Field repetition and local mapping in the hippocampus and the medial entorhinal cortex

Abstract

Hippocampal place cells support spatial cognition and are thought to form the neural substrate of a global “cognitive map.” A widely held view is that parts of the hippocampus also underlie the ability to separate patterns or to provide different neural codes for distinct environments. However, a number of studies have shown that in environments composed of multiple, repeating compartments, place cells and other spatially modulated neurons show the same activity in each local area. This repetition of firing fields may reflect pattern completion and may make it difficult for animals to distinguish similar local environments. In this review we 1) highlight some of the navigation difficulties encountered by humans in repetitive environments, 2) summarize literature demonstrating that place and grid cells represent local and not global space, and 3) attempt to explain the origin of these phenomena. We argue that the repetition of firing fields can be a useful tool for understanding the relationship between grid cells in the entorhinal cortex and place cells in the hippocampus, the spatial inputs shared by these cells, and the propagation of spatially related signals through these structures.

how locations in the outside world are represented in the brain has been a topic of intense research interest for almost 50 years, since the discovery of neurons in the rodent hippocampus—place cells—that fire in individual places in an environment (O’Keefe and Dostrovsky 1971). Following O’Keefe and Nadel’s (1978) conceptualization of the hippocampus as a cognitive map, much of the ensuing work has assumed that place cells comprise a representation of the entire environment in which the animal finds itself (although different reference frames are possible within this map; see, e.g., Gothard et al. 1996; Poucet 1993; Zinyuk et al. 2000). In the present review, we challenge this assumption of a global map in light of data indicating that many spatial cells are driven by local boundaries and a directional input. These influences provide an allocentric encoding of local spaces, which is only incidentally global.

Space is traditionally defined from two reference points. In the first, location within an environment is defined with “self-relative” directions, such as “on my left” or “20 feet in front of me.” This is egocentric space. In the second, locations are identified independent of the observer, for “halfway between the window and the door” or “behind the chair and toward the painting.” This is allocentric space. In the present review we are primarily concerned with how the latter is represented in the brain.

In humans, representations of space likely vary in terms of their scale and detail. For instance, a person can recognize his or her location within a given room of his/her house but also, simultaneously, where he/she is within a geographical region. Thus different types of spatial representations may operate, depending on the task at hand (Burgess 2006; Ekstrom et al. 2014). In the present review, we restrict our consideration to allocentric space as it is represented by (or as it correlates with) the firing fields of spatially tuned neurons in the rodent brain. Identifying the rules by which these operate may allow us to understand the interplay between location recognition and longer-range navigation.

Whether the mammalian brain maps space in local or global coordinates is an important issue because it likely constrains spatial cognition. From this perspective, there is evidence that certain types of spaces, such as repetitive local geometries, are more challenging than others for wayfinding. For example, city planners are discouraged from using repetitive street designs as they are considered disorienting (Rumbarger and Vitullo 2003). This effect is embodied in the repetitive streets of Brasília, which are challenging to navigate (Scott 1998). Difficulties in distinguishing locations can also be problematic for patients suffering from dementia. Such individuals can find long corridors confusing, especially those with repetitive elements (Netten 1989; van der Voordt 1993). There is also evidence that patients prefer “L”-shaped corridors to long straight ones (Elmståhl et al. 1997; Marquardt 2011; Passini et al. 2000; Rainville et al. 2002). As we consider below, such observations are consistent with the responses of spatially tuned neurons in the rodent brain to repetitive local environments.

Place and Grid Field Repetition

In the traditional view of place cells, each cell exhibits a unique firing field and together these place fields represent the animal’s entire environment (e.g., Barnes et al. 1997; see Fig. 1). One approach to studying place cells and other types of spatially tuned neurons has been to manipulate the animal’s environment and see how this affects firing fields (e.g., Acharya et al. 2016; Anderson and Jeffery 2003; Barry et al. 2007; Bostock et al. 1991; Chen et al. 2013; Krupic et al. 2015; Leutgeb et al. 2004, 2005a, 2005b; Lever et al. 2002b; Muller and Kubie 1987; O’Keefe and Burgess 1996). A second approach has looked at these cells during purposeful behavior. This work has shown that place cell firing is modulated by task demands (e.g., Hok et al. 2007; Markus et al. 1995; Moita et al. 2004; Wood et al. 2000) and by the internal state of the animal (Kennedy and Shapiro 2004, 2009; for review see Schiller et al. 2015). From the perspective of the hippocampus at least, the latter approach has indicated a function beyond the representation of space. In the ensuing discussion, however, we limit our consideration to studies focusing on the changes to the animal’s environment, although we acknowledge that the addition of task demands also influences place cell firing correlates.

