Environmental complexity modulates information processing and the balance between decision-making systems

Abstract

Behavior in naturalistic scenarios occurs in diverse environments. Adaptive strategies rely on multiple neural circuits and competing decision systems. However, past studies of rodent decision making have largely measured behavior in simple environments. To fill this gap, we recorded neural ensembles from hippocampus (HC), dorsolateral striatum (DLS), and dorsomedial prefrontal cortex (dmPFC) while rats foraged for food under changing rules in environments with varying topological complexity. Environmental complexity increased behavioral variability, lengthened HC nonlocal sequences, and modulated action caching. We found contrasting representations between DLS and HC, supporting a competition between decision systems. dmPFC activity was indicative of setting this balance, in particular predicting the extent of HC non-local coding. Inactivating mPFC impaired short-term behavioral adaptation and produced long-term deficits in balancing decision systems. Our findings reveal the dynamic nature of decision-making systems and how environmental complexity modulates their engagement with implications for behavior in naturalistic environments.

Keywords: behavior; decision making; dorsolateral striatum; environmental complexity; habit; hippocampus; medial prefrontal cortex; naturalistic environments; place cell; planning; task bracketing.

Figure 1:. Behavior in left/right/alternation task

(A) The left/right/alternation (LRA) foraging task. Three different wall types: open middle (Wall A), open left side (Wall B), and open right side (Wall C) were used to create intermediary decision-points. (B) Recording sites from dorsal mPFC, DLS, and dorsal HC. (C) Example histology and illustration of viral spread for DREADDs (red) and control virus animals (blue). (D) Example behavior from a single session and the hypothesized arbitration between deliberative and procedural decision-making systems. (E) Rat performance aligned to rule switches. (F) Example vicarious trial and error (VTE) and non-VTE paths. (G) z-Scored IdPhi aligned to rule switches. (H) Two example paths (orange and green) through the central maze segment showing low (top) and high (bottom) path stereotypy. (I) Trajectory stereotypy between pairs of paths through the central maze segment around rule switches. (E, G) black line and shaded error bars show mean±SEM.

Figure 2:. Quantification of environmental complexity

(A) Environmental complexity of different maze. The colored circles indicate the low, mid, and high complexity designations. (B) Illustration of hypothesized behavior and neurophysiology.

Figure 3:. Behavioral effects of environmental complexity

(A) Proportion of VTE laps (zIdPhi > 0.25) in low, mid, and high complexity environments. (B) Top: example of a low exploration session, and Bottom: example of a high exploration session. (C) Percentage of exploratory laps aligned to rule switches. Black line indicates mean, and the shaded gray region±SEM. (D) Distribution zIdPhi values at the choice point for non-exploratory and exploratory laps. Horizontal line denotes the boundary for identifying VTE events (Kruskal-Wallis test H(1)=46.41, p<0.001). (E) Percentage of exploratory laps per session broken down by maze complexity. (F) Examples of trajectory stereotypy in different environments. (G) Development of stereotypy around rule switches.

Figure 4:. Theta sequences in differently-complex environments

(A) Two example HC tuning curves. (B) Spatially binned activity of putative pyramidal cells on the linearized maze. (C) An example of HC ensemble representation of space. First row: The maze and an example trajectory (dark blue) along with raw (gray) and theta frequency filtered LFP (green). Second row: spike raster of simultaneously recorded HC cells on an example lap. Cells are ordered and colored by their respective centers. Gray line shows the actual trajectory of the rat. Third row: Decoded activity from the HC ensemble. Black line shows the actual trajectory of the rat in linearized space. (D, E) Examples of (D) high and (E) low score theta sequences. Left: HC cell spiking during a theta cycle ordered and colored by field location. Raw LFP (gray) and theta filtered LFP (green) are shown below with vertical lines indicating the peaks and trough of the theta cycle. Right: Bayesian decoding during the theta cycle. The arrow and cyan line show the actual position of the rat. (F) Average percent of significant theta sequences per session. (G) Distribution of theta sequence scores for the different maze complexity groups. Dashed lines indicate medians. (H) Session average theta sequence scores along the central path for the different maze complexity groups. (I) Differences in posterior probability of the decoding in the 1^st (peak to trough) and 2^nd half (trough to peak) of each identified theta cycle aligned to rat’s position. Positive values indicate more local and negative values indicate more non-local decoding. The red line indicates the shuffle distribution. Black lines indicate significant differences between pairs of environment complexity (top line low vs mid, middle line mid vs high; bottom line low vs high). (Repeated measures ANOVA for complexities across positions; Interaction: F(34)=51, p<0.001. Post-hoc on each position bin via two-sided Wilcoxon rank sum corrected for multiple comparisons).

Figure 5:. Relationship between theta sequence score and deliberative behaviors

(A) Distribution of correlations between zIdPhi and mean theta sequence score at the choice point. (B) Average theta sequence score at the choice point for laps identified as VTE and non-VTE. (C) Distribution of correlations between exploration amount and mean theta sequence score in the central path. (D) Average theta sequence score through the central path for laps that were identified as exploratory and non-exploratory. (E) Distribution of correlations between central path stereotypy and average theta sequence score in the central path. (A, C, E) Dashed lines indicate means.

