The study was approved by the regional ethics committee in Stockholm, Regionala etikprövningsnämnden (Dnr:2019-05226). All subjects gave written informed consent after being informed about the aim and scope as well as risks of the study. The experiment was conducted in strict compliance with the ethical principles outlined in the Declaration of Helsinki.

Demographics of the participants is shown in Table 1. This MEG study was designed to identify GPIAS parameters (interstimulus interval (ISI; 0, 60, 120, 240 ms), carrier (60 and 70 dB SPL), and pulse levels (from 70 to 95 dB SPL by steps of 5 dB)) that would achieve an optimal inhibition of the pulse response in presence of gaps, and compare the cortical evoked responses with the collected EOG. Normal hearing participants (n=26) were recruited via an online platform (accindi.se) for audiological assessment, MEG, and structural magnetic resonnance imaging (MRI). Exclusion criteria were pregnancy, sensitivity to sound (hyperacusis), psychiatric disorders, drug use, neurological disease, and non-removable metal implants. Sex was self-reported and well-balanced (13 females, 12 males and 1 other). Participants had a mean age of 28.4 (SD ± 5.8) and 88% of them had right handedness. Upon arrival at the national facility for magnetoencephalography (NatMEG) where the measures were performed in a single session, participants received detailed information on the procedure, signed an informed consent, confirmed they had no tinnitus, experienced no hearing difficulties, and filled in an on-line survey to assess stress, anxiety, depression and hyperacusis prior their auditory assessment (see section below). After successful MEG recording (total duration of 2.5–3 h), each participant was scheduled for a structural MRI scan (total duration 1 h) using a 3 Tesla GE MR750 Discovery at the MR Center, Karolinska Institutet. After completing both MEG and MRI sessions, all participants were compensated with 300 SEK for their time.

Swedish versions [46] of the Hyperacusis Questionnaire (HQ) [35], Perceived Stress Questionnaire (PSQ-30) [39], and the Hospital Anxiety and Depression Scale (HADS) [1] were delivered online (Sunet Artologik). Hearing thresholds were evaluated by high frequency Békésy-audiometry (0.125–16 kHz) with cut-offs for hearing loss defined at 25 dB HL following WHO guidelines (Astera2, Otometrics. Inc and HDA200 headphones, Sennheiser). Loudness discomfort levels (LDL) were tested in ascending steps of 5 dB at frequencies 0.5, 1, 2, and 4 kHz (Astera2, Otometrics. Inc and HDA200 headphones, Sennheiser).

Before MEG scanning, all participants changed into MEG-compatible (cotton only) clothing provided and were instructed to remove all metal objects such as hair pins, piercings, and jewelry. Standard MEG preparation was performed including placement of position indicators (cHPI coils) to control for movement, 3D registration of the scalp (Polhemus) for co-registration with structural MRI. All MEG data was collected using a 306-channel Elekta Neuromag TRIUX system. As part of routine MEG procedures, EOG and electrocardiogram (ECG) were recorded and digitalized together with the MEG data. Indeed, measuring ECG is critical in MEG studies, since the magnetic component of heart beats introduce pronounced artifacts in MEG. ECG was collected from electrodes placed on the collarbones, which were later used to identify and correct artifacts with ICA (see MEG processing pipeline below). While EMG is traditionally used for measuring the direct muscle activity of the orbicularis oculi in startle studies in humans, vertical EOG can also help inferring on the magnitude of the blinking response [25]. Horizontal EOG, which is more frequently used to assess retinal activity, was not used in the blinking analysis. To minimize sleepiness during the exposure to BBN carriers, the participants were instructed to sit relaxed and watch a silent nature film on a screen with a display size of a 72 \(\times \) 44 cm rectangle by an projector (FL35 LED DLP) situated outside the magnetically shielded room, while the different sound trials of the GPIAS paradigm were played. The MEG session concluded with a 5-min resting state recording in silence where the participants were instructed to sit relaxed, watching the movie as they had been for the previous part of the measurement. For the GPIAS paradigm, all sounds were presented pre-loaded from hardware (AudioFile stimulus processor; Cambridge research systems) through sound tubes (3.3 m, 12 mm and 4.5 mm wide) connected to one ADU1c acoustic driver unit (KAR audio) per ear.

