Fninf 08 00068
Fninf 08 00068
Fninf 08 00068
Edited by: Investigation of cellular and network dynamics in the brain by means of modeling
Andrew P. Davison, CNRS, France and simulation has evolved into a highly interdisciplinary field, that uses sophisticated
Reviewed by: modeling and simulation approaches to understand distinct areas of brain function.
Upinder S. Bhalla, National Center
Depending on the underlying complexity, these models vary in their level of detail, in
for Biological Sciences, India
Arnd Roth, University College order to cope with the attached computational cost. Hence for large network simulations,
London, UK single neurons are typically reduced to time-dependent signal processors, dismissing the
Werner Van Geit, EPFL, Switzerland spatial aspect of each cell. For single cell or networks with relatively small numbers of
Robert Andrew McDougal, Yale
University, USA
neurons, general purpose simulators allow for space and time-dependent simulations of
electrical signal processing, based on the cable equation theory. An emerging field in
*Correspondence:
Gillian Queisser, Computational Computational Neuroscience encompasses a new level of detail by incorporating the full
Neuroscience, Goethe Center for three-dimensional morphology of cells and organelles into three-dimensional, space and
Scientific Computing, Computer time-dependent, simulations. While every approach has its advantages and limitations,
Science and Mathematics, Goethe
University, Frankfurt am Main,
such as computational cost, integrated and methods-spanning simulation approaches,
Germany depending on the network size could establish new ways to investigate the brain. In this
e-mail: gillian.queisser@ paper we present a hybrid simulation approach, that makes use of reduced 1D-models
gcsc.uni-frankfurt.de using e.g., the NEURON simulator—which couples to fully resolved models for simulating
cellular and sub-cellular dynamics, including the detailed three-dimensional morphology
of neurons and organelles. In order to couple 1D- and 3D-simulations, we present
a geometry-, membrane potential- and intracellular concentration mapping framework,
with which graph- based morphologies, e.g., in the swc- or hoc-format, are mapped
to full surface and volume representations of the neuron and computational data from
1D-simulations can be used as boundary conditions for full 3D simulations and vice
versa. Thus, established models and data, based on general purpose 1D-simulators, can
be directly coupled to the emerging field of fully resolved, highly detailed 3D-modeling
approaches. We present the developed general framework for 1D/3D hybrid modeling
and apply it to investigate electrically active neurons and their intracellular spatio-temporal
calcium dynamics.
Keywords: calcium dynamics, electrical stimulation, hybrid, parallel, detailed modeling, intracellular signaling,
neuron, PDEs
necessity for modeling intracellular processes in a highly detailed an intra- and extra-cellular tetrahedral volume grid. In order to
fashion in order to capture the underlying physical concepts of associate grid nodes from the triangular surface grid with nodes
cellular signaling. from the compartment model geometry we developed Vm 2uG, a
The need for coupling different aspects of signal processing tool that uses a nearest neighbor-algorithm and kd-search algo-
in neurons has been recognized in the past (Kerr et al., 2008; rithm to perform the mapping of membrane potential data onto
Wils and De Shutter, 2009; Andrews et al., 2010; Cannon et al., the triangulated surface of the 3D-neuron.
2010; Oliveira et al., 2010). In addition to projects focussing In this paper we chose a setup consisting of a multi-
on the coupling of existing simulators for parallel computing compartment model of a CA1 stratum radiatum interneuron
architectures (Djurfeldt et al., 2010), the influence of spatial taken from Katona et al. (2011) and carried out simulations
channel distribution on the electrical properties (Cannon et al., for electrical signal processing in NEURON. We then used the
2010) or integrating reduced intra-cellular approximations of presented methods to couple 1D and 3D models and simu-
reaction-diffusion processes, (Resasco et al., 2012; Anwar et al., lated calcium dynamics in the full three-dimensional intracellular
2013; McDougal et al., 2013a), we focus on the topic of how space under various parameter settings. Based on these examples,
the three-dimensional intracellular architecture of neurons influ- we introduce our methods and tools for 1D/3D-hybrid modeling
ences intra-cellular signals and how the resulting models can be and show how existing models and data can be incorporated into
efficiently solved on different computing scales. Thus, the key highly detailed, three-dimensional simulations.
factors are to incorporate an accurate morphology of the neu-
ron including the three-dimensional intra-cellular architecture 2. RESULTS
(which can include active and passive organelles), model a multi- The methods described here couple one-dimensional electrical
ion system described by systems of partial and ordinary differen- models and three-dimensional models for intra-cellular signaling.
tial equations (these can include the exchange mechanisms across We chose the CA1 interneuron from Katona et al. (2011) and its
plasma- and organelle membranes) and allow bi-directional cou- neuron morphology made available on NeuroMorpho.org (Ascoli,
pling between one-dimensional membrane potential and three- 2006) and on ModelDB (Migliore et al., 2003). Simulations of
dimensional intra-cellular signaling computations. The complex- the membrane potential dynamics in 1D, i.e., on a compart-
ity of three-dimensional, detailed simulations of multi-ion sys- ment model level, were performed with NEURON (Hines and
tems is efficiently handled by the simulation framework uG, Carnevale, 2003; Carnevale and Hines, 2006), using a standard
where Finite Element or Finite Volume discretizations and multi- set-up defined in the Materials and Methods, section. Since
grid methods are used to solve the underlying large linear equa- intra-cellular processes are strongly regulated by calcium, e.g.,
tion systems on, potentially, massively parallel systems, (Heppner (Milner et al., 1998; Bading, 1998; West et al., 2002; Clapham,
et al., 2013; Vogel et al., in press). Making use of uG, we propose 2007; Tai et al., 2008), we chose calcium dynamics regulated by
a method to couple classical 1D-simulators for the computation plasma membrane-located calcium channels with a given den-
of membrane potentials, with detailed 3D-simulations within an sity, thus modeling effectively a channel conductance density, and
on-line 1D/3D-hybrid simulation framework. a diffusion-reaction process in the neuronal cytosol as a repre-
In this paper we demonstrate this newly developed approach sentative of three-dimensional, intracellular signaling in neurons.
using the example of intra-cellular calcium dynamics coupled 3D simulations were carried out in uG, Bastian et al. (1997);
to the membrane potential via voltage-gated calcium channels. Vogel et al. (in press). Note, that this is a representative setup
While 1D simulations can be carried out with classical simula- which is applicable to any other 1D simulations and 3D intracel-
tors, such as NEURON or Genesis, the simulations in 3D are lular processes. The coupling of both models here occurs on the
based on systems of partial and ordinary differential equations level of calcium channels—these require the membrane poten-
(PDEs and ODEs) describing distinct intracellular processes. In tial in space and time on the plasma membrane, the local intra-
order to merge the two into a 1D/3D-hybrid system, two things and extra-cellular calcium concentrations, as well as the geometry
need to be done. In the first step one needs to generate from itself. In this section we will introduce the models and the simula-
a 1D morphology (a compartment model geometry) a surface tion set-up, methods for grid generation and membrane potential
and volume grid for numerical simulations in 3D and in the sec- mapping, and will show simulation results for the described
ond step ensure a bi-directional coupling where the membrane 1D/3D hybrid simulation approach.
