Spatial Navigation: Head Direction Cells Are Anchored by Gravity

Abstract

Thalamic neurons that signal an animal’s direction of heading are crucial for spatial navigation. Both directional coding and flexible use of spatial memory are upended, however, when a rat has to find its way while upside down.

Adaptive behavior depends on predicting accurate relationships between percepts, actions, and outcomes. Some of these relationships generalize across niches, and neural mechanisms for computing predictions among them are conserved. Oculomotor systems predict how retinal images change with eye or head movements, and violating these predictions impairs cognition in surprising ways. I remember wearing prism lenses that reversed expected and actual horizontal visual movement and being utterly disoriented when I tried to walk across a room. Analogous systems use spatial computations to guide animals as they navigate about the environment, combining distance, heading, and location signals to track movement across the surface of the world. A paper published recently in Current Biology by Gibson et al. [1] reports evidence that these two-dimensional navigation systems are tethered by gravity.

Biological mechanisms for orienting in gravity begin in the otolithic organs of the inner ear, where hair cells transduce linear acceleration. The vestibular nuclei relay angular velocity signals to other brainstem, thalamic, and cerebellar nuclei that convey acceleration information to the spinal cord and the cerebral cortex. The dorsal tegmental and lateral mammillary nuclei integrate vestibular with other perceptual and motor signals across levels of processing. Gravity, equivalent to a constant acceleration, activates a subset of these hair cells even when the head is still [2].

Head direction cells integrate vestibular signals with other percepts to compute horizontal orientation. Head direction cells act like neuronal Geiger counters with receptive fields that span whole environments and have one trigger feature — the animal’s heading angle. First discovered in the rat presubiculum [3], head direction cells have been recorded in several interconnected structures that receive vestibular inputs, including mammilary nuclei, the anterior and lateral dorsal nuclei of the thalamus, and the retrosplenial and entorhinal cortices. Head direction cells fire at relatively low base rates until the rat’s head is pointing within about 60° of the cell’s preferred angle, when the firing rate can increase to 60 spikes per second. The preferred heading angle of a given head direction cell is acquired as animals explore an environment, is stable in a constant environment, and changes if prominent visual cues are altered.

Heading signals are crucial for ‘path integration’, the ability to track changing location by combining the distance and angles of self-generated movement. Vestibular signals allow head direction cell signals to persist in the dark, but they ‘drift’ over time until a familiar landmark is encountered that corrects or resets them. Head direction cell activity is weakened or abolished by lesions of the vestibular apparatus, especially the semi-circular canals; inactivation of the otoliths has less effect, which could imply that gravity plays a relatively unimportant role in heading signals. The new work of Gibson et al. [1] overturns that implication, however, by showing that both head direction cell activity in the anterodorsal thalamic nucleus and navigation are severely disrupted when animals walk upside down.

Gibson et al. [1] trained rats to cling to a circular wire mesh ‘ceiling’ and, while upside-down, walk from starting points at the edge of the apparatus to one of four potential escape hatches near the middle of each quadrant. The correct escape hatch let the rat climb above the mesh and stand up. After more than 100 training trials, the rats learned to move directly to the escape hatch from each of two different starting points, but they could not learn four. The task is operationally similar to the Morris water maze [4], in which rats learn to escape from the water by swimming from several starting points to a platform hidden in a constant location. The slow learning and limited performance in the clinging task contrasts markedly with the rapid learning and highly flexible performance in the water maze, where rats escape directly from four starting points after only a few trials. The tenfold difference in learning rates suggests that different mechanisms guide learning in the two navigation tasks.

Spatial behavior can be guided by several mechanisms, including taxons and cognitive mapping [5]. Taxons refer to directed responses with respect to a single stimulus, such as approaching a poster on a wall. Such cue approach strategies are impaired by lesions of the dorsolateral striatum that do not impair cognitive mapping [6]. Navigation by cognitive mapping combines spatial representations, defined by relationships among several stimuli, with path integration signals that indicate the direction and distance of movement. Rats demonstrate cognitive mapping by finding familiar spatial goals despite the removal of one or more visual cues, or after being placed in an unfamiliar start location [5]. Both types of behavioral flexibility are typical in water maze tests, suggesting that rats use relationships among distal spatial cues to navigate to the hidden platform.

Gibson et al. [1] assessed the strategy used by the inverted rats by placing the animals in an unfamiliar starting point at the center of the apparatus [1]. The rats headed in random directions and showed no sign of learning after several trials, consistent with their inability to learn to escape from more than two starting points. The inability to generalize in a highly familiar environment, together with slow learning in the familiar start task, suggest that the rats did not use a spatial mapping strategy, but rather used a taxon strategy to approach specific distal cues. Indeed, when a curtain blocked the distal cues from view, the upside down rats made random headings even from the familiar starting points, as though the rats no longer had a sense of direction. Indeed, head direction signals were lost when the rats were upside down.

