Intracortical depth analyses of frequency-sensitive regions of human auditory cortex using 7TfMRI

Abstract

Despite recent advances in auditory neuroscience, the exact functional organization of human auditory cortex (AC) has been difficult to investigate. Here, using reversals of tonotopic gradients as the test case, we examined whether human ACs can be more precisely mapped by avoiding signals caused by large draining vessels near the pial surface, which bias blood-oxygen level dependent (BOLD) signals away from the actual sites of neuronal activity. Using ultra-high field (7T) fMRI and cortical depth analysis techniques previously applied in visual cortices, we sampled 1mm isotropic voxels from different depths of AC during narrow-band sound stimulation with biologically relevant temporal patterns. At the group level, analyses that considered voxels from all cortical depths, but excluded those intersecting the pial surface, showed (a) the greatest statistical sensitivity in contrasts between activations to high vs. low frequency sounds and (b) the highest inter-subject consistency of phase-encoded continuous tonotopy mapping. Analyses based solely on voxels intersecting the pial surface produced the least consistent group results, even when compared to analyses based solely on voxels intersecting the white-matter surface where both signal strength and within-subject statistical power are weakest. However, no evidence was found for reduced within-subject reliability in analyses considering the pial voxels only. Our group results could, thus, reflect improved inter-subject correspondence of high and low frequency gradients after the signals from voxels near the pial surface are excluded. Using tonotopy analyses as the test case, our results demonstrate that when the major physiological and anatomical biases imparted by the vasculature are controlled, functional mapping of human ACs becomes more consistent from subject to subject than previously thought.

Figure 1

A diagram of non-human primate AC model, and a possible human homologue. (a) Non-human primate AC model. Note that the exact arrangement of tonotopic gradients may not be exactly as parallel to area boundaries as in this simplified presentation. However, in general, there are three core areas (auditory area 1, A1; rostral, R; rostrotemporal, RT) include roughly mirror-symmetric tonotopic gradients extending to the lateral (rostrotemporolateral, RTL; anterolateral, AL; middle lateral, ML; caudolateral, CL) and medial (rostrotemporomedial, RTM; rostromedial, RM; middle medial, MM; caudomedial, CM) belt cortices. (b) The corresponding human AC hypothesis, modified based on previous human fMRI (Humphries et al., 2010; Woods et al., 2010) and cytoarchitectonic studies (Hackett et al., 2001; Rivier and Clarke, 1997), is shown on a flattened patch of superior temporal cortex. Anatomical regions: Heschl’s gyrus (HG), planum temporale (PT), superior temporal gyrus (STG), and planum polare (PP).

Figure 2

Stimuli and task. Sound center-frequency changes after each echo-planar imaging (EPI) acquisition. Subjects heard binaural 6.5-s narrow-band noise tokens amplitudemodulated with speech envelope patterns. They were asked to press a button each time the temporal modulation pattern was similar to that presented before the preceding volume acquisition and to ignore the changes in the center frequency. The cycle direction shifted after every other run: an ascending cycle is displayed here.

Figure 3

A demonstration of the evenly spaced cortical surface depths in a representative subject. For clarity, only every other depth sampled in the present study, ranging from the white matter to the pial surface, is shown. The insert shows left hemisphere surfaces from areas including HG and the adjacent superior temporal and insular cortices. The fMRI data of the same subject are shown in Fig. 5.

Figure 4

AC high and low frequency regions identified using high-resolution 7T fMRI sampled across cortical depths. (a) A human AC model hypothesis shown on a flattened patch of left superior temporal cortex (for details, see Fig. 1). (b) Comparison of group analysis results using different analysis approaches. The dotted lines correspond to the main frequency gradients shown in the hypothetical model. An improvement of sensitivity and consistency with the hypothesis is observed when the signals from the pial voxels are excluded. This could reflect improved correspondence of results across the individual subjects. Poorest results are achieved with volume-based analyses (leftmost panel). The main effect of center frequency (100, 240, and 577 Hz vs. 1386, 3330, and 8000 Hz) is shown to help comparisons of frequency-sensitivity regions across subjects. The opaque color scale refers to post-hoc corrected (cluster-based Monte Carlo simulation tests, p<0.05) and the transparent color scale to uncorrected random-effects general linear model (GLM) results. For clarity, the boundaries of the clusters surviving the post-hoc correction have been marked with a white trace. (c) Individual fMRI results on flattened standard-brain AC patches. The upper row shows the data from the voxels intersecting the pial surface, the middle row those intersecting the white matter, and the lowest row shows cortical data with signals from the top layers excluded. Notably, the exclusion of signals from top layers improves the consistency of findings with the hypothesized model in most subjects (e.g., Subj. 8). Another important point to note is the relative weakness of individual-level observations in the white-matter surface: Despite this relative signal weakness, the group level result was statistically stronger than that from the pial surface, in line with our interpretation that the exclusion of pial voxels greatly improves the anatomical consistency of functional results across individual subjects.

Figure 5

Frequency sensitivity gradients in the individual left superior temporal cortex patch and AC volume of Subject 8. (a) In this subject, the more posterior high-frequency area depicted in the hypothesis (Fig. 1) and shown in the group estimate (Fig. 4) is basically lacking from the pial surface analysis but clearly evident in the white-matter and “all but top layer” cortical voxel analyses. (b) The underlying cause for this difference is shown in the 3D rendering: the stronger low-frequency activation is spread across the sulcal banks in voxels intersecting the pial surface. The location of voxel marked by the black arrow corresponds to the vertex location pointed by the white arrows in the surface patch shown in the panel a. (c) The 3D data demonstrate the results of the surface-based analysis resampled back to the corresponding voxels for demonstration purposes, not the native 3D analysis results.

Figure 6

Anatomical correspondence of tonotopy mapping across subjects estimated using the inverse variance of the phase angle. The results suggest that the inter-subject consistency of continuous tonotopy estimates is improved after the voxels intersecting the pial surface have been excluded from the analysis. The phase angle of the BOLD response was first determined in each individual subject at the stimulation-cycle frequency (1/60 Hz) by computing the FFT (Talavage et al., 2004), after which the inverse of the group variance of the phase angles was determined at each AC location (Berens, 2009). The individual-level phase angles utilized for this group analysis are shown in Fig. 7. The results are shown on a flattened patch of the left superior temporal cortex (Freesurfer fsaverage).

Figure 7

Phase-encoded analysis of tonotopy progressions at different depths of AC, shown on a flattened patch of the left standard-brain superior temporal cortex (Freesurfer fsaverage). Consistent with main analyses (Fig. 4), the continuous tonotopy mapping results become more consistent and coherent after the exclusion of signals from voxels intersecting the top layers of AC.

Figure 8

Within-subject reliability analyses shown on a flattened patch of the superior temporal cortex (Freesurfer fsaverage). (a) Statistical significance of ICCs calculated across the first and second half of experimental runs in all subjects. No trend of decreased reliability was observed in analyses based on voxels intersecting the pial surface, which would explain the group results in our main analyses (Figs. 4b, 6). (b) A qualitative comparison of frequency-sensitivity mapping results (main effect of high vs. low frequency blocks) in two subjects each scanned in two separate sessions. The overall pattern is consistent with the results of the group-level reliability analysis.