potentials of the 1D simulation are mapped onto the surface
of the 3D problem as boundary conditions for the numerical 2.1. THE 3D CALCIUM MODEL
problem and the computed intra-cellular ion concentrations are For this study we consider a calcium model on the continuum
mapped back to compute updated membrane potentials. For this scale, including the following components:
we developed automated tools to compute the necessary mor-
phology representations of a neuron and to allow bi-directional 1. Morphology: The morphology and thus the computational
coupling. domain is defined by a standard compartment model, e.g., in
For the neuron surface meshing in 3D, we employ a triangu- the hoc-format (see Supplemental Figure S1 for an example).
lation algorithm that makes use of the coordinates and radii of This morphology is then mapped to an equivalent three-
anatomical reconstructions, stored in e.g., hoc- or swc-format, to dimensional computational domain.
compute an equivalent surface mesh of the reconstructed neu- 2. Membrane potential: The membrane potential is an input
ron. We used TetGen (Si, 2009) to generate, from the surface grid, parameter for the calcium channel models and is computed
by the 1D simulations and provided to the calcium channel The spatio-temporal dynamics of buffering molecules are
models as input data. described by a diffusion-process
3. Calcium channels on the plasma membrane: Based on the
models by Borg-Graham (Graham, 1999), we define N-/L- ∂Bmobile (x)
= DBmobile · Bmobile (x) − B (7)
or T-type calcium channels (see Materials and Methods). ∂t
Channel densities can be space-dependent, thus inhomoge-
neous channel distribution is possible. For stationary buffers the diffusion coefficient DBmobile can be
4. Cytosolic calcium dynamics: In this study we consider diffu- set to zero. The parameter values of the model equations used
sion of calcium and reaction of calcium with buffers in the throughout the paper are listed in Table 1.
cytosol. The computed calcium concentrations are mapped to Note, that our framework is not limited to the above exam-
the 1D model to compute calcium-dependent currents. ple of calcium/buffer dynamics. The employed multi-physics
platform uG has been successfully used in a wide variety of appli-
We can formulate the above points mathematically as an initial- cations, ranging from, but not limited to, simulations of skin
value boundary problem for a diffusion-reaction model. To this permeability in pharmaceutical applications (Hansen et al., 2008;
end, let us denote the neuron geometry as , which is a compact Nägel et al., 2008, 2009; Muha et al., 2011) to groundwater flow
subset of R3 , such that studies (Grillo et al., 2010) and biogas reactor modeling (Muha
et al., 2012). The latter demonstrates the use of the framework for
• ⊂ R3 with plasma membrane boundary = ∂, large chemical reaction systems. The 3D intra-cellular model pre-
¯ := ∪ ∂ and
• sented here can thus be extended to include multiple ion species
• ∩ ∂ = ∅ that are non-linearly coupled, resulting in a non-linear system of
PDEs.
which defines the space-time cylinder
2.2. COMPONENTS OF THE HYBRID FRAMEWORK
Z := ×]0, T[⊂ R3 × R+ , ∀(x, t) ∈ Z : T > t > 0. (1) As mentioned earlier, our hybrid 1D/3D framework consists of
geometry matching and the coupling of 1D membrane poten-
is the problem-associated domain for the 3D model, recon- tial simulations and 3D intra-cellular simulations. Computational
structed from the original compartmental model geometry grids for numerical simulations in 3D rely on a geometry-defining
defined, in our case, by a hoc file. Furthermore let c(x, t) be surface grid and a discrete representation of the computational
the calcium concentration function. Then the diffusion-reaction domain, i.e., a volume grid. Approximation of unstructured
problem can be stated as domains is ideally done by a triangular surface and tetrahe-
dral volume grid. These grids need to be generated using the
∂c(x, t)
= D · c(x, t) + R(c) in (2)
∂t
Table 1 | Parameters used for the simulations using the 1D/3D hybrid
c(x, 0) = c0 (x), in (3)
framework.
∂c(x, t)
= (Vm (x, t), x, t), on ∂, (4) NEURON uG
∂ n
dt [ms] 0.1 0.1
where D denotes the diffusion coefficient for cytosolic calcium,
stim.dur [ms] 10 –
:= 3i=1 ∂ 2 /∂xi2 is the Laplace-Operator and R(c) a reaction
stim.amp [nA] 0.1 –
term that depends on the calcium concentration c. Note that,
timesteps [#] 10000 10000
the vector n denotes the direction perpendicular to the bound-
Boundary Condition – Vm from NEURON
ary surface. Equation (3) is the initial condition, i.e., a calcium
Calcium diffusion coefficient [µm2 · s−1 ] 20–100 20–100
distribution for t = 0 in the cytosol defined by the function c0 (x).
VGCC density [µm−2 ] 200–1000 200–1000
Function defines a Neumann flux boundary condition, regu-
Buffer kon [µM−1 · ms−1 ] - 0.09
lating the calcium flux through N-/L- or T-type calcium chan-
Buffer koff [ms−1 ] – 0.24
nels, and thus also depends on the space- and time-dependent
Buffer diffusion coefficient [µm2 · ms−1 ] – 0.043
membrane potential Vm (x, t). in this study is specified by Borg-
Total buffer concentration [µM] – 20
Graham model type calcium channels, which are listed in the
Extracellular calcium concentration [mM] 1.6 1.6
Materials and Methods section.
Intracellular calcium concentration [µM] 0.1 0.1
Cytosolic interaction of calcium with mobile and stationary
Temperature [K] 300 300
buffers is governed by reaction equations of the type:
The calcium diffusion coefficient was varied across the ranges documented in
B := kon,Bmobile [Ca2+ ] · [Bmobile ] − koff ,Bmobile · [CaBmobile ] (5) Allbritton et al. (1992) and the VGCC density was varied according to Pumplin
et al. (1981); Eggermann et al. (2011). Values for the buffer kinetics and diffusivity
as well as the conservation law for the buffer concentration: were taken from Nagerl et al. (2000) as well as extracellular calcium concentra-
tion from Egelman and Montague (1999) repsectively the intra-cellular calcium
[CaBmobile ] = Bmobile,total − [Bmobile ] (6) concentration from Helmchen et al. (1996); Maravall et al. (2000).
compartment geometry information, provided by the 1D model In Figure 2 we show the in hoc-style defined neuron over-
as, e.g., a hoc- or swc-file. To provide an automated way for gener- layed by the corresponding three-dimensional neuron, which can
ating surface and volume grids from anatomically reconstructed be used for the full-spatial simulation of intra-cellular processes.
