Persistent, memorandum-specific neuronal spiking activity has long been hypothesized to underlie working memory. However, emerging evidence suggests a possible role for ‘activity-silent’, synaptic mechanisms. This issue remains controversial because evidence for either view has largely relied on datasets that fail to capture single-trial population dynamics or on indirect measures of neuronal spiking. We addressed this by examining the dynamics of mnemonic information on single trials obtained from large, local populations of lateral prefrontal neurons recorded simultaneously in monkeys performing a working memory task.  Here, we show that mnemonic information does not persist in the spiking activity of neuronal populations during memory delays, but instead alternates between coordinated ‘On’ and ‘Off’ states. At the level of single neurons, Off states are driven both by a loss of selectivity for memoranda and a return of firing rates to spontaneous levels. Further exploiting the large-scale recordings, we show that mnemonic information is available in the patterns of functional connections among neuronal ensembles during Off states. Our results suggest that intermittent periods of memoranda-specific spiking coexist with synaptic mechanisms to support working memory. 

Subjects

Three adult male rhesus monkeys (Macaca mulatta) participated in the experiment. Monkey A, H, and J weighed 11, 14, and 12 kg, respectively. All surgical and experimental procedures were approved by the Stanford University Institutional Animal Care and Use Committee and were in accordance with the policies and procedures of the National Institutes of Health.

Behavioral task                         

Stimuli were presented on a VIEWPixx3D monitor positioned at a viewing distance of 60 cm using Psychtoolbox and MATLAB (MathWorks). Eye position was monitored at 1kHz using an Eyelink 1000 eye-tracking system (SR Research). On each trial, the animals were presented with a cue at one of 8 possible locations and reported this location after a brief memory delay to receive a fluid reward. Cues were square frames (green for Monkey A, black for Monkey H, white for Monkey J) measuring 1 degree of visual angle on a side and presented at 5-7 degrees of eccentricity (depending on the session).

Monkeys initiated behavioral trials by fixating a central fixation spot presented on a uniform gray background. After the monkeys maintained fixation for 600-800 ms (randomly selected on each trial), a cue appeared for 50 ms at one of 8 possible locations separated by 45 degrees around fixation. Cue presentation was followed by a delay period that varied randomly from 1400-1600 ms. After the delay period, the fixation spot disappeared, and the animal was presented with one of two possible response screens. On match-to-sample (MTS) trials, two targets appeared (filled blue circles, radius 1 DVA), one at the previously cued location, and the other at one of the 7 remaining non-cued locations. On memory-guided saccade (MGS) trials, no targets appeared. In either case, the animals received a juice reward for making an eye movement to within 5 degrees of visual angle (DVA) of the previously cued location and then maintaining fixation for 200 ms. MTS and MGS trials were randomly interleaved such that the animals could not predict the trial type. Monkey J was trained on and performed only the MGS task. The animals had to maintain their gaze within 3 DVA (monkey A) or 2 DVA (monkey H & J) from fixation throughout the trial until the response stage. The intertrial interval was 300-1000 ms after each correct response. Failures to acquire fixation, fixation breaks, and incorrect responses were not rewarded and were followed by a 2,000 ms intertrial interval.

Surgical Procedures and Recordings

Monkeys were implanted with a titanium headpost to immobilize the head and with a titanium chamber to provide access to the brain (see ref. 1 for full details). In a previous study, we identified the frontal eye field (FEF) based on its neurophysiological characteristics and the ability to evoke saccades with electrical stimulation at low currents. Here, we recorded from both area 8, within and anterior to the FEF, and the principal sulcus (9/46) (Monkey A: area 8; Monkey H: 8 and 9/46; Monkey J; 9/46) using primate neuropixels probes. During each session, we pierced the dura using a screw-driven 21 gauge pointed cannula and lowered a single probe through this cannula using a combination of custom 3D printed grids and motorized drives (NAN instruments). Probe trajectories spanned several cortical columns, as inferred from the broad distribution of preferred cue locations across neurons. Recordings were allowed to settle for ~30 minutes prior to the start of the experiment to mitigate drift. We configured probes to record from 384 active channels in a contiguous block, allowing dense sampling of neuronal activity along a 3.84 mm span.

Activity was filtered and digitized activity at the headstage (300 Hz high-pass filter, 30 kHz sampling frequency). Activity was monitored during experimental sessions and saved to disk using SpikeGLX (https://billkarsh.github.io/SpikeGLX/).

Data Preprocessing

Spiking in the action-potential band was identified and sorted offline using Kilosort3. As we were interested in population-level coding of memory, we analyzed both putative single- and multi-unit clusters identified by Kilosort. Spike times were aligned to a digital trigger on each trial indicating cue onset and corrected for a lag in stimulus presentation estimated offline using photodiode measurements from the stimulus display and the timing of the cue-evoked response. Neurons that fired fewer than 1000 spikes in the ~3 hour experimental sessions were excluded from further analyses. Spikes times were binned with a width and timestep of 1 ms.