How does hippocampus store memory

Learn Mem. Biol Psychiatry. Aging of the cerebral cortex differs between humans and chimpanzees. Anand KS, Dhikav V. Hippocampus in health and disease: An overview. Ann Indian Acad Neurol. Harvard Health Publishing. What is addiction? Myers DG. Exploring Psychology. New York: Worth Publishers; Your Privacy Rights. To change or withdraw your consent choices for VerywellMind. At any time, you can update your settings through the "EU Privacy" link at the bottom of any page.

These choices will be signaled globally to our partners and will not affect browsing data. We and our partners process data to: Actively scan device characteristics for identification. I Accept Show Purposes. Table of Contents View All. Table of Contents. Impact of Damage. Potential Pitfalls. What Is the Hippocampus? Location Because the brain is lateralized and symmetrical, you actually have two hippocampi. Was this page helpful? Thanks for your feedback! Sign Up. What are your concerns?

Verywell Mind uses only high-quality sources, including peer-reviewed studies, to support the facts within our articles. Read our editorial process to learn more about how we fact-check and keep our content accurate, reliable, and trustworthy.

Related Articles. A more incisive picture of the way memory works is slowly starting to emerge and these new findings will help propel further research into various kinds of memory. Simon Makin is a freelance science writer based in London. Already a subscriber?

Sign in. Thanks for reading Scientific American. Create your free account or Sign in to continue. See Subscription Options. Discover World-Changing Science. Get smart. Sign up for our email newsletter. Sign Up. Support science journalism. Knowledge awaits. Rolls et al. We demonstrated this by storing associated continuous and discrete representations in the same single attractor network and then by revealing that the representation in the continuous space can be retrieved by the discrete object that is associated with that spatial position and that the representation of the discrete object can be retrieved by providing the position in the continuous representation of space.

If spatial representations are stored in the hippocampus, an important issue arises in terms of understanding memories that include a spatial component or context of how many such spatial representations can be stored in a continuous attractor network.

One very interesting result is that, because correlations between the representations of places in different maps or charts where each map or chart might be of one room or locale by using, for example, cues in the room are generally low, very many different maps can be simultaneously stored in a continuous attractor network Battaglia and Treves a ; Rolls a.

We considered how spatial representations can be stored in continuous attractor networks and how the activity can be maintained at any location in the state space in a form of short-term memory when the external e. However, many networks with spatial representations in the brain can be updated by internal self-motion i.

The ways in which path integration can be implemented in recurrent networks such as the CA3 system in the hippocampus or in related systems including the entorhinal cortex see below are described elsewhere Giocomo et al. However, those who have focused on spatial and navigation processing in the hippocampus rather than memory processing do now envisage that attractor networks are involved in hippocampal function Hartley et al.

We hypothesize that the mossy fibre inputs force efficient information storage by virtue of their strong and sparse influence on the CA3 cell firing rates, in order to produce pattern separation, as described next. The strong effects likely to be mediated by the mossy fibres have also been emphasized by McNaughton and Morris and McNaughton and Nadel We Rolls , c , a , a ; Rolls and Treves ; Treves and Rolls hypothesize that the mossy fibre input is particularly appropriate in several ways.

First, the finding that mossy fibre synapses are large and located very close to the soma makes them relatively powerful in activating the postsynaptic cell. Second, the firing activity of dentate granule cells appears to be very sparse Jung and McNaughton ; Leutgeb et al. The hypothesis is that the mossy fibre sparse connectivity solution performs the appropriate function to enable learning to operate correctly in the CA3 to CA3 synaptic connections Cerasti and Treves ; Treves and Rolls Quantitative analysis shows that the perforant path input would not produce a pattern of firing in CA3 that contains sufficient information for learning Treves and Rolls The particular property of the small number of mossy fibre connections onto a CA3 cell, approximately 46 see Fig.

This is a pattern separation effect, which means, for example, that place cells in a given environment are well separated to cover the whole space and that any new object-place associations formed are different from earlier episodic memories. The result is that any one event or episode will set up a representation that is very different from other events or episodes, because the set of CA3 neurons activated for each event is random. This is then the optimal situation for the CA3 recurrent collateral effect to operate, because it can then associate together the random set of neurons that are active for a particular event for example, an object in a particular place and later recall the whole set from any part.

