Coding specificity in cortical microcircuits: a multiple-electrode analysis of primate prefrontal cortex

Abstract

Neurons with directional specificities are active in the prefrontal cortex (PFC) during tasks that require spatial working memory. Although the coordination of neuronal activity in PFC is thought to be maintained by a network of recurrent connections, direct physiological evidence regarding such networks is sparse. To gain insight into the functional organization of the working memory system in vivo, we recorded simultaneously from multiple neurons spaced 0.2-1 mm apart in monkeys performing an oculomotor delayed response task. We used cross-correlation analysis and characterized the effective connectivity between neurons in relation to their spatial and temporal response properties. The majority of narrow (<5 msec) cross-correlation peaks indicated common input and were most often observed between pairs of neurons within 0.3 mm of each other. Neurons recorded at these distances represented the full range of spatial locations, suggesting that the entire visual hemifield is represented in modules of corresponding dimensions. Nearby neurons could be activated in any epoch of the behavioral task (stimulus presentation, delay, response). The incidence and strength of cross-correlation, however, was highest among cells sharing similar spatial tuning and similar temporal profiles of activation across task epochs. The dependence of correlated discharge on the functional properties of neurons was observed both when we analyzed firing from the task period as well as from baseline fixation. Our results suggest that the coding specificity of individual neurons extends to the local circuits of which they are part.

Fig. 1.

Location of electrophysiological recording in dorsolateral prefrontal cortex, centered on area 46 and the frontal eye fields (area 8). This region included the caudal half of the principal sulcus and cortex lining the arcuate sulcus.

Fig. 2.

Oculomotor delayed response task and recording methodology. A, Trials began when the monkey fixated a central point on a screen for 500 msec. A target appeared in one of eight possible locations (0–315°) for 500 msec and was followed by a delay period of 3000 msec. When the fixation point was extinguished, the monkey saccaded to the location of the remembered target.B, Independently advancing electrodes were lowered in the monkey's cortex. C, Cross-view illustrating the electrode configurations most often used in this study. The top circle represents a single guide tube that contains four electrodes in a 2 × 2 matrix. The two bottom circles represent two guide tubes separated by 1 mm, each holding a single electrode.

Fig. 3.

Unit activity for two neurons recorded from electrodes ∼200 μm apart. The two outer panelsdisplay peristimulus time histograms representing neuronal activity on the ODR task. Histograms are arranged to indicate the location of the corresponding cue. The center panel presents the raw cross-correlation histogram with the shift predictor overlaid as a gray line. Horizontal lines represent CCH baseline and 0.001 confidence intervals. The neuron shown on theright was the reference cell for the construction of the CCH. PST histograms reveal spatially overlapping delay period activity in the lower contralateral visual field (315° position). The narrow peak centered on the zero time point of the cross-correlation histogram reflects synchronous neuronal firing. Spikes that contributed to CCH for Neuron 3083 and Neuron 3086 were 14,025 and 18,192, respectively. Percentages of total spikes represented in the peak (cross-correlation strength) were 2.3 and 1.8%, respectively. Bin size for CCH was 1 msec.

Fig. 4.

Unit activity is presented for neurons recorded from two electrodes ∼200 μm apart. Conventions are the same as for Figure 3, except for CCH bin size (2 msec). The neuronal pair shared spatially overlapping receptive fields, with maximal activation during the presentation of the cue. Spikes that contributed to CCH forNeuron 1437 and Neuron 1439 were 10,089 and 14,779, respectively.

Fig. 5.

Percentage of pairs exhibiting significant cross-correlation interactions as a function of electrode separation. The proportion of CCH peaks, both broad and narrow, decreased as the distance between electrodes increased to 1 mm.

Fig. 6.

Distribution of spatial tuning differences and rate correlation coefficients for neurons recorded simultaneously from electrodes 200–300 μm apart. The expected distributions, simulated by randomly pairing neurons recorded at different sessions, are shown on the right column. A, The number of pairs spatially tuned during the same task epochs is plotted as a function of their tuning difference (n = 136).B, The number of pairs is plotted as a function of the correlation coefficient computed for their mean firing rates in each task epoch (n = 286).

Fig. 7.

Incidence of narrow cross-correlation peaks as a function of spatial tuning. Left panel illustrates results from cross-correlation analysis based on the cue, delay, and saccade behavioral epochs. Most functional interactions were observed for neurons with the highest spatial tuning similarity. Right panel shows results from baseline fixation period. Pairs with similar tuning tended to exhibit narrow peaks during the fixation period, before the neurons were engaged by the task and the spatial tuning of the neuron could be determined.

Fig. 8.

Examples of neuronal pairs recorded 200–300 μm apart that displayed narrow CCH peaks. Polar plotsrepresent the spatial tuning of the neuron. The arrowrepresents the preferred location of each neuron as determined by a vector algorithm. Cross-correlation histogram is shown in thecenter of each panel. A, Spatial tuning of the pair was obtained in the cue period. B, Spatial tuning was obtained during the delay period. C, Spatial tuning was obtained in the saccade period.

Fig. 9.

Cross-correlation strength as a function of spatial tuning and epoch of activation. A, Eachpoint represents the mean CCH strength values for pairs with tuning difference that falls in a 20° wide bin centered around the point. Error bars represent SEM. Pairs with similar tuning exhibited stronger correlation peaks during ODR task epochs (left) and fixation period (right) when their spatial tuning was more similar. B, CCH strength is plotted as a function of the correlation coefficient computed by the mean firing rates of the two neurons in each task epoch.

Fig. 10.

Incidence of narrow CCH peaks as a function of epoch of activation. The percentage of pairs exhibiting narrow CCH peaks is shown for pairs of neurons: both of which are active in a task epoch (gray bars), only one neuron of which is active (white bars), or neither is active (black bars).