A neural network account of memory replay and knowledge consolidation.

Replay can consolidate memories through offline neural reactivation related to past experiences. Category knowledge is learned across multiple experiences, and its subsequent generalization is promoted by consolidation and replay during rest and sleep. However, aspects of replay are difficult to determine from neuroimaging studies. We provided insights into category knowledge replay by simulating these processes in a neural network which approximated the roles of the human ventral visual stream and hippocampus. Generative replay, akin to imagining new category instances, facilitated generalization to new experiences. Consolidation-related replay may therefore help to prepare us for the future as much as remember the past. Generative replay was more effective in later network layers functionally similar to the lateral occipital cortex than layers corresponding to early visual cortex, drawing a distinction between neural replay and its relevance to consolidation. Category replay was most beneficial for newly acquired knowledge, suggesting replay helps us adapt to changes in our environment. Finally, we present a novel mechanism for the observation that the brain selectively consolidates weaker information, namely a reinforcement learning process in which categories were replayed according to their contribution to network performance. This reinforces the idea of consolidation-related replay as an active rather than passive process.

A too-good-to-be-true prior to reduce shortcut reliance.

Despite their impressive performance in object recognition and other tasks under standard testing conditions, deep networks often fail to generalize to out-of-distribution (o.o.d.) samples. One cause for this shortcoming is that modern architectures tend to rely on ǣshortcutsǥ superficial features that correlate with categories without capturing deeper invariants that hold across contexts. Real-world concepts often possess a complex structure that can vary superficially across contexts, which can make the most intuitive and promising solutions in one context not generalize to others. One potential way to improve o.o.d. generalization is to assume simple solutions are unlikely to be valid across contexts and avoid them, which we refer to as the too-good-to-be-true prior. A low-capacity network (LCN) with a shallow architecture should only be able to learn surface relationships, including shortcuts. We find that LCNs can serve as shortcut detectors. Furthermore, an LCN's predictions can be used in a two-stage approach to encourage a high-capacity network (HCN) to rely on deeper invariant features that should generalize broadly. In particular, items that the LCN can master are downweighted when training the HCN. Using a modified version of the CIFAR-10 dataset in which we introduced shortcuts, we found that the two-stage LCN-HCN approach reduced reliance on shortcuts and facilitated o.o.d. generalization.

Mouth magnetoencephalography: A unique perspective on the human hippocampus.

Traditional magnetoencephalographic (MEG) brain imaging scanners consist of a rigid sensor array surrounding the head; this means that they are maximally sensitive to superficial brain structures. New technology based on optical pumping means that we can now consider more flexible and creative sensor placement. Here we explored the magnetic fields generated by a model of the human hippocampus not only across scalp but also at the roof of the mouth. We found that simulated hippocampal sources gave rise to dipolar field patterns with one scalp surface field extremum at the temporal lobe and a corresponding maximum or minimum at the roof of the mouth. We then constructed a fitted dental mould to accommodate an Optically Pumped Magnetometer (OPM). We collected data using a previously validated hippocampal-dependant task to test the empirical utility of a mouth-based sensor, with an accompanying array of left and right temporal lobe OPMs. We found that the mouth sensor showed the greatest task-related theta power change. We found that this sensor had a mild effect on the reconstructed power in the hippocampus (~10% change) but that coherence images between the mouth sensor and reconstructed source images showed a global maximum in the right hippocampus. We conclude that augmenting a scalp-based MEG array with sensors in the mouth shows unique promise for both basic scientists and clinicians interested in interrogating the hippocampus.

The relationship between hippocampal subfield volumes and autobiographical memory persistence.

