Locus coeruleus activity improves cochlear implant performance

Cochlear implants (CIs) are neuroprosthetic devices that can provide hearing to deaf people¹. Despite the benefits offered by CIs, the time taken for hearing to be restored and perceptual accuracy after long-term CI use remain highly variable^2,3. CI use is believed to require neuroplasticity in the central auditory system, and differential engagement of neuroplastic mechanisms might contribute to the variability in outcomes^4–7. Despite extensive studies on how CIs activate the auditory system^4,8–12, the understanding of CI-related neuroplasticity remains limited. One potent factor enabling plasticity is the neuromodulator noradrenaline from the brainstem locus coeruleus (LC). Here we examine behavioural responses and neural activity in LC and auditory cortex of deafened rats fitted with multi-channel CIs. The rats were trained on a reward-based auditory task, and showed considerable individual differences of learning rates and maximum performance. LC photometry predicted when CI subjects began responding to sounds and longer-term perceptual accuracy. Optogenetic LC stimulation produced faster learning and higher long-term accuracy. Auditory cortical responses to CI stimulation reflected behavioural performance, with enhanced responses to rewarded stimuli and decreased distinction between unrewarded stimuli. Adequate engagement of central neuromodulatory systems is thus a potential clinically relevant target for optimizing neuroprosthetic device use.

CIs represent the first successful example of using an electronic device to replace a human sense^1–5. However, the auditory benefits provided by CIs often take time to emerge. Some patients acquire a degree of speech comprehension a few hours after CI activation, but many patients require months or even years after implantation to achieve optimum levels of speech perception^2,3. There are many unresolved questions about the behavioural characteristics of this adaptation process and the underlying neurophysiological changes^13–15, owing to the challenges of monitoring and manipulating neural activity in humans and animal models of CI. However, studies in non-human primates and rodents have made important advances in surgical approaches and understanding how CIs stimulate the central auditory system^16–25. Here we address these issues by using our recently developed system for studying behaviourally and physiologically validated CI use in rats²², and examine neuromodulation and plasticity in learning and performance following cochlear implantation.

CI outcome variability

We initially tested how quickly hearing could be restored to deafened rats with CIs. We trained 16 normal-hearing Long-Evans tyrosine hydroxylase (TH)-cre rats on an auditory self-initiated go/no-go task^22,26,27. The rats were acoustically trained before deafening and cochlear implantation to separate procedural task structure learning from stages of CI perceptual learning. Rats were trained to self-initiate trials via nosepoke (Fig. 1a and Extended Data Fig. 1), and a tone of a given frequency was presented (0.5–32 kHz, 100 ms). In stage 1 of go/no-go training, rats could respond to a specific target frequency (4 kHz) for food reward; no non-target tones (foils) were played. They were then moved to stage two for training to withhold responses to non-target foils.

Fig. 1 |. Behavioural assessment of CI learning.

a, Behavioural training. Left, schematic of training for normal-hearing and CI rats. Right, trial structure: (1) rats self-initiate trials for (2) presentation of either target or foil stimulus; (3) behavioural responses lead to (4) food rewards on target hit trials, no outcome on correct withholds, or 7 s timeouts on miss trials or false positives. b, Left, schematic of bilateral deafening and unilateral placement of CI. Right, X-ray of CI electrode within the cochlea; note the full >360° turn of the CI. Scale bar, 1 mm. c, Responses in an example rat across CI channels on day 6 of behavioural training with CI, with the implant on (filled) or off (open). Arrowhead, target stimulus programmed to activate CI channel 4. d, Behavioural performance in rats trained with the CI, quantified between days 4 and 14. Performance decreased to chance when CI was turned off (on versus off, n = 16 rats, P < 0.0001; paired Wilcoxon signed-rank test). Data are mean ± s.e.m. e, Number of days to d′ ≥ 1.0 across rats (n = 16 rats). Data are median ± interquartile range. f, Performance (d′) with the CI over time. g, Hit rates. h, False positives. f–h, Grey dashed lines represent individual rats (n = 16 rats). Black dashed lines represent mean and error bars show s.e.m. i, Number of days to d′ ≥ 1.0 correlated with maximum d′ for each rat (n = 16 rats). **P < 0.01.

After the rats reached the training criterion (d′ ≥ 1.0, after 5 or more days; where d′ is the difference between z-scores for the distribution of responses to targets and for the distribution of responses to foils), they were bilaterally deafened and programmed with unilateral 8-channel CIs²² (3–8 active channels). These CI rats were then re-trained on the task, with stimulation delivered through a clinical speech processor and CI. New stimulus tones were chosen, and the speech processor was programmed so that each tone would be associated with the stimulation of a single electrode. One electrode was chosen as the target, with other electrodes being foils (Fig. 1a,b and Extended Data Fig. 2).

CI training occurred in two stages, paralleling the go/no-go task structure. In stage 1, only tones activating the target channel were presented. In stage 2, other tones activating 2–7 foil channels were introduced—that is, 3 or more active channels (one target and two or more foils) were required for stage two. As with human subjects, the initial performance of CI rats was variable and often low^2,3,13. Within 15 days of training, all implanted subjects reached d′ ≥ 1.0 (d′ = 1.7 ± 0.1 (mean ± s.e.m.), n = 16 rats; Fig. 1c,d and Supplementary Video 1).

To verify that the rats were deaf and performed the task via CI stimulation without residual hearing, we inactivated the CI on some trials; across all rats, behavioural performance dropped to chance (d′ = −0.05 ± 0.1, n = 16, P < 0.0001; Fig. 1c,d and Extended Data Figs. 3 and 4a,b). Examination of cochlear cytology after deafening showed significant reductions in the numbers of outer hair cells (OHCs) and inner hair cells (IHCs), including almost complete OHC loss across the cochlea (Extended Data Fig. 4c). Deafened rats had no detectable auditory brainstem responses (ABRs) across 0.5–16 kHz up to 90 dB sound pressure level (SPL) acutely post-deafening and weeks thereafter. Implant-evoked electrical ABRs (EABRs) and CI behavioural performance were robust in deafened rats lacking acoustic ABRs (Extended Data Fig. 4d–k).

Individual rats had different learning rates throughout stage two, measured by time to d′ ≥ 1.0 (3.0 ± 6.3 days (median ± interquartile); Fig. 1e). Some rats quickly recognized the behavioural meaning of CI stimulation within days (n = 9), but others took 5–15 days (n = 7; Fig. 1e,f). Similar to humans¹³, subject performance variability was not explained by differences in insertion depth, impedance, evoked compound action potential (ECAP) thresholds or normal hearing performance across rats (Extended Data Fig. 5a–f). Hit rates were similar across rats regardless of learning rate (Fig. 1g and Extended Data Fig. 5g). Instead, performance improvements were driven mainly by decreased false-positive rates (Fig. 1h), which was significantly lower in under-performing rats (Extended Data Fig. 5h). The overall performance obtained by each CI rat was inversely related to the number of days taken to reach d′ ≥ 1.0 (Pearson’s r = −0.61, P = 0.01; Fig. 1i), unlike performance on the acoustic task when these rats had normal hearing (Extended Data Fig. 5i). Performance on the first CI day was uncorrelated with performance on the last day of normal- hearing training (Extended Data Fig. 5j–l). Thus, similar to human subjects²⁸, early CI performance predicts eventual peak performance, but is largely independent of previous acoustic performance and experience.

LC activity during CI learning

We aimed to understand the central mechanisms contributing to individual variation in learning rates. One region important for early perceptual learning is LC, which is thought to broadcast a noradrenergic arousal signal throughout the central auditory system, including auditory cortex^26,29–32. We performed fibre photometry in four TH-cre rats virally expressing GCaMP6s to test how LC noradrenergic neurons were activated during CI learning, examining LC responses to toe pinch to confirm fibre placement and quality of photometry³³ (Fig. 2a–c). As rats were first conditioned only with targets (in stage 1) before adding foils (in stage 2), we studied the responses in these two stages separately.

Fig. 2 |. Dynamic LC activity during CI learning.

a, Schematic of LC fibre photometry in CI rats. DAQ, data acquisition system. b, GCaMP6s (green) in LC TH+ (red) cells, with an optical fibre above LC. 4V, fourth ventricle. Scale bar, 200 μm. c, LC activity evoked by toe pinch. d, Example LC activity from rat 1 during stage 1 CI training, aligned to tone onset or behavioural response, for high-miss rate session 1 (top) and low-miss rate session 5 (bottom). e, Stage 1 behaviour and LC responses in each rat for all behavioural sessions with photometry. Black, session miss rate. Green, dF/F tone-aligned (filled symbols, solid lines) or behavioural response-aligned (open symbols, dashed lines) on hit trials. f, Miss rates across stage 1 sessions were not correlated with tone-aligned normalized LC dF/F (top; n = 4 rats, 21 sessions) but correlated with dF/F when aligned with behavioural response (bottom). g, Example LC activity from rat 1 during stage 2, aligned to tone or behavioural response, for target (left, green) or foil (right, red) trials, during high false-positive session 3 and lower false-positive session 12. h, Stage 2 behaviour and LC responses from each rat for all behavioural sessions with photometry. Black, session false-positive rate. Green, dF/F tone-aligned (filled symbols, solid lines) or behavioural response-aligned (open symbols, dashed lines) on hit trials. Tone-aligned dF/F signals in stage 2 were highest when withholds were highest. i, Top, false positives negatively correlated with tone-aligned normalized LC dF/F (n = 4 rats, 40 sessions); higher LC activity predicted lower errors. Bottom, false positives were uncorrelated with behavioural response-aligned dF/F, indicating that LC activity shifted from being reward-driven in stage 1 to driven by predictive stimulus in stage 2. Data are mean ± s.e.m.

