How electrically evoked compound action potentials in chronically implanted guinea pigs relate to auditory nerve health and electrode impedance

Abstract

This study examined how multiple measures based on the electrically evoked compound action potential (ECAP) amplitude-growth functions (AGFs) were related to estimates of neural [spiral ganglion neuron (SGN) density and cell size] and electrode impedance measures in 34 specific pathogen free pigmented guinea pigs that were chronically implanted (4.9–15.4 months) with a cochlear implant electrode array. Two interphase gaps (IPGs) were used for the biphasic pulses and the effect of the IPG on each ECAP measure was measured (“IPG effect”). When using a stimulus with a constant IPG, SGN density was related to the across-subject variance in ECAP AGF linear slope, peak amplitude, and N1 latency. The SGN density values also help to explain a significant proportion of variance in the IPG effect for AGF linear slope and peak amplitude measures. Regression modeling revealed that SGN density was the primary dependent variable contributing to across-subject variance for ECAP measures; SGN cell size did not significantly improve the fitting of the model. Results showed that simple impedance measures were weakly related to most ECAP measures but did not typically improve the fit of the regression model.

I. INTRODUCTION

Studies in animals with cochlear implants (CIs) have demonstrated that nerve survival, specifically the density of spiral ganglion neurons in Rosenthal's canal (SGN density), is related to specific electrophysiological measures of CI function when using acute or short-term implanted animal models (Smith and Simmons, 1983; Stypulkowski and van den Honert, 1984; Hall, 1990; Shepherd and Javel, 1997; Ramekers et al., 2014; Ramekers et al., 2015). Ramekers and colleagues showed that, along with SGN density, SGN cell size independently contributed to some ECAP measures in the same animals (Ramekers et al., 2014). Generally speaking, abnormal cell size suggests perturbation of a homeostatic state which could result in abnormal function of surviving SGNs, independent of the number of SGNs remaining. A similar relationship is observed between SGN density and specific psychophysical and electrophysiological measures in chronically implanted and/or very long-term deafened guinea pigs (Pfingst et al., 2011; Pfingst et al., 2015b; Ramekers et al., 2015; Pfingst et al., 2017; Schvartz-Leyzac et al., 2019). Studies have also shown that ECAP responses in cochlear-implanted animals change significantly in the first thirty days following implantation, and typically become relatively stable by 100 days post-implantation. However, some animals' responses continue to change over time and it seems that relative stability is not achieved (Pfingst et al., 2015a; Schvartz-Leyzac et al., 2019).

We can use insight gained from these animal experiments to estimate neural conditions in humans with cochlear implants. ECAP measures are readily available for use in a clinical setting, are fairly quick to record, reflect the status of the auditory nerve, and have been shown to relate to speech recognition outcomes in cochlear implanted humans (Kim et al., 2010; DeVries et al., 2016; van Eijl et al., 2017; Schvartz-Leyzac and Pfingst, 2018). However, based on data from animal research, it is important to consider how ECAP measures relate to neural conditions in a long-term implanted model akin to humans with cochlear implants. One potential limitation of ECAP measures for assessment of neural health in animal or human subjects is that the recording of the evoked potential might be influenced by variables that are unrelated to neural health per se; here, we refer to these variables as “non-neural” variables. Electrode impedance is an example of a non-neural variable that has been shown to reflect ECAP recordings in humans. In some, but not all CI users, the across-site variation in simple impedance measures has been shown to be correlated with the ECAP amplitude and/or AGF slope measures for a constant duration IPG (7 μs) (Schvartz-Leyzac and Pfingst, 2016; Scheperle, 2017). Given that electrode impedance is readily and frequently measured in the clinic, it is an important factor to consider when determining the utility of ECAPs in the clinical setting.

In an effort to better control for non-neural factors, studies have examined how the ECAP response differs for stimuli with short versus long duration interphase gaps (IPGs) using biphasic electrical pulses in animals and humans (“IPG effect”) (Prado-Guitierrez et al., 2006; Ramekers et al., 2014; Ramekers et al., 2015; Schvartz-Leyzac and Pfingst, 2016; McKay and Smale, 2017; Schvartz-Leyzac et al., 2019). In humans, the ECAP IPG effect does not appear to be related to measures of simple impedance, unlike ECAPs measured with a fixed IPG (Schvartz-Leyzac and Pfingst, 2016). Taken together, results from acutely implanted and/or short-term deafened guinea pigs suggest that SGN density can account for as much as 60%–80% of IPG sensitivity when measured using the ECAP (Prado-Guitierrez et al., 2006; Ramekers et al., 2014). Additionally, Ramekers et al. found that, along with SGN density, perikaryal area (e.g., cell size area) also helped to explain additional variance in ECAP measures across animals (Ramekers et al., 2014). There is a need to better understand the effects of cell size and density in animals who have been implanted for an extended period of time, similar to humans with CIs.

Results in guinea pigs and humans have shown substantial changes in both impedance and ECAP values within the first hours and days following implantation (Brown et al., 2010; Chen et al., 2013; Pfingst et al., 2015a; Schvartz-Leyzac et al., 2019). In a recent study which measured ECAPs in cochlear-implanted guinea pigs, we found that responses using a fixed IPG as well as the ECAP IPG effect varied the most within 30 days post-implantation, and required as many as 100 days to reach a statistically stable response in some animals (Schvartz-Leyzac et al., 2019). Additionally, the same study found that changes in impedances were generally unrelated to changes in ECAP measures when followed for several months following cochlear implantation. Given the potential clinical utility of ECAP and impedance measures in chronically implanted human subjects, it is important to quantify how these measures reflect neural status during a stable period.

