Final Report Summary - SPACEBRAIN (Space coding in hippocampo-entorhinal neuronal assemblies)
Despite impressive advances in almost all fields of neuroscience our understanding of brain function has largely been confined to the brain's building blocks at the microscopic level, and to phenomenological descriptions at the macroscopic level. We know less about the link between these levels - how complex mental functions originate from electrical and chemical processes. Understanding this link requires a comprehensive and integrated multilevel analysis focused on neuronal assemblies and microcircuits where myriads of intricately connected neurons with different properties act together. The aim of SPACEBRAIN was to provide a multidisciplinary cross-level understanding of computation in cortical microcircuits.
Some 42 months of collaboration have produced significant advances in our understanding of how space is represented in neural networks of the entorhinal cortex and hippocampus, brain regions known to form internal maps of the local environment. Two papers with SPACEBRAIN funding have been published or accepted for publication in Nature, three in Science, one in Cell, and several are appearing in other high-impact journals such as Nature Neuroscience and Nature Methods. Key scientific advances include, first of all, new insights into the intrinsic wiring of the entorhinal cortex. The project has shown how grid cells (one of the key cells types) connect to other cell types in the circuit and in neighbouring regions, and we have learned that grid cells are strongly linked by inhibitory connections. Grid cells are not to be the only space-coding cell type. Border cells are entorhinal cells that fire along geometric edges and landmarks, whereas boundary vector cells in the subiculum fire at specific distances from such boundaries. The project showed that space-coding cells develop a lot earlier than researchers thought before and that a rudimentary spatial representation is present already when rats make their first navigational experiences outside the nest, at the beginning of the third postnatal week.
Among the known spatial cell types, direction-coding cells appear first, probably before the animals open their eyes for the first time. They are followed by place cells and grid cells, which need further maturation but nonetheless appear adult-like one-to-two weeks after the beginning of independent exploration. The SPACEBRAIN project has shown that grid cells and place cells encode horizontal and vertical space differently, with stronger spatial resolution in the horizontal dimension. The project has also shown that hippocampal neuronal assemblies follow principles of attractor networks and are organised within 125-ms time blocks referred to as theta cycles. Theoretical work has addressed the significance of this pattern of organisation.
The scientific progress has been accompanied by development of new research tools. Microdrives have been developed for recording electrical activity at multiple brain locations at the same time, and they have been miniaturised for recording in small animals. A portable high-resolution microscope has been developed for population studies of local neural networks during behaviour. The project reflects several unique features of SPACEBRAIN. One is the integration of experimental and theoretical work. From the beginning, experimentalists and modellers have been working on complementary subprojects, exchanging information as soon as data or models were available. Another is the explicit attempt to combine methods and concepts from multiple levels of analysis, ranging from mechanisms of synaptic transmission to interactions between large distributed circuits. The success is likely to reflect the open and positive working atmosphere of the consortium. Trust and an open attitude have laid the basis for constructive discussion and true collaboration.
Project context and objectives
Neuroscience is at a stage where integrative understanding of brain function can be reached. Tools for a cross-level analysis centred on microcircuit structure and activity in neuronal assemblies are becoming available and it has become possible to model large networks with connectional and dynamic properties similar to those of real brain circuits. This brought models closer to experimental observations and the benefits of mutual interaction between theoretical and experimental communities became apparent. The principal objective of SPACEBRAIN was to exploit these converging conceptual and technological developments to uncover some of the fundamental principles of neuronal assembly and microcircuit operation in the mammalian cortex. Not all areas of cortical investigation have matured to the stage where neuronal computation can be understood at the microcircuit level. Insights into circuit functions have been obtained for the early stages of sensory systems where the processing of signals can be followed through networks of increasing complexity from the receptor level to the primary sensory cortices.
These studies have largely clarified how neurons and neuronal networks extract features from the external world and refine these into cortical representations. How the brain generates its own codes, in the non-sensory parts of the cortex, has remained deeply mysterious. A new path was opened up when the grid cells of the entorhinal cortex were discovered by one of the participating groups in 2005. Grid cells are place-modulated neurons whose firing locations define a periodic triangular array covering the entirety of the animal's environment. They are thought to form an essential part of the brain's coordinate system for metric navigation. Grid cells have attracted enormous attention because the crystal-like structure underlying their firing fields does not arise out of the simplicity of their components, as it does in more familiar physical systems, but rather out of network complexity. It is a structure not imported from the outside world but created within the nervous system. Grid cells provide scientists with a direct window into some of the most fundamental operational principles of cell assemblies and microcircuits in the brain. The grid cells are part of the brain's system for spatial representation, which includes the hippocampus and the entorhinal cortex.
Through the discovery of place cells by one of the consortium groups in 1971 and the discovery of grid cells by one of the other groups in 2005, the hippocampo-entorhinal network is now, in the view of many neuroscientists, emerging as the brain system with the best potential for generating a comprehensive cross-level understanding of how non-sensory neural networks compute in the mammalian brain. We know that place cells form maps of the animal's local environment and that different maps are stored in the hippocampal circuit for different environments. These maps are different from the spatial maps generated by grid cells in the entorhinal cortex. Grid cells keep a constant spatial relationship across all environments; two grid cells with overlapping vertices in one environment will have overlapping vertices in another environment as well.
The entorhinal keeps a similar ensemble structure across all places. Hippocampal place-cell maps are environment-unique and strictly tied to memory functions of this brain structure. Studies of mechanisms by which neuronal circuits and cell assemblies in these two brain regions keep track of the animal's location as it is moving around in the environment may uncover general computational mechanisms employed widely in the cortex, not only in the entorhinal and hippocampal cortices. The aim of SPACEBRAIN was to begin the search for principles of microcircuit computation in the relatively accessible spatial representation system of rodents, using a powerful combination of novel computational, electrophysiological, optical and molecular research tools that have never been applied to the analysis of brain circuitry before.
Specific objectives
The consortium set out to provide a better understanding of how spatial location is computed and represented in microcircuits of the hippocampus and entorhinal cortex and how such representations arise during development of the nervous system.
The central objectives were as follows:
1. To provide a wiring diagram of entorhinal microcircuitry that contains sufficient quantitative detail to lay the foundation for a new generation of realistic computational models of dynamic spatial representation. The potential power of computational models in accounting for spatial representation in assemblies of grid cells has introduced a strong need for detailed quantitative analysis of the microcircuitry of the entorhinal cortex. A key objective of the project was to outline the intrinsic organisation of the area, with emphasis on its cell types, their intralaminar and interlaminar connections and the modifiability of the principal interconnections.
2. To determine how intrinsic and extrinsic inputs combine into a single spatial representation in entorhinal grid cells. Using a combination of experimental and theoretical approaches, the consortium set out to determine how motion-related information is combined with visual information about landmarks and geometry into a coherent representation of the animal's location in space.
3. To determine the mechanisms of neuronal assembly formation in hippocampus and entorhinal cortex. Experimental and theoretical approaches were combined to establish the operational principles of cell assemblies in place cells of the hippocampus and grid cells of the entorhinal cortex. Large numbers of cells were recorded simultaneously in the CA3 region of the hippocampus in order to determine the mechanisms for transformation between discrete spatial representations.
4. To determine the mechanisms of cell assembly formation in the temporal domain. The consortium used a combination of experimental and computational methods to determine the origin and neuronal mechanisms of hippocampal theta phase precession, one of the brain's clearest signatures of temporal sequence organisation.
5. To determine how spatial representations are formed in hippocampo-entorhinal cell assemblies during development. The consortium set out to establish how key components of the hippocampo-entorhinal representation system develop relative to each other during the first postnatal weeks, what their interrelation is, and whether and how maturation of the system interacts with early experience. Key objectives were to determine when connections between different stages of the hippocampal formation, the entorhinal cortex and the presubiculum reach adult patterns, whether different functional cell types develop in particular sequences, and whether this development reflects mainly maturational factors or is shaped by experience.
Understanding microcircuit function is strongly dependent on the development of new research tools. An additional objective of the SPACEBRAIN project was therefore to use and develop methodologies for behavioural assessment, computational modelling and high-density single-cell recording, as well as gene silencing and single-cell optical recording. Particular emphasis was placed on the development of multi-site recording technology for high-density studies of cell assemblies, as well as the development of a head-mounted two-photon fluorescence microscope for dynamic visualisation of cell assemblies. These technical developments were accompanied by the attempt to develop a new generation of anatomically and physiologically realistic computational models of entorhinal-hippocampal neural circuits. The project was motivated by its potential for translation to both clinical and industrial applications.
Topographical disorientation is one of the earliest and most prominent behavioural symptoms of Alzheimer's disease. Because the entorhinal cortex is the first brain region to be affected in many patients with Alzheimer's disease, functional insight into the spatial representation mechanisms of this brain region should provide clinical neurologists and health workers with essential tools for early diagnostics, prevention and treatment of Alzheimer's disease. Insights may not only reduce the prevalence of Alzheimer's disease and the enormous health care burdens associated with the disease, but may also have an important preventive impact on traffic accidents and human error in spatially demanding man-machine interactions caused by Alzheimer's disease at pre-diagnostic stages. Understanding space coding in the brain may also exert significant impact on the development of navigating robots. Despite decades of research, robotics has so far failed to develop a navigating agent capable of exploring an unknown or complex landscape at an efficiency that is anywhere near to that of a small rodent. By uncovering some of the fundamental algorithms of neural networks for spatial navigation in the rodent brain, the project is likely to provide the robotics industry with entirely new concepts for design of artificial navigating agents.
Project results
The project has led to significant advances in understanding of how space is represented in neural networks of the entorhinal cortex and hippocampus. Organisational principles of entorhinal circuit structure have been determined, back projections from the hippocampus have been shown to be essential for stable grid firing in the entorhinal cortex, grid scale has been shown to depend on the velocity of the moving animal, spatial cell types have been observed during early postnatal development, and cell ensembles have been found to operate as attractor networks within time frames defined by the hippocampal theta rhythm. These and many other findings are outlined in the section below. The work was organised in six work packages (WPs):
1. Attractor dynamics and circuit organisation in entorhinal cortex
The first scientific WP tested the idea that the entorhinal representation of self-location is based on continuous attractor dynamics. Attractors become continuous when the distribution of input cues to the system is continuous. In a continuous attractor network, a bump of activity is formed at a given location in the network defined by the input pattern. This bump of activity is then thought to move around in the network space in accordance with the speed and direction of the animal's movement in external space. It has been suggested that entorhinal networks have the ability to support the formation and translocation of attractor bumps. The continuous attractor idea makes a number of explicit assumptions about the underlying cellular and synaptic architecture and neural dynamics of the entorhinal cortex but the assumptions generally lack experimental support. The weak understanding of details of intrinsic wiring diagram of the entorhinal area, and physiological properties of cell assemblies in this network, motivated this WP. We sought a quantitative characterisation of the entorhinal microcircuitry. This was essential for the development of realistic computational models of spatial representation in the hippocampal-entorhinal system. The general aim was to provide such a characterisation and to test the basic assumptions of attractor models of the entorhinal cortex.