geometries, we developed novel tools and combined them with The quality of the volume grid influences the numerical accuracy
existing ones, which and stability. In particular the tetrahedral angles are a means of
determining the grid quality (Deuflhard and Weiser, 2011). For
1. Integrate NEURON to perform 1D simulations of membrane the grid used in this paper we measured minimal and maximal
activity in neurons. aspect ratios of the the triangular surface grid ARtri and the tetra-
2. Generate triangular surface grids from hoc-style neuron mor- hedral volume grid ARtet respectively, yielding values between
phology files. 0.14 ≤ ARtri ≤ 0.86 and 0.21 ≤ ARtet ≤ 0.8. As a guideline one
3. Map 1D simulation data of membrane potentials from can state that aspect ratios above 0.1 result in grid elements that
compartment geometries onto the two-dimensional neuron do not affect numerical stability (Shewchuk, 2002; Deuflhard and
membrane (Vm 2uG) which are used as boundary condi- Weiser, 2011; Thompson et al., 2012).
tions for membrane potential-dependent channels in the
3D model. 2.3. MEMBRANE POTENTIAL MAPPING—VM 2UG
4. Integrate ion concentrations on the equivalent surface grid to In 1D compartment models the membrane potential Vm is com-
feedback into the NEURON simulation. puted in one node per cylindrical compartment. In the 3D setting
each original cylinder is now represented by a segment of the
triangular surface grid, thus the membrane potential from one
The general workflow of the 1D/3D coupling is depicted in cylinder node needs to be mapped onto a calculated cluster of
Figure 1. Using the graph-structure data contained in neuron nodes in the triangular surface grid.
compartment geometry files, we generate an equivalent 3D geom- Associating each grid point y := (y1 , y2 , y3 ) of the 3D mor-
etry with the compartment model graph as its “backbone.” This phology with compartment model grid points x := (x1 , x2 , x3 )
procedure is fully automated and quality controlled and will be (Figure 1) and the corresponding membrane potentials Vm over
discussed in a forth-coming paper. Since our presented frame- the time-course of a simulation thus requires a mapping
work is not limited to the grid generator used here, we would
like to acknowledge the efforts of other authors addressing sur-
Vm 2uG : R3 × R+ × R → R3 × R+ × R (8)
face grid generation from point-diameter information, (e.g.,
McDougal et al., 2013b). (x, t, Vmx )
→ (y, t, Vmy )
FIGURE 1 | Workflow of the 1D/3D hybrid framework. The workflow can components of the model problem. In the simulation phase 1D membrane
be separated into 1D and 3D modeling and simulation and into a set-up and potential simulation data is computed and mapped to the 3D framework.
simulation phase. In the set-up phase a general purpose simulator is used to There, the membrane potential data is included in simulating intracellular
define the one-dimensional problem (e.g., a hoc-geometry and set-up file). processes, such as calcium signaling. The computed calcium data is then
The 3D setup consists of generating a three-dimensional representation of mapped back to the 1D problem, where it is used to compute
the defined neuron and of specifying the cellular and intra-cellular calcium-dependent membrane fluxes.
FIGURE 2 | Overlay of the multi-compartmental model defined by unphysiological intersections of dendrites at the soma in 3D, due to the
NEURON (red) and the equivalent tetrahedral volume mesh by uG fact that all dendrites in the multi-compartment description share a
(blue). (A) View of the whole neuron from Katona et al. (2011). Note, common soma node. (B) Magnified views of ROIs color-coded by
that close to the soma local optimization of the morphology leads to rectangular boxes within the overview for comparison, showing the
slight divergence between NEURON and uG morphology. Optimization is compartment model geometry defining the backbone of the generated
performed when the multi-compartment geometry model would cause surface grid.
that assigns a membrane potential Vmy to every grid point y. Vmy mapping process is dimension-independent and can be done for
is set to the potential Vmx associated with the nearest neighbor arbitrary dimensions by means of the underlying C++ library for
x := (x1 , x2 , x3 ) of grid point y with respect to all points con- multi-dimensional (binary) search trees (Mount and Arya, 2012),
tained in the hoc-file in every timestep t of the simulation. The a library that is used in Vm 2uG.
distance measure for calculating the nearest neighbor can be any
Minkowski metric and is interchangeable. In this study we used 2.4. ALGORITHMS
Algorithms 1 and 2 and chose the euclidean metric, or L2 -norm, The pairwise comparison is an exhaustive search that requires no
such that additional data structure but O(n · m) iterations which is imprac-
tical if the number n of provided points on the grid is large, and
d − 1 1/2 for the number of corresponding points in the hoc file holds n ≈
dist(p, q) := (pi − qi ) 2
(9) m, thus arriving at quadratic runtime complexity O(n2 ). For large
i=0 n, we therefore use the method of space partitioning (k-d trees,
Figures 3B,C) which leads to an improved average runtime com-
Because the geometry information provided by the hoc-file may plexity for a query of log nd , with d being the space dimension.
be very large, i.e., the number of provided points n is large, we With this approach we reach super-linear runtimes after the ini-
implemented and tested two strategies for determining the near- tial tree has been built. In contrast to the O(n2 ) method, we need
est neighbor. The first is an exact solution of the problem by space an additional structure, which is justified because of the overall
partitioning with multi-dimensional binary search trees, k-d trees speedup compared to the naïve approach. The overall runtime is
with a search complexity O(n · log(n)) (see Bentley, 1975 as well dominated by the utilized search algorithm used during the ini-
as Figure 3 for an example), the second strategy is the exact solu- tial build of the tree and the balancing of the tree itself, yielding a
tion by pairwise comparison with a search complexity O(n2 ). complexity O(n · log(n)) (Wald and Hvran, 2006; Cormen et al.,
Note that we consider a three-dimensional example here, but the 2011). The worst case arises if n 2d is satisfied and the runtime
q ← query point
2: NN ← getNearestNeighbor(q)
iDist ← NN.getDistance()
4: if dist ≤ cutoff then
NNs ← getNearestNeighbors(k)
6: for i = 1 → k, j = 1 → k,
l = 1 → k ∧ i = j = k = l do
det ← calculateDeterminant calculate determinant
8: p ← constructPlane(NN[i], NN[j],
NN[k], NN[l]) create plane
q ← projectOnPlane(q, p)
10: if getDistance(q , q) ≤ iDist then orthogonal project onto plane
iDist ← getDistance(q , q)
12: Vm ← p.interpBilinear(q ) bilinear interpolation
end if
14: end for
end if
FIGURE 3 | Basic operations. (A) General description of the nearest geometry. Interpolation is then carried out over n ∈ N+ pseudo
neighbor search task in 2D/3D: Let S be a set of #n points with dimension nearest neighbors to compute an optimal approximated value for
d := 2, 3. Let q be a query point. The task is to find a point r (green) in the
set S of all points (black), which is closest to the query point q [center of
the membrane potential which is then assigned to the given grid
sphere in (A) on the right] within a given bounding box, i.e., within a circle point.
