Neural activity during monkey vehicular wayfinding

Abstract

Navigation gets us from place to place, creating a path to arrive at a goal. We trained a monkey to steer a motorized cart in a large room, beginning at its trial-by-trial start location and ending at a trial-by-trial cued goal location. While the monkey steered its autonomously chosen path to its goal, we recorded neural activity simultaneously in both the hippocampus (HPC) and medial superior temporal (MST) cortex. Local field potentials (LFPs) in these sites show similar patterns of activity with the 15-30 Hz band highlighting specific room locations. In contrast, 30-100 Hz LFPs support a unified map of the behaviorally relevant start and goal locations. The single neuron responses (SNRs) do not substantially contribute to room or start-goal maps. Rather, the SNRs form a continuum from neurons that are most active when the monkey is moving on a path toward the goal, versus other neurons that are most active when the monkey deviates from paths toward the goal. Granger analyses suggest that HPC firing precedes MST firing during cueing at the trial start location, mainly mediated by off-path neurons. In contrast, MST precedes HPC firing during steering, mainly mediated by on-path neurons. Interactions between MST and HPC are mediated by the parallel activation of on-path and off-path neurons, selectively activated across stages of this wayfinding task.

Keywords: Cortex; Navigation; Neurophysiology; Wayfinding; hippocampus.

Fig. 1

(A) The monkey faced the front wall of the room from a cart-mounted chair. The monkey used an X-Y joystick to steer cart movement to the randomly selected one-of-eight goal positions. To initiate a trial, the display first showed the monkey's current position (start circle) and its goal location (square) for 1 s, thereafter showing only the goal position. The monkey steered to the goal, remaining there for 1 s to earn a liquid reward and trigger the next trial. (B) Path traces (lines) and vector fields (arrows) show the monkey's paths in trials starting at the rear-middle of the room (position 7) to a goal at the front-left corner (position 1). The monkey autonomously generated paths, mostly straight ahead and then left, or left-forward and then straight ahead. (C) Scatter plot of path characteristics showing shorter, single and dual axis tasks. Abscissa: $\mathbf{log}\left(\frac{\mathbf{observed\; length}}{\mathbf{shortest\; length}}\mathit{-}\mathbf{1}\right)$. Ordinate: Heading direction change (log total curvature). (D) Diagram indicating the location of the neurophysiological recording chamber (2 × 3 cm) centered at position (1.5 × 2.5 cm, center AP +3, ML ±15 mm, angle -12^o), and the path of microelectrode penetrations to HPC (blue arrow) and MST (red arrow) superimposed on a left coronal section (Fig. 97 in [49]). (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 2

Simultaneous recordings in HPC and MST yield neural-place maps. (A) Neural activity in HPC (top row) and MST (bottom row) during steering. Left: Place distribution of recorded response power across 4 LFP frequency bands (bottom to near top: 0–7, 8–14, 15–29, 30–100 Hz) and SNR firing rates (top) show greatest LFP power when the monkey is at the middle-rear of the room, with little place specificity in the SNRs. Middle: Weighting of LFP bands and SNR firing rates in an optimization of mutual information (MI) about the monkey's location in the room, a well-established approach to measuring the degree to which room location in encoded by neural signals [56]. In both HPC (top) and MST (bottom) the highest weight is from 15 to 29 Hz LFP power, the lowest is from SNR firing rates. Right: Optimized place maps from LFPs and SNRs showing highest neural activity in the middle-rear of the room (HPC top, MST bottom). (B) Left: MIs from all simultaneously recorded sites show concordance between HPC and MST with stronger MIs from higher frequency LFPs and little effect of SNRs. Middle: Neural-place MI optimizations yield highest coefficients from 15 to 100 Hz LFPs and negligible coefficients from SNRs. Right: MI optimizations from HPC and MST yield uneven distributions of high (red circles) and low (blue circles) with higher neural activity near the center rear, and lower neural activity around near start/goal locations in the front and left corners. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 3

Neural-place representations in start-turn-goal space. (A) MI place optimization of neural activity for all trials starting at the rear-middle of the room (position 7) and ending at the front-left of the room (position 1). Left: HPC (top row) and MST (lower row) show different room distributions of neural activity across LFP bands and SNRs. Middle: Optimization weights across LFPs and SNRs show strongest contributions from high frequency LFPs, with little effect from SNRs. Right: Optimized maps from HPC and MST both show lowest activity around the start position and highest activity around the goal position. (B) MI optimization of all recorded trials aligned in start-turn-goal coordinates. Left: Neural activity maps in start-turn-goal coordinates with the most distinct suppression at the start and activation at the goal being evident in the high frequency LFPs. Middle: Optimization weights across LFPs and SNRs confirm the strongest contributions from high frequency LFPs, with the least effect from SNRs. Right: Start-turn-goal MI optimized maps show least activity at the start position and secondarily around the turn position, and greatest activity at the goal position and secondarily at the far corners.

Fig. 4

Single neuron responses of all trial types plotted in start-turn-goal space. (A) Left: An MST neuron with activity concentrated along the paths steered between start and goal locations (β weight from linear regression of path vector field and neural activity map). Right: An HPC neuron with activity concentrated beyond the bounds of the paths steered between start and goal locations. (B) Distributions of β weights across all HPC and MST SNRs showing the shared continua from negative weights reflecting off-path activity to positive weights reflecting on-path activity.

Fig. 5

Combined HPC and MST SNR populations partitioned by on- vs off-path response relations. (A) Left: On-path SNRs show higher firing rates (red, yellow) concentrated within the range of the monkey's steered paths through start-turn-goal space. Right: Off-path SNRs show higher firing rates (red, yellow) concentrated beyond the bounds of the monkey's steered paths through start-turn-goal space. (B) Regression of the simultaneously recorded LFP bands and SNR activity in relation to the monkey's steered paths during those recordings. On-path SNR recording sites (left) and off-path SNR recording sites (right) are shown as bar graphs of the β weights for each LFP band and SNR sub-population firing. Left: On-path SNR recording sites show modest contributions across all LFP frequency bands with the strongest contribution from SNR activity. Right: Off-path SNR recording sites show greater contributions from higher frequency LFPs: the 15–30 hz frequency band with a strong negative β weight, the 30–100 hz frequency band with a strong positive β weight, and SNR activity with a strong β negative similar to of the 15–30 hz LFPs. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

Fig. 6

Granger causality analyses of SNR relations between HPC and MST suggest distinct inter-relations in on-path (upper row) and off-path (lower row) SNRs. Significant inter-relations (dashed lines, p < 0.05, corrected) are seen in all five stages of autonomous wayfinding. (A) Left: On-path HPC neuronal firing does not show significant relations with subsequent MST neuronal firing during any of the five wayfinding task periods. Right: On-path MST neuronal firing shows significant relations with subsequent HPC neuronal firing during the turn and steer wayfinding task periods. (B) Left: Off-path HPC neuronal firing shows significant relations with subsequent MST neuronal firing during the start and steer wayfinding task periods. Right: Off-path MST neuronal firing shows significant relations with subsequent HPC neuronal firing during the start, goal, and lost wayfinding task periods.