Patients were diagnosed to have BVF based on clinical examinations, bithermal caloric irrigation [bilateral hyporesponsiveness with mean peak slow phase velocity (SPV) of <5°/s on both sides], and quantitative head impulse recordings of the vestibulo-ocular reflex (VOR, reduced gain <0.7), absence of clinical signs for cerebellar disease, and normal cranial MRI. On clinical examination, all patients showed gait ataxia without significant consistency in lateropulsion/gait deviation. Gait ataxia severely increased with horizontal head movements while attempting to fixate targets at gaze straight ahead. Romberg’s test was pathological in all of them while the Unterberger test was not pathological (no consistent deviation) in any of the patients. A total number of 31 patients with chronic (>3 months, range: 3 months to 20 years) BVF were examined (mean VOR gain: 0.26). Nine patients had to be excluded due to comorbidity (polyneuropathy). This resulted in 22 eligible BVF patients [12 male; age: 64.0 ± 2.2 years (SE); disease duration: range 3 months to 20 years; mean 3.1 years]. The most common etiology of BVF was antibiotic ototoxicity (n = 13), unknown cause (n = 8), and sequential vestibular neuritis (n = 1). The patient and the HC group (n = 28, 17 male; age: 65.2 ± 1.7 years; mean gain 0.97 ± 0.02) did not differ significantly in age (two-sample t-test p = 0.68), gender (chi-square test p = 0.77), or Montreal Cognitive Assessment test score (two-sample t-test p = 0.52) [MoCA (30)].

Semicircular canal function was investigated by electronystagmography with caloric irrigation and quantitative head impulse testing; otolith function by static (background stationary) and dynamic (moving visual background) subjective visual vertical (SVV) (31) and cervical and ocular vestibular-evoked myogenic potentials [VEMP (32)]. Cervical VEMP were elicited by asking the lying participants to slightly lift their heads and maintain a tonic rotational position of their heads to the contralateral side while EMG activity was recorded from the mid portion of the sternocleidomastoid muscles. Unilateral AC tone bursts of 500 Hz were used and p13-n23 components were analyzed [for details see Ref. (32, 33)].

All participants were examined by quantitative head impulse test using video-oculography. Eye and head movements were recorded by the EyeSeeCam^® HIT System (Autronics, Hamburg, Germany) at a sampling rate of 220 Hz (34). VOR gain was determined by robust linear regression of eye and head velocity starting at head velocity >10°/s to 95% of peak head velocity using Matlab^®(The Mathworks Inc., Natick, MA, USA, version R2016a). For further details, see Ref. (35–37).

Posturography was recorded in the upright standing position with the hands hanging next to the trunk for 20 s. At baseline, subjects were asked to stand on the platform with feet (shoes) parallel to each other. We tested various experimental conditions (Figure 1) which differed in terms of (i) visual (eyes open/closed; EO/EC), (ii) graviceptive otolith (head tilted up, down vs. head erect, with the eyes open and closed) (38), (iii) SSC (horizontal head shaking) (iv) and proprioceptive (foam) input, (v) cognitive influence (dual task with backward counting), and (vi) complex motor challenging demands on postural control (tandem stance) (39). During horizontal head shaking (0.5 Hz) participants were asked to fixate a target 60 cm in front of the participants’ forehead. Head movements were recorded and monitored with the ZEBRIS system (CMS70P, Zebris Medizintechnik GmbH, Isny, Germany) at a sampling rate of 50 Hz (40). The system determines the position [specified by three values v = (x, y, z)] of an ultrasound-emitting marker relative to an array of three receivers. This condition was meant to compare the effects of vestibular semicircular canal stimulation on vestibulo-spinal postural control in patients with incomplete lesions of the vestibulo-ocular reflex, with and without visual feedback (eyes open vs. closed). Although this technique is not a selective stimulation of the horizontal canals, the effects on the horizontal SSC are expected to be much stronger than on the vertical SSC and on the utricles.

Head position was adjusted by an inclinometer (38). This recording assured that the different head positions (anteflection by 45°, upright head positions, 30° dorsoflection of the neck; with gaze straight ahead relative to head position) were maintained for the recording time. We used a slab of foam rubber (50 cm width, 60 cm length, height 10 cm, compression hardness: 3.3 kPa, volumetric weight: 40 kg/m³) for testing balance control under attenuated proprioceptive feedback under two conditions: (a) with the head erect, gaze fixation of LED at the gaze straight ahead position and (b) with the eyes closed.

