A Scale-Dependent Neural System for Human Spatial Navigation

It is currently unclear whether the human brain processes navigation tasks at different scales in the same way. According to the classical view, humans process navigation information using a hierarchical representation system in a unied way. Other theories, such as the dual and multiple systems theories, suggest that the processing of navigation tasks differs between spatial scales. In addition, previous human navigation studies have mainly focused on scales ranging from rooms to small neighborhoods. However, the neural mechanisms underlying the processing of large-scale spatial navigation tasks in the human brain, and the ways in which neural activation changes with scale, have not been examined in detail. In this study, we conducted a functional magnetic resonance imaging (fMRI) based multi-scale mental navigation experiment across six spatial scales. On the basis of cortical activity patterns, we discovered a scale-dependent neural system that included the parahippocampal gyrus, cingulate gyrus, parietal gyrus and temporal gyrus, revealing neural-based divisions based on spatial scale: small scale (room), medium scale (building), large scale (block), and huge scale (city, country and continent). In contrast, scale-free characteristics were observed in middle occipital gyrus. The current ndings provide new insight into the neural mechanisms of scale-driven spatial navigation in humans. , the room names as the location pairs. Similarly, for the building scale and the block scale, we used the building names and the block names, respectively. For scales larger than block, including the city, country and continent scales, city names were used for the location pairs, such as “Beijing, Shanghai” at the city scale, “Beijing, Tokyo” at the country scale, and “Beijing, London” at the continent scale. Training experiments were conducted before the formal navigation experiment to test whether all subjects were familiar with these locations.


Introduction
The mechanisms by which humans navigate at different spatial scales have been extensively studied over the past several years 1 . Understanding the nature of human multi-scale navigation is a critical issue for a range of applications, including navigation system design 2 , spatial ability evaluation 3,4 , and detection of cognitive impairment, such as preclinical Alzheimer's disease 5 . Classical theories have attempted to elucidate human spatial representations of navigational environments 6 , including unitary 7,8 , dual 9,10 and multiple 11 systems theories. These theories hold contrary views regarding whether or not internal representations change with spatial scales, leading to the need for further neural-level studies examining scale-dependent and scale-free characteristics that are latent in the neural system and involved in human spatial navigation. A number of previous studies have examined spatial navigation at the small scale (e.g., room, building, town) [12][13][14] , seeking to understand the spatial processes by which humans perceive, interpret, mentally represent, and interact with spatial characteristics during navigation.
Studies of rodents and birds have revealed that different neural mechanisms are involved in spatial navigation between small and large scales [15][16][17] . However, no empirical studies have elucidated the neural mechanisms underlying the performance of large-scale navigation in humans, and how differences in scale drive changes in human brain activation patterns during spatial navigation.
Human spatial navigation involves a range of processes, including visual characteristic processing, spatial distance and direction coding, spatial representation under different reference frames, and route planning. Visual characteristic processing is the basis of spatial navigation, and humans must acquire spatial knowledge to satisfy the information required for self-localization and self-orientation, landmark anchoring and route planning 18 . Room, building, and block are the main scales involved in daily movement for humans in urban environments. At these scales, individuals can directly perceive and process geospatial entities and structures through physical or visual experience 19 {Citation}. At the city scale and above, due to the large amount of spatial information and limitations of accessibility and visibility, humans typically cannot intuitively perceive a full picture of geographic space 11 . Instead, human cognitive processing of large-scale geographic space occurs through indirect learning of spatial knowledge via abstract expressions 20 . Although human visual processing in spatial navigation has been extensively studied, few studies have examined how the human brain processes scenarios across multiple scales.
In addition to visual feature processing, spatial distance and direction must be coded to measure spatial relationships between two isolated places. Studies examining regional brain activation have shown that the hippocampus and entorhinal cortex are strongly involved in distance processing [21][22][23][24] . In human navigation, Euclidean distances between two objects are reported to be processed in the entorhinal cortex, whereas, when it is necessary to detour, path distances are processed by the hippocampus 21 . For direction processing, besides head direction cells, which can measure spatial direction accurately 25 , the cingulate gyrus is an important brain region related to direction processing 26 . A previous study reported that, when humans process directions involving visible objects, the posterior parahippocampal place area (PPA) and retrosplenial complex (RSC) are activated. However, when locating isolated objects that are not in a visible range, such as a gate behind a building, anterior PPA and RSC are activated 6 . The study also explored how humans judge spatial distance across six common scales, revealing cortical gradient processing of spatial information from concrete to abstract 6 . No previous studies have examined how humans process spatial distance and direction during multi-scale navigation.
The reference frame is another fundamental issue in the cognitive neuroscience of spatial navigation.
The egocentric system is de ned as a viewer-centered perspective, which contributes to external object orientation via deviations in head direction 14 . In contrast, the allocentric system is independent of both the viewer and external objects, functioning as a global coordinate system 14 . According to small scale navigation studies, both egocentric and allocentric systems appear to represent space in parallel and function complementarily during navigation 27 . The egocentric system includes viewpoint imagination and mental rotation, and keeps the head direction in accord with the planned route. In contrast, the allocentric system helps an individual orient to unseen or distant objects. To form a reference frame, individuals must obtain su cient spatial knowledge about the navigation environment. However, as mentioned above, for scales larger than the neighborhood scale, spatial learning using maps differs substantially from spatial learning at smaller scales via real scene perception. Whether there are differences in the use of reference frames between navigation at different scales remains unclear.
In the current study, we hypothesized that the human brain uses a scale-dependent functional system to perform navigation at different scales. The aim of this study was to elucidate how and why differences in scale cause different brain functional activation patterns during spatial navigation. To this end, we tested subjects while they imagined multi-scale navigation, and compared differences in brain activity between scales. Speci cally, we focused on how humans process scenario information and choose spatial references during navigation at different scales.

