Physiologically Realistic Memory Network

[Ole Jensen - Lisman Lab]

Physiologically Realistic Memory Network
========================================
The following 4 papers have appeared together as a series
in the journal Learning and Memory (1996) 3: 243-287. 
They can be down-loaded in post-script format from 

http://eliza.cc.brandeis.edu/people/ojensen/
 
We have attempted to construct a physiologically realistic
memory model based on nested theta/gamma oscillation. The
network model can explain important aspect of data from human 
memory psychology (Lisman and Idiart, Science 267:1512-15)   
and place cell recordings (PAPER 4). 


Ole Jensen  (ojensen at cajal.ccs.brandeis.edu)  
Volen Center for Complex Systems
Brandeis University
Waltham
MA 02254

Come see our poster at the Neuroscience meeting,
Nov 19, Tuesday 1:00 PM, X-$, 549:14.  


PAPER 1:
--------
Physiologically Realistic Formation of Autoassociative Memory in 
Networks with Theta/Gamma Oscillations: Role of Fast NMDA Channels.

Learning and Memory (1996) 3:243-256. 

Ole Jensen, Marco A. P. Idiart, and John E. Lisman 

Recordings from brain regions involved in memory function show
dual oscillations in which each cycle of a low frequency theta
oscillation (5-8Hz) is subdivided into about 7 subcycles by high
frequency gamma oscillations (20-60Hz). It has been proposed
(Lisman and Idiart 1995) that such networks are a multiplexed
short-term memory (STM) buffer that can actively maintain about
7 memories, a capability of human STM. A memory is encoded by a
subset of principal neurons that fire synchronously in a particular
gamma subcycle. Firing is maintained by a membrane process
intrinsic to each cell. We now extend this model by incorporating
recurrent connections with modifiable synapses to store long-term
memory (LTM). The repetition provided by STM gradually modifies
synapses in a physiologically realistic way. Because different
memories are active in different gamma subcycles, the formation of
autoassociative LTM requires that synaptic modification depend on
NMDA channels having a time-constant of deactivation that is of
the same order as the duration of a gamma subcycle (15- 50 msec).
Many types of NMDA channels have longer time-constants (150
msec), as for instance those found in the hippocampus, but both fast
and slow NMDA channels are present in cortex. This is the first
proposal for the special role of these fast NMDA channels. The STM
for novel items must depend on activity-dependent changes intrinsic
to neurons rather than recurrent connections, which have not
developed the required selectivity. Because these intrinsic
mechanisms are not error correcting, STM will become slowly
corrupted by noise. This limits the accuracy with which LTM can
become encoded after a single presentation. Accurate encoding of
items in LTM can be achieved by multiple presentations, provided
different memory items are presented in a varied interleaved order.
Our results indicate that a limited memory capacity STM model can
be integrated in the same network with a high capacity LTM model.  

PAPER 2:
--------
Novel Lists of 7+/-2 Known Items Can Be Reliably Stored in an 
Oscillatory Short-Term Memory Network: Interaction with 
Long-Term Memory 

Learning and Memory (1996) 3:257-263. 

Ole Jensen and John E. Lisman 

This paper proposes a model for the short-term memory (STM) of
unique lists of known items, as for instance a phone number. We
show that the ability to accurately store such lists in STM depends
strongly on interaction with the pre-existing long-term memory
(LTM) for individual items (e.g. digits). We have examined this
interaction in computer simulations of a network based on
physiologically realistic membrane conductances, synaptic plasticity
processes and brain oscillations. In the model, seven short-term
memories can be kept active, each in a different gamma-frequency
subcycle of a theta frequency oscillation. Each STM is maintained
and timed by an activity-dependent ramping process. LTM is stored
by the strength of synapses in recurrent collaterals. The presence of
pre-existing LTM for an item greatly enhances the ability of the
network to store an item in STM. Without LTM, the precise timing
required to keep cells firing within a given gamma subcycle cannot
be maintained and STM is gradually degraded. With LTM, timing
errors can be corrected and the accuracy and order of items is
maintained. This attractor property of STM storage is remarkable
because it occurs even though there is no LTM that identifies which
items are on the list or their order. Multiple known items can be
stored in STM, even though their representation is overlapping.
However multiple, identical memories cannot be stored in STM,
consistent with the psychophysical demonstration of repetition
blindness. Our results indicate that meaningful computation
(memory completion) can occur in the millisecond range during an
individual gamma cycle. 

