When the First Pulse Fades: 75% Reduction in Precedence Effect in Cochlear Implanted Rats

Abstract

It is widely known that users of cochlear implants (CIs) often show deficits in binaural hearing. In particular, interaural time difference (ITD) discrimination tends to be substantially worse in CI users. However, only very few studies have investigated how CI stimulation affects the “precedence effect” (PE), and how this might be affected by abnormal hearing experience during development. We performed behavioral experiments to measure temporal weighting functions (TWFs) to quantify the PE in three different cohorts of rats: normally hearing (NH) acoustically stimulated, neonatally deafened (ND) CI-stimulated, and adult deafened (AD) CI-stimulated animals. Lateralization responses were recorded to bursts of eight stimulus pulses in which the ITD of each pulse varied independently, and probit analysis revealed the perceptual weight of each pulse. The NH and ND-CI cohorts were tested at 50, 300, and 900 pps; the AD-CI animals were tested only at 300 pps. The NH animals show a pronounced “onset dominance” at higher pulse rates, that is, the weighting of the first pulse was much larger than for subsequent pulses, exactly as one would expect given previous studies on human listeners. In contrast, the TWFs for the ND-CI and AD-CI animals were much flatter, with a reduction of onset dominance by ∼75%, especially for higher pulse rates. Since neither of our CI cohorts had any experience of reverberation in their electric hearing, we interpret these results to indicate that the PE may require domain-specific experience to develop.

Introduction

In our everyday environment, spatial cues allow us to locate and separate sound sources, giving us a better understanding of the complex auditory scene around us. In the real world, multiple competing sounds can arrive at overlapping times from different directions and are accompanied by echoes and reverberation, which greatly complicates spatial cue perception. Reverberant listening environments are known to be particularly challenging for cochlear implant (CI) users (Badajoz-Davila et al., 2020; Kerber & Seeber, 2013; Kressner et al., 2018; Xu et al., 2022). The healthy, normal auditory system is, in comparison, surprisingly well able to cope with such confounding factors. One heuristic used by the auditory system to avoid confusion caused by echoes and reverb is to take advantage of the fact that reverberant sound is reflected and therefore does not travel directly from the sound source to the ear, causing it to arrive with a slight delay. In reverberant environments, the auditory system therefore appears to rely heavily on the spatial cues encoded in “the first wave front” of the sound, as these cannot be contaminated by echoes. This reliance on spatial cues encoded in the sound onset is commonly referred to as the precedence effect (PE).

In a classic paradigm to study the PE, a pair of clicks, which are intended to mimic a direct sound source and a single echo (lead and lag click), are presented from loudspeakers positioned at each side of a listener in the free field (Wallach et al., 1949). The time delay between lead and lag click is decreased to a point where the listener only perceives a single auditory event, also termed fusion. This delay, around 5–10 ms for clicks and 50 ms for speech (Litovsky et al., 1999), is called the echo threshold. At delays below the echo threshold, the listeners localize the perceptually merged click pair in the direction of the leading click, a phenomenon also called localization dominance. The direction of the lagging click has very little or no influence on the listeners’ perceived direction of the fused sound. This remains true even if the lag comprises a rapid train of hundreds of clicks rather than just a single click (Freyman et al., 1991).

Thus, in rapid binaural click trains, the first (“onset”) click dominates the interaural time difference (ITD) perception, and the extent of this onset dominance can be elegantly quantified by behavioral measurements of a so-called temporal weighting function (TWF; Stecker & Hafter, 2002; Brown & Stecker, 2010; Stecker, 2014, 2018). To measure a TWF, participants are asked to lateralize a set of binaural click trains comprising a modest number (perhaps up to 20 or 30) clicks, in which the binaural cue value of each click in the train varies randomly and independently. Post-hoc statistical analysis is then performed to compute how strongly the first, second, …, Nth click, respectively, influenced the participant's overall lateralization judgments. Brown and Stecker (2010) used this approach to demonstrate that onset dominance (i.e., a much stronger weighting of the first relative to later clicks) becomes increasingly pronounced at higher click rates, while TWFs are comparatively flat when the click interval is large compared to the echo threshold.