It is because the representations in CA3 are unstructured or random, in this way, that large numbers of memories can be stored in the CA3 autoassociation system and that interference between the different memories is kept as low as possible, in that these memories are maximally different from each other Hopfield ; Rolls and Treves ; Treves and Rolls If some stored memory patterns were similar, they would tend to interfere with each other during recall.

For an episodic memory, each stored memory pattern should be different from the others, so that each episode can be separately retrieved. The requirement for a small number of mossy fibre connections onto each CA3 neuron applies not only to discrete Treves and Rolls but also to spatial representations and some learning in these connections, whether associative or not, can help to select out the small number of mossy fibres that may be active at any one time in order to chose a set of random neurons in the CA3 Cerasti and Treves Any learning may help by reducing the accuracy required for a particular number of mossy fibre connections to be specified genetically onto each CA3 neuron.

The optimal number of mossy fibres for the best information transfer from dentate granule cells to CA3 cells is in the order of 35—50 Cerasti and Treves ; Treves and Rolls The mossy fibres also make connections useful for feedforward inhibition Acsady et al.

On the basis of these and other points, we predicted that the mossy fibres may be necessary for new learning in the hippocampus but may not be necessary for the recall of existing memories from the hippocampus; existing memories can instead be implemented by the perforant path synapses that come directly from the entorhinal cortex and that make many more connections onto each CA3 neuron and are associatively modifiable Rolls a ; Rolls and Treves ; Treves and Rolls ; see below.

Experimental evidence consistent with this prediction about the role of the mossy fibres in learning has been found in rats with disruption of the dentate granule cells Lassalle et al. We Rolls and Kesner hypothesized that the nonassociative plasticity of mossy fibres i.

This plasticity and the competitive learning in the dentate granule cells would also have the effect that similar fragments of each episode e. This would have potential advantages in terms of economy of use of the CA3 cells in different memories and in making some link between different episodic memories with a common feature, such as the same location in space.

Consistent with this, dentate neurons that fire repeatedly are more effective in activating CA3 neurons Henze et al. As acetylcholine turns down the efficacy of the recurrent collateral synapses between CA3 neurons Giocomo and Hasselmo ; Hasselmo and Sarter ; Hasselmo et al.

If cholinergic activation at the same time facilitated LTP in the recurrent collaterals as it appears to in the neocortex , then cholinergic activation might have a useful double role in facilitating new learning at times of behavioural activation and emotional arousal, when presumably it may be particularly relevant to allocate some of the limited memory capacity to new memories. Acetylcholine may also facilitate memory storage versus recall by enhancing firing in dentate granule cells see Kesner and Rolls By calculating the amount of information that would end up being carried by a CA3 firing pattern produced solely by the perforant path input and by the effect of the recurrent connections i.

On the other hand, an autoassociative memory network needs afferent inputs to apply the retrieval cue to the network. We have shown that the perforant path system is likely to be the one involved in relaying the cues that initiate retrieval in CA3. The concept is that, in order to initiate retrieval, a numerically large input through associatively modified synapses is useful, so that even a partial cue is sufficient and that the retrieval cue need not be very strong, as the recurrent collaterals then take over in the retrieval process Rolls a ; Treves and Rolls In contrast, during storage, strong signals, in the order of millivolts for each synaptic connection, are provided by the mossy fibre inputs to dominate the recurrent collateral activations, so that the new pattern of CA3 cell firing can be stored in the CA3 recurrent collateral connections Rolls a ; Treves and Rolls The associatively modified synapses required in the perforant path to CA3 synapses make this a pattern association network.

The architecture and properties of pattern association networks are described briefly in Supplementary Material Box 2 and in more depth elsewhere Rolls a. These synapses need to be modified during the storage of an event memory, with the entorhinal input to CA3 becoming associated with whatever subset of neurons in CA3 is firing at that time Treves and Rolls Randomness sometimes referred to as noise is present in the spiking times of individual neurons, i.

The noise arises from synaptic and neuronal processes in ion channels, the quantal release of transmitter, etc. Faisal et al. A result is that, in an autoassociation network, if one population of neurons for one attractor or memory state has by chance more spikes from its neurons than the other populations, then the memory with more spikes is more likely to be recalled, especially when the recall cue or cues for the various neurons are relatively similar in strength.

The operation of such systems has been described in The Noisy Brain: Stochastic Dynamics as a Principle of Brain Science Rolls and Deco in the context of decision-making Wang but the approach applies equally to memory recall in an autoassociation memory, as the network architecture and operation for memory and for decision-making is the same Rolls a.