We quantified trial-averaged LC responses to stimuli (‘tone-aligned’), and responses aligned to nosepokes immediately preceding rewards on hit trials (‘response-aligned’). LC activity was not fixed, but instead markedly changed over CI learning. LC activity shifted from being driven by unexpected reward from responses to newly presented targets (in stage 1) to being evoked by target stimulus presentation—that is, tone-aligned (in stage 2)—predicting improvements across CI subjects for when each rat reduced misses in stage 1 and began to reduce false positives in stage 2.

The first effect is represented in Fig. 2d. Task-related LC responses preceded reliable CI behavioural responses to targets in stage 1. Initial LC responses were linked to nosepoke and reward, without clear stimulus-evoked signals (Fig. 2d, session 1 response and Fig. 2e, rat 1). Once this rat began responding consistently to targets with lower miss rates in stage 1, LC activity decreased (Fig. 2d, session 5 response and Fig. 2e). Across rats and sessions, there was negligible stimulus-aligned LC activity for hits (Fig. 2e,f, top; n = 4 rats, 21 sessions; correlation r between misses and tone-evoked dF/F = 0.09, P = 0.7) and miss trials (Extended Data Fig. 6a,b). By contrast, there was substantial behavioural response-aligned LC activity during earlier high-miss sessions, which decreased as performance improved (Fig. 2f; r = 0.67, P = 0.0008).

Rats were then moved to stage 2 of training with foil presentation. During stage 2 training (Fig. 2g; same rat as in Fig. 2h, rat 1), when false- positive rates were high during earlier sessions, tone-aligned and response-aligned LC activity remained low (Fig. 2g, top, session 3). As this rat began to withhold responses to foils, leading to lower false-positive rates, target-aligned LC activity emerged preceding reward (Fig. 2g, bottom, session 12 target trials). Across rats, tone- aligned LC activity was higher on hit trials throughout sessions with low false positives versus sessions with high false positives (Fig. 2h,i, top; n = 4 rats, sessions; correlation r between false positives and tone-evoked dF/F = −0.32, P = 0.04). There was negligible response-aligned activity (Fig. 2i, bottom; r = 0.17, P = 0.3), and LC activity was also low during false-positive, miss and withhold trials (Extended Data Fig. 6c–f).

LC plasticity reflected changes to internal representations of task variables, which was also observed in normal-hearing TH-cre rats on the acoustic version of the task, when the target tone was changed from 4 kHz to a different frequency (Extended Data Fig. 7). These results indicate that the LC neuromodulatory signal is initially driven by unexpected reward, but becomes linked to stimuli predicting this reward over the course of training. This is reminiscent of classic findings of dopamine neuron firing related to reward prediction error after training^34,35.

LC stimulation accelerates CI learning

We next explored whether we could harness LC activity to enhance CI learning. Imaging studies (Fig. 2) demonstrate that target tones selectively activated LC on correct trials after learning. We therefore hypothesized that pairing LC activation with target tones earlier in training might accelerate CI learning, analogous to the effects of LC stimulation on enhanced auditory cortical representations and perceptual learning in normal-hearing rats²⁶.

We stereotactically targeted LC in each rat and identified electrophysiological responses to toe pinch (Extended Data Fig. 8a,b). We then expressed the excitatory opsin ChETA in LC noradrenergic neurons of acoustically trained TH-cre rats. We confirmed ChETA expression in LC TH+ cells using immunohistochemistry, and examined LC placement of the optical fibre via micro-magnetic resonance imaging (μ-MRI)–micro-computed tomography (μ-CT) co-registration (Fig. 3a and Extended Data Fig. 8c). Rats with mistargeted fibres or those injected with control YFP virus were used as sham controls (same 16 ‘sham-paired’ rats from Fig. 1). Rats were deafened and fitted with CIs two weeks after injection, and behavioural training began several days later. Starting on the first day of CI training, optogenetic LC stimulation was paired with the target tone for 5–10 min before each session (Fig. 3b, offline LC pairing). We conducted pairing outside of behavioural context (that is, while rats were not engaged in a task) to examine the effects of potential longer-term central modifications to neural circuits induced by LC pairing, rather than more immediate changes to arousal level or brain state triggered by noradrenergic modulation.

Fig. 3 |. LC pairing enhances CI learning.

a, Optogenetic stimulation of noradrenergic LC. Left, optical fibre placement in LC was confirmed by μ-CT–μ-MRI co-registration. Scale bar, 3 mm. Right, ChETA expression (green) in LC TH+ cells (red). Scale, 200 μm. b, Schematic of offline LC pairing and CI training. c, Example rats receiving either LC pairing (top) or sham pairing (bottom), showing behavioural responses and d′ on days 1, 5 and 9 of CI training. Arrowhead, target tone activated CI channel 4 in both rats. Error bars show 95% confidence interval. d, LC-paired rats reached criteria (d′ ≥ 1.0) more quickly than sham-paired rats (LC-paired, n = 10 rats versus sham-paired, n = 16 rats, P = 0.04, unpaired two-tailed Mann–Whitney). Data are median ± interquartile range. All paired rats reached criteria within 1–3 days. Sham rats are from Fig. 1. e, LC-paired rats with 6 or more days of CI training had higher maximum d′ than sham-paired rats (LC-paired, n = 6 rats versus sham-paired, n = 14 rats, P = 0.01, unpaired two-tailed t-test). Data are mean ± s.e.m. f, Implant performance (d′) over time in LC-paired rats (n = 10; dark blue line shows mean (±s.e.m); light blue lines show individual rats) versus sham-paired rats (n = 16; black line shows mean (from Fig. 1f)). g, Hit rates over time in LC-paired rats (n = 10; dark blue line shows mean (±s.e.m); light blue lines show individuals) versus sham-paired rats (n = 16; black line shows mean (from Fig. 1g)). h, False positives over time in LC-paired rats (n = 10; dark blue line shows mean (±s.e.m); light blue lines show individuals) versus sham-paired rats (n = 16; black line shows mean (±s.e.m) (from Fig. 1h)). Four LC-paired rats and two sham-paired rats did not reach the six-day performance requirement to calculate maximum d′; these rats are shown in d,f–h but are excluded from e. *P < 0.05, **P < 0.01.

All rats receiving LC pairing reached d′ ≥ 1.0 within three days—that is, they learned to use the CI more quickly compared with the 16 sham-paired rats (Fig. 3c,d and Extended Data Fig. 8d–g; LC-paired rats: 2.0 ± 1.0 days to d′ ≥ 1.0 (median ± interquartile), n = 10; sham-paired: 3.0 ± 6.3 days to d′ ≥ 1.0, n = 16; unpaired two-tailed Mann–Whitney test, P = 0.04). Performance within the off-target fibre implantation group was not correlated with distance between LC and the mistargeted fibre (Extended Data Fig. 8d,e). LC-paired rats had higher levels of maximum performance compared with shams (Fig. 3e,f and Extended Data Fig. 8f–j; LC-paired rats: d′ = 2.9 ± 0.3 (mean ± s.e.m.); sham-paired: d′ = 2.0 ± 0.2; unpaired two-tailed t-test, P = 0.01). Enhanced learning in LC-paired rats was unrelated to differences in CI functionality assessed by insertion depth, number of active channels, estimated cochleotopic implant alignment, impedance, ECAP levels or behavioural task engagement (Extended Data Fig. 9a–h). LC- and sham-paired rats had similar poor performance in behavioural sessions when the implant was turned off (Extended Data Fig. 9i–j). Behavioural performance in these rats on the acoustic task prior to deafening did not predict CI performance (Extended Data Fig. 9k–m), indicating that these rats did not happen to be especially fast learners or high performers in general, and that good performance on the acoustic task does not seem to provide much advantage in terms of performance after implantation. Performance in LC-paired rats improved over time (Fig. 3f), with behavioural gains driven by a sustained high hit rate and progressive reduction in false-positive rates similar to the changes in behaviour in sham-paired rats (Fig. 3g,h).

Cortical representations of CI stimuli

LC projects throughout the central auditory system including auditory cortex, suggesting that differing degrees of noradrenergic modulation might lead to variable neural representations of CI channels. We focused on auditory cortical responses to CI stimulation based on evidence suggesting that behavioural improvements with CIs are paralleled by changes in auditory cortical responsiveness^10,36–38, pairing with auditory stimuli in normal-hearing rats leads to lasting changes in auditory cortex but not thalamus^26,29, and auditory cortical activity is required to perform this acoustic go/no-go task³⁹.

