Long-term consolidation switches goal proximity coding from hippocampus to retrosplenial cortex

Recent research indicates the hippocampus may code the distance to the goal during navigation of newly learned environments. It is unclear however, whether this also pertains to highly familiar environments where extensive systems-level consolidation is thought to have transformed mnemonic representations. Here we recorded fMRI while University College London and Imperial College London students navigated virtual simulations of their own familiar campus (> 2 years of exposure) and the other campus learned days before scanning. Posterior hippocampal activity tracked the proximity to the goal in the newly learned campus, but not in the familiar campus. By contrast retrosplenial cortex tracked the distance to the goal in the familiar campus, but not in the recently learned campus. These responses were abolished when participants were guided to their goal by external cues. These results open new avenues of research on navigation and consolidation of spatial information and help advance models of how neural circuits support navigation in novel and highly familiar environments. Significance Statement Historically, research on the hippocampal formation has focused on its role in long-term memory and navigation – often in isolation. No study to date has directly compared realistic navigation within familiar with recently learned environments, nor has it been explored how the neural substrates, along with computational codes, may change. In this study, we show for the first time, a shift from hippocampal to cortical coding of distance to a goal during active navigation. This study bridges the gap between memory consolidation and navigation, and paves the way for more functional and realistic understanding of the hippocampus.


Introduction
Understanding how the brain consolidates memories is a central question in neuroscience (1). Historically, research has focused on contextual and recognition memory in rodents and primates (2)(3)(4) and episodic memory in humans (5)(6)(7).
Despite substantial interest in the neural circuits that support navigation, little research has explored consolidation of spatial memories and representations of environments learned months or years ago ("familiar" environments) (8)(9)(10). This dearth of research is particularly surprising considering that current theories disagree about the contribution of the hippocampus to processing spatial representations over time: the standard consolidation theory (SCT) argues that initially the hippocampus is involved in processing the spatial memories and representations, and that over time the representations in neocortical regions are strengthened, reducing the demand on the hippocampus (11,12). By contrast multiple trace theory (MTT), and its offspring Trace Transformation Theory, argue that for detailed spatial memories and representations the hippocampus is always involved, or more specifically that an episodic hippocampal trace will exist in addition to the schematized representation in the cortex that can be activated depending on task requirements (9,(13)(14)(15)(16).
Neuropsychological evidence indicates that complex spatial memories acquired years in the past can become independent of the hippocampus (17-21) consistent with SCT. In several cases, however, hippocampal damage does appear to lead to impaired spatial memories for some detailed aspects of the environment Only two prior fMRI experiments have examined navigation of familiar environments.
One study involving London taxi drivers navigating a virtual simulation of London (UK) reported that the hippocampus is engaged at the start of navigating this highly familiar environment (22). The other study involved residents of Toronto mentally navigating this city and found no increased activity in the hippocampus (23).
Crucially, however, neither study directly compared navigation in familiar with recently learned environments. Although there is evidence from fMRI that the hippocampus is less implicated in mentally navigating in familiar environments (Hirshhorn, Grady, Rosenbaum, Winocur, & Moscovitch, 2012), exactly what the contribution of the hippocampus and other structures is and the nature of their computations is not known. Additionally, as these studies have relied on static, mental navigation, they may not engage hippocampal activity as would active, dynamic navigation in a virtual reality environment. Thus, for now, we cannot rule out whether the differences in hippocampal findings relate to the demands of navigating different cities, with London placing greater demands on mental simulation of future familiar routes than Toronto, or whether the structure of the environment is key to these differences (10).
One question not yet addressed is whether long-term consolidation changes the spatial information processed by brain regions during the navigation of an environment. In recently learned environments, the hippocampus has been shown to encode the distance to the goal (Balaguer, Spiers It is also possible that the involvement of brain regions will vary depending on how individuals plan their route or use certain strategies, with MTT/TTT predicting that the hippocampus, and in particular the posterior hippocampus, would play a more important role when perceptually detailed processing is required (39).
Here we combined fMRI and a virtual simulation of two university campuses to examine the brain regions coding the distance to the goal in highly familiar and recently learned environments within the same scan session. Students from two London universities (University College London and Imperial College London) navigated each campus, with the one they were not attending made familiar via 5 training material and a walking tour days before the fMRI session. During the scanning session they engaged in active navigation towards goal locations and subsequently reported on when they planned their routes. We aimed to test whether a) the hippocampus codes distances to the goal, b) this would depend on which environment they were in, and if this code may be represented elsewhere in the brain after extended exposure, and c) how these representations were related to route planning.