The current study estimated how both neural (e.g., SGN density and cell size) and a non-neural measure (e.g., simple impedance) relate to across-subject variance in ECAP recordings. We hypothesized that ECAP measures in chronically implanted animals would correlate with SGN density and cell size, but that impedance values would also contribute to the recorded response. We expected that impedance values would correlate with ECAP measures using a fixed IPG duration, but the same correlation would not be found when examining the ECAP IPG effect for which non-neural variables are fixed. We examined three different features of ECAP recordings (AGF slope, peak amplitude, and N1 latency) using two IPG durations. These measures were selected in order to closely compare the results shown here to those previously published in shorter-term deafened animal models. In the present study, we obtained these ECAP measures and simple impedance values in very long-term, chronically implanted guinea pigs with varying degrees of SGN density and cell size. ECAP recordings and impedance values were analyzed during a period of time in which responses were obtained closest to sacrifice for histological processing.

II. METHODS

A. Overview

Methods reported here are similar to those reported in previous publications from this laboratory (Pfingst et al., 2015a; Pfingst et al., 2017; Schvartz-Leyzac et al., 2019). The animals used in these experiments were specific pathogen free (SPF) pigmented guinea pigs that were chronically implanted (4.9–15.4 months) with a CI. After implantation ECAP AGFs were followed over time until stable, and data were obtained up to the time of sacrifice. These guinea pigs have been used in several experiments previously published or in progress. Only ECAP, impedance, and histological data are reported here.

Animals used for these studies were bred and maintained by the Unit for Laboratory Animal Medicine at the University of Michigan. The animal-use protocol was reviewed and approved by the University of Michigan Institutional Animal Care and Use Committee (IACUC). Altogether, data from 34 cochlear-implanted animals were used for the current study. Nine of these 34 animals were also reported on in a previous article (Schvartz-Leyzac et al., 2019), but the analyses and hypotheses were different than those addressed in the current article.

In the 34 animals that received a cochlear implant, some of the animals were implanted in a previously normal hearing, intact ear (Table I, group 1 “Implant only”). In this case, there was minimal loss of cochlear structures, which typically resulted in good hearing preservation in the region of the CI (Kang et al., 2010; Pfingst et al., 2011; Pfingst et al., 2017). The remaining implanted animals (N = 25) were deafened with neomycin prior to undergoing cochlear implantation but were divided between treatment with neurotrophins (Table I, group 2) and without neurotrophins (Table I, group 3). The goal of the current study was not to determine the effect of various deafening treatments on ECAP measures, but rather to create a group of animals with varied neural conditions. Previous to the present study, various neurotrophin treatments and delivery models were used in the lab over a period of time in order to test and evaluate the biological outcome of each (Budenz et al., 2015; Pfingst et al., 2017). These findings showed that the neurotrophin treatment approach used in the current study resulted in higher and lower SGN density, on average, relative to the neomycin-deafened only animals and implant only animals, respectively. However, these studies also showed that the neurotrophin treated animals exhibited a wide range of SGN densities. The specific methods related to neurotrophin treatments were not intended to be an independent variable for the current project, but are used to achieve a group of animals with varied, but stable, residual neural density.

TABLE I.

Characteristics for all animals.

Animal group characteristics
Group	Treatment name	Number of animals (Total N = 34)	Symbol	Average survival time in days post implantation (DPI)	Average SGN density (profiles A–C) cells/mm2	Average SGN size (profiles A–C) μm2	Number of animals with remaining IHCs (profiles A–C)
1	Implant only	9	*	269.3	853	199	9
2a	AAV.Ntf3	17	Δ	227.8	324	236	1
2b	AAV.BDNF	2	⋄	186.0	507	233	1
2c	AAV.Ntf3/BDNF	3	+	270.0	187	271	0
2d	Ad.BDNF	1	○	249.0	208	358	0
3	AAV.empty	2	□	254.3	56	250	0

Following deafening, animals in group 2 underwent gene therapy treatment with a variety of viral vectors and gene inserts (AAV.Ntf3, AAV.BDNF, AAV.Ntf3/BDNF, and Ad.BDNF; see Table I). Following deafening, group 3 (N = 2) received an empty adeno-associated virus (AAV.Empty), akin to a deafened, untreated ear. Details for deafening, inoculation, and implantation procedures for all groups are described below. Following implantation, ECAPs and impedances were measured until completion of all experiments and then animals were euthanized and histological analyses were performed.

In addition to the implanted animals, seven animals who received no treatment and were not implanted underwent histological analysis for comparative purposes. These animals were sacrificed at an average of 23.66 months of age (range: 7–46 months old), and approximated the range of age and weight of the implanted guinea pigs. Note that the non-implanted animals were used only as a reference for normal histology, and the implanted animals were the main focus of the current study.

B. Deafening, inoculation, and implantation procedures

All implanted animals were anesthetized with ketamine (40 mg/kg) and xylazine (10 mg/kg) and placed on a heating pad. A post-auricular incision was made, the temporal bone was exposed by blunt dissection, and the bulla was opened. For treatment groups 2 and 3, a small cochleostomy was made in the basal turn of the cochlea approximately 0.7 mm apical to the round window using a hand drill. Through this cochleostomy, neomycin sulfate (10 μl; 5% wt./vol) was slowly injected into the scala tympani (rate of 5 μl/min) using a cannula and an infusion pump. This was intended to destroy hair cells and deafen the ear. The infusion cannula was left in place for 20 min, and then removed. A drop of Healon (hyaluronic acid; 10 mg/mL; Abbott Medical Optics, Inc., Santa Ana, CA) was placed on the cochleostomy to promote penetration of administered substances through membranes, and then using a new cannula and the infusion pump, 5 μl of either AAV or Ad plus neurotrophin (Ntf3, BDNF or both) or an empty AAV were infused into the scala tympani at a rate of 1 μl/min. It should be acknowledged that administration of drugs via pump could itself cause damage, independent of the effect of a particular drug. However, the methods described here have been used in several studies previously in our lab with fairly consistent effects (Budenz et al., 2015; Pfingst et al., 2017; Swiderski et al., 2020). After the inoculation, the cannula was removed and the cochleostomy was either temporarily covered with a piece of muscle and allowed to sit for 30 min or the cochleostomy and the bulla were sealed with Durelon and left for two weeks before implantation. The total number of inoculated/implanted animals = 25. Of those 25 animals, 14 animals were inoculated and implanted immediately. The remaining animals (N = 11) were inoculated and implanted following a brief period of time. Nine of those 11 animals were inoculated and implanted approximately 14 days later. The two those 11 animals were inoculated and implanted at 20 or 64 days after inoculation.

