Single-cell spatial confusion upon returning to familiar context in AppNL-G-F rats.

A. Schematic of the familiar-novel context alternating task. B. Representative rate maps of CA1 place cells across sessions. C. Quantification of remapping during the transition between context A and context B. Left: Spatial correlation. Middle: Population vector correlation. Right: Rate overlap. D. Difference in the degree of remapping between the initial transition from context A to context B (A1–B1) and the return transition from context B to context A (B2–A2). Left: ΔSpatial correlation. Middle: ΔPopulation vector correlation. Right: ΔRate overlap. E. Proportion of confused cells in the total population. Confused cells are defined as neurons whose spatial correlation between B2 and A2 exceeded the 95th percentile of the A1B1 baseline distribution. F. Representative raster plots of place cell spike trains. G. Left: SVM decoding accuracy within session B2. Right: SVM decoding accuracy within session A2. Data are presented as mean ± SEM

Implicit pairwise cofiring fails to discriminate distinct contexts in AppNL-G-F rats.

A. Schematic of explicit place code and implicit cofiring code. Left: A single place cell forms a spatial rate map, discarding temporal information. Right: Kendall’s rank correlation (τ) is calculated between spike trains of cell pairs, incorporating temporal information. B. Representative heatmaps of pairwise τ values in each session. The bottom number shows the correlation of τ vectors across sessions, defined as the Population Cofiring Correlation (PCC). C. PCC between different sessions.

Cross-day pairwise cofiring gradually discriminates distinct contexts in AppNL-G-F rats.

A. Schematic of PCC analysis across sessions. Purple: Cross-context correlations (Dark purple: A1B1; Light purple: B2A2). Yellow: Same-context correlations (Dark yellow: B1B2; Light yellow: A1A2). The discrimination index (ΔPCC) is defined as the difference between same-context and cross-context PCCs. B. Evolution of PCC values across training days. C. Evolution of ΔPCC across training days. D. Schematic of the SVM decoding workflow. τ vectors from A1 and B1 served as the training set, while those from B2 and A2 served as the testing set. To ensure unbiased comparison, optimal hyperparameters were identified for each day and cross-validated across all days. E. SVM decoding accuracy. Left: Overall accuracy for combined B2 and A2 sessions. Middle: Accuracy specific to session B2. Right: Accuracy specific to session A2.

Explicit spatial discrimination stagnates and decouples from cofiring dynamics in AppNL-G-F rats.

A. Representative rate maps of a cell pair. Left: Rate maps in context A. PORA denotes the place field overlap ratio, and τA denotes the pairwise Kendall’s tau in context A. Right: Same for context B. ΔPOR and Δτ denote the differences between contexts. B. Scatter plot of ΔPOR versus Δτ for all cell pairs. R2 indicates the goodness of fit, reflecting the predictive power of spatial overlap for temporal cofiring. C. Distribution of daily goodness of fit (R2). D. Representative rate maps of place cells across training days for WT and AppNL-G-F rats. E. Spatial correlation between context A and B across training days. F. Population vector correlation between context A and B across training days.

Position-independent cofiring dynamics evolve to discriminate contexts with learning.

A. Schematic of the Position Tuning Independent Rate (PIR) calculation. The expected rate (pink line) is derived by projecting the animal’s real-time position onto the spatial rate map (pink box), which displays clear place fields. The observed rate (gray line) is the actual firing rate. All rate vectors were computed in 100 ms bins. PIR is defined as the subtraction of the expected rate from the observed rate. Re-mapping the PIR onto spatial coordinates (black box) confirms the absence of place fields. B. Representative heatmaps of pairwise τPIR calculated using PIR. C. PCC of τPIR during repeated exposure (i.e., B2A2). D. SVM decoding accuracy using τPIR vectors. The decoding strategy follows the workflow outlined in Figure 3D.

Progressive increase in reactivation consistency during rest periods in AppNL-G-F rats.

A. Schematic illustration showing the coordination of cell pair firing by theta rhythms. The time lag between spikes is analyzed across two running directions: from place field 1 (PF1) to PF2 (Reference direction) and from PF2 to PF1 (Opposite direction). B. Representative cross-correlograms (CCG) of a cell pair are plotted as a function of the time lag between spikes. The black and gray lines represent movement in the reference and opposite directions, respectively. Note that the periodicity of the peaks resembles the theta rhythm cycle (∼100 ms). C. Frequency power spectra of the CCG shown in B. The peak value within the theta band (6-12 Hz) is defined as the theta index. D. Quantification of theta indices across all recorded cell pairs. E. Schematic of the experimental timeline showing the behavior task in which rats rested for ten minutes between each running session. F. Schematic illustrating the computation of neuronal reactivation consistency. For each SWR event, a population vector was constructed consisting of the spike counts from all recorded neurons. Reactivation consistency was then quantified by computing the correlation between these population vectors across different SWR events. G. Representative examples showing the changes in reactivation consistency across training days for a WT rat and an App NL-G-F rat. H. Group data quantifying the changes in reactivation consistency across training days.