The consortium has acquired significant insight into the intrinsic wiring of the medial entorhinal cortex. Paired recordings from layer II principal cells have suggested that connectivity between principal cells within the layer is weak, averaging around 1 % of cell pairs or less, but it was shown that there is strong connectivity between principal cells and interneurons, pointing to alternative ways of synchronising the circuit. Targeted genetic manipulations have demonstrated that GABAergic interneurons in the medial entorhinal cortex project not only within the entorhinal circuit itself but also to subfields of the hippocampus, providing a direct means of synchronising the two structures with high temporal precision, such as during gamma network oscillations. The experimental work has been accompanied by theoretical work addressing the principles of attractor dynamics in hippocampus and entorhinal cortex, pointing the way to new experiments after completion of the SPACEBRAIN project.
Recurrent interconnectivity can in principle be present in cortical networks at two organisational levels, either within layers (intralaminar) or across layers (interlaminar). Previous studies have shown widespread interlaminar axonal arborisation of principal cells, connecting deep with superficial layers, as well as intralaminar connections in layers II, III, and V; however, the precision and detail of these studies have been far below the levels needed for realistic computational modelling of entorhinal networks. More than a hundred neurons in all layers of medial entorhinal cortex electrophysiologically were analysed through whole-cell current-clamp recordings, at the same time as these neurons were stained in order to determine their unique connection patterns. Excitatory interlaminar connections were found to arise from neurons in layers III and V but the study also showed that these connections may arise from layer VI principal cells. Based on an analysis of more than 500 potential pairs of stellate cells at different ages, it was shown that the ratio of monosynaptically coupled stellate cells in layer II is small. Less than 1 % of the recorded cell pairs had direct synaptic connections.
Connectivity ratios did not increase when cells were at larger distances from each other but stayed within the established average axonal arbor of 550 μm. Connections between principal cells and interneurons in layer II were substantially stronger. These findings indicate that the proposed excitatory interconnectivity underlying pure grid cell properties is weak and that, while inhibitory connections may play a primary role in synchronisation of the layer II network, the continuous attractor models accounting for grid activity in that layer need major revision. A new hypothesis has been formulated in which a denser recurrent connectivity in the deeper layers establishes coherence among conjunctive head direction/grid units, which then feed into a non-recurrently connected pools of pure layer II grid units to achieve sharper spatial resolution; a hypothesis that will be tested in computational models after the completion of the project. Behavioural tests in mice where the expression of Cre-recombinase is under the control of the parvalbumin promoter suggested that fast-spiking interneurons may be instrumental in synchronisation of network activity as well as the formation of precise spatial representations.
A critical assumption of continuous attractor models of grid cells in the entorhinal cortex was that descending and ascending connections between layer II and the deeper layers are asymmetric. Such asymmetry is necessary if conjunctive cells in the deeper layers are to translate the grid cell representation in layer II in accordance with the animal's movements. A key objective of the circuit analysis was to determine if asymmetry was present. We focused on connectivity between layers V and II. Out of more than a hundred individually stained neurons, approximately a third show a clear lateral shift in their axonal arbor, when compared to the dendritic arbor, in agreement with the translation idea. The findings suggest that also neurons in layers III and VI may have laterally shifted projections. Previous theoretical work indicated that cooperative activity of neocortical circuits is sensitive to short-term synaptic plasticity (STP) in connections between pyramidal neurons. Synaptic depression can generate fast changes in activity, while synaptic facilitation on the opposite can stabilise attractor states. An important objective of SPACEBRAIN was to determine the validity of this idea. The role of STP in stabilisation of attractor networks was tested in a 'ring model' applied to place and grid cells as well as a range of other problems. The results show that synaptic facilitation can dramatically slow down the drift velocity of the network, thus appearing as a strong stabilising factor in representation of space and memory.
2. Calibration of entorhinal assembly representation
Location-specific firing of grid cells is thought to be updated principally by a path integration-based mechanism, based on computations derived from the animal's movements through space. This suggestion is consistent with the constancy of grid maps across environments, their imperviousness to removal or displacement of external landmarks and the fact that grids are expressed immediately in a novel environment. While the basic structure of the grid may be determined by self-motion cues, grids are clearly influenced by external visual cues as:
(i) both phase and orientation of the grid are controlled by specific landmarks in the environment and
(ii) re-sizing of a familiar environment causes partial re-scaling of the entorhinal grid in the scaled dimension.
These observations implicate an interaction between path integration signals and visual inputs. The general objective was to determine how this integration is accomplished. A combined experimental-theoretical approach was taken in order to understand the relationship between motion inputs and other sensory inputs and the formation of grid cell patterns. The results provide important insights. It was shown with experimental approaches that back projections from the hippocampus are essential for maintaining stable grid representations, although they may not be required for the ensemble structure of the grid cells as such. Experiments have demonstrated that theta modulation of grid cells is dependent on the linear velocity of the animals, and that place and grid cells represent horizontal and vertical space differently, and computational modelling has investigated the significance of hippocampal back projections and other inputs in generating stable spatial periodicity. Previous work has shown that grid cells rescale when animals are placed in novel environments but the underlying mechanisms were not understood well. Experiments to tell us whether the grid expansion in a novel environment only applies in the absence of visual and other environmental cues we designed. Results suggest that grid expansion takes place even in novel environments with strong visual and odour cues. Continued testing in novel environments leads to relaxation back to the original grid scale suggesting that the expansion is due to novelty and not the absence of environmental information and that recalibration back to original scale takes place as new environments become familiar. Grid fields are larger in novel environments than they are in familiar environments, consistent with the notion of scaled-up grids in novel environments.
Using a combination of experimental and computational approaches, the research aimed to establish whether hippocampal back projections to the entorhinal cortex are necessary for maintaining and resetting grid representations in the entorhinal cortex. A role in maintenance and updating of representations would be expected if associations between path integrator coordinates and extrinsic landmarks were stored in the hippocampus and back projections to the entorhinal cortex were necessary for stabilising the grid throughout the environment. Associations between path integrator outputs and fixed landmarks might be formed within the medial entorhinal cortex itself by combining grid activity with specific sensory inputs received. To test these possibilities against each other we inactivated the hippocampus temporarily by infusing muscimol into both hippocampi before grid cells were recorded in a standard foraging condition. Hippocampal inactivation resulted in gradual decrease in firing rate of the grid cells, followed by disintegration of the grid pattern. The temporal coherence of the grid cell population remained stable, suggesting that the grid structure persisted and that the alignment to landmarks deteriorated. The results suggest that grid cells are dependent on stored associations in the hippocampus to anchor the grid to the external environment.
Output from the hippocampus may provide a resetting mechanism that constantly updates the network in the entorhinal cortex with regard to stored landmark features. This may be vital to the operation of oscillation-dependent mechanisms for grid cell formation; without frequent resetting, noise and variation in the period of the theta rhythm may be detrimental to the formation of spatial periodicity. During the project, simulations have been conducted in the context of the oscillatory interference model of grid cell firing. Under this model, the spatial organisation of the firing of grid cells arises from the interference between sub-threshold membrane potential oscillations in the theta band: the firing rate increases when the oscillations are in phase and decreases as they go out of phase. The model performs path integration by assuming that the frequencies of oscillations each reflect velocity in a specific direction. This scheme is susceptible to the accumulation of error in the inputs conveying velocity information. The simulations investigated how spatial stability can be attained by phase-reset of the oscillations in two ways. The first mechanism involves a phase-reset of all oscillations to be in-phase at a specific location, triggered by input from the place cells which fire at that location, given that the firing location of place cells is maintained in location by environmental sensory inputs. Spatial stability of grid cell firing was successfully attained under this model. The second mechanism involves a reset of all oscillations triggered by inconsistency between multiple oscillators. When velocity along more than two directions is integrated by multiple oscillators, different pairs of oscillators can produce different estimates of location. When these estimates become mutually inconsistent, all oscillators can be reset to correspond to the mean estimate. This provides a mechanism for error correction that is potentially intrinsic to entorhinal cortex. The effects of the two mechanisms were compared as a function of increasingly noisy estimates of velocity.
An objective was to seek the source of cortical and subcortical inputs to the grid cell system using anatomical tracing methods and to determine what sensory modalities influence the metric of the grid, particularly its spacing. Among the candidate inputs to the grid cell area in the medial entorhinal cortex, researchers focused on the retrosplenial cortex and interactions between inputs from this region and from postrhinal and pre-parasubicular cortical regions. Research showed that projections to layer V or medial entorhinal cortex originate mainly from ventral subdivision of retrosplenial cortex generally referred to as granular part A. Electron-microscopy analyses shows that this input is mainly excitatory, targeting mainly principal cells. The more dorsal subdivision b sends a massive input to layer III of the presubiculum, targeting preferentially principal cells that project to medial entorhinal cortex. Projections from the postrhinal cortex show a striking topographical organisation in that a narrow area of postrhinal cortex, close to the border with medial entorhinal cortex, projects to the directly adjacent portion of medial entorhinal cortex, which contains grid cells with relatively small firing fields. More dorsal portions of postrhinal cortex project more ventrally towards the area of medial entorhinal cortex where much coarser grid cells have been recorded. Postrhinal inputs target principal cells in layers II and III that project to the various subfields in the hippocampus. The majority of these projections are excitatory, targeting spines of projection neurons in the medial entorhinal cortex. Axons from the presubiculum target entorhinal neurons layers II, III and V, which project to the hippocampus. Inputs from pre- and parasubiculum converge onto single principal cells in the medial entorhinal cortex. In conjunction with the observations on retrosplenial inputs, the data break with older ideas in that information from retrosplenial cortex is shown to reach superficial layers of medial entorhinal cortex through a massive di-synaptic pathway involving layer III cells in presubiculum. The interaction between static and motion cues on place cells was explored in behaving animals on a linear track in which static somatosensory cues were shown to act as foci for the formation of place fields. During the project, these path integration studies were extended to the vertical dimension, by recording place and grid cells as animals explored environments having height as well as horizontal dimensions. It was found that place and grid cells show little modulation by height, suggesting that the basic frame of reference for the grid cell map is planar rather than volumetric. Behavioural studies showed that this anisotropy of encoding is mirrored in behavioural performance, in which rats also show a horizontal bias in both foraging and detour tasks.
3. Interactions between entorhinal and hippocampal cell assemblies
The aim was to establish mechanisms for interactions between neuronal assemblies within the hippocampus and between the hippocampus and the entorhinal cortex. The project focused on neuronal coding in large populations of hippocampal and entorhinal neurons. Experimental data generated have demonstrated that hippocampal assemblies follow principles of attractor networks and that attractor states are expressed within individual theta cycles, with relative independence between successive theta cycles. Competitive interactions between different ensemble states have been observed at the sub-second time scale during transitions from one representation to another. New microdrives for multi-tetrode ensemble recording with independently movable tetrodes, constructed by the industrial partner of the consortium will as soon as they are released on the market allow participants inside and outside the consortium to address large-scale population phenomena that were not readily accessible until now. Findings have been incorporated in computational models of attractor dynamics. The CA3 subfield of the hippocampus has many properties of an associative memory network. Most of the numerous synaptic inputs in this area are devoted to recurrent connections from other CA3 cells, pointing to a compact associative memory network with capacity to hold a large number of memory items. Within networks like this, memories may be stored as discrete attractors. In an attractor network, memories can be reactivated reliably from subsets of cues that were present when the memory was encoded, at the same time as interference from competing representations is minimised. Experimental evidence for competitive interactions between discrete attractor-like representations was provided and it was showed that dynamics are under strong control by the theta rhythm of the hippocampus.
