### Increased computational accuracy in multi-compartmental cable models (Lindsay et al. 2005)

Accession:129149
Compartmental models of dendrites are the most widely used tool for investigating their electrical behaviour. Traditional models assign a single potential to a compartment. This potential is associated with the membrane potential at the centre of the segment represented by the compartment. All input to that segment, independent of its location on the segment, is assumed to act at the centre of the segment with the potential of the compartment. By contrast, the compartmental model introduced in this article assigns a potential to each end of a segment, and takes into account the location of input to a segment on the model solution by partitioning the effect of this input between the axial currents at the proximal and distal boundaries of segments. For a given neuron, the new and traditional approaches to compartmental modelling use the same number of locations at which the membrane potential is to be determined, and lead to ordinary differential equations that are structurally identical. However, the solution achieved by the new approach gives an order of magnitude better accuracy and precision than that achieved by the latter in the presence of point process input.
Reference:
1 . Lindsay AE, Lindsay KA, Rosenberg JR (2005) Increased computational accuracy in multi-compartmental cable models by a novel approach for precise point process localization. J Comput Neurosci 19:21-38 [PubMed]
Model Information (Click on a link to find other models with that property)
 Model Type: Neuron or other electrically excitable cell; Brain Region(s)/Organism: Cell Type(s): Channel(s): I Na,t; I K; Gap Junctions: Receptor(s): Gene(s): Transmitter(s): Simulation Environment: NEURON; C or C++ program; Model Concept(s): Methods; Implementer(s):
Search NeuronDB for information about:  I Na,t; I K;
 / LindsayEtAl2005 readme.txt 03-192.pdf AnalyseResults.c BitsAndPieces.c CellData.dat CompareSpikeTrain.c Ed04.tex ExactSolution.dat GammaCode Gen.tex Gen1.tex Gen2.tex Gen3.tex Gen4.tex Gen5.tex Gen6.tex GenCom.c GenCom1.c GenCom2.c GenComExactSoln.c GenerateInput.c GenerateInputText.c GenRan.ran GetNodeNumbers.c Info100.dat Info20.dat Info200.dat Info30.dat Info300.dat Info40.dat Info400.dat Info50.dat Info500.dat Info60.dat Info70.dat Info80.dat Info90.dat InputCurrents.dat InputDendrite.dat JaySpikeTrain.c JayTest1.dat JayTest100.dat KenSpikeTrain.c KenTest1.dat * KenTest10.dat KenTest100.dat * KenTest10p.dat KenTest1p.dat * KenTest2.dat KenTest2p.dat KenTest3.dat KenTest3p.dat KenTest4.dat KenTest4p.dat KenTest5.dat KenTest5p.dat KenTest6.dat KenTest6p.dat KenTest7.dat KenTest7p.dat KenTest8.dat KenTest8p.dat KenTest9.dat KenTest9p.dat LU.c Mean50.dat Mean500.dat mosinit.hoc NC.pdf NC.tex NC1.tex NC2.tex NC3.tex NC4.tex NC5.tex NC6.tex NCFig2.eps * NCFig3.eps * NCFig4.eps * NCFig5a.eps * NCFig5b.eps * NCFig6.eps * NCPics.tex NeuronDriver.hoc NewComExactSoln.c NewComp.pdf NewComp.ps NewComp.tex NewComp.toc NewComp1.tex NewComp2.tex NewComp3.tex NewComp4.tex NewComp5.tex NewComp6.tex NewCompFig1.eps NewCompFig2.eps * NewCompFig3.eps * NewCompFig4.eps * NewCompFig5a.eps * NewCompFig5b.eps * NewCompFig6.eps * NewCompPics.tex NewComSpikeTrain.c NewRes.dat NewRes60.dat NewRes70.dat NewRes80.dat NewSynRes40.dat NewTestCell.d3 NResults.res OldComExactSoln.c out.res principles_01.tex rand Ratio.dat RelErr.dat ReviewOfSpines.pdf SpikeTimes.dat TestCell.d3 TestCell1.d3 TestCell2.d3 TestCell3.d3 TestCell4.d3 testcellnew2.hoc TestCGS.c TestGen1.c TestSim.hoc TestSim020.hoc TestSim030.hoc TestSim040.hoc TestSim050.hoc TestSim060.hoc TestSim070.hoc TestSim080.hoc TestSim090.hoc TestSim1.hoc TestSim100.hoc TestSim200.hoc TestSim300.hoc TestSim400.hoc TestSim500 TestSim500.hoc
\section{Distributed and point input to a segment}
In general, segments receive distributed and point sources of
input each of which require a different mathematical treatment.
The current supplied by distributed input such as intrinsic
voltage-dependent current or capacitative current is proportional
to the surface area of the segment on which it acts, whereas the
current supplied to a segment at a synapse or by an exogenous
point input is independent of the size of the segment. An implicit
assumption of a compartmental model is that distributed current
input to a segment is small by comparison with axial current
flowing along the segment.

