Cochlear implants are major biomedical devices, and an exemplary success story of the application of foundational neuroscience research and use of brain-machine interface neuroprosthetics to treat a widespread neurological condition: hearing loss1-5. However, the auditory benefits provided by a cochlear implant are not instantaneous, in contrast to the amplification of acoustic input provided by commercial hearing aids. Some patients acquire a degree of speech comprehension with the cochlear implant a few hours after activation, but many patients unfortunately require months or even years post-implantation to achieve optimum levels of speech perception2,3. There are many open questions about the behavioral characteristics of this adaptation process in human listeners and the underlying neurophysiological changes13,14. Measuring how cochlear implants activate the central auditory system or other brain areas is technically complicated due to significant limitations with imaging in patients with implanted metallic medical devices15. Historically there have also been considerable challenges with experimental animal models of cochlear implant use, especially with the aims of monitoring and manipulating neural activity in implanted freely-behaving subjects. Here we addressed these issues by utilizing our recently-developed system for studying behaviorally- and physiologically-validated cochlear implant use in rats16, and examined neuromodulation and plasticity for cochlear implant learning and performance.
Cochlear implant outcome variability
We initially aimed to determine how rapidly hearing could be restored to profoundly deafened rats with cochlear implants. We trained normal-hearing rats on an auditory self-initiated go/no-go task16-18. Animals were acoustically trained prior to deafening and cochlear implantation, in order to separate aspects of procedural task structure learning from stages of perceptual learning with the cochlear implant. In this task, rats were trained to self-initiate trials via nosepoke (Fig. 1a, Extended Data Fig. 1), and a tone of a given frequency was presented (0.5-32 kHz at one octave spacing, 100 msec duration). Rats then had opportunity to respond to a specific target frequency (4 kHz) for food reward, but were trained to withhold responses to non-target foil tones (0.5, 1, 2, 8, 16, and 32 kHz). Performance on this task was assessed by the discriminability index d’, which is the difference in z-scores for responses to targets (‘hits’) versus responses to foils (‘false positives’). Here, d’ = 0.0 indicates chance performance, while d’ ≥ 1.0 indicates good discriminability.
After animals reached criteria (d’ ≥ 1.0 for 5+ days), they were bilaterally deafened (via cochlear lesion) and fitted with unilateral 8-channel cochlear implants (3-8 active channels per animal)16. These cochlear implanted rats were then re-trained to perform this task, with acoustic stimuli activating different implant channels (Fig. 1a-b). In order to normalize audibility across animals, cochlear implant stimulation levels were based on evoked compound action potential (ECAP) thresholds from each active cochlear implant channel (see Methods). The target and foil tones used for the cochlear implant version of the task were adjusted in order to correspond to the number of active channels (as shown in the frequency allocation tables, Extended Data Fig. 2a-b). Each tone predominantly stimulated one active channel as confirmed by electrodograms (Extended Data Fig. 2c).
Cochlear implant training occurred in two stages, paralleling the go/no-go structure of the task. In stage one, only tones activating the target channel were presented (and rewarded upon nosepoke). In stage two, other tones activating 2-7 foil channels were introduced. As with human subjects, initial performance of cochlear implanted rats attempting to respond to acoustic percepts could be variable and was often quite low2,3,13. Within 15 days of training, all implanted subjects reached a d’ of at least 1.0 (Fig. 1c-f). Implanted animals on average responded to the tone activating the target channel and withheld responses to tones activating foil channels (d’ = 1.7 ± 0.1, mean ± s.e.m., N = 16 rats, d’ quantified between days 4-14; Fig. 1c,d, Supplementary Movie 1). To verify that animals were indeed profoundly deaf and performed the task via cochlear implant stimulation (i.e., there was no residual hearing), we turned off the cochlear implant for a subset of trials and observed that behavioral performance dropped to chance (d’ = -0.1 ± 0.1, N = 16, p < 0.0001; Fig. 1c-d, Extended Data Fig. 3,4).
