IUBio Biosequences .. Software .. Molbio soft .. Network News .. FTP

Dendritic Integration (was: high resolution intracellular recordings?)

yan king yin y.k.y at lycos.com
Wed Nov 12 01:47:40 EST 2003

I posted this question 2 months ago:
(Sorry I was busy with other stuff...)

> I'm trying to find intracellular recordings of the soma of
> in vivo neurons, with *temporal* resolutions in the sub-ms
> range (the finer the better). What I want to see is how the
> spatial and temporal integration actually take place in action.
> Any web page, paper etc...?

Matthew Kirkcaldie:

|There are many such papers, just look for whole-cell patch clamp 
|recordings in a lit search engine like PubMed.  They have recordings in 
|the picoampere/microsecond range routinely these days.  Generally they 
|can be difficult to read for "outsiders" (and I include myself here) 
|because they generally refer to inward and outward currents of ions 
|under current or voltage clamp conditions, and perform considerable 
|restrictions on the constituents of the nutrient bath to isolate certain 
|types of events. 

|I doubt you could find what you're looking for, which I think is a 
|recording showing how EPSPs combine to kick the membrane over threshold.  
|Believe it or not, the generation of an action potential is still far 
|from understood, despite the known presence of voltage-gated sodium and 
|potassium channels in the membrane near the start of the axon.  The work 
|of Stuart and Hausser is a recent and fascinating attempt to grapple 
|with the dynamics of these events - have a look at their papers.  The 
|story they reveal is daunting to say the least - the stereotype of 
|dendrites as passive antennas radiating EPSPs to the cell body, which 
|then fires if threshold is exceeded, is so over-simplified that it's 
|cartoon-like.  Instead, non-linear channel-mediated events can occur 
|throughout the dendrite field, generating lossless sodium or calcium 
|spike potentials which propagate to the cell body and interact on the 
|way.  In addition, when the cell fires, the action potential spreads 
|back into the dendrite field and interacts with synaptic events there.  
|It ain't simple (and I would venture it's not modellable in any 
|meaningful sense either!).
I've just read part of Stuart, Hausser & Spruston's book "dendrites"
and they talked about active dendritic spikes which was first
documented in 1950-60s. It looks like these spikes will add a
nonlinear component to dendritic integration. But the bottomline
remains that EPSPs are being 'integrated' (more or less additive)
by dendrites. So the classical picture has not changed.

The fact remains that distal EPSPs are small when they arrive at
the soma as compared to those from more proximal sites. This
accounts for what's called 'synaptic strengths'. From what I read,
a synapse on the soma can result in an EPSP ~5mV high, versus
a distal EPSP will attenuate to less than 0.5mV at the soma, ie
10 times. So this means synapses do have different strengths
(though I'm not sure what's the maximum range of synaptic
strengths). There's a theory that some kind of auto-regulation
will result in equalization of synaptic weights because only
those synapses that cause firing will remain in the long term.
But this does not take away the fact that synaptic weights take
a *range* of values. And that's probably the basis of memory.

Secondly, I think it has been firmly established that generation
of APs at the axon hillock is the result of the membrane voltage
exceeding the threshold, which has its basis in voltage-gated ion
channels. This implies that EPSPs are added up and compared
against the threshold. The nonlinearity may give rise to more
complexity which the Sigma-Pi model serves as a generic model.

So my point is that if we look at the sub-threshold membrane
voltage at the soma we can tell what kind of dendritic inputs
are being recieved at a time slightly earlier. You can't see
that with low sampling rates (ms range) but maybe with sub-ms
resolution it's possible. This is clearly doable with current
techniques but maybe no one has bothered to do that.

(In my original post I said it might be possible to discern
distal EPSPs with recordings at the soma, which is obviously
false because of the smearing effect. But it's still possible
to know the *result* of dendritic summation by recording

> I've read from Koch's "Biophysics of Computation" that it
> takes about 64 EPSP inputs *at* the soma close together to
> generate an action potential. (The threshold being about 16mV
> above resting potential and each EPSP typically around 0.3mV).

|Not really a very meaningful statement, since EPSPs don't typically 
|arrive on the cell body anyway - and their summation is a decidedly 
|non-linear process.  The voltages would be microvolts, by the way, and 
|threshold potential varies a great deal according to the type of neuron.

