Functional network topography of the medial entorhinal cortex

Abstract

The medial entorhinal cortex (MEC) creates a map of local space, based on the firing patterns of grid, head-direction (HD), border, and object-vector (OV) cells. How these cell types are organized anatomically is debated. In-depth analysis of this question requires collection of precise anatomical and activity data across large populations of neurons during unrestrained behavior, which neither electrophysiological nor previous imaging methods fully afford. Here, we examined the topographic arrangement of spatially modulated neurons in the superficial layers of MEC and adjacent parasubiculum using miniaturized, portable two-photon microscopes, which allow mice to roam freely in open fields. Grid cells exhibited low levels of co-occurrence with OV cells and clustered anatomically, while border, HD, and OV cells tended to intermingle. These data suggest that grid cell networks might be largely distinct from those of border, HD, and OV cells and that grid cells exhibit strong coupling among themselves but weaker links to other cell types.

Keywords: entorhinal cortex; grid cells; spatial coding; topography; two-photon microscopy.

Fig. 1.

Calcium imaging with the 2P miniscope in MEC. (A) Schematic of the miniscope. Table shows speed, numerical aperture (NA) specifications, and weight. Em. NA, emission NA; Exc. NA, excitation NA; HC-PCF, hollow-core photonic crystal fiber; TFB, tapered fiber bundle. (B) Sagittal brain section schematics at two medio-lateral (ML) positions. Green, GCaMP expression; red, retro-AAV expressing tdTomato. (Scale bar, 500 µm.) (C) Nissl-stained, sagittal brain slices from one animal. The approximate medio-lateral (ML) position is indicated. Green lines show the region accessible through GRIN + prism. The ventral implant border is indicated by black arrowheads. Mouse ID is indicated (Top Left). (Scale bar, 500 µm.) (D) Example maximum-intensity projection of one open-field session with a total of 213 detected cells (after SNR filtering). Colored regions show a subset of extracted cells. Numbers indicate Suite2p cell IDs. (Scale bar, 50 µm.) (E) Fluorescent (normalized ΔF/F, black traces) and deconvolved and filtered signal (green vertical bars) of 20 example cells (colored ROIs in D). Total session length, ∼1,200 s. The shaded region (60-s excerpt) is magnified (Right).

Fig. 2.

2P miniscope imaging of all functional cell types in MEC during unrestrained behavior. (A) Schematic of behavioral protocol. Initial head fixation on a wheel and foraging in a square open-field arena (“Standard protocol”). When screening for OV cells [“Object sessions (optional)”], we added either one object which was moved between two object sessions, or two objects in a single session. (B) Example path plots (animal number is indicated Top Left). Session time, exploration ratio, and exploration SD are indicated. Colored dots (orange and blue) are referenced in C. (Scale bars, 20 cm.) (C) Analysis of coverage (occupancy) of the open-field environment for 203 sessions recorded in 15 animals (80 × 80 cm² open-field arena, duration 1,453 ± 610.6 s, mean ± SD). Plotted is the exploration ratio (“ratio”) versus exploration SD (“std”). Stippled lines show arbitrary cutoffs, splitting the data into quadrants of “better” (Top Right) and “worse” coverage (Bottom Left). Sessions shown in B are indicated in orange and blue. (D–G) Example grid (D), border (E), HD (F), and OV (G) cells recorded during ∼20 min of free foraging in an 80 × 80 cm open field. (D) Grid cells. (D, Left) Spatial tuning maps for grid cells of a module with short grid spacing (Top) and two co-recorded grid cells from a module with larger spacing (Middle and Bottom). White lines indicate the position of a (single) cue card, fixed to the wall of the box. (D, Center) Two-dimensional autocorrelations of spatial tuning maps on the Left (GS, grid score). (D, Right) Path and deconvolved signal (amplitude is indicated by dot size, and session time is indicated by color). (E) Border cells. (E, Left) Spatial tuning maps of border cells with border score (BS) indicated above the spatial tuning map. The first two cells (animal 87244, cells 27 and 10) were co-recorded. (E, Right) Path and deconvolved signal (amplitude is indicated by dot size). (F) HD cells. Path and deconvolved signal (Left) and tuning curves (Right) of three example HD cells. Tuning strength (mean vector length; MVL) is displayed above the tuning curve. Deconvolved signal is shown color-coded by HD on top of the path plot. Tuning curves show normalized occupancy (gray line) and the cell’s directional tuning curve (black line). (G) OV cells. Spatial tuning maps of three OV cells, two with one field (Top and Middle), and one with two fields (Bottom). White squares with a black cross indicate the position of the object. OV score (OVS) is displayed above the Center tuning map. Color maps have the same limits across all sessions for one cell (range of baseline session).

