Fig. 4: Sweeps extend to never-visited locations.
From: Left–right-alternating theta sweeps in entorhinal–hippocampal maps of space
a, Decoded (internal) direction signals (PV correlation, based on head-direction tuning curves from the same session, all MEC–parasubiculum cells) point towards inaccessible and never-visited locations along an elevated linear track (left) or a wagon-wheel track (right). The decoded direction (arrows of constant length) is shown for successive theta cycles over the course of 3-s running segments. Alternating theta cycles are shown in different colours; black, running trajectory from the segment; grey, full trajectory. b, Sweeps on the wagon-wheel track decoded from a single grid module based on rate maps from a preceding open-field session (inset shows an example rate map). The decoded position during each theta cycle is plotted on top of the animal’s running trajectory. c, The LMT model allows decoding to include never-visited locations. The LMT model is fitted to the neural data by iteratively updating a latent 1D and 2D trajectory. Orange and blue arrows and lines show latent direction (top) and position trajectory (bottom) during 17 theta cycles from a wagon-wheel session at different stages of the model-fitting (left to right). The full running trajectory is shown in light grey. The latent direction and position signals are initialized with the rat’s actual head direction and running trajectory (iteration 1) but evolve into sweep-like trajectories that cover the 2D space surrounding the maze (iteration 150). d, Sequence of four successive sweeps (1–4) and concurrent internal direction during navigation on an elevated wagon-wheel maze (Bayesian decoder, based on fitted LMT tuning curves for all MEC–parasubiculum cells). Each video frame shows the internal direction (green arrow, length is fixed) and high-probability positions (coloured blobs, as in Fig. 1a, bottom) during one sweep. Note that sweeps travel into the inaccessible space inside and outside the navigable track. Scale bar, 0.5 m. e, Top, circular histogram showing head-centred distribution of fitted LMT internal direction values from one example session on the wagon-wheel track (left) and one example session on the linear track (right). Note that internal direction is bimodally distributed around the animal’s head direction (wagon wheel, 38.2° and 9.1° to either side, left–right alternation in 83.4% and 68.3% of theta cycles, n = 2 rats; linear track, 28.3° ± 8.3° to either side, left–right alternation in 78.1 ± 2.3% of theta cycles, mean ± s.e.m. from 7 rats), as in the open field (Fig. 2). Bottom, lines show sweeps averaged across each recording session during theta cycles that followed a left (red) or right (blue) sweep (wagon-wheel, 2 rats; linear track, 7 rats). f, Out-of-bounds sweeps consistently coincide with internal direction pointing towards the same location. Colour-coded 2D histograms of conditional occurrences of the two LMT latent variables (internal direction and sweeps in head-centred coordinates) for sweeps that terminate inside (top) or outside (bottom) visited portions of the environment in one animal (rat 25843, the same session as a–c). The circular correlation coefficient between head-centred internal direction and sweep directions was similar when analysis was confined to theta cycles in which sweeps terminated inside versus outside the wagon-wheel track: r = 0.83 versus r = 0.82. Similar results were obtained for a second rat with fewer cells (not shown): r = 0.34 versus r = 0.35. g, Firing-rate maps of a grid cell on the wagon wheel, based on either the original position coordinates (left) or the latent position fitted by the LMT model (right). h, Firing-rate maps of the same grid cell in an open-field session recorded on the same day. Note that the LMT model infers the continuation of grid-like periodic tuning to locations beyond the environment boundaries. Credit: rat, scidraw.io/Gil Costa.