Previous work showed that head direction cell coding was maintained when rats walked along a horizontal maze arm and then climbed up a vertical wall [7], suggesting that heading signals might be independent of gravity. Gibson et al. [1] recorded anterodorsal thalamic nucleus cells as rats explored a cylinder on the floor of the testing room before and after they performed the inverted navigation task [1]. They found that the anterodorsal thalamic nucleus cells had sharply-tuned directional firing correlates when the rats walked on the floor, but the signals were disrupted when the same cells were recorded when rats were upside down. The animals performed the inverted task reliably and followed consistent trajectories from the two familiar starting points, but the peak firing rates of the head direction cells declined, background firing increased, and directional tuning curves were noisy and inconsistent. During the center probe tests, the directional tuning declined further. Head direction cell activity returned to normal when the animals walked upright in the cylinder afterwards, and the strong tuning curves were the same before and after the inverted tests. Despite their ability to use distal cues to guide goal-directed behavior, the rats lost head direction signals and cognitive mapping when journeys started upside down.

Navigation requires several interconnected limbic structures [4,8]. Head direction cells interact with cells in the hippocampus that signal location, and medial entorhinal cortex cells that code distance [9,10]. Rats with head direction cell dysfunction do not learn the location of a hidden platform in the water maze, but readily learn to approach a visible platform [11]. Hippocampal dysfunction causes similar deficits [4]. Hippocampal neurons code locations via place fields, small regions in an environment that elicit high firing rates when visited by a rat [5]; 60 hippocampal cells can predict the location of a rat’s head within 2 cm² [12]. Like head direction cells, place fields are influenced by both external stimuli and path integration [13]. The neuropsychological parallels suggest that place and head direction signals interact, and indeed head direction cells contribute to stable place fields and vice versa. Anterodorsal thalamic nucleus lesions decrease the specificity and stability of hippocampal place fields [10], and hippocampal lesions reduce the stability of head direction cells [14].

Computational models of navigation propose that recurrent activation of place, distance, and heading encodes and retrieves spatial paths [5,13,15,16]. The models suggest that moving through an environment activates interconnected head direction, grid, and place units so that specific trajectories are represented by direction–distance–place code sequences and stored by synaptic plasticity. Familiar trajectories could be retrieved as place-direction-distance sequences; for example, when placed in a familiar start location in a water maze, a rat’s active hippocampal cells would signal that location and activate head direction cells predicting the platform direction. The head direction cells would activate grid cells that signal distance, and together these would activate the next hippocampal spatial representation on the way to the platform, and so on. Heading angles thereby predict future locations by activating spatial sequences.

These mechanisms explain how navigation is accomplished by path integration despite cue removal, when both head direction cell and place signals are maintained, and why navigation is impaired when heading signals are disrupted. Furthermore, the models predict that place fields, like head direction cells, should be disrupted in inverted rats, which has not yet been tested. Head direction cell dysfunction could untether the orientation of otherwise intact place field maps, or disrupt activation of place field sequences [16]. Computing novel routes requires associating goals with locations and headings, so would presumably depend upon head direction signals [17].

That head direction cells and navigation depend on gravity has important implications for understanding the flexible use of memory in general. In rats and other species, the hippocampus, entorhinal cortex, and dorsal thalamus are required for memory tasks that do not entail spatial navigation. Contextual memory retrieval of olfactory and visual associations, odor sequences, transitive inference, and trace eyelid conditioning are impaired by similar lesions in rats. People with homologous brain damage have problems learning new facts and remembering recent events [18]. Cognitive mapping theory proposes that spatial computations define the fundamental operations of memory systems, and that these other types of memory rely on these. Relational memory theory proposes that spatial memory exemplifies a general memory system that associates stimuli that overlap in time into events and sequential episodes [19]. The computational differences between spatial navigation and other forms of flexible memory remain unclear. Consider the example of rats trained to distinguish between two olfactory sequences that begin and end with distinct odors but contain an overlapping set of smells in the middle [19].

Compare that to another example in which rats are trained to distinguish between two spatial paths that begin and end in different places but have an overlapping middle section. Both tasks are impaired by hippocampal damage, but to what extent do they require the same computations? If non-spatial memory processing depends upon navigation mechanisms, then any manipulation that impairs navigation should also impair nonspatial memory. Head direction cells should be crucial for the spatial task, as heading (along with goal) information is crucial for linking successive places along each journey. If nonspatial memory processes depend upon navigation mechanisms, then head direction cells should be required for disambiguating olfactory sequences. Perhaps different odor sequences trigger different head direction cells, and ‘navigation’ proceeds through an olfactory space. If, however, head direction signals comprise one type of predictive, directed association among others, then head direction cells should be irrelevant to the olfactory sequence task.

From this view, non-spatial relational memory tasks engage parallel inputs to the hippocampus, so that disrupting spatial navigation selectively, by disrupting head direction cells, should leave intact other forms of relational memory. The study by Gibson et al. [1] provides a nice opportunity to distinguish these possibilities, simply by testing the rats in an explicitly nonspatial memory task while they are inverted. If relational memory rides atop spatial navigation mechanisms, then it should fall when rats are upside down and head direction cell signals are lost. If navigation exemplifies a broader set of memory computations, then the loss of head direction cell signals should spare non-spatial relational memory.