Parallel EEG assessment of different sound predictability levels in tinnitus

,


Introduction
Predicting the type and timing of upcoming events guides goaldirected behavior and is key to efficiently adapting to changes in an abundant sensory environment.In audition, predictions can take different arrangements.They can either be based on formal (type) or temporal (timing) stimulus characteristics (Bendixen et al., 2012;Mauk and Buonomano, 2004;Tavano et al., 2014).Formal predictions refer to the spectral information that is conveyed in an acoustic stimulus (Schwartze et al., 2012), while temporal predictions refer to the points in time when a stimulus event occurs (Schwartze et al., 2012).For example, 'what' type stimuli can be expected because of stimulus characteristics (e.g., tones with different frequencies), and be referred to as the formal dimension of predictability, while 'when' the next stimulus can be expected, refers to the temporal dimension.Next to when in time an event can be expected, temporal predictions may also concern predictions about the duration of an event (Coull et al., 2013), which was not explored in the current study.Nested in this basic distinction, predictions can be based on specific stimulus arrangements such as the repetitive binary stimulus grouping that is commonly used in 'sensory gating' (SG) studies to induce position predictions.
The fundamental importance of predictions is particularly evident when their underlying mechanisms change in maladaptive conditions.One condition in which altered auditory predictions seem to play a role is tinnitus (Brinkmann et al., 2021a;De Ridder et al., 2014;Hullfish et al., 2019;Roberts et al., 2013;Sedley et al., 2016).Tinnitus is typically described as 'ringing in the ears' and often experienced as a constant sound in the absence of a physical sound source (Axelsson and Ringdahl, 1989;Roberts et al., 2010).The most prominent risk factors for developing tinnitus are aging and hearing loss (Roberts et al., 2010).
In the general population, its prevalence ranges from 10 to 14 % in middle-aged adults and further increases with age (Jarach et al., 2022;Langguth et al., 2013).Persons with tinnitus experience a 'ringing in the ears' unlike persons without tinnitus.Persons with tinnitus can be further differentiated into those with decompensated or compensated tinnitus.The first group suffers from tinnitus while the second one is not substantially affected by it.
Peripheral tinnitus models propose that tinnitus might stem from aberrant cochlear activity (Mulders and Robertson, 2009;Sedley et al., 2016).However, tinnitus might persist even when input from the ear is removed by cutting the auditory nerve (see for example: House and Brackman, 1981).Subcortical models, on the other hand, suggest that tinnitus might be induced by increased spontaneous activity in the auditory system, also involving limbic areas (i.e., frontostriatal gating) (Leaver et al., 2011;Rauschecker et al., 2010;2015).Linked to the subcortical models is another approach that suggests that the discrepancy between the expected and the actual auditory input might activate subcortical modulations via auditory attention, which might ultimately contribute to the generation of tinnitus (Roberts et al., 2013).Taking a predictive coding perspective, chronic tinnitus might display altered default predictions, meaning that in chronic tinnitus, default predictions in audition change to represent 'something' instead of 'silence' (Hullfish et al., 2019;Sedley et al., 2019).Although the exact interplay of these factors is unknown, altered prediction mechanisms could lead to auditory phantom perception (Hullfish et al., 2019).As tinnitus is typically perceived as a tone with a constant pitch, it is assumed that formal predictions are most affected (Sedley et al., 2019).Additionally, altered temporal predictions in tinnitus have been suggested in thalamocortical dysrhythmia (De Ridder et al., 2015) and were subsequently discussed in terms of a predictive network hypothesis (Brinkmann et al., 2021a).The predictive network hypothesis focuses on the role of the auditory thalamus (i.e., the medial geniculate body (MGB)) in tinnitus and illustrates connectivity alterations between the MGB and other brain areas (i.e., bilateral auditory cortex, basal ganglia, frontal cortices, cerebellum).In tinnitus, the different firing modes of the MGB (i.e., burst vs tonic firing mode) might contribute to the altered processing of temporal information that travels from the MGB to the basal ganglia, frontal cortex, and bilateral auditory cortex (Brinkmann et al., 2021a).Furthermore, it is suggested that these subcortical changes influence not only temporal predictions, but also other forms of predictions, such as SG.SG reflects basic pre-attentive filtering of sensory input and can be assessed by investigating position predictions in audition.It should be noted that there are also behavioral measures to assess SG (Mohebbi et al., 2019b).Mohebbi et al. (2019b) showed altered behavioral SG in persons suffering from decompensated tinnitus, compared to persons with compensated tinnitus.Roberts et al. (2013) suggested that altered predictions might activate attentional processes in audition that contribute to the emergence of tinnitus.Indeed, prior research reported maladaptive (selective) auditory attention in tinnitus (Husain et al., 2015;Sharma et al., 2023;Sherlock and Brungart, 2021;Vasudevan et al., 2021), which could be considered as a deficit of executive control of attention (Heeren et al., 2014), but see Jensen et al. (2021).It was also suggested that maladaptive selective attention might contribute to the persistence of tinnitus (Tavanai and Mohammadkhani, 2018).Therefore, based on previous research and the described maladaptive (selective) attentional processing in persons with tinnitus, it was hypothesized that neurophysiological measures of SG should also be altered in tinnitus.However, it remains unclear whether other dimensions of auditory prediction are also affected.Accordingly, distinguishing dimensions of auditory prediction (i.e., formal-, temporal-, and position-predictions) combined with a differential assessment of their function in tinnitus might lead to a better and more comprehensive understanding of tinnitus beyond the level of formal predictions alone.
