Spatial Learning Drives Rapid Goal Representation in Hippocampal Ripples without Place Field Accumulation or Goal-Oriented Theta Sequences

Abstract

The hippocampus is critical for rapid acquisition of many forms of memory, although the circuit-level mechanisms through which the hippocampus rapidly consolidates novel information are unknown. Here, the activity of large ensembles of hippocampal neurons in adult male Long-Evans rats was monitored across a period of rapid spatial learning to assess how the network changes during the initial phases of memory formation and retrieval. In contrast to several reports, the hippocampal network did not display enhanced representation of the goal location via accumulation of place fields or elevated firing rates at the goal. Rather, population activity rates increased globally as a function of experience. These alterations in activity were mirrored in the power of the theta oscillation and in the quality of theta sequences, without preferential encoding of paths to the learned goal location. In contrast, during brief "offline" pauses in movement, representation of a novel goal location emerged rapidly in ripples, preceding other changes in network activity. These data demonstrate that the hippocampal network can facilitate active navigation without enhanced goal representation during periods of active movement, and further indicate that goal representation in hippocampal ripples before movement onset supports subsequent navigation, possibly through activation of downstream cortical networks.SIGNIFICANCE STATEMENT Understanding the mechanisms through which the networks of the brain rapidly assimilate information and use previously learned knowledge are fundamental areas of focus in neuroscience. In particular, the hippocampal circuit is a critical region for rapid formation and use of spatial memory. In this study, several circuit-level features of hippocampal function were quantified while rats performed a spatial navigation task requiring rapid memory formation and use. During periods of active navigation, a general increase in overall network activity is observed during memory acquisition, which plateaus during memory retrieval periods, without specific enhanced representation of the goal location. During pauses in navigation, rapid representation of the distant goal well emerges before either behavioral improvement or changes in online activity.

Keywords: hippocampus; place cell; replay; sharp-wave/ripple; spatial memory; theta oscillation.

Figure 1.

Improvement in behavioral performance across trials. A, Quantified behavioral trajectories (black line) across the first 8 trials for one session during goal-seeking (Goal) and random-foraging (Random) components, with start (green circle) and end (red square) denoted. B, E, H, For the first 15 trials across all sessions, mean ± SEM path length (B), latency (E), and mean velocity during active movement periods (H) during goal-seeking versus random-foraging segments. Pearson's linear correlation statistics for first 15 trials shown below graph. C, F, I, Mean ± SEM of path length (C), latency (F), and velocity (I) during Learning versus Retrieval periods defined by shaded regions in B, E, and H, respectively. D, G, J, Significance matrix calculated as in C, F, and I, but for different definitions of the Learning and Retrieval periods. Plotted is the p value for Learning versus Retrieval period. Highlighted square in D, G, and J represents the Learning and Retrieval period windows quantified in C, F, and I. K, For the first 15 trials, mean ± SEM number of times the rat entered a well partition during the random-foraging segment of the task for the Goal well (thick black line) and each other well (thin colored lines). L, Mean ± SEM of Goal well (black) and average non-Goal well (green) crossings during random-foraging portion of Learning versus Retrieval periods. Statistical tests and results: C, Two-way ANOVA. n = 8 sessions × 4 time points per group. Main effects: Learning versus Retrieval, F_(1,124) = 8.163, p = 0.005; Goal-seeking versus Random-foraging, F_(1,124) = 13.02, p = 0.0004; Interaction, F_(1,124) = 4.671, p = 0.0326. Post hoc multiple comparisons (Tukey's HSD): Learning-Goal versus Retrieval-Goal, adjusted p = 0.003; Retrieval-Goal versus Learning-Random, adjusted p < 0.0001; Retrieval-Goal versus Retrieval-Random, adjusted p = 0.0005; all other comparisons, adjusted p > 0.7. F, Two-way ANOVA. n = 8 sessions × 4 time points per group. Main effects: Learning versus Retrieval, F_(1,124) = 8.332, p = 0.0046; Goal-seeking versus Random-foraging, F_(1,124) = 4.231, p = 0.0418; Interaction, F_(1,124) = 6.684, p = 0.0109. Post hoc multiple comparisons (Tukey's HSD): Learning-Goal versus Retrieval-Goal, adjusted p = 0.001; Retrieval-Goal versus Learning-Random, adjusted p = 0.0036; Retrieval-Goal versus Retrieval-Random, adjusted p = 0.0072; all other comparisons, adjusted p > 0.9. I, Two-way ANOVA. n = 8 sessions × 4 time points per group. Main effects: Learning versus Retrieval, F_(1,124) = 0.7076, p = 0.4019; Goal-seeking versus Random-foraging, F_(1,124) = 4.086, p = 0.0454; Interaction, F_(1,124) = 2.213, p = 0.1394. Post hoc multiple comparisons (Tukey's HSD): all comparisons, adjusted p > 0.067. L, Two-way ANOVA. n = 8 sessions × 4 time points × 1 well for Goal groups, n = 8 sessions × 4 time points × 35 wells for Random groups. Main effects: Learning versus Retrieval, F_{(1, 2,300)} = 0.7216, p = 0.3957; Goal versus Random crossings, F_{(1, 2,300)} = 0.1569, p = 0.6921; Interaction, F_{(1, 2,300)} = 0.4626, p = 0.4965. **p ≤ 0.01. ***p ≤ 0.001. n.s., p > 0.05.