Individual rats had different learning rates throughout stage two, as measured by time to d' ≥ 1.0 (time = 3.0 ± 6.3 days, median ± interquartile range; Fig. 1e). Whereas some animals seemed to quickly recognize the behavioral meaning of cochlear implant stimulation in ≤ 3 days (N = 9), other animals took 5-15 days (N = 7; Fig. 1e-f). This subject performance variability was not explained by differences in insertion depth, impedance, ECAP thresholds, or normal-hearing performance across animals (Extended Data Fig. 5a-f), similar to observations in human studies13. Hit rates were also comparable across animals regardless of learning rate (Fig. 1g, Extended Data Fig. 5g). Instead, we noticed that improvements in performance were mainly driven by a decrease in false positive rates over days (Fig. 1h), which was significantly slower in under-performing animals (p = 0.01, Extended Data Fig. 5h). The overall performance obtained by each cochlear implanted animal was inversely related to the days to reach d’ ≥ 1.0 (Pearson’s r = -0.61, p = 0.01; Fig. 1i), which was not the case when these same animals initially learned the task when they were normal-hearing (Extended Data Fig. 5i). These results indicate that after cochlear implantation, early performance predicts peak performance, similar to reports in human cochlear implant subjects19.
Locus coeruleus activity during cochlear implant learning
We aimed to understand the central mechanisms contributing to individual variation in learning rates. One region important for early perceptual learning is the brainstem locus coeruleus, which is thought to broadcast a noradrenergic arousal signal throughout the central auditory system including auditory cortex17,20-23. To ask how locus coeruleus noradrenergic neurons might be activated during cochlear implant learning, we performed fiber photometry in tyrosine hydroxylase (TH)-Cre rats that had viral expression of GCaMP6s with AAVDJ-ef1a-DIO-GCaMP6s targeted unilaterally to locus coeruleus (Fig. 2a-b). We initially confirmed correct fiber placement and quality of fiber photometry recordings using tail pinch to elicit locus coeruleus responses24 (Fig. 2c). As animals were first conditioned just with target tones (in stage one) and foil tones added later (in stage two), we studied the responses in these two stages separately.
We quantified the trial-averaged locus coeruleus responses to the stimulus (‘tone-aligned’), and responses aligned to the nosepokes (‘response-aligned’) which immediately precede reward dispensation on hit trials. We found that locus coeruleus activity was not fixed, but instead dramatically changed over the course of cochlear implant learning. Specifically, locus coeruleus activity shifted from being driven by unexpected reward (in stage one) to being evoked by target stimulus presentation (in stage two), predicting improvements across cochlear implant subjects for when each animal reduced miss rates in stage one and began to have lower false positives in stage two.
The first effect we observed is represented by the example responses shown in Figure 2d. Task-related locus coeruleus responses emerged just prior to when this animal first began using the cochlear implant to respond to the target tone during stage one. These initial locus coeruleus responses were linked to reward and behavioral response, without clear stimulus-evoked signals (Fig. 2d, left, bottom). During this first stage of training, once this rat began responding consistently to the target tone with lower miss rates, locus coeruleus activity suddenly decreased (Fig. 2d, right, bottom). Across animals and sessions, there was negligible target tone-aligned locus coeruleus activity during earlier high-miss sessions and later low-miss sessions, both for the hit trials (Fig. 2e, top; N = 4 rats, n = 21 behavioral sessions split into quartiles, p = 0.71 comparing top versus bottom quartiles, unpaired two-tailed Student’s t-test) and for miss trials (Extended Data Fig. 6a,b). In contrast, there was substantial response-aligned locus coeruleus activity during earlier high-miss sessions, which decreased as animal performance increased (Fig. 2e, bottom; p < 0.02).
Animals were then moved to stage two of training when foil tones were also presented. In the animal shown in Figure 2f (same rat as Fig. 2d) during stage two training, when false positive rates were high during earlier sessions, the tone-aligned and response-aligned locus coeruleus activity both remained low (Fig. 2f, left, bottom). As this animal began to learn to withhold responses to foil tones leading to lower false positive rates, we observed target tone-aligned locus coeruleus activity that preceded reward (Fig. 2f, right, bottom). Across animals, tone-aligned locus coeruleus activity was higher on hit trials throughout behavioral sessions with low false positive rates compared to behavioral sessions with high false positive rates (Fig. 2g left, top; N = 4 rats, n = 40 behavioral sessions divided into quartiles, p = 0.01 comparing top versus bottom quartiles). There was negligible response-aligned activity (Fig. 2g, left, bottom; p = 0.85), and during false positive, miss, and withhold trials, locus coeruleus activity was similarly low (Fig. 2g, right, Extended Data Fig. 6c-d).