This noisy operation of the brain has been proposed to have many advantages, for example, in promoting the recall of different memories or different associations on different occasions, even when the inputs are similar and this is proposed to be an important contributor to original thought and creativity Rolls a ; Rolls and Deco Too much noise and therefore the instability of memory and decision systems might promote unstable attention and loose thought associations in schizophrenia Loh et al.

Too little noise and therefore too much stability may contribute to some of the symptoms of obsessive-compulsive disorder Rolls b ; Rolls et al. In both these cases, the combination of theoretical neuroscience approaches with experimental evidence concerning the transmitters present in these states is leading to interesting new approaches to understanding these disorders and perhaps to treating them more successfully Rolls b , a , together with the cognitive effects in normal aging Rolls and Deco We now turn to the hypothesis that the dentate granule cell stage of hippocampal processing, which precedes the CA3 stage, acts as a competitive network in a number of ways to produce, during learning, the sparse yet efficient i.

The properties of competitive networks are summarized in Supplementary Material Box 3 and in more detail by Rolls , a. An important property for episodic memory is that the dentate, by acting in this way, performs pattern separation or orthogonalization; Rolls b , a ; Rolls et al. The term pattern separation refers to the property that the output patterns are less correlated with each other than the input patterns, i.

As just described, the dentate granule cells might be important in helping to build and prepare spatial representations for the CA3 network.

The actual representation of space in the primate hippocampus includes a representation of spatial view E. Wirth in preparation; Rolls and Xiang , whereas in the rat hippocampus, it is of the place where the rat is. However, the spatial representations in the rat and primate could arise from essentially the same computational process as follows de Araujo et al.

The starting assumption is that, in both the rat and the primate, the dentate granule cells and the CA3 and CA1 pyramidal cells respond to combinations of the inputs received. In the case of the primate, a combination of visual features in the environment will result, because of the fovea providing high spatial resolution over a typical viewing angle of perhaps 10—20 degrees, in the formation of a spatial view cell, the effective trigger for which will thus be a combination of visual features within a relatively small part of space.

In contrast, in the rat, given the very extensive visual field that is subtended by the rodent retina and that may extend over — degrees, a combination of visual features formed over such a wide visual angle would effectively define a position in space that is a place de Araujo et al. The entorhinal cortex contains grid cells that have a high firing rate in the rat in a two-dimensional 2D spatial grid as the rat traverses an environment, with larger grid spacings in the ventral entorhinal cortex Moser et al.

This may be a system optimized for path integration McNaughton et al. How are the grid cell representations, which would not be suitable for the association of an object or reward with a place to form an episodic memory, transformed into a place representation that would be appropriate for this type of episodic memory? I have proposed that this might be implemented by a competitive network Rolls a in the dentate gyrus operating to form place cells and implemented by each dentate granule cell learning to respond to particular combinations of entorhinal cortex cells firing, where each combination effectively specifies a place; this has been shown to be feasible computationally Rolls et al.

Results of this competitive learning model are illustrated in Fig. Similar processes are involved in some later models of this transformation Giocomo et al. Simulation of competitive learning in the dentate gyrus to produce place cells from the entorhinal cortex grid cell inputs. The colours show the firing rates with blue being the lowest and red the highest in the test environment e.

After Rolls et al. In primates, there is now evidence for the presence of a grid-cell like representation in the entorhinal cortex, with neurons having grid-like firing as the monkey moves its eyes across a spatial scene Buffalo ; Killian et al.

Spatial view cells in primates represent a scene view allocentrically, as described below. How could such spatial view representations be formed in which the relative spatial position of features in a scene is encoded? I have proposed that this involves competitive learning analogous to that used to form place cells in rats but, in primates, operating on the representations of objects that reach the hippocampus from the inferior temporal visual cortex Rolls et al.

In this theory, it is the spatial asymmetry with respect to the fovea of different neurons that solves the binding problem, for the neurons indeed respond to an object and to its location with respect to the fovea Aggelopoulos and Rolls ; Rolls ; Rolls et al.

