Dynamical Systems Theory &

Neuroscience
Lecture # 8
May 22, 2012
Dynamical Systems Theory
 Definition: A dynamical system consists of a set of variables
that describe its state and a law that describes the evolution of
the state variables with time
 how the state of the system in the next moment of time depends on the
input and its state in the previous moment of time

Set of Variables Evolution Rule

Dani S. Bassett
An Example Application
Activity of a Single Neuron
 State Variables:
 Membrane potential
 Excitation variables, such as activation of a Na+ current.
 Recovery variables, such as inactivation of a Na+ current and activation of
a fast K+ current.
 Evolution Rule:
 Hodgkin-Huxley Model
 Integrate-and-Fire Models
 Etc.

Other Applications
 Motor Behavior (Meyer-Lindenberg et al. 2002)
 Neuronal Decision-Making (Briggman et al. 2005)
 Etc.

Dani S. Bassett
Neuronal Activity
 Identifying State Variables

The state of a neuron can be described by the membrane potential, V , and the activation
variable, n, of the persistent K+ current.

The evolution law would then be a 2-dimensional system of ordinary differential equations
describing how the state of the system in the next moment of time depends on the input
and its state in the previous moment of time
Dani S. Bassett
Phase Diagrams
 Understanding Neuron Dynamics in Stable Equilibrium
A neuron at rest: Its phase diagram:

• The axes of the phase diagram are the state variables.

• The trajectories in the phase diagram are determined by the evolution rule.
• Units of time may be thought of as plotted along each trajectory.
• Stable equilibrium is denoted by the black dot, called an attractor.

Dani S. Bassett
Phase Diagrams II
 Understanding Dynamics of an Active Neuron
An active neuron : Its phase diagram:

• Small perturbations result in small excursions from the stable equilibrium

point (e.g., postsynaptic potential).
• Large perturbations result in larger excursions that are amplified by the
neuron's intrinsic dynamics and result in the spike response.

Dani S. Bassett
Phase Diagrams III
 Understanding Dynamics of an Oscillating Neuron
A spiking neuron : Its phase diagram:

• A sufficiently strong current will cause the neuron to exhibit periodic

spiking activity, also known as a stable periodic orbit.
• A neuron can be switched from one mode (e.g., equilibrium) to another
(e.g., periodic spiking) by a transient input.

Dani S. Bassett
Bifurcations - Decisions
 How does a neuron decide to fire or not?
An active neuron : A spiking neuron :

• the transition corresponds to a bifurcation of neuron dynamics, i.e., a

qualitative change of phase portrait of the system.
• neurons are excitable because they are near bifurcations from resting to
spiking activity, so the type of the bifurcation determines the excitable
properties of the neuron.

Dani S. Bassett
Bifurcations – Type I
Phase portraits before, during, and after a transition at a
saddle-node bifurcation
Before During After

• As the magnitude of the injected current changes, a stable equilibrium

corresponding to the resting state is approached by an unstable equilibrium.
• They coalesce and annihilate each other.
• Since the resting state no longer exists, the trajectory jumps to the limit cycle
attractor, indicating that the neuron starts to fire tonic spikes.
Dani S. Bassett
What does this mean for a neuron?
 Neurons that display saddle-node bifurcations can be in either resting or
spiking states, and do not have subthreshold oscillations

Dani S. Bassett
Saddle-Node Bifurcations
 Neurons that go through saddle-node bifurcations are
defined as having “Class 2 excitability” – meaning bistable
integrator (all-or-none spikes) or resonator.

Dani S. Bassett
Example Neurons & SNBs
Fast spiking interneurons fire high-frequency tonic spikes with relatively constant
period. When the magnitude of the injected current decreases below a certain
critical value, they fire irregular spikes, switching randomly between resting and spiking
states.

Morphologically, these neurons are sparsely spiny or aspiny nonpyramidal cells (basket
or chandelier) providing local inhibition along the horizontal (intra-laminar) direction
of the neocortex.

Dani S. Bassett
Bifurcations – Type II
 Phase portraits before, during, and after a transition at a
saddle-node on invariant circle bifurcation
Before During After

• Similar to the saddle-node bifurcation

• Except that there is an invariant circle at the moment of bifurcation, which
then becomes a limit cycle attractor

Dani S. Bassett
What does this mean for a neuron?
 Neurons that display saddle-node on invariant circle are monostable and
do not have subthreshold oscillations.

Dani S. Bassett
Saddle-Node on Invariant Circle Bifurcations
 Neurons that go through saddle-node on invariant circle bifurcations are
defined as having “Class 1 excitability” – meaning. These are monostable
integrators that can continuously encode the strength of an input into the
frequency of spiking.

Dani S. Bassett
Example Neurons & SNICBs
Regular spiking neurons fire tonic spikes with adapting (decreasing) frequency in
response to injected pulses of DC current. The inter-spike frequency vanishes when
the amplitude of the injected current decreases.

Morphologically, these neurons are spiny stellate cells in layer 4 and pyramidal cells in
layers 2, 3, 5, and 6.

Dani S. Bassett
Understanding “Why?”
 In order to understand “why” these behaviors occur, we
need to model the neurons themselves and understand
their evolution rule (set of differential equations for the
evolution of state variables through time).

Set of Variables Evolution Rule

Dani S. Bassett
Conceptual Framework
 Going beyond models of single neurons, how can DST
help us to understand complex systems?
 Reconceptualizing Complex Systems:
 Treat the behavioral state of the system at any instant as a
point in the phase space of this system.
 Treat different behaviors as fixed points in phase space.
 Each axis represents a variable that measures the temporal
evolution of trajectories in this space.
 Single variables can not fully characterize the system
 Rather, the trajectory of the system in the space of several variables
characterizes the system
 These state variables can be linear combinations of observed
variables.
 Take all of your data and determine underlying “hidden” variables.

