## Introduction

Target-oriented navigation relies on visual information from the environment (allothetic cues), taking into account simultaneous vestibular and somatosensory afferent inputs (idiothetic cues) for continuous updating of one’s own position in space during locomotion1. Multisensory inputs are equally relevant for ego- and allocentric navigation strategies2,3. Egocentric navigation is based on a compass-like strategy, where the navigator’s current position in space is the absolute reference point for all the objects in the surrounding environment2,4,5. This strategy is often used to retrace familiar routes. In contrast, allocentric navigation relies on a map-like strategy, where different objects of the environment can be set in relation to each other independent of the navigator’s current position in space6,7. This strategy is most appropriate for recombining novel routes. During real-space navigation there is continuous overlap and change of ego- and allocentric strategies3. Navigation is guided by a widespread network of brain regions (prefrontal cortex, basal ganglia, thalamus, cerebellar regions, posterior parietal cortex, retrosplenial cortex, posterior parahippocampus, lingual gyrus, hippocampus, and entorhinal cortex)7,8. Creating a mental representation of a novel environment seems to depend critically on the hippocampus, entorhinal cortex, and retrosplenial cortex with its highly specialized cell ensembles (place cells, grid cells, and head direction cells)3,7,9.

Bilateral vestibular damage in rodents leads to severe and persistent navigation deficits in real space by disrupting the head direction cell code in the dorsal brainstem tegmentum, anterior thalamus, subiculum and entorhinal cortex, the place cell code in the hippocampus, and the grid cell code in the entorhinal cortex10,11,12,13. Patients with complete bilateral vestibulopathy (BVP) show spatial memory deficits in a desktop-based virtual variant of the Morris Water Maze Task (vMWMT) accompanied by a significant hippocampal volume loss14. Delayed spatial learning performance in vMWMT and a decrease in gray matter hippocampal/parahippocampal volume were also found in patients with incomplete BVP15. However, in a virtual cityscape navigation paradigm, patients with incomplete BVP had no significant performance deficits, and exhibited increased activations in the posterior cerebellum16. The authors interpreted these findings as a change of the prevailing navigation strategy towards sequence-wise learning of certain routes by loops in the cerebellum, basal ganglia, and prefrontal cortex17.

## Results

### Patient characteristics

BVP patients (n = 14, 54.1 ± 12.2 years, 7 female) were classified as complete BVP (cBVP, n = 6, horizontal vHIT gain: 0, caloric response: 0 deg/s, missing o/cVEMP response) or incomplete BVP (iBVP, n = 8, horizontal vHIT gain: 0.44 ± 0.13, caloric response: 2.2 ± 1.18 deg/s, oVEMP pathological in n = 5, cVEMP pathological in n = 6), and compared to age-matched healthy controls (HC) (n = 14, 55.1 ± 10.1, 7 female). Patients with cBVP were younger (46.5 ± 7.4 years) than those with iBVP (63.3 ± 10.5 years) (p = 0.02). The duration of bilateral vestibular failure was comparable between both groups (cBVP: 8.0 ± 2.6 years; iBVP: 8.3 ± 2.4 years; p = 0.78). The etiology of cBVP was bilateral vestibular neurectomy for treatment of vestibular schwannomas due to neurofibromatosis type 2 in all cases, while iBVP was classified as toxic (gentamicin) (n = 3) or idiopathic (n = 5) (Supplementary Table S1). Educational level did not differ between groups (BVP: 12.4 ± 2.2 years; HC: 12.5 ± 2.4 years; p = 0.93). Cognitive deficits were excluded by Montreal Cognitive Assessment (MOCA) (BVP: 29.3 ± 0.8; HC: 29.2 ± 0.8; p = 0.84).

### Navigation performance of BVP patients and healthy controls

Normalized error rates during real-space navigation (see “Methods”) were higher in BVP patients (29.0 ± 11.2%) than in controls (2.8 ± 5.3%) (p < 0.001). This was due to BVP patients making more errors when recombining novel routes (37.5 ± 16.1% versus 1.7 ± 5.8%) (t = 7.3, p < 0.001). On retraced, i.e., familiar routes, BVP patients and controls performed equally well (5.0 ± 12.4% versus 6.2 ± 11.9%) (t = 0.2, p = 0.82) (Fig. 2). Nevertheless, the navigation performance of the BVP patients was far better than random search (Supplementary Table S2). Gait velocity during steady-state locomotion was the same for BVP patients and controls (1.2 ± 0.1 m/s vs. 1.3 ± 0.1 m/s) (t = 0.4, p = 0.69). Correlation analysis revealed that the normalized error rate for recombined novel routes significantly increased with lower mean caloric response (Rho  = − 0.66, p = 0.03) and lower mean horizontal vHIT gain (Rho = − 0.91, p = 0.01) in iBVP patients (Supplementary Fig. 1). The normalized error rate for all routes was not significantly correlated with caloric response or horizontal vHIT gain.