Fig. 1.

Spatially modulated cell types in the mammalian brain. Top left: the firing rate map of a dCA1 (hippocampus) place cell. Action potentials and dwell time are binned, smoothed, and divided to give a spatial map of the cell’s firing rate. Generally, hot colors represent high firing rates, cold colors represent low firing rates, and white represents unvisited locations. This cell has an area of high firing located to the northeast of the environment, and this area is known as this cell’s “place field.” Reproduced from Grieves et al. (2016), licensed under CC BY 3.0. Top center: example of a medial entorhinal cortex (mEC) head direction cell. These “polar” plots show the action potentials of the cell, binned in terms of the animal’s head direction at the time and divided by the amount of time spent facing that direction overall. This cell fires at a high rate when the animal is facing to the north (90°) within the environment, and this is referred to as the cell’s preferred firing direction. Reproduced with permission from Harland et al. (2015). Top right: firing rate map of an mEC grid cell. This is produced with the same method as for the place cell. Multiple firing fields can be observed that form a triangular or hexagonal grid that spans the environment. Reproduced with permission from Harland et al. (2015). Middle: firing rate maps of a single subicular boundary cell recorded in 3 different environments, a circle, a diamond, and a square, placed in the same room. Note that the cell continues to fire along walls that subtend the rat at the same angle (northeasterly boundaries) even when the environment changes. Adapted with permission from Lever et al. (2009), Figure 3, cell 2d. Bottom left: firing rate map of a border cell recorded in the mEC. Adapted from Solstad et al. (2008) with permission from The American Association for the Advancement of Science. Bottom right: example of a modeled boundary vector cell (BVC), generated in the same way as in Hartley et al. (2000) (R. Grieves, unpublished data).

Within this domain, several findings suggest that when rats move between two or more similar maze rooms, a given place cell produces the same field in each room. For instance, Skaggs and McNaughton (1998) recorded dCA1 place cells while rats explored an environment composed of two identical compartments joined by a corridor. They found that place cells often showed similar firing fields in each of the two compartments (Fig. 2A; see also Fuhs et al. 2005). Thus instead of having unique representations of each compartment, as one would predict for a mapping of the entire environment, many place cells showed similar fields across compartments. The lack of remapping observed between compartments suggests that place cells are partly driven by local views.

Fig. 2.

Examples of local encoding by place cells. Firing rate maps utilize the color axis given below B. A: example dCA1 place cell recorded in the maze used by Skaggs and McNaughton (1998). Adapted with permission from Skaggs and McNaughton (1998), Figure 2. B: dCA1 place field repetition in the 4-compartment apparatus used by Spiers et al. (2015). Adapted from Spiers et al. (2015), Figure 3, licensed under CC BY 3.0. C: Derdikman et al.’s (2009) hairpin maze. Top: example of mEC grid field repetition is shown in firing rate maps recorded when the animal moved through the maze from left to right (left) and from right to left (right). Middle and bottom: similar example of dCA1 place field repetition. Adapted with permission from Macmillan Publishers Ltd., copyright 2009 (Derdikman et al. 2009, Figs. 1 and 6). D: 2 example dCA1 place cells recorded by Lever et al. (2002b) in a circular and square environment of the same size. Modified with permission from Macmillan Publishers Ltd., copyright 2002 (Lever et al. 2002b, Fig. 1).

In an elegant extension of the Skaggs and McNaughton study, Spiers et al. (2015) recorded dCA1 place cells as rats moved between four parallel maze compartments connected with an alleyway. They found that individual place cells tended to show similar place fields in all four compartments (Fig. 2B). Cells only formed a distinct representation for a specific box when its size or color was changed, and even in this case repetition of fields was found in the remaining three boxes. These findings were replicated by Grieves et al. (2016), as described in Repetition/Fragmentation of Firing Fields Depends on Direction.

