Nonhuman primate vestibuloocular reflex responses to prosthetic vestibular stimulation are robust to pulse timing errors caused by temporal discretization

Abstract

Electrical stimulation of vestibular afferent neurons to partially restore semicircular canal sensation of head rotation and the stabilizing reflexes that sensation supports has potential to effectively treat individuals disabled by bilateral vestibular hypofunction. Ideally, a vestibular implant system using this approach would be integrated with a cochlear implant, which would provide clinicians with a means to simultaneously treat loss of both vestibular and auditory sensation. Despite obvious similarities, merging these technologies poses several challenges, including stimulus pulse timing errors that arise when a system must implement a pulse frequency modulation-encoding scheme (as is used in vestibular implants to mimic normal vestibular nerve encoding of head movement) within fixed-rate continuous interleaved sampling (CIS) strategies used in cochlear implants. Pulse timing errors caused by temporal discretization inherent to CIS create stair step discontinuities of the vestibular implant’s smooth mapping of head velocity to stimulus pulse frequency. In this study, we assayed electrically evoked vestibuloocular reflex responses in two rhesus macaques using both a smooth pulse frequency modulation map and a discretized map corrupted by temporal errors typical of those arising in a combined cochlear-vestibular implant. Responses were measured using three-dimensional scleral coil oculography for prosthetic electrical stimuli representing sinusoidal head velocity waveforms that varied over 50–400°/s and 0.1–5 Hz. Pulse timing errors produced negligible effects on responses across all canals in both animals, indicating that temporal discretization inherent to implementing a pulse frequency modulation-coding scheme within a cochlear implant’s CIS fixed pulse timing framework need not sacrifice performance of the combined system’s vestibular implant portion.

NEW & NOTEWORTHY Merging a vestibular implant system with existing cochlear implant technology can provide clinicians with a means to restore both vestibular and auditory sensation. Pulse timing errors inherent to integration of pulse frequency modulation vestibular stimulation with fixed-rate, continuous interleaved sampling cochlear implant stimulation would discretize the smooth head velocity encoding of a combined device. In this study, we show these pulse timing errors produce negligible effects on electrically evoked vestibulo-ocular reflex responses in two rhesus macaques.

INTRODUCTION

Cochlear implants (CIs) intended to partially restore auditory sensation to individuals suffering from profound sensorineural hearing loss typically use a linear array of electrode contacts to electrically stimulate the cochlear nerve at different locations along the cochlea’s tonotopic axis. Early signal-processing schemes encoded individual frequency band waveforms via continuous-analog modulation of electrical current delivered concurrently to each electrode in the intracochlear array [reviewed by Loizou (1998) and Wilson et al. (1993)]. Experience ultimately revealed that simultaneous delivery of current across multiple channels can produce broad activation of spiral ganglion neurons, causing distortion of perceived auditory cues. To reduce current spread to nontarget cochlear nerve fibers, Wilson and colleagues developed the continuous interleaved sampling (CIS) strategy to asynchronously deliver pulses to each intracochlear contact (Wilson et al. 1991, 1993). In CIS, each electrode delivers stimulus pulses at a fixed pulse frequency amplitude modulated to encode power in the acoustic signal’s spectral band that electrode represents. To avoid overlapping stimulus current fields, CIS stimuli are presented one electrode at a time, in a nonoverlapping, “round-robin” fashion. This temporal compartmentalization of stimulation delivered via different electrodes yielded such an improvement in speech recognition over continuous-analog stimulation that nearly all CI systems now use a variation of the CIS approach, augmented by additional signal-processing schemes intended to enhance auditory perception by encoding the sound waveform’s fine structure or fundamental frequencies (Wilson and Dorman 2008; Wilson et al. 1993).

Efforts to develop vestibular implants (VIs) to treat bilateral vestibular hypofunction have lagged CI development, but ample evidence in animals and first-in-human trials over the past 15 years indicates that prosthetic electrical stimulation of vestibular afferent neurons can partially restore semicircular canal sensation of head rotation and the eye-, head-, and posture-stabilizing reflexes normally driven by vestibular sensation (Dai et al. 2011b, 2013; Della Santina et al. 2007; Golub et al. 2014; Gong and Merfeld 2000; Guinand et al. 2015; Nie et al. 2013; Perez Fornos et al. 2014; Phillips et al. 2015a, 2015b; van de Berg et al. 2015; Wall et al. 2007; reviewed by Guyot and Perez Fornos 2019). Ideally, a VI system based on this approach would be integrated with a CI system and use the same hermetically sealed stimulator retrofitted to include electrode arrays for both the cochlea and labyrinth, providing a means to simultaneously treat loss of both auditory and vestibular sensation. Such a combined system could provide a much-needed safety net during cochlear or vestibular implantation in patients who only have a deficit of either hearing or vestibular sensation: A surgeon could bank the unused electrode array in the mastoid cavity or temporalis muscle and then, should a new loss become apparent postoperatively, reoperate and insert the array to help recover function. Apart from the clinical advantages, basing VI system development on existing CI systems can also facilitate regulatory compliance and production while reducing the otherwise prohibitive costs of developing a new implantable stimulator from the ground up. Reflecting these factors, development of a VI for clinical treatment of bilateral vestibular hypofunction has so far relied on use of modified CI stimulators (Fridman and Della Santina 2012; Golub et al. 2014; Guyot and Perez Fornos 2019; Perez Fornos et al. 2014; Valentin et al. 2013).

