Translation between whole-body egocentric and allocentric reference frames
Egocentric to allocentric reference frames manipulation consists in transforming information coded in egocentric coordinates into an observer-independent representation.
The retrosplenial cortex is thought to be involved in the allocentric-to-egocentric and egocentric-to-allocentric transformation processes (Burgess et al., 2001a; Byrne et al., 2007; Vann et al., 2009). Retrosplenial inactivation has been shown to impair spatial memory in rodents (for review see (Troy Harker & Whishaw, 2004)) but the magnitude of spatial deficits are smaller than the magnitude of the deficits associated with hippocampal damage. The full impact of retrosplenial cortex lesions appears when animals are forced to shift modes of spatial learning, for example from dark to light (Cooper & Mizumori, 1999) or from local to distal cues (Vann & Aggleton, 2004; Pothuizen et al., 2008).
In humans, a functional imaging study specifically identified the retrosplenial as being involved in acquiring allocentric knowledge of an environment from ground-level observations (Wolbers & Buchel, 2005). Subjects repeatedly encountered a virtual environment composed of intersections and shops and were then tested on their knowledge of the topographical organization of the environment. The retrosplenial cortex was the only structure to parallel behavioural measures of map expertise of the subjects. In another study, Spiers and Maguire (2006) scanned participants navigating through a realistic virtual-reality simulation of central London to determine when during navigation different brain areas are engaged. Activity in the retrosplenial increased specifically when new topographical information was acquired or when topographical representations needed to be updated, integrated or manipulated for route planning (Spiers & Maguire, 2006). This could relate to the proposal that the retrosplenial acts as a short-term buffer for translating between egocentric and allocentric representations (Burgess et al., 2001a; Byrne et al., 2007).
The retrosplenial cortex has reciprocal projections with the hippocampal formation and thalamic nuclei (van Groen & Wyss, 1990; 1992; 2003). In rodents and primate it can be separated into dysgranular and granular region. The granular regions is distinguished by reciprocal connections with sites that contain head-direction cells (the lateral dorsal and anterior dorsal thalamic nuclei and the post-subiculum), whereas the dysgranular cortex region is more interconnected with visual areas through reciprocal connections (Vogt & Miller, 1983). It has been estimated that 10% of retrosplenial cortex cells are head-direction cells, equally distributed across the dysgranular and granular cortex; but interestingly, in the granular region only, the spatial tuning of some of these cells is modulated by velocity of locomotion (Chen et al., 1994). Immediate-early gene and lesion studies suggest that, in accord with the anatomical connections, the dysgranular retrosplenial cortex seems more important for visually guided spatial memory and navigation, whereas the granular retrosplenial cortex has a greater involvement in internally directed navigation (Vann & Aggleton, 2005; Pothuizen et al., 2009).
Place information: element of the mental representation of space
This section will describe the properties of three types of cells: head direction cells, grid cells and place cells. These cells were all discovered in rats but two of them have since been identified in humans. In 2003, Ekstrom et al. recorded from electrodes implanted in several regions as participants, epileptic patients, navigated in a virtual town. The authors found cells that could be confidently classified as place cells, amongst which the majority was found in the hippocampal region (Ekstrom et al., 2003). Grid cells have also been found in the entorhinal cortex and in the cingulate cortex using functional imaging (Doeller et al., 2010) and electrophysiological recordings (Jacobs et al., 2013).
Head direction (HD) cells fire when a rat’s head is facing a specific direction relative to the environment, irrespective of its location or whether it is moving or still (Taube et al., 1990a; Taube et al., 1990b). These cells were first discovered in the dorsal portion of the rat presubiculum (often referred to as the postsubiculum (PoS)) (Ranck, 1984) but have since been found in several regions belonging to the Papez circuit: the anterior dorsal thalamic nucleus (Taube, 1995a), the lateral mammillary nucleus (Stackman & Taube, 1998), the retrosplenial cortex (both granular and agranular regions) (Chen et al., 1994), the parasubiculum (Boccara et al., 2010) and entorhinal cortex (Sargolini et al., 2006). HD cells have also been identified in other non–Papez circuit areas, including the lateral dorsal thalamic nuclei (Mizumori & Williams, 1993), the dorsal striatum (Wiener, 1993) and the medial precentral cortex (also known as FR2 or AGm cortex) (Mizumori et al., 2005). Other areas in which they have been reported in smaller numbers include the medial prestriate cortex (Chen et al., 1994), CA1 hippocampus (Leutgeb et al., 2000), and the dorsal tegmental nucleus (DTN) (Sharp et al., 2001).
