Learning & Memory

Recent studies have shown that a shared neuronal ensemble in the hippocampus links distinct contextual memories encoded within a certain time window (specifically, 5 hours, 2 and 7 days). Here we explored the temporal dynamics of two contextual (neutral and aversive) memories linking and analysed neuronal ensembles in the hippocampus, amygdala and different cortical regions reactivated during retrieval. Firstly, we have found that memories integrated across different time-points including several hours, days and weeks but not if learning phases was separated by short-term and very long-term time intervals. Secondly, we have demonstrated a higher neuronal co-reactivation in the hippocampus and amygdala during retrieval in case of memories integration that supports the hypothesis that shared neuronal ensembles link distinct memories. Finally, we have elicited that proportions of reactivated neuronal ensembles in these brain regions are greater in case of contextual memories integration.

In associative fear learning, weak or temporally constrained training may fail to link a conditioned stimulus (CS) with an aversive unconditioned stimulus (US), particularly when the contextual representation is impaired (the immediate shock deficit). Here, we systematically tested behavioral conditions that enable linking of an initially neutral auditory memory trace to an aversive episode. Male C57BL/6 mice were studied in four experiments manipulating (i) preexposure to a CS tone, (ii) the duration of context exploration before immediate footshock, and (iii) whether CS memory was tested in a novel or a familiar-like context. A 5 s tone followed immediately by footshock did not induce reliable fear to either the CS or the training context. CS preexposure three days before conditioning did not facilitate CS aversive memory when animals were tested in a completely novel context. However, robust facilitation emerged when the CS memory was tested in a context similar to the preexposure/conditioning one, indicating strong contextual gating of CS retrieval. Extending context exploration before shock enabled CS fear learning, but reduced (and even reversed) the effect of CS preexposure, consistent with latent inhibition. Together, these results delineate behavioral constraints for linking an initially neutral cue memory to an aversive event and highlight contextual control over the coupling and expression of cue memory traces.

Aversive associative learning paradigms such as inhibitory avoidance (IA) are frequently used to examine episodic-like memories in rodents. In IA, rodents learn to associate a context with a footshock, followed by testing for memory strength in the original training context and for memory precision in a similar yet distinct neutral context. The present work assessed the effects of different contextual exposure procedures on memory strength and precision in IA at both recent and remote time points using male and female Long-Evans rats. An initial experiment found that rats kept in the lit (non-shock) compartment of the IA apparatus for 60 s during training, as opposed to 10 s, displayed enhanced memory strength, with discrimination between both chambers at the recent retention test and generalization at the remote retention test. Subsequent experiments investigated the effects of contextual pre-exposure the day before training. The results indicate that pre-exposure to the neutral context promoted generalization without altering memory strength compared to the first experiment. In contrast, pre-exposure to the aversive chamber promoted discrimination and enhanced memory strength. Notably, the different procedures yielded similar effects in both sexes. However, the results also indicate an overall pattern of greater contextual discrimination in females compared to males. These findings provide evidence for how different contextual exposures influence the degree of encoding at the time of training and a behavioral foundation for future studies examining the neurobiological mechanisms underlying memory strength and precision in IA, while highlighting the importance of using both sexes in initial behavioral work. Significance StatementStrength and precision are two fundamental properties of memory that can be simultaneously measured using inhibitory avoidance (IA), a type of context-footshock association task. However, little is known about how different context exposures alter rats encoding of these memories, thereby influencing subsequent memory strength and precision. Here, we found that pre-exposure to the neutral IA chamber decreased memory precision, whereas pre-exposure to the aversive IA chamber promoted memory strength and precision. Additionally, females demonstrated overall enhanced memory precision compared to males. These results indicate that different types of contextual exposures influence initial IA encoding and add to a limited body of research examining memory strength and precision in IA in both sexes.

