Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2007 Apr 19;54(2):219-35.
doi: 10.1016/j.neuron.2007.03.025.

neuroConstruct: A Tool for Modeling Networks of Neurons in 3D Space

Affiliations
Free PMC article

neuroConstruct: A Tool for Modeling Networks of Neurons in 3D Space

Padraig Gleeson et al. Neuron. .
Free PMC article

Abstract

Conductance-based neuronal network models can help us understand how synaptic and cellular mechanisms underlie brain function. However, these complex models are difficult to develop and are inaccessible to most neuroscientists. Moreover, even the most biologically realistic network models disregard many 3D anatomical features of the brain. Here, we describe a new software application, neuroConstruct, that facilitates the creation, visualization, and analysis of networks of multicompartmental neurons in 3D space. A graphical user interface allows model generation and modification without programming. Models within neuroConstruct are based on new simulator-independent NeuroML standards, allowing automatic generation of code for NEURON or GENESIS simulators. neuroConstruct was tested by reproducing published models and its simulator independence verified by comparing the same model on two simulators. We show how more anatomically realistic network models can be created and their properties compared with experimental measurements by extending a published 1D cerebellar granule cell layer model to 3D.

Figures

Figure 1
Figure 1
Overview of neuroConstruct (A) Schematic view of the main functionality of neuroConstruct. (B) Main neuroConstruct GUI showing a single abstract cell with a Na+ channel conductance density that varies on different parts of the cell membrane. (C) Main interface to neuroConstruct showing the visualization of a simple network using the transparency feature to highlight a single cell and its connections.
Figure 2
Figure 2
Detailed Cell Morphologies in neuroConstruct (A) A detailed reconstruction of a neocortical pyramidal cell (Mainen et al., 1995) imported into neuroConstruct from a NEURON morphology file. (B) Detail of a small part of a dendritic tree. Upper view: all 3D detail present in the original morphology file. Sections (between the blue spheres) contain a number of 3D points with associated diameter, each of which is the endpoint of a segment (small gray conical frusta). NEURON uses this information to compute membrane area and axial resistance, but only performs numerical integration at specific locations (red spheres; determined by nseg). Lower view: simpler representation of cell structure with fewer segments for mapping to GENESIS (Experimental Procedures).
Figure 3
Figure 3
Use of ChannelML for Specifying Cellular Mechanisms A ChannelML file (the code fragment shows the parameters needed to specify a double exponential time course synapse) can be converted into script files in the native language of various neuronal simulators (currently NEURON and GENESIS), using an XSL file for each mapping. HTML representations of the XML file provide a more readable view of the mechanism and associated metadata. Plots can be generated to view the mechanism's properties.
Figure 4
Figure 4
Test of the Simulator Independence of a neuroConstruct Model (A) Membrane potential responses to a 500 ms current pulse of 10 pA at a simulation time step of 0.01 ms for a granule cell model (Maex and De Schutter, 1998) implemented in neuroConstruct using ChannelML and run on NEURON and GENESIS. (B) Dependence of timing of last action potential on integration time step. (C) Dependence of the root-mean-square (rms) of the difference between traces on integration time step. The minimum at 0.001 ms is due to the peaks overlapping before converging at slightly different times in each simulator. The dotted line shows the rms error between the GENESIS model and one with a 1% difference in Na+ conductance density. (D) Values of some of the internal state variables as a function of time (ms) displayed as screenshots from NEURON (left, time units ms) and GENESIS (right, s).
Figure 5
Figure 5
Connectivity Schemes Used to Generate Network Connections between Cell Groups in neuroConstruct (A) Simplified morphology of a GrC (i) including soma and axon. Parallel fiber sections, highlighted in red, indicate presynaptic sections where synapses are permitted. Simplified morphology of a PC (ii) with red dendritic sections showing postsynaptic sections where synapses are permitted. Connections between multiple GrCs and a PC made using the morphology-based connection algorithm (iii). Green and red spheres show the sites of pre- and postsynaptic connection, respectively. (B) Simplified morphology of a cortical interneuron (i) including soma, dendrites, and a cylindrical volume (gray shading) defining boundaries of the axonal arbor. Simplified morphology of a cortical pyramidal cell (ii) with red dendritic sections showing postsynaptic sections where synapses are permitted. Three-dimensional connections between multiple interneurons and pyramidal cells made using the volume-based connection algorithm (iii). Sites of pre- and postsynaptic connections are linked by lines changing from green to red.
