Main

Pitch perception is an essential prerequisite for understanding music and speech intonation. Although there is common agreement that the perceived pitch of harmonic complex tones like instrumental sounds or vowels in the singing voice is closely related to the fundamental frequency (f0) of the sound spectrum, large individual differences in pitch and timbre perception challenge this one-to-one relationship1,2. In particular, if f0 is not physically present, pitch is perceived either as the missing f0 or as spectral pitch (fSP), corresponding to the dominant perceptual mode (historically referred to as 'synthetic' versus 'analytic' pitch perception3,4).

At the cortical level, both frequency processing and pitch perception have been found to correlate with neural activity changes in the auditory cortex, related to different processing stages of hierarchically organized auditory subareas5. Physical stimulus properties such as periodicity6, temporal regularity7,8 and frequency spectrum9 are encoded in both subcortical and cortical structures of the auditory ascending pathway. In primary auditory cortex, sound frequency is represented in mirror-symmetric tonotopic frequency maps9,10 by spatiotemporal integration11. The more complex the stimuli and the processing tasks, ranging from pitch perception12 to melody7, timbre or tonality13 processing, the more lateral and anterior are the main peaks of activation in Heschl's gyrus (HG) and anterior supratemporal gyrus (aSTG)13,14,15,16. Furthermore, the representation of pitch as a perceptual, rather than physical, stimulus property was found to correlate with neural activity changes in the non-primary auditory cortex15. Consistent with this finding, numerous functional imaging studies point to the existence of a 'pitch processing center'12 immediately anterolateral to primary auditory cortex within the lateral Heschl's gyrus (lHG), subserving the processing of fixed pitch16, pitch chroma17, pitch salience15, pitch direction18, pitch sequences19 and lively pitch20.

Both relative hemispheric asymmetries21,22,23,24,25,26 and the absolute magnitude of the neural auditory cortex substrate27 are important in enabling understanding of how brain structure maps with the observed functional specialization. In particular, recent functional imaging studies show a relative left-hemispheric specialization for rapid temporal processing23,25, whereas right auditory cortex shows a stronger sensitivity for spectral processing23 and a slower temporal processing mode25. Motivated by these findings of asymmetry and strong mutual correlations between early auditory evoked activity, anatomical size and behavioral predisposition in the anteromedial portion of HG27, we hypothesize here that f0 versus fSP perception may serve as a predictor reflecting both functional and structural aspects of the pitch-sensitive areas in lHG. The main purpose of this study was to investigate (i) individual psychometric differences in f0 and fSP perception in relation to musical aptitude, (ii) the neural basis for type of pitch perception by using MRI and magnetoencephalography (MEG) and (iii) the influence of relative hemispheric lateralization versus absolute magnitude of both gray matter volume structurally and auditory evoked activity functionally.

Results

Psychometric testing and grouping

A psychometric pitch test was designed on the basis of standard methodology1,2 for a parametric range that was extended with respect to the typical pitch range of musical instruments. The test required participants to state the dominant direction of pitch shift between items in each of a total of 144 pairs of complex tones with systematic variation of frequency (f), order of harmonics (n) (that is, a multiple integer of the fundamental frequency (fSP = n × f0)) and number (N) of present harmonics. The stimuli were such that the perceived direction of the shift in pitch between the two tones was dependent upon whether they are decoded in f0 or fSP pitch (Fig. 1a). Accordingly, for each individual a psychometric asymmetry coefficient was derived by recording the number of occasions of f0 listening versus fSP listening and computing an 'index of pitch perception preference' δp = (f0 − fSP) / (f0 + fSP). Analysis of the psychometric data of a large sample of 373 musicians including symphony orchestra musicians28 and 48 non-musicians showed that δp demonstrated a broad bimodal distribution (Fig. 1b), which allowed for a dichotomous classification of all subjects as belonging to one of two behavioral groups, either 'f0 listeners' (δp < 0) or 'fSP listeners' (δp > 0). The overall strong separation was most pronounced in the lower (Fig. 1c, spectral frequencies <1,500 Hz, F1,419 = 731.4, P < 0.0001) as compared with the higher spectral frequency range. Furthermore, the tendency to base direction of pitch judgments on the implied f0 increased systematically with increasing number (N) of components (Fig. 1c, F2,838 = 498.5, P < 0.0001) and with decreasing order (n) of harmonic number (Fig. 1d, F2,838= 352.8, P < 0.0001). The separation of f0 and fSP listeners was such that a two-component stimulus (N = 2) had even stronger fundamental pitch character for f0 listeners as a four-component stimulus (N = 4) had for fSP listeners.