Structural integrity of the human hippocampus is widely acknowledged to be necessary for the successful encoding and retrieval of autobiographical memories. However, evidence for an association between hippocampal volume and the ability to recall such memories in healthy individuals is mixed. Here we examined this issue further by combining two approaches. First, we focused on the anatomically distinct subregions of the hippocampus where more nuanced associations may be expressed compared to considering the whole hippocampal volume. A manual segmentation protocol of hippocampal subregions allowed us to separately calculate the volumes of the dentate gyrus/CA4, CA3/2, CA1, subiculum, pre/parasubiculum and uncus. Second, a critical feature of autobiographical memories is that they can span long time periods, and so we sought to consider how memory details persist over time by conducting a longitudinal study whereby participants had to recall the same autobiographical memories on two visits spaced 8 months apart. Overall, we found that there was no difference in the total number of internal (episodic) details produced at Visits 1 and 2. However, further probing of detail subcategories revealed that specifically the amount of subjective thoughts and emotions included during recall had declined significantly by the second visit. We also observed a strong correlation between left pre/parasubiculum volume and the amount of autobiographical memory internal details produced over time. This positive relationship was evident for particular facets of the memories, with remembered events, perceptual observations and thoughts and emotions benefitting from greater volume of the left pre/parasubiculum. These preliminary findings expand upon existing functional neuroimaging evidence by highlighting a potential link between left pre/parasubiculum volume and autobiographical memory. A larger pre/parasubiculum appears not only to protect against memory decay, but may possibly enhance memory persistence, inviting further scrutiny of the role of this brain region in remote autobiographical memory retrieval.

Nonmonotonic recruitment of ventromedial prefrontal cortex during remote memory recall.

Systems-level consolidation refers to the time-dependent reorganisation of memory traces in the neocortex, a process in which the ventromedial prefrontal cortex (vmPFC) has been implicated. Capturing the precise temporal evolution of this crucial process in humans has long proved elusive. Here, we used multivariate methods and a longitudinal functional magnetic resonance imaging (fMRI) design to detect, with high granularity, the extent to which autobiographical memories of different ages were represented in vmPFC and how this changed over time. We observed an unexpected time course of vmPFC recruitment during retrieval, rising and falling around an initial peak of 8-12 months, before reengaging for older 2- and 5-year-old memories. This pattern was replicated in 2 independent sets of memories. Moreover, it was further replicated in a follow-up study 8 months later with the same participants and memories, for which the individual memory representations had undergone their hypothesised strengthening or weakening over time. We conclude that the temporal engagement of vmPFC in memory retrieval seems to be nonmonotonic, revealing a complex relationship between systems-level consolidation and prefrontal cortex recruitment that is unaccounted for by current theories.

vmPFC Drives Hippocampal Processing during Autobiographical Memory Recall Regardless of Remoteness.

Our ability to recall past experiences, autobiographical memories (AMs), is crucial to cognition, endowing us with a sense of self and underwriting our capacity for autonomy. Traditional views assume that the hippocampus orchestrates event recall, whereas recent accounts propose that the ventromedial prefrontal cortex (vmPFC) instigates and coordinates hippocampal-dependent processes. Here we sought to characterize the dynamic interplay between the hippocampus and vmPFC during AM recall to adjudicate between these perspectives. Leveraging the high temporal resolution of magnetoencephalography, we found that the left hippocampus and the vmPFC showed the greatest power changes during AM retrieval. Moreover, responses in the vmPFC preceded activity in the hippocampus during initiation of AM recall, except during retrieval of the most recent AMs. The vmPFC drove hippocampal activity during recall initiation and also as AMs unfolded over subsequent seconds, and this effect was evident regardless of AM age. These results recast the positions of the hippocampus and the vmPFC in the AM retrieval hierarchy, with implications for theoretical accounts of memory processing and systems-level consolidation.

The Neural Dynamics of Novel Scene Imagery.