These cells are thus found in multiple structures which are anatomically interconnected. Lesion and electrophysiological studies have shown that the head-direction signal travels from the dorsal tegmental nucleus, which receives input from vestibular nuclei, to the hippocampus, through the hypothalamus (mammillary nucleus), the thalamus and the retrosplenial, subicular and entorhinal cortices (Taube, 2007). It is however important to keep in mind that most of the connections between these structures are bidirectional and that the transfer of head direction information is not necessarily unidirectional.
The primary correlate of head-direction cells is the orientation of the head in the horizontal plan. Pitch and roll appear to be relatively unimportant, as is the orientation of the rest of the body. Figure 8a shows the firing rate for one HD cell plotted as a function of the direction of heading. It has a single preferred direction, and firing falls off rapidly as the head direction rotates away from the preferred direction. The distribution of peak firing directions across the population of cells is uniform, with no direction preferred over any other. There does not appear to be any topography to the directions represented by neighbouring neurons. Importantly, unlike hippocampal place cells which can be active in only one environment, HD cells fire in all environments tested.
Identifying a place: hippocampal place cells
Since their discovery in 1971 (O’Keefe & Dostrovsky, 1971), there have been many technical and conceptual development in hippocampal single-unit recording which have shed light on different properties of place cells.
Their major behavioural correlate is the animal’s location. O’Keefe and Dostrovsky reported that these place cells were silent as the rat moved around the environment until it entered a small area of the environment, the place field, where the cell began to fire (O’Keefe & Dostrovsky, 1971). When a rat is first exposed to an environment, place cells acquire their place fields within a few minutes (Hill, 1978) and once their firing fields are established, their locations may be stationary for weeks or months (Muller et al., 1987; Thompson & Best, 1990). The activity of an ensemble of place cells covers a whole environment (Wilson & McNaughton, 1993). The same cells often fire in different environments, but the preferred locations are unrelated if the environments are sufficiently dissimilar from each other (Leutgeb et al., 2005). One notable feature of these place cells is that in unconstrained open fields – environments in which the animal is free to move in all directions – the cells fire in the place field irrespective of the direction in which the animal is facing (Muller et al., 1994). In environments that constrain the animal’s behaviour, for example a linear track or a radial maze, the cells become directionally sensitive and may be said to represent the successive locations along a path (McNaughton et al., 1983; O’Keefe & Recce, 1993; Muller et al., 1994). Whereas in the first situation each cell may be said to represent a location, in the second it might more properly be described as representing a serial position along a path. Place fields can be controlled by allothetic and idiothetic information (Jeffery et al., 1997).
Information from the environment comprises visual cues, such as distal cues (Cressant et al., 1997) or wall configuration of the maze (considered as intra-maze cues) (O’Keefe & Burgess, 1996). Indeed rotation of the box or rotation of a “cue-card” suspended prominently on the interior wall of the box, generally causes equal rotation of the place fields (Figure 12) (Kubie & Ranck Jr, 1983; Muller & Kubie, 1987). On multi-arm mazes with high walls, the fields sometimes take their reference from specific arms and not from the entire maze (Shapiro et al., 1997).
Recognizing a place: pattern completion, pattern separation
The different areas of the hippocampus, dentate gyrus (DG), CA3, CA2 and CA1, all contain place cells (even though in the DG these cells are granule cells and not pyramidal cells) (Muller et al., 1987; Jung & McNaughton, 1993; Mankin et al., 2013). Regarding place recognition, the CA3 and DG have been attributed two specific roles: “pattern completion” and “pattern separation”, respectively.