The high temporal resolution of the electroencephalogram (EEG) provides an excellent tool for investigating auditory predictions.The P50 and N100 event-related potential components (ERPs) peak around 50 and 100 ms post-stimulus, respectively, and are responsive to prediction.In neurophysiological research, SG is typically investigated by presenting pairs of identical sound stimuli (Adler et al., 1982).The response to the second stimulus leads to a suppressed P50 ERP response (i.e., 'gating out'), and indicates that the first stimulus is predictive of the second one (Adler et al., 1982;Cromwell et al., 2008).The P50 is generated in temporal and frontal cortices and interpreted as an indicator of SG (Korzyukov et al., 2007;Smith et al., 1994).Classically, P50 SG is conceived as a pre-attentional mechanism and mainly reflecting sensory processes (Jerger et al., 1992;Kho et al., 2003).However, the contribution of frontal cortices and the hippocampus to the generation of the P50 indicates the involvement of multiple lateral and medial cortical areas, potentially suggesting that SG might be a multistep process (Grunwald et al., 2003;Weisser et al., 2001).Modulation of the P50 response in SG might thus indicate the relative success of filtering out goal-irrelevant information (Jones et al., 2016).Accordingly, stronger P50 suppression is associated with better attentional orienting and inhibition (Wan et al., 2008).
The N100 is generated in the supratemporal plane of the auditory cortex (Näätänen and Picton, 1987) but also in the frontal cortex (Giard et al., 1994).The N100 reliably indicates formal predictions as assessed in pitch-based 'oddball' paradigms (Segalowitz and Barnes, 1993) and stands more for (selective) attentional processes (Thornton et al., 2007).Previous N100 research, that manipulated formal and temporal stimulus predictability, found differences between temporal and formal conditions (Schwartze et al., 2013).The N100 response to predictable stimuli becomes smaller over time and the interval between the stimuli is a factor that determines this decrease (Budd et al., 1998).In tinnitus research the N100 is a frequently investigated ERP component (Azevedo et al., 2020).Persons with tinnitus express smaller N100 amplitudes than persons without tinnitus, while latency differences are not always reported (e.g., Jacobson and McCaslin, 2003).Other research suggests that observed N100 changes might mirror a reduction of tinnitus symptoms after three weeks of customized sound therapy (Pineda et al., 2008).Moreover, frequency dependent changes for tones outside the tinnitus frequency range were observed in the N100 and show that N100 changes are not dependent on the individual tinnitus frequency of persons with tinnitus (Kadner et al., 2002).Based on the mentioned evidence and on the systematic review stating that latency and amplitude changes are commonly observed in tinnitus (Azevedo et al., 2020), the N100 serves as a comprehensive measure of tinnitus.Taken together, the existing evidence suggests that the P50 and N100 might indicate different stages of predictive sensory filtering, with successful SG leading to better task performance and protected higher-order cognitive functioning (Lijffijt et al., 2009;Venables, 1964).
Neurophysiological SG studies in adults with tinnitus have produced inconclusive results (Campbell et al., 2018;Dornhoffer et al., 2006).For example, it has been shown that the SG difference index of the Pa component that precedes the P50, correlates negatively with tinnitus severity, i.e., more severe tinnitus reduced the Pa suppression in response to the second tone in a tone pair (Campbell et al., 2018).Here tinnitus severity was assessed by using the tinnitus handicap inventory (THI) (Newman et al., 1996).However, the scores represented in the group of Campbell et al. (2018) only experienced mild tinnitus symptoms, which would refer to compensated tinnitus.Notably, no significant differences were observed between persons with tinnitus and those without for the Pa, P50, N100 or P200 gating effects (Campbell et al., 2018).Similarly, a study investigating P50 suppression in persons with and without tinnitus did not report group differences (Dornhoffer et al., 2006).Another study investigated habituation to repetitive auditory input and reported habituation to be dependent on tinnitus severity (Walpurger et al., 2003).In other words, persons with decompensated tinnitus (i.e., increased severity), displayed reduced stimulus habituation of the N100 and P200 amplitude differences compared to persons with compensated tinnitus (Walpurger et al., 2003) P. Brinkmann et al.Lastly, Sedley et al. (2019) manipulated formal predictability in a roving standard oddball paradigm and observed no P50 differences between persons with or without tinnitus when comparing their responses to standard and deviant tones, while deviant tones evoked larger N100 responses in both groups.To conclude, previous evidence reported reduced N100 -P200 amplitude differences during continuous repetitive stimulus presentation in persons with decompensated tinnitus (Walpurger et al., 2003) and that impaired Pa suppression might be linked to increased tinnitus severity (Campbell et al., 2018).However, for P50 SG no group differences were found (Dornhoffer et al., 2006;Sedley et al., 2019).Tinnitus was previously investigated by responses to stimulus sequences that incorporated manipulations of either formal or position predictions in isolation (Campbell et al., 2018;Dornhoffer et al., 2006;Sedley et al., 2019).However, experimental paradigms that allow the parallel assessment of different stimulus type dimensions and timing are needed to obtain a better understanding of how prediction influences the heterogeneity of tinnitus.Please note that tinnitus heterogeneity is defined as non-uniformity of at least one of four dimensions that encompass tinnitus perception, risk factors/comorbidities, distress, and/or treatment response (Cederroth et al., 2019).
Such a comprehensive approach could allow differentiating individual prediction capacities and identify which dimensions are dysfunctional in tinnitus.As SG is a filtering mechanism that might operate on all predictability dimensions, it also allows exploring possible interactions between temporal-, formal-and position-based SG.Finally, linking this approach to specific ERP markers, such as P50 and N100, might allow decomposing the underlying mechanisms of selective attention and inform how they look in persons suffering from tinnitus.
Along these lines, the current study assessed if and how ERP markers of auditory predictions are altered in persons with and without tinnitus.The experimental setup of the current study combined elements of previous studies (Schwartze et al., 2013;2011).This setup incorporated a paired-tone oddball design that simultaneously manipulated formal and temporal stimulus features to differentiate and relate different aspects of predictability.The aim of the current study was to assess different levels of auditory predictability (i.e., temporal-, formal-and position predictions) in persons with and without tinnitus to uncover potentially altered SG efficiency in the tinnitus group.Therefore, it was hypothesized that successful SG for temporal-, formal-and position predictions would result in smaller P50 and N100 amplitudes for predictable stimuli, but that in tinnitus, due to altered SG efficiency, increased P50 and N100 amplitudes should be observed.

