As noted above, the vestibular system detects head motion with five peripheral sensory organs: three semicircular canals that sense head rotation and two otolith organs that sense linear acceleration (Figure 1A, left). These sensory organs transmit head motion signals—via vestibular afferents within the VIII nerve—to central neurons within the vestibular nuclei to achieve essential functions (Figure 1A, right). When the sensory cells in the periphery (i.e. hair cells) are damaged, vestibular transduction is impaired, resulting in loss of vestibular functions. The most common known cause of hair cells damage is toxicity from systemic aminoglycoside drugs (e.g. gentamicin; Ward et al., 2013). In addition, those with congenital deafness due to loss of auditory hair cells can also lack properly functioning vestibular hair cells (Delmaghani and El-Amraoui, 2022). Vestibular prostheses aim to bypass these damaged sensors by directly detecting head motions and converting them into electrical stimulation of vestibular afferents. However, while these devices attempt to restore normal vestibular afferent activities during head motions, their development has only been guided to varying degrees by basic vestibular physiological principles listed below.

The goal of vestibular prostheses is to restore vestibular functions that are ultimately mediated by central vestibular pathways. In particular, the vestibulo-motor pathways that generate stabilizing eye and head movements are well characterized and provide easily quantified readouts of prosthesis performance. Vestibular afferents project from the vestibular periphery via the VIII nerve to the brainstem, where they directly synapse with neurons in the vestibular nuclei. Functionally, there are two main cell classes in the vestibular nuclei: those that help maintain gaze stability and those that contribute to postural stability and perception of self-motion (Figure 6; reviewed in Cullen, 2019). The circuits and functional roles of these two classes of central neurons, namely (i) VOR neurons and (ii) vestibular-only (VO) neurons, are each considered below.

VOR neurons are the middle link in the vestibular motor reflex pathways that ensure visual stability via the VOR (Figure 6A, top). This essential reflex provides visual stability during our everyday activities, generating eye movements that are equal in magnitude but opposite in direction to ongoing head movement. VOR neurons receive direct VIII nerve input and in turn project directly to extraocular motor neurons that control the eye muscles, constituting a very fast reflex pathway (~5 ms latency; Huterer and Cullen, 2002). These neurons both respond to vestibular stimulation and are sensitive to eye movements, consistent with their role in gaze stabilization. Although VOR neurons receive direct projections from both afferent types, their primary input is from regular afferents (Boyle et al., 1992; Goldberg et al., 1987; Highstein et al., 1987; Minor and Goldberg, 1991). Overall, the response dynamics of regular afferents are well matched to compensate for the delays within the VOR pathways (i.e. synaptic, neural, and muscle activation time) and the dynamics of the oculomotor plant to produce VOR eye movements that temporally match the time course of the head movement (Boyle et al., 1992; Goldberg et al., 1987; Minor and Goldberg, 1991). In addition, the spiking activities of regular afferents, and consequently VOR neurons, faithfully encode the detailed time course of head motion, a coding feature that is required for the generation of accurate compensatory VOR eye movements (Mackrous et al., 2020). Given that the VOR is highly robust and that VOR eye movements can be readily measured, to date, the vast majority of clinical and animal studies have focused on the VOR as the main behavioral output to quantify the efficacy of vestibular prosthetic stimulation.

VO neurons are the second primary class of neurons in the vestibular nuclei (Figure 6A, bottom). Their name reflects the fact that they respond to vestibular stimulation but are completely insensitive to eye movements. VO neurons serve a vital role in controlling postural stability via the vestibulo-spinal pathway. VO neurons send projections to the spinal cord—predominately via indirect pathways that include relay structures such as the interstitial nucleus of Cajal and reticular formation, as well as direct projections (Boyle et al., 1992; Goldberg et al., 1987; Highstein et al., 1987). In contrast to VOR neurons, VO neurons receive their primary input from irregular rather than regular afferents (Boyle et al., 1992; Goldberg et al., 1987; Highstein et al., 1987). The more dynamic responses of irregular afferents (i.e. larger gains and phase leads) are optimized to generate robust postural responses which must account for the larger inertia of the body relative to the eye (Minor and Goldberg, 1991; Peterson, 1998). Moreover, in contrast to VOR neurons, the majority of VO neurons do not faithfully encode the time course of head movements but instead optimize coding via temporal whitening—a property thought to contribute to ensuring postural control and perceptual stability during everyday activities (Mackrous et al., 2020). Similar to VOR, prosthesis-evoked vestibulo-spinal reflex (VSR) head movements are measured as the behavioral output of prosthetic stimulation (Mitchell et al., 2013; Wiboonsaksakul et al., 2023).