Forming a long-term memory requires changes in neuronal transcription. What happens, though, as the memory is forgotten? And how does the transcriptional state relate to the maintenance and recall of the long-term memory? To answer these questions we have been systematically tracing the time-course of transcriptional changes evoked by long-term sensitization in the marine mollusk Aplysia californica. Our approach captures transcriptional changes in neurons of known behavioral relevance using a within-subjects design, delineating patterns of transcriptional change that are comprehensive and reproducible. We have previously reported that within 1 day of long-term sensitization training there is a widespread transcriptional response involving robust changes in over 5% of tested transcripts (1,252 of [~]22k; Conte, 2017). Within 1 week, however, memory strength fades and nearly all transcriptional changes relapse to baseline (Perez, 2018). Here we report microarray analysis (N = 16) of transcriptional changes 5 days post-learning, a time-point when memory strength has weakened but is still robust. Remarkably, we find that at this intermediate behavioral stage nearly all transcriptional changes have fully decayed, even in subsets of animals that have shown very little forgetting. Thus, most transcriptional changes seem to decay more rapidly than memory expression. We discuss several possible ways that memory expression could become decoupled from detectable transcriptional regulation. HighlightsO_LILong-term sensitization training produces a memory that then fades over the course of a week, with behavioral expression at 5 days at an intermediate stage of partly forgotten with continued clear sensitization and considerable variety across animals. C_LIO_LIThe transcriptional response to sensitization training fades more quickly than behavioral expression, with nearly all transcripts regulated 1 day after training showing a statistically significant decline in regulation, even amongst animals that had shown little forgetting. C_LIO_LITranscription does not seem to have a straightforward relationship with the expression of sensitization memory, with a small set of transcripts consistently regulated even as behavioral expression changes and strong behavioral expression possible without most of the transcriptional changes observed during early maintenance. C_LI

The hippocampus plays a critical role in storing and retrieving spatial information. By targeting the dorsal hippocampus and manipulating specific "candidate" molecules using pharmacological and genetic manipulations, we have previously discovered that long-term active place avoidance memory requires transient activation of particular molecules in dorsal hippocampus. These molecules include amongst others, the persistent kinases Ca-calmodulin kinase II (CaMKII) and the atypical protein kinase C isoform PKC{iota} /{lambda} for acquisition of the conditioned behavior, whereas persistent activation of the other atypical PKC, protein kinase M zeta (PKM{zeta}) is necessary for maintaining the memory for at least a month. It nonetheless remains unclear what other molecules and their interactions maintain active place avoidance long-term memory, and the candidate molecule approach is both impractical and inadequate to identify new candidates since there are so many to survey. Here we use a complementary approach to identify candidates by transcriptional profiling of hippocampus subregions after formation of the long-term active place avoidance memory. Interestingly, 24-h after conditioning and soon after expressing memory retention, immediate early genes were upregulated in the dentate gyrus but not Ammons horn of the memory expressing group. In addition to determining what genes are differentially regulated during memory maintenance, we performed an integrative, unbiased survey of the genes with expression levels that covary with behavioral measures of active place avoidance memory persistence. Gene Ontology analysis of the most differentially expressed genes shows that active place avoidance memory is associated with activation of transcription and synaptic differentiation in dentate gyrus but not CA3 or CA1, whereas hypothesis-driven candidate molecule analyses identified insignificant changes in the expression of many LTP-associated molecules in the various hippocampal subfields, nor did they covary with active place avoidance memory expression, ruling out strong transcriptional regulation but not translational regulation, which was not investigated. These findings and the data set establish an unbiased resource to screen for molecules and evaluate hypotheses for the molecular components of a hippocampus-dependent, long-term active place avoidance memory.