Figure 2.

Spatial representation does not change during goal-seeking versus random-foraging. A, Spatial tuning maps (place fields) for 8 example neurons calculated using spikes and position information either during the entire session (left), restricted to the goal-directed segments of the task (middle), or restricted to the random-foraging segments of the task (right). Maximum firing rate listed for each tuning curve in top left. B, Left, Mean ± SEM spatial correlation between place fields for each cell calculated during only goal-directed segments and place fields calculated during only random-foraging segments. Middle, Correlation between entire-session place fields for all recorded cells. Right, Correlation between place fields of different cells with overlapping fields (spatial bin of maximal firing <10 cm apart). C, Mean ± SEM decoding error across the entire session (using nonoverlapping decoding windows of 250 ms) when decoding with place fields calculated during the entire session (left), during only goal-directed segments (middle), or during only random-foraging segments (right). D, Mean excitatory population firing rate at each well (Goal well in red) for all 4 rats on experimental days 1 (top) and 2 (bottom). E, Mean ± SEM place field size (left) and information per spike (right) for place cells with peak firing closer to the Goal well than any other well (“Goal cells”), place cells with peak firing nearest the eight wells adjacent to the Goal well (“Goal-Adjacent”), and place cells with peak firing elsewhere in the arena (“Goal-Distant”). Statistical tests and results: B, One-way ANOVA. n = 1172 field pairs for Goal/Random; 101,482 field pairs for All-Fields; 1135 field pairs for Nearby. Main effects: F_{(2, 103,786)} = 12,130.39, p < 10⁻⁶; post hoc multiple comparisons (Tukey's HSD): all comparisons adjusted p < 10⁻⁸. C, One-way ANOVA. n = 8 sessions per group. Main effects: F_(2,21) = 1.3429; p = 0.2826. E, Left, One-way ANOVA. n = 44 Goal cells, 235 Goal-adjacent cells, 909 Goal-distant cells. Main effect: F_(2,1185) = 0.4388; p = 0.6449. E, Right, One-way ANOVA. n = 44 Goal cells, 235 Goal-adjacent cells, 909 Goal-distant cells. Main effect: F_(2,1185) = 2.2784; p = 0.1029. ***p ≤ 0.001. n.s., p > 0.05.

Figure 3.