Another input to hippocampal spatial view cells may come from the parahippocampal place area Nasr et al. The associative modifiability in this connection helps the full information present in CA3 to be retrieved in the CA1 neurons Rolls ; Schultz and Rolls ; Treves ; Treves and Rolls Part of the hypothesis is that the various sub-parts of an episodic memory, which have to be represented separately in CA3 to allow for completion, can be combined together by competitive learning in CA1 to produce an efficient retrieval representation for the recall via the backprojection pathways to the neocortex of memories stored in the neocortex Rolls b , a ; Treves and Rolls The need for information to be retrieved from the hippocampus to affect other brain areas was noted in the Introduction.

The way in which this could be implemented via backprojections to the neocortex is now considered. The CA1 neurons would then activate, via their termination in the deep layers of the entorhinal cortex, at least the pyramidal cells in the deep layers of the entorhinal cortex see Fig. These entorhinal cortex layer 5 neurons would then, by virtue of their backprojections Lavenex and Amaral ; Witter et al.

The areas of neocortex in which this recall would be produced could include multimodal cortical areas e. The backprojections, by recalling previous episodic events, could provide information useful to the neocortex in the building of new representations in the multimodal and unimodal association cortical areas, which by building new long-term and structured representations can be considered as a form of memory consolidation Rolls a , b , c , a , b , a , or in organizing actions.

The hypothesis of the architecture whereby this multistage recall from the hippocampus to the neocortex is achieved is shown in Fig. The feedforward connections from association areas of the cerebral neocortex solid lines in Fig. The backprojections allow for divergence back to neocortical areas. The way in which I suggest that the backprojection synapses are set up to have the appropriate strengths for recall is as follows Kesner and Rolls ; Rolls a , b , c , a.

During the setting up of a new episodic memory, strong feedforward activity progresses towards the hippocampus. During the episode, the CA3 synapses are modified and, via the CA1 neurons and the subiculum, a pattern of activity is produced on the backprojecting synapses to the entorhinal cortex. Here, the backprojecting synapses from active backprojection axons onto pyramidal cells, being activated by the forward inputs to entorhinal cortex, are associatively modified. A similar process would be implemented at preceding stages of the neocortex, i.

The concept is that, during the learning of an episodic memory, cortical pyramidal cells in at least one of the stages would be driven by forward inputs from earlier cortical areas but would simultaneously receive backprojected activity indirectly from the hippocampus. This activity would, by pattern association from the backprojecting synapses to the cortical pyramidal cells, become associated with whichever cortical cells were being made to fire by the forward inputs.

Then, later on, during recall, a recall cue from perhaps another part of the neocortex might reach CA3, where the firing during the original episode would be completed. The resulting backprojecting activity would then, as a result of the pattern association learned previously in the hippocampo-cortical backprojections, bring back the firing in any cortical area that was present during the original episode.

Thus, retrieval involves the reinstating of the neuronal activity that was present in different cortical areas and that was present during the learning of an episode. The pattern association is also called heteroassociation in order to contrast it with autoassociation. The pattern association operates at multiple stages in the backprojection pathway, as is made evident in Fig. If the recall cue was an object, this might result in the recall of the neocortical firing that represented the place in which that object had been seen previously.

As noted elsewhere in this review and by McClelland et al. Overall, this is thus a theory of the way that different events, linked together in CA3 during the formation of an episodic memory, could produce completion in CA3 if only one of those events is presented later in recall.

This would then in turn via CA1 address, by multistage pattern association, the cortical areas in which activity was present during the original learning of the episodic memory and would reinstate the neocortical neuronal activity that was present when the episodic memory was formed.

This theory is supported by a computational neuroscience model of the operation of the whole of this system Rolls A plausible requirement for a successful hippocampo-directed recall operation is that the signal generated from the hippocampally retrieved pattern of activity and carried backwards towards the neocortex remains undegraded when compared with the noise attributable, at each stage, to the interference effects caused by the concurrent storage of other patterns of activity on the same backprojecting synaptic systems.

If p is equal to the number of memories held in the hippocampal memory, it is limited by the retrieval capacity of the CA3 network, p max. The above requirement is very strong: even if representations were to remain as sparse as they are in CA3, which is unlikely, to avoid degrading the signal, C HBP should be as large as C RC , i. If then C HBP has to be of the same order as C RC , one is led to a very definite conclusion: a mechanism of the type envisaged here could not possibly rely on a set of monosynaptic CA3-to-neocortex backprojections.

This would imply that, to make a sufficient number of synapses on each of the vast number of neocortical cells, each cell in CA3 has to generate a disproportionate number of synapses i. The required divergence can be kept within reasonable limits only by assuming that the backprojecting system is polysynaptic i.