Results
Our

fMRI Analysis
A categorical analysis of different event types and task blocks revealed more activity in lateral and medial parietal areas in familiar environments when actively navigating ("navigation" condition) compared to just following along a route ("control" condition), whereas recent environments did not show such a distinction ( Figure  S1). There were no clear differences when comparing familiar to recent navigation ( Figure  S2, and Supplementary Results). In order to examine how the neural responses relate to metrics in computational models we interrogated our fMRI data with parameters related to the distance and direction to the goal. We explored how path distance to the goal ( Figure  1A) was correlated with brain activity during events sampled during Travel Periods and at Decision Points. We focused on these events because we were able to sample >20 events per condition (familiar/recent, navigation/control).
Hippocampus and retrosplenial cortex track path distance to the goal during travel periods in recent and familiar environments, respectively During Travel Periods when navigating the recently learned environment, we found a significant negative correlation with the distance to the goal in the right mid-posterior hippocampus ( Figure  2A, Table  S4), indicating that hippocampal activity increased with proximity to the goal. This effect was absent in the familiar environment and the control routes ( Figure  2A, right panel). Moreover, right posterior hippocampal activity was significantly more correlated with proximity to the goal during navigation of the recently learned environment than the other conditions combined ( Figure  2C, Table   S4). Although the hippocampus did not appear to track the distance to the goal in the highly familiar environment, we found that the retrosplenial cortex did. During Travel Periods of navigation routes in the familiar environment we observed a significant positive correlation with distance in the retrosplenial cortex ( Figure  2B, Table  S4), indicating the activity was greatest when participants were farthest from their goal.
This response was absent in the recent environment and during the control conditions ( Figure  2B, right panel). Retrosplenial activity was significantly more correlated with the distance to the goal in familiar navigation routes than the other conditions combined ( Figure  2D, Table  S4). Even at liberal thresholds, activity in our other predicted ROIs (anterior cingulate cortex, caudate and parahippocampal cortices) was not significantly correlated with distance to the goal during Travel Periods in either recent or familiar environments ( Figure  S3).
In our previous study (Howard et al., 2014) we found that during the Travel Period hippocampal activity was positively, rather than negatively, correlated with the distance to the goal. Because the previous study used a much smaller sized environment (200-400m route lengths) than the current one (200-1000m route lengths), we examined the impact of down-sampling the range distances in our analysis, excluding those over 900m (accounting for the top 25% of distances, see Figure  S4A and Supplementary Methods). This analysis abolished the correlations between distance and activity in hippocampus and retrosplenial cortex, indicating that the larger distances were important in driving the relationship between distance and activity. Notably, the correlations, however, were not abolished when we downsampled the data by removing the same number of events, but randomly sampled across the whole distance range ( Figure  S4A, lower panel). Though in this study path distance and Euclidian distance were highly correlated (Table  S1), we ran a control model including both spatial parameters, and found both the hippocampal and retrosplenial results remained significant (Table  S4).
Hippocampal activity is positively correlated with the distance to the goal at decision points At Decision Points during the navigation of the recently learned environment right posterior hippocampal activity was positively correlated with the path distance to the goal ( Figure  3A, Table  S4). This response was observed when including inverse efficiency scores (IES) in the model (see Supplemental Methods), to account for differences in reaction times and accuracy across events. This response was absent in the familiar environment and control routes, and there was also no parametric hippocampal response to IES on its own, underscoring the fact that it was not behavioural differences between environments that was driving the hippocampal response. Right posterior hippocampal activity was significantly more correlated with the distance to the goal in recent navigation routes than the other conditions combined, albeit at a threshold of p < 0.005 uncorrected ( Figure  3B). Conducting the same analysis used for Travel Periods of excluding the events where the goal was a long distance away revealed that the significant correlation between activity and distance to the goal was dependent on these long distances ( Figure  S4B).  Table  S4). This same measure of planning was also correlated with hippocampal activity during Travel Periods in familiar environments ( Figure  4B, Table  S4). We found performance accuracy at Decision Points did not correlate with hippocampal activity during navigation routes, either in the recently learned environment or the familiar environment (familiar and recent environments both: r<0.3, p>0.1), and that the amount of planning reported was not correlated with performance accuracy (familiar and recent environments both: r<-.06, p>0.1).
Self-reported 'map-based' navigators have stronger correlations between hippocampal activity and the distance to the goal than 'route-based' navigators Prior evidence indicates that strategy use for navigation impacts on the engagement of different brain regions for navigation of simulated environments (40, 41). To test whether this is true for navigation of real-world environments, participants completed a questionnaire probing navigational strategy use (see Supplemental Methods). The questionnaire determines the extent to which a person uses a map-based approach to navigation or a sequential landmark-based approach to navigation. Participants with higher map-based navigation scores had significantly stronger negative correlations between right posterior hippocampal activity and the distance to the goal during Travel Period in the recently learned environments ( Figure  4C). We found no correlation between this self-reported strategy use and the amount of route planning (r<-0.18,p>0.1).