A similar phenomenon has been observed in grid cells—neurons from the entorhinal cortex (EC) and pre- and postsubiculum that exhibit multiple, regularly arranged fields within an environment (Hafting et al. 2005; Fig. 1). For example, Derdikman et al. (2009) recorded from grid cells and place cells in a zigzag alleyway or “hairpin” maze and found that both types of cell showed firing fields that repeated across alleyways facing the same direction (Fig. 2C). These fields did not repeat across alleyways that the animal entered in the opposite direction. Repeating, local representations persisted regardless of the large number of alleyways (5 in each direction), suggesting that self-motion information, such as distance traveled, did not inform the activity of these cells. The authors refer to this phenomenon as a “fragmentation” of the firing fields.

Repetition/Fragmentation of Firing Fields Depends on Direction

An important finding from Derdikman et al.’s (2009) hairpin maze experiment was that place and grid cell fields were modulated by the heading direction of the animal. Cells differentiated north-facing compared with south-facing alleyways, and the position of fields was also dependent on the direction in which the rat ran through the maze (Fig. 2C). As the zigzag route through the maze was continuous, the most parsimonious explanation for this finding is that the spatial cells were sensitive to the animal’s allocentric direction (e.g., McNaughton et al. 1983; Muller et al. 1994) as opposed to alternating between different motivational states (Smith and Mizumori 2006).

Supporting this interpretation, Whitlock and Derdikman (2012) recorded from medial entorhinal cortex (mEC) layers II, III, and V and showed that head direction cells, neurons in an interconnected series of brain regions that are tuned to individual allocentric directions (Taube et al. 1990a), maintained a stable firing direction throughout this apparatus. The head direction system is a defining input to both place cells and grid cells (Acharya et al. 2016; Leutgeb et al. 2000; Peyrache et al. 2016; Winter et al. 2015; Zhang et al. 2013; see also Rubin et al. 2014); one possibility is that such a directional input provides an invariant directional reference that contributes to repetition of spatial fields when an animal repeatedly faces the same direction across maze compartments. In this view, the head direction system provides a global reference frame across maze compartments (e.g., Taube and Burton 1995). This maintenance of orientation across compartments likely requires self-movement of the animal between compartments; when a rat is passively moved between different compartments or local features, the preferred firing direction of its head direction cells can switch from room to local cue anchors (Stackman et al. 2003; Taube et al. 2013).

The notion of a directional input to place cells is also supported by findings from Nitz (2011), who recorded dCA1 place cells in an alleyway that spiraled inward to a point. The cells had multiple fields in coils of the spiral that had the same angular relation to the center and that faced the same direction (Fig. 3B). Furthermore, as in Derdikman et al.’s (2009) hairpin maze, cells fired differently depending on the direction of travel through the alleyway. This is consistent with the finding of Fuhs et al. (2005) in a multicompartment environment. They replicated the two-box apparatus of Skaggs and McNaughton (1998) but also recorded dCA1 place cells in the same two compartments joined end to end and connected directly by a doorway (Fig. 3A). When the compartments were connected by a corridor, place cells showed the same activity in each. However, when the compartments were connected directly to one another, the cells formed a different representation for each compartment. Importantly, in the latter, the doorways were in different relative positions (south in one compartment, north in the other), whereas in the corridor situation the doorways were in the same position for both (e.g., west).

Fig. 3.

Place field repetition depends on direction. Top: maze schematics. Bottom: examples of the corresponding firing activity maps. Color bar next to A corresponds to C also. A: maze used by Fuhs et al. (2005). Left: example of dCA1 place field repetition when compartments were parallel and connected by a corridor (corridor data are ignored). Right: the same cell showed a lack of repetition when the compartments were rotated 90° and abutted each other. Adapted with permission from Fuhs et al. (2005). B: mazes used by Nitz (2011) and Cowen and Nitz (2014). Rats ran along a spiral path of either a square (left) or circular (right) maze. In both, linearized rate maps revealed that dCA1 place cells have multiple fields that occur when the animal is facing the same direction. Adapted with permission from Nitz (2011). C: mazes used by Grieves et al. (2016). Two example dCA1 place cells are shown, 1 per row. Left: place field repetition when animals navigate 4 parallel compartments connected by a corridor. Right: absence of place field repetition when the same compartments are arranged in a radial formation. Adapted from Grieves et al. (2016), licensed under CC BY 3.0.

