Neuron

Perspective Grid Cells and Neural Coding in High-End Cortices Edvard I. Moser1,* and May-Britt Moser1 1Centre for Neural Computation, Kavli Institute for Systems Neuroscience, Norwegian University of Science and Technology, 7491 Trondheim, Norway *Correspondence: [email protected] http://dx.doi.org/10.1016/j.neuron.2013.09.043

An ultimate goal of neuroscience is to understand the mechanisms of mammalian intellectual functions, many of which are thought to depend extensively on the cerebral cortex. While this may have been considered a remote objective when Neuron was launched in 1988, neuroscience has now evolved to a stage where it is possible to decipher neural-circuit mechanisms in the deepest parts of the cortex, far away from sensory receptors and motoneurons. In this review, we show how studies of place cells in the hippocampus and grid cells in the entorhinal cortex may provide some of the first glimpses into these mechanisms. We shall review the events that led up to the discovery of grid cells and a functional circuit in the entorhinal cortex and highlight what we currently see as the big questions in this field—questions that, if resolved, will add to our understanding of cortical computation in a general sense. The cerebral cortex is the multilayered sheet of neural tissue that covers the cerebral hemispheres. The size of the cerebral cortex has increased tremendously during mammalian evolution, and it is the growth of this brain structure that is thought to give rise to the widely expanded repertoire of intellectual abilities in primates. Complex cognitive processes such as memory, imagination, reasoning, planning, and decision making are examples of functions that depend on activity across widespread cortical networks. How these functions emerge as a product of activity in distributed neuronal assemblies is poorly understood, but with the current progress in neuroscience, we may be able to figure out parts of the mechanistic fundament of some of these functions in the not too distant future. Visual Cortex as an Early Gateway to Cortical Computation Much of what we know about cortical computation can be traced back to Hubel and Wiesel’s early work in the visual cortex. More than half a century ago, Hubel and Wiesel (1959) recorded activity of individual neurons in V1 of the cat visual cortex while patterns of light and dark were presented to the eyes of the animal. One of their key observations was that V1 neurons respond to elementary components of the visual scene. Many of their neurons fired specifically in response to bars or edges of particular orientations—some at discrete locations in the visual field (simple cells), others across a wider spatial range (complex cells) (Hubel and Wiesel, 1962). The discovery of these cells was accompanied by the first conceptual model for the formation of receptive fields in the visual cortex, in which a spatial summation mechanism accounted for the transformations from centersurround receptive fields in the thalamus to simple cells in the cortex and from simple cells to complex cells at subsequent stages (Hubel and Wiesel, 1962). These ideas created a fundament not only for visual neuroscience, but also for computational studies of the cortex. Hubel and Wiesel’s early studies were also important because they defined a functional architecture for visually responsive neurons in V1. The studies showed that in cats and monkeys,

V1 neurons are organized in layer-spanning left-eye and righteye ocular dominance bands as well as superimposed columns of cells that respond to similar features of the visual input, such as the orientation of the stimulus (Hubel and Wiesel, 1962; 1974; 1977). Subsequent work showed that orientation columns are arranged gradually around pinwheel centers (Bonhoeffer and Grinvald, 1991) and that, within orientation columns, cells are further organized according to direction preferences (Payne et al., 1981; Tolhurst et al., 1981; Weliky et al., 1996). The early studies in V1 were followed by descriptions of receptive fields at higher levels of the visual system (e.g., Gross et al., 1972; Desimone et al., 1984). In general, as the number of synaptic relays increased, visual receptive fields became larger and more selective, and the mechanisms that could generate those patterns became harder to access. At the top of the cortical hierarchy, where information is combined across sensory systems, it was often no longer possible to match the firing patterns to any experimentally defined stimulus patterns. Spatial Maps as a Window into High-End Cortices The fundament that the progress in visual systems neuroscience has laid for understanding cortical computation remains unequalled. The description of the neural elements of visual representations and their organization into functional circuits has been followed by advances in other cortical sensory systems, but in all of these systems, the biggest insights are, in general, still limited to the earliest stages of cortical processing. Less is known about the higher-order association cortices, where inputs cannot be traced back to particular sensory origins. One reason why the computational operations of most high-end association cortices remain terra incognita is that, for each synaptic relay that is added, neural activity becomes increasingly decoupled from the specific features of the sensory environment. With a lacking understanding of both afferent and efferent brain regions, and the ways that information is integrated across hierarchical levels, it may get difficult to find stimulus patterns that possess any predictable relationship to the firing pattern of the recorded cells. Yet, it is the high-end Neuron 80, October 30, 2013 ª2013 Elsevier Inc. 765

Neuron

Perspective Figure 1. Hierarchical Connectivity Map of the Cortex Visual input is shown at the bottom (RGC, retinal ganglion cells; LGN, lateral geniculate nucleus) and entorhinal cortex (ER) and hippocampus (HC) at the top. One of the goals of modern neuroscience is to understand pattern formation in highend cortices such as the entorhinal cortex and the hippocampus. Adapted, with permission, from Felleman and Van Essen (1991).

cells intermingle with head direction cells, which fire specifically when the animal faces certain directions, and border cells, which fire specifically when animals move along borders of the local environment (Sargolini et al., 2006; Savelli et al., 2008; Solstad et al., 2008). Collectively, these cell types form the elements of what we will refer to as the entorhinalhippocampal space circuit. In this review, we shall take a historical perspective and describe the unfolding of a system of elementary correlates for representation of space in the hippocampus and the entorhinal cortex. We shall discuss mechanisms that might generate this representation, many synapses away from the specific receptive fields of the sensory cortices, and we shall elaborate on how the evolution of a functional architecture within this system might benefit not only mapping of space, but also the formation of high-capacity memory.

cortices that we need to target if we want to understand the most complex cognitive functions. The hippocampus and entorhinal cortex have often been depicted as the apex of the cortical hierarchy (Felleman and Van Essen, 1991; Van Essen et al., 1992; Squire and Zola-Morgan, 1991; Lavenex and Amaral, 2000; Figure 1). On this background, it may come as a surprise that these systems contain a set of representations that perfectly match an attribute of the external world: the animal’s location in space. In the hippocampus, place cells fire specifically at one or a few locations in the animal’s environment (O’Keefe and Dostrovsky, 1971). In the medial part of the entorhinal cortex, grid cells fire at multiple locations that, for each cell, define a hexagonal array across the entire available space (Hafting et al., 2005; Moser et al., 2008; Figure 2). Grid 766 Neuron 80, October 30, 2013 ª2013 Elsevier Inc.

