The homogenous hippocampus: How hippocampal cells process available and potential goals

We present here a view of the firing patterns of hippocampal cells that is contrary, both functionally and anatomically, to conventional wisdom. We argue that the hippocampus responds to efference copies of goals encoded elsewhere; and that it uses these to detect and resolve conflict or interference between goals in general. While goals can involve space, hippocampal cells do not encode spatial (or other special types of) memory, as such. We also argue that the transverse circuits of the hippocampus operate in an essentially homogeneous way along its length. The apparently different functions of different parts (e.g. memory retrieval versus anxiety) result from the different (situational/motivational) inputs on which those parts perform the same fundamental computational operations. On this view, the key role of the hippocampus is the iterative adjustment, via Papez-like circuits, of synaptic weights in cell assemblies elsewhere.


Introduction
The hippocampus is an essentially unique structure in that it can be removed from the brain using blunt dissection and cut into transverse slices where the circuits, like the seeds of a banana, all appear the same.This within-slice circuitry is similar in rats and humans.But, the septal (dorsal rodent /posterior primate ) and temporal (ventral rodent /anterior pri- mate ) hippocampus are often seen as functionally distinct."The dominant view is that the [septal] hippocampus is implicated in memory and spatial navigation and the [temporal] hippocampus mediates anxietyrelated behaviours.However, this 'dichotomy view' may need revision.Gene expression studies demonstrate multiple functional domains along the hippocampal long axis, which often exhibit sharply demarcated borders.By contrast, anatomical studies and electrophysiological recordings in rodents suggest that the long axis is organized along a gradient" (Strange et al., 2014).
Here, we suggest, that the computations carried out by "slices" throughout the septo-temporal extent of the hippocampus are fundamentally the same at the algorithm level.They all compare aspects of available goals ("just checking") and generate functional output when those aspects are incompatible ("control mode").These aspects are sent to the hippocampus as efference copies of available goal representations.These representations are distributed between the dorsal (situation) and ventral (motivation) trends of the cortex and are sent in parallel to the septal and temporal hippocampus, respectively.Thus, septal hippocampus will detect discrepancies between goals that are primarily based on their situation (e.g.place) and temporal hippocampus will detect discrepancies that are based primarily on their motivation (e.g.attraction/repulsion controlling approach/avoidance, respectively).Computationally, then, the hippocampus can be seen as homogenous.
In what follows, we review data on the relationship of hippocampal cell firing to behaviour and the environment.We suggest that what are often called "place fields" should instead be viewed as "goal fields"; and we note problems with the cell "field" concept.We provide reminders of older data on eye blink conditioning and the orienting reflex that are not easily amenable to high level cognitive explanations; and we review data on "remapping", "splitting", and "flickering" that link to motivation and the future more easily than to the non-emotional episodic present.Finally, we present a model of homogenous hippocampal function throughout its septo-temporal extent that we believe is consistent with the reviewed data.

The place cell perspective
Cells in the hippocampus usually fire when the rat is in a particular place (O'Keefe and Dostrovsky, 1971).At first sight, it can seem obvious from this that they encode the place in which the rat is currently located.They appear to be part of the brain's GPS (Burgess, 2014;Moser and Moser, 2016).Likewise, the greater the experience of London taxi drivers with navigation about the city, the larger their septal (dorsal rodent /posterior primate ) hippocampus becomes (although notably also the smaller their temporal (ventral rodent /anterior primate ) hippocampus becomes) (Maguire et al., 2000).Together with the amnesia that appears to result when damage to the temporal lobe includes the hippocampus, this has historically been taken to suggest that hippocampal cells encode spatial memoriesand that the hippocampus contains an ever-expanding spatial map.
But the hippocampus is not purely spatial.Even 40 years ago, the first large scale theory based on hippocampal cell "place fields" was more general.It saw "The Hippocampus as a Cognitive Map" (O' Keefe and Nadel, 1978).Reduced hippocampal size in healthy elderly people relates to poor memory for non-spatial items (Golomb et al., 1996) and temporal lobe damage in humans affects non-spatial memory as much as spatial.More recently observation of single cell fields has led to the idea of hippocampal cell assemblies coding non-spatial items (Wood et al., 1999) including personal identity, e.g., of Jennifer Aniston (Quian Quiroga et al., 2005), as well as social hierarchies; with the suggestion that "the hippocampus organizes social information into a map-like format" (Schafer and Schiller, 2020, p 27).For this, it appears that area CA2 uses tuned structured geometries that differ depending on the social entity, allowing low dimensional generalization with novel conspecifics and high dimension memory storage with littermates (Boyle et al., 2024).
Thus, while the spatial map view continues to be incredibly influential, it has become increasingly clear in recent years that a role for the hippocampus in spatial memory fails to capture everything that the hippocampus is doing, leading to various alternative theories of hippocampal function that necessitate hippocampal representations that go beyond the purely spatial domain (e.g., Buzsáki and Llinás, 2017;Eichenbaum, 2013;Kelemen and Fenton, 2016;Manns et al., 2007;McKenzie et al., 2014;Morris, 2006;Redish, 1999;Samsonovich and McNaughton, 1997;Stachenfeld et al., 2017;Sugar and Moser, 2019;Whittington et al., 2020).These theories (and many others not cited) are notable for their variety and their general focus on memory and/or hippocampal cell firing (e.g., Barry et al., 2006).We do not feel we have space to compare the similarities and differences in the predictions they would make across the range of tasks that would differentiate them.However, we would emphasise that our view of the hippocampus differs in terms of being non-spatial and non-memorial (including hippocampal lesion effects on innate behaviours), while nonetheless accounting for spatial and memorial deficits.We also differ from many of these theories in accounting for the wide range of tasks where, after hippocampal lesions, active learning (including place learning, Whishaw, 1998) is intact but unlearning and inhibitory learning are dysfunctional (see, e. g., Gray and McNaughton, 1983) and for the strong resultant parallels between the effects of septo-hippocampal lesions and those of anxiolytic drugs (for a recent update, see McNaughton and Gray, 2024, Table 4.2, pp. 96-98).
Much of the work we will discuss below records only from single cells, but we would emphasise Hebb's (1949) view that information is coded in cell assemblies (see Park et al., 2011, for discussion of hippocampal ensemble place coding that is incompatible with a dedicated code; see Yuste et al., 2024, for an overview of "neuronal ensembles" and their relation to "cell assemblies" and "neuronal assemblies").An important point for our view of hippocampal processing is that such cell assemblies can be distributed between dorsal and ventral processing trends to encode different aspects of goals.
Both the original spatial map theory and the more recent expansions raise an important question about the size of the human hippocampus.It has remained relatively very small (see footnote 1 in §1.4) while the capacity of the human neocortex to encode information has become very large.How can such a small structure be such a large memory store, even if only a temporary one?If a large structure is required to code basic items how can a miniscule one code all the factorially more numerous relationships between many such items?A second difficulty for the spatial map theory is that, even after half a century of study, it is unclear how the firing patterns of hippocampal cells can uniquely code the cognitive constructs of episodic memory in general, or even places in particular.Even occasional "remapping" of subsets of currently active cells would be incompatible with a true map, whether spatial or social; and it is now universally recognised that "remapping" is a common feature of hippocampal cell activity (Bevandić et al., 2021;Kubie et al., 2020;Latuske et al., 2018;Low et al., 2021;Sanders et al., 2020;Schuette et al., 2020;Widloski and Foster, 2022, see also Section 7).
We should make clear, here, what we mean by a true (or conventional) map.With, e.g., a conventional country map, a single marker locates a single place and adjacent markers are in adjacent places.However, to take a set of object markers and deduce a central location requires only that the inputs to the decoding system be located in spacenot that the marker itself identifies a particular point in space.Indeed, the fact that a computational model can decode space more accurately with multiple place fields necessitates that a single field (marker) does not accurately represent a single location.Thus, the individual fields are not locations on a map and they represent inputs from which a map could be constructed by wherever is decoding the field informationwhich would be outside the hippocampus.We suspect the same result (capacity to locate in space with greater accuracy given more cell fields) would be true with recordings from a random selection of brain areas (particularly visual).

The available goal perspective
Here we argue, instead, that hippocampal cells receive information directly from the environment or in terms of their potential actions via memory about available goals rather than places or events per se.The presence of a single such goal will activate the relevant hippocampal cell assembly but, in this situation, absent any other competing representations, the hippocampus will be "just checking" and not produce any functional output.Given more than one such goal representation, the purpose of consequent hippocampal processing is to compare concurrent available or potential goals to determine any incompatibility.This could be incompatibility either (i) of situation as in turn left versus turn right options for a single motivation, or (ii) of motivation as in approachavoidance of a single location.If incompatibility is detected then the hippocampus enters "control mode" and sends output to other structures to resolve conflict between competing goals or interference between their recall.It is important to note, here, that excitotoxic medial septal or dentate gyrus lesions that impair hippocampal function, including spatial learning, leave hippocampal "place" fields intact (Leutgeb and Mizumori, 1999;McNaughton et al., 1989) consistent with the place fields mirroring input to the hippocampus rather than reflecting functional output after hippocampal processing.
We define a goal as a compound of a situationoften a "place" in rodent experimentsand a motivation.This particular use of the term "goal" has been described in detail previously (see §1.9 of Gray and McNaughton, 2000, particularly pp. 23-24).
"The most important feature of the term 'goal' is that it can be distinguished from both 'stimulus' and 'response' in their pure sense and represents a necessary conflation of the two.The goal of the rat running down the runway has both a stimulus (in this case, place) component and a response (the animal's tendency to run towards it) N. McNaughton and D. Bannerman or motivational component.Take either of these aspects away and it would cease to be a goal … We hold that the hippocampus deals only with conflicts between goals (approach-avoidance conflict in nominal tests of anxiety; S+ approach/S-approach conflict in nominal tests of memory) and not with conflicts between responses or motor acts as such.Thus, in the learning of mirror drawing there is considerable conflict between old and new muscular patterns required to achieve the goal; but there is only one goal and so no goal conflict, and there is motor system rather than hippocampal involvement.That is why someone like H. M., who lacks a hippocampus, can learn mirror drawing as well as anyone else."Thus, for the rat in the runway experiment the goal is "I need to move towards the end of the alley because there is food there and I am hungry."In many instances such a goal will be based on the memory of a previous experience in the same (or a similar) situation (e.g."last time I was in this place I moved towards the end of the alley and I received a food reward.").However, in some situations, goals can be based on innate drives such as foraging (i.e.explore in order to find something rewarding like food or water) or hard-wired avoidance responses (e.g.avoid open exposed places because there is a higher risk of predation).Hence, during unconditioned anxiety tests like the elevated plus maze, the competing goals are "explore the open arms as they are potentially rewarding" and "avoid the open arms they are potentially dangerous." In the pure memory domain, for example during episodic memory retrieval, a particular stimulus (or situation) might result in the retrieval of two overlapping or competing memories (which can be considered as available goals) only one of which is correct and therefore should be retrieved in the current situation.For example, "chocolate cake reminds me of my 12th birthday party which we celebrated in the restaurant in town," and "chocolate cake reminds me of my 13th birthday which we celebrated at my grandparents' house."In this case, there is now a conflict or overlap between the two goal memories that are evoked by the retrieval cue (i.e."on my 12th birthday I celebrated in the restaurant by eating chocolate cake" and "on my 13th birthday I celebrated at my grandparent's house by eating chocolate cake"), which we argue would require the hippocampus to select the correct memory for retrieval.
In spatial navigation, there is an inherent conflict at the level of how to respond to specific individual stimuli in the environment.For example, if a rodent starts at position X in an open field arena, in order to get to the reward location, the correct response might be to head towards the left of cue A located in the centre of the open field.However, if the animal starts from position Y on the opposite side of the arena, then now the correct response is to move towards the right of cue A. Thus, the available goals of "head towards the left of cue A to get food" and "head towards the right of cue A to get food" are in conflict with each other and thus require resolving by the hippocampus.Note, that the motivational component of the goal does not necessarily need to be directly associated with a physical reward per se.It could simply be the desire to move to position Z, for example during spatial exploration to investigate the possibility of a potential reward or it could reflect secondary reinforcement of a subgoal in a chain leading to a main goal.
Multiple cell assemblies coding potential available goals (detected in the environment or via memory) will be concurrently activated with the representation of what many would term "the goal" being the one that is currently in command of the motor system.(See §2 for details of our definition of a goal that allows a goal to have either positive or negative valence.) Critically, while firing patterns may suggest that hippocampal cell assemblies have "goal fields"; we argue that this does not involve encoding (in the strong sense) of any information.We argue that that the cells' firing patterns are efference copiesmirroring encoding in other parts of the brain of the key components of the goal; with the information that they mirror changing with ongoing circumstances.In many cases, this primary encoding will be in cell assemblies distributed between the dorsal and ventral trends of the neocortex; with the currently activated cell assemblies determined by particular configurations of environmental properties (determining situation and motivation).We echo Hebb's view that memories are stored in the neocortical areas that encode the equivalent immediate perceptions and cognitions (Hebb, 1949); and combine this cortical view of memory with Papez's view that the hippocampus is a key component of circuits controlling emotion (McNaughton and Vann, 2022;Papez, 1937).
Within this framework, the essential role for the hippocampus in detecting and resolving conflict between available goals can manifest following hippocampal lesions as either (i) a memory impairment in situations that generate high levels of interference due to the retrieval of overlapping or competing memory traces, or (ii) altered anxiety as a result of the potential threat associated with an ambiguous predictor of danger; and with the decision as to whether to approach or avoid a potentially dangerous situation.Indeed, these two scenarios merge to become one and the same when we consider conditioned anxiety, for example, as with contextual freezing.In 1992 Phillips and LeDoux (1992) reported the finding that hippocampal lesions will reduce freezing levels to the background context but have no effect on freezing to a punctate, foreground cue such as an auditory tone that reliably predicts the imminent delivery of a foot shock, a finding that has since been widely replicated.Many have interpreted deficits in contextual freezing as reflecting an inability of hippocampal lesioned animals to form a complex, multimodal representation of the environment (so requiring a form of spatial memory).But we suggest that the hippocampal deficit in contextual freezing reflects an inability to resolve the conflict generated by the presence of multiple contextual cues that have become associated with both shock and the absence of shock (i.e.there is the "potential" threat of shock in the context, but the threat is uncertain and not imminent).
Contextual freezing appears then to reflect conditioned anxiety and, notably, contextual fear conditioning (and rat strains selected for high contextual freezing) is being increasingly considered as a model of "generalised anxiety disorder" (Cruz et al., 2023;Lages, Maisonnette, Marinho, et al., 2021;Lages, Maisonnette, Rosseti, et al., 2021).Crucially, hippocampal deficits in conditioned freezing levels are also seen with non-spatial, punctate auditory or visual cues, which are ambiguous predictors of shock (e.g., the tone is followed by shock on only 50 % of trials; Glover et al., 2017; see also Tsetsenis et al., 2007), consistent with our hypothesis that it is goal conflict rather than the inherent complexity of the cue (e.g.spatial or contextual cues) that is crucial for hippocampal functional involvement (but not for place fields as such).
Thus, by framing hippocampal function in terms of the resolution of goal conflict we can potentially explain the apparently diverse effects of hippocampal lesions across a wide variety of different behavioural tasks.Tasks that assess not only spatial memory but also non-spatial memory and anxiety.By identifying such an algorithm or operation, we can begin to explain how the anatomically "homogenous hippocampus" can participate in what, on the face of it, would appear to be such disparate behavioursinvolving conflicts of situation, motivation, or both.

The role of anxiety and behavioural inhibition
As alluded to above, an account based on goal conflict processing rather than spatial memory also fits the facts that hippocampal lesions reduce unconditioned anxiety in ethological tests that generate an approach/avoidance conflictincluding the elevated plus maze, food hyponeophagia and novelty suppressed feeding tests, and tests of social interaction between rodents that are unknown to each other (e.g., Deacon et al., 2001).Importantly, these conflict-related deficits on ethological tests of anxiety can be entirely dissociated from any spatial memory deficits.Septal-but not temporal-hippocampal lesions impair spatial memory performance on tasks like the Morris water maze or radial maze (Bannerman et al., 1999;Moser et al., 1995;Pothuizen et al., N. McNaughton and D. Bannerman 2004); but the reverse is true for a battery of ethological anxiety tests (Bannerman et al., 2002;Bannerman et al., 2003;Bannerman et al., 2004;Kjelstrup et al., 2002;McHugh et al., 2004).Why should a deficit in spatial navigation (e.g., an impaired GPS) alter levels of social interaction with an unfamiliar conspecific; or why should a deficit in anxiety in the elevated plus maze (usually in experimentally naïve animals that have spent their entire lives in their home cages in the colony holding room) be considered a memory impairment?What prior experiences could have altered their behaviour?Instead, an inability to resolve the goal conflict that arises between the innate drive to explore and forage in novel and salient environments and the innate aversion to open, exposed places (i.e., an increased risk of predation) likely underpins the effects of hippocampal lesions on ethological anxiety tests like the elevated plus maze (Bannerman et al., 2004;Deacon et al., 2001) or predator defence (Blanchard and Blanchard, 1972;Blanchard et al., 1970;Pentkowski et al., 2006).
In sum, we believe that hippocampal cell assemblies, distributed between the septal and temporal hippocampus, only temporarily mirror situational (including spatial) information that is encoded elsewhere and they always combine this situational information with motivational/emotional information that, also, is encoded elsewhere.However, while claiming that the neocortex is the location of all stages of goal representation and memory for neocortically processed information, and claiming that hippocampal cell firing is motivationally loaded, we still assign the hippocampus a fundamental role that impacts on memory retrieval, updating and consolidation.We argue that hippocampal cell firing is an initial step in processes that can prevent "catastrophic hypermnesia" (McNaughton and Wickens, 2003).That is, hippocampal processing leads to the suppression of the incorrect retrieval of interfering competing memories to which simple associative networks are prone.
We argue that this memorial function of the hippocampus is an extension of a non-memorial role in suppressing responding to inappropriate competing innate goals while leaving responses to noncompeting goals intact.Hippocampal output thus leads only indirectly to the formation of correct memories and only in some cases.This memory-related inhibition is a specific case of hippocampal involvement in behavioural inhibition and inhibitory learning.Control of behavioural inhibition was the predominant theory of hippocampal function prior to the spatial/cognitive mapping hypothesis (Altman et al., 1973;Douglas, 1967;Kimble, 1969;Kimble et al., 1966;Niki, 1965;Simonov, 1974;Winocur and Mills, 1969) and it is worth reminding ourselves of the key data.The important point is that for a wide range of cases hippocampal lesions do not impair learning of a response but do impair inhibition of that response once it is learned (for detailed review of the behavioural effects of hippocampal lesions, see Gray and McNaughton, 1983; for detailed update see http://dx.doi.org/10.17605/OSF.IO/YW32K).Some examples are shown in Table 1.