Unfortunately, the cochlea and vestibular labyrinth employ different neural codes to represent sound and head movement, respectively. Whereas normal cochlear nerve activity is reasonably well emulated by a CI electrode array using an amplitude-modulated CIS stimulation paradigm with fixed pulse rates on electrodes that deterministically take turns firing in a round-robin sequence, primary vestibular afferents effectively use a rate code in which instantaneous spike rate represents head angular velocity (for the semicircular canals) or gravitoinertial linear acceleration (for the utricle and saccule). In other words, cochlear nerve fibers respond well to pulse amplitude modulation (PAM) at a fixed stimulation rate, but pulse frequency modulation (PFM) with or without PAM seems more appropriate for stimulation of vestibular nerve fibers. Of course, neural coding in both the cochlea and the vestibular labyrinth is more complex than CIS, PAM, or PFM because of the presence of large populations of hair cells and neurons with different sensitivities and temporal dynamics. Moreover, elaborations of the CIS coding strategy can vary from a simple rule mandating that electrodes always fire in the same fixed-rate sequence using a PAM code. Nonetheless, the fact remains that CIs typically do not implement PFM coding, which is the closest fit to the natural neural code implemented by vestibular primary afferent neurons.

Using PFM, with or without an additional PAM component, to encode head rotational velocity via prosthetic electrical stimulation has been successful in many preclinical (Chiang et al. 2011; Dai et al. 2011a, 2011b, 2013; Davidovics et al. 2013; Della Santina et al. 2007; Fridman et al. 2010; Gong and Merfeld 2000, 2002; Merfeld et al. 2007; Nie et al. 2013; Phillips et al. 2015a) and pilot clinical studies (Nguyen et al. 2016) of VI technology. Those studies used smooth modulation of stimulus pulse frequency above/below an adapted electrical stimulus rate to mimic the spike rate modulations about the spontaneous discharge rate of primary vestibular afferent fibers (Baird et al. 1988; Sadeghi et al. 2007). Although other studies have shown that modulation of pulse amplitude can successfully evoke compensatory vestibuloocular reflex (VOR) responses (Guinand et al. 2015; Nie et al. 2013; Perez Fornos et al. 2014; Phillips et al. 2015a, 2015b; van de Berg et al. 2015), integrating PFM stimulus-encoding methods into a CIS signal-processing framework would facilitate development of a future CI-VI combined device.

Merging a PFM-based VI and a CIS-based CI creates a trade-off because the temporal discretization necessary for the round-robin CIS approach prevents the smooth modulation of pulse frequency, creating errors in patterned timing of pulsatile stimuli. Those errors unavoidably distort the interpulse interval timing of PFM-modulated stimulus pulse trains used to encode head velocity; however, we hypothesized that they would produce negligible effects on the gain, phase, and misalignment of prosthetically evoked eye movements because rate codes generally, and the neural signaling and extraocular motor pathways mediating the VOR in particular, low-pass filter PFM-encoded stimuli in a way that smooths out the effects of temporal discretization.

To test this hypothesis, we created two head velocity-to-pulse rate maps to encode head motion: a smooth PFM (sPFM) map that continuously modulates stimulus pulse frequency with head velocity and a map corrupted by temporal discretization errors [discretized PFM (dPFM)] typifying errors that would occur in a CI-VI using CIS. We assayed responses to both mappings using a series of virtual head velocity waveforms (i.e., prosthetic vestibular stimulation under head-fixed conditions) and assayed evoked three-dimensional (3-D) VOR responses in two rhesus macaques.

METHODS

Surgical procedures.

Two female rhesus macaques (animals RhF20124B and RhF060738G, ~6 kg) were used for all experiments, which were performed in accordance with a protocol approved by the Johns Hopkins Animal Care and Use Committee. Methods of head cap assembly for head fixation and dual scleral magnetic search coil implantation in primates have been described previously (Chiang et al. 2011; Migliaccio et al. 2004; Minor et al. 1999; Robinson 1963). In brief, using general inhalational anesthesia (1.5–5% isoflurane) and sterile conditions, a head cap constructed from poly(ether ether ketone) was attached to the skull using poly(methyl methacrylate) and titanium bone screws. Two search coils were fabricated from Teflon-coated steel wire (Cooner Wire, Chatsworth, CA). Wire leads were tightly twisted to minimize artifacts induced in the magnetic search coil system. Connectors were fashioned out of 1-mm pitch pin headers and routed out of the animal’s head cap. Both coils were sutured to the sclera of one eye (left eye for animal RhF20124B and right eye for animal RhF060738G), with one coil implanted around the iris and the other positioned approximately orthogonal to the first. Postoperative procedures included treatment with analgesics and antibiotics for ~72 h postoperation.

After characterization of each animal’s natural vestibular function, an electrode array was implanted into the left labyrinth using a transmastoid approach analogous to that used for cochlear implantation, but with a transcutaneous connector embedded in the animal’s head cap (Dai et al. 2011a; Davidovics et al. 2013). Animal RhF20124B was implanted with an electrode array fashioned from pairs of Teflon-insulated platinum-iridium wire (Cooner Wire) for each stimulating and reference electrode. This electrode array used six stimulating electrodes [electrodes 0 and 1 (E0 and E1, respectively) in the left anterior canal, E2 and E3 in the left horizontal canal, and E5 and E6 in the left posterior canal] and two reference electrodes (E4 in the common crus and E7 near temporalis muscle). Animal RhF060738G was implanted with an electrode array designed using 3-D reconstructions of computed tomography images of rhesus macaque vestibular labyrinths (Chiang et al. 2011). This array, which was fabricated using a process similar to production of CI arrays and intended to facilitate surgical handling by eliminating the need to manipulate multiple interconnect wires individually, comprises nine stimulation electrodes (E1–3 in the left posterior canal, E4–6 in the left horizontal canal, and E7–9 in the left anterior canal) and two reference electrodes (E10 located near the temporalis musculature and E11 in the common crus), all embedded within a silicone carrier, with electrode contacts and leads made from Teflon-insulated 90:10 platinum-iridium wires flamed to ball electrodes at their ends.

