Integration of the sensory inputs to place cells: What, where, why, and how?

COMMENTARY

Integration of the Sensory Inputs to Place Cells:What, Where, Why, and How?

Kathryn J. Jeffery*

ABSTRACT: The hippocampal place cells are a highly multimodal classof neurons, receiving information from many different sensory sources tocorrectly localize their firing to restricted regions of an environment. Evi-dence suggests that the sensory information is processed upstream of thehippocampus, to extract both angular and linear metric information, andalso contextual information. These various kinds of information need tobe integrated for coherent firing fields to be generated, and the presentarticle reviews recent evidence concerning how this occurs. It is con-cluded that there is a functional dissociation of the cortical inputs, withone class of incoming information comprising purely metric informationconcerning distance and orientation, probably routed via the grid cellsand head direction cells. The other class of information is much more het-erogeneous and serves, at least in part, to contextualize the spatial inputsso as to provide a unique representation of the place the animal is in. Evi-dence from remapping studies suggests that the metric and contextualinputs interact upstream of the place cells, perhaps in entorhinal cortex.A full understanding of the generation of the hippocampal place represen-tation will require elucidation of the representational functions of theafferent cortical areas. VVC 2007 Wiley-Liss, Inc.

KEY WORDS: place cells; grid cells; sensory integration; space;context

INTRODUCTION

Since O’Keefe’s discovery of place cells, in the early 1970s, researchattention has been fairly evenly divided between trying to understandwhat these cells are for and in understanding how their firing propertiesarise in the first place; that is, in how they determine where they are. Evenbefore their place-specific firing properties were elucidated, it was recog-nized that hippocampal neurons respond to sensory information arrivingfrom a variety of different modalities (Ranck, 1973). Once it was appreci-ated that these neurons were interested in ‘‘place,’’ the question arose as tohow they could construct a place representation from such varied informa-tion. What form does this information take? Is the ‘‘placiness’’ of placecell firing locations (the place fields) inherent in the inputs, or does itsomehow arise from the computations that place cells perform? Indeed, isspace special to the hippocampus at all, or do place cells just look placey

because rats experience spatial stimuli more than theydo other kinds of stimuli?

To answer this question it is necessary to have a clearidea of what kinds of information place cells receive,and of what they do with this information once theyreceive it. The present article reviews data on how placecells integrate their sensory inputs and examineswhether these inputs are specifically spatial or whetherthere is another, superordinate category of descriptionthat encompasses both the spatial and nonspatial inputsknown to influence place cells.

Adding fuel to the spatial/supraspatial debate hasbeen the emergence in recent years of many reports ofnonspatial determinants of place cell activity, raisingthe question of whether place is really the fundamentalcorrelate of place cell activity at all. Observed nonspa-tial determinants of place cell activity have fallen intotwo broad classes. The first comprises those kinds ofstimuli which although not strictly spatial (in the senseof ‘‘metric’’: see below) in themselves, nevertheless maycontribute to the fleshing out of a place representation.An example might be the ‘‘color’’ of the environment,its texture, odor, or even (so it now appears) the expect-ations or intentions of the rat. These stimuli have incommon the property of being stable—that is, continu-ous in time and/or place—and of defining the environ-ment in a way that may assist the animal not only inknowing where it is but also what occurrences to expectthere or what actions it should perform. The secondclass of stimuli is a much more ill-defined and conten-tious collection, consisting of discrete events such astone conditioned stimuli (Berger and Thompson,1978), reward (Breese et al., 1989), shock (Moita et al.,2003), and discrete odor cues (Wood et al., 1999). Thedegree to which these discrete stimuli modulate placecell activity independently of place is still debated(O’Keefe, 1999), as is their functional significance.

For the remainder of this article, the term ‘‘spatial’’will be replaced with ‘‘metric’’ to remove any ambiguityabout what is meant by this class of stimuli. Asdescribed further in the next section, ‘‘metric’’ refers tosome kind of distance measure, either linear or angular.Thus, by ‘‘metric inputs’’ to place cells is meant ‘‘inputsto carrying information about distance and/or direc-tion’’. Nonmetric inputs are those that do not carrysuch information. It will be argued that there exists a

Department of Psychology, Institute of Behavioural Neuroscience, Uni-versity College London, London, United KingdomGrant sponsor: Biotechnology and Biological Sciences Research Council.*Correspondence to: Dr K.J. Jeffery, Department of Psychology, Instituteof Behavioural Neuroscience, University College London, 26 BedfordWay, London WC1H OAP, United Kingdom. E-mail: [email protected] for publication 1 May 2007DOI 10.1002/hipo.20322Published online 5 July 2007 in Wiley InterScience (www.interscience.wiley.com).

HIPPOCAMPUS 17:775–785 (2007)

VVC 2007 WILEY-LISS, INC.

functional and anatomical dissociation between metric inputsand (what we shall call) contextual, nonmetric inputs with theformer being selected, or ‘‘gated’’, by the latter. The article willconclude that evidence supports a functional differentiation ofspatial vs. nonspatial inputs to dorsal hippocampus, with spatialinputs being primary.

METRIC INPUTS

A metric, in mathematics, is defined by the Oxford EnglishDictionary as the ‘‘binary function of a topological space whichgives, for any two points of the space, a value equal to the dis-tance between them, or a value treated as analogous to distancefor the purpose of analysis.’’ The critical thing about a metric isthus that it provides, by one means or another, informationabout the distances between two identified points in a givenspace. For place cells, ‘‘distance’’ is taken to mean either ‘‘Euclid-ean distance’’ or ‘‘angular distance.’’ In the case of a rat in a re-cording box, this information might consist of informationdetermining how far apart (in Euclidean space), say, two placefields are.

Although place fields undeniably occur in specific regions ofspace, it was never a foregone conclusion that these locationsshould be specified metrically. They might be specified by somenonmetric means: for example, by some conjunction of cuessuch as odors, visual cues, or a mixture that only occurs at a par-ticular place. Alternatively they might be determined topologi-cally, so that a given place field would primarily be localized bythe fields around it, with preservation of adjacency relationshipsbetween all the fields in a given environment, but no explicitrepresentation of distance per se. In this case, there would ofcourse need to be some mechanism for anchoring the topologicalsheet of fields, as it were, to the real world, but this need not bea metric one. However it turns out, as described below, that agrowing body of evidence suggests that the environment doesindeed provide metric—distance and direction—information.This evidence falls into three main classes, the first weakly cir-cumstantial and the remaining two very strongly supportive ofthe notion.

1. Place cells are sensitive to metric transformations of theenvironment.

2. Place cells will accept replacement of one sensory stimulusby another, totally different kind of stimulus, as long asthese stimuli are equivalent (or isomorphic) in the metricdomain.

3. Cells in brain areas afferent to the place cells have angularand linear metric response properties.

Place Cells Respond Metrically to MetricTransformations of the Environment

The first indication that place field locations might be deter-mined by metric information came from Muller and Kubie’s

classic scaling experiment (Muller and Kubie, 1987), in whichexpansion of an environment was found to cause an expansionof place field size. Interestingly, however, the scaling was notquite proportional, suggesting an influence of other cues thansimple environment size.

