Effects of childhood hearing loss on the subcortical and cortical representation

1 Little is known about the effects of childhood mild-to-moderate sensorineural hearing loss (MM 2 HL) on the function of the auditory pathway. We aimed to examine the effect of childhood MM 3 HL and the benefit of frequency-specific amplification on both subcortical and cortical auditory 4 processing, and to relate it to speech-perceptual abilities. We recorded subcortical and cortical 5 responses to speech syllables in nineteen children with congenital MM HL (unamplified and 6 amplified), and sixteen children with normal hearing (unamplified sounds only). Speech 7 perception was measured behaviourally. Congenital HL led to smaller subcortical and cortical 8 responses to unamplified speech sounds. There was a significant benefit of amplification on 9 subcortical and early, but not late, cortical responses, with some effects differing across age. 10 No relationship was found between the neural and behavioural measures. Childhood MM HL 11 affects both subcortical and cortical processing of speech. Amplification mostly benefits 12 subcortical processing of speech in younger children. Childhood HL leads to functional changes 13 in the processing of sounds, with amplification differentially affecting subcortical and cortical 14 levels of the auditory pathway. 15 16 17 18 19 20 21 22 23 24 25


Introduction
Animal studies have shown biological and physiological alterations in the properties of neurons in the cochlea, subcortex, and cortex following even a temporary, mild-to-moderate (MM) hearing loss (HL) 1,2 .Similar changes have been documented in humans with congenital deafness [3][4][5] , or in older adults with age-related MM HL (for a review see 6 ).Nowadays, most individuals with MM HL who seek treatment receive auditory stimuli via hearing aids, which amplify the incoming auditory signal.To date, the bulk of studies investigating the effects of MM HL on the functional integrity of the auditory pathway (and the potential benefit of hearing aid amplification) has focused on older adults.However, congenital hearing loss affects up to 2 live births in 1000 7 , milder degrees of loss being more prevalent yet more often overlooked.
Here, we evaluate the effect of childhood MM HL and the benefit of amplification on simultaneously recorded subcortical and cortical responses to speech sounds.In the auditory system, speech sounds travel from the cochleae to the auditory cortices, undergoing increasingly complex levels of processing.At the level of the cochleae, sounds undergo a frequency analysis.The output of each channel can be separated into their relatively slowly fluctuating modulated signals known as the envelope (ENV), imposed upon a rapidly varying carrier waveform, known as the temporal fine structure (TFS).Both ENV and TFS cues are thought to be important in decoding speech.ENV cues support the robust identification of speech in quiet 8 , whereas TFS cues are thought to be especially useful in noisy backgrounds 9,10 .Indeed, individuals with typical hearing (TH) typically benefit from the TFS when they glimpse into the "dips" (i.e., local improvements in SNR) of a fluctuating background noise 11,12 .Therefore, speech intelligibility is much better in the presence of fluctuating than steady background noise, a phenomenon termed "masking release" 13,14 .However, adults with HL show reduced ability to use TFS 15 , a difficulty that limits the amount of masking release they experience 16,17 .Similarly to adults, children and adolescents with MM HL experience poorer speech intelligibility in noise [18][19][20] .They show reduced sensitivity to TFS compared to children/adolescents with TH 21 , which likely limits masking release of speech in the presence of a fluctuating background noise, even when they wear hearing aids 22 .
At the subcortical level of the auditory pathway, the frequency following response (FFR) reflects the periodicity of speech sounds 23 .Adding the neural responses to speech sounds presented in alternating polarities is thought to enhance the ENV of the response (henceforth FFRENV, here used as an index of fundamental frequency encoding), whereas subtracting them is considered to enhance the TFS (henceforth FFRTFS, an index of spectral harmonic encoding 24 ).Although the primary generator of the FFR appears to vary according to the fundamental frequency (F0) of the input stimulus, the current view is that the FFR emerges from multiple generators, including the cochlear nucleus, inferior colliculus and thalamus 25,26 .
Early studies suggested that the auditory brainstem reached maturity within the first 18 months of life 27 .However, more recent studies indicate that the FFR evoked by speech sounds keeps maturing well into the childhood years 28,29 .
At the cortical level, components of the late auditory event-related potentials (LAERs)     arise which are thought to reflect the initial detection (P1, N1, and P2), classification and discrimination (P2 and N2) of auditory stimuli, reflecting processing for auditory events ranging from isolated sounds to complex auditory scenes 30 .LAERs are thought to be generated in the auditory cortex (for a review, see 31 ), with additional contributors from non-auditory (e.g., prefrontal, premotor) areas 32 .The mismatch negativity (MMN) is thought to reflect the discrimination of a deviant sound, and is typically evoked by a comparatively rare deviant in a stream of repeated standards (i.e. an oddball paradigm; for a review see 33 ).The MMN is thought to have several additional generators, including bilateral temporal cortex, right inferior frontal gyrus, and bilateral frontal and centro-parietal regions (e.g., [34][35][36] ).In addition to distinct topographies, both LAERs and MMNs show distinct developmental trajectories.In typical development, P1 and N2 responses are thought to decrease in amplitude as children grow older, while N1 and P2 progressively appear in childhood, growing in amplitude until adolescence 37,38 .MMN amplitude is thought to increase throughout childhood 39 , but to be stable from around 10 years of age to adulthood 40,41 .
Both subcortical and cortical levels of the auditory pathway thus appear to follow a protracted trajectory, maturing until late adolescence.How MM HL affects the development of subcortical and cortical speech processing remains largely unknown.So far, studies have focused on either subcortical or cortical levels of processing, only providing fragmented information about the physiology of the auditory pathway in children/adolescents with MM HL.Yet using the right set of recording/analysis parameters, it is now possible to obtain subcortical and cortical responses during the same recording session 42,43 .Therefore, the first aim of this project was to evaluate the effect of childhood hearing loss on simultaneously recorded subcortical and cortical levels of processing.Noteworthily, the existing literature on this topic can be classified according to the strategy used to manage intensity levels across listener groups.