The cell field problem
Cell field descriptors must be treated with caution.To describe the firing of a hippocampal cell in terms of a "place field" may be appropriate but only if the descriptor is read as a colloquial form of observational shorthand.Early work on the visual system described the firing of cells in terms of "line" and "edge" fields (for review see Frisby, 1979).Such descriptions made the observed firing patterns easy to describe and to link to observed events.However, an important aspect of this kind of "feature detector" description is that it does not map to implied computational function (Sakurai, 1996b(Sakurai, , 1999)).For example, in the visual system "the labels 'edge detector', 'slit detector' and 'line detector' should not be taken too literally.They are convenient terms in that they describe the kinds of input features that optimally excite the simple cells: but they must not be taken any further than that" (Frisby, 1979, p 45).A cell with a conventional "edge field" has too coarse a tuning to be capable of detecting edges and is in fact, from a computational point of view, actually a preliminary filter for higher order mechanisms that may instead detect, for example, lines (Frisby, 1979).Likewise, variation in "place fields" with variation in parameters of the spatial environment shows only that, somewhere in the brain, place is being encodedit does not show that the hippocampal cell is where that encoding occurs.
Hippocampal unit discharges have been shown to be significantly correlated with structural components of every paradigm ever tested, including sleep (Pavlides and Winson, 1989)!But what can be made of such results?… By their very ubiquitousness, these discharges could not be considered as evidence for specific participation of the hippocampus … hippocampal units have behavioural correlates for tasks that can be performed even in the absence of an intact hippocampal system.(Wiener, 1996, p 338-339) Indeed, even place learning itself can survive hippocampal lesions with deficits in the place learning task attributable to a failure to resolve ambiguities (Whishaw, 1998).Similarly, while excitotoxic lesions centred on the medial septum appeared to reduce place cell "remapping" (Leutgeb and Mizumori, 1999), neither they nor dentate gyrus lesions (McNaughton et al., 1989) change location-specificity of hippocampal cell place fields but do produce significant spatial memory deficits.
The spatial map analogy, however, has proved so compelling that when the data are not consistent with a fundamental map, people talk about "remapping" (which can be spatially inconsistent, see below)rather than replacing the "place field" descriptor with something more accurate.In what follows, we will therefore focus on data that do not fit neatly into a spatial map frame.We will assume that changes in hippocampal "place fields" reflect changes in other areas of the brain that are undertaking the actual spatial encoding (which may, in any case, only be correlated with space) -and likewise for other types of hippocampal field.Importantly, the idea that hippocampal fields are reflections of available goal representations elsewhere accounts for the presence of place fields in place-oriented tasks while explaining why the very same cells can show "time fields" in time-oriented tasks (Young and McNaughton, 2000) (see also Colgin, 2020).
In what follows we will attempt to provide a single answer to a range of apparently disparate questions.
• How can a "field" that changes its spatial position when there is no change in the spatial aspects of its environment retain its encoding of space or memory?• If human memory is more complex than that of many other animals (with human cognition apparently requiring a massive expansion of neocortex) why is the human hippocampus relatively smallbeing

Table 1
The role of behavioural inhibition in sensitivity to hippocampal lesions.Tasks are shown in pairs where the primary difference is whether behavioural inhibition is required or not.Data extracted from McNaughton and Gray (2024), Table 4.2).only fractionally larger relative to human subcortex phylogenetically? 1 • How can the septal (dorsal rodent /posterior primate ) hippocampus appear to be dedicated to memory and the temporal (ventral/anterior) hippocampus appear to be dedicated to emotion when slices taken from any level of the hippocampus from a rat or human do not differ fundamentally in either primary circuitry or gross electrophysiology when tested in isolation in a tissue chamber?What is the common algorithm that the slices from different hippocampal levels would execute?• What have the very strong "eye blink fields" in the hippocampus got to do with memory?
Our single answer to all these questions has two aspects.First, hippocampal cell fields signal quite specifically and simply the activation by the environment and by its associations of available or potential goal representations.So, before proceeding to a review of the data, we will make clear what we mean by a goal representation, distinguishing it from higher order components of planning.In particular, we will argue that what we call goal representations in the brain can be both positive (generating attraction) and negative (generating repulsion).This departs from those meanings of "goal" that imply solely attraction.It treats approach-avoidance conflict as the result of the situational coincidence of two available goals that have opposite motivations.
Second, hippocampal cell assemblies do not encode or signal specific unique goals.They mirror neural activities elsewhere, of which they receive efference copies.This has two consequences: a) each hippocampal cell can appear to "encode" many different things, when in fact it is simply signalling that the motor system has a vague general class of intention (thus there is a many-cortex to onehippocampus mapping of connections); b) dorsal (/septal/posterior) and ventral (/temporal/anterior) hippocampus reflect different (dorsal/where and ventral/what) trend information that can be linked to the same goal, i.e., the hippocampal information is distributed among streams in the same way as cortical information.That is, a single goal-mirroring ensemble in the hippocampus will involve clusters of cells in both the septal and temporal portions.
Critically, a goal has a distributed representation in the neocortex and therefore different parts of the hippocampus code different aspects of a single goal.So, in the conclusion of our review of the data, we will spell out how distributed hippocampal circuitry processes neocorticallyencoded available goal information.

What is an available goal?
We often explain actions in terms of attempts to achieve a goal.In the spatial view of hippocampal function "routes imply goals which imply motivations" (O' Keefe and Nadel, 1978, p 83).But actions can also be explained by "taxes", tendencies to move along a proximal gradient (McFarland, 1987) of, say, light intensity (see, e.g., McNaughton et al., 2016, Fig. 1).Taxon explanations 2 cannot be used in relation to complex distal stimuli such as visual form (Hinde, 1966;McFarland, 1987) these embody goals and do so most clearly when absent from the current immediate environment and present only in memory.Goals, as explanations, are also internal representations.They should not be confused with the external object or event that activates the goal representation.
The internal representation of a goal has two, key, compounded, properties: "situation" and "motivation".In a typical spatial task, the situation is a place, while the motivation reflects the presence or memory of a reinforcer (e.g.food) coupled with appropriate drive (e.g.hunger).Note that a situation without the motivation would not constitute a goal.Situations can involve time and more complex cognitive constructs that can support conditioning such as stimulus configurations and dynamic patterns.Importantly, when a rat was passively transported to the location of its "place field", a cell did not fire if the rat was only allowed head movements and sniffing but showed appropriate firing if they could extend their limbs and make small shifts of posture but not locomote (Foster et al., 1989).
In our terms, activation of a goal representation generally reflects the presence of an available goalor its inference ( §7.7).In most situations there will be many available goals activated concurrently.Often, a winner-takes-all mechanism can determine "the goal" currently driving behaviour.But note that concurrent activation can affect arousal even when it does not impact on decision (Gray and Smith, 1969).However, when two or more available goals are in balanced conflict, a winner-takes-all mechanism cannot direct action; so conflict resolution is required and the hippocampus becomes involved.
A highly important feature of our particular use of "goal" to mean a situation-motivation compound is that the motivation component can be positive or negative.Rats dislike bright light and prefer the dark.A particular dark corner of a box will be "the goal", a point attractor, for a rat placed in any part of a surrounding bright area (Fig. 1A).Now suppose, instead, that we put the rat in the bright corner of an otherwise dark box (Fig. 1B).This will act as a point repulsor in the same way as the dark corner of Fig. 1A acts as a point attractor.That is, the rat's movements will follow the straight lines shown in Fig. 1 (provided there is only one available goal in the apparatus).Importantly, absent any other information, we can determine whether the goal is positive or negative from whether the rat's paths on multiple occasions converge (to an attractor) or diverge (from a repulsor).But, other than this directionality, driven by valence, the representations of the two types of goal, and the way they affect action, can have an identical neural form (for a demonstration of negative place fields, see Calvin et al., 2024).
Another important feature of a "goal" as an explanation is that it implies appropriate immediate changes in behaviour with changes in the environment or in bodily capacity to produce the original responses (see McNaughton et al., 2016, for a more extended version of the following).This contrasts with a strongly behaviourist S-R theory that "treats all instrumental behaviour as simple, elicited habits.This is certainly not the interpretation that our folk psychology gives for many of our instrumental actions.We typically regard our actions as purposive and explicitly selected and performed because of our knowledge of their beneficial consequences, and there is now good experimental evidence that we share this capacity for goal-directed, instrumental behaviour with other animals."(Balleine and Dickinson, 1998, p. 410) A behaviourist perspective can lead us to talk about an animal 1 The human hippocampus is only 4 times the size of an average insectivore's hippocampus; while the human neocortex is more than 150 times the size of an insectivore neocortex.Hippocampal size increases in step with neocortex in prosimians but not simians and "the wide scattering of the hippocampal indices for similar neocorticalization indicates further that there is no close size relationship between hippocampus and neocortex.Very high indices, up to 3.8, have already been reached in … progressive insectivores with a still poorly developed neocortex" Stephan & Andy, (1969).Quantitative comparative neuroanatomy of primates: an attempt at a phylogenetic interpretation.Annals of the New York Academy of Sciences, 167(1), 370-387.https://doi.org/10.1111/j.1749-6632.1969.tb20457.x.Your hippocampus is actually smaller than that of the aye-aye lemur from Madagascar. 2 As defined here these are much more primitive than the taxon system of O' Keefe, J., & Nadel, L. (1978).The hippocampus as a cognitive map.Clarendon Press.

N. McNaughton and D. Bannerman
producing a "conditioned response" after learningimplying that reinforcement has connected a specific previously unconditioned response to a conditional stimulus.However, a sheep trained to lift its head from a pad that delivers shock will immediately lift its leg, instead, if it is placed so the leg rather than head is on the pad (Cahill et al., 2001).Furthermore, in Miller's classic "Acquired Drive" experiment, rats were capable of learning different instrumental responses (e.g., running through a doorway, turning a wheel, pressing a lever) in order to escape from a section of the apparatus that had been previously associated with shock in order to get to a "safe" section (Miller, 1948).Crucially, the safe section acted not just as a goal for directed behaviour but also supported this new learning in the absence of any further punishing stimuli.
Likewise, when drugs (Towe and Luschei, 1981) or lesions prevent a rat from producing its usual coordinated avoidance response, it will immediately use new movements, such as rolling, that "are not stereotyped, but selected from variable patterns in such a manner as to bring the animal nearer the goal.Furthermore, the new patterns are directly and efficiently substituted without any random activity" (Hinde, 1966).It appears clear that the animal learns about the goal and then, on a particular occasion, selects an appropriate response to the goal's presence; it does not learn a fixed response to a simple stimulus.
At first glance, talk of goals appears very far from the associative memories or predictions, or future scenes and scenarios, or episodic future thinking, that are often seen as particularly hippocampal (see, e. g., Moscovitch et al., 2016).However, our view of the selection of one available goal from a set of competing ones is essentially the same as a view of the hippocampus in memory retrieval as selecting one from many competing retrieved memories for access to conscious recollection (a process that is most obvious with the tip-of-the-tongue phenomenon).That is, we view the rest of the brain as continually retrieving memories, and making them potentially accessible to prefrontal cortex for use in the generation of predictions, and the concatenation of subgoals into plans.The job of the hippocampus is to ride herd on this flock of retrieved items (and related newly synthesised equivalents).It does not use memories to make predictions "in house".The memories, effectively, are multiple competing predictions.If there is more than one equally primed goal/memory/plan (held in prefrontal cortex very often) and these are incompatible then you need the hippocampus (interacting with, say, prefrontal cortex) to eliminate the opposition and allow only one winner to take control of behavioural output.Note that the other available goals must remain latent or primed to allow fast switching if conditions change and it is only their output behaviour that is inhibited.

How does the brain code goals?
An important but, for many, unexpected feature of the brain's coding of information is that it is distributed: different parts of the brain code different aspects of what we think of as a single entity.A key major separation (Barbas and Pandya, 1991) is of information in a dorsal neural trend (placed mediodorsally in the cortex, evolving from primitive hippocampal archicortex, and coding "where" or more generally "situation") from information in a ventral neural trend (placed basoventrally, evolving from olfactory paleocortex, and coding "what" or more generally "motivation").
The visual system shows a particularly obvious dorsal trend/ventral trend separation.In the initial cortical stages of visual processing a ventral stream of structures codes "what" information (vision-for-action; form/colour) and this stream links the visual cortex with the temporal lobe.For example, faces receive specific dedicated processing in the inferotemporal cortex.Conversely, a dorsal stream of structures codes "where" information (vision-for-perception; motion/spatial relationships) and links the visual cortex with the parietal cortex (Bruno et al., 2008;Van Essen and Gallant, 1994).As a result, lesions to various parts of the visual system produce various distinct agnosias rather than all producing more general blindness.This separation of classes of information can lead to oddities such as "the man who mistook his wife for a hat" (Sacks, 1985).
This distributed perceptual processing is maintained in prefrontal cortex -and a large-scale separation of dorsal/where and ventral/what streams of processing is maintained.These streams also meet a third stream of information that is affect-driven and that we have suggested (Gray and McNaughton, 2000) can be seen as a mesial "why" trend in parallel with the dorsal and ventral trends.The three trends are interconnected allowing for some of the resultant cell assemblies (Hebb, 1949) to represent a what/where/why compound as the basis for action.
Frontal cortex also has an important caudal-rostral organisation.Simple sensori-motor acts are selected by isocortical areas posterior to the arcuate sulcus, more complex associative actions are selected by more anterior/rostral proisocortical areas, and goals are selected by more limbic areas (Chambers et al., 2009;Haber and Calzavara, 2009;Haegelen et al., 2009;Humphries and Prescott, 2010).So, we can see goal representations as being anterior/rostral and limbic, and more taxon-like or action-element representations as being more posterior/caudal and sensori-motor.
The hippocampus is connected primarily to more anterior/rostral and limbic parts of prefrontal cortex.Importantly, this connection is most extensive from the final output stage of the hippocampus: the subiculum.The connections of area CA1 (which provides input to the subiculum) are more limited, primarily to very anterior parts of the dorsal and ventral trends.
The connections of the subiculum and CA1, taken together, suggest that the output stages of the hippocampus are more concerned with the selection of goals than the selection of even complex actions.It also suggests that the hippocampus is concerned with both what and where, and particularly why, actions should be performed.
The link with goals and motivation is even clearer with output from area CA3 (Fig. 2).CA3 not only provides input for further processing to area CA1 but also provides a major source of hippocampal output to the subcortex and hypothalamus.For example, CA3 projections to the lateral septum provide access to dopaminergic neurons in the ventral tegmental area (VTA), a key driver of motivated behaviour (Luo et al., 2011).This massive output from the hippocampus through the fimbria/fornix is seldom mentioned in discussion of the role of the hippocampus in memory.The hippocampus as a whole, via septal relays (largely lateral septum), is topographically mapped (Fig. 2D) into hypothalamic areas (Risold and Swanson, 1996) and can, in turn, be seen as relaying topographically mapped information from the cortex (Fig. 3).
We have used the term "relay" in relation to Fig. 3 to emphasise the topographic organization of the pathways involved.However, at each step there is processing and integration of information.Thus, in the same way as the hippocampus carries out operations on its inputs such as "just checking" and can generate functional outputs as appropriate (i.e."control mode"), the lateral septum should be seen as an important area for further processing of hippocampal output.
"The lateral septum (LS) has been implicated in a wide variety of functions, including emotional, motivational, and spatial behavior, and the LS may regulate interactions between the hippocampus and other regions that mediate goal directed behavior.….We suggest that the lateral septum incorporates movement into the evaluation of environmental context with respect to motivation, anxiety, and reward to output an 'integrated movement value signal'.Specifically, hippocampally-derived contextual information may be combined with reinforcement or motivational information in the LS to inform task-relevant decisions."Wirtshafter and Wilson (2021, p. 544) As reviewed by Wirtshafter and Wilson (2021), LS cells have place-like fields (particularly in the dorsal part); and LS retains the hippocampal dorsal-ventral trend topography for its inputs.But the cells also correlate with reward, affect, and movementderived from input it receives from other structures.They conclude (p.554) that, "given the convergence of mood, movement, and motivation information in the LS system, it is worth considering whether the LS is a central nexus in the brain for integrating these disparate inputs.… We suggest that many of the functions attributed to the LS, including the modulation of anxiety, may instead be related to the LS's function integrating movement signals with different cues and contexts, and that septal damage therefore results in situationally inappropriate movement responses, which appear as heightened or lessened anxiety, aggression, or perseveration.The integration of these factors into a single firing-rate code in the LS may allow for downstream [hypothalamic]structures to weigh the costs and benefits of action and to optimize task performance across a conjunction of these factors, particularly in complex and ethologically relevant environments with multiple considerations." We would see the hippocampus, then, monitoring available goals, detecting cases of conflict, and passing this "contextual" information to LS to integrate with mood, movement and motivation and thence to hypothalamus for cost-benefit analysis to engage the mechanisms required to resolve the conflict.

Goal processing and interpretation of hippocampal cell fields
Both the compound nature and the distributed representation of available goals have implications for how we talk about hippocampal cell fields.As we have noted previously (Gray and McNaughton, 2000, Appendix 6), (1) goal fields will reflect the animal's view of its world (which we must determine experimentally) and will not necessarily map to the experimenter's intentions.(2) The same class of cell may appear to have a stimulus dominated field in one situation and a response dominated one in another situation.Both cases can be goals and even be the same goal (which is neither a stimulus nor a response as such but guides Fig. 2. Hippocampal outflow. A. 3-D schematic of location of the hippocampus as a bridge between cortex and subcortex in the rat brain (adapted from the human equivalent in Gastaut and Lammers, 1961).There is important outflow (arrows) both via the subiculum to the temporal cortex and via CA3 and the fimbria to the hypothalamus.B. Diagram of the hippocampal cell fields (courtesy of S. Kerr) within a slice (or functional lamella).C. The hippocampus can be viewed as a stack of essentially identical slices (Andersen et al., 1971;Andersen et al., 2000) with the circuitry of each slice as shown in B. D. The septal-temporal axis and these fields can be viewed, unfolded, as a simple flat map that is topographically connected (via the septum) to the hypothalamus (Risold and Swanson, 1996).Note that this topographic variation in connections from the hippocampus is matched by similar topographic variation in inputs to the hippocampus from both specific prefrontal (see Fig. 3) and entorhinal inputs, and also more diffuse neuromodulatory inputs.It is also mirrored by variation in neuromodulatory influence and hormonal regulation along the septotemporal hippocampal axis (Lathe, 2001; see also Strange et al., 2014).Thus, common circuits will execute a common algorithm but operating on different forms of information, on potentially different scales of time and space, and particularly with different emotion-related functions.to the many different responses that achieve a particular end).It follows that (3) we may sometimes wish to use stimulus-and sometimes response-loaded terms to describe our observations of hippocampal fields.
To talk of a goal allows, simultaneously, a stimulus and a response orientation.
"For example, in an approach avoidance conflict with an electrified water spout, it is usually easiest to describe the situation in terms of the responses to be made (approach, avoidance) than specific stimuli to be approached (since in the avoidance case there is only a single stimulus to avoid, but all of the apparatus, except the spout is safe).But it does no violence to normal usage to refer to approach and avoidance, respectively, as the subject's alternative goals.Likewise, in a discrimination task, it is usually easiest to describe the situation in terms of the stimuli to be responded to.But again, conventional usage allows us to refer to the left lever and the right lever as being the subject's alternative goals."(Gray and McNaughton, 2000, Appendix 6).
There are other aspects of hippocampal "goal fields" that we will have to consider in our interpretation of themtheir relation to hippocampal function, anticipation, changing situationsbut all these are best considered in the context of specific experimental data.It is to these we now turn.

