Introduction

The fMRI signal in the human auditory cortex is positively and negatively correlated to gamma and alpha band neuronal oscillations, respectively, during movie watching¹. Yet how BOLD signals at different cortical depths are correlated to neural signals under naturalistic acoustic stimuli remained unknown. Because of different input-output neuronal connections and vascular reactivities, the coupling between the fMRI signal and oscillatory neural activity is expected to vary across cortical depths. Specifically, with preferential anatomical feed-forward connections to the intermediate cortical depth in the primary sensory regions^2,3, we hypothesize that the hemodynamic response in intermediate cortical depth is more significantly correlated to neural signals than in superficial or deep layers. We further hypothesize that this correlation is more significant in the primary auditory cortex and most prominent at the right hemisphere⁴. Here we use cortical-depth dependent fMRI^5-7 and invasive electrode recording on epilepsy patients to study the correlation between neural oscillations and fMRI signals across cortical depths in the auditory cortex during music listening.

Methods

All patients and participants joined this study with written informed consent after the approval of the Institute Review Boards. Functional MRI data were acquired on a 7T (Terra, Siemens) and 3T MRI systems (Skyra, Siemens) with a 32-channel whole-head coil array and a customized 24-channel coil array fitted the right temporal lobe⁸, respectively. Structural and functional images from 3 (at 7T) and 16 (at 3T) healthy participants were acquired with MP2RAGE and a 0.8-mm (at 7T) and 1.5-mm (at 3T) isotropic resolution EPI, respectively. Nine cortical surfaces with equally spaced cortical thickness were reconstructed from the structural images using FreeSurfer ^9,10. Auditory stimulus including three songs (Song 1: “Doraemon” theme song, Song 2: “Brahms Piano Concerto No. 1”, and Song 3: “Lost stars” from Adam Levine). Each participant listened to each song twice in a randomized order.
Electrophysiological responses were measured from six epilepsy patients by stereotactic electroencephalography (sEEG). All patients had an electrode (Ad-Tech Medical Instrument, Oak Creek, WI, USA) with up to eight contacts (5-mm separation) at the temporal lobe. Pre-surgery and post-surgery structural MRIs were obtained from patients to identify electrode and contact locations. The same three songs were presented to each patient, who listened to each song twice in a randomized order.
The sEEG data were re-referenced to the average of each electrode. Frequency-specific oscillatory neural activities were estimated by first applying the Morlet wavelet transform (the central frequencies varying between 4 Hz and 150 Hz in steps of 2 Hz and 7-cycle width) to the sEEG time series and then taking the absolute values. At each central frequency, a modeled fMRI time series was created by convolving a canonical hemodynamic response function to the oscillatory neural activities at that frequency. Singular Value Decomposition was applied to the collection of the modeled fMRI time series from all contacts on the electrode at the temporal lobe. The first singular vector was taken as the average of the modeled fMRI time series elicited by frequency-specific oscillatory neural activities.
At each central frequency, a General Linear Model was used to correlate between the sEEG and cortical depth specific fMRI data from two different participant groups to reduce the concern of within-subject variability. We particularly focused on the core and noncore regions of the auditory cortex⁸.

Results

Figure 1 shows the average of the modeled fMRI time series elicited by frequency-specific oscillatory neural activities in listening to three different songs in two repetitions. Reproducible patterns were found below 30 Hz. At the auditory cortex, negative and positive correlations were found between the fMRI signals and neural activity at alpha/beta bands (8 to 24 Hz) and gamma band (36 to 76 Hz), respectively (Figures 2 and 3). These correlations were stronger at intermediate depth (normalize depth n.d.=0.5; Z > 5) and weaker at superficial (gray-pial boundary; n.d.=0.9; Z < 3) as well as deep (gray-white matter boundary; n.d.=0.1; Z < 3) depths (Figures 2 and 3). Comparing to the non-core auditory cortex, the core auditory cortex has more negative fMRI-alpha/beta bands neural activity correlations and more positive fMRI-gamma band neuronal activity (Figure 4). The core auditory cortex had a particularly positive correlation in the gamma band in the intermediate depth. At the non-core auditory cortex, the positive correlation in the gamma band was stronger at the superficial depth. The negative correlation in the alpha/beta band at the non-core auditory cortex was less than that at the core auditory cortex.

Discussion

We found the hemodynamic responses in human auditory cortex were significantly negatively and positively correlated with neuronal oscillation in alpha/beta band and gamma band, respectively, corroborating with previous studies of the auditory¹ and visual system⁷. Consistent with our hypothesis, the correlation between the fMRI signals and oscillatory neural responses was stronger in the core than the non-core area. At the core area, the intermediate cortical depth has particularly strong correlation in the gamma band between 50 Hz and 80 Hz. This finding is in line with the notion that gamma-band oscillations are related to the bottom-up information processing^11,12.

Acknowledgements

This work was partially supported by the Academy of Finland (No. 298131), and the Natural Sciences and Engineering Research Council of Canada (RGPIN-2020-05927).