So far, only very few studies have investigated the PE under electric stimulation, and its physiological basis remains almost completely unknown under any form of stimulation. One hypothesis is the possibility that the PE may be mediated by cochlear mechanics. Bianchi et al. (2013) argued that the PE could arise when ringing of the mechanical filters in the cochlea causes persistent effects of the initial wavefront, which then suppresses and shapes the response to subsequent sounds. This hypothesis would imply that there should be no PE in CI patients tested under direct electrical stimulation, given that this bypasses cochlear mechanics entirely. However, Brown et al. (2015) observed a PE in adult-onset deafness CI subjects that was broadly similar to that seen in a normally hearing comparison group. Likewise, van Hoesel (2008) also found evidence of a PE in TWFs measured in late deaf CI users. It therefore seems more likely that as yet unidentified central mechanisms are chiefly responsible for the PE.

Pecka et al. (2007) proposed that a bilateral network involving the dorsal nuclei of the lateral lemniscus and the inferior colliculi may aid in echo suppression. However, a study by Li et al. (2022), who studied TWFs in normally hearing rats both behaviorally and with cortical electrocorticogram responses, suggests that the PE may only emerge even later in the auditory pathway. They observed that, even at the level of the auditory cortex, neural evoked responses at individual cortical recording sites often failed to exhibit clear onset weighting, and a robust PE was only observed in a population-level analysis.

It is also worth considering that the PE may be only one aspect of a number of echo suppression mechanisms that the auditory pathway is capable of deploying, and that the auditory system can adapt rapidly to sudden and often profound changes in the reverberant environment, for example, when we move from a large room with soft furnishings, which has relatively weak echoes with relatively long delays, into a small, bare space where echoes will be much faster and stronger (Tsironis et al., 2024). These adaptations are likely driven by a rapid adjustment of neural tuning properties of auditory cortical neurons (Ivanov et al., 2022). Another reason to suspect that there is much more to the processing of echoes in mammals than simple echo suppression comes from the fact that a number of mammalian species, including bats and dolphins, are well known to be capable of quite impressive acoustic echolocation feats. This would be impossible if the information provided by echoes was suppressed before it could be evaluated, and even humans appear to be relatively easy to train to perform echolocation tasks (Wallmeier et al., 2013), and their PE appears to be strongly diminished while they engage in an echolocation task. Dealing with echoes and reverb can, therefore, be a sophisticated process, and we might predict that the neural circuits involved in the processing of reverberant sounds may require a lot of experience to fully develop, as supported by a recent study (Litovsky et al., 2025). This stands in contrast to a simple binaural cue sensitivity, which is heavily pre-wired by evolution and can emerge with little or no sensory experience (Buchholz et al., 2024; Rosskothen-Kuhl et al., 2021). The development of a neonatally deafened (ND) animal model, which is suitable for behavioral assessment of prosthetic binaural hearing, is now making it possible to start investigating the role of experience in shaping echo suppression circuits and how this affects the PE. With this aim in mind, we performed behavioral experiments on two cohorts of bilaterally implanted CI rats, one of which was ND, the other growing up with normal hearing experience before adult deafening (AD). Both were implanted in young adulthood and trained in a two-alternative forced choice (2-AFC) stimulus lateralization task (Rosskothen-Kuhl et al., 2021). All animals were then tested in a psychoacoustic task very similar to those described by van Hoesel (2008) and Li et al. (2022), which measures their TWF by statistical analysis of a large number of lateralization choices in response to eight-pulse bursts of binaural CI pulses in which the ITDs of each pulse varied randomly and independently. To assess whether normal hearing experience is necessary for normal TWFs with a pronounced PE to emerge, we measured TWFs for the ND-CI animals at pulse rates of 50, 300, and 900 pps and compared them to previously published rat behavioral TWF measurements (Li et al., 2022), which had been obtained using an identical paradigm, except that the animals in that study were normally hearing (NH) and acoustic pulses (clicks) were used instead of electrical CI pulses. In a second set of experiments, aiming to contrast the roles of prior acoustic hearing experience with later prosthetic hearing experiments, we measured TWFs of AD-CI rats at 300 pps, and compared the results from this cohort against those obtained from the NH and the ND-CI cohorts.