Following unilateral electrode implantation, each animal underwent bilateral intratympanic gentamicin treatment using a modified procedure typically performed in humans (Carey et al. 2002), adjusted with the use of general inhalational anesthesia (1.5–5% isoflurane) for ~30 min with the injected ear reoriented upward to promote diffusion into the inner ear. Each treatment consisted of ~0.5 ml of 26.7 mg/ml buffered gentamicin solution injected through the eardrum into the middle ear. VOR responses were assayed ~3 wk after treatment, and gentamicin injections were repeated until VOR gains across all treated canals were <10% of normal characterized gains. Typically, this required two to three injections per ear for each animal of this study. Reinjections such as this are not uncommon in human patients treated with gentamicin to reduce symptoms associated with Ménière’s disease (Nguyen et al. 2009).

Eye movement recording.

The search coil system used to record 3-D eye movements has been described in detail previously (Migliaccio et al. 2004; Robinson 1963). Each animal was seated in a plastic chair and restrained by the implanted head cap. The coil system superstructure comprised three pairs of field coils generating magnetic fields along the x (nasooccipital, +nasal)-, y (interaural, +left)-, and z (superoinferior, +superior)-axes of the monkey’s restrained head and oscillated at 79, 53, and 40 kHz, respectively. Each animal’s head was reoriented about the +y-axis either +15° (animal RhF20124B) or +2° (animal RhF060738G) to align the animal’s horizontal canal plane [as estimated using computed tomography scans acquired after head cap implantation (Della Santina et al. 2005)] with the Earth-horizontal plane. Currents induced on scleral search coils were demodulated to produce voltages proportional to the angles between each coil and the three magnetic fields. All signals were filtered using an analog eight-pole Butterworth low-pass filter with a corner frequency of 100 Hz and sampled at 1 kHz. Misalignments between positioning of the two coils were corrected using an algorithm for computing instantaneous angular position of the implanted dual coil pair when they are not exactly orthogonal (Tweed et al. 1990).

Data analysis and statistics.

Rotation vectors describing angular position in x (“roll”)-, y (“pitch”)-, and z (“yaw”)-coordinates were calculated from raw search coil data using standard techniques (Migliaccio et al. 2004; Robinson 1963; Straumann et al. 1995). The 3-D angular velocity vectors were computed from the rotation vectors (Haslwanter 1995), transformed into components along the mean anatomic axes of the left anterior-right posterior (LARP), right anterior-left posterior (RALP), and left horizontal-right horizontal (LHRH) coplanar canal pairs (Fig. 1), and smoothed using a running spline interpolation filter. All polarities are expressed using right-hand rule conventions. A custom semiautomatic algorithm detected quick phases of elicited vestibular responses, blanked out the detected region, and connected over blanked regions using spline interpolation. Responses along each 3-D component were separately cycle averaged after removing cycles corrupted by blinks.

Fig. 1.

Head-fixed coordinate system used to describe three-dimensional (3-D) vestibuloocular reflex (VOR) responses. The +x (“roll,” nasooccipital)-, +y (“pitch,” interaural)-, and z (“yaw,” superoinferior)-coordinates are mutually orthogonal stereotaxic axes. During data analysis, 3-D angular position and velocity data are computed in skull coordinates and transformed into a semicircular canal coordinate system. During data collection, each animal’s head is repositioned with either a +15° (animal RhF20124B) or a +2° (animal RhF060738G) pitch reorientation to align each animal’s positive left horizontal-right horizontal (+LHRH) plane with the Earth-horizontal plane. LARP, left anterior-right posterior; RALP, right anterior-left posterior. Reprinted from Dai et al. (2013) by permission from Springer Nature.

To compare VOR performance for each mapping, we computed peak eye velocity, phase lead, and misalignment for each stimulus condition. The peak eye velocity was determined by finding the largest-magnitude eye velocity about the correct canal axis with appropriate polarity. For example, excitation of the left horizontal ampullary nerve branch should encode head rotation to the left (a positive, right-hand rule rotation about the +LHRH axis by the coordinate frame convention in Fig. 1) and produce a compensatory eye movement to the right (negative by convention). In contrast, excitation of the left posterior or left anterior canal (as should occur during left-hand rule head rotations about the +RALP and +LARP axes in Fig. 1, which are assigned a negative polarity by convention) should produce a positive slow-phase eye velocity in the intended plane. The componentwise angular velocity magnitudes at this peak determined the VOR response vector. This vector was normalized by its length to produce a unit vector axis of rotation for the 3-D VOR response. A positive phase lead was defined where the VOR response leads the inverse of head velocity. Finally, misalignment was defined as the angle between the unit vector VOR axis of rotation and the ideal canal axis. All values are reported as means ± SD.

