Reduced Vestibular Function is Associated With Cortical Surface Shape Changes in the Frontal Cortex

ABSTRACT

Aging‐associated decline in peripheral vestibular function is linked to deficits in behaviors and cognitive abilities that are known to rely on the sensorimotor and frontal cortices, but the precise neural pathways are unknown. To fill this knowledge gap, this cross‐sectional study investigates the relationship between age‐related variation in vestibular function and surface shape alterations of the frontal and sensorimotor cortices, considering age, intracranial volume, and sex. Data from 117 older adults (aged 60+) from the Baltimore Longitudinal Study of Aging, who underwent end‐organ‐specific vestibular tests (cVEMP for the saccule, oVEMP for the utricle, and vHIT for the horizontal canal) and T1‐weighted MRI scans on the same visit, were analyzed. We examined a subset of 10 frontal and sensorimotor brain structures in the broader, distributed vestibular network: the middle‐superior part of the prefrontal cortex (SFG_PFC), frontal pole (SFG_pole), and posterior pars of the superior frontal gyrus (SFG), the dorsal prefrontal cortex and posterior pars of middle frontal gyrus (MFG_DPFC, MFG), the pars opercularis, pars triangularis, and pars orbitalis of the inferior frontal gyrus, as well as the precentral gyrus and postcentral gyrus (PoCG) of the sensorimotor cortex. For each region of interest (ROI), shape descriptors were estimated as local compressions and expansions of the population average ROI surface using Large Deformation Diffeomorphic Metric Mapping (LDDMM) surface registration. Shape descriptors were linearly regressed onto standardized vestibular variables, age, intracranial volume, sex, and in follow‐up analyses, multisensory function (hearing, vision, proprioception). We found that lower utricular function was linked with surface compression in the left MFG and expansion in the bilateral SFG_pole and left SFG. Reduced canal function was associated with surface compression in the right SFG_PFC and SFG_pole and left SFG. Both reduced saccular and utricular function correlated with surface compression in the posterior medial part of the left MFG. Our findings illuminate the complexity of the relationship between vestibular end‐organ function and the focal morphology in aging in areas of the frontal and sensorimotor cortices relevant to executive ability, motor planning, and self‐motion perception. An improved understanding of these pathways could help in developing interventions to enhance the quality of life in aging and populations with cognitive impairment.

We identified significant cortical surface shape changes in focal areas of the prefrontal cortex that correlate with age‐related vestibular dysfunction in 117 older adults from the Baltimore Longitudinal Study of Aging. These findings provide important insights into mechanisms linking vestibular decline, cognitive impairment, and fall risk in aging populations.

1. Introduction

Vestibular structure and function are known to decline with aging (Baloh and Honrubia 2001; Colebatch et al. 2013; Engström et al. 1974; Johnsson and Hawkins 1972; Rauch et al. 2001; Richter 1980; Rosenhall 1973) and may play a role in balance and cognitive phenotypes in aging and disease. The peripheral vestibular system, comprised of the saccule, the utricle, and the three semicircular canals, sends information about self‐motion relative to gravity to a widespread network of multi‐sensorimotor brain regions (Conrad et al. 2014; de Waele et al. 2001; Hitier et al. 2014; Lopez et al. 2012; Lopez and Blanke 2011; Ruehl et al. 2022; Zu Eulenburg et al. 2011). In older adults, age‐related vestibular loss is not only related to deficits in posture and balance, but also higher‐order behaviors, such as attention, visuospatial cognitive ability, executive ability, memory, self‐motion perception, and motor planning and execution (Anson, Pineault, et al. 2019; Bigelow and Agrawal 2015; Cullen 2014; Paul F. Smith 2019; Roditi and Crane 2012; Yoder and Taube 2014). Additionally, vestibular dysfunction has been linked with neurodegenerative diseases that impact these functions, such as dementia, multiple sclerosis, Huntington's disease, and Parkinson's disease (Berkiten et al. 2023; Biju et al. 2022; Bohnen et al. 2022; Carpinelli et al. 2021; Chastan et al. 2019; Cochrane et al. 2021; Cui et al. 2022; Harun et al. 2016; Park et al. 2022; Park and Kang 2021; Paul F. Smith 2018; Rüb et al. 2014; Wei et al. 2018; Wei et al. 2019). For all our expanding knowledge of the relationship between age‐related vestibular loss and both higher‐order behaviors and neurodegenerative diseases, significant gaps exist in our understanding of the involved neuroanatomical circuits.

While functional neuroimaging studies consistently implicate the brainstem, cerebellum, parieto‐insular vestibular cortex, and numerous other subcortical and cortical multisensory processing regions in vestibular processing (de Waele et al. 2001; Hitier et al. 2014; Lopez et al. 2012; Lopez and Blanke 2011; Zu Eulenburg et al. 2011), mounting research shows that portions of the frontal cortex and the precentral/postcentral gyri also respond to vestibular stimulation and contribute to higher‐order behaviors (e.g., executive control, motor planning, self‐motion perception) (Ferrè and Haggard 2020). Specifically, functional neuroimaging data suggest that the frontopolar cortex, dorsolateral prefrontal cortex, inferior frontal gyrus (IFG), frontal/supplemental eye fields, supplementary motor area, premotor cortex, and primary sensorimotor areas receive vestibular input via independent ascending thalamic projections (de Waele et al. 2001; Hitier et al. 2014; Lopez and Blanke 2011). These areas, sometimes referred to as part of the “vestibular cognitive/sensorimotor network,” are thought to integrate body‐ and motion‐related cues to support executive abilities, motor planning, attention, and spatial orientation (Ferrè and Haggard 2020). Despite the behavioral relevance of these regions, few studies have investigated whether structural variations in these regions can be tied to vestibular end‐organ function.

Although several studies have reported gray matter alterations in frontal, sensorimotor, and parietal regions in individuals with vestibular deficits (Hong et al. 2014; Hüfner et al. 2009; Padova et al. 2024; Wurthmann et al. 2017), most analyses rely on voxel‐based morphometry or gross volumetric measures. Such methods can miss localized expansions or compressions that might better capture subtle, spatially complex cortical alterations (Davatzikos 2004). Furthermore, many prior studies have treated vestibular function as a single composite measure, like those derived from balance measures, rather than assessing end‐organ‐specific associations (e.g., saccule vs. utricle vs. semicircular canals) (Hupfeld, McGregor, et al. 2022). Moreover, few studies have controlled for concurrent age‐related declines in other sensory modalities known to integrate with vestibular inputs, such as hearing, vision, or proprioception, that may confound or modulate cortical changes ostensibly linked to vestibular loss (De Dieuleveult et al. 2017). Overall, direct structural associations between vestibular end‐organ functions and focal cortical morphology, while controlling for multisensory functions, in frontal and sensorimotor regions remain poorly characterized in older adults.

