https://orcid.org/0000-0002-2214-3788
Figures
Abstract
Path integration changes may precede a clinical presentation of Alzheimer’s disease by several years. Studies to date have focused on how spatial cell changes affect path integration in preclinical AD. However, vestibular input is also critical for intact path integration. Here, we developed the vestibular rotation task that requires individuals to manually point an iPad device in the direction of their starting point following rotational movement, without any visual cues. Vestibular features were derived from the sensor data using feature selection. Machine learning models illustrate that the vestibular features accurately classified Apolipoprotein E ε3ε4 carriers and ε3ε3 carrier controls (mean age 62.7 years), with 65% to 79% accuracy depending on task trial. All machine learning models produced a similar classification accuracy. Our results demonstrate the cross-sectional role of the vestibular system in Alzheimer’s disease risk carriers. Future investigations should examine if vestibular functions explain individual phenotypic heterogeneity in path integration among Alzheimer’s disease risk carriers.
Citation: Coughlan G, Plumb W, Zhukovsky P, Aung MH, Hornberger M (2023) Vestibular contribution to path integration deficits in ‘at-genetic-risk’ for Alzheimer’s disease. PLoS ONE 18(1): e0278239. https://doi.org/10.1371/journal.pone.0278239
Editor: Seralynne Vann, Cardiff University, UNITED KINGDOM
Received: May 18, 2022; Accepted: November 14, 2022; Published: January 3, 2023
Copyright: © 2023 Coughlan et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: The raw and pre-processed data, demographic participant data, and the code for the machine learning algorithms are available at: https://github.com/WillPlumb/Vestibular-PreAD.
Funding: GC and MH were supported by the Faculty of Medical at the University of East Anglia. GC was also funded by the Alzheimer Society of Canada (GC, 22-08). MH was also supported by Alzheimer’s Research United Kingdom. PZ was supported by the Canadian Institutes of Health Research postdoctoral fellowship. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: No competing interests.
Abbreviations: PI, Path Integration; APOE, Apolipoprotein E; RF, Random Forest; SVM, Support Vector Machine; MLP, Multi-Layer Perception
Introduction
Alzheimer’s disease (AD) is a common neurodegenerative condition afflicting one’s ability to update self-motion information. This process, known as path integration, is partially mediated by spatial cells in the entorhinal cortex and hippocampus [1], both of which are altered by AD pathophysiology in the preclinical stage [2–4]. A major challenge at this stage, however, is to understand the mechanisms underlying path integration impairment in preclinical AD. Understanding modifiable factors that influence variability in path integration will help pinpoint pre-disease pathological changes and in turn guide preclinical treatment targets.
Beyond demographic factors such as age and sex [5], vestibular function is a predictor of declining PI in animal models, with substantial evidence that the vestibular system transmits specific types of information about self-motion to structures such as the hippocampus and entorhinal cortex [6–11]. Retrospective epidemiological studies in humans also show that vestibular dysfunction originating from the otolith and semicircular canals is associated with cognitive deficits in the elderly, including spatial deficits [12, 13]. Length of time with vestibular dysfunction may contribute to impairment severity [14]. Crucially, experimental studies in humans show that vestibular signals provide robust signals for path integration, suggesting that vestibular signals may contribute to impaired self-motion perception in preclinical AD, particularly given epidemiological research showing that vestibular dysfunction is three times more common in AD patients compared to age-matched controls [15, 16].
Vestibular dysfunction and its precipitating factors (such as hearing loss; see recent Review Smith, 2021, vestibular migrane and vertigo) are at least partially modifiable [14, 17, 18]. Thus, identifying and treating early vestibular dysfunction may serve a preventative role in AD. Building on the work of Mittelstaedt and colleagues [20], we developed a behavioural vestibular paradigm that is based on hand movements. Given the high-dimensional nature of this movement data, we applied machine learning algorithms to analyse patterns of movement, which we term ‘vestibular features’. The task was designed to test both path integration and vestibular function in mid-life adults.