Eye blink conditioning
It may seem surprising to start the detailed portion of a review that contrasts spatial with goal interpretations of hippocampal cell fields with a discussion of eye blink conditioning.However, the data on hippocampal cell firing in relation to conditioned eye blinks are as, or more, compelling than for any other type of experiment.
Eye blink conditioning is one of the simplest available conditioning paradigms and is carried out with the subject remaining stationary.A stimulus (e.g.tone or light) is presented and is followed shortly after by a puff of air to the eye, causing a blinkan essentially unidimensional response from fully open through to fully closed.A sensor over the eye can record the trajectory of the blink.After a number of trials, the conditional stimulus (CS) elicits a conditioned eye blink response in advance of the air puff.This sounds like a very mechanical process but "neurotic introverts usually condition faster and extinguish slower than other people [consistent with] susceptibility to fear and particularly its induction by punishment" (McNaughton & Corr, 2019, p. 122-123).
There are two common variations on this paradigm that are important for assessing hippocampal function.One is "delay" conditioning.Here, the CS duration is expanded so that its onset occurs some seconds before the delayed air puff, which is delivered at the time of CS offset.A conditioned blink to the onset of the CS would fail to avoid the punishment of the air puff, which would occur after the blink.Under these circumstances, animals including humans, learn to delay production of the blink until the CS terminates (i.e., at the time of air puff delivery).The second variation is "trace" conditioning.Here, the CS terminates some time before the air puff.Now the animal has only time as a guide to when to produce the conditioned blink response.Animals including humans can learn this form of blink conditioning too.Importantly, the CS in trace conditioning sets up the expectation of the air puff and so primes a blink response.But the trace period itself is, in external Fig. 3. Some selected structures and connections focussing on the role of the supramammillary area (SuM).SuM connects bidirectionally with rostral limbic cortex including the infralimbic (IL) and anterior cingulate (AC) cortex.As described in the text this cortex is topographically mapped into the medial septum and topographic mapping continues via the hippocampus, lateral septum and hypothalamus, as shown.(Note that dorsomedial septum is shown below the ventrolateral septum to simplify the topographic mapping.)The mapping is illustrated for only one dimension in each structure but is, in fact, two-dimensional (with respect to flat sheet representations of each level of the system).We postulate that the most extreme strands of this topographically organised system are specialised for more cognitive (AC, septal pole of hippocampus) and more emotional (IL, temporal pole of the hippocampus) processing-but with each representing goals and thus encoding conjunctions of cognitive and emotional features.This would be consistent with previous independent suggestions about the partitioning of function in the cortical structures (Heidbreder and Groenewegen, 2003) and the hippocampus (Bannerman et al., 2002;Bannerman et al., 2004;Kjelstrup et al., 2002;McHugh et al., 2004;McHugh et al., 2011;Richmond et al., 1999).It would also be consistent with, for example, the greater input from the amygdala to temporal portions of the hippocampus.In this descending scheme, the supramammillary area (SuM) is just one of an array of hypothalamic structures receiving topographically mapped input, presumed to be involved in the control of goal-related behaviour (Risold and Swanson, 1996).The direct connections of the cortex to the hypothalamus appear to show the same mapping as the relayed connections.It is likely that the topographic mapping of the various parts of the system extends to their connections with the central grey (CG).In contrast to the descending connections the ascending connections of SuM are quite diffuse and passage of information around these recurrent circuits could, we postulate, control cognitive-emotional interactions.Figure and legend taken from Pan and McNaughton (2004).
stimulus terms, just like the period before the CS and like the period after the US.So, blinking must be inhibited during the trace period.Thus, in terms of the theoretical framework outlined here, there is a goal conflict between "the tone means an air puff is coming so close the eye" versus "the end of the tone means the air puff is delayed so withhold closing the eye for the expected delay interval." The simplicity of the (unidimensional) response system and of the paradigms used to engage the hippocampus also allows more definitive conclusions than for many other cases.Importantly, we know a great deal about the entire circuit controlling the conditioned response, and much of the means through which hippocampal output can alter it.Although the eye blink is intrinsically simple, even simple conditioning of it as a sensorimotor output involves parallel plasticity in cognitive (hippocampal) and affective (amygdala) circuits, which can then modulate the basic conditioned response (Choi et al., 2001;Stanton, 2000).Eye blink conditioning is, then, a paradigm case that shows us how to interpret the hippocampal cell firing that occurs in more complex experimental situations that have much more complicated behavioural output (e.g., spatial memory tasks).
The first, obvious, point to make about the eye blink data, collected from a stationary subject, is that hippocampal cells do in fact often show conditioned-eye-blink-related responses.The fact that such "fields" occur at all suggests immediately that hippocampal processing in rodents extends far beyond location in space.Conversely, in humans where hippocampal processing clearly encompasses higher order cognition, the hippocampus still shows activation by eye blink conditioning (Cheng et al., 2008).The fact that eye blink fields can be observed relatively easily, coupled with the existence of responses to simple, non-spatial visual and auditory stimuli in habituation experiments (Vinogradova, 1975), and with temporal fields in temporal tasks (Young and McNaughton, 2000), emphasises that "there is something more than a little miraculous about constructing an ad hoc environment, sticking a wire into the middle of the brain and immediately encountering a place field" (Gray and McNaughton, 2000, p145).Like place fields, eye blink fields are clearest in the dorsal hippocampus (Weible et al., 2006).
The second point to make is that lesions of the hippocampus do not affect the basic form of the conditioned eye-blink when it occurs.The detailed circuit for conditioned eye blink (and, e.g.limb flexion) production in rodents and humans bypasses the hippocampus (Berger, Berry, et al., 1986;Blaxton et al., 1996;Logan and Grafton, 1995;Steinmetz, 2000).The hippocampal eye-blink-related activity simply echoes the simple conditioned response.Importantly, a hippocampal response does not occur at all with unconditioned eye blinks.Conditioning connects the previously neutral conditional stimulus to the basic eye blink reflex circuit via plasticity in cerebellum-related circuitry (Knowlton et al., 1988;Thompson, 1986; for review, see Thompson and Steinmetz, 2009).Notably, acquisition, maintenance, and even delay (but not trace, see below) of conditioned eye blinks all survive hippocampal lesions (Berger, Berry, et al., 1986) and the conditioned response even survives decerebration, emphasising its cerebellar nature (Mauk and Thompson, 1987).This is like simple fear conditioning, which bypasses the hippocampus and involves simple thalamic input becoming directly linked within the amygdala to the circuit that produced the unconditioned response (e.g., an undirected escape response) and now produces the conditioned response (e.g., blood pressure increase, freezing, LeDoux, 1994;Phillips andLeDoux, 1992, 1994).It should also be noted that the simple conditioned activity in the hippocampus does not generate similar activity in its primary subcortical target the mammillary bodies (Berger and Thompson, 1978).
The third point to make is that, while it is the conditioned stimulus that elicits the hippocampal activity, the pattern of cell firing can model the unidimensional trajectory of the response, rather than following the trajectory of the stimulus (Alger and Teyler, 1976; but see, Green and Arenos, 2007;Weiss et al., 1996).This hippocampal response disappears after lesions of nucleus interpositus in the cerebellum (Clark et al., 1984;Sears andSteinmetz, 1990, cited by Kim, Clark, andThompson, 1995) suggesting that the hippocampus receives its critical information from the cerebellum rather than remembering a stimulus that it then passes to the cerebellum.Also, prior to conditioning, dentate, CA3 and CA1 cells do not react to what will become the conditional stimulus either on initial presentation (unlike stimuli that produce a clear orienting response, see below) or after further unpaired presentations.In contrast all three areas showed activation during conditioning; and this activation can appear before clear conditioned eye blinks are observed at the behavioural level (Berger et al., 1983).The hippocampal response, then, reflects the compound of a situation (the conditioned stimulus) with a motivation (avoidance of a noxious air puff) that in our terms is a simple goalthere is no hippocampal response to either the situation or the motivation when these are separated.Critically, the hippocampal response is linked to the anticipation of the air puff and so represents the requirement to make some form of response to deal with the anticipated event.With trace conditioning, there is an early "additive neural response of the combined CS-and US-evoked neural activity similar to reflex facilitation or modification" (McEchron andDisterhoft, 1997, p 1037).Likewise, the eye blink can be enhanced by conditioning mediated by the amygdala (Choi et al., 2001;Neufeld and Mintz, 2001).On this view it is only because the response system is unidimensional that this goal representation can, at least sometimes, appear to model the form of the conditioned response.
How can the relatively small hippocampus come pre-wired to transform into a higher order code not only a plethora of neocortical information but, in this case, cerebellar information as well?The simplest answer, compatible with the simplicity of hippocampal size and structure, is that when a hippocampal cell fires it does not provide detailed information about the situation that is activating it.The cells multitask.A specific hippocampal ensemble can be activated by a range of available goals (hence multiple fields and remapping).Importantly, hippocampal firing signals only that a goal is availableno hippocampal output is required.It is only when more than one such signal reaches the hippocampusand their implied concurrent executions are incompatible that the hippocampus will shift from "just checking" to "control" mode.
It is the cerebellum (not the hippocampus) that houses the plasticity that links a particular stimulus to the eye blink response to produce conditioned eye blinks (Knowlton et al., 1988;Thompson, 1986).Importantly, this is true even with delay conditioning where the response is delayed relative to the onset of the stimulus but still occurs before the offset, so they overlap.
Notably, in contrast to simple and delay conditioning, trace eye blink conditioning is impaired by hippocampal lesions (Ivkovich and Stanton, 2001).Consistent with this, in humans, "comparable delay and trace activation was measured in the cerebellum, whereas greater hippocampal activity was detected during trace compared with delay conditioning" (Cheng et al., 2008, p 8108).As with other forms of hippocampal sensitive memory (e.g., the performance of the famous amnesic patient HM in the famous faces test), trace conditioning appears to have a retrograde gradient, becoming insensitive to hippocampal lesions if these are made a month after conditioning.We explain this as a result of consolidation creating "delay lines" (see below) that remove the need for suppression of interference by the hippocampus.Note that such delay lines will operate in parallel with the hippocampus to delay output of the eyeblink response; and that the eyeblink efference copy will continue to be sent to the hippocampus and produce blink-related firing even though the hippocampus is no longer necessary for the trace conditioning task.

"
The key question then is where the permanent memories are stored.Trace conditioning would seem to provide an animal model of this process of memory consolidation that is amenable to analysis because so much is known about the neural substrates of eye blink conditioning.The permanent store could be in the cerebellum, N. McNaughton and D. Bannerman perhaps in the cortex, or in neocortical areas and could be localized or distributed" (Kim et al., 1995, p 201).
The short answer to this question with respect to the eye blink response is that it cannot be the hippocampus and is in fact the cerebellar nucleus interpositus (Thompson and Steinmetz, 2009).But, is it reasonable to think that the hippocampus stores information in relation to trace but not delay conditioning?Why are there only modest differences in the way trace and delay are represented by cell firing (Green and Arenos, 2007; but see Kirsch et al., 2003)?How could the hippocampus transfer a temporary trace memory to nucleus interpositus (Thompson and Steinmetz, 2009) for more permanent storage?The hippocampus also sends output to the caudate but the caudate is involved as early in eye blink conditioning as the hippocampus (Flores and Disterhoft, 2009).
We can view the delay conditioned response as occurring through a direct link with the conditional stimulus and the trace conditioned response as requiring control via a longer "delay line" (see Miller, 1991, especially pp 160-162) that initially involves not only the hippocampus but also anterior cingulate (Weible et al., 2003) and prefrontal cortex (extending perhaps to working/active memory processes) (Berry and Hoffmann, 2011).Initially, premature responding to the start or end of the CS will compete with the correct trace responsewith a strong requirement for hippocampal behavioural inhibition of responding during the CS.However, as these earlier responses to the CS extinguish and the prefrontal generation of the correct trace response strengthens, hippocampal inhibition is required less and less.Consolidation of the prefrontal response over a period of weeks, therefore, accounts for the retrograde gradient of hippocampal involvement (Takehara-Nishiuchi et al., 2006;Takehara et al., 2003).
Thus, these eyeblink conditioning data provide us with a simple model that can also account for results with more complicated paradigms.For example, as mentioned previously ( §1) simple tone-fear conditioning is hippocampus-independent (Phillips and LeDoux, 1992) but if the tone is made an ambiguous predictor of shock then the hippocampus becomes involved (e.g., Glover et al., 2017), as it is in contextual fear-conditioning; which also likely reflects the fact that the context is an ambiguous, uncertain predictor of "potential" shock.As with trace eyeblink conditioning, there is a goal conflict resulting from the ambiguous meaning of the cue (either the tone that is partially associated with shock or the background context in a fear conditioning experiment).
Using the eye blink conditioning paradigm, latent inhibition (Solomon and Moore, 1975), blocking (Solomon, 1977), and reversal of stimulus discrimination between a CS+ and CS-cue (Orr and Berger, 1985) have all been shown to be impaired by non-fibre-sparing hippocampal lesions.At least in the case of reversal learning, hippocampal inhibition of prepotent responding appears to pass from the subiculum, via the retrosplenial cortex, to reach the cerebellar circuitry at the ventral pontine nuclei.
"Bilateral lesions of the retrosplenial cortex are associated with severe deficits in reversal learning that are very similar to those seen after bilateral damage to the hippocampal-subicular cortices.That is, lesions of either the retrosplenial cortex or the hippocampal formation produce no alteration in discrimination learning, but they do disrupt an animal's ability to reverse that discrimination.Moreover, animals with either lesion fail at the reversal phase of the task because of a continued high level of responding to [what was previously the CS+ and is now] the CS-; response rates to the CS+ are equivalent to those of control animals.These results also show that bilateral damage to the retrosplenial cortex produces deficits in reversal learning that are as severe in magnitude as those observed after bilateral hippocampectomy … animals with hippocampal damage display shorter latency and larger amplitude conditioned nictitating membrane responses than did control animals.… The hippocampal system [thus] modifies those conditioned reflexes in a manner that allows the organism to respond to more complex relations among environmental events, such as reversal of a previously learned response or conditional relations among stimuli … through the retrosplenial pontine projection" (Berger, Weikart, et al., 1986, p 804, 806) Although Berger et al do not say so, we might conclude that the crucial way the hippocampus "modifies those conditioned reflexes" is by temporarily or permanently inhibiting the upcoming response.First, there is no need for the hippocampus to encode and produce responsesthese will happen through simple learning in other circuits in any case and they are not sensitive to hippocampal lesions (see also §1.3, Table 1).Second, in both trace conditioning and reversal learning, the important thing is for the hippocampus to inhibit the pre-potent response that would otherwise occur.In the trace conditioning case this inhibition is temporary and bridges the trace interval.In the case of reversal learning of a stimulus discrimination task between a CS+ and CS-cue, the pre-potent response is to the original conditional stimulus (i. e. CS+; which during the reversal phase of the task is now the CS-) and inhibition needs to be permanent.
In contrast to reversal learning, "lesions confined to the retrosplenial cortex do not affect trace conditioning.… [while] lesions of the medial prefrontal cortex in rabbits impair trace eyeblink conditioning … [and] neurons in the medial prefrontal cortex respond early in trace eyeblink conditioning to the CS" (Green and Arenos, 2007, p281).So, in both trace conditioning and reversal learning we can see the hippocampus as immediately controlling response inhibition in parallel with the control by other areas of response acquisition.The hippocampus does not, as conventional memory accounts would suggest, much later pass some kind of previous temporarily stored information to the other areas for permanent storage.
An important point is that, even with the simple eye blink response, hippocampal control (Fig. 4) is coordinated by "theta frequency" synchrony across all of the different brain structures involved (Berry and Hoffmann, 2011).Theta-frequency rhythmicity, in and of itself, has been demonstrated to be functionally crucial in one of the best-accepted tests of hippocampal function, the Morris water maze (N.McNaughton et al., 2006).Hippocampus and cerebellum show strongly phase synchronous theta but this does not predict the rate of learning of trace eye blink conditioning (Wikgren et al., 2010).
The superficially linear picture of connections presented by Fig. 4 could cause one to ignore this important feature of the interactions between hippocampus and prefrontal cortexthe theta rhythmicity characteristic of the hippocampus and to which prefrontal cells (as emphasised in the figure) become entrained.The inter-burst intervals of theta pattern firing are similar to the round-trip times of the loops connecting the hippocampus with frontal cortex (Miller, 1989(Miller, , 1991)), which could contribute to recursive processing of the information involved (Gray and McNaughton, 2000, p. 24-27).Thus, the brief periods of inactivity provided by inhibitory theta pattern inputs provide a method for chunking time into distinct epochs of activity.This chunking would prevent interference between successive iterations of any recursive, real-time processing function (digital recursion pre-solves this problem).But theta does not just chunk information.It binds different brain structures into functional circuits, facilitating their cooperative interactions within a common timeframe.As originally suggested by (Miller, 1989), this timing may even determine which structures are currently interacting as functional circuits.
This section has dealt with one of the simplest behavioural reflex responses and a range of simple conditioning paradigms that modulate it.Perhaps because of its simplicity, the eye blink response makes clear a number of rules that we will find useful as we deal with more complex responses and paradigms: N. McNaughton and D. Bannerman 1.We can expect neural organisation to be hierarchical; with simple quick and dirty circuits dealing with simple immediate responses and higher order layers of various types becoming engaged as the task requirements become more complex.2. The apparent "fields" of hippocampal cells are not directly related to functional output from the hippocampus.They reflect information received by the hippocampus in what can be viewed as "just checking" mode 3 that is then required, together with other information, for functional output ("control" mode) only under certain circumstances.3. The hippocampus appears more involved with inhibitory control ( §1.3), preventing premature responses during trace conditioning and inhibiting prepotent responses during reversal learning, than with the production of responses (this is most obvious with reversal where the old and new response are the same but now attached to the alternative stimulus).4. The hippocampus appears to receive information about sit-uation+motivation compounds (in this case compounds of CS+US) and does not appear to react to a simple neutral situation (CS) or motivation (US) when these are not compounded, except when an orienting reaction is being generated (see §9).In our terms a sit-uation+motivation compound is a goaleven when the goal is a simple anticipatory eyeblink.Likewise, an otherwise simple stimulus can be a goal if it is sufficiently salient and novel or surprising, and so creates the motivation for an orienting response to it (see §9).