It was asked whether the attractor hypothesis can be directly confronted with experimental data in the way population activity in MEC / hippocampus responds to gradual or sudden changes in the environment. If place-specific activity is directly driven by external cues one would expect that population activity would faithfully represent the current constellation of cues that define the environment. If the activity is influenced by underlying attractor dynamics, one would expect interesting history-dependent effects of the change on population activity. The precise nature of these effects could depend on the number of environments stored in the attractor network and on the history dependence of synaptic transmission in the network. It was asked if transitions between representations are instantaneous or if intermediate mixed states exist, and researchers tried to determine the minimum temporal units for representation of distinct spatial environments. The dynamics of transitional states was investigated by developing a procedure where the transition moment was 'frozen' or extended in time by slowly or unexpectedly transforming the recording environment while ensembles of place cells were recorded from CA3 and CA1 of the hippocampus. Rats were exposed to two square environments (A and B) with common physical location for several days.
The only stimuli discriminating A from B was an array of intramaze light cues. Two distinct hippocampal representations were formed during training in these environments. Population vectors were defined for every theta cycle of the concurrently recorded EEG with cut points defined as phases of minimal spike activity. For each theta cycle, population vectors were compared with those at corresponding positions of the A and B environments, using both dot product and correlation measures. The teleportation event resulted in abrupt changes in representations under ongoing theta, often with a delay of less than two theta cycles. In most cases, the change was followed by a transient oscillation between the two states, with the current representation correlating with A and B on separate theta cycles. Network flickers were extremely fast, often with complete replacement of the active ensemble from one theta cycle to the next. Mixed states were rare; distributions of cycles correlated with A and B showed significantly fewer cycles correlated with both states than expected by chance. Within individual cycles, segregation was stronger towards the end, when firing started to decline, pointing to the theta cycle as a temporal unit for expression of attractor states in the hippocampus. Observations point to theta cycles as minimum units of hippocampal representation and imply that individual cycles, but not necessarily successive cycles, tend to remain within one state. Repetition of pattern-completion processes across successive theta cycles may facilitate error correction and enhance discriminative power in the presence of weak and ambiguous input cues.
In parallel with these experimental studies, researchers modelled an attractor network that stores continuously changing environments and analysed attractor states. A complete solution to the problem of continuous change was obtained, with a phase diagram indicating the possible regimes in the network. When the contribution of the intermediate environments is increasing, the network was found to undergo a transition into a combined representation of the whole sequence of environments, such that only cells with sufficiently close fields across the whole sequence remain active. Cells with far away initial fields are inhibited and lose their contribution to representing the environments. During screening of the deep layers of medial entorhinal cortex, the participants found a new entorhinal cell type that fires specifically when the animal is close to the borders of the proximal environment. The orientation-specific edge-apposing activity of these cells was shown to be maintained when the environment is stretched and during testing in enclosures of different size and shape in different rooms. Border cells were relatively sparse, comprising less than 10 % of the local cell population, but subsequent recordings in the other layers showed that these cells can be found in all entorhinal layers as well as the adjacent parasubiculum, often intermingled with head-direction cells and grid cells.
Under conditions that favour global remapping in the hippocampal place-cell population the border cells maintained their edge-related firing activity. When multiple border cells were recorded simultaneously, they maintained their geometric relationship. The remapping pattern of entorhinal border cells is different from the hippocampus, where place cells show statistically orthogonal firing patterns in different environments. Results suggest that the entorhinal cortex is predominated by codes that express similar geometric relationships irrespective of the specific features of the environment or the experiences. The discovery of border cells in the entorhinal cortex was accompanied by the discovery of 'boundary vector cells' in the subiculum by one of the other consortium groups. The two cell types have much in common and a question for future research is to determine the relationship between them. Microdrives for multi-site ensemble recording. With the objective of recording ensemble activity from multiple brain locations it became important to develop a multi-electrode microdrive for parallel recording in which electrodes could be moved independently at a desired spacing of electrodes. The industrial partner of the consortium has been responsible for this development. The multi-site microdrives were to be composed of at least two core clusters, with inter-cluster distance specifiable by the user. A computer-controlled milling machine was employed to manufacture miniature plastic components designed using the Rhino three-dimensional (3D) computer-aided design software package. A new drive mechanism, a new stable connector, a new head stage compatible with the Microdrive connectors and a new cannula attachment mechanism were developed. The simplest single-site drive is small, lightweight (less than 1 g) and the plastic components are sufficiently simple that they can be manufactured in bulk. The drives will be easier for users to load than current drives.
4. Visualisation of functional neuronal assemblies
It was aimed to develop a novel miniaturised two-photon fibrescope for imaging of activity patterns in local neuronal assemblies in freely moving rats. By using in vivo labelling with fluorescent calcium indicators, directly visualise place cell and grid cell activity was set out as the animal was navigating through the local environment. The goal was to unravel the fine-scale functional organisation of these neurons. Electrophysiological studies of the spatial structure of cell assembly activity are constrained by the spatial resolution of the tetrode technique. Tetrodes are not suitable for studying differences in activity between neighbouring microcircuits with sharp boundaries and they are suboptimal for identifying neurons that are not active in any of the test conditions. To overcome these problems a portable high-resolution multiphoton microscope was developed which can reveal activity patterns in local neural networks with single-cell and single-spike resolution.
Three key methodological procedures have been developed as part of the project:
(1) the in vivo preparation;
(2) labelling of hippocampal neurons with calcium indicator;
(3) two-photon imaging of hippocampal neurons in vivo.
Through these developments single-cell resolution optical recordings of hippocampal neurons in anesthetised animals were established. A surgical procedure was devised in anesthetised rats to routinely access the hippocampal formation for functional optical measurements, aiming at the CA1 region to begin with. Using visually guided blunt aspiration of a small block of overlying cortical tissue, it is possible to successfully identify the intact hippocampal formation. Researchers bolus-injected the calcium-sensitive dye Oregon-Green BAPTA 1-AM (OGB-1) and the astrocyte-specific dye sulforhodamine 101 (SR101) into the CA1 region of hippocampus. Neuron and astrocyte labelling under these conditions are visible about one hour after dye injection. Third, for imaging hippocampal cells researchers placed the anesthetised animal under a standard two-photon microscope. Cells in the hippocampus were imaged through a small gradient-index (GRIN) lens lowered onto the hippocampus, which mimics the situation with a miniature two-photon fibrescope.
The surgical preparation and bolus-labelling with calcium indicator have yielded good results in anesthetised and behaving rats. Researchers reliably stained and differentiated neurons and glial cells and with the two-photon microscope have visualised cell populations from the surface of the hippocampal formation through stratum oriens down to the pyramidal cell layer. Spontaneous activity was low in all functional measurements. But after reducing inhibition using bicuculline large and synchronised calcium transients were onsevered, presumably representing epochs of action potential activity in populations of pyramidal cells. These results directly demonstrate large-scale calcium indicator labelling and imaging of cell populations in the hippocampus in vivo. This is a prerequisite for all future imaging with this approach and clearly indicates that the ultimate goal is in reach.
5. Temporal codes in entorhinal cell assemblies
It was aimed to determine mechanisms and functions of theta phase precession. It can be argued that theta phase precession provides the most prominent experimental clue to sequence coding in the hippocampus. Although phase precession provides the hippocampus with a mechanism for storing extended sequences of places and experiences, the neuronal circuits and computations responsible for this effect are unknown. During the early stage, it was showed that phase precession is expressed also in grid cells in layer II (but not layer III) of medial entorhinal cortex, which is the origin of projections to the dentate gyrus and CA3. The presence of phase precession in one of the entorhinal inputs to the hippocampus points to the entorhinal cortex as a possible origin of phase precession. Found was that phase precession exists in principal cells of the medial entorhinal cortex and is generated there independently of the hippocampus, and computational models were developed to suggest mechanisms for how phase precession and grid patterns might emerge from intrinsic membrane properties of these neurons. Shown was how theta oscillations may contribute to formation of grid patterns and how these depend on specific velocity inputs to the region. A cellular mechanism for possible transformation has also been identified.
During the first year, researchers reported that grid cells in layer II of medial entorhinal cortex express phase precession. Layer III cells generally do not show this. The aim was to establish precession in layer II cells origates locally in the entorhinal cortex or is derived from the hippocampus. The presence of phase precession in medial entorhinal cortex would be consistent with the subthreshold oscillations of entorhinal stellate cells in layer II, which are faster than the extrinsic theta oscillations and so may cause spike activity to advance progressively across the theta cycle in a manner similar to the dual oscillator mechanism proposed for hippocampal place cells. An alternative would be that phase precession originates in the hippocampus or locally in both hippocampus and entorhinal cortex. Researchers distinguished between these possibilities by recording theta phase in entorhinal grid cells after more global inactivation of the dorsal hippocampus. If phase precession in entorhinal grid cells is inherited from hippocampus, precession in medial entorhinal cortex should be disrupted by inactivation of the hippocampus. An experiment showed that entorhinal phase precession is not blocked by inactivation of the hippocampus, suggesting that the phase advance is indeed generated in the grid cell network. The results pointed to possible mechanisms for grid formation and raised the possibility that hippocampal phase precession is inherited from entorhinal cortex. The phase-precession studies were accompanied by computation work in which an earlier model of phase precession in hippocampus was revisited. In this model, phase precession arises as a consequence of propagation of activity across place cells at each theta cycle due to asymmetry in synaptic connections in the direction of animal motion. STP was introduced in recurrent connections and effects on phase precession were analysed. It was found that synaptic depression that characterises inter-pyramidal connections in hippocampus can cause activity propagation in the network even when synaptic connections are symmetric and that during sharp-waves, replay of sequential activation of place cells can emerge.
An oscillatory interference model of grid cell formation was developed to formally identify the relationships of the theoretical membrane potential oscillations with the variables that can be experimentally measured in freely moving animals expressing spatially organised grid cell firing. Consortium participants have worked out the relationship of theoretical variables with the intrinsic firing frequency of grid cells as reflected in the spike train temporal autocorrelogram, and the theta rhythm of the extracellular local field potential. Support for a specific relationship between running speed and oscillation frequency in the theta-frequency band was obtained in a study. Recording in grid cells from mice with forebrain-specific knockout of HCN1, it was demonstrated that the dorsal-ventral gradient of the grid pattern is preserved in HCN1 knockout mice, but the size and spacing of the grid fields, as well as the period of the accompanying theta modulation, expanded at all dorsal-ventral levels, suggesting that HCN1 channels control the scale of the grid pattern. The speed-dependent modulation of theta frequency was strong in control mice but nearly absent in HCN1 knockout mice, implying that I(h) currents are critical for transformation of speed signals to spatially periodic firing fields. There was no change in theta modulation of simultaneously recorded entorhinal interneurons, suggesting that the expansion of grid scale is caused by lack of HCN1 channels in the grid cells themselves.