To fill these knowledge gaps, we investigated the associations between vestibular end‐organ function (saccule, utricle, horizontal canal) and cortical surface morphology in a subset of 10 frontal and sensorimotor regions previously implicated in higher‐order vestibular processing. We used data from 117 healthy, older adults from the Baltimore Longitudinal Study of Aging (BLSA) who underwent concurrent T1‐weighted MRI scanning and vestibular and multisensory testing (hearing, vision, proprioception). We utilized surface‐based morphometry based on Large Deformation Diffeomorphic Metric Mapping (LDDMM) to detect local surface expansions/compressions that might be overlooked by voxel‐based morphometry of volumetric approaches (Ratnanather et al. 2022). We aimed to answer two questions:

1. Is age‐related vestibular function related to prefrontal and sensorimotor cortex surface morphology in healthy, older adults?

2. Do vestibular‐associated morphological alterations in the prefrontal and sensorimotor cortices persist after accounting for multisensory involvement?

We hypothesized that higher functioning of the saccule, utricle, and horizontal semicircular canal is related to surface shape alterations in the regions of interest (ROIs), even after accounting for age, intracranial volume, sex, and multisensory function (hearing, vision, and proprioception). This cohort and its measurements were used in previous studies by our group (Jacob et al. 2020; Kamil et al. 2018) that explored distinct research questions involving a different cognitive network. Furthermore, this study significantly extends our previous study of prefrontal and sensorimotor volumes in this cohort (Padova et al. 2024) by using surface‐based morphometry and accounting for confounds from other age‐related sensory declines. By bridging end‐organ‐specific vestibular testing with surface‐based cortical shape analysis and multisensory testing, this study will improve the understanding of the consequences of aging on the vestibular pathways involved in vestibular‐mediated cognitive abilities and behaviors. An improved understanding of focal cortical markers will aid in developing rational strategies to preserve vestibular‐mediated behaviors in aging and populations with cognitive impairment.

2. Data and Methods

2.1. Study Sample

The data is a subset of 117 healthy older (aged ≥ 60 years) participants from the BLSA who had MRI brain scans and vestibular testing in the same visit between 2013 and 2015 (Shock and Gerontology Research Center 1984). All participants provided written informed consent. The BLSA study protocol (03‐AG‐0325) was approved by the National Institute of Environmental Health Sciences Institutional Review Board. Hearing loss, visual acuity loss, and proprioceptive loss were measured and included as confounding variables in follow‐up hypothesis tests. Hearing loss was measured as the speech‐frequency pure tone average of air‐conduction thresholds at 0.5, 1, 2, and 4 kHz from the better ear. Visual acuity loss, which refers to how much a pattern must differ in size to be seen, was measured as the angular deviation in logMAR units and ranges from 0.80 to −0.30 logMAR, where lower values indicate better acuity. Proprioceptive loss was measured as the degree of ankle deflection perceptible according to an established BLSA procedure (Ko et al. 2015). For analysis, the hearing, vision, and proprioceptive variables were treated as continuous variables and were negated so that increasing values indicate better function.

2.2. Vestibular Physiologic Testing

Vestibular function testing included measurement of saccular function using the cervical vestibular‐evoked myogenic potential (cVEMP) test, of utricular function using the ocular VEMP (oVEMP) test, and of horizontal semicircular canal function using the video head‐impulse test (vHIT), following established procedures (Harun et al. 2016; Li et al. 2014; Li et al. 2015; Nguyen et al. 2010).

2.2.1. cVEMP Test

The cVEMP test measures the function of the saccule (and inferior vestibular nerve) (Harun et al. 2016; Li et al. 2014; Li et al. 2015; Nguyen et al. 2010). Participants sat on a chair inclined at 30° above the horizontal plane. Trained examiners positioned EMG electrodes bilaterally on the sternocleidomastoid and sternoclavicular junction, with a ground electrode on the manubrium sterni. Participants were instructed to turn their heads to generate at least a 30 μV background response prior to delivering sound stimuli. Bursts of 100 auditory stimuli of 500 Hz and 125 dB were administered monoaurally through headphones (VIASYS Healthcare, Madison, WI). cVEMPs were recorded as short‐latency EMGs of the inhibitory response of the ipsilateral sternocleidomastoid muscle. To calculate corrected cVEMP amplitudes, nuisance background EMG activity collected 10 ms prior to the onset of the auditory stimulus was removed. The higher corrected cVEMP amplitude (unitless) from the left and right sides was used as a continuous measure of saccular function. A difference of 0.5 in corrected cVEMP is considered clinically relevant (Nguyen et al. 2010).

2.2.2. oVEMP Test

The oVEMP test measures the function of the utricle (and superior vestibular nerve) (Harun et al. 2016; Li et al. 2014; Li et al. 2015; Nguyen et al. 2010). Participants sat on a chair inclined at 30° above the horizontal plane. Trained examiners placed a noninverting electrode ≈3 mm below the eye centered below the pupil, an inverting electrode 2 cm below the noninverting electrode, and a ground electrode on the manubrium sterni. To ensure that symmetric signals are recorded from both eyes, participants were instructed to perform multiple 20° vertical saccades before stimulation. During oVEMP testing, participants were instructed to maintain an upward gaze of 20°. Head taps (vibration stimuli) applied to the midline of the face at the hairline and ≈30% of the distance between the inion and nasion using a reflex hammer (Aesculap model ACO12C, Center Valley, PA). oVEMPs were recorded as short‐latency EMGs of the excitation response of the contralateral external oblique muscle of the eye. The higher oVEMP amplitude (μV) from the left and right sides was used as a continuous measure of utricular function. A difference of 5 μV in oVEMP is considered clinically relevant (Nguyen et al. 2010).

2.2.3. vHIT

The vHIT measures the horizontal vestibular‐ocular reflex (VOR) (Agrawal et al. 2013; Agrawal et al. 2014; Harun et al. 2016) and was performed using the EyeSeeCam system (Interacoustics, Eden Prarie, MN) in the same plane as the right and left horizontal semicircular canals (Agrawal et al. 2014; Schneider et al. 2009; Weber et al. 2009). To position the horizontal canals in the plane of stimulation, trained examiners tilted the participant's head downward 30° below the horizontal plane and instructed participants to maintain their gaze on a wall target ≈1.5 m away. The examiner delivered rotations of 5°–10° (≈150°–250° per second) to the participant's head. The head impulses are performed at least 10 times parallel to the ground toward the right and left, chosen randomly for unpredictability. The EyeSeeCam system quantified eye and head velocity. VOR gain was calculated as the unitless ratio of the eye velocity to the head velocity. A VOR gain equal to 1.0 is normal and indicates equal eye and head velocities. The mean VOR gain from the left and right sides was used as a continuous variable. A difference of 0.1 in VOR gain is considered clinically relevant (Harun et al. 2016; Nguyen et al. 2010).

2.3. Structural MRI Acquisition

T1‐weighted volumetric MRI scans were acquired in the sagittal plane using a 3T Philips Achieva scanner at the National Institute on Aging Clinical Research Unit. The sequence used was a T1‐weighted image (WI) (magnetization prepared rapid acquisition with gradient echo [MPRAGE]; repetition time [TR] = 6.5 ms, echo time [TE] = 3.1 ms, flip angle = 8°, image matrix = 256 × 256, 170 slices, voxel area = 1.0 × 1.0 mm, 1.2 mm slice thickness, FOV = 256 × 240 mm, sagittal acquisition).