The theory of recall by the backprojections thus provides a quantitative account of why any neocortical area has as many backprojection as forward projection connections. Further aspects of the operation of the backprojecting systems are described elsewhere Rolls a. The theory described by McClelland et al. The particular model on which they focus for the learning of semantic representations by interleaved learning is the connectionist model of Rumelhart Rumelhart ; Rumelhart and Todd , which is trained by error backpropagation Rumelhart et al.

For some time, evidence has been available that the hippocampus plays a role in temporal order memory, perhaps for a sequence of spatial locations but also even when there is no spatial component Kesner and Rolls In humans, the hippocampus becomes activated when the temporal order of events is being processed Lehn et al.

One approach regarding the way that the hippocampus might be involved in temporal order memory is by encoding temporal order into each gamma cycle nested into a theta cycle Lisman and Buzsaki ; Lisman and Redish A very different approach is to use firing rate encoding in attractor networks Rolls ; Rolls and Deco and is based on evidence that neurons in the rat hippocampus have firing rates that reflect which temporal part of the task is current Macdonald et al.

In particular, a sequence of different neurons is activated at successive times during a time delay period. The tasks used include an object-odour paired associate non-spatial task with a 10 s delay period between the visual stimulus and the odour.

The evidence also shows that a large proportion of hippocampal neurons fire in relation to individual events in a sequence being remembered e. These interesting neurophysiological findings indicate that rate encoding is being used to encode time, i.

Eichenbaum ; Macdonald et al. These findings suggest several possible computational processes Kesner and Rolls ; Rolls First, because some neurons fire at different times in a trial of a temporal order memory task or delay task, the time in a trial at which an object e.

This would allow associations for the time at which the object was present to be formed. Given that time encoding neurons are also found in the medial entorhinal cortex Kraus et al. However, although lesions of CA3 impair temporal order-place representations, it is lesions of CA1 that impair temporal order-visual object and temporal order-odour representations Kesner and Rolls Thus, temporal timing and object information is possibly brought together by competitive learning in CA1 Kesner and Rolls , which receives inputs not only from CA3 but also directly from the entorhinal cortex see Fig.

Second, these associations would provide the basis for the recall of the object from the time in a trial or vice versa. The retrieval of object or temporal information from each other would occur in CA3 in a way that is analogous to that shown for recalling the object from the place or, vice versa, the place from the object Rolls et al.

Alternatively, if competitive learning in CA1 is the mechanism, generalization in the competitive learning Rolls a from either the object or the temporal order cue would retrieve the whole representation. In addition, if the time encoding neurons simply cycled through their normal sequence during recall, this would enable the sequence of objects or events associated with each subset of time encoding neurons to be recalled correctly in the order in which they were presented.

Third, we need a theory with respect to the origin of the temporal effect, whereby different hippocampal or potentially prefrontal cortex neurons fire in different parts of a trial or delay period. All utilize slow transitions between attractor states that can be a property of noisy attractor networks.

The first hypothesis is that an attractor network with realistic dynamics modelled at the integrate-and-fire level with a dynamical implementation of the neuronal membrane and synaptic current dynamics and with synaptic or neuronal adaptation can implement a sequence memory, as shown by Deco and Rolls The hypothesis is that there are several different attractors and that weak connections exist between the different attractors.

In the model, adaptation produces effects whereby, whatever sequence order of stimuli is presented in an individual trial, that order can be replayed in the same sequence, because as one attractor state dies as a result of the adaptation, the next attractor to emerge from the spontaneous firing because of the spiking-related noise is the one that has been active least recently and is the one that is least adapted Deco and Rolls The whole system operates at a rather slow timescale for the transitions between the attractors, partly because of the time for the noise to drive the system from one attractor state to another and partly because of the slow time course of the adaptation Deco and Rolls ; Rolls and Deco This implements a type of order memory.

The second hypothesis is analogous and is also implemented in a recurrently connected system such as the hippocampal CA3 system or local recurrent circuits in the neocortex Rolls and Deco This second theory is that, again, there are several attractors but that each attractor is connected by slightly stronger forward than reverse synaptic weights to the next.

In previous work, we have shown that, with an integrate-and-fire implementation with spiking noise, this allows slow transitions from one attractor state to the next Deco et al.