Figure 6
Figure 6
Implementation of Existing Network Models in neuroConstruct (A) Visualization of a 1D GrC layer network model from Maex and De Schutter (1998). MFs (bottom) are connected via excitatory synapses to GrCs (middle), which are in turn connected to GoCs (top). The GrCs receive inhibitory connections from GoCs. (B) Spike time histograms (bin size, 1 ms) as produced by the script files released with the original publication (left) and for the neuroConstruct model of the network (right). Spikes for the GrCs are in black and the GoCs are in red. (C) Replication of a network model of the dentate gyrus (Santhakumar et al., 2005). The model consists of (from the top down) 500 GrCs with two dendritic branches, 6 basket cells, 15 mossy cells, and 6 hilar cells. The 10,000+ synaptic connections have been removed for clarity. The network receives a brief perforant path focal stimulation, mainly on the central 100 GrCs. Cell coloring reflects network activity 110 ms after stimulation. (D) Raster plots of dentate gyrus GrC activity in the original published model and in the neuroConstruct implementation of the network.
Figure 7
Figure 7
Extension of a 1D Model of Granule Cell Layer to 3D (A) Visualization of a 3D cerebellar GrC model based on a published 1D model (Maex and De Schutter, 1998). MF terminals (blue), GrC somas (orange), and GoC somata (green) are packed in a 3D region (500μm in PF direction, 1 mm parasagittally, 50 μm in thickness) representing a section of the GrC layer of the cerebellar cortex. The ascending segments and parallel fibers of the GrCs extend into the molecular layer region, as do the single dendrites of the GoCs. (B) A single GoC and associated network connections highlighted using the transparency option in neuroConstruct. The range of connection lengths is larger than the experimentally reported values for the GoC dendritic tree (∼200 μm [Dieudonne, 1998]) due to the reduced cell density. (C) Histogram of the distribution of number of synaptic connections received by GrCs from MF terminals. Axis variables shown at bottom of window in (C)–(E). (D) Histogram showing the distribution of numbers of synaptic connections made to GrCs by the 96 MFs in the network. (E) Distribution of distances between connected MF terminals and GrC somata, corresponding to dendritic length.
Figure 8
Figure 8
Network Analysis of a 3D Granule Cell Layer Model (A) View of 3D cerebellar GrC layer model showing only the cell bodies. Two regions are identified, beam A and beam B, which have nonoverlapping sets of PFs. (B) Voltage traces of four GoCs at the end of a 4 s simulation run, with network connectivity as outlined previously and 50 Hz Poisson input to the MFs. Black trace (cell 31) and red trace (cell 5) are from GoCs in beam A. Cells 13 (blue) and 15 (green) are in beam B. Axis variables shown at bottom of window in (B)–(E). (C) Interspike interval histograms of the GrCs (black) and GoCs (red). The peak at approximately 40 ms reflects the observed average firing rate of the GoCs of 23.8 Hz, the single peak resulted from regular GoCs spiking. The GrCs have a lower average firing rate and do not fire on every GoC cycle, hence the multiple peaks in the histogram. (D) Crosscorrelation between cell 31 and the other four GoCs in beam A, each color graph representing a different cell. The y axis represents the probability of a spike occurring in the other cell with the specified offset (1 ms time window). (E) Crosscorrelation between cell 31 and the six GoCs in beam B, with identical axes to D.

Similar articles

See all similar articles

Cited by 90 articles

See all "Cited by" articles

References

    1. Alger B.E. Retrograde signaling in the regulation of synaptic transmission: focus on endocannabinoids. Prog. Neurobiol. 2002;68:247–286. - PubMed
    1. Ascoli G.A. Mobilizing the base of neuroscience data: the case of neuronal morphologies. Nat. Rev. Neurosci. 2006;7:318–324. - PubMed
    1. Ascoli G.A., Krichmar J.L., Nasuto S.J., Senft S.L. Generation, description and storage of dendritic morphology data. Philos. Trans. R. Soc. Lond. B Biol. Sci. 2001;356:1131–1145. - PMC - PubMed
    1. Attwell D., Iadecola C. The neural basis of functional brain imaging signals. Trends Neurosci. 2002;25:621–625. - PubMed
    1. Bednar J.A., Choe Y., De Paula J., Miikkulainen R., Provost J., Tversky T. Modeling cortical maps with Topographica. Neurocomputing. 2004;58-60:1129–1135.

Publication types

LinkOut - more resources

Feedback