To appreciate why this assumption is reasonable, consider a
cylindrical dendritic segment of radius $r$ (cm), length $h$ and
with membrane of constant conductance $g_\mathrm{M}$ (mS/cm$^2$).
Suppose that axoplasm has constant conductance $g_\mathrm{A}$
(mS/cm) and that a potential difference $V$ (mV) exists between
the segment boundaries, then the axial current along the segment
is $I_\mathrm{A}=\pi r^2 g_\mathrm{A} V/h$ ($\mu$A) and the total
distributed current crossing the membrane of the segment is
$I_\mathrm{M}=2\pi r h g_\mathrm{M}\,(V/2)$. The ratio of the
distributed current to the axial current is therefore
$$\label{pc1} \frac{\mbox{Distributed current}}{\mbox{Axial current}} =\frac{I_\mathrm{M}}{I_\mathrm{A}}=\frac{\pi r h g_\mathrm{M}\,V} {\pi r^2 g_\mathrm{A}\,(V/h)}=\frac{h^2 g_\mathrm{M}} {r g_\mathrm{A}}=\Big(\frac{h}{r}\Big)^2\, \frac{r g_\mathrm{M}}{g_\mathrm{A}}\,.$$
For a typical dendritic segment $r g_\mathrm{M}/g_\mathrm{A}$ is
small (say $\approx 10^{-5}$), and therefore distributed current
acting on a segment is small by comparison with axial current for
short'' segments. On the other hand, segments several orders of
magnitude longer than their radius can be expected to have
distributed and axial currents of similar magnitude. An important
property of a compartmental model is that segments are not
excessively long by comparison with their radius. In the treatment
of distributed current, the development of the new compartmental
model makes explicit use of the assumption that distributed
current is much smaller than axial current. Since this assumption
may not be valid for point sources of current, it will not be made
for the treatment of these current in the new compartmental model.

\subsection{Axial current in the absence of distributed and point current input}
Figure \ref{model} illustrates a dendritic segment of length $h$
(cm) where $\lambda\in[0,1]$ is the fractional distance of a point
of the segment from its proximal end ($\lambda=0$). Let
$r_\mathrm{P}$ and $r_\mathrm{D}$ be the radii of the segment at
its proximal and distal boundaries respectively, let
$V_\mathrm{P}(t)$ and $V_\mathrm{D}(t)$ be the membrane potentials
at these boundaries and let $I_\mathrm{PD}$ be the axial current
in the segment in the absence of transmembrane current.

\begin{figure}[!h]
\centering
\begin{tabular}{c}
\includegraphics[ ]{NewCompFig2.eps}
\begin{tabular}{p{2.55in}}
\caption{\label{model} A segment of length $h$ (cm) is illustrated. In
the absence of transmembrane current, membrane potentials $V_\mathrm{P}$
and $V_\mathrm{D}$ at the proximal and distal boundaries of the
segment generate axial current $I_\mathrm{PD}$.}
\end{tabular}
\end{figure}