The implantation procedure was similar for all subjects. For implant insertion, a cochleostomy approximately 0.7 mm apical to the round window was drilled with a diamond bur, bone dust and debris were removed from the area around the cochleostomy using a cotton pledget, and an eight-electrode scala tympani implant was partially or fully inserted to the point of first resistance resulting in five to eight intracochlear electrodes. In guinea pigs, the diameter of the scala tympani is considerably smaller compared to the human cochlea, and narrows rapidly past the first half turn. It is important that the electrode is not forced to create a deeper insertion, as this could cause damage to the cochlea and the electrode array. Therefore, caution is used when inserting the electrode as to not force the electrode array beyond initial resistance in order not to cause damage. The implant was secured to the bulla with a silk suture, the bulla opening was then sealed with Durelon cement and the incision was closed.

Anchor screws were placed at three points around bregma and used to secure a small inverted bolt (specified as the “anchor bolt”) on which the connector that interfaced with the stimulator would be mounted; this anchor bolt was also used for the ECAP stimulating ground. An additional screw was placed at the midline, 1 cm caudal to bregma (specified as the “vertex screw”) and used for the ECAP recording ground.

Following treatment and implant surgery, residual hearing was evaluated for each animal using acoustic psychophysical thresholds for pure tones from 50 Hz to 24 kHz and/or using auditory brainstem response (ABR) testing. Multiple measurements were taken within the first 45 dpi to assess and confirm hearing in all animal groups. If hearing was measurable during the first 45 dpi, then acoustic psychophysical thresholds were measured throughout the survival time of the animal. For details regarding the acoustic psychophysical threshold and ABR testing, please refer to previous publications from our lab that used the same methods (Pfingst et al., 2011; Pfingst et al., 2017).

C. Cochlear implants

The CIs (manufactured by Cochlear Corporation) consisted of either eight ring electrodes encircling a silicone rubber carrier (N = 31) or eight half-band electrodes on a silicone rubber carrier (N = 3). Different electrode types were used based on what was available from the manufacturer at that time. All electrodes were spaced at approximately 0.75 mm center to center. For the banded implants, the diameter of the implant near its apical end was 0.4 mm so that it filled the majority of the scala tympani at about 4.5 mm from the cochleostomy and could not be safely inserted further due to the rapidly decreasing diameter of the scala tympani apical to this point. For the half-band implants, the diameter of the implant near its apical end was 0.20 mm and tapered so this implant was able to be inserted deeper; on average around 6.5 mm. The primary electrode used for stimulation was the second most apical electrode (electrode 2 in Fig. 1) which was located an average of 2.7 mm apical to the cochleostomy (approximately the 18 kHz region) for the banded implants and 5.2 mm apical to the cochleostomy (approximately the 9 kHz region) for the half-banded implants. In both implant designs, the primary electrode typically sat close to the modiolar wall. The top panel in Fig. 1 shows a midmodiolar section of the guinea pig cochlea with the profiles used for data analysis labeled “A” through “F.” The banded implants were located in profile A and the half-banded implants were located in profiles A and B. This figure also appeared in a previous manuscript as it describes a standard method used in our lab for all experiments (Pfingst et al., 2017; Schvartz-Leyzac et al., 2019).

FIG. 1.

A peri-midmodiolar section of a guinea pig cochlea illustrating how half turns of the cochlea were labeled for reporting histological results is shown in the top part of the figure. The profiles were labeled A through F from base to apex. The apical end of an eight-electrode scala tympani implant is illustrated schematically in the lower part of the figure. Note that this figure also appears in previous publications with similar methods, and additional details are available in those publications (Pfingst et al., 2017; Schvartz-Leyzac et al., 2019).

D. Electrically evoked compound action potential (ECAP) amplitude-growth functions

ECAP AGFs were recorded in awake guinea pigs while the animals were standing in the test cage. Stimulation and recording for ECAPs utilized a MED-EL “Pulsar” CI100 receiver/stimulator connected to the implant through a percutaneous electrical connector. The receiver/stimulator was connected to a standard PC via a Research Interface Box (RIB II; University of Innsbruck). Measurement parameters were controlled using custom software developed in matlab (MATLAB, 2010). Monopolar electrode configurations were used for stimulation and recording. The default protocol for this study was to stimulate the second most apical electrode in the CI (electrode 2 in Fig. 1) and record from the adjacent, more basal, electrode (electrode 3). In cases where either of these electrodes was broken, another adjacent pair was used. The stimulating ground was anchor bolt and the ECAP-recording ground was vertex screw. The stimulus level was measured in current units (cu) where 1 cu was approximately equivalent to 1 μA.

The stimulus for ECAP recordings was a biphasic pulse, with 45 μs phase duration and 2.1 μs or 30 μs inter-phase gap (IPG). Pulses were delivered at 50 pps for 20 iterations. The recording amplifier was blanked for 213 μs following electrical pulse onset to avoid saturation artefacts. Prior to recording the ECAP, the maximum stimulus level (MSL) that could be used for each stimulation site was determined. The MSL was based on behavior (observed facial twitch or pinna movement) or by compliance limit of the stimulator/electrode combination, whichever was lower. The MSL for the ECAP experiments was typically set 10 to 20 cu below the current level eliciting a facial nerve response. To obtain an ECAP AGF, the program selected stimulus levels at 15 amplitudes evenly spaced from zero to the MSL and the order of presentation of the stimulus levels was permuted (8 1 9 2 10 15 5 6 14 13 12 7 3 4 11).