HL
When studying the effect of hearing loss on auditory processing, researchers often choose to investigate the unaided performance of individuals with HL.Stimuli are thus presented at a fixed intensity level for all individuals, with HL or not.Adult studies using this strategy to evaluate the effect of HL on subcortical processing have led to contradictory findings (for a review, see 44 ).While some authors reported smaller FFRENV and FFRTFS in adults with HL than adults with TH 45,46 , others did not show such group differences 47,48 .Note that a possible explanation for this discrepancy lies in the age-matching across groups 47,48 , or lack thereof 45,46 .
To our knowledge, no study has focused on the effects of childhood MM HL on subcortical processing of speech, using fixed intensity stimuli across groups of children/adolescents.At the cortical level, studies using fixed intensity levels across groups suggest that childhood MM HL leads to delayed and/or smaller LAERs evoked by speech or pure tones 49,50 .
Developmental changes following MM HL were also apparent in the MMN: While 8-11 year old children with MM HL showed age-appropriate MMN responses to sounds, this was no longer the case in 12-16 year old adolescents with MM HL, as shown both cross-sectionally and longitudinally 49 .This suggests that some effects of childhood MM HL might only reveal themselves in adolescence, highlighting the need to take age into account in developmental studies.

Neural processing of speech presented at similar sensation levels across groups
A common limitation of the above-mentioned studies using fixed intensity levels for all participants is that individuals with HL would have perceived the stimuli at reduced sensation levels (SL) relative to controls.Therefore, some researchers have attempted to equate for the effects of peripheral hearing loss by presenting listeners with TH and those with SNHL are presented with sounds at different intensities -more intense for individuals with SNHL (either by simulating the frequency-specific gain provided by individuals' hearing aids, or by recording the neurophysiological response while listeners wear hearing aids).With such adjustments for HL, adults with acquired MM HL show larger FFRENV to an amplified speech stimulus compared to an unamplified one 51,52 and in one study, than the FFRENV evoked by unamplified speech presented to TH listeners 47 .A similar benefit of amplification on the FFRENV evoked by the speech token /su∫i/ was recently reported in 18 children with HL aged 6 to 17 years, compared to the unamplified stimulus 53 .Note that older adults with acquired MM HL did not show a significant benefit of amplification on FFRTFS 47 .To our knowledge, the effect of amplification and childhood HL on FFRTFS has remained unexplored so far.At the cortical level, studies looking at the effect of childhood HL on LAERs by accounting for the effects of peripheral HL have led to contradictory findings.Pilot data from a small sample of 5 children aged 2 to 6 years show age-appropriate P1 and N2 evoked by speech stimuli when the children wore their hearing aids 54 .Another study focused on older, unaided children (9-12 years) with MM HL, to whom speech stimuli were presented on average 18 dB greater than in children with TH 55 .Note that here, the amplification provided to the group with HL was not frequency-specific -unlike hearing aids.Results indicate age-appropriate P1 and MMN responses in children with MM HL, but smaller (but not later) N2 responses compared to children with TH.However, 5 to 8 year-old children who wore their hearing aids (n = 7) showed significantly smaller LAERs evoked by complex tones than age-matched children with TH 56 .Interestingly, when re-tested three years later, the same children with HL showed age-appropriate LAERs.At both time-points, MMN amplitudes were age-appropriate in children with MM HL.To our knowledge, no study has directly investigated the intraindividual benefit of amplification on the cortical processing of speech in children/adolescents with MM HL.

Relationship between speech processing and perception of speech
A main stake of auditory neurophysiological studies is to relate neural processing to speech perception in a variety of contexts and clinical populations.Because the FFR reflects encoding of F0 and lower harmonics of the signal, that are particularly helpful in adverse listening environments, it is thought to relate to speech perception in noise.Better speech perception in noise has been associated with subcortical processing of complex sounds in quiet, as observed in children 57,58 , young 59 and older adults with TH 60,61 ; but see 62 for a study that did not find such an association).In fact, FFRENV evoked by amplitude-modulated tones presented in quiet even seems to predict sentence understanding in noise in adults with TH 63 .Recent evidence suggests that FFRENV (evoked by speech in quiet) correlates with speech perception in noise in unaided (but not in aided) 4 to 9 year-olds with congenital mild to moderate HL 64 .
Whether subcortical processing relates to speech perception in noise in older children/adolescents with congenital HL remains an open question.
Cortical processing of sounds is also thought to relate to behavioural speech perception, mostly in quiet.In 4-month old children with TH or HL, components of the LAER relate to parents' reports of auditory functional performance 65 .In adults, the MMN is thought to relate to perceptual speech discrimination, at least at the group level (for a review, see 66 ).However, no clear relationship has emerged between MMN amplitude and speech discrimination in children with TH so far [67][68][69] .To our knowledge, it remains unclear whether cortical EEG measures relate to speech perception in children/adolescents with congenital HL.

Study objectives
The main goal of this study was twofold: to evaluate (i) the effect of childhood HL and (ii) the benefit of amplification on both subcortical and cortical auditory processing.(i) So far, no study has investigated the effect of childhood MM HL on subcortical processing presented at fixed intensity levels across age-matched groups.Additionally, results from (older) adult studies have led to contradictory results.Therefore, this study will be a first exploration of the effect of MM HL on the processing of unamplified sounds throughout development.At the cortical level, we predict smaller/later LAER and MMN responses in children/adolescents with HL than those with TH, with a possible interaction between age and group on the MMN.(ii) Regarding the second objective, we expect larger FFRENV in children with HL when sounds are amplified than unamplified.At the cortical level, results from the literature are mixed, but have never compared LAER/MMN with and without amplification in the same children with HL.
We will thus explore the effect of amplification on the cortical responses evoked in children/adolescents with HL.
Given that subcortical and cortical processing have both been related to speech perception, an exploratory objective was (iii) to investigate how these neurophysiological responses relate to speech intelligibility in quiet and in noise.