6. Development of a spatial representation system
The objective was to determine how key components of the spatial representation system develop relative to each other, how the emergence of one function leads to the next, and how the maturation of the system interacts with experience of the animal during postnatal development. One study had systematically explored the development of spatial representations and the focus of this study was limited to the later stages of development in place cells. The development of entorhinal-hippocampal intrinsic connectivity was mapped. Electrophysiologal recording studies in two to three-week old rats have shown that entorhinal-hippocampal connections develop earlier than researchers thought and we know that a rudimentary spatial representation is already present at the time when rats make their first navigational experiences outside the nest (week 3). Place cells are present from the beginning and rudiments of grid cells can be seen too, although the number of grid cells with clear and stable firing fields takes longer time to develop.
Retrograde tracing studies revealed that the reciprocal connectivity between entorhinal cortex and hippocampus is present at birth. Preliminary voltage sensitive dye imaging data indicated that stimulation of entorhinal cortex results in adult-like patterns of activation at postnatal day P9, whereas only limited local activation can be seen at P5. Interactions between superficial and deep entorhinal layers are noticeable at P9 but not at P5. Intrinsic hippocampal circuits show slow development between P9 and P14. Results indicate the presence of functional connectivity earlier than has been assumed. With the use of in vitro slice preparations and voltage sensitive dye imaging, was established that inputs from pre and parasubiculum are not functional directly after birth. Intrinsic connectivity was investigated with simultaneous whole cell recordings from groups of 3 or 4 medial entorhinal stellate cells in horizontal brain slices from pups of different ages. Data show that connections between stellate cells in the early postnatal stages are sparse. A striking difference is the amount of synchronous firing of stellate cells. Whereas synchrony is limited during the early postnatal stages (P16-P21), this disappears later on.
These findings indicate that connectivity between stellate cells is set up during the early postnatal period and depends on monosynaptic connectivity between stellate cells and either the presence of common inhibitory inputs or common excitatory cortical inputs. The only study of place cells from young rats published before SPACEBRAIN suggested that place fields develop late, stabilising only around P52.
Recording in parallel from medial entorhinal cortex and hippocampus of the developing rat, two of the consortium groups set out to determine the following:
(i) If the late maturation of place cells in the early study was caused by an equally slow development of grid cells, which are thought to provide the spatial input to the place cells.
(ii) If grid patterns are expressed in medial entorhinal cortex at earlier stages of development, in which case the late expression of place cells may reflect slow maturation of the hippocampus itself.
Recordings from the medial entorhinal cortex were expected to establish whether the coherent firing of cells with similar spatial phase is present in the grid cell network from the beginning or develops after grid fields emerge in individual cells. It was found that a rudimentary map of space is present when 2.5-week old pre-weanling rats explore outside the nest for the first time. This map contains both place cells and grid cells but whereas place cells were present from the very beginning, the number of grid cells with strictly periodic fields remained surprisingly low until three to four weeks of age. Results showed convergent results for place cells and grid cells.
Both labs investigated development of head direction-modulated cells. Directionally modulated cells in the presubiculum were present in the earliest recordings. Head-direction cells in this region had adult-like properties from the outset. Directional firing preferences were stable within and between trials and differences in directional preferences between different cells were maintained. Slower development was observed in the entorhinal cortex. Early presence of head direction cells suggests that these cells may be instrumental in setting up the place and grid representations in the hippocampus and entorhinal cortex. Whether the rudimentary spatial periodicity of the earliest grid cells is sufficient to enable place-cell formation in the hippocampus remains to be determined. Joint results suggest that a preconfigured neural circuit for spatial representation is present at the initiation of outbound exploratory behaviour. The maturation of this neural circuit continues in parallel with the animal's first navigational experiences. Place and grid cells develop relatively independently of experience. First-time exploration was compared in animals at early and late states of ontogenetic development. The young group explored an open environment at the time when ratS normally begin leaving the nest. Older animals walked for the first time during the mid P20s. Place and grid fields were similar at comparable ages despite the relatively lengthy spatial experience of the former group. Data suggest that the presence of rudimentary forms of place and grid cells is determined primarily by maturational factors. The findings provide the first experimental support for Kant's more than 200-year old conception of space as an a priori faculty of mind.
Grid cells in the medial entorhinal cortex have been proposed to hold a continuous attractor-based representation of the animal's changing location in the environment by excitatory recurrent connections between cells with similar spatial phase; but does the development of structured connections precede or follow the development of grid fields in individual units? The view that grids reflect metrically organised connectivity in the adult tissue, is challenged by the fact that spatial phase is represented non-topographically in the grid cell network, so that connections between cells which later develop similar spatial phase need to be ordered ad hoc. An alternative possibility is that distant cells wire together instructed by an external 'teaching input' present only temporarily during development. As a result they develop similar spatial phase. The nature of such an input and its localisation in the brain is not known. Using a computational approach it was demonstrated that the conceptual feasibility of a model in which individual grid units first develop their multi-peak fields, and later those with similar spatial phase develop enhanced recurrent connections through Hebbian plasticity. In one of the models of grid formation the grid pattern of activity is elicited in individual units directly by the afferent inputs, conveying spatial information relative to one environmental context. The grids are co-aligned across different units if these are made to interact via recurrent weights modified with a model plasticity rule and may be suited to conjunctive units in the deeper mEC layers. Most microdrives are relatively large and heavy, making the ensemble difficult to carry for the smallest animals. A miniature multi-electrode Microdrive was developed which contains a prototype headstage with no AC coupling components and smaller op-amps. Another prototype was developed which reduces in the size of the tether cable between the animal and the recording system
Potential impact
Recent advances in our ability to record and analyse complex patterns of activity in specific brain circuits will lead towards a fragmented mechanistic appreciation of complex experience.
This will benefit European citizens in a number of ways:
- prevention and treatment of neurological diseases where spatial orientation and memory are impaired;
- a technological revolution in the analysis of neuronal microcircuits and the development of industry that can supply the new tools;
- new insight into algorithms that could be used for design of artificial navigating agents.
Translational potential and contribution to human health
An improved understanding of the cerebral activity responsible for spatial representation and memory is likely to have major consequences for improving the health of European citizens. Research on the functions and the organisation of the entorhinal cortex and its development has immense relevance for a number of devastating diseases and the ways we manage them. In the long run this knowledge is essential for improving diagnosis and treatment of diseases that affect the entorhinal-hippocampal space circuit, as well as other cortical circuits. An important target for translational research is Alzheimer's disease. Volume reduction of the entorhinal cortex is now considered a reliable measure for identifying individuals at risk for Alzheimer's disease. Functional insight into the working principles of the entorhinal cortex provides essential basic knowledge for early diagnostics, prevention and treatment. Results provide key insights into the operational principles of the normal entorhinal-hippocampal system. The consequences of early treatment of Alzheimer's disease for the population of European citizens are likely to be enormous. The economic benefits of reduced hospitalisation and institutionalisation would be enormous if these disorders could be diagnosed adequately or even prevented or cured.
Impact of the technological development
Several novel research tools have been developed. This benefits the European community both on scientific and commercial level. One of the developments resulted from SPACEBRAIN is the production of a user-customisable multi-core split-bundle microdrive which could enable researchers to collect data from large neuronal ensembles in distributed brain regions. As knowledge advances, ensemble recording is becoming an increasingly important tool. European researchers will gain a benefit by the availability of new technology for high-density single-unit recording from multiple areas. Functional imaging using fluorescence microscopy has become a standard technique for measurements in extracted tissues and brain slices, and the technology has contributed invaluable data for the characterisation of the physiology of individual neurons. Fibre-optic imaging tools for optical imaging in freely moving animals constitutes the next major step in optical recordings of neural activity and most likely will have impact on the research field. Also the pharmaceutical industry may become interested.
Exploitation: Relevance of proposed work to robotics and artificial intelligence (AI)
Establishing the network properties of brain systems offers the possibility of contributing to the development of artificial systems, particularly via the twin disciplines of robotics and artificial intelligence (R/AI). The solution to problems demonstrated in experimental work from the project is to use two types of information in tandem: external landmarks as a means of determining instantaneous position, and motion cues to update the position estimate, with periodic recalibration from landmarks. Despite decades of research, robotics has failed to develop a competent navigating agent. One problem may be that R/AI designers have not found a way to integrate static and motion cues in the way that animals do. The project has provided evidence for a neurobiological model according to which motion information in animals is processed via attractor networks instantiated in entorhinal cortex and hippocampus, and where landmark information interacts by 'resetting' the attractor when it is either undefined or has drifted due to accumulated errors from the motion detectors. The attractor networks would constitute the 'common domain' referred to above that allow multiple sensory modalities to converge on a single representation of place. A crucial feature in the mammalian design is the coding in a multiplicity of self-organised memory representations in the hippocampus.
While motion cues have a fixed meaning throughout the animal's lifetime, landmarks provide signals that are completely context-dependent and it would be disastrous to try to implement their fusion with motion cues by means of fixed registers. Evolution has addressed this problem by largely dedicating to landmark representation the most flexible system for generating arbitrary representations of virtually any content and immediately storing them in the hippocampus. The large storage capacity necessary for memorising a large variety of spatial contexts is ensured by the large number of independently modifiable synapses impinging on any given pyramidal cell - of the order of 104 'knobs' on each neuron, which can be tuned separately while exploring a new environment. This flexibility stands in marked contrast to the rigidity of representations in the nearby entorhinal cortex. What the recent discoveries are beginning to illustrate is a counterintuitive solution whereby motion cues are represented by rigid, essentially static codes, whereas static cues are dynamically assigned to new, unpredictable representations. The discovery of neurobiologists that attractors may lie at the heart of place representation suggests that R/AI may benefit from designing computationally analogous processes into their artificial navigating agents. It will be possible to extend biologically inspired robotic design into the sensor fusion domain, which can greatly enhance the design of artificial navigating agents.
Dissemination activities
Scientific results have been disseminated to the broad science community, industrial interests and the lay public. Several papers with SPACEBRAIN funding have been published or accepted for publication or are likely to be published during the coming years. The publication strategy has been to maximise the impact of the publications by withholding data until a relatively complete and coherent understanding is obtained and all interpretations are well supported by the data. Before formal publication in scientific journals, the results have also been made available as early as possible in preliminary form to the general neuroscience community at scientific meetings. An important objective was to inspire computational modelling of the hippocampo-entorhinal circuit. To enable the worldwide community of computational neuroscientists to develop models that match data as closely as possible. Shared data are available as Matlab files and include spike and position times as well as behavioural tracking data for a number of simultaneously recorded grid cells. To explore whether network processes similar to those studied in SPACEBRAIN can be built into navigating robots, we organised a workshop which was held during the final months of the consortium in order to maximise the transfer of mechanistic insight from mammalian microcircuits to the electronic circuits of artificial agents.
All participants in SPACEBRAIN have been committed to better public understanding of science and have been engaged in a number of activities relevant to this. Several participants are members of the European Dana Alliance for the Brain (EDAB). The mission of EDAB, and its American counterpart the Dana Alliance, is to inform the general public and decision-makers about the importance of brain research, to advance knowledge about the personal and public benefits of neuroscience, and to disseminate information on the brain in health and disease in an accessible and relevant way. IEDAB organises the yearly Brain Awareness Week, where hundreds of public events and activities worldwide bring scientific progress in the neurosciences out to the general public. We have used the Brain Awareness Week to disseminate knowledge about spatial representation and navigation, including results from the project. Schools and journalists have been contacted for informal talks about the brain at several of the participant institutions, and public lectures have been arranged. All major publications have been accompanied by press releases and journalists of national newspapers have been contacted by the authors. Members of the consortium are regularly in contact with newspapers and TV channels to increase public understanding of brain functions, brain mechanisms and brain disease.