For every amplitude step, responses to an anodic leading biphasic pulse and a cathodic leading biphasic pulse were averaged to reduce stimulation artifacts and the response to zero-amplitude stimulation was recorded and subtracted from this average to reduce non-stimulus related effects like switch-on artifacts. The resulting waveforms obtained for each of the 15 amplitude steps were plotted and analyzed using a custom-made matlab (MATLAB, 2010) program. This program picked all negative peaks and positive peaks, which were then verified by visual inspection of the waveforms. The N1 to P2 amplitudes in μV were plotted against stimulus current in μA to obtain input-output AGFs.

For the AGF slope calculation, matlab (MATLAB, 2010) was used to calculate linear regression fits using a custom method used previously by our lab (Schvartz-Leyzac et al., 2019; Schvartz-Leyzac et al., 2020). This method works well in our lab to comparatively assess ECAP AGFs in animals and humans, which often display various morphologies (e.g., sigmoidal, linear, etc). Based on the typical AGF shape, all points below 100 μV (noise floor) were excluded in the fit function. The AGF was linearized by approximating the slope of the linear region using the “gradient” function, and systematically removing the points that deviated by more than 20% of this slope. A linear regression model was fit to all of the remaining points and the resulting slope was calculated. For a constant IPG, the peak amplitude was taken as the largest voltage response of the AGF, and the N1 latency was recorded at this peak amplitude response. The IPG effect was calculated as the raw difference in slope, peak amplitude or N1 latency values between the 2.1 and 30 μs IPG conditions. Increasing the IPG often resulted in lower MSLs for all animals. Therefore, the IPG effects for the ECAP peak amplitude and N1 latency data were calculated using equal current levels for the two IPG conditions, which was typically the MSL for the 30 μs IPG condition (the lower of the two values).

ECAP recording schedules varied slightly depending on the animal and the laboratory protocol at the time of data collection. ECAPs were recorded frequently throughout the first 90 days after implantation and monthly thereafter. Some animals had ECAPs recorded more frequently in the first 90 days compared to other animals. One final ECAP was recorded on or close to the day that the animal was euthanized for histological examination.

The guinea pigs received intermittent electrical stimulation 4 to 5 days a week during test sessions to measure ECAP AGFs and psychophysical thresholds. These sessions started on the day of implantation and continued until the animal was euthanized. The animals received no other electrical stimulation.

E. Impedance measures

Prior to ECAP recording sessions, the impedance of all the electrodes relative to a ground ball placed in the neck muscle at the time to implant surgery was measured using a custom-built impedance meter with a 1 kHz sinusoidal test current at 1 μA rms. If the ground was broken, the head bolt was used. This was done to confirm the function of each electrode used in the ECAP recordings. Only impedances from the primary intracochlear stimulating electrode used in the ECAP measurements are shown in this manuscript. Average impedance values were calculated for each animal using values recorded on the same days as the ECAP recordings (+/− 1 day).

III. ANALYSIS

A. Histology

After collection of all functional data, cochlear implanted animals were anesthetized and perfused intravascularly with 4% paraformaldehyde. Temporal bones were extracted with the implant remaining in place in the implanted ear. Nine of the animals had their implants removed before the decalcification process. As such, there was insufficient structure remaining to mark the location of the primary electrode site (electrode 2) and rather the location was estimated from using the insertion angle re: to the round window. This value was then adjusted based on the amount of spiral lamina bone that remained following dissection. The remaining animals had their tissues decalcified in 3% EDTA for 3 to 6 months until the bone was soft. For these animals, when decalcification was complete and the cochlear-implant electrodes were visible through the bone, the cochlea was marked in the lateral wall at the location of the primary electrode that was used for stimulation in this study (electrode 2). The implant was then gently removed. For all animals (implanted and non-implanted), tissues were embedded in JB-4 (Electron Microscopy Sciences, Hatfield, PA, USA) and sectioned with glass knives to obtain 3 μm thick sections in a peri-midmodiolar plane centered at the location of the previously made mark designating the location of the primary stimulating electrode. Approximately 45 sections were collected per cochlea.

For cochlear implanted animals, the first section chosen for analysis was usually the section closest to the location of the primary stimulating electrode. If that section was damaged, the next closest section was used. Four other sections were then selected at intervals separated by a minimum of 6 sections (∼18 μm) in order to prevent counting a cell more than once. The sections used for SGN counts were stained with toluidine blue. The specimens were observed with a Leica DMRB epi-fluorescence microscope, although fluorescence light was not used for the analysis (Leica, Eaton, PA, USA) and photographed with a CCD Cooled SPOT-RT digital camera (Diagnostic Instruments). The cross-sectional area of the SGN cell bodies within Rosenthal's canal was determined using Image J software and the cell bodies in each of the five selected sections were counted. Density of SGNs was calculated by dividing the number of cells counted by this cross-sectional area. Only cells with diameters of 12 to 25 μm and nuclei of 5 to 9 μm or more in diameter were considered. Of these cells, only healthy-appearing cells were counted; cells that had a poorly defined cell membrane or that appeared to be shrunken or atrophying were not counted. For SGN cell size measurements, a blinded observer used ImageJ to outline and compute the enclosed area of each cell fitting the criteria for counting. It is important to note that the sectioning method described above did not influence the cell size measurements. The orientation of the sections was standardized in order to ensure that orientation was not skewed, and we restricted counting and measurement of cells to those sectioned near the center of the nucleus.

