Components of impedance in a cochlear implant animal model with TGFβ1-accelerated fibrosis

Abstract

Outcomes of cochlear implantation are likely influenced by the biological state of the cochlea. Fibrosis is a pathological change frequently seen in implanted ears. The goal of this work was to investigate the relationship between fibrosis and impedance. To that end, we employed an animal model of extensive fibrosis and tested whether aspects of impedance differed from controls. Specifically, an adenovirus with a TGF-β1 gene insert (Ad.TGF-β1) was injected into guinea pig scala tympani to elicit rapid onset fibrosis and investigate the relation between fibrosis and impedance. We found a significant correlation between treatment and rate of impedance increase. A physical circuit model of impedance was used to separate the effect of fibrosis from other confounding factors. Supported by preliminary, yet nonconclusive, electron microscopy data, this modeling suggested that deposits on the electrode surface are an important contributor to impedance change over time.

1 -. INTRODUCTION

A cochlear implant is the most successful neural implant available today. While advances in implant technology have led to generally good outcomes, there remains variability in the derived benefit (most notably understanding speech in noise and music appreciation). This variability in outcomes probably comes from the intersection of multiple factors (Moberly et al., 2016), including the growth of intrascalar tissue the scala tympani around the implant. The presence of fibrosis has been well documented (Linthicum et al., 1991, Migirov et al., 2011, Nadol and Eddington, 2004, Nadol et al., 2014, Seyyedi and Nadol, 2014) and has the potential to interfere with acoustical hearing (Quesnel et al., 2016), however it is unclear how this may affect electrical hearing (Swiderski et al., 2020). This uncertainty is in part because noninvasive methods to directly measure the amount of intrascalar tissue surrounding a cochlear implant are limited. Promising work has successfully correlated intrascalar tissue with an increase in electrical impedance, however tissue does not seem to be sufficient to cause increased impedance (Wilk et al., 2016). Regardless, fibrous growth may affect the flow of current from the electrode and therefore be noninvasively measurable by recording the change in impedance over time.

Impedance, in a general sense, is a measure of how much a system resists the flow of electrical current, either by a reduction in signal strength, or by a delay in time. When considering more complicated signals, such as a square pulse, impedance may affect each frequency of the signal differently. In its most general form, impedance at a single frequency is represented as a complex number. Sometimes referred to as “complex impedance”, this number can be broken down into a real component, the resistance, and an imaginary component, the reactance and roughly pertain to a reduction in signal strength and phase delay respectively. This complex number can be transformed in various other ways, leading to mathematically equivalent representations with various physical interpretations. Additionally, there are different ways of recording impedance, where two techniques may give the same underlying result but are obscured by mathematical refactoring. Thus, while impedance has been an area of active research, there are many different methods involving impedance in cochlear implant literature.

Most clinical applications use a simplistic measure of impedance; a known current pulse is passed through the electrode and the recorded voltage is used to calculate a single value (Hey et al., 2015). This measure can be readily recorded by modern clinical cochlear implant software and is well suited for monitoring implant integrity. Other studies use a biphasic pulse to derive an access resistance and polarization impedance (Huang et al., 2007, Linke et al., 2015, Newbold et al., 2004, Ni et al., 1992, Tykocinski et al., 2005, Wilk et al., 2016), descriptive measures of the resulting waveform. These two measures apply to a square pulse and are informative measures but lack clear electrochemical basis of their origin. The third method involves measuring the impulse response of the electrode to a current pulse which can derive the frequency response function of impedance. This can indeed fully characterize the impedance of a system at every frequency; however, the signal to noise ratio may not be constant from one frequency to another, limiting the accuracy of the measurement (Barsoukov and Macdonald, 2005). The most complete way to measure impedance is by presenting small amplitude (10 μA) sinusoidal current signals at multiple frequencies and measuring the resulting voltage wave, ensuring a high and consistent signal to noise ratio.