Importantly, in addition to mediating the VSR, VO neurons also make important contributions to ascending vestibular pathways that contribute to perception of self-motion and orientation (Figure 6B). In particular, VO neurons send projections via the posterior vestibulo-thalamic pathway, to the thalamus and cortex (reviewed in Cullen and Taube, 2017). This pathway targets the ventroposterior lateral thalamus of the thalamus, which in turn projects to multiple cortical and subcortical areas, including the parieto-insular vestibular cortex, which are thought to be involved in self-motion perception (reviewed in Cullen and Chacron, 2023). In contrast, the vestibular nuclei also send projections via the nucleus prepositus (McCrea et al., 1987) and subsequent relay nuclei to anterodorsal nucleus of the thalamus. In turn, the anterior vestibulothalamic pathway provides vestibular information to the head direction cell network, which contributes to regions involved in navigation and spatial memory including the retrosplenial cortex, entorhinal cortex, and the hippocampus (reviewed in Cullen and Taube, 2017).

While the VOR and VSR, as well as the vestibulo-thalamic pathways, receive differential input weighting from regular and irregular afferents, current vestibular prostheses deliver a single stimulation pulse (with large current amplitude) to recruit all afferents within the target semicircular canal. Consequently, prosthetic stimulation, unlike natural stimulation, is likely to evoke similar activation of both regular and irregular afferents during a specific head motion. To restore activities of both regular and irregular afferents to their (different) natural behaviors, the prosthesis must theoretically be able to independently stimulate each afferent group. While it has been proposed that such a directed approach could potentially be achieved by leveraging the lower recruitment threshold of irregular afferents (i.e. Wall et al., 2003), we speculate that it would be difficult to obtain precise afferent selectivity using currently available pulsatile stimulation-based technology.

As mentioned above, over the past decades of vestibular prosthesis development, the VOR has been extensively used to quantify performance (i.e. Figure 7A and B). In healthy animals and human participants, the VOR eye movements are generally quantified by calculating their gain and phase in response to sinusoidal head rotations across a range of frequencies. In healthy humans, the gain of the rotational VOR is ~1.0, meaning there is a compensatory eye velocity equal to the head velocity and in the opposite direction. Accordingly, to facilitate comparison with normal VOR function, studies of implanted nonhuman primates (Boutros et al., 2019b; Dai et al., 2011) as well as implanted patients (Boutros et al., 2019a; Perez Fornos et al., 2014) have measured the eye movements evoked by sinusoidal head rotations. In general, the VOR gain shows significant improvements when the prosthesis modulation is on vs. off (Figure 7C). This is accompanied by improved visual acuity during treadmill walking and head impulses when the prosthesis is turned on Ayiotis et al., 2022; Guinand et al., 2016; Starkov et al., 2020. Additionally, while oscillopsia is not directly correlated with VOR gain across unilateral vestibular loss, bilateral vestibular loss, and cerebellar patient populations (Palla et al., 2008; Geisinger et al., 2024; Morland et al., 1998; Grunfeld et al., 2000), patients in both the Johns Hopkins and Geneva-Maastricht clinical trials have reported reduced or absent oscillopsia symptoms when the prosthesis is turned on Ayiotis et al., 2024; Boutros et al., 2019a; Chow et al., 2021; van Stiphout et al., 2022. Taken together, these findings suggest that the prosthesis has significant potential to improve visual stability.

