Learning and Representation of Declarative Memories by Single Neurons in the Human Brain

Author: Rutishauser, Ueli

Year: 2008

Degree: Dissertation (Ph.D.)

Advisors: Schuman, Erin Margaret; Koch, Christof; Mamelak, Adam N.

Committee Members: Laurent, Gilles J.; Mamelak, Adam N.; Koch, Christof; Schuman, Erin Margaret; O'Doherty, John P.

Option: Computation and Neural Systems

DOI: 10.7907/GX3N-QD05

Abstract

Episodic memories allow us to remember not only that we have seen an item before but also where and when we have seen it (context). Neurons in the medial temporal lobe (MTL) are critically involved in the acquisition of such memories. Since events happen only once, the ability to distinguish novel from familiar stimuli is crucial in order to rapidly encode such events after a single exposure. Theoretically, this is a hard learning problem (single-trial learning). Yet, successful detection of novelty is necessary for many types of learning. During retrieval, we can sometimes confidently report that we have seen something (familiarity) but cannot recollect where or when it was seen. Thus episodic memories have several components which can be recalled selectively. We recorded single neurons and local field potentials in the human hippocampus, amygdala, and anterior cingulate cortex while subjects remembered, and later retrieved, the identity and location of pictures shown. We describe two classes of neurons that exhibit such single-trial learning: novelty and familiarity detectors, which show a selective increase in firing for new and old stimuli, respectively. The neurons retain memory for the stimulus for at least 24 h. During retrieval, these neurons distinguish stimuli that will be successfully recollected from stimuli that will not be recollected. Similarly, they distinguish between failed and successful recognition. Pictures which were forgotten by the patient still evoked a non-zero response. Thus, their response can be different from the decision of the patient. Also, we demonstrate that listening to these neurons (during retrieval) enables a simple decoder to outperform the patient (i.e., it forgets fewer pictures). These data support a continuous strength of memory model of MTL function: the stronger the neuronal response, the better the memory (as opposed to a dual-process model). I also describe specific power increases in specific frequencies of the local field potential that are predictive of later retrieval success. These neural signatures, recorded during learning, thus indicate whether plasticity was successful or not.

Files