Inner hair cells (IHCs) in each profile, if present, were also counted. To avoid counting the same hair cell in two sections, IHCs were only counted if both the nucleus and stereocilia were visible. The number of animals with surviving IHCs is reported in Table I. Mid-modiolar slices are not ideal in order to precisely count IHCs. Therefore, specific numbers are not provided but rather IHCs are classified as either present or absent.

B. Statistical methods

Data were analyzed using r version 3.4.3 (RCoreTeam, 2018) and a custom program developed in matlab (MATLAB, 2017). For the AGF slope calculation, matlab was used to calculate linear regression fits using a custom method described above. For analyses comparing SGN densities, cell size, and impedance to ECAP recordings, simple and multiple linear regression analyses were performed in r using the “lm” function. One-way ANOVAs and post hoc analyses were performed in r using the “aov” and “TukeyHSD” functions, respectively. Correlations were performed in r using the “cor.test” function.

The ECAP amplitude, AGF linear slope and N1 latency values were averaged across three time points which occurred closest to the time of sacrifice for each animal. Previous work has shown that, for some animals, slopes of ECAP AGFs fluctuated over the first 100 days post implantation (dpi), typically decreasing during the first few days after implantation and then increasing to a steeper slope which was relatively stable after about 100 days (Pfingst et al., 2015a; Schvartz-Leyzac et al., 2019). Therefore, in the current study all three time points were required to be >120 days dpi for each animal. While this criterion might not necessarily reflect a “stable” period for all animals, ECAP responses were obtained when cochlear status was close to that at the time of sacrifice.

IV. RESULTS

A. Histology

Average SGN densities are provided in Table I for all implanted animals. The range of SGN density in profiles A–C (cells/mm²) across all animals was 14–1070. By comparison, the range of SGN density (cells/mm²) in the non-implanted animals was 819–1205 (average = 938.54 cells/mm²). A one-way analysis of variance (ANOVA) was performed to determine if the SGN density counts differed across groups 1–3 as well as the non-implanted group. Group data are shown in the box plots Fig. 2. Results indicated a significant difference in SGN density between the groups [F(3, 37) = 33.02, p < 0.001]. Post hoc Tukey HSD analysis revealed that group 1 (853.87 cells/mm²) had significantly higher SGN density than group 2 (322.15 cells/mm²) and group 3 (56.17 cells/mm²) (p < 0.05). Results showed equivalent, average SGN densities between group 1 and the non-implanted group (p = 0.81). But, the non-implanted group had significantly higher SGN density when compared to group 2 or group 3 (p < 0.0001).

FIG. 2.

A boxplot showing SGN density (average of profiles A–C) information for all animals that participated in the experiments. Implanted animal groups are described in Table I. Within each box plot, the horizontal black line represents the median value within each group. The lower and upper limits to the box correspond to the first and third quartiles (the 25th and 75th percentiles). The lower and upper whisker extends from the 10th and 90th percentiles, respectively. The filled black symbols represent outliers. Horizontal brackets with asterisks reflect significant between-group differences as indicated by Tukey post hoc analysis (p < 0.05).

Average SGN size (μm²) for profiles A–C for each animal group are also shown in Table I. The cell size range across all animals was 118–397 μm². For comparison, the average SGN size for the group of non-implanted animals was 177 μm² (range = 102–227 μm²). A one-way ANOVA was performed to assess group differences in SGN size between all implanted groups as well as the group of non-implanted animals. Group data are shown in Fig. 3. Results showed a significant main effect of cell size [F(3, 37) = 3.162, p = 0.03]. Post hoc Tukey HSD analysis showed only one between-group difference, which was between the non-implanted animals (177 μm²) and group 2 animals (274 μm²) (p < 0.04).

FIG. 3.

Similar to Fig. 2, but with the y axis showing data for average cell size in profiles A–C. Note that group 3 does not have a box and whiskers because of the small sample size and lack of variation in this subject population.

Eleven implanted animals had some intact IHCs in profiles A–C. The majority of animals with residual IHCs (9/11) were in the implant only group (animals implanted in a non-neomycin treated ear).

B. Electrically evoked compounded action potentials (ECAPs)

Examples of acceptable ECAP waveforms are shown for one animal in Fig. 4 (580L1/group 1). Note that this figure is similar to one previously published and is used to demonstrate acceptable waveforms analyzed for the purposes of our studies (Schvartz-Leyzac et al., 2019).¹ Only responses with a clearly defined negative (N1) peak followed by a positive (P2) peak, as illustrated in Fig. 4, were used. In these examples, the waveform amplitude increased with increasing current level.

FIG. 4.

Figure showing raw ECAP waveforms from one animal (580L1 from group 1). This figure is shown to represent the quality of ECAP recordings that were deemed acceptable for analysis. Note a clearly visible N1-P2 peak is present at several current levels. A variant of this figure was also published in a previous manuscript for the same purpose (Schvartz-Leyzac et al., 2019).

1. Correlation between ECAP amplitudes, linear slopes and latencies for 2.1 and 30 μs IPG

Correlational analyses were performed using averaged ECAP responses over the final three data points collected before the animal was euthanized. Two-tailed, pairwise Pearson correlation analysis showed that data for the 2.1 μs IPG were highly correlated with those for the 30 μs IPG—AGF linear slope: r = 0.99, p < 0.0001; AGF peak amplitude: r = 0.95, p < 0.001; AGF N1 latency: r = 0.90, p < 0.001. Therefore, in an effort to limit multiple comparisons, we chose to focus on data for the 2.1 μs IPG stimuli and on the difference in such measures when increasing the IPG from 2.1 to 30 μs (IPG effect).

