Differential Activation of Fast-Spiking and Regular-Firing Neuron Populations During Movement and Reward in the Dorsal Medial Frontal Cortex

Abstract

The medial prefrontal cortex is thought to be important for guiding behavior according to an animal's expectations. Efforts to decode the region have focused not only on the question of what information it computes, but also how distinct circuit components become engaged during behavior. We find that the activity of regular-firing, putative projection neurons contains rich information about behavioral context and firing fields cluster around reward sites, while activity among putative inhibitory and fast-spiking neurons is most associated with movement and accompanying sensory stimulation. These dissociations were observed even between adjacent neurons with apparently reciprocal, inhibitory-excitatory connections. A smaller population of projection neurons with burst-firing patterns did not show clustered firing fields around rewards; these neurons, although heterogeneous, were generally less selective for behavioral context than regular-firing cells. The data suggest a network that tracks an animal's behavioral situation while, at the same time, regulating excitation levels to emphasize high valued positions. In this scenario, the function of fast-spiking inhibitory neurons is to constrain network output relative to incoming sensory flow. This scheme could serve as a bridge between abstract sensorimotor information and single-dimensional codes for value, providing a neural framework to generate expectations from behavioral state.

Keywords: anterior cingulate cortex; coding; inhibition; motivation; rat.

Figure 1.

Behavior methods and neuron classification. (A) The 3-choice, 2-cue behavioral apparatus used in the present experiments. At the ends of the 3 platform arms are speakers and LED lights, one of which signals the reward location on a given session. Rats shuttled between the central cue zone, where cue presentation was initiated, and feeder zones, where liquid food was delivered in the case of a correct choice, and a nonlocalized error sound was presented in the case of an incorrect choice. (B) Distribution of half-amplitude (depolarization) width and half-valley (after-hyperpolarization) width of each neuron's average waveform. Neurons that had inhibitory cross-correlation profiles with other neurons (black dots) tended to cluster into a group of cells with shorter after-hyperpolarization times. All neurons in this grouping were classified as putative inhibitory cells (example in Supplementary Fig. 1, neuron A). Those with wider waveforms were classified as putative excitatory projection cells (Supplementary Fig. 1, neurons B, C, and D). Insets are example of average waveforms for putative projection and inhibitory neurons. (C) Distribution across the neuron population of 2 features calculated from each neuron's 500-ms autocorrelation: Time that the autocorrelation reached a peak, and the magnitude of the downward slope. Autocorrelations include spikes from both rest and task session epochs. Black dots represent those neurons that were determined to be inhibitory from the cross-correlations (as in B). Visual inspection revealed roughly 3 groupings, which were classified as burst firing (early peaks but rapid decay; green), regular firing (later peaks and often slower decay; orange), and fast spiking (moderately early peaks but slow decay; blue). Inset shows average autocorrelations of all cells within each group. (D) The reconstructed electrode positions of all recorded neurons overlaying coronal (left), and sagittal (right) illustrations from Paxinos and Watson (1998; plates are 3.4 mm anterior to bregma and 0.9 mm lateral, respectively). Neuron recording sites that appear to be beyond the boundaries of the medial prefrontal cortex can be explained by the limitations of projecting 3D positions onto single sections. (E) Frequency of recorded neurons along medial–lateral and dorsal–ventral axes. Each panel contains a frontal coronal plate from Paxinos and Watson, overlaid with a color map describing the number of neurons recorded relative to the number of sessions an electrode occupied the position. Hotter colors indicate a higher proportion of neurons recorded at a given location. The increased frequency of burst-firing neurons (left) at a particular medial–lateral position (bottom panel) is consistent with observations that burst-firing neurons are more frequently observed in layer IV/upper-layer V (see Discussion). Regular-firing (middle) and fast-spiking (right) neurons were observed throughout medial–lateral and dorsal–ventral positions. The 267 neurons that did not fall within the boundary classifications were distributed throughout these regions. (F) Autocorrelation features as in C, but generated from spikes during the task epoch only. Many fast-spiking neurons formed a tight cluster with peaks around 20 ms. Average autocorrelations of neurons in this group exhibited a 40- to 70-Hz (gamma-frequency) rhythm (inset).

Figure 2.