2.4. MRI Processing Pipeline

Scans were automatically segmented using MRICloud (https://www.mricloud.org/) with the T1 multi‐atlas set “BIOCARD3T_297labels_10atlases_am_hi_erc_M2_252_V1.” The MRICloud multi‐atlas segmentation pipeline relies on the LDDMM image registration framework (Beg et al. 2005; Ceritoglu et al. 2013). Briefly, a template image I is aligned to a target image J by minimizing over vector fields the objective function calculated as the sum of the geodesic transformation energy and the data matching energy of the distance between the deformed template $I\circ {\phi }_1^{-1}$ and the target J

$$\frac{1}{2}{\int }_0^1{\left|v\left(t\right)\right|}_V^2\mathit{dt}+\frac{1}{2{\sigma }^2}{‖I\circ {\phi }_1^{-1}-J‖}_{L^2}^2,\text{with}\frac{d}{\mathit{dt}}{\phi }_t=v_t\circ {\phi }_t,{\phi }_0=\mathit{Id},t\in \left[0,1\right]$$

where the diffeomorphism $\phi$, the three‐dimensional velocities $v_t$ belong to the fixed Reproducing Kernel Hilbert Space (RKHS) V of 1‐time continuously differentiable‐in‐space functions decaying to zero at infinity, ${\left|\cdot \right|}_V$ is the RKHS norm, and ${\sigma }^{-2}$ is a regularization parameter. The final whole‐brain segmentations were attained using a Bayesian multi‐atlas likelihood fusion method (Tang et al. 2013).

Our analysis focuses on the 10 ROIs relevant to our hypothesis and shown in Figure 1. These ROIs include the middle‐superior part of the prefrontal cortex (SFG_PFC), frontal pole (SFG_pole), and posterior pars of the superior frontal gyrus (SFG), the dorsal prefrontal cortex and posterior pars of middle frontal gyrus (MFG_DPFC, MFG), the pars opercularis, pars triangularis, and pars orbitalis of the IFG, as well as the precentral gyrus (PrCG) and postcentral gyrus (PoCG) of the sensorimotor cortex. Intracranial volume was comprised of bilateral cerebral volumes, cerebellum, brainstem, and cerebrospinal fluid. We followed a procedure similar to those described in previous studies investigating subcortical changes associated with mild cognitive impairment and Alzheimer's disease (Miller et al. 2015; Qiu, Fennema‐Notestine, et al. 2009; Younes et al. 2014), Huntington's disease (Faria et al. 2016), attention deficit hyperactivity disorder (Qiu, Crocetti, et al. 2009), and schizophrenia (Qiu et al. 2010; Qiu et al. 2008). Figure 2 depicts an overview of the neuroimaging pipeline.

FIGURE 1.

Putative vestibular‐thalamocortical and cortico‐cortical circuits. Vestibular information from the semicircular canals, otoliths, and vestibular nuclei reaches the precentral and postcentral gyri of the sensorimotor cortex and the frontal gyrus via the thalamocortical and cortico‐cortical circuits. The red arrow indicates the ventral lateral nucleus of the thalamus which putatively receives vestibular input. CAWorks (www.cis.jhu.edu/software/caworks) was used for visualization. Pfc: prefrontal cortex; SCC: semicircular canals.

FIGURE 2.

Neuroimaging pipeline. Using T1‐weighted MRI scans as inputs, MRICloud automatically outputs a whole‐brain parcellation using a study‐appropriate multi‐atlas and LDDMM. Taking the binary image segmentations of the subset of regions of interest (eight subregions of the frontal cortex, the precentral gyrus, and the postcentral gyrus), 3D surfaces for each structure created using a restricted Delaunay triangulation. Then the collection of surfaces for each structure is uploaded to the MRICloud Shape Analysis pipeline (Ratnanather et al. 2022; Wu and Mori 2023) to perform surface template estimation and subsequently template‐to‐population mapping. The output vertex‐wise deformation descriptors (the logarithms of the surface and normal Jacobians) are then reduced to k descriptors based on spectral clustering for downstream statistical testing. Quality control was performed at each stage.

2.5. Shape Analysis

For each 3D segmented ROI, surface meshes with ≈800 vertices were generated using a restricted Delaunay triangulation. Using the MRICloud surface template generation pipeline, the collection of ROI surfaces was used to estimate left‐ and right‐side population templates (i.e., the average shape) agnostic to diagnostic criteria by an LDDMM‐based surface template estimation procedure after rigid alignment (Ma et al. 2010). The MRICloud template‐to‐population surface mapping pipeline was used to register each participant's surface to the population template, first rigidly then diffeomorphically using LDDMM for surface registration, called surface LDDMM (Vaillant and Glaunes 2005). By using surface LDDMM to generate a correspondence between the template and each target surface, high dimensional shape statistics are generated. The surface LDDMM algorithm generates an optimal diffeomorphism between the template surface S _temp and the target surface S _targ by minimizing the objective function representing the sum of the geodesic transformation energy and the data matching energy of the distance between the template‐mapped‐to‐target ${\phi }_1\cdot S_{\mathit{\text{temp}}}$ and the target

$$\frac{1}{2}{\int }_0^1{\left|v_t\right|}_V^2\mathit{dt}+d\left({\phi }_1\cdot S_{\mathit{\text{temp}}},S_{\mathit{\text{targ}}}\right),\text{with}\frac{d}{\mathit{dt}}{\phi }_t=v_t\circ {\phi }_t,t\in \left[0,1\right]$$

where $d\left({\phi }_1\cdot S_{\mathit{\text{temp}}},S_{\mathit{\text{targ}}}\right)$ denotes a square normed distance between surfaces. Then, shape statistics were derived from the optimal diffeomorphic transformation.

Surface shape alterations were measured by the logarithms of surface and normal Jacobian determinants of the diffeomorphic transformation at each vertex. The surface Jacobian is calculated as the ratio between the surface area of the faces attached to a vertex pre‐ and post‐transformation. The normal Jacobian is the ratio between the full Jacobian and the surface Jacobian. Whereas the surface Jacobian refers to change in surface area pre‐ and post‐transformation, the normal Jacobian refers to the change in normal distance pre‐ and post‐transformation. A positive (negative) surface log‐Jacobian value denotes an expansion (contraction) of the template around that vertex in the direction tangent to the surface to fit the subject. Similarly, a positive (negative) normal Jacobian value denotes an expansion (contraction) of the template around that vertex in the direction normal to the surface to fit the subject. We analyze the surface and normal log‐Jacobians independently. To increase the power of the analyses and to improve computational efficiency, the surfaces were spectrally clustered into $k\in \left\{10,\dots ,20\right\}$ clusters of size ≈150–400 mm² based on the surface geometry of the template, as described previously (Faria et al. 2016). Thus, the k shape descriptor variables attached with each subject structure were used as separate outcome variables for hypothesis testing.