Project website: http://www.ntnu.no/cbm/spacebrain
Some 42 months of collaboration have produced significant advances in our understanding of how space is represented in neural networks of the entorhinal cortex and hippocampus, brain regions known to form internal maps of the local environment. Two papers with SPACEBRAIN funding have been published or accepted for publication in Nature, three in Science, one in Cell, and several are appearing in other high-impact journals such as Nature Neuroscience and Nature Methods. Key scientific advances include, first of all, new insights into the intrinsic wiring of the entorhinal cortex. The project has shown how grid cells (one of the key cells types) connect to other cell types in the circuit and in neighbouring regions, and we have learned that grid cells are strongly linked by inhibitory connections. Grid cells are not to be the only space-coding cell type. Border cells are entorhinal cells that fire along geometric edges and landmarks, whereas boundary vector cells in the subiculum fire at specific distances from such boundaries. The project showed that space-coding cells develop a lot earlier than researchers thought before and that a rudimentary spatial representation is present already when rats make their first navigational experiences outside the nest, at the beginning of the third postnatal week.
Among the known spatial cell types, direction-coding cells appear first, probably before the animals open their eyes for the first time. They are followed by place cells and grid cells, which need further maturation but nonetheless appear adult-like one-to-two weeks after the beginning of independent exploration. The SPACEBRAIN project has shown that grid cells and place cells encode horizontal and vertical space differently, with stronger spatial resolution in the horizontal dimension. The project has also shown that hippocampal neuronal assemblies follow principles of attractor networks and are organised within 125-ms time blocks referred to as theta cycles. Theoretical work has addressed the significance of this pattern of organisation.
The scientific progress has been accompanied by development of new research tools. Microdrives have been developed for recording electrical activity at multiple brain locations at the same time, and they have been miniaturised for recording in small animals. A portable high-resolution microscope has been developed for population studies of local neural networks during behaviour. The project reflects several unique features of SPACEBRAIN. One is the integration of experimental and theoretical work. From the beginning, experimentalists and modellers have been working on complementary subprojects, exchanging information as soon as data or models were available. Another is the explicit attempt to combine methods and concepts from multiple levels of analysis, ranging from mechanisms of synaptic transmission to interactions between large distributed circuits. The success is likely to reflect the open and positive working atmosphere of the consortium. Trust and an open attitude have laid the basis for constructive discussion and true collaboration.
Project context and objectives
Neuroscience is at a stage where integrative understanding of brain function can be reached. Tools for a cross-level analysis centred on microcircuit structure and activity in neuronal assemblies are becoming available and it has become possible to model large networks with connectional and dynamic properties similar to those of real brain circuits. This brought models closer to experimental observations and the benefits of mutual interaction between theoretical and experimental communities became apparent. The principal objective of SPACEBRAIN was to exploit these converging conceptual and technological developments to uncover some of the fundamental principles of neuronal assembly and microcircuit operation in the mammalian cortex. Not all areas of cortical investigation have matured to the stage where neuronal computation can be understood at the microcircuit level. Insights into circuit functions have been obtained for the early stages of sensory systems where the processing of signals can be followed through networks of increasing complexity from the receptor level to the primary sensory cortices.
These studies have largely clarified how neurons and neuronal networks extract features from the external world and refine these into cortical representations. How the brain generates its own codes, in the non-sensory parts of the cortex, has remained deeply mysterious. A new path was opened up when the grid cells of the entorhinal cortex were discovered by one of the participating groups in 2005. Grid cells are place-modulated neurons whose firing locations define a periodic triangular array covering the entirety of the animal's environment. They are thought to form an essential part of the brain's coordinate system for metric navigation. Grid cells have attracted enormous attention because the crystal-like structure underlying their firing fields does not arise out of the simplicity of their components, as it does in more familiar physical systems, but rather out of network complexity. It is a structure not imported from the outside world but created within the nervous system. Grid cells provide scientists with a direct window into some of the most fundamental operational principles of cell assemblies and microcircuits in the brain. The grid cells are part of the brain's system for spatial representation, which includes the hippocampus and the entorhinal cortex.
Through the discovery of place cells by one of the consortium groups in 1971 and the discovery of grid cells by one of the other groups in 2005, the hippocampo-entorhinal network is now, in the view of many neuroscientists, emerging as the brain system with the best potential for generating a comprehensive cross-level understanding of how non-sensory neural networks compute in the mammalian brain. We know that place cells form maps of the animal's local environment and that different maps are stored in the hippocampal circuit for different environments. These maps are different from the spatial maps generated by grid cells in the entorhinal cortex. Grid cells keep a constant spatial relationship across all environments; two grid cells with overlapping vertices in one environment will have overlapping vertices in another environment as well.
The entorhinal keeps a similar ensemble structure across all places. Hippocampal place-cell maps are environment-unique and strictly tied to memory functions of this brain structure. Studies of mechanisms by which neuronal circuits and cell assemblies in these two brain regions keep track of the animal's location as it is moving around in the environment may uncover general computational mechanisms employed widely in the cortex, not only in the entorhinal and hippocampal cortices. The aim of SPACEBRAIN was to begin the search for principles of microcircuit computation in the relatively accessible spatial representation system of rodents, using a powerful combination of novel computational, electrophysiological, optical and molecular research tools that have never been applied to the analysis of brain circuitry before.
Specific objectives
The consortium set out to provide a better understanding of how spatial location is computed and represented in microcircuits of the hippocampus and entorhinal cortex and how such representations arise during development of the nervous system.
The central objectives were as follows:
1. To provide a wiring diagram of entorhinal microcircuitry that contains sufficient quantitative detail to lay the foundation for a new generation of realistic computational models of dynamic spatial representation. The potential power of computational models in accounting for spatial representation in assemblies of grid cells has introduced a strong need for detailed quantitative analysis of the microcircuitry of the entorhinal cortex. A key objective of the project was to outline the intrinsic organisation of the area, with emphasis on its cell types, their intralaminar and interlaminar connections and the modifiability of the principal interconnections.
2. To determine how intrinsic and extrinsic inputs combine into a single spatial representation in entorhinal grid cells. Using a combination of experimental and theoretical approaches, the consortium set out to determine how motion-related information is combined with visual information about landmarks and geometry into a coherent representation of the animal's location in space.
3. To determine the mechanisms of neuronal assembly formation in hippocampus and entorhinal cortex. Experimental and theoretical approaches were combined to establish the operational principles of cell assemblies in place cells of the hippocampus and grid cells of the entorhinal cortex. Large numbers of cells were recorded simultaneously in the CA3 region of the hippocampus in order to determine the mechanisms for transformation between discrete spatial representations.
4. To determine the mechanisms of cell assembly formation in the temporal domain. The consortium used a combination of experimental and computational methods to determine the origin and neuronal mechanisms of hippocampal theta phase precession, one of the brain's clearest signatures of temporal sequence organisation.
5. To determine how spatial representations are formed in hippocampo-entorhinal cell assemblies during development. The consortium set out to establish how key components of the hippocampo-entorhinal representation system develop relative to each other during the first postnatal weeks, what their interrelation is, and whether and how maturation of the system interacts with early experience. Key objectives were to determine when connections between different stages of the hippocampal formation, the entorhinal cortex and the presubiculum reach adult patterns, whether different functional cell types develop in particular sequences, and whether this development reflects mainly maturational factors or is shaped by experience.
Understanding microcircuit function is strongly dependent on the development of new research tools. An additional objective of the SPACEBRAIN project was therefore to use and develop methodologies for behavioural assessment, computational modelling and high-density single-cell recording, as well as gene silencing and single-cell optical recording. Particular emphasis was placed on the development of multi-site recording technology for high-density studies of cell assemblies, as well as the development of a head-mounted two-photon fluorescence microscope for dynamic visualisation of cell assemblies. These technical developments were accompanied by the attempt to develop a new generation of anatomically and physiologically realistic computational models of entorhinal-hippocampal neural circuits. The project was motivated by its potential for translation to both clinical and industrial applications.
Topographical disorientation is one of the earliest and most prominent behavioural symptoms of Alzheimer's disease. Because the entorhinal cortex is the first brain region to be affected in many patients with Alzheimer's disease, functional insight into the spatial representation mechanisms of this brain region should provide clinical neurologists and health workers with essential tools for early diagnostics, prevention and treatment of Alzheimer's disease. Insights may not only reduce the prevalence of Alzheimer's disease and the enormous health care burdens associated with the disease, but may also have an important preventive impact on traffic accidents and human error in spatially demanding man-machine interactions caused by Alzheimer's disease at pre-diagnostic stages. Understanding space coding in the brain may also exert significant impact on the development of navigating robots. Despite decades of research, robotics has so far failed to develop a navigating agent capable of exploring an unknown or complex landscape at an efficiency that is anywhere near to that of a small rodent. By uncovering some of the fundamental algorithms of neural networks for spatial navigation in the rodent brain, the project is likely to provide the robotics industry with entirely new concepts for design of artificial navigating agents.
Project results
The project has led to significant advances in understanding of how space is represented in neural networks of the entorhinal cortex and hippocampus. Organisational principles of entorhinal circuit structure have been determined, back projections from the hippocampus have been shown to be essential for stable grid firing in the entorhinal cortex, grid scale has been shown to depend on the velocity of the moving animal, spatial cell types have been observed during early postnatal development, and cell ensembles have been found to operate as attractor networks within time frames defined by the hippocampal theta rhythm. These and many other findings are outlined in the section below. The work was organised in six work packages (WPs):
1. Attractor dynamics and circuit organisation in entorhinal cortex
The first scientific WP tested the idea that the entorhinal representation of self-location is based on continuous attractor dynamics. Attractors become continuous when the distribution of input cues to the system is continuous. In a continuous attractor network, a bump of activity is formed at a given location in the network defined by the input pattern. This bump of activity is then thought to move around in the network space in accordance with the speed and direction of the animal's movement in external space. It has been suggested that entorhinal networks have the ability to support the formation and translocation of attractor bumps. The continuous attractor idea makes a number of explicit assumptions about the underlying cellular and synaptic architecture and neural dynamics of the entorhinal cortex but the assumptions generally lack experimental support. The weak understanding of details of intrinsic wiring diagram of the entorhinal area, and physiological properties of cell assemblies in this network, motivated this WP. We sought a quantitative characterisation of the entorhinal microcircuitry. This was essential for the development of realistic computational models of spatial representation in the hippocampal-entorhinal system. The general aim was to provide such a characterisation and to test the basic assumptions of attractor models of the entorhinal cortex.
The consortium has acquired significant insight into the intrinsic wiring of the medial entorhinal cortex. Paired recordings from layer II principal cells have suggested that connectivity between principal cells within the layer is weak, averaging around 1 % of cell pairs or less, but it was shown that there is strong connectivity between principal cells and interneurons, pointing to alternative ways of synchronising the circuit. Targeted genetic manipulations have demonstrated that GABAergic interneurons in the medial entorhinal cortex project not only within the entorhinal circuit itself but also to subfields of the hippocampus, providing a direct means of synchronising the two structures with high temporal precision, such as during gamma network oscillations. The experimental work has been accompanied by theoretical work addressing the principles of attractor dynamics in hippocampus and entorhinal cortex, pointing the way to new experiments after completion of the SPACEBRAIN project.
