing, but on an overall covariance in the neuron's spike trains that
was due to a _neural_ process (i.e., the JPSTH correction was used
here).  

I find it interesting here that these data (if I remember the ref
correctly) were used by Ferster and Spruston (their ref. 13) in a
recent Science perspective to support the view that precise spike
timing is a potential code for the CNS.  As large bins were used in
the JPSTHs, one can NOT conclude that precise spike times were the
basis for this neural correlation!

I want to point again that correlational analysis can not directly
address the issue of neural coding.  That is, if one is to conclude
that a pattern of spike activity is a putative code for some neural
process related to stimulus or behavior, then one must show that the
pattern of activity is _predictive_ of the stimulus or behavior.  One
can only make meaningful predictions, in the sense of statistical
pattern recognition, based on trial-by-trial, or stimulus-by-stimulus,
analyses.  

Correlational data are of utility, however, in that they allow one to
examine the interactions between elements in a neural ensemble. For
example, if one finds that a neural code is redundant across an
ensemble of neurons and that the cells are cross-correlated, then one
can interpret the data as evidence for coding distributed across the
ensemble of neurons.  One can then address the basis of this
redundancy by examing, for example, whether the cells had correlated
noise (e.g., Gawne and Richmond, J. Neurosci., 13:2758-2771).   

There has been a lot of discussion about population activity and
emergent coding.  However, redundancy is also interesting, especially
when the redundant neurons are located in different regions of a
structure or in different parts of the CNS.  Such redundancy may mean
that common information is "broadcast" over large regions of the CNS.

Finally, I think that recent developments for large-scale parallel
recording (pioneered by our lab) and for methods for the analysis of
temporal variations by single neurons and spatiotemporal variations by
neural ensembles have led us to a new "paradigm" for neurophysiology.
It is no longer adequate to account for neural activity in terms of
the percent of cells that responded under different conditions.  We
need to address the statistics of neural response properties to
further our understanding how neurons and neural ensembles process
information.

-----------------------------------
Mark Laubach
Dept. of Physiology & Pharmacology
Bowman Gray School of Medicine
Wake Forest University
Winston-Salem, NC 27157
laubach at biogfx.neuro.wfu.edu
-----------------------------------




More information about the Neur-sci mailing list