The impedance of a cochlear implant may be affected electrochemical reactions between the implant and tissue, including the various mediums the electrical signal may encounter: perilymph, neural tissue, bone, and other tissue; each with different electrical properties. Additionally, electrochemical reactions at the electrode surface may also affect impedance, for example by the formation of an electrolytic double at the surface of the electrode (Duan et al., 2004, Linke et al., 2015, Newbold et al., 2004). It is very likely that the impedance measured from a cochlear implant electrode is a combination of all of these; thus, to parse the contribution of only fibrosis it would be useful to create a model of these sources of impedance, which has been done previously for a general bioelectric system (Franks et al., 2005). We build upon that work by expanding the model for a specific use for a cochlear implant. The electrical components of this model were fit to our recorded impedance data. By determining how these components changed over time, we could better identify what components of the system were causing the change in impedance and relate those components to the extent of fibrosis. We hypothesized that fibrosis would change the resistivity of the medium surrounding the implant to such an extent that we would be able to measure a change in the spreading resistance over time. To explore this potential relationship, we measured impedance in two groups of animals: one where rapid onset intrascalar tissue was induced by via Ad.TGF-β1 treatment and one untreated control that we expected to have little to no tissue formation at the end of the experiment. We hypothesized that there would be measurable changes at the electrode surface that would affect impedance; therefore, the electrodes were viewed with electron microscopy to identify any physical changes. We found that the increase in impedance observed after implantation was not correlated to the amount of fibrosis but was most likely due to the presence of biological deposits, such as protein, on the electrode surface.

2 -. METHODS

2.1. Animals

Ten pathogen-free, pigmented, adult male guinea pigs were used in this study. The guinea pigs ranged from 300–600g and were housed and maintained by the Unit for Laboratory Animal Medicine at the University of Michigan. The study’s animal use protocol was reviewed and approved by the University of Michigan Committee on the Use and Care of Animals.

2.2. Study Design

All animals were implanted with an activated, two-channel platinum ball electrode implant. Half of the animals were inoculated with Ad.TGF-β1(Leask and Abraham, 2004) prior to implantation to induce fibrosis(Kawamoto et al., 2003). Briefly, the adenovirus (serotype 5 with E1A/E1B and part of E3 genes deleted) had a CMV promoter driving expression of the transgene. The vector was prepared by the Viral Vector Core at the University of Michigan Medical School. The adenovirus vector stock contained 10% glycerol and was stored at −80°C until thawed for use. The injected solution had a titer of 10¹⁰ particles/ml. The remaining five animals were untreated and implanted as a control. Complex impedances were measured in two sessions per day for two weeks, a time point by which the scala tympani in TGF-β1-treated animals has been shown to be filled by fibrosis (Kawamoto et al., 2003). After each animal’s last recording session, it was euthanized, perfused with fixative and decapitated. The temporal bones were carefully dissected to determine implant placement, then prepared for histological analysis, as described below.

2.3. Electrode Activation

To reduce initial impedance, prior to implantation, cochlear implants were activated by inserting the electrode array and ground wire in a saline bath and applying a cyclic voltage waveform over each electrode. The stimulus was presented for a minimum of five minutes, after which the magnitude of impedance at 1 kHz was recorded.

2.4. Implantation and Inoculation Procedure

Animals were anesthetized with ketamine (40 mg/kg) and xylazine (10 mg/kg), then placed on a self-regulating heating pad. Following a post-auricular incision, the temporal bone was exposed, and the bulla opened. For the animals receiving TGF-β1 treatment, a small cochleostomy was made with a hand drill approximately 0.7 mm apical to the round window on the basal turn of the cochlea. Through the cochleostomy, 2 μL of Ad.TGF-β1 was injected into the scala tympani using a cannula and infusion pump at a rate of 1 μL per minute. The cannula was left in place for approximately 20 minutes while the support screws for the implant connector were placed. Three screws were inserted into the skull around bregma to secure an inverted “anchor” bolt. This bolt was used for securing the intermediate connector between the impedance meter and the implant connector (mounted on the skull with acrylic). Then, the cannula was removed and the cochleostomy enlarged with a diamond burr and the surrounding area cleaned of debris with a cotton pledget. The electrode array (2 platinum balls imbedded in a silastic carrier; MED EL, Innsbruck Austria) was then inserted approximately 4 – 4.4 mm into scala tympani and secured to the bulla with a silk suture; the implant sealed the cochleostomy sufficiently such that further sealing was not necessary. The opening in the bulla was sealed with Durelon cement and the incision was sutured shut. Animals that did not receive TGF-β1 treatment were implanted using the steps above, except for the omission of those pertaining to the cannulation and inoculation procedure.