Memories are often categorized into types, reflecting their behavioral, anatomical and molecular diversity: these classifications both aid understanding of the differences among varieties of memory and help delineate the unifying cross-species principles underlying them. In the nematode worm Caenorhabditis elegans, we find that an associative memory of the pairing of the normally attractive odorant benzaldehyde and starvation depends on de novo translation, is independent of CREB, and is produced by massed training: a pattern which does not correspond to any of the well-characterized molecular categories of invertebrate memory. Further, as has been shown for many memories in vertebrates, but not previously in nematodes, we find that formation of this memory continues after removal of the stimuli initially causing it, and that it is labile to disruption through protein synthesis inhibition following training, but that inhibition of proteasomal activity does not extend the duration of the memory. Previous findings have implicated insulin pathway signaling, which canonically regulates the transcription factor DAF- 16, as a key component of this benzaldehyde/starvation memory, however our results suggest that transcriptional inhibition has, at most, only moderate effects on memory formation. We find that insulin signaling instead acts to regulate phospholipase C, which in turn regulates memory through diacylglycerol signaling. These findings better characterize this model associative memory in relation to other invertebrate memory types and identify ways in which it both shares their traits and differs from them, as well as revealing a more complete picture of the molecular pathway underlying it.

The progress of learning is usually quantified by averaging responses across participants and/or multiple trials within a block. However, such approaches obscure the trial-by-trial progress of learning, which has been shown recently to express a rich variety of dynamics. An alternative approach which does not suffer from this problem is the detection and analysis of points of behavioral change, i.e., change-point analysis. Using change-point analysis, we reanalyzed data from human participants in different predictive learning tasks in which learned contingencies underwent reversal. We find that responses of individual participants were more accurately characterized by behavioral change points than the average learning curve. Importantly, change points significantly shifted to later trials during reversal learning indicating that reversal learning is more difficult than the initial learning. In a computational model based on deep reinforcement learning, we show that the change point shift required the replay of previous experiences, which in turn depends on the hippocampus. This finding is consistent with studies showing that lesions of the hippocampus yield faster reversal learning. In summary, we reaffirm the importance of the analysis of single participant responses, show that phenomenological learning rates are slower during reversal learning, and provide a theoretical account for this difference.

Although many studies have addressed the role of the amygdala in modulating long-term memory, it is not known whether weak training plus amygdala stimulation can transform a short-term memory into a remote memory. Object place recognition (OPR) memory after strong training remains hippocampus-dependent through the persistent action of PKM{zeta} for at least 6 days, but it is unknown whether weak training plus amygdala stimulation can transform short-term memory into an even longer memory, and whether such memory is stored through more persistent action of PKM{zeta} in hippocampus. We trained rats to acquire OPR and 15 min or 5 h later induced a brief pattern of electrical stimulation in basolateral amygdala (BLA). Our results reveal that a short-term memory lasting < 4 h can be converted into remote memory lasting at least 3 weeks if the BLA is activated 15 min, but not 5 h after learning. To examine how this remote memory is maintained, we injected ZIP, an inhibitor of atypical PKCs (aPKCs), PKM{zeta} and PKC{iota}/{lambda}, into either hippocampal CA1, dentate gyrus (DG), or anterior cingulate cortex (ACC). Our data reveal amygdala stimulation produces consolidation into remote memory, not by persistent aPKC activation and capture by synaptic tagging processes in the hippocampal formation, but in ACC. Our data establish a powerful modulating role of the BLA in forming remote memory and open a path in the search for neurological restoration of memory, based on enhancing synaptic plasticity in aging or neurodegenerative disorders such as Alzheimers disease.

The acquisition of temporally proximate information can impair the brains ability to consolidate earlier experiences, resulting in retroactive interference (RI). Recognition-based behavioral paradigms are well-suited for investigating RI in rodents, particularly those involving sequential learning episodes. The medial prefrontal cortex (mPFC) integrates multimodal information relevant to the regulation of memory interference and is strongly modulated by the serotonergic system. Serotonin 2A receptors (5-HT2AR), which are densely expressed in the mPFC, have been shown to influence the retrieval of competing object-recognition memories. However, their role in other phases of memory processing, particularly in modulating RI, remains unclear. Using a modified version of the object recognition task designed to induce RI, combined with pharmacological manipulation of 5-HT2AR, we demonstrate that RI specifically impairs the object-related component of memory. Moreover, serotonin signaling through 5-HT2AR is necessary to prevent RI. Strikingly, activating 5-HT2AR before retrieval can rescue the expression of memories affected by RI, suggesting that RI may not erase memory traces but rather hinder their access.