From Place Cells to Grid Cells The study of spatial representation and spatial navigation started long before neuroscientists approached the cortex. The notion of an internal spatial map can be traced back to Edward C. Tolman, who in his cognitive theory of learning suggested that behavior was guided by a map-like representation of stimulus relationships in the environment, rather than by chains of stimulus-response sequences of the type envisaged by Thorndike and Hull (Tolman, 1948). The internal map was thought to enable animals to navigate flexibly in the environment, taking shortcuts and making detours when previously traveled routes were less effective. Tolman’s ideas remained controversial for decades, partly because scientists did not have tools to determine if the cognitive entities proposed by Tolman actually existed. Tolman’s ideas were revitalized many years after his death, after the development of microelectrodes for extracellular recording from single neurons in behaving animals. This development led Ranck (1973) and O’Keefe and Dostrovsky (1971) to monitor activity from single neurons in the hippocampus of

Neuron

Perspective Figure 2. A Grid Cell from the Medial Entorhinal Cortex of the Rat Brain The gray trace is the trajectory of a rat that is foraging in a 2.2 m wide enclosure. Spike locations of the grid cell are superimposed on the track. Each black dot corresponds to one spike. Data were recorded by Kristian Frøland.

freely moving rats. Both laboratories found reliable links between neural firing and the animal’s behavior, but it was O’Keefe and Dostrovsky who found that the firing depended on the animal’s location in the environment. Cells with location-dependent firing were termed place cells, and their specific firing locations were referred to as place fields. Different cells were found to have different place fields (O’Keefe, 1976). The place representation was shown to be nontopographic in the sense that place fields of neighboring cells appeared no more similar than place fields of more widely spaced neurons. The fact that each location in the environment was associated with a unique combination of active place cells pointed to the place cells of the hippocampus as a physical manifestation of Tolman’s cognitive map (O’Keefe and Nadel, 1978). This idea was later reinforced when new technology made it possible to record simultaneously from many dozens of place cells and the trajectory of the animal could be reconstructed from the cumulative firing of these cells (Wilson and McNaughton, 1993). The discovery of place cells was followed by three decades of studies focusing, among other questions, on the properties of the environment that determined the localized firing of the place cells (Muller, 1996). The neural origin of the signal remained deeply enigmatic, however. Much of the challenge was related to the relative isolation of the hippocampus in the functional brain map. The hippocampus was encircled by areas that were poorly characterized structurally as well as functionally. The major

cortical input and output of the hippocampus, the entorhinal cortex, was no exception. It is only now that the entorhinal cortex is beginning to peek out from the dark. At the turn of the millennium, entorhinal activity from freely moving animals had been reported in only a handful of studies. Of particular interest is the report by Quirk et al. (1992) in which the authors recorded activity of individual neurons in medial entorhinal cortex while rats were foraging in a cylindrical environment identical to the ones used by the same authors for place-cell recording in the hippocampus. The neurons had spatial firing preferences, but the firing fields appeared larger and noisier than in hippocampal neurons, and the coactivity patterns did not, like place cells, respond to geometric transformations of the environment. Together with two studies that showed similarly dispersed firing fields in linearized environments (Barnes et al., 1990; Frank et al., 2000), the observations of Quirk et al. (1992) suggested that some location-specific firing exists prior to the hippocampus. However, the confined nature of the firing was thought to originate within the hippocampus itself. The idea that place fields evolved within the hippocampal circuit led us to monitor activity in place cells from CA1, the output stage of the hippocampus, after all input from other hippocampal subfields was disconnected (Brun et al., 2002). Somewhat to our surprise, small, well-defined place fields were still present in CA1, suggesting either that place fields were generated within the local network of CA1 or they were derived, primarily, from spatially responsive cells in upstream cortical regions with direct inputs to the CA1. We decided to explore the latter alternative. A key event in the search for cortical origins of the place-cell signal was the recognition that the hippocampal-entorhinal system is functionally organized along its dorsoventral axis. Our own awareness to this issue was raised by the observation that spatial learning in a water maze navigation task is impaired significantly more by lesions in the dorsal part of the hippocampus than by equally large lesions in the ventral part (Moser et al., 1993; 1995). This observation directed us to studies of Menno Witter, who in the 1980s provided evidence for rigid topographical organization along the hippocampal-entorhinal dorsoventral axis. Witter and colleagues showed that dorsal parts of the hippocampus connect to dorsal parts of the entorhinal cortex and Neuron 80, October 30, 2013 ª2013 Elsevier Inc. 767

Neuron

Grid spac. (cm)

Perspective 120 100 80 60 40

Rat: 15708 TT2

0

TT5

5

0

Rat: 15314 TT3

TT9

0 10 20 10 20 0 10 20 Dorso-ventral position (cell number, ranked)

ventral parts of the hippocampus to ventral parts of the entorhinal cortex (Witter and Groenewegen, 1984; Witter et al., 1989). Dorsal and ventral entorhinal regions were in turn linked to different parts of the rest of the brain (Witter et al., 1989; Burwell and Amaral, 1998). The discovery of entorhinal-hippocampal projection topography raised the possibility that previous recordings in the entorhinal cortex had not targeted those regions that had the strongest connections to the dorsal quarter of the hippocampus, where nearly all place-cell activity had been recorded at that time. With this mismatch in mind, we decided, together with Menno Witter, to approach the dorsalmost parts of the medial entorhinal cortex. The move paid off; recordings from this region showed firing fields that were as sharp and confined as in the hippocampus (Fyhn et al., 2004). The difference was that each cell had multiple firing fields that were scattered around in the entire recording arena. In order to visualize the spatial organization of the firing fields of each cell, we next decided to test the animals in larger environments, where a larger number of fields could be displayed (Hafting et al., 2005). It could now be seen that the fields formed a hexagonal array, with equilateral triangles as a unit, like the arrangement of marble holes on a Chinese checkerboard (Figure 2). We termed the cells grid cells. The grid pattern was similar for all cells, but the spacing of the fields, the orientation of the grid axes, and the x-y location of the grid fields (their grid phase) might vary from cell to cell. The pattern persisted when the room lights were turned off and was not abolished by variations in the speed and the direction of the animal, pointing to self-motion signals as a major component of the mechanism that determined the firing locations. The continuous adjustment for changes in speed and direction suggested that grid cells had access to path-integration information (Hafting et al., 2005; McNaughton et al., 2006). However, at the same time, grid fields appeared in the same locations on successive trials, and the fields rotated in correspondence with rotated landmarks, suggesting that firing positions are also responsive to the configuration of landmarks in each environment. More extensive recordings soon showed that grid cells intermingle with other cell types. While grid cells predominated in layer II of the medial entorhinal cortex, intermediate and deep layers also contained a large fraction of head direction cells (Sargolini et al., 2006). Head direction cells, originally described in the dorsal presubiculum (Ranck, 1985), are cells that fire specifically when animals face a certain direction, regardless of the animal’s position (Taube et al., 1990a; 1990b). In the medial entorhinal cortex, many head direction cells were also grid cells, 768 Neuron 80, October 30, 2013 ª2013 Elsevier Inc.