2.5. Complex Impedance Measurements

Complex impedance measurements were performed in awake animals in two sessions per day (morning and evening) for approximately 14 days, replicating the timeframe of histology described in (Kawamoto et al., 2003). We performed three recordings per session, using an Agilent E4980A precision LCR meter. Although most animals were tested for 14 days, one animal suffered a short between the recording electrode and ground, rendering the implant nonviable and was terminated earlier while another animal was tested on day 15. Complex impedance was recorded from the most apically inserted electrode with respect to a remote ground set at the anchor bolt. Complex impedance was determined using a 10 μA sinusoidal input signal presented at frequencies ranging from 0.1 to 100 kHz. The 28 sessions were used to test whether temporal patterns in impedance differed between treatment groups. Occasionally, due to a poor connection or disconnect, the values of the recording would fluctuate. To combat this variability, we recorded three separate measurements at each frequency at each time, of which the median was used for further analysis.

2.6. Circuit Modeling

A general equivalent circuit model of electrode impedance was taken from Franks and others (Franks et al., 2005) and modified for this experiment. The resulting circuit model consisted of four electrical components (Figure 1), each with an electrochemical basis briefly described here. The two parallel components are attributed to processes local to the surface of the electrode. The charge transfer resistance (Rct) accounts for the net effect of all faradaic reactions occurring at the surface of the electrode (Antano-Lopez et al., 2001). The constant phase element (CPE) accounts for impedance due to the electrolytic double layer capacitance formed at the surface of the electrode (Barsoukov and Macdonald, 2005); its charging and discharging is just a redistribution of ions near the electrode; there is no chemical change (Brummer and Turner, 1975). The CPE is mathematically described as a pseudo capacitor with a constant phase delay differing from the 90-degree phase delay of an ideal capacitor (Equation 2). The deviation from an ideal capacitor likely stems from a non-uniform surface geometry of the electrode (Pajkossy, 2005); however, the exact mechanism is not fully understood. The spreading resistance (Rs) accounts for the resistance of the bulk medium surrounding the electrode (Tykocinski et al., 2005). Finally, a series capacitor (C) represents the complex resistivity of intermediate tissue and other tissue anisotropy; complex behavior of conductivity has been observed in ovine skeletal muscle (Kwon et al., 2017), and is expected to be present in other mammalian tissues. The mathematical formulation of the circuit model can be seen in Equation 1.

$$Z_{eqiv}={\left(\frac{1}{Rct}+{(j\omega Q)}^n\right)}^{-1}+R_s+\frac{1}{j\omega C}$$ Eq 1.

For each impedance recording, the values for the components of the equivalent circuit model were fit using the function “lsqcurvefit” in MATLAB which employs the Levenberg Marquardt algorithm, a form of gradient decent (Moré, 1977). Our impedance recording data were nonlinear as a function of frequency and changes over time may be exaggerated or diminished depending on the frequency of interest. The equivalent circuit model was fit to the frequency dependent recording at each timepoint. For simplicity in analysis, we choose to only look at the impedance values at one frequency, 1 kHz.

Figure 1. Equivalent Circuit model:

The equivalent circuit model of impedance has been adapted from a general model of aqueous biomedical electrodes. The model may be subdivided into location based on the components electrochemical basis and is mathematically defined by Eq 1.