We acquire perceptual skills through experience to adapt ourself to the changing environment. Accomplishing an effective skill acquisition is a main purpose of perceptual learning research. Given the often observed learning effect specificity, multiple perceptual learnings with shared parameters could serve to improve the generalization of the learning effect. However, the interference between the overlapping memory traces of different learnings may impede this effort. Here, we trained human participants on an orientation discrimination task. We observed a proactive interference effect that the first training blocked the second training at its untrained location. This was a more pronounced effect than the well-known location specificity in perceptual learning. We introduced a short reactivation of the first training before the second training and successfully eliminated the proactive interference when the second training was inside the reconsolidation time window of the reactivated first training. Interestingly, we found that practicing an irrelevant task at the location of the second training immediately after the reactivation of the first training could also restore the effect of the second training but in a smaller magnitude, even if the second training was conducted outside of the reconsolidation window. We proposed a two-level mechanism of reactivation-induced memory integration to account for these results. The reactivation-based procedure could integrate either the previously trained and untrained locations or the two trainings at these locations, depending on the activated representations during the reconsolidation process. The findings provide us with new insight into the roles of long-term memory mechanisms in perceptual learning.

Learning from aversive experiences often generalizes beyond the context in which they occurred. In rodents, a strong aversive event can induce retrospective memory linking (RLI), whereby fear generalizes to a previously neutral context encountered days earlier. Although prior work has shown that RLI is associated with increased co-activity of hippocampal CA1 neurons across neutral and aversive contexts, it remains unclear how broader representational changes support generalization without affecting the ability to discriminate between contexts. Here, we reanalyzed calcium imaging data from dorsal CA1 during RLI to examine how hippocampal representational geometry changes during fear generalization. Using robust, non-parametric measures of population similarity, we show that in mice exhibiting RLI, the representation of the neutral context not only changes over time but becomes more similar to the aversive context during recall. Beyond this similarity increase, we provide evidence for a higher-dimensional geometric transformation consistent with a shared "fear" operation that can be applied across contexts while preserving their identity. Crucially, these two representational signatures dissociate by behavioral state: similarity to the aversive context emerges during freezing, whereas a shared transformation is expressed during active exploration. Together, these findings demonstrate that hippocampal representations support retrospective fear generalization through state-dependent geometric transformations, highlighting representational geometry as a key computational mechanism to resolve the apparent tension between generalization and discriminability.

Contextual fear conditioning is a classical laboratory task that tests associative memory formation and recall. Techniques such as multi-photon microscopy and holographic stimulation offer tremendous opportunities to understand the neural underpinnings of these memories. However, these techniques generally require animals to be head-fixed. There are few paradigms that test contextual fear conditioning in head-fixed mice, and none where the behavioral outcome following fear conditioning is freezing, the most common measure of fear in freely moving animals. To address this gap, we developed a contextual fear conditioning paradigm in head-fixed mice using virtual reality (VR) environments. We designed an apparatus to deliver tail shocks (unconditioned stimulus, US) while mice navigated a VR environment (conditioned stimulus, CS). The acquisition of contextual fear was tested when the mice were reintroduced to the shock-paired VR environment the following day. We tested three different versions of this paradigm and, in all of them, observed an increased conditioned fear response characterized by increased freezing behavior. This was especially prominent during the first trial in the shock-paired VR environment, compared to a neutral environment where the mice received no shocks. Our results demonstrate that head-fixed mice can be fear conditioned in VR, discriminate between a feared and neutral VR context, and display freezing as a conditioned response, similar to freely behaving animals. Furthermore, using a two-photon microscope, we imaged from large populations of hippocampal CA1 neurons before, during, and following contextual fear conditioning. Our findings reconfirmed those from the literature on freely moving animals, showing that CA1 place cells undergo remapping and show narrower place fields following fear conditioning. Our approach offers new opportunities to study the neural mechanisms underlying the formation, recall, and extinction of contextual fear memories. As the head-fixed preparation is compatible with multi-photon microscopy and holographic stimulation, it enables long-term tracking and manipulation of cells throughout distinct memory stages and provides subcellular resolution for investigating axonal, dendritic, and synaptic dynamics in real-time.