(black) or sphere. (A) Pairwise comparison of all points in a radius search. B:
Space-partitioning approach using k-d trees (multi-dimensional binary 2.5. WORKFLOW AND BIDIRECTIONAL COUPLING
search trees). (C) An exemplary multi-dimensional binary search tree of
dimension d = 2 (enhanced and modified according to Wikipedia, 2012),
We use a direct data coupling mechanism termed a type A problem
representing the space partition in (B). in Heterogenous Multiscale Modeling, see (Weinan et al., 2003),
utilizing an online algorithm within uG. The NEURON library is
used as a plugin/shared library within the uG project. The sim-
Algorithm 1 | Linear interpolation
ulation workflow is defined in uG by a lua-script (Ierusalimschy
et al., 2013) (see Figure 4 for the script used in this paper). We
q ← query point provide fully flexible control over NEURON within uG by means
2: NN ← getNearestNeighbor(q) of a custom API or by directly including the hoc-script (which is
iDist ← NN.getDistance() then executed by the HOC language interpreter). For the latter we
4: if dist ≤ cutoff then developed an additional C++ wrapper.
NNs ← getNearestNeighbors(k) The workflow defined in the lua-script performs the steps
6: for i = 1 → k, j = 1 → k ∧ i = j do
illustrated in Figures 1, 4. The numerical procedures are defined
g ← constructLine(NNs[i], NNs[j]) create straight line
12: end if
points, which is guaranteed by our framework.
end for
Data exchange between 1D and 3D model is bidirectional,
14: end if where membrane potential data is passed from 1D to 3D via the
mapping algorithm presented in the previous section. The com-
puted ion concentrations in the 3D model are then passed from
1 3D to 1D in the following way:
complexity increases per query to O d · n(1− d ) (Lee, 1997).
Consider a cylindrical compartment Ci of the 1D model. By
Yet, the curse of dimensionality is not an issue in the presented ∂(Ci ∩ ∂) we denote the portion of the cell surface ∂ that is
application, since our geometry is defined in three-dimensional associated with compartment Ci . We can then compute the intra-
space. cellular calcium concentration for Ci as
Vm 2uG offers linear or bilinear membrane potential interpo-
lation (Algorithms 1 and 2). This can be useful if the compart-
∂Ci
mental representation of the neuron is very coarse and map- [Ca2+ ]Ci = uCa2+ (x)
∂(Ci ∩ ∂) ∂(Ci ∩∂)
ping distances become large, i.e., 3D surface grid points lie far
away from the computed nearest neighbor on the compartment for x ∈ ∂(Ci ∩ ∂) and ∀i = 0, . . . , N (10)
FIGURE 4 | Script file illustrating the setup of the NEURON and uG the uG-setup. Note that one can specify the geometry and stimulation
model, here divided into a NEURON setup phase, a uG setup phase, protocol in distinct files and combine this with using the hoc interpreter from
chemical model specifications and the simulation run control. One can uG to refine the setup by statements of the form:
make use of the hoc interpreter features within the lua script files that define HocSetup:execute(“command string”).
where uCa2+ (x) is the calcium concentration in node x of the the values computed by Equation (10) shows good agreement
3D grid and N is the number of compartments representing between the two models, see Supplemental Figure S2 as well as
the 1D geometry. . symbolizes the size of ∂Ci and ∂(Ci ∩ ∂) Supplemental Figure S3.
respectively.
∂Ci
The factor ∂(C i ∩∂)
accounts for the fact, that the surface size 2.6. SIMULATIONS USING THE 1D/3D HYBRID METHOD
of ∂(Ci ∩ ∂) and the surface of Ci are not necessarily identical. We applied the described method for coupling 1D and 3D sim-
Computing the calcium fluxes in uG and in NEURON using ulations to a CA1 interneuron from Katona et al. (2011), taken
FIGURE 6 | Time series for a simulation of intra-cellular calcium signaling calcium-dependent membrane potential in NEURON. (A–F) covers 1s of real
with a calcium diffusion coefficient set to 100 µm2 /s, VGCC density of time. Time points shown are at 0, 0.01, 0.02, 0.03, 0.1, and 1 s. Note, the
1000 µm−2 and no obstacle present in the dendrite. The basal calcium maximum single-channel conductance was set to 60 pS according to Graham
concentration is set to 100 nM. (A–F) show the propagation of calcium in a (1999). (G) Highlighted measuring points close to the plasma membrane (pm)
dendrite, triggered by the the stimulation protocol listed in Table 1. in between and at the center used in (H). (H) Evolvement of the calcium
Membrane potentials were computed in NEURON and mapped to the plasma concentrations at three different points over a period of 20 ms. Note that the
membrane in order to compute the VGCC-dynamics regulating the calcium VGCCs close at the peak amplitude and the calcium profile quickly adjust to a
exchange. Calcium concentrations were then mapped back to compute the uniform value based on the intra-cellular diffusion.
Cobstacle
H := ∈ [0, 1] (11)
Ccylinder
Table 3 | Hindrance factors. level or in a networks with relatively small numbers of neu-
Hindrance [%] Obstacle volume [µm3 ] Obstacle surface area [µm2 ]
rons (Xylouris et al., 2007; Wittmann et al., 2009; Xylouris,
2013).
0 0.0000 0.0000 Given the limitations in computing resources, it is currently
25 0.0022 0.3040 not feasible to model and simulate large networks of neurons
50 0.0344 1.9000 with full single cell, three-dimensional detail. To make use of
75 0.0945 3.7240 the advantages of highly detailed three-dimensional models and
100 0.4076 8.6401
to cope with the complexity that comes with modeling large
Listed above are the surface and volume sizes of the obstacles that were placed networks of neurons, we developed a method to couple state
in the dendrite for the simulations shown in Figure 7 with corresponding hin- of the art general purpose neuron simulators, e.g., NEURON,
drance factors. A hindrance factor of 100 % represents a full blockage of the with three-dimensional models of single neurons. This 1D/3D
dendritic cross-section. hybrid method includes automated tools to either reconstruct
three-dimensional neuron morphologies from raw microscopy
3. DISCUSSION data (Broser et al., 2004; Queisser et al., 2008; Jungblut et al.,
Modeling and simulating cellular and network processes are 2011), from anatomically recorded data (Wolf et al., 2013) or
constrained by the complexity of the computational problem from graph-structure morphologies as used, e.g., in the NEURON
and the availability of experimental data to validate the mod- Simulator in the form of hoc-files or swc-files.