Results
We compared experimental trials with a baseline condition at the whole-brain level. The results are shown in Figure 1 To investigate which brain regions contribute to scale-driven navigation, we directly compared brain responses during spatial navigation at different scales. The effects of spatial navigation in the brain driven by differences in scale are shown in Figure 2 and 3. Four brain regions exhibited signi cant differences between scales, namely the parahippocampal gyrus (Figure 2 (a-b)), cingulate gyrus ( Figure  2 (c-d)), parietal gyrus (Figure 3 (a-b)) and temporal gyrus (Figure 3 (c-d)) (all p < 0.05, AlphaSim corrected). Between the room and building scales, we found greater activation at the building scale in the parahippocampal gyrus, anterior cingulate gyrus and middle temporal gyrus. Between the building and block scales, we found greater activation at the building scale in the parahippocampal gyrus, posterior cingulate gurus, postcentral gyrus and middle temporal gyrus. Between the building and country scales, we found greater activation at the building scale in the parahippocampal gyrus, anterior cingulate gyrus and posterior cingulate gyrus, whereas greater activation at the country scale was observed in the middle cingulate gyrus, postcentral gyrus and middle temporal gyrus. Between the block and country scales, we found greater activation at the country scale in the postcentral gyrus and middle temporal gyrus. There were no signi cant differences between the block and city scales, or between the city, country and continent scales. In general, we summarized brain regions on the basis of scale dependency. Figure 4 (a) shows scale-dependent brain regions, including the parahippocampal, cingulate, parietal and temporal gyrus, which exhibited activation that was dependent on changes in the spatial scale of navigation environments. Figure 4 (b) shows the scale-free activation observed in the middle occipital gyrus, which was independent of changes in the spatial scale of navigation environments, but related to navigation tasks.