2.6. Statistical Modeling

For participants with missing vestibular data, we carried over data from an adjacent prior or subsequent visit using an external longitudinal dataset comprised of the same participants (Mongin et al. 2019). Whereas our original dataset had 58, 64, and 91 observations for cVEMP, oVEMP, and VOR, respectively, the imputed dataset had 95, 100, and 107 observations for cVEMP, oVEMP, and VOR, respectively. Using this imputed dataset, multiple linear regression adjusted for age, intracranial volume, and sex was used to investigate the relationship between local shape descriptors and vestibular function. The null hypothesis, $H_{0A}$, in Equation (1) predicts the (normal, surface) Jacobian $j{\mathit{ac}}_i$, for participant i, $i=1,\dots ,N$. The alternate hypothesis $H_1$, predicts the (normal, surface) Jacobian $j{\mathit{ac}}_i$ using a vestibular variable, ${\mathit{\text{vest}}}_i$ in Equation (2), such as best corrected cVEMP, best oVEMP, and mean VOR gain as continuous independent variables,

$$H_{0A}:j{\mathit{ac}}_i=c_0+c_2{\mathit{age}}_i+c_3{\mathit{\text{isFemale}}}_i+c_4{\mathit{icv}}_i+{\mathrm{ϵ}}_i$$ (1)

$$H_1:j{\mathit{ac}}_i=c_0+c_1{\mathit{\text{vest}}}_i+c_2{\mathit{age}}_i+c_3{\mathit{\text{isFemale}}}_i+c_4{\mathit{icv}}_i+{\mathrm{ϵ}}_i$$ (2)

To test whether the addition of the function of hearing, vision, or proprioception either explains away or masked vestibular relationships, we performed three additional bivariate sensory hypothesis tests. The null hypothesis, $H_{0B}$, in Equation (3) and the alternative hypothesis, $H_2$, in Equation (4) additionally covary for the sensory variable, which represents hearing function, vision function, or proprioceptive function,

$$H_{0B}:j{\mathit{ac}}_i=c_0+c_2{\mathit{age}}_i+c_3{\mathit{\text{isFemale}}}_i+c_4{\mathit{icv}}_i+c_5\mathit{\text{sensor}}y+{\mathrm{ϵ}}_i$$ (3)

$$H_2:j{\mathit{ac}}_i=c_0+c_1{\mathit{\text{vest}}}_i+c_2{\mathit{age}}_i+c_3{\mathit{\text{isFemale}}}_i+c_4{\mathit{icv}}_i+c_5\mathit{\text{sensor}}y+{\mathrm{ϵ}}_i$$ (4)

In hypothesis tests $\left\{H_{0A},H_{0B},H_1,H_2\right\}$, $c_0$ corresponds to the global average, ${\mathit{age}}_i$ is the age in years of subject $i$, ${\mathit{\text{isFemale}}}_i$ is a binary indicator variable for the sex of subject $i$ (1 = female, 0 = male), and ${\mathit{icv}}_i$ denotes the intracranial volume of subject $i$. We assumed that the log‐Jacobian of the surface transformation depends linearly on age. We also assumed that the measurement noise ${\mathrm{ϵ}}_i$ is independently and identically distributed zero‐mean Gaussian with unknown, common variance. The unknown effects $\left\{c_0,c_1,c_2,c_3,c_4,c_5\right\}$ were estimated via least‐squares. To determine whether the study sample is stable and that our individual results are not driven by outliers or extreme values, we performed permutation testing according to an established procedure (Jacob et al. 2020). The vestibular variable was permuted across all clusters on a surface under the null hypotheses, $H_{0A}$ and $H_{0B}$, for 10,000 simulations. The maximum test statistic, calculated as the maximum of the ratio of maximum squared errors of the null to the alternative model, was calculated for both the real and simulated models. The overall permutation p‐value, p _perm, is calculated as the proportion of simulated max test statistics greater than true (non‐simulated) max test statistics. We rejected the null hypothesis if p _perm < 0.05. Thus, the p‐values from testing across clusters are corrected for Family‐Wise Error Rate (FWER) at the 0.05 level. Furthermore, a cluster k is significant if the true test statistic is greater than the 95th percentile of simulated test statistics. For clusters which rejected the null hypotheses, $H_{0A}$ ($H_{0B}$), 95% confidence intervals were calculated by bootstrapping model residuals under the alternative hypothesis, $H_1$ ($H_2$), to mitigate the effects of outliers. Bootstrapped studentized confidence intervals were computed by bootstrapping model residuals with 10,000 simulations using the bootci function in Matlab. All analyses were implemented in Matlab.

3. Results

3.1. Characteristics of the Study Sample

Table 1 shows the characteristics for the study sample from the BLSA. Two‐sided t‐tests show that bivariate partial correlations of vestibular function and vision/proprioception function are insignificant (p < 0.05) while controlling for age (Table 2). Additionally, the bivariate correlation between hearing and vestibular functions was significant while controlling for age (p = 0.042), but fell below significance when additionally controlling for gender ($\rho =-0.19$ (p = 0.066)).

TABLE 1.

Characteristics of the study sample, presented on their original scale (N = 117).

Characteristic	Mean (SD)	N (%)
Age (years)	77 (8.7)
Sex
Male		79 (67.5)
Female		38 (32.5)
Education (years)	17.1 (2.5)
Best corrected cVEMP amplitude	1.2 (0.75)
Best oVEMP amplitude (μV)	13.6 (10.1)
Mean VOR gain	0.997 (0.16)
Best four‐frequency PTA (dB)	32 (14.9)
Visual acuity (logMAR)	0.11 (0.13)
Proprioception threshold (degrees)	1.71 (1.73)

Abbreviations: n: the number of participants with a visit where both the characteristic and MRI data were available; PTA: four‐frequency (0.5, 1, 2, 4 kHz) pure tone average from the better ear; SD: standard deviation; %: 100 (n/N) percent.

TABLE 2.

Bivariate Pearson partial correlation coefficients, with p‐values in parentheses, between vestibular and multisensory function (N = 117).

	Best corrected cVEMP amplitude	Best oVEMP amplitude	Mean VOR gain
Best four‐frequency PTA a	−0.21 (0.042*)	−0.00014 (1.0)	0.11 (0.25)
Visual acuity a	0.11 (0.31)	−0.064 (0.54)	0.0050 (0.97)
Proprioception threshold a	−0.060 (0.57)	−0.13 (0.20)	−0.090 (0.36)

Abbreviation: PTA: four‐frequency (0.5, 1, 2, 4 kHz) pure tone average from the better ear.

^a Variables have been negated such that increasing values indicate better function.

^* p < 0.05.