However, there remains significant room for improvement in evoking sufficient VOR via vestibular prostheses. The gains of the evoked VOR eye movements are not fully compensatory (i.e. gain<1; Figure 7C). Notably, there is considerable variability across both implanted nonhuman primates (Wiboonsaksakul et al., 2022) and patients (Boutros et al., 2019a), such that VOR gains typically range between ~0.2 and 0.6. One obvious approach to improve the VOR gain is to alter the mapping scheme between the detected head motion and stimulation modulation, such that the same head motion gives rise to greater modulation, e.g., by increasing the gain in Figure 5 (Perez Fornos et al., 2014). However, a disadvantage of this strategy is that it comes at the cost of a reduction in the dynamic range that the prosthesis can encode before entering cut-off/saturation. Fortunately, congruent extra-vestibular signals (i.e. visual, tactile, proprioception) can augment prosthetic stimulation to improve the VOR gain. For example, recent work in nonhuman primates has demonstrated greater VOR gain (though still not fully compensatory) from prosthetic stimulation in conjunction with physical rotation compared to prosthetic stimulation alone (Dai et al., 2012; Wiboonsaksakul et al., 2022). The effect is supra-additive, such that the gain during the combined condition was more than the sum of the gains in each separate condition alone.

A classic study by Suzuki and Cohen, 1964, demonstrated that electrical stimulation of the afferents evoked stereotyped postural reflexes in the neck and limbs, in addition to compensatory VOR eye movements. However, measuring and quantifying these evoked compensatory postural responses is inherently more complex relative to those of the evoked VOR. As a result, studies of the effects of vestibular prostheses on posture and balance are at a more nascent stage. One essential class of vestibular postural reflexes—the vestibulocollic reflex—moves the head on the body to stabilize the head in space as the body moves under it (reviewed in Goldberg and Cullen, 2011). Short pulse trains applied via the vestibular prostheses evokes not only VOR eye movements, but also head motion contralateral to the stimulated side in implanted monkeys (Mitchell et al., 2013). Notably, the amplitude of the evoked compensatory head movement increases with stimulation current amplitude and pulse rate (Mitchell et al., 2013), and the activation of the neck muscles driving these compensatory head movements has been measured via EMG in patients (Fornos et al., 2019).

The second essential class of vestibular postural reflexes are those which generate limb movements to stabilize the whole body in response to a head motion perturbation. Experiments in implanted monkeys have shown improvements in posture when stimulation is provided by a vestibular prosthesis encoding head motion via pulse rate modulation during gaze shifts (Figure 8A; Thompson et al., 2016). Moreover, in implanted patients, short pulse trains applied via the vestibular prostheses to each semicircular canal can generate whole-body postural responses that are biased toward the stimulated canal’s plane (Figure 8B; Phillips et al., 2013). Similarly, more recent studies in implanted patients have demonstrated that modulated vestibular prosthetic stimulation improves performance on both tests that directly assess vestibular function, such as the modified Romberg, and more general tests of posture and balance, including the timed walk and timed up and go tests (Figure 8C; Chow et al., 2021). While the latter two tests are not exclusively vestibular-dependent, the improvements in these tasks underscore the potential of vestibular prosthetic stimulation to enhance overall balance and posture in daily life.

Functional outcomes from animal studies and clinical trials highlight promising advancements in the development of a vestibular prostheses. However, most studies rely on VOR gains as a performance metric, which, though integral to the development of vestibular prostheses, do not always correlate with broader vestibular functions. For instance, prior studies have reported the uncoupling of VOR function from self-motion perception in trained dancers (Nigmatullina et al., 2015). VOR gains also poorly predict oscillopsia (Grunfeld et al., 2000) and fall risk (Dobbels et al., 2020) in patients with bilateral vestibular loss. Furthermore, individuals with vestibular implants can exhibit significant improvements in balance and gait despite suboptimal VOR gains (Chow et al., 2021). These findings underscore the need to assess higher-order vestibular functions, such as self-motion perception and fall risk, in prosthesis optimization. Additionally, current implantation criteria do not explicitly exclude patients based on co-morbidities, except those that would preclude surgery (van de Berg et al., 2020). However, evidence suggests chronic conditions, such as diabetes, worsen self-motion perception thresholds, postural stability, and fall risk over time (La Scaleia et al., 2024). Incorporating co-morbidity assessments into patient selection criteria could significantly enhance future clinical outcomes.

