The Neural Representation of Prospective Choice during Spatial Planning and Decisions

Abstract

We are remarkably adept at inferring the consequences of our actions, yet the neuronal mechanisms that allow us to plan a sequence of novel choices remain unclear. We used functional magnetic resonance imaging (fMRI) to investigate how the human brain plans the shortest path to a goal in novel mazes with one (shallow maze) or two (deep maze) choice points. We observed two distinct anterior prefrontal responses to demanding choices at the second choice point: one in rostrodorsal medial prefrontal cortex (rd-mPFC)/superior frontal gyrus (SFG) that was also sensitive to (deactivated by) demanding initial choices and another in lateral frontopolar cortex (lFPC), which was only engaged by demanding choices at the second choice point. Furthermore, we identified hippocampal responses during planning that correlated with subsequent choice accuracy and response time, particularly in mazes affording sequential choices. Psychophysiological interaction (PPI) analyses showed that coupling between the hippocampus and rd-mPFC increases during sequential (deep versus shallow) planning and is higher before correct versus incorrect choices. In short, using a naturalistic spatial planning paradigm, we reveal how the human brain represents sequential choices during planning without extensive training. Our data highlight a network centred on the cortical midline and hippocampus that allows us to make prospective choices while maintaining initial choices during planning in novel environments.

Fig 1. Task.

(A) During a 3.25-s planning phase, participants had to infer the shortest path from the starting point in maze (a red square) to the goal location (green square) and remember the chosen direction for each choice point along the shortest path. Half of the mazes (shallow mazes) had two paths and only one choice point (red square), whereas the other half (deep mazes) had four paths and two choice points (red square and another point further in the maze). After 3.25 s, a choice point was highlighted (choice highlight) for 250 ms. The highlighted location could either be the red square or the second choice point along the shortest path for deep mazes. In shallow mazes, only the red starting location was highlighted. Crucially, for deep mazes, participants were tested on one choice point before starting the next trial. Subsequently, the choice period featured a first-person viewpoint of the highlighted location, where participants had a maximum of 1.5 s (Deep Maze mean: ~545 ms; Shallow Maze mean: ~440 ms) to choose the correct direction on the shortest path (left, forward, right, or equal) with a button box. Immediately following the button press, an intertrial interval (ITI) screen appeared for 1.5 s before a new trial began. (B) Left: Example shallow maze trial with a large path length difference (less demanding choice). Right: Example shallow maze trial with a small path length difference (demanding choice). (C) Left: Example deep maze trial with a small path length difference (demanding) initial choice at the red square and large (easy) path length difference at the second (prospective) choice point. Right: Example deep maze trial with a large (easy) path length difference initial choice at the red square and small (demanding) path length difference at the second (prospective) choice point. Deep mazes contained a combination of small, medium, or large path length differences at first (initial) and second (prospective) choice points. (D) Overhead view (not shown during the experiment) of three example mazes showing which path lengths contribute to each parametric regressor in our fMRI analyses: initial (left), prospective (centre), and unchosen path length differences in deep mazes. Initial path length differences in shallow mazes represent the difference between the only two available paths. The black line highlights shortest path.

Fig 2. Behavioural Results.

(A) Accuracy during choice phase. Left: Significant interaction (p < 0.001) for initial versus prospective path length differences in deep mazes. Deep mazes are split by small–small, small–medium, small–large, medium–small, medium–medium, medium–large, large–small, large–medium, and large–large path length differences at the initial choice point (i.e., the shortest options for either choice at the starting location) and the two available paths at the prospective choice point, respectively. Right: Significant positive linear trend in accuracy (p < 0.001) with increasing path length differences for shallow mazes. Shallow mazes are split by small, medium, and large path length differences. (B) Log RT during the choice phase. Left: Significant interaction (p < 0.001) for initial versus prospective path length differences in deep mazes. Deep mazes are split by small–small, small–medium, small–large, medium–small, medium–medium, medium–large, large–small, large–medium, and large–large path length differences at the initial choice point (i.e., the shortest options for either choice at the starting location) and the two available paths at the prospective choice point, respectively. Right: Significant negative trend (p < 0.001) in log RT with increasing path length differences in shallow mazes. Shallow mazes are split by small, medium, and large path length differences. See S1 Data for participant data.