Discussion
In this study, we identi ed four scale-related brain activation patterns during the processing of spatial navigation, with signi cant differences in brain activation patterns across scales. These ndings indicate the existence of unique human navigation strategies at four spatial scales, namely the small scale (between rooms), medium scale (between buildings), large scale (between blocks) and huge scale (between cities, countries and continents). The ndings and implications are summarized as below.
First, the results of the comparison between the task and baseline conditions indicated that the middle occipital gyrus was the most important brain region for completing the tasks in this experiment. A previous study reported that the middle occipital gyrus is specialized in processing Chinese characters 28 .
Thus, the observed activation in this region may have been caused by requiring participants to read Chinese words during the experiment. However, another study reported that activation in the middle occipital gyrus was strongly related to successful navigation 29 . In addition, some previous research suggests that the middle occipital gyrus is involved in object processing [30][31][32] , indicating that participants processed knowledge of landmarks located in the navigation scenarios. Furthermore, the middle occipital gyrus is also reported that it plays a role in processing objects at different levels of speci city 33 , suggesting that participants processed landmarks at different scales with the involvement of the middle occipital gyrus, and that the speci cation of processed landmarks is dependent on the scale of the navigation scenario. Taken together with these previous ndings, the current results suggest that the middle occipital gyrus is a scale-free brain region for human navigation.
The parahippocampal gyrus is an important brain region in spatial processing. The results indicate that activation of the parahippocampal gyrus at the medium scale was signi cantly greater than that at the small and larger scales. The parahippocampal gyrus is considered to play an important role in view encoding in spatial navigation, particularly for views of buildings 34 , which may have caused the greater activation observed at the building scale in the current study. Hassabis et al. reported that activation in the parahippocampal gyrus discriminated between different environments 35 , in accord with the current results, suggesting that the medium scale is a key demarcation in human spatial navigation. These studies provide evidence suggesting that the parahippocampal gyrus discriminates between processing views at different spatial scales. Moreover, the parahippocampal gyrus was reported to be involved in the process of perceiving and using landmarks 18 , and was strongly activated when participants attempted to increase navigational accuracy 36 . The current ndings suggest that participants paid substantially more attention to retrieving and reusing landmarks based on their memory of routes to complete navigation tasks at the medium scale. In addition, the parahippocampal gyrus is considered to be responsible for the processing of global information about the environment in humans 37 , receiving egocentric information and converting it to allocentric representations, and helping navigation in an allocentric view 36,38 . The current ndings suggest that participants tended to use an allocentric reference to complete the navigation tasks.
The cingulate gyrus is also a scale-dependent brain region. We found signi cant differences in anterior cingulate gyrus activation between the medium and huge scales, but not between the small and large scales. Previous studies reported that the anterior cingulate gyrus is related to backtracking strategy 39 . The anterior cingulate gyrus is also reported to exhibit more activation when participants experience di culties in navigation tasks 36 . Therefore, our ndings suggest that participants experienced more di culty and performed more backtracking at the small to large scales. In contrast, at the huge scale, navigation appeared to be easier and more direct. We also found signi cant differences in posterior cingulate gyrus activation between the medium scale and above, but no differences between the small and medium scales. The posterior cingulate gyrus is thought to be involved in transforming allocentric references to egocentric references 40 . The current ndings and our oral investigation of the route training sessions indicate that, at the small to large scale, participants tended to imagine an abstract structure of the entire environment using an allocentric reference. In addition, at the small to medium scales, participants also switched spatial references frequently. Thus, participants may have been imagining real processes of navigation and processing local information with an egocentric reference.
Activation was in uenced by spatial scale in several other brain regions. First, we found a scaledependent difference in activation in the parietal gyrus. The results indicated that activation in parietal gyrus increased monotonically with spatial scale, in accordance with a previous report by Peers et al. that stated that the parietal gyrus exhibited more activation at large scales 6 . The parietal gyrus is involved in processing egocentric references 41,42 . Therefore, our ndings suggested that, at the huge scale, participants completed navigation by imagining maps in an egocentric reference frame, in which routes were simpli ed on the basis of the paths of ights or ships. Second, we found that the temporal gyrus was strongly activated at the huge scale. The temporal gyrus is related to the processing of spatial contextual information for route planning, and distinguishing between navigational episodes 43 . Some previous studies examining spatial navigation among patients with Alzheimer's disease suggested that patients suffered from spatial disorientation due to neurodegeneration in the medial temporal gyrus and parietal gyrus [44][45][46][47] . In addition, several studies reported that the parietal gyrus is related to the processing of direction 41,48 . Therefore, although participants were able to easily plan and navigate along a route, they may have found it harder to determine accurate directions because they naturally retrieved and processed fewer details about the environment compared with the small to large scales.
At all of the scales we investigated, we observed a scale-dependent neural system for human spatial navigation, and there are four typical scale-dependent brain activation patterns and corresponded navigation strategies. In the current study, at the small scale, participants used both egocentric and allocentric references, but processed real and detailed scenario views more, indicating that participants mainly used egocentric references during navigation. At the medium scale, participants used both egocentric and allocentric references, with no obvious differences, and tended to switch reference type frequently to cope with different scenarios during a single navigation task. Participants processed both speci c landmark information and abstract route networks. This type of navigation strategy enabled participants to process information comprehensively and navigate more accurately, but also naturally led to a higher level of di culty in completing tasks. At the large scale, participants mainly used an allocentric reference and processed abstract spatial information, because the amount of information at this spatial scale was overly large for participants to process, and they had to rely on abstract information by simplifying the spatial information into route networks and several key decision points. At the huge scale, participants mainly used an egocentric reference to process undetailed and abstract mental maps. Our ndings are consistent with the multiple system theory for spatial representation 11 , and also verify the fact that human activate a speci c neural system 6 to accomplish different navigation strategies proper to spatial scales 49 .
In conclusion, the current study provided whole-brain, voxel-based evidence supporting the role of various brain regions in scale-driven spatial navigation. The current results revealed, for the rst time, a scaledependent neural system that included the parahippocampus, cingulate, parietal and temporal gyrus, and revealed a neural-based division of spatial scale: small scale (room), medium scale (building), large scale (block), and huge scale (city, country and continent). These four scales were associated with obvious differences from the perspective of spatial information processing and spatial referencing, revealing the neuronal basis of the in uence of scale on human navigation and spatial cognition. In addition, activity in the middle occipital gyrus was found to be independent of scale in human spatial navigation. Furthermore, this nding provides potentially useful information regarding the design of auto-adapted navigation systems for multi-scale navigation, contributing to the investigation of novel methods for detecting preclinical Alzheimer's disease via multi-scale navigation tests.