A structural MRI was recorded on a separate day after having participated in the MEG session. The MRI data consist of 3D T1-weighted magnetization-prepared rapid gradient-echo (MPRAGE) sequence structural images (voxel size, 1 \(\times \) 1 \(\times \) 1 mm; field of view, 256 mm; RT, 2300 ms; ET, 2.98 ms) obtained on a GE Discovery 3.0T MR scanner. MRI data were processed with FreeSurfer (version 7) for cortical surface reconstruction and for creating source space models for MEG source reconstruction.

The critical parameters theorized to influence (cortical) inhibition by silent gaps were evaluated by modulations of, namely, carrier noise, pulse-level, and interstimulus interval (ISI). Trials were constructed with a uniform broadband white noise as carrier in MATLAB 2019b and exported as mono wav-files at 44.1 kHz sample rate. Level was calibrated using a sound level meter (Brüel & Kjær, 2235) coupled with a pre-amplifier (Brüel & Kjær, 2619) to an artificial external ear (Brüel & Kjær, 4157) at a level equivalent to fast, A-weighted dB SPL as measured for 15 s of continuous presentation - dB(A) SPL is hereafter referred to as dB for sake of brevity.

Two broadband carrier background of either 60 or 70 dB were used to present firstly increasing pulses (pulse only (PO); 20 ms with no rise and fall time) from 70 to 95 dB or 75 to 95 dB (with two ranges of relative intensities: +10 to +35 dB and +5 to +25 dB, respectively), secondly with silent gaps (duration of 50 ms with a 2 ms sinusoidal rise and fall ramp) in absence (gap only (GO)) or in presence of a 90 dB pulse (gap pulse (GP), to avoid exposure to louder sound intensities), using ISIs of 0, 60, 120, or 240 ms. In one 5-min block, all trials were presented 5 times each in random order with a 2-s interval (jittered by 0.5 sec), first in a 60 dB carrier white noise, followed by presentation in a carrier noise of 70 dB.

Trial randomization and inter-trial jitter, as well as synchronization of trigger channels, were performed by a custom script in Presentation software (Version 18.0, Neurobehavioral Systems, Inc.). After each block, we communicated with the participant via an intercom system to make sure they were comfortable and assessed sleepiness using the Karolinska Sleepiness Scale (KSS) [55]. Ten blocks were repeated for a total of 50 presentations of each trial type, and the total scanning time was around 90 min.

Of note, this study did not include any PPI experiments, since measurements from structures of the basal ganglia such as the LGP cannot be obtained using MEG. Neurons in the LGP are not neatly aligned in parallel as the synapses of the pyramidal neurons in cortex are. As a consequence, we assumed there would not be enough summation of the electric/magnetic fields to generate a strong enough signal to be measured outside the head. Would the anatomy not have been a hindrance, the LGP is too distant from the MEG sensors outside the head to be picked up, as the signal decay of the magnetic field over distance is the inverse cubed power.

The MEG data was recorded with a sampling rate of 5000 Hz. We generated signal-space projection (SSP) vectors [68] from both gradiometer and magnetometer sensors, using two empty-room recordings taken on the same day as the subject’s experiment, before and after the experiment. These SSP vectors were then employed to eliminate environmental noise originating from sources outside the subject’s body and the MEG system in the data. Next, we utilized the signals from continuous head position indicator (cHPI) coils to track the subject’s head position over time, which is used for subsequent compensation in the recordings. Afterwards, we automatically identified and corrected noisy and flat MEG channels, as well as crosstalk compensation among the well-functioning MEG channels [64]. We concluded this by performing spatiotemporal signal-space separation (tSSS) on the data in 10-s chunks for computing temporal projections [30, 64]. Subsequently, the data was downsampled to 250 Hz. Prior to downsampling to prevent aliasing, a low-pass filter with a cutoff frequency of 500 Hz was applied to the recordings. The MEG recordings were then bandpass filtered within the range of 0.1 to 40 Hz.