Methods
The study was approved by the ethics committee of the Maastricht University Medical Center + (MUMC+) with the code 2019-0970.Due to the COVID pandemic, data collection was paused and then continued intermittently under strict safety regulations.

Recruitment and inclusion
Participants were recruited via leaflets, word of mouth, and an existing database of persons with tinnitus.Persons with and without (subjective) tinnitus were included when they were between 18 and 69 years old and had an audiogram and a bilateral high tone Fletcher Index lower than 60 dB.The Fletcher Index was calculated as the average hearing levels of for 1, 2, and 4 kHz for both ears (Wang et al., 2018).Exclusion criteria were objective tinnitus (i.e., pulsatile tinnitus), a maximum air-bone gap of more than 20 dB, or a history of ear surgery, brain surgery or brain/ear implants.If available, participants provided their audiograms, all but seven obtained within the last year, otherwise an audiogram was obtained by trained personnel before testing.The two groups were matched for sex at the participant level and for age and education at the group level.

Participants
Fifty-two persons with tinnitus and without participated (Table 1).Education was scored on 8 levels in persons with and without tinnitus, where 8 was the highest level.Hearing loss (HL) was assessed using the average pure tone audiometry (PTA) in dB for the left and right ears for persons with tinnitus and without.A more detailed description of the audiograms is presented in the Supplementary Figure 1.Tinnitus burden was assessed with the Dutch version of the Tinnitus Questionnaire (TQ) (Goebel and Hiller, 1994;Meeus et al., 2007) in the tinnitus group (M TQ = 37.3, SD TQ = 17.6), scores between 31 and 46 points indicate mild tinnitus burden (Grade II).Hence, persons with tinnitus can be classified as persons with compensated tinnitus (TQ < 47) and decompensated tinnitus (TQ ≥ 47).For the current tinnitus group, n = 10 persons experienced decompensated tinnitus, while the correlation between TQ and bilateral hearing loss was non-significant (r = 0.25, p = .22).Tinnitus duration was assessed in months (M TinDuration = 82.8,SD TinDuration = 76.4).The tinnitus group suffered from chronic tinnitus, considering that tinnitus is 'chronic' if it is experienced for at least three months (Snow, 2004).

Procedure
Upon arrival at the laboratory, participants signed the informed consent, answered questions about their tinnitus (i.e., tinnitus duration etc.), and filled in the TQ.They then entered an electronically shielded and soundproof booth for the EEG recordings.Before the start of the experiment, each participant listened to a short excerpt of each tone condition.The volume was adjusted according to the perceived binaural loudness and if necessary adapted to the perceived loudness for each ear in each participant.

Experimental design and stimuli
The two stimulus sequences each consisted of 1152 standard (600 Hz) and 288 deviant (660 Hz) tones (50 ms duration including 5 ms rise and fall times) corresponding to a 4:1 standard to deviant ratio.The total duration of each tone sequence was 12 min and multiple checks were in place to ensure that participants stayed alert during that time.For example, participants were given a short break after 6 min and were continuously monitored during the EEG session by checking their activity levels via video and online EEG monitoring.Alternating short and long intervals between tones ensured that the sequences resembled typical paired stimulus SG paradigms.Predictability regarding formal tone characteristics (frequency) was manipulated in terms of the standard to deviant ratio over the whole sequence, while the frequency of the standard and deviant tones was fixed.The resulting sequence consisted of the following tone pair combinations: standard-standard, deviant-standard, and standard-deviant (Fig. 1).At the beginning of each sequence, 3 standard-standard tone pairs were presented and there were no two deviants in a row.The intervals between the tones of a pair (intra-pair-interval, intra-PI) were 200 ms in the fully predictable isochronous sequence and between 100 and 300 ms in the random sequence.The intervals between pairs (inter-pair-interval, inter-PI) were 700 ms in the isochronous sequence and between 350 and 1050 ms in the random sequence.The random sequence was designed so that participants still perceived the tones in pairs, but the intra-PIs (100 ms, 150 ms, 200 ms, 250 ms, 300 ms) and inter-PIs (350 ms, 525 ms, 700 ms, 875 ms, 1050 ms) varied.The order of these time intervals was randomized using a Williams design (Williams, 1949), thereby ensuring that each interval occurred equally frequently and that each interval was preceded by each other interval equally.

EEG recording and pre-processing
EEG was recorded from 128 active electrodes (actiCAP, Brain Products GmbH), mounted into an elastic cap at 1000 Hz sampling rate, while impedances were kept at ≤10 kOhm.FCz was used as the online reference and the audio signal was recorded with the EEG.Data were then downsampled to 500 Hz and a bandpass filter (1-44 Hz) was applied using EEGlab (Delorme and Makeig, 2004).To detect and reject bad channels clean_rawdata was used (Mullen et al., 2015).Rejected channels (M = 5.5, SD = 4.9) were spherically interpolated, then the online reference was added back to the data and data were re-referenced to the average (following Foti et al., 2009).Artifact subspace reconstruction (ASR) was performed using clean_rawdata and data were re-referenced again as suggested by Miyakoshi (2022) to reorganize the data to be zero-sum across channels.Then ICA (runica; 30 pca components) was performed (Makeig et al., 1997).IC components (M = 4.3, SD = 1.2) reflecting horizontal and vertical eye movements, muscle activity, heart rate, line noise or channel noise were rejected using IClabel (Pion-Tonachini et al., 2019).

ERP analysis
Epochs lasting from − 50 ms to 145 ms relative to stimulus onset were created, baseline corrected (− 50 to 0 ms), and then averaged per participant.To avoid bias or double dipping about time and spatial distribution of the ERP components (Kriegeskorte2009; Luck and Gaspelin, 2017; Miyakoshi 2022), analyses followed a data-driven approach.A two-step temporal-spatial PCA (tsPCA) was performed using the EP toolkit (version 2.95) (Dien, 2010;2012).This procedure decomposes the data based on covariances between voltages at sampling points and sampling sites and aims to identify and disentangle components that are transparently and objectively extracted (Dien and Frishkoff, 2005).Following the guidelines formulated by Dien (2012), first, a temporal PCA was performed on the averaged data, using participants, stimulus types, and recording sites as observations.A (oblique) Promax rotation was used (Hendrickson and White, 1964) and seven factors were extracted after inspection of the scree plot (Cattell, 1966) with the help of a parallel test that compares the scree plot obtained with the experimental data with one that is derived from random data (Horn, 1965).Second, a spatial (orthogonal) Infomax ICA was performed on each temporal component that survived the first step, and seven spatial components were extracted (Bell and Sejnowski, 1995;Delorme and Makeig, 2004).Please consult the Supplementary Fig. 2. for a depiction of the elicited ERPs.