Participants
Thirty-five children aged 8 to 16 years were included in the study.Nineteen (8 female) had a diagnosis of mild-to-moderate SNHL (HL group) and sixteen (8 female) were agematched children with typical hearing (TH group).All participants were monolingual British English natives and obtained a nonverbal IQ (NVIQ) score of at least 85 (Block Design subtest of the Weschler Abbreviated Scale of Intelligence 70 ).No child had any known medical, psychological, neurological or developmental disorders other than SNHL.Individual airconduction audiometric thresholds were measured bilaterally in all children at octave frequencies from 0.125 to 8 kHz using an Interacoustics AC33 audiometer.The study was conducted with the verbal assent of the participants, the written consent of their parents, and was approved by the UCL Research Ethics Committee (2109/004).It was performed in accordance with the Declaration of Helsinki.Participants in the HL group had better-ear pure tone average (BEPTA) thresholds of 21-70 dB HL across 0.25-8 kHz (see Figure 1), with on average 7 dB difference in PTA between the ears.All children in the HL group were spoken language users (i.e., non-signers), and sixteen had prescription hearing aids, but these were not used in this study.Children in the HL group were individually matched in age (± 6 months) to those in the TH group, and mean age did not differ significantly between the two groups (see Table 1).None of the children in the TH group had a known history of hearing loss (including otitis media with effusion), educational difficulties, or speech and language problems (based on parent/guardian report).
All had mean PTA thresholds ≤ 20 dB HL across octave frequencies 0.25-8 kHz in both ears and obtained thresholds no higher than 20 dB HL at any given frequency.
Group comparisons are presented in Table 1.Note that some children did not complete all behavioural and electrophysiological tasks.With respect to the behavioural tasks, two children (1 TH and 1 HL) did not perform the speech identification in noise tasks.Due to technical difficulties, two children (1 TH and 1 HL) did not perform the speech identification in a fluctuating noise task.Two children with HL did not perform the speech discrimination in quiet task.With respect to the EEG measurement, one child (HL) was only tested in the amplified, but not the unamplified EEG condition.The sampling frequency of the EEG recording was accidentally set to 2048 Hz for another child (HL).For this child, we could obtain the cortical measures, but not the subcortical measures.In all these cases, results were treated as missing data.

Data acquisition and pre-processing
Stimuli were digitised /bɑ/ (standard) and /dɑ/ (deviant) syllables (see Figure 2) originally spoken by a native British female speaker, and identical to those used in 21,49,71 .The intonation contours were made monotone at 220 Hz using Praat 72 and the final stimuli were RMS equalised with GoldWave 73 .Vowel formant frequencies were approximately 800, 1340 and 2700 Hz for formants F1, F2 and F3 respectively, although there were small differences between the two syllables as the stimuli were based on natural utterances.Both syllables had rising F1s over ~50 ms with a much larger transition for the /dɑ/ (~230 Hz) than the /bɑ/ (~115 Hz).The phonetic contrast was cued primarily by differences in the properties of the initial release burst (more intense for the /dɑ/, especially at higher frequencies) and on the formant transitions into the vowel.No F2 transition was present for the /bɑ/ but the /dɑ/ had a significant falling transition for F2 of ~350 Hz.
Each condition consisted of a total of 1320, 175-ms stimuli, presented in a passive oddball paradigm (deviant probability: 10%).Half of all stimuli were presented in positive polarity and the other half in negative polarity.The order of presentation of each polarity was randomized.The inter-stimulus interval was jittered randomly between 275 and 375 ms to limit neural adaptation 74 .EEG was recorded using a BioSemi ActiveTwo system at a sampling rate of 8192 Hz, from 32 scalp electrodes in the standard 10/20 configuration 77 .Additional electrodes were placed on each mastoid and recordings were re-referenced offline to the average of activity at the mastoid electrodes.Electrode offsets were consistently < 50 mV.

Subcortical response analysis
Subcortical response analyses were performed only on the standard (/bɑ/), to avoid the confound of small spectral differences between the standard and the deviant syllables.1188 epochs (standard probability: 90%) were obtained by applying a band-pass filter from 90 to 1600 Hz (zero-phase, finite impulse response, -6dB/octave) to the EEG data evoked by the standard syllables (/bɑ/) at Cz, and extracting data 20-170 ms relative to stimulus onset, which covers the duration of the steady portion of the vowel.Trials exceeding ± 100 µV were rejected before averaging.There was no significant difference in the proportion of trials that were rejected across groups [TH-U: 18.5%; HL-U: 19.1%; HL-A: 20.3%; F(2, 47) = 0.723, p = 0.491, The averaged responses to positive and negative polarities were then added together to enhance the envelope of the response (FFRENV), and subtracted from each other (i.e.positive minus negative polarity) to enhance the TFS of the response (FFRTFS).Spectral amplitudes were calculated using a fast Fourier transform (FFT), after applying zero-padding to the sampling rate to obtain an FFT with a resolution of 1 Hz.Spectral peaks at F0 (220 Hz) and harmonics (H2: 440 Hz; H3: 660 Hz; H4: 880 Hz; H5: 1120 Hz; and H6: 1320 Hz) were computed as the maximum amplitude within an 8-Hz bin centred on F0 and subsequent harmonics, following visual inspection of the grand average subcortical response, and in line with previous studies (e.g. 78).Neural background noise amplitudes were calculated by taking the mean amplitude of the fast Fourier transform across 100-Hz windows surrounding each spectral peak (50 Hz on each side, excluding the ±70 Hz adjacent bins).

