Intracortical myelination in musicians with absolute pitch: Quantitative morphometry using 7‐T MRI

Abstract Absolute pitch (AP) is known as the ability to recognize and label the pitch chroma of a given tone without external reference. Known brain structures and functions related to AP are mainly of macroscopic aspects. To shed light on the underlying neural mechanism of AP, we investigated the intracortical myeloarchitecture in musicians with and without AP using the quantitative mapping of the longitudinal relaxation rates with ultra‐high‐field magnetic resonance imaging at 7 T. We found greater intracortical myelination for AP musicians in the anterior region of the supratemporal plane, particularly the medial region of the right planum polare (PP). In the same region of the right PP, we also found a positive correlation with a behavioral index of AP performance. In addition, we found a positive correlation with a frequency discrimination threshold in the anterolateral Heschl's gyrus in the right hemisphere, demonstrating distinctive neural processes of absolute recognition and relative discrimination of pitch. Regarding possible effects of local myelination in the cortex and the known importance of the anterior superior temporal gyrus/sulcus for the identification of auditory objects, we argue that pitch chroma may be processed as an identifiable object property in AP musicians. Hum Brain Mapp 37:3486–3501, 2016. © 2016 Wiley Periodicals, Inc.

Supporting information for: Kim and Knösche, Intracortical myelination and absolute pitch neighbor interpolation. The individual qT1 maps on native surfaces were then registered to a template mesh contained in FreeSurfer (called 'fsaverage' in MNI305 space) using spherical mapping that aligns local curvature.
Finally, qR1 (= 1/qT1) values were computed from the surface-mapped qT1 values and bounded within a range between 0.25 and 10 s -1 , which corresponds to the qT1 range between 100 and 4000 msec.

Correlation between local curvature and cortical myelin
Spurious correlations between cortical myelin density, cortical thickness, and local curvature were reported from histological samples (Annese et al., 2004) and a large-scale in-vivo MRI dataset (Shafee et al., 2015). Those relationships are known to be negative (greater myelination in a thinner or more convex region of the cortex) (Shafee et al., 2015). To construct a grid system free of such a bias, Waehnert et al. (2014) implemented a novel layering method that preserves local volume so that the modeled layers mimic the physical bending process of cortical sheets. Although such layer modeling is highly attractive and worked better in ex-vivo images at 0.14-mm isotropic resolution, as shown in Waehnert et al. (2014), the result of the isovolumetric model was very similar to that of other methods, such as an isopotential model using the Laplacian or an equidistant model, suggesting that the benefit of the novel layering method would be limited in our dataset at 0.7-mm isotropic resolution. Thus, we used an equidistant model (i.e., equal thickness of 'layers' at each point) by fixed proportion of cortical thickness.
In the present data, we indeed found a negative correlation of qR1 values with curvature and thickness as previously reported in the literature (Dick et al., 2012;Lutti et al., 2014;Sereno et al., 2013). However, the effect sizes were very small and negligible (correlation between qR1 and curvature was -0.0058 ± 0.0026; correlation between qR1 and thickness was -0.0757 ± 0.0083). The significance level was very high (p< 10 -16 ), but this was mainly due to the very large number of measures (more than 140,000 vertices for each hemisphere). This also explains why the 'de-curved/de-thickened' qR1 values were very similar to the demeaned qR1 values in those studies (Dick et al., 2012;Sereno et al., 2013). Consequently, we did not use the geometrical covariates in our statistical analysis.

Effects of demographic variables
To determine whether covarying demographic variables is necessary, we tested effects of age, sex, and ethnicity using a very simple model as: where D is either age, sex, or ethnicity. We found significant effects over extensive areas including right inferior frontal gyrus, right anterior cingulate gyrus, and right superior temporal gyrus (age effect; Figure S1, upper row), right superior temporal sulcus (sex effect; Figure S1, middle row), and left planum temporale and supra marginal gyrus (ethnicity effect; Figure S1, lower row). Given the significant effects of demographic variables, we incorporated them as covariate terms in order to control possible confounding.
Supporting information for: Kim and Knösche, Intracortical myelination and absolute pitch

Controlling for the effect of ethnicity
Due to the imbalanced design of the current study in terms of ethnicity (i.e., 8 musicians with absolute pitch [AP] = 5 Asian musicians + 3 European musicians; 9 musicians without AP = 9 European musicians), one might wonder if it is possible to separate the effect of ethnicity from that of absolute pitch. To compare the bias in the current sample with that of an ideal scenario, a numerical simulation was carried out. As stated in the main text, the group effect was estimated using the model (Eq. S2): (Eq. S2) From the peak vertex at the middle depth (50% of cortical thickness) in the right planum polare, estimated coefficients were 0.4437 (interceptor), 0.0004 (age), 0.0040 (sex), -0.0093 (ethnicity), and 0.0169 (AP) s -1 . The standard deviation of the residuals was 0.0033 and the mean was 10 -14 s -1 . The Kolmogorov-Smirnov test did not reject the null hypothesis that the distribution of residuals is normal (p = 0.845). Therefore, we used normal distribution to generate noise with the observed mean and standard deviation. Assuming the model (Eq.2) as well as effect size and noise level estimated from the current dataset as "ground truth", simulated datasets (k = 10,000) were generated for different design matrices varying the number of European AP musicians. Because the number of European AP musicians controls the separation between the ethnicity and AP variables, its change varies the collinearity between ethnicity and AP. The procedure of simulation is as follows: 1. Change the original design matrix X to make it more or less balanced by changing the ethnicity vector 2. Repeat 1 for all possible design matrices.
Boxplots of relative errors of estimated effect sizes of AP over the correlation between ethnicity and AP are given in Figure S2. Because the generated noise was unbiased, the median of estimated β i was very close to the true β i in all cases. However, as the design becomes more imbalanced (i.e., greater correlation), the reliability of estimation decreases (i.e., more distant range between outliers and quantiles), and vice versa.
The result indicates that the reliability of estimating the effect size of AP under the current design (relative error= -0.09 ± 13.16 %; the blue squared boxplot in Figure S2) is still comparable to a perfectly balanced design (relative error= 0.05 ± 9.41 %, the leftmost boxplot in Figure S2) and much better than the worst case (relative error= -0.39 ± 21.39 %; the rightmost boxplot in Figure S2). Additionally, in terms of the minima and maxima of relative errors from 10,000 randomizations, the current design shows reasonable bounds (