PAPER 3:
--------
Theta/Gamma Networks with Slow NMDA Channels Learn Sequences and 
Encode Episodic Memory: Role of NMDA Channels in Recall 

Learning and Memory (1996) 3:264-278.

Ole Jensen and John E. Lisman 

This paper examines the role of slow NMDA channels (deactivation
about 150 msec) in networks that multiplex different memories
in different gamma subcycles of a low frequency theta oscillation.
The NMDA channels are in the synapses of recurrent collaterals and
govern synaptic modification in accord with known physiological
properties. Because slow NMDA channels have a time-constant
that spans several gamma cycles, synaptic connections will form
between cells that represent different memories. This enables brain
structures that have slow NMDA channels to store heteroassociative
sequence information in long-term memory (LTM). Recall of this
stored sequence information can be initiated by presentation of
initial elements of the sequence. The remaining sequence is then
recalled at a rate of 1 memory every gamma cycle. A new role for
the NMDA channel suggested by our finding is that recall at gamma
frequency works well if slow NMDA channels provide the dominant
component of the EPSP at the synapse of recurrent collaterals: the
slow onset of these channels and their long duration allows the firing
of one memory during one gamma cycle to trigger the next memory
during the subsequent gamma cycle. An interesting feature of the
readout mechanism is that the activation of a given memory is due
to cumulative input from multiple previous memories in the stored
sequence, not just the previous one. The network thus stores sequence
information in a doubly redundant way: activation of a memory
depends on the strength of synaptic inputs from multiple cells of
multiple previous memories. The cumulative property of sequence
storage has support from the psychophysical literature. Cumulative
learning also provides a solution to the disambiguation problem that
occurs when different sequences have a region of overlap. In a final
set of simulations, we show how coupling an autoassociative network
to a heteroassociative network allows the storage of episodic
memories (a unique sequence of briefly occurring known items). The
autoassociative network (cortex) captures the sequence in
short-term memory (STM) and provides the accurate,
time-compressed repetition required to drive synaptic modification
in the heteroassociative network (hippocampus). This is the first
mechanistically detailed model showing how known brain
properties, including network oscillations, recurrent collaterals,
AMPA channels, NMDA channel subtypes, the ADP, and the AHP
can act together to accomplish memory storage and recall. 


PAPER 4:
--------
Hippocampal CA3 Region Predicts Memory Sequences: Accounting for 
the Phase Precession of Place Cells 

Learning and Memory (1996) 3:279-287 

Ole Jensen and John E. Lisman 

Hippocampal recordings show that different place cells fire at
different phases during the same theta oscillation, probably at the
peak of different gamma cycles. As the rat moves through the place
field of a given cell, the phase of firing during the theta cycle
advances progressively (O'Keefe and Recce 1993; Skaggs et al.
1996). In this paper we have sought to determine whether a recently
developed model of hippocampal and cortical memory function can
explain this phase advance and other properties of place cells.
According to this physiologically based model, the CA3 network
stores information about the sequence of places traversed during
learning. Here we show that the phase advance can be understood if
it is assumed that the hippocampus is in a recall mode that operates
when the animal is already familiar with a path. In this mode,
sensory information about the current position triggers recall of the
upcoming 5- 6 places (memories) in the path at a rate of one
memory per gamma cycle. The model predicts that the average
phase advance will be one gamma cycle per theta cycle, a value in
reasonable agreement with the data. The model also correctly
accounts for 1) the fact that the firing of a place cell occurs during
$\sim$7 theta cycles (on average) as the animal crosses the place
field 2) the observation that the phase of place cell firing depends
more systematically on position than on time 3) the fact that
traversal of an already familiar path produces further modifications
(shifts the firing of a cell to an earlier position in the path). This
later finding suggests that recall of previously stored information,
strengthens the memory of that information. In the model, this
occurs because of a novel role of NMDA channels in recall. The
general success of the model provides support for the idea that the
hippocampus stores sequence information and makes predictions of
expected positions during gamma-frequency recall.