Statistical analysis
For the demographics, independent t-tests were performed.A chisquare test was performed for sex differences and the robust counterpart of the t-test as implemented in the WRS2 (version 1.1-4) package (Mair and Wilcox, 2020) was used when assumptions were violated.All analyses were performed in R (version 4.2.0) using Rstudio (version 2022.07.1).For the EEG data, after inspection of the time course and the spatial distribution of the components of interest, the microvolt scaled amplitudes of the max and min peaks were analyzed for two time windows (i.e., TF1SF1: 90-130 ms, TF3SF1: 30-70 ms; with TF denoting temporal factor and SF spatial factor).Microvolt scaled reconstruction was chosen to simplify subsequent interpretation and the time-windows enhanced the reliability of the component activity peaks.The component-wise reconstructed microvolt scaled tsPCA scores were exported using the ERP PCA Toolkit (Dien, 2010) and used for statistical analysis.For the figures, we performed the reconstruction by first multiplying the variance with the factor loadings for the spatial PCA and the temporal PCA.Then, we multiplied them by the factor scores.Subsequently, 2 (without Tinnitus vs. Tinnitus) x 2 (Isochronous vs. Random) x 2 (Standard vs. Deviant) x 2 (Position 1 vs Position 2) mixed ANOVAs were performed, separately for TF1SF1 and TF3SF1 using the rstatix (version 0.7.0) package (Kassambara, 2021).
When checking the assumptions, some outliers were identified and two extreme outliers excluded (i.e., exceeding the interquartile range by a threefold), as they were outliers for several combinations of factors.Two participants without tinnitus were correspondingly excluded.Levene's tests were not significant and normality was assumed based on the central limit theorem.All effects are reported as significant at p < .05.Effect sizes are reported as generalized eta squared (η 2 G ). Non-normally distributed variables such as the Hearing loss (HL) and duration of tinnitus in months underwent square root transformation (see Supplementary material for further correlation analyses).Correlations between the SG difference values of the P50-like and N100-like activity with HL and duration of tinnitus were explored (see Supplementary Figs. 4. & 5. for scatter plots).

N100-like activity (TF1SF1)
There were significant main effects of temporal structure (i.e., isochronous versus random) indicating more negative values for the 11.17, p = .002,η 2 G = .008.For the temporal structure x position interaction, post-hoc tests showed significant differences between the

