Pre-ataxic and ataxic SCA3 mutation carriers (N = 18) were enrolled between July 2019 and February 2022. The presence and severity of ataxia was assessed using the Scale for the Assessment and Rating of Ataxia (SARA); clinically manifest ataxia was defined by a SARA score ≥ 3 [19]. The examination of neurological signs other than ataxia was done using the Inventory of Non-Ataxia Signs (INAS) [20]. The presence and severity of depressive symptoms was assessed with a Rasch-based depression screening (DESC) [21]. Further, the Montreal Cognitive Assessment (MoCA) was applied to screen for cognitive impairment. [22]

The control group comprised N = 18 healthy adults and was carefully matched for age. None of the participants reported any history of hearing impairment or psychiatric or neurological disorders.

Written informed consent was obtained from every subject and they were not paid for participation. The experimental procedures were conducted in accordance with the ethical standards as laid down in the 1964 Declaration of Helsinki and its later amendments and approved by the local ethics committee (Medical Faculty, University of Heidelberg, S-242/2019).

A 3-alternative forced choice task [23] was conducted, with two runs per listener, to determine the frequency up to which Huggins pitch could be heard by the individual participants. Each trial consisted of three 800-ms-long broadband Gaussian noise bursts (bandpass-filtered at 10–10.000 Hz and ramped on/off with 20 ms cosine ramps; inter-stimulus interval: 500 ms) that were presented to the listeners in random order. In one of the noises, a 2π phase transition was imposed in the right-ear channel to elicit Huggins pitch, and subjects were asked to identify this stimulus from the three noise bursts. The bandwidth of the phase transition was set to 10% of the Huggins pitch frequency (HP-f0). The HP-f0 starting value was set to 400 Hz; HP-f0 increased after correct responses in two successive trials, and it decreased after one incorrect response. The HP-f0 change factor for an increase or decrease was set to 1.5 and was set to 0.75 after two reversals. The upper frequency limit of Huggins pitch was determined as the mean HP-f0 of the last 6 reversals in the task. The task and all sounds for the MEG experiment described below were built using MATLAB 7.1 (The Mathworks, Inc. Natick, MA). Stimuli were presented at 65 dB SPL and with 48,000 Hz sampling rate, using Beyerdynamic DT 770 PRO headphones attached to an external digital-analog converter and a headphone amplifier (RME ADI 2-DAC FS). To avoid fine structure cues due to frozen noise, each single noise was generated freshly from running noise. HP-f0 differences between groups were statistically analyzed using the Wilcoxon-test.

The MEG experiment comprised two binaural and one monaural conditions which were recorded in separate blocks, each block with a total duration of 15 min. In each block, stimuli were played 200 times, with inter-stimulus intervals randomly distributed between 800 and 900 ms. Within conditions, each single stimulus segment had a length of 750 ms and was equipped with 50-ms Hanning windows at its onset and offset. All sounds were generated using in-house MATLAB scripts, converted by a 24-bit sound card (RME ADI 8DS AD/DA interface), attenuated by Tucker Davis Technologies PA-5, and delivered to the listeners via shielded Etymotic Research (ER3) earphones, attached to 90-cm plastic tubes and foam inserts. The earphones were driven by a Tucker Davis Technologies HB-7 headphone buffer.

Besides the Huggins pitch condition, there was also a binaural condition in which the interaural correlation of two successive broadband Gaussian noise bursts changed from + 1 to − 1, or vice versa, with square-rooted 10-ms Hanning ramp crossfading to avoid monaural cues at the transition. Changing the interaural correlation induces a diffuse change in the perceived spatial location of the noise, but not a pitch percept. The monaural condition was made of iterated rippled noise (IRN) [24] which is built by adding the copy of a noise back to the original signal with a given time delay. IRN pitch corresponds to the reciprocal of the time delay, and pitch salience increases with the number of iterations. In this study, we employed IRN with 20 iterations. In the IRN and Huggins pitch conditions, stimuli were assembled to short triplets, in an effort to separate the neuromagnetic energy onset response (EOR) from pitch onset (POR) and pitch change responses (PCR) (e.g., Andermann et al. 2021) [25]. Each triplet began with a noise burst without pitch, followed by two stimuli with different pitch. The first pitch corresponded to 880 Hz, and the second pitch was either a perfect fifth up or down, relative to the first pitch.

Neuromagnetic field gradients in response to the stimulation were acquired using a Neuromag-122 MEG gradiometer system (Elekta Neuromag Oy, Finland) [26] inside a shielded room (IMEDCO, Hägendorf, Switzerland). Data were sampled at 1000 Hz and low-pass filtered at 330 Hz. Prior to the recordings, fiducials and 100 surface points were digitized using a Polhemus 3D-Space Isotrack2 system to determine the head shape and position relative to the MEG gradiometers. During the recordings, participants watched a silent movie of their own choice to maintain stable vigilance, and they were asked to ignore the sounds in the earphones.

Gradiometer data were analyzed using the BESA 5.2. software (BESA GmbH, Gräfelfing, Germany), with a spherical head model and a homogeneous volume conductor. After removing noisy channels, epochs with amplitudes > ± 8000 fT/cm and gradients > ± 800 fT/cm/ms were automatically removed using the BESA rejection tool. Across all conditions, on average, 85.9% (SD = 10.4) of the sweeps remained in the analysis. Prior to spatio-temporal source analysis [27, 28], data were zero-phase filtered at 1–30 Hz. For all conditions, neuromagnetic responses to the transitions between the noise segments were pooled, and a source model with one equivalent dipole per hemisphere was fitted on the prominent N100m POR in the IRN condition, with the fit interval centered about 30–50 ms around its peak. No constraints were applied for fitting, but symmetry was applied to stabilize the fit when necessary. The source models were then applied as spatio-temporal filters to derive the source waveforms within the unfiltered IRN, Huggins pitch, and interaural correlation conditions; here, principal component analysis [29] based on the last 100 ms of the epochs was used for drift compensation. The resulting source waveforms were exported to MATLAB for further graphical and statistical analysis; similarly, dipole coordinates were exported in approximate Talairach space. [30]

N100m peak amplitudes were extracted by using the grand-average N100m peak latency as a reference and averaging the individual waveform in a 50-ms window around this value, separately for every participant. This procedure ensured that amplitude values could be determined for each single listener even when the waveform did not show a prominent peak response (which was the case for a few listeners in the Huggins pitch condition). For between-group comparison of the N100m amplitude, the non-parametric Wilcoxon test was chosen from the R statistical package (Version 4.1.0). Correlations were computed using the Spearman rank correlation coefficients.

Neurophysiological data is available in OSF (https://osf.io/rmshp/).