3.2. Vestibular Effects on Prefrontal and Sensorimotor Cortex Morphology

Figures 3, 4, 5 illustrate the spatial distribution of the significant vestibular‐only effects from the alternative hypothesis $H_1$ and of the significant vestibular effects from the alternative hypothesis $H_2$ which additionally covaried for hearing, vision, or proprioception function. Adding hearing function to the model reduced the saccular and canal, but not utricular, function model sample sizes from 95 and 107 subjects to 94 and 106 subjects, respectively. Adding vision function reduced the saccular, utricular, and canal function model sample sizes from 95, 100, and 107 subjects to 90, 95, and 100 subjects, respectively. The addition of proprioception function reduced the saccular and canal, but not utricular, function model sample sizes from 95 and 107 subjects to 94 and 106 subjects, respectively.

FIGURE 3.

Spatial distribution of the significant saccular effects on the shapes of the frontal and sensorimotor cortices visualized on the population template. (A) shows the saccular‐only results, and (B and C) show the saccular‐hearing results. Regions that are colored red (blue) indicate a significant surface expansion (compression) in the direction tangent to the surface (surface Jacobian) with higher saccular function. MFG, posterior pars of middle frontal gyrus; PoCG, postcentral gyrus of the sensorimotor cortex.

FIGURE 4.

Spatial distribution of the significant utricular effects on frontal cortex shape visualized on the population template. (A–C) show the utricular‐only results, (D and E) show the utricular‐hearing results, and (F and G) show the utricular‐proprioception results. Red (blue) indicates a region of significant surface expansion (compression) in the direction tangent/normal to the surface (surface/normal Jacobian) with higher utricular function. MFG, posterior pars of middle frontal gyrus; SFG, posterior pars of the superior frontal gyrus; SFG_PFC, the middle‐superior part of the prefrontal cortex; SFG_pole, frontal pole.

FIGURE 5.

Spatial distribution of the significant horizontal canal effects on frontal cortex shape visualized on the population template. (A–C) show the utricular‐only results, (D and E) show the utricular‐hearing results, and (F–H) show the utricular‐proprioception results. Red (blue) indicates a region of significant surface expansion (compression) in the direction tangent/normal to the surface (surface/normal Jacobian) with higher canal function. MFG, posterior pars of middle frontal gyrus; SFG, posterior pars of the superior frontal gyrus; SFG_PFC, the middle‐superior part of the prefrontal cortex, SFG_pole, frontal pole.

3.2.1. Prefrontal Cortex

In the vestibular‐only analyses, several relationships between vestibular end‐organ function and surface shape alterations in the prefrontal cortex were significant according to permutation testing. A 1 standard deviation (SD) increase in saccular function was associated with a 0.031% expansion tangent to the cortical surface in the medial left posterior MFG (p ≈ 0.04, CI: (−0.028, 0.091)). A 1SD increase in utricular function was associated with a 0.008% expansion tangent to the cortical surface in the medial left posterior MFG (p ≈ 0.018, CI: (−0.048, 0.064)), a 0.009% compression normal to the cortical surface in the rostral lateral region of the left SFG (p ≈ 0.047, CI: (−0.041, 0.023)), a 0.023% compression normal to the cortical surface in the caudal dorsal region of the left SFG_pole (p ≈ 0.0019, CI: (−0.054, 0.007)), and a 0.027% compression normal to the cortical surface in the dorsal region of the right SFG_pole (p ≈ 0.031, CI: (−0.060, 0.008)). A 1SD increase in canal function was associated with a 0.008% expansion tangent to the cortical surface in the medial rostral region of the left SFG (p ≈ 0.034, CI: (−0.033, 0.050)), a 0.008% expansion tangent to the cortical surface in the dorsal lateral region of the right SFG_PFC (p ≈ 0.042, CI: (−0.037, 0.054)), and a 0.018% expansion normal to the cortical surface in the dorsal region of the right SFG_pole (p ≈ 0.035, CI: (−0.014, 0.052)). Figures 3, 4, 5 show the spatial distribution of these associations.

In the multisensory analyses of otolith function, many relationships persisted and others were found (see Figures 3 and 4, and Table S1). Figure 3B shows a 1SD increase in saccular function was associated with a 0.025% expansion tangent to the cortical surface in the medial left posterior MFG after accounting for hearing function (p ≈ 0.0283, CI: (−0.007, 0.11)). A 1SD increase in saccular function was associated with a 0.025% expansion tangent to the cortical surface in the medial left posterior MFG after accounting for hearing function (p ≈ 0.0283, CI: (−0.007, 0.11)) (see Figure 3B). In models additionally covarying for hearing function, a 1SD increase in utricular function was associated with a 0.008% expansion tangent to the cortical surface in the medial left MFG (p ≈ 0.0164, CI: (−0.038, 0.069)), a 0.006% expansion tangent to the dorsal lateral surface of the right SFG_PFC (p ≈ 0.0157, CI: (−0.055, 0.025)) (Figure 4D), and a 0.023% compression normal to the caudal surface of the left SFG_pole (p ≈ 0.0025, CI: (−0.021, 0.045)) (Figure 4D). In models additionally covarying for proprioceptive function, a 1SD increase in utricular function was associated with a 0.014% compression normal to the cortical surface in the rostral lateral left SFG (p ≈ 0.0312, CI: (0.027, 0.105)), a 0.025% compression normal to the cortical surface in the rostral dorsal left SFG_pole (p ≈ 0.0025, CI: (−0.026, 0.041)), and a 0.031% compression normal to the caudal surface of left SFG_pole (p ≈ 0.037, CI: (−0.096, −0.012)) (see Figure 4F,G).

In the multisensory analyses of canal function, many relationships persisted and numerous others were found (see Figure 5 and Table S1). Figure 5D,E shows a 1SD increase in canal function was associated with a 0.002% expansion tangent to the cortical surface in the medial rostral region of the left SFG (p ≈ 0.035, CI: (−0.043, 0.044)), a 0.007% expansion normal to the cortical surface in the medial rostral region of the left SFG (p ≈ 0.037, CI: (−0.02, 0.033)), and a 0.006% expansion tangent to the cortical surface in the dorsal lateral region of the right SFG_PFC (p ≈ 0.014, CI: (−0.042, 0.052)), after accounting for hearing function. In the canal‐proprioception function models, a 1SD increase in canal function was associated with a 0.003% compression tangent to the cortical surface in the medial left posterior MFG, a 0.008% expansion tangent to the cortical surface in the medial rostral region of the left SFG (p ≈ 0.033, CI: (−0.025, 0.051)), and a 0.018% expansion normal to the cortical surface in the medial rostral region of the left SFG (p ≈ 0.027, CI: (−0.011, 0.047)) (see Figure 5F–H).

No relationships between vestibular function and the shape of the IFG (pars opercularis, pars triangularis, pars orbitalis), or the MFG_DPFC survived permutation testing at the 0.05 level. Notably, these findings persist and more are uncovered after accounting for hearing function, vision function, and proprioceptive function in individual bivariate analyses (see Figures 3, 4, 5 and Table S1). Furthermore, all relationships between vestibular function and the shapes of the frontal and sensorimotor cortices were attenuated when correcting for vision function, but many relationships still showed strong trends in the left MFG (canal function: p ≈ 0.066), left SFG (canal function: p ≈ 0.056), and right SFG_PFC (canal function: p ≈ 0.087). Moreover, we note several strong trends in the multisensory analyses that did not survive permutation testing at the 0.05 level (see Table S1).