The third hypothesis is that the mechanism for the time encoding neurons lies in the entorhinal cortex where there are ring attractors, as described below. The possibility that the recurrent collateral connections in, for example, CA3 could be used to store long sequences by employing discrete timesteps Cheng seems implausible, for an important property of attractor networks is that when implemented with integrate-and-fire neurons, the dynamics become continuous and the whole attractor network settles very fast into its basin of attraction, in 1.

Temporal order memory has been suggested to be implemented in the hippocampus as described above and might make an important contribution to episodic memory in which several events linked in the correct order might form an episode. The theory shows how items in a particular temporal order could be separated from each other, a property that we have referred to as the temporal pattern separation effect Kesner and Rolls For the order to be correctly implemented in the semantic neocortical store, a similar mechanism involving, for example, stronger forward than reverse synaptic weights between long-term memory representations in attractors might build an appropriate long-term order memory Rolls and Deco The entorhinal cortex contains grid cells that have a high firing rate in the rat in a 2D spatial grid as the rat traverses an environment see Fig.

Computational approaches to this system model it as a set of linked ring continuous attractors Giocomo et al. These are the CANNs described above.

The concept is that, as the rat locomotes, the peak of the firing in the continuous attractor moves and, after a certain distance has been navigated, the place represented returns to the same set of neurons, completing the ring. The position of the peak in the ring continuous attractor is updated, for example, by self-motion or possibly by time for at least some neurons.

By having different ring attractors that cover large to small distances with one pass through the ring, the system provides, with its multiscale representation, information that, when read out, appropriately provides a coarse and fine representation of position. The phases of the different ring attractors must be locked for this to work. The use of ring attractors could, in this way, implement a representation of the position of the rat in a 2D environment; this representation would be self-generating and so would work in any environment, if it is updated by self-motion or time.

Indeed, one theory of the underlying mechanism is that neuronal or synaptic adaptation could be used to make the continuous attractor move its peak of activity continuously round the ring as a function of time Kropff and Treves A fast adaptation mechanism would produce small rings for the grid, whereas a slow adaptation mechanism would produce large rings for the grid.

Part of the interest in this suggestion is that grid cells formed by using this adaptation process would effectively be time cells, different cells of which would fire at different times in a trial, as have now been described in the rat entorhinal cortex Kraus et al.

A set of various modelling approaches for the grid cells have been described by Giocomo et al. The system may be used therefore not only for spatial path integration McNaughton et al. A fundamental question about the function of the hippocampus in rodents and primates including humans is whether the hippocampus is for memory or navigation. Strong emphasis is placed on navigation as a function of rodent place cells Burgess et al. In one approach to the function of the hippocampal system in rodents, attractor dynamics for path integration have been suggested to be implemented in the entorhinal cortex for which the evidence is good; Giocomo et al.

External inputs are then held to learn to link correctly onto the appropriate part of this preconfigured map Colgin et al. According to this spatial map theory of the rat hippocampus, there would be no episodic learning of associations between objects and places in hippocampal networks such as CA3 for episodic or event memory and no attractor dynamics within the hippocampus. The discovery of hippocampal cells that respond first to one location and then to another in an ambiguous visual environment is usually however taken as evidence that attractor dynamics exist within the hippocampus Jezek et al.

The purely spatial navigation approach to hippocampal function is also inconsistent with the presence of object-related information in the hippocampus, with object-place association information in the primate hippocampus, with the evidence in rats indicating that one-trial object-place associations are hippocampus-dependent Day et al.

Spatial information is almost always part of an episodic memory and thus spatial representations in the hippocampus may be useful for navigation. For example, episodic memories of particular journeys could help to build neocortical maps that would require many journeys to elaborate. Such maps may be found in the neocortex, given the evidence that lesions to the neocortex can produce topographical agnosia and the inability to navigate Kolb and Whishaw Further, the right hippocampus in humans is activated during mental navigation in recently learned but not highly familiar environments Hirshhorn et al.

Mental navigation in familiar environments activates cortical areas, such as the lateral temporal cortex, posterior parahippocampal cortex, lingual gyrus and precuneus Hirshhorn et al. Given these data, a consideration of the role of the hippocampus in navigation is of interest. First, any model of navigation based on place cells in rodents cannot provide an adequate model of the role of the primate hippocampal cortex in navigation, in view of the presence of spatial view cells in primates, which by their firing provide a basis for the representation of places other than where an individual is located, i.