In a more tightly-controlled test of scaling, O’Keefe and Bur-gess (1996) manipulated the metric properties of a recordingenvironment—a wooden box—in each of the two dimensions,configuring the recording box as a small square, a rectangle ori-ented east–west or north–south, or a large square. Because allthe box configurations used the same physical walls, the changeswere subtle, and so the place cells did not reorganize their firingas they do if large, noticeable changes are made to an environ-ment. Place cells also reacted in a subtle way, usually just shiftingtheir fields slightly. Different cells reacted in different ways, as ifnot all of the cells were privy to the same information. By andlarge, fields tended to be most influenced by the walls closest tothem so that they ‘‘followed’’ those walls in preference to themore distant ones. Taken together, the data suggested that agiven cell processed information primarily from a subset ofnearby walls. O’Keefe and Burgess proposed that a given cell wasin receipt of metric information from two or more orthogonalwalls in the form of a Gaussian tuning curve peaked at a set dis-tance close to each of those walls. A combination of two suchGaussians from two orthogonal walls proved very effective inmodeling the way in which the fields reacted to the changes inthe environment.

Although the O’Keefe and Burgess model postulated an ex-plicitly metric input to place cells, the stretchy-box results mightstill be explained by reference to some kind of sensory inputfrom the walls that varies linearly with distance but does not ex-plicitly encode distance (that is, could not in itself be used toperform metric computations such as trigonometry). Arguingagainst this is the second indication that place field inputs havean explicitly metric component.

Place Cells Will Accept Replacement of OneSensory Stimulus by Another Totally DifferentKind of Stimulus if These Stimuli AreEquivalent in the Metric Domain

It is a now well-established finding that place cells will acceptstimulus substitution if the stimuli are metrically equivalent. Anexample of what this means is shown in the angular domain inFigure 1. A place field, recorded in a rotationally symmetricalenvironment, ‘‘followed’’ rotation of the visual directional cue, alighted cue card, irrespective of signals from the rat’s internalsense of direction. If the card was removed so that no visualpolarizing landmarks remain, the cells now followed, equallyreadily, rotations of the rat itself. Thus, the cells can use either avisual landmark or the internal ‘‘direction sense’’ of the rat—both polarizing stimuli—to orient their fields. Similarly,although place cells generally follow visual polarizing cues, in asloping environment they can use the slope as an orienting cue ifthe visual cue is removed (Jeffery et al., 2006). This is despite

776 JEFFERY

Hippocampus DOI 10.1002/hipo

the fact that slope is determined via touch and proprioceptionwhereas the cue card is processed visually.

Stimulus substitution has also been found in the linear domain.Quirk et al. (Quirk et al., 1990) found that place fields that werewell established in a lighted environment could persist in the darkif the animal had been in the environment when the lights wereswitched off but not if it had not been removed first, and Gothardet al. found that place fields would be determined by movementcues (by a process commonly known as ‘‘path integration’’, dis-cussed below) if the rat was far from a visual landmark but visualcues if it were close to it (Gothard et al., 1996).

This switch between two very different types of sensory infor-mation, such as between vision and path integration, is one ofthe most compelling arguments to be made for the metric natureof (at least some of ) the inputs to place cells. Put another way, inthe metric domain, and only in the metric domain, vision andmotion cues are equivalent, or isomorphic, and thus able to sub-stitute for each other. Such data argue very convincingly for afundamentally metric basis to place field inputs.

Cells in Brain Areas Afferent to the Place CellsHave Linear and Angular Metric Properties

The final demonstration that place cells receive metric informa-tion comes from observations that cells in brain areas surroundingthe hippocampus, and which are either directly or indirectly afferentto it, have intrinsic metric properties that seem to be independentof environmental cues, except insofar as external cues are needed toanchor the metric to the real world in a reproducible way. Theseobservations have been made in both angular and linear domains.

In the angular domain, the relevant discovery is that of headdirection cells, made by Ranck and Taube in the late 1980s andfirst published in 1990 (Taube et al., 1990a,b). Head directioncells fire only when the animal’s head points in a particular direc-tion, and are strongly controlled by visual landmarks. Like placecells, and unlike more primary sensory neurons, the cells can

switch easily between vision and path integration (Taube, 1998)and thus seem to respond to information that is isomorphic inthe (angular) metric domain.

Evidence for linear metric encoding is only very recent, andstems from the remarkable discovery made by Hafting et al. of aclass of cells in the entorhinal cortex known as ‘‘grid cells’’(Hafting et al., 2005). Grid cells occur in dorso-medial EC, aregion of cortex afferent to the place cells, and while they haveplace fields that individually resemble hippocampal place fields,these fields are circular (rather than ovoid as place fields generallyare) and they recur at regularly spaced intervals across the envi-ronment. These repeating fields form a pattern long familiar tophysicists and mathematicians as hexagonal closest packing (thisbeing the most efficient way of tiling a planar surface withcircles). The pattern of fields of a given cell forms a grid coveringthe whole of the traversable environment, and hence the name‘‘grid cells’’ for these neurons. The constant interpeak interval ofgrid cells is independent of the environment (with some caveats,detailed later) and must therefore derive from some internalproperty of the network, rather than from external landmarks(though external cues must subsequently attach the grid to theoutside world).

The finding of something looking very much like a metricgrid (akin to a sheet of graph paper) in a cortical region directlyafferent to place cells strongly suggests that place cells receive in-formation conveying something about the metric properties ofthe environment the animal is in. Depending on which dorso-ventral region of EC is examined, the interpeak spacing can varyfrom about 30 cm dorsally to some number of meters in themost ventral regions. Interestingly, in a given animal in a givenenvironment, the orientation of the grids appears to be the same,although the exact location of peaks varies form cell to cell.Thus, the grid cell network as a whole seems to receive the sameorientational information. The obvious source of this informa-tion is the nearby head direction network, known to project tothese areas (Sargolini et al., 2006).

INTEGRATION OF METRIC CUESWITH EACH OTHER

Having established that place cells appear to receive metricinputs, the question arises as to how the cells integrate theseinputs. There are two classes of metric input, directional and lin-ear, and so the relevant interactions comprise directional inputswith each other, linear inputs with each other, and directionalinputs with linear inputs. At this stage, the discussion is aboutpurely metric information—the question of how nonmetric in-formation (including that from objects and landmarks) is incor-porated into the representation will be dealt with later.

Integration of Directional InputsWith Each Other

Directional inputs are those that provide information aboutthe orientation of the environment, or of the rat within the envi-

FIGURE 1. Stimulus substitution by place cells in the direc-tional domain [see (Jeffery and O’Keefe, 1999) for details]. Placecells were recorded as a rat foraged in a square box. Between trials,the rat was covered and rotated, and the visual cue was either rotatedby a multiple of 908 or removed altogether. In the presence of thecue card, the place field of the cell (shown in smoothed contour-plotform) always rotated with the card, regardless of the rat’s internaldirection sense (top row). In the absence of the card, however, thefield rotated with the rat’s internal sense of direction (bottom row).