Figure 3. Modular Organization of Grid Cells

TT10

300

10

20

Grid spacing is shown as a function of position along the entorhinal dorsoventral axis. Each dot corresponds to one cell. Cells on different tetrodes (TT) are shown in separate diagrams. Note stepwise increases in grid spacing. Up to four or five steps were identified in individual animals, each corresponding to a separate module. Modules differed on multiple parameters, including grid scale, grid orientation, grid symmetry, and theta modulation. Adapted from Stensola et al. (2012).

firing only when the animal passed through the grid vertices with its head in a certain direction (Sargolini et al., 2006). Two years later, grid cells and head direction cells were found to colocalize with a third type of cell: border cells (Savelli et al., 2008; Solstad et al., 2008). These cells fired specifically when the animal was near one or several borders of the local environment, such as a wall or an edge. The firing fields followed the walls when the walls were moved, and when a new wall was inserted, a new firing field often emerged along the insert. Grid cells, head direction cells, and border cells were found to coexist not only in the medial entorhinal cortex, but also in the adjacent presubiculum and parasubiculum (Boccara et al., 2010). Collectively, these observations pointed to a second internal map of space, different from the place-cell map described in the hippocampus. Grid cells, head direction cells, and border cells may be key elements of this map. The clearest difference between these cell types and the place cells in the hippocampus is perhaps the invariance of the activity patterns in the entorhinal cortex. Entorhinal cells appear to fire in all environments, and many cells maintain their phase and orientation relationships from one environment to the next. For example, two grid cells with similar vertex locations in one environment may fire at similar positions also in other environments (Fyhn et al., 2007). The persistence of coactivity patterns also applies to head direction cells (Taube et al., 1990b; Taube and Burton, 1995; Yoganarasimha et al., 2006; Solstad et al., 2008) and border cells (Solstad et al., 2008). Until recently, studies of entorhinal cell types focused mainly on single-cell properties. Recent developments have made it possible to record activity from many dozens of grid cells at the same time. Up to 180 grid cells could be recorded per animal (Stensola et al., 2012). With such extensive recordings, it was possible to show, in individual animals, that grid cells are organized into a small number of maps with discrete properties (Figure 3), as had been suspected based on recordings with small cell numbers (Barry et al., 2007). Within the area of entorhinal cortex that could be sampled, four or five different grid modules were identified. Each module had a unique grid spacing. The smallest values predominated at the dorsal end of the medial entorhinal cortex. Modules with larger spacing were added successively as recording electrodes were advanced ventrally. There was a strict scale relationship between modules, with grid scale increasing, on average, by a factor of 1.4 from one module to the next, as in a geometric progression. A modular organization with geometric scaling has been shown in theoretical analyses to be the one that best allows position to be estimated from grid cells (Mathis et al., 2012).

Neuron

Perspective With the finding that the grid map is modular, a functional architecture for the representation of space is beginning to unfold, but many questions remain. For example, the cellular substrate of the grid modules has not been determined. The distribution of grid modules does not correspond to any familiar molecular expression pattern, and we do not know whether and how grid cells in the same module are linked to each other. If cells from the same module are connected, when and how do these connections develop? Are cells from the same grid module derived from the same population of progenitor cells, as reported for cells with similar orientation preferences in the visual cortex (Li et al., 2012; Ohtsuki et al., 2012)? Or do functional modules develop by activity-dependent mechanisms in response to specific patterns of experience (Ko et al., 2013)? These possibilities are not mutually exclusive (Ko et al., 2013). Answers to such questions will increase our understanding of how functional architecture arises, not only in the entorhinal cortex, but in the cortex in general. Three Questions In the remainder of this review, we shall highlight three questions that we believe will be central to investigations of entorhinal spatial map formation in the years to come: (i) the mechanisms of the grid pattern, (ii) the mechanisms for transformation between entorhinal and hippocampal firing fields, and (iii) the mechanisms for transformation of a rigid population response in the entorhinal cortex to a wide spectrum of uncorrelated representations in the hippocampus, a property that may be crucial to the formation of high-capacity episodic memory. Question 1: How Are Grid Cells Generated? Since grid cells were discovered in 2005, a number of mechanisms have been proposed for these cells. These mechanisms could generally be sorted into two classes, both of which assume that grid cells perform path integration in response to incoming velocity signals (Moser et al., 2008; Giocomo et al., 2011). One category of models, oscillatory interference models, has suggested that grid patterns emerge within cells as a result of interference between oscillators of different frequencies, where the frequency of one of the oscillators is determined by the instantaneous running speed of the animal in a particular direction (Burgess et al., 2007; Hasselmo et al., 2007; Blair et al., 2007; Burgess, 2008). The simultaneous appearance of these oscillators within a cell or among the inputs to a cell generates an interference pattern in the membrane potential of the cell along the orientation of the velocity-controlled oscillator. Because the frequency of this pattern is constantly modulated by velocity, the oscillation is transformed to a spatial oscillation. If there are three oscillators, and their preferred orientations are somehow separated by 60 degrees, a hexagonal spatial firing pattern is generated. Experimental evidence has not generally supported the specific mechanisms for grid patterns proposed in the oscillatory interference models. Two key assumptions have recently been tested. One is that grid cells require theta oscillations. Grid cells have now been recorded in two species in which theta oscillations occur only intermittently. In bats (Yartsev et al., 2011) and monkeys (Killian et al., 2012), grid patterns were as prominent in the absence of theta oscillations as in their presence, suggest-