Methods

All procedures involving experimental animals reported here were approved by the City University of Hong Kong Animal Research Ethics Sub-Committee and conducted under license by the Department of Health of Hong Kong (#16–52 DH/HA and P/8/2/5) and/or the Regierungspräsidium Freiburg (#35-9185.81/G-17/124, #35-9185.81/G-22/067), as well as by the appropriate local Ethical Review Committee. We confirm that all of our methods were performed in accordance with the relevant guidelines and regulations and that our study is reported in accordance with the ARRIVE guidelines.

Subjects and Surgery

In total, data from 12 CI-implanted, eight ND, and four AD, as well as four NH female Wistar rats, contributed to this study. The data from the NH animals have been previously published by Li et al. (2022). Eight of the CI-supplied rats were neonatally deafened by daily intraperitoneal (i.p.) injections of 400 mg/kg kanamycin from postnatal days 8 to 20, inclusively, as described by Rosskothen-Kuhl and Illing (2012) and Rosskothen-Kuhl et al. (2018, 2024, Figure 1), while the remaining four had normal acoustic hearing experience and were only deafened during the CI surgery in young adulthood. All 12 animals underwent CI surgeries as described by Rosskothen-Kuhl et al. (2021). Briefly, the animals were anesthetized with i.p. injections of ketamine (80 mg/kg) and xylazine (12 mg/kg). The cochlea was exposed using a retroauricular surgical approach. Tympanic membrane and ossicles were removed, and a cochleostomy was performed (0.6 mm diameter drill) over the middle turn of the rat cochlea. At this point, the adult deafening took place in the four AD-CI animals. To induce acute hair cell death, we adapted protocols by Dai et al. (2017) and Balaram et al. (2019): a Hamilton needle was inserted into the cochleostomy and 6 μL of kanamycin solution (50–60 mg/mL Sigma-Aldrich Co.) were injected into the perilymphatic fluid, 3 µL were applied in the apical direction, and 3 µL in the basal direction. This acute kanamycin injection into the cochlea was repeated 3 times every 2 min. Afterwards, cochlear implantation preceded exactly as for the ND-CI cohort as follows: dedicated CI animal arrays, obtained either from Peira (Beerse, Belgium) or from MED-EL Medical Electronics (Innsbruck, Austria), were implanted through a cochleostomy window into the middle turn of each cochlea such that the tip electrode, used for intracochlear stimulation, sits approximately in the 4–8 kHz region. All CI animals were implanted bilaterally simultaneously and received binaurally synchronized stimulation from the beginning. Electric or acoustic auditory brainstem responses (eABRs/ABRs) were assessed to ensure normal hearing in the NH cohort, or hearing loss and successful CI insertion in the ND-CI and AD-CI cohorts, respectively.

Figure 1.

Experimental design. (A) Schema of behavioral setup. Rats were bilaterally implanted with cochlear implants (CIs) via cochleostomy into the middle turn of the cochlea. CI animals initiate trials by licking the central start spout (“S”). This triggered electric pulse bursts to their left (blue) and right (red) ears. Pulse bursts consist of eight pulses displayed in (B) and (C). To receive a water reward, the animals had to lick response spouts on the side (left “L” or right “R”). In this example, the left ear is leading. (B) Example “Honesty Trial” with only the left leading ITDs and random jitter per pulse. Responses to honesty trials were only rewarded if the side of the leading pulse burst was chosen. (C) Example “Probe Trial” with interaural time differences (ITDs) jumping randomly between the left and right ear leading. Responses to probe trials were always rewarded, independently of the animal's choice. Illustration created with BioRender.com.

Behavioral Training Setup

Animals were trained on a 2-AFC lateralization task, using the behavioral setup and protocols described in several previous studies (Buchholz et al., 2024; Buck et al., 2023; Rosskothen-Kuhl et al., 2021; Schnupp et al., 2025), and as schematically illustrated in Figure 1A. Briefly, on testing days, the animals received access to drinking water through the behavioral setup, and the animals were trained to initiate binaural pulse burst stimuli by licking a center spout (“S” in Figure 1A) and then choose either the left or right response spouts based on the ITD stimulus presented to obtain drinking water as a reinforcer.

