Action Potential bandwidth

Richard L. Hall rhall at
Sun Aug 20 10:16:59 EST 2000

Another Richard "speaks":

There may be other approaches to this question.

Consider that the vast majority of electrical activity in the brain 
involves graded or electrotonic synaptic transmissions which are 
independent from spiking or firing of action potentials.  Thus, any 
approach to determining power requirements and information processing 
that relies on the current densities in axons projecting information 
to distant loci greatly underestimates matters.  In other words, 
counting action potentials falls far short of providing an answer.

Retinal processing (and thus energy consumption) is essentially 
constant in light and in dark..  Why?  Receptive fields function as 
either center on or center off .  Some are turned on with 
illumination and the others are turned off by illumination.  Most of 
the energy in the retina is due managing the massive sodium leak into 
photoreceptor cells and supporting graded synaptic transmission 
between photo receptors and bipolar cells and between bipolar cells 
and ganglion cells.  The retina basically enhances contrast and is 
there for "reductionist" in the sense that 100,000,000 photo 
receptors input into one million ganglion cells per eye.   Ganglion 
cells produce action potentials transmitting visual information to 
the thalamus (>95%) and superior colliculus (<5%).  In theory you 
should be able to model information processing to the level of the 
thalamus and superior colliculus easily.  Energy costs could be 
estimated  by spectrophotometrically  determining oxygen consumption 
in a retina using something akin to pulse oximetry or if you have the 
bucks use PET.

But what happens next in terms of information processing gets vastly 
more complex.   Thalamic relays do not appear to be more than simply 
grouping visual fields for routing to the primary visual cortex.  The 
processing within the superior colliculus is much more substantial 
since it triggers complex motor reflexes for tracking, regulating 
pupillary and accommodation reflexes, and activating all sorts of 
somatic motor responses for the head and neck.  The primary visual 
cortex has a host of receptive fields and relays scattered throughout 
seven cortical layers organized in columns.  Information begins to be 
simultaneously consolidated and elaborated.  Then relayed to 
secondary visual centers and then to association areas and then who 
knows where?

Similar modelling is possible for auditory systems, corticospinal and 
extrapyramidal (sorry but I like the older terminology) motor 
systems, and even autonomic responses.

The bottom line is this:  We can measure oxygen consumption in the 
CNS and thus have indirect estimates of metabolic costs...about 600 
Kcal per day = one good slice of cake per day.   On the other hand, 
estimates of the calculating power of the brain is more difficult. 
One method would be to use the retina as a basic model (equaling 
three layers of cortex) and use a mass scaling algorithm defining the 
brain in retinal calculating units.  Another approach would be to 
analyze the EEG of a discrete region of the brain, assume that the 
EEG measures graded potentials (not action potential), integrate 
and"divine" a scalar to predict the calculating power per second of 
the entire neocortex then fudge a number for the subcortical nuclei 
and brain stem.

 From posts by the other Richard (Norman) and many, many non-Richards 
over the years, we usually end with statements like "the brain 
consists of massively parallel megabit processors".  That may be an 
over estimate if the retinal calculations are correct!  Then again, 
we have not really answered the older question of how many angels can 
dance on the head of a pin.  But it is fun to run through matters 
from time to time.


>Thanks for the reply.  Let me clarify a little more of what I was
>attempting to get at.
>My understanding is that a "firing" of a neuron causes an action
>potential to propagate down the axon.  I think I misused the term
>"action potential".  Perhaps you could tell me the right term for
>the event which is a neuron triggering and propagating an AP down
>the axon.  I'll call it a "firing" here.
>Anyway, it seems I ought to be able, at least in principle, to count
>the number of neuron "firings" per second in the entire human brain.
>I'd like to know this number because I'd like to compare it to the
>bandwidths at various points inside a computer.  Here, I'll give you
>the other number: a 1 GHz Pentium-III has about 10^15 transistor gate
>voltage changes per second (30 watts, 1 GHz, operating at 2 volts,
>and the average transistor is about 5 or 10 fF).  My guess is that
>this number is about the same as the number of times, per second,
>that a gated ion channel turns on or off in the human brain.  My
>guess is that there are maybe 10^11 neuron firings per second.
>I thought perhaps I could get to this number by using an energy
>consumption argument.  The voltages and currents and resistances in
>the membrane might be changing, but in the end, if sufficient charge
>(sodium and potassium ions) crosses a capacitor (the cell membrane)
>to change the field across that capacitor from one voltage to
>another, that dissipates an amount of energy that I can calculate --
>I do it all the time when designing CMOS.
>I think the chemical work of ions moving down gradients is exactly
>the electrical work of charge moving across a capacitor.  And to
>provide the energy for that work, the brain burns ATP->ADP to pump
>the ions back up these gradients to recharge the capacitor for the
>next firing.  When you point out the other work the brain does, in
>processing and reprocessing the synaptic transmitters, you're right,
>and I didn't know how to account for that except with a fudge factor.
>You point out that there is a lot of variation in neuron surface
>area, and in the current density on that surface, so maybe that's
>the wrong way to chase down the number I'm looking for.
>You also point out that gated ion channels typically see currents
>of a few pA, which would lead to energy dissipation of a few fJ per
>action potential per gated ion channel.  And assuming that this
>energy dissipation is 10% of the brain's power dissipation, that
>gives about 10^15 of these events per second for a 25 watt brain.
>-Iain McClatchie                            650-364-0520 voice
>                       650-364-0530 FAX
>iain at                             650-906-8832 cell

Richard L. Hall, Ph.D.
Comparative Animal Physiologist

University of the Virgin Islands
2 John Brewers Bay
St. Thomas, U.S.V.I. 00802

340-693-1385 FAX

rhall at

"Live life on the edge...the view is always better"  rlh


More information about the Neur-sci mailing list