Discussion
This study investigated modulations of the P50-and N100-like activity in response to three different dimensions of stimulus predictability in persons with and without tinnitus.Manipulations of predictability altered the formal and temporal structure as well as the position of tones presented in pairs.This binary grouping made the setup comparable to classical binary SG paradigms.P50-like activity indicated expected effects of formal structure, temporal structure, and position in both groups.In other words, P50 amplitudes were smaller for standard than deviant tones, smaller for isochronous than random stimulus timing, and smaller in response to the second than the first tone of tone pairs, confirming a classical SG effect in both groups.These results confirm the effectiveness of a paradigm to simultaneously assess formal-, temporaland position predictability.Additionally, there was a significant difference between the temporal conditions for the second tone in pairs, while this was not the case for the first tone.This again indicates a stronger amplitude reduction in the isochronous than the random condition in both groups.This overall pattern repeated in the N100-like activity, while globally, amplitudes for the isochronous sequence were more negative than for the random sequence.However, amplitudes in N100like activity were larger for deviant than for standard tones in the isochronous timing condition in persons without tinnitus only.Previous research reported that participants respond faster when stimuli are presented in a temporally regular than irregular context (Lange, 2009;Rohenkohl et al., 2012) and that temporal predictability facilitates stimulus detection (Lawrance et al., 2014).At the neurophysiological level, the suppression of early ERP responses to expected stimuli is a well-established phenomenon (Bendixen et al., 2012;Costa-Faidella et al., 2011;Lange, 2009;Schwartze et al., 2013).Therefore, the current findings for temporal predictability are in line with these previous findings.
However, the current study did not reveal differences for position predictions (SG) between persons with and without tinnitus in P50-and N100-like activity.Previously, Campbell et al. (2018) found a correlation for tinnitus severity, suggesting decreased gating in the Pa component when tinnitus burden increased.However, similar to the Fig. 3. Left: Temporal-spatial factor reflecting N100-like activity for persons without tinnitus (top) and persons with tinnitus (bottom).Right: Topographical depiction of the activity of the temporal factor (TF) 1 in combination with the spatial factor (SF) 1 (i.e., TF1SF1) for all conditions and stimuli, and both groups.Abbreviations: Sta: standard, Dev: Deviant, ISO: isochronous condition, RAN: random condition.current results, they reported no group differences in SG for the Pa, P50, N100 or P200 components (Campbell et al., 2018).Similar P50 results were reported by Dornhoffer et al. (2006).Compared to the current study, participants in the Campbell et al. (2018) study had similar hearing thresholds but were younger (i.e., on average between 20 and 22 years) and their tinnitus handicap was very low (i.e., 0 -14).A tinnitus handicap inventory (THI) score can range between 0 -100 and scores between 0 -16 indicate no or only a slight handicap (Lee et al., 2014;Newman et al., 1996).Here, we administered the TQ and not the THI, even though both instruments have shown high convergent validity (Baguley et al., 2000).In addition, the paradigms differed in terms of the intra-PI (500 ms) and the inter-PI (7 s) (Campbell et al., 2018).As the  current paradigm used shorter intra-PIs and inter-PIs and to avoid double dipping, we applied tsPCA, a data-driven analysis that can delineate overlapping processes and can therefore enhance the signal-to-noise ratio (Dien, 2012;Foti et al., 2009).Moreover, in the current study, N100-like activity was frontally located, which might be explained by age in the current participant sample (Paitel and Nielson, 2021).Paitel and Nielson (2021) observed frontal temporal PCA components for the N200 and the P300 in the older but not in the younger group.Although Paitel and Nielson (2021) investigated later ERPs assessing successful inhibitory control, the current results might suggest more frontal recruitment for slightly earlier components as well.In comparison, in Dornhoffer et al. (2006), the age range was similar to the current study, while another tinnitus severity questionnaire (i.e., tinnitus severity index questionnaire) was administered.Dornhoffer et al. (2006) applied different intra-PIs (i.e., 250 ms, 500 ms, 1000 ms) and no group differences for P50 SG were observed.Overall, the current study thus produced similar results as Campbell et al. (2018) and Dornhoffer et al. (2006), despite methodological differences between the studies.However, contradictory to classical SG studies, we observed more negative amplitudes for S2 compared to S1 for the N100-like activity where one would expect a less negative S2 amplitude compared to S1 (Lijffijt et al., 2009;Rentzsch et al., 2008a;Rentzsch et al., 2008b).Underlying reasons could stem from the current study design and the short inter-PIs applied.It was shown that N100 amplitudes were slightly more negative for shorter inter-PIs (i.e., 8 s vs 2 s), but the amplitudes for S2 were consistently smaller than for S1 (Rentzsch et al., 2008a).Applying a data-driven analysis method such as tsPCA can disentangle latent ERPs, while increasing the inter-PI interval may reverse the observed SG effect in the N100 in future studies.