Stimulus Design and Behavioral Paradigm

The stimulus design and behavioral paradigm were very similar to those previously used by Li et al. (2022) to measure TWFs in NH rats. The previous acoustic study used trains of eight binaural, broad-band (2–22 kHz) pulses (Li et al., 2019, 2022) delivered over tube phones at pulse rates of 50, 300, or 900 Hz. For the CI animals, the acoustic pulses were simply replaced by biphasic electric current pulses (duty cycle: 40.96 μs positive, 40.96 μs at zero, and 40.96 μs negative). All pulses had the same amplitudes. For NH rats, click trains were presented at an average binaural acoustic level of ∼80 dB SPL, while for CI rats, a stimulus intensity of 2–7.5 dB above each animal's eABR threshold was chosen. Electric current pulses were generated with a Tucker Davis Technologies IZ2 programmable constant current stimulator and delivered to the active intracochlear electrodes. The animals had to lateralize bursts of eight pulses in a 2-AFC task, and the ITDs for each pulse in the burst were chosen uniformly and independently from the set {−120, −80, −40, 0, +40, +80, +120} μs. The ITDs tested were, therefore, all within the animal's physiological range of −120 to +120 μs (Koka et al., 2008). Our convention here is to use negative ITD values to denote left ear leading ITDs and vice versa. In order to investigate a possible dependence of the PE on pulse rate, we tested the ND-CI and acoustic cohort with bursts of either 50, 300, or 900 pps. Due to time constraints, the AD-CI cohort was tested at 300 pps only as a translationally relevant pulse rate.

In order to monitor and control the animals’ behavior, the testing sessions randomly interleaved “honesty” and “probe” trials. Examples of the two trial types are shown in Figure 1B and C. In honesty trials, the ITD choice of the eight pulses in the stimulus was constrained such that they all had the same sign, that is, all ITDs pointed to the same side but still varied in magnitude. Each honesty trial, therefore, had only one objectively correct response, and animals were rewarded with a small amount of drinking water only if they responded accordingly. Incorrect responses during honesty trials triggered a time-out of ∼15 s duration during which no further trials could be initiated. In probe trials, the ITDs for each click varied without constraint, and probe trials thus routinely contained pulses with both left-ear leading and right-ear leading ITDs. How such a probe trial “ought to be” lateralized does, of course, depend on the animal's own TWF, which was a priori unknown and which we were trying to infer from the animal's perceptual choices. In order not to bias these choices during probe trials, we rewarded probe trial responses irrespective of which side was chosen. Honesty trials outnumbered probe trials at a ratio of 4:1 in order to “keep the animals honest,” as this ensured that the animals needed to keep evaluating the binaural cues provided by the stimuli if they wanted to avoid frequent time-outs, preventing them from reaping rewards at a high rate. To ensure data quality, we only included testing sessions for which honesty trial performance exceeded 75% in the final analysis, and we ensured that there was no way for the animals to know whether a given trial was an honesty or a probe trial. Note that meeting this 75% correct honesty trial performance criterion is challenging for the animals, given that rats typically have ITD thresholds close to 50 μs, and two-thirds of the pulses in the honesty trials (those with ITDs of ± 40 or 80 μs) can, therefore, be considered “near threshold.” To make it a little easier for the animals to reach this criterion, we presented each burst consisting of eight binaural pulses not just once, but repeated the burst between 5 and 12 times in quick succession, with a 100 ms silent interval between repeats. The number of repeats of each burst was adjusted as needed for each animal and pulse rate to achieve the required ≥ 75% correct honesty trial performance criterion.

Analysis

To compute the perceptual “weights” of each pulse in the burst, we used a regression analysis approach similar to that used by van Hoesel (2007) and identical to that previously described by Li et al. (2022). Specifically, we fitted a probit regression model to the behavioral data. The equation for the probit analysis is given by the following equation:

$$P({}^{″}\mathrm{right}{}^{″}\mathrm{response})=\mathrm{\Phi }({\beta }_0+{\beta }_1{\mathrm{lTD}}_1+{\beta }_2{\mathrm{lTD}}_2+{\beta }_3{\mathrm{lTD}}_3+{\beta }_4{\mathrm{lTD}}_4+{\beta }_5{\mathrm{lTD}}_5+{\beta }_6{\mathrm{lTD}}_6+{\beta }_7{\mathrm{lTD}}_7+{\beta }_8{\mathrm{lTD}}_8).$$