2.7. Histology

At the end of the study period, animals were anesthetized as described above, then transcardially perfused with 4% paraformaldehyde. Animals were decapitated, and the heads were immersed in 4% paraformaldehyde for three days to enhance fixation. The heads were carefully dissected to remove the temporal bones without disturbing the position of the electrode. The temporal bones were decalcified in 3% EDTA until the bones were soft, approximately three months, and the decalcified bone was thinned allowing for visual determination of the stimulating electrode location. This location was marked using physical scoring of the decalcified bone, and the array explanted and stored in phosphate buffered saline at 4°C to preserve any adhering tissue. The decalcified cochleae were dehydrated in ethanol, embedded in JB-4 (Electron Microscopy Sciences, Hatfield, PA, USA) and sectioned with glass knives at 3 μm thickness parallel to the midmodiolar plane passing through the location of the electrode. A series of 42 sections including the midmodiolar plane was collected for each cochlea. One section from the first 10 was chosen at random for analysis, along with every third section of the next 12, for a total of 5 sections spanning 39 μm. The fibrosis in the basal turn of scala tympani was scored on a qualitative scale from 0 (no fibrosis) to 5 (dense fibrosis) averaged across the five sections to produce an individual’s score for statistical analyses.

2.8. Environmental Scanning Electron Microscopy of the Electrode Surface

Explanted cochlear implants were secured to a Peltier cooling stage with double sided carbon adhesive tape and scanned with a FEI Quanta 3D e-SEM/FIB in environmental mode. In contrast to the high vacuum conditions with conventional scanning electron microscopy (SEM), environmental SEM allows for higher temperatures and pressures, as well as imaging non-conductive and wet material which allows for better preservation of biological material while imaging. Images of each electrode on an array were taken and saved for further analysis. An electrode array that had never been implanted was also imaged to serve as reference. Images were examined by three blinded observers who rank ordered the subjects by the extent of observable material adhering to the electrode surface.

2.9. Statistics and Data Analysis

Due to the small number of subjects in this experiment, we elected to use a Bayesian approach to our regressions. Bayesian statistics can better estimate coefficients of interest when limited by a small number of samples compared to more traditional, frequentist, methods (McNeish, 2016, Muthen and Asparouhov, 2012, Zhang et al., 2016, van de Schoot et al., 2015). This is because Bayesian regressions allow us to use hypotheses to influence the outcome of the regression in the form of a prior distribution. As data are added, the prior distributions are updated through Bayes theorem to reflect the new information (Jeffreys, 1957). These updated distributions are called the posterior distributions. As more data are added, the posterior distributions are further updated to more confidently reflect the true value of the regression. By intelligently selecting a prior distribution, we can avoid the power issues of frequentist statistics and come to a better estimate of our coefficients of interest (McNeish, 2016, Muthen and Asparouhov, 2012, Zhang et al., 2016, van de Schoot et al., 2015).

In this experiment we were interested in how certain variables change with respect to time. We fit a generalized linear model (GLM) as seen in Equation 2 where Y is a datum taken at time “$t$” in days post implantation and $T$ is a Boolean representing inoculation of Ad.TGF-β1. We have two groups of terms, the first, containing alpha and nu, denote the change over time of our data where alpha is the change over time for the control group and alpha plus nu is the slope of the treated group. In a similar fashion, the second grouped term denotes the starting values, or intercept, of our data at time zero where beta is the intercept of the control and beta plus tau is the intercept of the Ad.TGF-β1 treated group. From these we set the prior distribution for tau to be a normal distribution with mean zero and variance 2 as we expect no difference in intercept between the two groups as the expression of the transforming growth factor does not begin until days after inoculation (Kawamoto et al., 2003). We expect the intercepts of our regression to be positive in all cases, thus the prior for beta was set to a half-cauchy distribution with scale parameter of 50 (Gelman, 2004). We have no prior belief on what the value of the parameters denoting the change over time thus we set both to a relatively uninformative normal distribution with a zero mean and standard deviation of 100. All numbers in the data were scaled to the first order of magnitude and represented by the appropriate prefix.

$$Y=(\alpha +vT)\cdot t+(\beta +\tau T)$$ Eq 2.