Retrieval of long-term episodic memories is characterized by synchronized neural activity between hippocampus and ventromedial prefrontal cortex (vmPFC), with additional evidence that vmPFC activity leads that of the hippocampus. It has been proposed that the mental generation of scene imagery is a crucial component of episodic memory processing. If this is the case, then a comparable interaction between the two brain regions should exist during the construction of novel scene imagery. To address this question, we leveraged the high temporal resolution of MEG to investigate the construction of novel mental imagery. We tasked male and female humans with imagining scenes and single isolated objects in response to one-word cues. We performed source-level power, coherence, and causality analyses to characterize the underlying interregional interactions. Both scene and object imagination resulted in theta power changes in the anterior hippocampus. However, higher theta coherence was observed between the hippocampus and vmPFC in the scene compared with the object condition. This interregional theta coherence also predicted whether imagined scenes were subsequently remembered. Dynamic causal modeling of this interaction revealed that vmPFC drove activity in hippocampus during novel scene construction. Additionally, theta power changes in the vmPFC preceded those observed in the hippocampus. These results constitute the first evidence in humans that episodic memory retrieval and scene imagination rely on similar vmPFC-hippocampus neural dynamics. Furthermore, they provide support for theories emphasizing similarities between both cognitive processes and perspectives that propose the vmPFC guides the construction of context-relevant representations in the hippocampus.SIGNIFICANCE STATEMENT Episodic memory retrieval is characterized by a dialog between hippocampus and ventromedial prefrontal cortex (vmPFC). It has been proposed that the mental generation of scene imagery is a crucial component of episodic memory processing. An ensuing prediction would be of a comparable interaction between the two brain regions during the construction of novel scene imagery. Here, we leveraged the high temporal resolution of MEG and combined it with a scene imagination task. We found that a hippocampal-vmPFC dialog existed and that it took the form of vmPFC driving the hippocampus. We conclude that episodic memory and scene imagination share fundamental neural dynamics and the process of constructing vivid, spatially coherent, contextually appropriate scene imagery is strongly modulated by vmPFC.

Imaging the human hippocampus with optically-pumped magnetoencephalography.

Optically-pumped (OP) magnetometers allow magnetoencephalography (MEG) to be performed while a participant's head is unconstrained. To fully leverage this new technology, and in particular its capacity for mobility, the activity of deep brain structures which facilitate explorative behaviours such as navigation, must be detectable using OP-MEG. One such crucial brain region is the hippocampus. Here we had three healthy adult participants perform a hippocampal-dependent task - the imagination of novel scene imagery - while being scanned using OP-MEG. A conjunction analysis across these three participants revealed a significant change in theta power in the medial temporal lobe. The peak of this activated cluster was located in the anterior hippocampus. We repeated the experiment with the same participants in a conventional SQUID-MEG scanner and found similar engagement of the medial temporal lobe, also with a peak in the anterior hippocampus. These OP-MEG findings indicate exciting new opportunities for investigating the neural correlates of a range of crucial cognitive functions in naturalistic contexts including spatial navigation, episodic memory and social interactions.

The time course of systems consolidation of spatial memory from recent to remote retention: A comparison of the Immediate Early Genes Zif268, c-Fos and Arc.

Systems consolidation is a process involving the stabilisation of memory traces in the neocortex over time. The medial prefrontal cortex becomes increasingly important during the retrieval of older memories, however the timescale of its involvement is unclear, and the contribution of other neocortical brain regions to remote memory have received little attention. The Immediate Early Genes (IEGs) Zif268, c-Fos and Arc have been utilised as markers of neural activity during spatial memory retrieval, however the lack of a direct comparison between them hinders the interpretation of results. To address these questions, we examined the expression of Zif268, Arc and c-Fos protein in the medial prefrontal cortex, as well as the hippocampus, and the entorhinal, perirhinal, retrosplenial and parietal cortices of male Wistar rats following a probe trial of the Morris water maze either one day, seven days, 14 days or 30 days after acquisition. Activity of the medial prefrontal cortex during retrieval, as measured by all three IEGs, increased in correspondence with the age of the memory, reaching significance between 14 and 30 days. Similar increases in c-Fos and Arc were observed over the course of consolidation in other neocortical and parahippocampal areas, however this pattern was not observed with Zif268. Activity of the hippocampus remained largely unchanged across retention intervals. These findings suggest that systems consolidation of spatial memory takes at least two weeks, are consistent with an ongoing role for the hippocampus in the retrieval of spatial memory, and suggest that c-Fos and Arc may be a more sensitive measure of neural activity in response to behavioural tasks than Zif268.

