Suprathreshold Psychoacoustics and Envelope-Following Response Relations: Normal-Hearing, Synaptopathy and Cochlear Gain Loss

The perceptual consequences of cochlear synaptopathy are presently not well understood as a direct quantification of synaptopathy is not possible in humans. To study its role for human hearing, recent studies have instead correlated changes in basic suprathreshold psychoacoustic tasks with individual differences in subcortical EEG responses, as a proxy measure for synaptopathy. It is not clear whether the reported missing relationships between the psychoacoustic quantities and the EEG are due to the adopted methods, or to a minor role of synaptopathy for sound perception. We address this topic by studying the theoretical relationship between subcortical EEG and psychoacoustic methods for different sensorineural hearing deficits.


Introduction
The role of cochlear synaptopathy( i.e., the loss of innerhair-cell auditory-nervefiber synapses due to noise exposure or aging; or hidden hearing loss)f or suprathreshold hearing has been heavily contested in recent human studies [1,2,3,4] even though animal studies showclear histological evidence for synaptopathy [5,6,7]. It is not clear whether the cause of the missing correlations between subcortical EEG measures, as anon-invasive tool to quantify synaptopathy, and the suprathreshold psychoacoustic tasks stems from methodological confounds. It might be that the adopted subcortical EEG methods (e.g. the envelopefollowing response, EFR and auditory brainstem response, ABR)a re not sensitive markers of synaptopathyi nh umans, or,that the EEG methods are not targeting the same mechanisms involved in the psychoacoustic task, resulting in differential effects of synaptopathyonboth measures.
To address these issues, we study the theoretical relationship between the EFR and twoc ommon supra-threshold hearing tasks: tone-in-noise (TiN)a nd amplitudemodulation (AM) detection for different degrees of sensorineural hearing loss. We employacomputational model of the human auditory periphery that simulates neural responses to quantify psychoacoustic detection cues and subcortical EEG metrics [8]. We simulate howdifferent aspects of sensorineural hearing loss (synaptopathy, cochlear gain loss and combinations)affect the theoretical relationship between the EFR and psychoacoustic metrics to assess their sensitivity in quantifying synaptopathyi nh umans.
Psychoacoustic stimuli were delivered monaurally using insert ER-2 earphones connected to aT DT HB7 and Fireface UCX Soundcard and were calibrated using B&K 2669, 2610, 4153, 4134 products. All measurements were conducted in asound-proof booth and consisted of aprac-tice run followed by a3-alternative forced choice, 1-up-2down procedure with 3repetitions (AFC software). Stimuli were 500-ms long, followed by 500 ms of silence and thresholds were calculated as the mean overthe last 6reversals at the smallest step size. TiNd etection:s tep sizes were 8-4-2-1 dB. A65-dB SPL 4-kHz tone wasembedded in ao ne-octave wide white noise masker (i.e., the reference)ofvarying level(SNR within one NH 4-kHz equivalent rectangular bandwidth wasthe tracking variable). AM detection:T he initial modulation depth (MD) was −6dB re 100% modulation and stepsizes were 10, 5, 3, 1dB. The carrier wasa70-dB, 4-kHz tone, the modulation frequency 100 Hz and stimulus levels for different MDs were normalized to remove loudness cues.
EFRs were recorded on a3 2-channel Biosemi amplifieru sing magnetically shielded ER-2s for sound delivery while subjects watched as ilent movie in ar eclining chair.T he 16-kHz sampled Cz data wasr e-referenced to the offline averaged earlobe electrodes. Each of 600 stimulus repetitions lasted 600 ms followed by auniformly distributed random silence jitter (>90 and <110 ms). Stimuli were 100% modulated 120-Hz AM signals. Fort he TiN experiment, EFRs were recorded to a4-kHz centered oneoctave white noise carrier of 75 dB, whereas the carrier wasa70-dB 4-kHz pure-tone for EFRs in the AM experiment. Recordings were averaged, base-line corrected and filtered between 60 and 650 Hz before epoching and bootstrapping wasperformed to calculate the individual noise floors and confidence intervals [9]. The FFT wasc alculated from the averaged −0.01 to 0.6 sw indowr et rigger onset and EFR amplitudes were calculated by adding up spectral EFR peaks (ret ot he noise floor)a tt he modulation frequencya nd all available harmonics (inµ V).T he AM frequencyinthe psychoacoustic and EFR experiment were not identical butb oth greater than 80 Hz, consistent with brainstem generators of AM encoding [10].
Ac omputational model of the human auditory periphery wasa dopted [8] to simulate a7 0-dB, 4-kHz tone and 120-Hz AM tones of different modulation depths. Additionally,20different one-octave wide noise iterations with or without an embedded 4-kHz, 70-dB tone were averaged and simulated for arange of SNRs. Population responses were computed from simulating neural activity at the Inferior Colliculus (IC) stage of the model and by summing up time-waveforms across 401 simulated CFs spanning the human cochlear partition. Figure 1A shows an example IC population response to the 4-kHz pure tone and a 120-Hz AM tone. The psychoacoustic detection cue was derivedf rom the difference signal between the IC population response to the pure-tone and AM tone. Figure 1B shows the difference signal from IC population responses to anoise and atone embedded in noise at different SNRs. The rms of the difference signal wascomputed and transformed to dB to yield the detection cues plotted in Figures   to the scalp-recorded potential. Eight hearing profiles were simulated: (i) aN Hm odel with normal Q ERB sa nd 3l ow (1 spike/s; LS), 3medium (10spikes/s; MS)a nd 13 high (70s pikes/s, HS)s pontaneous rate AN fibers synapsing at each of the 401 inner-haircells (IHCs), (ii) as elective synaptopathymodel where all LS and MS fibers were removed(i.e., LS loss), (iii-iv) asynaptopathymodel where all LS&MS fibers as well as 50% or 75% of the HS fibers were removed( LS50%HS and LS75%HS). Lastly,( v)-(viii)H Im odels with Q ERB sc orresponding to ah ighfrequencys loping audiogram (above 1-kHz)u pt o3 5dB HL at 8kHz and synaptopathyprofiles as in (i)-(iv).