We performed electrophysiological recordings in anaesthetized rats after they had been deafened and trained to use CIs, and we compared the neural responses to implant stimulation in trained rats to responses from other untrained rats that were similarly deafened and implanted. Auditory cranial nerve (CN VIII) ECAPs were measured and multi-unit or whole-cell recordings were made from auditory cortex neurons in response to stimulation of individual CI channels (Fig. 4a). Recordings were obtained from 16 rats; 4 rats were LC-paired (4 out of 8 rats from Fig. 3), 8 rats were sham-paired (8 out of 16 rats from Figs. 1 and 3), and 4 rats were untrained.

Fig. 4 |. Auditory cortical responses in CI rats.

a, Schematic of experimental recording. ACx, auditory cortex. b, CI-evoked auditory cortical multi-unit activity (MUA) in LC-paired (left), sham-paired (middle) and untrained (right) rats. Dashed line, behavioural CI target channel. Unresponsive sites are indicated with X. Sites are ordered by responsiveness (trained, to target; untrained, overall). c, CI-evoked cortical responses from sample rats. Arrowhead, CI stimulus artefact. Bottom right, trial-by-trial first-spike latency jitter (untrained (UT): 6.7 ± 1.4 ms, n = 4 rats; trained (T): 2.4 ± 0.6 ms, n = 12 rats; P = 0.009, unpaired two-tailed t-test; black, sham-paired; blue, LC-paired), coefficient of variation (untrained: 0.63 ± 0.10; trained: 0.34 ± 0.05; P = 0.003). The horizontal line indicates overall mean, dots show mean from individual rats. d, In vivo whole- cell voltage-clamp recording from auditory cortical neuron in an untrained rat. Left, EPSCs and IPSCs evoked by CI stimulation across channels. Right, synaptic excitatory–inhibitory tuning profile (rEI = 0.04). e, Currents in a neuron from trained rat. Grey line, behaviour (d′ = 1.4, target channel 4). Overall excitatory–inhibitory co-tuning was high (rEI = 0.85). f, Excitatory–inhibitory tuning correlations (rEI) for implanted (CI) rats (untrained rEI = −0.21 ± 0.11, n = 4 neurons from 2 rats; trained rEI = 0.72 ± 0.12, n = 4 neurons from 3 rats; P = 0.001, unpaired two-tailed t-test) versus normal-hearing (NH) young (postnatal day 12–16 rEI = 0.30 ± 0.06, n = 34 neurons) and adult rats (rEI = 0.71 ± 0.05, n = 25 neurons). g, Multi-unit responses (top) and behaviour (bottom) for sample LC-paired (left, n = 19 recording sites), sham-paired (right, n = 6 recording sites) rats. Data are mean ± s.e.m. (top) or response rate ± 95% confidence interval (bottom). h, Across LC-paired (n = 4) and sham-paired rats (n = 8), neural and behavioural d′ values were correlated. i, Multi-unit responses to target correlated with behavioural performance (n = 12 rats). j, Foil channel responses were uncorrelated with behaviour (n = 12). k, Coefficient of variation to foils was negatively correlated with behaviour (n = 12). **P < 0.01.

Peripheral responses were similar between trained and untrained rats, but cortical responses were substantially different. Auditory nerve ECAPs were similar across target and foil channels as well as across groups (Extended Data Fig. 10a,b), indicating that overall implant-evoked auditory peripheral nerve responses were similar regardless of training. This was different from cortical responses, which were substantially affected by experience. The number of sites with multi-unit responses to CI stimulation was higher in trained versus untrained rats (Extended Data Fig. 10c). Multi-unit activity magnitude at responsive sites was similar between untrained and trained rats (Fig. 4b; z-scored activity at best channel, trained rats: 2.4 ± 0.5, n = 12; untrained rats: 1.6 ± 0.5, n = 4; P = 0.35, unpaired two-tailed t-test), but trial-by-trial evoked responses were more erratic in untrained rats, with higher first-spike latency jitter and higher variability in evoked response size quantified by coefficient of variation—that is, the s.d. normalized by the mean response (Fig. 4c, bottom right).

Previous studies of deafening and CI stimulation suggest that inhibitory transmission might be specifically affected after hearing loss and CI stimulation^40–45. To determine whether the changes that we observed in spiking output resulted from adjustments of CI-evoked synaptic inputs, we obtained in vivo whole-cell recordings from auditory cortical neurons in trained or untrained anaesthetized adult rats^26,46.

We measured CI-evoked excitatory and inhibitory postsynaptic currents (EPSCs and IPSCs) in voltage clamp, and measured overall correlation between EPSC and IPSC amplitude across implant channels (rEI). Neurons from untrained rats had imbalanced excitatory and inhibitory implant tuning profiles, but neurons from trained rats showed high degrees of co-tuning (Fig. 4d–f). CI-evoked cortical EPSCs and IPSCs in trained rats were similarly co-tuned or ‘balanced’, as with tone-evoked excitatory–inhibitory tuning measured in normal-hearing adult rats. This co-tuning likely underlies the more reliable and precise spike generation in trained or experienced rat auditory cortex^26,46. By contrast, CI-evoked excitation and inhibition in untrained rats were even more disorganized than in young normal-hearing rats just after hearing onset⁴⁶ (Fig. 4f). As inhibitory responses emerge entirely within auditory cortex, these changes to CI-evoked inhibitory tuning imply that auditory cortex is a major site of neuroplasticity with CI training. Furthermore, learning rules of synaptic plasticity for CI inputs may be different from those engaged by acoustic experience during postnatal development, requiring enhancement of plasticity mechanisms via neuromodulation.

Finally, we tested how cortical multi-unit activity related to behavioural performance across trained rats. Responses to target channel stimulation seemed higher in LC-paired versus sham-paired rats (Fig. 4b,c,g). We calculated neural d′ values for each rat, and found that neural and behavioural d′ values were highly correlated across rats (Pearson’s r = 0.77, P = 0.003; Fig. 4h and Extended Data Fig. 10d,e). This correlation was due to two main indicators of enduring cortical plasticity after CI training and LC pairing, as in the example LC-paired rat in Fig. 4g, left, with good CI performance (d′ = 2.2). Recordings from this rat showed strong cortical responses to the target channel and responses across foil channels that were similar to each other. Conversely, a sham-paired rat with lower performance had more variable responses over all channels (Fig. 4g, right). Across rats, behavioural performance was correlated with target channel response magnitude, but was less clearly correlated with foil channels (targets: Pearson’s r = 0.67, P = 0.02; Fig. 4i; foils with outlier included: Pearson’s r = 0.19, P = 0.55; foils with outlier excluded via Grubbs’ test: r = 0.82, P = 0.002; Fig. 4j).

This correlation does not imply that target channel responses alone are responsible for go/no-go task performance. Behavioural outcomes were also inversely correlated with the variability of foil channel responses (Pearson’s r = 0.65, P = 0.02; Fig. 4k). The reduction in variation means that different foil channels have similar evoked responses to each other, in line with their similar behavioural meaning. Therefore, CI stimulation can effectively engage the auditory cortex in trained rats, with cortical responses shaped via mechanisms of neuromodulator-enabled plasticity to represent behavioural categories of different input. Specifically, reward-predictive stimuli evoked stronger neural responses, whereas responses to unrewarded stimuli were grouped together and became less distinct. Our results indicate that classification of foils and consequent behavioural regulation of ‘no-go’ withholding responses might be an especially important aspect of cortical function on this auditory task.

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 observed that LC activity is a key indicator of hearing restoration with clinical-grade multi-channel CIs in deaf rats. LC responses could predict when each rat first began performing reliably, as well as their overall performance with the CI. Furthermore, our results indicate that central neuromodulatory systems may not have been adequately engaged in those rats that learned more slowly (here, a fraction of rats in the sham-paired group, but not in the LC-paired group). Modulatory areas such as LC receive sensory input in order to appropriately regulate brain state and behavioural processes such as learning^26,27,31,32, and we showed that optogenetic LC stimulation led to rapid CI learning rates in all rats. Our results provide a potential path for accelerating and improving outcomes with CIs as well as other neuroprosthetic devices, combined with real-time behavioural monitoring to enhance prosthetic device use during optimal periods of attention and arousal.

Online content

Any methods, additional references, Nature Portfolio reporting summaries, source data, extended data, supplementary information, acknowledgements, peer review information; details of author contributions and competing interests; and statements of data and code availability are available at https://doi.org/10.1038/s41586–022-05554–8.

Methods

Animal use and ethics

All procedures were approved under NYU Langone Institutional Animal Care and Use committee protocols. A total of 59 adult female Long- Evans rats were used for these studies: 16 TH-cre rats for training in Fig. 1 (which are the 16 sham rats in Fig. 3, and 8 of these rats were used for recordings in Fig. 4), 4 other TH-cre rats for photometry with the CI in Fig. 2, 2 other TH-cre rats for photometry in normal-hearing rats, 10 other TH-cre rats for optogenetics in Fig. 3 (4 of these rats were used for recordings in Fig. 4), 14 rats used for measuring ABRs or EABRs pre- and post-deafening (for Extended Data Fig. 4; 11 wild-type and 3 TH-cre rats), 4 untrained wild-type rats for multi-unit recordings (in Fig. 4), 5 rats for in vivo whole-cell recording (2 untrained and 3 trained in Fig. 4; 2 of the trained rats were TH-cre), and 4 untrained wild-type rats for cochleogram analysis. The number of rats required in the LC pairing experiments was determined by performing a power analysis with power of 0.8 based on control behavioural data in King et al.²². Based on the positive control fibre photometry data, a sample size of at least three was estimated be required to detect changes in LC activity at half the level activity during strongly aversive stimuli. Rats were randomly assigned to groups after reaching criteria for baseline behavioural training. Behavioural blinding was not performed in these experiments owing to the requirement of the same rat to be followed for several independent surgeries and daily behavioural testing.