The results of Tanila (1999) are consistent with these findings. Tanila recorded dCA3 place cells in a similar apparatus—two compartments connected directly by a doorway. Similar to the results with CA1 cells, 91% of the place fields in CA3 cells differed between compartments. Again, as the rats actively moved between the compartments, it was likely that the doorway between the two served as a distinguishing landmark.

To directly assess the impact of compartment orientation as a distinguishing cue, Grieves et al. (2016) recorded place cells in a four-compartment apparatus similar to the one used by Spiers et al. (2015). In addition to this “parallel” configuration, an alternative maze was used in which a 60° angle was introduced between the compartments (Fig. 3C). The same actual compartments were used in both situations, and they differed only in their orientation and the shape of the connecting alleyway for each. In the parallel configuration dCA1 place cells fired similarly in every compartment, as observed by Spiers et al. (2015). However, similar to the results of Fuhs et al. (2005), when compartments were at a 60° angle to one another place field repetition was not observed. These results again suggest that directional reference allows place cells to disambiguate otherwise visually and geometrically identical local environments.

Repetition of Spatial Fields May Constrain Spatial Learning

As noted above, human navigation performance decreases as directional and geometric cues become invariant, such as in long repetitive corridors or streets (Marquardt 2011). Might repetition of the activity of spatial cells underlie such difficulties in navigation? To test this, Grieves et al. (2016) trained naive animals on a conditional odor discrimination task in either the parallel or radial version of their four-compartment maze (Fig. 3C). In this task, an identical set of four odorized sand wells was present in each box and a different odor was rewarded in each one. Thus rats had to discriminate between the compartments to find the food efficiently. In the parallel configuration, where field repetition was found, animals were significantly impaired in learning compared with the group trained in the radial configuration where field repetition was absent. These results suggest that local environments in which place field repetition is observed are more difficult for animals to discriminate compared with those in which place field repetition is not observed. Although it was not examined in the Grieves et al. (2016) experiments (where separate rats were used in the recording and behavioral experiments), it is also possible that learning to discriminate maze compartments yields more unique place cell fields across compartments.

Can a bias toward local mapping be overcome with experience? Although rats in the Grieves et al. (2016) study were impaired in parallel compartments, some did eventually learn the task. Thus it is possible that with repeated experience of connected environments a global representation replaces local maps. A recent study by Carpenter et al. (2015) provides evidence for this. They recorded grid cells in the mEC as rats explored two parallel, connected compartments similar to those of Skaggs and McNaughton (1998) although larger (90 cm instead of 60 cm square) in order to reveal the grid firing structure. During initial exposure to this environment, grid cells often fired similarly in both compartments. However, after multiple exposures to the environment, cells tended to possess fields that formed a continuous grid across the two compartments (Fig. 4A). This suggests that with experience the encoding of local compartments gives way to a representation of the entire enclosure. Whether this slow change in grid firing is accompanied by a change in place cell activity is not known, although such a relationship has been observed in other experiments (Fyhn et al. 2007; Jeffery 2011). If grid and place cells behave similarly, it might also be predicted that grid fields are less local in compartments that face different directions.

Fig. 4.