Recurrent interconnectivity can in principle be present in cortical networks at two organisational levels, either within layers (intralaminar) or across layers (interlaminar). Previous studies have shown widespread interlaminar axonal arborisation of principal cells, connecting deep with superficial layers, as well as intralaminar connections in layers II, III, and V; however, the precision and detail of these studies have been far below the levels needed for realistic computational modelling of entorhinal networks. More than a hundred neurons in all layers of medial entorhinal cortex electrophysiologically were analysed through whole-cell current-clamp recordings, at the same time as these neurons were stained in order to determine their unique connection patterns. Excitatory interlaminar connections were found to arise from neurons in layers III and V but the study also showed that these connections may arise from layer VI principal cells. Based on an analysis of more than 500 potential pairs of stellate cells at different ages, it was shown that the ratio of monosynaptically coupled stellate cells in layer II is small. Less than 1 % of the recorded cell pairs had direct synaptic connections.
Connectivity ratios did not increase when cells were at larger distances from each other but stayed within the established average axonal arbor of 550 μm. Connections between principal cells and interneurons in layer II were substantially stronger. These findings indicate that the proposed excitatory interconnectivity underlying pure grid cell properties is weak and that, while inhibitory connections may play a primary role in synchronisation of the layer II network, the continuous attractor models accounting for grid activity in that layer need major revision. A new hypothesis has been formulated in which a denser recurrent connectivity in the deeper layers establishes coherence among conjunctive head direction/grid units, which then feed into a non-recurrently connected pools of pure layer II grid units to achieve sharper spatial resolution; a hypothesis that will be tested in computational models after the completion of the project. Behavioural tests in mice where the expression of Cre-recombinase is under the control of the parvalbumin promoter suggested that fast-spiking interneurons may be instrumental in synchronisation of network activity as well as the formation of precise spatial representations.
A critical assumption of continuous attractor models of grid cells in the entorhinal cortex was that descending and ascending connections between layer II and the deeper layers are asymmetric. Such asymmetry is necessary if conjunctive cells in the deeper layers are to translate the grid cell representation in layer II in accordance with the animal's movements. A key objective of the circuit analysis was to determine if asymmetry was present. We focused on connectivity between layers V and II. Out of more than a hundred individually stained neurons, approximately a third show a clear lateral shift in their axonal arbor, when compared to the dendritic arbor, in agreement with the translation idea. The findings suggest that also neurons in layers III and VI may have laterally shifted projections. Previous theoretical work indicated that cooperative activity of neocortical circuits is sensitive to short-term synaptic plasticity (STP) in connections between pyramidal neurons. Synaptic depression can generate fast changes in activity, while synaptic facilitation on the opposite can stabilise attractor states. An important objective of SPACEBRAIN was to determine the validity of this idea. The role of STP in stabilisation of attractor networks was tested in a 'ring model' applied to place and grid cells as well as a range of other problems. The results show that synaptic facilitation can dramatically slow down the drift velocity of the network, thus appearing as a strong stabilising factor in representation of space and memory.
2. Calibration of entorhinal assembly representation
Location-specific firing of grid cells is thought to be updated principally by a path integration-based mechanism, based on computations derived from the animal's movements through space. This suggestion is consistent with the constancy of grid maps across environments, their imperviousness to removal or displacement of external landmarks and the fact that grids are expressed immediately in a novel environment. While the basic structure of the grid may be determined by self-motion cues, grids are clearly influenced by external visual cues as:
(i) both phase and orientation of the grid are controlled by specific landmarks in the environment and
(ii) re-sizing of a familiar environment causes partial re-scaling of the entorhinal grid in the scaled dimension.
These observations implicate an interaction between path integration signals and visual inputs. The general objective was to determine how this integration is accomplished. A combined experimental-theoretical approach was taken in order to understand the relationship between motion inputs and other sensory inputs and the formation of grid cell patterns. The results provide important insights. It was shown with experimental approaches that back projections from the hippocampus are essential for maintaining stable grid representations, although they may not be required for the ensemble structure of the grid cells as such. Experiments have demonstrated that theta modulation of grid cells is dependent on the linear velocity of the animals, and that place and grid cells represent horizontal and vertical space differently, and computational modelling has investigated the significance of hippocampal back projections and other inputs in generating stable spatial periodicity. Previous work has shown that grid cells rescale when animals are placed in novel environments but the underlying mechanisms were not understood well. Experiments to tell us whether the grid expansion in a novel environment only applies in the absence of visual and other environmental cues we designed. Results suggest that grid expansion takes place even in novel environments with strong visual and odour cues. Continued testing in novel environments leads to relaxation back to the original grid scale suggesting that the expansion is due to novelty and not the absence of environmental information and that recalibration back to original scale takes place as new environments become familiar. Grid fields are larger in novel environments than they are in familiar environments, consistent with the notion of scaled-up grids in novel environments.
Using a combination of experimental and computational approaches, the research aimed to establish whether hippocampal back projections to the entorhinal cortex are necessary for maintaining and resetting grid representations in the entorhinal cortex. A role in maintenance and updating of representations would be expected if associations between path integrator coordinates and extrinsic landmarks were stored in the hippocampus and back projections to the entorhinal cortex were necessary for stabilising the grid throughout the environment. Associations between path integrator outputs and fixed landmarks might be formed within the medial entorhinal cortex itself by combining grid activity with specific sensory inputs received. To test these possibilities against each other we inactivated the hippocampus temporarily by infusing muscimol into both hippocampi before grid cells were recorded in a standard foraging condition. Hippocampal inactivation resulted in gradual decrease in firing rate of the grid cells, followed by disintegration of the grid pattern. The temporal coherence of the grid cell population remained stable, suggesting that the grid structure persisted and that the alignment to landmarks deteriorated. The results suggest that grid cells are dependent on stored associations in the hippocampus to anchor the grid to the external environment.
Output from the hippocampus may provide a resetting mechanism that constantly updates the network in the entorhinal cortex with regard to stored landmark features. This may be vital to the operation of oscillation-dependent mechanisms for grid cell formation; without frequent resetting, noise and variation in the period of the theta rhythm may be detrimental to the formation of spatial periodicity. During the project, simulations have been conducted in the context of the oscillatory interference model of grid cell firing. Under this model, the spatial organisation of the firing of grid cells arises from the interference between sub-threshold membrane potential oscillations in the theta band: the firing rate increases when the oscillations are in phase and decreases as they go out of phase. The model performs path integration by assuming that the frequencies of oscillations each reflect velocity in a specific direction. This scheme is susceptible to the accumulation of error in the inputs conveying velocity information. The simulations investigated how spatial stability can be attained by phase-reset of the oscillations in two ways. The first mechanism involves a phase-reset of all oscillations to be in-phase at a specific location, triggered by input from the place cells which fire at that location, given that the firing location of place cells is maintained in location by environmental sensory inputs. Spatial stability of grid cell firing was successfully attained under this model. The second mechanism involves a reset of all oscillations triggered by inconsistency between multiple oscillators. When velocity along more than two directions is integrated by multiple oscillators, different pairs of oscillators can produce different estimates of location. When these estimates become mutually inconsistent, all oscillators can be reset to correspond to the mean estimate. This provides a mechanism for error correction that is potentially intrinsic to entorhinal cortex. The effects of the two mechanisms were compared as a function of increasingly noisy estimates of velocity.
An objective was to seek the source of cortical and subcortical inputs to the grid cell system using anatomical tracing methods and to determine what sensory modalities influence the metric of the grid, particularly its spacing. Among the candidate inputs to the grid cell area in the medial entorhinal cortex, researchers focused on the retrosplenial cortex and interactions between inputs from this region and from postrhinal and pre-parasubicular cortical regions. Research showed that projections to layer V or medial entorhinal cortex originate mainly from ventral subdivision of retrosplenial cortex generally referred to as granular part A. Electron-microscopy analyses shows that this input is mainly excitatory, targeting mainly principal cells. The more dorsal subdivision b sends a massive input to layer III of the presubiculum, targeting preferentially principal cells that project to medial entorhinal cortex. Projections from the postrhinal cortex show a striking topographical organisation in that a narrow area of postrhinal cortex, close to the border with medial entorhinal cortex, projects to the directly adjacent portion of medial entorhinal cortex, which contains grid cells with relatively small firing fields. More dorsal portions of postrhinal cortex project more ventrally towards the area of medial entorhinal cortex where much coarser grid cells have been recorded. Postrhinal inputs target principal cells in layers II and III that project to the various subfields in the hippocampus. The majority of these projections are excitatory, targeting spines of projection neurons in the medial entorhinal cortex. Axons from the presubiculum target entorhinal neurons layers II, III and V, which project to the hippocampus. Inputs from pre- and parasubiculum converge onto single principal cells in the medial entorhinal cortex. In conjunction with the observations on retrosplenial inputs, the data break with older ideas in that information from retrosplenial cortex is shown to reach superficial layers of medial entorhinal cortex through a massive di-synaptic pathway involving layer III cells in presubiculum. The interaction between static and motion cues on place cells was explored in behaving animals on a linear track in which static somatosensory cues were shown to act as foci for the formation of place fields. During the project, these path integration studies were extended to the vertical dimension, by recording place and grid cells as animals explored environments having height as well as horizontal dimensions. It was found that place and grid cells show little modulation by height, suggesting that the basic frame of reference for the grid cell map is planar rather than volumetric. Behavioural studies showed that this anisotropy of encoding is mirrored in behavioural performance, in which rats also show a horizontal bias in both foraging and detour tasks.
3. Interactions between entorhinal and hippocampal cell assemblies
The aim was to establish mechanisms for interactions between neuronal assemblies within the hippocampus and between the hippocampus and the entorhinal cortex. The project focused on neuronal coding in large populations of hippocampal and entorhinal neurons. Experimental data generated have demonstrated that hippocampal assemblies follow principles of attractor networks and that attractor states are expressed within individual theta cycles, with relative independence between successive theta cycles. Competitive interactions between different ensemble states have been observed at the sub-second time scale during transitions from one representation to another. New microdrives for multi-tetrode ensemble recording with independently movable tetrodes, constructed by the industrial partner of the consortium will as soon as they are released on the market allow participants inside and outside the consortium to address large-scale population phenomena that were not readily accessible until now. Findings have been incorporated in computational models of attractor dynamics. The CA3 subfield of the hippocampus has many properties of an associative memory network. Most of the numerous synaptic inputs in this area are devoted to recurrent connections from other CA3 cells, pointing to a compact associative memory network with capacity to hold a large number of memory items. Within networks like this, memories may be stored as discrete attractors. In an attractor network, memories can be reactivated reliably from subsets of cues that were present when the memory was encoded, at the same time as interference from competing representations is minimised. Experimental evidence for competitive interactions between discrete attractor-like representations was provided and it was showed that dynamics are under strong control by the theta rhythm of the hippocampus.
