Page 1 of 1

Fixed temperature -- custom initialization required?

Posted: Sat Oct 18, 2008 12:04 am
by kathryn
I've been reading about custom initializations in this forum (e.g. ... 841&p=2890) and in chapter 8 of The NEURON Book. Despite my reading, I'm still not sure whether custom initialization is necessary when the global 'celsius' is fixed at a value other than the default temperature of 6.3 degrees. I want the temperature to remain constant for the entire simulation, so I set it to the desired temperature in my hoc code:

Code: Select all

celsius = 35
Is that OK? [Edit: Maybe not, given the following...

At 6.3 degrees a basic model cell (hh mechanism but no other channels) fires a few action potentials when injected with 0.1nA of depolarizing current. The action potentials are abolished at 35 degrees, a result which is not consistent with our in vitro experiments on peripheral neurons. (We perform all wet experiments at 34-35 degrees.) At the latter temperature it takes an excessive stimulus (3nA!) to provoke a dodgy action potential. I realize the hh mechanism was derived from experiments on the squid giant axon (at much lower temperatures), but is there any hope for my model?]

Re: Fixed temperature -- custom initialization required?

Posted: Sat Oct 18, 2008 10:29 am
by ted
First, a general comment: celsius affects only those mechanisms whose source code specifications explicitly use celsius. For mechanisms specified with NMODL, this means that the ASSIGNED block must declare
celsius (degC)
and that some equation block (e.g. FUNCTION, PROCEDURE, DERIVATIVE, BREAKPOINT) contain an assignment statement whose right hand side involves celsius.

Next a comment about temperature effects on spikes: warming makes spikes smaller for two reasons. It speeds up gNa activation and inactivation, so depolarization produces briefer inward current and less net depolarization (dV = dq / C, where dq = total charge entry). It also speeds up gK activation, which diminishes the net depolarization that the Na current can produce (plot iNa and iK vs. t and you'll see that, even at 6.3 deg C, iK is only slightly delayed compared to iNa--indeed, unless you look very closely, their time courses seem to overlap almost completely).

The experiments from which the parameters of the Hodgkin-Huxley squid axon model were derived were performed on cold-adapted invertebrates in very cold sea water. That model is not going to generate fast spikes, and it will not produce robust spikes at warm temperatures. If you need fast spikes, or spikes that hold up at warm temperatures, you'll have to use mechanisms based on experiments on vertebrate neurons, preferably mammalian, at warm temperatures.

A caveat: there are very few studies of the effects of temperature on voltage-gated channels, and there is good reason to believe that their Q10s are not 3 (see
How to implement temperature dependent rates

Posted: Tue Oct 21, 2008 3:23 am
by kathryn
Given the number of vertebrate neurons included in the ModelDB database (a subset of which are presumably mammalian), I'm surprised my various web/literature/book/ModelDB searches have failed to find such a mechanism. I'm attempting to model a cell type in the enteric nervous system, which is -- unfortunately for me as a newcomer to NEURON -- always studied near 37 degrees Celsius.

I haven't given up just yet, because I'm not certain I know how to find the mechanism I need (if it exists). Any suggestions?

Modeling, if nothing else, is good at pointing out the gaps in our scientific data and knowledge.

Re: Fixed temperature -- custom initialization required?

Posted: Tue Oct 21, 2008 7:44 am
by ted
The weaknesses of any search strategy are its propensities to false positives and false negatives. Searching ModelDB for entries that contain both celsius and mod (i.e. enter
+celsius +mod
in the search field) is vulnerable to obvious false positives and false negatives (among the latter are all non-NEURON implementations that might have useful parameters, and everything that omits the keyword "celsius"). However, it's a start, and as of this moment it turns up 62 model entires, at least some of which look promising. As part of your screening before testing any code, it would be highly advisable to read at least the methods sections of the associated paper(s).