### Cerebral glucose metabolism during navigation in BVP patients and healthy controls

A direct comparison of rCGM during navigation versus locomotion of HC and BVP patients revealed a relative decrease of rCGM in the right anterior hippocampus and bilateral insular cortex, and a relative increase of rCGM in the posterior parahippocampal cortex and lingual gyrus (i.e., parahippocampal place area, PPA) bilaterally in BVP patients (Fig. 5). Directly comparing the two subgroups cBVP versus iBVP revealed a significantly higher rCGM in the pontine brainstem tegmentum, vestibulocerebellum, anterior thalamus, posterior insular, and retrosplenial cortex in iBVP patients. In contrast, iBVP exhibited lower rCGM in the superior and medial frontal gyrus, the subgenual prefrontal cortex, superior temporal gyrus, and caudate nucleus as compared to cBVP (Fig. 6). Similar results were found, when correlating residual vestibular function by vHIT gain with rCGM across the entire BVP group (Supplementary Fig. S2). Correlation of the percentage use of shortcut with rCGM revealed an increased activation of the pontine brainstem tegmentum, vestibulocerebellum, and right anterior thalamus, as well as a decreased activation of the prefrontal cortex areas with more frequent use of the shortcut route (Supplementary Fig. S3).

## Discussion

The major findings for navigation performance, visual exploration behaviour, search path, and brain activations in patients with BVP were as follows: (1) BVP resulted in a selective impairment of recombining novel routes in a real-space environment, which correlated with the degree of vestibular hypofunction. (2) BVP patients exhibited higher gait fluctuations with a “stop-and-go like locomotion pattern”, spent less time at crossroads and used a possible shortcut route less frequently. (3) Patients showed significantly fewer object fixations and horizontal head movements, when recombining novel routes. (4) The described alterations in navigation performance, search path, visual fixations and horizontal head movements were accompanied by reduced navigation-induced activations in mesiotemporal brain regions such as the hippocampus and entorhinal cortex; in contrast, there were higher activations in posterior mesiotemporal and temporooccipital (posterior parahippocampus and lingual gyrus) brain regions. (5) Navigation-induced brain activations significantly differed in patients with cBVP compared to those with residual vestibular function. iBVP patients exhibited higher activations in the pontine brainstem tegmentum, anterior thalamus, retrosplenial and posterior insular cortex, while cBVP patients showed higher activations in the caudate nucleus, dorsolateral prefrontal cortex and subgenual prefrontal cortex.

### Real-space navigation performance and behaviour in BVP patients

In the current real-space navigation task, we cannot completely exclude an effect of the order of the routes on navigation performance, which means that BVP patients perform worse than HC on the later routes for example due to faster and more prominent tiredness. However, this seems rather unlikely, since BVP patients had no problems with their physical condition during stereotyped hallway locomotion for 20 min on a second date. Furthermore, none of the BVP patients showed signs of or reported relevant tiring or fatigue during or after the navigation task. We also can definitely exclude (verbal) memory deficits as a confounder for the observed navigation deficits in BVP patients, as each patient could recall the five target items after the exploration and navigation phase without problems. In line with this, MOCA screening revealed normal cognitive abilities in each subject within both groups (BVP and HC).

### Navigation-induced brain activation patterns in BVP patients

A complex and widely distributed cerebral network guides specific navigation strategies in humans: the parahippocampal place area (PPA) for landmark processing9,22,36; the posterior parietal cortex (PPC), (pre)frontal cortex, striatum, and cerebellum for distance estimation17,29; the retrosplenial cortex (RSC), precuneus, and anterior dorsal nucleus (ADN) of the thalamus for direction computations via the head direction cell code37,38,39; and the anterior hippocampus, and entorhinal cortex for the creation of a metric cognitive map of the environment by specific cell types, such as place and grid cells40,41,42,43,44,45.