Figure 1: Psychometric testing and grouping.
figure 1

(a) Experimental design: participants were required to state the dominant direction of pitch shift between tone pairs. Solid lines represent the harmonics of the test tones and dashed lines the harmonics which are not physically present, such as the missing fundamental, indicated as number 1. (b) Bimodal distribution of fundamental (f0) and spectral pitch (fSP) listeners. (c,d) Perceptual changes as a function of the acoustic variables. The dependence on frequency (f) and lowest order of harmonics are depicted in separate curves for stimuli composed of N = 2, 3 and 4 adjacent harmonics. n indicates the order of harmonics in a complex tone; that is, an integer multiple of the fundamental frequency (fSP = n × f0). N is the number of physically present harmonics (on the left of each curve).

Full size image

Neural basis of pitch perception

MRI of brain structure and functional MEG of neural activity in response to harmonic complex tones were performed in a subgroup of 34 f0 and 53 fSP listeners and demonstrated a strong neural basis for type of pitch perception. The individual surfaces of all 87 left and right auditory cortices were segmented and reconstructed three dimensionally from the T1-weighted MRI slices (Fig. 2a). The pronounced oblique crescent-shaped gyral entity in anterior auditory cortex, including HG in its mediolateral extent and aSTG anterolaterally, was always identified by detection of the first complete Heschl's sulcus (cHS) as posterior boundary of HG and the first transverse sulcus (FTS) as the anterior boundary (Fig. 2b–d, red). HG may include incomplete duplications by a sulcus intermedius (SI)29,30,31,32 indenting locally its crown or a medial Heschl's sulcus (mHS), not reaching the lateral end. We next calculated the gray matter volume along the medial-lateral-anterior progression of this anterior gyrus (Fig. 2b, dashed lines), by marking the corresponding gray values successively in cross-sectional slices perpendicular to the orientation of HG and aSTG. Overall, the gray matter volume increased successively from mHG to aSTG (Fig. 2e,f). When comparing the volumes of the left and the right hemispheres, we found a characteristic asymmetry exclusively within the lateral aspect of HG irrespective of musical aptitude. The f0 listeners demonstrated a pronounced leftward asymmetry (Fig. 2e), whereas fSP listeners demonstrated a pronounced rightward asymmetry (Fig. 2f). The asymmetry started at the lateral border of mHG, peaked within the lateral edge of HG and was absent in aSTG.

Figure 2: Neural basis of fundamental and spectral pitch perception.
figure 2

(a) 3D reconstruction of an individual auditory cortex and auditory evoked activity (blue and red dipoles). (b–d) The top view of three individual 3D surface reconstructions of right auditory cortex shows the pronounced gyral entity including HG and aSTG (colored red), bordered anteriorly by the first transverse sulcus (FTS) and posteriorly by the first complete Heschl's sulcus (cHS). HG may include a sulcus intermedius (SI), indenting locally the crown of HG, or a medial Heschl's sulcus (mHS), not reaching the lateral end. Gray matter volume was successively calculated in cross-sectional slices along the medial-lateral-anterior progression. (e,f) Fundamental pitch listeners demonstrated a pronounced leftward asymmetry of gray matter volume and spectral pitch listeners a rightward asymmetry, peaking within the lateral one-third of HG. (g,h) Functional asymmetry of the auditory evoked P50m source activity of lHG in response to harmonic complex tones. P50m magnitude was relatively larger in the left hemisphere for f0 listeners and vice versa for fSP listeners, irrespective of musical aptitude. Prof, professional musician; non, non-musician, in all figures.