Statistical analysis was performed in MATLAB (MathWorks, Natick, Massachusetts) and R (RStudio, Boston, MA) to evaluate differences between the sPFM and dPFM mappings. Eye movement responses often did not meet normality criteria necessary for standard analysis of variance (ANOVA) and post hoc t-test methods. For this reason, we used the aligned rank transform (ART) procedure (Higgins and Tashtough 1994; Salter and Fawcett 1993; Wobbrock et al. 2011) to perform nonparametric, multifactorial repeated-measures ANOVA hypothesis testing and post hoc Wilcoxon rank sum tests for pairwise comparisons where necessary, with statistical significance set to P < 0.05. In these tests we used factors of mapping (“sPFM” and “dPFM”) and either frequency (0.1–2 Hz) or intensity (50–400°/s).

Stimulation paradigm.

Electrical stimulation was delivered using a MED-EL PULSAR CI100 stimulator, a hardware interface box [either Research Interface Box developed at the University of Innsbruck (Bahmer et al. 2010) or MED-EL MAX Programming Interface], and a custom MATLAB software package in a manner analogous to work described previously (Valentin et al. 2013). The stimulator was programmed to deliver biphasic, charge-balanced 150 µs/phase current pulses with no interphase gap delivered to the electrode arrays implanted in each canal using the common crus electrode as the return/reference contact. Calibration experiments using mock electrode arrays with series sense resistors confirmed that the stimulator faithfully delivered current pulse waveforms.

For all experiments described here we wanted to isolate the effect of each PFM mapping and remove any contributions from residual vestibular function, visual input, or any nonprosthetic influence to the evoked responses. The animal’s head was held stationary within the plastic chair, the superstructure was locked in place, and the lights in the experimental chamber were all turned off. An LED was briefly illuminated directly in front of the animal between trials to recenter gaze.

At the start of each experiment, the animal was adapted to a constant current amplitude, 94 pulse/s electrical stimulus aimed to mimic the resting discharge rate of primary vestibular afferent fibers in rhesus macaques (Sadeghi et al. 2007). The active electrode and current amplitude within each canal were chosen to maximize evoked VOR magnitude and minimize 3-D misalignment as defined above. Electrode contacts and current amplitudes are outlined in Table 1. Animals were acclimated to this tonic electrical stimulus in light with an Earth-fixed target and brief intermittent periods in darkness to isolate eye movement responses. This procedure lasted between 60 and 90 min and continued until the evoked slow-phase nystagmus dropped below 5°/s (Dai et al. 2011b; Davidovics et al. 2013).

Table 1.

Summary of electrode contacts and current levels for discretized pulse frequency modulation experiments

	RALP		LHRH		LARP
Animal	Electrode contact	Pulse amplitude, µA	Electrode contact	Pulse amplitude, µA	Electrode contact	Pulse amplitude, µA
RhF20124B	E5	150	E3	150	E0	150
RhF060738G	E2	375	E8	200	E5	275

Both animals were implanted with intralabyrinthine electrode arrays targeting each ampullary nerve branch in the left ear. Differences between each electrode array are described in methods. All stimulation waveforms in this study used 150 μs/phase biphasic, charge-balanced current pulses. E5, electrode 5; LARP, left anterior-right posterior; LHRH, left horizontal-right horizontal; RALP, right anterior-left posterior.

After the animal was acclimated to constant current/rate stimulation, virtual pulsatile stimulus waveforms encoding a series of sinusoidal head velocity waveforms tested VOR responses using both the sPFM and dPFM mappings with the animal’s head stationary. Each virtual head velocity waveform targeted a single canal via modulation of the baseline rate on that canal’s active electrode while holding the pulsatile rate of the active electrodes in the other canals at 94 pulses/s. Virtual head velocity waveforms tested each mapping encoding different peak head velocities (50–400°/s at 0.5 Hz) and different stimulus frequencies (0.1–5 Hz at 300°/s peak velocity). Cycle-averaged responses from ~10 cycles were averaged for lower-frequency stimuli (0.1 and 0.2 Hz, from 15 stimulus presentations), and ~15 cycles were accepted for higher frequencies (0.5–5 Hz, 30 presentations).

dPFM mapping.

Each mapping was generated using a sigmoid curve approximating mean firing rates of regular and irregular vestibular afferents recorded in normal rhesus monkeys (Sadeghi et al. 2007) with a pulse rate corresponding to no motion (0°/s) of 94 pulses/s and a maximum input velocity of 450°/s corresponding to a peak pulse rate of ~425 pulses/s (Dai et al. 2011b; Valentin et al. 2013). For the dPFM map (Fig. 2, red trace), we modeled a combined CI-VI with a maximum aggregate pulse rate of 50,000 pulses/s (i.e., the maximum overall pulse rate of the MED-EL PULSAR device), a total of 12 CI channels spanning the cochlea’s tonotopic axis, and 3 VI channels for each canal in the left ear. Such a device would be capable of stimulating a single electrode at a maximum per-channel rate of ~3,333 pulses/s [i.e., a minimum interpulse interval (IPI₀) of 300 μs]. The central processing unit of a potential CI-VI device would determine whether the stimulator delivers a pulse at integer multiples of 300 μs on each channel in a round-robin, CIS-like fashion. For example, to approximate a desired pulse rate of ~335 pulses/s, the dPFM mapping would deliver a pulse after a wait of 10 × IPI₀ (i.e., deliver a pulse every 10 × 300 μs, or 3 ms, constructing an ~333 pulse/s instantaneous pulse rate). The dPFM mapping uses this integer multiple of IPI₀ to approximate all target pulse rates between 317 and 351 pulses/s (and thus encoding head velocities between ~248–292°/s according to the sPFM map).

Fig. 2.