Here, P is the probability of an animal choosing the “right” side spout, and β₀ is a constant offset that can capture a possible bias, or idiosyncratic preference, that an animal may have for one of the sides. ITD₁, ITD₂, …, ITD _N correspond to the ITD, in ms, of the 1st, 2nd, …, Nth pulse of the pulse burst presented during a given trial. The probit coefficients β₁, β₂,…, β₈, quantify the relative contribution, or “weight,” of each of the eight pulses towards the animal's behavior decision. Within our probit framework, the interpretation of these coefficients is straightforward: an ITD of x ms at the i-th click is predicted to increase the probability that the animal responds on the right-hand side by x⋅β _i standard deviations on a cumulative standard normal distribution, and the weighted effects of each click are assumed to be additive. The Probit regression analysis was performed for each animal and each pulse rate separately. The β coefficients were computed using the Python library function statsmodels.api.Probit(), which also estimated p-values for whether each β was significantly different from zero. Only probe trials were included in the probit analysis, and only from sessions in which the honesty trial performance exceeded 75% correct.

The phenomenon of the PE leads to the expectation that the coefficient of the first pulse, β₁, should be substantially larger than the β values later in the pulse train. To quantify the extent to which this expectation is met, we defined the “onset index” as β₁–mean(β₃, β₄, β₅, β₆). This metric computes how much more strongly the animal's behavioral response weighted the first pulse compared to the pulses in the middle of the burst. The size of the onset index can thus be taken as a measure of the strength of the PE.

The relationship between pulse rate and onset index was analyzed using a linear mixed-effects model. Pulse rate was log-transformed prior to analysis. Fixed effects included log(pulse rate), cohort (ND-CI vs. NH-acoustic), and their interaction. To account for repeated measurements within animals, animal ID was included as a random effect. Models were fit using the function “mixedlm()” from the “statsmodels” Python library. The fitted model took the form: OnsetIndex = β₀ + β₁ × log(PulseRate) + β₂ × Cohort + β₃ × [log(PulseRate) × Cohort], where Cohort is an indicator variable taking the value 0 for ND-CI animals, and 1 for NH-acoustic animals. In addition to allowing us to determine whether Cohort was a significant factor, the model thus also effectively fitted a regression with slope β₁ to the ND-CI animals and a regression with slope β₁ + β₃ to the NH-acoustic animals, while correcting for the random effects introduced by the fact that each animal contributed multiple data points (one for each pulse rate). For visualization, population-average (marginal) predictions were generated from the fixed-effect coefficients, yielding cohort-specific linear trends across the range of tested pulse rates. Uncertainty in these slope estimates was quantified using standard errors derived from the fixed effects variance–covariance matrix.

Results

Using the methods just described, we were able to collect data from a cohort of eight ND-CI animals at 50, 300, and 900 pps. Three ND-CI animals at 900 pps failed to reach our performance criterion of 75% correct in honesty trials and had to be excluded. The TWFs measured for the ND-CI animals are plotted in red on the right-hand side of Figure 2, next to the TWFs from the NH animals (previously published by Li et al., 2022) shown in blue on the left. Probit coefficients, which are significantly different from zero at p < .05, are plotted as dark filled circles, and coefficients that are not significant are plotted as small circles. The average TWFs across animals for each pulse rate are shown as a dotted magenta line, with the range of ±SEM of the probit coefficients indicated by the shaded areas.

Figure 2.

TWFs for ND-CI and NH cohorts. The top, middle, and bottom rows show results at 50, 300, and 900 pps, respectively. Each circle (NH-acoustic: blue; ND-CI: red) shows the temporal weighting (probit regression coefficient in 1/ms) for each of the eight pulses within one burst. Dark filled circles show coefficients that are significantly different from zero (p < .05), small circles show non-significant coefficients. Data from each animal are shown in a different hue. Error bars indicate SEM of individual coefficients. Magenta dotted lines show the mean across animals in each condition, and the grey shading shows the SEM across animals. NH-acoustic data originally from Li et al. (2022).

