Representational drift as a result of implicit regularization.

Recent studies show that, even in constant environments, the tuning of single neurons changes over time in a variety of brain regions. This representational drift has been suggested to be a consequence of continuous learning under noise, but its properties are still not fully understood. To investigate the underlying mechanism, we trained an artificial network on a simplified navigational task. The network quickly reached a state of high performance, and many units exhibited spatial tuning. We then continued training the network and noticed that the activity became sparser with time. Initial learning was orders of magnitude faster than ensuing sparsification. This sparsification is consistent with recent results in machine learning, in which networks slowly move within their solution space until they reach a flat area of the loss function. We analyzed four datasets from different labs, all demonstrating that CA1 neurons become sparser and more spatially informative with exposure to the same environment. We conclude that learning is divided into three overlapping phases: (i) Fast familiarity with the environment; (ii) slow implicit regularization; and (iii) a steady state of null drift. The variability in drift dynamics opens the possibility of inferring learning algorithms from observations of drift statistics.

Representational drift as a result of implicit regularization.

Recent studies show that, even in constant environments, the tuning of single neurons changes over time in a variety of brain regions. This representational drift has been suggested to be a consequence of continuous learning under noise, but its properties are still not fully understood. To investigate the underlying mechanism, we trained an artificial network on a simplified navigational task. The network quickly reached a state of high performance, and many units exhibited spatial tuning. We then continued training the network and noticed that the activity became sparser with time. Initial learning was orders of magnitude faster than ensuing sparsification. This sparsification is consistent with recent results in machine learning, in which networks slowly move within their solution space until they reach a flat area of the loss function. We analyzed four datasets from different labs, all demonstrating that CA1 neurons become sparser and more spatially informative with exposure to the same environment. We conclude that learning is divided into three overlapping phases: (i) Fast familiarity with the environment; (ii) slow implicit regularization; (iii) a steady state of null drift. The variability in drift dynamics opens the possibility of inferring learning algorithms from observations of drift statistics.

Flexible coding of time or distance in hippocampal cells.

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.

Active experience, not time, determines within-day representational drift in dorsal CA1.

Memories of past events can be recalled long after the event, indicating stability. But new experiences are also integrated into existing memories, indicating plasticity. In the hippocampus, spatial representations are known to remain stable but have also been shown to drift over long periods of time. We hypothesized that experience, more than the passage of time, is the driving force behind representational drift. We compared the within-day stability of place cells' representations in dorsal CA1 of the hippocampus of mice traversing two similar, familiar tracks for different durations. We found that the more time the animals spent actively traversing the environment, the greater the representational drift, regardless of the total elapsed time between visits. Our results suggest that spatial representation is a dynamic process, related to the ongoing experiences within a specific context, and is related to memory update rather than to passive forgetting.

Are grid cells used for navigation? On local metrics, subjective spaces, and black holes.

The symmetric, lattice-like spatial pattern of grid-cell activity is thought to provide a neuronal global metric for space. This view is compatible with grid cells recorded in empty boxes but inconsistent with data from more naturalistic settings. We review evidence arguing against the global-metric notion, including the distortion and disintegration of the grid pattern in complex and three-dimensional environments. We argue that deviations from lattice symmetry are key for understanding grid-cell function. We propose three possible functions for grid cells, which treat real-world grid distortions as a feature rather than a bug. First, grid cells may constitute a local metric for proximal space rather than a global metric for all space. Second, grid cells could form a metric for subjective action-relevant space rather than physical space. Third, distortions may represent salient locations. Finally, we discuss mechanisms that can underlie these functions. These ideas may transform our thinking about grid cells.

Spatial coding in the hippocampus and hyperpallium of flying owls.

The elucidation of spatial coding in the hippocampus requires exploring diverse animal species. While robust place-cells are found in the mammalian hippocampus, much less is known about spatial coding in the hippocampus of birds. Here we used a wireless-electrophysiology system to record single neurons in the hippocampus and other two dorsal pallial structures from freely flying barn owls (Tyto alba), a central-place nocturnal predator species with excellent navigational abilities. The owl's 3D position was monitored while it flew between perches. We found place cells-neurons that fired when the owl flew through a spatially restricted region in at least one direction-as well as neurons that encoded the direction of flight, and neurons that represented the owl's perching position between flights. Many neurons encoded combinations of position, direction, and perching. Spatial coding was maintained stable and invariant to lighting conditions. Place cells were observed in owls performing two different types of flying tasks, highlighting the generality of the result. Place coding was found in the anterior hippocampus and in the posterior part of the hyperpallium apicale, and to a lesser extent in the visual Wulst. The finding of place-cells in flying owls suggests commonalities in spatial coding across mammals and birds.

