Grid cells provide a neural representation of space, by discharging when an animal traverses through the vertices of a periodic hexagonal grid spanning the environment1. Although grid cells have been characterized in detail in rats1, 2, 3, 4, 5, 6, the fundamental question of what neural dynamics give rise to the grid structure remains unresolved. Two competing classes of models were proposed: network models, based on attractor dynamics7, 8, 9, and oscillatory interference models, which propose that interference between somatic and dendritic theta-band oscillations (4–10 Hz) in single neurons transforms a temporal oscillation into a spatially periodic grid10, 11, 12, 13. So far, these models could not be dissociated experimentally, because rodent grid cells always co-exist with continuous theta oscillations4, 5, 6, 14. Here we used a novel animal model, the Egyptian fruit bat15, 16, to refute the proposed causal link between grids and theta oscillations. On the basis of our previous finding from bat hippocampus, of spatially tuned place cells in the absence of continuous theta oscillations17, we hypothesized that grid cells in bat medial entorhinal cortex might also exist without theta oscillations. Indeed, we found grid cells in bat medial entorhinal cortex that shared remarkable similarities to rodent grid cells. Notably, the grids existed in the absence of continuous theta-band oscillations, and with almost no theta modulation of grid-cell spiking—both of which are essential prerequisites of the oscillatory interference models. Our results provide a direct demonstration of grid cells in a non-rodent species. Furthermore, they strongly argue against a major class of computational models of grid cells.
At a glance
To elucidate the cellular and network mechanisms of grid cells in mammalian entorhinal cortex, we conducted electrophysiological recordings in a megabat, Rousettus aegyptiacus (Egyptian fruit bat). This bat species15 belongs to a different suborder of bats than the big brown bat16, the hippocampus of which we studied previously17, 18. Therefore, to set the foundations for subsequent entorhinal recordings in this bat species, we first examined neural activity in hippocampal area CA1 (Fig. 1a). Bats crawled in a large arena (Fig. 1b) in search of food, and the activity of individual neurons was recorded using tetrodes (Supplementary Figs 1 and 2). Of all the well-isolated excitatory neurons in CA1, 36% were classified as place cells, becoming active when the bat entered a particular region of the environment (23 out of 64 cells with spatial information >0.5 bits per spike). Locations of individual place fields spanned the entire experimental arena (Fig. 1c), similar to place cells in rats19. High-frequency ripple oscillations (120–160 Hz) were present in the local field potential (LFP) during sleep (Supplementary Fig. 3a), and were very similar in their properties to ripples in rat hippocampus20 (Supplementary Fig. 3 and Supplementary Text). Hippocampal theta oscillations in the LFP occurred in short, intermittent bouts (Fig. 1d), lasting typically ~1 s, and these bouts in the behaving bat were separated with an average 19-s interval (Supplementary Fig. 4a–d)—markedly different from the continuous theta oscillations observed in locomoting rats21, but similar to the intermittent theta bouts in monkeys22 and humans23. Because of the sparse occurrence of these bouts, theta oscillations were not evident in spectral analysis of the LFP, neither during sleep (Fig. 1e, left), during behavioural sessions (right), nor as a function of the bat’s velocity (Fig. 1f). Thus, in agreement with our previous findings in big brown bats17, we found that in Egyptian fruit bats, hippocampal place cells existed in the absence of continuous theta oscillations in the LFP.
We next turned to studying the medial entorhinal cortex (MEC) of bats, the region where grid cells are most prevalent in rats1, 2, 3, 4, 5, 6. As a first step, we determined the anatomical location and borders of MEC (Fig. 2a and Supplementary Figs 5 and 6). Immunohistochemical staining revealed that: (1) similar to rats24, the dorsal border of bat MEC (postrhinal border) was clearly identifiable by an abrupt change in layer structure (Supplementary Fig. 6b, d) and an abrupt transition in parvalbumin and calretinin staining (Fig. 2a and Supplementary Figs 5 and 6); and (2) similar to rats, the bat MEC had a densely packed layer II, sparser layers III and V, and a layer IV (lamina dissecans) with very few cell bodies (Supplementary Figs 5 and 6). Thus, the overall structure of bat MEC was very similar to that of rat MEC.