It is apparent from Figure 2 that the TWFs from the NH cohort (previously published by Li et al., 2022) are comparable with previous studies on NH humans (Brown & Stecker, 2010) in that onset dominance was present, and was particularly prominent at the higher pulse rates tested. The onset dominance at 50 pps was much less pronounced, but perhaps nevertheless larger than one might have expected, given that, at this relatively low pulse rate, the inter-pulse intervals are quite large relative to typical click echo thresholds reported for human listeners. It is possible that the observed, modest up-weighting of the initial pulses at these low pulse rates might therefore result less from echo suppression effects than from some kind of “primacy” effects during the animals’ perceptual decision making. In any event, a very similar onset dominance was also observed under electric stimulation in the ND-CI cohort at 50 pps, but for the ND-CI animals, this onset dominance did not grow as pulse rates increased to 300 and 900 pps. Instead, it declined slightly, showing an opposite trend to what would be expected based on previous NH rat or human studies. At 900 pps, the average probit coefficient for the first pulse for ND-CI animals was only about one-fourth the size of that observed in the NH animals.

That the difference in these opposing trends is statistically robust can be seen in Figure 3. It plots the onset indices (differences between the first weight and the mean of weights 3 to 6, see Methods section) for each of the TWFs shown in Figure 2 as a function of log(pulse rate). Data are plotted in blue for the NH-acoustic animals, and in red for the ND-CI animals.

Figure 3.

Onset indices computed for each animal and condition (ND-CI and NH-acoustic) from the TWFs shown in Figure 2, as a function of pulse rate (50, 300, and 900 pps). Solid lines indicate population average fits derived from the fixed effects of a linear mixed-effects model with random intercepts and random slopes for pulse rate at the level of individual animals. The slopes of these fixed effects are shown alongside their standard error estimates.

Regression fits derived from a linear mixed-effects model are also shown. The slope relating onset index to pulse rate was negative for the ND-CI cohort, indicating a decrease in onset index with increasing pulse rate, whereas the NH-acoustic cohort exhibited a positive slope (Figure 3). The difference in slopes between cohorts was statistically significant (t = 4.90, p < .001).

In addition, we computed a two-way ANOVA with pulse rate and cohort as the independent factors and onset index as the dependent variable. This revealed a significant main effect of cohort (F = 102.5, p = 4.9⁻¹¹, partial η² = 0.78), as well as a significant interaction between pulse rate and cohort (F = 19.2, p = 1.4⁻⁴, partial η² = 0.40). But the main effect of pulse rate on its own failed to reach significance (F = 1.05, p = .31), presumably because the increasing trend for the NH-acoustic cohort, previously reported by Li et al. (2022), was cancelled out by the negative trend for the ND-CI cohort to a sufficient extent, so that the mean onset index, when averaged across both cohorts, did not change significantly as a function of pulse rate. Overall, the two-way ANOVA thus provided further confirmation that the dependence of PE on pulse rate and on cohort, which we observed in Figures 2 and 3, are statistically robust.

Our data thus clearly indicate that, while the NH acoustically stimulated animals exhibit a strong PE, which increases in strength with pulse rate, the ND-CI animals do not. In order to shed further light on whether these differences are due to a lack of acoustic hearing experience in the ND-CI animals during their infancy, we set up an additional cohort of four AD-CI implanted animals and measured their TWFs at 300 pps, the pulse rate that resulted in the highest ITD sensitivity for rectangular pulses (Buck et al., 2023). Figure 4 compares their TWFs against those seen in the NH-acoustic and the ND-CI cohorts. The figure suggests that the TWFs observed in the AD-CI animals are more similar to those seen in ND-CI animals than those in the NH-acoustic cohort, indicating that growing up with normal hearing experience is not the key driver of a strong PE when tested with prosthetic binaural stimulation later in life. Again, we computed and compared onset indices for each animal in each condition. A one-way ANOVA showed a highly significant cohort effect (F = 14.20, p = .0005, η² = 0.69), and a pairwise post-hoc Tukey-HSD test computed with the onset indices for the ND-CI and AD-CI cohorts were not significantly different (p = .98), but did reveal a significant difference between ND-CI and NH-acoustic (p = .0007) and AD-CI and NH-acoustic (p = .0018). Figure 5 summarizes the onset indices for all animals by cohort at 300 pps. It can be seen that the large majority (10 out of 12) of CI animals had onset indices above zero. If there was no PE at all in the CI animals, then their onset indices should be scattered randomly around zero. Under this null hypothesis, the probability of observing 10 or more positive onset indices (the cumulative binomial probability of 10 or more “hits” in 12 trials with 50/50 odds) is only .0193. Our data thus show that, on average, the PE was not completely absent in the CI-stimulated animals, but it was substantially weaker than in the NH cohort, and in two out of the 12 deafened CI animals, no PE could be observed. Specifically, the mean onset index in the NH cohort was 8.2 ms⁻¹, and that in the ND-CI and AD-CI cohorts were 2.2 and 2.0 ms⁻¹, respectively, which implies a reduction in the size of the PE by ∼75% in the CI conditions.