Figure 6:. DLS ensemble activity across the linearized maze and the arbitration between deliberative and procedural decision-making

(A) Top row: Average firing rate of all putative dorsolateral striatum medium spiny neurons over the linearized track averaged across the different complexity groups. Firing rates are plotted as a function of laps aligned to the start of rule blocks. Bottom row: Average firing rates averaged across laps and maze group. Solid lines indicate mean, and the shading±SEM. Dashed lines indicate the internal wall locations projected onto the linearized maze. (B) Quantification of task-bracketing. (C) Average task-bracketing score for each session broken down by maze complexity group. (D) Spatial cross correlation of DLS ensemble activity broken down by maze complexity group. (E) Blue: Mean HC theta sequence score across the central path aligned to rule switches. Orange: DLS Task-bracketing aligned to rule switches. For both plots, solid lines indicate mean, and the shading±SEM.

Figure 7:. Representational transitions after rule switches and predictions of prospective representations

(A) Lap-by-lap correlation of mPFC, DLS, and HC population activity aligned to rule switches. Correlation matrices are averaged across the rule changes. The vertical and horizontal dashed lines correspond to the last lap of the previous rule block. (B) Distribution of physiology-based change points aligned to the task rule switch laps. (C) Distribution of the difference in physiology-based change points for simultaneously recorded rats (n_{Rats HC-mPFC}=5 and _{Rats DLS-mPFC}=2). Black dots indicate mean physiology-based change point lap difference between region in non-simultaneously recorded rats . (D) Distribution of the posterior probabilities of mPFC activity predicting binned theta sequence score. The red line indicates actual data. The blue lines indicate the different shuffles. The dashed line indicates uniform probability. (E) Posterior probability of mPFC activity predicting binned theta sequence score aligned to the first error lap at or closest to rule switches. Solid lines indicate mean (red: actual, blue: mPFC spike time shuffle), and shading±SEM. (F) Distribution of correlations between zIdPhi and mean posterior probability of mPFC prediction of HC theta sequence score at the choice point. Red dashed line indicates the median of the actual data, the blue lines indicate different shuffle medians.. (G) Distribution of correlations between exploration amount and mean posterior probability of mPFC prediction of HC theta sequence score in the central path. Representation as in (F). (H) z-Scored mPFC lap firing rate aligned to the first error lap at or closest to rule switches. Solid line indicates mean, and the shading±SEM. (I) Distribution of z-scored mPFC lap firing rate for correct (green) and error (red) trials. Dashed lines indicate the median. (J) z-Scored mPFC firing rate restricted to the side feeder zones aligned to the first error lap at or closest to rule switches. Representation as in (H). (K) z-Scored mPFC firing rate restricted the central path (start of maze exit to choice point entry) aligned to the first error lap at or closest to rule switches. Representation as in (H).

Figure 8:. Behavioral effects of mPFC inactivation with DREADDs

(A) Example histology (right) and illustration of viral spread (left) for DREADDs virus animals (red) and control virus animals (blue). (B) Distribution of rat performance for mPFC inactivated rats (DREADDs: DCZ dark pink) and control virus animals (light pink). (C) Distribution of minimum distance between behavioral change points and rule switch laps for DREADDs virus rats under DCZ, post DCZ (2^nd day saline repeat), and VEH, as well as all control virus animals. (D) Distribution of proportion of VTE events per session for inactivated animals (DREADDs: DCZ, red), for 2^nd day after inactivation (DREADDs: post DCZ, blue), and controls (Control, light pink). (E) Difference in the proportion of VTE events between 1^st day manipulation and 2^nd day rebound (Day 2 – Day 1) for DREADDs (dark pink) and control (light pink) animals with respect to drug condition (ANOVA virus*DCZ F(1)=3.86, p=0.05). (F) Per session distribution of the percentage occupied space in the central segment of the maze for inactivated animals, for 2^nd day after prefrontal inactivation, and controls. (G) Difference in the percentage of the central segment of the maze occupied between 1^st day manipulation and 2^nd day rebound (Day 2 – Day 1) for DREADDs and control animals with respect to drug condition (ANOVA virus*DCZ F(1)=25.41, p<0.001). (H) zIdPhi aligned to rule switches for DREADDs and control virus animals. The horizontal dashed line indicates the threshold for which events were identified as binary VTE and non-VTE events. The vertical dotted line corresponds to the last lap of the previous rule block. The black line above indicates laps in which there was a significant difference between Active and Control groups. Data are shown as mean±SEM. (I) Proportion of exploratory laps aligned to rule switches for DREADDs and control animals. Representation as in (H). (J) Trajectory stereotypy between pairs of paths through the central maze segment around rule switches for DREADDs and control virus animals. (K) Rat performance across the experimental weeks for DREADDs and control animals. Each ‘week’ is an 8-day sequence (Supplementary Fig. 8C).