We targeted our tetrode recordings to the dorsal-most part of MEC: the region where we expected to find grid cells with the tightest and most pronounced grid structure, as in rats1, 2. The positions of all tetrodes were verified histologically by reconstructing the tetrode tracks and by electrolytic lesions (Fig. 2b–d and Supplementary Figs 7–9). Many MEC neurons were active at multiple locations in the arena (for example, Fig. 2e, f), and a hexagonal grid structure was revealed by computing the spatial autocorrelogram of their firing-rate map (Fig. 2g). To quantify the degree of 60° hexagonal rotational symmetry for each neuron, we employed a commonly used ‘gridness’ index2 (Supplementary Fig. 10): a higher index indicates a more hexagonal firing field. To test statistically whether a neuron is a grid cell, we used a standard shuffling procedure4, 5, 6 (Fig. 2h, i). Of the 70 well-isolated, behaviourally active neurons recorded in MEC, 36% were classified as grid cells (25 out of 70).
How similar are grid cells in bats versus rats? First, similar to rats1, grid vertices were separated by 60° angles (Fig. 3a), as expected from a hexagonal structure, and individual firing fields were almost equally spaced (mean standard deviation of grid spacing across the six inner peaks: 6.8 cm; n = 25 grid cells). Second, as in rats1, 2, 3, 4, 5, 6, co-localized grid cells in superficial layers of MEC shared similar grid orientation (tilt) and spacing (Fig. 3b and Supplementary Information). Third, as in rats1, cross-correlating the rate maps of simultaneously recorded grid cells revealed that their maps were offset in phase (Fig. 3c, central peaks of cross-correlograms are offset from the white cross), and the amount of offset spanned all possible phases (Fig. 3d). Fourth, as in rats1, the grid spacing of individual cells increased with the cell’s distance from the postrhinal border (Fig. 3e); this correlation was significant across all neurons (r = 0.47, P < 0.025), but was even more apparent when analysing separately each bat’s data (bat 1, r = 0.98, P < 0.02; bat 2, r = 0.67, P < 0.05; bat 3, r = 0.60, P = 0.07). Fifth, as in rats2, the firing rate increased with movement velocity (Fig. 3f; t-test, t24 = 2.9, P < 0.001). This correlation of firing rate and velocity probably explains the relatively low firing rates of MEC neurons in crawling bats (Fig. 3e, g) compared to rats1, 2, 3, 6, because bats crawled rather slowly (Supplementary Fig. 11). Finally, we found that in bat MEC, the diversity of spatial cell types was very similar to that known from rat MEC1, 2, 3, 4, 5, 6, 25 and contained: (1) ‘pure grid cells’1, 2 (Fig. 3g, lower-right quadrant; for example, cells 3, 4, 5), with no sensitivity to the animal’s head direction1, 2; (2) ‘conjunctive grid cells’2, 6 (Fig. 3g, upper-right quadrant; cell 2), which are grid cells tuned to a specific head direction2, 6; (3) head-direction cells2 that have no grid structure (Fig. 3g, upper-left quadrant; cells 8, 9; see also Supplementary Fig. 12); and (4) ‘border cells’25, which fired along geometrical borders of the environment (for example, Fig. 3g, cell 6). Some cells that did not cross the shuffling thresholds still exhibited clear spatial patterns characteristic of grid cells (Fig. 3g, cell 10) or head-direction cells (cell 7). Taken together, these results indicate that the detailed properties of grid cells (and other cell types) in bat MEC were very similar to those in rat MEC.
