doi: 10.1038/s41593-025-01928-z.
Epub 2025 Apr 14.
Hippocampal output suppresses orbitofrontal cortex schema cell formation
Affiliations
- PMID: 40229506
- PMCID: PMC12081290
- DOI: 10.1038/s41593-025-01928-z
Item in Clipboard
Hippocampal output suppresses orbitofrontal cortex schema cell formation
Nat Neurosci.
2025 May.
Abstract
Both the orbitofrontal cortex (OFC) and the hippocampus (HC) are implicated in the formation of cognitive maps and their generalization into schemas. However, how these areas interact in supporting this function remains unclear, with some proposals supporting a serial model in which the OFC draws on task representations created by the HC to extract key behavioral features and others suggesting a parallel model in which both regions construct representations that highlight different types of information. In the present study, we tested between these two models by asking how schema correlates in rat OFC would be affected by inactivating the output of the HC, after learning and during transfer across problems. We found that the prevalence and content of schema correlates were unaffected by inactivating one major HC output area, the ventral subiculum, after learning, whereas inactivation during transfer accelerated their formation. These results favor the proposal that the OFC and HC operate in parallel to extract different features defining cognitive maps and schemas.
© 2025. This is a U.S. Government work and not under copyright protection in the US; foreign copyright protection may apply.
Conflict of interest statement
Competing interests: The authors declare no competing interests.
Figures
a, Schematic illustrating the events of a trial in the odor-sequence task. The illumination of two overhead house lights indicated the start of each trial. After poking into the central odor port and sampling the presented odor, rats could respond with a ‘go’ to obtain a sucrose reward or a ‘no-go’ to avoid a prolonged intertrial interval. b, Odor-sequence task illustrated as two virtual figure-of-eight mazes. Ten odors were organized into two sequence pairs (S1 and S2), each comprising two subsequences (a and b). Each subsequence consists of four trials or positions (P1–P4) indicated by odor numbers. Red +, rewarded; black −, nonrewarded; 0–9, odor identities; arrows indicate sequence transitions. c, Reconstruction of recording locations in the lateral OFC. The approximate extent of recording locations in each rat is represented by red squares. d, Virus expression. An adeno-associated virus (AAV) carrying the soma-targeted GtACR2-FusionRed construct under the CaMKIIa promoter was injected into the ventral subiculum (vSub) bilaterally. GtACR2-expressing neurons were identified using immunohistochemistry (red, GtACR2; blue, DAPI). GtACR2-expressing neurons were found in the vSUB and dentate gyrus (DG). This experiment was independently repeated across all eight animals, yielding consistent results. Individual neurons expressing GtACR2 are magnified in the dashed white box. Scale bars, 1 mm (left) and 10 µm (right). e, Reconstruction of GtACR2 expression and optical fiber placements in the vSUB. The maximal and minimal extents of GtACR2 expression are indicated by purple and green colors, respectively, and red dots indicate optical fiber placement. f,g, Percentage correct (f) and latency to poke into the odor port to initiate a trial after light onset (g) on each trial type in S1a, S2a (above y axis), S1b, S2b (below y axis) for control (left) and GtACR2 (right) sessions (gray, maze 1; green, maze 2). The error bars represent the s.e.m. Four-way ANOVAs confirmed the significant main effects of position on both measures (percentage correct: F(3,1405) = 145.5, P = 4.2 × 10−82, = 0.24; poke latency: F(3,1405) = 889.1, P = 1.0 × 10−323, = 0.66; n = 45 sessions for control; n = 44 sessions for GtACR2), with reward driving more accurate and faster performance. Further regression analyses on the latency to initiate trials showed that this measure was affected by whether the reward was to be delivered on both the current and the next trials (g, right for control and GtACR2). Notably, in these analyses, there were no effects of inactivation (F < 0.82; P > 0.36; < 0.0006; n = 45 sessions for control; n = 44 sessions for GtACR2). The error bars are the s.e.m. (see Extended Data Fig. 1). Source data
a–c, Heatmaps of the OFC neurons showing epoch-specific (a), reward-specific (b) and position-specific (c) firing in the figure-of-eight task. In each panel, the heatmap shows average activity in each epoch at each position in one maze. Individual squares corresponding to each epoch are magnified in the black dashed box at the top. Arrows represent sequence directions. A red asterisk marks the reward epoch on rewarded trial types (P3 and P4), whereas a black asterisk marks the reward epoch for nonrewarded trial types (P1 and P2). d–f, Plots show the percentage of the OFC neurons with firing that was significantly modulated by epoch (d), reward (e) and position (f) (ANOVA, P < 0.01), with each neuron assigned to the condition of maximal firing. There were no effects of inactivation (χ2 < 1.42; P > 0.23; degree of freedom (d.f.) = 1; χ2 test). Red denotes the chance level.