In clinical trials of vestibular prostheses, the onset of baseline stimulation initially induces an asymmetry in vestibular input between the left and right ears that is interpreted as a head movement, and thus causes nystagmus and postural instability (Boutros et al., 2019a; Chow et al., 2021; Lewis et al., 2013b). Patients then adapt to this unilateral baseline stimulation rate over a relatively rapid time course (Figure 9A). It has generally been assumed that adaptation to this asymmetric vestibular input is mediated by central pathways, notably the commissural vestibular pathways (Figure 9B, green; Chiang et al., 2011; Della Santina et al., 2007; Merfeld et al., 2006; Merfeld et al., 2007). Indeed, experiments in rats and mice suggest plasticity in commissural pathways along with cerebellar and cortical vestibular areas that produces changes in functional connectivity within minutes to hours following unilateral vestibular loss, a condition similar to the onset of baseline stimulation with a vestibular prosthesis, which induces asymmetry in the input to central vestibular pathways (Cirelli et al., 1996; Grosch et al., 2021; Kai et al., 2022; Kitahara et al., 1997; Zwergal et al., 2016). In this view, adaptation is beneficial and results in a new setpoint, from which a unilateral implant can up- or down-modulate to encode head motion in two opposing directions.

Importantly however, recent studies exploring the effect of prosthetic stimulation on central pathways have instead established that adaptation to prosthetic stimulation is mediated earlier in the pathway, at the afferent-vestibular nuclei synapse (Figure 9B, red; Mitchell et al., 2016; Mitchell et al., 2017). Specifically, experiments in implanted monkeys have revealed that prosthetic baseline stimulation actually leads to the rapid and marked (>50%) reduction in efficacy of the afferent-vestibular nuclei synapse, which in turn substantially reduces efficacy of central vestibular pathways within minutes after stimulation onset (Figure 9C; Mitchell et al., 2016; Mitchell et al., 2017). Adaptation at this synapse is fundamentally problematic because, instead of resulting in a new setpoint (as for central adaptation), it attenuates all prosthetic inputs, including motion-modulated stimulation encoding the head movement information that the prosthesis is trying to restore to the brain. This maladaptive prosthesis-induced reduction in synaptic efficacy occurs at the level of afferent input to both the vestibular nuclei neurons that generate the VOR (Mitchell et al., 2016) and those that control VSR (Mitchell et al., 2017). Rapid complementary plasticity within inhibitory commissural nuclei network contributes to lessen this maladaptive effect of stimulation. However, this rapid plasticity only partially improves behavioral VOR and VSR performance (Mitchell et al., 2016; Mitchell et al., 2017). Critically, only a few minutes of baseline prosthetic stimulation causes reductions in both VOR and VSR behavioral responses that can persist for more than 8 hr in absence of head motion or further prosthetic stimulation (Mitchell et al., 2016; Mitchell et al., 2017). Overall, we speculate that the maladaptive reduction in the efficacy of the afferent-vestibular nuclei synapse is a key reason why implanted patients show incomplete restoration of function (Boutros et al., 2019a). How to effectively reduce this undesirable early adaptation and what stimulation parameters—i.e., current amplitude (at sub- or peri-threshold levels) and pulse shapes—contribute to this maladaptation remain open questions for future investigations.

The adaptation at the afferent-vestibular nuclei synapse described above is fundamentally problematic. This is because it will not only attenuate central neuron responses to baseline prosthetic stimulation but also to head motion modulated stimulation. As a result, this maladaptation reduces the efficacy of head motion information transfer to the reflex pathways as well as those that contribute to self-motion perception (Figure 1A). This reduction in efficacy is likely due, at least in part, to the synchrony that is induced across vestibular afferents by prosthetic stimulation. The spiking activity of stimulated afferents synchronizes with prosthetic-driven pulse trains because their action potentials are tightly linked to the applied pulses (Mitchell et al., 2016; Steinhardt et al., 2022). Importantly, this synchrony contrasts with what is observed across afferent population responses to natural head movement stimuli (Dale and Cullen, 2013; Yu et al., 2014). In vitro studies have demonstrated that repeated pulsatile stimulation of vestibular afferents induces long-term depression at the afferent-vestibular nuclei synapse (McElvain et al., 2010), thereby providing a mechanism for this observed reduction in central pathway efficacy (Mitchell et al., 2016; Mitchell et al., 2017). Correspondingly, it has been established that cochlear implant stimulation induces artificial synchrony in the firing of auditory afferents (Babalian et al., 2003; Hu et al., 2010; Sachs et al., 1983; Wu et al., 2016) as well as their target neurons in the cochlear nucleus (Babalian et al., 2003). Furthermore, as in the vestibular system, the synchrony induced by cochlear implant stimulation reduces the efficacy of central auditory pathways over time (Hu et al., 2010; Runge et al., 2018).