We next turned to address the central question of this study, and asked whether the grids can exist without key elements of the oscillatory interference models10, 11, 12, 13—namely, without continuous theta oscillations in the LFP, and in the absence of theta modulation of grid-cell firing. As a first step in studying the LFP in MEC, we examined non-theta-band, high-frequency ripple oscillations, which are most prominent during sleep20 (Supplementary Fig. 3d). Ripples in bat MEC were very similar in their properties to ripples in rat MEC (Supplementary Fig. 3d–f and Supplementary Text). However, despite the similarity in ripple oscillations, bats differed markedly from rats in the nature of theta oscillations. First, unlike rats4, 5, 6, 14, 26, we never observed a prominent continuous theta-band oscillation in the LFP, irrespective of the recording site in MEC and the type of reference used (Fig. 4a, b and Supplementary Information). The LFP power-spectrum showed no theta peak, neither during sleep (Supplementary Fig. 13, left), during behaviour (Fig. 4a,b, ‘Behav.’), as a function of the animal’s velocity (Fig. 4a, b, coloured panels), nor as a function of the bat’s echolocation mode17 (Supplementary Fig. 14a–d). Second, unlike the continuous theta oscillations typically observed in rat MEC4, 5, 14, 26, theta oscillations in bat MEC occurred in short intermittent bouts, both during behaviour (Fig. 4c–e and Supplementary Fig. 4f, h) and during sleep (Supplementary Fig. 4e, g), similar to bat CA1; 92% of theta bouts during behaviour had duration ≤1 s (Fig. 4d). Theta bouts were separated by very long inter-bout intervals (Fig. 4e; average interval: 37 ± 2 s; 18% of intervals were >1 min). The bat’s velocity and echolocation rate were not different during theta bouts versus non-theta epochs (Supplementary Fig. 14e, f). Third, we computed the spike-train temporal autocorrelations for individual grid cells (Fig. 4f, top and Supplementary Fig. 15), and examined the degree of theta modulation in these autocorrelations by computing a standard ‘theta index’ that was used in previous studies in rats4, 5, 6, 26; this index is based on the relative theta-band power in the Fourier transform of the temporal autocorrelation (Fig. 4g and Supplementary Information). Using either the same criterion that was employed previously to identify theta modulation in rat grid cells6 (theta index >5), or using a statistical shuffling procedure for each individual spike train, we found that nearly all grid cells in bat MEC (24 out of 25) did not exhibit theta-modulated firing (Fig. 4h, left; theta index across all grid cells 1.29 ± 0.82 (mean ± s.d., maximal value 2.9)), irrespective of the bat’s velocity (Supplementary Fig. 16). Because, in the rat MEC, neurons recorded in layers II and III show the most pronounced theta modulation of neuronal firing6, 27, we also analysed separately the data recorded from layers II and III of bat MEC; these analyses showed that no significant theta modulation was present in any of the 35 neurons (of all classes) recorded in layers II and III and in any of the multi-units from these layers (Supplementary Fig. 17). Fourth, because the firing of most neurons in rat MEC (especially in superficial layers) is theta-modulated6, and they have similar phases28, we also examined the temporal periodicity of the multi-unit activity, where firing rates are much higher than in individual neurons, and hence oscillations might be detected more robustly. We found that 100% of multi-unit sites where grid cells were recorded (17 out of 17) did not exhibit theta-modulated firing (Fig. 4h, right; theta index 0.97 ± 0.56; maximal value 2.1; see also Supplementary Figs 15–17). Fifth, because rat MEC neurons are often locked to a specific phase of the theta cycle14, 27, 28, we examined whether spikes are locked to bats’ theta phase during theta bouts. We found that bat MEC neurons indeed exhibited a clear, albeit weak, phase-locking during theta bouts (Fig. 4i, bottom, grey; see also Supplementary Fig. 18). Importantly, no phase locking could be observed outside the theta bouts (Fig. 4i, red). The contrast between phase locking of spikes during theta bouts versus lack of locking outside the bouts (Fig. 4i) indicates that theta bouts in bats are truly discrete and locally generated events. Sixth, to examine the possible contribution of theta bouts to grid formation, we removed in each grid cell all theta-bout epochs, and re-computed the firing-rate maps and two-dimensional autocorrelograms; this did not cause substantial alterations in the grid pattern (Fig. 4j, k and Supplementary Fig. 19). Notably, only a minority of spikes emitted by any single grid cell occurred during theta bouts (4.4 ± 0.75%). In fact, in some grid cells, the grid field existed in the absence of any spikes emitted during theta bouts (Fig. 4k, right; zero spikes emitted during all the theta bouts). Population analysis confirmed that theta-bout removal did not lead to significant changes in gridness values, in any of the grid cells (Fig. 4l, 100% of the cells showed changes in gridness that did not exceed the 95% confidence intervals). This suggests that the grids are maintained during times when theta oscillations are not present.