Mixed evidence for pattern repetition changes with learning. A, top: diagram shows a floor plan of the maze used by Carpenter et al. (2015). Second row: representative rate maps from 1 mEC grid cell for the 2 compartments in an early session (session 4), where it fires similarly in 2 compartments. Third row: maps for the same cell in a later session (session 19). Here it fires with a global representation—the grid pattern extends between the environments as if the wall between them was not present. Bottom: scatterplot shows the result of subtracting the measure of local encoding from one of global encoding for all grid cells that were recorded at differing session intervals of exposure. As animals were exposed for more sessions their representation became more global, and thus the line corresponds to a linear increase. Adapted from Carpenter et al. (2015), licensed under CC BY 4.0. B, top: diagram shows a floor plan of the maze used by Singer et al. (2010). Second row: firing rate map of a dCA1 place cell that shows pattern repetition. Third row: the same data when the color map is capped at 3 Hz. Bottom: bar graph shows the normalized overlap or similarity of place cell firing (when linearized) for cells recorded by Singer et al. (2010) in their multiarm maze. Greater overlap here is suggestive of pattern repetition in the maze arms, and this seems to increase with training. Adapted with permission from Singer et al. (2010). C, top: diagram shows a schematic of the maze used by Grieves et al. (2016). Bottom: plot shows the average level of correlation between compartments as a function of recording session. Correlations between compartments in the parallel version of the task were consistently higher than those in the radial version. Moreover, the level of correlation in either configuration did not change significantly over the course of the experiment. Adapted from Grieves et al. (2016), licensed under CC BY 3.0. D, top: a mock firing rate map for a cell recorded in the maze used by Spiers et al. (2015). The numbers show the distance of each compartment (in compartments) from the one with the highest firing rate. Bottom: plot shows the highest compartment firing rate (compartment 0) and firing rates of every other compartment ranked in order of their distance from this (compartments 1–3) found by Spiers et al. (2015). This relationship is shown for the first day of recording and the last. Because this analysis selects the highest firing rates for compartment 0, this value is significantly higher. If some form of rate coding or remapping was present, the other compartment distances would also be distinguishable in terms of firing rate. However, this was not the case, and this effect did not develop with training. Adapted from Spiers et al. (2015), licensed under CC BY 3.0.

In contrast to the spatial deficits reported by Grieves et al. (2016) and the gradual transformation toward a global map reported by Carpenter et al. (2015), some research suggests that a form of place field repetition increases with spatial learning. This evidence comes from studies by Frank et al. (2000, 2001) and Singer et al. (2010) examining the activity of spatial cells while animals navigated mazes composed of multiple, parallel alleyways. As in Derdikman et al.’s (2009) hairpin maze, dCA1 and dCA3 place cells and neurons in the EC (superficial and deep layers of mEC) fired similarly in multiple alleyways (Fig. 4B). Furthermore, these representations were also dependent on the direction of the animal’s movement. In agreement with the view of the hippocampus as a pattern separator, this field repetition was observed more in EC neurons than in hippocampal place cells. Frank et al. (2000, 2001) and Singer et al. (2010) termed this field repetition “path equivalence” and suggested that it represents encoding of the relationship between behavior and location. In support of this, the frequency of path equivalence appeared to increase as animals learned a task (Fig. 4B). To account for this, it may be speculated that in well-learned tasks spatial cells also begin to reflect common elements of different paths, perhaps via inputs from regions such as the retrosplenial cortex (e.g., Alexander and Nitz 2017).

Visual, Geometric, and Directional Inputs to Spatial Cells

Because of the strong control the geometry of the environment has over place cell activity (Barry and Burgess 2007; Lever et al. 2002a, 2002b; O’Keefe and Burgess 1996; see Fig. 2D, Fig. 5, A and B), it has been proposed that place fields arise from the activity of cells sensitive to boundaries, termed “boundary vector cells” (BVCs) (Barry et al. 2006; Hartley et al. 2000). These cells were originally predicted to be sensitive to boundaries at a specific direction and distance from the animal (Fig. 5C). Actual cells resembling BVCs were subsequently observed in the subiculum (Barry et al. 2006; Brotons-Mas et al. 2017; Lever et al. 2009; Solstad et al. 2008) (Fig. 1 and Fig. 5D), the presubiculum and parasubiculum (Boccara et al. 2010), the mEC (Bjerknes et al. 2014; Savelli et al. 2008; Solstad et al. 2008), and recently in the anterior claustrum (Jankowski and O’Mara 2015) and the rostral thalamus (Jankowski et al. 2015). These “boundary cells” are sensitive to walls, low ridges, or even vertical drops (Fig. 5, C and D) (Lever et al. 2009). The directional component of boundary cells is presumably informed by the head direction system (Peyrache et al. 2016 but see Burgess et al. 2001; Byrne et al. 2007; Julian et al. 2015a). Importantly, in multiple, geometrically identical, similarly oriented compartments the firing of a single boundary cell is expected to be identical (Carpenter et al. 2015; Lever et al. 2009). If place cells are driven by local borders (e.g., Zhang et al. 2014), identical place fields would be observed in each compartment. In this view, as the angle between identical compartments or alleyways increases, boundary cell firing should correspondingly start to differentiate them. It is also possible, however, that other types of spatially tuned neurons represent the shape of local environments (e.g., Brotons-Mas et al. 2017) and thereby contribute to repetition of spatial firing fields.

