The theta rhythm is important for many cognitive functions including spatial processing, memory encoding, and memory recall. The information processing underlying these functions is thought to rely on consistent, phase-specific spiking throughout a theta oscillation that may fluctuate significantly in baseline (center of oscillations), frequency, or amplitude. Experimental evidence shows that spikes can occur at specific phases even when the baseline membrane potential varies significantly, such that the integrity of phase-locking persists across a large variability in spike threshold. The mechanism of this precise spike timing during the theta rhythm is not yet known and previous mathematical models have not reflected the large variability in threshold potential seen experimentally. Here we introduce a straightforward mathematical neural model capable of demonstrating a phase-locked spiking in the face of significant baseline membrane potential fluctuation during theta rhythm. This novel approach incorporates a degenerate grazing bifurcation of an asymptotically stable oscillation. This model suggests a potential mechanism for how biological neurons can consistently produce spikes near the peak of a variable membrane potential oscillation.

Spatial working memory enables behavioral flexibility and depends upon the hippocampus. The hippocampal dentate gyrus (DG) is hypothesized to act as a pattern separator to disambiguate similar memories. We used activity-dependent labeling strategies in DG to tag multiple memory-associated neuronal ensembles recruited during the course of learning a spatial working memory task. We termed the tagged ensembles "nonspecific" because they arose from experiences across days, maze arenas, and the home cage. Optogenetic stimulation of nonspecific ensembles during the encoding phase within random trials of the task acutely impaired performance during light-on trials only. The degree of reactivation in the dentate hilus, but not the blade, inversely correlated with task accuracy. While our results do not rule out the possibility that the stimulation of random, sparse DG ensembles could lead to similar impairment, they are consistent with the proposed role of DG in pattern separation, completion, and spatial working memory.

Spatial disorientation is emerging as an early cognitive biomarker of dementia, but the underlying neural mechanisms remain undefined. The anterodorsal nucleus of the thalamus (ADn) exhibits early and selective vulnerability to pathological misfolded forms of tau, a major hallmark of Alzheimer’s disease. As the ADn contains a high density of head direction (HD) cells, we hypothesized that ptau disrupts HD cell activity, promoting spatial disorientation. To test this, we virally expressed human mutant tau in the ADn of adult wild-type and C1ql2-Cre mice. ADn-tau mice were defined by phosphorylated and oligomeric forms of tau in ADn somata and axon terminals. During initial spatial learning, ADn-tau mice exhibited increased looping behavior, indicative of spatial disorientation. Using in vivo extracellular recordings, we found that ADn cells had reduced directionality, lower directional coherence, and altered burst firing. These findings suggest that early alterations in HD signaling are predictive of future cognitive decline.

INTRODUCTION:Previous simulations of Hebbian associative memory models inspired the malignant synaptic growth hypothesis of Alzheimer's disease (AD), which suggests that cognitive impairments arise due to runaway synaptic modification resulting from poor separation between encoding and retrieval. METHODS:We computationally model presynaptic inhibition by the recently identified interaction of soluble amyloid precursor protein alpha (sAPPα) with γ-aminobutyric acid type B receptor (GABABR) as one potential biological mechanism that can enhance separation between encoding and retrieval. RESULTS:Simulations predict that the dual effect of sAPPα on long-term potentiation and presynaptic inhibition of glutamatergic synapses maintains effective associative memory function and prevents runaway synaptic modification. Moreover, computational modeling predicts that sAPPα, which interacts with the 1a isoform of GABABR, is more effective than the GABABR agonist baclofen at stabilizing associative memory. DISCUSSION:Molecular mechanisms that enhance presynaptic inhibition, such as sAPPα-GABABR1a signaling, are potential therapeutic targets for preventing cognitive impairments in AD. HIGHLIGHTS:Computational modeling of Hebbian associative memory provides a framework for understanding the functional basis of Alzheimer's disease. Soluble amyloid precursor protein (sAPPα) presynaptic activation of γ-aminobutyric acid B (GABAB) receptors prevents runaway synaptic modification in associative memory models. sAPPα is more effective than baclofen at stabilizing associative memory.

A new model of how we plan pathways through the world shows that populations of neurons that code our current position, including entorhinal grid cells and head direction cells, could interact to aid planning by scanning alternating forward trajectories through the environment.

