Sum of the Parts:Musings on the Function

of the Hippocampo-Entorhinal System

NaK GroupSeptember 24, 2003

“Computational models of the hippocampal region: linking incremental and

episodic memory”

MA Gluck, M Meeter and C.E. MyersTRENDS in Cognitive Sciences

June 2003

Connection Overview

Interesting Features:Layer II perforant path splits: one to DG, one to CA3Layer III projects to CA1Unidirectional“Tri-synaptic Circuit”

Incremental (Multiple Trial) Learning:Gluck and Myers

Hippo system performs info processing that transforms stimulus representations according to specific rules with series of connected nets

Compresses (makes similar) co-current or redundant inputDifferentiates inputs that predict different future eventsPasses new assemblies to LTM networks in the neocortex and cerebellum, where error between predicted (idealized or random) hippo output and actual output is used to update weights

Recently, specifically proposed EC’s anatomy and physiology could compress representations of co-current stimuli

Incremental Learning (Multiple Trials): Schmajuk and DiCarlo

Hippo region crucial for forming new stimulus configs (A and B diff from AB)

Cortex combines cue info to allow configural learning, then cerebellum learns to map this configural info into a behaviorHippo region calculates error between prediction and actual, then sends the error measure to neocortex and cerebellum, as well as sending predictions to cerebellum

Recently, proposed that the prediction signal, used by the hippo region, originates in the EC

Episodic (One Time) Memory Models

Stores random vectors (not unreasonable since info may only appear once)Hippo system orthogonalizes conjunctive, overlapping, neocortical patterns by forming relatively sparse patterns, reducing interference from similar memoriesUsing computational arguments, this storage may be temporary (during theta), with older memories passed to the neocortex (during SPWs)GABAergic modulation has been argued to facilitate encoding new memories while not interfering with retrieval of oldHippo system may also play a role in sequence learning and spatial navigation

Episodic Memory:Brain Substrates

CA3Ideal for binding diff parts of one pattern (autoassociative) or binding diff patterns in sequence (heteroassociative), due to high recurrent collateral

CA1“Decodes” hippo patterns, allowing association with the cortical pattern from which it originatedMay be a pattern separator

Dentate Gyrus“Sparsifier” enabling pattern separation

Entorhinal CortexMay extract regularities of longer time intervals, forming a familiarity signal

“Physiological patterns in the Hippocampo-

entorhinal cortex system”

JJ Chrobak, A Lorincz and G BuzsakiHIPPOCAMPUS

2000

Theta and Sharp WavesTheta Waves

Seen during exploration, sensory input, etc.EC Layer II and III neurons fire in theta-modulated gamma freq, projecting to the dentate, CA3, CA1, and the subiculumDentate and CA1 neurons also independently fire theta-modulated gamma as they receive EC inputEC Layers V and IV are relatively quietThought to allow EC neurons to alter synaptic connectivity in the hippo

Sharp WavesSeen during consummatory behaviors, sitting quietly, etc.EC Layers V and IV neurons fire 140-200 Hz as they receive sharp waves from the hippo (originating primarily in CA3)This discharge coincides with neocortical activity (perirhinal and medial prefrontal cortexes)EC Layers II and III do not increase their firing ratesThought to allow hippo neurons to alter connectivity of neocortical neurons

EC to Hippocampus Projections

EC Layer II project to DG and CA3 by perforant path

Stellate cells and pyr cells form islands, which may represent functional clustersStellates exhibit theta-freq, sub-threshold membrane oscillations, firing (gamma) spike clusters on depolarizing phases, conveying patterns to DG/hippo targets

EC Layer III projects to CA1 and SubiculumPrimarily pyr cells, similar to neocortical neuronsPossibly “high fidelity” pattern transmitters of cortical input

Hippocampus to EC Projections

Layer V and IV are primary receivers of hippo output, which then project to cortical, subcortical (amygdala, septum, etc.) targets, and Layers II and III

Occurring primarily during SPWsLayer III only slightly increased their firing rates and Layer II showed no change

Experimental stimulation of deep layers produces inhibition of superficial layersLesion of Layer III will allow propogation of epileptiform bursts from deep layers to Layer II, implying Layer III may act as a “gate”

The gating may be controlled by input from the amygdala, which projects to Layers III and V and is excited by SPWs via CA1

Working Together: EC-Hippocampal Cooperation

“Novelty” (error)-detecting “reconstruction network”Assumptions:

Neocortical patterns are primarily projected to the hippo via Layer IIIThe hippo reconstructs the neocortical template in order to optimize the pattern into temporal sequencesLayer II compares neocortical inputs and feedback inputs from the hippocampus

Steps:1. Primary input from the

neocortex and Layer V-transformed output from the hippocampus is compared by Layer II

2. Layer II “calculates” error or novelty, and this is sent to the DG and CA3, where alterations (plasticity) occur in the CA3 network

3. If there is no novelty, then the hippocampus simply reproduces previously stored patterns

4. This will continue until the error is minimized

Working Together:Summary

“Hippocampus as comparator: role of two

input and two output systems of the hippocampus in selection and registration

of information”

OS VinagradovaHIPPOCAMPUS

2001

Two InputsReticulo-Septal

“Attention” mechanismTheta-modulated, allowing “packeting”May organize hippocampal responses to sensory input and protect them from interference

Cortico-HippocampalCortical areas, as well as EC, areas gather sensory infoDG prelim “mixer,” that generalizes and simplifies (sparsifies?) the signal before CA3

Two OutputsCA3 to

Septum (and on to the brain stem)Regulates level of arousal by inhibiting the Reticular FormationDuring novel stimulus, the RF is released (arousal) due to decreased output of CA3, which is being used for processing and therefore subject to more inhibitory controlWhen novelty is lost, CA3 activity increases again, suppressing RF

CA1Schaffer Collateral as “filter”Thought to shunt dendritic APs, possibly through local I cells, blocking cortical signalsOnly CA1 cells not receiving CA3 input participate in processing and transmission

CA1-Subiculum to Limbic Circuit to NeocortexEncoding preserved as outputs and more differentiated the farther away from hippoThis additional processing may be crucial for permanent storage in cortex

Putting It All TogetherCA3 “compares” cortical (via DG) and brain stem (via Septum) inputs 1. In constant state (no cortical input), CA3 indirectly

suppresses RF2. Change causes regulatory inhibition to dominate CA3,

which releases RFTheta activatedCortical pathways to some CA1 cells blocked

3. Cortical signal develops with delayCA3 response starts to habituateCA1 output passes to limbic circuit, which is additionally processed at each higher level and eventually stored

4. CA3 completely habituates as novelty is lost, returning the system to “closed” state

5. If familiar signal appears, the system briefly “opens” again, but quickly closes

Comparator System