%\begin{figure}[!h]
%\centering
%\begin{tabular}{c}
%\begin{mfpic}[1][1]{-40}{140}{180}{300}
%\pen{0.5pt}
%\dotspace=4pt
%\dotsize=1pt
%\pen{1pt}
%\dotspace=4pt
%\dotsize=1.5pt
%%
%% LH cylinder
%\parafcn[s]{-180,180,5}{(100-21*sind(t),240+28*cosd(t))}
%\lines{(0,288),(100,268)}
%\lines{(0,192),(100,212)}
%%
%% Partial cylinder on left
%\dotted\parafcn[s]{0,180,5}{(36*sind(t),240+48*cosd(t))}
%\parafcn[s]{0,180,5}{(-36*sind(t),240+48*cosd(t))}
%%
%% Annotation of LH cylinder
%\dashed\arrow\lines{(0,240),(100,240)}
%\tlabel[bl](50,250){\large $I_\mathrm{PD}$}
%\tlabel[bc](0,250){$V_\mathrm{P}$}
%\tlabel[cc](0,180){\large $\lambda=0$}
%\tlabel[bc](0,295){\textsf{P}}
%\arrow\lines{(0,235),(0,200)}
%\tlabel[cr](-5,220){\textsf{$r_\mathrm{P}$}}
%%
%% Annotation of RH cylinder
%\tlabel[bc](100,250){$V_\mathrm{D}$}
%\tlabel[cc](100,180){\large $\lambda=1$}
%\tlabel[cc](100,280){\textsf{D}}
%\arrow\lines{(100,235),(100,216)}
%\tlabel[cl](105,228){\textsf{$r_\mathrm{D}$}}
%\arrow\lines{(60,228),(95,228)}
%\arrow\lines{(40,228),(5,228)}
%\tlabel[cc](50,228){$h$}
%\end{mfpic}
%\begin{tabular}{p{2.55in}}
%\caption{\label{model} A segment of length $h$ (cm) is illustrated. In
%the absence of transmembrane current, membrane potentials $V_\mathrm{P}$
%and $V_\mathrm{D}$ at the proximal and distal boundaries of the
%segment generate axial current $I_\mathrm{PD}$.}
%\end{tabular}
%\end{figure}

The membrane of the segment in Figure \ref{model} is formed by
rotating the straight line PD about the axis of the dendrite to
form the frustum of a cone of radius
$$\label{mp1} r(\lambda)=(1-\lambda)r_\mathrm{P}+\lambda r_\mathrm{D} \,,\qquad \lambda\in[0,1]\,.$$
Assuming that the segment is filled with axoplasm of
constant conductance $g_\mathrm{A}$ and that no current
crosses its membrane, then the relationship between
$V_\mathrm{P}$, $V_\mathrm{D}$ and $I_\mathrm{PD}$ can be
constructed by integrating
$I_\mathrm{PD}=-\frac{g_\mathrm{A}\pi}{h}\,\Big[\, (1-\lambda)r_\mathrm{P}+\lambda r_\mathrm{D}\,\Big]^2\, \frac{dV}{d\lambda}$
with boundary conditions $V(0)=V_\mathrm{P}$ and
$V(1)=V_\mathrm{D}$. This calculation shows that the potentials
$V_\mathrm{P}$ and $V_\mathrm{D}$ give rise to axial current
$$\label{mp2} I_\mathrm{PD}= \frac{\pi g_\mathrm{A} r_\mathrm{P} r_\mathrm{D}}{h}\,\big(\,V_\mathrm{P}-V_\mathrm{D}\,\big)$$
in the absence of distributed and point currents, and that the
potential at point $\lambda$ is
$$\label{mp3} V(\lambda) = \frac{V_\mathrm{P}\,(1-\lambda)\, r_\mathrm{P}+V_\mathrm{D}\,\lambda\,r_\mathrm{D}} {(1-\lambda)\,r_\mathrm{P}+\lambda\,r_\mathrm{D}}\,.$$
Expressions (\ref{mp2}) and (\ref{mp3}) are estimates of the axial
current flowing along a segment and the potential distribution within
the segment in the absence of transmembrane current.