Full size image

This structural asymmetry was paralleled by a corresponding functional asymmetry. In particular, we performed an MEG study in which subjects were instructed to listen passively to harmonic complex tones covering the large parametric range of the pitch test. Auditory evoked fields were recorded continuously over both hemispheres. The source activity was calculated from the sensor distribution by modeling one equivalent dipole in each hemisphere. When fitted to the secondary P50m response peaking 50 ms after tone onset (P50m), the dipoles localized in the lateral portion of HG in most cases. Figure 2g,h shows the group-averaged source waveforms for professionals and non-musicians. P50m magnitude was larger in the left hemisphere for f0 listeners (professionals: factor 1.3 ± 0.2, F1,33 = 6.2, P < 0.01; non-musicians: factor 1.5 ± 0.2, F1,9 = 14.5, P < 0.01) and in the right hemisphere for fSP listeners (professionals: factor 1.3 ± 0.2, F1,29 = 22.4, P < 0.0001; non-musicians: factor 1.6 ± 0.3, F1,9 = 7.7, P < 0.01). No significant hemispheric asymmetry was observed for the early P30m (ref. 27) and the pitch-sensitive N100m response10,33.

Morphometry of auditory cortex subregions

Based on this specific connection of structural and functional asymmetry in relation to pitch perception, we identified objective criteria to define lHG. After normalization34, a grand-average auditory cortex map of all 87 brains was calculated from the individual landmarks of FTS, cHS and posterior border of planum temporale (PT)35. The top view of this map is depicted in Figure 3. Despite large individual differences in shape and progression of HG, the grand average over all 87 right and left auditory cortices demonstrated a completely symmetric organization with respect to angulation, extent and transition from HG to aSTG. However, the PT clearly showed the expected leftward asymmetry. Secondly, functional activation peaks in relation to pitch perception8,15,16,17,19,20,23,36 were plotted on this averaged auditory cortex map (Fig. 3, magnified HG). The main activation peaks concerning pitch perception are all confined to the lateral portion of HG (x-range: ± 45 to ± 65; y-range: −20 to 0). Melody-specific activation7,13,16,23 originates more anterior (y > 0) in aSTG. Our own findings on lateralization (Fig. 2e,f), the current knowledge of primary auditory cortex extent16,29,30,31,37,38,39,40 and the functional separation of adjacent pitch-sensitive areas14,15,16,17 were used as criteria to define objective boundaries of lHG: (i) a line perpendicular to the progression of HG at the mediolateral two-thirds to separate the region of mHG and lHG; (ii) a line at y = 0 to separate the region of lHG (y < 0, highlighted) and aSTG (y > 0, colored in Fig. 3). By strict application of these boundary definitions, the gray matter volumes of mHG, lHG, aSTG and PT were calculated. Only lHG demonstrated a strong leftward asymmetry in f0 listeners and rightward asymmetry in fSP listeners (Table 1). The individual auditory cortex morphology of 32 professional musicians, eight amateur musicians and eight non-musicians (Fig. 4) illustrates (i) the large individual differences with respect to angulation and progression of HG, (ii) the differences in structural left-right asymmetry of lHG in relation to pitch perception and (iii) the structural enlargement of the entire anterior crescent-shaped gyrus in musicians as compared with non-musicians. The frequency of duplications or sulci depends on hemisphere and on perceptual preference (Table 2).

Figure 3: Averaged landmarks of 87 auditory cortices (top view, standard stereotaxic coordinates34; line width of the dashed landmarks corresponds to averaged s.e.m.).
figure 3

The crescent-shaped anterior gyrus including HG and anterior supratemporal gyrus (aSTG) showed a completely symmetric shape and progression with respect to angulation, extent, duplications and curvature. A complete posterior duplication (PD) was considered to be part of the planum temporale (PT). The main activation peaks of functional imaging studies are plotted on the magnified map. Black open circles indicate the averaged localization of the auditory evoked P50m response in lHG measured by MEG. Pitch-specific activation localized in lHG, melody-specific activation more anterior in aSTG. In key at right, superscript numbers refer to references.

Full size image
Table 1 Gray matter volume of auditory cortex subregions
Full size table
Figure 4: Individual HG morphology.
figure 4

f0 listeners demonstrate a larger left lHG and fSP listeners a larger right lHG in most cases (lHG highlighted, red and blue arrows). The occurrence of sulci and duplications (SI, asterisks; mHS, '+' symbols; PD, black open circles) depends on hemisphere and pitch perception preference. Professional musicians and amateurs (A) showed greater gray matter volume of the entire anterior convolution including HG and aSTG (colored structure) than non-musicians (bottom).