INTEGRATION OF THE SENSORY INPUTS TO PLACE CELLS 777


ronment, and probably, as discussed above, act via the headdirection system. Two main classes of directional input havebeen described: visual and movement-related (sometimes calledidiothetic).

Visual landmarks seem to be powerful cues for orienting theplace and head direction cells. In the early studies of place cells,it was found that rotation of a prominent cue card within anotherwise rotationally symmetrical environment would causeconcomitant rotation of place fields (Muller and Kubie, 1987).Later studies found that cue card rotation also caused head direc-tion cells to rotate their firing orientations (Taube et al., 1990b),and that place cell and head direction cell rotation rarely, if theyever become uncoupled (Knierim et al., 1995; Hargreaves et al.,2007, this issue), leading to the proposition that place fieldsassume the orientation they do because of the head directioncells. Head direction cells respond better to polarizing visualstimuli that are located at some distance from the rat (Zugaroet al., 2001), and it is assumed that the directional system hasevolved to prefer cues that are more reliable indicators of globalorientation (distant cues being better because their relative angu-lar position changes little with the animal’s movements).

Relatively little is known about how information from cueensembles is integrated in the directional domain. In a studyusing two cue cards, Fenton et al. found that moving the twocards apart by a small amount caused a topological deformationof place fields, with all fields being influenced by both cards to agreater or lesser extent, but a given field being most affected bythe card closest to it (Fenton et al., 2000). This is an interestingexperiment and raises the question of how head direction cellswould react to such a manipulation—would head direction cellfiring directions also ‘‘pull apart’’, or would the representationmaintain coherence, as predicted by attractor models of the headdirection system (Zhang, 1996)? The latter would necessitate apartial uncoupling of the head direction and place cells, which,although surprising, is not inconsistent with recently reporteddata from double cue dissociations (Yoganarasimha et al., 2006),in which head direction cell coherence was maintained but placefields dissociated.

Directional cues occur in modalities other than vision. Earlierwe discussed two other kinds: movement-based cues (path inte-gration), revealed by slowly rotating an animal in darkness andobserving that place fields rotated too, and terrain slope, whichprovides vestibular and proprioceptive information. Conflictexperiments provide useful information about how these differ-ent forms of information interact. When vision and path integra-tion are placed in conflict, place cells generally prefer to followthe visual stimulus (Jeffery, 1998). Two exceptions to this rulehave been noted. The first is that if the rats saw the cue cardmove and the movement was large (i.e., introduced a big con-flict), the cells are less likely to follow it (Rotenberg and Muller,1997). The second is that if the rat had previously seen that thecue card could move, then the cells are subsequently disinclinedto follow it even if the conflict is now introduced out-of-sight ofthe rat (Jeffery and O’Keefe, 1999). These experiments showthat the place system somehow weights the two sources of infor-mation and makes a decision about which is more, as it were,

‘‘trustworthy.’’ Furthermore, this weighting is plastic and can bemodified by experience.

It remains to be determined how and where these multifarioussources of directional information are integrated, but it is likelythat the above conflicts are resolved in the head direction system,which seems, as far as we can tell, always to generate an unam-biguous, holistic directional signal.

Integration of Linear Inputs With Each Other

Place fields have a two-dimensional structure, and so theirlocations must be determined by at least two one-dimensionalmetric parameters. These might consist of a vector together withthe distance from some origin (a landmark, say), and thus takethe form of a polar coordinate system. Alternatively, they mightconsist of two distance parameters aligned orthogonally, andthus comprise a Cartesian coordinate system in which the twoalignment directions form a reference frame, analogous to x andy axes. Although a polar coordinate system might seem more in-tuitive for a biological agent, particularly one known to possessdirectionally selective neurons, evidence in fact is more support-ive of the latter.

Support for the Cartesian hypothesis of place field localizationfirst came from O’Keefe and Burgess’s (1996) stretchy-boxexperiment, described earlier, in which it appeared that placefields were formed by the intersection of distance inputs comingfrom orthogonally aligned walls. In an extension of this model,Hartley et al. showed that a similar effect would arise from mul-tiple inputs from irregularly oriented walls (Hartley et al., 2000),and incorporation of plasticity into their model accounts for de-velopment of fields with experience (Barry and Burgess, 2007,this issue). The implication is that the underlying coordinateframework is generated by the intersection of linear metrics,rather than combination of a single linear metric together withan angular metric, that is, it is Cartesian rather than polar.

In a test of the polar vs. Cartesian hypotheses, we progressivelydeconstructed a four-walled recording environment until only asingle landmark and a single directional cue remained. Theexperiment was conducted in near-darkness and the rat was con-fined to the centre of the room by a ‘‘noise boundary’’ (i.e., itwas punished with aversive white noise if it ventured beyond theboundary: Barry et al., 2006). Figure 2 shows what happenedwhen the walls were removed one by one: place fields progres-sively disintegrated. Although the remaining features carriedenough information to uniquely define any point in a polarcoordinate-defined space, this information was apparently insuf-ficient to define stable place fields (with the occasional rareexception, see below). This suggests that the coordinate system isnot inherently polar.

The wall-removal experiment above provides some supportfor O’Keefe and Burgess’s boundary vector model, inasmuch asremoval of walls caused place fields to break down. It should benoted, however, that in a number of cases a field was able—atleast for a short time—to be supported by a single wall, andoccasionally, though very rarely, even a single pillar, a findingthat remains to be explained.

778 JEFFERY


In an attempt to isolate the individual inputs—the putativeboundary vectors—to place cells, an additional experiment wasconducted in which one of the walls of the four-walled box waspulled out from the others just enough to let the rat escape intothe surrounding room (Barry et al., 2006), thus enabling a com-parison of the enclosed space (surrounded by four walls, albeitone slightly separated from the rest) and the open space (eachpart bounded by a single wall). It was thought that this proce-dure might uncover the component vectors driving a given placefield. In one case it seemed as if this might be so (Fig. 3 Cell 3),but generally, cells were reluctant to fire outside the box, whilecontinuing to maintain their fields inside it. It thus seems thatplace cells prefer to fire in enclosed spaces. One possible explana-tion for this is that the boundedness offers more possibilities forsummation of inputs to above firing threshold. This is a far fromproven hypothesis, however. It may also be that the intrabox fieldis somehow able to suppress other fields in the connected spaceoutside the box by a nonlocality mechanism discussed later.