\subsection{Motivation for partitioning point current input - model
accuracy}\label{assertion}

One inescapable feature of a traditional compartmental model is
that small variations in the location of segment boundaries, as
might occur when a dendrite is represented by segments, may exert
a large influence on the solution of the resulting mathematical
model. Consider, for example, a point input close to a segment
boundary. A small variation in the position of that boundary may
change the assigned location of this input from the centre of one
segment to that of an adjacent segment. With respect to the
mathematical model, the location of this input is therefore
determined only to an accuracy of half a segment length, and this
indeterminacy will in turn generate a model solution that is
particularly sensitive to segment boundaries -- small changes in
these boundaries may lead to large changes in the model solution.
Of course, with a small number of point sources of input, this
problem can be avoided in a traditional compartmental model by
arranging that only one point input falls on a segment, and that
the location of this input coincides with the centre of the
segment. However, this strategy is not feasible when dealing with
large scale point input. What is required is a procedure that
describes the effect of point input on a dendritic section in a
way that is largely insensitive to how that section is represented
by segments. It is essential to recognise that there are two
primary sources of error in the construction of a compartmental
model; the first is the well-documented effect of discretising a
continuous dendrite, and the second pertains to error introduced
by the placement of input on this dendrite. In a traditional
compartmental model with $n$ compartments, the first type of error
is $O(1/n^2)$ (by analogy with the finite difference
representation of derivatives), but it is not widely recognised
that the second type of error is $O(1/n)$. Since the accuracy of
any model must be governed by the least accurate contribution to
the model, it is clear that \emph{in practice} a traditional model
is $O(1/n)$ accurate. This theoretical observation is supported by
the simulation exercises of Subsections \ref{sim1} and \ref{sim2}.
By contrast with a traditional compartmental model, the new
compartmental model describes the influence of input to an
accuracy of $O(1/n^2)$, and therefore one would anticipate that it
does not degrade the overall accuracy of the model. This assertion
is testable by a simulation exercise.

\subsection{Partitioning rule for transmembrane current}
In compartmental modelling the effect of input current enters the
mathematical model at points, or nodes, at which the membrane
potential is known. In a traditional model, these nodes are at the
centres of segments, whereas in the new model they are at the
boundaries of segments. In the new model, input at any location is
partitioned between the nodes at the proximal and distal
boundaries of the segment on which the input acts. This procedure
ensures that the solution of the mathematical model is insensitive
to small changes in the location of segment boundaries simply
because changes in these boundaries also affects how the input is
partitioned between nodes.