3.2.2. Sensorimotor Cortex

No relationships between vestibular function and the shape of the PrCG or the PoCG survived permutation testing at the 0.05 level (i.e., all p _perm ≥ 0.05) in the non‐multisensory analysis. Importantly, a relationship was found after accounting for hearing function, but not vision or proprioceptive function in separate bivariate analyses (see Figure 3C and Table S1). Permutation testing revealed a significant relationship between saccular function and tangent surface shape in the right poCG when additionally covarying for hearing function. Figure 3C shows that a 1SD increase in saccular function correlated with approximately 0.020% expansion tangent to the cortical surface in the posterior ventrolateral surface of the right PoCG (p ≈ 0.0242, CI: (−0.075, 0.010)). Despite not surviving permutation testing at the 0.05 level, there were several strong trends in the multisensory analyses, in particular in the left PoCG (canal‐vision function model: p ≈ 0.079), left PrCG (saccular‐vision function model: p ≈ 0.091), and right PrCG (canal‐proprioception function model: p ≈ 0.1) (see Supplementary Table S1).

4. Discussion

In this study of healthy, older adults, we found that reduced vestibular function is associated with shape alterations in a subset of 10 prefrontal and sensorimotor ROIs of the broader multisensory vestibular network that are thought to subserve higher‐order cognitive and sensorimotor abilities. The ROIs investigated include the middle‐superior part of the prefrontal cortex (SFG_PFC), frontal pole (SFG_pole), and posterior pars of the SFG, the dorsal prefrontal cortex and posterior pars of middle frontal gyrus (MFG_DPFC, MFG), the pars opercularis, pars triangularis, and pars orbitalis of the IFG, as well as the PrCG and PoCG of the sensorimotor cortex. Specifically, we found associations between reduced saccular function and significant cortical surface compression in the MFG, reduced utricular function and MFG compression and expansion of the SFG and SFG_pole, respectively, and reduced canal function and surface compression of the SFG, SFG_PFC, and SFG_pole. After additionally adjusting for measures of hearing and proprioception, we observed shape alterations in the MFG, SFG, SFG_PFC, SFG_pole, and PoCG with poorer end‐organ functions. However, additionally adjusting for vision function attenuated the observed relationships, albeit they exhibited strong trends toward significance. This finding likely stems from a power loss resulting from a redistribution of explained variance, thereby reducing the vestibular‐only effect size. Additionally, the loss of power is likely influenced by a small reduction in degrees of freedom (e.g., adding vision reduced the saccular, utricular, and canal function model sample sizes from 95, 100, and 107 subjects to 90, 95, and 100 subjects, respectively). This loss of power raises the detectable vestibular‐only effect size for our sample size, and thus a larger sample size would be needed to detect vestibular effects in the presence of vision effects. Furthermore, a ceiling/floor effect of vision function could lead to overestimation of the vision effect, further exacerbating the issue with the detectable vestibular effect size. Importantly, given that vestibular and vision functions were insignificantly correlated, and a larger sample size would allow the vestibular‐only effects to be revealed in the presence of vision function, we suspect that the relationship between vestibular function and local frontal cortex morphology may have an effect that is independent of vision function. The significant structures are known to exhibit robust activations to artificial and naturalistic vestibular stimulation as well as structural alterations in aging and vestibular syndromes (Hong et al. 2014; Hüfner et al. 2009; Hupfeld, McGregor, et al. 2022; Lopez et al. 2012; Nakul et al. 2021; Padova et al. 2024; Wurthmann et al. 2017; Zu Eulenburg et al. 2011). Our findings align with previous links between vestibular function and the structures of the somatosensory (Hüfner et al. 2009; Hupfeld, McGregor, et al. 2022; Padova et al. 2024), motor (Hupfeld, McGregor, et al. 2022; Wurthmann et al. 2017), and prefrontal cortices (Hong et al. 2014; Hupfeld, McGregor, et al. 2022; Padova et al. 2024; Wurthmann et al. 2017) and clarify previous inconsistent reports (Göttlich et al. 2016; Helmchen et al. 2009; Hong et al. 2014; Hüfner et al. 2009; Hupfeld, McGregor, et al. 2022; Kremmyda et al. 2016; Padova et al. 2024; Wurthmann et al. 2017; Zu Eulenburg et al. 2010).

4.1. Prefrontal Cortex

Our findings support the initial hypothesis that diminished vestibular function correlates with structural changes in the prefrontal cortex, largely independent of multisensory functions. Specifically, reductions in saccular and utricular functions are associated with surface compression in the medial left MFG, irrespective of auditory function. Additionally, decreased canal function correlates with compressions in the medial rostral region of the left SFG, independent of both hearing and proprioception functions, and in the dorsal lateral region of the right SFG_PFC, independent of hearing function alone. In the context of age‐related auditory changes, reduced canal function is associated with compression in the most rostral medial region of the left SFG. Conversely, considering age‐related proprioceptive changes, reduced canal function is linked to both an expansion in the rostral medial surface of the left MFG and a compression in the rostral lateral surface of the left SFG. Interestingly, the compressive effect on the dorsolateral surface of the right SFG_PFC, associated with diminished utricular function, is insignificant when accounting for age‐related hearing changes. Unexpectedly, reduced utricular function is also associated with expansions in the rostral lateral region of the left SFG, the caudal dorsal region of the left SFG_pole, and the rostral dorsal region of the right SFG_pole, independent of proprioceptive functions. These expansions, which may reflect age‐related alterations in vestibular sensitivity, are moderated by auditory function, as evidenced by attenuations in the expansions of the right SFG_pole and left SFG (Jahn et al. 2003; Zu Eulenburg et al. 2017).

The MFG and SFG, which contain the premotor cortex, supplementary motor area, and frontal eye fields, are crucial for motor control, planning, and initiating visuospatial movements. These regions are interconnected with various brain regions, including other prefrontal areas, premotor, cingulate, somatosensory, and insular regions, facilitating the coordination of working memory for actions and complex planning sequences (Rolls et al. 2023a). Vestibular inputs to these areas, supported by animal and human studies, suggest significant vestibular influence on regions near the frontal eye fields and the supplementary motor area (Çavdar et al. 2023; Ebata et al. 2004; Lopez and Blanke 2011). Evidence from subclinical and clinical studies further suggests that vestibular impairments correlate with structural changes in these cortical areas, emphasizing their role in vestibular processing (Hong et al. 2014; Hupfeld, McGregor, et al. 2022; Wurthmann et al. 2017). Moreover, the frontal pole's extensive connectivity and role in episodic memory (lateral subregion) and cognitive switching (anterior medial subregion) suggest its critical involvement in managing deficits in cognitive‐motor dual‐tasking observed in vestibular patients (Danneels et al. 2023; Gilbert et al. 2006; Peng et al. 2018; Rolls et al. 2023b). Despite no detected relationship in the literature at the time of writing between dual‐task gait performance and the SFG_pole (Hupfeld, Geraghty, et al. 2022), we speculate that the frontal pole utilizes utricle‐ and canal‐transduced linear and angular acceleration data to coordinate complex body and visuomotor actions (e.g., locomotion, reaching, and grasping) while balancing various cognitive, interoceptive, and emotional demands.

