Grid-Layout and Theta-Modulation of Layer 2 Pyramidal Neurons in Medial Entorhinal Cortex

See allHide authors and affiliations

Science  21 Feb 2014:
Vol. 343, Issue 6173, pp. 891-896
DOI: 10.1126/science.1243028

Entorhinal Cell Clusters

There is considerable interest in understanding the function of neurons in layer 2 of the medial entorhinal cortex and how they generate their unique firing patterns, which are important in the recall of facts and past events (see the Perspective by Blair). Ray et al. (p. 891, published online 23 January) investigated principal cells in layer 2 by immunoreactivity, projection patterns, microcircuit analysis, and assessment of temporal discharge properties in awake, freely moving animals. In tangential sections, pyramidal neurons were clustered into patches arranged in a hexagonal grid—very similar to the patterns observed in grid cell spatial firing. These patches received selective cholinergic innervation, which is critical for sustaining grid cell activity. Kitamura et al. (p. 896, published online 23 January) found that these cells drive a hippocampal circuit by projecting directly to the hippocampal CA1 area and synapsing with a distinct class of inhibitory neurons. This circuit provides feed-forward inhibition in combination with excitatory inputs from layer 3 cells of the medial entorhinal cortex, projecting to CA1 pyramidal cells to determine the strength and time window of temporal associative inputs.


Little is known about how microcircuits are organized in layer 2 of the medial entorhinal cortex. We visualized principal cell microcircuits and determined cellular theta-rhythmicity in freely moving rats. Non–dentate-projecting, calbindin-positive pyramidal cells bundled dendrites together and formed patches arranged in a hexagonal grid aligned to layer 1 axons, parasubiculum, and cholinergic inputs. Calbindin-negative, dentate-gyrus–projecting stellate cells were distributed across layer 2 but avoided centers of calbindin-positive patches. Cholinergic drive sustained theta-rhythmicity, which was twofold stronger in pyramidal than in stellate neurons. Theta-rhythmicity was cell-type–specific but not distributed as expected from cell-intrinsic properties. Layer 2 divides into a weakly theta-locked stellate cell lattice and spatiotemporally highly organized pyramidal grid. It needs to be assessed how these two distinct principal cell networks contribute to grid cell activity.

Temporal (13) and spatial (4) discharge patterns in layer 2 of the medial entorhinal cortex (MEC) are related through phase precession (5) and the correlation of gridness (hexagonal regularity) and theta-rhythmicity (2). Layer 2 principal neurons divide into pyramidal and stellate cells, the latter of which have been suggested to shape entorhinal theta (6, 7) and grid activity (8) by their intrinsic properties. Progress in understanding entorhinal microcircuits has been limited because most though not all data (911) stem from extracellular recordings of unidentified cells. Such recordings have characterized diverse functional cell types (1214) in layer 2. Clustering of grid cells (15) points to spatial organization. It is not clear, however, how functionally defined cell types correspond to stellate and pyramidal cells (7, 16), which differ in conductances, immunoreactivity, projections, and inhibitory inputs (6, 1720). We combined juxtacellular labeling with principal cell identification (20) to visualize microcircuits in the MEC (Fig. 1A).

Fig. 1 Grid-like arrangement of calbindin+ pyramidal cells in the MEC.

(A) Posterior view of a rat cortical hemisphere. LEC, lateral entorhinal cortex; PaS, parasubiculum; Per, perirhinal cortex; Por, postrhinal cortex. (B) Calbindin-immunoreactivity (brown precipitate) in a parasaggital section reveals patches with apical dendrites of calbindin+ pyramidal cells forming tents (white arrows) in layer 1. (C) Tangential section showing all neurons (red, NeuN-antibody) and patches of calbindin+ neurons (green). Bracket, dashed lines indicate the patch-free stripe of MEC. (D) Inset from (C). (E) Two-dimensional spatial autocorrelation of (D) revealing a hexagonal spatial organization of calbindin+ patches. Color scale, –0.5 (blue) through 0 (green) to 0.5 (red); grid score is 1.18. Scale bars, (A) 1 mm; (B) 100 μm; (C) to (E) 250 μm. D, dorsal; L, lateral; M, medial; V, ventral.