Excitatory population activity increases across learning. A, For the first 15 trials across all sessions, mean ± SEM in-field firing rate per putative excitatory cell (black) or overall firing rate per putative inhibitory cell (blue). B, Mean ± SEM excitatory in-field (left) or inhibitory (right) firing rate during Learning and Retrieval periods defined by shaded regions in A. C, Significance matrix as in Figure 1D for excitatory in-field (top) or inhibitory (bottom) firing rate. D, Same as in A, separated by Goal cells (black), Goal-adjacent cells (red), and Goal-distant cells (cyan). E, Mean ± SEM in-field firing rate for Goal cells (left), Goal-adjacent cells (middle), and Goal-distant cells (right) during Learning and Retrieval periods defined by shaded regions in A. F, Mean ± SEM excitatory in-field firing during goal-seeking (black) or random-foraging (green) segments during Learning and Retrieval periods defined by shaded regions in A. Statistical tests and results. B, Left, Wilcoxon rank sum test. n = 1188 cells. p < 10⁻¹⁰. B, Right, Wilcoxon rank sum test. n = 50 cells. p = 0.4380. E, Two-way ANOVA. n = 44 Goal cells, 235 Goal-adjacent cells, 909 Goal-distant cells × 4 time points × 8 sessions. Main effects: Learning versus Retrieval, F_{(2, 6970)} = 76.53, p < 0.0001; Cell type, F_{(2, 6970)} = 2.052, p = 0.1286; Interaction, F_{(2, 6970)} = 2.054, p = 0.1284. Post hoc multiple comparisons (Tukey's HSD): all Learning versus Retrieval for same cell type, adjusted p < 0.0007; all across-cell-type for Learning, adjusted p > 0.14; all across-cell-type for Retrieval, adjusted p > 0.9. F, Two-way ANOVA. n = 1188 cells. Main effects: Learning versus Retrieval, F_{(1, 4187)} = 40.07, p < 0.0001; Goal-seeking versus Random-foraging, F_{(1, 4187)} = 2.145, p = 0.1431; Interaction, F_{(1, 4187)} = 3.056, p = 0.0805. Post hoc multiple comparisons (Tukey's HSD): all Learning-Goal versus Retrieval-Goal, adjusted p = 0.0109; Learning-Random versus Retrieval-Random, adjusted p < 0.0001; Learning-Goal versus Retrieval-Random, adjusted p < 0.0001; Learning-Random versus Retrieval-Goal, adjusted p = 0.0047; all other comparisons, adjusted p > 0.12. *p ≤ 0.05. ***p ≤ 0.001. n.s., p > 0.05.

Figure 4.

Theta power increases across learning. A, E, I, For the first 15 trials across all sessions, mean ± SEM theta power (A), percent of recorded cells active per theta oscillation (E), and number of spikes per participating cell per theta oscillation (I). B, F, J, Mean ± SEM of theta power (B), cell participation (F), and spikes per participating cell (J) during Learning versus Retrieval periods. C, G, K, Significance matrix as in Figure 1D for theta power (C), cell participation (G), and spikes per participating cell (K). D, H, L, Same as in B, F, and J, separated into goal-seeking (black) and random-foraging (green) segments. Statistical tests and results: B, Wilcoxon rank sum test. n = 8 sessions × 4 time points per group. p = 0.003975. D, Two-way ANOVA. n = 8 sessions × 4 time points per group. Main effects: Learning versus Retrieval, F_(1,124) = 11.66, p = 0.0009; Goal-seeking versus Random-foraging, F_(1,124) = 0.0578, p = 0.8105; Interaction F_(1,124) = 0.6253, p = 0.4307. Post hoc multiple comparisons (Tukey's HSD): Learning-Goal versus Retrieval-Goal, adjusted p = 0.0130; Learning-Random versus Retrieval-Random, adjusted p = 0.0404; all other comparisons, adjusted p > 0.1399. F, Wilcoxon rank sum test. n = 8 sessions × 4 time points per group. p = 0.2192. H, Two-way ANOVA. n = 8 sessions × 4 time points per group. Main effects: Learning versus Retrieval, F_(1,124) = 1.582, p = 0.2111; Goal-seeking versus Random-foraging, F_(1,124) = 0.001244, p = 0.9719; Interaction, F_(1,124) = 0.007203, p = 0.9325. J, Wilcoxon rank sum test. n = 8 sessions × 4 time points per group. p = 0.004707. L, Two-way ANOVA. n = 8 sessions × 4 time points per group. Main effects: Learning versus Retrieval, F_(1,124) = 11.71, p = 0.0009; Goal-seeking versus Random-foraging, F_(1,124) = 0.3030, p = 0.5831; Interaction, F_(1,124) = 0.07663, p = 0.7824. Post hoc multiple comparisons (Tukey's HSD): Learning-Goal versus Retrieval-Goal, adjusted p = 0.0371; Learning-Random versus Retrieval-Random, adjusted p = 0.0457; Learning-Random versus Retrieval-Goal, adjusted p = 0.0386; all other comparisons, adjusted p > 0.1549. *p ≤ 0.05. **p ≤ 0.01. n.s., p > 0.05.