Where do the above-described linear metrics come from? Inother words, how does a place cell determine how far the animalis from a given boundary? In a familiar environment, much ofthis distance information may be visual, for example, the animalmay know from experience how high the walls are, and thus beable to use their apparent height to infer their distance. In anovel environment, however, or in darkness, it must be able touse something else. A great deal of evidence now suggests thatthis ‘‘something else’’ comprises movement cues generated as theanimal moves through space; this is the process of ‘‘path integra-tion’’ referred to earlier [see (Etienne and Jeffery, 2004), forreview]. Path integration has been shown to influence metric

integration in the linear domain, and this was initially suggestedby the well-established observation that place fields persist in theabsence of vision (Quirk et al., 1990; Save et al., 1998). It wasshown explicitly by Gothard et al. (Gothard et al., 1996), whorecorded place cells as rats shuttled on a linear track and foundthat fields were determined by the closest wall, even if this wasthe one behind the rat. However, it was not until the discoveryof grid cells (Hafting et al., 2005), however, that the importanceof path integration as the basis of the hippocampal metric reallybecame established beyond doubt, because it seemed that theconstancy of the grid cell metric across environments could onlybe established by an internal, movement-based process. In otherwords, because the grid spacing of a grid cell occurs at its charac-

FIGURE 2. Degradation of place fields as their metric deter-minants are progressively removed. (A) A four-walled environmentwas deconstructed, wall by wall, until only a single wall remained.The quality of the place field steadily deteriorated too. (B) For apopulation of cells, place field quality, as measured by compactness(are over which the cell fired, left) and coherence (degree to which

the cell’s activity was concentrated in one place, right), was relatedto the number of walls in the arena. This supports the notion thatplace fields are determined by summation from boundaries, ratherthan from polar-coordinates requiring only an origin and a direc-tion. Adapted from Barry et al., 2006.

FIGURE 3. Recording of three place cells in a closed (leftmaps) and a surrounded but slightly open (right maps) spacedefined by four walls. Place fields tended to be compact within thebounded area. When the south wall was moved to allow the rat toescape into the surrounding area, the cells continued to fire asstrongly or even more strongly within the bounded region, andonly occasionally was a field seen in the unbounded region (e.g.,Cell 3). This observation supports the boundary summation hy-pothesis of place cell integration. Adapted with permission fromBarry et al., Rev Neurosci, 2006, 17, 71–97, ' Freund PublishingHouse.



teristic spatial frequency even in a never-before-visited environ-ment, and even in the dark, this spacing could not be deter-mined by static environmental cues (which are arbitrary, andunique to a given environment) but only by dynamic, motion-derived information (which is constant for all environments).

Despite the evident importance of motion-derived cues forestablishing a grid cell’s grid, there is now overwhelming evi-dence that the grid cell firing pattern in a familiar environmentmust arise from a learned interaction between the internal metricand environmental cues. Some of this evidence was presented byHafting et al. in their original report, in which they showed thatrotation of a cue card caused a concomitant rotation of the grids.This reveals landmark control for the angular domain, at least.The fact that grids in repeated trials always have the same offsetwith respect to the environment also implies some kind ofanchoring process that aligns the grids to the environment in thesame way upon repeated trials. Another clue comes from the ob-servation that the grids tend not to be perfectly regular, and thatthe irregularities also persist across trials, suggesting that a givenpeak in a grid is somehow uniquely specified in a givenenvironment.

That this anchoring is learned, rather than hard-wired, hasnow been shown by an experiment of ours (Barry et al., 2007) inwhich grids that were established in an environment of one sizewere able to be deformed (or ‘‘rescaled’’) by subtle changes inthe dimensions of the enclosure (Fig. 4). This rescaling could beinduced in either direction: if the rat was familiarized with alarge square enclosure, compression of one of the axes to form arectangle caused a partial compression of the grid in that axis;conversely, if the rat was familiarized with a rectangle, expansionalong the short axis to form a square caused expansion of thegrid ion that axis, but not the other. Because the rescaling wasonly partial, the implication is that deforming the environmentsets up a conflict between the intrinsic path integration metricand the extrinsic learned boundary cues. Understanding howboth intrinsic and extrinsic cues can influence grid spacing intandem like this is a current priority.

Integration of Directional and Linear Inputs

Regardless of whether the coordinate system employed byplace cells is polar or Cartesian, it is necessary for there to besome kind of integration of directional and linear inputs. In thepolar case, this is because direction is one of the two componentsof the vector identifying locations, and in the Cartesian case it isbecause identical walls can only be distinguished on the basis oftheir directions. Additionally, the finding that place cells can‘‘path integrate’’ in order to update their position calculationsindicates that somewhere, angular and linear distance informa-tion must be processed together.

The head direction cells are the obvious source of directionalinformation to place cells, but because the head direction areasdo not project directly to the hippocampus, it has always beenunderstood that there must be some mediating structure. It nowseems likely that the grid cells are the site of this mediation.Grids have a clear and reproducible orientation in a specific envi-

ronment (Hafting et al., 2005) and are in receipt of direct inputsfrom head direction areas. In addition, many grid cells, mostprominently in the deeper layers of entorhinal cortex, have direc-tional firing properties, and many straightforward head directioncells are also found in these areas (Sargolini et al., 2006). Mostintriguing of all, inactivation of hippocampus, which feeds backinto the deep layers of entorhinal cortex, causes some grid cells tolose their grids but retain their directionality, becoming in effect,head direction cells (Bonnevie et al., 2007). These findings pointto the importance of directional information in the grid cellnetwork and support the notion that that linear and angular infor-mation are combined here and passed thence to the place cells.

NONMETRIC (OR ‘‘CONTEXTUAL’’) INPUTS

The foregoing discussion focused on how place cells integratethe various sources of information that specify where a place fieldshould be located. However, as noted earlier, a growing body ofevidence points to the responsiveness of place cells to nonmetricinputs. Are these inputs of a qualitatively different kind, or areboth metric and nonmetric inputs merely variants of some otherkind of superordinate category of information? For reasons of

FIGURE 4. Deformation of grid cell grids by deformation of apreviously-learned environment. Rats were familiarized with eithera large square environment (A) or rectangular environment (B)and grid cells recorded, first in the familiar environment and thenin a resized environment that was either compressed (A) orextended (B) along one or both axes. The raw grid cell activity inthe familiar environments (outlined in red) versus the rescaledenvironments is shown on the left: the spatial autocorrelationfunctions (Hafting et al., 2005) on the right. The crosses in theautocorrelation maps show the centers of the central peaks.Observe that the peaks become closer together along the com-pressed axis and further apart along the extended axis, but by alesser amount than the actual rescaling. This provides evidencethat the grids, although intrinsic initially, have become partially‘‘attached’’ to the walls of the environment. [Color figure can beviewed in the online issue, which is available at www.interscience.wiley.com.]

780 JEFFERY


clarity and conciseness, we shall refer to the nonmetric inputs,discussed in this section, as ‘‘contextual’’.

