"In order to understand the brain, we have used the computer as a model for it. Perhaps it is time to reverse this reasoning. To understand where we should go with the computer, we should look to the brain for some clues,"
Robert Noyce - co-inventor of the integrated circuit.
The hippocampus' beautiful neuronal architecture and its
importance in several cognitive functions as for spatial navigation and
memory formation have made it a central topic of research
during the past 40 years. The many recent exciting discoveries and development
in experimental and in computational studies, are bringing keys to break the
neural code of this central brain area.
This tutorial aims to provide the current state of the art
of our understanding of how the hippocampus manages
to perform these two functional roles; memory formation and spatial navigation.
To this aim, a state of the art of current biological evidences will be
given. Among others, this will focus on :
how the hippocampus processes information from an anatomical point of view;
how different rhythms are involved in the processing of hippocampal information
(from slow rhythms (~1Hz) to fast ripples (~200Hz));
what are the specific firing activity of hippocampal pyramidal cells (place cells,
direction cells, grid cells and more recently the border cells);
how firing activity is modulated by the hippocampal rhythms;
how learning occurs and its relation to the various form of synaptic plasticity;
To decipher the hippocampal cognitive functions, biological data alone are
not enough. Computational neuroscientists are playing a key role in giving insights on
how to put these biological data together in an integrative view. To this aim, computational hypotheses
are proposed and simulated using computational models. In turn, these models are proposing
new predictions which lead to new experiments. In this tutorial, several of these models will be
reviewed and compared. Since these models involve different neural architectures (e.g. feedforward and recurrent),
different types of learning rules (e.g. asymmetric Hebbian learning and long term potentiation), different types
of neural units (e.g. McCullogh and Pitts or oscillator units), we expect these models to be insightful
for researchers in the field of artificial intelligence and neural information
processing.
Overview
I decided three years ago to prepare and give a tutorial on my research field (originally it was prepared with Naoyuki Sato).
One aim was to force me to review 'exhaustively' the literature and to confront my hippocampal models with other existing models.
Each year this tutorial is updated based on the last trends in the field and also based on comments I received from the persons who followed it the year before.
PART I: Background
Cognitive roles of the Hippocampus from lesion studies (both human and animals)
The famous HM patient (bilateral removal of temporal lobe)
Role in episodic memory (what means memory ? overview of the classification proposed by L.Squire)
Role in navigation (london taximen)
Distinction between human hippocampus and animals models
Anatomy and organization of the Hippocampus
Where is the hippocampus (in humans and animals)
What distinguish the hippocampus from the rest of the brain
The famous tri-synaptic loop
Sending and receiving data : afferent and efferent brain regions
The different cell types: excitation and inhibition
Evidence for Hippocampus' synaptic plasticity and learning
LTP/LTD and other types of learning
Memories and their permanence (long vs short - synaptic plasticity vs neurodynamics)
Modeling and artificial neural networks
Rapid review of a neural model: the architecture, the activation unit and the learning of the synaptic weights
David Marr's legacy to the field
The importance of brain's rhythms
Theta, gamma, sharp waves ripples, slow oscillation (up down states). Neural evidence and their link to behavioral/cognitive states
Do we need to model them or not? A possible need for oscillatory units?
PART II: Hippocampus behavioral and neurophysiological data with associated neural models
Very recent data and models will be presented (including non published data from last SFN meeting to highlight new trends to the field). When appropriate my own models will be presented and confronted with other models
Real data manipulation to get a better feeling of what is going on (data received from Buzsaki's laboratory)
Models for path integration and the cognitive map
The components: grid cells , place cells, head direction cells and the phase precession mechanism
The head direction network
Models of place cells formation
Models of grid cells formation
Does phase precession originates from a network, a unit or a cellular mechanism ?
More integrated models for path integration and the cognitive map
Why the dentate gyrus? Its possible role in remapping and orthogonalization of the information
Memory formation, consolidation and retrieval
The hypothesis
Episodic memories and asymmetric learning rule
Sharp waves and memory consolidation (forward and reverse replay)
More integrated models
Examples of what will appear during the tutorial
On one hand, this tutorial will show real biological data, how these data are analyzed and what
these data tell us about the hippocampus cognitive functions.
On the other hand, several computational models will be presented and explained. To facilitate the course, when possible, schematic
pictures will be provided. Below are a few of these schematic dynamic pictures programmed with flash.
Rat
running in an environment. By inserting electrodes in the hippocampus
(DG-CA3-CA1), we can record the activity of individual cells. It has been
demonstrated that most of these cells have place selectivity: the cell fire
only when the rat is at a specific portion of the environment (O'Keefe
Dostrovsky - 1971). Quite surprisingly, it has been observed a few years ago
that medial enthorinal cells (located one synapse upstream the hippocampal
cells), are firing at multiple locations in the environment; these locations
forming a grid (Hafting et al. (2005)).
Flash animation: Use your Mouse to click on the rat's brain to insert electrodes and to see the
resulting receptive fields on the behavioral space.
Example
of a computational model for memory encoding during exploratory behavior and
their reactivation during awake and sleeping sharp waves. (Molter et al. (2007))
During exploratory behavior, place fields are crossed sequentially. This sequence
is encoding into the CA3 recurrent network in form of asymmetric connections due to an
asymmetric Hebbian learning rule.
After memory encoding, its reactivation during sharp wave events will occur
according to two different patterns. First, in term of forward replay during
sleep, second in term of reverse replay during awake sharp waves.
Flash animation: Push on the play/replay button to (re)simulate the encoding part. After encoding,
use your mouse to click on the rat
eating or sleeping images to simulate sharp wave events (forward replay during sleep and reverse replay during awake state).
NB: a real computational model is implemented behind this graphical view to generate the behavior.
When running in an environment, hippocampal cells are driven by a local field potential
oscillating at the frequency of ~8Hz (the theta rhythm). At a first glance,
the firing activity of hippocampal pyramidal cells is phase locked to this theta rhythm.
A more precise observation has demonstrated that these cells are firing with 'phase precession':
the phase of the firing activity continuously precesses while the rat is crossing the place field.
Flash animation: Click on the place fields buttons to add/remove them from the view. Select the
type of firing rate model you want to test by selecting a radio button (you can choose between with or without theta phase precession).
You can click on the phase precession image to enlarge it. This figure shows an example of phase precession figure made with real data.
Example
of a computational model explaining how hippocampal place cells can be formed
from the conjunctive activity of entorhinal grid cells. (Molter and Yamaguchi (2008))
Our model is the first model giving a functional role for the entorhinal phase precession. We propose that
this temporal code enables to coordinate the activity of multiple grid cells having different
scales and offsets (spatial phase). As a result, place fields are formed one synapse downstream.
Hebbian plasticity helps to further stabilizes the place fields representations.
Flash animation: it enlights our computational hypothesis by showing a schematic view of two grid cells connected to one DG cell.
To face our model with other models not including theta phase precession, you can choose entorhinal cells to fire with or without phase precession.
Click on the help button (top left) to understand how to manipulate the flash animation.
Selected References
In Construction.
For Biological papers, click here .
Computational models, click here.
Hippocampal models code and related materials
Available only for the participants of the tutorial; Click here to access the page.
<UNDER CONSTRUCTION – Deadline February 2009>