Fig. 3.

Grid cells and OV cells occupy discrete regions of MEC. (A and B) Anatomical distribution of grid and OV cells and spatial tuning map examples in two different animals (animal name and recording date are at Top Left). Cells are color-coded by tuning strength (OV score, blue; grid score, red). Cells not meeting cutoff criteria are shown in gray. (Scale bars, 50 µm.) (A) Recording with 98 grid cells and 3 OV cells. (B) Recording with 27 OV cells and 9 grid cells. Tissue orientation is indicated by a cross (D, dorsal; L, lateral; M, medial; V, ventral). (C) Fraction of (pure) grid versus OV cells in 36 sessions across 13 animals. Each dot represents a session and animals are coded by color. Spearman’s correlation is shown (Top Right) (Spearman’s r = −0.596, n_Grid = n_OV = 36; ***P = 0.00013). (D) Summary of correlations as in C, but for all cell types. Spearman’s r is color-coded and shown as text in the center of each box (filtered by MEC, minimum cutoff of 15 cells after filtering, only pure cell types). A thick gray border around some squares indicates significance (P < 0.05) in Spearman’s correlation result. See SI Appendix, Extended Results for detailed statistics. (E) Fraction of (pure) grid and OV cells and the discrimination index over grid and OV cells in each session across six example animals (minimum cutoff of 15 cells), with higher fractions of grid cells than OV cells (Top) or the opposite (Bottom). Pie charts show the average percentage of grid and OV cells across sessions. Boxes in boxplots extend from lower to upper quartiles of the data; the horizontal line indicates population median; whiskers indicate 1st to 99th percentile; dashed line indicates 0. (F, Left) Summary across all sessions and animals shown in E. Grid: six animals, Mann–Whitney U = 320, n_Grid = n_OV = 18, ***P = 3.82e-07, two-sided; OV: six animals, Mann–Whitney U = 3, n_Grid = n_OV = 16, ***P = 2.70e-06, two-sided. (F, Right) Fraction of cell types (OV or grid) and discrimination indices in comparison with a shuffled distribution. Asterisks indicate results of two-sided Wilcoxon signed-rank tests of the data versus the population mean of the shuffled data (*P < 0.05, **P < 0.01, ***P < 0.001). Boxplot properties are as in E. See SI Appendix, Extended Results for detailed statistics.

Fig. 4.

Anatomical separation of grid cells from other functional cell types in MEC. (A) Example cell maps of two recordings (OV animals 87245 [322 cells] and 88106 [309 cells]), showing the distribution of HD (blue), grid (orange), border (green), and OV (pink) cells. Cells are color-labeled if they crossed shuffling cutoff criteria for not more than one cell class. All other corecorded cells are shown in light gray if they crossed none or dark gray if they crossed more than one cutoff criterion. (Scale bars, 50 µm.) (B) Schematic of the NN analyses shown in this figure. The NN distances were extracted from size-matched (subsampled) populations of starter cells. Lines indicate NN (A to B) distances (line thickness indicates distance; thinner = longer). (C) Representative cell maps and NN distance analyses for six sample pairwise comparisons across multiple animals. Graphs below cell maps show the cumulative distribution and boxplots over NN distances (black) in comparison with a shuffled distribution (gray). Numbers in Insets indicate population averages. Minimum number of starter cells is five; minimum cell center distance is 10 µm. Boxes extend from lower to upper quartile values of the data, with a vertical line indicating the population average; whiskers indicate 1st to 99th percentile. (Scale bars, 50 µm.) (D) Normalized mean NN distances over all sessions and animals. (D, Left) Scatterplot showing one dot per dataset (95th percentile shuffling cutoffs were used throughout). Statistics above indicate results of Mann–Whitney U test (*P < 0.05, **P < 0.01). (D, Right) Color-coded average of results on the Left. Asterisks indicate results of two-sided one-sample t test against a population mean of 1 (*P < 0.05, **P < 0.01, ***P < 0.001); ns, not significant (P > 0.05). See SI Appendix, Extended Results for detailed statistics. (E) Spring-loaded network model; each node represents one cell class. Edges in the graph are color-coded by weight (1/normalized NN distance). (E, Top) Graph shows an example result of one simulation after 1,000 simulation steps. (E, Middle) More examples with increasing ratios. (E, Bottom) Results of graph distance ratios after shuffling weights (1,000 permutations) versus data (shuffled: lighter gray, median ratio 0.8 [small dashed line in histogram]; data: darker gray, median ratio 5.0 [large dashed line in histogram]). Ratios of the example graphs shown above (graphs A–D) are indicated.