a–d, Heatmaps (left) and mean firing rate at each position and epoch (right) for OFC neurons showing generalization of activity related to epoch (a), reward (b) and position (c) or a combination of factors (d). Heatmaps plot activity as described in Fig. 2. Line plots show the average firing rate in each epoch at each position in each maze, ordered according to the reward on the current and next trials. The gray line represents maze 1 and the green line maze 2. The firing rates were not significantly different between maze 1 and maze 2 at all epochs in each example (P > 0.10; two-sided Wilcoxon’s rank-sum test; n = 8 positions for each maze of each neuron). The error bars are the s.e.m. (Extended Data Figs. 2–4).
a, Correlation in firing across mazes for all OFC neurons recorded in control and GtACR2 sessions. The plots show the distribution of r scores with the neurons that met the arbitrary cutoff for classification as schema cells (r > 0.8, P < 0.01, correlation coefficients) shown in orange, nonschema cells (r ≥ 0.4 and r ≤ 0.8, correlation coefficients) in dark gray and noncoding cells (r < 0.4, correlation coefficients) in light gray. b, Percentage of schema neurons at different thresholds for categorization. There was no difference between the two groups in the proportion of neurons at any threshold value (χ2 = 0.67; P = 0.41; d.f. = 1; χ2 test). c, Percentage of schema (Sch.), nonschema (Nonsch.) and noncoding (Noncod.) neurons from control and GtACR2 sessions on each day of training (using thresholds in a). There was no difference between the two groups in the proportion of neurons at any day (χ2 < 2.1; P > 0.15; d.f. = 1; χ2 test). d–f, Explained variance, averaged across neurons, for each factor (epoch, reward, position) within maze in the schema (d; n = 877 units for control; n = 910 units for GtACR2), nonschema (e; n = 771 units for control; n = 722 units for GtACR2) and noncoding (f; n = 208 units for control; n = 202 units for GtACR2) populations. There were no effects of inactivation (P > 0.09; two-tailed Student’s t-test). g, Accuracy of decoding position across all epochs by individual schema cells, where → denotes chance decoding of 12.5%. One-way ANOVA showed that accuracy was similar for decoding within and across mazes for neurons in control (F(1,3710) = 0.068; P = 0.79; = 1.8 × 10−5) and GtACR2 sessions (F(1,366) = 0.016; P = 0.90; = 4.4 × 10−6) and there was no significant effect of inactivation (within: F(1,368) = 0.45; P = 0.50; = 1.2 × 10−4; across: F(1,3688) = 0.35; P = 0.55; = 9.5 × 10−5). h, Accuracy of decoding position within each epoch by ensembles of schema cells, where the dotted line denotes chance decoding of 12.5%. A one-way ANOVA showed that accuracy was similar for decoding within and across mazes for neurons in control (F(1,16) = 0.02; P = 0.88; = 1.5 × 10−3) and GtACR2 sessions (F(1, 16) = 0.31; P = 0.58; = 0.02) and there was no significant effect of inactivation (within: F(1,16) = 0.37; P = 0.55; = 0.023; across: F(1, 16) = 0; P = 0.96; = 1.6 × 10−4; Supplementary Figs. 2–5 and 10).