Discussion
Using a virtual simulation of two London university campuses and fMRI we examined how, during navigation, the distance to the goal is represented by brain regions when the environment is highly familiar or recently learned. We found that in recently learned environments, the right posterior hippocampal activity is correlated with the distance to the goal, while in familiar environments distance to the goal was correlated with activity in the retrosplenial cortex. For recently learned environments, the more participants reported that they navigate via map-like strategies, the stronger the correlation between hippocampal activity and the distance to the goal.
While overall we found that hippocampal activity was specific to a recently learned environment, participants who reported more route planning in a post-scan debrief showed increased hippocampal activity when navigating a familiar environment.
These results help inform debates about memory consolidation, navigation systems of the brain, and the functional differentiation of the long-axis of the hippocampus.

Systems consolidation of spatial representations
Our finding that hippocampal activity tracks the distance to the goal in recently learned, but not familiar environments, is consistent with standard consolidation models which argue that with the passage of time and repetition, information initially processed by the hippocampus becomes consolidated within neocortical structures (4,42,43). It is also consistent with evidence that lesions to the hippocampus lead to deficits in navigating new environments (44, 45), but not highly familiar environments (10,(17)(18)(19)(20)(21). As such, upon first inspection, our primary result would appear to conflict with the prediction from multiple trace theory (MTT) that detailed spatial processing, such as the distance to the goal, should always require the hippocampus no matter how familiar the environment (16). However, we speculated, in accord with TTT, that participants might alter how they navigate as the environment becomes more familiar and examined this with a post-scan debrief. Consistent with this prediction, we found that participants reported more route planning in the recently learned environment than the familiar environment. Moreover, we found that the more route planning a person reports in the familiar environment at decision points, the more hippocampal activity they express, which aligns well with the proposal in MTT/TTT that the hippocampus plays a continual role in supporting navigation when detailed processing is required (16). This finding is also broadly consistent with the argument that the hippocampal role in many tasks derives from a need to construct or re-construct scene representations (46). Our findings suggest that the amount of time spent in an environment, making it familiar, may not be the key determinant of the brain regions engaged, rather it is the shift in how an individual navigates the environment.
The brain regions responsible for processing consolidated spatial representations have remained somewhat mysterious (10). Here we found that the retrosplenial cortex codes spatial information in highly familiar environments rather than in recently learned environments. This finding is consistent with fMRI evidence that hippocampal activity declines with learning the layout of a virtual environment, but the activity in the retrosplenial cortex increases, tracking the new learning of stable spatial relationships in the environment (47-49). It is also consistent with reports of disorientation in highly familiar environments after retrosplenial lesions (10, 50, 51).

