Rapid spatial reorientation and head direction cells

Abstract

It is surprising how quickly we can find our bearings when suddenly confronted with a familiar environment, for instance when the lights are turned on in a dark room. Subjectively, this appears to occur almost instantaneously, yet the neural processes permitting this rapid reorientation are unknown. A likely candidate is the head direction (HD) cell system. These limbic neurons found in several brain regions, including the thalamus and the hippocampus, discharge selectively when the head of an animal is oriented in a particular ("preferred") direction. This neuronal activity is independent of position and ongoing behavior and is thus likely to constitute a physiological basis for the sense of direction. Remarkably, although the HD cell system has properties resembling those of a compass, it is independent of geomagnetic fields. Rather, the preferred directions of the HD cells are strongly anchored to visual cues in the environment. Here, we bring evidence for the first time that a fundamental component of the capacity to rapidly reorient in a familiar environment may be brought about by updating of HD cell responses as rapidly as 80 msec after changes in the visual scene. Continuous attractor networks have been used successfully to model HD cell ensemble dynamics. The present results suggest that after large rotations of the surrounding landmarks, activity in such networks may be propagated in abrupt jumps rather than in a gradually progressive manner.

Fig. 1.

Experimental procedure. A, Because the rat remains immobile oriented in the (previously determined) preferred direction while drinking water from the reservoir (left), full response curves cannot be sampled. Rather, only the cell responses corresponding to this particular head direction can now be recorded (black circle in right panel). B, The light is turned off, and the card is rotated by 90° along the cylinder wall.C, The light is turned back on. This triggers a shift in the directional response curve of the neuron because this activity is anchored to visual cues (right panel). Accordingly there should be a marked decrease in firing rate (compare the filled circles inB and C, right panels). D, The light is turned off again, and the card is returned to the standard position. E, The light is turned back on. The preferred direction shifts back to its initial orientation (right panel). This corresponds to a marked increase in discharge frequency (compare the filled circles in D and E, right panels). Steps B through E are repeated until the rat is satiated and no longer remains immobile at the center.

Fig. 2.

Latency of preferred direction updates in HD cells. Raster plots (above), peri-event histograms (middle), and cumulative histograms (below) (bin width = 10 msec) of action potentials recorded from all of the HD cells analyzed. Time 0 indicates when the lights were turned on again. After light onset, the preferred directions return to their initial orientations (A) or shift to the rotated (nonpreferred) orientations (B). To determine the average latency of the preferred direction update, least-squares estimates were computed from the cumulative histograms using the first 250 msec of data after light onset (thick curves) (Friedman and Priebe, 1998). Transition points are at 80 ± 10 msec (A) for returns to the preferred orientation and 140 ± 10 msec (B) for shifts to the nonpreferred orientation. Brackets indicate trials from the same cell within a given session; the variations in spike density among the rows of rasters reflect differences in peak and background firing rates among the neurons.

Fig. 3.

A typical analysis of update latency on a trial by trial basis with the method of Seal et al. (1983). Raster plots show action potentials recorded from a single HD cell during a single session, when the firing rate increases (A) or decreases (B) after card rotation. Light onset occurs at time 0. For each trial, the update latency is computed as the maximum likelihood estimator of the change point in the mean interspike interval (thick vertical bars).

Fig. 4.

Two possible mechanisms for dynamic updating in continuous attractor networks. A, Network connectivity. Each cell (circle) sends excitatory signals (triangles) to its neighbors and inhibitory signals (bars) to all of the cells in the network (for clarity, only the connections from one prototypic cell are shown). B, Progressive updating of the ensemble response of the HD system. The firing rate of each formal cell is proportional to the height of the vertical bar. The hill of activity migrates progressively to the target population. C, Abrupt updating of the ensemble response. The hill of activity jumps to the target firing pattern without activation of intermediate neurons.