ing that the grid mechanism is theta independent (but see Barry et al., 2012). A second prediction was that when theta oscillations occur, grid fields should coincide with theta-interference waves in the membrane potential. This prediction remains largely unsupported, as whole-cell recordings from grid cells fail to show any association between grid vertices and changes in the amplitude of theta oscillations in the cell’s membrane potential (Domnisoru et al., 2013; Schmidt-Hieber and Ha¨usser, 2013). Finally, the oscillatory interference models share the theoretical limitation that the 60-degree separation—the very phenomenon to be explained—is put in by hand, i.e., 60-degree separation is supposed to be present already in the inputs to the grid cells (Moser et al., 2014). Taken together, these experimental and theoretical considerations have suggested to many researchers that theta oscillations and theta interference are not necessary for the formation of spatial periodicity. The recent downturn of the oscillatory interference models has raised increased interest in the other major class of grid cell models. This class of models suggests that hexagonal firing patterns emerge as an equilibrium state in competitive attractor networks with strong recurrent excitatory and inhibitory connections (Fuhs and Touretzky, 2006; McNaughton et al., 2006; Burak and Fiete, 2009; Moser et al., 2014). Neural activity is moved across such networks in response to velocity signals, in agreement with the animal’s movements through the environment. During the early days of grid cells, the recurrent connections were thought to be excitatory, with an inhibitory surrounding. However, this assumption does not fit with the connectivity of the cell type that apparently expresses the most periodic grid pattern: the stellate cells of layer II in the medial entorhinal cortex. Stellate cells are interconnected almost exclusively via inhibitory interneurons (Dhillon and Jones, 2000; Couey et al., 2013; Pastoll et al., 2013). If grid patterns are generated in stellate cells, the competitive interactions must therefore be exclusively inhibitory. In favor of this possibility, recent modeling has shown that in networks where each neuron has an inhibitory output of a constant magnitude and a fixed radius, activity will self-organize into a stable hexagonal grid pattern (Couey et al., 2013; Bonnevie et al., 2013; Pastoll et al., 2013) (Figure 4). One condition for this to occur is for the network to receive steady excitatory input from an external source. Without such input, the firing of a grid cell would be determined by its external inputs, such as directional signals from the head direction system. Experimental work supports this prediction. Removal of one of the major excitatory inputs (from the hippocampus) leads to disruption of the grid pattern at the same time that directional modulation is increased (Bonnevie et al., 2013). The attractor models receive indirect support from a number of research lines. The strongest indication of an attractor mechanism is perhaps the observation that, within a grid module, the spatial relationship between pairs of grid cells persists across environments and environmental manipulations, despite substantial changes in the sensory input (Fyhn et al., 2007; Yoon et al., 2013). The fact that cell-cell relationships are better preserved across environments than responses of single cells speaks in favor of a network organization in which grid-cell activity falls into a low number of internally generated stable states (Yoon et al., 2013). An additional line of support for this Neuron 80, October 30, 2013 ª2013 Elsevier Inc. 769

Neuron

Perspective Figure 4. Spontaneous Formation of Hexagonal Firing Patterns in a Model of Grid Cells A hexagonal grid forms spontaneously on a twodimensional lattice of stellate cells with all-or-none inhibitory interconnections. Each pixel corresponds to one cell in the lattice. Neurons are arranged according to the x-y locations of their grid fields. Rings indicate the area of inhibition around two example cells (crosses). Firing rate is color coded as shown in the scale bar. Reproduced from Couey et al. (2013).

idea is the fact that grid cells are organized in modules with similar grid spacing and grid orientation (Stensola et al., 2012). This is an implicit and necessary assumption of all existing attractor models for grid cells. A modular organization makes it possible to maintain a constant relationship between velocity of movement and displacement in the neural sheet such that the hexagonal organization of the grid network is reflected in the activity of individual neurons. Both sets of observations—internal coherence and modularity—are consistent with the notion that individual grid modules operate as attractor networks, but they do not prove it. It is important to be aware that the current evidence does not directly address the core ideas of the models, the mechanisms for hexagonal pattern formation and speed-dependent network translation. In future work, it will be necessary to test these mechanisms by direct observation, e.g., by monitoring spatial firing patterns in large networks of cells with known connectivity. It will be necessary to more directly address key assumptions, such as the proposed connectivity preference between grid cells with similar firing locations. The recent availability of tools for the tracing of connectivity in functionally identified neurons may allow such experiments to be realized in the not too distant future. Finally, the extent to which the network is robust against noise in functional connectivity must be determined (Moser et al., 2014). Variations in strength of input and output may cause unwanted drift that destroys the periodicity of the grid pattern. It is currently not known how networks circumvent such drift, although interesting proposals have been made (Itskov et al., 2011). In the absence of clear answers to these challenges, it may be fair to conclude that the available evidence speaks in favor of some sort of attractor mechanism, but the detailed implementation is certainly not well understood. 770 Neuron 80, October 30, 2013 ª2013 Elsevier Inc.

Question 2: How Are Grid Cells Transformed to Place Cells? How are outputs from grid cells and other entorhinal cells transformed to place signals in the hippocampus? One of the first neural code transformations to be investigated in the cortex was the conversion of concentric receptive fields in the lateral geniculate nucleus to orientation-specific linear receptive fields in simple cells of the visual cortex (Hubel and Wiesel, 1959). This transformation was explained by a simple spatial summation mechanism (Hubel and Wiesel, 1962). However, with the single-spine resolution of modern imaging technologies, it seems clear that, at least in layers II–III, the synaptic inputs to individual orientation-selective V1 cells span a wide range of orientations, although the average tuning across this wide range is similar to that of the somatic output (Jia et al., 2010; Chen et al., 2013). The shaping of an orientation-selective output may thus be a more complex process than previously thought, involving dendritic amplification as well as local circuit mechanisms. Similarly complex mechanisms may be involved in the formation of place signals from entorhinal spatial outputs. In the earliest models for grid-to-place transformation, place fields were thought to be generated by a Fourier mechanism in which periodic fields from grid cells with different grid spacing and orientation were linearly combined to yield a single-peaked place field (O’Keefe and Burgess, 2005; Fuhs and Touretzky, 2006; McNaughton et al., 2006; Solstad et al., 2006). The resulting signal was also periodic, but because different wavelengths were combined, large-amplitude signals were expected only at widely spaced locations—too far from each other for repeated activity to be seen in an experimental setting. In their reliance on summation of inputs from specific classes of neurons, this family of models bears some similarity to the early models for formation of linear orientation-specific receptive fields in the visual cortex. The idea that place cells are generated by outputs from grid cells with specific properties raises the question of whether other entorhinal cell types are not relevant to the formation of place cells. We have tested which entorhinal cell types project to the hippocampus using an optogenetic-electrophysiological strategy in which the light-sensitive channel protein channelrhodopsin-2 is expressed selectively in the hippocampus-targeting