Cortical response analysis
Evoked potentials considered to be cortical in origin were obtained by band-pass filtering the EEG recorded at electrode Fz between 0.5 Hz and 35 Hz (zero-phase, finite impulse response, -6dB/octave) and creating epochs from -100 ms to 500 ms relative to each stimulus onset time.1320 epochs (1188 standards and 132 deviants) were baseline corrected using the mean value from -100 to 0 ms.Trials exceeding ± 100 µV were rejected before averaging.
P1 and N2 grand average latency were quantified by means of the half-area latency measure of the standard waveform recorded in the TH group 79 , a computation that is less sensitive to high-frequency noise than traditional latency peak measures 80,81 .The window for computation of the area under the curve of the P1 ranged from 55 to 135 ms and from 250 to 350 ms for the N2, based on visual inspection of the TH grand average.P1 and N2 amplitudes were then computed for each participant and condition as the mean amplitude within a 40-ms window centred around the half-area latency (P1: 96 ms post-stimulus onset; N2: 329 ms poststimulus onset).P1 and N2 latencies were computed for each participant and condition as the half-area latency for the local peak to appear, within the 40-ms window centred around the TH grand average evoked by standards.Note that N1 and P2 were not analysed because they could not be reliably identified in the younger participants, which is likely due to a protracted maturation of these responses until adolescence 37,82 .
The MMN was computed as the differential wave obtained by subtracting the neural response evoked by the standard to that evoked by the deviant, for each participant and condition.The MMN was then statistically assessed in two ways.First, to determine whether there was evidence for neural discrimination of speech deviants, the presence of an MMN was evaluated for the three combinations of group and amplification conditions.Point-to-point comparison of the differential wave amplitudes was performed to determine the latency period over which the waveforms were significantly smaller than zero, if any.One-sided t-tests were computed within the 100-500 ms post-stimulus-onset time window which was identified as the region most likely to contain the MMN by visual inspection of the grand average waveforms.
Because adjacent points in the waveform are highly correlated, potentially leading to spurious significant values in short intervals, an MMN was considered present when p < .01(one-tailed) for more than 20 ms at adjacent time-points 83,84 .Half-area MMN latency was computed on the differential waveform recorded in the TH group.MMN amplitude for each participant and condition were then computed as the mean amplitude in a 100-ms window centred around the grand averaged differential waveform half-area latency peak observed in children with TH (at 426 ms post-stimulus onset 79 .MMN individual latencies were computed as the half-area latency within a 100-ms window centred around the grand average differential waveform of the children with TH.

Speech discrimination in quiet
The speech discrimination in quiet task was the same as that used in previous studies from our laboratory 21,85 .The two endpoints of the continuum were based on the same digitised /bɑ/ and /dɑ/ used in the electrophysiological recordings.A continuum of 100 stimuli was then constructed (including the endpoints), using the morphing capabilities of the programme STRAIGHT 86 .The task was delivered via a child-friendly computer game, incorporating an adaptive, three-interval, three-alternative, forced-choice paradigm.Stimuli were presented binaurally via headphones (Sennheiser 25 HD) at a fixed intensity (70 dB SPL), on a tablet, meaning identical levels presented in the two groups.Children were required to detect the oddone-out, where two of the intervals corresponded to the /bɑ/ end of the continuum, and the third, target stimulus fell somewhere along the rest of the continuum.Participants received visual feedback regarding the accuracy of their responses.A three-down, one-up procedure was used to select, on each trial, the appropriate target sound from a continuum, tracking a performance level of 79.4% correct 87 .This was preceded by an initial one-down, one-up procedure until the first reversal 88 .The initial step size was 15 stimulus places along the continuum, which reduced to five after the first reversal.The threshold was the arithmetic mean of the last four reversals in direction of the adaptive track, expressed as the stimulus number (0-99) of the target stimulus along the continuum.Therefore, smaller numbers indicate better performance.The task was preceded by five practice trials which contained standard-target difference that had previously been deemed suprathreshold for adults with TH 21 .

Speech perception in noise
The rationale for choosing the stimuli for this task was to provide the most direct comparison to the neurophysiological stimuli, which were speech syllables.Another advantage of syllables is that they minimise the role of linguistic and cognitive factors on auditory performance, allowing to focus more on lower level auditory properties.Speech reception thresholds (SRTs) were thus measured adaptively for 13 consonants presented in vowelconsonant-vowel (VCV) logatomes, e.g./imi/, /apa/.Vowels were /i/, /ɑ/, and /u/, and consonants were /p/, /k/, /b/, /d/, /g/, /f/, /m/, /n/, /l/, /v/, /w/, /y/, and /z/.Both the vowels and consonants varied from trial to trial, but only the response for the consonant was scored.Stimuli were recorded by a native British English female speaker in an anechoic chamber and were digitized via a 16-bit analogue-to-digital converter at a 44.1-kHz sampling frequency.The logatomes were presented together with two types of background noise: steady (i.e. unmodulated, but see 89 ) or fluctuating in amplitude sinusoidally at 8 Hz, which corresponds to a syllabic modulation rate 90 .The modulation depth was fixed at 1.0 and the starting phase was randomized between 0° and 360° on each trial.The noise was shaped on-and-off using a raisedcosine with 50-ms rise-fall times.
Participants were seated in a quiet room and the stimuli were presented binaurally over Etymotics ER-2 earphones at an overall level of 70 dB SPL, again meaning identical levels presented in the two groups.They were asked to repeat the logatomes as best as they could.
The experimenter scored the participants' responses using a graphical interface which showed the 13 possible consonants.No feedback was provided.
The expected wide variability in intelligibility across consonants at a given SNR would not allow their reliable use in a simple adaptive procedure.Therefore, SRTs for each combination of consonant and vowel were estimated from a prior study of young adults with

Speech perception in quiet
As a baseline and familiarisation for the speech perception in noise task, we also included a speech perception in quiet task.Percentage correct responses were measured for 78 consonants presented at a fixed intensity of 70 dB SPL, using the same VCV logatomes and procedure for recording the participants' responses as those used in the speech perception in noise task.