Simple discrimination learning and memory
The eye blink experiments have at their heart an extremely simple "reflex" circuit activated in an animal that is usually restrained.As we have emphasised, hippocampal responses are linked to conditioning of the eye blink not its simple reflex production.In this section we focus on experiments that assess hippocampal responses explicitly during the learning of CS+/CS-discriminations and in more complex memory tests.In these experiments the animal is usually free to move and so it is hard to determine how far hippocampal responses are related to specific, unmonitored, behaviours.Nonetheless, the observations in these more complex situations largely parallel the picture we have developed with the eye blink data.
Early reports of hippocampal responses with simple CS+/CSdiscrimination from Olds's laboratory found an initial inhibitory response to stimuli (see discussion of orienting in §9) that, as with eye blink conditioning, became gradually excitatory and anticipatory as learning progressed.
"The size of this anticipatory type of response across a number of experiments could be shown to vary somewhat between anticipation of food versus water (Olds et al., 1969) but not between lever press for food versus lever press for shock avoidance (Fuster and Uyeda, 1971).It was also greater for a CS+ than for a CS- (Hirano et al., 1970;Sideroff and Bindra, 1976).Taken together, these results are all consistent with the idea that hippocampal reactions in area CA1 and CA3 are anticipatory of goals.For example, with the reinforcer case, food and drinking are different classes of goal and so produced different electrophysiological reactions, while in the food/shock case the immediate goal was a lever press in both cases and so produced a similar reaction despite the difference in reinforcer."Fig. 4. Schematic depicting essential and modulatory circuitry for trace eyeblink classical conditioning (EBCC) and activity profiles underlying theta-related behavioral effects.Averaged local field potentials (LFPs) for hippocampus (HIPP), interpositus nucleus (IPN), and Larsell's hemispheric lobule 6 of cerebellar cortex (HVI) show robust, time-locked theta oscillations at 6-7 Hz during the trace and post-US periods for trials triggered in the explicit presence of naturally occurring hippocampal theta (T+).Furthermore, these theta oscillations reset at short latency following the stimulus events such that the perturbation of theta is minimized.In contrast, animals that received trace EBCC trials in the explicit absence of theta (T) display less robust hippocampal theta during the trial period and little to no theta rhythmicity in the cerebellar LFPs.Under optimal hippocampal states, in which theta is present, GABAergic and cholinergic projections from medial septal nucleus (MSN) provide pacemaker inputs to HIPP to regulate theta activity.HIPP theta activity, in turn, is known to modulate prefrontal cortex (PFC) unit firing properties.Contiguity of PFC-driven trace-related activity via lateral pontine nuclei (LPN) with US-related information from inferior olive (IO) within the cerebellum is thought to drive learning.Rapid behavioral acquisition occurs with theta phase-locked rhythmic processing in cerebellum (IPN and HVI) and such patterns are disrupted in animals that learn more slowly.Large and small tick marks below LFP traces indicate onset and offset of conditioned stimulus (CS) and unconditioned stimulus (US), respectively.Additional abbreviations: CF, climbing fibers; CR, conditioned response; MF; mossy fibers; PF, parallel fibers. 3Hence intact place fields after lesions that produce behavioural dysfunction as discussed earlier.
A similar focus on stimuli is present in various memory experiments in monkeys.Importantly, the monkeys are restrained with their heads fixed and so cannot alter their spatial location and so there is a limited capacity for their cell firing to reflect place fields, with neurons instead responsive to spatial view (Robertson et al., 1998;Robertson et al., 1999;Rolls et al., 1997;Rolls et al., 1998).Spatial position relative to the monkey is also kept uncorrelated with reinforcement to ensure that firing differences are related to the differences in presented stimuli.Here hippocampal cell assemblies can also appear to reflect available goals.Unlike inferotemporal cells (which are stimulus specific), hippocampal cells showed stimulus-general incremental (readiness-like) firing during the delay of non-spatial delayed matching to sample task (Colombo and Gross, 1994).Hippocampal cell firing can also react to the sample stimulus, the delay, the test stimulus, and combinations of these (Riches et al., 1991;Watanabe and Niki, 1985).With non-spatial conditional and go/no-go tasks stimulus-linked firing has been shown to be context-dependent (Brown, 1982;Sakurai, 1994) and response-linked (Sakurai, 1990(Sakurai, , 1996a)).While it could be that various sorts of complex field are being observed, across these different experiments, an alternative conclusion is that the fields reflect goals, including sub-goals within a sequence (see Wiener, 1996) or subgoal sequences (if these can be represented by a single engram), encoded by cell assemblies.
It has been suggested that hippocampal cell fields represent relations between stimuli (Cohen and Eichenbaum, 1993;Eichenbaum and Cohen, 1988).That is "hippocampal neurons fire in association with various nonspatial task relevant stimuli and conjunctions of such stimuli … In instrumental paradigms, some CA1 neurons have been observed to fire in association with discriminative stimuli in any sensory modality auditory, visual or olfactory.Others have found CA1 cells to fire in relation to conditioned appetitive movements… [but not] in relation to simple sensory or motor events.[They] reflect higher order relationships, beyond the multimodal processing of the neocortical areas that project to the hippocampal system.… To our way of thinking, this reflects the processing of relationships among the task constrained objects or events with which the animal is confronted, the task defined relevance or significance of those objects or events, and the behavioral responses made under these constraints.This is truly relational processing."(Cohen and Eichenbaum, 1993, pp. 115-118, a large number of citations omitted; see also Young et al., 1994) However, while this seems reasonable in complex tasks, it does not obviously fit with the eye-blink results we have considered where there can be response-related patterning and where firing occurs in relation to blinks independent of the nature of the ongoing conditioning.Nor does it appear to be true with Eichenbaum's own data in an odor discrimination task: "Three major categories of cells were identified: (1) "Cue sampling" cells fired after onset of odor cue sampling.Response magnitude was related to cue valence on both the current and past trials.(2) "Goal approach" cells fired prior to arrival at either the odor sampling port or reward cup.A number of sampling and approach cells also had place correlates.However, detailed analyses indicated that specific behaviors associated with increased firing reliably occurred at the same place.Unit activity was at least as well described by behavioral as spatial parameters.(3) "Theta" cells fired at high rates in strict relation to the ongoing limbic theta rhythm."(Eichenbaum et al., 1987, p. 716) The desire, nonetheless, to attribute complex functions to hippocampal cell fields remains strong.In a review covering "five decades of hippocampal place cells", (Colgin, 2020, p. 54) says "Our understanding of place cells has since expanded to include findings showing that hippocampal "complex-spike cells," the term used at the time to describe putative pyramidal cells, also represent many nonspatial aspects of experience.Initial influential studies from the late Bob Muller, his long-time collaborator John Kubie, and other colleagues showed that changing the stimuli associated with an environment (e.g., color, shape, light vs darkness) could change the firing patterns of place cells (Bostock et al., 1991;Muller and Kubie, 1987;Quirk et al., 1990).In 1999, Emma Wood, Paul Dudchenko, and the late Howard Eichenbaum published a groundbreaking study recording responses of complex spike cells in rats performing an olfactory memory task (Wood et al., 1999).They found that cells in hippocampal subregions CA3 and CA1 responded selectively to salient, nonspatial aspects of the animal's experience, including odor and trial condition (i.e., match or nonmatch).Since then, subsequent studies have shown that place cells represent tactile (Gener et al., 2013), auditory (Aronov et al., 2017), and taste (Herzog et al., 2019) information." Note here the use of the term "place cells" for cells that are not coding places and even for cases where the apparent field is both simple in stimulus terms and not in any way linked to place.Indeed, the key conclusion of Aronov et al was that in a sound-control task hippocampal cells mapped the key elements of the task including particular sound frequencies.The Herzog paper is of particular interest in that hippocampal "taste" responses could be unrelated to place and, even when they were related, occurred after taste responses in other brain areas, and appeared linked to the acquisition of hedonic value.Thus, the cells coded the association of place with hedonic value, which we would argue reflects the situation+motivation coding of goals.
Importantly, in addition to the somewhat paradoxical spatial "remapping" that place fields can undergo (see §7), the same cells that show non-spatial fields in non-spatial tasks can show place fields in spatial ones (Wiener et al., 1989).Likewise, Shapiro et al. (1997, p. 624) found with varying "distal stimuli and with distinct local tactile, olfactory and visual cues covering each arm [of a four-arm maze] … [that] different hippocampal neurons encoded individual local and distal cues, relationships among cues within a stimulus set, and the relationship between local and distal cues.Double rotation trials, which maintained stimulus relationships within distal and local cue sets, but altered the relationship between them, often changed the responses of the sample neuronal population and produced new representations.After repeated double rotation trials, the incidence of new representations increased, and the likelihood of a simple rotation with one of the cue sets diminished.Cue scrambling trials, which altered the topological relationship within the local or distal cue set, showed that the cells that followed one set of controlled stimuli responded as often to a single cue as to a constellation.These cells followed the single cue when the stimulus constellation was scrambled, but often continued firing in the same place when the stimulus was removed or switched to respond to other cues.When the maze was surrounded by a new stimulus configuration, all the cells either developed new place fields or stopped firing."All these results appear complex when viewed in simple stimulus terms and, often, when taking responses into account; but are comprehensible in terms of goals, which predict upcoming stimulus-response combinations.

What are "place fields"; how can a place be remapped?
Even in the more complex experiments where place fields are usually demonstrated, close inspection suggests that the cells are in fact coding goals (i.e., upcoming targets of action that the animal "has in mind") and not the specific location in space at which the cell is firing.There are a N. McNaughton and D. Bannerman number of different phenomena that lead to this conclusion.

Conditional place fields
Location can be a necessary but not a sufficient condition for "place cell" firing.Firing can depend on the particular point of view of stimuli (Breese et al., 1989;McNaughton et al., 1983;Robertson et al., 1998;Robertson et al., 1999;Rolls and O'Mara, 1995;Rolls et al., 1997;Rolls et al., 1998); place fields can become absent if reinforcement conditions change (Gothard et al., 1996) or if a transparent barrier is placed across part of the field (Muller and Kubie, 1987); or they can cluster in locations where reinforcement is regularly delivered, e.g., accumulating at the goal location in an annular water maze provided the location is not variable (Hollup et al., 2001).These conditionalities are incompatible with simple spatial coding by place cells.

Multi-place fields
Place fields seem to appear instantly in a new apparatus (Hill, 1978) and yet the same cell can have such fields in each of several different apparatus.A cell can even have multiple "place" fields within a single apparatus with no apparent "organizing or over-riding principles for the position of a single cell's place fields across environments" (Kubie and Ranck, 1983, p. 438), albeit with some retention of spatial and non-spatial structural knowledge (Whittington et al., 2020).These data are incompatible with the idea that a place field literally signals the single place the animal is currently in, which would require each cell to only ever fire in one single place in the universe (and would make such cells very hard to find).They are also incompatible with a signal linked to a single place even within a single circumscribed environment.

Non-local representations and extra-field spiking
A key advance came with computational approaches that allowed decoding of neural activity on much faster time scales than had been used previously (on the order of neural oscillations like theta activity rather than on the order of minutes).In a seminal study, Johnson and Redish (2007) trained rats to navigate different versions of a T-maze task (a multiple-T task and a cued-choice T-maze task) for food rewards whilst recording hippocampal CA3 neuronal activity.In the multiple-T task, the relationship between cues and the location of food rewards was consistent within a day but varied across days.Brief periods of non-local cell firing outside the usual place field ("flickering") were seen at key choice points (see also §11.3 below and Redish, 2016, for review).

Remapping
Not only do "place" fields cluster at locations where reinforcement is delivered, they can then subsequently shift when the location of reward delivery changes.The resulting shift of a cell's largely stable place field from one location to another is usually termed remappingthough it should be noted that the place fields of other cells in the same spatial context often do not change and so it is not that a completely new map is being created for all cells.Indeed, there can be cases where some fields change spatial position while others do not without any change in the spatial organisation of the environment (e.g., unmoving sets of spouts where what is changed is where water is available).With 6 geometrically morphed environments experienced daily over weeks, cells showed differently located, sometimes multiple, place fields across the environments with continuous representational change over weeks but with the relative structure linked to an environment unchanged (Keinath et al., 2022).Allowing for rotation of field structure relative to the environment, about half of a population remain 'stable' between sessions, while the others 'remap'; further, the stable field structures did not reliably use arena or extra-maze cues for their alignment (Kinsky et al., 2018).These data are incompatible with any kind of truly spatial dimensional mapping.

Field-specific remapping
The remapping with a reinforcer change, sometimes called "goaloriented remapping", notably can occur in CA1 with no change in the CA3 subfield (Breese et al., 1989;Dupret et al., 2010;Dupret et al., 2013).Likewise, across trials in a new environment, "place fields in CA1 emerge rapidly but tend to shift backwards from trial-to-trial and remap upon re-exposure to the environment a day later.In contrast, place fields in CA3 emerge gradually but show more stable trial-to-trial and day-to-day dynamics" (Dong et al., 2021, p. 1); and, across dissimilar virtual environments, medial entorhinal cortex cells remap even more than hippocampal cells (Cholvin et al., 2021).Such "re-mapping" is clearly incompatible not only with a spatial map but with any nominally stable form of cognitive map.Conversely, if the fields are seen as reflecting efference copies of goals, their alteration, e.g., when the location of reinforcement is altered, would be expected.In humans, fMRI shows that, during learning of successive goal locations, "map-like representations of the environment emerged in both hippocampus and neocortex.Cognitive maps in hippocampus and orbitofrontal cortices were compressed so that locations cued as goals were coded together in neural state space, and these distortions predicted successful learning.… [So,] goals warp the neural representation of space in macroscopic neural signals."(Muhle-Karbe et al., 2023, p. 1) 7.6.Where is the map?
The usual view of place fields requires them to indicate locations within a topographic map.Perceptual brain areas of all kinds have such map-like (retinotopic, tonotopic, etc.) anatomical organisation.However, historically there has been limited evidence for such a topographic organisation in a typical flat arena of the 2D space within the hippocampus (but see B. L. McNaughton et al., 2006) and, indeed, exposure to an apparatus in the dark can cause non-linear remapping of place fields that is then maintained when the light is switched back on (Quirk et al., 1990).With calcium imaging, it has been shown "that for cells separated by more than a few tens of micrometers, there was no strong relationship in our experimental paradigm between the location of place fields in the virtual reality environment and the position of the corresponding place cells in area CA1.Nearby cells (less than <35 μm separation) showed enhanced correlation" (Dombeck et al., 2010(Dombeck et al., , p. 1433)).Wirtshafter and Disterhoft (2023) replicated the lack of topography, and noted that the local clustering was of cells with place fields close to reward locationsconsistent with the clusters forming goal ensembles.Interestingly, across radically distinct apparatus in which punishment rather than reward is given, there is overlap of activated ensembles if testing in both contexts occurs within a day but much less so with a week separationlinked to the generalization of the conditioning event per se despite the difference in conditioning "context" (Cai et al., 2016).
In many ways the entorhinal cortex has more map-like properties with place-linked activity of cells forming a topographic spatial grid but with variation in both scale and orientation that is different from a conventional surveyors map (Leutgeb et al., 2005).Interestingly, the size of "place fields" appears to vary steadily from under 10 % of an environment at the septal end of the hippocampus to over 80 % at the temporal end (Jung et al., 1994;Kjelstrup et al., 2008); and so the property being "mapped" by this dimension of the hippocampus does not appear to be directly related to spatial location.Note also that comparing rats with monkeys and then humans the temporal hippocampus has expanded much more than the septal hippocampus.Nevertheless, while the temporal portion of the hippocampus appears more linked to motivation than location (Bannerman et al., 2004;Bannerman et al., 2014), even in humans its activation can apparently be strongly linked to spatial processing (Barry et al., 2019).
We suggest that rather than mapping space or even the wider

N. McNaughton and D. Bannerman
physical context of the environment, temporal-end cells are coding functionally for the motivational aspects of situation/motivation compounds, including any emotional component of the response.Adjacent situation components would be likely to be associated with the same motivational component (i.e. the motivation cluster would contribute to more than one "goal" ensemble, see Yuste et al., 2024).As a result, what is observed as an apparently single temporal "place field" will likely be spread around a wider area within the environment and be linked to several distinct "place fields" in septal hippocampus.During contextual fear conditioning, for example, if the animal feels anxious at position X in a context previously paired with shock, it seems highly likely that it will also feel anxious at position Y 30 cm away in the same context.If the ventral hippocampal cell is coding for motivation (e.g. the feeling of anxiety) then it will fire at both spatial locations, and indeed potentially throughout the entire context, generating a much larger place field.
A similar account could also explain why "place field" size often scales with arena size (albeit in different ways and one or both dimensions, O'Keefe and Burgess, 1996).Place fields can scale in size (likely by orders of magnitude) with the scale of their local environment, are larger when landmark cues are absent rather than present, and are smaller at locations near wallspotentially linked to whiskering (see Geva-Sagiv et al., 2015, for review including rat-bat comparisons).For example, in bats, 3D place fields are significantly smaller in the presence of obstacles (Wohlgemuth et al., 2018).In this context, it is worth noting that in rodent experiments cells that have a single moderate-sized field in a smaller rectangular testing box (180 x 120 cm) can have multiple larger and smaller fields irregularly distributed in a very large (530 x 350 cm) space (Harland et al., 2021).Similarly, those with a single small field in a small circular (68 cm diameter) featureless cylinder, had multiple larger, irregularly arranged, fields in a somewhat larger (150 x 140 cm) rectangular space with added features like stairways and water spouts (Fenton et al., 2008).To the external observer (or decoded by some other part of the brain) this could "suggest that the ensemble population of subfields form a multi-scale representation of space within the dorsal hippocampus" (Harland et al., 2021, p. 1).However, an alternative possibility is to invoke activation by efference copies of particular types of goal (e.g.those eliciting whiskering or rearing during exploration, including exploration directed to a single local object cue).Or in the case of ventral hippocampal place fields, they could represent motivational states as a component of a goal representation (e.g.anxiety, hunger, thirst) which transcend the entire arena.

"Remapping as hidden state inference"
Sanders et al. ( 2020) extensively review the wide range of problems with the remapping concept, and the inconsistent results obtained across laboratories when studying remapping.In particular, they note that "change of context" as a cause is a very malleable conceptas indeed is the nature of the remapping that can occur.They conclude that "remapping does not directly reflect objective, observable properties of the environment, but rather subjective beliefs about the hidden state of the environment" (p. 1).The internal nature of our goal concept is the same as their view that a specific "context … doesn't necessarily have to be physically the same environment, as long as the animal infers that it is the same environment" (p.2).The key to their solution to the remapping problem is to frame questions "in terms of hidden state inference: new memories are formed when an animal has inferred that it has encountered an unfamiliar (previously unvisited) state, and old memories are updated when it has inferred that it has encountered a familiar state.As we formalize below, these inferences can be calculated using Bayes' rule, which computes a posterior probability distribution over hidden states by integrating prior beliefs about the hidden states with the likelihood of those hidden states given the animal's observations.The hidden states are sometimes interpreted as latent causes, to emphasize the idea that the animal is forming beliefs about the causal structure of the environment." (p. 3) This perspective is very close to ours; except that it retains the key (very complex) processing in the hippocampus and retains the view that hippocampal cell fields are encoding the critical information.As presented below, we would see any inferences as being generated in other areas of the brain and the key role of the hippocampus as preventing interference between multiple concurrent inferences.

Non-spatial fields
As we have already noted at several points above, cells with place fields in a spatial task can also have non-spatial fields in non-spatial tasks (Wiener et al., 1989).Of course, in the relatively simple, non-navigational experiments that we discussed in previous sections such as eye blink conditioning, the animal was often unable to, or did not critically need to, translate itself from one place to another (i.e., there was no requirement for navigation).Importantly, the predictors of events (particularly reinforcers) are often non-spatial (e.g., time in the case of differential reinforcement of low rates of response, Young and McNaughton, 1997).It is not too surprising that place fields were not generally found in these experiments.But, from a pure place perspective, it is undoubtedly surprising that cells in general had fields and that these were non-spatial.In contrast, from a reinforcer/goal point of view, it is expected.

Non-spatial remapping
Importantly, even in more complex spatial situations, "the functions of place cells extend well beyond a specific role in mapping the physical space.Place cells have been shown to respond to various nonspatial sensory inputs, and to alternate between multiple representations in the same location, reflecting both salient physical properties of the place and events associated with the place, either at present or in the past" (Leutgeb et al., 2005, p 738).Importantly firing in a specific place can vary depending on the upcoming choice and flag whether a specific choice has just been rewarded (Takamiya et al., 2021) as well as switching rapidly between representations of present and past environments (Jezek et al., 2011, see also §11.3).