We used a Kistler force platform (Model 9260AA6, Kistler Instrumente AG, Winterthur, Switzerland; 50 cm width, 60 cm length, height 10 cm) equipped with piezo-electric 3-component force sensors for recording postural changes during the above mentioned experimental conditions in a similar way as described elsewhere (41, 42). Postural sway signals were bidirectionally filtered (50 Hz Gaussian filter) to eliminate low amplitude recording noise (43). The platform recorded torques and sheer forces with six degrees of freedom using force transducers with an accuracy better than 0.5 N. The displacement of the center of pressure in the medio-lateral (ML) and the anterior–posterior (AP) directions were recorded and the sum vector calculated using Matlab^®. Results are given as the mean postural sway speed (PSS, in centimeter per second), calculated from the AP and ML movements: PSS=mean((APi−APi−1)2+(MLi−MLi−1)2*SamplingRate)

Postural sway was recorded in intervals of 20 s duration for off-line analysis (sampling frequency 250 Hz) (39).

Statistical analyses were performed with SPSS (22.0.0.2; IBM Corp., Somer, NY, USA). Analyzing the postural sway speed, the factors TARGET VISIBILITIY (eyes open/closed), HEAD POSITION, HEAD SHAKING, DUAL TASK (counting), PROPRIOCEPTION (foam), and TANDEM STANCE were taken as within-subject factors and group as between-subjects factor. Analyzing Romberg’s ratio the factor TARGET VISIBILITY was eliminated, therefore all other factors were included in the ANOVA. In some comparisons sphericity requirement was violated. Therefore, we report F-values with Greenhouse-Geisser correction but report degrees of freedom (df) uncorrected in order to show the factorial analysis design. Statistical comparisons were performed parametric unless stated otherwise.

Multi-factorial ANOVA with the above mentioned factors were performed. Significance levels of these tests were Bonferroni corrected for multiple testing. Statistical differences were regarded as significant for values p < 0.05. Error bars indicate SEM. Correlation analyses were performed using Spearman-Rho coefficient unless otherwise stated. The effects of visual deprivation on postural stability were determined by Romberg’s ratio computing PSS with the eyes closed/eyes open (22).

The mean VOR gain was reduced to 0.26 ± 0.04 indicating severe bilateral vestibulopathy. Mean peak SPV of caloric nystagmus was 4.5 ± 0.8°/s. oVEMP were recorded in 22 patients and 23 HC subjects; they were absent in 12 patients and revealed reduced amplitudes in the other 10 patients: peak amplitude differed significantly between groups (Mann–Withney U test, p = 0.003, median patients: 3.8 µV, median HC subjects: 6.95 µV). cVEMP were recorded in 22 patients and 23 HC subjects (median: 24.4 µV); they were absent in 17 patients and showed significantly reduced amplitudes in the other five patients (median 8.0 µV; p = 0.028). SVV did not show pathological tilts (>2.5°) and did not differ between patients and controls, neither during dynamic nor static SVV.

Generally, postural sway speed differed between paradigms (F(13,36) = 71.716, p = 0.001) and revealed an interaction of CONDITION × GROUP (F(13,36) = 2.559, p = 0.038). ANOVA showed a significant group difference (F(1,48) = 7.596, p = 0.008), i.e., BVF patients (n = 22) showed on average larger PSS than HC participants (n = 28).

There was a main effect for GROUP (F(1,48) = 6.08; p = 0.015) and TARGET VISIBILITY (Eyes open/eyes closed; EO/EC) (F(1,48) = 73.85; p < 0.001), i.e., PSS in patients and controls (solid platform, parallel feet, head upright) was significantly larger during eye closure than during eyes open. With the eyes open, the between-group analysis of PSS, however, did not reveal differences between both groups (p = 0.295). There was a significant interaction for TARGET VISIBILITY × GROUP (F(1,48) = 6.35; p = 0.015), i.e., PSS increased on eye closure more in patients than in controls. The difference for Romberg’s ratio (PSS ratio of EC/EO) between both groups (patients: 3.40 ± 0.0.44; controls: 2.49 ± 0.16) failed to reach significance level (T(48) = 1.96; p = 0.061). In short, Romberg’s ratio at baseline standing condition was larger in BVF.