2. Predicting ECAP responses based on neural factors

Results of each simple linear regression between SGN density and ECAP measures are shown in Fig. 5. Presence or absence of IHCs are coded by color (blue = No IHC in profiles A–C; red = IHC present in profiles A–C). Regression results are shown within each graph. Bonferroni adjustments were applied due to multiple comparisons (p = 0.008) to calculate the criteria for statistical significance. These results demonstrate that there was a significant and positive relationship between SGN density and most ECAP measures examined at a time close to tissue collection for histological analysis to end of life and histological analysis. The IPG effect for N1 latency was the only measure not dependent on SGN density measured in the same animals.

FIG. 5.

(Color online) Linear regression analyses showing the relationships between SGN density and ECAP measures. The left column shows data for a 2.1 μs IPG and the right column shows data for the IPG effect (data for a 30 μs IPG minus data for a 2.1 μs IPG). (a) ECAP AGF linear slope; (b) IPG effect for ECAP AGF linear slope; (c) ECAP AGF peak amplitude; (d) IPG effect for ECAP AGF peak amplitude; (e) ECAP AGF N1 latency; (f) IPG effect for ECAP AGF N1 latency. The symbols are identical to those used in Table I. Colors (red and blue) indicate if inner hair cells (IHCs) were present or absent, respectively. Solid regression lines are for analyses using data from all animals, regardless of IHC status. Regression statistics are shown in each figure.

Previous studies showed that both SGN density and cell size help to explain variance in ECAP responses in cochlear implanted guinea pigs (Ramekers et al., 2014). In order to explore these relationships in the present cohort of animals, we first performed correlational analyses (Pearson's correlation) to explore interrelationships prior to further analysis via multiple regression analysis. Results of the correlational matrix are shown in Table II. It can be observed that SGN density and cell size are significantly correlated with one another, and this relationship shows that cell size decreases as SGN density increases (r = −0.449, p = 0.009). In most cases, each neural measurement (SGN density and cell size) shows a significant relationship with ECAP measurements, without correction for multiple comparisons (p < 0.05). All ECAP measures of interest with the exception of one (i.e., ECAP N1 latency IPG effect) show a significant, positive relationship with SGN density, similar to results shown in Fig. 5. The ECAP AGF linear slope and peak amplitude, along with the IPG effect for each of these measures, show a significant, negative relationship with cell size. Based on these data, a linear multiple regression analysis was performed using a Forward method (p < 0.05 to enter). Based on the correlational matrix shown in Table II, the predictor variable of SGN density was entered first and cell size was entered second, if the p < 0.05 criterion was met. Multiple regression analyses were performed for each of the six ECAP measures shown in Table II. Collinearity diagnostics were performed by examining the variance inflation factor (VIF); VIFs of less than 2.5 were acceptable which was true in every comparison.

TABLE II.

Correlation matrix for independent and dependent variables of interest.

	SGN density (cells/mm2)	SGN size (μm2)	Impedance (kOhm)
SGN density (cells/mm2)	—	—	—
SGN size (μm2)	r = −0.449	—	—
	p = 0.009a	—	—
Impedance (kOhm)	r = −0.173	r = 0.211	—
	p = 0.328	p = 0.082	—
ECAP AGF linear slope (2.1 μs IPG)	r = 0.662	r = −0.519	r = −0.388
	p < 0.001a	p = 0.002a	p = 0.023
ECAP peak amplitude (2.1 μs IPG)	r = 0.645	r = −0.550	r = −0.340
	p < 0.001a	p = 0.001a	p = 0.049
ECAP N1 latency (2.1 μs IPG)	r = 0.514	r = −0.048	r = 0.017
	p = 0.002a	p = 0.791	p = 0.922
ECAP AGF linear slope IPG effect	r = 0.604	r = −0.424	r = −0.269
	p < 0.001a	p = 0.014	p = 0.042
ECAP peak amplitude IPG effect	r = 0.635	r = −0.381	r = −0.249
	p < 0.001a	p = 0.020	p = 0.043
ECAP N1 latency IPG effect	r = −0.089	r = −0.099	r = −0.065
	p = 0.617	p = 0.585	p = 0.715

^aSignificant findings following Bonferroni correction for multiple comparisons (p < 0.05/8 = 0.006).

Generally, results of the regression analyses revealed that variance in SGN density across animals was the primary neural factor that helped to account for across-subject variances observed in most ECAP measures examined. Variance in cell size only significantly contributed to the regression model for the ECAP AGF peak amplitude measure, in which case SGN density was the primary factor (R² = 0.471, p < 0.001) and entering cell size as a second factor significantly improved the multiple regression model (change in F statistic, p = 0.03) (R² = 0.545, p < 0.001). As shown in Fig. 5, for the ECAP AGF linear slope (R² = 0.415, p < 0.001) and N1 latency measures (R² = 0.252, p < 0.0001), SGN density helped to significantly explain variance in both measures but cell size did not significantly contribute to the model (p > 0.05). As shown in Fig. 5, for the ECAP IPG effect measures, results of the analysis showed that SGN density accounted for a significant proportion of variance for the AGF peak amplitude (R² = 0.40, p < 0.001) and AGF linear slope (R² = 0.346, p < 0.001), but the factor of cell size did not significantly contribute to either model (p > 0.05). As expected, based on the results of the correlation matrix in Table II, the model examining the IPG effect for N1 latency was not significant when SGN density was entered as an independent variable.

3. Contribution of impedance variance to ECAP measures

Impedance values for the three groups are shown in Fig. 6. Note that the solid horizontal lines in the boxplots shown in Fig. 6 represent the median value within each group. Results of a one-way ANOVA revealed no significant difference between the average impedance values between groups [F = 7.89 (2, 32), p = 0.681].

FIG. 6.

A boxplot showing simple impedance information for each animal group. The format details are identical to those in Fig. 2.