Neuron

Perspective subset of entorhinal projection neurons (Zhang et al., 2013). With this strategy, directly projecting neurons could be identified in medial entorhinal cortex as neurons that responded at minimal latencies to a local light stimulus. As expected, a large number of spatially modulated entorhinal projection cells were grid cells; however, the entorhinal projection also contained other cell types, including border cells and head direction cells as well as many cells with no detectable spatial correlate. The results suggested that the hippocampus receives direct input from a broad range of entorhinal functional cell types, conveying information from a variety of sources that contain both path-integration and landmark-based information (O’Keefe, 1976; O’Keefe and Burgess, 1996; Gothard et al., 1996; Terrazas et al., 2005). In the presence of such diversity of inputs, it is perhaps unlikely that place cells are generated exclusively from grid cells. If the spectrum of inputs to an individual cell is broad, sharply confined place fields may only be generated after the addition of local mechanisms, such as recurrent inhibition (de Almeida et al., 2009; Monaco and Abbott, 2011), changes in synaptic strength (Rolls et al., 2006; Savelli and Knierim, 2010), or active dendritic properties (Smith et al., 2013). If this turns out to be true, the mechanisms for place field refinement would, in part, have returned to the hippocampus, where our search started more than 10 years ago. The difference, however, is that now we have some knowledge of the functional nature of the hippocampal inputs. This brings us closer to deciphering the mechanisms by which those inputs are converted to place-cell signals. The formation of place cells from grid cells and other entorhinal outputs may share key properties with the mechanism for receptive field formation in the sensory cortices. In orientation-selective neurons of the visual cortex, a broad range of orientation inputs is transformed into a specific orientation preference in the firing pattern (Jia et al., 2010: Chen et al., 2013), and in the auditory cortex, a wide distribution of frequency preferences in the synaptic inputs is converted to a narrow range in the cell’s output (Chen et al., 2011). The mechanisms for these transformations remain to be determined, but with the availability of methods that can monitor activity across synaptic inputs and dendritic segments at the same time as the cell’s output, significant advances may soon take place. A similarly sophisticated set of transformation mechanisms in the entorhinal-hippocampal circuit would definitely enhance the computational power of hippocampal representations. Convergent input from a broad spectrum of cell types would enable individual place cells to respond dynamically, favoring different types of input under different behavioral circumstances (Zhang et al., 2013; 2014). Question 3: From Space to Memory A third topic that we consider fundamental for future studies is the relationship between space and memory. The observation that grid cells are organized as discrete modules is important not only because it provides a neural architecture for space, but also because of its putative consequences for the formation of representations in downstream brain areas, such as the hippocampus. Simultaneous recording from multiple grid modules has shown that when the local environment is geometrically deformed, some modules rescale in accordance with the deformation, whereas others do not (Stensola et al., 2012). If individual mod-

ules respond independently to changes in the environment, the coactivity pattern among grid cells may be changed at all locations in the recording environment, and a different subset of place cells is likely to be recruited at each place (Fyhn et al., 2007; Stensola et al., 2012; Buzsa´ki and Moser, 2013). Independent module responses might thus give rise to a very large number of coactivity patterns in the hippocampus in the same way that a combination lock with only five digits may give rise to a hundred thousand unique patterns with only ten response alternatives per module (Rowland and Moser, 2014). Computational simulations have verified that extensive diversity can be generated with a number of modules that correspond closely to the experimental data (Monaco and Abbott, 2011). This expansion of neuronal patterns may have been the mechanism that during evolution allowed the hippocampus to take on an increasingly important role in high-capacity episodic memory formation (Buzsa´ki and Moser, 2013). The proposed link between grid modules and hippocampal memory capacity remains to be tested, however. We know that individual grid maps maintain their functional structure from one environment to the next (Fyhn et al., 2007; Solstad et al., 2008), whereas hippocampal representations are diverse, showing complete independence across pairs of recording environments (Muller and Kubie, 1987; Leutgeb et al., 2004). Whether this transformation from a small number of entorhinal maps to a large number of hippocampal maps is evoked by independent responses among grid modules remains to be tested. Similarly, the neural mechanisms that could enable such a transformation and the detailed consequences for memory formation remain elusive. Evolution of the Space Circuit The fundamental properties of the entorhinal-hippocampal space circuit seem to be preserved across mammalian evolution. While most studies of this system use rodents, grid cells have also been found in bats, which are phylogenetically distant from rodents (Yartsev et al., 2011). All cardinal features of rodent grid cells are preserved in bats, including not only hexagonal firing, but also the coherence of grid orientation and grid spacing, the apparent lack of phase topography, the increase in grid scale along the dorsoventral axis, and the modulation of firing rate by velocity. In bats, as in rodents, grid cells colocalize with head direction cells and border cells. More recently, grid cells have been reported in monkeys, but here the hexagonal firing was determined by where the monkey fixated on a visual image (Killian et al., 2012). The dependence on view location in monkey grid cells is reminiscent of earlier work suggesting that in monkeys, hippocampal and parahippocampal cells fire when the animal looks at certain locations, independently of where the animal is located (Rolls and O’Mara, 1995; Rolls et al., 1997). Collectively, these findings suggest that in primate evolution, grid cells and place cells became responsive not only to changes in the speed and direction of locomotion, but also the velocity of the animal’s eye movements. Whether grid cells of monkeys are driven only by saccades or also by locomotion remains to be determined. The fact that grid cells have been reported in humans performing a virtual reality task (Jacobs et al., 2013) reinforces the view that, in primates, grid activity can be evoked by a spectrum of Neuron 80, October 30, 2013 ª2013 Elsevier Inc. 771