The task was administered to apolipoprotein (APOE) ε4 allele carriers, of whom almost half develop Alzheimer’s disease by the average age of 76, compared to only 20% of non-carriers who convert to Alzheimer’s disease by the average age of 84 [19]. Using machine learning, we aimed to (i) determine whether vestibular features classify ε4 carriers from non-carrier controls with and without the inclusion of demographic information and (ii) determine whether vestibular features classify ε4 carriers from non-carrier controls with and without the inclusion of PI accuracy. We hypothesised that each machine learning model would distinguish at-genetic risk APOE ε4 carriers from non ε4 carriers. Given the subtle nature of cognitive deficits in mid-life adults, we expected that classification accuracy would increase with each consecutive task trial. As machine learning techniques are not commonly applied to cognitive data in preclinical AD research, we applied three machine learning algorithms to explore if each algorithm consistently produced a similar classification accuracy.
Materials and methods
Participants and procedure
One hundred and fifty participants between 50 and 75 years of age were recruited to participate in a research study at the University of East Anglia, Norwich, UK. Written consent was obtained from all participants and ethical approval was obtained from Faculty of Medicine and Health Sciences Ethics Committee at the University of East Anglia, Reference FMH/2016/2017–11. Participants with a history of psychiatric or neurological disease, substance use disorder or motor control disorder were excluded. Participants receiving anti-depression or anti-anxiety medication at the time of screening were excluded. Saliva kits were sent to participants’ home and returned to the university on the same day the saliva sample was taken to determine APOE genotype status. Sensor data were collected on the iPad-based assessment tool (see The Vestibular Rotation Task for details), These data were only collected during the follow-up visit of the study, 18-months after the baseline assessments that are published. As just 25% of the population carry an APOE ε4 allele (23% APOE ε3ε4, 2% APOE ε4ε4 [19], 31 ε3ε4 carriers were detected. We selected a subset of ε3ε3 carriers that form the majority of the population (75%) to match the ε3ε4 group for age and sex. Ten ε3ε4 did not complete task at the follow-up timepoint due to study attrition and technological problems, leaving the final sample size at 53 participants (32 ε3ε3 carriers and 21 ε3ε4 carriers). We did not include a third genetic subgroup of homozygous APOE ε4 carriers, because they were too rare (n = 3), although their scores are reported in S1 File. APOE ε2 carriers (15% of the UK population) were also excluded as it is unclear how the ε2 allele acts on cognitive performance or the further development of AD. The sample size required for the present study was determined based on similar studies [20–23].
APOE genotyping
A saliva sample was collected for direct genotyping of APOE. See [24] for further details.
Vestibular rotation task
To isolate the input of vestibular signals to performance, the task was administered in the complete absence of external visual or auditory cues. Given the subtle nature of cognitive deficits in mid-life adults, the complexity of the task increased in each consecutive task trial (rotations are detailed in Table 1). During the task: 1) Participants sat in the rotating chair (feet not touching the floor) and held an iPad flat in their hands (Fig 1A). 2) The participant was blindfolded and given earplugs on to ensure the test tapped into the vestibular system with no external stimuli. 3) Participants were told to remember the object in front of them (i.e. door) as their reference point. 4) The x, y, z co-ordinates of the reference point were recorded on the iPad. 5) During the trials, the examiner rotated the participant in the chair (see purple arrows in Fig 1B and 1C) and always remained behind them to avoid serving as a location cue. 6) Three seconds following the turn completion, participants were asked to point the iPad as accurately as possible in the direction of the reference point, while still wearing their blindfold and earplugs (see green arrows in Fig 1B and 1C). 7) During the pointing back movement response, the iPad sensors recorded accelerometer, gyroscopic, and compass information along the x-axis (forward/backwards motion) y axis (left/right motion) and z axis (up/down motion) (AppleInc 2021). 8) The distance between the reference point and participant response (i.e., the end error) was recorded and used as a proxy measure of path integration, similar to previous studies [20, 21].