Full size image
Table 2 Frequency of HG duplications and sulci
Full size table

Structural, functional and perceptual asymmetry

Corresponding to the definition of perceptual asymmetry, the structural and functional asymmetry of lHG was measured in terms of a 'structural asymmetry index' δs = (RlHG − LlHG)/(RlHG + LlHG) and a 'functional P50m asymmetry index' δf = (RP50m − LP50m)/(RP50m + LP50m). The correlation of δs versus δp was calculated separately by systematically including or excluding HG duplications: (i) all duplications including complete posterior duplications (PDs) were included (r = 0.47, P < 0.0001); (ii) right and left PD's were excluded; that is, cHS was always the posterior boundary (r = 0.77, P < 0.0001); (iii) only left PDs were excluded, (r = 0.81, P < 0.0001); (iv) all duplications posterior to cHS and mHS were excluded (r = 0.70, P < 0.0001); or (v) all duplications posterior to cHS, mHS and SI were excluded; that is, the anterior HG27 was considered (r = 0.38, P < 0.001). Overall, the correlation remained strongly robust irrespective of inclusion or exclusion of duplications. However, with respect to a symmetric definition of boundaries, the effect was most pronounced if cHS was always the posterior boundary (case 2, Fig. 5a). The scatter plot of P50m asymmetry (δf) versus pitch perception asymmetry (δp) demonstrated again a robust effect (r = 0.63, P < 0.0001, Fig. 5b). As a consequence, structural and functional asymmetry of lHG was strongly linked (r = 0.55, P < 0.001). However, if absolute magnitude of the neural substrate was considered instead of the relative hemispheric asymmetries, the correlation dropped to insignificance (P50m dipole amplitude versus gray matter volume of lHG: r = 0.04, n.s.).

Figure 5: Pitch perception preference and neural asymmetries.
figure 5

(a) Correlation of pitch asymmetry (δp) versus structural asymmetry (δs) of gray matter in lHG. (b) Correlation of pitch asymmetry (δp) versus functional asymmetry (δf) of P50m magnitude. The correlations are strong irrespective of musical aptitude.

Full size image

Influence of musical ability

With respect to the absolute magnitude of the neural lHG substrate, large group-specific differences were found, corroborating and extending earlier findings27. The crescent-shaped anterior convolution of auditory cortex including mHG, lHG and aSTG, colored in red (right hemisphere) and in blue (left hemisphere, Figs. 2b–d and 4), demonstrated strongly enlarged gray matter volume in professional musicians as compared with non-musicians (Table 1). The gray matter volume correlated significantly with musical aptitude as measured by the Advanced Measure of Music Audiation (AMMA) test27,41 (lHG: r = 0.71, P < 0.0001). Furthermore, the auditory evoked P50m response showed a fivefold larger magnitude in professionals as compared with non-musicians (Fig. 2e,f; professional: 25.1 ± 1.9 nAm, non-musician: 5.3 ± 1.2 nAm, factor 4.7 ± 0.8, F1,85 = 51.2, P < 0.0001) and correlated with the intensity of musical practice during the last ten years (r = 0.80, P < 0.0001). No significant correlation was found between any musical ability parameters and neural (δf, δs) or perceptual (δp) asymmetries. As a consequence, the correlations shown in Figure 5a,b remained strong for the non-musicians (δs versus δp: r = 0.75, P < 0.0001; δf versus δp: r = 0.67, P < 0.0001).

Discussion

Application of an auditory judgment task that is known to produce large perceptual differences across individuals1,2 in the investigation of a large sample of musicians and non-musicians enabled a systematic categorization in f0 and fSP listeners. The f0 listeners tended to base direction of pitch change judgments on the implied fundamental frequency, whereas the fSP listeners performed the pitch change judgment on the basis of the spectral envelope rather than fundamental frequency. By using MRI and MEG we found a strong neural basis of both types of pitch perception, which corroborated the functional specialization of left auditory cortex for rapid temporal23,25 and the right hemisphere for spectral23 processing.