els. Experimental techniques are evolving, shedding new light on In this paper we introduced a framework for geometry and
the filigreed functioning of neurons and networks. Spatial and membrane potential and intra-cellular ion concentration map-
time resolutions are becoming fine enough to resolve positions ping between 1D simulations and the equivalent 3D model. For
of single proteins and receptors at synapses and the intra-cellular this we used the NEURON simulator to compute electrical sig-
domain can be investigated at increasing levels of detail. On the nals in a compartment neuron and uG to simulate intracellular
other hand computational complexity increases when a model calcium dynamics in 3D. With this approach we were able to
includes biological and spatial detail, resulting in a highly detailed exploit the modeling and computational advantages that a gen-
three-dimensional model of neurons. Computational complexity eral purpose simulator for large networks brings, as well as the
further increases when not only one neuron, but entire networks necessary tools to investigate intra-cellular (or even extra-cellular)
need to be modeled. processes on a very fine scale. By including the detailed mor-
To cope with network complexity, several approximation phology of neurons—which can be subject to temporal adaption
methods for cellular and network function have been devel- due to neuronal activity (Silver et al., 1990; Korkotian and Segal,
oped over the last decades. These approximations yield zero- 1999; Muller et al., 2002; Van Aelst and Cline, 2004; Hayashi and
or one-dimensional models in space, describing neurons with Majewska, 2005; Tada and Sheng, 2006; Wittmann et al., 2009;
point- or compartment neuron models. Numerous specialized Kanamori, 2013)—the interplay between cellular/organelle mor-
and general purpose neuron simulators incorporate these basic phology and cellular function can be systematically investigated.
abstraction techniques (e.g., Bower and Beeman, 1997; Hines Using defined stimulation protocols in NEURON, we ran 1D-
and Carnevale, 2003; Gewaltig and Diesmann, 2007) and allow simulations mapping the results onto the three-dimensional mor-
the user to simulate large and complex neural network struc- phology as boundary conditions for a cellular calcium model
tures. These models and simulators were developed and evolved and vice versa. The calcium model consisted of a local distribu-
not only around the limitations of computational power, but tion of N-type voltage-gated calcium channels, regulated by the
also around the availability of experimental data to validate the membrane potential and intracellular diffusion and reaction of
models. The evolution of experimental methods, computational calcium ions with a mobile buffer. After presenting the function-
resources and numerical mathematics motivates the develop- ality of the 1D/3D hybrid method on a reference parameter set,
ment of neuron models that include greater physical detail. we varied the density of voltage-gated calcium channels, the cal-
This brings us to three-dimensional models that include the cium diffusion coefficient and introduced intracellular obstacles.
detailed morphology of neurons, either by modeling physical The results show that for one, the presented method is designed in
processes using stochastic (e.g., Andrews et al., 2010; Oliveira a general fashion and is thus applicable to a broad range of neuro-
et al., 2010) or deterministic models. We make use of the lat- biological questions and that the effect of intra-cellular obstacles,
ter approach with the additional feature of being able to inte- locally changing channel densities and cytosolic diffusivity have a
grate obstacles and intracellular organelles, which yields models substantial effect on intracellular signals.
formulated on the continuum scale by means of systems of par- For future research, problem-specific models can be used
tial differential equations. Solving these equations numerically within the presented framework. For instance, not only single
requires advanced mathematical tools. One general purpose plat- neurons, but entire networks could be simulated on the 1D level,
form is uG (Bastian et al. 1997), which we used in this paper. where a small subset of “key neurons” in the network can be
uG offers numerical discretization methods and efficient solvers resolved in full three-dimensional detail (Xylouris et al., 2007).
for systems of partial differential equations on highly unstruc- Furthermore intra-cellular models for different ionic species,
tured grids and runs on massively parallel systems (Heppner detailed models of intra-cellular organelles, such as mitochondria
et al., 2013; Vogel et al., in press). As shown in the past, and endoplasmic reticulum calcium exchangers, protein synthesis
these numerical tools are applicable and efficient on a cellular and synapses could be included on the 3D scale.
where τx,0 = 0 for x = n, m, h. The values for αx and βx are F(V, [Ca]i , [Ca]o )
derived from (Hodgkin and Huxley, 1952):
z2 F 2 V [Ca2+ ]i − [Ca2+ ]o exp( − z FV
RT )
= pCa (27)
0.01(V + 10) RT 1 − exp( − z FV
RT )
αn = V − 10
(18)
e 10 −1
V where pCa is the permeability and for N-type calcium channels is
βn =
e 80
(19) set to the value 10−8 cm3 /s. F denotes the Faraday constant, R the
8 gas constant and T the temperature. For the VGCCs used in the
0.1(V + 25) simulations, the gating function G is defined as
αm = V + 25 (20)
e 10 − 1
V G(V, t) = k(V, t)l2 (V, t) (28)
βm = 4e 18 (21)
V
αh = 0.07e 20 (22) Here, k and l fulfill the ordinary differential equation (15) and eqs.
(16, 17), using the following values:
1
βh = V + 30
(23)
e 10 +1
zx γx (V − V1/2,x )F
αx (V) = Kx exp
Solving a multi-compartment model where for each cylindrical RT
compartment an equation of the type
−zx (1 − γx )(V − V1/2,x )F
βx (V) = Kx exp (29)
dVm RT
= −im + Ielectrode (24)
dt
The parameters V1/2 , z, γ , K, τ0 also depend on the particular
is computed using the NEURON simulator. Numerically, the aris- channel and are documented in Graham (1999). We set the
ing sets of ordinary differential equations are solved with the values accordingly:
V1/2,k := − 21 mV a preconditioner along with BiCGstab as the base solver for the
linear part (inner loop) and a Newton solver for the non-linear
zk :=2
part (outer loop), see e.g., (Hackbusch, 1985, 2003). The numeri-
γk :=0 cal setup and simulations were carried out in uG [see Bastian et al.
(1997); Vogel et al. (in press)].
Kk :=1.7 ms
τk,0 :=1.7 ms ACKNOWLEDGMENTS
V1/2,l := − 40 mV Numerous gratitude is given to Andreas Vogel pointing out
important properties of the solvers and their implementation
zl :=1 in uG, as well as Anton Vadimovich Chizhov of the Physical-
γl :=0 technical Institute Ioffe (Russian Academy of Sciences) in St.
Petersburg, Russia for providing data for validation of the Borg
Kl :=70 ms Graham Channel model (Graham, 1999) used in conjunction
τl,0 :=70 ms with our hybrid method.
It is sufficient to solve Equation (15) using the first order explicit SUPPLEMENTARY MATERIAL
Euler method, in which the required membrane potential is taken The Supplementary Material for this article can be found online
from the membrane potential mapping (Vm 2uG) in each time at: http://www.frontiersin.org/journal/10.3389/fninf.2014.
step. 00068/abstract
4.3. NUMERICAL APPROACH Supplemental Movie 1 | Simulation of 200 ms real time, perspective 1.
The calcium model used in the presented simulations is based
on a cytosolic diffusion equation with boundary conditions that Supplemental Movie 2 | Simulation of 200 ms real time, perspective 2.
include the plasma membrane calcium exchange mechanisms,
i.e., Neumann boundary conditions representing an ionic flux
REFERENCES
over a given surface area of the plasma membrane. This can be Allbritton, N., Meyer, T., and Stryer, L. (1992). Range of messenger action of
stated as: calcium ion and inositol 1,4,5-triphosphate. Science 258, 1812–1815. doi:
10.1126/science.1465619
∂
∇ · (D∇u(x, t)) = u(x, t) in ⊂ R3 (30) Andrews, S. S., Addy, N. J., Brent, R., and Arkin, A. P. (2010). Detailed simula-
∂t tions of cell biology with Smoldyn 2.1. PLoS Comput. Biol. 6:e1000705. doi:
10.1371/journal.pcbi.1000705
with boundary conditions Anwar, H., Hepburn, I., Nedelescu, H., Chen, W., and De Schutter, E. (2013).