Imaging spatial learning in the brain using immediate early genes: insights, opportunities and limitations.

Immediate early gene (IEG) imaging offers a sophisticated approach to study the neural basis of learning in the brain. Demonstrating a high degree of spatial resolution, the activation of entire neuronal ensembles at multiple time-points can be observed. IEG imaging techniques have revealed a high level of responsiveness to spatial exploration within the hippocampus and other brain regions. The pattern of IEG activation is tightly linked with specific environments and appears to be involved in the subsequent consolidation of spatial information. This incidental learning is a potential confounding factor in studies investigating the neural correlates of spatial learning in both the radial arm maze and water maze. Although both these tasks increase hippocampal IEG expression from baseline levels, where control groups have fully explored the apparatus in the absence of task demands, or where animals are performing a non-spatial task, IEG expression in this region is comparable to spatially trained groups. However, the relationship between IEG expression and task performance, as well as the pattern of brain activation has been shown to differentiate between experimental and control groups. Inconsistencies between training protocols appear to contribute to the discrepancies between reported findings, and the role of IEG expression in the retention of spatial memory tasks remains unclear. Further investigation of the time course and dynamics of IEG expression during learning and retention is required to fully interpret observed results.

Watching Movies Unfold, a Frame-by-Frame Analysis of the Associated Neural Dynamics.

Our lives unfold as sequences of events. We experience these events as seamless, although they are composed of individual images captured in between the interruptions imposed by eye blinks and saccades. Events typically involve visual imagery from the real world (scenes), and the hippocampus is frequently engaged in this context. It is unclear, however, whether the hippocampus would be similarly responsive to unfolding events that involve abstract imagery. Addressing this issue could provide insights into the nature of its contribution to event processing, with relevance for theories of hippocampal function. Consequently, during magnetoencephalography (MEG), we had female and male humans watch highly matched unfolding movie events composed of either scene image frames that reflected the real world, or frames depicting abstract patterns. We examined the evoked neuronal responses to each image frame along the time course of the movie events. Only one difference between the two conditions was evident, and that was during the viewing of the first image frame of events, detectable across frontotemporal sensors. Further probing of this difference using source reconstruction revealed greater engagement of a set of brain regions across parietal, frontal, premotor, and cerebellar cortices, with the largest change in broadband (1-30 Hz) power in the hippocampus during scene-based movie events. Hippocampal engagement during the first image frame of scene-based events could reflect its role in registering a recognizable context perhaps based on templates or schemas. The hippocampus, therefore, may help to set the scene for events very early on.

Testing covariance models for MEG source reconstruction of hippocampal activity.

Beamforming is one of the most commonly used source reconstruction methods for magneto- and electroencephalography (M/EEG). One underlying assumption, however, is that distant sources are uncorrelated and here we tested whether this is an appropriate model for the human hippocampal data. We revised the Empirical Bayesian Beamfomer (EBB) to accommodate specific a-priori correlated source models. We showed in simulation that we could use model evidence (as approximated by Free Energy) to distinguish between different correlated and uncorrelated source scenarios. Using group MEG data in which the participants performed a hippocampal-dependent task, we explored the possibility that the hippocampus or the cortex or both were correlated in their activity across hemispheres. We found that incorporating a correlated hippocampal source model significantly improved model evidence. Our findings help to explain why, up until now, the majority of MEG-reported hippocampal activity (typically making use of beamformers) has been estimated as unilateral.

A novel control condition for spatial learning in the Morris water maze.