Plasticity of locus coeruleus activity seemed to reflect changes to internal representations of task variables and was not modality specific just for cochlear implant use. We also performed fiber photometry from locus coeruleus in normal-hearing TH-Cre animals trained on the acoustic version of this go/no-go task (Extended Data Fig. 7). We observed similar patterns and dynamics of locus coeruleus activity when the target tone was changed from 4 kHz to a different frequency. These results indicate that the locus coeruleus neuromodulatory signal is initially driven by unexpected reward, but becomes linked to stimuli predicting this reward over the course of training.
Locus coeruleus stimulation accelerates cochlear implant learning
We next wondered if we could harness the activity of the locus coeruleus to enhance learning with the cochlear implant. The imaging studies of Figure 2 demonstrate that once animals learned the task, target tones selectively activated locus coeruleus on correct trials. We therefore hypothesized that pairing locus coeruleus activation with presentation of the target tone earlier in training could accelerate cochlear implant learning, analogous to the effects of locus coeruleus stimulation on enhanced auditory cortical representations and perceptual learning in normal-hearing rats17.
Locus coeruleus was stereotaxically targeted in each animal and identified via electrophysiological responses in vivo to toe pinch (Extended Data Fig. 8a,b). We then expressed the excitatory opsin ChETA in noradrenergic locus coeruleus neurons via injection of AAV5-ef1a-DIO-ChETA in acoustically trained TH-Cre rats. We confirmed ChETA expression in TH+ locus coeruleus cells using immunohistochemistry and examined placement of the optical fiber within locus coeruleus via MRI/CT co-registration performed by blinded observers (Fig. 3a, Extended Data Fig. 8c). Viable animals with mistargeted optical fibers were used as sham controls. Animals were deafened and fitted with a cochlear implant two weeks after injection, and behavioral training began several days later. Starting on the first day of cochlear implant training, optogenetic stimulation of locus coeruleus was paired with the target tone for 5-10 minutes prior to each session (‘offline LC pairing’, Fig. 3b). We conducted pairing outside of behavioral context to focus on the impact of potential longer-term central modifications to neural circuits induced by locus coeruleus pairing, rather than more immediate changes to arousal level or brain state triggered by noradrenergic modulation.
Remarkably, all animals receiving locus coeruleus pairing reached d’ ≥ 1.0 within three days, i.e., they learned to use the cochlear implant more quickly compared to ‘sham paired’ animals injected with control YFP virus (AAV5-ef1a-DIO-eYFP) or off-target fiber implantation (Fig. 3c,d, Extended Data Fig. 9a; locus coeruleus paired animals: 2.0 ± 1.0 days to d’ ≥ 1.0, median ± interquartile range; sham paired animals: 3.0 ± 6.3 days to d’ ≥ 1.0; Mann-Whitney test, p = 0.04). Locus coeruleus paired animals also had higher levels of maximum performance (Fig. 3e,f, Extended Data Fig. 9b-e; locus coeruleus paired animals: d' = 2.9 ± 0.3, mean ± s.e.m; sham paired animals: d’ = 2.0 ± 0.2; unpaired two-tailed Student’s t-test, p = 0.01). Enhanced learning in locus coeruleus paired animals was unrelated to differences in cochlear implant functionality assessed by insertion depth, impedance, ECAP levels, or behavioral engagement in the task (Extended Data Fig. 10a-e). Performance in locus coeruleus paired animals improved over time (Fig. 3f), with behavioral gains driven by a sustained high hit rate and progressive reduction in false positive rates similar to the changes in behavior in sham paired animals (Fig. 3g,h).
We conclude that outcome variability across individual cochlear implant users was not entirely due to differences in insertion quality or level of peripheral auditory system activation. Specifically, none of these factors were different across the groups of cochlear implant rats examined here (Extended Data Figs. 5,10). Instead, variable learning rates across individuals seem to be due to differential engagement of modulatory systems (e.g., locus coeruleus) known to promote long-lasting plasticity and enhance perceptual learning. Although we artificially augmented neuromodulatory activity with optogenetic stimulation in locus coeruleus paired animals, our results from Figure 2 and Extended Data Figure 7 show that this system is endogenously active during acoustic or cochlear implant training, perhaps at different levels in various individuals. Such a continuum of neuromodulatory engagement might then account for behavioral variability across the population of cochlear implant users.