Unexpectedly, no significant correlations were found between vestibular function and the shapes of the pars opercularis, pars orbitalis, and pars triangularis of the IFG, or the MFG_DPFC. This absence of expected correlations, despite extensive literature highlighting the involvement of these areas in vestibular processing (Nakul et al. 2021), suggests two possible explanations. The first is that we may be more likely to detect volume‐based or cortical thickness‐based changes in these regions. The second is that there are potential compensatory mechanisms within the brain that adjust to age‐related changes in vestibular sensitivity (Jahn et al. 2003; Zu Eulenburg et al. 2017). Furthermore, the peripheral vestibular information processed through thalamic‐limbic‐striatal‐frontal circuits likely integrates with multi‐sensorimotor data before reaching the prefrontal cortex. This integration could explain the lack of observed structural changes as compensatory adaptations or differential sensitivities to combined sensory inputs, influenced by aging and neuroplasticity. Together, an age‐related and multisensory involvement could explain how these regions respond to otolith or canal information and show no relationship with structural alterations (and language function (Bosmans et al. 2022; Wei et al. 2020)) in older adults. Overall, our results underscore a significant link between age‐related declines in vestibular function and morphological variations in the frontal cortex, suggesting a broader impact of sensory integration on cognitive and motor functions in older adults (Bigelow and Agrawal 2015; Grabherr et al. 2011; Moser et al. 2017; Preuss et al. 2014).

4.2. Sensorimotor Cortex

It is unclear why, in the vestibular‐only analysis, there were no associations of saccular, utricular, or horizontal semicircular canal function with the PrCG or PoCG in either hemisphere, given these regions are implicated in the vestibular cognitive network (de Waele et al. 2001; Ferrè and Haggard 2020; Hitier et al. 2014; Lopez and Blanke 2011). It may be the case that we are more likely to detect changes in volume or thickness in these regions. However, in the multisensory analysis, reduced saccular function correlated with compression in the posterior ventrolateral region of the right PoCG when accounting for age‐related hearing loss. Previous studies of vestibular stimulation reported robust neural responses in the primary and secondary somatosensory cortex in rats (Rancz et al. 2015) and in humans (Lopez and Blanke 2011). Because the posterior ventrolateral region of the right PoCG may contain the representation of the mouth and larynx, it is connected with the multi‐sensorimotor speech and language network (e.g., the supramarginal gyrus), and speech activates both the auditory and vestibular systems (Emami 2023; Emami et al. 2012; Gattie et al. 2021), this finding may imply the importance of saccular and hearing function in the context of speech planning and execution. We speculate that this finding may also be important for self‐other voice discrimination, which relies on auditory, somatosensory (e.g., bone‐conducted vibration signals, mouth proprioception), and vestibular processing (Emami 2023; Emami et al. 2012) by the PoCG. Additionally, the connectivity of the posterior ventrolateral region of the PoCG, approximately corresponding to BA 1 and BA 2, with key vestibular network regions (e.g., the insula) implies an important role more broadly in somatosensation, bodily self‐consciousness and control, and motor planning which involves self‐motion perception and social cognition (Anson, Pineault, et al. 2019; Deroualle and Lopez 2014; Roditi and Crane 2012; Rolls et al. 2023; Stiles and Smith 2015).

Whether the PoCG may use information about linear acceleration of the head in the horizontal plane transduced by the utricle or about angular acceleration of the head in the horizontal plane (yaw) transduced by the horizontal semicircular canal must be elucidated. Older adults with reduced vestibular function as measured by the standing on foam with eyes closed balance (FOEC) test (i.e., more sway) were observed to have poorer sensorimotor cortex structure (Hupfeld, McGregor, et al. 2022). This is important as one study of older adults found that age‐related horizontal canal dysfunction is associated with decreased performance on the FOEC test (Anson, Bigelow, et al. 2019). Additionally, older adults were observed to have significantly shallower sulcal depth (i.e., worse brain structure) in the sensorimotor, supramarginal, insular, and superior frontal and parietal cortices with poorer dual‐gait performance (Hupfeld, Geraghty, et al. 2022).

4.3. Strengths of This Study

We report several strengths of this study. One such strength is that the relationships examined were hypotheses‐driven based on converging evidence from structural and functional neuroimaging in humans. A second strength is that we use a state‐of‐the‐art brain mapping pipeline. This pipeline utilizes a study‐appropriate multi‐atlas and LDDMM, a well‐established framework of nonlinear image and shape registration techniques (Ashburner and Friston 2011). LDDMM is known for its desirable properties, including the ability to handle large deformations, conserve momentum, remove nuisance transformations, and achieve low‐variance mappings, often yielding superior results compared to other diffeomorphic registration methods (Ashburner and Friston 2011; Miller et al. 2006; Tward et al. 2018). Moreover, we employed LDDMM‐based surface diffeomorphometry (Vaillant and Glaunes 2005) which can achieve superior performance compared to other surface‐based registration methods, such as CARET or Freesurfer (Zhong et al. 2010), and can overcome the limitations encountered by other vestibular neuroimaging studies that used low‐strength MRI, voxel‐based morphometry, or volume‐based morphometry, such as the tendency to miss effects that are subtle, non‐focal, or non‐uniformly spatially distributed across the ROI (Davatzikos 2004). Surface diffeomorphometry provides a sensitive measure of cortical shape variation and has been used to track sub‐voxel structural alterations in aging and disease (Faria et al. 2016; Jacob et al. 2020; Miller et al. 2015; Qiu, Crocetti, et al. 2009; Qiu, Fennema‐Notestine, et al. 2009; Qiu et al. 2010; Qiu et al. 2008; Younes et al. 2014). A third strength is that our quality control pipeline involved manual inspections of the data at each step of processing. Also, our statistical testing pipeline accounts for multiple comparisons as well as for outliers using permutation testing and bootstrapping, respectively. In contrast to vestibular neuroimaging studies that stimulate the end‐organs in a combination of ways (e.g., galvanic vestibular stimulation; caloric stimulation), we use individual measurements of the utricle, saccule, and horizontal semicircular canal to capture end‐organ specific relationships with brain morphology. This is important for aging studies because the hair cells in the cristae of the semicircular canals decline with age earlier than those of the otolithic maculae. Thus, their individual contributions to the aging of the central vestibular pathways may be different. We also used specific clinical assessments of hearing, vision, and proprioception function to determine multisensory involvement, rather than use a composite clinical test based on gait or balance/posturography.

4.4. Limitations of This Study

We note several limitations to this study. First, although most of the frontal cortex regions identified here show activations during vestibular stimulation in functional neuroimaging studies, several regions frequently highlighted as key vestibular processing hubs in such studies were not significantly associated with structural changes in this study (e.g., IFG). This discrepancy raises an important caveat: the structural changes observed here might not directly reflect functional alterations, or may reflect only a subset of the full frontal/sensorimotor vestibular processing network. Although gray matter generally declines in aging, cortical structure can also undergo plastic changes associated with either compensatory functional gains or maladaptive functional losses, neither of which may neatly map onto structural (e.g., volumetric, shape) changes. Future studies should incorporate direct functional and additional structural assessments (e.g., local thickness) to further elucidate these relationships.