Surgical protocols

All rats were kept in a vivarium on a 12/12-h light/dark cycle and housed individually or in pairs. Female Long-Evans wild-type or TH-cre rats 3–5 months old were anaesthetized with intramuscular ketamine (40 mg kg⁻¹) and dexmedetomidine (0.125 mg kg⁻¹). Atropine (0.02 mg kg⁻¹) and dexamethasone (0.2 mg kg⁻¹) were subcutaneously applied immediately after anaesthesia induction to minimize bronchial secretions, inflammation, and intracranial pressure. Body temperature was maintained slightly hypothermic at 34–35 °C with a DC temperature controller heating pad throughout the procedure and eye ointment (Puralube Vet Ointment, Dechra) used to prevent corneal drying. The rat was positioned prone to optimize respiratory function.

Acoustic training

Rats were trained on a go/no-go task^{22,26,27,39,47} to nosepoke in response to a target tone frequency for a sugar pellet reward (Bio-Serv) in operant conditioning chambers (Med Associates). Each chamber contained a speaker calibrated across frequencies at 70 dB, a food dispenser on the left wall, and three nosepoke ports (two on either side of the food dispenser and one on the wall opposite). Each chamber was placed in a larger wood enclosure and insulated with foam. The measured background noise in each chamber was <30–40 dB.

Rats were food restricted to maintain the weights at 80–85% of their initial pre-training weights. First, rats were shaped with two to three days of training to nosepoke for one food pellet. Next, rats were trained to nosepoke within 2.5 s after a target tone was played (stage 1: go training). When the rats had hit rates of >80%, stage 2 (go/no-go) training began, and three foil tones were introduced (2–16 kHz at one octave intervals excepting the target frequency). Rats were trained to hit rates >90%, along with false-positive rates <40%. Finally, the foil tones were expanded to six total (0.5–32 kHz at one octave intervals excepting the target frequency), and rats were trained to the same criteria. Target and foil pure tones were 100 ms in duration presented in a pseudo-random order at 70 dB. For correct trials, each trial ended at either the time of food pellet delivery (hit trials for targets) or 2.5 s after the tone (correct withhold trials for foils). On error trials, failure to respond (miss trials for targets) as well as incorrect responses (false-positive trials for foils) were punished with a time-out of 7 s before the next trial could be initiated. Random nosepokes were punished with time-out as well. Rats self-initiated the trials by nosepoking in a different port than the ‘response’ port. After 0.5–1.5 s, either a target or foil tone was played. Behavioural performance was measured using the discriminability index d′ (the difference in the z-scores for the distribution of responses to targets and for the distribution of responses to foils). d′ values were computed as the difference in z-scores between hits and false positives: d′ = z(hit rate) – z(false-positive rate).

Rats that achieved criterion behavioural performance on the baseline task with the target tone of 4 kHz underwent surgery as described above, had optical fibres chronically implanted in left LC, and were allowed to recover for two weeks during viral expression. Rats then underwent bilateral deafening and unilateral cochlear implantation²². After three days of recovery, rats began behavioural training with the CI.

Deafening and cochlear implantation

Cochlear implantation was conducted 2–4 weeks after viral injection as previously described²². For both deafening and cochlear implantation, a postauricular incision was made and the superficial fascia of the neck was dissected. The sternocleidomastoid muscle and posterior belly of the digastric muscle were retracted from the tympanic bulla. The tympanotomy was begun ventrocaudally to the trunk of the facial nerve with a 0.5-mm diamond burr and continued dorsally until the stapedial artery overlying the round window was fully visualized.

The cochleostomy site was identified ~0.5 mm directly below the lip of the round window in the basal turn of the cochlea. The site was drilled with a 0.1-mm diamond burr. For the deafened only side, a sham CI array (4-channel, Cochlear) was inserted and left in the cochlea for 10 min to induce full field deafness. A 4-channel array was used for deafening as these devices are more durable and re-usable, and was inserted into the cochlea at ~180 degrees. The array was removed and the cochleostomy site is closed with a tympanic bulla muscle graft followed by 2-octyl cyanoacrylate (Surgi-Lock 2oc, Meridian Animal Health). For the cochlear implanted side, prior to performing the cochleostomy and inserting the array (8-channel, Cochlear), the array lead and connector were cemented to the head cap securing the previously implanted optic fibre. Blunt dissection was used to expand the postauricular incision towards the head cap. The ground was sutured into a small muscle pocket in the trapezius. The array was inserted without resistance using AOS forceps (Cochlear) until all the platinum-iridium contacts were within the scala tympani. The array and cochleostomy were sealed using 2-octyl cyanoacrylate. Before closure, dexamethasone was applied to the root of the facial nerve to prevent inflammation. The portion of the lead wires exiting the skull connector as reinforced with additional silicone insulation, and further protected with silicone sealant (Kwik-Cast, WPI) and dental epoxy (Triad Gel, Dentsply).

We measured ABRs to clicks and tones (0.5, 1, 2, 4, 8 and 16 kHz) in a subset of rats before the deafening procedure and/or cochlear implantation, immediately post-deafening, 2–6 weeks postoperatively. Subdermal needle electrodes were placed at the cranial vertex (recording electrode), behind each pinna, and at the base of the spine above the tail (reference and grounds). ABRs were recorded with a preamplifier (DAM50; World Precision Instruments) connected to an amplifier (MultiClamp 700A; Molecular Devices) and digitizer (Digidata 1440A; Molecular Devices); acoustic stimuli were presented at 70, 80, and 90 dB with a digital signal processor and speaker (Tucker-Davis Technologies) calibrated with an ACO 7017 microphone (ACO Pacific). ABR waveforms were recorded using Clampex 10.7 (Molecular Devices) and data analysed with Matlab R2019b (MathWorks). Tone duration was 3 ms (1 ms rise/fall). All stimuli were presented for 300 sweeps at 20 Hz to reduce adaptation.

A similar setup was used for EABRs in CI rats. A CI24RE CI was driven by a Freedom speech processor connected through a programming pod to a Windows personal computer running the Custom Sound Suite 4.0 programming software (Cochlear). Custom Sound Suite’s EABR function was used (5 charge-balanced biphasic pulses, 25 μs/phase, 900 Hz stimulation frequency, 300 sweeps at 20 Hz) to stimulate the CI. A modified cable with stereo jack and Bayonet Neill-Concelman connectors was used to connect the programming pod to the trigger input of the Digidata 1440A (Molecular Devices), facilitating coordination of each sweep of the electrical stimulus (stimulation intensity was controlled through Custom Sound) with the ABR recording setup in Clampex 10.7 (Molecular Devices).

CI programming

Impedance and ECAP threshold measurements were obtained using Custom Sound EP and used for the initial programming of the speech processor as previously described²². In Custom Sound Suite 4.0, ECAP thresholds were measured and used to set the dynamic range. The ECAP threshold was used as the comfort level. The threshold level was set to ~6 dB (~300 μA) below the ECAP threshold, as our previous studies showed this level of stimulation correlates to cortical thresholds²². The implant impedances and ECAP thresholds were checked every 3–5 days and changes to comfort and threshold levels made accordingly. A standard frequency allocation table spanning the frequency range from 188 to 7,938 Hz was used (Extended Data Fig. 2). The pulse width was set to 25 μs, stimulation rate to 900 Hz, and stimulus maxima to 1. In order to activate specific channels, tones that coincided with the centre frequency of each analysis filter in the speech processor map were used, combined with the use of a single maximum (n = 1 in the ACE cochlear strategy, which is an n-of-m stimulation strategy).

Cochleograms

Histological confirmation of cochlear lesioning was completed via cochleogram analysis by CILcare. Rats were unilaterally deafened in the left ear using the deafening protocol described above and allowed to recover for 4–7 days. Rats were then humanely euthanized, and temporal bones were removed and fixed in 10% neutral buffered formalin for 48 h. The temporal bones were then transferred to 1× HBSS and sent to CILcare for processing. Temporal bones were decalcification in 0.125 M EDTA for 7 days. The samples were bisected along the modiolar axis, excess material was removed, and the half-turns of the cochlear were separated. Immunostaining was completed with Myo7a (primary antibody 1:400, rabbit anti-myosin VIIa, Axxora PTS-25–6790-C050; secondary antibody 1:1,000, goat anti-rabbit Alexa Fluor 488, Abcam AB150077) and samples were scanned confocally for OHC and IHC counts. The right ear was intact and used as the normal-hearing comparison for hair cell counts compared to the left deafened ear.