Instead of simple values, the output of these regressions are distributions reflecting what we believe the true value of the free variable to be based on the data. To perform these regressions, we use the package PyMC3 in python (Salvatier et al., 2016) using the no U-turn sampler (Hoffman and Gelman, 2014). Convergence is checked using the Gelman-Rubin statistic and by inspecting autocorrelation plots of our samples (Gelman and Rubin, 1992). All reported regressions achieved convergence. R squared values are calculated from a method described by Gelman and others (Gelman et al., 2017). P-values were derived by the probability mass function of the posterior distribution evaluated at zero. Each recorded datum point is the median of three points at an individual time point; this helped to reduce random outliers and increase the overall accuracy of the analysis.

To analyze the condition of our explanted electrodes, each electrode surface was subjectively ranked from least to most coverage by three blinded individuals using a unimplanted electrode as a reference. The rankings were averaged, and a Spearman’s correlation was employed to correlate biological coverage to various other recorded data.

3 -. RESULTS

3.1. Histology

To assess levels of fibrosis, cochleae were fixed and assessed by a treatment-blind individual (figure 2). The scala tympani of all Ad.TGF-β1 treated animals were completely filled with varying density of tissue surrounding the implant, whereas a majority of untreated control animals had minimal fibrosis. Due to unexpected insertion trauma, one control animal exhibited extensive tissue growth despite no additional treatment. As previously reported, we observed a hypertrophied and thick Reissner’s membrane in animals treated with Ad.TGF-β1 (Kawamoto et al., 2003) (data not shown), while no other gross histological abnormalities were observed in either group. While there was a statistically significant difference in the extent of fibrosis between each treatment group (p = 0.016), there was overlap in outcomes for each group, complicating the relationship to impedance as there was no significant correlation between the extent of fibrosis and change in impedance from baseline at implantation day (p = 0.51).

Figure 2. Histological comparison of impedance data to tissue rank.

Exemplar midmodiolar sections of untreated implanted animals (A) and implanted animals treated with Ad.TGF-β1 (B). The change in impedance from implantation for each animal, grouped by treatment and against intrascalar tissue (C) did not significantly correlate to the change from baseline of recorded impedance magnitude at 1 kHz. The extents of intrascalar fibrosis shown here were near the middle of the tissue range for each respective treatment group, with both examples exhibiting similar change in impedance from implantation.

3.2. Recorded Complex Impedance

In each animal, we recorded the impedance in complex form as a function of frequency: containing the real value known as resistance, and the imaginary value, termed reactance. These values represent the electrical and chemical interactions between a medium and electrical current. These values may not be stable and can change over time corresponding to changes in the surrounding environment. Fibrosis formation may induce such a change and therefore may temporally affect impedance.

To investigate potential trends over time, we fit our time series impedance data at specific frequencies with a generalized linear model using Bayesian techniques with the python package PyMC3 (Salvatier et al., 2016). This statistical model fit estimators for the change over time (slope) and initial value (intercept) for the control group, while also accounting for the additive effect to each from the Ad.TGF-β1 treatment. The animals starting impedance values were not statistically different between groups (figure 3A, 3C), suggesting any effect from Ad.TGF-β1 was not instantaneous, however the rate of impedance growth was significantly higher for the untreated control group (p = 9.65e-15, figure 3A, 3B). These effects were consistent for each frequency at which impedance was recorded (figure 3D, 3E), however the extent of this difference diminished inversely proportional to frequency, suggesting that changes in some capacitive-like process may be driving this change. To simplify the analysis, we will highlight a single frequency for all further statistics.

Figure 3. Hierarchical modeling of Impedance Magnitude:

The time series recorded impedance data (A) at 1 kHz for the control animals (Orange) and Ad.TGF-β1 treated animals (Blue); posterior predictive regression are overlayed on the time series data. The posterior distribution of the slope (B) and intercept (C) of a linear model fit with Bayesian methods at 1 kHz. The slope and intercept of the control group were fit directly (α and β respectively) while the slope and intercept distributions of the Ad.TGF-β1 treatment group were calculated based on the calculated difference (ν and τ) due to treatment, (α + ν) and (β + τ) respectively. The impedance of the control animals increased at a significantly faster rate than the Ad.TGF-β1 treated animals. This significant increase in slope was consistent over every frequency (D), while the intercepts of each group are not statistical different below 10 kHz (E).