Spatial view cells provide a basis for the representation of scenes, landmarks in scenes and locations of objects and rewards in scenes Rolls a ; Rolls and Xiang This type of representation is likely to be crucial in primates, including humans, for computations involved in navigating to new places in which the individual has not been located previously.

Moreover, spatial view neurons are found not only in CA3 and CA1 but also in the parahippocampal cortex Rolls and Xiang Second, hippocampal place cells in rodents and spatial view cells in primates can be updated by idiothetic self-motion inputs, for example, by moving the eyes to a different location in a scene in the dark Robertson et al.

The basis for this is the idiothetic update of attractor networks of grid cells on the entorhinal cortex Giocomo et al. These processes may employ head direction cells found in the presubiculum of rodents and primates Robertson et al. To summarize, the evidence described in this review indicates that the hippocampus is involved in episodic unstructured memory by utilizing a single attractor network in CA3 for one-trial object-place and related associations, that the dentate system prepares the inputs for storage by performing pattern separation and that the backprojections to the neocortex are used for memory retrieval.

This system might be useful in navigation, at least in new environments where episodic information may be helpful. In addition, a system of attractor networks exists in the entorhinal cortex for path integration, which may be of value for idiothetic navigation and for idiothetic update of the place being represented in the hippocampal memory system.

A useful theory should make predictions that can then be tested to substantiate the theory or to show ways in which it should be developed or modified. This section illustrates the important and rich interplay that occurs between theory and experiment, which is essential for understanding the manner in which the brain computes. Further developments have been described Kesner and Rolls The theory predicts that the dentate granule cell mossy fibre system of inputs to the CA3 neurons is necessary to store spatial memories but not to recall them Rolls a ; Treves and Rolls , Lassalle et al.

The theory predicts that pattern separation is performed by the dentate granule cells. Evidence consistent with this has been found neurophysiologically in the small sparsely encoded place fields of dentate neurons Jung and McNaughton ; Leutgeb and Leutgeb and their reflection in CA3 neurons Leutgeb and Leutgeb Selective dentate lesions in rats Gilbert and Kesner ; Gilbert et al.

If adult neurogenesis in the dentate gyrus Clelland et al. Consistent with the dentate spatial pattern separation hypothesis Rolls b , c , ; Treves and Rolls , , in mice with impaired dentate neurogenesis, spatial learning in a delayed non-matching-to-place task in the radial arm maze is impaired for arms that are presented with little separation, whereas no deficit is observed when the arms are presented farther apart Clelland et al. Consistently, impaired neurogenesis in the dentate also produces a deficit for small spatial separations in an associative object-in-place task Aimone and Gage ; Clelland et al.

In other cortical systems, the synapses involved in storage and recall are the same and are associatively modified e. The theory predicts that the direct perforant path input from the entorhinal cortex to the CA3 cells which bypasses the dentate granule cells is involved in the recall of memory from the CA3 system. Lee and Kesner obtained evidence consistent with this in a Hebb-Williams maze recall task by showing that lesions of the perforant path impair retrieval Lee and Kesner Much evidence has been gained from subregion analyses involving the disruption of CA3 showing that CA3 is necessary for arbitrary associations between places and objects or rewards Gilbert and Kesner ; Kesner and Rolls Similar impairments have been obtained following the deletion of CA3 NMDA receptors in mice in the acquisition of an odour-context paired associate learning task Rajji et al.

The theory predicts that the CA3 is especially important in object-place or reward-place completion tasks in which associations must be completed from a part of the whole. The theory predicts that the CA3 system is especially needed in rapid one-trial object-place learning and recall. In subregion studies, Kesner and colleagues have shown that CA3 lesions produce chance performance on a one-trial object-place recall task Kesner et al.

For example, CA3 lesions produce chance performance on both a one-trial object-place recall and a place-object recall task Kesner et al. This is evidence that CA3 supports arbitrary associations and episodic memory based on one-trial learning.

A control fixed visual conditional-to-place task with the same delay is not impaired, showing that it is recall after one-trial or rapid, episodic learning that is impaired Kesner et al. As described above , we have shown that primate hippocampal CA3 neurons reflect the computational processes necessary for one-trial object-place event memory, used as a model for episodic memory Rolls and Xiang Place cells Hartley et al.