CI training

Rat intracochlear electrodes were connected to a clinical CI (CI24RE, Cochlear) via a custom commutator (Exmore) placed at the top of the behaviour box. The CI24RE was driven by a clinical speech processor (Freedom, Cochlear). The microphone of the speech processor was oriented toward the speaker. The speaker then played pure tones (1,000 ms duration) corresponding to the centre frequency associated with the electrode to be stimulated (Extended Data Fig. 2b). Speakers were calibrated based on electrodograms, which represent the sequence of stimulation pulses (stimulated electrode, stimulation current, and pulse duration) as a function of time. In devices manufactured by Cochlear, this pulse magnitude sequence can be obtained by capturing and decoding the radio frequency (RF) output of the speech processor. The same model of speech processor programmed with the same processor settings used for behavioural testing was used for obtaining the electrodograms. Med-PC IV 4.2 software (Med Associates) was used to play the centre frequency for each electrode across sound intensities (50–90 dB, 2 dB steps) while an automated Matlab routine was used to record the RF output from the speech processor. RF output was captured using the Clinical Programming System (CPS) connected to a PC by an IF5 card (Cochlear) and controlled with the NICCaptureClient commands built into the Nucleus Implant Communicator 2 software package (Cochlear). These commands record the pulse sequence represented in the RF output. The speech processor and CPS unit were placed with the microphone oriented towards the speaker in the operant conditioning box, as done during behavioural testing. The electrodograms were then analysed, and the centre frequency intensity for each electrode that generated 80% maximum pulse amplitude while minimally stimulating other electrodes was selected for behavioural training (Extended Data Fig. 2c).

Rats were re-trained on the behavioural task, beginning with the nosepoke training to confirm recovery and task engagement. Rats had to use the CI to respond to the centre frequency of the target channel (stage 1). Once a rat reached a hit rate of >80% to the target, the two most distal channels were introduced as foils (stage 2). After a rat reached criteria of d′ = 1.0 on the two-foil version of the task, the centre frequencies of the remaining active channels were introduced as foils. Since the activation of the CI was acoustic, it was imperative that the rat was acoustically deaf, as confirmed by performance dropping to chance with the CI turned off (Fig. 1 and Extended Data Figs. 3, 4 and 10), and compared to the d′ in the session immediately preceding or following with the CI on. For the analysis of Fig. 1, maximum d′ was the highest d′ value across all days of stage 2 after all active foil channels were used for training. For comparisons in Figs. 3 and 4 and Extended Data Fig. 9, maximum d′ was computed only in rats that had at least six days of behavioural training to ensure comparisons with steady-state performance rather than initial learning rates. All rats received CI stimulation only during behavioural training and pairing. Specifically, the sham-paired and LC-paired rats received the same amount of CI stimulation during pairing, and comparable levels during training.

At least three active channels (one target, two or more foils) were required for stage 2. All rats had 4+ active channels up to d′ ≥ 1, and if an active channel used as target or foil went bad, we re-trained rats with either the nearest-available neighbour channel as target or foil (respectively), as long as 3+ channels remained active overall. Training had to be discontinued in an rat once all electrodes within an implant were broken (impedances >30 kΩ). This eventually occurred in 14/26 of rats from Figs. 1 and 3, resulting in 12/26 trained rats included in Fig. 4. This is probably owing to rats inducing mechanical damage to the lead wires of the implant. For six rats in Fig. 3d, this occurred after only 1–2 days of training; since these rats did not have adequate experience with the implant, we could not fairly calculate their ‘maximum’ performance for Fig. 3e.

Viral injections and optic fibre placement in LC

Viral injections into female rat LC²⁷ were performed using stereotaxic coordinates (from lambda, in mm: 3.7–3.8 posterior, 1.2–1.4 lateral, 5.6–6 ventral) with the head at a 15° downward angle. A craniotomy was placed over left LC and location was verified during surgery by measuring multi-unit spontaneous activity and responses elicited by noxious stimuli (toe pinch) and confirmed afterwards using histological methods and μ-CT/μ-MRI co-registration (Extended Data Fig. 6). Injections were performed with a 5 μl Hamilton syringe and a 33-gauge needle. We used two 1.0 μl injections at 0.1 nl s⁻¹ into LC: one at the most dorsal point where noxious stimuli responses were observed and the second 250 μm ventral to that location.

For viral expression in noradrenergic LC neurons, we used TH-cre Long-Evans rats to restrict expression of Cre-inducible viruses⁴⁸. For optogenetic stimulation, we injected AAV5-ef1α-DIO-ChETA-EYFP (viral titre: 1.5 × 10¹³); for sham stimulation we injected AAV5-ef1α-DIO-EYFP (viral titre: 1.0 × 10¹³); and for fibre photometry we injected AAVDJ-ef1α-DIO-GCaMP6s (viral titre: 1.1 × 10¹³ (Deisseroth laboratory)).

A calibrated optical fibre ferrule (200 μm for optogenetic experiments, 400 μm for fibre photometry experiments, ThorLabs) was implanted in LC, and craniotomy and implant were sealed with silicone sealant and Metabond (Parkell).

Fibre photometry

To perform fibre photometry (n = 4 CI rats, n = 2 normal-hearing rats), 210 Hz sinusoidal blue light (465 nm) and 330 Hz ultraviolet light (405 nm; to control for motion and photobleaching) were delivered via the optical fibre from an LED (100–200 μW) for GCaMP6s excitation (Doric). We collected the emitted light via the same optical fibre, using an integrated fluorescence mini cube (Doric) to direct emitted light to a femtowatt silicon photoreceiver (Newport), and data were recorded using an RZ6 real-time processor and Synapse 96 software (Tucker-Davis Technologies). The analogue readout was then low-pass filtered at 10 Hz. Behavioural boxes (Med Associates) were synchronized with the fibre photometry system using TTL pulses to align nosepokes and tone onset to LC activity. A custom dual commutator system was built to allow for simultaneous fibre photometry recording and CI behaviour (Doric, Exmore).

Recordings were analysed using a custom Matlab script. Data from both channels were smoothed using a moving average. The GCaMP signal was then corrected by normalizing each channel using a least-squares regression. The dF/F was generated by subtracting the fitted 405 nm signal from the 465 nm signal in order to reduce movement and photobleaching artefacts. To compare across rats and recording sessions, the signal for each trial was z-scored using the 3 s prior to trial initiation. Across the four rats, there were 2–4 days spent in stage 1, and 3–4 days spent in stage 2. Each day consisted of 1–5 sessions, where each session was 11–68 individual trials, with an average number of trials per session of 43.4 ± 2.9 trials for stage 1 sessions and an average of 46.5 ± 1.9 trials for stage 2 sessions. Traces were trial-averaged based on behavioural response: hits, false positives, misses, or withholds.

Optogenetic pairing

Starting on the first day of target tone training with the CI, the target tone (centre frequency of target channel) was paired with activation of the LC with blue light (n = 10 rats for LC pairing, n = 16 rats for sham pairing). For optogenetic stimulation, LC-tone pairing was conducted at a rate of 0.33 Hz, for 5–10 min daily immediately prior to behavioural testing, for a total of 100–200 pairings per session. Optogenetic LC stimulation began at tone onset, tone duration was 1,000 ms (same as during behavioural training), and optogenetic LC stimulation was 500 ms duration at 10 Hz with 5 ms pulses. This pairing procedure was performed each day throughout training with the CI. Pairing was conducted outside of the context of behaviour for two reasons: (1) we were interested primarily on the longer-term effects of LC pairing for neuroplasticity in cortex and elsewhere for representations of the paired stimulus, rather than the more immediate neuromodulatory effects of LC activation; and (2) to reduce the potential impact of LC activation on behavioural performance, such as freezing and changes in attention, as electrical or optogenetic LC stimulation can sometimes lead to behavioural arrest in the moment, particularly at rates of ≥10 Hz (which are most effective for cortical plasticity)^26,49.

Micro-computed tomography imaging for LC and CI targeting confirmation

At the end of behavioural experiments, rats were scanned on CT to confirm that the CI was intracochlear and for co-registration analysis of optical fibre placement for optogenetic rats. The number of intracochlear electrodes was estimated from individual planes of the CT scan. Based on this estimate, the dimensions of the arrays used, and a place-frequency equation⁵⁰, we were able to estimate the tonotopic location of the target channel. From this, we derived the mismatch between the CI target and the previous normal-hearing target (4 kHz).

The localization of the implanted optical fibres was assessed in vivo using μ-CT scans in post-implanted rats, followed by co-registration with a μ-MRI rat brain atlas (Extended Data Fig. 8a, c). Unlike μ-MRI, μ-CT imaging can be performed on subjects with metal implants. We combined registration of post-implant μ-CT with a three-dimensional μ-MRI rat brain atlas (as lack of soft tissue contrast in μ-CT limits the anatomical detail required to precisely verify optical fibre placement). Three-dimensional (3D) μ-MRI brain datasets were acquired from four female 3–5 months old Long-Evans rats. These datasets were generated with no slice gap in order to compare LC location to a widely used digital MRI rat brain atlas obtained from an adult Sprague-Dawley rat based on the Waxholm space coordinate system (https://www.nitrc.org/projects/whs-sd-atlas). Datasets were generated with a 3D RARE sequence acquired with 91 μm isotropic resolution on a 7-Tesla scanner using various T₂-weighting contrast with echo times ranging from 30–50 ms to delineate LC. After careful examination, we did not observe any noticeable structural difference within the LC region, which led us to use the Waxholm space atlas throughout this study after its manual segmentation and colour-coding based on the Paxinos and Watson Rat Brain Atlas⁵¹.