Dani S. Bassett
Leech Behavior - Brigmann et al. 2005 Science
 Take all variables and use a data-reduction technique like PCA to
determine unique state variables.
crawling

swimming

 Use the trajectory in this phase space to determine whether the

trial will be a crawling or a swimming trial as the decision is being
made.

Dani S. Bassett
Mvt Planning - Churchland et al. 2010 Nat Neurosci
 Again, use GPFA (a factor analysis, similar to PCA) to reduce
data to 2 principal dimensions/factors.

 The trajectory in this phase space clearly follows behavior.

 The stimulus was a reach target (135°, 60 mm distant), with no
reach allowed until a subsequent go cue.
Dani S. Bassett
Finger Mvt – Meyer-Lindenberg et al. 2002 PNAS
 Moving fingers in mirror movements is much easier than
moving fingers in parallel movements at high frequencies.
 The coordination law governing the relative phase of
hand movement

 State variable: phase φ. The frequency is a.

 Two fixed points exist, corresponding to the mirror, in-
phase (φ=0) and the parallel, out-of-phase (φ=π)
movement.
 The relative stability of these patterns is governed by the
term b/a.
Regions of Interest
 The regions of interest in this study are those that show an
interaction between type of movement (mirror or parallel)
and frequency of movement (high or low).

Control
regions – no
interactions
(M1/S1)

ROIs– strong
interactions
(PMA/SMA)
TMS can switch brain between 2 states
 TMS at the regions of interest (that show frequency x mvt type
interactions) causes switches from the less stable movement (parallel) to
the more stable movement (mirror).
Relationship between frequency and TMS intensity
 If the behavior is very unstable (parallel movements at high frequencies),
only a small amount of TMS is needed to cause a switch to mirror
movements.
 If the behavior is more stable, larger amounts of TMS are needed.
Sustained Changes
 On the trials where a switch occurred, the phase showed a sustained
change (for ~ 30 cycles).
 Suggests that this represents a large-scale behavioral reorganization, rather
than simply errors in movement performance and sequencing.
What do these studies have in common?
 Treat the brain as a dynamical rather than a static system.
 Study trajectories rather than states.
 Look for state variables (that might rely on collective interactions).
 Attempt to understand stable behaviors and transitions between them
(decision-making, movements, etc.).
Other Resources

 Dynamical Systems in Neuroscience (2007) by Izhikevich

 From which a significant portion of the content in this presentation
was taken.
 Focuses on single neuron applications

 Mathematical Foundations of Neuroscience (2010) by

Ermentrout & Terman
 Includes more on networks of coupled neurons

 Complex and Adaptive Dynamical Systems: A Primer

(2011) by Gros
 For more of the mathematics behind the theory.

Dani S. Bassett
Discussion Points for Brigmann et al. 2005
 The paper frames the problem such that the leech's nervous system is a dynamical system with three stable
states: swimming, crawling or whole-body shortening. Evidence for these stable states can be seen both in
the leech's behavior and in distinct patterns of neuronal firing.
 Is there another state in which the leech is doing none of these three things? Does its system wander
through phase space (producing a "do nothing" behavior) until it enters one of the stable states, or is it
constantly doing one of these behaviors and only switches when a biasing input moves the system?
 t_{nerve} is the time at which the state of the system is unambiguous based on peripheral motor neuron
firing patterns.The authors found 17 cells that can predict behavior before t_{nerve} and consider them
candidates for "decision-making" cells, but these cells were not the most highly weighted by LDA. Is there
any theory about how the leech nervous system might make use of something like LDA weights? Weights in
a multivariate analysis are highly dependent on the classifier used. Is there any reason to believe the nervous
system would prefer LDA weighting over something like weights from a naive Bayes classifier or SVM?
 How exactly did they get individual cell weights if they performed LDA on the PC-transformed data?
 The authors make a note that some trials begin with a trajectory toward swimming, but then diverges and
ends up in the crawling state. There was a talk at SfN last year where a very similar analysis was performed
with a monkey that could move a cursor to one of two cued positions on a screen. While the monkey
fixated, "trap doors" on the screen would appear and block the path to one of the possible targets. A similar
dynamical system model was built using PCA, and the trajectory of the movement plan could be seen to
move from one trajectory to the other.
 Are all the connections between neurons in the leech known? The multivariate analysis applied here does
not model any connectivity between neurons.
 When modelling a biological system, some sort of data is collected from the system to test the model. I
don't know much about modern optical imaging, but voltage-sensitive dyes can be toxic to cells. Also,
impaling cells will change their electrical properties. Are the small number of trials in this experiment due
to the leeches dying? Is this really a study of the injured leech nervous system?
Discussion Points for Meyer-Lindenberg et
al. 2002
 What would a split-brain patient do? Would they be able
to have more than 2 fixed points?
 Why is there such a big lateralization effect for PMA?
 What factors could effect individual variability in
transition frequency?
 Are behavioral states coded for by different neuronal
ensembles? Or the same ensembles acting differently?
Bifurcations – Type III
 Phase portraits before, during, and after a transition at a
saddle-node bifurcation
Before During After

• A small unstable limit cycle shrinks to a stable equilibrium and makes it lose
stability.
• Because of instabilities, the trajectory diverges from the equilibrium and
approaches a large-amplitude spiking limit cycle or some other attractor.

Dani S. Bassett
Bifurcations – Type IV
 Phase portraits before, during, and after a transition at a
saddle-node bifurcation
Before During After

Dani S. Bassett