Neural mechanisms of self-location

Abstract

The ability to self-localise and to navigate to remembered goals in complex and changeable environments is crucial to the survival of many mobile species. Electrophysiological investigations of the mammalian hippocampus and associated brain structures have identified several classes of neurons which represent information about an organism's position and orientation. These include place cells, grid cells, head direction cells, and boundary vector cells, as well as cells representing aspects of self-motion. Understanding how these neural representations are formed and updated from environmental sensory information and from information relating to self-motion is an important topic attracting considerable current interest. Here we review the computational mechanisms thought to underlie the formation of these different spatial representations, the interactions between them, and their use in guiding behaviour. These include some of the clearest examples of computational mechanisms of general interest to neuroscience, such as attractor dynamics, temporal coding and multi-modal integration. We also discuss the close relationships between computational modelling and experimental research which are driving progress in this area.

Figure 1

Neural representations of self-location in the hippocampal formation. (A) Left, schematic of single unit recording. A rodent with chronically implanted extracellular electrodes forages in an open environment, with surrounding sensory cues for orientation (not shown). Tracking data from an overhead camera are synchronized with neural data. Middle, raw data from a place cell. The animal’s path is indicated by the black line, and action potentials are superimposed in red at the locations where they were emitted. Right, a firing rate map of the raw data; binned spike count is divided by binned dwell time and locally smoothed to calculate average firing rate. ‘Hotter’ colours indicate higher firing rates reaching a maximum of 8.3 Hz (indicated above the map), dark blue indicates low rate (0 to 20% of the peak rate), white bins are unvisited. This CA1 place cell is only active when the animal occupies a small area on the west of the environment. (B) Raw data (left) and firing rate map (middle) for a mEC grid cell. The multiple circular firing fields are arranged in a close packed hexagonal lattice. Right, the regular grid-like firing pattern is characterised by its orientation, spacing, and offset. (C) Two head direction cells recorded from the deep layers of mEC; similar directional responses are exhibited by head direction cells found in other brain regions. The polar plots show firing rate as a function of head direction; the cell on the left has a peak firing rate of 26.8 Hz achieved when the animal was facing an orientation of 42° relative to the environment (measured anti-clockwise from the horizontal axis). (D) A boundary vector cell in the subiculum, showing the raw data (left) and firing rate map (middle). The boundary vector cell fires whenever there is an environmental boundary a short distance to the south. The boundary vector cell shows a second firing field after an east–west oriented barrier is put into the environment (right).

Figure 2

The boundary vector cell model of place cell firing. (A) A boundary vector cell responds when a boundary (black line) occupies its receptive field which is peaked at a preferred distance and allocentric direction from the animal (a short distance to the northeast in this example, left). This boundary vector cell will have a firing rate map with raised firing along the northeast boundary of the environment (right). (B) The activity of a place cell is modelled as the summed and thresholded activity of a population of heterogeneous boundary vector cells (with different preferred distance and direction tunings, above). The place cell’s firing (below) can be estimated for any geometric arrangement of environmental boundaries. (Adapted with permission from .)

Figure 3

Schematic continuous attractor networks. (A) Left: the head direction system can be modelled as a circular continuous attractor network. Head direction cells (grey circles) are shown arranged according to their preferred firing direction, with raised firing rates indicated by warmer colours. Cells are reciprocally connected with their neighbours such that those with more similar preferred firing directions have stronger connectivity (schematic connections from the most active cell shown in red: thicker arrows indicating stronger connectivity). Connectivity is translation invariant such that all cells have similar connectivity profiles (connectivity for a second cell is shown in light grey). Activity in the head direction cell population will settle into a stable state, comprising a single activity ‘bump’, signalling the animal’s facing direction (schematic activity ‘bump’ shown in red, below). Middle and right: the activity bump can be moved around the circle by introducing an asymmetry into inter-cellular interactions. To track the animal’s heading, the strength of this asymmetric component must be proportional to the angular velocity of the animal’s head. This can be achieved by ‘shifter cells’ whose firing is modulated by head direction and angular velocity. Asymmetric connectivity producing an anti-clockwise rotation is shown (above) with the activity bump (below). (B) Left: populations of grid cells with grid-like firing patterns of the same orientation and scale can be modelled as a two-dimensional continuous attractor network: grid cells are shown arranged in a sheet according to the relative offset of their firing patterns. Cells are reciprocally connected with their neighbours so that those with closer firing patterns have stronger connectivity; this connectivity pattern is translations invariant across the sheet of cells. Again, a stable activity bump will form, signalling a static location. Right: the activity bump can be moved across the sheet of cells by asymmetric connectivity. To track the animal’s movement, the strength and direction of this asymmetrical component must be proportional to the animal’s velocity. This can be achieved by ‘shifter cells’ with grid-like firing that is also modulated by running velocity (that is, similar to conjunctive grid cells recorded from layers III–VI of the mEC [29]). The periodic nature of the spatial firing patterns corresponds to toroidal boundary conditions such that, as the activity bump moves off one side of the sheet, it appears on the other side.

Figure 4

Theta-band oscillations structure spatial activity in the hippocampal formation. (A) Local field potential (LFP) recorded from the CA1 pyramidal cell layer of a moving rat. Black trace, raw mean-normalised LFP, in which the 8 Hz theta modulation is visible along with higher frequency gamma oscillations (the signal band-pass filtered in the 6–12 Hz range is shown in red). (B) Theta phase precession in a CA1 place cell: each point indicates, for a single action potential, the theta phase (y axis) and animal’s location (x axis); data from multiple runs through the place field, moving left to right. Red line indicates the circular-linear regression of phase on position. (C) Schematic of the oscillatory interference model showing two components: a baseline oscillation (blue, with frequency fb) and a velocity controlled oscillator (red) whose frequency (fa1) varies from baseline proportionate to the animal’s running speed in direction φi. (D) Interference pattern generated between the active and baseline oscillations in (C). Spikes are emitted at the peaks of the carrier (black) which is the sum of the two oscillations, showing a repeating periodic pattern. (E) If velocity controlled oscillator frequency fa varies around the baseline frequency with the animal’s movement proportional to the speed and the cosine of direction relative to a preferred direction (radial black arrow), then the baseline and velocity controlled oscillator sum to produce a spatially stable striped pattern perpendicular to the preferred direction. Multiple velocity controlled oscillators with preferred firing directions selected to differ by multiples of 60° produce a grid-like firing pattern. (Adapted with permission from .)

Figure 5

Potential arrangements of self-motion and environmental inputs to grid cells and place cells. The neural representations of self-location expressed by place and grid cells reflect both environmental and self-motion information. Grid cell activity (as well as head direction cell activity, not shown) is strongly influenced by self-motion cues which are proposed to originate from conjunctive shifter cells or velocity controlled oscillators. Place cells are strongly influenced by environmental information, including that relating to environmental boundaries (boundary vector cells), and to local cues (potentially from lateral entorhinal cortex) as well as non-spatial ‘contextual’ inputs that may ‘gate’ spatial inputs . Regions containing grid cells and place cells are reciprocally connected, which may allow both representations to reflect an optimal combination of self-motion and environmental information . Alternatively, self-motion and environmental information may respectively reach place cells and grid cells directly (dashed lines).