While cortical surface shape analysis provides sensitive measures of morphology, it only describes how the surface is altered. Thus, structural changes within the structure of interest are missed. While volume measures the size of the broader structure and complements the local shape measures when the changes are uniform across (not within) the structure, cortical thickness may provide a complementary sensitive measure of cortical morphology, and when paired with equivolumetric theory, can describe what happens within each structure of interest in terms of layer thicknesses (Ratnanather et al. 2019). The normal Jacobian measure used in this study is qualitatively different from a cortical thickness measure, which is often defined as the length of a streamline connecting two opposing points on the cortical ribbon. Recent studies have highlighted the importance of cortical microarchitecture and layer‐specific relationships in cognitive networks (Paquola et al. 2022). The small local changes in shape could be compensated for by shape changes in the other direction in the rest of the broader structure (even if the latter changes are each nonsignificant). This has been shown before in studies of children with ADHD in whom basal ganglia volume changes in some cases conflicted with the direction of local shape changes (Seymour et al. 2017; Tang et al. 2019).

Notably, the reproducibility of findings is a challenge for several reasons. Anatomical definitions can vary between atlases and experts, adding to the great variability in the appearance of cortical parcels at high granularity. The quality and smoothness of surface triangulations, as well as the choice of surface mapping algorithm parameters, may also impact the results. The surface clustering approach may impact the results. The choice of the number of surface clusters determines the spatial extent of the cluster, which implicitly corresponds to an assumption about the size of the region related to the effect. In this study, we chose the number of clusters to balance the number of patches and, therefore, the number of comparisons and the spatial extent of the effects. To consider a continuum of surface cluster sizes, a threshold‐free cluster enhancement procedure for surfaces could be developed. Additionally, the boundaries of the clusters may overlap regions where the true effect lies and noise in such a way as to mask the true effect. Moreover, our population templates are created based on this particular sample; thus, they would be different when creating a new population template based on a different sample. Although the test statistic we used was based on maximum squared errors, which is generally less robust than the sum of squared errors, it is a conservative approach typically employed in our group. This approach is conservative because it compares our results to the least favorable outcome, rather than average outcomes, under permuted vestibular function. Due to our nonparametric testing procedure, the validity of our p‐values is independent of the data distribution and of our choice of test statistic, whereas the power of the tests is dependent.

Although our large multi‐atlas set spans our investigated age range to capture the anatomical variability of adult brains, this multi‐atlas set lacks modern cytoarchitectonic definitions that have relevance to brain function. Another limitation is that we did not examine the potential role of the cerebellum, the brainstem, the hypothalamus, or the thalamus in modulating the effects of age‐related vestibular loss in the cortex. However, robust measures of cerebellar and brainstem structures are being developed. While we did reuse data, we did not account for dataset decay (Thompson et al. 2020) because we explored a distinct research question compared to previous studies from our group that use this cohort and measurements (Jacob et al. 2020; Kamil et al. 2018). This may limit the strength of our findings, and future confirmatory studies may be needed to reinforce our findings. Additionally, our findings may not generalize to the broader and younger population due to the age range used in this study and the propensity of BLSA participants to have higher levels of education and socioeconomic status than typical adults. These are important caveats to the interpretation of our findings, as higher education and socioeconomic status may be associated with frontal and sensorimotor structure. Finally, these results may not generalize to younger adults who have weakened vestibular function.

4.5. Future Work

To understand the neuroanatomical underpinnings of aging on vestibular‐mediated behaviors, several studies will be needed. Longitudinal studies incorporating gray matter volume (Padova et al. 2021), shape, and cortical thickness and white matter microstructural integrity of the limbic system, temporo‐parietal junction, and frontal cortex will help to understand the relationships over time. Because the brain vestibular network is plastic and compensates for vestibular loss to maintain behavioral function, we aim to use changepoint analysis to identify subtle nonlinearities in the trends of brain structure alterations that may be missed by gross aging trends (Bethlehem et al. 2022). Then a precedence graph can be created that highlights the sequence of regional structural changes in relation to each other. To investigate causal hypotheses between vestibular loss and structural changes in the multi‐sensorimotor vestibular network, our group plans to use longitudinal structural equation modeling that accounts for possible confounding by multi‐sensorimotor function. Notably, structural equation modeling can test hypotheses regarding the relationship between vestibular‐mediated behaviors and intermediating brain regions (e.g., brainstem, hypothalamus, cerebellum, thalamus). By harmonizing our atlas definitions with modern brain atlases based on cytoarchitecture (Smith et al. 2023; Zachlod et al. 2023) or multimodal parcellations (Eickhoff et al. 2018; Huang et al. 2022; Smith et al. 2023), new insights into structure–function relationships can be gleaned. Altogether, this future work can reveal the sequence and causal direction of changes in the multi‐sensorimotor vestibular network.

5. Conclusion

Our findings highlight subtle associations between age‐associated vestibular loss and the structure of the frontal cortex—a key region in the vestibular cognitive network that receives multi‐sensorimotor vestibular information—in line with previous neuroimaging studies of vestibular function. Furthermore, these findings may provide the neuroanatomical links between vestibular loss and higher‐order cognitive deficits observed in the aging population and in people with dementia or Parkinson's disease. Future work will need to determine the temporal and spatial flow of structural alterations in brain regions that receive vestibular information and that are involved in vestibular‐mediated behaviors, such as self‐motion perception, motor planning, and executive function. Bolstering the understanding of the involvement of peripheral and central vestibular loss in self‐motion perception, motor planning, and executive function will be vital for the development of sensible interventions.

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1. Significant and strongly trending results of the multisensory regression models, in which hearing, vision, and proprioception function were included as covariates. A tangent expansion (compression) corresponds to a positive (negative) log‐surface Jacobian, and similarly for a normal expansion (compression) and a positive (negative) log‐normal Jacobian. Key: MFG: posterior middle frontal gyrus; SFG: posterior superior frontal gyrus; SFG_PFC: prefrontal cortex part of the superior frontal gyrus; SFG_pole: frontal pole of the superior frontal gyrus; PrCG: precentral gyrus; PoCG: postcentral gyrus; CI: confidence interval; ** p < 0.01, * p < 0.05.

Padova, D. , Ratnanather J. T., Faria A. V., and Agrawal Y.. 2025. “Reduced Vestibular Function is Associated With Cortical Surface Shape Changes in the Frontal Cortex.” Human Brain Mapping 46, no. 10: e70251. 10.1002/hbm.70251.

Data Availability Statement

Data from the BLSA are available on request from the BLSA website (blsa.nih.gov). All requests are reviewed by the BLSA Data Sharing Proposal Review Committee and are also subject to approval from the NIH institutional review board.