## Current injections and cell voltage response

Anything that doesn't fit elsewhere.
maya
Posts: 6
Joined: Mon Sep 27, 2010 9:10 am

### Current injections and cell voltage response

I am injecting square negative current at a specific place in a reconstructed cell in order to discover the distribution of the channels in the cell (by comparing the results to current injections in a real cell from my experiment).
The problem is that no matter what I do in the model, I receive about half the voltage change I get in the real cell.
I tried to change Ra, g_pas, channel distributions, e_pas.

What do you think is the problem?

ted
Posts: 5768
Joined: Wed May 18, 2005 4:50 pm
Location: Yale University School of Medicine
Contact:

### Re: Current injections and cell voltage response

maya wrote:I am injecting square negative current at a specific place in a reconstructed cell in order to discover the distribution of the channels in the cell (by comparing the results to current injections in a real cell from my experiment).
This is a classical optimization problem in experimental neurophysiology (discussed in more detail at the end of this post). The usual problem is nonunique fits, i.e. a wide range of parameters gives nearly identical simulation results. But you're having the opposite problem--
The problem is that no matter what I do in the model, I receive about half the voltage change I get in the real cell.
I tried to change Ra, g_pas, channel distributions, e_pas.
Here are a couple of hints that may help.

First, are you sure that your model's spatial grid is fine enough for good spatial accuracy? Make a very simple model that contains only the anatomical specifiation, and set cm and Ra to the largest values that you are willing to accept as "plausible"--maybe 1.5 or 2 uf/cm2, and 150 or 200 ohm cm, respectively. Then use the d_lambda rule to adjust nseg of each section so that no segment is longer than 0.1 length constant at 100 Hz.

Using these values for nseg, try to fit a completely passive model to your experimental data. Let cm, g, and Ra be uniform throughout the cell and see if you can get a decent fit. Once you have tried that, you will be ready to introduce the additional complexities of active currents and nonuniform membrane properties.

Comments on the problem of tuning computational models of neurons to experimental observations

In the simplest case, the perturbation of membrane potential is small so that membrane conductance can be treated as linear and time-invariant. Under such conditions, given very clean recordings of membrane potential (under current clamp) or clamp current (under voltage clamp) it can be possible to adjust a model's parameters (specific membrane capacitance cm and conductance g, and cytoplasmic resistivity Ra) so as to get a very close fit to experimental data.

However, the usual experimental data are obtained by injecting and recording at a single site, with very simple stimulus waveforms. Furthermore, noise degrades the quality of the recordings. This has two serious consequences. First, the resulting data are usually insufficient to constrain all of the model's parameters. Underconstrained optimization results in nonunique fits. Second, it turns out that goodness of fit is much less sensitive to Ra than it is to cm or g (particularly when current injection and voltage recording are done at only a single location).

Another factor that must be considered, but is often overlooked by those whose background is in engineering, math, or physics, is the quality of the anatomical data on which the computational model is based. Tissue shrinkage, incomplete dye fills of cells, amputation of branches during brain slice preparation or post-experiment tissue processing, light scatter by tissue, difficulty resolving the diamters of fine branches that are only 2 or 3 times wider than the wavelength of light (in cells that have extensive dendritic trees, more than half of the membrane surface area may be located in branches with diameters on the order of 1-2 um and are therefore subject to signifcant blurring)--all of these combine to limit the experimental accuracy and precision of anatomical data. Yet another factor that is almost universally ignored is the fact that most models treat diameter measurements as if cell branches all have circular cross-sections--but anyone who has looked at electron micrographs knows that almost no cell processes have circular cross-sections.

So "non-uniqueness" and "approximate results" are to be expected under the best of circumstances.

But there is another practical issue. In the best of all possible worlds, one's model should combine anatomical and biophysical experiments obtained from the same cell. This is technically very difficult; only a few papers report the results of such experiments.

So given all of that, why does anyone bother? Well, it turns out that even approximate results can be useful. One example is the work of
Chen, W.R., Shen, G.Y., Shepherd, G.M., Hines, M.L., and Midtgaard, J.
Multiple modes of action potential initiation and propagation in mitral cell primary dendrite.
J. Neurophysiol. 88:2755-2764, 2002.
and
Shen, G.Y.Y., Chen, W.R., Midtgaard, J., Shepherd, G.M., and Hines, M.L.
Computational analysis of action potential initiation in mitral cell soma and dendrites based on dual patch recordings.
J. Neurophysiol. 82:3006-3020, 1999.
These authors used a model with simplified anatomy that captured the essential architecture of mitral cells, and constrained the biophysical properties of the model by fitting it to data obtained by simultaneous recordings from soma and dendrite (this helps particularly with Ra) under four different experimental protocols:
weak or strong stimulus applied to the soma
weak or strong stimulus applied to the dendrite

Another group has used a somewhat different approach that is described in these papers
Keren, N., Peled, N., and Korngreen, A.
Constraining compartmental models using multiple voltage recordings and genetic algorithms.
J. Neurophysiol. 94:3730-3742, 2005.
Gurkiewicz, M. and Korngreen, A.
A numerical approach to ion channel modelling using whole-cell voltage-clamp recordings and a genetic algorithm. PLoS Computational Biology 3:1633-1647, 2007.
Keren, N., Bar-Yehuda, D., and Korngreen, A.
Experimentally guided modelling of dendritic excitability in rat neocortical pyramidal neurones.
Journal of Physiology-London 587:1413-1437, 2009.
You'll find more papers by Korngreen in the list of papers that used NEURON http://www.neuron.yale.edu/neuron/stati ... ednrn.html, but these should give you an idea of what they did.

Working source code for many of the above cited papers are available from ModelDB http://senselab.med.yale.edu/modeldb/