Figure 4.

TWFs for the NH-acoustic, ND-CI, and AD-CI cohorts at 300 pps, plotted as shown in Figure 2. Left (NH-acoustic; blue), middle (ND-CI; red), and right (AD-CI; yellow). Data from each animal is plotted in a different hue. NH-acoustic data originally from Li et al. (2022).

Figure 5.

Onset indices for all animals, by cohort (AD-CI, ND-CI, and NH-acoustic) at 300 pps pulse rate. Significant differences between the cohorts are shown in black. *** p < .001, ** .001 ≤ p < .01.

Discussion

To date, only a limited number of studies have investigated the PE in CI users. Notably, van Hoesel (2008) and Brown et al. (2015) both found that human CI users do exhibit a PE, although potentially with altered temporal weighting characteristics. For example, van Hoesel (2008) reported that TWFs for interaural level differences were considerably flatter than those for ITDs, a pattern that contrasts with the steep early onset weighting for binaural cues typically observed in NH listeners (Litovsky et al., 1999). None of the studies conducted so far included prelingually deaf patients. Therefore, a great deal of uncertainty remains regarding how different the PE is in patients relying on prosthetic hearing from that observed in NH listeners. In particular, it is not yet clear to what extent such deviations reflect fundamental differences in how binaural cues are encoded and integrated through electric rather than acoustic stimulation, or to what extent they are driven by experience. Given that reverberation can substantially affect sound localization performance in patients with bilateral CIs (Kerber & Seeber, 2013), answering these questions is not of purely academic interest, but could open avenues toward CI strategies that would enable CI patients to operate more efficiently in challenging acoustic environments.

This knowledge gap provides a key motivation for this study, which aims to elucidate the mechanisms and the effect of hearing experience on the perception of PE under bilateral CI stimulation. We observed that, while a PE appears to be present in our CI rats, it is clearly substantially and significantly weaker under CI stimulation than in NH animals tested under nearly identical conditions, and this is the case regardless of whether the rats were neonatally deafened or were raised with normal acoustic hearing experience (Figure 5). Also striking is the fact that, in our CI animals, the strength of the onset weighting did not increase with pulse rate (Figure 3), as is commonly observed in NH humans (Brown & Stecker, 2010) and in our previously published NH animals (Figure 2 left, see Li et al., 2022).

Our data thus seem to contrast with the small number of previous studies of the PE in human CI users. Both van Hoesel (2008) and Brown et al. (2015) report evidence of a PE in CI users that appears to be rather similar to that seen in NH populations. However, the study by van Hoesel (2008) lacks a direct NH comparison group and collected and analyzed the data in a manner that makes it easier to reveal whether a PE was present in each individual participant, but harder to compare data across participants and with results previously published for NH human listeners. Nevertheless, there appears to be clear qualitative similarity between TWFs of CI data reported by van Hoesel (2008) and those published for NH participants in studies such as Brown and Stecker (2010). Brown et al. (2015) did include an NH comparison group in their study investigating the PE in human CI users, and while they observed on average slightly shorter echo thresholds and slightly weaker PEs in the CI group, these differences were too small relative to the large within-cohort variability to reach statistical significance. Neither the van Hoesel (2008) nor the Brown et al. (2015) study would therefore have led us to predict the substantially diminished PE that we observed in both our ND-CI and AD-CI cohorts.

Even though our results are surprising, we can nevertheless have a high level of confidence that they are robust. The random interleaving of a high proportion of “honesty trials” into the experimental design, coupled with a stringent 75% honesty trial performance criterion, ensured that the small weights for the first pulse seen in the TWFs of our CI animals cannot be attributed to signal-to-noise problems that would arise if the animals were just randomly guessing, rather than diligently listening. The CI and the NH rats all performed the lateralization task competently and with good accuracy, but while the NH rats based their lateralization judgments very heavily on the first pulse, our CI animals based theirs on a more evenly weighted average of the binaural cue values of all the pulses in the burst, in apparent contrast to the human CI participants by van Hoesel (2008) and Brown et al. (2015).