An ANOVA on the PSS with the within-subject factors HEAD POSITION and TARGET VISIBILITY and the between-subject factor GROUP revealed main effects for GROUP (F(1,48) = 6.070, p = 0.017), TARGET VISIBILITY (F(1,48) = 67.340, p < 0.001), HEAD POSITION (F(2,47) = 6.086, p = 0.004), and an interaction of TARGET VISIBILITY × GROUP (F(1,48) = 6.635, p = 0.013) but no interaction of HEAD POSITION × GROUP (F(1,48) = 2.161, p = 0.124) or HEAD POSITION × TARGET VISIBILITY (F(2,47) = 2.974, p = 0.061) and no triple interaction (p > 0.9).

A separate ANOVA on PSS with the eyes open revealed no main effect of GROUP but a main effect of HEAD POSITION (F(2,47) = 9.845, p = 0.001): PSS increased in the head down (nose down) (p < 0.001) and head up (nose up) position (p = 0.001) with no difference between the gravity-dependent (up vs. down) head positions (Figure 2A). With the eyes closed, there was a main effect for GROUP (F(1,48) = 6.453, p = 0.014), HEAD POSITION (F(2,47) = 3.821, p = 0.027) but no interaction HEAD POSITION × GROUP (p > 0.4). Analyzing Romberg’s ratio (Figure 2B) there were main effects of GROUP (F(1,48) = 6.748, p = 0.012) and HEAD POSITION (F(2,47) = 7.758; p = 0.001) but no interaction of HEAD POSITION × GROUP (F(2,48) = 0.793; p > 0.45). In BVF patients, Romberg’s ratio was lower in the head up position (p = 0.033) and the head down position (p = 0.017). In HC, Romberg’s ratio was lower in the head down (p = 0.039) but not the head up position. Thus, gravity-dependent tonic head positions in the pitch plane increased postural sway in both groups (no interaction of HEAD POSITION × GROUP) but the increase in postural sway was larger in BVF on eye closure.

Analyzing PSS during HEAD SHAKING there was a trend for a main effect of GROUP (F(1, 48) = 3.887, p = 0.054), a main effect for TARGET VISIBILITY (F(1, 48) = 50.138, p < 0.001) but no interaction, i.e., higher PSS in the eyes closed condition (Figure 3A). Romberg’s ratio during HEAD SHAKING did not differ between groups (p > 0.8).

Comparing HEAD SHAKING to baseline (head erect, parallel stance) there was a main effect for GROUP (F(1,48) = 5.242, p = 0.026), TARGET VISIBILITY (F(1,48) = 72.111, p = 0.001) as well as for the interactions TARGET VISIBILITY × GROUP (F(1,48) = 4.335, p = 0.043) and HEAD SHAKING × TARGET VISIBILITY (F(1,48) = 4.303, p = 0.043) but no triple interaction HEAD SHAKING × TARGET VISIBILITY × GROUP (F(1,48) = 0.023, p > 0.8). Romberg’s ratio during HEAD SHAKING did not change to baseline condition (F(1,48) = 3.164, p = 0.082) and revealed no group-related differences (F(1,48) = 1.920, p = 0.172). In summary, postural unsteadiness during head shaking did not differ between groups.

Analyzing PSS during DUAL TASK, there was a main effect of TARGET VISIBILITY (F(1,48) = 32.827, p < 0.001) but no main effect of GROUP (p > 0.15) and no interaction of GROUP × TARGET VISIBILITY (p > 0.08), showing higher PSS for eyes closed condition (Figure 3B). Romberg’s ratio did not differ between groups (p > 0.13).