Neuron

Perspective sensory inputs and that the grid network may be used for multiple purposes. Exploration of the variety of functions potentially served by grid cells in primates should certainly have priority. Conclusion The mammalian space circuit is one of the first nonsensory cognitive functions to be understood in mechanistic terms. With the presence of grid cells, and with the availability of new tools for selective activation and inactivation of circuit elements, it has become possible to study neural computation at the high end of the cortical hierarchy, far away from sensory inputs and motor outputs. A huge benefit of studying these cells is the close correspondence between the firing pattern and a property of the external world: the animal’s location in the environment. This correspondence provides researchers with easy experimental access to high-end neuronal coding within the circuits where the codes are generated. Understanding how space is created in this circuit may provide important clues about general principles for cortical computation that extend well beyond the domain of space, touching on the realms of thinking, planning, reflection, and imagination. ACKNOWLEDGMENTS We thank the European Research Council (‘‘CIRCUIT’’ Advanced Investigator Grant, Grant Agreement 232608; ‘‘ENSEMBLE’’ Advanced Investigator Grant, Grant Agreement 268598), the Louis-Jeantet Prize for Medicine, the Kavli Foundation, and the Centre of Excellence scheme and the FRIPRO and NEVRONOR programs of the Research Council of Norway for support. REFERENCES Barnes, C.A., McNaughton, B.L., Mizumori, S.J., Leonard, B.W., and Lin, L.H. (1990). Comparison of spatial and temporal characteristics of neuronal activity in sequential stages of hippocampal processing. Prog. Brain Res. 83, 287–300. Barry, C., Hayman, R., Burgess, N., and Jeffery, K.J. (2007). Experiencedependent rescaling of entorhinal grids. Nat. Neurosci. 10, 682–684. Barry, C., Bush, D., O’Keefe, J., and Burgess, N. (2012). Models of grid cells and theta oscillations. Nature 488, E1–E2. Blair, H.T., Welday, A.C., and Zhang, K. (2007). Scale-invariant memory representations emerge from moire´ interference between grid fields that produce theta oscillations: a computational model. J. Neurosci. 27, 3211–3229. Boccara, C.N., Sargolini, F., Thoresen, V.H., Solstad, T., Witter, M.P., Moser, E.I., and Moser, M.-B. (2010). Grid cells in pre- and parasubiculum. Nat. Neurosci. 13, 987–994. Bonhoeffer, T., and Grinvald, A. (1991). Iso-orientation domains in cat visual cortex are arranged in pinwheel-like patterns. Nature 353, 429–431. Bonnevie, T., Dunn, B., Fyhn, M., Hafting, T., Derdikman, D., Kubie, J.L., Roudi, Y., Moser, E.I., and Moser, M.-B. (2013). Grid cells require excitatory drive from the hippocampus. Nat. Neurosci. 16, 309–317. Brun, V.H., Otnass, M.K., Molden, S., Steffenach, H.A., Witter, M.P., Moser, M.-B., and Moser, E.I. (2002). Place cells and place recognition maintained by direct entorhinal-hippocampal circuitry. Science 296, 2243–2246.

Burgess, N., Barry, C., and O’Keefe, J. (2007). An oscillatory interference model of grid cell firing. Hippocampus 17, 801–812. Burwell, R.D., and Amaral, D.G. (1998). Cortical afferents of the perirhinal, postrhinal, and entorhinal cortices of the rat. J. Comp. Neurol. 398, 179–205. Buzsa´ki, G., and Moser, E.I. (2013). Memory, navigation and theta rhythm in the hippocampal-entorhinal system. Nat. Neurosci. 16, 130–138. Chen, X., Leischner, U., Rochefort, N.L., Nelken, I., and Konnerth, A. (2011). Functional mapping of single spines in cortical neurons in vivo. Nature 475, 501–505. Chen, T.W., Wardill, T.J., Sun, Y., Pulver, S.R., Renninger, S.L., Baohan, A., Schreiter, E.R., Kerr, R.A., Orger, M.B., Jayaraman, V., et al. (2013). Ultrasensitive fluorescent proteins for imaging neuronal activity. Nature 499, 295–300. Couey, J.J., Witoelar, A., Zhang, S.J., Zheng, K., Ye, J., Dunn, B., Czajkowski, R., Moser, M.-B., Moser, E.I., Roudi, Y., and Witter, M.P. (2013). Recurrent inhibitory circuitry as a mechanism for grid formation. Nat. Neurosci. 16, 318–324. de Almeida, L., Idiart, M., and Lisman, J.E. (2009). The input-output transformation of the hippocampal granule cells: from grid cells to place fields. J. Neurosci. 29, 7504–7512. Desimone, R., Albright, T.D., Gross, C.G., and Bruce, C. (1984). Stimulusselective properties of inferior temporal neurons in the macaque. J. Neurosci. 4, 2051–2062. Dhillon, A., and Jones, R.S. (2000). Laminar differences in recurrent excitatory transmission in the rat entorhinal cortex in vitro. Neuroscience 99, 413–422. Domnisoru, C., Kinkhabwala, A.A., and Tank, D.W. (2013). Membrane potential dynamics of grid cells. Nature 495, 199–204. Felleman, D.J., and Van Essen, D.C. (1991). Distributed hierarchical processing in the primate cerebral cortex. Cereb. Cortex 1, 1–47. Frank, L.M., Brown, E.N., and Wilson, M. (2000). Trajectory encoding in the hippocampus and entorhinal cortex. Neuron 27, 169–178. Fuhs, M.C., and Touretzky, D.S. (2006). A spin glass model of path integration in rat medial entorhinal cortex. J. Neurosci. 26, 4266–4276. Fyhn, M., Molden, S., Witter, M.P., Moser, E.I., and Moser, M.B. (2004). Spatial representation in the entorhinal cortex. Science 305, 1258–1264. Fyhn, M., Hafting, T., Treves, A., Moser, M.-B., and Moser, E.I. (2007). Hippocampal remapping and grid realignment in entorhinal cortex. Nature 446, 190–194. Giocomo, L.M., Moser, M.-B., and Moser, E.I. (2011). Computational models of grid cells. Neuron 71, 589–603. Gothard, K.M., Skaggs, W.E., and McNaughton, B.L. (1996). Dynamics of mismatch correction in the hippocampal ensemble code for space: interaction between path integration and environmental cues. J. Neurosci. 16, 8027–8040. Gross, C.G., Rocha-Miranda, C.E., and Bender, D.B. (1972). Visual properties of neurons in inferotemporal cortex of the Macaque. J. Neurophysiol. 35, 96–111. Hafting, T., Fyhn, M., Molden, S., Moser, M.-B., and Moser, E.I. (2005). Microstructure of a spatial map in the entorhinal cortex. Nature 436, 801–806. Hasselmo, M.E., Giocomo, L.M., and Zilli, E.A. (2007). Grid cell firing may arise from interference of theta frequency membrane potential oscillations in single neurons. Hippocampus 17, 1252–1271. Hubel, D.H., and Wiesel, T.N. (1959). Receptive fields of single neurones in the cat’s striate cortex. J. Physiol. 148, 574–591. Hubel, D.H., and Wiesel, T.N. (1962). Receptive fields, binocular interaction and functional architecture in the cat’s visual cortex. J. Physiol. 160, 106–154.