Taken together, these findings provide the first report on grid cells in a non-rodent species, which supports the generality of the grid-; more importantly, our findings causally dissociate the link between the existence of grids and the existence of continuous theta-band oscillations in the mammalian entorhinal cortex. The similarities in the anatomy and grid-cell properties between bats and rats strongly suggest that similar underlying neural mechanisms generate the grid, indicating that the functional dissociation between theta and grids generalizes across mammalian species. Although it is possible that continuous theta oscillations would be found in bats during flight, we emphasize that, under the specific crawling conditions of our experiments, we observed simultaneously the existence of grids without continuous theta oscillations in the LFP or in the spiking activity, which strongly argues against the theta-based class of computational models of grid cells10, 11, 12, 13, but is consistent with the other models which do not rely on theta oscillations7, 8, 9.
Recently, two studies showed that inactivation of medial septum inputs to rat MEC disrupts grid fields, and also disrupts theta oscillations in MEC28, 29. Additionally, these studies found that, after medial septum inactivation, the firing rates of MEC grid cells dropped by >40%, on average, and in many cases the firing rates dropped threefold and even fivefold28, 29. These observations have been interpreted as supporting the oscillatory interference models28, 29, but they equally well support network models of grid cells: it is well known in the theory of neural networks that the removal of a major input to a network, if accompanied by a marked decrease in firing rates, can drive the network into a very different activity regime30, which could disrupt grid formation. Thus, inactivation of a major input to a brain network28, 29 cannot serve to dissociate oscillation-based models from network models. In contrast, our study in bats did not manipulate the inputs to the entorhinal network, and thus provided a unique opportunity to causally challenge the oscillatory interference models of grid cells. More generally, we provide here a rare example of causally disproving a major class of computational models of a higher brain area.
Immunohistochemical stainings were conducted to delineate the anatomical location and borders of the medial entorhinal cortex (MEC) in Egyptian fruit bats (Rousettus aegyptiacus). Single neuron activity and local field potentials (LFP) were recorded from hippocampal area CA1 and MEC of five bats (two and three bats, respectively), using tetrodes17, 18, 19. Neuronal activity and positional data were collected while bats foraged in a large arena (117 × 117 cm) in search of food. Place cells19 in CA1 were identified using a criterion of spatial information >0.5 bits per spike (ref. 17). Grid cells and head-direction cells were identified using a standard gridness index2 and mean vector length index of the head-direction tuning4, 5; significance of these two indices was tested using a random shuffling procedure similar to that described previously in rats4, 5. To quantify grid properties, we computed grid spacing, orientation, phase and velocity-modulation of the cell’s firing1, 2. High-frequency ripple oscillations in the LFP during sleep were detected as transients in the power of the filtered LFP trace (80–160 Hz) exceeding 7 s.d. above the mean power17. Theta oscillation epochs were defined as 2-s windows in which the ratio between the power in the theta (4–8 Hz) and delta (2–4 Hz) frequency ranges exceeded 2.0. Theta modulation of neuronal firing was assessed using a standard theta index, which is based on the spectral power of the spike train temporal autocorrelogram4, 5, 6. Detailed experimental and analytical procedures are provided in the Supplementary Information.
- Microstructure of a spatial map in the entorhinal cortex. Nature 436, 801–806 (2005) , , , &
- Conjunctive representation of position, direction, and velocity in entorhinal cortex. Science 312, 758–762 (2006) et al.