Calbindin immunoreactivity (20) identifies a relatively homogeneous pyramidal neuron population in MEC layer 2. Parasagittal sections stained for calbindin (Fig. 1B) showed that calbindin-positive (calbindin+) pyramidal cells were arranged in patches (21). Apical dendrites of calbindin+ pyramidal cells bundled together in layer 1 to form tent-like structures over the patches (Fig. 1B). The patchy structure is well defined at the layer 1/2 border, whereas a “salt-and-pepper” appearance of calbindin+ and calbindin cells is observed deeper in layer 2 (fig. S1). Patches contained 187 ± 70 cells (111 ± 42, ~60% calbindin+; 76 ± 28, ~40% calbindin cells; counts of 19 patches from four brains). We double-stained tangential sections for calbindin (green) and the neuronal marker NeuN (red) to visualize patches in the cortical plane. Calbindin+ (green/yellow) patches covered the MEC except for a 400- to 500-μm-wide patch-free medial stripe adjacent to the parasubiculum (Fig. 1C). Clustering was not observed in calbindin neurons (red) (Fig. 1C). We noted a striking hexagonal organization of calbindin+ patches (Fig. 1, C and D) and characterized this organization by means of three techniques. (i) We used two-dimensional spatial autocorrelation analysis (4), which captures spatially recurring features and revealed a hexagonal regularity (Fig. 1E). (ii) We modified grid scores (12) to quantify hexagonality also in elliptically distorted hexagons (22), distortions that result from tissue curvature and anisotropic shrinkage. Grid scores range from –2 to +2, with values >0 indicating hexagonality. The example in Fig. 1D had a grid score of 1.18, suggesting a high degree of hexagonality. (iii) We assessed the probability of hexagonal patch arrangements given preserved local structure (14) by means of a shuffling procedure. We found that the strongest Fourier component of the sample (Fig. 1D) exceeded that of the 99th percentile of shuffled data, suggesting that such hexagonality is unlikely to arise by chance.

We retrogradely labeled neurons from ipsilateral dentate gyrus (Fig. 2A) using biotinylated dextran amine (BDA) (Fig. 2B) or cholera toxin B (Fig. 2C) to investigate the arrangement of layer 2 principal cells with identified projection patterns and immunoreactivity (20). Although most retrogradely labeled neurons were stellate cells (16, 23), a small fraction had pyramidal morphologies, but these neurons appeared larger than calbindin+ pyramidal cells (Fig. 2B). Calbindin+ neurons did not project to the dentate gyrus (only 1 double-labeled out of 313 neurons in Fig. 2, C to E) (20). Calbindin+ patches were hexagonally arranged (Fig. 2, C, D, and F), whereas dentate-gyrus–projecting neurons (red) were uniformly distributed (Fig. 2, E and G). Reconstructions of calbindin+ and calbindin cells labeled in vivo confirmed their pyramidal and stellate morphologies, respectively. Calbindin+ dendrites were largely confined to patches, whereas calbindin stellates cells had three times larger dendritic trees (7.6 versus 2.6 mm average total length, P < 0.03), which extended unrelated to patches (Fig. 2, H and I). Differentiating layer 2 neurons by calbindin and reelin immunoreactivity confirmed patchy hexagonality of calbindin+ cells and scattered distribution of reelin+ cells without overlap between these neurons (fig. S2) (20).

Fig. 2 Calbindin+ pyramidal but not dentate-projecting stellate neurons form patches.