Statistical analyses
Regarding the subcortical data, we first ran an outlier detection algorithm.Data points that fell outside the mean ± 3 SD (within each combination of group, amplification, FFR component and harmonic) were considered outliers and excluded from further analyses.In total, 2.7% of data points were excluded from the analyses, most of them from the HL group.After excluding outliers, the subcortical data were normally distributed (Kolmogorov-Smirnov test: all ps < .05).Note that for both FFRENV and FFRTFS, there were significant differences in variance between groups as evidenced by Levene's test for equality of variance (both ps < 0.001).
Second, paired two-tailed t-tests with Bonferroni-Holm correction were conducted separately for each group (THU, HLU, HLA), FFR component (FFRENV and FFRTFS) and harmonic (F0 to H6) to determine whether the FFR responses were significantly above the neural noise floor.FFR responses are considered present if at least one of the three groups showed a significant difference between target frequency peak and spectral noise floor in each component, at each harmonic.For FFRENV, peaks were significantly larger than the noise floor only at F0 and H2 (all ps < 0.001).For FFRTFS, peaks were significantly larger than the noise floor at H3, H4, H5 and H6 (all ps < 0.05).
Third, a linear mixed model was used to determine whether amplitude of the spectral noise floor differed across groups (THU, HLU, HLA), FFR component (FFRENV and FFRTFS) or harmonic (F0 to H6).As expected, the only significant predictor was harmonic (p < 0.001).In the absence of a significant group effect on the baseline (p > 0.10), subsequent analyses were performed on the target frequency peak amplitude.
Regarding the cortical data, we found no outliers (± 3 SD from the mean) in the amplitude of either P1 or MMN.The cortical data were normally distributed (Kolmogorov-Smirnov test: all ps < .05).For both P1 and MMN amplitudes, there were no significant differences in variances between groups (Levene's test for equality of variance: both ps > .05).
Linear mixed models (lmerTest package of R 92 ) were used to evaluate group differences in both cortical (P1, N2 and MMN) and subcortical (FFRENV and FFRTFS) measures.Fixed effect predictors and all their interactions (all orders) were included in the initial model.In all models, the factor listener was used as a random intercept, and random slopes for all the appropriate predictors (harmonic and amplification) that did not lead to convergence issues or inestimable models.Backward stepwise reduction was then used to simplify the models based on F-tests, eliminating insignificant terms, but always keeping lower-order terms that were significant in higher-order interactions.This was done first for the random effects, and then the fixed effects.Only significant results are reported below.Analyses of group differences aimed to answer three main questions.
-First, is there a significant group difference in electrophysiological responses to unamplified speech stimuli?Here, the predictors were group (THU vs. HLU) and age (continuous variable), with harmonic as an additional predictor (FFRENV: F0 and H2; FFRTFS: H3 to H6) to model the subcortical data.
-Second, for children with HL only, is there a benefit of amplification on the neural encoding of sounds?Here, the predictors were amplification (HLU vs. HLA), BEPTA and age, again with harmonic as an additional predictor for the subcortical data.
responses that are similar to the control group (THU vs. HLA)?Again, the predictors

Results
A summary of the main effects of hearing loss (HLU versus THU) and of amplification (HLA versus HLU) is provided in Table 2, for subcortical, cortical and behavioural measures.First, group differences in the subcortical processing of unamplified speech were examined (THU vs. HLU; see Supplementary Table 1).FFRENV were significantly smaller (by 0.082 µV) in children with HLU than in children with THU (p = .001)for both harmonics, with no effect of age.Similarly, FFRTFS were significantly smaller (by 0.028 µV) in children with HLU than in children with THU (p = .015).Note that in neither analysis was there a significant group x harmonic interaction.

Subcortical responses
Next, the effects of amplification were examined by comparing the subcortical responses of children with HL presented with unamplified versus amplified speech stimuli (HLU vs. HLA; see Supplementary Table 2).Overall, FFRENV were significantly larger (by 0.118 µV) in response to amplified compared to unamplified speech sounds (p = 0.011).The amplification × age interaction got excluded during model reduction (p = 0.051), yet we observed a marginally significant decreasing benefit of amplification with age on FFRENV amplitude (p = 0.019).With respect to the FFRTFS, we observed a significant amplification × harmonic interaction.FFRTFS were significantly larger in response to amplified than unamplified speech sounds at H4, H5 and H6, but not at H3. Last, we compared the amplitude of the responses obtained when amplified sounds were presented to children with HL versus unamplified sounds presented to children with TH (HLA vs. THU; see Supplementary Table 3).The amplitude of the FFRENV did not significantly differ between HLA and THU (p = 0.168).With respect to the FFRTFS, we observed a group × harmonic interaction (p < 0.001).FFRTFS were significantly larger in the HLA than THU groups at both H4 and H6, but not at H3 and H5.First, we assessed group differences in the cortical processing of unamplified speech (THU vs. HLU; see Supplementary Table 4).As shown in Figure 6, P1 was smaller (by 2.05 µV) and later (by 2.79 ms) in the HLU than the THU group (both ps < 0.05).N2 was also smaller (by 3.79 µV; p < 0.001) but not later in the HLU than the THU group.Additionally, children with HLU had significantly smaller MMN responses (by 3.47 µV) than children with THU (p = 0.003).

Cortical responses
Next, we sought to determine whether amplification benefited the cortical processing of speech sounds in children with HL (HLU vs. HLA; see Supplementary Table 5).For P1 amplitude, we found a significant amplification × BEPTA interaction (p = 0.028; see Supplementary Figure 1).Perhaps not surprisingly, the P1 amplitude in children with better BEPTA thresholds was not changed by the relatively small amplification they received, whereas the P1 amplitude for children with poorer BEPTA thresholds (and who received greater degrees of amplification) was increased so as to make P1 amplitude more-or-less independent of BEPTA.P1 latency was longer in the unamplified than the amplified condition (p = 0.027).It was also longer for individuals with higher BEPTA thresholds (p = 0.024) and for younger children (p = 0.005), irrespective of the condition for both.For N2 latency, there was a significant main effect of amplification (p = 0.033).Even though the amplification × BEPTA interaction was also significant (p = 0.030), post-hoc contrasts were not significant (p = 0.842).There was no significant effect of amplification on MMN amplitude or latency (both ps > 0.10).Note that irrespective of the condition, MMN amplitude decreased with age in children with HL (p = 0.040).
Last, we evaluated the potential benefit of amplification for cortical processing of speech in children with HL to that of children with TH (HLA vs. THU; see Supplementary Table 6).There was no significant effect of group on P1 amplitude or latency (both ps > 0.10).Yet there remained a significant group difference when looking at N2.Despite amplification, N2 amplitude was smaller (by 2.43 µV; p = 0.004), but not later, in the group with HLA than the group with THU.Similarly, children with HLA showed significantly smaller MMN than the group with THU (p = 0.002).Note that, although the group × age interaction did not come out significant (p = 0.146), post-hoc comparisons indicate a significant decrease of MMN amplitude with age in HLA (p = 0.038) but not in THU (p > 0.10).
Supplementary Figure 1: Amplitude (µV) of the P1 for the HL group as a function of BEPTA.Individual (shapes) and group (lines) data are shown for both unamplified (HLU, plain lines) and amplified (HLA, dotted lines) raw.Shaded lines represent the 95% confidence intervals.

