Background: Direct electrical stimulation of the amygdala can enhance declarative memory for specific events. An unanswered question is what underlying neurophysiological changes are induced by amygdala stimulation. Objective: To leverage interpretable machine learning to identify the neurophysiological processes underlying amygdala-mediated memory, and to develop more efficient neuromodulation technologies. Method: Patients with treatment-resistant epilepsy and depth electrodes placed in the hippocampus and amygdala performed a recognition memory task for neutral images of objects. During the encoding phase, 160 images were shown to patients. Half of the images were followed by brief low-amplitude amygdala stimulation. For local field potentials (LFPs) recorded from key medial temporal lobe structures, feature vectors were calculated by taking the average spectral power in canonical frequency bands, before and after stimulation, to train a logistic regression classification model with elastic net regularization to differentiate brain states. Results: Classifying the neural states at the time of encoding based on images subsequently remembered versus not-remembered showed that theta and slow-gamma power in the hippocampus were the most important features predicting subsequent memory performance. Classifying the post-image neural states at the time of encoding based on stimulated versus unstimulated trials showed that amygdala stimulation led to increased gamma power in the hippocampus. Conclusion: Amygdala stimulation induced pro-memory states in the hippocampus to enhance subsequent memory performance. Interpretable machine learning provides an effective tool for investigating the neurophysiological effects of brain stimulation.

People often forget information because they fail to effectively encode it. Here, we test the hypothesis that targeted electrical stimulation can modulate neural encoding states and subsequent memory outcomes. Using recordings from neurosurgical epilepsy patients with intracranially implanted electrodes, we trained multivariate classifiers to discriminate spectral activity during learning that predicted remembering from forgetting, then decoded neural activity in later sessions in which we applied stimulation during learning. Stimulation increased encoding-state estimates and recall if delivered when the classifier indicated low encoding efficiency but had the reverse effect if stimulation was delivered when the classifier indicated high encoding efficiency. Higher encoding-state estimates from stimulation were associated with greater evidence of neural activity linked to contextual memory encoding. In identifying the conditions under which stimulation modulates memory, the data suggest strategies for therapeutically treating memory dysfunction.

The hippocampus and surrounding medial-temporal-lobe (MTL) structures are critical for both memory and spatial navigation, but we do not fully understand the neuronal representations used to support these behaviors. Much research has examined how the MTL neurally represents spatial information, such as with “place cells” that represent an animal's current location or “head-direction cells” that code for an animal's current heading. In addition to behaviors that require an animal to attend to the current spatial location, navigating to remote destinations is a common part of daily life. To examine the neural basis of these behaviors, we recorded single-neuron activity from neurosurgical patients playing Treasure Hunt, a virtual-reality spatial-memory task. By analyzing how the activity of these neurons related to behavior in Treasure Hunt, we found that the firing rates of many MTL neurons during navigation significantly changed depending on the position of the current spatial target. In addition, we observed neurons whose firing rates during navigation were tuned to specific heading directions in the environment, and others whose activity changed depending on the timing within the trial. By showing that neurons in our task represent remote locations rather than the subject's own position, our results suggest that the human MTL can represent remote spatial information according to task demands. Tsitsiklis et al. record single-unit activity from epilepsy patients during spatial navigation. They find that the firing rates of MTL neurons vary with the locations of spatial targets, heading direction, and serial position. This suggests that the human MTL represents multiple types of spatiotemporal information to support spatial cognition.