The data was further broken down into independent components (ICs) for the purpose of artifact correction, utilizing the Infomax method [2]. We determined the minimum number of components necessary to collectively account for at least 95% of the data’s variance. Specifically for the ECG channel, we applied a bandpass filter between 8 and 16 Hz and identified the relevant ICA component, which was subsequently removed using the cross-trial phase statistics method [13]. In the case of muscle artifacts, we used the approach detailed in Dharmaprani et al. [16] to identify and remove any associated ICA components. Notably, ICA components representing vertical eye movements were retained in the data. This decision was made because not all the activities detected by ICA were actual blinks; some were ocular (startle) responses to the auditory stimuli.

Consequently, after epoching the data, we established a custom threshold (500 \(\mu v\)) to selectively remove epochs with blinks exceeding the threshold, effectively eliminating “voluntary” blinks while preserving ocular responses to the auditory stimuli. This is based on evidence showing that voluntary blinks exhibit a significantly greater amplitude compared to spontaneous (involuntary) blinks [14]. The recordings are segmented into epochs, covering a time window of 300 ms before the stimulus onset to 300 ms after the stimulus onset. The period ranging from 300 ms before the stimulus until the stimulus onset is designated for baseline correction. Epochs exhibiting a peak-to-peak signal amplitude less than 1 femtotesla or exceeding 4 picotesla (for magnetometers) are discarded. Similarly, epochs with a peak-to-peak signal amplitude below 1 fT/cm or exceeding 400 pT/cm (for gradiometers) are also excluded from subsequent processing.

We performed the anatomical cortical surface reconstructions of the MRIs using FreeSurfer software [21]. Following that, we generated boundary element model (BEM) surfaces, including the inner skull, outer skull, and outer skin (scalp), utilizing the watershed algorithm [30, 54]. We established a surface-based source space with bilateral hemisphere representation, recursively dividing it with octahedron spacing. For each subject, we constructed a BEM model and its corresponding solution using the linear collocation method [30]. To ensure proper alignment, we performed co-registration between the MRI and the head model, utilizing three fiducial points provided in the MRI as an initial solution. This transformation was further refined through 40 iterations of the iterative closest point (ICP) algorithm [10]. Any outlier points exceeding a distance of 5 cm were excluded, and the fitting process was repeated. For each subject, we computed the forward solution using the source model, BEM model, and co-registration information. We calculated the noise covariance of the recordings, focusing on the pre-stimulus periods of the pulse-only stimuli, as these periods are devoid of event-related brain activity. We used both the empirical [30] and shrunk [38] methods to estimate the noise covariance and selected the best estimator based on log-likelihood and cross-validation with unseen data [19]. Using the noise covariance and the forward solution, we applied the linear minimum-norm inverse method, known as dSPM to determine the inverse solution [30]. This enabled us to obtain source time courses for each vertex in the source space. To facilitate group-level analysis and ensure that spatial locations could be compared uniformly across all subjects, we morphed each individual’s source space onto a template brain. This involved a linear mapping of cortical surface values from each individual subject to those in a FreeSurfer template brain [21]. Next, we generated a single time course for each brain label by averaging the source time courses within the vertices located within that specific brain label, regardless of their orientation. The brain labels and cortical parcellation were derived from the Desikan-Killiany Atlas [15] with 68 bi-hemispheric parcels.