It was asked whether the attractor hypothesis can be directly confronted with experimental data in the way population activity in MEC / hippocampus responds to gradual or sudden changes in the environment. If place-specific activity is directly driven by external cues one would expect that population activity would faithfully represent the current constellation of cues that define the environment. If the activity is influenced by underlying attractor dynamics, one would expect interesting history-dependent effects of the change on population activity. The precise nature of these effects could depend on the number of environments stored in the attractor network and on the history dependence of synaptic transmission in the network. It was asked if transitions between representations are instantaneous or if intermediate mixed states exist, and researchers tried to determine the minimum temporal units for representation of distinct spatial environments. The dynamics of transitional states was investigated by developing a procedure where the transition moment was 'frozen' or extended in time by slowly or unexpectedly transforming the recording environment while ensembles of place cells were recorded from CA3 and CA1 of the hippocampus. Rats were exposed to two square environments (A and B) with common physical location for several days.
The only stimuli discriminating A from B was an array of intramaze light cues. Two distinct hippocampal representations were formed during training in these environments. Population vectors were defined for every theta cycle of the concurrently recorded EEG with cut points defined as phases of minimal spike activity. For each theta cycle, population vectors were compared with those at corresponding positions of the A and B environments, using both dot product and correlation measures. The teleportation event resulted in abrupt changes in representations under ongoing theta, often with a delay of less than two theta cycles. In most cases, the change was followed by a transient oscillation between the two states, with the current representation correlating with A and B on separate theta cycles. Network flickers were extremely fast, often with complete replacement of the active ensemble from one theta cycle to the next. Mixed states were rare; distributions of cycles correlated with A and B showed significantly fewer cycles correlated with both states than expected by chance. Within individual cycles, segregation was stronger towards the end, when firing started to decline, pointing to the theta cycle as a temporal unit for expression of attractor states in the hippocampus. Observations point to theta cycles as minimum units of hippocampal representation and imply that individual cycles, but not necessarily successive cycles, tend to remain within one state. Repetition of pattern-completion processes across successive theta cycles may facilitate error correction and enhance discriminative power in the presence of weak and ambiguous input cues.
In parallel with these experimental studies, researchers modelled an attractor network that stores continuously changing environments and analysed attractor states. A complete solution to the problem of continuous change was obtained, with a phase diagram indicating the possible regimes in the network. When the contribution of the intermediate environments is increasing, the network was found to undergo a transition into a combined representation of the whole sequence of environments, such that only cells with sufficiently close fields across the whole sequence remain active. Cells with far away initial fields are inhibited and lose their contribution to representing the environments. During screening of the deep layers of medial entorhinal cortex, the participants found a new entorhinal cell type that fires specifically when the animal is close to the borders of the proximal environment. The orientation-specific edge-apposing activity of these cells was shown to be maintained when the environment is stretched and during testing in enclosures of different size and shape in different rooms. Border cells were relatively sparse, comprising less than 10 % of the local cell population, but subsequent recordings in the other layers showed that these cells can be found in all entorhinal layers as well as the adjacent parasubiculum, often intermingled with head-direction cells and grid cells.
Under conditions that favour global remapping in the hippocampal place-cell population the border cells maintained their edge-related firing activity. When multiple border cells were recorded simultaneously, they maintained their geometric relationship. The remapping pattern of entorhinal border cells is different from the hippocampus, where place cells show statistically orthogonal firing patterns in different environments. Results suggest that the entorhinal cortex is predominated by codes that express similar geometric relationships irrespective of the specific features of the environment or the experiences. The discovery of border cells in the entorhinal cortex was accompanied by the discovery of 'boundary vector cells' in the subiculum by one of the other consortium groups. The two cell types have much in common and a question for future research is to determine the relationship between them. Microdrives for multi-site ensemble recording. With the objective of recording ensemble activity from multiple brain locations it became important to develop a multi-electrode microdrive for parallel recording in which electrodes could be moved independently at a desired spacing of electrodes. The industrial partner of the consortium has been responsible for this development. The multi-site microdrives were to be composed of at least two core clusters, with inter-cluster distance specifiable by the user. A computer-controlled milling machine was employed to manufacture miniature plastic components designed using the Rhino three-dimensional (3D) computer-aided design software package. A new drive mechanism, a new stable connector, a new head stage compatible with the Microdrive connectors and a new cannula attachment mechanism were developed. The simplest single-site drive is small, lightweight (less than 1 g) and the plastic components are sufficiently simple that they can be manufactured in bulk. The drives will be easier for users to load than current drives.
4. Visualisation of functional neuronal assemblies
It was aimed to develop a novel miniaturised two-photon fibrescope for imaging of activity patterns in local neuronal assemblies in freely moving rats. By using in vivo labelling with fluorescent calcium indicators, directly visualise place cell and grid cell activity was set out as the animal was navigating through the local environment. The goal was to unravel the fine-scale functional organisation of these neurons. Electrophysiological studies of the spatial structure of cell assembly activity are constrained by the spatial resolution of the tetrode technique. Tetrodes are not suitable for studying differences in activity between neighbouring microcircuits with sharp boundaries and they are suboptimal for identifying neurons that are not active in any of the test conditions. To overcome these problems a portable high-resolution multiphoton microscope was developed which can reveal activity patterns in local neural networks with single-cell and single-spike resolution.
Three key methodological procedures have been developed as part of the project:
(1) the in vivo preparation;
(2) labelling of hippocampal neurons with calcium indicator;
(3) two-photon imaging of hippocampal neurons in vivo.
Through these developments single-cell resolution optical recordings of hippocampal neurons in anesthetised animals were established. A surgical procedure was devised in anesthetised rats to routinely access the hippocampal formation for functional optical measurements, aiming at the CA1 region to begin with. Using visually guided blunt aspiration of a small block of overlying cortical tissue, it is possible to successfully identify the intact hippocampal formation. Researchers bolus-injected the calcium-sensitive dye Oregon-Green BAPTA 1-AM (OGB-1) and the astrocyte-specific dye sulforhodamine 101 (SR101) into the CA1 region of hippocampus. Neuron and astrocyte labelling under these conditions are visible about one hour after dye injection. Third, for imaging hippocampal cells researchers placed the anesthetised animal under a standard two-photon microscope. Cells in the hippocampus were imaged through a small gradient-index (GRIN) lens lowered onto the hippocampus, which mimics the situation with a miniature two-photon fibrescope.
The surgical preparation and bolus-labelling with calcium indicator have yielded good results in anesthetised and behaving rats. Researchers reliably stained and differentiated neurons and glial cells and with the two-photon microscope have visualised cell populations from the surface of the hippocampal formation through stratum oriens down to the pyramidal cell layer. Spontaneous activity was low in all functional measurements. But after reducing inhibition using bicuculline large and synchronised calcium transients were onsevered, presumably representing epochs of action potential activity in populations of pyramidal cells. These results directly demonstrate large-scale calcium indicator labelling and imaging of cell populations in the hippocampus in vivo. This is a prerequisite for all future imaging with this approach and clearly indicates that the ultimate goal is in reach.
5. Temporal codes in entorhinal cell assemblies
It was aimed to determine mechanisms and functions of theta phase precession. It can be argued that theta phase precession provides the most prominent experimental clue to sequence coding in the hippocampus. Although phase precession provides the hippocampus with a mechanism for storing extended sequences of places and experiences, the neuronal circuits and computations responsible for this effect are unknown. During the early stage, it was showed that phase precession is expressed also in grid cells in layer II (but not layer III) of medial entorhinal cortex, which is the origin of projections to the dentate gyrus and CA3. The presence of phase precession in one of the entorhinal inputs to the hippocampus points to the entorhinal cortex as a possible origin of phase precession. Found was that phase precession exists in principal cells of the medial entorhinal cortex and is generated there independently of the hippocampus, and computational models were developed to suggest mechanisms for how phase precession and grid patterns might emerge from intrinsic membrane properties of these neurons. Shown was how theta oscillations may contribute to formation of grid patterns and how these depend on specific velocity inputs to the region. A cellular mechanism for possible transformation has also been identified.
During the first year, researchers reported that grid cells in layer II of medial entorhinal cortex express phase precession. Layer III cells generally do not show this. The aim was to establish precession in layer II cells origates locally in the entorhinal cortex or is derived from the hippocampus. The presence of phase precession in medial entorhinal cortex would be consistent with the subthreshold oscillations of entorhinal stellate cells in layer II, which are faster than the extrinsic theta oscillations and so may cause spike activity to advance progressively across the theta cycle in a manner similar to the dual oscillator mechanism proposed for hippocampal place cells. An alternative would be that phase precession originates in the hippocampus or locally in both hippocampus and entorhinal cortex. Researchers distinguished between these possibilities by recording theta phase in entorhinal grid cells after more global inactivation of the dorsal hippocampus. If phase precession in entorhinal grid cells is inherited from hippocampus, precession in medial entorhinal cortex should be disrupted by inactivation of the hippocampus. An experiment showed that entorhinal phase precession is not blocked by inactivation of the hippocampus, suggesting that the phase advance is indeed generated in the grid cell network. The results pointed to possible mechanisms for grid formation and raised the possibility that hippocampal phase precession is inherited from entorhinal cortex. The phase-precession studies were accompanied by computation work in which an earlier model of phase precession in hippocampus was revisited. In this model, phase precession arises as a consequence of propagation of activity across place cells at each theta cycle due to asymmetry in synaptic connections in the direction of animal motion. STP was introduced in recurrent connections and effects on phase precession were analysed. It was found that synaptic depression that characterises inter-pyramidal connections in hippocampus can cause activity propagation in the network even when synaptic connections are symmetric and that during sharp-waves, replay of sequential activation of place cells can emerge.
An oscillatory interference model of grid cell formation was developed to formally identify the relationships of the theoretical membrane potential oscillations with the variables that can be experimentally measured in freely moving animals expressing spatially organised grid cell firing. Consortium participants have worked out the relationship of theoretical variables with the intrinsic firing frequency of grid cells as reflected in the spike train temporal autocorrelogram, and the theta rhythm of the extracellular local field potential. Support for a specific relationship between running speed and oscillation frequency in the theta-frequency band was obtained in a study. Recording in grid cells from mice with forebrain-specific knockout of HCN1, it was demonstrated that the dorsal-ventral gradient of the grid pattern is preserved in HCN1 knockout mice, but the size and spacing of the grid fields, as well as the period of the accompanying theta modulation, expanded at all dorsal-ventral levels, suggesting that HCN1 channels control the scale of the grid pattern. The speed-dependent modulation of theta frequency was strong in control mice but nearly absent in HCN1 knockout mice, implying that I(h) currents are critical for transformation of speed signals to spatially periodic firing fields. There was no change in theta modulation of simultaneously recorded entorhinal interneurons, suggesting that the expansion of grid scale is caused by lack of HCN1 channels in the grid cells themselves.
6. Development of a spatial representation system
The objective was to determine how key components of the spatial representation system develop relative to each other, how the emergence of one function leads to the next, and how the maturation of the system interacts with experience of the animal during postnatal development. One study had systematically explored the development of spatial representations and the focus of this study was limited to the later stages of development in place cells. The development of entorhinal-hippocampal intrinsic connectivity was mapped. Electrophysiologal recording studies in two to three-week old rats have shown that entorhinal-hippocampal connections develop earlier than researchers thought and we know that a rudimentary spatial representation is already present at the time when rats make their first navigational experiences outside the nest (week 3). Place cells are present from the beginning and rudiments of grid cells can be seen too, although the number of grid cells with clear and stable firing fields takes longer time to develop.