Modulation of spatial activity by nonspatial cues is manifest asthe phenomenon known widely as ‘‘remapping,’’ an example ofwhich is shown in Figure 5. Remapping occurs when a highly sa-lient change is made to an environment, and was originally consid-ered to reflect recruitment of a new environmental representation(Muller and Kubie, 1987). In classic, so-called ‘‘complete’’ remap-ping, all cells alter their activity simultaneously, some stop firing,some previously silent cells develop fields and some, about 50%,lose a field in one place and develop a new one in a different place.More recent reports have suggested that the distribution of on–offvs. location-shift remapping differs depending on where the cellsare located, with CA3 cells more likely to on–off (or ‘‘rate’’) remapand CA1 cells more likely to shift their fields (Guzowski et al.,2004; Leutgeb et al., 2004). The question remains open as towhether these are qualitatively different phenomena.

Remapping can be induced by a variety of different kinds ofenvironmental change, including changing the color1 of a cuecard (Bostock et al., 1991), the shape of the box (Lever et al.,2002), color and odor of the box (Anderson and Jeffery, 2003),or placing the recording box in a new place (Hayman et al.,2003) or room (Leutgeb et al., 2005). It can, however, also beelicited by a different kind of stimulus such as changing the task

the animal is performing in a given chamber (Markus et al.,1995) or changing the route it is traveling (Wood et al., 2000;Frank et al., 2000). These are particularly interesting becausethey do not involve making physical changes to the environ-ment—the change is all in the mind of the rat, so to speak. Weshall refer to these classes of stimuli as ‘‘external’’ and ‘‘internal,’’respectively.

Contextual Stimuli External to the Animal

External stimuli known to invoke remapping are those thatcause some kind of salient, easily sensed change to the environ-ment. Some kinds of environmental change invoke completeremapping, i.e., change to the whole observed place cell popula-tion simultaneously, whereas others involve only some cells (par-tial remapping). Initial studies of remapping looked mainly atcomplete remapping. The strong sense of observers that the placecell network was ‘‘flipping’’ into a new state during this processprompted the development of attractor-based models (Samsono-vich and McNaughton, 1997; Tsodyks, 1999) to explain the sud-den, network-wide transition in activity patterns. In attractormodels, the assemblage of incoming sensory data is sharedamong neurons in the highly interconnected CA3 network andeventually settles to a stable state, specific to that environmentand generating a unique pattern of activity across all place cells.By this view, because an individual CA3 neuron gets most of itsinputs not directly from sensory sources, but rather, from otherhippocampal neurons (Amaral and Witter, 1989), the sensorycorrelates (such as responsiveness to particular cues) arise asemergent phenomena from network processes, rather than fromdirect sensory drive. More recent findings of partial remapping(e.g. Anderson and Jeffery, 2003) challenge this view somewhat,because they require at least two networks of neurons (in somecases four), one of which remaps and the remainder which donot, in each one of the environments.

The conditions under which partial vs. complete remappingoccur have not yet been defined. Our experience is that even in agiven environment with a given protocol, remapping patternscan vary, sometimes frustratingly. The likely reason is thatbecause remapping is the output of network computations, rela-tively small changes in parameters, including animal-specific var-iations, can affect whether the system responds to the stimuli ornot. One such parameter is probably how salient the stimuli areinitially. If a context change is made highly salient by using morethan one cue (e.g., shape combined with color), remapping willsubsequently still occur even if the context change is now re-stricted to just one of those cues (Wills et al., 2005). One inter-pretation of this is that the initially salient changes enable thenetwork to carve out two attractor basins after which the systemmore easily settles into one of these basins when nudged by justone of the original cues. However, stimulus salience cannot bethe only factor, because we have observed that highly salientstimuli (as ascertained by the reaction of the rat; for example,rearing in response to a strong odor) do not always cause remap-ping (Caswell Barry, unpublished data). Another factor may behow familiar the original environment was before the contextual

FIGURE 5. Interaction of contexts in remapping. Place cellswere recorded in a square box that was located in either a northposition or a south position until they remapped between the twolocations. The box color was then changed (in this example, fromblack to white) and recordings again made in the north and southpositions. It can be seen that cells that discriminated the locationsin the familiar box no longer discriminated in the unfamiliarbox—the discrimination is thus context-specific. This can only beexplained by assuming an interaction between the two kinds ofcontextual cue, location, and color upstream of the place cell.Adapted from Hayman et al., 2003.

1Rats have essentially monochromatic vision and so the visual stimuli inthe experiments described here have always involved variants of black,white and grey. Strictly speaking, these are not ‘‘colors,’’ they are degreesof brightness, or luminance. However, the term ‘‘color’’ is used herebecause it is a term we more usually use to describe wall and floor cover-ings. If the reader is offended by this mild inaccuracy then please men-tally replace ‘‘color’’ with ‘‘brightness’’ throughout.



change was made: in a familiar environment a change may besurprising, perhaps attracting attention from the rat. Indeed,there is some evidence that attention may be a factor in whetherstimuli induce place cell changes (Kentros et al., 2004).

It is worth noting here that although a distinction has beenmade above between metric and contextual cues, this distinctionmay not be clear-cut; it may be that metric cues can also act ascontextual cues under some circumstances. An obvious exampleis the two-box-location experiment of Hayman et al. (2003), inwhich place cells remapped when the recording box was trans-lated between a north and south location. Although the changemade was a spatial one, it could be that the effect was to changethe contextual aspects of the environment. We draw this conclu-sion because place fields never appeared to be dragged aroundwithin the box by the environment translations (by analogy withhow they were dragged by walls in the O’Keefe and Burgess1996 experiment); rather, they either remained in the same loca-tion relative to the box or switched on and off. A similar space-as-context explanation may account for the findings of Leveret al. (2002) that changing the box from a circle to a squarecaused sudden transitions between states.

Contextual Stimuli Internal to the Animal

As mentioned earlier, another kind of stimulus can induceremapping; these are stimuli that are not properties of the physi-cal environment per se, but rather are more abstract stimuli,such as the task the animal is performing at a given time(Markus et al., 1995) or which of two overlapping routes it iscurrently executing (Frank et al., 2000; Wood et al., 2000; Fer-binteanu and Shapiro, 2003). These kinds of stimuli are particu-larly interesting because they are not detectable to an observerand must not therefore arrive at the hippocampus via a simplesensory route. Because these observations have been made whenthe animals are performing highly repetitive tasks, some observ-ers have considered that the salient features of the stimuli aretheir sequential nature: that is, the cells are considered to beresponding to temporally organized sequences of actions, and arethus perhaps encoding a simple form of episodic memory(Wood et al., 2000; Shapiro et al., 2006). An alternative view-point, however, is that the parameters in these tasks that controlremapping are those that could be considered to be a part of the‘‘context’’, that is, they are aspects of the environment that arerelatively stable, ongoing (albeit switching back and forth), anddefine it in terms of the actions that must be performed there.By this view, the reason the hippocampus is interested (in thesense of responding to) in these stimuli is because they form apart of the spatial context, or, as it were, the situation the animalis in. In support of the latter characterization, remapping to taskparameters is a fickle phenomenon and does not always occur insituations where the rat is clearly executing a task with sequentialcomponents (Lenck-Santini et al., 2001; Bower et al., 2005). Bythe contextual-encoding view, the fickleness arises from the like-lihood that for whatever reason, some conditions favor the for-mation of two attractors to represent the context, and othersuperficially similar ones favor only one. The finding that inter-

nal factors can cause environment-specific remapping leads tothe notion that the hippocampal encoding of ‘‘place’’ is highlymultidimensional, extending far beyond the mere physical layoutand metric parameters thereof.