Fig 3. Prefrontal responses to prospective path length differences.

(A) Regions significantly responding to smaller prospective path length differences. Left: Coronal image showing rd-mPFC/superior frontal gyrus (SFG). Centre: Sagittal image showing dACC/pSMA. Right: Coronal image showing left lFPC. (B) Pregenual anterior cingulate cortex/ventromedial PFC (pgACC/vmPFC) region significantly engaged by larger prospective path length differences. (C) Effect size for an 8-mm sphere around the rd-mPFC/SFG (left), dACC/pSMA (centre), and left lFPC (right) peak voxels that responded to smaller prospective path length differences displayed in A for three parametric modulators: initial (including both deep and shallow mazes), prospective, and unchosen path length differences (mean ± standard error of the mean [SEM]). (D) Effect size for an 8-mm sphere around the pgACC/vmPFC peak voxel that responded to larger prospective path length differences. For both C and D, asterisks indicate a significant correlation (p < 0.05) with path length differences. A positive effect size represents a BOLD correlation with larger path length differences, whereas a negative effect size represents a correlation with smaller path length differences. (E) Images centred on rd-mFPC peak in A, which was the only region featured in a that significantly responded to decreasing prospective versus initial choice path length differences in shallow mazes. All highlighted regions survived cluster-level FWE correction at p < 0.05 and are displayed at an uncorrected statistical threshold of p < 0.005. Corresponding coordinate from the Montreal Neurological Institute (MNI) template image listed below each brain image. See S2 Data for individual effect size data.

Fig 4. fMRI activations related to unchosen path length difference.

(A) Regions that significantly responded to larger unchosen path length differences between the shortest and unchosen path. Top left: Sagittal image showing posterior PCC. Top centre: Coronal image showing right striatum. Top right: Sagittal image showing right angular gyrus. Bottom: Effect size for an 8-mm sphere around the PCC (left), right striatum (centre), and right angular gyrus (right) regions that responded to larger unchosen length differences displayed in A, for three parametric modulators: initial (deep and shallow mazes), prospective, and unchosen path length differences (mean ± SEM). Asterisks indicate a significant correlation (p < 0.05) with path length differences. A positive effect size represents a positive BOLD correlation with larger path length differences, whereas a negative effect size represents a correlation with smaller path length differences. (B) dACC region significantly responding to smaller unchosen path length differences. All highlighted regions survived cluster-level FWE correction at p < 0.05 and are displayed at an uncorrected statistical threshold of p < 0.005. See S3 Data for individual effect sizes.

Fig 5. Hippocampal contributions to planning.

(A) Coronal image showing higher left hippocampal activity (circled in blue) during planning prior to correct versus incorrect choices. Subthreshold right hippocampal activity that was higher for correct choices is also visible. (B) Sagittal image showing ventral temporal activity extending into right posterior hippocampus (circled in blue) that positively correlated with the distance of the shortest route between the starting and goal location. (C) Top: Sagittal image showing posterior right hippocampal activity during planning that positively correlated with subsequent log RT. Bottom: Effect size for an 8-mm sphere around right posterior hippocampus peak voxel showing that the correlation with log RT is significantly higher (p < 0.05) for deep versus shallow planning trials. All hippocampal peak voxels presented survive correction for multiple comparisons (p < 0.05) across the whole hippocampal volume, but clusters are shown at p < 0.005 uncorrected for visualization purposes. (D) Top: Sagittal image showing medial extent of rd-mPFC (peak voxel same as 3A) that exhibited increased functional connectivity with the right posterior hippocampus in deep versus shallow maze planning trials. Bottom left: Effect size for an 8-mm sphere around rd-mPFC peak voxel (mean ± SEM) showing significantly increased functional connectivity with hippocampus for deep versus shallow planning trials. Bottom right: Hippocampal rd-mPFC functional connectivity (mean ± SEM) was significantly higher during deep planning trials prior to correct versus incorrect choices. See S4 Data for individual effect sizes presented in 5C and 5D.