Friday 12th August 2022

The hippocampus as the switchboard between perception and memory


Imagine spotting a familiar face at a (real) conference. As your acquaintance approaches, you frantically try to recall the last time the two of you met and—without sneakily glancing at the nametag—remember what their name was. This example illustrates how adaptive behavior often requires us to shift our focus from external sensory information to internal mnemonic representations. In experimental terms, this scenario constitutes a cued recall task, where a reminder cue may or may not trigger recall of associated mnemonic target information. How does our brain accomplish the feat of converting an external reminder into a target memory?

According to computational models, the hippocampus links disparate cortical representations into a coherent memory trace (1, 2). It retains pointers to the cortical sites involved in the initial experience (3, 4) such that presenting a partial reminder prompts reinstatement of the entire association via hippocampal pattern completion (5, 6). In support of these models, human functional MRI (fMRI) studies linked hippocampal activation with cortical reinstatement of mnemonic target representations during successful recall (713). However, the relatively poor temporal resolution of the fMRI signal leaves open whether the hippocampus precedes or follows mnemonic reinstatement, let alone whether hippocampal engagement would mark the rapid switch from perceptual cue to mnemonic target representations.

Moreover, the cognitive complexity and representational richness of memory recall likely requires concerted engagement of wider brain networks (14, 15). Indeed, beyond the hippocampus, neuroimaging work has consistently implicated a particular set of cortical regions in episodic memory tasks (16, 17), herein referred to as the “cortical retrieval network” (CRN). The CRN overlaps with the “default mode network” (18) and includes posterior parietal regions as well as medial prefrontal cortex. It has been linked to retrieval success across multiple stimulus domains (19) as well as to episodic (re)construction processes (20, 21). Critically, a recent study employing “lesion network mapping” suggests that the hippocampus serves as a functional hub linking these cortical nodes in service of memory processes (22). While these results indicate that successful memory relies on intricate hippocampal–cortical interactions, the temporal dynamics within the CRN are challenging to resolve with fMRI alone, hampering understanding of different CRN regions’ contributions (16).

To overcome these limitations, we used intracranial electroencephalography (iEEG) complemented by high-density scalp EEG to reveal 1) the role of the hippocampus in the conversion of perceptual cues to mnemonic targets and 2) the ensuing dynamics in the frontoparietal retrieval network.



We used the same memory paradigm (Fig. 1) in an iEEG study (n = 11) and a high-density scalp EEG study (n = 20). In addition, we conducted “localizer” runs (Fig. 1A) to train a classifier to distinguish brain patterns of object vs. scene representations (see below). In the memory experiment (Fig. 1B), participants were presented with pairs of object and scene images during encoding. During retrieval, a cued recall task was employed in which only one of the images was shown (“cue”), with the question whether the associated image (“target”) was also remembered. Catch trials were interspersed in which participants were prompted to describe the target image after giving a “Remember” response.

Fig. 1.

Experimental paradigm. (A) In a perceptual localizer session, participants saw trial-unique images of objects and scenes and indicated the category of the given image. This part served as an independent training dataset for multivariate pattern analyses. (B) The main experiment employed an object–scene memory task, consisting of an encoding phase (Top) and a cued recall phase (Bottom). During encoding, participants saw trial-unique object–scene pairs and indicated whether the given combination was plausible or implausible. During cued recall, participants were given either the object or the scene image as the cue and were asked to recall the paired target (scene or object image, respectively). The key conditions were 1) trials in which participants indicated they did remember the target image (“Remember” trials) and 2) trials in which participants indicated they did not remember the target image (“Forgot” trials). Labels below denote the cue-target (memory) status of trials. O = object, S = scene, R = remember, F = forgot.

In the iEEG study, accuracy on the localizer task was on average 95% (SEM = 2%) correct (mean reaction time [RT] = 1.40 s, SEM = 0.21). During the cued recall task, iEEG participants indicated they remembered the target on 67% of trials (SEM = 5%). During catch trials, accuracy was 94% (SEM = 2%). RTs were faster for “Remember” trials (mean [M] = 2.59 s, SEM = 0.26) than for “Forgot” trials [M = 5.95 s, SEM = 0.40; t (10) = 8.46, P < 0.001]. “Remember” RTs did not differ significantly for object vs. scene targets [t (10) = 0.40, P = 0.695].

In the scalp EEG study, participants remembered 60% (SEM = 3%) of target images. Accuracy on catch trials was 92% (SEM = 2%). RTs were faster for “Remember” trials (M = 1.61 s, SEM = 0.08) than for “Forgot” trials [M = 2.37 s, SEM = 0.17; t (19) = 5.30, P < 0.001]. Again, RTs did not differ significantly for object vs. scene targets [t (19) = 0.73, P = 0.476]. Given average RTs for “Remember” trials, iEEG and scalp EEG data were analyzed from −0.5 s to 2.6 s and 1.6 s, respectively.

A Hippocampal Recall Signal at ∼500 ms.

Our first analysis examined spectral power in the hippocampus (Fig. 2A) during successful vs. unsuccessful cued recall (“Remember” vs. “Forgot”). As shown in Fig. 2B, we observed an extended cluster in the gamma frequency range (55 to 110 Hz, 570 to 1,730 ms, peak frequency: 85 Hz) in which “Remember” trials elicited greater power than “Forgot” trials [Pcluster = 0.011, summed cluster t (10) = 2,434]. The gamma effect was followed by a power decrease for “Remember” trials relative to “Forgot” trials below 30 Hz, with a distinctive peak in the alpha band [2 to 29 Hz, 680 to 2,600 ms, peak frequency: 10 Hz; Pcluster = 0.004, summed cluster t (10) = −6,391]. These hippocampal gamma/alpha effects were highly reliable across participants (11/11) and across subselections of hippocampal contacts (anterior hippocampus, posterior hippocampus, or nonpathological tissue according to clinical diagnostics; SI Appendix, Fig. S1). Hippocampal gamma and alpha power time courses are shown in Fig. 2C, averaged across the peak ranges of 80 to 90 Hz for gamma and 8 to 12 Hz for alpha (note though that results remain the same when using a weighted average across the entire clusters of significant frequencies; SI Appendix, Fig. S2). Across participants, there was a significant negative correlation between the earlier gamma effect (80 to 90 Hz, averaged from 500 to 1,500 ms) and the later alpha effect (8 to 12 Hz, averaged from 1,500 to 2,500 ms) [Pearson r (11) = −0.71, P = 0.015; see SI Appendix, Fig. S2C for a two-dimensional correlation map]. This finding replicates a previous report in which we found a gamma power increase followed by an alpha power decrease for successful vs. unsuccessful associative recognition memory (23) and extends it to a cued recall paradigm.


Read More:The hippocampus as the switchboard between perception and memory