(A) Schematic of retrograde labeling from dentate gyrus. (B) Such retrograde labeling (BDA, brown) stains neurons (most with stellate morphologies) in a parasaggital MEC section. (C) Tangential MEC section showing calbindin+ neurons (green) and retrogradely labeled neurons (red) after dentate-gyrus–cholera–toxin-B injection. (D and E) Insets from (C). (F) Two-dimensional spatial autocorrelation of (D) reveals regular organization of calbindin+ patches; grid score is 0.32. The strongest Fourier component of the sample exceeded that of the 99th percentile of shuffled data confirming hexagonality. (G) Two-dimensional spatial autocorrelation of (E) reveals no spatial organization; grid score is –0.03. (H and I) Superimposed reconstructions of dendritic morphologies of 5 calbindin+ pyramidal (green) and 5 calbindin stellate neurons (black) in the tangential plane. Morphologies were “patch-centered” aligned according to orientation and the center of the nearest calbindin+ patch (gray outlines). Scale bars, (B) 100 μm; (C) to (E) and (G) to (I) 250 μm. D, dorsal; L, lateral; M, medial; V, ventral.

To investigate the organization of calbindin+ patches across the MEC, we prepared flattened whole-mount preparations. Patches had similar arrangements throughout the dorsoventral extent of the MEC (fig. S3). At the layer 1/2 border, we consistently observed hexagonal arrangements in well-stained specimens. We quantified patch size and spacing in 10 largely complete MEC whole-mounts. Patch density was similar throughout the MEC, whereas patch diameter slightly increased toward ventral (fig. S3). We estimated 69 ± 17 patches across the entire MEC (n = 10 hemispheres). Calbindin patches stained also positive for cytochrome-oxidase activity (9). However, the two staining patterns were not the same because calbindin patches were more sharply delineated than were spots revealed by cytochrome-oxidase activity, and cytochrome-oxidase staining revealed many more patches than did calbindin staining in the MEC (9). Moreover, the staining patterns did not correspond at all in the parasubiculum.

Calbindin+ patches shared a roughly 60° symmetry of their axes (Fig. 3A). One axis runs parallel to the dorsoventral axis of the parasubiculum (Fig. 3, A and B). Lines fitted through the dorsoventral axis of the parasubiculum, and the most medial column of calbindin+ patches had the same orientation (Fig. 3B). A second consistent axis was tilted ~60° relative to the dorsoventral axis. This calbindin+ patch axis curved ventrally at more lateral positions and aligned with the orientation of overlaying layer 1 myelinated axons (Fig. 3, C to F). Thus, the line connecting diagonally neighboring calbindin patches (revealed by spatial autocorrelation) (Fig. 3, D and E) aligned with the orientation of layer 1 axons (Fig. 3F). We quantified the orientation of axonal segments by a polar plot shown in Fig. 3G and confirmed that layer 1 axons share one main orientation in the MEC (9, 24, 25).

Fig. 3 Alignment of the calbindin grid to parasubiculum, layer 1 axons, and cholinergic markers.

(A) Section from Fig. 1C. Dashed white lines indicate axes of the calbindin+ grid (angles are indicated). Axes aligned with parasubiculum (B) and layer 1 axons [(C) to (G)]. (B) (Left) Schematic of calbindin patches and parasubiculum from (A). The orange line fits the dorsoventral axis of the parasubiculum, and the green line fits the most medial column of patches (red); the angle between these lines is indicated. (Right) Fitted lines and their relative angles for four other brains. (C) Tangential section processed for calbindin (green) and myelin basic protein (red). (D) Inset from (C). (E) Two-dimensional spatial autocorrelation of (D). Dashed black lines indicate grid axes. (F) Inset from (C). (G) Axonal segments in (F) were manually traced from left to right, and we computed a polar plot (red) of the orientations of the axonal segments. The orientations of axonal segments aligned with one axis of the grid of calbindin patches [superimposed dashed lines from (E)]. (H) Tangential section stained for acetylcholinesterase activity. (I) Section from (H) costained for calbindin. (J) Overlay of (H) and (I) shows overlap between acetylcholinesterase and calbindin staining. Scale bars, (A), (C) to (F), (H), and (I) 250 μm; (J) 100 μm. D, dorsal; L, lateral; M, medial; V, ventral.