Cortical representations of cochlear implant stimuli
The locus coeruleus projects throughout the central auditory system including the auditory cortex, suggesting that differing degrees of noradrenergic modulation might lead to variable neural representations of cochlear implant channels. We focused on neural responses to implant stimulation in the auditory cortex, based on evidence suggesting that behavioral improvements with cochlear implants are paralleled by changes in auditory cortical responsiveness10,25-27. We specifically asked how cortical responses might relate to the behavior of rats trained with cochlear implants either with or without locus coeruleus stimulation. Locus coeruleus pairing with auditory stimuli in normal-hearing rats leads to lasting changes in auditory cortex but not the auditory thalamus17,20, and auditory cortical activity is required to perform this acoustic go/no-go task28.
We performed electrophysiological recordings in anesthetized rats after these animals had been deafened and trained to use cochlear implants; some animals were locus coeruleus paired while other animals were sham paired. Auditory cranial nerve (‘CN VIII’) ECAPs were measured and multi-unit recordings from auditory cortical responses to each individual cochlear implant channel were recorded (Fig. 4a). These methods are comparable to the types of electrophysiological recordings that can be conducted in human subjects; ECAPs are routine in human users, and multi-unit activity reflects a population-level responsiveness similar to electrocorticography and electroencephalography but with higher spatial resolution and signals closer to underlying single-unit representations.
Auditory nerve ECAPs were comparable across target and foil channels as well as across groups (Fig. 4b-c). This indicates that overall auditory nerve activity was broadly similar for both groups and stimulus types. In contrast, we noticed that the cortical multi-unit responses to target channel stimulation seemed higher on average in locus coeruleus paired animals versus sham paired animals (Fig. 4d-f). Based on this observation, we calculated neural d’ values (mean target z-score – mean foil z-score across all multi-unit recording sites) for each animal and compared this to their behavioral performance. We found that neural and behavioral d’ values were highly correlated across both groups of animals, such that animals (either sham or locus coeruleus paired) with high behavioral performance also had higher neural d’ values (Pearson’s r = 0.77, p = 0.003; Fig. 4g).
This correlation was due to two main indicators of enduring cortical plasticity resulting from cochlear implant training (augmented in some animals by explicit locus coeruleus pairing). These effects can be observed in the example animal with locus coeruleus pairing shown in Figure 4f (left), which had good performance using the cochlear implant to perform the task (d’ = 2.2); recordings from this animal showed 1) strong cortical responses to the target channel, and 2) very similar responses across foil channels. In contrast, a sham paired animal with lower performance had more variable responses over all channels (Figure 4f, right). Across animals, behavioral performance was correlated with responses to the target channel but not foil channels (targets: Pearson’s r = 0.67, p = 0.02; Fig. 4h; foils: Pearson’s r = 0.19, p = 0.55; Fig. 4i). Instead of the mean response to foils, behavioral outcomes were inversely correlated with the variability of foil channel responses among the different active channels (as quantified by the coefficient of variation, the average standard deviation across foils normalized by the average mean response; Pearson’s r = 0.65, p = 0.02; Fig. 4j). The reduction in variation means that the different foil channels have similar evoked responses to each other, over the active channels and throughout the multi-unit recording sites from each animal. Therefore, cochlear implant stimulation can effectively engage the auditory cortex in trained rats, with cortical responses shaped via mechanisms of neuromodulator-enabled plasticity to represent behavioral categories of different input. Specifically, reward-predictive stimuli evoked stronger neural responses, whereas responses to unrewarded stimuli seemed to be grouped together and became less distinct from each other.
For enhancement of perceptual and cognitive abilities, adequate interface between neuroprosthetics and neural tissue requires adaptation by the host biological circuitry to the signals provided by the neural implant. Here we discovered that locus coeruleus activity is a key indicator of hearing restoration with clinical-grade multi-channel cochlear implants in profoundly deaf rats. Responses in the locus coeruleus could predict when each animal first began hearing with the implant as well as overall performance levels. Optogenetic stimulation augmented the natural engagement of locus coeruleus to normalize performance across all animals. Failure to optimally use the cochlear implant in some subjects was therefore not due to poor activation of the auditory system, and indeed we observed robust cortical responses to implant stimulation in every animal. Training and neuromodulation seemed to refine cortical activity in a manner related to the behavioral meaning of the sounds, leading to an effective perceptual categorization in auditory cortex of the best performing animals. Our results provide a potential path for accelerating and improving outcomes with cochlear implants as well as other neuroprosthetic devices. Targeted neuroplasticity combined with neuromodulatory recruitment could be achieved with devices such as vagus nerve stimulators (believed to activate locus coeruleus)29,30, combined with real-time behavioral monitoring to enhance prosthetic device use during optimal periods of attention and arousal.