Order-dependent modulation of directional signals in the supplementary and presupplementary motor areas

Abstract

To maximize reward and minimize effort, animals must often execute multiple movements in a timely and orderly manner. Such movement sequences must be usually discovered through experience, and during this process, signals related to the animal's action, its ordinal position in the sequence, and subsequent reward need to be properly integrated. To investigate the role of the primate medial frontal cortex in planning and controlling multiple movements, monkeys were trained to produce a series of hand movements instructed by visual stimuli. We manipulated the number of movements in a sequence across trials, making it possible to dissociate the effects of the ordinal position of a given movement and the number of remaining movements necessary to obtain reward. Neurons in the supplementary and presupplementary motor areas modulated their activity according to the number of remaining movements, more often than in relation to the ordinal position, suggesting that they might encode signals related to the timing of reward or its temporally discounted value. In both cortical areas, signals related to the number of remaining movements and those related to movement direction were often combined multiplicatively, suggesting that the gain of the signals related to movements might be modulated by motivational factors. Finally, compared with the supplementary motor area, neurons in the presupplementary motor area were more likely to increase their activity when the number of remaining movements is large. These results suggest that these two areas might play complementary roles in controlling movement sequences.

Figure 1.

Serial reaction time task. A, Spatiotemporal sequence of the task. Each frame shows a visual target (gray disk) and the feedback cursor (black dot). B, Four different target paths and four rewarded locations (gray squares). The asterisk indicates the path shown in D. C, An example sequence consisting of six movements that ends in the rewarded location in the top left corner of the grid. The thick line corresponds to the first movement in the sequence. D, Another example sequence in which NRM for the first movement (thick arrow) remains ambiguous (1 or 7) before the onset of the third target.

Figure 2.

Effects of the NRM and the OP on reaction times. The average reaction times were computed separately for each combination of NRM and OP, and are connected by a line for each sequence length (SL = 2, 4, 6, or 8). The error bars (SEM) (data not shown) are all smaller than the symbols. The horizontal dotted line and the empty circles correspond to the average reaction time for movements with ambiguous NRM (Fig. 1D).

Figure 3.

An example neuron in SMA showing activity negatively correlated with NRM. Activity of this neuron was best accounted for by the additive NO factor model. Top, Raster plots showing the activity separately for different movement directions and NRM (0, 2, 4, and 6). Middle, Spike density functions (convolved with a Gaussian kernel, σ = 40 ms) estimated separately for different NRM and different movement directions. Bottom, Average firing rates (red) during a 500 ms window centered at target onset plotted as function of NRM separately for different movement directions and residual spike counts (blue) obtained from the kinematic model during the same period. Error bars indicate SEM.

Figure 4.

An example neuron in pre-SMA showing activity positively correlated with NRM. Activity of this neuron was best accounted for by the multiplicative N linear model. The format is the same as in Figure 3.

Figure 5.

Distribution of regression coefficients associated with NRM and OP. The different symbols indicate whether the effect of NRM or OP was statistically significant or not (t test, p < 0.05). Numbers indicate the figures that illustrate the corresponding neurons.

Figure 6.

An example neuron in pre-SMA showing activity negatively correlated with NRM. Activity of this neuron was best accounted for by the additive NO factor model. The format is the same as in Figure 3.

Figure 7.

An example neuron in pre-SMA showing activity negatively correlated with NRM. Activity of this neuron was best accounted for by the multiplicative N factor model. The format is the same as in Figure 3.

Figure 8.

An example neuron encoding both NRM and OP (A) and another neuron encoding only NRM (B). The average firing rates for each combination of NRM and OP during a 500 ms window centered at the time of target onset are plotted as a function of NRM (top) or OP (bottom). The values obtained from the trials of the same sequence length (SL) are connected by a line. These two neurons are the same neurons illustrated in Figures 3 (A) and 7 (B), respectively. Error bars indicate SEM.

Figure 9.

Fraction of variance in the spike counts during the 500 ms window beginning 250 ms before target onset that was accounted for by the kinematic model (abscissa) and by the best model that incorporated the effect of NRM and OP (ordinate). The different symbols indicate whether the best model included NRM or OP. Numbers indicate the figures that illustrate the corresponding neurons.

Figure 10.

Effect of uncertainty about NRM on neural activity. This analysis was performed separately according to whether the neurons showed significant negative (top) or positive (bottom) correlation with NRM. The number of neurons in each group is indicated in the plot. Within each group, normalized activity was averaged separately for unambiguous movements as a function of NRM (filled circles) and for movements with ambiguous NRM (empty circles). Error bars indicate SEM. *p < 0.05 (t test).