MEC function and grid cell activity (26, 27) depend on medial septum inputs (28, 29) and cholinergic transmission (30). We observed a patchy pattern of acetylcholinesterase labeling at the layer 1/2 border (Fig. 3H), which colocalized with the cores of calbindin+ patches (Fig. 3, H to J). Axonal terminals positive for the vesicular acetylcholine transporter (VAChT) were closely apposed to calbindin+ cells, and their density was twofold larger in calbindin+ patches than between patches (fig. S4). We also stained for m1 muscarinic receptors and observed a diffuse labeling without colocalization of these receptors to VAChT puncta. Moreover, we analyzed the apposition and distribution of presynaptic VAChT puncta relative to dendrites of in vivo filled calbindin+ and calbindin layer 2 cells by means of confocal microscopy. VAChT puncta were much more abundant around calbindin+ than calbindin layer 2 cells, but proximity histograms of VAChT puncta and dendrites did not indicate a direct targeting of calbindin+ cell dendrites by cholinergic synapses (fig. S4). Both the m1 receptor labeling and our dendrite- VAChT puncta colocalization analysis are in line with a volumetric action of acetylcholine in the MEC (3133).

Last, we assessed in freely moving animals how activity of identified neurons related to the entorhinal theta-rhythm. We recorded 31 layer 2 neurons in rats trained to explore open fields and classified them by morphology and immunoreactivity. Calbindin+ neurons (n = 12) were pyramidal cells, whereas calbindin neurons (n = 19) had stellate morphologies. Firing rates were not different (calbindin+ = 2.1 ± 1.1 Hz; calbindin = 2.3 ± 1.5 Hz; P > 0.5, Mann-Whitney test). We found, however, that calbindin+ neurons (Fig. 4, A to C) showed stronger theta-rhythmicity of spiking than that of calbindin cells (P < 0.01, unpaired t test) (Fig. 4, D to G). Theta-rhythmicity was associated with locomotion of the animal (fig. S5). A similar twofold difference in theta-rhythmicity between calbindin+ (n = 14) and calbindin (n = 20) cells was observed under urethane-ketamine anesthesia (P = 0.0003, Mann-Whitney test) (Fig. 4H), which preserves cortico-hippocampal theta-rhythmicity (3, 34). Pharmacological blockade of cholinergic transmission suppressed theta-rhythmicity in both calbindin+ and calbindin cells (Fig. 4I). Specifically, we observed that cholinergic blockade led to a loss of the distinct peak at theta-frequency in the power spectra of spike discharges (fig. S6). Cells also differed in their phase-locking to entorhinal field potential theta: Calbindin+ cells were more strongly phase-locked (average Rayleigh vector length = 0.54 versus 0.22 in calbindin cells; P < 0.0012, Mann-Whitney test) and fired near the trough of the theta-oscillation, whereas locking was weaker and more variable in calbindin cells (Fig. 4J).

Fig. 4 Theta-modulation of calbindin+ positive and calbindin cells.