3.3. Equivalent Circuit of Complex Impedance

Due to the inverse relationship between impedance and frequency, it is likely that there was some capacitive like effect on impedance by the system, however the exact extent cannot be elucidated without further analysis. To explore these relationships, we compose an equivalent circuit model of electrode impedance using circuit components with a preexisting electrochemical basis (Barsoukov and Macdonald, 2005). By fitting the components of this model to the recorded impedance data and investigating how the values of these components change over time, we may gain insight to the underlying processes affecting impedance change.

Our equivalent circuit model expands on a more general model of impedance for aqueous, biomedical electrodes. The model was found to fit with good agreement to the recorded impedance data, with a representative example shown in figure 4. In a similar fashion to the recorded impedance magnitude, we fit a GLM for the components of the equivalent circuit model and investigated how each changed over time. The results of these regressions can be seen in figure 5. The two components that had statistically significant differences in slopes between treatment groups were the constant phase element (CPE) (p = 4.63E-06, figure 5) and charge transfer resistance (R_ct) (p = 2.98e-07, figure 5), both which are believed to correspond with the condition of the surface of the electrode, while series resistance (R_s) and capacitance (C), which are believed to correspond with the impedance of the bulk tissue, did not significantly change over time. This result suggests that the components driving a change in impedance are the CPE and Rct, implying that changes at the surface of the electrode are responsible for change in impedance. Furthermore, we performed multiple analyses of variance investigating observed tissue and the change in each element value from baseline, but we found that bulk fibrosis was not predictive of the change in element values of any element in the equivalent circuit model.

Figure 4. Exemplar fit of equivalent circuit model.

(A) An exemplar fit of our equivalent circuit model plot against recorded impedance resistance (left) and reactance (right) compared to a more general model of aqueous biomedical electrode impedance. (B) The accuracy of our model fit is significant increased as seen by a drop in AIC for each impedance data compared to the general model.

Figure 5. Hierarchical modeling of Equivalent Circuit Components:

Posterior distributions of a linear model fit with Bayesian methods for each element of the equivalent circuit model when fit to time series impedance data at 1 kHz. As in Figure 3, the slope and intercept of each element’s control group were fit directly (α and β respectively) while the slope and intercept distributions of the Ad.TGF-β1 treatment group were calculated based on the calculated difference (ν and τ) due to treatment, (α + ν) and (β + τ) respectively. The values of the constant phase element (CPE) and charge transfer resistance (R_ct) of animals treated with Ad.TGF-β1 (Blue) increase was significantly slower when compared to untreated controls (Orange), while there was no significant difference in rate of change for the spreading resistance (R_s) and capacitance (C). There was no significant difference in the intercepts of each model between groups. Furthermore, the extent of tissue growth was not predictive of the total change in any element from baseline.

3.4. Scanning Electron Microscopy of Implant Surfaces

Figure 6 shows representative environmental SEM images of cochlear-implant ball electrodes. Unidentified material was observed on some of the explanted CI electrodes to a greater or lesser extent. Electrodes were ranked by three blinded individuals based on extent of material coverage independent of treatment. The ranks were averaged and correlated to the recorded impedance values of each corresponding animal by a Spearman’s correlation. A significant correlation was found between the ranking and the constant phase element (ρ = 0.79, p = 0.02). There were not significant correlations between the ranking and the spreading resistance (ρ = 0.22, p = 0.61) or the charge transfer resistance (ρ = 0.57, p = 0.13).

Figure 6. Scanning electron micrographs of cochlear implant ball electrodes.

The surface of an unimplanted electrode (A) can be compared to explanted electrodes from an untreated control animal (B) and one treated with Ad.TGF-β1 (C).

4 -. DISCUSSION

Fibrosis was found to varying degrees in both the Ad.TGF-β1 and control treatment group, but the largest extent of fibrosis was found in the Ad.TGF-β1 treatment group. Despite this, the impact on complex impedance from this fibrosis on the predicted circuit elements for both treatment groups were marginal. Instead, the differences in complex impedance across animals at the end of the observation period (about two weeks after implantation) were associated with higher CPE impedance, suggesting changes in the electrode-electrolyte interface, possibly due to material deposited on the electrode surface. This result stems from the equivalent circuit modeling, where the only components with similar time course of impedance, and treatment difference in recorded impedance, were the components theoretically associated with the surface of the electrode.

