Place Cell-Like Activity in the Primary Sensorimotor and Premotor Cortex During Monkey Whole-Body Navigation

Abstract

Primary motor (M1), primary somatosensory (S1) and dorsal premotor (PMd) cortical areas of rhesus monkeys previously have been associated only with sensorimotor control of limb movements. Here we show that a significant number of neurons in these areas also represent body position and orientation in space. Two rhesus monkeys (K and M) used a wheelchair controlled by a brain-machine interface (BMI) to navigate in a room. During this whole-body navigation, the discharge rates of M1, S1, and PMd neurons correlated with the two-dimensional (2D) room position and the direction of the wheelchair and the monkey head. This place cell-like activity was observed in both monkeys, with 44.6% and 33.3% of neurons encoding room position in monkeys K and M, respectively, and the overlapping populations of 41.0% and 16.0% neurons encoding head direction. These observations suggest that primary sensorimotor and premotor cortical areas in primates are likely involved in allocentrically representing body position in space during whole-body navigation, which is an unexpected finding given the classical hierarchical model of cortical processing that attributes functional specialization for spatial processing to the hippocampal formation.

Figure 1

Wheelchair navigation setup and examples of spatially-selective neurons. (A) View from the top of the experimental room, where the monkey navigated in a motorized wheelchair under BMI control from one of three possible starting locations (shown as green circles) to the grape dispenser (marked by a ‘+’). The four walls of the experiment room are labeled as “front”, “back” “right” and “left”, which correspond to the monkey facing the grape dispenser and navigating from the back to the front of the room. Red semicircle labels the docking zone upon which an auto-pilot took over the wheelchair control. Black semicircle labels the reach zone within which monkeys initiated reach movements for the grape. The position of the monkey’s body and orientation of the head (gray ellipse) and trunk (tan ellipse) were evaluated using the following coordinates: $\theta$- the angle of the wheelchair’s location with respect to the dispenser; r – the distance from the wheelchair to the dispenser; $\alpha$– the angle of the dispenser location with respect to the monkey’s head direction; and $\beta$– the angle of the dispenser location with respect to the wheelchair’s direction. (B–D) Spatial tuning diagrams in representative neurons from monkey K and monkey M (E–G), axis labels as in (A). Color represents trial-average z-scored firing rates for different room locations. Cortical areas, where the neurons were recorded, are indicated.

Figure 2

Neuronal tuning to room location and head orientation. (A–D) Modulation of spatial tuning patterns by head orientation in two neurons from monkey K. Conventions as in Fig. 1B–G. The color plots correspond (see key on top) to the right from the grape dispenser (α from 45 to 135 degrees; A, C) or to the left from the dispenser (α from −45 to −135 degrees; B, D). The spatial tuning pattern of one PMd neuron did not substantially change with the head orientation (A,B), whereas the pattern of another M1 neuron dramatically changed (C,D). (E–I) Scatterplots of tuning depth (TD) to room location versus TD to orientation (α and β) for monkey K’s (E–G) or monkey M’s (H-I) neurons with significant mutual information to either parameter from all sessions, across cortical areas: S1 (E, H), M1 (F, I) and PMd (G). The diagonal lines show where space-TD equals to orientation-TD. The dashed ellipses below and above the diagonal lines illustrate clusters of highly-tuned position-preferring and orientation-preferring neurons, respectively.

Figure 3

Place fields and preferred-directions for the neuronal ensembles. (A,B) Heatmaps of ensemble-average place fields for monkey K (A) and monkey M (B); plots represent averages across all sessions. Only position-preferring neurons were included. Each neuron’s place field was defined as the room locations where the neuron’s tuning function (as fitted by GAM) exceeded the median value over all room locations. A histogram of place-fields was constructed from all position-preferring cells’ place-fields and normalized by the number of position-preferring cells in each session. These session histograms were then averaged to produce the final heatmap. (C,D) Circular histograms of preferred $\alpha$ directions averaged over sessions for monkey K (C) and monkey M (D). For each session, the preferred $\alpha$ directions of all orientation-preferring neurons were extracted from their $\alpha$-tuning curves (as fitted by GAM). The preferred directions were then represented as circular histograms. Radial direction represents proportion of neurons. (E,F) Circular histograms for $\beta$. Conventions as in (C,D).

Figure 4

Individual neurons with consistent position tuning across multiple sessions. (A) The position-tuning diagram of monkey K’s PMd neuron shown in Fig. 1C for all nine experimental sessions. Session ordered from left to right, then top to bottom. The first session displayed corresponds to Fig. 1C. (B) Sample brain-controlled navigation trajectories for the same session as the data shown in Fig. 1C. Colors represent normalized firing rates along the trajectories. Black ‘*‘ mark the location at the end of each trial, and black ‘o’ mark the starting locations of each trial. (C) The position-tuning diagram of monkey K’s M1 neuron shown in Fig. 1B, for the first six experimental sessions. Session number increases from left to right. The first session is the same as shown in Fig. 1B. (D) Position tuning diagrams of monkey K’s PMd neuron for all nine experimental sessions. Session ordered from left to right, then top to bottom. (E) Position tuning diagrams of monkey M’s S1 neuron shown in Fig. 1F for nine consecutive experimental sessions, just before the session shown in Fig. 1F. Session ordered from left to right, then top to bottom.

Figure 5

Decoding of room location from neuronal ensemble activity. (A) Box plots showing the distribution of decoding accuracy and chance decoding accuracy (left and right boxes within each group, respectively) measured by the mean prediction error (MPE) in meters, in monkey K and monkey M using different population of neurons. Only sessions with decoding accuracy significantly above chance are shown. The distributions of chance-level decoding accuracy are constructed from the lower-end of the 95% confidence interval of the selected sessions’ chance-level error. (**) indicates significant p-value < 0.01 for post-hoc Tukey’s multiple comparison following Kruskal-Wallis test. (B–G) Color plots of MPE as a function of room location in one representative session each for monkey K (B–D) and monkey M (E–G) – from using all neurons (B,E), position-preferring neurons (C,F), and orientation-preferring neurons (D,G).