Vestibular features
The sensor data that was derived from accelerometer, gyroscope, and compass underwent pre-processing steps that are detailed in the S1 File. Following the data preparation phase, feature selection was used to generate latent vestibular features from the movement data, which in turn served as proxy measures of primary vestibular graviceptors [25]. This process was guided by theoretical knowledge on path integration, the vestibular system and cognitive impairment [24, 26–29]. Vestibular features are listed in Table 2.
Statistical analysis
Simple two-tailed t-tests were used to test the significance of any demographic or neuropsychological differences between the genetic groups (ε3ε3 and ε3ε4 carriers).
Machine learning algorithms
To determine whether vestibular features could distinguish between the genetic groups, classification models were created for each task trial. Each trial involves a different rotational movement, and it is not possible to combine all the trials. Classification accuracy of three machine learning algorithms, including Random Forest (RF) [30], Support Vector Machines (SVM) [31] and Multi-Layer Perceptron (MLP) [32] were computed. Based on the novelty of the task our choice of model variants was based on standard models that have shown to be well generalized. In the case of RF, we applied the generic approach based on [33]. Similarly, we used a standard SVM approach and examined linear and radial basis function (RBF) kernels. Linear was used as a baseline approach and radial basis function was used to capture the potential complex variation in movement features. Finally, MLP was applied as a classical Neural Network approach to classification. Using a grid search, we evaluated performance with different model parameters. To moderate against over fitting we also applied a grid search over regularization parameter represented as C in the SVM model and α in MLP model. C parameter within the SVM provides the degree of flexibility a decision boundary can have while classifying training points. Low C identifies large margins separating groups. A higher C ensures more points are classified correctly with more complex boundaries. We evaluated performance over C values of 0.5, 1, 3, 5, 10 and 20. Similarly, α combats overfitting by restricting the complexity of decision boundaries. A lower α creates a stricter boundary and searches over α values of 0.0001, 0.0005, 0.001 and 0.002.This array of standard models is used in this first experimentation to minimise assumptions and to understand which basic method is most effective in each consecutive trial. Classification is based on a standard class metric, the F1 score. The F1 score is a measure of the model’s performance that considers precision (true positives/true positives + false positives) and recall (true positives/true positives + false negatives):
Thus, the F1 score determined the classification performance for APOE status (i.e., ε3ε3 vs ε3ε4). A random classification F1 score is equal to 0.57, with a score of above 0.5 suggesting improved efficacy/sensitivity. Accuracy scores were also calculated as a secondary metric. Accuracy represents the percentage of correctly classified ε3ε3 and ε3ε4 participants. Two sets of analyses were performed: in the first set of models, all features were included (blue line in Fig 2A). In the second set of models, end error was excluded to assess the independent precision of the movement-based features to the prediction of genetic risk (red line in Fig 2A). An overview of the machine learning algorithms is provided in S1 File.
A F1 scores for the best performing algorithm are shown. A random predictor would score 0.57, with a score of above 0.57 representing better-than-chance APOE status classification performance. Blue line includes all features. Red line excludes the path integration feature, end error. B Importance scores are represented by the circle diameter and were derived for the best performing model on each of the trials shown. Scores vary between 0 and 1 depending on the proportion of influence the feature has for that trial. RF = Random Forest, SVM = Support Vector Machine, MLP = Multi-Layer Perception.
Cross-validation
A control validation resampling procedure evaluated the machine learning models [34]. Five-fold validation was used using four groups as training data and one group for testing. We selected these splits such that each participant was included in either the training or test dataset. Five F1 scores for each algorithm and for each trial were computed and averaged. Each of five groups are created using stratified folds to preserve the samples in both APOE groups.
Results
Demographic characteristic
Demographic and neuropsychology characteristics were not significantly different between genetic groups (Table 3). Secondary characteristics are presented in S1 Table in S1 File.