The pitch test allowed systematic analysis of the influence of acoustic stimulus properties on pitch perception. First, the separation of f0 and fSP listeners was much stronger within a spectral frequency range below 1,500 Hz and decreased towards the higher frequencies. This may be reflected in the psychometric 'dominance principle'42, stating that the center of harmonic order which is dominant for pitch perception decreases from n = 5 (f0 = 200 Hz) to n = 1 (f0 = 2,000 Hz). Second, the general increase of f0 dominance with decreasing n and increasing N (Fig. 1c,d) corroborates the current knowledge of fundamental pitch saliency2,10,42. However, our data emphasize that pitch perception depends on both spectral envelope and fundamental frequency information with different weighting and cannot be explained by a simple one-to-one relationship between perceived pitch and fundamental frequency6,10. A functional separation of periodicity coding and spectral integration at the level of auditory cortex may account for the different pitch percepts, varying between individuals by up to three or four octaves, when the same sound was presented.

The large perceptual dissimilarity in pitch perception was paralleled by a strong inter-individual structural and functional variability in auditory cortex. To visualize this pronounced variation, which was anticipated a long time ago from myelogenetic studies40, we depicted for the first time the full shape and progression of HG and aSTG, forming together a crescent-shaped gyral entity of the anterolateral stream in human auditory cortex14. The huge morphological differences with respect to angulation, extent of HG and its oblique transition towards aSTG were superposed by a conspicuous increase of gray matter volume in musicans27,43 (Fig. 4). Averaged over 87 subjects, the shape was completely symmetric, whereas some morphometric studies show asymmetric average maps29,32. However, these and our studies are not sufficiently comparable owing to differences in sample size, definition of boundaries and extent of the region of interest. Furthermore, detailed morphometric analysis showed that the correlation between preference of f0 versus fSP perception and neural asymmetry was confined to lHG and was not present for the entire body of HG or aSTG. In particular, f0 listeners demonstrated an asymmetry favoring the left lHG in terms of both cortical gray matter volume and auditory evoked P50m activity, whereas fSP listeners showed the opposite asymmetry. This corroborated the functional specialization of lHG as a pitch processing center12. Our results imply a left-hemispheric specialization for (missing) fundamental pitch perception and a right hemispheric specialization for spectral pitch perception, consistent with a recent functional imaging study comparing the neural processing of spectral and temporal variation23. Furthermore, left auditory cortex is sensitive to short time scales (25–50 ms)25,44 and right auditory cortex to slower time scales (200–300 ms)25. The fundamental pitch (f0) of an instrumental sound (f0 > 25 Hz) reflects its periodicity6 T = 1/ f0, corresponding to time segments shorter than 40 ms. Thus, the existence of two pitch centers may facilitate the extraction of fundamental pitch in left auditory cortex and spectral pitch in right auditory cortex. Indeed, most professional musicians perceive simultaneously both fundamental and spectral pitch from an ambiguous tone, and the subjective differences are rather relative than absolute24. Here, these relative perceptual differences were found to correlate strongly to neural asymmetries, as anticipated by earlier studies on cerebral dominance45. Thus, a greater volume on the left may predispose one to hear the f0 in an ambiguous tone, and vice versa, a greater volume on the right may lead to a dominant perception of spectral pitch or single harmonics. A psychophysical study on patients with temporal lobe lesions demonstrates fourfold higher thresholds for determining the direction of pitch changes in patients with right hemisphere lesions that encroached on HG. This study concludes that detecting the direction of pitch changes may depend largely on the right HG18. However, there is no conflict with our results, because the magnitudes of pitch changes used in our study were largely above the f0 discrimination thresholds investigated in the lesion study. Thus, the latter may reflect in particular the accuracy of pitch direction judgment relative to the magnitude of f0 discrimination, irrespective of general left-right hemispheric lateralization effects in pitch perception.

The direct link between structural and functional asymmetries reported here seems to be confined to a local mechanism within the pitch center of lateral HG, present only in a small time range of 30–40 ms around P50m activity of the MEG recordings. The main activation peaks from comparable fMRI15,16 studies depict the activity within a much larger time frame of several seconds, including other activities which originate partially from lateral HG, in particular the pitch-sensitive N100 activity10,33 and the pitch onset response36 occurring about 130 ms after tone onset. Overall, functional asymmetries are not always observed in every task, subject, stimulus condition or specified processing level. In some cases or individuals the activation maps really are symmetric14,16,20. However, with respect to pitch- or melody-related tasks, the majority of functional imaging studies14,16,23,24 showed right-sided asymmetries. Multimodal functional imaging with professional musicians may help to clarify under which conditions the observed pitch asymmetries arise.

