Neural correlates of real-world route learning

Abstract

Classical theories of spatial microgenesis (Siegel and White, 1975) posit that information about landmarks and the paths between them is acquired prior to the establishment of more holistic survey-level representations. To test this idea, we examined the neural and behavioral correlates of landmark and path encoding during a real-world route learning episode. Subjects were taught a novel 3 km route around the University of Pennsylvania campus and then brought to the laboratory where they performed a recognition task that required them to discriminate between on-route and off-route buildings. Each building was preceded by a masked prime, which could either be the building that immediately preceded the target building along the route or immediately succeeded it. Consistent with previous reports using a similar paradigm in a virtual environment (Janzen and Weststeijn, 2007), buildings at navigational decision points (DPs) were more easily recognized than non-DP buildings and recognition was facilitated by in-route vs. against-route primes. Functional magnetic resonance imaging (fMRI) data collected during the recognition task revealed two effects of interest: first, greater response to DP vs. non-DP buildings in a wide network of brain regions previously implicated in spatial processing; second, a significant interaction between building location (DP vs. non-DP) and route direction (in-route vs. against-route) in a retrosplenial/parietal-occipital sulcus region previously labeled the retrosplenial complex (RSC). These results indicate that newly learned real-world routes are coded in terms of paths between decision points and suggest that the RSC may be a critical locus for integrating landmark and path information.

Introduction

A strategy for successful navigation in large-scale environments is to follow routes defined by landmarks and the spatial relationships between them. Such a strategy is likely to be especially useful for urban navigation where landmarks are plentiful and the space is generally too large to be perceived in its entirety from any one vantage point. Siegel and White (1975) proposed that spatial knowledge develops in a series of stages where landmark and route knowledge are the precursors of survey knowledge; a scheme that has been labeled the dominant framework (Montello, 1998). Neuroscientific work on spatial navigation, on the other hand, has often focused on the systems that support flexible route generation using cognitive maps, or the following of “well-worn” routes that are coded in terms of stimulus-response contingencies, leaving the neural systems that encode newly learned routes, which are not yet well-worn or incorporated into a cognitive map, relatively unexplored. Here we address this lacuna by examining fMRI activity related to the coding of landmarks and spatial relationships between landmarks along a newly learned route.

Subjects were trained on a 3.8 km route through the University of Pennsylvania campus and the surrounding territory and later scanned while viewing photographs of buildings from the route. Our design is a real-world adaptation of a paradigm previously employed by Janzen and Weststeijn (2007) to study route learning in a virtual (i.e. videogame) environment. Using a masked prime recognition task, these authors found behavioral and neural evidence that individuals distinguish between objects at navigational decision points (i.e. locations such as intersections where there is possibility for a change in direction) and other objects; furthermore, the direction of travel is specifically encoded at decision points. fMRI results revealed that the posterior parahippocampal gyrus responded preferentially to objects at decision points and that the superior parietal lobe, anterior cingulate and right caudate nucleus were sensitive to the direction of travel at these locations (Janzen & van Turennout, 2004, Janzen & Weststeijn, 2007). Although we expected to replicate the behavioral results from this earlier study, we anticipated the possibility that different neural systems would be involved in route learning under natural learning conditions, where real-world routes are populated with ecologically valid landmarks (i.e. buildings rather than small objects on tabletops) and are too complex to be learned through stimulus-response contingencies.

We were especially interested in determining the involvement of the parahippocampal place area (PPA) (Epstein and Kanwisher, 1998) and retrosplenial complex (RSC) in landmark and route learning. The PPA and RSC respond preferentially to stimuli of potential navigational relevance, such landscapes, cityscapes and rooms, and are critical nodes of the larger network of regions that are typically activated during navigation tasks (Aguirre et al., 1996, Ghaem et al., 1997, Maguire et al., 1998, Hartley et al., 2003, Rosenbaum et al., 2004, Spiers & Maguire, 2007). Previous work suggests that the PPA plays a particularly important role in the coding of scenes and buildings (Epstein et al., 1999); furthermore, insofar as the PPA is coterminous with the posterior parahippocampal gyrus, the previous work on decision points suggests that its response to even nonscene/non-building objects can be modulated by the navigational relevance of these objects (Janzen & van Turennout, 2004, Janzen & Weststeijn, 2007). The RSC, on the other hand, may play a key role in the coding of directional information that allows us to understand the spatial relationship between different locations (Sato et al., 2006, Byrne et al., 2007, Epstein, 2008). Thus, we predicted that PPA should respond strongly to buildings from the path, and we set out to determine if this response would be modulated by the position of buildings at decision points. We also set out to substantiate a possible role for RSC in the coding of path direction. To anticipate, both these predicted functions were supported and elaborated by the data.