a–c, Scatter plots showing the correlation coefficients of each neuron from the control (left) and GtACR2 (right) sessions. The y axes plot the correlation coefficients from unshuffled data and the x axes the mean correlation coefficients obtained after shuffling data (1,000×) to disrupt contributions of information related to the epoch (a), reward (b) or position (c). Orange, gray or black cells had actual correlation coefficients >99% of the shuffled results, indicating a significant contribution of the shuffled type of information to the correlated firing patterns. These populations, the percentage of the total of that category noted on the panels, were not affected by inactivation (χ2 < 3.4; P > 0.066; d.f. = 1; χ2 test). Orange denotes schema cells and gray nonschema cells. d, Venn diagrams summarizing data from a to c, showing the fraction of schema neurons recorded in control and GtACR2 sessions that were affected by the shuffling of information related to epoch (light gray), reward (light green) and position (dark gray). The sizes of circles are normalized to the total number of neurons recorded in each group and proportions in each category that overlap between categories were not affected by inactivation (χ2 < 0.40; P > 0.54; d.f. = 1; χ2 test). e, Food consumption across trials in the neophobia task. Lines show new food consumed per trial as a percentage of familiar food. Light green, control and deep green, GtACR2. A three-way ANOVA revealed a significant main effect of novelty (F(1,104) = 9.11; P = 0.0032; = 0.081; n = 5 trials for both groups) and a significant interaction between the novelty and group (F(1,104) = 4.05; P = 0.047; = 0.038; n = 5 trials for both groups). Further testing showed a significant difference between groups on the last three (F(1,34) = 7.21; P = 0.01; = 5.6 × 10−3; n = 3 trials for both groups) but not the initial two trials (F(1,22) = 0.95; P = 0.34; = 0.042; n = 3 trials for both groups). The error bars are the s.e.m.
a,b, Percentage correct (a) and trial initiation latencies (b) across days of learning for rats in the control and GtACR2 groups. The ANOVAs revealed significant effects of session, trial type, group, an interaction between session and trial type, and an interaction between trial type and group (F > 5.4; P < 0.021; > 0.03; n = 10 d for both control and GtACR2) in the percentage correct, reflecting quicker development of the no-go response on nonrewarded positions in the inactivated group at the early stages of learning (days 2–6) (− in scatter plots: t53 = 2.7; P = 9.4 × 10−3; two-tailed Student’s t-test; n = 28 for control; n = 27 for GtACR2) and a significant main effect of trial type and an interaction between group and trial type (F > 5.8, P < 0.0007, > 0.043; n = 10 d for both control and GtACR2) in the trial initiation latencies, reflecting a failure of rats in the inactivated group to distinguish the two nonrewarded positions (P1 versus P2 in the scatter plots) (control: t104 = 5.2, P = 1.2 × 10−6; n = 53 sessions; GtACR2: t104 = 0.29, P = 0.77; two-tailed Student’s t-test; n = 53 sessions). −, nonrewarded trials; +, rewarded trials. The error bars are the s.e.m. (***P < 0.001; **P < 0.01).
a–c, Percentage of cells recorded on each day that met criteria as schema (a), nonschema (b) and noncoding (c) neurons in each group. The prevalence of schema and nonschema neurons in the two groups were similar initially and then diverged thereafter, with schema neurons increasing more rapidly (overall: χ2 = 78.9, P = 6.6 × 10−19; days 1–2: χ2 = 1.6; P = 0.21; days 3–10: χ2 = 83.0, P = 8.0 × 10−20; d.f. = 1; χ2 test) and nonschema neurons declining more rapidly (overall: χ2 = 59.2, P = 1.4 × 10−14; days 1–2: χ2 = 0.87; P = 0.35; days 3–10: χ2 = 66.3, P = 3.9 × 10−16; d.f. = 1; χ2 test) in the inactivated group, with no effects of learning or inactivation on the noncoding neurons (χ2 < 1.8; P > 0.18; d.f. = 1; χ2 test). d–f, Average explained variance for each factor (epoch, reward, position) within the maze in the schema (d; n = 260 for control; n = 863 for GtACR2), nonschema (e; n = 516 for control; n = 758 for GtACR2) and noncoding (f; n = 181 for control; n = 330 for GtACR2) populations. Inactivation resulted in modest but significant increases in all three kinds of information in the schema and nonschema neurons (schema: P = 0.036; 1.5 × 10−4; 2.6 × 10−7; two-tailed Student’s t-test; nonschema: P = 5.8 × 10−8; 2.1 × 10−5; 3.6 × 10−9; two-tailed Student’s t-test; noncoding: P = 0.87, 0.40, 0.76, two-tailed Student’s t-test). No adjustments were made for multiple comparisons (***P < 0.001; *P < 0.05; Supplementary Figs. 8 and 9).
