Single Neuron Coding of Identity in the Human Hippocampal Formation

REY, HERNAN G.; GORI, BELEN; CHAURE, FERNANDO J.; COLLAVINI, SANTIAGO; BLENKMANN, ALEJANDRO O.; SEOANE, PABLO; SEOANE, EDUARDO; KOCHEN, SILVIA; QUIAN QUIROGA, RODRIGO

CURRENT BIOLOGY

CELL PRESS

Lugar: United States; Año: 2020

0960-9822

SUMMARYExperimental findings show the ubiquitous presenceof graded responses and tuning curves in theneocortex, particularly in visual areas [1?15]. Amongthese, inferotemporal-cortex (IT) neurons respond tocomplex visual stimuli, but differences in the neurons?responses can be used to distinguish thestimuli eliciting the responses [8, 9, 16?18]. The ITprojects directly to the medial temporal lobe (MTL)[19], where neurons respond selectively to differentpictures of specific persons and even to their writtenand spoken names [20?22]. However, it is not clearwhether this is done through a graded coding, as inthe neocortex, or a truly invariant code, in which theresponse-eliciting stimuli cannot be distinguishedfrom each other. To address this issue, we recordedsingle neurons during the repeated presentation ofdifferent stimuli (pictures and written and spokennames) corresponding to the same persons. Usingstatistical tests and a decoding approach, we foundthat only in a minority of cases can the different picturesof a given person be distinguished from theneurons? responses and that in a larger proportionof cases, the responses to the pictures were differentto the ones to the written and spoken names. Weargue that MTL neurons tend to lack a representationof sensory features (particularly within a sensory modality),which can be advantageous for the memoryfunction attributed to this area [23?25], and that afull representation of memories is given by a combinationof mostly invariant coding in the MTL with arepresentation of sensory features in the neocortex.RESULTSExperimental Paradigm and Neural RecordingsWe recorded single neuron activity during 16 sessions in six patientswith drug-resistant epilepsy, who were candidates forsurgical treatment and implanted with intracranial electrodes(STAR Methods). For each identity used in an experimental session,a total of five stimuli were presented: three different pictures,the name written on the screen, and the name spokenby a computer-synthesized voice. Each stimulus was presented30 times in pseudorandom order (in contrast to the limited six trialsused in many previous works). From the 450 units recorded inthese sessions, the response criterion (STAR Methods; FigureS1) led to a set of 159 responses in 35 units (16 from amygdalaand 19 from hippocampus).Invariant Responses in the Human MTLFigure 1A shows an example of a unit from the hippocampus respondingto all the stimuli associated to the actor ??Jackie Chan.??From an average baseline firing of 3.65 Hz (SD 3.89), upon presentationof these stimuli, the neuron increased its firing to anaverage of 14.19 Hz (SD 5.66,) with peaks of up to 30 Hz, inthe response window. Figure 1B shows another example of aunit responding to all the stimuli associated to a TV host inArgentina. In this case, from an average baseline firing of3.05 Hz (SD 2.76), the neuron had an average firing of 11.67 Hz(SD 4.96) in response to these stimuli. Although most units respondedselectively to a single identity, there were several neurons(7 out of 35) showing responses to more than one identity.One example is shown in Figure S2, where a hippocampalneuron responded to the stimuli associated to three TVcelebrities.The normalized firing rate and response latency for all 159response-eliciting stimuli are shown in Figure 2A. First, the responsesto text and sound stimuli exhibit larger latencies thanthose to pictures. Moreover, the responses to the pictureshave been sorted by the latency of the 47 response-elicitingidentities, and although they cover a wide 250-ms latency rangefor the whole dataset, it can be seen that the responses within agiven identity are clustered together.To further understand the response characteristics of theseneurons, we first analyzed the degree of visual invarianceobserved in the single unit responses (Figure 2B). For eachresponse-eliciting identity, we compared the spike count associatedwith the three pictures against the one of three randomlyselected pictures from the other identities (STAR Methods). We