Representation of the distance to the goal
Our finding that hippocampal activity was correlated with the distance to the goal in recently learned environment agrees with prior fMRI reports of similar coding (26, 27, 29, 30) and supports models which argue the hippocampus computes information about the future path to goal for navigational guidance (52-54). The observation that this was specific to navigation routes and not present in control routes is consistent with prior evidence (26) that such distance tracking is not automatic and requires goal-directed navigation. It is possible that the hippocampal activity correlated with distance to the goal relates to the pre-activation of place cells along a route to the goal, or 'forward-sweeps' of activity towards the goal (55). However, such responses may also relate to the recently discovered path distance coding neurons in the CA1 Our data provide further support for this perspective in that the right posterior hippocampal activity is modulated by both the propensity for route planning in familiar environments and the extent to which a person reports using map-based strategies for navigation.

Conclusion
In summary, our fMRI study provides the first comparison of active navigation of both a recently learned and a highly familiar environment. Our data supports models in which there is change in demand on brain regions with extended consolidation of the memories of an environment, with the hippocampus representing the distance to the goal in recently learned and the retrosplenial cortex in familiar environments. We also find support for the view that detailed spatial processing of an environment will involve the hippocampus even when recalling highly familiar environments (6, 16), and that this will entail the right posterior hippocampus (39, 62). Future research will be useful to determine how neuronal-level activity in the hippocampus and retrosplenial cortex may give rise to the fMRI signal dynamics reported here. Participants were trained on both the recent and familiar campuses in real life with a guided tour by an experimenter, with a strict set of rules, which were as follows: 1)

Participants
Each road had to be walked past twice, in both directions, and each goal location had to be visited twice. 2) A probe for the name of each goal location was asked once before each visit (experimenter pointed in the direction of the nearest goal location before it became visible). 3) After arriving at each goal location, its name was read to the participant, and the direction of the start location was also given if the goal location was also a starting point. 4) On five occasions, participants were asked to point out the directions of two goal locations that they had visited twice. 5) The name of each street was asked twice, while the participant was not on it, before and after visiting it. 6) At the end of training, participants were asked about the directions of 10 goal locations and the names of the streets where they were located.
The order in which participants were trained on the campuses was counterbalanced across participants and familiarity, and was done to ensure that the familiar campus was also recently visited in its entirety, thereby removing any confounding effects of just the recency of exposure (rather than the age of the memory itself). See Figure  1 for summary.

Task Design
The task in the scanner was designed to simulate walking through the campuses, by Therefore, we will focus only on Travel periods and Decision Points, as they have a sufficient number of events (>20 per condition).

Post-scan Debrief
Immediately after the scan there was a brief interview. All navigation routes that each participant was tested on were replayed in the same order as in the scanner.
Participants were instructed to report what they remembered thinking during the navigation, not what they should have done, and to answer questions posed by the experimenter. At the start of each route they were asked "Were you oriented from the beginning?" - this was during the screen shown at the street entry. After that the experimenter pressed the play button. The navigation automatically paused whenever a New Goal Event appeared. Before and after each junction participants were told the responses made in the scanner and the experimenter would ask "Were you planning the route to the goal at this point during the scanning?". At detours they were asked "Were you re-planning at this point?". They were also asked if they were lost after detours. To this end, we acquired data at the following events (per familiar and recent campus): oriented, lost, and planning (at New Goal Events, Decision Points and Detours). Participants were also asked to report any salient memory at any point during navigation. All interviews were audio recorded.