a, Accuracy of decoding position across all epochs by individual schema cells, where → denotes chance decoding of 12.5%. One-way ANOVA showed that accuracy was similar for decoding within and across mazes for neurons in control (F(1,518) = 0.12, P = 0.73, = 2.3 × 10−4) and GtACR2 rats (F(1,724) = 3.0, P = 0.083, = 0.0017), whereas inactivation increased accuracy of decoding (within: F(1,1121) = 12.48, P = 4.0 × 10−4, = 0.011; across: F(1,1121) = 36.9, P = 1.7 × 10−9, = 0.032). b, Accuracy of decoding position within each epoch by ensembles of schema cells, where dotted line denotes chance decoding of 12.5%. One-way ANOVA showed thta accuracy was greater within than across mazes for neurons in control (F(1,16) = 12.1, P = 3.1 × 10−4, = 0.43) but not GtACR2 rats (F(1,16) = 2.1, P = 0.17, = 0.11), and inactivation caused better decoding across (F(1,16) = 23.4, P = 2.0 × 10−4, = 0.59) but not within the maze (F(1,16) = 0.52, P = 0.48, = 0.032). c–e, Percentage of schema neurons with correlated activity across mazes affected by shuffling (as in Fig. 5a–c) to disrupt information related to epoch (c), reward (d) or position (e). No significant differences between the two groups were observed for either epoch or position (χ2 < 3.3, P > 0.067, d.f. = 1; χ2 test), whereas the influence of reward grew modestly but significantly faster with inactivation (overall schema: χ2 = 17.1, P = 3.6 × 10−5; days 1–2: χ2 = 2.2, P = 0.14; days 3–10: χ2 = 8.9, P = 0.0028; d.f. = 1; χ2 test). f, Venn diagrams summarizing data from c–e, showing the fraction of schema neurons recorded in control and GtACR2 sessions that were affected by shuffling of information related to epoch (light gray), reward (light green) and position (dark gray) as in Fig. 5d. Sizes of circles are normalized to the total number of neurons recorded in each group, averaged across days (see Extended Data Fig. 6 for the same illustration by day). The proportions in each category and overlap between categories were affected by inactivation, with an increase in those affected by epoch and reward (χ2 = 9.8; P = 0.0018; d.f. = 1; χ2 test) and a corresponding decrease in those affected by epoch only (χ2 = 18.4; P = 1.8 × 10−5; d.f. = 1; χ2 test). The error bars are the s.e.m. (***P < 0.001, **P < 0.01; Extended Data Figs. 6 and 7 and Supplementary Figs. 10–12).
Reconstruction illustrating the recording sites within the lateral orbitofrontal cortex (OFC) containing axons from transfected VSub neurons, expressing GtACR2. The approximate extent of recording sites in each rat is depicted by red squares. This experiment was independently repeated with all six animals, producing consistent results. OFC axons from VSub neurons expressing GtACR2 are showing a higher magnification within the dashed white box and indicated by white arrows. Scale bars are provided at 100 µm and 10 µm.
A pair of heatmaps are shown for each neuron, plotting average activity in each epoch at each position in each maze. Arrows represent sequence directions. Red * marks the reward epoch on rewarded trial types (P3 and P4), while black * marks the reward epoch for non-rewarded trial types (P1 and P2). The correlation coefficients before and after shuffling of epoch, reward, and position are presented in the lower panel of each figure; numbers shown in red were significantly affected by shuffling.
A pair of heatmaps are shown for each neuron, plotting average activity in each epoch at each position in each maze. Arrows represent sequence directions. Red * marks the reward epoch on rewarded trial types (P3 and P4), while black * marks the reward epoch for non-rewarded trial types (P1 and P2). The correlation coefficients before and after shuffling of epoch, reward, and position are presented in the lower panel of each figure; numbers shown in red were significantly affected by shuffling.
A pair of heatmaps are shown for each neuron, plotting average activity in each epoch at each position in each maze. Arrows represent sequence directions. Red * marks the reward epoch on rewarded trial types (P3 and P4), while black * marks the reward epoch for non-rewarded trial types (P1 and P2). The correlation coefficients before and after shuffling of epoch, reward, and position are presented in the lower panel of each figure; the numbers shown in red were significantly affected by shuffling.