The anticipation of extrinsic reward facilitates memory formation. However, it is unclear how reward following memory retrieval influences the information that is retrieved and later remembered. Here, we conducted four behavioral experiments (N=42 male/female young adults per experiment) in which we manipulated retrieval practice reward delivery. Across all experiments, participants studied word-image pairs and then completed two rounds of retrieval practice, followed by a final recognition test. Participants made vividness judgments during retrieval practice and in three of four experiments each response had a 50% chance of yielding positive feedback. We find that repeated rewards following retrieval practice facilitate later memory whereas low vivid retrieval practice impairs later memory. Together, these results suggest that the benefit of both retrieval practice and reward may be dependent on the strength of the memory that is retrieved.

New information is rarely learned in isolation, instead most of what we experience can be incorporated into or uses previous knowledge networks in some form. However, most rodent laboratory tasks assume the animal to be naive with no previous experience influencing the results. Previous knowledge in form of a schema can facilitate knowledge acquisition and accelerate systems consolidation: memories become more rapidly hippocampal independent and instead rely more on the prefrontal cortex. Here, we developed a new spatial navigation task where food locations are learned in a large, gangway maze - the HexMaze. Analysing performance across sessions as well as on specific trials, we can show simple memory effects as well as multiple effects of previous knowledge accelerating both online learning and performance increases over offline periods. Importantly, we are the first to show that schema build-up is dependent on how much time passes, not how often the animal is trained.

Stress and emotional arousal interfere with encoding of temporal context memory for episodic events. However, it remains unclear how stress affects more fine-grained temporal memory, such as episodic events sequences and event times. Here, 86 healthy participants (M age = 22.5; 46% women, 54% men) were subjected to either a stress condition (socially evaluated cold pressor test) or a control condition, directly after or at a delay of 30 minutes they were presented the temporal structure of four virtual days. In these virtual days, time was scaled and participants could use clock cues to construe the passage of time within a day. We examined whether acute stress would interfere with encoding of episodic event sequences and temporal memory. Our results show that when learning took place directly after a stressor, virtual time estimates were more strongly biased towards a generalized timeline but temporal memory overall was not differentially affected between the stress and control groups. Exploratory analyses suggest that memory accuracy improved in men and deteriorated in women as a function of subjective stress levels following acute stress. In conclusion, acute stress amplified memory generalization but we found no stress related differences in memory accuracy across levels of temporal granularity.

Traumatic stress can cause long-lasting changes in cognition and affect, sometimes leading to diagnoses such as post-traumatic stress disorder (PTSD). The stress-enhanced fear learning (SEFL) model recapitulates understudied components of PTSD, such as stress-induced sensitization of fear learning. The SEFL procedure entails exposing mice to footshock stress followed later by fear conditioning in a different context. When tested later for recall of fear conditioning, previously stressed mice exhibit enhanced freezing compared to non-stressed controls. Studies have shown that dorsal and ventral dentate gyrus (DG) generates neural ensemble representations of contextual fear, such that fear recall involves reactivation of a sparse set of "engram cells" that were active during fear memory acquisition. How stress affects these hippocampal ensemble representations is unknown. We used SEFL and activity-dependent neuronal tagging with FosTRAP2 mice to investigate effects of stress on fear memory ensembles in rostral and caudal hippocampal DG. FosTRAP2/Ai6 mice received footshock stress or equivalent context exposure without shock in Context A on day 1. Five days later, mice received 1-shock conditioning in Context B and immediately received an injection of 4-OHT (55mg/kg) to tag fear acquisition neurons with the zsGreen reporter. One day later, mice were tested for fear recall in Context B and were perfused 90 minutes after testing. Confirming prior studies, prior stress potentiated 1-shock conditioning in Context B, with stressed mice displaying higher freezing in the Context B test session than non-stressed mice. At the level of neural activity, results showed stress had no effect on the number of zsGreen+ fear ensemble cells or the number of cfos+ recall-activated cells in rostral or caudal DG. However, stress increased reactivation (percentage of zsGreen+ cells expressing cfos) in the caudal but not rostral DG. The results suggest stress potentiates later fear learning by enhancing fear representations in caudal hippocampus, a region of the hippocampus specialized for integrating emotional and motivational valence into memory.