Subjects.
We recruited 11 human subjects ( ve males, aged 19-24 years) for the experiments. All of the subjects were Chinese undergraduate or graduate students at universities in Beijing, China. All subjects had a high level of education and had good navigation skills. Four subjects majored in geography or geographyrelated subjects. All of the subjects were trained in navigation tasks before fMRI scanning. All subjects provided written informed consent before the experiment. All experimental procedures were performed in accordance with the Declaration of Helsinki. The experiment was reviewed and approved by the Beijing Normal University Research Ethics Committee.

Experimental stimuli.
Location pairs were used as experimental stimuli, which were displayed as text on the screen during scanning. Six scales were set as the conditions of the experiment, including room, building, block, city, country and continent, in accordance with a previous study. All of the location pairs comprised two locations at the same scale. For example, at the room scale, we used the room names as the location pairs. Similarly, for the building scale and the block scale, we used the building names and the block names, respectively. For scales larger than block, including the city, country and continent scales, city names were used for the location pairs, such as "Beijing, Shanghai" at the city scale, "Beijing, Tokyo" at the country scale, and "Beijing, London" at the continent scale. Training experiments were conducted before the formal navigation experiment to test whether all subjects were familiar with these locations.
The location pairs were displayed with a top-bottom layout. The starting points were located at the top and the end points were located at the bottom.
Experimental paradigm: The experimental task required subjects to plan a route on the basis of the displayed location pair, and to imagine themselves navigating the planed route. The fMRI experiment conformed to a randomized block design paradigm. Each block lasted 12.5 s followed by a 7.5 s xation period. In each block, the stimuli were displayed, and subjects began the navigation task. Subjects stopped the task once they saw a xation cross. The experiment included four runs, and each run comprised 24 blocks (four blocks per scale condition).
Statistical analysis. SPM 12 was used for statistical analysis of fMRI data. First, we conducted individual-level analysis using the general linear model to estimate the beta values of different contrasts between scales. Bloodoxygenation-level-dependent signals were averaged for all blocks containing each scale across all runs. Second, we conducted one-sample t-tests for group level analysis. Beta plots were also created by averaging the beta values calculated in the random-effects general linear model analysis across all subjects.

Declarations
Data availability.
All the material will be available on request from the corresponding author.