Our findings demonstrate a strong correlation between the relative hemispheric lateralization of structure and function and perceptual preference, as confirmed by morphometric31,45,46 and functional imaging studies25. In contrast, absolute magnitudes of the neural HG substrate depend on musical expertise consistent with previous studies27,33,47,48. At a more fundamental level, a recent post-mortem study observed characteristic asymmetries in auditory belt areas at the level of the underlying microanatomical architecture26. However, the exact relation between absolute changes at the microanatomical and macroanatomical level still remains unclear and needs further clarification. Use-dependent subcortical changes such as dendritic arborization changes or interdigitation of neuronal clusters26 may enhance the magnitude of synchronized postsynaptic potentials as measured by MEG in professional musicians without changing, however, the cortical thickness or volume of the underlying gray matter tissue. Likewise, a larger HG gray matter volume implies a larger neural network per se, independent of neural connectivity, and may reflect a greater potential of musical aptitude. A strong relation between absolute magnitudes of structure and function was observed at the early automatic processing level in anteromedial HG27 and obviously disappears in secondary auditory cortex. Post-mortem studies29,30 showed considerable variation between structure and function when comparing the boundaries of individual microanatomical structure and macroanatomical visible magnetic resonance landmarks. These results suggest that the functional areas do not correspond to areas defined on the basis of macroscopic boundaries. Nevertheless, when calculating micro- and macroanatomical probability maps, the corresponding centroids of the location were found to be almost identical14,16,29. As a consequence, despite large individual variability29, the medial two-thirds of anterior HG were considered a reliably good approximation of the anterolateral extent of primary auditory cortex16,24,37, as confirmed by cytoarchitectonic38, histochemical39 and myelogenetic40 studies. Overall, we conclude that the relative hemispheric lateralization of functional and structural size reflects the type of pitch processing irrespective of musical aptitude, whereas the absolute magnitudes of the neural HG substrate depends on musical expertise. Further studies may clarify whether the observed lateralization is linked to a preference for characteristic physical sound properties (in particular, the faster temporal structure of percussive sounds versus the slower time scale of sustained sounds25,49,50) therefore influencing musical instrument preference and musical performance.

Methods

Subjects.

A large sample of 420 right handed healthy subjects (125 professional musicians including members of the Royal Liverpool Philharmonic Orchestra28, 181 graduate students in music, 66 amateur musicians and 48 non-musicians) were recruited for the psychometric evaluation of pitch perception and evaluation of musical aptitude. A subgroup of 87 subjects participated in the MRI and MEG measurements (out of 51 professionals, including 21 members of the Royal Liverpool Philharmonic Orchestra, 16 amateurs and 20 non-musicians, 34 were f0 and 53 fSP listeners). Averaged over the groups, no significant differences in age, sex and head size were observed. Experimental procedures were approved by the relevant local research ethics committee.

Pitch test.

The pitch test included 144 different pairs of harmonic complex tones. Each tone pair consisted of two consecutive harmonic complex tones (duration 500 ms, 10-ms rise-fall time, interstimulus interval 250 ms). Each test tone comprised two, three or four adjacent harmonics, leaving out the fundamental frequency. Overall, the tone pairs were designed with six different upper component frequencies (293, 523, 932, 1,661, 2,960 and 5,274 Hz) chosen to be equidistant on a logarithmical frequency scale corresponding to the musical interval of a major ninth, beginning with D3 (293 Hz) up to C8 (5,274 Hz). The upper component frequency of both tones in each tone pair was identical to minimize the perception of edge pitch. Furthermore, the lowest presented harmonic number transitions (n1,min → n2,min) within a tone pair was one of the following four conditions: 2 → 3, 3 → 4, 4 → 6 or 7 → 9. Thus, the spectral components ranged between 146 and 5,274 Hz and f0 between 29 to 1,318 Hz. The magnitude of f0 pitch changes ranged between a factor of 1.1 (major second) to 1.8 (minor seventh); the fSP pitch changes of the lowest harmonic number ranged from a factor of 1.1 to 3.1. Thus, the magnitude of these pitch changes was considerably larger than pitch discrimination threshold18. By using additionally complete harmonic complex tones (nmin = 1) as reference tones (conditions 2 → 1, 3 → 1, 4 → 1 and 5 → 1), the pitch test allowed detection of octave-shifted fundamental pitches (for instance, one octave above f0). This case occurred only significantly for three-component stimuli within the higher spectral range (>1,000 Hz) and was not considered to be part of f0 perception. However, if fundamental pitch perception would mean both f0 and octave-shifted f0 perception, our results would not qualitatively change. All stimuli were presented binaurally in pseudorandomized order using a Hammerfall DSP Multiface System with a stimulus level of 50 dB nSL to avoid the interfering superposition of combination tones. Each tone pair was repeated once and the next new tone pair presented after a pause of 2 s. Subjects were instructed to select the dominant pitch direction or to answer according to the first, spontaneous impression, if either both directions were perceived at the same time or if tones lacked a clear pitch. Test duration was 22 min. All subjects were tested on an identical set of stimuli. A subgroup of 37 subjects repeated the pitch test about 6 months later and demonstrated strong individual re-test reliability (r = 0.96, P < 0.0001).