In keeping with the historical focus of this special issue of Hippocampus, this paper reviews the history of my development of the SPEAR model. The SPEAR model proposes that separate phases of encoding and retrieval (SPEAR) allow effective storage of multiple overlapping associative memories in the hippocampal formation and other cortical structures. The separate phases for encoding and retrieval are proposed to occur within different phases of theta rhythm with a cycle time on the order of 125 ms. The same framework applies to the slower transition between encoding and consolidation dynamics regulated by acetylcholine. The review includes description of the experimental data on acetylcholine and theta rhythm that motivated this model, the realization that existing associative memory models require these different dynamics, and the subsequent experimental data supporting these dynamics. The review also includes discussion of my work on the encoding of episodic memories as spatiotemporal trajectories, and some personal description of the episodic memories from my own spatiotemporal trajectory as I worked on this model.

A hallmark of human intelligence is the ability to infer abstract rules from limited experience and apply these rules to unfamiliar situations. This capacity is widely studied in the visual domain using the Raven's Progressive Matrices. Recent advances in deep learning have led to multiple artificial neural network models matching or even surpassing human performance. However, while humans can identify and express the rule underlying these tasks with little to no exposure, contemporary neural networks often rely on massive pattern-based training and cannot express or extrapolate the rule inferred from the task. Furthermore, most Raven's Progressive Matrices or Raven-like tasks used to train neural networks consist only of symbolic challenges, whereas humans can flexibly solve both symbolic and perceptual challenges. In this work, we present an algorithmic approach to rule detection and application using feature detection, affine transformation estimation and search. We applied our model to a simplified Raven's Progressives Matrices task, previously designed for behavioral testing and neuroimaging in humans. The model exhibited one-shot inference and achieved near human-level performance in the symbolic reasoning condition of the simplified task. Furthermore, the model can express the relationships discovered and generate multi-step predictions in accordance with the underlying rule. Finally, the model can handle perceptual challenges containing continuous patterns. We discuss our results and their relevance to studying abstract reasoning in humans, as well as their implications for improving intelligent machines.

Numerous scientific advances and discoveries have arisen from research on the hippocampal formation. This special issue provides first-person historical descriptions of these advances and discoveries in hippocampal research, written by those directly involved in the research. This is the first section of a special issue that will also include future articles on this topic. Here, we discuss some of the factors that motivated this special issue, and the major themes of hippocampal research that are addressed.

Neurons in the retrosplenial (RSC) (Alexander et al 2020a, LaChance & Hasselmo 2024) and postrhinal cortex (POR) respond to environmental boundaries and configurations using egocentric coordinates relative to an animal's current position. Neurons in these structures and adjacent structures also respond to spatial dimensions of self-motion such as running velocity (Carstensen et al 2021, Robinson et al 2023). Experimental and modeling data suggest that these responses could be essential for guiding behaviors such as obstacle avoidance and goal-directed navigation (Erdem & Hasselmo 2012, Erdem & Hasselmo 2014). However, these findings still leave the unanswered question: What specific features, and in what coordinate frames, drive these egocentric neural responses? Here we present models of the potential circuit mechanisms generating egocentric responses in RSC. One model posits that neurons encode internal representations of barriers in head-centered coordinates, defined by distance and angle, which are modulated by running velocity to enable trajectory planning and obstacle avoidance. We contrast this with a complementary hypothesis in which neurons respond to retinotopic features-such as the top, bottom, or edges of walls, which may serve as precursors to head-centered representations. Additional hypotheses include trajectory-based forward scanning (e.g., ray tracing) for barrier detection or comparing optic flow across the visual field. These hypotheses generate complementary modeling predictions about how changes in environmental parameters could alter the neural responses of egocentric boundary cells that are presented here.