Compared to baseline there was no main effect for the DUAL TASK condition (F(1,48) = 0.105, p = 0.747) and no GROUP difference (F(1,48) = 4.002, p > 0.051) but larger PSS during eye closure [TARGET VISIBILITY (F(1,48) = 59.567, p = 0.001)]. There were interactions of TARGET VISIBILITY × DUAL TASK (F(1,48) = 12.765, p = 0.001) and TARGET VISIBILITY × GROUP (F(1,48) = 5.30, p = 0.026) but no DUAL TASK × GROUP interaction (F(1,48) = 1.843, p = 0.181). There was a main effect for Romberg’s ratio in GROUPs during DUAL TASK (F(1,48) = 4.783, p = 0.034) with larger ratios for BVF. Furthermore, Romberg’s ratio was lower for DUAL TASK condition (F(1,48) = 16.759, p < 0.001) while there was no interaction DUAL TASK × GROUP (p > 0.32). All in all, cognitive distraction in the dual task paradigm did not dissociate postural performance of patients and controls.

Four patients required postural assistance and were excluded from this analysis. Analyzing PSS during sensory deprivation by foam, there were main effects of GROUP (F(1,42) = 19.023, p < 0.001) and of TARGET VISIBILITY (F(1,42) = 133.218, p < 0.001) but no interaction—showing higher PSS for patients and for eyes closed condition (Figure 4A). Romberg’s ratio during FOAM did not differ between groups (p > 0.6). Comparing PSS to baseline, there was a main effect of the FOAM paradigm (F(1,42) = 138.025, p < 0.001) and of TARGET VISIBILITY (F(1,42) = 169.573, p < 0.001), with GROUP differences (F(1, 42) = 14.278, p < 0.001), significant interactions for TARGET VISIBILITY × FOAM (F(1,42) = 34.104, p < 0.001), for FOAM × GROUP (F(1,42) = 14.654, p < 0.001) and for TARGET VISIBILITY × GROUP (F(1,42) = 4.661, p < 0.037). Romberg’s ratio on FOAM showed no significant interaction FOAM × GROUP (p = 0.135) and no main effects of FOAM (p > 0.076) or group differences (p > 0.39). With the eyes open, patients showed larger PSS compared to HC (T(42) = −3.454, p = 0.003). Alltogether, postural sway of patients increased during proprioceptive deprivation by foam compared to controls, with additional significant enlargements in the patients in the absence of visual control (target visibility) on postural stability.

Eight patients required postural assistance and were excluded from this analysis. Analyzing PSS during tandem stance, there were main effects of GROUP (F(1, 37) = 6.164, p = 0.016) and of TARGET VISIBILITY (F(1,37) = 128.554, p < 0.001) but no interaction—showing higher PSS for patients and for the eyes closed condition (Figure 4B). Romberg’s ratio during tandem stance was higher for HC than for patients (T(36) = 2.141, p = 0.039). Compared to baseline, there was a main effect of TANDEM STANCE (F(1,37) = 164.119, p < 0.001), GROUP (F(1,37) = 5.149, p = 0.029) and TARGET VISIBILITY (F(1,37) = 169.792, p < 0.001), an interaction for GROUP × TANDEM STANCE (F(1,37) = 7.022, p = 0.012), an interaction for TARGET VISIBILITY × TANDEM STANCE (F(1,37) = 32.609, p < 0.001) but no triple interaction for GROUP × TANDEM STANCE × TARGET VISIBILITY (p > 0.2). For Romberg’s ratio, there was no interaction of TANDEM STANCE × GROUP (p > 0.12) and no main effects of GROUP (p > 0.42) and TANDEM STANCE (p > 0.27). Thus, patients showed larger postural instability on tandem stance than controls, irrespective of visual control (target visibility).

In a multiple regression using all conditions postural sway (PSS) explained 70% of the variance of the VOR gain (R² = 0.704, F(14,38) = 4.085, p = 0.001). The best—and the only significant—predictor for vestibular hypofunction (VOR gain) was the standing on foam condition with the eyes closed (R² = 0.358, F(1,38) = 20.642; p < 0.001). Accordingly, PSS of BVF patients increased in the foam paradigm on eye closure with the severity of vestibular impairment (VOR gain reduction, n = 18, r = −0.486, p = 0.041) (Figure 5) but not with disease duration (p > 0.055). In none of the experimental conditions, PSS (p = 0.557) or Romberg’s ratio (p = 0.558) correlated with disease duration.

