NEURON implementation of spiny neuron model

Address questions to Ted Carnevale
ted.carnevale@yale.edu

The following comments show how to operate these programs, and discuss the implementation of the models and how to reuse this code.

Usage

1. Unzip spinycell.zip into an empty directory.
2. Compile the mod files with nrnivmodl (or mknrndll if you're using MSWin)
3. Follow the instructions below.

Figs. 2 and 3A

Figs. 3B and C

to bring up a plot that reproduces Fig. 3B

or Fig. 3C

Fig. 4

Fig. 6

Fig. 6a

Fig. 6b

1. The first 200-300 ms of each trace is a "warmup" or "stabilization" interval during which the model reaches a dynamic equilibrium appropriate for the baseline synaptic drive.
2. The increase of synaptic drive starts at 400 ms.
3. The pseudorandom afferent spike trains are different on each run, so each press of these buttons produces a slighly different family of traces. However, the envelopes of the individual trajectories are essentially similar from run to run.
   Example: repeat "Fig. 6b run"
   
4. Mean synaptic conductance appears to be the primary determinant of the membrane potential trajectories. Other details of synaptic input (negexp ISI spike trains driving the relatively simple ExpSyn mechanism here, vs. the more complex, empirically derived model of synaptic activity used by Gruber et al.) affect only the low-level fluctuations of the individual traces.

Figs. 7A and B

Fig. 7A

Fig. 7B

Fig. 10A and B

• the membrane potential of the spiny neuron
• the cortical input spike train
• the time course of the modulatory action of dopaminergic input

Example 1: weak cortical input, target unrewarded

Example 2: strong cortical input, target rewarded

1. During the first few hundred ms, the model reaches a dynamic equilibrium appropriate for the background synaptic drive.
2. Sequential runs under the same conditions will differ because a different afferent spike train is used each time.

Model implementation

Among these files there are actually two different implementations of a spiny neuron model. The first is used to generate the figures that illustrate the basic properties of the model (Figs. 2 - 4 and 7), and the second is used for those figures that show how dopamine affects the model's response to afferent spike trains (Figs. 6 and 10).

Both implementations use the same membrane currents, which are specified by these files:

  caL.mod
  kir2.mod
  ksi.mod
  leak.mod

Some of the programs also use the mechanisms specified in one or both of these files:

  damsg.mod
  gspike.mod

Implementation 1: a "stand-alone" model cell

These were set up with NEURON's GUI, and they are recreated by having NEURON execute the code in model_mu_1.ses. Think of this file as a "virtual experimental preparation," somewhat analogous to a single cell in tissue culture or an acutely isolated soma.

This file is loaded by the following programs :

  figs2and3A.hoc
  fig3BC.hoc
  fig4.hoc
  fig7a.hoc
  fig7b.hoc

msg = 1

soma for (x,0) { // skip the nodes at 0 and 1
        setpointer mu_caL(x), msg
        setpointer mu_kir2(x), msg
}

For fig7a.hoc and fig7b.hoc, computational efficiency requires a different approach. Unlike other simulations involving the "stand-alone model," these need msg to have nontrivial dynamics. Using a hoc scalar and modifying NEURON's main computational loop to force incremental changes in msg at every fadvance() is not the best way to do this.

Therefore fig7a.hoc and fig7b.hoc employ a DAsyn point process. This "net-ready" synaptic mechanism has a state variable msg which, in this implementation, has first order dynamics and can be driven by spike events. The file fig7syn.hoc contains the statements that link the "mu" POINTERS of caL and kir2 to this msg.

DAsyn is also used in the program that reproduces Fig. 10 (see below).

Implementation 2: a "networked" model cell

  fig6.hoc
  fig10ab.hoc

The NetReadyCellGUI tool that specifies the properties of this "biological model cell", and the ArtCellGUI tool that specifies the properties of the artificial spiking cells that generate the afferent spike trains, were created using NEURON's GUI and saved to session files called fig6netcells.ses and fig10netcells.ses.

As noted above, the conductance fluctuations used in the simulations reported by Gruber et al. are quite different from those generated by spike trains with negexp statistics converging on an ExpSyn mechanism. Even so, the simulation results produced by these files are qualitatively quite similar to the published figures. Mean synaptic conductance, rather than its detailed fluctuations, appears to be the principal factor that governs the time course of membrane potential.

fig6.hoc uses a simple hoc scalar called msg to represent dopaminergic effect, and the mu POINTERs of the caL and kir2 mechanisms are linked to it in the file fig6.hoc. As noted previously, the simulations that generate Figure 10 use a DAsyn point process so that msg can have its own dynamics and yet be computationally efficient. fig10syn.hoc contains the statements that link the mu POINTERs of caL and kir2 to the DAsyn's msg.

The "spike" mechanism used by Gruber et al. 2003 for Fig. 10 requires an implementation that demands fig10ab.hoc be run only with fixed time step integration (see gspike.mod).

Reusable hoc code

c_cell.hoc contains a template (class definition) for model neurons with properties identical to the one that was used to generate Fig. 10, complete with both an ExpSyn (to receive "cortical input spikes") and a DAsyn (to receive "dopaminergic input spikes"). It can be used, with or without NEURON's GUI, to spawn new instances of this model neuron that are suitable for use in network simulations.

load_file("c_cell.hoc")
objref foo
foo = new List()
for i=0,99 foo.append(new C_Cell())
foo.count()  // returns 100
// now there are 100 of these cells ready to be connected into a network

  celsius = 20

How I created c_cell.hoc

1. Clicked on the "Hoc File" button of fig10ab.hoc's NetGUI tool and specified fig10network.hoc as the name of the output file
2. Extracted the template for the C_Cell class from fig10network.hoc
3. Finally, made one small but important modification to the C_Cell template:
   adding a "forsec all" loop to proc synapses() so that the mu POINTERs of every instance of caL and kir2 will be linked to the DAsyn's msg variable

Here's the modified proc synapses()--

proc synapses() {
  /* ExpSyn0 */   soma syn_ = new ExpSyn(0.5)  synlist.append(syn_)
    syn_.tau = 3
  /* DAsyn1 */   soma syn_ = new DAsyn(0.5)  synlist.append(syn_)
  // attach all POINTERs to DAsyn.msg
  forsec all {
    if (issection("caL")) {
      for (x,0) { // skip the nodes at 0 and 1
	setpointer mu_caL(x), synlist.object(1).msg
      }
    }
    if (issection("kir2")) {
      for (x,0) { // skip the nodes at 0 and 1
	setpointer mu_kir2(x), synlist.object(1).msg
      }
    }
  }
}