The μ-CT datasets were acquired using the μ-CT module of a Multi-Modality hybrid micro-Positron Emission Tomography (μ-PET) / μ-CT Inveon Scanner (Siemens Medical Solutions). The Inveon scanner is equipped with a 165-mm × 165-mm X-ray camera and a variable-focus tungsten anode X-ray source operating with a focal spot size of less than 50 μm. The scan consisted of a 30-min whole-head acquisition over an axial field of view of 44 mm and a transaxial of 88 mm with a resolution of 21.7 μm pixels binned to 43.4 μm. 440 projections were acquired using a 1 mm aluminium filter, a voltage of 80 kV, and a current of 500 μA. The datasets were reconstructed using the Feldkamp algorithm⁵².

The hybrid scanner is equipped with a M2M Biovet module used to monitor continuously vital signs. All rats were monitored continuously throughout the scanning session via a respiration sensor pad and electrocardiogram. The imaging scan consisted of initially placing each rat in an induction chamber using 3–5% isoflurane exposure during 2–3 min until the onset of anaesthesia. The rat was then subsequently positioned laterally along the bed palate over a thermistor heating pad in which 2.0% to 3.0% isoflurane was administered via a 90° angled nose cone throughout the scan. The head of each subject was judiciously oriented perpendicular to the axis of the rat body so that the extracranial part of the implanted electrode could be easily kept away from the field of view of the μ-CT image acquisition. Importantly, the large extracranial metal components and dental cement of the implant can cause beam hardening that can appear as cupping, streaks, dark bands or flare in the μ-CT^53–55. To this effect, the head positioning helped reduce the risks of image artefacts that could be induced by the implant along the path of the X-ray beam.

LC was manually segmented and colour-coded with guidance from the Paxinos and Watson Rat Brain Atlas⁵¹. This region was set as the target of reference. A rigid co-registration between the acquired μ-CT and the modified μ-MRI atlas images was systematically performed using a commercial software Amira 5.5 (Thermo Fisher Scientific). Both datasets were overlaid to match the intracranial space between both imaging modalities with location of bregma and lambda used as cranial anchors towards stereotaxic localization. Visual analysis helped determine the sub-millimetric localization of the electrode tip. This analysis was conducted by two individuals blind to the behavioural and viral status of the rat. Three rats were excluded from the localization analysis of Extended Data Fig. 8c as they lacked immunohistochemical data on optical fibre placement; one of those rats had no viable μ-CT signal, and the other two rats were not imaged for μ-CT (including for intracochlear electrode placement).

Immunohistochemistry

At the end of the behavioural, imaging, and electrophysiological studies, rats were perfused with 4% paraformaldehyde, brains recovered, and embedded in Optimal Cutting Temperature compound prior to freezing at −80 °C. Afterwards, 20 μm thick slices were cut from the brainstem and stained using standard immunohistochemistry histological methods. Staining for tyrosine hydroxylase (primary antibody 1:500, chicken anti-TH, Aves Labs TYH; secondary antibody, 1:1,000, goat anti-chicken, Alexa Fluor 568, Abcam ab175477) was co-localized with YFP (primary antibody 1:500, rabbit anti-GFP, Abcam ab290; secondary antibody 1:1,000, goat anti-rabbit, Alexa Fluor 488, Abcam AB150077).

Electrophysiology

ECAP thresholds from the auditory nerve were measured using autoNRT (Custom Sound Suite 4.0, Cochlear) an automated system for the Nucleus Freedom speech processor⁵⁶. Traces were analysed for ECAP magnitude using a custom Matlab script.

For CI-evoked multi-unit cortical responses, rats were anaesthetized as described above. Following partial resection of the temporalis muscle, the temporal skull was removed to expose auditory cortex contralateral to the implanted ear. In vivo multi-unit recordings were made with a Multiclamp 700B amplifier (Molecular Devices). Multi-unit recordings were obtained from 500–700 μm below the pial surface with tungsten microelectrodes (0.5 MΩ). An individual tungsten electrode was repeatedly inserted into the cortex across multiple sites, to obtain measurements of implant-evoked responses for as long as neural tissue and the rat remained viable. We used cortical vasculature to guide placement of the electrode, as the rats were bilaterally deafened and tonotopy could not be evaluated. We recorded 20 responses per implant channel, for each recordings site. Across all rats, there were between 5 and 8 active channels (7.3 ± 0.3 channels per rat) and between 7 and 29 recordings sites for each rat (19.0 ± 1.4 recording sites per rat).

In vivo whole-cell recordings from anaesthetized rat auditory cortex were obtained as previously⁴⁶. Patch pipettes (resistance of 5–7 MΩ) contained (in mM): 130 Cs-methanesulfonate, 1 QX-314, 4 TEA-Cl, 0.5 EGTA, 10 phosphocreatine, 10 HEPES, 4 Mg-ATP, 0.3 Na-GTP (osmolality, 285 mOsm; pH 7.32 adjusted with CsOH). The pressure of the patch pipette was monitored with a manometer, 15–20 mbar pressure was applied when the patch pipette was lowered into the brain, and pressure was adjusted to 1.5–2 mbar when the pipette reached the targeted depth from the pial surface. Recordings in voltage-clamp mode were obtained with a Multiclamp 700B amplifier (Molecular Devices) and data were acquired with Clampex 10.7 (Molecular Devices), low-pass filtered at 1 kHz, high-pass filtered at 100 Hz and digitized at 20 kHz.

The CI24RE implant was driven by a Freedom speech processor connected through the clinical programming pod to a Windows personal computer running the Custom Sound Suite 4.0 software (Cochlear). Custom Sound Suite’s evoked ABR function was used (5 charge-balanced biphasic pulses, 25 μs per phase, 900 Hz stimulation frequency, 57 μs pulse duration, 20 sweeps at 0.9 Hz) to stimulate the individual electrodes of the array. A modified cable with stereo jack and Bayonet Neill-Concelman connectors was used to connect the Programming Pod to the trigger input of the Digidata 1440A (Molecular Devices), facilitating coordination of each sweep of the electrical stimulus (stimulation intensity was controlled through Custom Sound) with the multi-unit activity recording setup in Clampex 10.7 (Molecular Devices). For cortical analysis, z-scored CI-evoked responses and neural d′ were calculated using a custom Matlab script, quantifying number of spikes in the 100 ms post-stimulus period compared to pre-stimulus baseline. Spike thresholding was based on the RMS of the baseline period. Auditory cortical recording sites with no z-scored evoked activity >0.4 were considered to be non-responsive and not included in analyses. The neural d′ was computed as the difference in the z-score of the behavioural target channel and the average of the z-scores of the behavioural foil channels. Spike timing jitter was defined as the standard deviation of the trial-by-trial latency to the first spike evoked by stimulation of the best channel per site⁴⁶. The coefficient of variation was calculated as the standard deviation of z-scored activity for all foil channels divided by the average of the z-scored activity for all foil channels.

Statistics

Statistical analysis was conducted using Graphpad Prism 9.2.0 and Microsoft Excel 16.65. Behavioural data from sham-paired CI rats in Fig. 1e–h also served as the control for Fig. 3c–g. Impedance measurements from sham-paired CI rats in Extended Data Fig. 5c was also shown for comparison in Extended Data Fig. 9d. Paired two-tailed Student’s t-tests were performed in Fig. 1d and Extended Data Fig. 4a,b,g, unpaired two-tailed Student’s t-tests were performed in Figs. 3e and 4c,f and Extended Data Figs. 8f, 9b, e–g and 10c, and paired one-tailed Student’s t-tests were performed in Extended Data Fig. 4c. Unpaired two-tailed Mann–Whitney tests were used in Figs. 3d, 5b, 8g and 9a,b. Pearson’s correlations were computed in Figs. 1i, 2f,i and 4h–k and Extended Data Figs. 5d–l, 6b,c,e,f, 7, 8d,e, 9c,h,k–m and 10d,e. A mixed-model two-way ANOVA with Tukey’s multiple comparisons correction was conducted for Extended Data Fig. 10b and Supplementary Table 1. Error bars and shading on line plots denote ± s.e.m. unless otherwise stated.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Extended Data

Extended Data Fig. 1 |. Auditory conditioning on go/no-go task in normalhearing rats.

a, Normal-hearing behavioral response curves from three example rats that reached training criteria. Arrowhead, target tone was 4 kHz for all animals. Error bars, response rate ± 95% confidence intervals. b, Average initiation rates for final five days of normal-hearing behavioral performance (N = 16 rats). Error bars, mean ± s.e.m. c, Average hits and false positive rates for final five days (N = 16 rats). d, Average behavioral performance (d’) for final five days (N = 16 rats). Error bars, mean±s.e.m. e, Days to d’ ≥ 1.0 (N = 16 rats). Error bars, median±interquartile range.