Figure 5.

Spatial encoding across entire theta oscillations is unchanged by learning. A, Ensemble activity during movement was decoded using a single decoding window for each theta oscillation. For representative trials across a single session, the mean decoded posterior probability across all theta oscillations within that trial (colormap) and the rat's behavioral trajectory (cyan line). For clarity, only the goal-directed portion of each trial is displayed. B, E, For the first 15 trials across all sessions, mean ± SEM decoding error (B) and maximal posterior probability (E) for each theta oscillation per trial. C, F, Mean ± SEM of decoding error (C) and maximal posterior probability (F) during Learning versus Retrieval periods. Black represents all theta oscillations during active movement. Magenta and green represent theta oscillations during either the goal-directed navigation (magenta) or random-foraging (green) component of the task. D, G, Significance matrix as in Figure 1D for decoding error (D) or maximal posterior probability (G). Statistical tests and results: C, Two-way ANOVA. n = 8 sessions × 4 time points per group. Main effects: Learning versus Retrieval F_(2,186) = 1.326, p = 0.2510; Trial type F_(1,186) = 0.5138, p = 0.5991; Interaction F_(1,186) = 0.00847, p = 0.9916. F, Two-way ANOVA. n = 8 sessions × 4 time points per group. Main effects: Learning versus Retrieval, F_(2,186) = 2.726, p = 0.1004; Trial type, F_(1,186) = 0.03736, p = 0.9633; Interaction, F_(1,186) = 0.2129, p = 0.8084. n.s., p > 0.05.

Figure 6.

Theta sequences extend to movement direction, but not Goal location across learning. A, Each theta oscillation during movement (velocity ≥ 10 cm/s) was decoded in 15° steps and normalized to rat's current position and movement direction. Plotted is the probability histogram of theta phase with maximal posterior probability at each position relative to rat position (2 cm bins) for all movement-related theta oscillations during goal-seeking (Goal) or random-foraging (Random) components of Trials 1 and 10. B, For the first 15 trials across all sessions, mean ± SEM slope of weighted best-fit line of forward portion of decoded theta sequences oriented to rat's movement direction. C, Mean ± SEM of forward slope during Learning versus Retrieval periods. Black represents all theta oscillations during active movement, including Retrieval period with subsampled spike data (Ret. Sub.). Magenta and green represent theta oscillations during either the goal-directed navigation (magenta) or random-foraging (green) component of the task. D, E, Significance matrix as in Figure 1D for forward slope of weighted best fit line (D) or quadrant score (E). F, Same as in B, for theta sequences when rat was within 25 cm of Goal well and the difference between the rat's movement direction and the direction to the Goal well was >30° and <180°. Each theta oscillation was oriented to either the rat's current movement direction (solid line) or to the direction to the Goal (dashed line). G, Mean ± SEM of forward slope during Retrieval phase for theta sequences oriented to rat's movement direction (black) versus the direction to the goal well (gray). Statistical tests and results: C, Left, One-way ANOVA. n = 13,551 Learning oscillations, 10,313 Retrieval and Retrieval Subsample oscillations. F_{(2, 34,174)} = 51.109; p < 10⁻¹⁰). Post hoc multiple comparisons (Tukey's HSD): Learning versus Retrieval, adjusted p < 10⁻⁹; Learning versus Ret. Sub., adjusted p < 10⁻⁹; Retrieval versus Ret. Sub., adjusted p > 0.13. C, Right, Two-way ANOVA. n = 6892 Learning-Goal oscillations, 2953 Retrieval-Goal oscillations, 6659 Learning-Random oscillations, 7752 Retrieval-Random oscillations. Main effects: Learning versus Retrieval, F_{(1, 24,252)} = 82.56, p < 0.0001; Goal-seeking versus Random-foraging, F_{(1, 24,252)} = 1.243, p = 0.2649; Interaction, F_{(1, 24,252)} = 9.421, p = 0.0021. Post hoc multiple comparisons (Tukey's HSD): Retrieval-Goal versus Retrieval-Random, adjusted p = 0.5961; Learning-Goal versus Learning-Random, adjusted p = 0.0043; all other comparisons, adjusted p < 0.0001. G, Wilcoxon rank sum test. n = 1162 oscillations. p < 10⁻¹⁰. **p ≤ 0.01. ***p ≤ 0.001. n.s., p > 0.05.