Increased navigation-induced activation of the bilateral PPA in BVP patients might reflect a compensatory mechanism. In consequence of deficient hippocampal place learning, increased visual scene processing and visual analyses of the novel environment seem to be a reasonable strategy to overcome the aforementioned deficits. One might argue that fewer fixations towards objects feasible as landmarks during recombining novel routes in the BVP group contradict this view. However, previous studies showed that the PPA is critically involved in processing novel visual scenes and the selection of visual cues as landmarks59,60. A higher activation of PPA thus would be compatible with a more appropriate selection instead of a higher quantity of landmarks. The incorporation and optimal employment of selected landmarks for precise navigation is not driven by the PPA, but instead by the RSC9,22.

## Conclusions

The current study expands our knowledge of the contribution of the vestibular system to higher cognitive functions. Recombining novel routes were explicitly affected in BVP, which was paralleled by a disintegration of hippocampal (place cell) and brainstem-thalamic (angular head velocity/head direction cell) networks. Future research is warranted to further clarify the long-term functional consequences of higher vestibular network dysfunction for cognitive domains.

## Methods

### Subjects

Fourteen patients with BVP (according to the diagnostic criteria of the Bárány Society)67 and 14 age-matched healthy controls (HC) with normal neurological status and comparable educational levels were included in the study. All subjects underwent a comprehensive neuro-otological examination, including horizontal video head-impulse test (vHIT), caloric cold (32°) and hot (42°) water testing of the horizontal semicircular canals (SCC), ocular and cervical vestibular-evoked myogenic potentials (o/cVEMP), and posturography. Based on these tests, BVP patients were further divided into two subgroups: 6 with complete BVP (cBVP) (no vestibular response from otoliths and SCCs) and 8 with incomplete BVP (iBVP) (with residual vestibular response). Additional hearing loss was documented in the 6 patients with complete BVP (cBVP) due to previous bilateral vestibular neurectomy to treat vestibular schwannomas. Concomitant visual loss (visual acuity < 0.5), or clinically relevant polyneuropathy (vibration sense at the medial malleolus < 5/8) was excluded by clinical neurological examination. Ischemic lesions, microvascular changes (Fazekas > 1), brain atrophy, or other structural brain pathologies were ruled out by structural magnetic resonance imaging (MRI with T2, FLAIR, DWI). Relevant cognitive deficits that might interfere with navigation performance were excluded by Montreal Cognitive Assessment (MOCA).

### Standard protocol approvals and patient consent

Subjects gave their informed, written consent to participate in the study, and for publication of identifying image in an online open-access publication. The protocol was approved by the local Ethics Committee (Ludwig-Maximilians-Universität München) in accordance with the Declaration of Helsinki and the German Federal Office for Radiation Protection.

All participants performed a well-established navigation paradigm in a complex and unfamiliar spatial environment to test their spatial orientation performance (for details see21):

In the exploration phase, the examiner showed the participant the exact location of five different items (pictures of a ball/mushroom/flower/train/house, placed within an outpatient clinic) in a defined sequence. The investigator-guided exploration walk took exactly 10 min for all participants. This was controlled by a stopwatch, which was carried with the investigator. During the exploration phase, the participant followed the investigator to the items in a defined order and along defined routes (start position → ball → mushroom → flower → train → house, and opposite sequence: house → train → flower → mushroom → ball → start position) (Fig. 1). Some items were placed in niches (e.g., ball, flower, house), while others (mushroom, train) were behind doors. The participants had to enter the doors in order to see those items. All items were placed at a height of 1.7 m. The investigator chose a gait velocity of about 1.0 m/s, which allowed controls and patients to reach the items without any discomfort. At the respective target items, a stop of 30 s was included to allow visual exploration. The subjects had to walk a distance of about 330 m to approach all items and return back to the starting point. The participant was instructed to explore the environment intensely, in preparation of the following self-reliant navigation phase.

For the subsequent navigation phase (duration: 10 min), the participants were asked to navigate to the target items by a fully self-determined strategy. They were not motivated to go as fast as possible or as optimal as possible. Subjects were instructed verbally (e.g., “please, go to the ball”) and visually (picture of the ball) to approach the next target item, once they reached the previous one. When a participant did not find the requested target item within a time limit (of 2-times mean duration of the respective route in a previous control sample), the following target item was requested to be approached from the current position. There was no return to the starting position. This strategy was followed consistently in all subjects.