Animals rely on both sensory perception and memory when navigating relative to learned allocentric locations. Incoming sensory stimuli, which arrive from an egocentric perspective, must be integrated into an allocentric reference frame to allow neural computations that direct an animal toward a learned goal. This egocentric-allocentric spatial transformation has been proposed to involve projections from the rodent postrhinal cortex (POR), which receives strong visual input, to the medial entorhinal cortex (MEC), which contains allocentric spatial cell types such as grid and border cells. A major step toward understanding this transformation is to identify how POR and MEC spatial representations differ during place navigation, which is currently unknown. To answer this question, we recorded single neurons from POR and MEC as rats engaged in a navigation task that required them to repeatedly visit a learned uncued allocentric location in an open field arena to receive a randomly scattered food reward. While neurons in both regions displayed strong tuning to the spatial structure of the environment, neither showed bias toward the goal location despite strongly biased behavior. Critically, when local visual landmarks were manipulated to place the visual scene in conflict with the learned location, POR neurons adjusted their tuning preferences to follow the visual landmarks, while MEC neurons remained in register with the true global reference frame. These findings reveal a strong dissociation between POR and MEC spatial reference frames during place navigation and raise questions regarding the mechanisms underlying integration of POR egocentric signals into the MEC allocentric spatial map.

Recent experimental studies have discovered diverse spatial properties, such as head direction tuning and egocentric tuning, of neurons in the postrhinal cortex (POR) and revealed how the POR spatial representation is distinct from the retrosplenial cortex (RSC). However, how these spatial properties of POR neurons emerge is unknown, and the cause of distinct cortical spatial representations is also unclear. Here, we build a learning model of POR based on the pathway from the superior colliculus (SC) that has been shown to have motion processing within the visual input. Our designed SC-POR model demonstrates that diverse spatial properties of POR neurons can emerge from a learning process based on visual input that incorporates motion processing. Moreover, combining SC-POR model with our previously proposed V1-RSC model, we show that distinct cortical spatial representations in POR and RSC can be learnt along disparate visual pathways (originating in SC and V1), suggesting that the varying features encoded in different visual pathways contribute to the distinct spatial properties in downstream cortical areas.

Complex sensory information arrives in the brain from an animal's first-person ('egocentric') perspective. However, animals can efficiently navigate as if referencing map-like ('allocentric') representations. The postrhinal (POR) and retrosplenial (RSC) cortices are thought to mediate between sensory input and internal maps, combining egocentric representations of physical cues with allocentric head direction (HD) information. Here we show that neurons in the POR and RSC of female Long-Evans rats are tuned to distinct but complementary aspects of local space. Egocentric bearing (EB) cells recorded in square and L-shaped environments reveal that RSC cells encode local geometric features, while POR cells encode a more global account of boundary geometry. Additionally, POR HD cells can incorporate egocentric information to fire in two opposite directions with two oppositely placed identical visual landmarks, while only a subset of RSC HD cells possess this property. Entorhinal grid and HD cells exhibit consistently allocentric spatial firing properties. These results reveal significant regional differences in the neural encoding of spatial reference frames. Whether and how the postrhinal (POR) and retrosplenial (RSC) cortices interact with each other and impact downstream allocentric representations are not fully understood. Here authors present single neuron recordings from freely moving rats exploring different environments to reveal distinct egocentric (self-centered) and allocentric (world-centered) coding frameworks for landmarks and boundaries in interconnected cortical regions.

We develop a framework to learn bio-inspired foraging policies using human data. We conduct an experiment where humans are virtually immersed in an open field foraging environment and are trained to collect the highest amount of rewards. A Markov Decision Process (MDP) framework is introduced to model the human decision dynamics. Then, Imitation Learning (IL) based on maximum likelihood estimation is used to train Neural Networks (NN) that map human decisions to observed states. The results show that passive imitation substantially underperforms humans. We further refine the human-inspired policies via Reinforcement Learning (RL) using the on-policy Proximal Policy Optimization (PPO) algorithm which shows better stability than other algorithms and can steadily improve the policies pre-trained with IL. We show that the combination of IL and RL match human performance and that the artificial agents trained with our approach can quickly adapt to reward distribution shift. We finally show that good performance and robustness to reward distribution shift strongly depend on combining allocentric information with an egocentric representation of the environment.