Psychoacoustics Figures 2A&B depict simulated psychoacoustic detection cues for the AM and TiNd etection experiment and
showt hat synaptopathyh as ag reater influence on shifting the NH curved ownward than ah igh-frequencys loping cochlear gain loss. In fact, the AM detection cue is somewhat stronger in the HI models for the same degree of synaptopathy. In the model, this is explained by  Figure 3. (Colour online) A:S imulated and recorded AM detection thresholds (70-dB, 4-kHz AM tone)a nd EFRs to 100% modulated 120-Hz 4-kHz tones [in dB re best NH EFR] for different hearing loss profiles. B:Simulated tone-in-noise detection thresholds (70-dB SPL tone)and EFRs to 100% modulated 120-Hz 4-kHz tones [in dB re best NH EFR] for different hearingloss profiles. Reference EFRs were recorded to 75-dB SPL, oneoctave wide, 4-kHz centered 100%, 120-Hz modulated white noises.
al ower effective drive to the IHC-AN complexc aused by cochlear gain loss, resulting in less saturated AN responses and enhanced AM sensitivity,c orroborating observations in the chinchilla AN [11]. The psychoacoustic threshold for the NH model wass et to the detection cue corresponding to the modulation depth at which NH people performed (i.e., -29.5 dB; 2 nd best human NH performer in Figure 3A). The gray threshold line in Figure 2A shows that the AM threshold shifted by 8dBand even by 15 dB for the LS50%HS and LS75%HS models respectively.S imilarly,F igure 2B predicts the need for a4 -dB stimulus SNR increase for the LS50%HS models to reach the reference NH detection cue amount and performance.
Figures 3A&B summarize the simulated detection cue shifts (black lines, filled markers)a nd EFR amplitude reductions (colored lines, filled markers)a longside human reference data (open markers)f or NH and HI participants who performed significantly worse on all measures (p<0.01). As the simulated detection cue shifts for the HI models were similar to those of the NH models, we conclude that synaptopathyrather than cochlear gain loss was responsible for the degraded detection cues. The range of simulated AM detection thresholds caused by synaptopathyand the spread in the reference data corresponded well. The best thresholds in the reference AM experiment were in line with those in [12]   age-related reductions between 5and 10 dB in the absence of cochlear gain loss [13,14]. Our simulations suggest that synaptopathycan explain alarge degree of the individual performance despite co-existing elevated hearing thresholds. The absence of ar elationship between the experimental 4-kHz pure-tone and AM detection thresholds supports this notion (R 2 =0.3; p=0.09). The ≈7dB spread in the TiNreference data corresponded well to the shift predicted by synaptopathya nd corroborate reported 5-10 dB TiNthreshold shift in afixed50-dB-SPLbroadband noise when more than 60% of the IHC population is lost [15]. However, in contrast to the AM thresholds, degraded TiN detection performance wasa lso related to elevated puretone thresholds (R 2 =0.3; p=0.02).