Effect of frequency discrimination threshold
We tested a GLM (Eq. 3 in the main text) as: where FDT=F1/F0, F1 is the minimal frequency of a target that one can discriminate, and F0 is the frequency of a reference. See Micheyl et al. (2006) for further details of the behavioral experiment. The results are given in Figure   S3 and Table S  See Table S-II for details of significant clusters.

Subcortical white matter myelination
The main findings of altered cortical myelination in the anterior part of the right supratemporal plane (STP) that related to absolute pitch (AP) performance could either be due to long-range connections from/to the anterior STP or due to intracortical connections within that region. If increased myelination was also found in the white matter underneath the anterior STP, the additional cortical myelin in the AP musicians could be interpreted as part of a long-range connection. However, qR1 mapping is not an optimal choice to investigate white matter myelination compared to magnetization transfer imaging or diffusion-weighted imaging. Here, we only explored such a possibility from qR1 data, but it should be further examined using white matter myelin imaging in a future study.
We sampled qR1 values from subcortical white matter at 25, 50, and 75 % of cortical thickness along the inward normal vectors pointing to the white matter. All sampling points were visually inspected (see Figure S4 for an example) and computationally confirmed that coordinates of all sampling points are confined in the white matter mask in 3-D volume. Supporting information for: Kim and Knösche, Intracortical myelination and absolute pitch Using a general linear model (GLM) (Eq. S2) that was used to test AP effect in intracortical myelination, we found an AP effect in the right angular gyrus (AG) (max T(12) = 6.29, p = 0.018, area =25 mm² at the -75 % projection level, peak 42] mm) as seen in Figure S5 (upper row).
However, we did not find any differences in subcortical white matter near the right anterior STP, suggesting that the greater cortical myelination in the AP musicians might reflect horizontal, local connections within the cortex rather than vertical, long-range connections.
In addition, we investigated FDT effect in qR1 values in the subcortical white matter using a GLM (Eq. S3) because intersubject variability in FDT might be related to variability in myelination of acoustic radiation (i.e., thalamic projection into primary auditory cortex (PAC). However, we only found a positive correlation between FDT and subcortical myelin in the left orbital sulci (max T(12) = 9.40, p = 0.0002, area = 254 mm 2 at the -25 % projection; max T(12) = 8.41, p = 0.0158, area = 48 mm 2 at the -50% projection) as shown in Figure S5 (lower row) but no effect of FDT in qR1 values beneath the PAC. Significant clusters for the effects of AP and FDT are listed in Table S-III. However, as mentioned earlier, more evidence from other imaging techniques that are better at localizing alteration of white matter myelination is needed.

Intersubject variability of octave errors
Octave errors are defined as answers with correct pitch chroma, but incorrect pitch height. Because nonabsolute pitch (AP) musicians did not give a sufficient number of answers with correct pitch chroma (by definition), we only analyzed octave errors of AP musicians (n=8). We computed the octave error rate (OER), which is a ratio of the number of octave errors to the number of answers with correct pitch chroma, for pure tones and piano tones separately.
As discussed in the Behavioral results subsection in the main text, AP musicians made more octave errors for pure tones compared to piano tones (T(7) = 5.68, p = 0.0007). The OERs for pure tones were greater than 50% for all AP musicians whereas the OER for piano tones were less than 50% as shown in Figure S6 A. In accordance with known dissociation between the acuities in pitch chroma and pitch height (Deutsch, 2013), we did not find any correlation between the OER and the absolute pitch score (APS) (p = 0.48, pure tone; p = 0.56, piano tone) as in Figure S6 B, C. Moreover, no significant correlation between the octave errors and fine-grained relative pitch perception (i.e., FDT) was found (p = 0.5 for pure tones; p = 0.92 for piano tones) as in Figure S6 D, E.
Because the number of AP musicians in the current dataset is limited (n=8), it was difficult to fit a GLM with demographic covariates reasonably (i.e., degrees of freedom = 3). Because age, sex, and ethnicity were still influential also within the AP musicians, we believe controlling the demographic variables is necessary. Further investigation on the octave errors would be possible with a greater dataset of AP musicians in the future.