Vestibular features distinguish ε4 carriers from non-carrier controls
Next, we sought to examine if the vestibular features distinguished the genetic groups by applying a machine learning approach. The three chosen algorithms examined the classification performance of the vestibular features for APOE genotype status. The vestibular features are listed in Table 2. Six out of the nine trials achieved a cross-fold average F1 score of above 0.6 using one or more of the algorithms, suggesting stable differences in vestibular function among APOE ε3ε3 and ε3ε4 carriers. Accuracy percentages ranged from 65% to 79% depending on task trial and algorithm. Excluding age, sex and occupation data produced similar F1 scores, suggesting that vestibular function classified APOE status irrespective of demographic variation (see S2 Table in S1 File). Please refer to S3 Table in S1 File for results including a small number of highest-genetic-risk ε4ε4 carriers who were included in the APOE ε4 carrier group, subsequently producing greater F1 scores (reaching 0.77) across all trials. We then examined the influence of the vestibular function when the path integration proxy ‘end error’ was excluded. Excluding this feature led to a best prediction accuracy of 0.75, suggesting that the movement-based features alone maintained good classification performance for APOE status. Accuracy ranged from 65%-75%. The highest performing algorithm for each trial including and excluding path integration is presented in Fig 2A. F1 scores and accuracy scores across all trials and the best performing algorithm are presented in Table 4.
Identifying the most influential vestibular features
Features with the best classification precision and accuracy varied across task trials (Fig 2B). Demographics are included as a means of comparison.
Discussion
Our results show that a novel movement-based vestibular task can classify vestibular changes in at-genetic-risk of AD, irrespective of demographic background. Machine learning algorithms achieved good performance (F1 and accuracy scores of up to 0.72 and 0.75, respectively) based on the derived vestibular features of the task. We also replicated previous path integration deficits in APOE ε4 gene carriers when task novelty was high [20, 24]. Our findings highlight the need for a broader computational perspective to understand the at-genetic-risk AD phenotype.
In more detail, we developed the Vestibular Rotation Task and applied a feature selection approach to reduce the compass, accelerometer and gyroscope data into meaningful vestibular features that characterise the human vestibular system, as well as basic path integration performance. The features engineered included: ‘end error‘ (the basic path integration performance measure), angular displacement, total tilt, change in tilt, acceleration, hesitations, and average jerk. Three machine algorithms tested the classification accuracy of the features to detect APOE ε4 carriership status, achieving a prediction accuracy of up to 75%. After removing the PI accuracy measure, ‘end error’, the best prediction accuracy reduced by just 3% (to 72% accuracy), suggesting that the motion features distinguished APOE ε4 carriers from non-carriers.
Dysregulation of the vestibular system is associated with path integration deficits
Evidence from animal models has consistently demonstrated the importance of the vestibular system to path integration. Vestibular lesions impair spatial memory and the ability to rerun to a goal location following passive transport [35]. This effect is exacerbated in darkness [36]. Interestingly, rodents with vestibular lesions path integrate successfully with the aid of external visual cues [37]. This has striking similarities to Bierbrauer and colleagues who found that a path integration deficit in APOE ε4 carriers emerges only if external visual cues are not available [20]. These findings are also consistent with limited studies that show path integration impairment in vestibular deficient humans emerges when external cues are not available [38], supporting the emerging theory that vestibular function plays a mechanistic role in path integration deficits previously observed in adults at-genetic-risk AD.
Clinical implications
Vestibular signals that influence path integration in preclinical AD may help pinpoint pre-disease pathological changes and in turn guide treatment. Identifying vestibular contributions to the cognitive phenotype of preclinical AD is important because vestibular dysfunction is often present with treatable hearing loss [39], recently cited as a modifiable risk factor for AD [40]. Moreover, vestibular balance training such as Intensive Slackline-Training improves path integration and vestibular function and implanted vestibular prosthesis (that reproduces information normally provided by semi-circular canals) improves spatial orientation in monkeys with severe vestibular damage [33], suggesting adults with vestibular dysfunction (and path integration impairments), may respond to a vestibular implant and or vestibular intensive training. Moreover, because the vestibular system has extensive connections to AD vulnerable brain regions including the hippocampus, cingulate cortex and parietal lobe, vestibular stimulation may indeed improve cognitive performance related to the integrity of these brain regions, including disorientation and memory loss. Our approach to investigating vestibular contributions should be considered a steppingstone towards more tailored treatment programs for preclinical AD, that can be combined with pharmacological, non-pharmacological sensory stimulation, gamma-induction, and dietary treatment strategies [41, 42].