Further analysis steps included dividing the continuous recording in epochs ± 600 ms relative each pulse trigger onset for each separate stimuli type. The data was then inspected for each stimuli type, and high variance trials were manually excluded for both gradiometer and magnetometer channels. All trials were compiled, and ECG channel was used to identify components of the heartbeat that could create artifacts. These were removed from the full dataset using independent component analysis (ICA) in a standard pipeline. The data was not processed to remove ICA components, corresponding to vertical eye movements from the EOG channels as high correlation between eye blinks and any response of interest related to the pulse stimuli was expected. The EOG channel, which records vertical eye movements to detect blinks, was analyzed independently to compare the traditional muscle reflex with cortical responses. The EOG channel, which captures a mixture of eyelid movements (blinks) and vertical eye movements, was analyzed independently to compare cortical responses with peripheral signals related to blink-like activity. EOG signals were recorded using the standard NatMEG electrode placement setup, as detailed at the following link. The vertical EOG electrode was positioned above the eye to capture signals related primarily to eyelid movements, particularly blinks, though contributions from vertical eye movements cannot be entirely ruled out. Signals were digitized at 250 Hz and bandpass filtered between 1 and 10 Hz using a Hanning-windowed finite impulse response (FIR) filter with a 0.5 Hz transition bandwidth. This filter was applied bidirectionally to ensure zero-phase distortion. The selected frequency range was chosen to isolate blink-related transients from slower ocular movements. Blinks were automatically detected using a peak detection algorithm, with the detection threshold defined as one-fourth of the EOG signal’s amplitude range: \((max(EOG) - min(EOG)) / 4\). EOG responses surpassing a certain threshold (i.e., 500 \(\mu v\)) were assigned as spontaneous blinks, resulting in the removal of such trials. Data from individual gradiometer sensors and those with the highest event-related field (ERF) amplitude among most participants were examined separately on the left and right sides of the array. All MEG processing analyses were performed using MNE software (Version 1.6.1).

Responses for all trials (i.e., EOG, ERP peak latency an amplitude) were analyzed in a defined time window of interest of 85–115 ms with a two-way analysis of variance (ANOVA). To identify a response as a blinker, we used a thresholding response whereby an EOG signal exceeding the average EOG signal before the stimulus onset value + 10 * the median absolute deviation of the same pre-stimulus period was considered a blink. The percentage of inhibition for the four different ISIs were calculated as [1-(GP/PO)], where GP represent area under curve in trials with a gap preceding the pulse and PO trials with a pulse only. The resulting values denoting the percentage of inhibition were then also subjected to two-way ANOVA. For the individual contrasts between conditions, post hoc tests and P-values reported are corrected for multiple comparisons using the Tukey method. All statistical analyses have been performed with Python (Version 3.10.9) and the packages SciPy (Version 1.13.0). Statistical test assumptions were checked before testing and the global two-tailed significance level was set to p< 0.05.

Blinking and cortical responses to increasing pulse intensities. a Distribution of EOG area under curve values for multiple pulse levels starting from 70 to 95 dB with 60 dB (upper panel, orange) and 70 dB (bottom panel, purple) broadband noise carrier. b Pie plots representing the number of subjects with maximal activity in left hemisphere (left panel) and right hemisphere (right panel) brain parcels. c Distribution of peak amplitude values in the left (blue) and right (red) transverse temporal gyrus for multiple pulse levels within 70 and 95 dB; upper and lower panels display peak amplitude values with 60 and 70 dB broadband noise carrier level, respectively. d Grand averaged topographic distribution in the canonical N1-time window in response to 90 dB pulse stimulus. e Average EOG responses for pulse levels between 70 and 95dB in steps of 5 dB with 60 dB (upper panel, orange) and 70 dB (bottom panel, purple) broadband noise carrier. Highest intensity responses (95 dB) are colored in black. f Average source level analysis of left and right transverse temporal gyrus show an expected increase of activation with rising pulse level in both 60 (upper panels) and 70 dB (bottom panels) broadband carrier noise for both left (blue lines) and right (red lines) transverse temporal gyrus. Box plots show individual values together with median ± 1.5 x IQR (inter quartile range), n = 22–26