Patten completion is a phenomenon through which initial association between multiple stimuli allows the subsequent retrieval of the whole encoded event when the subject is presented with just a subset of the original stimuli. It is hypothesised to be an important role for CA3 due to its intrinsic network of recurrent collaterals. Experimental evidence supporting the role of CA3 in pattern completion comes from studies involving mice with a deletion of the NR1 gene specific to the CA3 region (resulting in no NMDA receptor mediated plasticity) which were unable to navigate in a water maze task when a subset of their previously learned cues were removed (Nakazawa et al., 2002). This corresponds to the early hippocampal model of David Marr in which he proposed that the area CA3 was ideally suited as a site for memory storage and retrieval because of this region’s recurrent connections which could allow storing patterns of activity and later retrieving these same patterns if all or part of the original pattern re-occurred (Marr, 1971).
Table of contents :
PART I Navigation and Mental representation of space
1. Mental representation of space
1.1. Tolman’s “Cognitive maps”
1.2. The discovery of hippocampal place cells, substrate for Tolman’s cognitive map
1.3. From cognitive maps to mental representations of space
2. Building a mental representation of space – processes and neural substrates
2.1. Multi-modal information integration
2.1.1. Allothetic information
2.1.2. Idiothetic information
2.2. Reference frame manipulations
2.2.1. Head-to-body and head-to-world transformations
2.2.2. Translation between whole-body egocentric and allocentric reference frames
2.3. Place information: element of the mental representation of space
2.3.1. Head-direction cells
2.3.2. Grid cells
2.3.3. Boundary cells
2.3.4. Identifying a place: hippocampal place cells
2.3.5. Recognizing a place: pattern completion, pattern separation
3. Using the mental representation of space – processes and neural substrates
3.1. Goal information
3.2. Planning & Decision-making
3.3. Organization of action sequences
PART II Neural Dynamics across Task Learning
1. Learning stages
2. Network dynamics across learning
2.1. Imaging results
2.2. Theoretical models
3. Learning stages in rodent studies
3.1. Neural dynamics in goal-directed and habitual behaviours
3.2. Neural dynamics in spatial learning
PART III From mental representations of space to spatial memory
1. Different forms of memory
1.1. Sensory memory
1.2. Short-term memory
1.3. Long term memory – storage and consolidation
2. Different memory systems in long-term memory
2.1. Non-declarative memory
2.2. Declarative memory
2.2.1. Episodic memory
2.2.2. Semantic memory
PART IV: Navigation Strategies: Identifying the mental representation of space
1. Navigation strategies and computational processes
1.1. Path integration
1.2. Goal or Beacon approaching
1.3. Stimulus-response strategy
1.4. Map-based strategy
1.5. Reinforcement learning to model stimulus-response and map-based strategies
2. Tasks to identify strategies and their neural basis
2.1. Stimulus-response and map-based strategies
2.2. The starmaze task and sequence-based navigation
3. Sequence-based navigation (or sequential egocentric strategy): what computational processes and what neural basis?
3.1. Sequential egocentric strategy
3.2. Neural basis of sequence learning in rodents
3.3. Hippocampus and sequential activity
3.3.1. Retrospective and prospective firing
3.3.2. Sequence representation
3.3.3. Time cells
Conclusion and experimental question
METHODS FOCUS: FOS IMAGING
1. Fos imaging
2. Network analysis
2.1. Functional connectivity
2.2. Graph theory
Article 1: Complementary Roles of the Hippocampus and the Dorsomedial Striatum during Spatial and Sequence-Based Navigation Behaviour
2. Main results
Article 2, in preparation: Functional connectome and learning algorithm for sequence-based navigation
2.1. Behavioural study
2.2. Fos imaging
2.3. Computational learning study
5. Supplementary material
1. Hippocampus and sequence-based navigation
2. Model-free reinforcement learning with memory and sequence-based navigation
3. Network underlying sequence-based navigation