Relation between psychoacoustics and EFRs
Figure 3c ompares simulated and recorded EFR amplitudes re the reference NH EFR amplitude (blue and red). The spread of simulated and recorded EFR amplitude reductions around their mean coincide, although the HI EFRs showed overall greater reductions than predicted, suggesting that synaptopathyd i ff erences may only partly explain the individual spread in the human data. To test whether psychoacoustic metrics can be used as areplacement for EFRs in the diagnosis of synaptopathy, we studied their relationship. While AM detection and EFRs both rely on arobust coding of temporal envelope information, the sensitivity to small modulation depths (the psychoa-Vol. 104 (2018) coustic task)may not be apredictor of the EFR amplitude to 100% modulated stimuli. Regression fitsb etween individual EFR and psychoacoustic metrics (irrespective of their NH or HI status)a re depicted in Figure 4. The best simulated NH psychoacoustic threshold wasm atched to that of the best performing NH listener,w hile simulated EFR amplitudes were not scaled in the analysis. Simulated detection cues and EFR amplitudes related well (R 2 >0.9) and the fit wassomewhat better for the AM detection task due to stimulus similarity.I nt he model, the regression is generally predictive of the degree of synaptopathy( not cochlear gain loss). The experimental results showalarger spread around the regression line (R 2 of 0.3: TiNand 0.4: AM)than predicted by the model, butnevertheless showa significant relationship. The observed relation for the AM experiment extend the reported NH correlation between the EFR and AM detection [1] to HI listeners. The experimental relationship between TiNdetection SNR at threshold and the EFR amplitude has not been reported earlier, butits existence is supported by the model simulations.

Discussion
Both suprathreshold psychoacoustic tasks were strongly affected by synaptopathyand only mildly by cochlear gain loss for the considered stimulus configurations, suggesting that these tasks may differentially diagnose synaptopathy in NH and HI listeners. Even though simulations are inherently limited by the quality of the model (which does not account for plasticity or cognitive factors), we propose that the effect of synaptopathyonsuprathreshold psychoacoustsics is much greater than so farassumed. Signal detection theory predicts a1 .5-dB or 5-dB shift in the TiN detection threshold for arespective loss of 50% and 90% of the available AN fibers [2], whereas we observed that a 70% fiber loss (i.e., LS50%HS)i naf unctional model of the human auditory periphery causes at hreshold shift of 4dB. ForA Md etection, a7 0% or 85% (i.e., LS75%HS) fiber loss predicted ar espective 8a nd 15-dB threshold shift, which matched the individual variability in the combined NH and HI reference data well. Controversially,we propose that the reason whyt he HI listeners performed worse than the NH listeners, wasdue to their synaptopathy and AN fiberloss and not because of their coexisting outer haircell loss deficits. This latter aspect can be confirmed experimentally,asage-related synaptopathywas shown to occur before outer haircell loss [16]. If our predictions are correct, ageing listeners with normal audiometric thresholds suffering from synaptopathyshould showEFRs, TiN and AM detection thresholds in range with those of HI participants.