a-b, Percentage of schema neurons on each day of retraining after surgery from the rats comprising the control and GtACR2 groups in the learning experiment (a) and from control sessions on the well-learned task (b). There were no differences between the rats in the two groups on any prior day of training (retraining: χ2 < 6.0; P > 0.10; d.f. = 1; χ2 test; well-learned problem: χ2 < 3.94; P > 0.38; d.f. = 1; χ2 test). False discovery rate (FDR) and Benjamini-Hochberg (BH) corrections were applied to correct for multiple comparisons. c, Effect of inactivation of ventral subiculum on prevalence of schema cells during learning, excluding data from the 2 replacement rats added during this phase; without data from these rats, the prevalence of schema cells in the two groups was similar initially and then diverged thereafter (Overall: χ2 = 84.4, P = 4.0 ×10−20; days 1-2: χ2 = 4.0; P = 0.05; d.f. = 1; χ2 test; days 3-10: χ2 = 83.5, P = 6.4 ×10−20; d.f. = 1; χ2 test). (***P < 0.0001; NS, not significant).
Venn diagrams show the percentage of schema neurons that were recorded during control and GtACR2 sessions and how they were influenced by information related to epoch (light gray), reward (light green), and position (dark gray) across days. Size of figures reflect overall proportion of neurons that categorized as schema cells, which increased more rapidly in the inactivated group, with an expansion of the reward group and its overlap with epoch and position.
a-c, Percentage of non-schema neurons on each day of learning, whose correlated activity across mazes was affected by shuffling to disrupt information related to epoch (a), reward (b), or position (c). d-f, Percentage of non-coding neurons on each day of learning, whose correlated activity across mazes was affected by shuffling to disrupt information related to epoch (d), reward (e), or position (f). No significant differences between the two groups were observed (χ2< 4.8; P > 0.028; d.f. = 1; χ2 test). (NS, not significant).
The positions in principal component space for each maze are shown for both the Control (a) and GtACR2 (b) groups during each epoch of the well-learned task. For each epoch, gray lines represent Maze 1, while green lines correspond to Maze 2. The four dots on each line, which signify points P1, P2, P3, and P4, transition in color from light to dark in this order. Additionally, the marker size increases progressively from P1 to P4. P1 and P2 represent the common arms that share the same odor, while P3 and P4 correspond to the unique arms with distinct odors. c, Quantification of Procrustes analysis for the Control and GtACR2 groups in (a) and (b) was used to measure the geometric similarity between Maze 1 and Maze 2. No significant difference in geometric similarity between Maze 1 and Maze 2 was observed between the two groups across epochs (t35 = 1.1, P = 0.28; two-tailed Student’s t-test; n = 36 for Control and GtACR2). Error bars are SEM.
Geometric similarity of task representations across mazes during new task learning. The positions in principal component space for each maze are shown for the Control (a) and GtACR2 groups (b) at each learning epoch. Gray lines represent Maze 1, and green lines represent Maze 2. In each epoch, four dots on each line, corresponding to positions P1, P2, P3, and P4, transition from light to dark, with marker sizes increasing progressively from P1 to P4. P1 and P2 represent the common arms, both sharing the same odor, while P3 and P4 correspond to the unique arms, each with distinct odors. c, Procrustes analysis was performed for the Control and GtACR2 groups in (a) and (b) to evaluate the geometric similarity between Maze 1 and Maze 2. Across epochs, no significant differences in geometric similarity between the two mazes were observed between the groups (t35 = 4.4, P = 1.1 × 10−4; two-tailed Student’s t-test; n = 36 for Control and GtACR2). Error bars are SEM.
Scatter plots of the first two averaged LDAs are shown for both the well-learned task (a-b) and the learning task (c-d), combining all sessions. The task includes 16 trial types, each represented by a colored dot. Ten odors were organized into two mazes (Maze 1 and Maze 2), each consisting of two subsequences (a and b). Each subsequence contains four positions (P1–P4). For example, M1a1 refers to Maze 1, sequence a, position 1. Trials from each position clustered together. The positions are classified as P1, P2, P3, and P4, as indicated in the figure legend. b, Quantification of the mean silhouette value for each trial type was performed for both the Control and GtACR2 groups during the well-learned task. No significant difference between the two groups was observed (t30 = 0.24, P = 0.81; two-tailed Student’s t-test). d, Quantification of the mean silhouette value for each trial type in the learning task showed that the GtACR2 group had significantly higher silhouette values compared to the Control group (t30 = 3.5, P = 1.6 × 10−3; two-tailed Student’s t-test). Error bars are SEM. (NS, not significant; **P < 0.01).
References
MeSH terms
Substances
Grants and funding
LinkOut - more resources
Full Text Sources