Spatial Parameters
Calculation  Table  S1 for correlation between spatial parameters at each event type. The aim was to create routes where the spatial parameters were maximally decorrelated. However, due to the nature and layout of the campuses, there were limits on the flexibility of route design. Therefore, we entered path distance independently as a parametric regressor in our analyses, as this was the main focus of our task. However, we also checked our results when including ED along with PD, to establish the robustness of our findings, even with highly correlated regressors. Spatial parameter values were scaled between 0 and 1.  Table  S2 for a description of the models, events included and regression parameters (if applicable). Note for parametric modulation models, the event of interest was modelled with the corresponding spatial parameter regressors (i.e., Path Distance, Euclidian Distance, and Egocentric Goal Direction), but also included the other events in order to fully account for activity relating to stimulation. Additionally, we also included a task block regressor, which indicated whether the task was performed in a familiar or recent environment, and navigation or control. Only the implicit baseline (fixation period) of 17 seconds between routes was not included in the model.  The training protocol consisted of two stages. First, a self-guided learning of landmarks and campus layouts with printed training materials. Two days before the fMRI scan participants were taken on an intensive guided tour of one of the campuses, and the next day, they were taken around the other campus. The order was counterbalanced across subjects and could start at either their home campus or the new one. On the day of scanning, participants completed routes in both campuses, in both active navigation and a control condition that involved following directions. During the debrief session, they filled out questionnaires regarding navigation strategies, and performed a behavioural task where they indicated where along the route they engaged in planning while navigating. C) Excerpt from a "navigation" route at UCL. At the start of every route, participants were shown the street they are on, as well as their facing direction. Next they were given a New Goal Location, and subsequently asked to indicate the general direction of that landmark. They travelled down the road until they reach a junction (the duration and number of images of this was variable depending on the length of the street), at which point they were asked to indicate the correct turn to take towards the goal (Decision Point). Here, the program automatically advanced on a pre-defined route, which may have been correct, or it may have been an unplanned detour (occurring less frequently). After a variable interval, a new street was entered and the participant was informed again of their location and facing direction. Note the jittered interval after Decision Points is to allow for separating signals relating to Decision Points and Turns (or Detours). Analysis of fMRI data was constrained to the Travel and Decision Point periods. For fMRI analysis, Travel periods were taken as the mid-point between two events, and modelled as a punctate event. For simplicity, the above figure only depicts part of a route, and Travel is shown as two frames, but could range between 3-48 frames (on average: mean 31±18). Figure  2. Travel distance in familiar and recent environments is coded in different brain areas. A) During Travel in recent environments, there was a significant negative correlation with path distance, such that there was higher BOLD activity in the hippocampus when participants were closer to the goal location. B) During Travel in familiar environments, there was a significant positive correlation with path distance, such that there was higher BOLD activity in the retrosplenial cortex when participants were further away from the goal location. In each plot, from left to right: parameter estimates (PE) extracted from a categorical model (binned by distance), the BOLD activity for the relevant condition, and the PE from the peak voxel in the ROI for each condition. All effects survive small-volume correction, including when ED (Euclidian distance) is added to the models. C/D) Show the data when the GLM included weighted regressors for the effects seen in A/B, respectively. For example, in C, the contrast was 1 -3 1 1, testing for an overall effect of correlation with path distance during Travel, for the recent navigation condition compared to all other conditions. The hippocampal and retrosplenial effects from A and B, respectively, are replicated indicating that the correlations with path distance are selective to these areas. *=p<0.05 SVC, †p<0.1 SVC Figure  3: Path distance coding at Decision Points. A) At Decision Points, there was a significant positive correlation with path distance, such that there was higher BOLD activity in the right hippocampus when participants were further away from the goal location in recent environments. The effect plotted is corrected for IES (inverse efficiency), and is significant with and without this correction, thus underscoring that it is not a RT (or difficulty) effect. It also survives small-volume correction, including when ED (Euclidian distance) is added to the model. B) Brain activity when the GLM included weighted regressors for the effects seen in A. The contrast was -1 3 -1 -1, testing for an overall correlation with path distance during Decision Points, for the recent navigation condition. *=p<0.05 SVC, †p=0.005 u.c.   Table  S2: Details of GLM parameters for the fMRI models *Inverse efficiency: trial-by-trial RT/mean accuracy to control for differences in RT between familiar and recent condition during navigation (see behavioural results below) **Note that the GLM was run post-hoc after finding a significant correlation using extracted parameter estimates per person (in the peak hippocampal voxel for the Travel PD effect (recent navigate)) and mapping scores.  (2) x Nav (2) PD, ED DP Environment (2) x Nav (2) PD, ED, IES* Travel - long distances removed (25% of total) Environment (2) x Nav (2) PD DP - long distances removed (25% of total )

