Persistent neural activity is a putative mechanism for the maintenance of working memories. Persistent activity relies on the activity of a distributed network of areas, but the differential contribution of each area remains unclear. We recorded single neurons in the human medial frontal cortex and the medial temporal lobe while subjects held up to three items in memory. We found persistently active neurons in both areas. Persistent activity of hippocampal and amygdala neurons was stimulus-specific, formed stable attractors, and was predictive of memory content. Medial frontal cortex persistent activity, on the other hand, was modulated by memory load and task set but was not stimulus-specific. Trial-by-trial variability in persistent activity in both areas was related to memory strength, because it predicted the speed and accuracy by which stimuli were remembered. This work reveals, in humans, direct evidence for a distributed network of persistently active neurons supporting working memory maintenance.

Memory-based decisions are often accompanied by an assessment of choice certainty, but the mechanisms of such confidence judgments remain unknown. We studied the response of 1065 individual neurons in the human hippocampus and amygdala while neurosurgical patients made memory retrieval decisions together with a confidence judgment. Combining behavioral, neuronal and computational analysis, we identified a population of memory-selective (MS) neurons whose activity signaled stimulus familiarity and confidence as assessed by subjective report. In contrast, the activity of visually selective (VS) neurons was not sensitive to memory strength. The groups further differed in response latency, tuning, and extracellular waveforms. The information provided by MS neurons was sufficient for a race model to decide stimulus familiarity and retrieval confidence. Together, this demonstrates a trial-by-trial relationship between a specific group of neurons and declared memory strength in humans. We suggest that VS and MS neurons are a substrate for declarative memories.

The human amygdala is a key structure for processing emotional facial expressions, but it remains unclear what aspects of emotion are processed. We investigated this question with three different approaches: behavioural analysis of 3 amygdala lesion patients, neuroimaging of 19 healthy adults, and single-neuron recordings in 9 neurosurgical patients. The lesion patients showed a shift in behavioural sensitivity to fear, and amygdala BOLD responses were modulated by both fear and emotion ambiguity (the uncertainty that a facial expression is categorized as fearful or happy). We found two populations of neurons, one whose response correlated with increasing degree of fear, or happiness, and a second whose response primarily decreased as a linear function of emotion ambiguity. Together, our results indicate that the human amygdala processes both the degree of emotion in facial expressions and the categorical ambiguity of the emotion shown and that these two aspects of amygdala processing can be most clearly distinguished at the level of single neurons.

Humans can self-monitor errors without explicit feedback, resulting in behavioral adjustments on subsequent trials such as post-error slowing (PES). The error-related negativity (ERN) is a well-established macroscopic scalp EEG correlate of error self-monitoring, but its neural origins and relationship to PES remain unknown. We recorded in the frontal cortex of patients performing a Stroop task and found neurons that track self-monitored errors and error history in dorsal anterior cingulate cortex (dACC) and pre-supplementary motor area (pre-SMA). Both the intracranial ERN (iERN) and error neuron responses appeared first in pre-SMA, and ~50 ms later in dACC. Error neuron responses were correlated with iERN amplitude on individual trials. In dACC, such error neuron-iERN synchrony and responses of error-history neurons predicted the magnitude of PES. These data reveal a human single-neuron correlate of the ERN and suggest that dACC synthesizes error information to recruit behavioral control through coordinated neural activity.

Holding information in working memory (WM) is an active and effortful process that is accompanied by sustained load-dependent changes in oscillatory brain activity. These proportional power increases are often reported in EEG studies recording theta over frontal midline sites. Intracranial recordings, however, yield mixed results depending on the brain area being recorded from. We recorded intracranial EEG with depth electrodes in 13 patients with epilepsy that were performing a Sternberg WM task. Here, we investigated patterns of theta power changes as a function of memory load during maintenance in three areas critical for WM: dorsolateral prefrontal cortex (DLPFC), dorsal anterior cingulate cortex (dACC) and hippocampus. Theta-frequency power in both hippocampus and dACC increased during maintenance. In contrast, theta-frequency power in the DLPFC decreased during maintenance and this decrease was proportional to memory load. Only the power decreases in DLPFC, but not the power increases in hippocampus and dACC, were predictive of behavior in a given trial. The extent of the load-related theta power decreases in the DLPFC in a given subject predicted a subject’s reaction times, revealing that DLPFC theta explains individual differences in WM ability between subjects. Together, this data reveals a pattern of theta power decreases in the DLPFC that is predictive of behavior and that is opposite of that in other brain areas. This result suggests that theta band power changes serve different cognitive functions in different brain areas and specifically that theta power decreases in DLPFC have an important role in maintenance of information.