Burak, Y., and Fiete, I.R. (2009). Accurate path integration in continuous attractor network models of grid cells. PLoS Comput. Biol. 5, e1000291.

Hubel, D.H., and Wiesel, T.N. (1974). Sequence regularity and geometry of orientation columns in the monkey striate cortex. J. Comp. Neurol. 158, 267–293.

Burgess, N. (2008). Grid cells and theta as oscillatory interference: theory and predictions. Hippocampus 18, 1157–1174.

Hubel, D.H., and Wiesel, T.N. (1977). Ferrier lecture. Functional architecture of macaque monkey visual cortex. Proc. R. Soc. Lond. B Biol. Sci. 198, 1–59.

772 Neuron 80, October 30, 2013 ª2013 Elsevier Inc.

Neuron

Perspective Itskov, V., Hansel, D., and Tsodyks, M. (2011). Short-Term Facilitation may Stabilize Parametric Working Memory Trace. Front Comput Neurosci 5, 40.

Pastoll, H., Solanka, L., van Rossum, M.C., and Nolan, M.F. (2013). Feedback inhibition enables q-nested g oscillations and grid firing fields. Neuron 77, 141–154.

Jacobs, J., Weidemann, C.T., Miller, J.F., Solway, A., Burke, J.F., Wei, X.X., Suthana, N., Sperling, M.R., Sharan, A.D., Fried, I., and Kahana, M.J. (2013). Direct recordings of grid-like neuronal activity in human spatial navigation. Nat. Neurosci. 16, 1188–1190.

Payne, B.R., Berman, N., and Murphy, E.H. (1981). Organization of direction preferences in cat visual cortex. Brain Res. 211, 445–450.

Jia, H., Rochefort, N.L., Chen, X., and Konnerth, A. (2010). Dendritic organization of sensory input to cortical neurons in vivo. Nature 464, 1307–1312. Killian, N.J., Jutras, M.J., and Buffalo, E.A. (2012). A map of visual space in the primate entorhinal cortex. Nature 491, 761–764. Ko, H., Cossell, L., Baragli, C., Antolik, J., Clopath, C., Hofer, S.B., and MrsicFlogel, T.D. (2013). The emergence of functional microcircuits in visual cortex. Nature 496, 96–100. Lavenex, P., and Amaral, D.G. (2000). Hippocampal-neocortical interaction: a hierarchy of associativity. Hippocampus 10, 420–430. Leutgeb, S., Leutgeb, J.K., Treves, A., Moser, M.-B., and Moser, E.I. (2004). Distinct ensemble codes in hippocampal areas CA3 and CA1. Science 305, 1295–1298. Li, Y., Lu, H., Cheng, P.L., Ge, S., Xu, H., Shi, S.H., and Dan, Y. (2012). Clonally related visual cortical neurons show similar stimulus feature selectivity. Nature 486, 118–121. Mathis, A., Herz, A.V., and Stemmler, M. (2012). Optimal population codes for space: grid cells outperform place cells. Neural Comput. 24, 2280–2317. McNaughton, B.L., Battaglia, F.P., Jensen, O., Moser, E.I., and Moser, M.B. (2006). Path integration and the neural basis of the ‘cognitive map’. Nat. Rev. Neurosci. 7, 663–678. Monaco, J.D., and Abbott, L.F. (2011). Modular realignment of entorhinal grid cell activity as a basis for hippocampal remapping. J. Neurosci. 31, 9414– 9425. Moser, E.I., Moser, M.-B., and Andersen, P. (1993). Spatial learning impairment parallels the magnitude of dorsal hippocampal lesions, but is hardly present following ventral lesions. J. Neurosci. 13, 3916–3925. Moser, M.-B., Moser, E.I., Forrest, E., Andersen, P., and Morris, R.G.M. (1995). Spatial learning with a minislab in the dorsal hippocampus. Proc. Natl. Acad. Sci. USA 92, 9697–9701. Moser, E.I., Kropff, E., and Moser, M.-B. (2008). Place cells, grid cells, and the brain’s spatial representation system. Annu. Rev. Neurosci. 31, 69–89. Moser, E.I., Moser, M.-B., and Roudi, Y. (2014). Network mechanisms of grid cells. Philos. Transact. R. Soc. London B Biol. Sci., in press. Muller, R. (1996). A quarter of a century of place cells. Neuron 17, 813–822. Muller, R.U., and Kubie, J.L. (1987). The effects of changes in the environment on the spatial firing of hippocampal complex-spike cells. J. Neurosci. 7, 1951– 1968. O’Keefe, J. (1976). Place units in the hippocampus of the freely moving rat. Exp. Neurol. 51, 78–109. O’Keefe, J., and Burgess, N. (1996). Geometric determinants of the place fields of hippocampal neurons. Nature 381, 425–428. O’Keefe, J., and Burgess, N. (2005). Dual phase and rate coding in hippocampal place cells: theoretical significance and relationship to entorhinal grid cells. Hippocampus 15, 853–866. O’Keefe, J., and Dostrovsky, J. (1971). The hippocampus as a spatial map. Preliminary evidence from unit activity in the freely-moving rat. Brain Res. 34, 171–175.