Different neuron classes exhibit different patterns of trial phase selectivity. (A) Activity across trial phases in an example fast-spiking inhibitory neuron (blue), regular-firing projection neuron (orange), and burst-firing projection neuron (green). All 3 neurons were simultaneously recorded on the same electrode. Firing rates are averaged across trials and displayed for each trial phase from the rat's departure from a feeder zone (phase 1) until obtaining subsequent reward again at a subsequent feeder zone (phase 8). Insets show average waveforms for each neuron (left) and 500-ms autocorrelograms (right). (B) Distribution of trial phase selectivity for inhibitory neurons and both regular-firing and burst-firing groups of excitatory neurons. Selectivity was measured in bits and log-normalized for illustration and statistical purposes. Both regular- and burst-firing projection neuron groups were much more selective for trial phase than interneurons. (C) Across-rat averages of relative trial phase activity for fast-spiking inhibitory and regular- and burst-firing projection neurons. Firing rates across trial phases were z-score normalized then averaged within rats for each cell type to show relative activity levels between trial phases (error bars are SEM across 5 rats). Fast-spiking neurons (blue) tended to fire when cues were presented, while regular-firing neurons (orange) tended to be least active during this time, and most active when rats acquired reward. (D) Similar patterns as in C can be observed by tabulating the number of neurons that fired maximally in each individual trial phase. (E) Mean, normalized cross-correlation of all 68 pairs of simultaneously recorded inhibitory and regular-firing projection with significant, reciprocal, inhibitory–excitatory interactions (only neurons with higher than 1-Hz firing rates included). (F) Trial phase activity among only those neurons within the 68 pairs (regular firing: orange, inhibitory: black). (G) Tabulated number of neurons with maximal firing in each individual trial phase. (H) A matrix of correlation coefficients, averaged across rats, between the firing rate vectors for each trial phase and for the rest epoch on the x and y axes. Firing rate vectors included all neurons recorded in a rat, where each element of the vector was the across-trial average firing rate of a neuron within the specific trial phase indicated. Hotter colors represent higher correlations. White lines around the fourth and eighth rows mark the correlation coefficients for cue and reward phases that are also presented in different graphic form in I. (I) Line graphs illustrate the same correlation coefficient data as presented in H for the cue (upper graph) and reward (lower graph) periods, with error bars representing SEM across 5 rats. The population activity was most distinct between cue and reward periods, and this was consistent across rats.

Figure 3.

Relationship between fast-spiking inhibitory neuron firing with task stimuli and movement. (A) Raster plot (top) and PETH (bottom) of an example fast-spiking, putative inhibitory neuron. Across all trials, firing rates increased following presentation of the auditory, decision-cue (central axis), and activity was sustained for longer than 1 s. (B) Average, normalized PETH aligned to the auditory decision cue across all neurons that were both fast-spiking and putative inhibitory cells (blue trace), plotted with the cue-locked averages of the rats' acceleration when the neurons were recorded (green). Many neurons exhibited either a transient or sustained activity increase following cue presentation; often these firing rate patterns matched the times at which rats accelerated as they searched for and oriented toward the cues. (C) Average PETHs and acceleration patterns as in B, aligned to when rats initiated movement toward the selected feeder. (D and E) Same as in B, with PETHs and movement acceleration aligned to reward delivery and error sound presentation, respectively. (F) Performance during the first (black) and second (gray) half of each session plotted across days from the reward contingency switch. Only sessions in which fast-spiking or putative inhibitory neurons were recorded were included in figure (error bars are SEM across these sessions). Average performance ranged from over 80% correct before the switch to nearly 40% correct during the half-session following the switch. (G) Average firing rates for fast-spiking or putative inhibitory neurons during the 500-ms follow cue presentation for the first (black) and second (gray) half of each session plotted across days as in F. Although different neurons may have been recorded between days, firing rates did not appear to change between the first and second half-sessions. Probability of reward therefore did not appear to influence firing during the cue period; responses were therefore more likely to reflect attention and movement initiation processes.

Figure 4.