From a clinical point of view, it is important to realize that postural control in BVF was indistinguishable from HCs as long as patients can use proprioceptive and visual feedback. Postural control of our BVF patients heavily depended on visual feedback as they showed a strong increase of postural sway on eye closure in all (even in the baseline) conditions compared to the age-matched HCs. This is in line with previous studies (22, 46–50). This increase is reflected by Romberg’s ratio which is used as an indicator of visual and proprioceptive contribution to postural stability (42). In the baseline condition, it was larger in BVF. This dissociates postural control in BVF from patients with vestibulo-cerebellar disorders, e.g., downbeat nystagmus whose increase in postural sway on eye closure (Romberg’s ratio) does not differ from HCs (39). Thus, postural behavior in postural ataxia in degenerative vestibulo-cerebellar disorders and BVF can be distinguished based on (i) baseline standing and (ii) Romberg’s ratio.

Weakening proprioceptive feedback on postural control by standing on foam showed much stronger postural imbalance (PSS) in BVF compared to controls, even with visual feedback support. This is in line with the enhanced proprioceptive dependence of postural control in chronic BVF (22, 65) and the destabilizing effect of additional diseases affecting proprioceptive feedback control on posture, e.g., in polyneuropathy (66). In combined deprivation of visual and proprioceptive feedback signals (foam condition with the eyes closed), some BVF patients required short postural assistance and needed to be excluded. This explains the high sensitivity (80%) of the “Romberg’s test on foam rubber” in BVF (67) as it provokes a stronger dependence of postural control on vestibular input. In healthy subjects, normal VOR is sufficient to maintain balance under these multisensory deprivations but BVF patients fall off the mattress if VOR is heavily impaired. Accordingly, there was an increase of postural sway the stronger VOR gain was reduced (Figure 5). Therefore, patients with severe BVF should be informed about increasing postural unsteadiness and risk of falls when they lack firm support beneath their feet or suffer from additional polyneuropathy.

Dual postural-cognitive task conditions have been used to study the relationship between attention and postural control. This relation is highly age-dependent (68): older subjects have higher attentional demands for postural control and show slower reaction times during combined postural-cognitive task (69). This leads to a higher risk of falling during standing and walking while talking (70). Our BVF patients could maintain postural control during attentional distraction in the dual task condition indistinguishable from age-matched HCs as long as visual and proprioceptive feedback was assured. This distinguishes BVF patients from the elderly (71), cerebellar patients (72) or patients with phobic postural vertigo (PPV) (73). The increased and inadequate use of sensory feedback in PPV patients suspected to cause their postural imbalance normalizes by distracting cognitive tasks (74, 75). This is not the case in BVF patients who largely rely on closed-loop mechanisms of postural control. This dependence on proprioceptive feedback may probably be even stronger as they showed a higher Romberg’s ratio in the dual task condition compared to controls. Unfortunately, severity of postural imbalance of our BVF patients did not allow us to investigate whether they maintain stance under more challenging dual task conditions (e.g., on foam).

Increased multisensory and motor postural demands (tandem stance) heavily destabilized BVF patients. Postural control of BVF patients was highly impaired compared to HCs, even when visual and proprioceptive input is used. Additional visual deprivation elicited a stronger postural imbalance compared to the HC group. Tandem stance requires multisensory integration, including vestibular input as visual and proprioceptive feedback is not sufficient to stabilize stance in BVF. This is in line with the concept of vestibular compensation in which postural control in vestibular failure is compensated by improving the sensory weight of unaffected sensory systems (24), i.e., they rely stronger on visual and proprioceptive feedback sources to maintain postural control (60). Accordingly, patients with uni-sensory deficit have a smaller risk of falling that patients with impairment of multiple sensory inputs required for postural control (66). It remains to be investigated whether the increased risk of falls in BVF (66) is related to increased co-contractions of antagonistic muscle groups as found in patients with cerebellar disease (72) and PPV (75).

Individual BVF patients may vary in the extent they exercise vestibular rehabilitation and accordingly they may vary in the magnitude of vestibular compensation. At the time of recording, vestibular compensatory mechanisms should have been established with respect to the average disease duration of our patients (3.1 years), if they developed at all. Therefore, we cannot specify how individual exercise influenced the variability of postural sway but can only refer to the group effects.