Abstract
We use computational methods to study how neurons and microcircuits in
the brain operate. We are interested in the interaction between fundamental
properties like morphology or homeostatic regulation and neural functions
like information processing or learning. Most of our models concern the
cerebellum as this brain structure has a relatively simple anatomy and
the physiology of its main neurons has been studied extensively, allowing
for detailed modeling at many different levels of complexity.
Research Goals
1) The importance of morphology for function at multiple scales.
Many neuroscientists ignore morphology in their analysis of experimental
data. Examples are the application of unrestricted one-dimensional diffusion
equations to the analysis of high resolution imaging data recorded in spines
and dendrites or the view that any neuron can be treated as a point neuron.
Often the simulator tools to properly model the effects of morphology are
simply absent. This is for example the case for reaction-diffusion modeling
of signaling pathways involved in synaptic plasticity.
Within this topic we focus on two levels:
- importance of submicroscopic morphology for reaction-diffusion systems
applied to the induction of long-term depression of parallel fiber synapses
in Purkinje cells.
- synaptic processing by active dendrites and realistic modeling of the
growth of active dendrites, again mostly focused on the Purkinje cell.
2) The importance of homeostasis in neuronal excitability and synaptic
networks.
Traditionally neurons are seen as standardized assembly line products where
a fixed combination of a particular set of features, in particular ion
and synaptic channels, determines their function. Recently we have come
to realize that there is actually a large variability in the expression
of channels for neurons with identical firing behavior and that this can
be self-regulated if changes in activity occur. This has resulted in the
concept of homeostasis for both voltage-gated and synaptic channels.
The homeostasis concept has big implications for modeling of neurons and
networks, which we want to investigate in detail. What is the spread and
structure of parameter space for neuron models and how can this be regulated
in a dynamic way? What is the impact of synaptic homeostasis on learning
rules in circuits where several learning mechanisms co-exist? What is the
proper way to incorporate homeostatic rules in neuron and network models
and are these rules constant over development?
3) Neural coding at multiple levels in the olivocerebellar and corticocerebellar
systems.
The cerebellum is interesting from a functional viewpoint as it is both
very important, considering its size and number of neurons, but rather
poorly understood, especially its function in cognition. In man the corticopontine
projection to cerebellum is the biggest cortical output, suggesting an
important interaction between cortical and cerebellar operations.
While the olivocerebellar system has been extensively modeled in the context
of motor control theory, little use has been made of spiking neuron models.
Combined modeling and experimental work has led to our hypothesis that
pauses in Purkinje cell simple spiking are an important coding principle.
How does this coding principle extend to the population level and what
is the predicted direct effect on neurons in deep cerebellar nuclei and
indirectly on the inferior olive? How does this interact with learning
in the olivocerebellar system?
Where it has been studied in detail, like the tactile projections to the
cerebellar hemispheres of the rat, it was found that primary sensory and
cortical sensory neurons project to the same area of the cerebellum with
overlapping receptive fields. This implies that the cerebellar neurons
receive two versions of the same input with a time delay between them.
What is the predicted interaction between these temporally delayed inputs
and can the comparison of these two representations be used in control
theory?
Strategies
We use computational methods in close collaboration with experimental labs
that can both provide the data needed to constrain the models and verify
the modeling predictions. We mostly use compartmental modeling strategies
but will extend these further, e.g. our new Purkinje cell model will have
molecularly identified channels. In some cases we also use simpler spiking
models for large network simulations. In general we are quite interested
in multi-scale modeling.
Our innovative research often requires software development. Currently
we are working on STEPS, a simulator for stochastic reaction-diffusion
systems in realistic morphologies, HSOLVE, a module for efficient compartmental
simulation of cellular models, and NeuroFitter for automated parameter
searches for cellular models.
|
|