Goal selective ("splitter") cells 4
A "place field" is nominally determined by firing that occurs in a particular place (such as on a maze arm, including prior to a choice point).However, in apparatus like a T (or, more clearly, "M"/"W") maze a cell can 'split' its activity between trialsfiring at a high rate when traversing one trajectory to one place and at a low rate when traversing another to a different place (for review and use of the term 'splitter', see Dudchenko and Wood, 2014).This is akin to remapping ( §7.4, §7.6) but is stable when linked to the ultimate goal and simply varies from trial to trial, depending on the trajectory being followed, potentially reflecting latent state inferences about the environment, actions, and outcomes (Duvelle et al., 2023, which see also for recent splitter review).Stable firing has also been observed linked to specific laps in a square maze where the rat repeatedly visits the start box but receives reward only on every 4th lap (Sun et al., 2020).In this latter case the cell firing appears to reflect specific subgoals in a sequence (i.e."run the X th lap of the maze" with the ultimate goal of obtaining reward after running 4 laps) and in the absence of any explicit choices between goals or subgoals.
Initial demonstrations of splitters (Ferbinteanu and Shapiro, 2003;Frank et al., 2000;Wood et al., 2000) subsequently appeared to depend on the particular task demands; with later work suggesting that "whether a place cell shows differential firing likely depends, in part, upon whether reward is provided at the start or end of overlapping routes or whether the animal's goal choices are restricted with a barrier during training.Each of these manipulations could affect how an animal solves the alternation task, and this in turn could influence the incidence of splitter activity in the hippocampus" (Dudchenko and Wood, 2014, p. 257).Thus the cells appear to fire in relation to intention not location.It should be noted that simple alternation without any delay (e.g. during continuous running maze paradigms) is not sensitive to hippocampal lesions (Ainge, van der Meer, et al., 2007).Without a delay splitter activity tends to be retrospective (coding where the rat has come from); while, with a delay, performance is sensitive to hippocampal lesions and splitter activity tends to be prospective (coding where the rat is going to).Retrospective activity is thought to appear "more strongly towards the start of the track (where the recent past splits) and prospective activity more strongly closer to the choice point (where the close future splits …)" (Duvelle et al., 2023, p. 20).Splitter fields appear early during learning (albeit later than place fields), show firing at earlier and earlier portions of the start segment within a recording session, but are highly stable and spatially consistent across days (Kinsky et al., 2020).
That prospective splitter cells could reflect the anticipatory activation of goal representations, as such, is shown by a multiple choice point (successive Ys) maze with a single start box leading to 4 reward locations (Ainge, Tamosiunaite, et al., 2007).Each location was consistently rewarded for a block of 10 trials; and then the reward location was switched to a different goal box (balanced across the 4 boxes).Successful choice change was sensitive to hippocampal lesions.Interestingly, "place fields" were observed in much greater numbers at the start box and initial segment than elsewhere in this particular study.Importantly, about half of these cells fired at higher (x 5) rates in advance of journeys to a specific goal box compared to the other three locationswith different cells predicting different goal boxes.Some cells with "place fields" before the second choice point also showed similar single-goal box selectivity.
We argue that the conditional effect of the specific goal box on the cell's firing rate reflects the fact that the hippocampus is receiving an efference copy from anticipatory goal processing areas (i.e."move to goal box X to obtain reward").Similar firing would occur if only one goal box were ever being rewarded but, in "just checking" mode, this would make no contribution to behavioural output; albeit providing the experimenter with information about both position and ultimate goal.Because in this experiment the identity of the rewarded goal box is constantly changing every 10 trials, there is a conflict generated between the currently rewarded and previously rewarded goals, which thus requires the hippocampus to enter "control mode" (hence the task is sensitive to hippocampal lesions).The same appears true of effects linked to "contexts", cue-dependent responding, and the rapid (procedural learning is slow) development of splitter reactions (Dudchenko and Wood, 2014, their §10.6-10.9).
That said, with two alternatives routes to a single goal box (with a transparent barrier forcing a particular route to the same final location for a particular block of trials), the bulk of cells that fire in the start arm have been reported to show route-dependent rather than goaldependent firingthat is they fire in advance of one route only, with different cells firing for different routes, even though both routes lead to the same final goal box (Grieves et al., 2016).However, it should be noted that the design of the task reinforces a particular route (even if it finishes at the same ultimate goal from the experimenter's point of view), with a higher number of errors at choice points where there are two possible routes to the same goal than cases where there is only one correct route to a goal.In our terminology, we might consider the two different routes to the same final reward location as two different subgoals ("go into the left arm at the Y junction to get reward" and "go into the right arm at the Y junction to get reward"), and these two subgoals are in conflict with each other, with the correct subgoal determined by the current block of trials.
Note, this is a variant of the conflict between different goal locations across different trial blocks.The distinction between the two subgoals that reflect taking different routes to the same final, rewarded location emphasises our specific and nuanced definition of a goal as the combination of situation and motivation.Here the two subgoals differ in their situation (left choice arm versus right choice arm) which is defined by the temporal context provided by the current trial block in the same way as the choices between the four goal boxes, even though from the experimenter's point of view the ultimate goal is the same in both cases.So, assuming goal-subgoal scaffolding, it seems likely that the cells are activated (and behaviour controlled) by a route-specific subgoal that delivers 100 % reward as opposed to what the experimenter sees as the main goal (which would deliver 50 % frustration during initial learning).A similar subgoal relationship could underlie the lap-specific firing we mentioned in the first paragraph of this section, where the lapspecific cells fire at a non-rewarded point in a sequence that ultimately results in reward (Sun et al., 2020).
Of particular interest here is the finding of dynamic synchronization with stepping (Joshi et al., 2023).While running to the choice point in the M/W maze, rats' stepping was rhythmic and matched ~8 Hz hippocampal modulation.
"We also discovered precisely timed coordination between the time at which the forelimbs touch the ground ('plant' times of the stepping cycle) and the hippocampal representation of space.Notably, plant times coincide with hippocampal representations that are closest to the actual position of the nose of the rat, whereas between these plant times, the hippocampal representation progresses towards possible future locations.This synchronization was specifically detectable when rats approached spatial decisions.Together, our results reveal a profound and dynamic coordination on a timescale of tens of milliseconds between central cognitive representations and peripheral motor processes.This coordination engages and disengages rapidly in association with cognitive demands and is well suited to support rapid information exchange between cognitive and sensory-motor circuits."Joshi et al. (2023, p. 125) The finding of both retrospective and prospective splitter cells also in entorhinal cortex (Frank et al., 2000;Lipton et al., 2007) is important, here too (see §11.2 for further discussion of entorhinal input to hippocampus).Frank et al. (2000) found that prospective coding appeared more in the deep layers of entorhinal cortex that send output to cortical areas.Lipton et al. (2007, p. 5787) used the continuous T maze and focussed on the shallow layers (that provide input to the hippocampus) and found that entorhinal cells "more strongly distinguished left-turn from right-turn trials compared with CA1 neurons, whereas CA1 neurons more selectivity encoded places [that would be] traversed within each route".Dudchenko and Wood suggested that one possibility for this difference is that activity of "CA1 splitters may result from converging spatial (place) information directly from MEC and trajectory information … regarding memory strategy and task rules [relayed by nucleus reuniens] from medial PFC, as well as current goal information from orbitofrontal cortex" (Dudchenko and Wood, 2014, p. 269).In many respects the orbitofrontal cortex can be seen as a mirror of the hippocampus (Kennedy and Shapiro, 2004) but we would focus on the reverse that the small hippocampus mirrors filtered key elements of information held in the orbitofrontal cortex and other cortical and subcortical areas, including cerebellum.
So far we have looked at cells that tell us where the rat is going but not necessarily why it is going there.It may seem obvious that the conditional start box fields signal the intention to visit reinforced end goal boxes (i.e."goals") but this (as opposed to the simple flagging of a N. McNaughton and D. Bannerman specific upcoming motor action) has not been formally tested in the experiments we have describedalthough differentiation of anticipated places is clear.
In addition to the data that we have just considered there is a large amount of evidence for "goal approach" cells and for an overrepresentation of CA1 "place fields" at goal locations.For example, this is true for probe trials in an annular water maze, where the rewarding escape platform is never visible during either training or in probe trials "analyses were limited to the first minute of probe trials when the platform was unavailable to the rat's senses, implying that the sensory environment was identical for rats trained with the platform at different locations.Despite this fact, the goal location attracted a disproportionate number of firing fields, regardless of its location.The only difference between locations in which fields accumulated and locations in which they did not was whether the animal expected to find the platform there.The fact that place fields were more abundant in the segment preceding the goal than in the succeeding segment further suggests that expectancy may have contributed to the firing in the platform area."(Hollup et al., 2001(Hollup et al., , p. 1641, emphasis added) , emphasis added) Consistent with a strong role for expectancy, "place cells" can fire selectively when a rat is waiting for 2 s at an unmarked location in order to obtain a food pellet that is then dropped anywhere in the apparatus.Hok et al. (2007) trained rats on a place preference task during which they were required to enter an unmarked circumscribed zone in a cylindrical arena and then wait there for at least 2 s which then triggered delivery of reward from an overhead dispenser.As the pellet was dropped from above, it could end up anywhere in the arena and the rat then had to forage for the pellet.The rat also had to wait, without reward, for 3 s before re-entering the goal area which would then deliver a further pellet.With reward effectively occurring at multiple locations within the cylinder, the place cell firing can be attributed to the expectation of reward delivery (or preparation for action) rather then the prior association of a specific spatial location with reward (Hok et al., 2007).In a "honeycomb" maze, where hexagonal platforms can be raised to provide the rat with a choice between either of two steps to a distant goal, rats scan round their current platform allowing the firing of cells to be linked to the rat's heading.It has been shown that a cell's directional firing can converge on a location (a "ConSink") and that, across cells, these tend to cluster in the region of the goal platform (with their average location close to the goal location).With a new goal, Consinks now became centred on the new goal.Such cells could allow "animals to select optimal paths to destinations from any location in the environment" (Ormond and O'Keefe, 2022, p. 741).
There are also increases in firing of entorhinal grid cells close to reward locations and restrucuring of the entorhinal map to incorporate remembered reward locations (Boccara et al., 2019;Butler et al., 2019) that, together with the CA1 data, suggest one is "navigating for reward".In a review paper with this title, Sosa and Giocomo (2021, p. 472) suggested that "The hippocampus and the entorhinal cortexare ideally suited to coordinate this larger network by representing both physical and mental space as a series of states.These states may be linked to reward via neuromodulatory inputs to the hippocampus-entorhinal cortex system.Hippocampal outputs can then broadcast sequences of states to the rest of the brain to store reward associations or to facilitate decision-making, potentially engaging additional value signals downstream".The same will be true (but with opposite behavioural consequences) of punishment associations that depend on different changes in the same neuromodulators (Asaad and Eskandar, 2011;Niv et al., 2012;Schultz and Dickinson, 2000;Stewardson and Sambrook, 2023).
However, as always we would see the key information as being encoded outside the hippocampus and, rather than broadcasting "sequences of states to the rest of the brain to store reward associations", we think the hippocampus interacts with the circuits that normally store such associations to prevent storage and/or current expression of unwanted associations in cases where there is interference and so goal conflict.For example, with ConSinks (Ormond and O'Keefe, 2022), we would see the hippocampal activity as reflecting the activation of the neocortical goal representation and the directionality reflecting the fact that this representation becomes more active when the animal is focussed on the relevant location (or situation in non-spatial tasks).Likewise, with theta-linked "look ahead" sequences (that predict the sequence of the rat's future subgoal locations, Wikenheiser and Redish, 2015) (i.e., "future paths to remembered goals", Pfeiffer, 2022), we would see the theta synchrony across brain regions as important for ensuring stable readout from the cortex to the hippocampus (and vice versa for the results of hippocampal computation).Further, (as with place fields remaining after lesions that produce spatial learning dysfunction) goal-oriented ripple sequences do not change with learning, although their activation can increase (Pfeiffer, 2022).
The key point is that if the hippocampus deals with goals as we are suggesting (related to place or otherwise) then its fields (which are receiving efference copies) will "care" about motivation not just situation and, critically, will reflect the properties of both.This will potentially be relatively more situation-oriented in dorsal (stream) hippocampus and more motivation-oriented in ventral (stream) hippocampus but (especially with entorhinal mixing) will involve both.The differential effects of dorsal and ventral hippocampal lesions reflect the fact that the goal conflict can occur either between different situations with a common motivation (e.g.spatial choices on a maze task looking for reward), or between different motivations in a common situation (e. g. approach or avoid the same alley goal box in which both reward and punishment have previously been experienced).
Notably, motivational state (hunger versus thirst) affects which dorsal hippocampal cells fire in a particular place during a well-learned "contextual retrieval" task but not during random foraging.Moreover, as the authors stated "'prospective coding' in the contextual retrieval task was not influenced by allocentric spatial trajectory, but rather by the animal's deprivation state and the associated, non-spatial target, suggesting that hippocampal coding includes a wide range of predictive associations.[We conclude] that beyond coding spatiotemporal context, hippocampal representations encode the relationships between internal states, the external environment, and action to provide a mechanism by which motivation and memory are coordinated to guide behavior" (Kennedy and Shapiro, 2004, p. 10805).
But note the use of "encode".As we have said before, how can a relatively (x 40) small hippocampus do so much that, apparently, a large neocortex cannot.In all this literature we suggest that "encode" should be replaced with "echo".Can we conclude that "the mammalian hippocampal formation contains several distinct populations of neurons involved in representing self-position and orientation.These neurons, which include place, grid, head direction, and boundary cells, are thought to collectively instantiate cognitive maps supporting flexible navigation.However, to flexibly navigate, it is necessary to also maintain internal representations of goal locations, such that goal directed routes can be planned and executed" (Nyberg et al., 2022, p. 394, emphasis added).Yes, such masses of information (together with complex social equivalents, Boyle et al., 2024;Schafer and Schiller, 2020) must be available in the brainbut we think cannot fit in the tiny hippocampus.Rather hippocampal cells receive condensed efference copies with which they multitask.

The hippocampus and habituation
The analysis of cell firing fields above has not provided us with a clear perspective on how the different parts of the hippocampus interact to process information.Indeed, a simple "firing field" approach makes it very unclear why the hippocampal "map" would involve a series of distinct cytoarchitectonic zones, a primary "trisynaptic" set of connections between the zones, a massive intermediate output from CA3 via the lateral septum (often ignored by theorists), and topographic variation in, e.g., firing field size as one goes from the septal to temporal pole.
Importantly, labels such as "place field" fail to capture the concept of the hippocampal system as a key node in what essentially comprise two large, circular subsystems ("delay line" and "regulator" in Fig. 5) with distinct (although related) functions.Given the relatively limited number of concurrent multi-cytoarchitectonic-zones recordings until quite recently, and the different interpretations that can be provided of any particular field type, and of the available data on complex fields, we have chosen in this section to focus on the orienting reflex (OR) and its habituationa process and paradigm that is even simpler than that of eye-blink conditioning and for which data have been obtained that are intended to track processing through the circuits of the hippocampal formation.This habituation of the OR is likely supported by a distinct sub-circuit within the hippocampal system from that which underpins trace eye blink conditioning, although both primarily involve inhibition of prepotent responses.It also involves simpler processing than the conflict in choices between alternative goals (see §11).Before proceeding to the detailed cell firing data on the habituation of the OR, we will briefly describe related types of habituation, the orienting response, and the effects of hippocampal lesions.

Types of habituation
There are a wide variety of ways in which "habituation" (a decrease in "response" to the repetition of a specific stimulus) can occur.This, in part, depends on the nature of the initially-novel stimulus but also depends on the nature of its context.One important distinction is between a transient stimulus that elicits an explicit stimulus-specific, essentially reflexive, response (withdrawal, startle, etc) and a transient stimulus that elicits a more general goal-oriented orienting response (OR).A second important distinction is between such novel transient punctate stimuli (eliciting an immediate brief specific or general orienting response) and the static presence in the environment of a novel object or situation, which elicits exploratory behaviours oriented to it.While this latter case involves static stimuli and extended responding to them, it can also be seen as requiring orienting to novelty and so we will not distinguish it from the simple OR.
The OR is an immediate, directed response to the occurrence of certain novel or surprising stimuli.This directionality distinguishes it from the undirected startle reflex and from the lack of any discernible behavioural effect with some other novel stimuli where habituation might reflect sensory adaptation at a non-instrumental level of stimulus processing.Repeated presentation of novel stimuli often produces habituation of the OR in normal subjects, which can be viewed as one of the simplest forms of stimulus-specific learning.Notably, HPC lesions do not block or prevent the OR.For example, rats with HPC lesions exhibit clear orienting responses to punctate auditory and visual cues delivered in an operant chamber (Honey andGood, 2000a, 2000b).
As with simple and delay eyeblink conditioning, the simplest form of habituation of the OR does not require a hippocampus (Honey et al., 1998); being an essentially mono-synaptic reduction in activation that can be observed in, e.g.Aplysia (Castellucci and Kandel, 1974).McNaughton and Vann (2022).Presentation of a novel stimulus that requires orienting (illustrated by a tone, ) will immediately and non-specifically generate an arousal reaction in the reticular formation (RF) and so in the ascending theta (θ) circuit (blue shading and lines).This will cause firing of CA3 cells.Repetition of the stimulus (absent any links with reinforcement) will result in habituation of the CA3 response as a result of development of a planned response (including the plan to ignore the stimulus and do nothing) in frontal circuits that receive specific information about the stimulus.Information that appropriate action/inaction is planned is passed via the informational circuit to the entorhinal cortex (EC) and then dentate gyrus (DG) which sends input to CA3 that cancels the CA3 response, generating habituation.The change in CA3 response is passed round the regulator circuit to lateral septum (LS) and then median raphe (MR), which inhibits the production of the orienting response.EC, dentate gyrus (DG), CA3, CA1, subiculum (SUB), and retrosplenial cortex (RSp) connect unidirectionally.CA1, SUB, and RSp build successive slower and more elaborate models that further damp down orienting providing better integration of the eliciting stimuli and delay lines.SUB and RSp send output to the mammillary bodies (MB) and the anteroventral thalamus (AVT) that is then passed to dorsal and ventral prefrontal (PFCd, PFCv) then perirhinal (Peri) and parahippocampal (Para) cortex complete the informational circuit in EC.RSp also sends output to memory processing areas that interact bidirectionally with PFC to generate plans.The circuitry is greatly simplified and details of the multiple parallel Papez-like circuits involved are provided by McNaughton and Vann (2022).pendix 6 to Gray, J. A., & McNaughton, N. (2000).The Neuropsychology of Anxiety: An enquiry into the functions of the septo-hippocampal system (2 ed.).Oxford University Press.also available as Appendix 6 of McNaughton, N., & Gray, J. A. (2024).The Neuropsychology of Anxiety: An enquiry into the functions of the septo-hippocampal system (3 ed.) (Vol. in press).Oxford University Press.Parts of the text are verbatim from the original but, for simplicity, have not been marked as direct quotes.See the appendix (at http:// dx.doi.org/10.17605/ OSF.IO/ YW32K) for a more detailed coverage, particularly of the contents of Table 2.
However, hippocampal lesions were found to have a profound effect on the expression of the OR in more complex situations.For example, Marshall et al., (2004) assessed which of two visual light cues (V1 and V2), presented simultaneously, rats chose to look at, based on how recently the two cues had been presented previously (V1 was presented 60 s before, and V2 was presented 10 s before, the choice trial).While the controls oriented slightly more to the less recent light cue V1, as expected, the hippocampal lesioned rats did the opposite, orienting to a much greater extent to the more recent cue V2.Furthermore, in experiments in which rats are repeatedly presented with sequences of pairs of auditory and visual cues, whereas controls oriented more to an unprimed target cue compared to a primed target, as expected, HPC lesioned animals exhibited the opposite pattern of behaviour, orienting more to the primed cue than the unprimed stimulus (Honey and Good, 2000).So, HPC lesioned animals can show a change in what they orient to, based on prior experience, but in a way that is opposite to that of controls (Honey and Good, 2000a).Thus, HPC lesioned animals do have a stimulus-specific memory but its expression is different from intact animals, and expression of the OR is clearly altered in these subjects.
With both eye-blinks and the OR, then, primary response production does not require the hippocampus.In contrast, hippocampal cell firing (which is essentially as an efference copy) occurs in relation to primary response production that does not require the hippocampus.But partial or complete inhibition of such primary responses does require the hippocampus.Thus, hippocampal-dependent habituation can be considered in terms of inhibition of the OR.
Importantly, the hippocampal-dependent regulation of the OR, like trace conditioning, can be seen as involving a different form of goal conflict.Normally goal conflict involves a difference between two situations with the same motivation or between two motivations in the same situation.However, with habituation of the OR, both the situation (in the form of the novel stimulus) and the motivation (to orient to the novel stimulus) will enter into conflict with the requirement to temporarily or permanently inhibit the OR as the stimulus gains familiarity.Likewise, the addition of a trace requirement to eyeblink conditioning generates a conflict between the need for a blink response and the need to temporarily inhibit the blink response (during the trace interval).The same is true of extinction of a simple running response down an alley for a food reward (Jarrard et al., 1986).The animal has a conditioned pre-potent response to run down the alley for reward from acquisition (i.e. the goal is "run down alley and get reward") which then comes into conflict with "run down alley means no reward" during extinction, with the hippocampus acting to inhibit the original "run down alley for reward" goal.

Simple habituation of the OR as a key to hippocampal processing
Table 2 details neural activity changes linked to the habituation of the OR to an explicit stimulus that appears against a static background.It includes both experimental parameters and parts of the hippocampal formation in which recordings were made.There are three general points that emerge: 1) many septo-hippocampal neurons are multimodalresponding to stimuli of several sensory modalities; 2) in some areas, unit responses are elicited by novel stimuli and are then subject to habituation with a time course which closely resembles the one described by Sokolov (1960) for the OR in the intact behaving organism; 3) novel sensory stimuli do not always elicit hippocampal responses suggesting the possibility that, as with eye blinks, the cells may be coding orienting responses more than stimulus onset that could be unrelated to an OR.
The bulk of the habituation studies we describe are fairly old and taken from Olga Vinogradova's laboratory.She used a consistent paradigm across a very wide range of brain areas and so her results are worth considering largely as a coherent but isolated block, and as a guide to how plasticity can be transferred from one hippocampal field to another.It is a pity that recent work has not taken the same reductive approach.
Where no reference is given for a fact it is taken from her own reviews (Vinogradova, 1975(Vinogradova, , 1995(Vinogradova, , 2001;;Vinogradova and Brazhnik, 1978) and a more detailed review is provided in the appendix referred to in footnote 3.
Rabbits were confined in a box in which the presentation of a nonspatial punctate stimulus such as a brief tone or light was under rigorous experimental control against a uniform background.So, cell firing was assessed in terms of the modality of the stimuli, their rate of presentation etc.However, we will argue that it is the orienting response elicited by a stimulus that is likely to have been the critical factor in determining firing not the properties of the stimulus as such.
Sensory stimuli produced a variety of specific firing patterns."Phasic" responses were bound to the stimulus, occasionally involving an "on" and "off" response at the start and the end of the stimulus, respectively."Tonic" responses continued after the termination of the stimulus.All of the above (including on-off) could be initially excitatory or initially inhibitory.An advantage of the simple paradigm employed is that latency data, as well as firing patterns, were used in the analysis of the system.As Wiener (1996, pp. 350-351) notes, similar responses to simple sensory stimuli have been obtained in restrained rats, cats and monkeys and also, apparently, in freely moving rats.
In what follows, we will generally conflate the different types of response and concentrate on the relationship of the responses to stimulus presentation and the progressive changes in the hippocampal circuitry with repeated presentation.However, it should be noted that the average firing rates reported by Vinogradova are often in the 10-40 spike/s range associated with cells that fire in a theta pattern in the rat (see below).There is disagreement as to whether high rates and thetarelated firing are a mode of cell activation or reflect a type of cell (Rivas et al., 1996;Vinogradova et al., 1993).Whichever is the case, high rate cells in the rat (where they are presumed to be interneurons) do not have the same receptive fields as low rate cells (presumed pyramidal cells, Jung et al., 1994).Thus, Vinogradova's results may apply to only a subpopulation of hippocampal cells; and, indeed, her conditions may have been such as to activate only that subpopulation, since many principal cells of the hippocampus (granule and pyramidal cells) may be silent except under quite limited conditions (see, e.g., Jung and McNaughton, 1993, p. 182).This subpopulation may have been interneurons.Although Vinogradova (2001, p. 581-582), herself, says "it is necessary to note that in the rabbit, CA3 and CA1 pyramidal neurons are characterized by high level of mean frequency (18.0 ± 1.5 [spikes/s]) and presence of theta-modulation.Low level of activity with intermittent complex spikes is usually observed during somnolence with cortical inactivation and is better expressed in the sequence CA3, CA1, subiculum.Multiple series of experiments with registration of activity in the pyramidal layer of the hippocampus, and special analysis using several criteria, confirmed this conclusion [ (Vinogradova et al., 1992)]." We would expect variation in responses not only with cell type within an area but also across different areas or subfields of the hippocampal formation.In principle we would expect each of these areas to have some sort of comparator function but with area CA3 and CA1/ subiculum being the most prominent as they are the strongest sources of hippocampal output.Table 2 progresses through the HPC formation region by region, following the flow of the "tri-synaptic" path that originates in the entorhinal cortex and can be simplified as passing through the dentate gyrus, CA3, CA1 and subiculum to a final hippocampal formation zone, the retrosplenial cortex (Fig. 5).Note that areas CA2 and CA4 can be thought of as embedded within this main path; and also, that direct entorhinal input is sent in parallel to all the hippocampal zones.But in addition to this main hippocampal formation system, we need to consider a second direct topographically-mapped parallel input to all these structures (including the entorhinal cortex) that arises in the reticular formation, inputs to the medial septum/diagonal band complex, and supplies the phasic input that times rhythmic slow activity N. McNaughton and D. Bannerman ("theta") in all these structures.