Ultimately, overcoming the maladaptive effect of vestibular prosthesis induced afferent synchrony will be essential for improving these devices. One general strategy that has been used for cochlear implants is high-frequency subthreshold stimulation. In this strategy, stimulation is delivered at a frequency well beyond that of the afferents’ maximum firing rate, typically 5 kHz (e.g. Babalian et al., 2003; Litvak et al., 2003). Two main approaches have been used to deliver high-frequency stimulus. The first uses a high-frequency as the primary carrier for amplitude-modulated stimulus pulses (Babalian et al., 2003; Hu et al., 2010; Rubinstein et al., 1999), while the second adds a high-frequency, subthreshold ‘buzz’ on top of lower-frequency modulated stimuli (Litvak et al., 2003). The goal of both approaches is to exploit the afferents’ fixed refractory period to produce more variability in response timing across afferents. A second strategy that has been successful in improving the performance of cochlear implants is the use of altered pulse waveforms, most notably delayed pseudomonophasic (DPM) asymmetric stimulus pulses. In particular, improvements in auditory thresholds have been reported for DPM stimulation characterized by a brief, high-amplitude cathodic phase followed by a long, low-amplitude anodic phase (Quass et al., 2020; van Wieringen et al., 2008). We speculate that these improvements are also mediated via a reduction in synchrony and that combining these two general approaches (i.e. high-frequency subthreshold stimulation methods and novel pulse shapes) will further reduce the overall synchrony of vestibular afferent responses, in turn improving vestibular prosthesis performance.

Finally, it is noteworthy that both vestibular prostheses and cochlear implants are designed to directly stimulate afferents in their respective systems, bypassing the hair cells of the sensory epithelium which are no longer functional (reviewed in Shibata et al., 2011). Thus, these devices operate on the assumption that their respective afferents remain fundamentally healthy. However, neurophysiological studies of the auditory system have shown that hair cells provide both neurotrophic support and excitatory inputs to afferents. Accordingly, the loss of auditory hair cells not only significantly reduces the sensitivity of afferents fibers (e.g. Hartmann et al., 1984; Kiang and Moxon, 1972; Suthakar and Liberman, 2021) but also results in afferents fiber demyelination and ultimately loss of afferent fiber and their somas within the spiral ganglia over months to years (Kurioka et al., 2016; Nadol, 1997; Spoendlin, 1975; Webster and Webster, 1981). Likewise, work to date focusing on changes in the vestibular periphery following hair cell loss has reported a similar reduction in vestibular afferent somas (Ishiyama et al., 2004). However, promisingly for vestibular as well as cochlear implantation, electrical stimulation of auditory afferents via a cochlear prosthesis has been shown to protect auditory afferents from further degeneration (reviewed in Shibata et al., 2011). Thus, we speculate that implanting as early as possible after vestibular hair cell loss may help improve patient outcomes. Furthermore, a recent study in monkeys found that vestibular implants can effectively evoke eye movements, even when the prosthesis was implanted 10 years after bilateral labyrinthectomy, demonstrating that vestibular implants are likely to be effective even if some afferents have been lost (Wiboonsaksakul et al., 2023). An important area of research moving forward will be to improve our understanding of how to optimally preserve healthy afferent function following the loss of hair cells in both systems (reviewed in Shibata et al., 2011).

Vestibular prosthesis development has been largely guided by our fundamental knowledge of the anatomy and physiology of vestibular afferents. Yet, as reviewed above, the vast majority of both clinical trials and animal studies have employed mappings that did not consider how vestibular afferents respond during natural everyday activities, or were implemented based on the limited knowledge available at the time. In general, while prior studies have accounted for the fact that afferents modulate their firing rates around a resting baseline and show cut-off/saturation nonlinearities (see Figure 5, top), they did not account for the high-pass response dynamics of the afferents (i.e. increasing gain and phase lead as a function of frequency; Figure 3B). Instead, the prevailing approach in ongoing clinical trials has been to simply map instantaneous head velocity to a specific pulse rate (i.e. flat gain and no phase lead). Because natural head motion can reach frequencies as high as 20 Hz (Carriot et al., 2014; Carriot et al., 2017), it is essential to consider the high-pass dynamics of vestibular afferents. For example, the phase of VOR eye movements evoked by head motion remains compensatory across the physiological range of motion even at frequencies exceeding 20 Hz (Huterer and Cullen, 2002; Ramachandran and Lisberger, 2005). In theory, without the phase compensation provided by the high-pass dynamics of afferents, VOR eye movements would increasingly lag head motion reaching a lag of >50 degrees for higher frequencies within physiological range. This is because an increase in afferent phase lead as a function of frequency is required to compensate for the fixed delays in the VOR pathways (e.g. synaptic transmission and muscle activation, see Discussion in Huterer and Cullen, 2002).

