Effects of gaze shifts on maintenance of spatial memory in macaque frontal eye field

Abstract

The activity of 91 neurons in the frontal eye fields (FEFs) of two macaque monkeys was recorded while the animals performed a delayed spatial match-to-sample task. During the delay, the animals were required to shift their gaze to one of four eccentric locations. Neuronal activity during the delay was analyzed for sensitivity to cue location and eye position. One-third of the neurons showed significant delay activity selective for cue location, whereas slightly more than one-half of the neurons showed significant modulation of delay activity when the gaze was shifted to an eccentric location. Despite this modulation, the neurons continued to signal their preferred cue location during most of the delay. However, after recentering saccades, the memory signal was temporarily abolished and then reemerged over a period of few hundred milliseconds. This is consistent with the idea that spatial working memory is buffered outside of the FEF. For most neurons, delay activity tended to increase when the gaze was shifted away from the preferred location and to decrease when the gaze was shifted toward the preferred location. This pattern of modulation is consistent with a vector subtraction mechanism that allows for the superposition of multiple saccade plans.

Figure 1.

Delayed spatial match-to-sample task. a, Timing of trial events: fixation (FIX) (100 msec), cue presentation (CUE) (100 msec), delay [fixation (FIX) (250 msec), eccentric/central fixation (ECC FIX) (1500 msec), and refixation (RE-FIX) (500 msec)], and choice saccade (SACC) (500 msec). b, Schematic of visual display during task. Black squares indicate fixation target (thin lines are for illustration only; they were not visible on display) or choice targets. White square indicates cue. Arrow indicates choice saccade. Target and Distractor are the two identical choice stimuli that appeared simultaneously near the end of the trial. On/off indicate the visibility of the stimulus. E_H and E_V are horizontal and vertical eye position, respectively. The horizontal black bar groups the panels that belong to the delay period.

Figure 2.

Example of middle-delay and late-delay interval activity and tuning dynamics. a–c, Spike rasters and histograms (smoothed with a Gaussian; width, 12 msec). E_H and E_V are horizontal and vertical eye position, respectively, for a single representative trial. d–f, Responses as a function of cue direction and tuning vectors for middle-delay activity. g–i, Spatial tuning for late-delay activity. sp/s, Spikes per second.

Figure 3.

Population vector dynamics for early-delay and middle-delay activity. Each row represents a different eye position during the delay. Small arrows are tuning vectors (magnitude, 20×) for individual neurons. Large arrows are population vectors. Horizontal dashed lines indicate preferred cue direction for center eye position during delay. Vertical dashed lines indicate the early and middle delay. Asterisks indicate significance level of Rayleigh uniformity test for neuronal tuning vector directions (1) and Hotelling–Lawley test for population vector amplitude (2). sp/s, Spikes per second; Sacc, saccade.

Figure 4.

Population vector for middle-delay and late-delay activity. Each row represents a different eye position during the delay. Small arrows are tuning vectors (magnitude, 20×) for individual neurons. Large arrows are population vectors. The leftmost arrow in each row indicates middle-delay population vectors. Horizontal dashed lines indicate preferred cue direction for center eye position during delay. Vertical dashed lines indicate the middle and late delay. Asterisks indicate significance level of Rayleigh test (1) and Hotelling–Lawley test (2). sp/s, Spikes per second; Sacc, saccade.

Figure 5.

Population vector amplitude for entire delay. Each row represents a different eye position during the delay. Horizontal dashed lines indicate average delay activity for trials in which fixation remained at center. Vertical dashed lines separate the early, middle, and late delay periods. The filled circles, triangles, and squares represent the population vector amplitude at different moments in time during the early, middle, and late delay, respectively. sp/s, Spikes per second; Sacc, saccade.

Figure 6.

Two examples of neurons with cue and eye position modulation during the delay. a, Cell with preferred eye position opposite preferred cue location. Thick black lines represent delay activity as a function of cue location (error bars are ±1 SEM). Each tuning curve is offset to reflect the eye position during the delay. Filled arrows are the center-of-mass vector for each tuning curve. Open arrows indicate an example of trials with matching saccade. Dotted lines indicate the tuning curves for activity during the 100 msec cue interval. b, Cell with enhanced response for all of the eccentric eye positions. The conventions are the same as those in a. sp/s, Spikes per second.

Figure 7.

Comparison of best cue direction versus best eye direction. a, Arrows are preferred eye vectors for individual neurons. The PEVs have been rotated by subtracting the direction of the preferred cue vector of the cell. The radial axis is in spikes per second, and the angular axis is in degrees. b, Distribution of direction differences between preferred eye and preferred cue vectors. The radial axis is number of cells.

Figure 8.

Delay activity for central versus eccentric fixation using saccade-matched trials. Downward arrows indicate the means of the respective distributions. The open bars represent the worst eye position cases, and the filled bars represent the best eye position cases.

Figure 9.

Superposition of saccade plans model. Filled symbols and solid lines in the middle plot represent the spatially tuned response for central fixation. Gray lines in the outer ring of plots represent the gainfield-like component. Open squares with thin lines represent the average of the tuned and gainfield response components.

Figure 10.

Normalized delay activity and model predictions. Open circles represent the average delay activity sorted by cue direction and eye position compared with the acitvity predicted by the weighted saccade plan model. The set of activities for each cell was normalized to the maximum for that cell. Solid line is the least-mean squares regression (slope = 1.1; intercept = –0.03). r is the sample correlation coefficient.

Figure 11.

Plot of the weights (W) given to each saccade plan by each neuron. Weights were found by fitting the responses of each neuron with the superposition model. Solid line is least-mean squares regression. Numbers in parentheses are the slope and intercept, respectively. Each open triangle represents the pair of weights for a single neuron. r is the correlation coefficient of the sample.