Fig. 5.

Pattern repetition likely reflects environmental geometry. Color bar below A applies to A, B, and D, and color bar below C applies to C and E. A: example of a dCA1 place cell recorded in an environment where the walls could be moved to change its size. In the small square (top left) the cell has a field at top left. When the square’s length was extended (bottom left), the cell’s firing remains unchanged. However, when the square’s width was extended (top right), the place cell’s field extended proportionally. When the environment was extended isometrically the cell’s field faintly extends equally in all directions (bottom right). These results show that place cell firing is at least partly dictated by boundaries in the animal’s environment and that some boundaries exert more control over a given cell than others. Reproduced with permission from Macmillan Publishers Ltd., copyright 1996 (O’Keefe and Burgess 1996). B, middle: plot shows the firing rate map of a dCA1 place cell recorded in a square environment with a bisecting wall. Note that the cell has 2 fields, 1 on each side of the barrier. Bottom: plot of a modeled place cell generated using boundary vector cell (BVC) inputs that shows the same pattern of firing. Adapted with permission from Barry and Burgess (2007). C: firing rate maps of an example modeled BVC in 4 different-shaped environments. This cell maintains the same preferred firing direction (roughly northwest) and distance in all environments (modeled with the BVC model, Barry et al. 2006). Note that in plot at top right, where a barrier bisects the environment the BVC’s firing is also bisected and takes on a repetitive appearance (R. Grieves, unpublished data). D: example boundary cell recorded from the rat subiculum in a 3-platform environment. The cell fires along the west boundary of each platform, which in this case is a vertical drop. Adapted from Stewart et al. (2013), licensed under CC BY 3.0. E: a dCA1 place cell recorded in an elevated platform maze composed of 4 parallel alleyways. In this maze we can see that vertical drops are also sufficient to drive pattern repetition in place cells (Grieves 2015). This cell does not fire in the far right arm of the maze, and this is consistent with the findings of Spiers et al. (2015) and Grieves et al. (2016), which suggest that place field repetition is a continuous phenomenon. In repetitive environments, many place cells exhibit repeating fields in every subcompartment, but some only exhibit them in a minority of compartments and some do not exhibit repeating fields at all. This suggests that the strength of different inputs (e.g., geometry, self-motion) may vary for different place cells. Adapted with permission from Grieves (2015).

As an alternative, visual inputs could account for spatial field repetition. If the corners of a compartment or alleyway can function as visual cues, then parallel compartments or alleyways may fall on the retina in similar patterns at the same head direction. If the angle between these compartments is increased, however, this relationship will decrease. Thus place field repetition could arise from the congruence of visual and directional inputs. As with boundary cells, neurons that are sensitive to a conjunction of head direction and position can also be found in the retrosplenial cortex (Cho and Sharp 2001). Grid cells are also sensitive to visual and olfactory contextual changes (Chen et al. 2016; Marozzi et al. 2015; Pérez-Escobar et al. 2016), and changes in grid fields are correlated with remapping in place cells (Fyhn et al. 2007; Jeffery 2011; Miao et al. 2015; Monaco and Abbott 2011).