Use it or lose it.

Blocking the activity of neurons in a region of the brain involved in memory leads to cell death, which could help explain the spatiotemporal disorientation observed in Alzheimer's disease.

The chicken and egg problem of grid cells and place cells.

Place cells and grid cells are major building blocks of the hippocampal cognitive map. The prominent forward model postulates that grid-cell modules are generated by a continuous attractor network; that a velocity signal evoked during locomotion moves entorhinal activity bumps; and that place-cell activity constitutes summation of entorhinal grid-cell modules. Experimental data support the first postulate, but not the latter two. Several families of solutions that depart from these postulates have been put forward. We suggest a modified model (spatial modulation continuous attractor network; SCAN), whereby place cells are generated from spatially selective nongrid cells. Locomotion causes these cells to move the hippocampal activity bump, leading to movement of the entorhinal manifolds. Such inversion accords with the shift of hippocampal thought from navigation to more abstract functions.

Territorial blueprint in the hippocampal system.

As we skillfully navigate through familiar places, neural computations of distances and coordinates escape our attention. However, we perceive clearly the division of space into socially meaningful territories. 'My space' versus 'your space' is a distinction familiar to all of us. Spatial frontiers are social in nature since they regulate individuals' access to utilities in space depending on hierarchy and affiliation. How does the brain integrate spatial geometry with social territory? We propose that the action of oxytocin (OT) in the entorhinal-hippocampal regions supports this process. Grounded on the functional role of the hypothalamic neuropeptide in the hippocampal system, we show how OT-induced plasticity may bias the geometrical coding of place and grid cells to represent social territories.

Hippocampal sub-networks exhibit distinct spatial representation deficits in Alzheimer's disease model mice.

Not much is known about how the dentate gyrus (DG) and hippocampal CA3 networks, critical for memory and spatial processing, malfunction in Alzheimer's disease (AD). While studies of associative memory deficits in AD have focused mainly on behavior, here, we directly measured neurophysiological network dysfunction. We asked what the pattern of deterioration of different networks is during disease progression. We investigated how the associative memory-processing capabilities in different hippocampal subfields are affected by familial AD (fAD) mutations leading to amyloid-ß dyshomeostasis. Specifically, we focused on the DG and CA3, which are known to be involved in pattern completion and separation and are susceptible to pathological alterations in AD. To identify AD-related deficits in neural-ensemble dynamics, we recorded single-unit activity in wild-type (WT) and fAD model mice (APPSwe+PSEN1/ΔE9) in a novel tactile morph task, which utilizes the extremely developed somatosensory modality of mice. As expected from the sub-network regional specialization, we found that tactile changes induced lower rate map correlations in the DG than in CA3 of WT mice. This reflects DG pattern separation and CA3 pattern completion. In contrast, in fAD model mice, we observed pattern separation deficits in the DG and pattern completion deficits in CA3. This demonstration of region-dependent impairments in fAD model mice contributes to understanding of brain networks deterioration during fAD progression. Furthermore, it implies that the deterioration cannot be studied generally throughout the hippocampus but must be researched at a finer resolution of microcircuits. This opens novel systems-level approaches for analyzing AD-related neural network deficits.

Dori Derdikman.

Interview with Dori Derdikman, who studies spatial memory and navigation at the Technion - Israel Institute of Technology.

Directional tuning in the hippocampal formation of birds.

Birds strongly rely on spatial memory and navigation. Therefore, it is of utmost interest to reveal how space is represented in the avian brain. Here we used tetrodes to record neurons from the hippocampal formation of Japanese quails-a ground-dwelling species-while the quails roamed in an open-field arena. Whereas spatially modulated cells (place cells, grid cells, border cells) were generally not encountered, the firing rate of about 12% of the neurons was unimodally and significantly modulated by the head azimuth-i.e., these were head-direction cells (HD cells). Typically, HD cells were maximally active at one preferred direction and minimally at the opposite null direction, with preferred directions spanning all 360° across the population. The preferred direction was independent of the animal's position and speed and was stable during the recording session. The HD tuning was broader compared to that of HD cells in rodents, and most cells had non-zero baseline firing in all directions. However, similar to findings in rodents, the HD tuning usually rotated with the rotation of a salient visual cue in the arena. Thus, these findings support the existence of an allocentric HD representation in the quail hippocampal formation and provide the first demonstration of HD cells in birds.