Head velocity-to-pulse rate maps encoding head motion using pulse frequency modulation (PFM). Each animal was tested using both a smooth PFM (sPFM, black trace) and a discretized PFM (dPFM, red trace) map representing the discretization inherent to a cochlear implant using a continuous interleaved sampling signal-processing strategy. The dPFM map was constructed by assuming that the maximum pulse rate possible on the stimulator used in the study [PULSAR CI100 with a maximum total pulse rate of 50,000 pulses/s (pps)] was divided among 15 electrodes in a round-robin-type fashion (3 electrodes dedicated to vestibular PFM stimulation and 12 electrodes assumed to be routed to a linear cochlear array). This reduces the maximum per-channel rate to ~3,333 pulses/s [thus a minimum interpulse interval (IPI₀) of ~300 μs]. Thus, to deliver high pulse rates used to encode large positive angular velocities, the dPFM mapping approximates the desired pulse rate using integer multiples of IPI₀, creating stair step discontinuities.

This maximum per-channel rate is advantageous for the CI portion of the proposed device, as high, constant pulse rate stimulation allows smooth amplitude modulation to encode the envelope of sound for a given CI frequency band. Yet for the VI portion of a combined implant, using a dPFM-coding scheme constrains pulse rates to 1/(K × IPI₀), where K is an integer. For low stimulus pulse rates (and thus long interpulse intervals) this does not pose a substantial problem; the desired rate can be closely approximated (i.e., Fig. 2 between −450 and 0°/s). For high pulse rates (or short interpulse intervals; Fig. 2 between 100 and 450°/s), only a single integer multiple of IPI₀ can approximate the desired rate for a range of input velocities. This approximation discretizes the sPFM map, creating “stair step” discontinuities as seen in Fig. 2.

The discretization seen when approximating high-rate stimuli is exacerbated when using the dPFM mapping to encode low-frequency head velocity stimuli where the pulse frequency needs to stay in high-pulse rate regions of the head velocity-to-pulse rate mapping for sustained modulation periods. For example, when encoding a 0.1-Hz (Fig. 3A) or 1-Hz (Fig. 3B), 50°/s sinusoidal head velocity waveform, the dPFM map closely approximates the instantaneous pulse rate used to encode the head velocity waveform. The 0.1-Hz stimulus spends more time in the highly discretized region at the peak of the waveform compared with the 1-Hz stimulus, though both accurately approximate the desired stimulus waveform.

Fig. 3.

Examples of pulse frequency modulation (PFM) waveforms using the smooth PFM (sPFM) and discretized PFM (dPFM) mappings. Example pulsatile waveforms using PFM to encode sinusoidal head rotations generated using the sPFM (●) and dPFM (red circles) at 0.1 Hz (A and C) and 1 Hz (B and D). When encoding lower-amplitude 50°/s stimuli (A and B), the dPFM mapping closely approximates the sPFM waveform. In contrast, when encoding large-amplitude stimuli (300°/s, C and D), the dPFM creates stair step discontinuities that discretize the prosthesis output. Inst., instantaneous; pps, pulses per second.

When peak rotational head velocity amplitude increases into the heavily discretized region of the dPFM mapping, the resulting PFM waveform no longer accurately approximates the sPFM signal creating substantial temporal errors. For example, when coding 300°/s sinusoidal stimuli (Fig. 3, C and D), the instantaneous pulse rate delivered to encode the head velocity waveform becomes jagged with stair step discontinuities and overshoots the peak pulse rate (i.e., the sPFM map called for a pulse rate of 356 pulses/s coding a 300°/s head velocity, where the closest dPFM approximation is 370 pulses/s).

We compared dPFM and sPFM mappings by measuring electrically evoked VOR (eeVOR) responses of two rhesus monkeys using each mapping during 1) a sinusoidal intensity series using prosthetic stimuli intended to encode 0.5-Hz, 50–400°/s peak virtual head velocity stimuli and 2) a sinusoidal frequency series using prosthetic stimuli intended to encode 0.1–5-Hz sinusoidal head rotations with 300°/s peak head velocity.

RESULTS

dPFM produces robust, selective eye movements.

Sinusoidal modulation of pulse frequency delivered to primary vestibular afferents in both animals produced robust 3-D eye movements that rotated about axes approximating the anatomic canal axes of the targeted canals (demonstrated with animal RhF060738G in Fig. 4). Virtual head velocity waveforms representing 0.5-Hz, 300°/s head motions about the RALP (Fig. 4, A and B), LHRH (Fig. 4, C and D), and LARP (Fig. 4, E and F) canal axes produced robust (~150°/s peak velocity), directionally appropriate slow-phase compensatory responses that were dominated by the intended canal component using both the sPFM (Fig. 4, A, C, and E) and dPFM (Fig. 4, B, D, and F) mappings. Additionally, large opposite-polarity quick phases during excitatory half cycles are seen in all examples of both mappings, consistent with PFM stimulation creating afferent spike patterns activating downstream VOR circuits in a manner comparable to the healthy vestibular periphery.

Fig. 4.

Raw/processed three-dimensional angular vestibuloocular reflex (VOR) data using smooth pulse frequency modulation (sPFM) and discretized pulse frequency modulation (dPFM) mappings. Electrically evoked VOR responses to 300°/s virtual 0.5-Hz sinusoidal stimuli about the right anterior-left posterior (RALP; A and B), left horizontal-right horizontal (LHRH; C and D), and left anterior-right posterior (LARP; E and F) anatomic axes in animal RhF060738G. Eye velocity is decomposed into LARP (green), RALP (blue), and LHRH (red) components. The pulse frequency modulation instantaneous pulse rate for trials using sPFM (A, C, and D) and dPFM (B, D, and F) is plotted in black for reference; pps, pulses per second.