Extended Data Fig. 2 |. CI programming.

a, Depiction of center frequencies for individual CI channels. b, Frequency allocation tables used to select tones for behavioral conditioning based on center frequency of channels with different electrode configurations. c, Example electrodograms, showing that only the CI channel for selected center frequency was activated by the tone.

Extended Data Fig. 3 |. Deafened animals used the CI to perform the auditory task.

a-p, Behavioral performance for all 16 rats from Fig. 1. Each subpanel is a separate animal. Upper left, behavioral response rates across CI channels with the CI turned on (black) or turned off (red). Arrowhead, target tone programmed to activate channel 3 or 4. Error bars, response rate ± 95% confidence interval. Upper right, d’ over days on stage two. Lower left, hit rate over time. Lower right, false positive rate over time. No difference in d’ for rats with 7–8 active channels (N = 8) vs 3 active channels (N = 8) on testing day for implant on vs off (7–8 active channel implant on d’: 1.8 ± 0.2, 3 active channel implant on d’: 1.6 ± 0.2, p = 0.61, unpaired two-tailed t-test; 7–8 active channel implant off d’: 0.0 ± 0.1, 3 active channel implant off d’: −0.1 ± 0.1, p = 0.39).

Extended Data Fig. 4 |. Behavioral and electrophysiological confirmation of deafness in implanted rats.

a, Hit rates were lower and false positives were higher in rats when the CI was off (N = 16 rats, on vs off, hits: p < 0.0001; false positives: p = 0.02; paired Wilcoxon signed-rank test). b, Initiation rates decreased when the CI was turned off (N = 16 rats, on vs off, p = 0.0002; paired Wilcoxon signed-rank test). c, Hairs cells were lesioned by the deafening process. Representative immunohistochemistry from normal-hearing cochlea from the right ear (‘NH’) and deafened cochlea from the left ear (‘Deafened’) stained with hair cell marker Myo7a. Scale bar, 100 μm. Summary of hair cell counts showed that deafening significantly reduced the number of OHCs (‘NH’, 2746 ± 116 OHCs in normal-hearing animals; ‘Deaf’ 52 ± 46 OHCs in deafened animals; 98.1% loss of OHCs, N = 4, p = 0.0001, Student’s one-tailed paired t-test) and IHCs (‘NH’ 787 ± 34 IHCs; ‘Deaf’ 387 ± 118 IHCs; 50.8% loss of IHCs, N = 4, p = 0.04, Student’s one-tailed paired t-test). d, ABRs were gone both acutely (immediately after deafening) and weeks later. Example waveforms from the same rat with 4 kHz stimuli at 70 dB SPL, 80 dB SPL, and 90 dB SPL; chronic post-deafening ABRs measured 41 days post-deafening. e, EABRs intact both acutely and weeks later. Stimulation was at ECAP threshold. f, Standard deviation (SD) of acoustically-evoked ABRs (solid lines) and baseline noise (dashed lines) across frequencies from sample rat displayed in d,e. Red, acute deafness; blue, chronic deafness 41 days later. g, Summary of ABR/EABR recordings with stimuli of 4 kHz at 90 dB SPL or ECAP threshold in 14 rats pre- and post-deafening (6 pre and acute post; 4 chronic post; 4 pre, acute, and chronic post). ABRs were equivalent to baseline noise after deafening (‘Pre’, before deafening noise SD: 1.1 ± 0.2 μV, before deafening ABR SD: 3.2 ± 0.3 μV, N = 10, p = 0.0003,; ‘Acute’ just after deafening noise SD: 1.2 ± 0.1 μV, just after deafening ABR SD: 1.2 ± 0.3 μV, N = 10, p = 0.61; ‘Chronic’ weeks after deafening noise SD: 1.6 ± 0.2 μV, weeks after deafening ABR SD: 1.6 ± 0.1 μV, N = 8, p = 0.55; Student’s paired two-tailed t-tests). EABRs were significantly evoked (‘Acute’ just after deafening noise SD: 1.0 ± 0.1 μV, just after deafening EABR SD: 3.9 ± 0.7 μV, N = 10, p = 0.001; ‘Chronic’ weeks after deafening noise SD: 1.2 ± 0.1 μV, weeks after deafening EABR SD: 4.6 ± 0.8 μV, N = 8, p = 0.002, Student’s paired two-tailed t-test). Chronic measurements made between 13–42 days after deafening. h-k, Behavioral performance of four animals from g showing behavioral responses ± 95% confidence interval and d’ values post-deafening on stage 2 when CI is on (black) vs off (red) (upper left), d’ over time (upper right), hit rates and false positive over time (middle), and sample ABR and EABR traces (bottom). *, p < 0.05; **, p < 0.01.

Extended Data Fig. 5 |. Individual variability with CI use was related to false positive rate but not insertion depth, impedance, ECAP thresholds, hit rates, or normal-hearing performance.

a, Example x-rays of full insertion (8 channels) and partial insertion (4–7 channels). b, Days to d’ ≥ 1.0 did not differ based on CI insertion depth (full insertion: N = 9 rats vs partial insertion: N = 7 rats, p = 0.32, unpaired two-tailed Mann–Whitney test). Error bars, median±interquartile range. c, Average impedance of active CI channels over time. Grey dashed lines, individual rats (N = 16). Black, mean±s.e.m. d, Days to d’ ≥ 1.0 did not correlate with initial impedance values (N = 16 rats, Pearson’s r: 0.23, p = 0.40). e, Days to d’ ≥ 1.0 did not correlate with ECAP threshold (N = 16 rats, Pearson’s r: −0.21, p = 0.44). f, CI learning days to d’≥1.0 did not correlate normal-hearing learning days to d’ ≥ 1.0 (N = 16 rats, Pearson’s r: 0.07, p = 0.79). g, Days to d’ ≥ 1.0 did not correlate with hit rate (N = 16 rats, Pearson’s r: 0.16, p = 0.56). h, Days to d’ ≥ 1.0 correlated with false positives (N = 16 rats, Pearson’s r: 0.61, p = 0.01). i, During normal-hearing training, days to d’ ≥ 1.0 did not correlate with maximum d’ performance (N = 16 rats, Pearson’s r: −0.37, p = 0.16). j, Hit rates on first CI day were uncorrelated with hit rates on last normal-hearing day (N = 16 rats, Pearson’s r: 0.02, p = 0.93). k, False positive rates on first CI day were uncorrelated with hit rates on last normal-hearing day (N = 16 rats, Pearson’s r: 0.03, p = 0.90). l, d’ values on first CI day were uncorrelated with hit rates on last normal-hearing day (N = 16 rats, Pearson’s r: −0.03, p = 0.92).

Extended Data Fig. 6 |. Fiber photometry miss/withhold analysis.

a, Example LC activity aligned to tone onset during stage one CI training on miss trials, showing dF/F in high-miss rate behavioral session 1 (top) and in later low-miss behavioral session 5 (bottom). Error bars, mean ±s.e.m. b, In stage one, LC dF/F signals during miss trials were highest in sessions where the miss rates were highest (N = 4 rats, n = 21 sessions, Pearson’s r: 0.61, p = 0.003). c, LC signals were not predictive of false positive trials in stage two (tone-aligned: N = 4 rats, n = 39 sessions, Pearson’s r: −0.03, p = 0.83; response-aligned: N = 4 rats, n = 38 sessions, Pearson’s r: −0.30, p = 0.06). d, Example LC activity aligned to tone onset during stage two (foil and target training). Miss and withhold trials in high-false positive (F+) behavioral session 2 (top); miss and withhold trials in low-F+ behavioral session 12 (bottom). Error bars, mean ±s.e.m. e, Tone-aligned normalized dF/F LC signals during miss trials over all stage two sessions (N = 4 rats, n = 36 sessions, Pearson’s r: −0.01, p = 0.97). f, Tone-aligned normalized dF/F LC signals during withhold trials over all sessions (N = 4 rats, n = 40 sessions, Pearson’s r: 0.24, p = 0.14).

Extended Data Fig. 7 |. LC activity in normal-hearing reversal learning.

a, Schematic of go/no-go auditory behavioral task in normal-hearing rats when target tone is changed to a different frequency. After training to response to one target tone (black) while withholding from foil tones (green/red), one of the previously unrewarded tones (green) became the rewarded tone and the previously rewarded tone (black) became unrewarded. b, Example of animal performance on this task to first and second rewarded tones. Black arrowhead, first rewarded tone; green arrowhead, second rewarded tone. Error bars, response rates ± 95% confidence intervals. c, Tone-aligned LC activity, response-aligned LC activity, and miss rates across behavioral sessions in an example animal. Black, miss rates. Green, dF/F responses either tone-aligned (filled symbols, solid lines) or behavioral response-aligned (open symbols, dashed lines). d, Miss rates across all sessions were not correlated with tone-aligned normalized LC dF/F (N = 2 rats, n = 23 behavioral sessions, Pearson’s r: −0.05, p = 0.8). e, Miss rates across all sessions were correlated with dF/F when aligned to behavioral response (N = 2 rats, n = 23 behavioral sessions, Pearson’s r: 0.45, p = 0.03). f, Tone-aligned LC activity, response-aligned LC activity, and false positive rates across behavioral sessions in an example animal. Black, false positives. Green, dF/F responses either tone-aligned (filled symbols, solid lines) or behavioral response-aligned (open symbols, dashed lines). G, False positive rates across all sessions were not correlated with tone-aligned normalized LC dF/F (N = 2 rats, n = 23 behavioral sessions, Pearson’s r: −0.34, p = 0.1). h, False positive rates across all sessions were negatively correlated with dF/F when aligned to behavioral response (N = 2 rats, n = 23 sessions, Pearson’s r: −0.45, p = 0.03). i, Tone-aligned LC activity, response-aligned LC activity, and d’ across behavioral sessions in an example animal. Black, d’. Green, dF/F responses either tone-aligned (filled symbols, solid lines) or behavioral response-aligned (open symbols, dashed lines). j, d’ across all sessions correlated with tone-aligned normalized LC dF/F (N = 2 rats, n = 23 behavioral sessions, Pearson’s r: 0.52, p = 0.01). k, d’ across all sessions was not correlated with dF/F when aligned to behavioral response (N = 2 rats, n = 23 behavioral sessions, Pearson’s r: −0.11, p = 0.6).