During hippocampal inactivation, grid cells maintain synchrony, even when the grid pattern is lost.

The grid cell network in the medial entorhinal cortex (MEC) has been subject to thorough testing and analysis, and many theories for their formation have been suggested. To test some of these theories, we re-analyzed data from Bonnevie et al., 2013, in which the hippocampus was inactivated and grid cells were recorded in the rat MEC. We investigated whether the firing associations of grid cells depend on hippocampal inputs. Specifically, we examined temporal and spatial correlations in the firing times of simultaneously recorded grid cells before and during hippocampal inactivation. Our analysis revealed evidence of network coherence in grid cells even in the absence of hippocampal input to the MEC, both in regular grid cells and in those that became head-direction cells after hippocampal inactivation. This favors models, which suggest that phase relations between grid cells in the MEC are dependent on intrinsic connectivity within the MEC.

Dissociation between Postrhinal Cortex and Downstream Parahippocampal Regions in the Representation of Egocentric Boundaries.

Navigation requires the integration of many sensory inputs to form a multi-modal cognitive map of the environment, which is believed to be implemented in the hippocampal region by spatially tuned cells [1-10]. These cells encode various aspects of the environment in a world-based (allocentric) reference frame. Although the cognitive map is represented in allocentric coordinates, the environment is sensed through diverse sensory organs, mostly situated in the animal's head, and therefore represented in sensory and parietal cortices in head-centered egocentric coordinates. Yet it is not clear how and where the brain transforms these head-centered egocentric representations to map-like allocentric representations computed in the hippocampal region. Theoretical modeling has predicted a role for both egocentric and head direction (HD) information in performing an egocentric-allocentric transformation [11-15]. Here, we recorded new data and also used data from a previous study [16]. Adapting a generalized linear model (GLM) classification [17]; we show that the postrhinal cortex (POR) contains a population of pure egocentric boundary cells (EBCs), in contrast with the conjunctive EBCs × HD cells, which we found downstream mostly in the parasubiculum (PaS) and in the medial entorhinal cortex (MEC). Our finding corroborates the idea of a brain network performing an egocentric to allocentric transformation by HD cells. This is a fundamental building block in the formation of the brain's internal cognitive map.

Mitochondrial Regulation of the Hippocampal Firing Rate Set Point and Seizure Susceptibility.

Maintaining average activity within a set-point range constitutes a fundamental property of central neural circuits. However, whether and how activity set points are regulated remains unknown. Integrating genome-scale metabolic modeling and experimental study of neuronal homeostasis, we identified mitochondrial dihydroorotate dehydrogenase (DHODH) as a regulator of activity set points in hippocampal networks. The DHODH inhibitor teriflunomide stably suppressed mean firing rates via synaptic and intrinsic excitability mechanisms by modulating mitochondrial Ca2+ buffering and spare respiratory capacity. Bi-directional activity perturbations under DHODH blockade triggered firing rate compensation, while stabilizing firing to the lower level, indicating a change in the firing rate set point. In vivo, teriflunomide decreased CA3-CA1 synaptic transmission and CA1 mean firing rate and attenuated susceptibility to seizures, even in the intractable Dravet syndrome epilepsy model. Our results uncover mitochondria as a key regulator of activity set points, demonstrate the differential regulation of set points and compensatory mechanisms, and propose a new strategy to treat epilepsy.

ASCT1 (Slc1a4) transporter is a physiologic regulator of brain d-serine and neurodevelopment.

d-serine is a physiologic coagonist of NMDA receptors, but little is known about the regulation of its synthesis and synaptic turnover. The amino acid exchangers ASCT1 (Slc1a4) and ASCT2 (Slc1a5) are candidates for regulating d-serine levels. Using ASCT1 and ASCT2 KO mice, we report that ASCT1, rather than ASCT2, is a physiologic regulator of d-serine metabolism. ASCT1 is a major d-serine uptake system in astrocytes and can also export l-serine via heteroexchange, supplying neurons with the substrate for d-serine synthesis. ASCT1-KO mice display lower levels of brain d-serine along with higher levels of l-alanine, l-threonine, and glycine. Deletion of ASCT1 was associated with neurodevelopmental alterations including lower hippocampal and striatal volumes and changes in the expression of neurodevelopmental-relevant genes. Furthermore, ASCT1-KO mice exhibited deficits in motor function, spatial learning, and affective behavior, along with changes in the relative contributions of d-serine vs. glycine in mediating NMDA receptor activity. In vivo microdialysis demonstrated lower levels of extracellular d-serine in ASCT1-KO mice, confirming altered d-serine metabolism. These alterations are reminiscent of some of the neurodevelopmental phenotypes exhibited by patients with ASCT1 mutations. ASCT1-KO mice provide a useful model for potential therapeutic interventions aimed at correcting the metabolic impairments in patients with ASCT1 mutations.