After removing quick phases from 3-D traces and cycle averaging responses uncontaminated by blinks (Fig. 5), the sPFM and dPFM mappings produced nearly equivalent 3-D VOR waveforms across each canal with minimal cross talk to nontarget canal afferents. Additionally, these examples demonstrate an excitation-inhibition asymmetry, where excitatory phases for each virtual head velocity waveform (representing ipsiversive head motion and corresponding to an increase in the stimulus pulse rate) produce larger-magnitude eye movements compared with the inhibitory half cycle (encoding contraversive head velocities by decreasing pulse rates).

Fig. 5.

Cycle-averaged three-dimensional vestibuloocular reflex responses to the smooth pulse frequency modulation (sPFM) and discretized pulse frequency modulation (dPFM) mappings. Three-dimensional cycle-averaged responses from animal RhF060738G from 0.5-Hz, 300°/s stimuli delivered to electrode contacts in each canal. Virtual head velocity traces are shown in black, whereas dynamics of instantaneous pulse rate are shown in gray. Note that because of the right-hand rule convention used in data analysis, the head velocity waveform for left anterior-right posterior (LARP) and right anterior-left posterior (RALP) stimulation is shown as negative leading, whereas the left horizontal-right horizontal (LHRH) waveform is positive leading. Inst., instantaneous; pps, pulses per second.

Effect of dPFM mapping on encoding head velocity.

Sweeps of PFM intensity to encode virtual 0.5-Hz sinusoids with increasing peak velocity amplitude evaluated the effect of temporal discretization on encoding head velocity (Fig. 6). For this experiment, 0.5 Hz was chosen as a trade-off between using a slow enough stimulus modulation frequency to add sufficient durations of discretization and the time needed to acquire a complete data set during experimental sessions. Virtual peak head velocities ranged from 50 to 400°/s, where larger-amplitude peak velocities inherently produce more temporal discretization compared with slower head motions (Fig. 3, C and D vs. A and B). In this case, we expect lower-amplitude stimuli using the dPFM mapping to evoke similar eye movements compared with the sPFM mapping, since in this region of the dPFM mapping, pulse train patterns closely approximate the smooth, sPFM version. In contrast, at high-amplitude peak head velocities, the peak amplitude region of the stimulus creates longer durations of “flat” transient steps in pulse frequency.

Fig. 6.

Responses to 0.5-Hz sinusoidal head velocity waveforms encoding an intensity series of peak head velocities show negligible difference between the discretized pulse frequency modulation (dPFM) and smooth pulse frequency modulation (sPFM) mappings. Virtual head velocity waveforms processed using both the sPFM (solid lines) and dPFM (dashed lines) produced nearly identical three-dimensional vestibuloocular reflex responses in both animals RhF060738G (A–C and G–I) and RhF20124B (D–F and J–L). Differences in peak slow-phase velocity (Vel.) and misalignment angle were not significant [aligned rank transform 2-way repeated-measures ANOVA; right anterior-left posterior (RALP): F_1,262 = 1.55, P = 0.21; left horizontal-right horizontal (LHRH): F_1,308 = 0.64, P = 0.43; left anterior-right posterior (LARP): F_1,293 = 2.54, P = 0.11].

The eeVOR produced nearly identical peak slow-phase eye velocities about the intended canal axis when stimuli were processed using both the sPFM and dPFM mappings in both animals (Fig. 6, A–F). Peak eye velocity steadily increased with virtual head velocity up to the 300°/s stimulus intensity and plateaued or slightly decreased when tested with 400°/s peak velocity virtual sinusoids, showing a possible saturation of the eeVOR. Using both mappings, VOR peak amplitudes reached >120°/s across all canals, whereas stimuli delivered to the left anterior canal in animal RhF060738G (Fig. 6C) and the left horizontal canal in animal RhF20124B (Fig. 6E) reached ~200°/s.

To assay the 3-D alignment of the evoked 3-D VOR response, we computed the angle between evoked 3-D VOR axes of rotation and the ideal anatomic canal axis (Fig. 6, G–L). In most cases, the misalignment angle remained <10° as the VOR magnitude grew, consistent with selective activation of the targeted canal afferent nerve branches. Responses from both animals and mappings (e.g., Fig. 6, G and I–K) sometimes had increasing misalignment mean and variance for decreasing mean response magnitudes below ~50°/s. This is the expected effect of measurement noise/distortion caused by nystagmus quick-phase removal and spline interpolation to estimate the time-varying 3-D slow-phase velocity vector. When the true, noiseless response magnitude is relatively small, the effect of this distortion on misalignment angle can become large. Numerical simulation revealed that for a true 3-D slow-phase velocity vector of magnitude 20°/s (approximating the smallest responses in Fig. 6, G and I–K), addition of a zero-mean normally distributed distortion error to each component of 3-D slow-phase velocity with standard deviation 2, 5, or 10°/s (estimated from raw data traces in Fig. 4A) would yield misalignment angles of 15 ± 7, 35 ± 20, or 52 ± 30° (means ± SD), respectively. The same simulation revealed that for true 3-D slow-phase velocity vector magnitudes >100°/s, misalignment due to this distortion should be <10°, and measured misalignments >10° therefore likely represent real, physiologic effects. In animal RhF20124B during modulation of the electrical stimulus delivered to the left anterior canal (Fig. 6L) the misalignment remained ~30° through most levels tested, consistent with spurious stimulation of nontarget canal nerve branches producing unintended RALP and LHRH 3-D components with both mappings.

