Dissociable retrosplenial and hippocampal contributions to successful formation of survey representations

Abstract

During everyday navigation, humans encounter complex environments predominantly from a first-person perspective. Behavioral evidence suggests that these perceptual experiences can be used not only to acquire route knowledge but also to directly assemble map-like survey representations. Most studies of human navigation focus on the retrieval of previously learned environments, and the neural foundations of integrating sequential views into a coherent representation are not yet fully understood. We therefore used our recently introduced virtual-reality paradigm, which provides accuracy and reaction-time measurements precisely indicating the emergence of survey knowledge, and functional magnetic resonance imaging while participants repeatedly encoded a complex environment from a first-person ground-level perspective. Before the experiment, we gave specific instructions to induce survey learning, which, based on the clear evidence for emerging survey knowledge in the behavioral data from 11 participants, proved successful. Neuroimaging data revealed increasing activation across sessions only in bilateral retrosplenial cortices, thus paralleling behavioral measures of map expertise. In contrast, hippocampal activation did not follow absolute performance but rather reflected the amount of knowledge acquired in a given session. In other words, hippocampal activation was most prominent during the initial learning phase and decayed after performance had approached ceiling level. We therefore conclude that, during navigational learning, retrosplenial areas mainly serve to integrate egocentric spatial information with cues about self-motion, whereas the hippocampus is needed to incorporate new information into an emerging memory representation.

Figure 1.

Example views from the virtual environment. Left, Aerial view of the environment (not shown to participants). Arrows and numbers indicate the traveled route; road sections were visited in ascending order. Letters serve to illustrate the difference between direct, close, and remote pairs. Note that, during encoding, participants were moved throughout the entire environment, thereby encountering all 12 landmarks. Top right, Ground-level view of one of the 12 landmarks. Bottom right, Example of an image used for retrieval. Subjects were to indicate by button press the relative position of the small building, imagining that they were standing in front of the large building. Given the route depicted to the left, six combinations with landmark “A” as the large building were possible. Direct pair, A-B (target building visited in immediate temporal sequence). Close pairs, A-C/A-D (target building on an adjacent intersection visited immediately after the large building). Remote pairs, A-E/A-F/A-G (target building on an adjacent intersection not visited immediately after the large building).

Figure 2.

Behavioral results during retrieval. Top, Performance and reaction time data for the 11 participants who met our criteria for survey learning. Significant changes across sessions were observed for direct, close, and remote pairs, indicating the emergence of survey knowledge. Bottom, Performance and reaction time data for the six participants who did not meet our criteria for survey learning. Because of the absence of learning for all three of the categories, these subjects were excluded from additional analyses.

Figure 3.

Performance-related increase (n = 11). Areas showing a significant activation increase across encoding sessions that paralleled behavioral performance are indicated. Left, Mean ± SEM regression coefficients of the peak voxel in the retrosplenial cortex, along with the fitted learning curve for remote pairs. Results of the random-effects analysis are displayed with a threshold of p < 0.05 (corrected) on the averaged MNI template brain. Regions with increasing activation across encoding control sessions were excluded (for an uncorrected activation map omitting this masking procedure, see supplemental material, available at www.jneurosci.org). Note that, because of previous anatomical hypotheses, correction for multiple comparisons in the retrosplenial cortex was based on a reduced search volume.

Figure 4.

Learning-related change (n = 11). Top, Left hippocampal region with activation reflecting the amount of knowledge acquired in any given encoding session. Results of the random-effects analysis are displayed with a threshold of p < 0.05 (corrected) on the averaged MNI template brain (for an uncorrected activation map, see supplemental material, available at www.jneurosci.org). Note that, because of previous anatomical hypotheses, correction for multiple comparisons in the hippocampus was based on a reduced search volume. Bottom, Fitted learning curves for remote pairs in two representative subjects (left). Right, The corresponding partial derivatives, indicating the amount of learning in a given session, and the regression coefficients of left hippocampal peak voxels. Whereas hippocampal activation in subject 11 is strongest in the initial learning stage and decays rapidly after performance has approached ceiling level in session 3, the slower learning process in subject 03 is paralleled by stronger hippocampal activation in the second half of the experiment. As a consequence, hippocampal activation seems to be the most prominent whenever substantial performance improvements are observable.