Stochastic calcium mechanisms cause dendritic calcium spike variability. J.
∂u Neurosci. 33, 15848–15867. doi: 10.1523/JNEUROSCI.1722-13.2013
= g(Channels) (31)
∂n Arellano, J. I., Benavides-Piccione, R., DeFelipe, J., and Yuste, R. (2007).
Ultrastructure of dendritic spines: correlation between synaptic and spine
where u denotes the intracellular calcium concentration, D the morphologies. Front. Neurosci. 1, 131–143. doi: 10.3389/neuro.01.1.1.010.2007
diffusion tensor (a 3 × 3-Matrix) and g(Channels) the calcium Ascoli, G. A. (2006). Mobilizing the base of neuroscience data: the case of neuronal
morphologies. Nat. Rev. Neurosci. 7, 318–324. doi: 10.1038/nrn1885
flux depending on the channels defined in the simulation (see Bading, H. (1998). Transcription-dependent neuronal plasticity: the nuclear cal-
previous section). To solve Equation 30, we discretized in space cium hypothesis. Eur. J. Biochem. 267, 5280–5283.
by means of first order finite volumes, treating the time derivative Balls, G. T., and Baden, S. B. (2004). “A large scale monte carlo sim-
by an implicit Euler method. ulator for cellular microphysiology,” in Proceedings of the 18th
International Parallel and Distributed Processing Symposium (San
For this we discretized the plasma membrane geometry with a
Diego, CA: California University). doi: 10.1046/j.1432-1327.2000.
triangular surface grid and the volume of the cell with a tetrahe- 01565.x
dral volume grid. In the following we denote the discrete volume Bastian, P., Birken, K., Johannsen, K., Lang, S., Neuss, N., Rentz-Reichert, H., et al.
of the neuron as and each arising control volume box as Bi , i =
(1997). UG a flexible software toolbox for solving partial differential equations.
1, . . . , n. Therefore = n1 Bi . The Finite Volume method (cf. Comp. Vis. Sci. 1, 27–40. doi: 10.1007/s007910050003
Bentley, J. L. (1975). Multidimensional binary search trees used for associative
Eymard et al., 2000) defines an ansatz space Vh ⊂ H 1 () which
searching. Commun. ACM 18, 509–517. doi: 10.1145/361002.361007
is spanned by piecewise linear ansatz functions i (xi ) = δij and Blackwell, K. T., Wallace, L. J., Kim, B., Oliveira, R. F., and Koh, W. (2013). Modeling
is used as the approximation for the solution u. The approxima- spatial aspects of intracellular dopamine signaling. Methods Mol. Biol. 964,
tion of u is then defined as the linear combination of these ansatz 61–75. doi: 10.1007/978-1-62703-251-3_5
functions, resulting in a system of linear equations Bower, J., and Beeman, D. (1997). The Book of GENESIS: Exploring Realistic Neural
Models with the GEneral NEural SImulation System. New York, NY: Springer.
Brandi, M., Brocke, E., Talukdar, H. A., Hanke, M., Bhalla, U. S., Kotaleski,
n
J. H., et al. (2011). Connecting MOOSE and NeuroRD through MUSIC:
Aij uj = bi (32)
towards a communication framework for multi-scale modeling. BMC Neurosci.
j 12(Suppl. 1):P77. doi: 10.1186/1471-2202-12-s1-P77
Broser, P. J., Schulte, R., Roth, A., Helmchen, F., Waters, J., Lang, S., et al. (2004).
where A is the stiffness matrix. To solve the linear system of equa- Nonlinear aniostropic diffusion filtering of three-dimensional image data from
tions, we applied a geometric multigrid solver (smoother ILU) as 2-photon microscopy. J. Biomed. Opt. 9, 1253–1264. doi: 10.1117/1.1806832
Burette, A. C., Lesperance, T., Crum, J., Martone, M., Volkmann, N., Ellisman, Katona, G., Kaszas, A., Turi, G. F., Hajos, N., Tamas, G., Vizi, E. S., et al.
M. H., et al. (2012). Electron tomographic analysis of synaptic ultrastructure. (2011). Roller coaster scanning reveals spontaneous triggering of dendritic
J. Comp. Neurol. 520, 2697–2711. doi: 10.1002/cne.23067 spikes in CA1 interneurons. Proc. Natl. Acad. Sci. U.S.A. 108, 2148–2153. doi:
Cannon, R. C., O Donnel, C., and Nolan, M. F. (2010). Stochastic ion channel gat- 10.1073/pnas.1009270108
ing in dendritic neurons: morphology dependence and probabilistic synaptic Kerr, R. A., Bartol, T. M., Kaminsky, B., Dittrich, M., Chang, J. C., Baden, S. B., et al.
activation of dendritic spikes. PLoS Comp. Biol. 6:e1000886. doi: 10.1371/jour- (2008). Fast monte carlo simulation methods for biological reaction-diffusion
nal.pcbi.1000886 system in solution and on surfaces. SIAM J. Sci. Comput. 30, 3127–3249. doi:
Carnevale, N. T., and Hines, M. L. (2006). The NEURON Book. Cambridge UK, 10.1137/070692017
Cambrdige University Press. doi: 10.1017/CBO9780511541612 Kits, K. S., Dreijer, A. M. C., Lodder, J. C., Borgdorff, A., and Wadman, W. J.
Chen, X., Winters, C., Azzam, R., Li, X., Galbraith, J. A., Leapman, R. D., et al. (1997). High intracellular calcium levels during and after electrical discharges in
(2008). Organization of the core structure of the postsynaptic density. Proc. molluscan peptidergic neurons. Neuroscience 79, 275–284. doi: 10.1016/S0306-
Natl. Acad. Sci. U.S.A. 133, 4453–4458. doi: 10.1073/pnas.0800897105 4522(96)00651-3
Clapham, D. E. (2007). Calcium signaling. Cell 131, 1047–1058. doi: Korkotian, E., and Segal, M. (1999). Release of calcium from stores alters the mor-
10.1016/j.cell.2007.11.028 phology of dendritic spines in cultured hippocampal neurons. Proc. Natl. Acad.
Cormen, T. H., Leiserson, C. E., and Rivest, R. L. (2011). Introduction to Algorithms. Sci. U.S.A. 96, 12213–12215. doi: 10.1073/pnas.96.21.12068
Cambridge, MA: MIT Press; Massachusetts Institute of Technology. Lee, D. T. (1997). Worst-case analysis for region and partial region searches in
Deuflhard, P., and Weiser, M. (2011). Numerische Mathematik 3. Berlin: De Gruyter multidimensional binary search trees and balanced quad trees. Acta Inf. 9,
Lehrbuch 2011. 23–29.