In an effort to better understand how non-neural factors contribute to ECAP recordings, we analyzed the relationship between impedance and each ECAP measure (i.e., the mean of ECAP measures at three time points closest to animal sacrifice) across animal groups. Correlational analyses are shown in Table II. Results suggest that impedance is weakly correlated with some of the ECAP measures, however, not after correcting for multiple comparisons using the Bonferroni criterion (p = 0.008). Based on these observations, multiple linear regression models were calculated to determine if, in each of these cases, impedance values significantly improved the regression model between SGN density and the specified ECAP response. For each model, SGN density was entered as the first factor and impedance values were entered as the second factor using a Forward approach and similar methods to those noted previously for the cell size analysis.

For the first model (ECAP AGF linear slope), both the primary (SGN density) and secondary (impedance) factors contributed significantly to the overall model (R² = 0.52, p < 0.001). As observed in Table II and Fig. 5, SGN density had a significant and positive relationship with ECAP AGF linear slope. However, the factor of impedance had a significant and negative relationship with ECAP AGF linear slope. These results suggest that, while higher SGN density relates to steeper AGF linear slope values, increasing impedance values have the opposite relationship (a higher impedance value is associated with a lower AGF slope value). Although the final models (F and p values) are significant for all other analyses, the impedance variable did not significantly contribute to the model (p > 0.05). That is, impedance did not account for additional variance in the ECAP response (IPG effect for AGF linear slope of peak amplitude and ECAP AGF peak amplitude).

V. DISCUSSION

A. Comparing SGN density and ECAP measures when using a constant or changing IPG

In this study, SGN density was significantly correlated with ECAP AGF slope and peak amplitude for a constant IPG and also for the IPG effect, but the proportion of variance explained by SGN density was always less for the IPG effect (Fig. 5). The IPG effect is complex, but it stands to reason that the health and/or properties of the stimulated neurons may affect these particular ECAP measures differently compared to other ECAP measures that use a constant IPG stimulus. Logically, the IPG effect involves temporal properties of the membrane as the cell depolarizes or hyperpolarizes depending on the excitatory phase of the biphasic pulse. Some studies have noted morphological changes in the SGNs following insult or injury (Ylikoski et al., 1974; McFadden et al., 2004; Versnel et al., 2007) that seem to be somewhat independent of the total SGN density (van Loon et al., 2013). There is evidence that cell size is correlated with membrane capacitance (Limon et al., 2005). It could be reasoned that in a healthy ear both the cell count and cell properties would be fairly normal, showing robust ECAP responses evoked by a stimulus with either a constant or changing IPG. However, following insult to the ear resulting in hearing loss it is possible that these two attributes would be differentially affected. The density of neurons might remain stable which should be reflected by the ECAP response for a constant IPG, while the function of the cells might be abnormal which could be reflected in the IPG effect. There also might be unpredictable effects of neurotrophic factors on SGN morphology and function. Results here show that SGN density and cell size were significantly related (Table II); animals with higher SGN density demonstrated smaller cell size. Additionally, we found that cell size did not help to explain additional proportion of variance in ECAP measures beyond what was accounted for by variance in SGN density. These findings and related issues are further discussed below.

B. Potential effects of neurotrophic treatment on ECAP responses

In the current study the deafening and treatment paradigms were used simply as a method to derive a long-term implanted animal model with a wide range of pathology across subjects. Regardless, it is quite possible that the varied deafening/treatment methods used in the current study affected the outcomes. Neurotrophic factors appear to typically reduce the number of neurons that are lost as a consequence of deafening and preserve or enhance other features of normal cochlea morphology (e.g., peripheral neurites, supporting cells, stria vascularis, spiral ligament) (Shibata et al., 2010; Wise et al., 2011; Budenz et al., 2012). However, research suggests that cell size is sometimes abnormally large in BDNF treated animals (Agterberg et al., 2008; van Loon et al., 2013), and overexpression of NT3 in normal hearing guinea pigs can cause poorer hearing and disorganized peripheral fibers compared to untreated animals (Lee et al., 2016). Perhaps most relevant to the current study, there is evidence that BDNF and NT3 treatment cause abnormal firing of the neurons measured in vitro. Such dysfunction was associated with abnormal expression of select ion channels including voltage-gated potassium channels and calcium-activated potassium channels (Adamson et al., 2002). Needham and colleagues (Needham et al., 2012) measured the effect of combined BDNF and NT-3 treatment in cultured rat SGNs, and found that the firing adaption and resting membrane potentials were normal, but spike latency and firing threshold were affected by the treatment. Taken together, these findings could have important implications for interpretation of the ECAP IPG effect using a neurotrophin-treated animal model. In particular, this approach could help to explain some of the divergent findings when compared to studies which used different deafening techniques as further discussed below.

In the present study, we did find that cell size in the neurotrophin treated animals (group 2) was quite variable, and significantly larger than that of the non-implanted animals (Fig. 3). However, it is important to note that cell size observed in group 2 was not statistically significantly different from that obtained in the implant only group (group 1). Although neurotrophin treatment was not a specific independent variable for the present study, it certainly could have had an effect on the reported results. Abnormal cell size does suggest disturbance of a homeostatic state which could be suggestive of some abnormal function of those cells. Regardless, the statistical interpretation of the data used in the present study does not suggest that cell size is a primary factor contributing to the suprathreshold ECAP measures examined here, for either a fixed or changing IPG. However, the correlational analyses shown in Table II do indicate that cell size is significantly correlated with most of the ECAP measures reported (all except the N1 latency), suggesting that suprathreshold ECAP measures are at least somewhat related to cell size density, perhaps just to a lesser extent than SGN density.

C. Comparison to previous findings

Previous studies from our laboratory found that insertion trauma and recovery functions for ECAP measures (linear slope and peak amplitude) obtained with a constant IPG duration during the first months after implantation were related to the long-term neural health of the implanted ear (Pfingst et al., 2015a; Schvartz-Leyzac et al., 2019). Here, we assessed how ECAPs measured in a stable, post-recovery period (>100 days post implantation) reflect neural and non-neural attributes assessed within the same time period, or shortly thereafter.