INTEGRATION OF CONTEXTUALINPUTS WITH EACH OTHER

Because ‘‘context’’ is such a multifarious phenomenon, thequestion of interest regarding integration of contextual inputs iswhether this integration occurs upstream of the hippocampus,implying a ‘‘context processor’’ somewhere in the neocortex, orwhether the place cells themselves are the site of contextual inte-gration. This is important for understanding how the fullyfleshed-out place representation is ultimately constructed.

We explored this issue by creating multielement contexts(each a ‘‘color’’ paired with an odor) and observing remappingpatterns to determine whether the cells would respond to thecontexts en bloc, which would suggest that the inputs had al-ready been integrated upstream, or whether they responded tothe elements individually (Anderson and Jeffery, 2003). Wefound the latter; that is, the place cells responded heterogene-ously to the contexts, each responding to some unique color/odor combination. The implication is that rather than receivinga preprocessed, holistic context signal, the contextual informa-tion must have arrived at the place cells in elemental form, to becombined in the hippocampus.

It seems that place cells are not hard-wired with (all of ) theircontextual inputs: these can be learned. In a pilot experiment inwhich a box enclosure was moved back and forth between twolocations, place cells acquired the ability to discriminate the loca-tions over repeated trials spanning several days (Jeffery, 2001).This discrimination, which we take to be a contextual discrimi-nation (for reason discussed earlier), was subsequently found toshow a second-order context specificity, inasmuch as the contex-tual discrimination occurring in one context (e.g., a black box)did not transfer to another context (a white box; Fig. 5)(Hayman et al., 2003). The context-specific behavior (where‘‘context’’ is location) was itself context-specific (where ‘‘con-text’’ now means color). Our interpretation of this surprisingfinding is that the inputs enabling the place cells to discriminatecontexts in the locational domain were themselves selected bythe inputs in the color domain. This implies an interactionbetween the contexts – location and color – upstream of theplace cells.

INTEGRATION OF METRIC ANDCONTEXTUAL CUES

How do metric and contextual cues interact and where? Isthere a reason to think these kinds of stimuli are functionallydifferent?

Although both metric and contextual cues influence placecells, the phenomenology of their responses to contextual and

782 JEFFERY


metric manipulations suggests that the nature of this influence isdifferent. Whereas variation of metric cues (by virtue of the ani-mal’s walking around in the environment) causes a steadyincrease or decrease of firing rate as the animal moves into andout of a place field, variation of contextual cues often causes anabrupt, discontinuous change (‘‘remapping’’): the cells start orstop firing or the place field jumps from one part of the environ-ment to another. It could be argued that these different behaviorsmerely reflect the different way in which metric vs. nonmetriccues vary in the real world. An animal is able to experiencesmoothly varying metric cues because it locomotes smoothlyaround the environment, whereas it usually only experiencescontextual changes in a sudden way. Perhaps, if we were able tosmoothly vary contextual cues, we would see that place cellsrespond smoothly in the contextual dimension too. By exten-sion, perhaps, as far as the hippocampus is concerned, contextualcues are no different from metric cues.

Over the past several years we have begun to explore the issueof whether contextual and metric cues are functionally dissoci-able, and our impression is that they are. In the first such experi-ment, we looked at the remapping induced when a recordingbox was changed from black to white (Jeffery and Anderson,2003). This simple, nonmetric change induced complete remap-ping, with approximately half of the cells switching off or onand the remainder developing new firing locations, as usual. Thelatter class of cells, which shifted fields, was especially interestingin this experiment because the cells were expressing different spa-tial coordinates depending on the box color, which enabled us totest how they might determine which coordinates to use. Thequestion we asked was why did changes occur to the spatial firingof the cells when the box was changed in color? Was this becausethe cells received inputs specifying how far they should fire fromblack-walls and different inputs specifying how far they shouldfire from white-walls? In other words, were they receiving ‘‘color-coded’’ metric inputs? This proposition was tested by creatinghybrid boxes in which some of the walls were black and somewhite. We hypothesized that new fields should develop, havinghybrid x–y coordinates, so to speak: that is, the new fields shouldbe determined in the x-dimension by walls of the x-color and inthe y-dimension by walls of the y-color. Thus, the new fields inthe hybrid boxes should be predictable on the basis of knowingthe field location in each of the single-colored boxes.

This manipulation was so singularly ineffective that it wasquickly abandoned; the fact was that despite their robust remap-ping to complete color changes, the cells responded little to thepartial wall color changes. We made changes to the floor colorinstead and found that whatever combination of wall and floorcolor we used, the cells nearly always, with rare exceptions,adopted the place field pattern appropriate to the floor, irrespec-tive of the walls. It is as if the cells did not even receive wall-colorinformation. And yet, they clearly receive wall information ofsome sort, because place fields will readily follow walls if thewalls are moved. The seemingly inescapable conclusion is thatplace cells use walls and floor differently; they use walls as a basisfor computing distances, but the floor to tell them which envi-ronment they are in, and therefore which field to express.

On the basis of these, and other, findings, we have proposed afunctional distinction between contextual and metric inputs,whereby metric inputs (from boundaries) specify the location ofa place field in the manner proposed by Burgess and colleagues(O’Keefe and Burgess, 1996; Hartley et al., 2000), whereas con-textual inputs select which of these spatial inputs will be activein a given environment (Fig. 6). By this view, rate remappingcan be understood as a variation in the strength of the drivethrough a given set of metric inputs produced by small changesin context, and ‘‘global’’ (or locational) remapping as when largecontext changes cause one set of inputs to be silenced altogether(its contextual modulators no longer active at all) and a new set(possibly) activated instead.

The above scheme is an abstract formulation of the underlyinglogic of the input structure. How might such a scheme be imple-mented in reality, particularly in light of what we have recentlylearned about grid cells and their interaction with place cells? Aninteresting and relevant observation is that although place cellsfrequently express independent fields in two different environ-ments, such cells tend not ever to express both fields together inany environment. (Note that place cells frequently do have morethan one place field; however these always remap together andbehave as if they are, in effect, the same field spread across twolocations. What place cells don’t seem to express is two independ-ently-remappable fields at once). It is almost as if two the fields‘‘know’’ about each other and thus know not to cooccur, andyet, how can a place field in one location be suppressed byanother field which, though potentially active in that environ-ment, is not actually active (because the rat is not at that locationat that moment)?