Regular-firing neuron activity on correct and incorrect trials. (A) Firing patterns across trial phases of a representative sample of 4 neurons in rewarded (red) and unrewarded (black) trials. Many individual cells discriminated error and reward trials prior to delivery of the outcome. The four examples include neurons that apparently predicted reward or error outcome (first and second plot), a neuron with a bidirectional field that fired on the way to and from the feeder zones (third plot), and a neuron that was sharply tuned to a particular trial phase, and may have generally been more activity during a phase of the task associated with more errors (as firing rates differed even before cues were delivered). (B) Proportion of regular-firing projection neurons that significantly differentiated between correct and incorrect trials at each 500-ms trial phase (Wilcoxon rank-sum test, α = 0.01). Black trace represents average proportions across 5 rats, error bars are SEM across rats. Prior to presentation of decision cues, the number of neurons differentiating between a correct and incorrect trial approximates chance, given the chosen acceptance levels (α = 0.01). Many regular-firing projection neurons fired differently on correct versus incorrect trials after presentation of the error sound or click of the feeder solenoid. Many (∼15%) also fired differently on correct versus incorrect trials immediately before outcomes were presented (trial phase 6), suggesting that rats had knowledge about whether they may be committing an error. (C) Net levels of activity across the population of regular-firing projection neurons for rewarded (orange) and error (brown) trials. Firing rates for individual neurons were first z-score normalized across all trials and trial phases, then averaged across rewarded or error trials, and then averaged within rats. Error bars are SEM. Activity levels only discriminated between reward and error trials in the final trial phase, 500–1000 ms after outcomes were presented, suggesting that increases in net activity were not due to reward or its expectation alone, and instead could be said to relate either to outcome more generally, or to other information related to the circumstances in which rewards were encountered in the past. (D) Average PETHs of regular-firing projection neuron firing rates locked to outcome presentation, binned at 10 ms. There is no evidence of discrimination between trial types for the first 150 ms. (E) Net levels of activity across the population of fast-spiking inhibitory neurons for rewarded (light blue) and error (dark blue) trials. In contrast with regular-firing projection neurons, activity on the two trial types was significantly different during the first 500 ms following outcome presentation. (F) Average PETHs of fast-spiking inhibitory neuron firing rates locked to outcome presentations, binned at 10 ms. Discrimination between error and reward trials appeared to be slightly earlier than in the population of regular-firing projection neurons.

Figure 5.

Neuron activity in task versus rest epochs. (A) Distribution of firing rates for fast-spiking (blue), regular-firing (orange), and burst-firing (green) neurons during performance of the task (x-axis) and during rest (y-axis) epochs. Average firing rates were high for fast-spiking neurons, and these cells tended to fire at similar rates between epochs (though slightly reduced during rest). Firing rates were lower for regular-firing neurons, and were less correlated between task and rest epochs. Most regular-firing and fast-spiking neuron rates fell below the graph diagonal, suggesting that firing rates tended to be higher during the task epoch. (B) Firing rates normalized across epochs, averaged within groups. Both fast-spiking and regular-firing neurons fired at higher rates during the task epoch (error bars SEM across 5 rats). (C) Correlation coefficients between epochs for population firing rate vectors across neuron classes. In fast-spiking and burst-firing neurons, population firing rate vectors were highly correlated between task and rest epochs; in regular-firing neurons, populations were less correlated, meaning that a relatively distinct set of regular-firings neurons was active during the 2 epochs, and therefore that the population represented the 2 behavioral conditions differently.

Figure 6.

Shift in network activity from cue zones to reward sites at more ventral recording locations. (A) Electrodes localized to the prelimbic cortex, ventral to the dorsal anterior cingulate cortex, exhibited a higher proportion of regular-firing neurons that were most active during the reward phase of the decision trial. (B) While, on average, regular-firing neurons in the medial prefrontal cortex fired more during the task epoch, neurons in the dorsal prelimbic cortex fired more during the rest epoch (y-axis is a score relating the firing rate in the task epoch compared with firing rates in the rest epoch with range [−1, +1]). (C) A subset of fast-spiking neurons tended to fire at the frequency of gamma during the task epoch; as with the larger set of fast-spiking neurons, the average activity in these neurons was higher when cues were presented, and lower during the reward. This tendency gradually shifted as recording electrodes were lowered more ventrally into the medial prefrontal cortex, as illustrated by the relationship between recording depth (y-axis) and the ratio of activity at feeder versus cue zones (x-axis). (D) There was a significant relationship between recording region and relative activity of the gamma-frequency, fast-spiking neurons during cue versus outcome periods. (E) The relative strength of LFP gamma during the cue period compared with during the outcome period was reduced in the most ventral and most dorsal recording electrodes. One explanation for why gamma power patterns at more dorsal sites did not match firing patterns of fast-spiking neurons may be that the LFP signals were contaminated by proximity to the reference electrode. (F) Consistent with patterns observed in fast-spiking neurons, relative gamma amplitude during the cue versus outcome trial phases changed across the regions that the recordings were made. (G) Gamma-frequency neuron firing patterns over trial phases resembled that of the larger set of fast-spiking neurons (error bars SEM across rats; rw, reward; er, error). (H) Relative amplitude of LFP gamma from all recording electrodes (averaged within rats, then across rats; trial phases 8 and 9 omitted due to licking artifact at the feeder). Gamma amplitude was also higher when rats were at the cue zone and lower when rats entered the feeder zone, although unlike the population of inhibitory neurons, amplitude was relatively lower when rats initiated a return back from feeder zones.