Spatial navigation involves the formation of coherent representations of a map-like space, while simultaneously tracking current location in a primarily unsupervised manner. Despite a plethora of neurophysiological experiments revealing spatially-tuned neurons across the mammalian neocortex and subcortical structures, it remains unclear how such representations are acquired in the absence of explicit allocentric targets. Drawing upon the concept of predictive learning, we utilize a biologically plausible learning rule which utilizes sensory-driven observations with internally-driven expectations and learns through a contrastive manner to better predict sensory information. The local and online nature of this approach is ideal for deployment to neuromorphic hardware for edge-applications. We implement this learning rule in a network with the feedforward and feedback pathways known to be necessary for spatial navigation. After training, we find that the receptive fields of the modeled units resemble experimental findings, with allocentric and egocentric representations in the expected order along processing streams. These findings illustrate how a local and self-supervised learning method for predicting sensory information can extract latent structure from the environment.

Ascending neuromodulatory projections from deep brain nuclei generate internal brain states that differentially engage specific neuronal cell types. Because neurovascular coupling is cell-type specific and neuromodulatory transmitters have vasoactive properties, we hypothesized that the impulse response function (IRF) linking spontaneous neuronal activity with hemodynamics would depend on neuromodulation. To test this hypothesis, we used optical imaging to measure (1) release of neuromodulatory transmitters norepinephrine (NE) or acetylcholine (ACh), (2) Ca 2+ activity of local cortical neurons, and (3) changes in hemoglobin concentration and oxygenation across the dorsal surface of cerebral cortex during spontaneous neuronal activity in awake mice. A canonical convolution model with a stationary IRF (e.g., the convolution kernel) describing evolution of total hemoglobin (HbT, reflective of dilation dynamics) with respect to Ca 2+ , resulted in a poor fit to the data. However, the HbT time-course was well predicted, pixel-by-pixel, by a weighted sum of Ca 2+ and NE time-courses. The weighting coefficients, calculated using linear regression, varied smoothly across the cortical space. Consistent with this result, modeling HbT as a weighted sum of stationary Ca 2+ - and NE-specific IRFs convolved with the respective time-courses dramatically improved the fit compared to the invariant IRF. In both the linear regression and the Double-IRF convolution models, Ca 2+ and NE weighting was positive and negative, respectively. In contrast to NE, ACh was largely redundant with Ca 2+ and therefore did not improve HbT estimation. Because NE covaried with arousal, we observed instances of the diminished hemodynamic coherence between cortical regions during high arousal despite coherent behavior of the underlying neuronal Ca 2+ activity. We conclude that while neurovascular coupling with respect to neuronal Ca 2+ is a dynamic and seemingly complex phenomenon, hemodynamic fluctuations can be captured by a simple linear model with stationary IRFs with respect to the underlying dilatory and constrictive forces. In the current study, these forces were captured by the positive Ca 2+ (dilation) and negative NE (constriction) coefficients. Without accounting for NE neuromodulation and the associated vasoconstriction, diminished hemodynamic coherence, commonly referred to as ″functional (dys)connectivity″ in BOLD fMRI studies, can be falsely interpreted as neuronal desynchronizations.

The hippocampus and medial entorhinal cortex (MEC) form a cognitive map that facilitates spatial navigation. As part of this map, MEC grid cells fire in a repeating hexagonal pattern across an environment. This grid pattern relies on inputs from the medial septum (MS). The MS, and specifically GABAergic neurons, are essential for theta rhythm oscillations in the entorhinal-hippocampal network; however, the role of this population in grid cell function is unclear. To investigate this, we use optogenetics to inhibit MS-GABAergic neurons and observe that MS-GABAergic inhibition disrupts grid cell spatial periodicity. Grid cell spatial periodicity is disrupted during both optogenetic inhibition periods and short inter-stimulus intervals. In contrast, longer inter-stimulus intervals allow for the recovery of grid cell spatial firing. In addition, grid cell phase precession is also disrupted. These findings highlight the critical role of MS-GABAergic neurons in maintaining grid cell spatial and temporal coding in the MEC.

The hippocampus is a higher-order brain structure responsible for encoding new episodic memories and predicting future outcomes. In the absence of external stimuli, neurons in the hippocampus track elapsed time, distance traveled, and other idiothetic variables. To this day, the exact determinants of idiothetic representations during free navigation remain unclear. Here, we developed unsupervised approaches to extract population and single-cell properties of more than 30,000 CA1 pyramidal neurons in freely moving mice. We find that spatiotemporal representations are composed of a mixture of idiothetic and allocentric information, the balance of which is dictated by task demand and environmental conditions. Additionally, a subset of CA1 pyramidal neurons encodes the spatiotemporal distance to rewards. Finally, distance and time information is integrated postsynaptically in the lateral septum, indicating that these high-level representations are effectively integrated in downstream neurons.