Supplementary Results
Behaviour In the pre-training assessment, we wanted to confirm that participants knew their own campus well, and were unfamiliar with the other campus. In the familiar environment participants were on average 48% and 86% accurate on street names and landmarks, respectively. Conversely, for the recent campus, they were only 10% accurate on the street names, and 22% for landmarks. Thus, there was a significant effect of environment (F(1,24)=140.8,p<0.001), type of information probed (F(1,24)=48.9,p<0.001), and interaction (F(1,24)=14.8,p=0.001). Post-hoc paired ttest were also all significant (all t>3, p<0.004), underscoring the notion that participants did indeed have better knowledge about their own campus, and mainly its landmarks.
Our main interest was performance on the task performed during the scanning, which had a 2x2 factorial design (environment x navigation). We compared reaction times and accuracy at both New Goal Events (NGE) and Decision Points (DP), between familiar and recent environments, for both navigation and follow conditions using repeated-measures ANOVAs. At NGEs, there was a main effect of environment (F(1,24)=4.5,p=0.045), and a paired-t test revealed a trend for significantly lower RT in familiar navigate compared to recent navigate only (t=-1.9,p=0.064). For NGE accuracy, there was an overall effect of navigation (F(1,24)=41.8,p<0.001), with follow conditions having overall higher accuracy (as would be expected given participants were given the correct answer). For DP RTs, there was a main effect of navigation, environment and a significant interaction (all F>13.3, p=0.001). Paired-t tests confirmed that in the navigation condition, there was a significantly lower RT to DPs in familiar environments (t=-4.7, p=0.001). The same pattern was seen for DP accuracies. To control for potential differences in brain activation due to reaction time differences in the familiarity condition, we included a trial-by-trial inverse efficiency score (RT/mean accuracy) as a regressor when modelling the DPs.
For the debrief session, we compared responses to familiar and recent environments and found that although participants were equally oriented in both environments during NGEs (t=1.7, p>.1) and were only occasionally lost (less than 1%, but more so in the recent: t=-2.1, p=0.043), they did report more planning at NGEs, DPs, and Detours in the recent environment (all t>-2.9, p<0.006). We also found that participants who planned more at NGEs and DPs in familiar environments also planned more in recent ones (NGE: r>.88,p<0.001;; DP: r=.6,p=0.001), and there was a correlation between how much one planned between NGEs and DPs in recent environments, (r=0.41,p=0.042), but not in familiar environments. In relation to the questionnaire measures, there was a trend towards a correlation between the NSQ & being lost in familiar environments (r=-.37, p=0.06), indicating that participants with higher scores on the questionnaire tended to be less lost. Relating performance in the scanner to debrief responses we found that NGE RT in familiar environments correlated with amount of planning at recent DPs (r=.46, p=0.02), such that when they were quicker to respond to goal, it meant less planning at decision points. Additionally, DP RT in recent environments correlated with amount of planning at DPs (both familiar and recent;; >r=.42, p=0.03), such that less planning at DPs meant quicker responses. However, there was no correlation between planning and accuracy (all r<-.16, p>0.1).
We also looked at behaviour and its relation to spatial measures (for the navigation condition only), and found that both DP RTs in both environments, and NGE RTs in recent environments correlated significantly with PD (r>0.24, p<0.05), such that larger path distances incurred longer RTs. See Table  S3 below for correlation coefficients. Finally, we investigated the strategies questionnaires. Participants scored an average of 4.2 (range: 2.7-5.7) on the SBSDS. On the NSQ, they scored an average of 7.2 points (out of 14, where the maximum indicates only map-based navigation strategies). There was no correlation between the scores on these two questionnaires, between these scores and performance in the scanner, or to the amount of planning reported. We explored global differences in brain activity, at each event of interest relating to the navigation vs follow condition, in both environments (see Figure  S3). For familiar environments there was significantly greater activity in the precuneus for Travel, New Goal Events and Detours, and in the retrosplenial cortex for New Goal Events and Decision Points (Travel periods also showed this pattern but did not survive small volume correction [SVC]). New Goal Events also activated the visual cortex and intraparietal attention areas during navigation. The anterior cingulate region was also more active during navigation at Detours. This pattern of results is broadly consistent with the networks found during navigation in a similar paradigm reported by Howard et al (2014), however, our a-priori hypothesis was that this study would be more comparable to the 'recent' environment. Activity in recent environments in the current study were virtually absent when contrasting navigation vs follow conditions, perhaps reflecting engagement of similar networks, or continued encoding, when in a newly learned environment, regardless of the task demands.
We were also interested how navigation in familiar vs recent environments may rely on different neural substrates. When looking at what areas were more active in the familiar environments (compared to recent, and only during navigation), we found again the precuneus during Detours and the retrosplenial cortex during Travel periods. Additionally the anterior cingulate cortex was more active at Decision Points. For the reverse contrast, the retrosplenial cortex showed increased activity during Detours. Activity at Decision Points was characterized by widespread temporal lobe and insular activation, as well as a more posterior to the ACC, dorsomedial PFC activation. More strikingly, when looking at overall session blocks of navigation in familiar vs recent, there was a clear and significant left hippocampal activation. All these contrasts are reported in Figures  S2  and  S3, as well as Table  S4, and all effects reported are thresholded at p=0.001 uncorrected.

