Copyright © 2008 Elsevier B.V. All rights reserved.
Neuronal encoding of meaning: Establishing category-selective response patterns in the avian ‘prefrontal cortex’
Received 30 July 2007;
Forebrain association areas interweave perceived stimuli with acquired representations of own actions and their outcome. Often, relevant stimuli come in a bewildering variety of shapes and sizes and we slowly have to learn to group them into meaningful categories. Therefore, the aim of the present study was twofold: First, to reveal how single units in the pigeon's nidopallium caudolaterale (NCL), a functional analogue of the mammalian prefrontal cortex (PFC), encode stimuli that differ in visual features but not in behavioral relevance. The second aim was to understand how these categorical representations are established during learning. Recordings were made from NCL neurons while pigeons performed a go–nogo categorization paradigm. Responses during presentation of the two S+ stimuli and non-responding during presentation of the two S− stimuli were followed by reward. We recorded from two pigeons at different learning stages. In the beginning of the learning process, neurons were active during and shortly before reward, but only in go trials. These data suggest that during the early phase of learning avian ‘prefrontal’ neurons code for rewards associated with the same behavioral demand, while ignoring feature differences of stimuli within one category. When learning progressed, (1) category selectivity became stronger, (2) responses selective for nogo stimuli appeared, and (3) reward-related responses disappeared in favor of category-selective responses during the stimulus phase. This backward shift in time resembles response patterns assumed by the temporal difference (TD) model of reinforcement learning, but goes beyond it, since it reflects the neuronal correlate of functional categories.
Keywords: Pigeon; Columba livia; Forebrain; Nidopallium caudolaterale (NCL); Multiple single-unit activity; Functional categorization
- 1. Introduction
- 2. Materials and methods
- 2.1. Behavioral task
- 2.2. Recording procedures
- 2.3. Data analysis
- 2.3.1. Behavior
- 2.3.2. Neuronal responses
- 2.4. Histology
- 3. Results
- 3.1. Behavior
- 3.2. Electrophysiology
- 4. Discussion
Fig. 1. Behavioral task. Inset on the left side: The pigeon sat in a foam couch looking on a monitor. The head was fixed and the reward could be delivered into a small plastic container below the beak of the pigeon. Beak movements (mandibulations) were detected by an infrared light barrier. The task consisted of two trial types (go and nogo), each containing two different stimulus types (go: heart or lightning; nogo: triangle or cross). Main figure: All trials began with a cue phase of 1.5 s where the cue light switched on. Afterwards one out of four possible visual stimuli were presented, instructing the animal about the type of trial (go vs. nogo). The stimulus phase took maximally 5 s. Mandibulation during the presentation of one of the go stimuli or rejection from mandibulation during the presentation of one of the nogo stimuli resulted in delivery of water for 1.5 s. Misses of go trials and mandibulation during nogo trials caused a mild punishment (all lights and stimuli off for 3 s).
Fig. 2. Frequency of response types of the BEGINNER (left) and the EXPERT (right) for go trials (hits: solid black bars; misses: black/white striped bars) and nogo trials (correct rejections: solid grey bars; false alarms: grey/white striped bars). The error bars represent standard error of the mean, significance level was 5%.
Fig. 3. Response classes of the neurons recorded from the BEGINNER (left) and the EXPERT (right). The fractions given in percentage were based on the total number of the level above, respectively. First, the neurons were divided in “responsive” and “unresponsive” neurons, depending on whether they responded to events within the task. Second, they were classified in “categorical” and “non-categorical”, depending on whether they responded to stimuli within on category only or not. Finally, they were analyzed regarding the trial phase in which the neuronal response occurred.
Fig. 4. Examples of neuronal responses collected from the BEGINNER (A) and the EXPERT (B). In each subplot the first four rows contain the raster diagram of the four different stimuli (from top to bottom: heart, lightning, triangle and cross; green: go, red: nogo). The grey dots within the raster diagram represent the mandibulations of the pigeon. The plot at the bottom of each subplot depict the histograms (binwidth 50 ms, filtered with a gaussian kernel) of the four raster diagrams above with the same color code. The dotted horizontal line represents the response threshold (mean ITI discharge rate +2 SD). In the subplots A1, B1, and B2 the raster diagrams and the histograms were aligned to reward onset, in the subplots A2 and B3 to stimulus onset and in subplot A3 to the first mandibulation after stimulus onset. The light blue area represents the time span of reward delivery. A1 (Go-REW) shows a neuron which responded to the reward delivery in go trials (arrowhead). A2 (Nogo > Go) shows a neuron which showed a general difference between go and nogo trials, without being triggered by a specific event (arrow). A3 (before first mandibulation and Go-REW) shows a neuron the activity of which was suppressed about 500 ms before the first mandibulation occurred (arrow) and then responded to the reward in go trials only (arrowhead). Although in this plot the data for the nogo trials were obtained from incorrect trials (otherwise the mandibulation had not occurred), and therefore the blue reward area is not valid for the nogo trials, this neuron did not respond to the reward in correct nogo trials (not shown). B1 (Go-PRE-REW) shows a neuron which increases discharge rate before reward onset (arrowhead) and peaked at reward delivery in go trials only. B2 (Go-STIM and All-REW) shows a neuron which responded to the go stimuli onsets with a first peak (arrow) at about 800 ms after stimulus onset (depicted by the grey vertical line at timepoint −2 s) and peaked a second time after reward onset in all conditions (arrowhead). The grey vertical line at −5.5 s represents the nogo stimuli onsets. B3 (Go-STIM) shows a neuron which responded to the stimulus onset on go trials only (arrow). (For interpretation of the references to color in this figure legend, the reader is referred to the web version of the article.)
Fig. 5. Schematic sagittal sections of the pigeon brain recording sites. The uppermost drawing represents the dorsal view on a pigeon brain with caudal left and rostral right. Lines depict the position of section planes shown in the six subplots at the bottom. The line drawing in the middle represents a section at L5.00 to illustrate the region enhanced in the subplots. Each symbol represents the recording site of one neuron. CDL: area corticoidea dorsalis; DA: tractus dorso-arcopallialis; N: nidopallium; NCL: nidopallium caudolaterale. Figures adapted from graphs in the pigeon brain atlas .
Fig. 6. Time course of category selectivity, using a sliding ROC analysis across all specifically responding categorical neurons in the BEGINNER (13 neurons, left column) and the EXPERT (20 neurons, right column) over two trial epochs: Around stimulus onset and around reward delivery. Each row represents the AUC time course of one neuron; the figures were constructed by sorting the neurons (y-axis) by their mean AUC value in the stimulus epoch. Category selectivity was stronger in the EXPERT than in the BEGINNER.
Fig. 7. Examples of the activity of three neurons which were classified as STIM-neurons correlated to motor behavior. Mandibulations take place at the origin of the abscissa (vertical grey line). An activity peak around 80 ms prior to beak movement would be characteristic of premotoric activity. Top: raster diagram, bottom: histogram (binwidth: 50 ms).
Upper row: mean percentages with standard error of correct responses in the four trials types. Lower row: mean response times with standard error of correct responses (hits) in go trials and of incorrect responses (false alarms) in nogo trials.