The results shown here (Fig. 5) largely agree with those reported by Ramekers and colleagues (Ramekers et al., 2014) but some differences are notable. Ramekers and colleagues (Ramekers et al., 2014) found that the IPG effect for ECAP AGF slope and N1 latency were significantly related to measures of neural health (SGN density and cell size) but a similar effect was not found for the IPG effect on the peak amplitude measure. In the present study we found that SGN density accounted for a significant proportion of the variance in the IPG effect for ECAP AGF linear slope and peak amplitude, but not the N1 latency measure. We also found that cell size did not help to explain additional variance in ECAP measures.

Methodological differences might help explain the different effects on N1 latency in the present studies and those reported by Ramekers and colleagues (Ramekers et al., 2014), which could stem from differences in neural health and properties between animals used in either study. Note that some of the discussion points related to this argument have been reviewed above. First, the animals examined by Ramekers and colleagues (Ramekers et al., 2014; Ramekers et al., 2015) were either acutely implanted or chronically implanted animals, but ECAP measures were obtained only through 12 weeks post implantation. Animals in the current study were implanted for about 4–15 months before data was examined and histological analysis performed. Further, as noted above, recently published data suggest that ECAP responses change over time in cochlear implanted guinea pigs (Pfingst et al., 2015a; Schvartz-Leyzac et al., 2019); most animals show stable responses by three months post implantation, while in others ECAP responses seem to continuously change over time. As stated above, it is possible the dissimilar approaches for deafening/treatment could have affected the results.

The IPG effect for N1 latency is probably particularly sensitive to the cochlear-health related variables that might have differed across the two studies. The effect of IPG manipulation might be minimal if the excitatory phase of the biphasic pulse is the leading phase. There is evidence in animal subjects of different species that either a cathodic or anodic phase can drive the ECAP response (Miller et al., 1998; Matsuoka et al., 2000; Miller et al., 2001). Some have hypothesized that these differences in animals might be caused by species dependent differences such as cochlear anatomy affecting the physical distance and orientation of the electrodes with respect to the stimulated neurons (Miller et al., 1998; Matsuoka et al., 2000). The excitatory phase may also depend on neural health and anatomy; specifically, if the peripheral process is present and/or myelinated (Rattay, 1999; Rattay et al., 2001). If the excitatory phase differs across animals used in the current study, then the use of an alternating stimulus polarity (used in both the present study and by Ramekers et al., 2014) makes it difficult to meaningfully interpret how the IPG effect for N1 latency might be related to SGN density and might help to explain discrepancies observed between the two studies.

D. Contribution of impedance values to ECAP measures

The present study showed that, while variance in SGN density across animals accounted for a significant proportion of variance in ECAP AGF linear slope (Fig. 5), variance in simple impedance measures across animals also contributed significantly to the overall model for the linear slope measure. However, given that this finding was the only significant regression model, impedance does not seem to be a dominant factor that contributes to suprathreshold ECAP measures. As expected based on some results obtained in human CI users (Schvartz-Leyzac and Pfingst, 2016; Scheperle, 2017), results showed that a greater number of neurons accounted for larger ECAP AGF slope values, but higher impedance values were associated with lower ECAP AGF slope values. Higher impedances might reflect the accumulation of denser tissue on or near the recording or stimulating electrode, resulting in lower ECAP AGF values. ECAP AGF slope values for short and long duration IPGs should be affected in similar ways by the conditions near the electrodes and those effects would cancel when calculating the difference between two growth functions at a given recording site. Impedance values across animals were not related to the ECAP IPG effect for any measure in the current study (Table II). However, these findings also show that impedance measures were not necessarily strongly correlated with ECAPs measured using a fixed IPG duration. Hence, these results somewhat contrast with previous studies performed in humans in which approximately 40% of the cases, across-site variation in impedance values was correlated with across-site variation in ECAP AGF linear slope values, but this was not true in any subject when examining the relationship between impedance and the IPG effect for ECAP AGF linear slope (Schvartz-Leyzac and Pfingst, 2016; Scheperle, 2017).

The simple impedance measures used here encompass resistance and reactive components which might depend on the frequency of the signal, the geometry of the electrode surface, or the properties of fluids or tissue surrounding the electrode (Tykocinski et al., 2001; Duan et al., 2004; Franks et al., 2005). These factors could contribute to discrepancies between the contribution of impedance measures to ECAP measures assessed in humans and animals. Further work is needed to determine the extent to which specific impedance components influence ECAP recordings in both animal and human subjects.

E. Conclusions

Results of the current study extend those from previous studies that suggest that nearly 50% of the variance observed in some ECAP measures can be accounted for by variation in SGN density in the animal model. A growing number of studies have shown that postoperative speech recognition is related to certain ECAP measures to various degrees (Kim et al., 2010; Kim et al., 2017; Scheperle, 2017; van Eijl et al., 2017; Schvartz-Leyzac and Pfingst, 2018). This suggests that the neural status, as reflected in these ECAP measures, is indeed important for speech recognition in cochlear implanted humans. Currently, ECAP measures have low clinical utility, but they have the potential to be used more extensively in clinic, perhaps in a site-selection approach similar to previous work with psychophysical measures [e.g., Garadat et al. (2013)]. Based on the results presented here and previously (Schvartz-Leyzac and Pfingst, 2016, 2018), we conclude that a suprathreshold ECAP measure which examines the IPG effect is relatively unaffected by non-neural factors that are related to impedance values (e.g., tissue growth). Further work is needed in human CI recipients to better understand if ECAP measures can be efficiently and effectively used to improve programming and CI outcomes.