In the current study, we showed that for the P50-like activity, predictions regarding formal structure and position are largely intact in persons with and without tinnitus in a temporally predictable context, indicating intact binary auditory stimuli filter mechanisms in persons with tinnitus.However, the suppression pattern was reversed for the N100-like activity for persons with and without tinnitus.This partly confirms the robustness of the results, suggesting that SG dysfunctions (i.e., as assessed by the P50) in tinnitus are likely more subtle than assumed.
Another point to consider is that the paradigm used tones of 600 and 660 Hz, whereas tinnitus tones often arise in higher frequency ranges (i.e., 3000-7000 Hz) (Neff et al., 2019).The rationale underlying the choice of these specific tones was that the corresponding frequency range is generally not much affected by age-related hearing loss that might influence sound processing (König et al., 2006).Further, we aimed to maintain comparability of the current findings with our own prior results obtained in healthy older adults tested with the same standard and deviant pitches (Brinkmann et al., 2021b).There is unfortunately only very limited research on sensory gating in persons experiencing tinnitus.Therefore, considering the possibility of frequency-dependent sensory gating in tinnitus is challenging.Campbell et al. (2018) investigated sensory gating in persons with mild tinnitus using even lower tone frequency (i.e., 250 Hz).Their results indicated intact sensory gating in persons with and without tinnitus.Other research used broadband clicks that contain a wide range of frequencies and observed decreased sensory gating in persons with tinnitus (Morse and Vander Werff, 2024).These findings suggest that sensory gating might at least in part depend on the usage of different frequency stimuli.However, the tinnitus severity levels varied between studies.Therefore, it is possible that next to the respective stimuli, tinnitus severity contributed to gating capacities.Moreover, the tinnitus frequency varies between affected individuals.Thus, to test if sensory gating is indeed frequency-dependent, a future goal would be to match stimuli taking tinnitus frequency into account.
For the N100-like activity, however, a different pattern emerged for deviance processing in the two timing conditions in persons with and without tinnitus.This observation can be linked to different attentional processing for deviant events in persons with and without tinnitus.Altered attention in persons with tinnitus has previously been proposed by Roberts et al. (2013).Roberts et al. (2013) proposed a qualitative model, in which attention allocation changes following a mismatch between the incoming auditory input and the sound representation generated in the auditory cortex.Other research on cognitive performance and tinnitus indicates that tinnitus is linked to decreased auditory attention, executive function, processing speed, general short-term and working memory, and learning and retrieval (Clarke et al., 2020;Husain et al., 2015;Sharma et al., 2023;Tavanai and Mohammadkhani, 2018).More specifically, it has been suggested that persons with tinnitus exhibit more specific changes in the executive control network of attention, when assessed with the Attention Network Test (ANT), which was also correlated to tinnitus duration (Heeren et al., 2014).However, other research investigating persons with severe tinnitus reported contradicting results (Jensen et al., 2021).How tinnitus severity is classified is relevant and both studies recruited persons with chronic decompensated tinnitus.Therefore, the conflicting results are likely not explained by tinnitus severity.While a link between tinnitus severity and psychological distress in older persons has been found, the heterogeneity of the available studies calls for further evidence to establish a link between tinnitus, emotional wellbeing and cognitive capacities in older adults (Malesci et al., 2021).A meta-analysis investigated behavioral and electrophysiological measures of attention in persons  with tinnitus and confirmed that later attentional processes are altered in persons with tinnitus as indicated by reduced mismatch negativity (MMN) and P300 amplitudes, although the heterogeneity of the tinnitus populations and methods limit precise interpretations of the underlying mechanisms (Vasudevan et al., 2021).Another study assessing MMN activity for frequency deviants found higher amplitudes for persons with decompensated tinnitus in comparison to persons with compensated tinnitus or control participants (Mohebbi, et al., 2019a).This underlines the importance of tinnitus severity in tinnitus investigations.