1. Speech discrimination in quiet
Figure 7 shows the behavioural performance of the two groups on the speech perception tasks.For speech discrimination of the speech sounds /bɑ/ vs /dɑ/ in quiet (Figure 7A), there was no significant effect of group, age, or their interaction on response thresholds (all ps > .10;see Supplementary Table 7A).

2. Speech identification in quiet and noise
Figure 7B presents the percentage correct consonant identification in quiet for the two groups (also see Supplementary Table 7B).Children with HL were significantly poorer than children with TH (p < 0.001).Note that for both groups, performance significantly improved with age (p = .003),although ceiling effects meant only a very small increase in percentage correct for the TH group.Finally, Figure 7C presents the SRTs in steady and fluctuating background noise for the two groups (also see Supplementary Table 7C).The significant main effects of group and noise type (both ps < .001)need to be considered in light of the significant group × noise type interaction (p = .006).As expected, both groups experienced masking release when they were presented with fluctuating noise (both ps < 0.001).However, the magnitude of this effect was larger in children with TH (8.5 dB SNR) than in children with HL (3.9 dB SNR).

Relationship between the neural and behavioural responses
The relationship between electrophysiological activity and behavioural responses was investigated by means of a set of correlations.Holm-Bonferroni correction was applied to correct for multiple comparisons where appropriate.
Given the evidence that perception of speech in quiet is dominated by the encoding of the envelope of the signal, we investigated the relationship between the amplitude of the FFRENV at both F0 and H2 and behavioural performance of speech identification in quiet, separately for children with THU and children with HLU.None of the correlations were significant (all ps > 0.05).
Adequate perception of speech in noise requires precise spectro-temporal processing of speech, particularly in the dips of a fluctuating background noise.In fact, the amplitude of speech-evoked FFR in quiet relates to the perception of speech in noise (see introduction).
Therefore, we investigated the relationship between amplitude of the speech-evoked FFRTFS at H4 and H6 (in quiet) and behavioural speech perception in steady and fluctuating noise, as well as masking release, separately for children with TH and children with HL (in both unamplified and amplified conditions).None of the correlations were significant (all ps > 0.05).
The speech stimuli used for the electrophysiological recordings were the endpoints /bɑ/ and /dɑ/ of the continuum used for the speech discrimination in quiet behavioural task.As such, we hypothesised that a relationship might be observed between the amplitude of the MMN and the stimulus distance threshold measured in the speech discrimination in quiet behavioural task.
However, there was no significant correlation between speech discrimination in quiet and MMN amplitude, as evaluated separately for both groups, and separately for the amplified and unamplified conditions in children with HL (all ps > 0.05).
Last, we investigated the relationship between subcortical and cortical responses, separately for both groups (and separately for HLA and HLU).In children with TH, larger FFRENV at F0 were associated with smaller N2 responses (p = 0.004).There was no other significant correlation between FFRENV at F0 and P1, N2 or MMN (all ps > 0.05).

Discussion
This study investigated the effect of childhood MM HL and the benefit of amplification on both subcortical and cortical processing of speech sounds.When sounds were presented at the same intensity across groups, both subcortical and cortical responses were smaller in children with HL than those with TH.When sounds were amplified, the subcortical responses increased in children with HLA compared to HLU.In fact, they could even be larger in children with HLA than THU (although only significantly so for the FFRTFS).At the cortical level, an amplification benefit was also observed on P1 amplitude but not on the amplitude of later responses (N2 or MMN).In fact, at the group level, children with HL did not show a significant MMN (unlike children with TH), whether the sounds were amplified or not.When looking at individual responses, the amplitude of the MMN tended to decrease with age in children with HL but not in those with TH.Behaviourally, children with HL had poorer speech intelligibility than those with TH both in quiet and in noise, and they showed less benefit from background noise fluctuations.Yet, the relationship between the neurophysiological responses and the behavioural performance remains unclear.

Subcortical processing
When presented with unamplified sounds, children with MM HL showed significantly smaller FFRENV and FFRTFS than those with TH.This finding replicates and extends previous results from older adults with age-related HL 45,46 -but contrasts with some studies that did not find group differences on unamplified FFR measures 47,48 .The etiology of the HL might thus modulate its effect on subcortical processing of unamplified sounds.When presented with amplified sounds, children with MM HL showed a clear benefit from the frequency-specific gain applied to the sounds.In fact, the amplified FFRENV of children with MM HL did not significantly differ from that of children with THU.Furthermore, the amplified FFRTFS was even larger than that of children with THU.Our results thus fit with existing results from the adult and child literature to indicate a clear benefit of amplification at the subcortical level 47,51- 53,93,94 .
We were initially surprised by the spectacular increase in amplitude of the FFRTFS in the higher harmonics following amplification.Indeed, the amplitude of neural spectral peaks typically decreases with increasing frequency (e.g. 47,95,96).We believe this increase stems from applying the individualized, frequency-specific NAL gains calculated from audiograms with sloping hearing losses, which leads to increasing gains for higher frequency harmonics.To our knowledge, the only other study that used the NAL formula analysed FFRTFS response up to ~700 Hz.In this frequency range, there was no significant difference in the FFRTFS of older adults with HL (in quiet, with or without amplification) and age-matched adults with TH 47 .In our study, FFRTFS was only above noise floor at frequencies at H3 (660 Hz) and above.In fact, our results show significantly larger FFRTFS in HLA compared to both HLU and THU at H4 and H6, which are close to the first and second formant peaks of our stimulus.In adults with TH, FFRTFS are most clearly detected at harmonics that are close to formant peaks, up to 1500 Hz 24 .
Large FFRTFS at H4 and H6 might therefore reflect the interaction between frequency-specific NAL amplification and the formants of our speech stimuli.Yet a larger FFRTFS in children with HLA than those with THU might indicate hyperactivity at the subcortical level following childhood HL.
Hyperactivity has been widely reported as a consequence of SNHL (for a review, see 97 ), and is thought to compensate for reduced inputs from damaged peripheral structures to central brain regions, to maintain sensation 98 .Such compensation fits within the theoretical framework of homeostatic plasticity, which states that neurons regulate their physiological properties to compensate for persistent changes in their activity level 99 .Homeostatic mechanisms are thought to operate from synaptic to network levels of sensory systems.That FFR amplitude was larger in children with HL than those with TH, but only when sounds were amplified suggests that compensatory activity is intensity-dependent at the brainstem level.In mice, changes in auditory cortex responsiveness following conductive HL were only observed in sounds above a certain threshold, or in certain frequency ranges 100 .Future studies are warranted to investigate the effect of audibility on homeostatic plasticity of subcortical processing.