Humans and other animals are able to quickly generalize latent dynamics of spatiotemporal sequences, often from a minimal number of previous experiences. Additionally, internal representations of external stimuli must remain stable, even in the presence of sensory noise, in order to be useful for informing behavior. In contrast, typical machine learning approaches require many thousands of samples, and generalize poorly to unexperienced examples, or fail completely to predict at long timescales. Here, we propose a novel neural network module which incorporates hierarchy and recurrent feedback terms, constituting a simplified model of neocortical microcircuits. This microcircuit predicts spatiotemporal trajectories at the input layer using a temporal error minimization algorithm. We show that this module is able to predict with higher accuracy into the future compared to traditional models. Investigating this model we find that successive predictive models learn representations which are increasingly removed from the raw sensory space, namely as successive temporal derivatives of the positional information. Next, we introduce a spiking neural network model which implements the rate-model through the use of a recently proposed biological learning rule utilizing dual-compartment neurons. We show that this network performs well on the same tasks as the mean-field models, by developing intrinsic dynamics that follow the dynamics of the external stimulus, while coordinating transmission of higher-order dynamics. Taken as a whole, these findings suggest that hierarchical temporal abstraction of sequences, rather than feed-forward reconstruction, may be responsible for the ability of neural systems to quickly adapt to novel situations.

The hippocampus is a higher-order brain structure responsible for encoding new episodic memories and predicting future outcomes. In absence of external stimuli, neurons in the hippocampus express sequential activities which have been proposed to support path integration by tracking elapsed time, distance traveled, and other idiothetic variables. On the other hand, with sufficient external sensory inputs, hippocampal neurons can fire with respect to allocentric cues. Previously, these idiothetic codes have been described in conditions where running speed is clamped experimentally. To this day, the balance of idiothetic and allocentric representations in freely moving conditions remains unclear. Additionally, whether CA1 and CA3 temporal and distance codes are transmitted downstream to the lateral septum has not been established. Here, we develop an unsupervised model trained to compress neural information with minimal loss, and find that we can efficiently decode elapsed time and distance travelled from low-dimensional embeddings of neural activity in freely moving mice. We also developed unbiased information metrics that are minimally sensitive to quantization parameters and enable comparisons across modalities and brain regions. In more than 30,000 CA1 pyramidal neurons, we show that spatiotemporal information is represented as a mixture of self-motion, idiothetic as well as allocentric information, the balance of which is dictated by task demand and environmental conditions. In particular, we find that a subset of CA1 pyramidal neurons encode the spatiotemporal distance relative to rewards. Single cell and population statistics across the hippocampal-septal circuit reveal that idiothetic variables emerge in CA1 and are integrated postsynaptically in the lateral septum. Finally, we implement a computational model trained to replicate real world neural activity, and find that grid cells could provide a plausible input for CA1 representations of time and distance. Altogether, our results suggest that hippocampal CA1 continuously integrates both idiothetic and allocentric signals depending on task demand and available cues, and these high-level representations are effectively transmitted to downstream regions.

Analysis of neuronal activity in the hippocampus of behaving animals has revealed cells acting as ‘Time Cells’, which exhibit selective spiking patterns at specific time intervals since a triggering event, and ‘Distance Cells’, which encode the traversal of specific distances. Other neurons exhibit a combination of these features, alongside place selectivity. This study aims to investigate how the task performed by animals during recording sessions influences the formation of these representations. We analyzed data from a treadmill running study conducted by Kraus et al., 2013, in which rats were trained to run at different velocities. The rats were recorded in two trial contexts: a ‘fixed time’ condition, where the animal ran on the treadmill for a predetermined duration before proceeding, and a ‘fixed distance’ condition, where the animal ran a specific distance on the treadmill. Our findings indicate that the type of experimental condition significantly influenced the encoding of hippocampal cells. Specifically, distance-encoding cells dominated in fixed-distance experiments, whereas time-encoding cells dominated in fixed-time experiments. These results underscore the flexible coding capabilities of the hippocampus, which are shaped by over-representation of salient variables associated with reward conditions.