The primary outcome parameter was the error rate in both groups (BVP versus HC). An error was registered, when a participant did not find the requested target item within a time limit (2-times mean duration of the respective route in previous control sample)21, passed by or ignored the target item. This definition was chosen to avoid a random-like search strategy. If a subject used a non-optimal route but reached the requested item within the above-mentioned time limit, this was accepted. Error rates were normalized to a total of 15 routes in our task (normalized error rate = (15 – correctly found items)/15 × 100%). Some participants with very good navigation performance successfully completed more than 15 routes in 10 min, which resulted in negative normalized error rates (e.g., for 16 correctly approached target items: (15–16)/15 × 100% equals a normalized error rate of − 6.7%). Error rates were further separated for retraced familiar and recombined novel routes and normalized to the number of target items expected to be found in 10 min (retraced familiar routes: n = 5, recombined novel routes: n = 10). The percentage use of the shortcut route was defined by the number of actually used shortcuts/number of all possible shortcuts. The parameter travelled distance/optimal path (%) was added as an indirect measure for the use of inefficient routes (Table 1).

### Recording of navigational path and visual exploration behaviour

All participants wore a gaze-in-space measuring device throughout the experiment to document their visual exploration. This consisted of a mobile infrared video-eye-tracking system with goggles, a head-fixed camera, and an inertial measurement unit with a triaxial accelerometer, gyroscope, and magnetometer to record head movements (for details see68). The sampling rate for eye tracking was 220 Hz. A 5th order 21 sample Savitzky-Golay (SG) FIR smoothing filter was applied to preserve high-frequency detail in the signal, while maintaining both temporal and spatial information about local maxima and minima69.

Analysis of saccades and fixations was carried out using MATLAB 2012a (Mathworks, Natick, MA, USA) software based on the established algorithm70. Raw eye movements in the x and y-axes were converted into degrees in the respective axis and displayed as heat maps. The overall distance traversed by gaze was determined as:

$${\text{Distance}} = \sqrt {\left( {x_{t1} - x_{t2} } \right)^{2} + \left( {y_{t1} - y_{t2} } \right)^{2} }$$

As a fixation we defined, when the gaze was directed towards a certain point or object with a duration of more than 100 ms at a velocity and acceleration cut-off of less than 240 deg/s and 3000 deg/s2, respectively. The total frequency of fixations (Hz) was computed quantitatively throughout the whole navigation task. We analyzed overall saccades and fixations for the total task and separately for the retraced familiar and recombined novel routes in a group-wise manner. All saccades and fixations were annotated manually post-hoc to viewed objects by an experienced investigator. In total, 447 viewed separate objects were identified and classified in four major object categories based on their assumed relevance to guide spatial orientation:

• Category 1: non-specific saccades and fixations (e.g., ground floor, bare wall, ceiling)

• Category 2: task-inherent saccades and fixations (e.g., investigator, instructions, target items)

• Category 3: fixations reflecting non-specific general search behaviour (e.g., doors, views into rooms/corners/niches)

• Category 4: object fixations suitable as landmarks for (re)orientation, reflecting specific search behaviour (e.g., distinct fixed pictures, objects and furniture)

By this diligent post-hoc coding of all the object fixations during the exploration and navigation task in each single subject, we were able to depict the average fixation behaviour for both groups (BVP and HC). Therefore, we specifically focused on the object fixations to potential landmarks (category 4, n = 40 objects included). From these objects, we extracted the ones that were fixated at least twice and calculated the mean rate of fixations to single objects on a group level (BVP and HC). The 25 most frequently fixated objects per group were plotted on a ground map of the navigation space as circles (diameter proportionate to the average number of views to the respective object), to depict the spatial distribution of potential landmarks in each group (see Fig. 3a,b). We then analyzed the overlap of object fixations within groups (i.e., BVP-group: exploration versus navigation, HC-group: exploration versus navigation) and between groups (i.e., exploration: BVP- versus HC-group, navigation: BVP- versus HC-group) (see Fig. 3c), to illustrate the selection strategy and recall of landmarks for both groups.

The search path during the navigation task was mapped by accumulating time at a specific place and was analyzed quantitatively for mean gait speed, use of shortcuts, and time spent at crossroads. Gait velocity was calculated in each subject based on the post-hoc coding of the subject position in time. By means of the magnetometer, horizontal head movement velocity (deg/s) was depicted separately throughout the whole task and further differentiated for all the retraced familiar and recombined novel routes.