Limitations, future research directions and conclusion
Vestibular contributions to the cognitive phenotype of at-genetic-risk AD warrant further investigation. Our study included an APOE genotyped sample that was a similar sample size to previous studies [21, 24]. Nonetheless, this sample is moderate in the context of machine learning approaches. A further challenge when applying supervised machine learning methodology is the correct creation and selection of movement features. As more research is conducted into the vestibular system in the context of cognitive impairment, additional movement-based features can be created and used as predictors. To prevent the high dimensionality of features, a naïve correlation feature selection was applied to reduce the number of input dimensions. Moreover, signals from the device can contain noise and unwanted artifacts. We implemented a bespoke code to pre-process to counteract this. Simple heuristics applied to the raw signal included flipping movement when a participant held the iPad rotated 180° and applying a threshold of 80° for sequential compass values. To extract potential hesitations, the raw data was filtered by a moving window averaged over 100 data points and smoothing the raw signal such that we could implement a peak detection for hesitations. From a theoretical perspective, further research is needed to understand how degraded graviception contributes to deficits in self-motion perception and the role played by head direction and grid cell dysfunction in graviception.
In conclusion, we introduce a novel movement-based task of human vestibular function and path integration. The application of machine learning revealed movement-based vestibular changes in at-genetic-risk adults, which should be a priority for basic scientific research in AD. Our findings may accelerate objective, high frequency and passive digital phenotyping of at-genetic-risk AD and help elucidate the mechanisms by which the human vestibular system contributes to cognitive impairment in preclinical AD. Future studies may examine whether grid cell function, detectable in task-based fMRI, is causally linked to vestibular dysfunction in at-genetic-risk AD and secondly, whether vestibular changes precede, or succeed, the emergence of tauopathy in the brainstem and entorhinal cortex during the preclinical stage of AD.
Acknowledgments
We thank Tom Percy for programming the sensor task and the many participants who gave their time to this research.
References
- 1. Hafting T, Fyhn M, Molden S, Moser M-B, Moser EI. Microstructure of a spatial map in the entorhinal cortex. Nature. 2005;436: 801–806. pmid:15965463
- 2. Gomez-Isla T. Neurond Loss Correlates with but Exceeds Newofibriiary T a n g l e s in Alzheimer’s Disease. Ann Neurol. 1997;41: 17–24.