Vinogradova's model of the hippocampus and habituation
Before looking at Vinogradova's model we must first draw the most obvious conclusions from the detailed data presented in Appendix 6 (see footnote 5) and summarised in Table 2.
First, even in this simplest of paradigms there was a large variety of different patterns of responses, varying in whether the unit was initially activated or inhibited by a stimulus and, the nature and time course of subsequent changes.However, when looking at the general pattern of changes across areas, some consistency emerged, with progressive changes with passage around the tri-synaptic circuit.
Second, CA3 appeared to be the focus of a hippocampal habituation process.Habituation was not seen extensively in the three structures that project to CA3 (medial septal nucleus, entorhinal cortex and dentate gyrus).It was, however, seen in the two structures that receive output from CA3 (the lateral septal nucleus and, less regularly, CA1).Thus, hippocampal habituation took place in CA3 and was then passed on to the latter two structures, as well as, potentially, to the VTA, hypothalamus and other target areas.An active transfer of habituation, as opposed to a loss of previous excitation, was suggested by the fact that, if the connection between CA3 and the lateral septum was severed, habituation no longer occurred in the lateral septal area; on the contrary, unit responses tended to increase with stimulus repetition (Vinogradova, 1975;Vinogradova and Brazhnik, 1978).Causal evidence for the crucial role of CA3 in hippocampal habituation is provided by manipulation of glutamate receptor function within this subfield (Bygrave et al., 2019;Kilonzo et al., 2022).
Third, habituation in CA3 appeared to be based on detection of a match in time between medial septal ("wake up" / something happened) and entorhinal ("it's under control" / something is expected to happen) inputs.If the septum was disconnected from the hippocampus or if the entorhinal cortex was disconnected from the hippocampus, the ultimate response of CA3 cells was a gradual increase rather than decrease in firing rate.Further, the initial stimulus-nonspecific responses in CA3 (essentially an efference copy of OR generation) must come largely from the medial septum, while the stimulus-specificity of the habituation (driven by the planning of action/inaction) must come from the entorhinal cortex (directly or filtered by the dentate).Note that a mismatch in either direction resulted in CA3 output, and that a stimulus-specific signal could cancel a multimodal signal.This makes sense if the purpose of an OR is to determine the nature of an unexpected stimulus and so whether action is required; and the OR becomes unnecessary as soon as action/inaction is already planned for that particular stimulus.In both cases, our hypothesis sees what is being processed in the hippocampus as intended-goal information (with the OR having the general goal of information collection about a specific stimulus) not cognitive (stimulus-related) information and, in both cases, with the hippocampus receiving efference copies from goal systems and not acting to control goals directly (hence the lack of any effect of hippocampal lesions on the OR itself, as opposed to its habituation).
Fourth, concurrent with habituation to a familiar stimulus in CA3 (8-15 presentations), there was a build-up in the response first in the entorhinal cortex (2-12 presentations of the stimulus) and then in the dentate gyrus (15-20 presentations).This augmentation of response is most simply explained as the result of neural plasticity, probably starting in cortical areas that build up a model of the stimulus, and progressing through the entorhinal cortex to the dentate gyrus.This potentiation could correspond to a build-up of "familiarity" (Barkus et al., 2014;Vinogradova and Brazhnik, 1978) and invocation of a consequent "familiar-ignore" command.
Vinogradova encapsulated her results in a model where "the hippocampal system is regarded as consisting of two large circular subsystems (Fig. [5, see legend for abbreviations]).The first of them, linking the hippocampus (especially the CA3 area) to the brain-stem structures through the relay nuclei of the septum, is regarded as a regulatory circuit.This system introduces into CA3 primary information about the changes in a relatively stable environment, and controls the general level of brain activity.Its increase by release of RF excitatory influences is necessary for arousal, orienting-response, attention-providing conditions for effective processing and fixation of information in the neocortex.Its suppression by the hippocampus through the intermediary of [the median raphe] is necessary for the "disconnection" of attention and its switching to other stimuli.
The [second] system, which is linked mainly through the CA1 area, introduces into the hippocampus the same signals preprocessed in the neocortex and, after additional complex transformations of the information at its multiple relays (subiculum, [MB], AVT, [RSp]), returns it back to the neocortex as a final order for its registration in nonprimary areas.This system is regarded as an informational circuit, because the neuronal responses in its links retain the qualitative characteristics of the stimuli.The important feature of this system is the incremental dynamics of responses, which are slower, the further a relay structure is from CA1.This allows us to regard the structures of the main limbic circuit as a chain of successively linked integrators, in which each next link became active only after the signal is shaped at the preceding link, and as a delay line, which prevents rigid fixation of spurious, low-probability signals and helps to obtain the best organization of the classificatory system of trace storage in the long-term memory.
Both systems are connected through the CA3 area, which is regarded as a comparator device, detecting the novelty of a stimulus (i.e., absence of its trace in the memory system) on the basis of signals in its two inputs: those from the brain stem and cortex.Both signals are additionally preprocessed at the symmetrical relay structures at the entrance to CA3: MS-DB and [DG].At both these relays, the additional procedure of secondary simplification and generalization of the input signals is performed."(Vinogradova, 2001, p 594) Thus, according to Vinogradova, the same simple stimulus can elicit responses that produce both non-specific and specific activity that affect the hippocampus through two separate routes.On first and subsequent

Table 2
Neural activity linked to the habituation of the OR (Vinogradova, 1975(Vinogradova, , 1995(Vinogradova, , 2001;;Vinogradova and Brazhnik, 1978).presentations, the occurrence of a stimulus activates the reticular system to generate an OR and affects the hippocampus via input from the medial septum.Initially, input also activates specific sensory cortical areas, but not the temporal lobe.Successive presentations allow a predictive "cortical model" of the stimulus to be built (depending on neural plasticity mechanisms).Once this cortical model is sufficiently complete, its prediction of upcoming stimulus occurrence flags an appropriate response (or lack of response) and affects the hippocampus through the second route, the entorhinal cortex.This second specific predictive input cancels the effects of the first general input, resulting in the observed habituation in CA3 and so cancellation of the orienting reaction.Interestingly, Vinogradova (pers. comm.) has shown that if LTP is artificially induced by perforant path stimulation, the hippocampus proper becomes totally unresponsive to natural stimuli that had previously elicited a response (see also Miller et al., 1995;Vinogradova, 2001).If LTP in the dentate represented the simple medium to long-term storage of a stimulus memory, the opposite result might be expectedi.e. an increase in responses to natural stimuli.Thus, there is an important practical reason for preferring this, essentially functionally inhibitory, view of dentate gyrus LTP to the view that it reflects associational learning or map building (see discussion in Elliott and Whelan, 1978, p. 407 et seq;and, e.g., McNaughton and Morris, 1987;O'Keefe and Nadel, 1978, p. 230).
The results presented by Vinogradova's group provide a satisfyingly coherent account of how the septo-hippocampal system performs at least one of the functions, processing of novel stimuli, which has been attributed to the hippocampus.There is also evidence from fMRI and PET scans that the hippocampus (and other components of the limbic circuit) is activated by novelty of stimuli in human beings (Kumaran and Maguire, 2005;Tulving et al., 1994).But Vinogradova herself has warned that "it is curious how we find in the brain what we are looking for" (in Elliott and Whelan, 1978, p. 197).There has, indeed, been some question as to how far Vinogradova's results can be replicated (Best and Best, 1976;Hirano et al., 1970;Lidsky et al., 1974aLidsky et al., , 1974b;;Mays and Best, 1975; but see also Segal, 1974).This variability of results is likely to depend on the extent to which different stimuli result in an orienting response; and we will argue that the habituation experiments are better explained in terms of processing of goals (and hence, potentially, responses) than processing only of stimuli.Critically, habituation can likely occur at different levels within the brainranging from what we would see as sensory adaptation to loss of interest in a plan or goal.Justifying Vinogradova's warning, Wiener and co-workers obtained sensory correlates of hippocampal unit activity in freely moving rats of a similar type to Vinogradova's, and noted that "these discharges had no location-selective or task-related correlates.… These were not simply novelty responses since the rats had experienced these stimuli in many training sessions.… [They appear] linked, perhaps in an indirect manner, with movements triggered by the sensory stimuli.… [For example,] visual stimuli could trigger orienting responses like eye movements; the latter have been shown to be correlated with hippocampal activity in the monkey" (Wiener, 1996, p. 351-352) (see also Givens, 1996 on the predominance of response related correlates in the medial septum; Korshunov et al., 1996).

Eyeblinks versus orienting
It is useful at this point to compare and contrast the eyeblink and orienting data.We will refer to these below as conditioning and habituation experiments, respectively.Both involve simple non-spatial paradigms and have provided comparable data for different parts of the hippocampal formation.We have already noted that, in both cases, processes related to the primary response being measured produce hippocampal cell fields that are essentially efference copies but that the hippocampus only becomes involved in the control of those responses when there is a requirement to inhibit the response (trace conditioning in the case of eyeblinks, habituation in the case of orienting).
Conditioning and habituation results are directly comparable with respect to the fact that both hippocampal and septal responses were recorded in restrained rabbits responding to simple, non-spatial stimuli like tones.The conditioning experiments differ in that they include a paired air-puff UCS (and so also required an unpaired CS/US control group) and provided a less sensorily-deprived environment.
A major similarity between the conditioning and habituation studies is the gradual development of dentate gyrus responses with repeated trials.In the case of the conditioning experiments, these responses have the temporal topology of the motor response.This is likely to have also been the case for the (unmeasured) responses in the habituation experiments if this involved a simple orienting reaction.
A theoretically important difference between the results obtained in the habituation and conditioning experiments, respectively, appears in both CA3 and CA1.In the conditioning experiments responding increases, in both CA3 and CA1, in parallel with the increase in the dentate.This increase parallels the development of conditioned responding.In the habituation experiments, responding decreases in CA3 and CA1, apparently suppressed by the increase seen in the dentate.In both cases, therefore, the dentate "model" (Berger and Alger, 1976) is closer to the stimulus events, while the CA3 and CA1 "model" is closer to the response events (changing from non-response to response in the conditioning experiments, and changing from response to non-response in the habituation experiments).
Two additional points arise from §5 that should be emphasised here.Both are important for our understanding of hippocampal cell fields and for hippocampal control of behaviour.Critically, together they lead to the conclusion that the superficial characteristics after which hippocampal cell fields are named are not a good guide to what the hippocampus does with the information it receives.
First, the circuit that is the basis for both the actual eyeblink conditioning process and for the conditioned response itself has been well worked out and does not include any part of the septo-hippocampal system.Conditioning depends on a simple reflex system eliciting the unconditioned response.The unconditioned response is transformed into a conditioned response by plasticity in a circuit involving the deep cerebellar nucleus and the red nucleus.This circuit connects neurons which respond to the CS with neurons in the UR circuit, and so closely parallels the mechanism of simple avoidance conditioning in the amygdala (Bauer et al., 2001;Lamprecht et al., 2009;LeDoux, 1994).
Second, lesions of the hippocampus (or even decerebration) do not usually interfere with the conditioned nictitating membrane response; nor with its acquisition in the basic form of the paradigm; nor when a delay procedure is used (in which the CS overlaps the US in time).Here, we should remember that, while area CA1 shows conditioned responses, the mammillary bodies do not (Berger and Thompson, 1978).The mammillary bodies receive their hippocampal information from the subiculum rather than CA1, and from the lateral septum rather than CA3.This suggests that the conditioned reactions in CA1 and CA3 are not transferred to the subiculum and lateral septum, respectively, unless additional criteria are met.(We have already seen such a lack of transfer within the hippocampus in that the dentate reaction is not matched in CA3 and CA1 if the stimuli prove unimportant).That is, each step in the CA3-CA1-subiculum-retrosplenial passage of information provides a logical gate to determine the nature of any functional hippocampal formation outputwith areas like perirhinal cortex able to directly alter such information transfer (Fig. 5).
These results, with the "modelling" or representation by the hippocampus of the simple conditioned response, remind us of a point that we started with: that the most obvious correlate of an increase in a cell's response may not be a good indicator of function.Consistent with an inhibitory view of the hippocampus (Gray andMcNaughton, 1983, 2000), stimulus reversal of the nictitating membrane response is impaired after hippocampal lesions due to continued production of the response to the original stimulus (which was originally learned at a N. McNaughton and D. Bannerman normal rate by the hippocampal-lesioned animals, Weikart and Berger, 1986), that is to a failure to inhibit responses to the previously correct, but now incorrect, stimulus, rather than due to failure to acquire the newly correct response.The missing inhibition could, when the hippocampus is intact, in theory be supplied either via the CA3 output to the lateral septum, via the subicular output to the mammillary bodies, via the subicular output to posterior cingulate, or via more minor efferents (Berger, Berry, et al., 1986).Consistent with the idea of a CA1/subiculum/posterior cingulate route, Berger, Weikart, et al. (1986) showed that lesions of the posterior (retrosplenial) cortex produced a similar loss of inhibition of incorrect responses during stimulus reversal of the nictitating membrane response.
In contrast, the habituation experiments are most easily described in terms of hippocampal activity coding for familiarity of stimuli, with the crucial output from area CA3 permitting novel stimuli to activate the lateral septal area and the mammillary bodies (Luo et al., 2011).Given the hypothalamic targets of this output, we can then say that the output from area CA3 allows novel stimuli access to response mechanisms.However, novelty itself may not be sufficient, given the lack of response to simple stimuli observed in the conditioning experiments; and, as suggested by Wiener, the lateral septal activity may reflect input from subcortical orienting-related mechanisms.In other words, the hippocampus will only become involved if the novelty or surprise is such as to generate an orienting response.
The conditioning experiments are most easily described in terms of hippocampal activity coding for upcoming responses, rather than for the stimuli that elicit the responses.However, this coding is not necessary for production of the responses being coded; nor does it appear to produce significant output from the hippocampal formation; nor does it accompany the unconditioned responses.
What is needed, then, is a view of habituation of orienting that is more response-focussed (as opposed to sensory adaptation for example), coupled with a view of the conditioning experiments that is more stimulus-focussed.This can be achieved by rephrasing both sets of data in terms of goalsnoting that a component of habituation (but not all habituation) occurs at the level of the goal and that even very simple conditioning can be viewed as goal acquisition rather than stimulusresponse linking (McNaughton et al., 2016).
We propose that, with a stimulus that is sufficiently intense or significant to elicit an orienting reflex, medial septal and hence hippocampal cells react.Here the hippocampus is detecting neither the stimulus nor its novelty, as such, but the requirement to produce an orienting response directed to the source of the stimulus (possibly indicated by the activation, but not release, of subcortical response systems), i.e., it has detected the presence of a available goal.After suitable neural plasticity in a variety of structures, including finally the entorhinal cortex and dentate gyrus, a model of the currently preplanned response (action or inaction) to the goal arrives in the hippocampus and (if the stimulus is a neutral one) cancels the effect of the septal response.This model of the goal, by flagging a pre-planned response, effectively codes familiarity in purely cognitive terms.Provided it is not accompanied by other inputs that would indicate preparation for action, it also indicates unimportance.A significant feature of this circuit is that a subsequent failure of the medial septal input to match the model, i.e., omission of the expected stimulus, will also result in hippocampal output, reflecting detection of a mismatch and providing a "surprise" signal.Thus, not only does the entorhinal input gain the capacity to block the effects of the medial septal input, but the septal input concurrently gains the capacity to block the effects of the entorhinal input.
We should also note that the multicellular records suggest that the increase in firing rate, at least in area CA3, is quite general.It occurs even when the classically conditioned stimulus is superimposed on quite different instrumentally conditioned baselines.Thus a CS for shock produces an increase in CA3 firing rate, and continues to do so both when superimposed on an aversive baseline (when the CS increases behavioural responding) and on an appetitive baseline (when the CS decreases behavioural responding), as was shown by Laroche et al. (1987).Berger et al. (1983Berger et al. ( ), p. 1206) ) reported "that 83 % (40/48) of all pyramidal neurons significantly increased firing rate within either the CS period, or the UCS period or both periods".Some of their individual cells showed the same modelling of the conditioned response as the multiunit record (see their Fig.5a), but others showed much more restricted firing which suggested that "a complete unit representation of the conditioned [nictitating membrane] response may be produced by the simultaneous activities of many pyramidal cells" (i.e., by cell assemblies; op.cit., p. 1208).

What does the hippocampus do?
There is a popular, widely held, view that memories start off in the hippocampus before being consolidated and passed on to the cortex (see reviews by Inostroza and Born, 2013;Moscovitch and Gilboa, 2021).Sleep has been suggested as a key state for this consolidation; but it may take months/years for memories to become independent of the hippocampus (Ogden and Corkin, 1991, p. 200-201).A detailed picture of the hippocampus recycling information, particularly during sleep, for consolidation by the cortex is presented by Inostroza and Born (2013).
Here we present the details of an alternative view intended to fit with the data reviewed so far.

The hippocampus is not an associative memory store
While several authors have argued for a memory engram in the hippocampus (e.g., Hainmueller and Bartos, 2018;Sweis et al., 2021;Tanaka and McHugh, 2018), we think there are good reasons to believe that there is never associative memory "storage" as such in the hippocampus.Most obvious is its relatively small size and lack of any relation of this to the size of the neocortex (see footnote 1).It is also unclear how the coding of information by the brain would allow transfer of an engram consisting of a pattern of synaptic weights in a cell assembly 6 in the hippocampus to a cell assembly in the cortex.Engrams result from plasticity of connections within cell assemblies.It is the location of the assembly within neocortex or subcortex (e.g.amygdala for fear conditioning) that determines its meaning.Its meaning is not (as with a computer) determined by a bit pattern independent of location.Transfer of an engram from one location to another, then, seems impossible.Activation of a neocortical or subcortical cell assembly encoding a goal engram will result in output to appropriate action programming systems.This output, in the form of an efference copy (potentially simplified or compressed, see §11.2) can activate the hippocampus and, of necessity, will have all the environmental correlates of the activation of the goal.It will echo the activation of the cortical engram but not encode it.
Furthermore, it is notable that more recent studies have suggested that the neuronal ensembles that are thought to constitute the engram in the hippocampus (e.g., Park et al., 2011) are much more dynamic and fluid than previously thought to be the case (Sweis et al., 2021).For example, in vivo electrophysiological recording studies show that the ensemble activity correlation of place cells is approximately 90 % when animals are returned to the same spatial environment minutes later, but is 60 % or lower when the retention interval is across days.Likewise, immediate early gene studies demonstrate approximately 90 % 6 We use the term cell assembly rather than neuronal ensemble or neuronal assembly (see Box 1 in Yuste, R., Cossart, R., & Yaksi, E. (2024).Neuronal ensembles: Building blocks of neural circuits.Neuron.https://doi.org/10.1016/j.neuron.2023.12.008 for definitions) to emphasise the link with Hebb, D. O. (1949).The organization of behavior: a neuropsychological theory.Wiley-Interscience.Our use includes both the static and dynamic aspects of ensembles/assemblies.