In the mathematical model, the effect of input to a segment is
treated as perturbations $I_\mathrm{P}$ and $I_\mathrm{D}$ to the
axial current $I_\mathrm{PD}$ at the proximal and distal
boundaries of a segment. Axial current
$I_\mathrm{PD}+I_\mathrm{P}$ is assumed to leave the proximal
boundary of a segment in the direction of its distal boundary,
while axial current $I_\mathrm{PD}+I_\mathrm{D}$ is assumed to
arrive at the distal boundary of a segment from the direction of
its proximal boundary. The perturbations $I_\mathrm{P}$ and
$I_\mathrm{D}$ must satisfy the conservation of current condition
$$\label{potc1} (I_\mathrm{PD}+I_\mathrm{D})-(I_\mathrm{PD}+I_\mathrm{P})+h\int_0^1 J(\lambda,t)\,d\lambda=0\quad\rightarrow\quad I_\mathrm{P}-I_\mathrm{D}=h\int_0^1 J(\lambda,t)\,d\lambda$$
where $hJ(\lambda,t)\,d\lambda+o(d\lambda)$ is the transmembrane
current crossing the segment in $(\lambda,\lambda+d\lambda)$. The
task is to construct expressions for $I_\mathrm{P}$ and
$I_\mathrm{D}$ that satisfy (\ref{potc1}) for all constitutive
forms for $J(\lambda,t)$. In the new compartmental model,
transmembrane current acting at point $\lambda$ is divided between
the proximal and distal boundaries of a segment in inverse
proportion to the resistance of the segment lying between the
point $\lambda$ and that boundary. If $R_\mathrm{P}(\lambda)$ is
the axial resistance of the portion of segment lying between the
point $\lambda$ and the proximal boundary of the segment, and
$R_\mathrm{D}(\lambda)$ is the axial resistance of the portion of
segment lying between the point $\lambda$ and the distal boundary
of the segment, then
$$\label{potc2} R_\mathrm{P}(\lambda) = \frac{\lambda h} {\pi g_\mathrm{A} r_\mathrm{P} r(\lambda)}\,,\qquad R_\mathrm{D}(\lambda) = \frac{(1-\lambda)h} {\pi g_\mathrm{A} r_\mathrm{D} r(\lambda)}\,,\qquad R_\mathrm{P}(\lambda)+R_\mathrm{D}(\lambda) = \frac{h} {\pi g_\mathrm{A} r_\mathrm{P} r_\mathrm{D}}\,.$$
The rule for partitioning transmembrane current now leads to the
expressions
$$\label{potc3} I_\mathrm{P} = h\int_0^1 \frac{(1-\lambda)\,r_\mathrm{P}\, J(\lambda,t)\,d\lambda}{(1-\lambda)\,r_\mathrm{P} +\lambda\,r_\mathrm{D}}\,,\qquad -I_\mathrm{D} = h\int_0^1 \frac{\lambda\,r_\mathrm{D}\,J(\lambda,t)\,d\lambda} {(1-\lambda)\,r_\mathrm{P}+\lambda\,r_\mathrm{D}}\,,$$
which clearly satisfy identically condition (\ref{potc1}) for the
conservation of current.

\subsection{Specification of transmembrane current}
Transmembrane current is usually assumed to consist of four
distinct components: capacitative current, intrinsic
voltage-dependent current, synaptic current and exogenous current.
Total transmembrane current is represented by
$$\label{tc1} \int 2\pi r \,c_\mathrm{M}\,\frac{\partial V}{\partial t}\,dx +\int 2\pi r\,J_\mathrm{IVDC}(V)\,dx+\sum J_\mathrm{SYN}(V_\mathrm{syn}) +\sum I_\mathrm{EX}$$
where the integrals and summations are taken over the length of a
segment. In this expression $c_\mathrm{M}$ ($\mu$F/cm$^2$) is the
specific capacitance of the segment membrane, $V(x,t)$ is the
distribution of membrane potential at time $t$ (msec),
$J_\mathrm{IVDC}(V)$ ($\mu$A/cm$^2$) is the density of
transmembrane current due to intrinsic voltage-dependent channel
activity, $J_\mathrm{SYN}(V_\mathrm{syn})$ ($\mu$A) describes
synaptic input and $I_\mathrm{EX}$ ($\mu$A) describes exogenous
input. Although the specific capacitance of dendritic membrane is
normally taken to be constant in neuronal modelling, it will be
treated here as a function of position to show how transmembrane
current of this type may be incorporated into the new
compartmental model. For a segment of length $h$, the expression
for $J(\lambda,t)$ corresponding to formula (\ref{tc1}) is
$$\label{tc2} \begin{array}{rcl} h J(\lambda,t) & = & \ds 2\pi h r(\lambda)\,c_\mathrm{M}(\lambda)\,\frac{\partial V(\lambda,t)}{\partial t}+2\pi h r(\lambda)\,J_\mathrm{IVDC}(V(\lambda,t))\\[10pt] &&\qquad \ds+\;\sum_k J_\mathrm{SYN}(V_\mathrm{syn})\,\delta(\lambda-\lambda_k) + \sum_k I_\mathrm{EX}(t)\,\delta(\lambda-\lambda_k) \end{array}$$
where $\lambda_k$ denotes the relative location of the $k^{th}$
synapse or exogenous input with respect to the proximal boundary
of the segment ($\lambda=0$).