Retrograde tracing studies revealed that the reciprocal connectivity between entorhinal cortex and hippocampus is present at birth. Preliminary voltage sensitive dye imaging data indicated that stimulation of entorhinal cortex results in adult-like patterns of activation at postnatal day P9, whereas only limited local activation can be seen at P5. Interactions between superficial and deep entorhinal layers are noticeable at P9 but not at P5. Intrinsic hippocampal circuits show slow development between P9 and P14. Results indicate the presence of functional connectivity earlier than has been assumed. With the use of in vitro slice preparations and voltage sensitive dye imaging, was established that inputs from pre and parasubiculum are not functional directly after birth. Intrinsic connectivity was investigated with simultaneous whole cell recordings from groups of 3 or 4 medial entorhinal stellate cells in horizontal brain slices from pups of different ages. Data show that connections between stellate cells in the early postnatal stages are sparse. A striking difference is the amount of synchronous firing of stellate cells. Whereas synchrony is limited during the early postnatal stages (P16-P21), this disappears later on.
These findings indicate that connectivity between stellate cells is set up during the early postnatal period and depends on monosynaptic connectivity between stellate cells and either the presence of common inhibitory inputs or common excitatory cortical inputs. The only study of place cells from young rats published before SPACEBRAIN suggested that place fields develop late, stabilising only around P52.
Recording in parallel from medial entorhinal cortex and hippocampus of the developing rat, two of the consortium groups set out to determine the following:
(i) If the late maturation of place cells in the early study was caused by an equally slow development of grid cells, which are thought to provide the spatial input to the place cells.
(ii) If grid patterns are expressed in medial entorhinal cortex at earlier stages of development, in which case the late expression of place cells may reflect slow maturation of the hippocampus itself.
Recordings from the medial entorhinal cortex were expected to establish whether the coherent firing of cells with similar spatial phase is present in the grid cell network from the beginning or develops after grid fields emerge in individual cells. It was found that a rudimentary map of space is present when 2.5-week old pre-weanling rats explore outside the nest for the first time. This map contains both place cells and grid cells but whereas place cells were present from the very beginning, the number of grid cells with strictly periodic fields remained surprisingly low until three to four weeks of age. Results showed convergent results for place cells and grid cells.
Both labs investigated development of head direction-modulated cells. Directionally modulated cells in the presubiculum were present in the earliest recordings. Head-direction cells in this region had adult-like properties from the outset. Directional firing preferences were stable within and between trials and differences in directional preferences between different cells were maintained. Slower development was observed in the entorhinal cortex. Early presence of head direction cells suggests that these cells may be instrumental in setting up the place and grid representations in the hippocampus and entorhinal cortex. Whether the rudimentary spatial periodicity of the earliest grid cells is sufficient to enable place-cell formation in the hippocampus remains to be determined. Joint results suggest that a preconfigured neural circuit for spatial representation is present at the initiation of outbound exploratory behaviour. The maturation of this neural circuit continues in parallel with the animal's first navigational experiences. Place and grid cells develop relatively independently of experience. First-time exploration was compared in animals at early and late states of ontogenetic development. The young group explored an open environment at the time when ratS normally begin leaving the nest. Older animals walked for the first time during the mid P20s. Place and grid fields were similar at comparable ages despite the relatively lengthy spatial experience of the former group. Data suggest that the presence of rudimentary forms of place and grid cells is determined primarily by maturational factors. The findings provide the first experimental support for Kant's more than 200-year old conception of space as an a priori faculty of mind.
Grid cells in the medial entorhinal cortex have been proposed to hold a continuous attractor-based representation of the animal's changing location in the environment by excitatory recurrent connections between cells with similar spatial phase; but does the development of structured connections precede or follow the development of grid fields in individual units? The view that grids reflect metrically organised connectivity in the adult tissue, is challenged by the fact that spatial phase is represented non-topographically in the grid cell network, so that connections between cells which later develop similar spatial phase need to be ordered ad hoc. An alternative possibility is that distant cells wire together instructed by an external 'teaching input' present only temporarily during development. As a result they develop similar spatial phase. The nature of such an input and its localisation in the brain is not known. Using a computational approach it was demonstrated that the conceptual feasibility of a model in which individual grid units first develop their multi-peak fields, and later those with similar spatial phase develop enhanced recurrent connections through Hebbian plasticity. In one of the models of grid formation the grid pattern of activity is elicited in individual units directly by the afferent inputs, conveying spatial information relative to one environmental context. The grids are co-aligned across different units if these are made to interact via recurrent weights modified with a model plasticity rule and may be suited to conjunctive units in the deeper mEC layers. Most microdrives are relatively large and heavy, making the ensemble difficult to carry for the smallest animals. A miniature multi-electrode Microdrive was developed which contains a prototype headstage with no AC coupling components and smaller op-amps. Another prototype was developed which reduces in the size of the tether cable between the animal and the recording system
Potential impact
Recent advances in our ability to record and analyse complex patterns of activity in specific brain circuits will lead towards a fragmented mechanistic appreciation of complex experience.
This will benefit European citizens in a number of ways:
- prevention and treatment of neurological diseases where spatial orientation and memory are impaired;
- a technological revolution in the analysis of neuronal microcircuits and the development of industry that can supply the new tools;
- new insight into algorithms that could be used for design of artificial navigating agents.
Translational potential and contribution to human health
An improved understanding of the cerebral activity responsible for spatial representation and memory is likely to have major consequences for improving the health of European citizens. Research on the functions and the organisation of the entorhinal cortex and its development has immense relevance for a number of devastating diseases and the ways we manage them. In the long run this knowledge is essential for improving diagnosis and treatment of diseases that affect the entorhinal-hippocampal space circuit, as well as other cortical circuits. An important target for translational research is Alzheimer's disease. Volume reduction of the entorhinal cortex is now considered a reliable measure for identifying individuals at risk for Alzheimer's disease. Functional insight into the working principles of the entorhinal cortex provides essential basic knowledge for early diagnostics, prevention and treatment. Results provide key insights into the operational principles of the normal entorhinal-hippocampal system. The consequences of early treatment of Alzheimer's disease for the population of European citizens are likely to be enormous. The economic benefits of reduced hospitalisation and institutionalisation would be enormous if these disorders could be diagnosed adequately or even prevented or cured.
Impact of the technological development
Several novel research tools have been developed. This benefits the European community both on scientific and commercial level. One of the developments resulted from SPACEBRAIN is the production of a user-customisable multi-core split-bundle microdrive which could enable researchers to collect data from large neuronal ensembles in distributed brain regions. As knowledge advances, ensemble recording is becoming an increasingly important tool. European researchers will gain a benefit by the availability of new technology for high-density single-unit recording from multiple areas. Functional imaging using fluorescence microscopy has become a standard technique for measurements in extracted tissues and brain slices, and the technology has contributed invaluable data for the characterisation of the physiology of individual neurons. Fibre-optic imaging tools for optical imaging in freely moving animals constitutes the next major step in optical recordings of neural activity and most likely will have impact on the research field. Also the pharmaceutical industry may become interested.
Exploitation: Relevance of proposed work to robotics and artificial intelligence (AI)
Establishing the network properties of brain systems offers the possibility of contributing to the development of artificial systems, particularly via the twin disciplines of robotics and artificial intelligence (R/AI). The solution to problems demonstrated in experimental work from the project is to use two types of information in tandem: external landmarks as a means of determining instantaneous position, and motion cues to update the position estimate, with periodic recalibration from landmarks. Despite decades of research, robotics has failed to develop a competent navigating agent. One problem may be that R/AI designers have not found a way to integrate static and motion cues in the way that animals do. The project has provided evidence for a neurobiological model according to which motion information in animals is processed via attractor networks instantiated in entorhinal cortex and hippocampus, and where landmark information interacts by 'resetting' the attractor when it is either undefined or has drifted due to accumulated errors from the motion detectors. The attractor networks would constitute the 'common domain' referred to above that allow multiple sensory modalities to converge on a single representation of place. A crucial feature in the mammalian design is the coding in a multiplicity of self-organised memory representations in the hippocampus.
While motion cues have a fixed meaning throughout the animal's lifetime, landmarks provide signals that are completely context-dependent and it would be disastrous to try to implement their fusion with motion cues by means of fixed registers. Evolution has addressed this problem by largely dedicating to landmark representation the most flexible system for generating arbitrary representations of virtually any content and immediately storing them in the hippocampus. The large storage capacity necessary for memorising a large variety of spatial contexts is ensured by the large number of independently modifiable synapses impinging on any given pyramidal cell - of the order of 104 'knobs' on each neuron, which can be tuned separately while exploring a new environment. This flexibility stands in marked contrast to the rigidity of representations in the nearby entorhinal cortex. What the recent discoveries are beginning to illustrate is a counterintuitive solution whereby motion cues are represented by rigid, essentially static codes, whereas static cues are dynamically assigned to new, unpredictable representations. The discovery of neurobiologists that attractors may lie at the heart of place representation suggests that R/AI may benefit from designing computationally analogous processes into their artificial navigating agents. It will be possible to extend biologically inspired robotic design into the sensor fusion domain, which can greatly enhance the design of artificial navigating agents.
Dissemination activities
Scientific results have been disseminated to the broad science community, industrial interests and the lay public. Several papers with SPACEBRAIN funding have been published or accepted for publication or are likely to be published during the coming years. The publication strategy has been to maximise the impact of the publications by withholding data until a relatively complete and coherent understanding is obtained and all interpretations are well supported by the data. Before formal publication in scientific journals, the results have also been made available as early as possible in preliminary form to the general neuroscience community at scientific meetings. An important objective was to inspire computational modelling of the hippocampo-entorhinal circuit. To enable the worldwide community of computational neuroscientists to develop models that match data as closely as possible. Shared data are available as Matlab files and include spike and position times as well as behavioural tracking data for a number of simultaneously recorded grid cells. To explore whether network processes similar to those studied in SPACEBRAIN can be built into navigating robots, we organised a workshop which was held during the final months of the consortium in order to maximise the transfer of mechanistic insight from mammalian microcircuits to the electronic circuits of artificial agents.
All participants in SPACEBRAIN have been committed to better public understanding of science and have been engaged in a number of activities relevant to this. Several participants are members of the European Dana Alliance for the Brain (EDAB). The mission of EDAB, and its American counterpart the Dana Alliance, is to inform the general public and decision-makers about the importance of brain research, to advance knowledge about the personal and public benefits of neuroscience, and to disseminate information on the brain in health and disease in an accessible and relevant way. IEDAB organises the yearly Brain Awareness Week, where hundreds of public events and activities worldwide bring scientific progress in the neurosciences out to the general public. We have used the Brain Awareness Week to disseminate knowledge about spatial representation and navigation, including results from the project. Schools and journalists have been contacted for informal talks about the brain at several of the participant institutions, and public lectures have been arranged. All major publications have been accompanied by press releases and journalists of national newspapers have been contacted by the authors. Members of the consortium are regularly in contact with newspapers and TV channels to increase public understanding of brain functions, brain mechanisms and brain disease.
Project website: http://www.ntnu.no/cbm/spacebrain