Morphometry.

The three-dimensional (3D) gray matter surface reconstructions of all individual auditory cortices were calculated from T1-weighted structural MRI data (Siemens, Symphony, 1.5-T) after segmentation using BrainVoyager software (Brain Innovation). All brains were rotated in direction of the antero-posterior commissural line and normalized by unfiltered transformation in Talairach space34. Using standard definitions of the anatomical auditory cortex landmarks29,30,31,32,35,37, the sagittal MRI slices of the individual auditory cortices were segmented along the Sylvian fissure to obtain PT, HG and aSTG. The inclusion range of image gray values was calculated in a normalized box around left and right auditory cortex. For gray matter surface reconstruction and morphometry, the 'gray value inclusion range' was defined individually from the intensity histogram of gray values for each left and right auditory cortex, by identifying (i) the half-amplitude side-lobe of the gray matter peak distribution towards cerebral spinal fluid and (ii) the saddle point between the gray and white matter peak. All gray value voxels inside this inclusion range were marked and used for 3D reconstruction and morphometry. The non-automated parts of this structural analysis (in particular, the identification of individual landmarks from the individual 3D surface reconstructions of auditory cortex) were obtained by observers who were blind to subject group and hemisphere.

HG subregions.

Here we used the most obvious and well-accepted definition of HG29,30 by identifying the first complete Heschl's sulcus (cHS) as its posterior and the crescent-shaped first transverse sulcus (FTS)29,30,31,32 as its anterior boundary. cHS was identified by virtue of having a clear lateral indentation (Fig. 2d), large mediolateral extent and pronounced depth, and divided auditory cortex in two parts: (i) an anterior auditory stream including HG and aSTG and (ii) a posterior stream including the PT. Based on normalization, the pronounced crescent-shaped gyrus anterior to cHS was subdivided systematically in mHG, lHG and aSTG. By using functional and structural criteria (Fig. 3), the y = 0 line was found to be an appropriate borderline to separate HG and aSTG. Individual extrapolation of FTS towards the lateral end of HG, as proposed by some morphometric studies31,32,37, was impossible in our sample owing to large individual differences with respect to angulation and asymmetric progression (Fig. 4). The extent of mHG was defined by the medial two-thirds of HG along the mediolateral direction of HG, similar to the estimated extent of primary auditory cortex16,29,30. lHG was the remaining part of the gyrus between aSTG and mHG. The gray matter volume of each specified subregion was calculated by marking and counting all included gray values of the individual gray matter peak distribution (see above). Finally, the volume of lHG was the difference between the volumes of HG and mHG.

Magnetoencephalography.

Using a Neuromag-122 whole-head MEG system, we recorded auditory evoked fields to twelve harmonic complex sounds covering the parametric range as used for the pitch test (f0: 100 and 500 Hz, lowest number of harmonics: 1, 4 and 10, complete spectrum and three adjacent harmonics). Subjects were instructed to listen passively to the sounds, each of which was presented 240 times in pseudorandomized order. Cortical responses were averaged for each frequency using the BESA program (MEGIS Software) and collapsed into an individual grand average for source analysis (3,600 averages). The source activity of the auditory evoked P50m response was separated from the earlier P30m and later N100m response by spatiotemporal source modeling27, using one equivalent dipole in each hemisphere. The fitting intervals were adjusted to the individual source waveforms in time intervals around the peaks defined by their half–side lobes. Signal strength was calculated for each peak relative to a 100 ms baseline.