Overall, both mappings produced consistent 3-D VOR responses throughout the entire programmed head velocity input range. Small differences in 3-D VOR responses were not significant for each frequency tested (ART 2-way repeated-measures ANOVA; RALP: F_1,262 = 1.55, P = 0.21; LHRH: F_1,308 = 0.64, P = 0.43; LARP: F_1,293 = 2.54, P = 0.11) indicating that temporal discretization present in the dPFM mapping does not affect vestibular prosthetic encoding of angular head velocity peak amplitude.

Effect of dPFM mapping on encoding head motion frequency.

In addition to encoding peak head velocity, VI technology that aims to prosthetically replicate natural vestibular function should faithfully encode frequency response characteristics of the vestibular periphery and drive a compensatory VOR. This goal is complicated when using a dPFM map, where low-frequency head motion stimuli produce PFM pulsatile waveforms with larger durations of temporal discretization (i.e., more time per cycle in the “discretized” region of the dPFM head velocity-to-pulse rate mapping shown in Fig. 2, exemplified in Fig. 3). To examine the effect of temporal discretization on frequency response characteristics of the eeVOR, we assayed 3-D VOR responses to virtual sinusoidal head velocity stimuli between 0.1 and 5 Hz with 300°/s peak head velocities. This high amplitude was chosen to examine an extreme case of temporal errors and maximize the discretization effects of the mapping.

Data acquired with sPFM and dPFM mappings produced nearly equivalent magnitude responses across both animals and all six tested canals (Fig. 7, A–F). Consistent with data reported previously, prosthetic vestibular stimulation using sinusoidal PFM encoding produced a more high-pass response characteristic and greater phase leads relative to the normal rhesus VOR response to whole body rotation (Dai et al. 2011b, 2017), as might be expected if the pulsatile stimuli preferentially excited irregular primary vestibular afferent neurons in relatively greater proportion compared with less electrically sensitive regular afferents (Goldberg et al. 1984). In both animals, electrodes in the left posterior canal (Fig. 7, A and D) produced the largest gains, seen at 2 and 5 Hz. Animal RhF060738G produced a gain of 1 (and thus an angular velocity of ~300°/s) at these frequencies, whereas animal RhF20124B generated gains >1 at 5 Hz.

Fig. 7.

Frequency response characteristics using discretized pulse frequency modulation (dPFM) vs. smooth pulse frequency modulation (sPFM) mappings show minimal differences. Three-dimensional vestibuloocular reflex (VOR) responses from both animals RhF060738G (A–C, G–I, and M–O) and RhF20124B (D–F, J–L, and P–R) were practically indistinguishable using the sPFM (solid lines) and dPFM (dashed lines) mappings. Residuals between the two mappings were not found to be significant [aligned rank transform 2-way repeated-measures ANOVA; right anterior-left posterior (RALP): F_1,311 = 2.15, P = 0.14; left horizontal-right horizontal (LHRH): F_1,413 = 0.70, P = 0.40; left anterior-right posterior (LARP): F_1,258 = 1.94, P = 0.17].

Phase lead and misalignment response characteristics from both mappings were approximately equal for the frequency band tested. Response profiles indicate a 20–30° phase lead that decreased as a function of frequency, with a slight phase lag at 5 Hz. Response misalignment remained low across frequency for animal RhF060738G, indicating well-aligned 3-D VOR responses. Animal RhF20124B produced less aligned responses consistently across both mappings.

Differences in VOR frequency response characteristics were not significant between the dPFM and sPFM mappings (ART 2-way repeated-measures ANOVA; RALP: F_1,311 = 2.15, P = 0.14; LHRH: F_1,413 = 0.70, P = 0.40; LARP: F_1,258 = 1.94, P = 0.17). The negligible changes in VOR gain, phase response, and 3-D alignment indicate that reducing temporal resolution should have a minimal effect on VI performance of a future combined CI-VI device.

DISCUSSION

Integrating signal-processing strategies from CIs and VIs into a combined CI-VI may require implementing a PFM-coding scheme within the context of the temporally discretized and interleaved modulation scheme used by CIs employing the CIS stimulation strategy. The mismatch of temporal precision necessary for smooth modulation of pulse frequency and the fixed rate used in CIS may create temporal errors for the VI component of a combined CI-VI. This study modeled such a combined system by implementing a dPFM head velocity stimulus-encoding mapping and comparing prosthetically evoked 3-D VOR responses for each of the three canals in the left ear of two rhesus macaques. Overall, results show that VOR magnitude, phase response, and 3-D VOR misalignment are minimally affected by temporal discretization representing integration of a PFM-based VI with typical CI stimulator capabilities. (The MED-EL PULSAR stimulator used in this study has a maximum overall pulse rate of 50,000 pulses/s across all output channels).

A combined CI-VI that alleviates the temporal discretization errors modeled in the dPFM mapping used here will likely benefit from advances in both stimulation hardware capabilities and signal-processing schemes. The PULSAR CI100 device used in this study contains the I100 electronics platform, which integrates individual current sources for each stimulating electrode channel but only a single current sink used as a reference electrode (typically routed to a metal contact on the CI stimulator housing in clinical applications and routed to a platinum-iridium electrode implanted in the common crus in this study; Hochmair et al. 2006). CIs using independent current source-sink pairs may leverage a higher per-channel stimulation rate compared with the PULSAR device used in this study. An increase in the maximum rate achieved on individual stimulus channels as a result of independent source-sink pairs would effectively decrease the IPI₀ in our dPFM model and provide the microcontroller controlling this device with finer temporal control over pulse rate approximation. This makes our dPFM model a liberal estimate of potential temporal discretization effects and therefore a conservative estimate of the performance of a CI-VI with faster electronics.

