INTRODUCTION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Recognition of sound patterns must be largely independent of level, but level may also contain important information, e.g., about sound sources. Furthermore sound targets should be recognized in spite of background sounds, even if the background sound levels exceeds that of the target. Therefore the relationship of pattern analysis and level coding is an important issue especially in auditory cortex (AC).

AC is generally assumed to harbor key mechanisms of pattern recognition in multiple functional fields (for reviews, see Ehret 1997; Kaas and Hackett 1998; Kaas et al. 1999; Merzenich et al. 1977; Pandya 1995; Rauschecker 1998; Scheich 1991; Schreiner 1992; Suga 1990). The patterns of which level-dependent responses are studied here in human AC are unidirectional frequency modulation (FM) sweeps, upward or downward. These auditory patterns are information bearing elements in numerous environmental sounds, species-specific vocalizations and human speech. A common observation in AC of animals is the high incidence of units with sensitivity to various parameters of FM, including range, direction, and speed of modulation (Heil and Irvine 1998; Heil et al. 1992a,b; Horikawa et al. 1998; Kowalski et al. 1995; Mendelson and Cynader 1985; Mendelson and Grasse 1992; Mendelson et al. 1993; Phillips et al. 1985; Schulze et al. 1997; Shamma et al. 1993; Tian and Rauschecker 1994, 1998). Therefore it may be assumed that human auditory cortex also shows specializations with respect to FM parameters. The present study addressing level coding of FM sweeps over a large range of intensities is part of an ongoing project with functional magnetic resonance imaging (fMRI) and electrophysiology to elucidate FM specializations of different areas of AC (Brechmann et al. 2000; Ohl et al. 2001).

Three specific questions were pursued in the present study. The first was whether primary and secondary fields in human AC might be differentiated on the basis of sound-level-dependent activation. Blood oxygen level dependent contrast (BOLD)-MRI is based on a local change in oxygenation level of blood due to metabolic demands of increased neuronal activity. The exact causality involving deoxyhemoglobin level and blood volume is still under discussion (e.g., Buxton et al. 1998; Hess et al. 2000; Ogawa et al. 1992), but the BOLD signal seems to be proportional to the average firing rate of neurons (Heeger et al. 1999; Rees et al. 2000). From electrophysiological studies in animals, it is known that primary and nonprimary auditory cortical fields can have different proportions of neurons with monotonic and nonmonotonic rate-level functions of firing (Phillips and Orman 1984; Recanzone et al. 2000). If this was also the case for human AC fields, a differentiation on the basis of BOLD signal intensity of these areas as a function of level of FM-sweeps might be expected because the firing rate of monotonic neurons increases over a larger range of stimulus intensities than the firing rate of nonmonotonic neurons.

The second question was whether areas of human AC possibly specialized for certain aspects of FM like range, direction, or speed of modulation are insensitive to level changes of these stimuli. This question relates to hypotheses about specializations of higher order sensory fields, namely that fields processing specific aspects of a pattern might be largely invariant for the processing of nonspecific aspects of the same stimulus. For instance, Frackowiak (1994) argued on the basis of a positron emission tomography (PET) study by Price et al. (1992) that increasing word repetition rate should lead to an increase of activation of auditory cortex on Heschl's gyrus but not in Wernicke's area. The insensitivity of Wernicke's area to word repetition rate was taken as a sign of specialization of this area for word comprehension invariant of segmentation speed. As caudal areas of human AC appear to be sensitive for certain aspects of FM like sweep direction (Brechmann and Scheich 2000) or spectral motion (Thivard et al. 2000), we tested whether activation of these areas showed invariance with respect to FM level.

The third question relates to the previous finding that an area anterior of Heschl's gyrus was more specifically involved in a foreground-background decomposition task than any other auditory area on the dorsal surface of the temporal lobe (Scheich et al. 1998). In that study, notes of different musical instruments had to be compared when masked by a much louder background of a continuous sinusoidal FM. A similar situation was introduced in the present experimental design by including a very low sound pressure level (36 dB) of FM sweeps that were 12 dB below the continuous scanner noise. This allowed to address questions of background noise which are of general relevance for fMRI studies (Bandettini et al. 1998; Hall et al. 2001; Shah et al. 1999; Talavage et al. 1999; Yang et al. 2000) and more specifically whether activation in any AC area is differentially influenced by stimulus levels higher and lower than the scanner noise.

The present work addresses these questions with special low-noise (48 dB) fMRI (Scheich et al. 1998). Rising and falling FM sweeps were varied in frequency range and over a large range of sound pressure levels from 36 to 102 dB. A detection task served to maintain attention. As in our previous fMRI work (Baumgart et al. 1999; Gaschler-Markefski et al. 1998; Scheich et al. 1998), we parcellated AC into four landmark-related territories for an individual analysis of each brain.

	METHODS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Subjects

Ten right-handed subjects (Edinburgh Handedness Inventory) with normal hearing participated in this study. Subjects (8 females, 2 males; mean age, 26 yr) gave written informed consent to the study which was approved by the ethics committee of the University of Magdeburg.

