Data were analyzed from three male rhesus monkeys (Macaca mulatta), ages 5–9 years old, weighing 5–12 kg. None of these animals had any prior experimentation experience at the onset of our study. Monkeys were either single-housed or pair-housed in communal rooms with sensory interactions with other monkeys. All experimental procedures followed guidelines set by the U.S. Public Health Service Policy on Humane Care and Use of Laboratory Animals and the National Research Council’s Guide for the Care and Use of Laboratory Animals and were reviewed and approved by the Wake Forest University Institutional Animal Care and Use Committee under protocol number A14-196.
Monkeys sat with their heads fixed in a primate chair while viewing a monitor positioned 68 cm away from their eyes with dim ambient illumination and were required to fixate on a 0.2° white square appearing in the center of the screen. During each trial, the animals were required to maintain fixation on a 0.2° white square appearing in the center of the screen while visual stimuli were presented either at a peripheral location or over the fovea, in order to receive a liquid reward (typically fruit juice). Any fixation break immediately terminated the trial and no reward was given. Eye position was monitored throughout the trial using a non-invasive, infrared eye position scanning system (model RK-716; ISCAN, Burlington, MA). The system achieved a < 0.3° resolution around the center of vision. Eye position was sampled at 240 Hz, digitized and recorded. The visual stimulus display, monitoring of eye position, and synchronization of stimuli with neurophysiological data was performed with in-house software implemented on the MATLAB environment (Mathworks, Natick, MA), utilizing the Psychophysics Toolbox 33.
Behavioral Task
Pretraining Task. Following a brief period of fixation training and acclimation to the stimuli, monkeys were required to fixate on a center position while stimuli were displayed on the screen. The stimuli shown in the pre-training, passive, spatial task consisted of white 2° squares, presented in one of nine possible locations arranged in a 3 × 3 grid with 10° of distance between adjacent stimuli. The stimuli shown in the pre-training passive shape task consisted of white 2° geometric shapes drawn from a set comprising a circle, diamond, the letter H, the hashtag symbol, the plus sign, a square, a triangle, and an inverted Y-letter. The stimuli analyzed here were always presented at the center location of the 3 × 3 grid.
Presentation began with a fixation interval of 1 s where only the fixation point was displayed, followed by 500 ms of stimulus presentation (referred to hereafter as cue), followed by a 1.5 s “delay” interval where, again, only the fixation point was displayed. A second stimulus (referred to hereafter as sample) was subsequently shown for 500 ms. In the spatial task, this second stimulus would be either identical in location to the initial stimulus, or diametrically opposite the first stimulus. In the shape task, this second stimulus would appear in the same location to the initial stimulus and would either be an identical shape or the corresponding non-match shape (each shape was paired with one non-match shape). Only one nonmatch stimulus was paired with each cue, so that the number of match and nonmatch trials were balanced in each set. In both the spatial and shape task, this second stimulus display was followed by another “delay” period of 1.5 s where only the fixation point was displayed. The location and identity of stimuli was of no behavioral relevance to the monkeys during the “pre-training” phase, as fixation was the only necessary action for obtaining reward.
Post-training task. The monkeys were then trained to perform working memory tasks that involved the presentation of identical stimuli as the spatial and shape tasks during the “pre-training” phase. Now monkeys were required to remember the spatial location and/or shape of the first presented stimulus, and report whether the second stimulus was identical to the first or not, via saccading to one of two target stimuli (green for matching stimuli, blue for non-matching). Each target stimulus could appear at one of two locations orthogonal to the cue/sample stimuli, pseudo-randomized in each trial.
Surgery and neurophysiology
A 20 mm diameter craniotomy was performed over the PFC and a recording cylinder was implanted over the site. The location of the cylinder was visualized through anatomical magnetic resonance imaging (MRI) and stereotaxic coordinates post-surgery. Electrode penetrations were mapped onto the cortical surface. We identified 6 lateral PFC regions: a posterior-dorsal region that included area 8A, a mid-dorsal region that included area 8B and area 9/ 46, an anterior-dorsal region that included area 9 and area 46, a posterior-ventral region that included area 45, an anterior-ventral region that included area 47/12, and a frontopolar region that included area 10 34. Only posterior dorsal, mid-dorsal and posterior-ventral areas were sufficiently sampled and were included in these analyses.