> Is it possible to actually discern individual contributions
> of EPSPs from such recordings? What about background noise
> that is not from EPSPs?

|EPSPs are easily resolved, in fact events called "minis" are routinely 
|resolved, which are the result of spontaneous release of a *single* 
|transmitter vesicle at a synapse!  Generally patch clamps achieve a 
|gigohm seal, meaning that the recordings are pretty robust and noise 
|isn't a problem in a good rig with a good preamp.

|Sorry for the negativity - I often feel that people who are confident 
|that the brain can be computationally modelled at the cellular level, 
|just don't understand the problem sufficiently.  I don't mean to say 
|that the brain can't be computationally modelled using abstractions or 
|analogies, but there is a very real gulf between the biological and the 
|discrete which is not surmountable by any means I'm familiar with.

I'm saying if the dendritic integration picture is basically
correct, it'll be possible to model neurons with current models.
Unless there're some problems I've missed??

Thanks for you reply, hope you're still there =)

Richard S Norman:

|Matthew Kirkcaldie has already give you some valuable information.
|But here is some other stuff you should realize.

|First, the cell membrane has a relatively large capacitance resulting
|in a time constant at least several msec in duration.  That is, the
|membrane potential is not easily changed in times much shorter than
|this. I am talking about "passive" potentials, here which includes the
|spread and summation of psp's.  Active potentials are different. In
|the production of an action or a synaptic potential, open channels
|produce a high conductance or low resistance membrane, hence a fast
|time constant.  In other words, the passive properties of the membrane
|act as a low-pass filter and events in the sub-millisecond range are
|drastically reduced in amplitude.  No such activity spreads very far
|down a dendrite.  Note:  this whole discussion ignores the active
|responses of dendrites that Matthew talks about.  Still, it is a good
|first approximation to what happens.

This is totally correct. I made a mistake in the original post...

|Second, intracellular recording technique with microelectrodes
|introduces its own limitations. With an electrode resistance in the
|tens of megohms, a stray capacitance of a few picofarads produces
|another time constant and a low pass filter so that the signal is
|limited in bandwidth.  In addition, the high electrode resistance
|produces random "thermal" noise that obscures small signals.  In other
|words, it is very, very difficult to see small, fast signals, even if
|they really are there.  Patch clamping, again described by Matthew,
|uses very different electrodes and has very different constraints, but
|often the type of data you want is best seen by the older
|intracellular microelectrode.

I figured that with metal electrodes, and _intra_cellular recording,
the resolution could be much higher (electrode resistance practically
zero, frequency response practically perfect). I'm not sure why that's
not being done?

|Then you have the problem of actually seeing individual events in the
|smear of membrane potential.  The "classical" view of synaptic
|integration is a "simple" spatial/temporal summation of potentials on
|a distribution R-C cable.   Again, Matthew shows why this is
|inadequate, but again it is a good first approximation.  The
|properties of the cable equation (the low-pass filtering of the
|membrane I referred to earlier) means that the psp's produced by
|distant synapses reach the soma as a blur.  The potentials are
|drastically reduced in ampltude an drastically slowed and spread out
|in duration.  It is not at all possible to separate out individual
|contributions.  And the cell may not even care about individual
|potentials -- in may be the "mass effect" in changing the general
|excitability of the cell that is the "purpose" of this wiring pattern.

|You may have in mind a model of synaptic integration that works like a
|logic equation:
|  Fire an action potential IF (A AND B) OR (C AND D AND E) BUT NOT (F)
|or something of the kind but with hundreds of terms.  You want to see
|the data with enough resolution to work out each individual term.  It
|doesn't work that way.

As I said above, I believe the Sigma-Pi model is general enough
to cover all aspects.

It may be possible to record V_soma and deduce from it what the
*sum* of the synpatic inputs have been. With a lot of trials it
may be possible to *regress* those parameters (synaptic weights).
If this is feasible it may have some very useful applications =)

Thanks a lot,
Yan King Yin

More information about the Neur-sci mailing list

Send comments to us at biosci-help [At] net.bio.net