### [18F]-FDG PET imaging of navigation- and locomotion-induced brain activations

To investigate the brain activation pattern during navigation, BVP patients and HC were examined by [18F]-FDG-PET following a previously established protocol (for details, see72,73): [18F]-FDG was injected at the start of the 10-min navigation phase. Afterwards subjects rested in a supine position for 20 min and image acquisition started 30 min after tracer administration (Fig. 1c). Each subject was scanned while in a fasting state > 6 h (checked by means of blood glucose concentration). This paradigm was chosen because the cerebral glucose utilisation is weighted to the 10 min following [18F]-FDG injection and is integrative due to intracellular trapping of the tracer74. It therefore allows an estimation of neuronal activation specific to the task performed in a time period of 10 min immediately after [18F]-FDG injection. A second PET scan was acquired 4 weeks later during hallway locomotion (control condition in a different spatial layout without spatial orientation). Image acquisition was again started 30 min post-injection on an ECAT EXACT HR+ PET scanner (Siemens/CTI, Knoxville, TN, USA). The scanner acquires 63 contiguous transaxial planes, simultaneously covering 15.5 cm of an axial field of view. The transaxial and axial resolutions (full width at half maximum) of the PET system were 4.6 mm and 4.0 mm, respectively, at the centre and 4.8 mm and 5.4 mm, respectively, at a radial offset of 10 cm. The patient’s head was secured to a foam cushion and adequately positioned in the gantry. The emission recording consisted of three frames (10 min per frame, 3-D acquisition) covering the period from 30 to 60 min post injection, after which a transmission scan was obtained using a rotating [68Ge] point-source. For further evaluation, the frames were added to a single frame (30 min acquisition). Images were reconstructed as 128 × 128 matrices of 2 × 2 mm voxels by filtered back-projection using a Hann filter with a cut-off frequency of 0.5 Nyquist and corrected for random, dead time, scatter, and attenuation. The reconstructed [18F]-FDG images were transformed to NIfTI format for further processing.

### Analysis of [18F]-FDG PET data acquired during navigation and locomotion

Data processing and statistical analysis were performed using SPM8 software (Wellcome Department of Cognitive Neurology, London) implemented in MATLAB 2012a following an established protocol21,72. All the reconstructed [18F]-FDG-PET images were linearly co-registered to the corresponding MRI using automated SPM8 algorithms. Anatomical brain MRIs were spatially normalised into the MNI standard template (McGill University, Montreal QC, Canada) using an affine transformation (12 parameters for rigid transformations), whose parameters were applied to the co-registered [18F]-FDG-PET images. Then the spatially normalised images were blurred with a Gaussian filter (FWHM 12 mm) to account for regional inter-subject variability. All scans were analyzed after normalisation to the white matter72,73. The normalisation prior to voxel-based statistics was done with an anatomical mask (centrum semiovale) in MNI space, to remove the effects of different overall counts. Images of the spatial orientation paradigms were compared voxel-wise with those of the control condition (paired t-test) and between groups (unpaired t-test). Correlation analyses of regional cerebral glucose metabolism (rCGM) were done with vestibular function tests (mean vHIT gain, mean caloric response), normalized error rates, % use of the shortcut route, duration at crossroads, and object-specific fixation frequency on recombined novel routes. rCGM increases and decreases were considered significant for a p < 0.005, similar to previous studies21.

### Statistical analysis

Behavioural and navigational measurements of the three groups (cBVP, iBVP, HC) were analyzed using SPSS 24 (IBM, Armonk New York). Independent sample t-tests were used to compare the demographic data, navigation performance, visual search parameters, search path (i.e., duration at crossroads, use of shortcut route), head movements, and gait speed between BVP patients and HC. Kolmogorov–Smirnov was applied to guarantee the normal distribution of all the parameters analyzed by the independent sample t-test. To correct for multiple testing of all the different eye movement (i.e., fixations and saccades) and head movements parameters from Table 1, Bonferroni-correction was applied post hoc. P-values < 0.05 were considered statistically significant. Spearman’s rank correlation was analyzed for normalized error rates during navigation against the degree of vestibular impairment (quantified by mean vHIT gain, mean caloric response) and considered significant for Rho > 0.5 and p < 0.05.