Control Analyses for effects of Path Distance
For all analyses regarding path distance we chose to focus on Travel Periods and Decision Points as they occurred most frequently, and contributed a minimum of 20 trials per condition.
We also ran control analyses, in which we included Euclidian distance (ED) in the models investigating parametric effects of path distance (PD). We replicated the PD results, emphasizing the robustness of these effects (see Table  S4), despite highly correlated parametric regressors (see Table  S1).
As our original Travel analysis included both travel midpoints as well as New Street Entry events, we also ran two additional models, which separated these events. When looking at Travel midpoints only, we replicated the right hippocampal effect in recent navigation, though this only survived a mid-hippocampal SVC (Table  S4). We did not find the retrosplenial cortex in familiar navigation, but there was a positive correlation of PD in the precuneus and parietal-occipital sulcus. These effects were absent from the New Street Entry only models, emphasizing that the correlation with PD was specific to the Travel periods.
We also checked whether there was any evidence that the negative correlation with PD coding in the hippocampus was related to approaching the goal when it was straight ahead. We modeled Travel periods when this was the case, point-by-point, for segments that were at least 7 seconds long (minimum 2 TRs, which was 3.4 seconds). Two subjects did not have segments long enough (this depended on the routes they were given and some routes had shorter 'goal approach' segments) and were not included in the analysis. We did not find any evidence for coding in the hippocampus for recent navigation, which may be due to the reduced power (overall less samples), or perhaps that given the visualization of the task, the goal was not actually visible on each panorama image and as such it wasn't as clear as it would have been if it were a continuous movie stream. Finally, we explored how the topology of the routes may have affected distance coding. We looked at Travel events that were in associated with either few or many upcoming turns. For Travel periods involving 'few' upcoming turns (<3), the hippocampal and retrosplenial PD effects, in recent and familiar environments, respectively, were replicated, but the ROIs did not survive statistical thresholding. Interestingly, there were no significant effects in recent environments for 'many' upcoming turns, however, in familiar environments there was a significant hippocampal cluster, positively correlated with PD. Retrosplenial cortex was also correlated with PD but this was not significant. When the environment is complex, encompassing many turns or fragments, the hippocampus may be involved even in familiar environments. However the lack of this effect in the recent environment precludes any conclusive suggestions as to how PD may be coded depending on environmental complexity when there is a strong need for planning ahead. In addition we also ran a model in which the number of upcoming turns was included as a parametric regressor for Travel (instead of PD), and found no effects. Please see Table  S4 for z-scores and cluster sizes for significant effects. For small-volume corrections of the hippocampus and retrosplenial cortex, we used anatomical masks. The hippocampal ROI encompassed the mid and posterior right hippocampus (modified from Howard et al 2014). Egocentric Goal Coding Because of recent evidence of goal-related egocentric and proximity combined modulation of activity in the hippocampus (Howard et al., 2014;; Sarel et al., 2017) we also explored whether there was any such modulation at Decision Points. We found the mid hippocampus was more active the further away, and less directly ahead the goal was, in recent environments. We also found evidence of medial superior parietal activity correlated with the egocentric direction to the goal during Travel Period in the familiar environment, broadly consistent with prior findings (26, 28). Note egocentric goal direction was also highly correlated with path distance (Table  S1;; see TS4 for details of activation).