N. McNaughton and D. Bannerman
ensemble activation correlation after 20 mins, falling to 30-50 % across days.Although the stability of hippocampal representations may depend to some extent on the specific subfield (e.g., Hainmueller and Bartos, 2018), these results suggest that these putative hippocampal engrams, particularly in CA3 and CA1 subfields, are less stable than previously envisaged.
Indeed, Tanaka et al. (2018, p. 2-3) showed that even when hippocampal "engram" neurons maintained some activity on their second visit to the same environment, "contrary to expectations of a spatial memory trace, many of these cells shifted their firing locations [i.e.remapped] during the second visit.In [c-fos] positive neurons, the average correlation of the firing-rate maps from the encoding and recall sessions was close to zero.… [Thus,] engram cells do not necessarily represent reliable spatial information about the external world, but rather through their net activity serve as an index to episodic information stored elsewhere in the brain." As they note, this is very close to the original indexing theory of the hippocampus (Teyler and Discenna, 1986;Teyler and Rudy, 2007) but places it in parallel with cognitive mapping, rather than excluding it.In a more detailed exposition of this idea (Tanaka and McHugh, 2018), they argue that indexing of context (where the context itself is held elsewhere) is important for reinstatement of activity in the cortical cell assemblies activated during initial encoding.Note that, like the original indexing theory, this does not by itself account for the fact that in cases like H.M. longer term memories are intact.It needs the further assumption that consolidation (particularly during sleep) would involve 'continual reactivation of a particular hippocampal index [that] would have a slowly incrementing effect on the cortical circuitry it indexes [and] the hippocampal index, with time, would become redundant' (Teyler and Discenna, 1986, p. 150;updated by Teyler and Rudy, 2007) (see also Goode et al., 2020).
It is unclear with such theories why an 'index' is required in the first place if it is not needed later.This is particularly true for indexing of context as opposed to indexing of the engram/cell assembly itself.However, on the view that background context makes tasks hippocampal-sensitive by generating multiple possible interfering alternatives (compared with a simple foreground stimulus), 'indexing' would reflect the suppression of incorrect alternatives (and so be linked to context in a general sense) -with consolidation creating long-term suppression of the interfering associations rendering the correct alternative easy to recall without hippocampal involvement.
It is also worth noting that Hebb's original view, which has stood the test of time, saw different types of memory as different modes of activation of the same type of cell assembly (Hebb, 1949).The only obvious problem with the immediate cortical storage of memories (see Hebb's Fig. 7 for an example in visual cortex) is the high potential for interference from related previous associations and for, linked to that to some extent, catastrophic forgetting (McNaughton and Wickens, 2003).This problem is solved if storage occurs in the cortex but retrieval is supported by hippocampal suppression of interference until consolidation has sufficiently strengthened the target memory relative to its competitors (McNaughton and Wickens, 2003).We discussed data supporting this view when considering trace eyeblink conditioning.Note that such consolidation could include recycling of information through the hippocampus but the hippocampus would not create the strengthened memory or pass it to the cortex.As we will discuss in detail in §11.5, the hippocampus plays a modulatory role, permitting the potentiation of desired connections by repetition, while encouraging depression of interfering connections (both of which are stored in the cortex via Hebbian mechanisms).
Despite, in particular, the idea that the hippocampus never acts as even a temporary information store for associative memories, our model can be mapped to current ideas about involvement of the hippocampus in hidden-state inference ( §7.3; see also §11.3), and information recycling (Inostroza and Born, 2013).In our model, consolidation involves retrieval and re-remembering, with information being refined as it loops between cortex and HPC.
On our view the hippocampus is critical for goal processing, including but not limited to the temporary storage and retrieval of memories of goals.Note that a goal (situation plus motivation) here is very different from an action per se and would not, therefore, involve the hippocampus in procedural learning or procedural memory.The model of hippocampal network function provided below has as its foundation the ideas (Gray and McNaughton, 2000;McNaughton and Gray, 2024) that the hippocampus: 1) receives, and so has cell fields linked to, much goal information (from neocortex, subcortex, cerebellum) to which for much of the time it might produce no functional output because it is "just checking" for conflict between goalsthat is, the hippocampus is comparing features between multiple distinct available goal fields, not reacting to any single one; but then, importantly, 2) when it does detect conflict between goals, the hippocampus acts immediately to selectively block the actions programmed by currently competing goals and then to primarily resolve conflict via external (exploration, risk assessment) or internal (memory) information-gathering.

The nature of the entorhinal input to hippocampus and the role of grid cells
Entorhinal cortex is often seen as the primary input to the hippocampus (particularly to the dentate gyrus).It also receives hippocampal output (particularly from the subiculum).So, its prime purpose could be viewed as mediating two-way communication between the hippocampus and neocortex.
However, it has extensive intrinsic circuitry that does not appear focussed on hippocampal output but rather on substantial integration of the dorsal (where) and ventral (what) trend information it receives (Nilssen et al., 2019); and it has reciprocal connections via perirhinal/parahippocampal cortex to the dorsal/ventral trends of neocortex.Thus, entorhinal cortex likely provides more than just a relay station between neocortex and hippocampus.Notably, there are important differences in the anatomical projections to lateral and medial entorhinal areas from perirhinal and parahippocampal/post-rhinal cortices respectively (in turn, potentially reflecting more input from the ventral and dorsal visual streams to perirhinal and parahippocampal/post-rhinal regions respectively; Eichenbaum et al., 2007, see their Fig. 3).Based on these anatomical differences, Manns and Eichenbaum (2006) suggested that these parallel processing pathways convey separate information regarding (i) the spatial context of sensory information (more generally the situation) to the medial entorhinal cortex (MEC), and (ii) the non-spatial identity of stimuli such as objects (what we have labelled motivation) to the lateral entorhinal area (LEC).These distinct streams of information are then combined in the DG and CA3 subfields of the hippocampus, with Manns and Eichenbaum (2006) suggesting that this could be important for item-context associations.Elaborating on this framework, Whittington et al. (2020) suggested that hippocampal representations reflect the conjunction between sensory input from the LEC and structural knowledge about the world from the MEC (which includes not only knowledge of spatial structures but also non-spatial models), allowing first presentation inferences to be made without any previous experience of the specific sensory details and to generate predictions as to the next sensory experience.
Thus, Whittington et al. (2020), suggested that there is a structural model of the world in MEC (which also potentially reflects the input to this region from prefrontal cortical areas) that is combined with sensory information from LEC to generate predictions which can then be relayed to the hippocampus.In our terms, then, the entorhinal cortex integrates motivation and situation to code upcoming goal-related responding.So, an efference copy sent to the hippocampus (and linked to recall) would signal, as we put it in §9.4,"it's under control" / something is expected to happen.Thus, there appears to be a model of the structuring of goals in the world in MEC that is combined with object-specific, particularly motivational, information from LEC to generate predictions.This all suggests that entorhinal cortex provides a direct contribution to the selection of courses of action and does not act merely as a two-way relay for the, anatomically somewhat separated, hippocampus.In this context, it is particularly interesting, that cells in MEC layer 5a send an efference copy of their telencephalic outputs to area CA1 (Tsoi et al., 2022).Thus, both the primary entorhinal input to the dentate gyrus (potentially affecting subcortical outflow from CA3) and this input to CA1 (potentially affecting cortical outflow from the subiculum) may function to permit monitoring by the hippocampus of available goal selection that is being undertaken by entorhinal cortex.
There is a comprehensive literature on the entorhinal cortex and, in particular, its "grid cells" (for a brief overview of grid cells in the context of place cells, see Burgess, 2014).Here we will ignore further details of more general entorhinal functions (linked to its output to neocortex) and focus on its output to the hippocampus.In contrast to the reciprocal connections of the deep (essentially final output) layers of both LEC and MEC to their cortical output targets, the shallower layers send unidirectional input to all the hippocampal subfields, with a unidirectional return from the hippocampal circuit, from CA1 and the subiculum to the deep entorhinal layers (Nilssen et al., 2019, see their Fig. 1).This would be consistent with hippocampus, as a whole, monitoring (as in "just checking") the ongoing processing in both LEC and MEC.One indicator of the nature of this processing is the "grid" pattern of cell firing in entorhinal cortex.
Grid cells can be thought of as cells with recurring "place fields" that, viewed as a heat map, form a hexagonal array (Hafting et al., 2005).This map is more regular and topographic than a place field map in terms of the spacing of its multiple fieldsbut it does rescale and so it is not locked to space as such.Furthermore, (as with 'place fields') a grid pattern is also present in non-spatial situationsfor example, auditory (Aronov et al., 2017) and conceptual (Constantinescu et al., 2016).So, what is the repetitive grid pattern related to?Aronov et al. (2017, p. 722, emphasis added) suggest that "in more complex [hippocampal-sensitive] tasks (for example, those containing memory-guided decision points), the hippocampal-entorhinal system might similarly represent arbitrary behavioural states.In this framework, task performance activates a sequence of neural activity, in which firing fields are elicited parametrically with progress through behaviour.Neighbouring and partially overlapping fields therefore represent the order and adjacency of behavioural states.This could be useful for linking events in episodic memory and for planning future actions (for example, via simulated continuous neural sequences)." We would suggest that it is not "behavioural states", per se, that are being processed.Behaviour can be highly variable from occasion to occasion.Rather the patterns will reflect sequences of upcoming goals and subgoals (where choice of behaviour, as such, by the motor system will occur on the fly and not be exactly repeated but goals will recur across occasions).Critically, hippocampal firing must be generalised and predictive (i.e."memory-guided").The requirement for generality (so a current case can be recognised as "the same" as a previous case, Coutureau et al., 2002) suggests that a key process driven by interactions between LEC and MEC could be a reduction in the amount of information coding for "a goal" -reducing it to features that will easily transfer between occasions (and also be easily matched with the "wake up" / something happened input to hippocampus from the medial septum).
Along similar lines, Stachenfeld et al. (2017Stachenfeld et al. ( , p. 1643) ) argue that hippocampal cells are involved in predictive coding, supporting "learning of a predictive map that represents each state in terms of its successor states" (see also Fuhs and Touretzky, 2007, §8, §11.3).They also argue that entorhinal grid cells are predictive but "encode a low-dimensionality basis set for the predictive [hippocampal] representation, useful for suppressing noise in predictions and extracting multiscale structure for hierarchical planning … [that is, they] encode a low-dimensional decomposition of the predictive map, useful for stabilizing the map and discovering subgoals."For hippocampus, we would see this as uncovering specific goals and subgoals from a plethora of competing alternativesnot discovering isolated examples.That is, hippocampus would become involved in "noise suppression" when there is conflict between competing and/or conflicting predictive maps (see also §11.3).Baram et al. (2021) find that, with both entorhinal cortex and ventromedial prefrontal cortex the stability of the cortical representation across multiple reinforcement learning problems in humans requires preservation of the task structure.Not only does this show generalisation and task-relatedness, it suggests that the entorhinal cortex is dealing with generalised task-related information that is also held in frontal (ventro-medial in this case) cortex and striatum.So, we can see the entorhinal filter as applying rules that unidirectionally feed the hippocampus a more general picture of the nature of the current goals (tending to blur to the general way a problem is being solved rather than the current detailed specifics).This generalisation before feeding to hippocampus accounts for the hippocampus having multiple place fields (in that they are all the same type of problem) and (since its representations are compacted) for the limited expansion of the hippocampus relative to neocortex (see footnote 1).Constantinescu et al. (2016) found that, at least in a two-dimensional conceptual task, human anterior cingulate/medial prefrontal cortex and entorhinal cortex generate a grid-like signalthe strength of which correlates with task accuracy.This signal in the ventromedial prefrontal cortex was consistent on repeat testing after half an hour and also after more than one week.This is consistent with the results of Baram et al. (2021) and emphasises the role of the ventromedial prefrontal cortex in the long term.Critically, therefore, the grid feature is not a spatial thing per se.Space just happens to be continuous like the other variables manipulated by Constantinescu et al. (2016) and Baram et al. (2021)."Our findings suggest that global relational codes may be used to organize nonspatial conceptual representations and that these codes may have a hexagonal grid-like pattern when conceptual knowledge is laid out in two continuous dimensions."(Constantinescu et al., 2016, p. 1).
Thus, there is growing evidence as to exactly what sort of informationa predictive mapis being represented by ensembles of entorhinal neurons.These predictive maps of available goals and subgoals provide, likely in a further condensed form, the information being received by hippocampus across both spatial and non-spatial domains.The idea that the hippocampus is representing predictions of likely future events or scenarios is also analogous to its proposed role in "episodic future thinking" (Johnson and Redish, 2007;Redish, 2016).For example, Eleanor Maguire and colleagues used fMRI to demonstrate hippocampal activation when subjects were asked to imagine fictitious experiences (Hassabis et al., 2007a).Notably, activation was observed in distributed networks that overlapped considerably with those regions that were activated during episodic memory recall (including the hippocampus), implicating associative retrieval processes in these imagined scenarios.In a parallel study, Hassabis et al., (2007b) also found that the ability to imagine fictitious future scenarios was impaired in hippocampal amnesic patients.The key question is what does the hippocampus then do with this information?
At this point it is worth noting that these theories as to the nature of hippocampal representations come largely from correlative studies recording neural activity during behaviour in either humans (indirectly) or animals (directly).One notable feature of hippocampal research over many years, as mentioned previously, is the apparent discordance that sometimes occurs between the conclusions drawn from studies recording hippocampal activity during behaviour and those arising from the effects of experimental manipulations that disrupt hippocampal function on behaviour (e.g., intact CA1 place fields despite lost place learning; intact running despite lost theta or absent hippocampus).

N. McNaughton and D. Bannerman
Indeed, across the memory domain, we have seen that there are numerous examples of hippocampal lesion effects (or the effects of other manipulations which disrupt hippocampal function, Bannerman et al., 2012) on reversal or extinction of an acquired behaviour, whereas the initial acquisition of the task was unaffected (Table 1).If the role of the hippocampus was solely in generating a prediction and responding accordingly, it is not immediately obvious as to why this would be important for task reversal and not for the original task acquisition.
The answer may be found in another well-established electrophysiological signature in the hippocampus, namely hippocampal flickering; which may be linked to the role of theta phase in separating/disambiguating between competing goal representations "to allow extinction of prior learned associations that do not match current input" (Hasselmo et al., 2002, p. 793; but note that we see the key functional changes as occurring in cortex not hippocampus).Note, more complex explanations may be required for the phenomena of serial reversal learning (Fuhs and Touretzky, 2007).

Flickering and the selection between competing goals
We have already briefly mentioned ( §7.3) the phenomenon of "competitive flickering".We focussed on it as evidence for a lack of local spatial stability and so as a reason for not accepting the idea of hippocampal cells having "place" fields.However, it can also be viewed as strong positive evidence for the idea that hippocampal cells are responding to efference copies of the activations of goal fields, and then potentially to the resolution of conflict between competing goals by the hippocampus.Note that while these "fields" may often be linked to a place (e.g., a specific choice point in a maze) competing goals must be in the futurethey are potential targets of upcoming, incompatible, action sequencesand they provide the data on which the upcoming choice between them will be made.
As we briefly noted earlier, Johnson and Redish (2007) exposed rats to the same multiple-T maze across days while varying reward location from day-to-day and while recording from large ensembles of hippocampal cells.They determined conventional "place fields" (averaged over long periods) and, critically, looked at brief periods of scattered spiking when cells fired outside their usual field ("nonlocally").In their study and across several other studies "when animals pause at the choice point, hippocampal sequences within any given theta cycle proceed along a single path towards a single goal, one at a time, first towards one goal and then towards the other" (Redish, 2016, p 151).This firing may reflect "the consideration of potential alternatives, and might be indicative of a trajectory planning process" (Johnson and Redish, 2007, p.12176), particularly when occurring at critical choice points and during error correction.Importantly, the sequences did not track from a goal back to the rat nor in both directions simultaneously.
"That extrafield spiking maintained spatial organization across multiple simultaneously recorded cells during extrafield activity implies that the extrafield spiking may coincide with transient, nonlocal representations of events or locations.In these tasks, all essential information can be projected onto space (i.e., location of the animal).As a result, memory (sequences and episodes) and any decision-related processes present within hippocampal ensemble dynamics can be made observable by examining dynamic changes in the spatial representation.It is critical to note the reconstruction analysis makes no fundamental assumption regarding spatial representation within the hippocampus, only that memory and decisionmaking signals may be coherently projected onto space in these tasks.… [Further,] the hippocampus may only be providing the prediction component; evaluation of the value of that prediction and the making of the decision may happen downstream of the hippocampal prediction process" Johnson and Redish (2007, p. 12183-12184) The distinction between goal control and automated action is important herewith rat responding becoming more automated and stereotyped with more experience.Matching this, "in the hippocampus, [flickering] sequences of firing of hippocampal place cells [initially] represent sweeps ahead of the animal, serially exploring the paths towards the available goals.[But] as behaviour automates, these sweeps transition from going in both directions to going only in one direction, and then vanish" (Redish, 2016, p 148).
Along similar lines, Jezek et al. (2011) recorded hippocampal neural activity in the CA3 subfield while rats were "teleported" between distinct spatial contexts.They found "that instantaneous transformation of the spatial context does not change the hippocampal representation all at once but is followed by temporary bistability in the discharge activity of CA3 ensembles.Rather than sliding through a continuum of intermediate activity states, the CA3 network undergoes a short period of competitive flickering between preformed representations of the past and present environment before settling on the latter.Network flickers are extremely fast, often with complete replacement of the active ensemble from one theta cycle to the next.Within individual cycles, segregation is stronger towards the end, when firing starts to decline, pointing to the theta cycle as a temporal unit for expression of attractor states in the hippocampus.Repetition of patterncompletion processes across successive theta cycles may facilitate error correction and enhance discriminative power in the presence of weak and ambiguous input cues."Jezek et al. (2011, p246, emphasis added) In all this, you should remind yourself of the relative lack of expansion of the human hippocampus in phylogeny and avoid the temptation to assign functional value to our superficial description of firing patterns.We see the hippocampus as holding a mirror up to activity elsewhere.During learning, associative processes in planning areas such a prefrontal cortex will activate cell assemblies linked to both old and new situations, essentially concurrently.Decision mechanisms presented with this input will have a focus that flickers between the alternatives.The alternating ascendency of the goals from which output inhibition must be released to generate action thus send "flickering" efference copies to the hippocampus.This flickering may be a key hippocampal signature of the presence of goal conflict.
For example, Dupret and colleagues trained rats on a cheeseboard maze task during which the animals had to find three hidden food rewards at three different spatial locations.The animals were required to learn a new set of three goal locations on each day of testing.Thus, behavioural testing essentially involved a series of spatial reversals during which one set of goal locations must be suppressed at the same time as a new set of goal locations are activated (c.f.discussion of reversal learning with eye blinks in §5) in an otherwise highly familiar spatial environment.Hippocampal neural activity was recorded in the CA1 subfield of the hippocampus (see Dupret et al., 2010;2013).Their idea was that "cell assembly patterns can flicker rapidly between the representation of different maps across consecutive theta oscillatory cycles when environmental cues or task parameters are abruptly changed.It is possible that such flickering may also take place between old and newly-formed representations during spatial learning.This could enable competitive processes in which old and new maps initially vie for prominence, with the new maps dominating in later stages of learning.… [Consistent with this] within many earlier trials, both the old and the new pyramidal assembly representations were expressed in nonoverlapping theta cycles, with later trials dominated by the new patterns.Moreover, the expression strength of the new assemblies improved during the course of learning, suggesting their refinement." N. McNaughton and D. Bannerman Dupret et al. (2013, p. 166, p. 168) 11.4.Iterative processing7

An initial iterative analogy
A key feature of our model of hippocampal operation is iterative interaction.Iteration is a general key to the production of major information processing output with minimal computational infrastructure.Modified output is returned to a network as input for further, progressive, modification.Fig. 6 provides an analogy from simple visual processing.It shows a small part of a simple network devised by Marr and Poggio (see Frisby, 1979, p. 150) that models "global stereopsis".
The key features of this network are: (a) the input patterns from the two eyes are "clamped" onto the input lines of adjacent sides of the square lattice; (b) lateral facilitation and lateral inhibition within the network then change the activity at each of the nodes where the two types of input line intersect; and, most importantly, (c) the result of this change is, in turn, fed back to the network and the process is iterated (potentially hundreds of times) until the network settles into a stable state at which point the viewer sees the embedded object ("B.I.S" in the example shown if viewed through a stereoscope).Importantly, because of the use of iteration, only a few (2 or 3) layers of neurones are required for all of these calculations for any particular point in space.Although we are using this model purely as an analogy for any form of iterative conflict resolution, it is interesting to note that the human brain mechanisms supporting global stereopsis appear to involve theta-frequency chunking (~4 Hz, Deng et al., 2023).
We would see the hippocampus as interacting, in computational but not functional terms, in roughly the same way.So, the areas that send it information via the multiple parallel loops shown in Fig. 5 then receive modified information back from the hippocampus.This iterative processing is packaged by rhythmical slow activity ("theta") that seems tuned to the round-trip time for prefrontal-hippocampal loops (Miller, 1989(Miller, , 1991)).The same is true of the somewhat different theta-packaged eye blink conditioning circuits (Berry and Hoffmann, 2011;Cicchese and Berry, 2016).
This iterative perspective is consistent with the idea that cell fields may reflect "attractor networks", which have the key property of being able to produce sharp output transitions at a point in the middle of a smoothly varying transformation of the stimulus input to the network (Leutgeb et al., 2005).While linked to pattern separation and pattern completion (in the sense of recovering the original from a degraded input) this property would also clearly be important for decision making under conditions of goal conflict or for recovery of a key memory against a background of interference.