It seems unlikely to us that fundamental species differences in the auditory pathways might provide a good explanation for the flatter TWFs that we observed in our CI rats compared to those reported for human CI users. For example, the 300 and 900 pps TWFs of our NH rats conform very closely to what one would expect to see in NH humans based on previously published studies, and it would be very surprising if species differences in the function of the auditory pathway only manifested under electric but not under acoustic stimulation. Also, a number of other studies indicate that binaural hearing in rats is generally quite similar to that in humans. Admittedly, rats are not as sensitive to low sound frequencies as humans, which may reduce their sensitivity to ongoing interaural phase differences at low frequencies (Wesolek et al., 2010), but their sensitivity to high-frequency “envelope ITDs” is excellent and similar to that seen in humans (Li et al., 2019). The sensitivity of CI rats to ILDs is similarly very good and comparable to that seen in highly performing CI patients (Buchholz et al., 2024). Furthermore, the ITD performance of our CI rats is extraordinarily better than that seen in human CI listeners (Rosskothen-Kuhl et al., 2021).

In our view, the most likely explanation for the apparently much weaker PE we observed in our CI rats compared to that seen in NH rats and humans, as well as in human CI users, is that our CI rats had no experience of trying to use their prosthetic hearing to explore a complex auditory environment involving echoes or competing sound sources. We find it particularly interesting that the very pronounced PE that is seen in NH, acoustically stimulated animals, did not carry over into prosthetic hearing in our AD-CI animals. To us, this suggests not only that the PE is probably highly plastic, but also that it is quite context or domain-specific. Although our AD-CI group had normal acoustic hearing experience prior to implantation, they never experienced reverberation or echoes under electric hearing conditions, and in this respect, their experience was essentially the same as that of the ND-CI group. It is therefore interesting that our CI animals nevertheless exhibited a small PE, even though their electric hearing experience was free of the “adaptive pressures“ that normally favor the development of a strong PE for echo suppression. Nevertheless, their TWFs were much “flatter” than those of the NH animals. Both our ND-CI and our AD-CI animals used their CIs exclusively for the purpose of lateralizing “anechoic” electrical pulse trains, and no part of their training with CI stimuli made the binaural cues in the later pulses of a burst less reliable than the initial pulse. Under these conditions, a flat TWF is advantageous. Physical and physiological noise limit the accuracy with which the binaural cues of any one pulse in a pulse train can be determined by the auditory pathway, but as long as later parts of the signal are not contaminated by the effects of reverberation, averaging cue values over subsequent parts of the signal should improve the signal-to-noise ratio. Developing a temporal weighting strategy that does not focus almost exclusively on onset cues was therefore adaptive for our ND- and AD-CI animals alike.

In any event, hearing with CIs is clearly a very different experience from acoustic hearing, and this may also entail that late deaf CI patients may have to relearn echo suppression as part of their rehabilitation. If this is indeed the case, then that raises the question whether contemporary clinical CI processor strategies provide CI patients with the rich and precise temporal information that would facilitate the learning of optimal echo suppression strategies. The documented difficulties that CI patients experience when localizing sounds in reverberant environments (Kerber & Seeber, 2013) suggest that this is an area where further research could point to paths towards significant improvements.

Conclusion

Our data strongly point to the fact that the emergence of a strong PE in CI hearing appears to depend on highly specific experience in the appropriate auditory modality. Prior experience of echoic listening conditions through acoustic hearing in our AD-CI cohort did not transfer to a strong onset upweighting in their TWFs in comparison to the ND-CI cohort. We, therefore, predict that listening experience with CIs in echoic conditions is likely necessary for a strong PE to emerge in bilaterally implanted CI patients, even if these patients had extensive normal binaural hearing experience early in life. Reverberation is known to pose significant challenges to CI patients. Our data suggest that the auditory system adapts in highly specific ways to the reliability of binaural cue information present at different time points of incoming signals, which raises the question of whether CI technology could be optimized to provide patients with sensory input that makes them better able to adapt to the specific challenges posed by hearing in echoic environments with their prosthetic devices.