Figure 7.

Ripple rate, but not other ripple properties, increases across learning. A-F, Same as in Figure 4A-C, for ripple rate per time immobile (A), ripple duration (B), ripple power (C), percent of recorded cells that participate in each ripple (D), firing rate of participating cells per ripple (E), and percent of ripples that encode statistically significant replay (F). Statistical tests and results: A, Wilcoxon rank sum test. n = 8 sessions × 4 time points per group. p = 8.495 × 10⁻⁶. B, Wilcoxon rank sum test. n = 8 sessions × 4 time points per group. p = 0.5836. C, Wilcoxon rank sum test. n = 8 sessions × 4 time points per group. p = 0.3647. D, Wilcoxon rank sum test. n = 8 sessions × 4 time points per group. p = 0.5134. E, Wilcoxon rank sum test. n = 8 sessions × 4 time points per group. p = 0.1458. F, Wilcoxon rank sum test. n = 8 sessions × 4 time points per group. p = 0.3964. ***p ≤ 0.001; n.s., p > 0.05.

Figure 8.

Representation of a goal emerges rapidly in ripples. A, Across all sessions, mean representation of the Goal well (red), the previous Random well (black), and all other wells (gray) in all statistically significant replay events occurring in Trials 1-4 (top), Trials 8-11 (middle), and Trials 15-18 (bottom). For each replay, representation of the well currently occupied by the rat was eliminated from analysis. *p < 0.05 (statistical outlier, Grubb's test). B, Representation of the Goal well (red), previous Random well (solid black), or average ± SEM of all other wells (dashed black) across the first 30 trials, using a four-trial average. *p < 0.05, Goal well is a statistical outlier from all other wells for that four-trial window (Grubb's test). The previous Random well is not a statistical outlier for any data point. C, Same as in B for ripples that do not encode statistically significant, spatially smooth trajectories. D, For both Learning (top) and Retrieval (bottom) phases, mean latency to return to Goal well following ripples which occurred while the rat was away from the Goal well and which represented the Goal well (“Goal Encoding,” red) or did not represent the Goal well (“No Goal Encoding,” black). Statistical tests and results: D, Two-way ANOVA. n = 10 Learning/Goal-encoding ripples, 16 Retrieval/Goal-encoding ripples, 27 Learning/non-Goal-encoding ripples, 56 Retrieval/non-Goal-encoding ripples. Main effects: Learning versus Retrieval, F_(1,105) = 17.53, p < 0.0001; Goal-encoding versus non-Goal-encoding, F_(1,105) = 11.12, p = 0.0012; Interaction, F_(1,105) = 3.266, p = 0.0736. Post hoc multiple comparisons (Tukey's HSD): Learning/Goal-encoding versus Learning/non-Goal-encoding, adjusted p = 0.0086; Retrieval/Goal-encoding versus Learning/not-Goal-encoding, adjusted p < 0.0001; Learning/non-Goal-encoding versus Retrieval/non-Goal-encoding, adjusted p < 0.0001; all other comparisons, adjusted p > 0.5. *p ≤ 0.05.