A possible mechanism for this, as it were, nonlocality, lies inthe grid cells, whose rigid metric spacing means that if a grid cellis firing in one particular place at a given moment, its firing in adifferent place at a future time is constrained by whether thatlocation also lies on a grid peak. This constrained firing patternmeans that distant parts of the environment are, in a sense, con-nected via the underlying grid structure. Suppose that place fieldsare formed by the summation of overlaid grids; it thus follows if

FIGURE 6. Model of how spatial inputs might be gated bycontextual inputs. A place cell has two sets of metric inputs, onedetermining each field. Different contexts (in this case, black vs.white) activate different sets of metric inputs. Adapted from Jefferyand Anderson, 2003.



a number of grid peaks overlie each other at one location, theplaces at which they next overlie each other will be constrained,to a distance specified by the period of the largest-spacing grid.If the grids contributing to a given place field are sufficientlylarge, then the point of next summation will lie outside the re-cording box. Given that the smallest reported grid spacing isaround 30 cm, and most boxes are somewhere between 1 and 2m, this seems likely. This, therefore may explain why place cellstend to have only one field in an environment (though notalways).

How could this grid-conferred nonlocality explain remapping?In a recent experiment, Fyhn et al. found that following manipu-lations that induced global remapping in place cells, grid cellswould simultaneously shift and rotate their grids in concert(Fyhn et al., 2007). On the assumption that the shift in gridscaused the shift in fields, an explanation put forward by theseauthors (see their Supplementary Fig. 12) is that any manipula-tion that displaced and/or rotated the grids would shift the align-ment point, described above, out of the range of the box enclo-sure and cause the place field to disappear from inside the box.Were that shift simultaneously to move another alignment pointinto the box, the result would be an apparent shift of the field.By this view, the reason a place cell can never express both itsfields at once is that grid cells can only express one grid at once.

Such a scheme could explain complete remapping; explainingpartial remapping is a little more complex. In order for only someplace fields to respond to contextual changes, there must be somedegree of partial selection of the grid cell inputs by the contextualcues, either because perhaps only some grids realign following apartial context change, or because the subset of grid cells thatdrives a given place cell can be switched (gated) by context cues.Further experiments are needed to untangle these possibilities.

Putting all this together, we can postulate that the metric ofthe grid forms the metric component of our metric/contextualgating model, and some aspect of the mapping between grid cellsand place cells corresponds to the contextual component. Thequestion of how contextual cues drive place cells has thusbecome a new question—how do contextual cues drive gridcells? Recordings of grid cells in multielement contexts will beneeded to answer this question.

CONCLUSION

This article has argued for the view that metric and nonmetricinputs to place cells are functionally dissociable, with metricinputs specifying where a cell’s field should go, and nonmetric(or ‘‘contextual’’) inputs specifying whether or not that fieldshould be expressed in a given environment. Metric inputs areintegrated from information arriving in a variety of sensorymodalities, and it is likely—based on the responses of grids tocontextual, metric, and directional manipulations—that much ofthis integration occurs in entorhinal cortex. Similarly, contextualinputs are also multimodal, sometimes to a high level of abstrac-

tion (such as the parameters of a given task) and again, evidencesuggests this integration may take place in entorhinal cortex.

If so much integration occurs upstream of the place cells, thenwhat is the function of the place cells themselves? Evidence fromgrid cell recordings, such as the hippocampal inactivation studyof Bonnevie et al. (2006), suggests that one function may be toprovide the anchoring that attaches the grids to the environment.Another function may be to sculpt the attractor basins thatunderlie the processes of pattern separation and completionpostulated to occur in dentate gyrus and CA3. And finally, itmay be that place cells provide the combinatorial functions thatallow multiple different contexts to be associated with the sameset of spatial parameters (Anderson et al., 2006).

While the details of multimodal integration are rapidly beinguncovered by studies of place and grid cells, one of the big mys-teries is how these neurons contribute to the other function ofthe hippocampal system, episodic memory. It seems clear thatthe foundation of the place and grid cell representation is a spa-tial one, but whether the more detailed and transient parts of aremembered episode are incorporated at this level, or by someother brain region, remains to be determined.

REFERENCES

Amaral DG, Witter MP. 1989. The three-dimensional organization ofthe hippocampal formation: A review of anatomical data. Neuro-science 31:571–591.

Anderson MI, Jeffery KJ. 2003. Heterogeneous modulation of placecell firing by changes in context. J Neurosci 23:8827–8835.

Anderson M, Killing S, Morris C, O’Donoghue A, Onyiagha D,Stevenson R, Verriotis M, Jeffery K. Behavioral correlates of thepopulation coding of spatial context. Hippocampus 16(9):730–742.

Barry C, Burgess N. 2007. Learning in a geometric model of placecell firing. Hippocampus. Vol 17 (in press).

Barry C, Lever C, Hayman R, Hartley T, Burton S, O’Keefe J, JefferyKJ, Burgess N. 2006. The boundary vector cell model of place cellfiring and spatial memory. Rev Neurosci 17:71–97.

Barry C, Hayman R, Burgess N, Jeffery KJ. Experience-dependentrescaling of entorhinal grids. Nat Neurosci 10(6):682–684.

Berger TW, Thompson RF. 1978. Neuronal plasticity in the limbicsystem during classical conditioning of the rabbit nictitating mem-brane response. I. The hippocampus. Brain Res 145:323–346.

Bonnevie T, Fyhn M, Hafting T, Moser E, Moser M-B. 2006. Mis-alignment of entorhinal grid fields after hippocampal inactivation.Soc Neurosci Abstr 68.1/BB9.

Bostock E, Muller RU, Kubie JL. 1991. Experience-dependent modifi-cations of hippocampal place cell firing. Hippocampus 1:193–205.

Bower MR, Euston DR, McNaughton BL. 2005. Sequential-context-dependent hippocampal activity is not necessary to learn sequenceswith repeated elements. J Neurosci 25:1313–1323.

Breese CR, Hampson RE, Deadwyler SA. 1989. Hippocampal placecells: Stereotypy and plasticity. J Neurosci 9:1097–1111.

Etienne AS, Jeffery KJ. 2004. Path integration in mammals. Hippo-campus 14:180–192.

Fenton AA, Csizmadia G, Muller RU. 2000. Conjoint control of hip-pocampal place cell firing by two visual stimuli. I. The effects ofmoving the stimuli on firing field positions. J Gen Physiol116:191–209.

Ferbinteanu J, Shapiro ML. 2003. Prospective and retrospective mem-ory coding in the hippocampus. Neuron 40:1227–1239.

784 JEFFERY


Frank LM, Brown EN, Wilson M. 2000. Trajectory encoding in thehippocampus and entorhinal cortex. Neuron 27:169–178.

Fyhn M, Hafting T, Treves A, Moser MB, Moser EI. 2007. Hippo-campal remapping and grid realignment in entorhinal cortex.Nature 446:190–194.

Gothard KM, Skaggs WE, McNaughton BL. 1996. Dynamics of mis-match correction in the hippocampal ensemble code for space:Interaction between path integration and environmental cues.J Neurosci 16:8027–8040.