Djurfeldt, M., Hjorth, J., Eppler, J. M., Dudani, N., Helias, M., Potjans, T. C., et al. Maravall, M., Mainen, Z. F., Sabatini, B. L., and Svoboda, K. (2000).
(2010). Run-time interoperability between neural network simulators based on Estimating intracellular calcium concentrations and buffering without wave-
the MUSIC framework. Neurinform 8, 43–60. doi: 10.1007/s12021-010-9064-z length ratioing. Biophys. J. 78, 2655–2667. doi: 10.1016/S0006-3495(00)
Egelman, D. M., and Montague, P. R. (1999). Calcium dynamics in the EC space 76809-3
of mammalian neural tissue. Biophys. J. 76, 1856–1867. doi: 10.1016/S0006- McDougal, R. A., Hines, M. L., and Lytton, W. W. (2013a). Reaction-
3495(99)77345-5 diffusion in the NEURON simulator. Front. Neuroinform. 7:28. doi:
Eggermann, E., Bucurenciu, I., Goswami, S. P., and Jonas, P. (2011). Nanodomain 10.3389/fninf.2013.00028
coupling between Calcium channels and sensors of exocytosis at fast mam- McDougal, R. A., Hines, M. L., and Lytton, W. W. (2013b). Water-tight mem-
malian synapses. Nat. Rev. Neurosci. 13, 7–21. doi: 10.1038/nrn3125 branes from neuronal morphology files. J. Neurosci. Methods 220, 167–178. doi:
Eymard, R., Gallouet, T. R., and Herin, R. (2000). “Parabolic equations, Chapter 4,” 10.1016/j.jneumeth.2013.09.011
in The Finite Volume Method in Handbook Of Numerical Analysis, Vol. 7. eds P. Migliore, M., Morse, T. M., Davison, A. P., Marenco, L., Shepherd, G. M.,
G. Ciarlet and J. L. Lions (Elsevier), 713–1020. and Hines, M. L. (2003). ModelDB: making models publicly accessible
Gewaltig, M. O., and Diesmann, M. (2007). Nest (neural simulation tool). to support computational neuroscience. Neuroinformatics 1, 135–139. doi:
Scholarpedia 2, 1430. 10.1385/NI:1:1:135
Graham, L. (1999). Interpretations of Data and Mechanisms for Hippocampal Milner, B., L.R., S., and E.R., K. (1998). Cognitive neuroscience and the study of
Pyramidal Cell Models. New York, NY: Plenum Publishing Corporation. memory. Neuron 20, 445–468. doi: 10.1016/S0896-6273(00)80987-3
Grillo, A., Logashenko, D., Stichel, S., and Wittum, G. (2010). Simulation Mount, D., and Arya, S. (2012). ANN Library.
of density-driven flow in fractured porous media. Adv. Water Resour. 33, Muha, I., Grillo, A., Heisig, M. Schöneberg, M., Linke, B., and Wittum, G. (2012).
1494–1507. doi: 10.1016/j.advwatres.2010.08.004 Mathematical modeling of process liquid flow and acetoclastic methanogenesis
Hackbusch, W. (1985). Multi-Grid Methods. Leipzig: Springer. under mesophilic conditions in a two-phase biogas reactor. Biores. Technol. 106,
Hackbusch, W. (2003). Multi-Grid Methods and Applications in Springer 1–9. doi: 10.1016/j.biortech.2011.11.087
Computational Mathematics. Leizpig: Springer. Muha, I., Nägel, A., Stichel, S., Grillo, A., Heisig, M., and Wittum, G. (2011).
Hansen, S., Henning, A., NÃd’gel, A., Heisig, M., Wittum, G., Neumann, D., Effective diffusivity in membranes with tetrakaidekahedral cells and implica-
et al. (2008). In-silico model of skin penetration based on experimentally tions for the permeability of human stratum corneum. J. Memb. Sci. 368, 18–25.
determined input parameters. Part i: Experimental determination of parti- doi: 10.1016/j.memsci.2010.10.020
tion and diffusion coefficients. Eur. J. Pharm. Biopharm. 68, 352–367. doi: Muller, D., Nikonenko, I., Jourdain, P., and Alberi, S. (2002). LTP,
10.1016/j.ejpb.2007.05.012 memory and structural plasticity. Curr. Mol. Med. 2, 605–611. doi:
Hayashi, Y., and Majewska, A. (2005). Dendritic spine geometry: functional 10.2174/1566524023362041
implication and regulation. Neuron 46, 529–532. doi: 10.1016/j.neuron.2005. Murase, S., and Schuman, E. M. (1999). The role of cell adhesion molecules
05.006 in synaptic plasticity and memory. Curr. Opin. Cell Biol. 11, 549–553. doi:
Helmchen, F., Imoto, K., and Sakmann, B. (1996). Ca2+ buffering and action 10.1016/S0955-0674(99)00019-8
potential-evoked ca2+ signaling in dendrites of pyramidal neurons. Biophys. J. Nägel, A., Hansen, S., Neumann, D., Lehr, C. M., Schaefer, U. F., Wittum, G., et al.
70, 1069–1081. doi: 10.1016/S0006-3495(96)79653-4 (2008). In-silico model of skin penetration based on experimentally determined
Heppner, I., Lampe, M., Nägel, A., Reiter, S., Rupp, M., Vogel, A., et al. (2013). input parameters. Part ii: Mathematical modelling of in-vitro diffusion experi-
“Software framework uG4: parallelmultigrid on the hermit supercomputer,” in ments. identification of critical input parameters. Eur. J. Pharma. Biopharm. 68,
High Performance Computing in Science and Engineering, eds W. E. Nagel, D. H. 368–379. doi: 10.1016/j.ejpb.2008.11.009
Kröner, and M. M. Resch (Berlin-Heidelberg: Springer), 434–449. Nägel, A., Heisig, M., and Wittum, G. (2009). A comparison of two- and three-
Hines, M. L., and Carnevale, N. T. (2003). “The NEURON simulation environ- dimensional models for the simulation of the permeability of human stra-
ment,” in The Handbook of Brain Theory and Neural Networks, ed M. A. Arbib tum corneum. Eur. J. Pharm. Biopharm. 72, 332–338. doi: 10.1016/j.ejpb.