Cortical processing
Despite the medium to large benefit of amplification on subcortical responses, there was only a small benefit of amplification on cortical responses (P1), or no benefit at all (N2, MMN).
Individualized frequency-specific gain increased P1 amplitude (which reflects improved sound detectability, 31 ), but only for children with poorer BEPTA.Irrespective of BEPTA, P1 was earlier in HLA than HLU.In fact, the P1 amplitude and latency of HLA did not significantly differ from that of children with THU.This is in line with studies that show age-appropriate P1 when sounds are presented at similar sensation level to children with HL and those with TH 54- 56 .
Despite preserved P1, our results indicate that childhood HL leads to smaller (but not later) N2 even when sounds are amplified, in line with 55 , but contrary to 54,56 .Furthermore, we did not observe a significant MMN at the group level on children with HL, even when sounds were amplified.At the individual level, the MMN amplitude of children with HLA remained smaller than those with THU.Note however that there was no group difference in MMN latency.
Our finding of absent/smaller MMNs in 8 to 16 year-olds with HL contrast with those of ageappropriate MMNs in 5-12 year-old children 55,56 .Part of the explanation for this discrepancy might lie in the developmental effects of childhood HL on the central auditory pathway (see below for a discussion of the developmental effects observed in this study).
Why does the P1 amplitude benefit from amplification, but not the N2 or MMN?One possible explanation might lie in the different generators of the P1, N2 and MMN.The P1 generator has been reliably located within the secondary auditory cortex 101,102 , whereas MMN generators likely include an extended thalamocortical circuitry projecting to primary auditory cortex 103 .Little is known about the generators of the N2, but early intracranial recordings also seem to point towards a subcortical (thalamic) source 104 .Although anatomically close, the primary and secondary auditory cortices receive input from distinct portions of the medial geniculate body.Whereas the lemniscal pathway connects the ventral medial geniculate body to the primary auditory cortex, the nonlemniscal pathway connects the medial/dorsal geniculate body to both primary and secondary auditory cortices 105 .Given that our results indicate smaller N2 and MMN even when children with HL were presented with amplified sounds, these could indicate a specific alteration of the nonlemniscal pathway connecting the medial portion of the geniculate body to the primary auditory cortex.Our cortical findings thus contrast with the subcortical evidence of hyperactivity following childhood HL.

The effect of HL on the development of the central auditory pathway
In fact, our cortical findings also contrast with findings of hyperactivity in the auditory cortex as a consequence of age-related HL.Indeed, studies have reported some degree of enhancement of the cortical responses to sounds in older adults with age-related HL compared to age-matched individuals with TH 106-108 .To our knowledge, none of the pediatric studies investigating cortical processing of (amplified or unamplified) sounds in children with HL found such indications of increased cortical responsiveness 49,50,[54][55][56][109][110][111] . In mice,the same population of neurons showed distinct mechanisms of homeostatic adaptation if visual deprivation occurred in juvenile or adult animals 112 .Such an age-dependent effect of homeostatic plasticity might also be observed in the central auditory pathway following auditory deprivation.
Noteworthily, these age-dependent effects might already be operating between 8 and 16 years of age.Indeed, we observed a significant decrease in the benefit of amplification with age on FFRENV amplitude.Additionally, the amplitude of the MMN decreased with age in children with HLA (but not in those with THU) -although note that the group × age interaction was not significant.Both results seem to indicate that adolescence could be a period of heightened plasticity for functional changes following auditory deprivation.Future studies are warranted to investigate this claim in a larger sample size, ideally taking puberty into account, as it could act as a biological trigger to those alterations [113][114][115] .

Multiple gain mechanisms within the auditory system
To our knowledge, only three studies so far have combined subcortical and cortical measures from the same individuals to investigate the effects of HL on auditory processing.
Speech-evoked subcortical and cortical responses were recorded successively in older adults, with and without hearing aids.When sounds were presented at 65 dB SPL, hearing aid amplification lead to significantly larger subcortical, but not cortical responses 93

Relationship between speech processing & perception
An exploratory objective of our study was to investigate how changes in neurophysiological responses following HL might relate to speech intelligibility in quiet and in noise.To do so, we chose stimuli that were as similar as possible for both neurophysiological and behavioural tasks.However, two methodological choices are worth mentioning here: Unlike the neurophysiological task, stimuli were only presented unamplified in the behavioural tasks.Additionally, syllables were only presented in quiet in the neurophysiological tasks; yet speech perception was investigated in quiet and in noise.
Behaviourally, children/adolescents with HL have higher (i.e.poorer) thresholds in noise than children/adolescents with TH 19,20 .Their thresholds improved in the presence of fluctuating compared to steady background noise, but to a lesser extent than the group with TH.
Similarly to adults with HL [15][16][17] , children with MM HL seem less able to listen in the "valleys" of a fluctuating background noise to improve their speech perception.However, we did not find any significant relationship between measures of speech processing and perception.This contrasts with numerous studies that report small to medium correlations between subcortical responses and speech intelligibility, especially in noise 57-61, 63, 64 .Some correlations have also been observed between cortical responses and speech discrimination 65,66 , although the nature of this relationship remains unclear [67][68][69] .Again, our sample size might be too small to adequately investigate such possible correlations.Future studies including larger sample sizes are warranted to clarify the relationship between speech processing and perception in the context of congenital HL.These studies would benefit from a direct investigation of the effect of childhood HL on subcortical processing of speech in noise and its relationship with performance.