Head-fixed linear treadmill tasks have been used to study hippocampal physiology in mice. Although some hippocampal neurons establish place fields along linear treadmills, it is not clear if the hippocampus is required for spatial memory on this task. Using a Designer Receptors Exclusively Activated by Designer Drugs (DREADDs) approach, we found that silencing hippocampal output on rewarded treadmill tasks impaired search for rewards signaled by spatial cues but did not impair search for rewards signaled by local cues, recapitulating findings from other behavior tasks. These findings serve to contextualize data on hippocampal physiology from mice performing this task.

In extinction learning, contextual renewal occurs when an extinguished behavior reemerges after a context change. A key question is how stimuli become integrated as contextual cues. While contingency, the predictive relationship between stimuli and outcomes; is known to be important, its precise contribution remains unclear. Using a ABA renewal design with pigeons in operant chambers, we systematically violated contingency by probabilistically reinforcing responses during extinction. Our results show that partial violations of contingency modulated extinction learning but did not abolish contextual renewal. Instead, pigeons developed meta-learning strategies, adapting their behavior across sessions to optimize reward despite extinction conditions. These findings highlight that context formation is sensitive to contingency levels, but also that animals can flexibly reorganize their learning strategies when contingency is unstable.

While forgetting has been studied extensively in various organisms, its precise nature has often been unclear. Here, we used behavioral experiments in Drosophila to determine that a significant aspect of forgetting consists of a decrease in the ability of a memory to induce an appropriate behavior. We tested flies for memory retention at various times after training and then separately retested both flies that chose correctly and those that chose incorrectly. Although the ability to choose correctly decreased over time, we could not measure any differences in memory between flies that initially chose correctly and those that chose incorrectly upon retest. This suggests that forgetting is unlikely to consist of a spontaneous loss of a memory but instead consists of a decrease in the probability of flies that remember choosing the correct behavioral response. Thus, although flies maintain memory over time, there is an increase in uncertainty associated with this memory. We find that forgetting of long-term memories and accelerated forgetting in old flies occur in a similar manner.

Mammalian brain has evolved to infer from past experiences and elicit context relevant novel behavioural responses hitherto unexpressed by the animal. However, little is known about how prior knowledge influences the emergence of such responses. Remarkably, the brain not only arrives at these responses through logical inferences based on previous leanings, but also acquire new related information, without causing catastrophic interference. Mental schemas have often been proposed as the framework for this phenomenon. In this study, using mice as a model animal, we show that schematic networks not only enhance the cognitive load handling capacity (CLHC) and prevent catastrophic interference, but also facilitate the generation of novel, contextually relevant responses. Interestingly, when the animals were trained in a paradigm that did not invoke the pre-formed mental schema, we observed neither an enhancement to CLHC nor a generation of novel context relevant responses. Based on the principles of mental schemas discovered in our animal experiments, we developed a biologically plausible artificial neural network (ANN) that avoids catastrophic interference and captures the learning properties observed in our experiments. The custom architecture of this ANN enables it to generate responses similar to those of animals in novel scenarios. Significance StatementLittle is known about the role of mental schemas in preventing memory interference--a process in which overlapping or similar memories hinder the acquisition and retention of related information. In this study, we demonstrate that mental schemas enhance cognitive load handling capacity and improve the ability to solve novel but related problems. Using mice as a model, we show that the mere existence of a mental schema is not enough for improved cognitive load handling; instead, the relationship between existing and new information must be explicitly established during the learning process. Based on these findings, we developed a minimalistic artificial neural network (ANN) that effectively mimics this behaviour. These insights pave the way for developing more efficient learning and teaching strategies.