Indeed, a recent study has confirmed the theoretical scenario above (Wiboonsaksakul et al., 2022). Conventional mappings that did not account for high-pass dynamics of vestibular afferents resulted in decreasing gain and increasingly lagging phase of the evoked VOR, while biomimetic mappings that incorporated high-pass tuning resulted in more robust gain and more accurate phase of the evoked VOR (Figure 5, bottom; Wiboonsaksakul et al., 2022). Further exploration of the parameter space revealed that more extreme tunings were not beneficial in that they produced saturation and unnatural phase advances. Moreover, mapping strategies biased toward the response dynamics of regular afferents yielded better VOR performance than a mapping strategy based on the response dynamics of irregular afferents (Figure 10, blue and purple vs. red). This finding is consistent with prior experimental findings demonstrating that regular afferents make a more significant contribution to VOR pathway than do their irregular counterparts (Boyle et al., 1992; Goldberg et al., 1987; Highstein et al., 1987; Minor and Goldberg, 1991). Future work in vestibular prosthesis development should focus on the relative improvements in performance produced by different biomimetic mappings for different vestibular motor pathways (e.g. regular mapping for VOR vs. irregular mapping for postural control and perception).

To date all but a handful of studies have focused on vestibular prostheses that aim to restore semicircular canals rather than otolith function. As described below, this semicircular canal focused approach is currently well warranted because of the technical challenges of implanting an otolith prosthesis. However, since natural head movements generally comprises both rotational and linear motion, inputs from both the semicircular canal and otolith are typically processed simultaneously and integrated for both behavior and perception (reviewed in Angelaki and Cullen, 2008). Thus, an important future direction for improving functional outcomes in patients with bilateral vestibular loss is restoring otolith as well as semicircular canal input. The otolith organs, however, are a more challenging target for prosthesis development. Notably, although it is straightforward to activate semicircular canal hair cells in a directionally precise manner (i.e. Figure 2), this is not the case for the utricle and saccule.

A key feature of the otoliths is that their hair cells are distributed over the neuroepithelia surface in a systematic manner that comprises a directional tuning map, with the axis of greatest sensitivity perpendicular to the curved line across the otolith organs (Figure 11A, orange lines). This hair cell alignment allows the utricle and saccule to sense linear motion in the horizontal and vertical plane, respectively, and together detect linear forces in three dimensions. For any given head motion direction, different hair cells are excited or inhibited to many varying degrees depending on how their axis of sensitivity aligns with the head motion. Consequently, stimulating the otoliths in a directionally specific manner requires a great number of electrodes that can selectively and independently stimulate a very focused area of the organ. Despite these challenges, the Della Santina laboratory has recently implanted a multichannel planar stimulating array in the otoliths of chinchillas (Figure 11A, gray circles). Stimulation of the otoliths using the implant successfully generated torsional and vertical eye movements consistent with vestibulo-ocular responses to tilt and translation (Hageman et al., 2020), indicating successful stimulation of sections of the otolith sensitive to different axes of head motion. However, precisely positioning the arrays within the otoliths is difficult, requiring surgical approaches that risk damaging other structures. Furthermore, current spread across an implanted otolith limited both the magnitude and spatial selectivity of the device. Thus, significant work remains to be done in developing an otolith prosthesis.