(A) (Left) Micrograph (tangential section) of a calbindin+ neuron recorded in a freely moving animal. Green, calbindin; red, neurobiotin. (Right) Soma in red, green channel, and overlay. (B) Autocorrelogram of spike discharges for the calbindin+ neuron shown in (A). (C) Filtered (4 to 12 Hz) local field potential (top) and spiking pattern (bottom) of the neuron shown in (A). (D to F) Same as (A) to (C) but for a calbindin neuron. (G) Strength of theta-rhythmicity in calbindin+ and calbindin neurons in freely moving animals. Numbers are n of neurons. Error bars indicate SEM. (H) Same as (G) but for recordings under urethane-ketamine anesthesia (34). (I) Theta-rhythmicity in calbindin+ neurons (green, n = 8) and calbindin neurons (black, n = 7) under anesthesia before and after systemic cholinergic blockade with scopolamine (Wilcoxon signed rank test, P = 0.0078 for calbindin+, P = 0.0156 for calbindin cells). Dots indicate medians. (J) Polar plot of preferred theta-phase (theta-peak = 0°) and modulation strength (Rayleigh vector, 0 to 1, proportional to eccentricity) for calbindin+ (green) and calbindin (black); dots indicate single cells, and lines indicate averages. Scale bars, (A) and (D) 100 μm (left), 10 μm (right).

What is the cellular basis of theta-rhythmicity in MEC layer 2? Stellate cells have been prime candidates for theta discharges in layer 2 (6, 7) because intrinsic conductances make them resonate at theta-frequency (35, 36). We found, however, that calbindin+ pyramidal cells showed twofold stronger theta-rhythmicity and theta-phase-locking than calbindin stellate neurons. The stronger theta-rhythmicity of calbindin+ pyramidal neurons, which have weaker sag-currents (7, 20), is opposite from what had been predicted on the basis of intrinsic properties (8, 37). Hence, layer 2 theta-modulation is cell-type–specific but not distributed as expected from cell-intrinsic resonance properties. This finding agrees with other evidence that questioned the causal relationship between intrinsic properties and theta-rhythmicity in vivo (10, 37, 38). The membrane properties of calbindin+ neurons are not tuned to the generation of theta-rhythmicity (20). Their strongly rhythmic discharges suggest that calbindin+ neurons might correspond to a subset of cells with strong membrane potential theta-oscillations (11), which—in the absence of cell-intrinsic mechanisms—probably arise from synaptic interactions. Cholinergic innervation and effects of cholinergic blockade suggest cholinergic drive sustains theta-rhythmicity of calbindin+ cells.

We were not yet able to assess spatial modulation in a sufficient number of identified neurons to directly relate our results to grid cell function. The limited available evidence suggests that grid cells are recruited from a heterogeneous neuronal population in layer 2 (10, 11, 39), possibly indicating weak structure-function relationships (40). Yet, we observed similarities between calbindin+ neurons and grid cells: Calbindin+ patches receive cholinergic inputs, which are required for grid cell activity according to preliminary data (30); calbindin+ cells have strong theta-rhythmicity, a feature that correlates with grid cell discharge (2); and, like grid cells, calbindin+ cells are clustered.

We have hypothesized that calbindin+ neurons form a “grid-cell-grid” (41)—that their hexagonal arrangement might be an isomorphism to hexagonal grid cell activity, much like isomorphic cortical representations of body parts in tactile specialists (42, 43). However, hexagonality often results from spacing constraints and hence might be unrelated to grid cell activity. Determining the spatial modulation patterns of identified entorhinal neurons will help clarifying whether and how the calbindin+ grid is related to grid cell activity.

Supplementary Materials

Materials and Methods

Figs. S1 to S6

References (4457)

  • Present address: Werner Reichardt Centre for Integrative Neuroscience, Otfried-Müller-strasse 25, 72076 Tübingen, Germany.

References and Notes

  1. Acknowledgments: This work was supported by Humboldt Universität zu Berlin, Bernstein Center for Computational Neuroscience Berlin [German Federal Ministry of Education and Research (BMBF), Förderkennzeichen 01GQ1001A], NeuroCure, the Neuro-Behavior European Research Council grant, and the Gottfried Wilhelm Leibniz Prize of the Deutsche Forschungsgemeinschaft. We thank C. Ebbesen, M. von Heimendahl, R. Rao, J. Steger, J. Tukker, U. Schneeweiß, P. Turko, and I. Vida.

Stay Connected to Science

Navigate This Article