Role of the head-direction signal in spatial tasks: when and how does it guide behavior?

Since their discovery, mammalian head-direction (HD) cells have been extensively researched in terms of sensory origins, external cue control, and circuitry. However, the relationship of HD cells to behavior is not yet fully understood. In the current review, we examine the anatomical clues for information flow in the HD circuit and an emerging body of evidence that links neural activity of HD cells and spatial orientation. We hypothesize from results obtained in spatial orientation tasks involving HD cells that when properly aligned with available external cues, the HD signal could be used for guiding rats to a goal location. However, contradictory inputs from separate sensory systems may reduce the influence of the HD signal such that animals are able to switch between this and other systems according to their impact on behavior.

The Many Worlds of Plasticity Rules.

Two recent papers have tackled the fundamental questions of how place fields are formed in a new environment and what plasticity mechanisms contribute to this process. Bittner et al., in their recent publication, discovered a novel plasticity rule that, in contrast to previous rules, spans the behavioral, seconds-long, timescale. Sheffield et al. have monitored, for the first time, dendritic activity during place field formation, and show the emergence of spatially tuned local NMDA spikes in basal dendrites of CA1 neurons. Together, these papers suggest that multiple complementary dendritic plasticity mechanisms may contribute to place field formation in changing environmental contexts.

Consistency of Spatial Representations in Rat Entorhinal Cortex Predicts Performance in a Reorientation Task.

Goal-directed behavior can be affected by environmental geometry. A classic example is the rectangular arena reorientation task, where subjects commonly confuse opposite but geometrically identical corners [1]. Until recently, little was known about how environmental geometry shapes spatial representations in a neurobehavioral context [2] (although see [3]). In the present study, we asked: Under what circumstances does the internal cognitive map predict behavior? And when does it fail to do so? To this end, we developed a variant of the classical reorientation task that allows for investigation of temporal dynamics of reorientation. We recorded head-direction (HD) cells and grid cells in the medial entorhinal cortex (MEC) of rats before, during, and after performing the task. MEC cells showed a bimodal response of being either aligned or rotated, relative to the free-foraging open-field sessions. Alignment was remarkably stable between disorientations and indicative of corner choice as a function of current and past alignment of spatial representations. Accordingly, when the cells showed consistent and properly aligned readout across multiple trials, behavioral choices were better predicted by HD and grid cell readout, with a probability of more than 70%. This was not the case when the cells did not show a stable consistent readout. Our findings indicate that entorhinal spatial representations predict corner choice, contingent on the stability and reliability of their readout. This work sets the stage for further studies on the link between the reliability of the neuronal signal and behavior, with implications for many brain systems in many organisms.

Grid Cells Encode Local Positional Information.

The brain has an extraordinary ability to create an internal spatial map of the external world [1]. This map-like representation of environmental surroundings is encoded through specific types of neurons, located within the hippocampus and entorhinal cortex, which exhibit spatially tuned firing patterns [2, 3]. In addition to encoding space, these neurons are believed to be related to contextual information and memory [4-7]. One class of such cells is the grid cells, which are located within the entorhinal cortex, presubiculum, and parasubiculum [3, 8]. Grid cell firing forms a hexagonal array of firing fields, a pattern that is largely thought to reflect the operation of intrinsic self-motion-related computations [9-12]. If this is the case, then fields should be relatively uniform in size, number of spikes, and peak firing rate. However, it has been suggested that this is not in fact the case [3, 13]. The possibility exists that local spatial information also influences grid cells, which-if true-would greatly change the way in which grid cells are thought to contribute to place coding. Accordingly, we asked how discriminable the individual fields of a given grid cell are by looking at the distribution of field firing rates and reproducibility of this distribution across trials. Grid fields were less uniform in intensity than expected, and the pattern of strong and weak fields was spatially stable and recurred across trials. The distribution remained unchanged even after arena rescaling, but not after remapping. This suggests that additional local information is being overlaid onto the global hexagonal pattern of grid cells.