The rating of fibrous tissue in the inner ear was not significantly correlated with change in complex impedance. Surprisingly, the large amount of tissue from the Ad.TGF-β1 treatment seemed to have slowed the increase in impedance. This might be because the type of tissue induced by Ad.TGF-β1 had different electrical properties when compared to fibrosis from the normal immune response. Yet, that is unlikely, as the observed tissues are morphologically similar between the two treatment groups. Thus, it is likely that the fibrous tissues in both treatment groups are the same and that their electrical properties, and thus their effect on impedance in identical amounts, are equivalent.

Even though the differences in fibrosis rating between the two treatment groups were notable, the fibrosis rankings were not related to impedance. This lack of correlation may be due to using a distant ground in this study. For a distant ground, the spreading resistance likely reflects the lumped electrical characteristics of all tissue between the stimulating electrode and ground, including bone, nervous tissue, and muscle among others. Considering how small an effect the amount of fibrotic growth that the cochlea may have compared to the vast amount of tissue in the rest of the current path, it is likely that the extent to which the intracochlear fibrosis contributed to the spreading resistance was small.

Then what explains the difference in impedance between the two treatment groups? One explanation involves the interface between the surface of the electrode and the immediately surrounding media. It has been observed that a thin layer of perilymph exhibits similar conductance to bulk saline solution which suggests that increases in impedance are related to changes of the electrolyte close to the electrode surface (Duan et al., 2004). For platinum electrodes, faradaic reactions, and capacitive charging, are methods of charge injection (Cogan, 2008). Charge injection necessitates the presence of water (Bockris et al., 1963); consequently, anything displacing water from the metal electrode would affect impedance. Duan and colleagues hypothesized that a deficiency of perilymph adjacent to stimulating electrodes, likely from cell growth on the electrode, explained impedance increases in normally implanted cats (Duan et al., 2004).

We observed an increase in complex impedance due to the double layer capacitance, represented mathematically by the constant phase element in the equivalent circuit impedance model (Duan et al., 2004, Franks et al., 2005, Weiland and Anderson, 2000). This increase was highly correlated with the overall complex impedance, suggesting that changes to the double layer capacitance was the driving force behind the change in recorded complex impedance over time. This coincides with previous work by Newbold and colleagues, who found a correlation between impedance and the total coverage of fibroblasts on gold electrodes in vitro (Newbold et al., 2004). Durisin and colleagues found evidence of cell adhesion on explanted human electrodes with abnormally high impedances, but also found adhesion on some electrodes with impedances in the normal range (Durisin et al., 2011). Additionally, Duan and colleagues found that cleaning explanted cochlear implant electrodes of a residual protein biofilm increased the double layer capacitance, which in turn lowered impedance (Duan et al., 2004). In agreement with these observations, we found a positive correlation between the coverage of explanted electrodes by unidentified deposit and impedance related to the double-layer capacitance, but no obvious evidence of whole cells. It may be that protein adsorption is causing the impedance increase. Additionally, it is plausible that the rate of this protein adsorption is correlated to the rate of change of the double layer capacitance. A 60% increase in the polarization impedance from standard has been observed by protein adsorption on gold electrodes in vitro (Newbold et al., 2010).

What is not clear is why the Ad.TGF-β1 treated group exhibited lower and more stable impedances when compared to the control group. It may be the case that the rate of fibrosis development differed between the two treatment groups. The rapid onset of fibrosis by the Ad.TGF-β1 group may have displaced perilymph and the associated proteins before appreciable adsorption of biological material on the electrode surface occurred.

5 –. CONCLUSIONS

This experiment provides evidence to suggest that fibrosis in the inner ear minimally affects cochlear implant impedance. While the observed impedance changes are better explained by changes at the metal electrode surface or changes in the electrolytes surrounding the electrode, more work is needed to fully characterize this increase.