Conclusions
When presented with unamplified sounds, children with HL show impaired subcortical and cortical responses to speech syllables.Amplification was beneficial for subcortical and early (P1) but not late (N2, MMN) cortical responses.Our results support the existence of multiple gain mechanisms that could underlie homeostatic plasticity throughout the auditory pathway.Responses elicited at different levels of the auditory pathway appear to display idiosyncratic changes following HL, and may be age-and level-dependent.Even though speech perception was poorer in the group with HL than TH (in quiet and in noise), there was no clear relationship between the neurophysiological and behavioural measures.Even mild to moderate congenital HL induces functional alterations to the central auditory pathway, each level showing a specific alteration of its typical response.Future studies are needed to clarify the relationship between these functional changes and the behavioural performance.

Figure 1 :
Figure 1: Pure-tone air-conduction audiometric thresholds for children with HL (orange) and children with TH (blue and white).Audiometric thresholds are shown across octave frequencies from 0.25 to 8 kHz in the left and right ears.Individual thresholds for the HL group are shown as lines, and the group mean is shown as a bold line.Mean thresholds for the TH group are marked in white, with the shaded blue area representing the range for the TH group.

Figure 2 :
Figure 2:Top row: Spectrograms of the standard and deviant speech syllables in their unamplified form.Bottom row: Spectra of the vocalic portion (20 to 170 ms) of the standard syllable /bɑ/ in both unamplified and NAL-R amplified versions.Children with TH were only presented with unamplified versions of the stimuli.Children with HL were presented with both versions of the stimuli, with the amplification being dependent upon each individual's audiogram.For visualisation purposes, the amplified stimulus here was generated based on the average audiogram of the HL group.It can clearly be seen that, as expected, the level of the harmonics below ~2 kHz is unchanged because there is minimal hearing loss at those frequencies, with amplification of the higher harmonics in the frequency region in which there is substantial HL.
TH (unpublished), allowing the calculation of adjustment factors in SNR to give all stimuli the same nominal SRT91 .These factors were then used during the adaptive procedure to adjust the presented SNR.The signal-to-noise ratio (SNR) was varied adaptively by varying the level of the target and fixing the level of the noise at 70 dB SPL.The first logatome was presented at a nominal SNR of -10 dB (taking into account that token's adjustment factor).Depending on the participant's response (correct/incorrect), the SNR of the second logatome was decreased/increased by 10 dB.Again, depending on the participant's response (correct/incorrect), the SNR of the third logatome was decreased/increased by 6 dB.For each subsequent logatome, the SNR increased/decreased by 2 dB for incorrect/correct responses, respectively.Measurement stopped after either 12 reversals or 78 trials, whichever occurred first.The SRT was computed by taking the mean SNR (dB) across all track reversals.

Figure 3
Figure3shows the grand-average time waveforms and spectra of the subcortical responses

Figure 3 :
Figure 3: Grand average of the subcortical responses at Cz.The upper row represents the FFRENV in the time domain.The middle and lower rows represent the FFRENV and the FFRTFS evoked by the standard sound /bɑ/, presented either unamplified to children with TH (70 dB SPL, left panel), unamplified to children with HL (70 dB SPL, middle column), or amplified to children with HL (with a frequency-specific gain tailored to each child's hearing thresholds, right panel).

Figure 4 :
Figure4: Scatterplots of the amplitude (µV) of the FFRENV (F0 and H2) and FFRTFS (H3 to H6) over age, for each group (THU, HLU and HLA).Regression lines are fit on the basis of each combination of group, FFR component and harmonic.

Figure 5 :
Figure 5: Grand average waveforms at Fz, evoked by the standard (lighter line) and deviant (darker line) presented either unamplified to both children with THU (70 dB SPL, left column) and children with HLU (70 dB SPL, middle column) or amplified to children with HLA (right column).Voltage maps show the mean MMN activity during the 100-500 ms post-stimulus time window.Negative values of the MMN are shown in blue, and positive values in red.The horizontal line represents the duration of a significant MMN in the period of the 100-500 ms post-stimulus-onset epoch evoked in children with TH.No significant MMNs were observed in either the unamplified or amplified conditions for the children with HL as a group.

Figure 6 :
Figure 6: Amplitude (µV; left column) and latency (ms; right column) of the P1, N2 and MMN for each group and condition.Individual (shapes) and group (colour) data are shown for both unamplified (HLU, orange) and amplified (HLA, red) data.Shaded lines represent the 95% confidence intervals.Latencies (right column) are illustrated as group boxplots because there was no effect of age.

Figure 7 :
Figure 7: Boxplots of (A) stimulus distance (B) percentage correct consonant identification in quiet and (C) SRTs for consonant identification in steady (light grey) and fluctuating (dark grey) background noises.Children with TH are shown in blue, HL children are shown in orange.The boxes show the interquartile range (25 th -75 th percentile) of the data with horizontal line indicating the median.The whiskers indicate values that fall within 1.5 times the interquartile range.

Table 1 :
Mean (SD) participant characteristics for the HL and TH groups.Comparisons are independent-samples t-tests.Effect sizes are Cohen's d.Significant effects are shown in boldface.

Table 2 :
Summary of the main effects of hearing loss (HLU versus THU) and amplification 469perception in noise, a linear mixed model was computed with group (TH, HL), noise type 470 (steady, fluctuating), age, and all their interactions as fixed factors, and listener as a random 471 effect.Backward stepwise reduction was then applied, as for the EEG measurements.472Finally,the relationship between subcortical and cortical measures, as well as that 473 between electrophysiological and behavioural measures were investigated using correlations, 474 conducted separately on the TH and HL groups, and separately for unamplified and amplified 475 conditions in the latter group.478 (HLA versus HLU) for subcortical, cortical and behavioural measures.* p < 0.05, ** p < 0.01, 479 *** p < 0.001.
119n gerbils with developmental conductive HL, age-appropriate subcortical, but not cortical responses were recorded to amplitude modulated sounds presented at similar sensation level as in animals with TH116.More recently, Hutchison et al., (2023)117recorded subcortical and cortical neurophysiological responses in young adults with TH before, throughout and after fitting them with earplugs.The results indicate an increase in the subcortical response, but a reduction of cortical activity ipsilateral to the deprived ear.Together with our observations, the literature suggests that multiple gain mechanisms might underlie homeostatic plasticity throughout the auditory pathway118.Neurons from each level of the auditory pathway are thought to display idiosyncratic changes following HL (for a review, see119).