Another type of test of the autoassociation or attractor hypothesis for CA3 has been to train rats in various environments, e. Evidence consistent with the hypothesis has been found Wills et al. In a particularly dramatic example, Jezek et al. This is an indication, predicted by Rolls and Treves , that autoassociative memory recall can take place sufficiently rapidly to be complete within one theta cycle ms and that theta cycles might provide a mechanism for a fresh retrieval process to occur after a reset caused by the inhibitory part of each theta cycle.

Thus, the memory can be updated rapidly to reflect a continuously changing environment and not remain too long in an attractor state. Tests of the theory reveal quantitatively and analytically the way that memories can be retrieved from the hippocampus to the neocortex Treves and Rolls Memory retrieval has been shown, by the simulation of the multistage hippocampal system, including the entorhinal cortex, dentate, CA3 and CA1 and the return to the entorhinal cortex for recall, to be quantitatively realistic Rolls Many further tests of the theory are described elsewhere Kesner et al.

The hippocampal processes described here for primates include recalling objects from spatial view recall cues. Simonides of Ceos lived to tell the story of how, when a banquet hall collapsed in an earthquake, he could identify all the victims by recalling who had been sitting at each place at the table Cicero 55 BC.

This way of remembering items was developed into what has become known as ars memoriae by Roman senators who presented complex legal arguments in speeches that might last a whole day; they achieved this feat by associating each step in their argument with a location in a spatial scene through which their memory could progress from one end to the other during the speech, thus enabling them to recall each item in the correct order Yates Empirical work has demonstrated that the method of loci is efficacious De Beni and Cornoldi ; Moe and De Beni Moreover, the activity of neurons in the human medial temporal lobe has been related to object-place memory and recall Ison et al.

The new theory Rolls is that this type of memory, ars memoriae , is implemented in the CA3 region of the hippocampus in which, in primates, spatial view cells can be found that allow a particular view to be associated with a particular object in an event or episodic memory. Given that the CA3 cells, with their extensive recurrent collateral system connecting different CA3 cells and with their associative synaptic modifiability, form an autoassociation or attractor network, the spatial view cells with their approximately Gaussian view fields become linked in a continuous attractor network.

As the view space is traversed continuously for example, by self-motion or imagined self-motion across the scene , the views are therefore successively recalled in the correct order, with no view missing and with low interference between the items to be recalled. Given that each spatial view has been associated with a different discrete item, the items are recalled in the correct order, with none missing. The theory provides a foundation for understanding the implementation of the key feature of ars memoriae , namely the ability to use a spatial scene to encode a sequence of items to be remembered Rolls A summary and clarification of where memories are stored in the hippocampal system and the roles of spatial representations in the theory described here might be helpful at this point.

The theory is that the CA3 receives spatial and object information and can bring such information together by CA3-CA3 associative synaptic modification. Because this is a fast learning process, taking place in one trial, it is an unstructured memory about a particular event or episode and not a structured semantic memory.

During storage, at least at one stage of the backprojection pathway to neocortex after CA3, associative learning between the backprojected information and the incoming information would occur to enable the correct neocortical representations in, for example, object or spatial cortical areas to be retrieved.

To facilitate the latter retrieval, CA1 may then remap the separate parts of an event memory to a single representation with the parts no longer separate for the whole memory by using competitive learning in order later to provide an efficient recall cue Kesner and Rolls ; Rolls a.

The dentate granule cells may operate as a competitive network to contribute to pattern separation before the CA3 cells and may use this mechanism to remap grid cells to place or spatial view cells. The connectivity from the dentate granule cells to the CA3 cells via the mossy fibres has a low probability of connectivity for contributing to pattern separation in CA3.

Because the main function of the dentate to CA3 synapses is pattern separation and not information storage, these synapses are not associatively modifiable and, therefore, the neurogenesis of dentate granule cells can help pattern separation. This is the storage process. Recall takes place in CA3 when a partial retrieval cue is applied, for example, the place, so that the whole memory is recalled by completion in the CA3 autoassociation or attractor network.

The object information reaches the hippocampus from the inferior temporal visual cortex via the perirhinal and lateral entorhinal cortex. Reward information reaches the hippocampus from the orbitofrontal cortex and amygdala via the perirhinal and entorhinal cortex. Spatial information reaches the hippocampus from the parietal cortex including the precuneus and also the posterior cingulate and retrosplenial cortex via the parahippocampal gyrus areas TF and TH and medial entorhinal cortex.

The entorhinal cortex grid cell system has multiple attractors that perform idiothetic update path integration in the dark. In rodents, the spatial information is primarily about the place in which the rodent is located.