Quirk, G.J., Muller, R.U., Kubie, J.L., and Ranck, J.B., Jr. (1992). The positional firing properties of medial entorhinal neurons: description and comparison with hippocampal place cells. J. Neurosci. 12, 1945–1963. Ranck, J.B., Jr. (1973). Studies on single neurons in dorsal hippocampal formation and septum in unrestrained rats. I. Behavioral correlates and firing repertoires. Exp. Neurol. 41, 461–531. Ranck, J.B. (1985). Head direction cells in the deep cell layer of dorsal presubiculum in freely moving rats. In Electrical Activity of the Archicortex, G. Buzsa´ki and C.H. Vanderwolf, eds. (Budapest: Akademiai Kiado), pp. 217–220. Rolls, E.T., and O’Mara, S.M. (1995). View-responsive neurons in the primate hippocampal complex. Hippocampus 5, 409–424. Rolls, E.T., Robertson, R.G., and Georges-Franc¸ois, P. (1997). Spatial view cells in the primate hippocampus. Eur. J. Neurosci. 9, 1789–1794. Rolls, E.T., Stringer, S.M., and Elliot, T. (2006). Entorhinal cortex grid cells can map to hippocampal place cells by competitive learning. Network 17, 447–465. Rowland, D.C., and Moser, M.-B. (2014). From cortical modules to memories. Curr. Opin. Neurobiol. 24, 22–27, http://dx.doi.org/10.1016/j.conb. 2013.08.012. Sargolini, F., Fyhn, M., Hafting, T., McNaughton, B.L., Witter, M.P., Moser, M.-B., and Moser, E.I. (2006). Conjunctive representation of position, direction, and velocity in entorhinal cortex. Science 312, 758–762. Savelli, F., and Knierim, J.J. (2010). Hebbian analysis of the transformation of medial entorhinal grid-cell inputs to hippocampal place fields. J. Neurophysiol. 103, 3167–3183. Savelli, F., Yoganarasimha, D., and Knierim, J.J. (2008). Influence of boundary removal on the spatial representations of the medial entorhinal cortex. Hippocampus 18, 1270–1282. Schmidt-Hieber, C., and Ha¨usser, M. (2013). Cellular mechanisms of spatial navigation in the medial entorhinal cortex. Nat. Neurosci. 16, 325–331. Smith, S.L., Smith, I.T., Branco, T., and Ha¨usser, M. (2013). Dendritic spikes enhance stimulus selectivity in cortical neurons in vivo. Nature, in press. Solstad, T., Moser, E.I., and Einevoll, G.T. (2006). From grid cells to place cells: a mathematical model. Hippocampus 16, 1026–1031. Solstad, T., Boccara, C.N., Kropff, E., Moser, M.-B., and Moser, E.I. (2008). Representation of geometric borders in the entorhinal cortex. Science 322, 1865–1868. Squire, L.R., and Zola-Morgan, S. (1991). The medial temporal lobe memory system. Science 253, 1380–1386. Stensola, H., Stensola, T., Solstad, T., Frøland, K., Moser, M.-B., and Moser, E.I. (2012). The entorhinal grid map is discretized. Nature 492, 72–78. Taube, J.S., and Burton, H.L. (1995). Head direction cell activity monitored in a novel environment and during a cue conflict situation. J. Neurophysiol. 74, 1953–1971. Taube, J.S., Muller, R.U., and Ranck, J.B., Jr. (1990a). Head-direction cells recorded from the postsubiculum in freely moving rats. I. Description and quantitative analysis. J. Neurosci. 10, 420–435.

O’Keefe, J., and Nadel, L. (1978). The Hippocampus as a Cognitive Map (Oxford: Clarendon Press).

Taube, J.S., Muller, R.U., and Ranck, J.B., Jr. (1990b). Head-direction cells recorded from the postsubiculum in freely moving rats. II. Effects of environmental manipulations. J. Neurosci. 10, 436–447.

Ohtsuki, G., Nishiyama, M., Yoshida, T., Murakami, T., Histed, M., Lois, C., and Ohki, K. (2012). Similarity of visual selectivity among clonally related neurons in visual cortex. Neuron 75, 65–72.

Terrazas, A., Krause, M., Lipa, P., Gothard, K.M., Barnes, C.A., and McNaughton, B.L. (2005). Self-motion and the hippocampal spatial metric. J. Neurosci. 25, 8085–8096.

Neuron 80, October 30, 2013 ª2013 Elsevier Inc. 773

Neuron

Perspective Tolhurst, D.J., Dean, A.F., and Thompson, I.D. (1981). Preferred direction of movement as an element in the organization of cat visual cortex. Exp. Brain Res. 44, 340–342. Tolman, E.C. (1948). Cognitive maps in rats and men. Psychol. Rev. 55, 189–208. Van Essen, D.C., Anderson, C.H., and Felleman, D.J. (1992). Information processing in the primate visual system: an integrated systems perspective. Science 255, 419–423. Weliky, M., Bosking, W.H., and Fitzpatrick, D. (1996). A systematic map of direction preference in primary visual cortex. Nature 379, 725–728. Wilson, M.A., and McNaughton, B.L. (1993). Dynamics of the hippocampal ensemble code for space. Science 261, 1055–1058. Witter, M.P., and Groenewegen, H.J. (1984). Laminar origin and septotemporal distribution of entorhinal and perirhinal projections to the hippocampus in the cat. J. Comp. Neurol. 224, 371–385. Witter, M.P., Groenewegen, H.J., Lopes da Silva, F.H., and Lohman, A.H.M. (1989). Functional organization of the extrinsic and intrinsic circuitry of the parahippocampal region. Prog. Neurobiol. 33, 161–253.

774 Neuron 80, October 30, 2013 ª2013 Elsevier Inc.

Yartsev, M.M., Witter, M.P., and Ulanovsky, N. (2011). Grid cells without theta oscillations in the entorhinal cortex of bats. Nature 479, 103–107. Yoganarasimha, D., Yu, X., and Knierim, J.J. (2006). Head direction cell representations maintain internal coherence during conflicting proximal and distal cue rotations: comparison with hippocampal place cells. J. Neurosci. 26, 622–631. Yoon, K., Buice, M.A., Barry, C., Hayman, R., Burgess, N., and Fiete, I.R. (2013). Specific evidence of low-dimensional continuous attractor dynamics in grid cells. Nat. Neurosci. 16, 1077–1084. Zhang, S.J., Ye, J., Miao, C., Tsao, A., Cerniauskas, I., Ledergerber, D., Moser, M.-B., and Moser, E.I. (2013). Optogenetic dissection of entorhinal-hippocampal functional connectivity. Science 340, 1232627. Zhang, S.J., Ye, J., Couey, J.J., Witter, M.P., Moser, E.I., and Moser, M.-B. (2014). Functional connectivity of the entorhinal-hippocampal space circuit. Philos. Transact. R. Soc. London B Biol. Sci., in press.