Extended Data Fig. 8 |. LC targeting and behavioral comparison between sham-paired YFP-injected animals vs fiber-mistargeted animals.

a, Surgical approach for targeting LC. Multi-unit recordings were conducted to locate LC and then viral injection and optic fiber placement were based on these coordinates. b, Example LC multi-unit activity evoked by toe pinch. c, Optical fiber placement based on histology and μ-CT/ μ-MRI co-registration. Top, fiber placement in LC-paired animals. Bottom, fiber placement in sham-paired animals (red, mis-targeted fibers outside of LC; black, YFP-injected controls). Scale bar, 1 mm. d, In fiber mis-targeted animals, there was no significant correlation between distance of probe tip for optical stimulation and days to d’ ≥ 1.0 (N = 11 rats, Pearson’s r: 0.17, p = 0.63) e, In fiber mis-targeted animals with at least six days of CI training, there was no significant correlation between distance of probe tip for optical stimulation and maximum performance with CI (N = 10 rats, Pearson’s r: −0.02, p = 0.95). f, Days to d’ ≥ 1.0 was similar between the two sub-groups of sham-paired animals with either YFP-only expression in LC or when fiber was mis-targeted outside LC (YFP: N = 5 rats, mis-targeted: N = 11 rats, p = 0.38, unpaired two-tailed Mann–Whitney test). Error bars, median±interquartile range. g, Sham-paired animals in each subgroup with at least six days of CI training had similar maximum d’ (YFP: N = 4 rats vs mis-targeted: N = 10 rats, p = 0.71, unpaired two-tailed Student’s t-test). h, CI performance (d’) over time in YFP: N = 5 rats vs mis-targeted: N = 11 rats. One YFP animal in f,h and one mis-targeted animal shown in d,f,h did not reach the six-day requirement for maximum performance analysis; these animals are not displayed in e,g. i, Hit rates over time in YFP: N = 5 rats vs mis-targeted: N = 11 rats. j, False positives over time in YFP: N = 5 rats vs mis-targeted: N = 11 rats. Data are mean±s.e.m. except in f. One YFP rat and one mis-targeted rat did not reach the six day performance requirement to calculate maximum d’. This mis-targeted animal is displayed in d, but excluded from e, and both are displayed in f, but excluded from g.

Extended Data Fig. 9 |. LC-paired vs sham-paired animals had comparable CI insertions, impedances, ECAPs, behavioral initiation rates, and lack of residual hearing.

a, Number of intracochlear electrodes as assessed by x-ray was similar between LC-paired rats and sham-paired rats (LC-paired, N = 8 rats vs sham-paired, N = 16 rats, p = 0.93, unpaired two-tailed Mann–Whitney test). Blue, LC-paired animals. Black, sham-paired animals. Error bars, median±interquartile range. b, Degree of insertion did not predict performance across sham-paired (black) and LC-paired (blue) rats (full insertion: 8-channels, N = 14 rats vs partial insertion:4–7 channels, N = 10 rats, p = 0.36, unpaired two-tailed Mann–Whitney test). Measure of center, median. C, No significant correlation between estimated frequency mismatch and learning rate in both sham-paired (black, N = 16 rats, Pearson’s r: 0.39, p = 0.1) and LC-paired (blue, N = 8 rats, Pearson’s r: −0.37, p = 0.4) animals. Two LC-paired animals self-explanted their CIs prior to x-ray assessment of insertion; these animals are not displayed in a,b,c. d, Average impedances of CI channels over time in LC-paired (blue, N = 10) and sham-paired rats (black, N = 16). Dashed lines, individual rats. Solid lines, mean±s.e.m. e, Initial and final impedance values were similar in LC-paired and sham-paired rats (LC-paired: N = 10 rats vs sham-paired: N = 16 rats, initial: p = 0.32; final: p = 0.27; unpaired two-tailed Student’s t-test). f, ECAP thresholds during stage one and stage two training did not differ between LC-paired and sham-paired rats (LC-paired: N = 10 rats vs sham-paired: N = 16 rats, stage one: p = 0.91; stage two: p = 0.64; unpaired two-tailed Student’s t-test). g, Initiation rates were similar between LC-paired and sham-paired rats (LC-paired: N = 10 rats vs sham-paired: N = 16 rats, p = 0.35; unpaired two-tailed Student’s t-test). h, Learning rates with CIs (days to d’ ≥ 1) were not significantly correlated with initiation rates across LC-paired (blue, N = 10 rats, Pearson’s r: −0.49, p = 0.2) or sham-paired rats (black, N = 16 rats, Pearson’s r: −0.23, p = 0.4). i, Behavioral response rates were comparable in sham-paired (N = 16 rats) vs LC-paired (N = 10 rats) animals for sessions when the CI was turned off. (Blue, LC-paired: N = 10 rats vs black, sham-paired: N = 16 rats, hit rate: p = 0.72; false positives: p = 0.61; unpaired two-tailed Student’s t-test). j, d’ values were ~0 for both sham-paired (black, N = 16 rats) and LC-paired (blue, N = 10 rats) animals when the CI was turned off (p = 0.84; unpaired two-tailed Student’s t-test). Animals with high hit rates in I tended to also have high false positive rates; similarly, animals with low hit rates tended to have low false positive rates. k-m, Relating performance before and after deafening on CI performance vs acoustic normal-hearing (NH) task for last NH day vs first CI day in LC-paired animals. No significant correlation between hit rates (N = 10 rats, Pearson’s r: −0.05, p = 0.9) (k), false positives (N = 10 rats, Pearson’s r: −0.26, p = 0.5) (l), or d’ (N = 10 rats, Pearson’s r: −0.27, p = 0.4) (m). Data are error bars, mean±s.e.m. except in a, b.

Extended Data Fig. 10 |. Electrophysiological recordings from the auditory periphery and auditory cortex of implanted rats.

a, Example ECAPs in CN VIII from an LC-paired rat (left), a sham-paired rat (middle), and an untrained rat (right). b, Average ECAP amplitudes (P1-N1) were similar across groups and target/foil channels (LC-paired target ECAP amplitude: 120.0 ± 5.8 μV, LC-paired foil: 94.0 ± 5.2 μV, N = 4 rats; sham-paired target: 124.9 ± 19.6 μV, sham-paired foil: 111.9 ± 16.8 μV, N = 8 rats; untrained: 114.5 ± 3.9 μV, N = 4 rats). There was no significant difference between LC-paired, sham-paired, and untrained animal ECAPs (comparing LC-paired target vs foil, p = 0.27; sham-paired target vs foil, p = 0.55; LC-paired vs sham-paired target, p = 0.59; LC-paired vs sham-paired foil, p = 0.99; untrained vs LC-paired target, p = 0.99; untrained vs LC-paired foil, p = 0.61; untrained vs sham-paired target, p = 0.92; untrained vs sham-paired foil, p = 0.41; two-way ANOVA across all groups with Tukey’s multiple comparisons correction). c, Relative fraction of unresponsive multi-unit sites was greater in untrained vs trained rats (fraction of unresponsive sites in: untrained rat auditory cortex, N = 4 rats, 21.3 ± 10.1% vs trained rat auditory cortex, N = 12 rats 5.5 ± 2.1%, p = 0.03, Student’s unpaired two-tailed t-test). Total number of sites recorded from was comparable in untrained rats (18.5 ± 1.0 sites/animal, N = 4 rats), LC-paired trained rats (17.5 ± 2.8 sites/animal, N = 4 rats), and sham-paired trained rats (20.0 ± 2.4 sites/animal, N = 8 rats). d, Neural vs behavioral d’ values across animals as in Fig. 4h, but with neural d’ values computed using only sites tuned to the target channel (N = 12 rats, Pearson’s r: 0.36, p = 0.25). e, As d, but using only sites where foils were the best channel (N = 12 rats, Pearson’s r: 0.85, p = 0.0005).

Data availability

The data that support the findings of this study are further available on Zenodo (https://doi.org/10.5281/zenodo.7226424) or the NYU Data Catalogue (https://datacatalog.med.nyu.edu/dataset/10584). Source data are provided with this paper.