Another research direction that could restore vestibular information to the brain in a more naturalistic way is the use of direct current (DC) stimulation, which has gained popularity in recent years for use in transcutaneous and transcranial stimulation (reviewed in Aplin and Fridman, 2019; Fridman and Della Santina, 2013). Unlike pulsatile stimulation that can only evoke more firing in the targeted nerve (and thus requires baseline stimulation rate in order to encode motion in opposing directions), DC stimulation can both increase and decrease the firing. However, DC stimulation is not generally used due to its electrochemical reactions that are harmful to tissue when used with traditional metal electrodes. Rather than directly stimulating tissue with a metal electrode, safe DC stimulators make use of ionic currents in saline with a switching mechanism that allows for continuous DC flow while also maintaining charge balance at the tissue interface (Figure 11B). In the context of the vestibular prosthesis, safe DC stimulation of the semicircular canals has been shown to evoke robust VOR eye movements consistent with the canal plane being stimulated in chinchillas (Aplin et al., 2019). Specifically, cathodic and anodic currents evoked eye movement in the excitatory and inhibitory (opposite) directions, respectively. Importantly, when compared to traditional pulsatile stimulation, DC stimulation provided a larger dynamic range of the evoked eye movements due to the ability to also inhibit activity. Additionally, electrophysiology (Manca et al., 2019) and modeling (Steinhardt and Fridman, 2021) studies have both shown that DC stimulator modulates afferent firing rate in a manner that also maintains their firing properties (i.e. irregularity and asynchronous behavior of firing). Critically, this preservation of firing properties would result in a more naturalistic activation of the afferent population, which could reduce any adverse effects resulting from sustained pulsatile stimulation (e.g. Figure 9).

Because DC stimulation does not modulate afferent activity in a manner that directly links stimulation pulses to the afferent firing rate (like pulsatile stimulation does), computing the current modulation that yields the desired naturalistic encoding requires knowledge of the relationship between stimulation and neuronal output. Fortunately, recent studies, quantifying afferent responses to transcranial DC stimulation of the vestibular system (i.e. galvanic vestibular stimulation), have shown that a transfer function between current delivered and afferent firing can be readily constructed and accounted for to evoke desired firing and head motion encoding within the afferents (Forbes et al., 2020; Forbes et al., 2023; Kwan et al., 2019; Figure 11C and D). Critically, this finding underscores how neurophysiological insights can directly guide translational and clinical applications. Ultimately, safe DC stimulation may become a stimulator of choice for vestibular prosthesis (alone or together with pulsatile stimulation) as it provides a more naturalistic way to control neural firing. Currently, the size and complexity of the required hardware limit the application of DC stimulation in the clinic, nevertheless, this remains an exciting area of active research.

The unique properties of the vestibular system have driven rapid progress in the development of vestibular prostheses in recent years, both in animal models and clinical trials. These devices have shown promising functional outcomes, significantly improving the quality of life for patients who previously could not compensate through rehabilitation exercises alone. Such patients would otherwise continue to experience unsteadiness while walking or standing, blurred vision (oscillopsia) during head movements, or worsening instability in low-light conditions or uneven terrain. However, to further advance this technology, it is essential to fully incorporate our current understanding of vestibular neurophysiology into both prosthesis design and clinical application. A key step forward lies in leveraging natural afferent encoding strategies to tailor treatments to each patient’s specific needs. Personalized prosthesis mappings could be developed to target the most debilitating symptoms. For instance, patients primarily concerned with visual stability could benefit from a regular afferent mapping, while those focused on improving postural stability might see better results with an irregular afferent mapping (Figure 5). Nevertheless, it is also important to recognize that since current pulsatile prostheses will coincidentally stimulate all afferent types, addressing one problem with a specific mapping could potentially impair other functions that rely on a different mapping. While the full potential of this personalized approach has yet to be explored, the selection of afferent encoding during a patient’s initial ‘device activation’ visit may become a key aspect of prosthesis customization. Additionally, we note that future advancements in surgical techniques, though outside the scope of this review, will play a pivotal role in further enhancing vestibular prosthesis outcomes, as highlighted in a recent review by Stultiens et al., 2023.

Aside from furthering and leveraging our understanding of vestibular neurophysiology, gaining even more insights into how neural populations, along each stage of vestibular processing, respond to prosthetic stimulation will be key for addressing current limitations of prosthesis designs and clinical applications. In particular, future efforts should explore how higher-order brain areas, including the cerebellum, ascending thalamocortical pathways (i.e. the anterior head direction network and posterior thalamic pathway), and cortex respond and adapt to prosthetic inputs. Once the innovative engineering and the foundational neurophysiology are fully integrated, we believe the field will move forward and ultimately improve patient outcomes.