Fig. 5.

Grid cells cluster anatomically. (A) Schematic of NN analyses. The mean NN distances in C and E were extracted from groups of five starter cells (blue in schematic). Lines indicate distances to the five nearest neighbors of the example cell in the center. This group size (“NN group”) is systematically varied in D. (B–D) Grid animals (Top) and OV animals (Bottom). The minimum number of cells above the cutoff for each panel was set at 15. (B) Two sessions with multiple grid cells (“Grid animal”; Top) or multiple OV cells (“OV animal”; Bottom) (color-coded grid score or OV score); gray shows cells that did not meet cutoff criteria. (Scale bars, 50 µm.) (C) Normalized NN distances. (C, Left) Lines indicate recording sessions and colors represent animals (mean ± SEM is shown by a black line). (C, Right) Boxplots: Boxes extend from lower to upper quartiles; the orange line indicates the median, and whiskers indicate 1st to 99th percentile; outliers are shown as open black circles, and black plus signs indicate the mean. Statistics above line plots indicate results of two-sided Mann–Whitney U test, and above boxplots indicate two-sided Wilcoxon signed-rank test (against 1) (*P < 0.05, **P < 0.01, ***P < 0.001); ns, not significant (P > 0.05). See SI Appendix, Extended Results for detailed statistics. (D) Mean NN distance over varying numbers of neighbors (NN group) for grid cells in grid animals (Top) and OV cells in OV animals (Bottom). Thin black and blue lines show single recordings and thick lines show group average ± SEM. Significance (*P < 0.05, **P < 0.01, ***P < 0.001) indicates the result of two-sided Mann–Whitney U test, data vs. reference. Vertical labels show the total number of animals and sessions that were used in each comparison (for both C and D). See SI Appendix, Extended Results for detailed statistics. (E) Normalized, mean NN distances and statistics as in C for HD cells (Top) and border cells (Bottom) including all animals. See SI Appendix, Extended Results for detailed statistics.

Fig. 6.

Functional cell types cover distinct and stable anatomical territories. (A) Four example topographic tuning maps, color-coded by score, are shown for one animal with multiple FOVs aligned as in SI Appendix, Fig. S7C. Left to Right: grid score (Grid), boundary vector score (Border), HD tuning expressed as MVL (HD), and spatial information content (“Spatial information”). (A, Bottom) Distribution of shuffled Moran’s I values for each tuning map and actual values (“Data”; red); 95th and 99th percentile cutoffs are shown as dashed lines. (Scale bar, 50 µm.) (B) Overlay and correlation of topographic tuning maps for different tuning properties. Maps were first smoothed with a Gaussian kernel with sigma 2 bins. The number of combined sessions in each example is shown above (n). Diagonal lines indicate which maps were combined and smoothed (composites in A). (B, Bottom) Pearson’s correlation (Pearson’s r) between the different tuning maps, compared with a shuffled distribution. Actual values are shown in red (Data), and 95th and 99th percentile cutoffs are shown as dashed lines. (Scale bar, 50 µm.) (C) Topographic tuning map correlation results (dots indicate animals). Data are shown as the difference of the actual Pearson’s r value (Data) and the median of the shuffled distribution of Pearson’s r values (“Shuffled”). Statistics at the top of the figure indicate statistically significant (P < 0.05) results of two-sided Wilcoxon signed-rank test against a mean of zero (*P < 0.05, **P < 0.01, ***P < 0.001). See SI Appendix, Extended Results for detailed statistics. (D) Summary of topographic tuning map correlation results. Data are shown as a color-coded representation of the difference in the number of results that are above the 95th percentile shuffling cutoff and the number of results that are below the 5th percentile cutoff over all sessions (same data as analyzed in C). Grid × border −0.53, grid × HD −0.67, grid × OV −0.1, border × HD 0.07, border × OV −0.1, HD × OV −0.2. Grid/border as well as grid/HD territories are most strikingly anticorrelated compared with the rest.