Vol. 2, (Cambridge MA: MIT Press), 769–773. 2008.11.009
Hodgkin, A. L., and Huxley, A. F. (1952). A quantitative description of membrane Nagerl, U. V., Mody, I., and Vergara, J. L. (2000). Binding kinetics of calbindin-
currents and its application to conduction and excitation in nerve. J. Physiol. d(28k) determined by flash photolysis of caged ca(2+). Biophys. J. 79,
117, 550–544. 3009–3018. doi: 10.1016/S0006-3495(00)76537-4
Ierusalimschy, R., Celes, W., and de Figueiredo, L. H. (2013). Lua: The Programming Oliveira, R., Terrin, A., Di Benedetto, G., Cannon, R. C., Koh, W., Kim,
Language. Rio: MIT License. M., et al. (2010). The role of type 4 phosphodiesterases in generating
Jungblut, D., Queisser, G., and Wittum, G. (2011). Inertia based filtering of microdomains of cAMP: large scale stochastic simulations. PLoS ONE 5:e11725.
high resolution images using a gpu cluster. Comp. Vis. Sci. 14, 181–186. doi: doi: 10.1371/journal.pone.0011725
10.1007/s00791-012-0171-2 Popov, V. I., Kleschevnikov, A. M., Klimenko, O. A., Stewart, M. G., and Belichenko,
Kanamori, T. (2013). Compartmentalized calcium transients trigger dendrite prun- P. V. (2011). Three-dimensional synaptic ultrastructure in the dentate gyrus and
ing in Drosophila sensory neurons. Science 340, 1475–1478. doi: 10.1126/sci- hippocampal area ca3 in the ts65dn mouse model of down syndrome. J. Comp.
ence.1234879 Neurol. 519, 1338–1354. doi: 10.1002/cne.22573
Pumplin, D. W., Reese, T. S., and Llinas, R. (1981). Are the presynaptic membrane Weinan, E., Engquist, B., and Huang, Z. (2003). Heterogenous multiscale method:
particles the calcium channels? Proc. Natl. Acad. Sci. U.S.A. 11, 7210–7213. doi: a general methology for multiscale modeling. Phys. Rev. B 67, 367–450. doi:
10.1073/pnas.78.11.7210 10.1103/PhysRevB.67.092101
Queisser, G., Wittmann, M., Bading, H., and Wittum, G. (2008). Filtering, recon- West, A., Griffith, E., and Greenberg, M. (2002). Regulation of transcription
struction, and measurement of the geometry of nuclei from hippocampal factors by neuronal activity. Nat. Rev. Neurosci. 3, 921–931. doi: 10.1038/
neurons based on confocal microscopy data. J. Biomed. Opt. 13:014009. doi: nrn987
10.1117/1.2829773 Wikipedia (2012). K-d tree — Wikipedia, The Free Encyclopedia.
Ray, S., and Bhalla, U. S. (2008). PyMOOSE: interoperable scripting in Python for Wils, S., and De Shutter, E. (2009). STEPS: modeling and simulationg com-
MOOSE. Front. Neuroinf. 2:6. doi: 10.3389/neuro.11.006.2008 plex reaction-diffusion systems with python. Front Neuroinf. 3:15. doi:
Resasco, D. C., Gao, F., Morgan, F., Novak, I. L., Schaff, J. C., and Slepchenko, B. M. 10.3389/neuro.11.015.2009
(2012). Virtual cell: computational tools for modeling in cell biology. Wiley Wittmann, M., Queisser, G., Eder, A., Wiegert, J., Bengtson, C., Hellwig, A.,
Interdiscip. Rev. Syst. Biol. Med. 4, 129–140. doi: 10.1002/wsbm.165 et al. (2009). Synaptic activity induces dramatic changes in the geometry of
Shewchuk, J. R. (2002). “What is a good linear finite element? - interpola- the cell nucleus: interplay between nuclear structure, histone H3 phospho-
tion, conditioning, anisotropy and quality measures,” Proceedings of the 11th rylation, and nuclear calcium signaling. J. Neurosci. 29, 14687–14700. doi:
International Meshing Roundtable (Berkeley, CA). 10.1523/JNEUROSCI.1160-09.2009
Si, H. (2009). TetGen: a quality tetrahedral mesh generator and three-dimensional Wolf, S., Grein, S., and Queisser, G. (2013). Employing NeuGen 2.0 to auto-
delaunay triangulator. Int. J. Num. Methods Eng. 75, 856–880. matically generate realistic morphologies of hippocampal neurons and neu-
Silver, R. A., Lamb, A. G., and Bolsover, S. R. (1990). Calcium hotspots caused by ral networks in 3D. Neuroinformatics 11, 137–148. doi: 10.1007/s12021-012-
l-channel clustering promote morphological changes in neuronal growth cones. 9170-1
Nature 343, 751–754. doi: 10.1038/343751a0 Xylouris, K. (2013). Generalized cable equation for signal processing in neurons.
Spacek, J., and Harris, K. M. (1997). Three-dimensional organization of smooth PhD Thesis.
endoplasmic reticulum in hippocampal CA1 dendrites and dendritic spines of Xylouris, K., Queisser, G., and Wittum, G. (2007). A three-dimensional mathe-
the immature and mature rat. J. Neurosci. 24, 4233–4241. matical model of active signal processing in axons. Comput. Visual. Sci. 13,
Tada, T., and Sheng, M. (2006). Molecular mechanisms of dendritic spine morpho- 409–418. doi: 10.1007/s00791-011-0155-7
genesis. Curr. Opin. Neurobiol. 16, 95–101. doi: 10.1016/j.conb.2005.12.001
Tai, C., Kim, S., and Schuman, E. (2008). Cadherins and synaptic plasticity. Curr. Conflict of Interest Statement: The authors declare that the research was con-
Opin. Cell Biol. 5, 567–575. doi: 10.1016/j.ceb.2008.06.003 ducted in the absence of any commercial or financial relationships that could be
Takahashi, A., Camacho, P., Lechleiter, D. J., and Herman, B. (1999). Measurement construed as a potential conflict of interest.
of intracellular calcium. Physiol. Rev. 79, 1089–1125.
Thompson, J. F., Soni, B. K., and Weatherill, N. P. (2012). Handbook of Grid Received: 09 January 2014; accepted: 28 June 2014; published online: 29 July 2014.
Generation. Leipzig: CRC Press. Citation: Grein S, Stepniewski M, Reiter S, Knodel MM and Queisser G (2014) 1D-3D
Van Aelst, L., and Cline, H. (2004). Rho GTPases and activity-dependent hybrid modeling—from multi-compartment models to full resolution models in space
dendrite development. Curr. Opin. Neurobiol. 14, 297–304. doi: and time. Front. Neuroinform. 8:68. doi: 10.3389/fninf.2014.00068
10.1016/j.conb.2004.05.012 This article was submitted to the journal Frontiers in Neuroinformatics.
Vogel, A., Reiter, S., Rupp, M., Nägel, A., and Wittum, G. (in press). uG 4 - A Copyright © 2014 Grein, Stepniewski, Reiter, Knodel and Queisser. This is an open-
novel flexible software system for the simulation of PDE-based models on high access article distributed under the terms of the Creative Commons Attribution License
performance computers. Comput. Vis. Sci. (CC BY). The use, distribution or reproduction in other forums is permitted, provided
Wald, I., and Hvran, V. (2006). “On building fast kd-trees for ray tracing and the original author(s) or licensor are credited and that the original publication in this
on doing that in o(NlogN),” IEEE Symposium on Interactive Ray Tracing journal is cited, in accordance with accepted academic practice. No use, distribution or
(Amsterdam). doi: 10.1109/RT.2006.280216 reproduction is permitted which does not comply with these terms.