BACKGROUND: Assessing markers of neural activity in rodent behavioural tasks benefits from meaningful comparison with matched control conditions. The Morris water maze is a widely used learning and memory task, but currently implemented control conditions are not optimal. NEW METHOD: We created a novel control condition involving random escape platform placement for each trial, while increasing the frequency of platforms throughout the training period. These control rats could therefore escape the maze by swimming in a random fashion and were not required to learn a platform location. Crucially, because the number of available escape platforms increased throughout training, their escape latencies should decline in line with rats who were learning a fixed platform location. RESULTS: No evidence of place learning was observed in this novel condition, but their swimming behaviour, including escape latency, distance and velocity were well matched to spatially-trained rats throughout training. Further, they did not display stress-related behaviour. COMPARISON WITH EXISTING METHODS: We contrasted this new control condition against a frequently used control where animals swim freely in the maze and showed behaviour was more closely matched to spatially-trained animals. CONCLUSIONS: This novel control condition represents a significant advance from those currently available and may assist in the interpretation of task-related neural activity.

Remote Memory and the Hippocampus: A Constructive Critique.

The hippocampus is known to be recruited during the recall of experiences from our distant past, despite evidence that memory traces in this region vanish over time. Extant theories of systems-level consolidation have yet to accommodate both phenomena. We propose that the hippocampus reconstructs remote memories in the absence of the original trace. It accomplishes this by assembling consolidated neocortical elements into spatially coherent scenes that form the basis of unfolding memory events. This reconstruction is likely facilitated by input from the ventromedial prefrontal cortex. This process-oriented approach to hippocampal recruitment during remote recollection is consistent with its increasingly acknowledged role in constructing mental representations beyond the domain of memory.

Temporal dynamics of immediate early gene expression during cellular consolidation of spatial memory.

The consolidation of newly acquired memories on a cellular level is thought to take place in the first few hours following learning. This process is dependent on de novo protein synthesis during this time, which ultimately leads to long-term structural and functional neuronal changes and the stabilisation of a memory trace. Immediate early genes (IEGs) are rapidly expressed in neurons following learning, and previous research has suggested more than one wave of IEG expression facilitates consolidation in the hours following learning. We analysed the expression of Zif268, c-Fos and Arc protein in a number of brain regions involved in spatial learning either 90min, 4h or 8h following training in the Morris water maze task. Consistent with the role of IEGs in the earliest stages of consolidation, a single wave of expression was observed in most brain regions at 90min, however a subsequent wave of expression was not observed at 8h. In fact, Zif268 expression was observed to fall below the levels of naïve controls at this time-point in the medial prefrontal and perirhinal cortices. This may be indicative of synaptic downscaling in these regions in the hours following learning, and an important marker of the consolidation of spatial memory.

Segmenting subregions of the human hippocampus on structural magnetic resonance image scans: An illustrated tutorial.

BACKGROUND: The hippocampus plays a central role in cognition, and understanding the specific contributions of its subregions will likely be key to explaining its wide-ranging functions. However, delineating substructures within the human hippocampus in vivo from magnetic resonance image scans is fraught with difficulties. To our knowledge, the extant literature contains only brief descriptions of segmentation procedures used to delineate hippocampal subregions in magnetic resonance imaging/functional magnetic resonance imaging studies. METHODS: Consequently, here we provide a clear, step-by-step and fully illustrated guide to segmenting hippocampal subregions along the entire length of the human hippocampus on 3T magnetic resonance images. RESULTS: We give a detailed description of how to segment the hippocampus into the following six subregions: dentate gyrus/Cornu Ammonis 4, CA3/2, CA1, subiculum, pre/parasubiculum and the uncus. Importantly, this in-depth protocol incorporates the most recent cyto- and chemo-architectural evidence and includes a series of comprehensive figures which compare slices of histologically stained tissue with equivalent 3T images. CONCLUSION: As hippocampal subregion segmentation is an evolving field of research, we do not suggest this protocol is definitive or final. Rather, we present a fully explained and expedient method of manual segmentation which remains faithful to our current understanding of human hippocampal neuroanatomy. We hope that this 'tutorial'-style guide, which can be followed by experts and non-experts alike, will be a practical resource for clinical and research scientists with an interest in the human hippocampus.