- 3. Kunz L, Schröder TN, Lee H, Montag C, Lachmann B, Sariyska R, et al. Reduced grid-cell-like representations in adults at genetic risk for Alzheimer’s disease supplementary materials. Science (1979). 2015;350: 430–433. pmid:26494756
- 4. Braak H, del Tredici K. The preclinical phase of the pathological process underlying sporadic Alzheimer’s disease. Brain. 2015;138: 2814–2833. pmid:26283673
- 5. Smith PF, Agrawa Y, Darlington CL. Sexual dimorphism in vestibular function and dysfunction. J Neurophysiol. 2019;121: 2379–2391. pmid:31042453
- 6. Cheng Z, Gu Y. Vestibular system and self-motion. Frontiers in Cellular Neuroscience. Frontiers Media S.A.; 2018. p. 456. pmid:30524247
- 7. Brandt T, Schautzer F, Hamilton DA, Brüning R, Markowitsch HJ, Kalla R, et al. Vestibular loss causes hippocampal atrophy and impaired spatial memory in humans. Brain. 2005;128: 2732–2741. pmid:16141283
- 8. Cohen HS, Sangi-Haghpeykar H. Walking speed and vestibular disorders in a path integration task. Gait Posture. 2011;33: 211–213. pmid:21131202
- 9. Wallace DG, Hines DJ, Pellis SM, Whishaw IQ. Vestibular information is required for dead reckoning in the rat. Journal of Neuroscience. 2002;22: 10009–10017. pmid:12427858
- 10. Jacob PY, Poucet B, Liberge M, Save E, Sargolini F. Vestibular control of entorhinal cortex activity in spatial navigation. Front Integr Neurosci. 2014;8: 1–9. pmid:24926239
- 11. Kamil R. Vestibular function and Hippocampal Volume in the Baltimore Longitudinal Study of Aging (BLSA). Otology and Neuotology. 2018. pmid:29889787
- 12. Semenov YR, Bigelow RT, Xue QL, Lac S Du, Agrawal Y. Association between Vestibular and Cognitive Function in U.S. Adults: Data from the National Health and Nutrition Examination Survey. Journals of Gerontology—Series A Biological Sciences and Medical Sciences. 2016;71: 243–250. pmid:26219850
- 13. Bigelow RT, Semenov YR, Du Lac S, Hoffman HJ, Agrawal Y. Vestibular vertigo and comorbid cognitive and psychiatric impairment: The 2008 National Health Interview Survey. J Neurol Neurosurg Psychiatry. 2016;87: 367–372. pmid:25886779
- 14. Smith PF. Hearing loss versus vestibular loss as contributors to cognitive dysfunction. J Neurol. 2021. pmid:33387012
- 15. Harun A, Oh ES, Bigelow RT, Studenski S, Agrawal Y. Vestibular Impairment in Dementia. Otology and Neurotology. 2016. pmid:27466890
- 16. Previc FH. Vestibular loss as a contributor to Alzheimer’s disease. Med Hypotheses. 2013;80: 360–367. pmid:23375669
- 17. Dordevic M, Gruber J, Schmitt FC, Mueller N. Impairments in path integration, rotational memory and balancing in patients with temporal lobe epilepsy. BMJ Neurol Open. 2020;2: e000077. pmid:33681800
- 18. Karmali F, Haburcakova C, Gong W, della Santina CC, Merfeld DM, Lewis RF. An implanted vestibular prosthesis improves spatial orientation in animals with severe vestibular damage. Journal of Neuroscience. 2021;41: 3879–3888. pmid:33731447
- 19. Liu C-C, Kanekiyo T, Xu H, Bu G. Apolipoprotein E and Alzheimer disease: risk, mechanisms and therapy. Nat Rev Neurol. 2013;9: 106–118. pmid:23296339
- 20. Bierbrauer A, Kunz L, Gomes CA, Luhmann M, Deuker L, Getzmann S, et al. Unmasking selective path integration deficits in Alzheimer’s disease risk carriers. Sci Adv. 2020;6: eaba1394. pmid:32923622
- 21. Kunz L, Schröder TN, Lee H, Montag C, Lachmann B, Sariyska R, et al. Reduced grid-cell-like representations in adults at genetic risk for Alzheimer’s disease. Science (1979). 2015;350: 430–433. pmid:26494756
- 22. Zimmermann JF, Butler CR. Accelerated long-term forgetting in asymptomatic APOE ε4 carriers. Lancet Neurol. 2018;17: 394–395.
- 23. Coughlan G, Zhukovsky P, Puthusseryppady V, Gillings R, Minihane M, Cameron D, et al. Functional connectivity between the entorhinal and posterior cingulate cortices associated with navigation impairment following path integration in at-genetic-risk Alzheimer’s disease. Neurobiol Aging. 2020.
- 24. Coughlan G, Coutrot A, Khondoker M, Minihane AM, Spiers H, Hornberger M. Toward personalized cognitive diagnostics of at-genetic-risk Alzheimer’s disease. Proc Natl Acad Sci U S A. 2019;116: 9285–9292. pmid:31015296
- 25. Hall MA. Correlation-based Feature Selection for Machine Learning. 1999.