Neural recordings
Recordings were performed from the prefrontal cortex both before and after training in the working memory tasks. Subsets of the data presented here were previously used to determine the individual properties of neurons in the posterior-dorsal, mid-dorsal, anterior-dorsal, posterior-ventral, and anterior-ventral PFC subdivisions 17. Extracellular recordings were performed with multiple microelectrodes that were either glass- or epoxylite-coated tungsten, with a 250 μm diameter and 1–4 MΩ impedance at 1 kHz (Alpha-Omega Engineering, Nazareth, Israel). A Microdrive system (EPS drive, Alpha- Omega Engineering) advanced arrays of up to 8-microelectrodes, spaced 0.2–1.5 mm apart, through the dura and into the PFC. The signal from each electrode was amplified and band-pass filtered between 500 Hz and 8 kHz while being recorded with a modular data acquisition system (APM system, FHC, Bowdoin, ME). Waveforms that exceeded a user-defined threshold were sampled at 25 μs resolution, digitized, and stored for off-line analysis. Neurons were sampled in an unbiased fashion, collecting data from all units isolated from our electrodes, with no regard to the response properties of the isolated neurons. A semi-automated cluster analysis relied on the KlustaKwik algorithm, which applied principal component analysis of the waveforms to sort recorded spike waveforms into separate units.
LFP recordings
LFP data were acquired in the aforementioned areas of the PFC, both before and after training in each working memory task, through extracellular recordings with the same microelectrodes used to record single-neuron activity. The signal from each electrode was amplified and band-pass filtered between 0.5 - 200 Hz for LFP processing, with the same modular data acquisition system (APM system, FHC, Bowdoin, ME).
Data analysis
Neural Data. Data analysis was implemented with the MATLAB computational environment (R2014-2021, Mathworks, Natick, MA). A trial-averaged peristimulus time histograms (PSTHs) for illustrations were calculated by convolving the spiking events a 50 ms steps apart. Neurons were identified to be selective in any the task epoch by virtue of significantly different responses to the spatial location (or shape) of the stimulus, using a one-way ANOVA test performed on the firing rates of each neuron obtained across trials during the task epoch in question. To avoid false positives among neurons with a handful of spikes, we also required that a selective neuron exhibit a firing rate of at least 2 spikes/s for its best stimulus location or shape, in the task period that the ANOVA test indicated a significant main effect.
LFP Analysis. LFP recording were preprocessed by using custom MATLAB code in MATLAB R2019a (MathWorks) and the FieldTrip toolbox 35. A bandpass filter between 0.5-200 Hz with a zero-phase sixth-order Butterworth filter were used on single-trial LFP traces; we also used a notch filter at 60 Hz with a bandwidth of 0.1 Hz to remove the power line artifacts. Further, single- trial LFP traces underwent artifact rejection. The Chronux package 36 was used for time-frequency analysis. We used a multi-taper method to perform a power spectrum analysis of LFPs. The spectrogram of each single trial between 0.5 and 128 Hz was computed with 8 tapers in 500 ms time windows; the spectrograms were estimated with a temporal resolution of 2 ms. We also used the mean filter corresponding to 1.95 Hz and 20 ms for smoothing the spectrogram of each single trial. We relied on induced power of the LFP in all of our analysis, which is computed by first performing a power computation in each trial and then power across trials is averaged. Induced power thus determines power at specific frequencies that may not necessarily be synchronized with specific task events across trials. Power was expressed relative to the mean power recorded during the inter-trial interval. Time-resolved plots (spectrograms) were constructed and plotted after dividing the power of the signal by the mean inter-trial interval power at each frequency (which is equivalent to subtracting the baseline power in logarithmic, dB, scale). Raw LFP signal was represented as dB levels of the data acquisition system’s analog-to-digital unit.
Statistical analysis. Statistical testing of differences between conditions was performed in the following fashion. First we calculated power across an entire epoch: fixation period, cue presentation, first delay, sample presentation, second delay (rather than at every time point, as illustrated in spectrograms). Secondly, we averaged power values in these epochs from all trials of every electrode site, essentially treating each LFP site as one observation. We then constructed a 3-way ANOVA model, with factors selective- partially-selective- nonselective sites; prefrontal subdivision; and task epoch. In every case, this analysis was performed for the gamma frequency band, defined as 45-100 Hz; effects of training on other LFP properties have been reported elsewhere 12, 13.