Guzowski JF, Knierim JJ, Moser EI. 2004. Ensemble dynamics of hip-pocampal regions CA3 and CA1. Neuron 44:581–584.

Hafting T, Fyhn M, Molden S, Moser MB, Moser EI. 2005. Micro-structure of a spatial map in the entorhinal cortex. Nature 436:801–806.

Hargreaves EL, Yoganarasimha D, Knierim S. 2007. Cohesiveness ofspatial and directional representations recorded from neural ensem-bles in the anterior thalamus parasubiculum, medial entorhinal cor-tex, and hippocampus. Hippocampus. Vol 17. (in press).

Hartley T, Burgess N, Lever C, Cacucci F, O’Keefe J. 2000. Modelingplace fields in terms of the cortical inputs to the hippocampus.Hippocampus 10:369–379.

Hayman RM, Chakraborty S, Anderson MI, Jeffery KJ. 2003. Con-text-specific acquisition of location discrimination by hippocampalplace cells. Eur J Neurosci 18:2825–2834.

Jeffery KJ. 1998. Learning of landmark stability and instability by hip-pocampal place cells. Neuropharmacology 37:677–687.

Jeffery KJ. 2001. Plasticity of the hippocampal cellular representationof space. In: Holscher C, editor. Neuronal Mechanisms of MemoryFormation. Cambridge: Cambridge University Press. pp 100–121.

Jeffery KJ, O’Keefe J. 1999. Learned interaction of visual and idiot-hetic cues in the control of place field orientation. Exp Brain Res127:151–161.

Jeffery KJ, Anderson MI. 2003. Dissociation of the geometric andcontextual influences on place cells. Hippocampus 13:868–872.

Jeffery KJ, Anand RL, Anderson MI. 2006. A role for terrain slope inorienting hippocampal place fields. Exp Brain Res 169:218–225.

Kentros CG, Agnihotri NT, Streater S, Hawkins RD, Kandel ER.2004. Increased attention to spatial context increases both placefield stability and spatial memory. Neuron 42:283–295.

Knierim JJ, Kudrimoti HS, McNaughton BL. 1995. Place cells, headdirection cells, and the learning of landmark stability. J Neurosci15:1648–1659.

Lenck-Santini PP, Save E, Poucet B. 2001. Place-cell firing does notdepend on the direction of turn in a Y-maze alternation task. Eur JNeurosci 13:1055–1058.

Leutgeb S, Leutgeb JK, Treves A, Moser MB, Moser EI. 2004. Dis-tinct ensemble codes in hippocampal areas CA3 and CA1. Science305:1295–1298.

Leutgeb S, Leutgeb JK, Barnes CA, Moser EI, McNaughton BL, MoserMB. 2005. Independent codes for spatial and episodic memory inhippocampal neuronal ensembles. Science 309:619–623.

Lever C, Wills T, Cacucci F, Burgess N, O’Keefe J. 2002. Long-termplasticity in hippocampal place-cell representation of environmentalgeometry. Nature 416:90–94.

Markus EJ, Qin YL, Leonard B, Skaggs WE, McNaughton BL, BarnesCA. 1995. Interactions between location and task affect the spatialand directional firing of hippocampal neurons. J Neurosci 15:7079–7094.

Moita MA, Rosis S, Zhou Y, LeDoux JE, Blair HT. 2003. Hippocam-pal place cells acquire location-specific responses to the conditioned

stimulus during auditory fear conditioning. Neuron 37:485–497.

Muller RU, Kubie JL. 1987. The effects of changes in the environ-ment on the spatial firing of hippocampal complex-spike cells.J Neurosci 7:1951–1968.

O’Keefe J. 1999. Do hippocampal pyramidal cells signal non-spatial aswell as spatial information? Hippocampus 9:352–364.

O’Keefe J, Burgess N. 1996. Geometric determinants of the placefields of hippocampal neurons. Nature 381:425–428.

Quirk GJ, Muller RU, Kubie JL. 1990. The firing of hippocampalplace cells in the dark depends on the rat’s recent experience.J Neurosci 10:2008–2017.

Ranck JB Jr. 1973. Studies on single neurons in dorsal hippocampalformation and septum in unrestrained rats. I. Behavioral correlatesand firing repertoires. Exp Neurol 41:461–531.

Rotenberg A, Muller RU. 1997. Variable place-cell coupling to a con-tinuously viewed stimulus: Evidence that the hippocampus acts as aperceptual system. Philos Trans R Soc Lond B Biol Sci 352:1505–1513.

Samsonovich A, McNaughton BL. 1997. Path integration and cogni-tive mapping in a continuous attractor neural network model.J Neurosci 17:5900–5920.

Sargolini F, Fyhn M, Hafting T, McNaughton BL, Witter MP, MoserMB, Moser EI. 2006. Conjunctive representation of position,direction, and velocity in entorhinal cortex. Science 312:758–762.

Save E, Cressant A, Thinus-Blanc C, Poucet B. 1998. Spatial firing ofhippocampal place cells in blind rats. J Neurosci 18:1818–1826.

Shapiro ML, Kennedy PJ, Ferbinteanu J. 2006. Representing episodesin the mammalian brain. Curr Opin Neurobiol 16:701–709.

Taube JS. 1998. Head direction cells and the neurophysiological basisfor a sense of direction. Prog Neurobiol 55:225–256.

Taube JS, Muller RU, Ranck JB J. 1990a. Head-direction cellsrecorded from the postsubiculum in freely moving rats. I. Descrip-tion and quantitative analysis. J Neurosci 10:420–435.

Taube JS, Muller RU, Ranck JB J. 1990b. Head-direction cellsrecorded from the postsubiculum in freely moving rats. II. Effectsof environmental manipulations. J Neurosci 10:436–447.

Tsodyks M. 1999. Attractor neural network models of spatial maps inhippocampus. Hippocampus 9:481–489.

Wills TJ, Lever C, Cacucci F, Burgess N, O’Keefe J. 2005. Attractordynamics in the hippocampal representation of the local environ-ment. Science 308:873–876.

Wood ER, Dudchenko PA, Eichenbaum H. 1999. The global recordof memory in hippocampal neuronal activity. Nature 397:613–616.

Wood ER, Dudchenko PA, Robitsek RJ, Eichenbaum H. 2000. Hippo-campal neurons encode information about different types of memoryepisodes occurring in the same location. Neuron 27:623–633.

Yoganarasimha D, Yu X, Knierim JJ. 2006. Head direction cell repre-sentations maintain internal coherence during conflicting proximaland distal cue rotations: Comparison with hippocampal place cells.J Neurosci 26:622–631.

Zhang K. 1996. Representation of spatial orientation by the intrinsicdynamics of the head-direction cell ensemble: A theory. J Neurosci16:2112–2126.

Zugaro MB, Berthoz A, Wiener SI. 2001. Background, but not fore-ground, spatial cues are taken as references for head directionresponses by rat anterodorsal thalamus neurons. J Neurosci 21:RC154.



Documents

Integration of the sensory inputs to place cells: What, where, why, and how?