Overview of the experimental tasks.

(a) Virtual environment (VE). (Left) An aerial view of the park with four rooms, R1-R4, arranged orthogonally. (Middle) Subject’s perspective at the entrance and center of the VE. (Right) Subject’s perspective from outside and inside the rooms. The directions N, E, S, and W represent north, east, south, and west, respectively. (b) Experimental timeline. The experiment was conducted over 4 non-consecutive days. The intervals between each behavioral training and fMRI sessions was ≤ 2 days, while the interval between the 1st fMRI scan session and the 2nd behavioral training session was ≤ 8 days. RA = response accuracy. (c) Object familiarization task. Subjects learned about the objects used in the VE through an object learning phase and an object memory test. (d) Exploration task. Subjects freely explored the VE to learn its layout and object locations. (e) Object finding task. Subjects were instructed to locate specific objects within the VE. (f) Judgment of relative direction (JRD) task. Subjects judged the direction of one object relative to another. The blue icons indicate the tasks being conducted outside the scanner, and the green icon indicates a task being conducted inside the scanner.

Illustration of the relationships among the three hierarchical levels, reference frames (RFs), and the learning process in the experiment.

The local level of the environment captures spatial relationships between the objects within each room, based on the local RF. After learning, the subjects may strengthen and generalize these relationships to similar spatial structures. The categorical level reflects the semantic distinction between rooms (R1-R4). The global level represents the spatial layout of the entire park using the global RF, with spatial relationships across rooms becoming clearer after learning.

Behavioral performance and brain activation in the judgement of relative direction (JRD) task.

(a) Interaction between spatial information and learning stage on response time (RT). EDref-tar represents the Euclidean distance between the reference and target objects, which was divided into four equal bins based on the quartiles of the distance distribution for ease of visualization. Bin1, Bin2, Bin3, and Bin4 represent increasing distances from closest to furthest. represent the local directions of a reference object. Because only two orthogonally-aligned rooms were included in each fMRI session, we set the Room variable as a binary factor, indicating whether the room was aligned along the north-south or east-west axis. represent the global direction of a reference object. The boxplots indicate the interquartile range (IQR), with horizontal lines within the boxes representing the median, whiskers extending to 1.5 times the IQR, and dots showing outliers beyond 1.5 times the IQR. Asterisks above each learning session denote the main effect of spatial information, and asterisks across learning stages indicate interaction effects between learning stage and spatial information. *, p < .05, **, p < .01, ***, p < .001. The results of post-hoc analysis are provided in Table S1 (Supplementary Data). (b) Brain regions with significant activations in the JRD task. (c) Significant activation in the left retrosplenial cortex (RSC) for the contrast of post-learning > pre-learning stage in the JRD trials. No regions showed significant activation for the contrast of post-learning < pre-learning stages in the JRD task. (d) Brain regions showing the activation associated with EDref-tar in the JRD task. The warm/cold colors indicate positive/negative correlation. L/R indicates the left/right hemisphere. (e) Significant brain-behavior correlation. The x-axis represents the random effect coefficients (η) of the learning stage from the linear mixed-effects model (LMM), and the y-axis represents the β-values from the general linear model (GLM) analysis for the JRD trials and EDref-tar. The coordinates indicate the centers of ROIs. The shaded area indicates the 95% confidence interval (CI). The detailed brain-behavior correlation for all brain regions involved in the JRD task are listed in Table S2 (Supplementary Data).

Representational similarity analysis (RSA).

(a) Schematic of the RSA. Two theoretical representational dissimilarity matrices (RDMs), one for global and one for local levels, were built separately for both the learning stages (pre-learning and post-learning) to capture the spatial relationships of the reference objects. Considering that each room contained 8 objects and each learning stage used two rooms (16 objects total), we built the RDMs as 16 × 16 matrices. The left panel is an example of a pair of reference objects (highlighted in yellow in the matrices) in a single learning stage. Black (gray) rectangles indicate rooms that were (not) used in the learning stage. The global RDM and local RDM were built using the Euclidean distance between the reference objects within the global reference frame and the local reference frame , respectively, with larger distances indicating greater dissimilarity. For both the pre-learning and post-learning stages, the neural RDM was built by (1 - rij), where rij represents Pearson’s correlation coefficient between the β-values for the ith and jth reference objects within a given brain region. Light (dark) colors in the matrices indicate high (low) similarity. Corresponding to the two hierarchical levels and two learning stages, four Spearman’s rank correlations were computed by estimating the correlation between the neural RDMs and each theoretical RDM. This yielded four correlation coefficients: , and (b) Significant results of the ROI-based RSA. The coordinates represent the gravity centers of ROIs. Pre = pre-learning stage, Post = post-learning stage. (c) Brain regions showing significant differences in the correlation coefficients. Brain regions outlined by yellow and green dashed lines showed significant and respectively. Abbreviations: HIP = hippocampus, RSC = retrosplenial cortex, l/mOFC = lateral/medial orbitofrontal cortex, d/vACC = dorsal/ventral anterior cingulate cortex, THA = thalamus, MiFG = middle frontal gyrus, INS = insula, MTG = middle temporal gyrus. L/R represents left/right hemisphere.

Fixed-effects ANOVA results obtained from the linear mixed-effects model (LMM).