Are these inputs functionally different? Research suggests that there are differences in how visual information and boundaries are used. Field repetition can be observed in environments whether or not a distal visual cue is provided (Derdikman et al. 2009; Grieves et al. 2016), if proximal cues are provided (Fuhs et al. 2005), and even in the dark (Grieves 2015). This striking perseveration suggests that perhaps only local visual cues such as those utilized by Spiers et al. (2015) are enough to drive pattern separation and overcome field repetition, which would be suggestive of a contextual input, such as that from the EC. This is supported by the finding that in many environments humans and animals primarily utilize geometric information to orient themselves while ignoring contextual visual information (Cheng 1986; Hermer and Spelke 1994; Krupic et al. 2016; but see Hupbach and Nadel 2005; Learmonth et al. 2002). Furthermore, mice have been observed to utilize contextual visual cues to recognize an environment while continuing to make systematic heading errors, suggesting that contextual and geometric information may be processed and utilized by two separate systems (Julian et al. 2015b). One possibility is that place cell firing is largely and primarily dictated by geometric inputs from boundary cells but this input is mediated by a contextual input from EC, similar to the contextual gating model proposed by Hayman and Jeffery (2008).

The view proposed here is that on initial exposure to an environment a rapid process is initiated that relies heavily on geometric inputs from boundary cells to orient and arrange both place and grid fields. In a repetitive environment these inputs are identical in each local area and hippocampal pattern separation fails, resulting in repeating place fields. However, with greater exposure to an environment, information accumulated through path integration drives the repeating grid fields toward a global representation with low levels of field repetition (Carpenter et al. 2015) and this development in turn could potentially drive increasingly global (spatially unique) place fields. Evidence for rapid mapping based on geometry can be seen when comparing the timescales at which spatial cells develop their firing patterns. In novel environments boundary and head direction cells develop stable firing patterns instantaneously (Jankowski et al. 2015; Taube et al. 1990b; Taube and Burton 1995), whereas hippocampal place cells require 5–10 min to form stable place fields (Bostock et al. 1991; Frank et al. 2004; Hill 1978; Wilson and McNaughton 1993) and grid cells require a number of hours to stabilize (Barry et al. 2012). Visual inputs also play an important role within this framework. For instance, when large contextual changes occur within an environment, like the color change of a subcompartment, EC cells locally remap, which allows for greater pattern separation in the hippocampus in the altered compartment.

Remaining Challenges

A central theme of this review is that place cells, and to an extent grid cells, are driven by local boundaries and a directional input. If these are congruent across maze compartments, repetition of firing fields is observed. This suggests that, at least initially, the mapping of external, allocentric space in the mammalian brain is local and not global.

Grid cell field fragmentation and place field repetition are strikingly similar and would appear to represent the same phenomenon. However, several questions remain. First, as place fields are still present after grid cell firing is abolished (Brun et al. 2008; Hales et al. 2014), does inactivation of the mEC affect hippocampal field repetition (or vice versa)? Second, do inputs from the subiculum, where many boundary cells reside, affect firing in either the mEC or the hippocampus? Indirect evidence for this is found in work showing that grid cells may be sensitive to border cell inputs (Hardcastle et al. 2015) and that lesions of the subiculum contribute to spatial navigation deficits (Morris et al. 1990). Third, what effects does disruption of the head direction system have on border/boundary cells (Burgess et al. 2001; Byrne et al. 2007)? Finally, does disruption of the head direction system affect place field repetition?

Given the framework of this review, without head direction input place cells should be reduced to relying purely on visual inputs, assuming that boundary cells require the head direction system. Do grid cells immediately form a global representation in radial compartments as place cells do, and how do contextual changes in local compartments influence grid cells? One prediction is that grid cells remap immediately after a compartment context change and that this is accompanied by remapping in place cells, but this has yet to be shown in a multicompartment environment. With a better understanding of these relationships we should gain insight into processing between the hippocampus, the EC, and the surrounding structures. Ultimately, this may inform the design of repetitive environments to minimize spatial ambiguity.

GRANTS

This work was supported by Biotechnology and Biological Sciences Research Council Grant BB/P001725/1.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

R.M.G., E.R.W., and P.A.D. conceived and designed research; R.M.G. performed experiments; R.M.G. and E.R.W. analyzed data; R.M.G., É.D., E.R.W., and P.A.D. interpreted results of experiments; R.M.G. and É.D. prepared figures; R.M.G., É.D., E.R.W., and P.A.D. drafted manuscript; R.M.G., É.D., E.R.W., and P.A.D. edited and revised manuscript; R.M.G., É.D., E.R.W., and P.A.D. approved final version of manuscript.