When assessing the N100 component in persons with tinnitus, reduced N100 amplitudes were observed for persons with decompensated tinnitus (i.e., higher severity) (Jacobson and McCaslin, 2003).Moreover, persons with low and high levels of tinnitus severity, as characterized by tinnitus (de)compensation, showed differences in N100 activity (Delb et al., 2008).More specifically, in an unattended condition, in which participants had to ignore tones and think of something pleasant, persons with decompensated tinnitus showed a more negative N100 response compared to persons with compensated tinnitus.Moreover, more negative N100 response was also observed in persons with decompensated tinnitus compared to persons without tinnitus in the attended condition.Here, N100 amplitudes of persons with decompensated tinnitus were more negative than those of persons without tinnitus (Delb et al., 2008).Other research on auditory attention in tinnitus (Roberts et al., 2013), suggests facilitatory cholinergic neuromodulation in cortico-subcortical pathways at the level of the ventromedial prefrontal cortex and basal forebrain, which could reinforce aberrant neural synchrony in persons with tinnitus.Interestingly, it was reported that activity in fronto-parietal regions differs in persons with tinnitus when they are presented with tinnitus-specific frequency sounds as opposed to a control frequency (Salvari et al., 2023).These findings may indicate why a more efficient adaptation to deviant tones was observed in persons without tinnitus than in persons with tinnitus and suggests altered auditory attention allocation in response to deviant tones in the tinnitus group.Along these lines, the current findings may indicate a shifting of attentional bias toward the tinnitus percept or altered redirection of selective auditory attention away from it.Future research should therefore delineate how auditory predictions are influenced by tinnitus-specific frequencies (i.e., regularly 2-4 kHz in tinnitus linked to noise induced hearing loss (Eggermont and Roberts, 2004) for the P50-and N100-like components, respectively.
Lastly, when working with a clinical population such as persons with tinnitus, controlling for hearing loss, age, and/or hyperacusis at the same time is challenging.Tinnitus is more prevalent in older persons and often accompanied by some degree of hearing loss (Axelsson and Ringdahl, 1989;Eggermont and Roberts, 2004;Knipper et al., 2013;Nelson and Chen, 2004), although it can also occur without altered hearing thresholds (Roberts et al., 2010;Weisz et al., 2006).Therefore, another limitation of the study was that hearing loss differed between groups.The average hearing loss in the current tinnitus sample was mild and associated with no impairment or slight difficulties for everyday functioning (Olusanya et al., 2019).To ensure that each individual perceived the stimuli loud enough and equally loud in both ears, we adjusted the volume of the stimuli for each participant if necessary.However, individual thresholds for stimulus loudness perception based on hearing acuity were not obtained and therefore are considered as a limitation.Further, research that includes persons with tinnitus with and without hearing loss and comparisons to hearing-matched persons without tinnitus remains scarce at this stage.The relatively small sample size might add to this limitation and reduced overall power.Considering the heterogeneity of tinnitus and the presence of comorbidities that might influence results, future studies should consider these aspects in their study design/ a-priori power estimations.
The results show that the adopted paradigm permits assessing three dimensions of auditory prediction (i.e., formal-, temporal-, and position).Classic position-based SG was not altered in persons with and without tinnitus for the P50-like activity.However, for the N100-like activity, deviance processing was further modulated by temporal regularity only in persons without tinnitus but not in persons with tinnitus.Hence, it seems likely that temporally regular and thus fully predictable stimulation facilitates deviance processing in persons without tinnitus and that this mechanism is altered in persons with tinnitus.Auditory filtering as indexed by classical SG effects for binary auditory stimuli seems thus not substantially different in persons with and without tinnitus.It rather seems that tinnitus alters attention-allocation in response to the deviant, i.e., unpredicted or at least less predictable auditory events.This finding might indicate a shifting attentional bias towards the tinnitus sound that may be accompanied by dysfunctional allocation of selective auditory attention to other sounds.Linking the current findings to existing models of tinnitus might suggest that underlying predictive auditory processing is altered in tinnitus (as suggested by Roberts et al. (2013), Hullfish et al. (2019) or by the predictive network hypothesis (Brinkmann et al., 2021a).As we neither assessed limbic areas, nor thalamocortical coupling, we cannot link the current results directly to the noise cancellation theory (Rauschecker et al., 2010) or the thalamocortical dysthymia hypothesis (De Ridder et al., 2015).However, classical SG (i.e.position predictions) seems to be unaffected in tinnitus, which would be in line with the model of Roberts et al. (2013).Lastly, individual differences in executive control and tinnitus heterogeneity might have influenced the results.