- 26. Stangl M, Kanitscheider I, Riemer M, Fiete I, Wolbers T. Sources of path integration error in young and aging humans. Nat Commun. 2020;11: 1–15. pmid:32457293
- 27.
Segen V, Ying J, Morgan E, Brandon M, Wolbers T. Path integration in normal aging and Alzheimer’s disease. Trends in Cognitive Sciences. Elsevier Ltd; 2022. pp. 142–158. https://doi.org/10.1016/j.tics.2021.11.001 pmid:34872838
- 28. Cullen KE, Taube JS. Our sense of direction: Progress, controversies and challenges. Nat Neurosci. 2017;20: 1465–1473. pmid:29073639
- 29. Smith PF. Hearing loss versus vestibular loss as contributors to cognitive dysfunction. J Neurol. 2021. pmid:33387012
- 30. Breiman L. Random Forests. Mach Learn. 2001;45: 50.
- 31.
Evgeniou T, Pontil M. Support vector machines: Theory and applications. Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). Springer Verlag; 2001. pp. 249–257. https://doi.org/10.1007/3-540-44673-7_12
- 32. Murtagh F. Multilayer perceptrons and regression. 1991.
- 33. Breiman L. Random Forests. 2001.
- 34.
Berrar D. Cross-validation. Encyclopedia of Bioinformatics and Computational Biology: ABC of Bioinformatics. Elsevier; 2018. pp. 542–545. https://doi.org/10.1016/B978-0-12-809633-8.20349-X
- 35. Miller S, Potegal M, Abraham L. Vestibular involvement in a passive transport and return task. Physiological Psychology. 1983;11: 1–10.
- 36. Baek JH, Zheng Y, Darlington CL, Smith PF. Evidence that spatial memory deficits following bilateral vestibular deafferentation in rats are probably permanent. Neurobiol Learn Mem. 2010;94: 402–413. pmid:20736074
- 37. Stackman RW, Herbert AM. Rats with lesions of the vestibular system require a visual landmark for spatial navigation. Behavioural Brain Research. 2002;128: 27–40. pmid:11755687
- 38. Heimbrand S, Müller M, Schweigart G, Mergner T. Perception of horizontal head and trunk rotation in patients with loss of vestibular functions. Journal of Vestibular Research. 1990;1: 291–298. pmid:1670161
- 39. Bosmans J, Jorissen C, Cras P, van Ombergen A, Engelborghs S, Gilles A, et al. Impact of hearing loss and vestibular decline on cognition in Alzheimer’s disease: A prospective longitudinal study protocol (Gehoor, Evenwicht en Cognitie, GECkO). BMJ Open. 2020;10: 1–9. pmid:32948575
- 40. Livingston G, Huntley J, Sommerlad A, Ames D, Ballard C, Banerjee S, et al. The Lancet Commissions Dementia prevention, intervention, and care: 2020 report of the Lancet Commission. The Lancet. 2020;6736. pmid:32738937
- 41. Lee J, Ryu S, Kim HJ, Jung J, Lee B, Kim T. 40 Hz acoustic stimulation decreases amyloid beta and modulates brain rhythms in a mouse model of Alzheimer’s disease. bioRxiv. 2018.
- 42. Etter G, van der Veldt S, Manseau F, Zarrinkoub I, Trillaud-Doppia E, Williams S. Optogenetic gamma stimulation rescues memory impairments in an Alzheimer’s disease mouse model. Nat Commun. 2019;10: 1–11. pmid:31757962
- 43. Angelaki DE, Klier EM, Snyder LH. A Vestibular Sensation: Probabilistic Approaches to Spatial Perception. Neuron. 2009;64: 448–461. pmid:19945388
- 44. de Winkel KN, Soyka F, Bülthoff HH. The role of acceleration and jerk in perception of above-threshold surge motion. Exp Brain Res. 2020;238: 699–711. pmid:32060563