Hippocampal iteration
When the hippocampal formation receives information about only one prepotent goal, it is, effectively, in "just checking mode".When, in addition, it receives information about a second conflicting goal, it detects the multiplicity and effectively enters "control mode".Note that no real change in the nature of goal processing occurs between these two modes.At all times information enters a comparator (in CA3 or CA1) and is integrated but, when the summation of inputs passes some threshold, it also produces output.
The output from the hippocampal formation involves the return of information to those areas whose activity has given rise to conflictproviding a simple, quite general, affectively-negative biasing function.The effect is to increase the valence of affectively negative stimuli and associations of stimuli (memories) and, particularly with chronic conflict, reduce the valence of affectively positive stimuli.This output, therefore, shifts approach-avoidance conflict in the avoidance direction, suppressing approach, on both the current and future occasions.With retrieval from memory, it shifts correct target versus interfering targets conflicts in the correct direction suppressing the "approach" to the incorrect ones (flagged by emotionally negative "error" signals).The net result of this process of negative bias and suppression is determined, however, not by the hippocampal formation but by the results of its effects on its target structures at the current point in cycles of iteration.The details of how this is achieved determine our interpretation of the functional role of RSA/"theta".
The mechanism we postulate relies upon the concept of iterative networks.Using Marr and Poggio's (1976) model solely as an analogy, we can see the power and apparent sophistication that iteration can produce in the processing of information by even simple parallel networks of the type we postulate as linking the hippocampal formation with goal processing structures.Similar iterative circuitry is used by Hinton et al. (1993) for "clean up" units that adjust discrepancies between words and semantic units in a neural network simulating reading.
Our picture is of successive, dynamic, adjustments to the relative capacity of one goal or another of what may be a large set to capture motor output systems.At the start, by definition, at least two incompatible goals will be similarly activated (or else the most strongly activated would immediately generate a response).Output for the next iteration will cause alteration of motivational values and, potentially, release of new information from memory; and so adjust the balance between the goals.This will be repeated until the conflict is resolved.
The common computational core postulated in our model as lying at the heart of hippocampal function, whether this is manifest in spatial cognition, memory or anxiety, comprises a comparator, to detect conflict between concurrent activation of incompatible goal-seeking mechanisms, coupled to iterative networks that link the hippocampus to memory stores (in the temporal lobes), complex motor programming circuits (in the basal ganglia and frontal cortex), "fixed action pattern" circuits (in, e.g., the amygdala and hypothalamus) and thalamocortical perceptual systems, with all of which the hippocampus interacts so as to increase the negative weighting of items with greater need for suppression (including those that produce interference in memory tasks).
The hippocampus produces such an increase in negative weighting at every cycle of a series of computations that progressively increase bias (and hence suppress goals) until one or other goal is sufficiently predominant to take control of the motor mechanisms.This increase in negative bias has two consequences.It affects current motor output directly, and it affects future motor output indirectly through its biasing of associations (including those that are just being formed).Additionally, while its iterative operations continue to fail to resolve the conflict, the hippocampus produces additional output that engages exploratory mechanisms designed to resolve the conflict through the obtaining of new information about the environment (thus preventing the animal from getting locked into an impasse).

need for phasic control
In Marr and Poggio's model, since it is carried out on a digital computer, each cycle of computation is quite discrete.However, the effects of opening ion channels on hippocampal neurones can change neuronal excitability for periods of time that equal or exceed the time it takes information to circulate from the hippocampus to its targets and back (Miller, 1989(Miller, , 1991)).The effect of RSA/"theta" (and particularly the imposition of a regular period of inhibition on hippocampal cells by the medial septal input) will be to quantize hippocampal processing and thus reduce the chances that the effects of information received on a previous cycle interfere with the comparator's assessment of inputs on the current cycle.From this follows the conclusion that RSA/"theta" will be important, particularly for complex or evenly balanced conflicts; but, since it only increases the acuity of processing in the temporal domain, RSA/"theta" will not be essential for the hippocampus to perform at least some functions where a mere degradation in performance is not crucial.
Here we should note also that the extensive circuitry controlling the frequency of RSA/"theta" and gating its occurrence implies that particular levels of that activity will not be universally beneficial (or RSA/"theta" would occur all the time at a fixed level).Nor, therefore, should we assume that the interaction of the hippocampus with other brain structures will always be beneficial.Certainly, there are some behavioural tasks (e.g., two-way active avoidance in a shuttlebox) where lesions of the hippocampus produce an improvement in performance in percent correct (if not adaptive) terms.

The full model
Much of what we have said above has treated the hippocampus as operating similarly along its length, which seems reasonable in terms of its lamellar structure (Fig. 2B, C).However, we have already noted that projections from hippocampal fields and its septo-temporal extent map topographically into subcortex (Fig. 3A) and that its primary cortical outputs (from CA1 and subiculum) go to anterior, goal processing, areas of prefrontal cortex (Fig. 2A).We also noted that medial frontal cortex closes the Papez circuits to the hippocampus via topographically organised connections relayed in the medial septum (Fig. 3).
In Fig. 7, we extend these topographic ideas to the nature of the goal processing in different parts of the hippocampus.With a few caveats (Houser et al., 2021) we can see the iterative computations being carried out by much the same algorithm in all lamellae.However, it is the essence of the topographic organisation that the inputs to different parts of the hippocampus will carry different information; and the outputs affect different goal systems.Our suggestions, based on Moser and Moser (1998); (see also Strange et al., 2014), is that the septal pole of the hippocampus processes largely dorsal trend information, detecting conflicts where goals differ in their situation (e.g.arm colour in a T-maze), while the temporal pole processes largely ventral trend information, detecting conflicts where goals differ in their motivation (e.g., food versus shock) but not situation.Intermediate portions would detect conflicts where both situation and motivation differ.

Overview
We have presented a picture of hippocampal cell firing as mirroring the activation of cell assemblies that encode goal representations."Encode", here, sounds computational and cognitive.But we define a goal as a situation-plus-motivation compound ( §2), with distributed coding in the dorsal and ventral cortical streams (respectively, §3); where the motivation component entails a tendency to action.The hippocampus, therefore, receives available-goal-related information that will be motivationally positive or negative and can be more stimulus-or more response-loaded ( §4).Indeed, with eyeblink conditioning ( §5), some hippocampal fields mirror motor output (which is unidimensional) as much as they do perceptual or cognitive input.Functionally, the hippocampus becomes important when there are concurrently available or potential incompatible goals, with the hippocampus acting to resolve this goal conflict.
One obvious advantage of this account compared to many other theories of hippocampal function is that it readily provides an explanation for the effects of hippocampal manipulations on tests of unconditioned anxiety responses (e.g., reduced freezing to but superior avoidance of a cat, Blanchard and Blanchard, 1972); with ventral but not dorsal hippocampal lesions affecting unconditioned anxiety but not unconditioned fear (Pentkowski et al., 2006).In a range of different tasks, while initiation of unconditioned immobility is not greatly affected, maintenance of immobility (i.e.behavioural inhibition) which is required as part of the anxiety response is reduced by hippocampal lesions.All of these results are less well explained by either purely spatial or purely mnemonic accounts than by a goal conflict account.
Furthermore, the goal conflict account provides a simple explanation of the inhibition-selective differences shown in Table 1.Many accounts, particularly memory-based accounts, would not obviously predict an effect of hippocampal manipulations on inhibition of responding but, at the same time, predict no effect on the original acquisition of the response.
Notably, the relational memory account explicitly excludes "studies on orientation, distraction, exploration, motor patterns, operant schedules, emotion, and species-specific behaviors" (Eichenbaum et al., 1994, p. 450), despite the fact that hippocampal lesions can often have quite profound effects on these behaviours.It does however provide an account both of impaired reversal learning generally, and improved olfactory reversal: "whether the information must be represented in a flexible way to permit its expression in novel testing situations [i.e., be declarative] or whether, instead, the representation need only support performance in repetitions of the original learning situation [i.e., be procedural]" (Cohen and Eichenbaum, 1993, p. 265, see pp. 137-141 for discussion of the olfactory case).But this does not appear to us (McNaughton, 1994) to work for simple runway extinction, passive avoidance, 2-way active avoidance, nor for the innate reactions described above.
The eyeblink data also make clear that the hippocampus receives much goal information (during simple conditioning and delayed conditioning, which are insensitive to hippocampal lesions) for which it does not generate output.However, when inhibition of output from the goal representation is required (e.g., during trace conditioning or reversal learning, which are sensitive to hippocampal lesions) it detects the conflict between pre-planned action and currently-required inaction and produces output that feeds back to the original goal representation.The same is true of simple discrimination and memory experiments where the hippocampus appears to respond to available or retrieved goals ( §6) but is not functionally involved unless discrimination is successive or subjected to reversal or memory is subject to interference ( §11.1).
None of these data fit into the classic description of hippocampal cells as having "place fields".We argue ( §7) that "place field" is a misnomer and hippocampal cell fields are incompatible with a true spatial metric.We discuss: conditional place fields; multi-place fields; non-spatial fields; non-local representations and extra-field spiking; remapping; non-spatial remapping; and field-specific remapping.None of these are compatible with a "conventional" view of a spatial map and many are simply unrelated to space as such at all.Further, we ask "Where is the map?".The organisation of firing fields within the hippocampus has none of the required properties for a true spatial read out.It has been suggested (Sanders et al., 2020) that "remapping" involves hidden state inference reflecting subjective beliefs not external reality.This is close to our view of a goal (which is an internal construct).
A spatial/cognitive map perspective also leaves unclear why the hippocampal formation should consist of a superficially linear chain of structures (Fig. 5) with multiple parallel connections (including phasic rhythmic input).We briefly covered ( §9) the main attempt to determine the flow of processing through this chain (Vinogradova, 1975, 1995, Fig. 7. Dorsal-ventral topography of the primary connections of the hippocampus.Attractors (green) and repulsors (red) are activated by situations (S a … S z ) and send output that can generate specific motor programs guided by the situation (M i S a …M vi S z ) with efference copies to the hippocampus.The dorsal trend of frontal cortex (e.g., ACC in medial frontal cortex) carries primarily situational information to dorsal hippocampus; while ventral frontal cortex (e.g., ILC in medial frontal cortex) carries primarily motivational information to ventral hippocampuswith the goal representation distributed between the trends both in cortex and hippocampus.The dorsal trend detects conflict that depends more on situation (go left/go right) than motivation; and the ventral detects conflict that depends more on motivation (food/shock) than situation.The middle portion of the hippocampus processes conflicts where both situation and motivation differ between the goals.The 3 parts are not equal and there may be others.Based on Fig. 1 in Moser and Moser (1998); (see also Strange et al., 2014).Note that internal connections of dorsal and ventral hippocampus can differ (Houser et al., 2021).There is amygdala input to the whole length (strong/thick arrows to ventral, weak/thin arrows elsewhere), so motivational/goal processing occurs at all points but greater input to ventral hippocampus means that it is more to do with motivation conflict than situation conflict; and vice versa for cortical input to dorsal.

Fig. 1 .
Fig. 1.Examples of a positive goal (an attractor) and a negative goal (a repulsor).The arrows show the rat's attraction to dark (A) and repulsion from bright light (B).See McNaughton et al. (2016), Fig. 2, p. 28) for the combining of negative and positive goals to create a trajectory.
Fig. 2. Hippocampal outflow.A. 3-D schematic of location of the hippocampus as a bridge between cortex and subcortex in the rat brain (adapted from the human equivalent inGastaut and Lammers, 1961).There is important outflow (arrows) both via the subiculum to the temporal cortex and via CA3 and the fimbria to the hypothalamus.B. Diagram of the hippocampal cell fields (courtesy of S. Kerr) within a slice (or functional lamella).C. The hippocampus can be viewed as a stack of essentially identical slices(Andersen et al., 1971;Andersen et al., 2000) with the circuitry of each slice as shown in B. D. The septal-temporal axis and these fields can be viewed, unfolded, as a simple flat map that is topographically connected (via the septum) to the hypothalamus(Risold and Swanson, 1996).Note that this topographic variation in connections from the hippocampus is matched by similar topographic variation in inputs to the hippocampus from both specific prefrontal (see Fig.3) and entorhinal inputs, and also more diffuse neuromodulatory inputs.It is also mirrored by variation in neuromodulatory influence and hormonal regulation along the septotemporal hippocampal axis(Lathe, 2001; see alsoStrange et al., 2014).Thus, common circuits will execute a common algorithm but operating on different forms of information, on potentially different scales of time and space, and particularly with different emotion-related functions.Figure and legend adapted from McNaughton and Gray (2024), Fig. 1.4).
Fig.4.Schematic depicting essential and modulatory circuitry for trace eyeblink classical conditioning (EBCC) and activity profiles underlying theta-related behavioral effects.Averaged local field potentials (LFPs) for hippocampus (HIPP), interpositus nucleus (IPN), and Larsell's hemispheric lobule 6 of cerebellar cortex (HVI) show robust, time-locked theta oscillations at 6-7 Hz during the trace and post-US periods for trials triggered in the explicit presence of naturally occurring hippocampal theta (T+).Furthermore, these theta oscillations reset at short latency following the stimulus events such that the perturbation of theta is minimized.In contrast, animals that received trace EBCC trials in the explicit absence of theta (T) display less robust hippocampal theta during the trial period and little to no theta rhythmicity in the cerebellar LFPs.Under optimal hippocampal states, in which theta is present, GABAergic and cholinergic projections from medial septal nucleus (MSN) provide pacemaker inputs to HIPP to regulate theta activity.HIPP theta activity, in turn, is known to modulate prefrontal cortex (PFC) unit firing properties.Contiguity of PFC-driven trace-related activity via lateral pontine nuclei (LPN) with US-related information from inferior olive (IO) within the cerebellum is thought to drive learning.Rapid behavioral acquisition occurs with theta phase-locked rhythmic processing in cerebellum (IPN and HVI) and such patterns are disrupted in animals that learn more slowly.Large and small tick marks below LFP traces indicate onset and offset of conditioned stimulus (CS) and unconditioned stimulus (US), respectively.Additional abbreviations: CF, climbing fibers; CR, conditioned response; MF; mossy fibers; PF, parallel fibers.Figure and legend taken from Berry and Hoffmann (2011) with permission.

Fig. 5 .
Fig. 5. Main circuits and concepts of Vinogradova's model updated from Figure 10 ofVinogradova (2001) based onMcNaughton and Vann (2022).Presentation of a novel stimulus that requires orienting (illustrated by a tone,) will immediately and non-specifically generate an arousal reaction in the reticular formation (RF) and so in the ascending theta (θ) circuit (blue shading and lines).This will cause firing of CA3 cells.Repetition of the stimulus (absent any links with reinforcement) will result in habituation of the CA3 response as a result of development of a planned response (including the plan to ignore the stimulus and do nothing) in frontal circuits that receive specific information about the stimulus.Information that appropriate action/inaction is planned is passed via the informational circuit to the entorhinal cortex (EC) and then dentate gyrus (DG) which sends input to CA3 that cancels the CA3 response, generating habituation.The change in CA3 response is passed round the regulator circuit to lateral septum (LS) and then median raphe (MR), which inhibits the production of the orienting response.EC, dentate gyrus (DG), CA3, CA1, subiculum (SUB), and retrosplenial cortex (RSp) connect unidirectionally.CA1, SUB, and RSp build successive slower and more elaborate models that further damp down orienting providing better integration of the eliciting stimuli and delay lines.SUB and RSp send output to the mammillary bodies (MB) and the anteroventral thalamus (AVT) that is then passed to dorsal and ventral prefrontal (PFCd, PFCv) then perirhinal (Peri) and parahippocampal (Para) cortex complete the informational circuit in EC.RSp also sends output to memory processing areas that interact bidirectionally with PFC to generate plans.The circuitry is greatly simplified and details of the multiple parallel Papez-like circuits involved are provided byMcNaughton and Vann (2022).

Fig. 6 .
Fig.6.Marr and Poggio's computational solution to the global stereopsis problem (for more detailed description, including simulation outputs, seeFrisby, 1979, p. 150-151).A. Random dot stereograms.B. A slice taken horizontally at the point marked by the filled square at the left of the left-hand image in A. C. Small portions of B taken from the equivalent areas for the left and right eye and superimposed as input onto a computational lattice.The excitatory and inhibitory connections within the lattice then, over a large number of cycles, eliminate activity in the nodes representing incorrect matches, leaving the activity representing correct matches.Note that it is not the initial level of activity in any node that determines whether it will remainrather it is the relation between it and the recursive changes in activity within the network.D. The resultant "ground" (left) and "figure" (right) separated one from the other.
Fig.7.Dorsal-ventral topography of the primary connections of the hippocampus.Attractors (green) and repulsors (red) are activated by situations (S a … S z ) and send output that can generate specific motor programs guided by the situation (M i S a …M vi S z ) with efference copies to the hippocampus.The dorsal trend of frontal cortex (e.g., ACC in medial frontal cortex) carries primarily situational information to dorsal hippocampus; while ventral frontal cortex (e.g., ILC in medial frontal cortex) carries primarily motivational information to ventral hippocampuswith the goal representation distributed between the trends both in cortex and hippocampus.The dorsal trend detects conflict that depends more on situation (go left/go right) than motivation; and the ventral detects conflict that depends more on motivation (food/shock) than situation.The middle portion of the hippocampus processes conflicts where both situation and motivation differ between the goals.The 3 parts are not equal and there may be others.Based on Fig.1inMoser and Moser (1998); (see alsoStrange et al., 2014).Note that internal connections of dorsal and ventral hippocampus can differ(Houser et al., 2021).There is amygdala input to the whole length (strong/thick arrows to ventral, weak/thin arrows elsewhere), so motivational/goal processing occurs at all points but greater input to ventral hippocampus means that it is more to do with motivation conflict than situation conflict; and vice versa for cortical input to dorsal.Abbreviations: DBBh/v = Diagonal Band of Broca, horizontal/vertical; EC = entorhinal cortex; GAS = Goal Attraction System; GRS = Goal Repulsion System; MS = medial septum; ACC = anterior cingulate cortex; ILC = infralimbic cortex.(a) Figure and adapted legend from McNaughton and Gray (2024).