Fig. 1 .
Fig. 1.Schematic representation of the stimulus sequences.Pairs of sinusoidal tones were continuously presented in two sequences, one in which the intervals between tones and pairs were fixed (isochronous pairs) and one in which both intervals varied in duration with the constraint that shorter and longer intervals alternated (random pairs).Black bars represent standard tones (S), gray ones deviant tones (D).Standard and deviant tones differed in pitch and were presented with an overall 4:1 ratio.

Fig. 2 .
Fig. 2. Overview of the temporal factors.Microvolt scaled temporal factors based on temporal-spatial principal component analysis of ERP data.

Fig. 4 .
Fig. 4. Left: Temporal-spatial factor reflecting P50-like activity for persons without tinnitus (top) and persons with tinnitus (bottom).Right: Topographical depiction of the activity of the temporal factor (TF) 3 in combination with the spatial factor (SF) 1 (i.e., TF3SF1) for all conditions and stimuli, and both groups.Abbreviations: Sta: standard, Dev: Deviant, ISO: isochronous condition, RAN: random condition.

Fig. 5 .
Fig. 5. Depiction of the peak amplitudes for the P50-like activity.Peak amplitudes in microvolts for the third temporal and first spatial factor (TF3SF1) for the temporal structure x formal structure (left) and temporal structure x position (right) interaction.Abbreviations: Sta: standard, Dev: deviant, ISO: isochronous condition, RAN: random condition, *** p ≤ 0.001.

Fig. 7 .
Fig. 7. Depiction of the peak amplitudes for the group differences for the N100-like activity.Peak amplitude differences in microvolt for the first temporal and spatial factor (TF1SF1) for the simple pairwise comparisons.Abbreviations: ISO: isochronous condition, RAN: random condition, C: control Group/persons without tinnitus, T: persons with tinnitus, *** p ≤ 0.001.

Table 1
Demographics of the study participants.