Implementations of the CIS stimulus-encoding scheme have advanced since its introduction >20 years ago. For example, modern techniques such as fine structure processing (FSP) deviate from the CIS approach to improve auditory perception. In contrast to standard CIS processing (where fixed-rate, interleaved pulsatile stimulation is amplitude modulated to encode slowly varying envelopes of band pass-filtered signals), FSP replicates the high-frequency carrier signal within a frequency band via modulation of pulse rate to match the fine structure frequency content with a constant current amplitude (similar to PFM encoding used in prototype VIs). Incorporating FSP on more apical (i.e., low-frequency) CI electrodes with traditional CIS-like schemes on higher-frequency channels has been successful in encoding information critical for complex auditory percepts such as pitch and timbre (Kleine Punte et al. 2014; Roy et al. 2015; Zirn et al. 2016). Adapting strategies such as FSP to stimulate the VI electrodes in a CI-VI would offer an alternate approach to circumventing temporal discretization errors such as those modeled by the dPFM mapping we tested.

Although data presented here indicate that dPFM did not significantly differ from sPFM with regard to 3-D eeVOR performance, it is not clear whether jagged PFM steps seen at high-amplitude head velocities and/or low-frequency stimuli would have any noticeable perceptual differences with the sPFM mapping. In other words, although response dynamics and filtering characteristics of central VOR neuronal circuits, extraocular muscles, and the eyes may effectively smooth evoked compensatory VOR responses, a human subject using a combined CI-VI device might notice distortion of head motion sensation when using the dPFM mapping. Future work assaying perceptual differences between sPFM and dPFM mappings in human subjects should evaluate perception in addition to assaying VOR performance when characterizing the effects of temporal discretization with the goal of determining the minimal temporal resolution necessary for an effective VI. Furthermore, quantifying a vestibular perceptual equivalent of a “flicker fusion frequency” (Hecht and Shlaer 1936; Simonson 1959) in terms of the minimal quantization error necessary for smooth stimulus encoding may provide insight into how peripheral vestibular activity is integrated over time within central vestibular, cerebellar, and oculomotor circuits.

Single-unit electrophysiologic recordings of activity in primary vestibular afferents, central vestibular neurons, and oculomotor neurons may provide complementary information on how temporal precision changes as sensory signals traverse VOR pathways. Without single-unit recording from regular and irregular primary vestibular afferents, we cannot know the relative sensitivities of those two classes of neuron in this stimulation paradigm. However, the work of Goldberg et al. (1984), particularly their data on responses of regular and irregular squirrel monkey primary vestibular afferent fibers to “short shocks,” suggests that irregular afferents are more sensitive than regular afferents to the pulsatile electrical stimuli we delivered. That hypothesis is supported by the high-pass filter frequency dependence of VOR gain in Fig. 7, A–F, relative to normal (Dai et al. 2011b, 2017) and the phase leads for 0.1–2 Hz in Fig. 7, G–L (Dai et al. 2011b).

In a study of eeVOR responses in guinea pigs (Saginaw et al. 2011), spectral components of the slow-phase VOR response were detected at the 250-Hz baseline pulse rate about which they modulated their prosthetic stimuli. That was an especially intriguing finding because the physiologic bandwidth of the VOR in both primates and rodents is typically assumed to be less than ~50–100 Hz based on the risetime of responses to transient, high-acceleration head rotations. We examined short-term Fourier transform (STFT) representations of sPFM and dPFM stimuli and the corresponding VOR response data, seeking frequency response components driven by the spectral differences between stimuli generated using the sPFM and dPFM mappings. However, effects due to nystagmus quick phases could have masked any discernible differences in STFT between responses to sPFM and dPFM stimuli. We repeated the STFT analysis using data for which quick phases had been removed and the resulting time segments spline interpolated, but we again could not discern a significant difference between sPFM and dPFM data that we could confidently attribute to physiologic responses. We conclude that the use STFT analysis in the Saginaw et al. study was facilitated by the absence of voluntary saccades in guinea pigs. In contrast, our rhesus monkeys made such frequent voluntary saccades, in addition to nystagmus quick phases, that none of the responses during sPFM or dPFM stimulation had sufficient saccade/quick phase-free data to make the STFT analysis tractable.

GRANTS

This work was supported by National Institute on Deafness and Other Communication Disorders Grants R01-DC-009255, R01-DC-13536, and T32-DC-000023.

DISCLOSURES

C. C. Della Santina in an inventor of pending and awarded patents related to technologies discussed in this manuscript, and he holds an equity interest in and is the CEO/CSO of Labyrinth Devices, LLC. The terms of this arrangement are managed in accordance with Johns Hopkins University policies on conflict of interest.

AUTHOR CONTRIBUTIONS

P.J.B. and C.C.D.S. conceived and designed research; P.J.B., N.S.V., K.N.H., C.D., and D.R. performed experiments; P.J.B. and C.C.D.S. analyzed data; P.J.B. and C.C.D.S. interpreted results of experiments; P.J.B. prepared figures; P.J.B. drafted manuscript; P.J.B. and C.C.D.S. edited and revised manuscript; P.J.B., N.S.V., K.N.H., C.D., D.R., and C.C.D.S. approved final version of manuscript.