- Experience-dependent rescaling of entorhinal grids. Nature Neurosci. 10, 682–684 (2007) , , &
- Development of the spatial representation system in the rat. Science 328, 1576–1580 (2010) et al.
- Development of the hippocampal cognitive map in preweanling rats. Science 328, 1573–1576 (2010) , , &
- Grid cells in pre- and parasubiculum. Nature Neurosci. 13, 987–994 (2010) et al.
- A spin glass model of path integration in rat medial entorhinal cortex. J. Neurosci. 26, 4266–4276 (2006) &
- Path integration and the neural basis of the ‘cognitive map’. Nature Rev. Neurosci. 7, 663–678 (2006) , , , &
- Accurate path integration in continuous attractor network models of grid cells. PLOS Comput. Biol. 5, e1000291 (2009) &
- An oscillatory interference model of grid cell firing. Hippocampus 17, 801–812 (2007) , &
- Temporal frequency of subthreshold oscillations scales with entorhinal grid cell field spacing. Science 315, 1719–1722 (2007) , , &
- Grid cell firing may arise from interference of theta frequency membrane potential oscillations in single neurons. Hippocampus 17, 1252–1271 (2007) , &
- Scale-invariant memory representations emerge from Moiré interference between grid fields that produce theta oscillations: a computational model. J. Neurosci. 27, 3211–3229 (2007) , &
- Hippocampus-independent phase precession in entorhinal grid cells. Nature 453, 1248–1252 (2008) , , , &
- Optimal localization by pointing off axis. Science 327, 701–704 (2010) , , &
- What the bat’s voice tells the bat’s brain. Proc. Natl Acad. Sci. USA 105, 8491–8498 (2008) &
- Hippocampal cellular and network activity in freely moving echolocating bats. Nature Neurosci. 10, 224–233 (2007) &
- Dynamics of hippocampal spatial representation in echolocating bats. Hippocampus 21, 150–161 (2011) &
- Dynamics of the hippocampal ensemble code for space. Science 261, 1055–1058 (1993) &
- High-frequency oscillations in the output networks of the hippocampal-entorhinal axis of the freely behaving rat. J. Neurosci. 16, 3056–3066 (1996) &
- Theta oscillations in the hippocampus. Neuron 33, 325–340 (2002)
- Hippocampal theta activity in monkeys. Brain Res. 538, 59–63 (1991) &
- Human hippocampal theta activity during virtual navigation. Hippocampus 15, 881–889 (2005) et al.
- Anatomical organization of the parahippocampal-hippocampal network. Ann. NY Acad. Sci. 911, 1–24 (2000) , , &
- Representation of geometric borders in the entorhinal cortex. Science 322, 1865–1868 (2008) , , , &
- Theta modulation in the medial and the lateral entorhinal cortices. J. Neurophysiol. 104, 994–1006 (2010) , , &
- Theta oscillations provide temporal windows for local circuit computation in the entorhinal-hippocampal loop. Neuron 64, 267–280 (2009) , , &
- Reduction of theta rhythm dissociates grid cell spatial periodicity from directional tuning. Science 332, 595–599 (2011) et al.
- The spatial periodicity of grid cells is not sustained during reduced theta oscillations. Science 332, 592–595 (2011) , , &
- Modeling feature selectivity in local cortical circuits. In Methods in Neuronal Modeling: from Synapses to Networks 2nd edn (eds & ) 499–567 (MIT Press, 1998) &
We thank D. Derdikman, S. Romani, M. Ahrens and Y. Cohen for comments on the manuscript, M. Melcón for initial assistance with hippocampal CA1 recordings, M. Weinberg for veterinary oversight, and R. Eilam and C. Ra'anan for histology. This study was supported by research grants from the Israel Science Foundation and Minerva Foundation to N.U., by a Lev-Zion predoctoral excellence fellowship to M.M.Y., as well as by grants from the Norwegian Research Council and the Kavli Foundation to M.P.W.
- Supplementary Information (18.1M)
The file contains Supplementary Figures 1-22 with legends, Supplementary Methods, Supplementary Text and Data and additional references.