Stimuli and task

Thirty-two different linearly FM tones (16 upward and 16 downward), each covering one octave in 500 ms (500-1,000 Hz, 600-1,200 Hz in steps of 100 Hz up to 2,000-4,000 Hz and the reverse) were presented at five different SPLs (36, 48, 72, 96, 102 dB). Pure tones of 500-ms duration varied between 800 and 1,600 Hz in steps of 100 Hz and served as infrequent target tones the detection of which had to be reported by mouse click. The FM and pure tones had a linear rise/fall time of 10 ms.

Each block at a given sound pressure level consisted of 45 randomized stimuli (repetition rate, 1 Hz), including seven or eight pure-tone targets.

One experimental session consisted of 12 alternating stimulus and "silence" blocks of 45 s resulting in 18-min total duration. In the first session, stimulus blocks were presented at 48, 72, and 96 dB SPL with four repetitions each. In the second session, 1 wk later, stimulus levels were 36, 72, and 102 dB SPL. The order of blocks of stimulus level was randomized with no immediate repetition of the same level.

The study was split into two experimental sessions to avoid weariness and head motion of the subjects due to long scanning periods. The second session was designed to test extreme sound levels, both very high (102 dB) and very low (36 dB). To assess comparability, the 72-dB sound level was examined in both sessions.

Scanning procedure

fMRI experiments were carried out on a BRUKER 3T/60 head scanner equipped with a quadrupolar birdcage head coil. Pilot scans were used for orientation of three contiguous 6-mm slices covering the superior temporal plane in both hemispheres by following the course of the sylvian fissure on both sides as closely as possible. One hundred twenty functional images of each slice were collected in 18-min runs using a gradient echo sequence [echo time (TE), 30.7 ms; repetition time (TR), 161 ms; flip angle, 15°, in-plane resolution, 2.8 * 2.8 mm]. With these settings, a single volume containing three slices required a scan time of 9 s. The beginning of each stimulus and silence block coincided with the acquisition of a volume.

The relatively slow conventional fast low angle shot (FLASH) sequence was modified by using long gradient-ramp rise time (6,000 µs), which reduced the scanner noise (see following text). In contrast to echo planar imaging (EPI) sequences that acquire a complete functional image during each repetition, the gradient echo sequence acquires one line in k-space during one repetition (TR). High T1-contrast imaging was used to obtain anatomical landmarks and immediately followed the fMRI. The subject's head was fixated with a vacuum cushion with attached ear muffs containing the fMRI compatible headphones (Baumgart et al. 1998).

Calibration of sound pressure level and measurement of scanner noise

For sound pressure level (SPL) measurements of the stimuli and the scanner's gradient noise, the ear muffs containing the headphones (Sennheiser, model HD 475) were sealed together with a condensor microphone (Brüel and Kjaer, model 4155) positioned in the middle. The entire set was put into the vacuum cushion and placed inside the scanner at the position corresponding to normal operation. Measurements were made with a sound level meter (Brüel and Kjaer, model 2233) using the fast time weighting (125-ms time constant) and the A-frequency weighting.

Pure tones in the frequency range of FM sweeps used in the present study served to measure the frequency response curves of the headphones at each SPL without the scanner noise. According to these frequency response curves the sound pressure level of each FM tone was equalized. If necessary, the peak SPL of each FM tone was then adjusted to 36, 48, 72, 96, and 102 dB.

The reduced scanner noise as measured with the same arrangement had a peak SPL of 48 dB and an average root mean square of 40 dB SPL with most energy <2 kHz. This sound was continuously present throughout the 18-min duration of an experimental session.

Data analysis

Subject movement was monitored using the AIR package (Woods et al. 1998) but images were not corrected for movement. The output of AIR was merely used to asses the amount of movement. Continuous movements exceeding one voxel in at least one direction were used as criterion for data exclusion. This was the case for one subject in experiment 2. After motion analysis, image matrix size was increased to 128 *128 by pixel replication followed by smoothing with a Gaussian filter [full-width half maximum (FWHM) = 2 pixel, Kernel = 5 pixel]. Functional activation was analyzed by correlation analysis (Bandettini et al. 1993) to obtain a statistical parametric map. A simple trapezoid function served as correlation vector, roughly modeling the expected BOLD response. Because of the rather coarse temporal resolution (each volume of 3 slices was acquired in 9 s), a more exact model is not appropriate. The first image of each stimulus and silence block was set to half-maximum values. This takes into account that the full evaluation of the BOLD response and the return to baseline takes a few seconds (Bandettini et al. 1997; Buxton et al. 1998; Rosen et al. 1998). The remaining images acquired during silence were set to minimum values, and the remaining images acquired during stimulus periods were set to maximum values. In both sessions, activated voxels (P < 0.05) in each slice were attributed to one of the four territories T1a, T1b, T2, and T3 on an individual basis (Fig. 1). For each of these territories, activated volumes (AV), i.e., the number of activated voxels and the average BOLD signal intensity (SI) of these voxels, i.e., the percent change in image signal between stimulus and silence conditions were determined.

View larger version (130K): [in this window] [in a new window]   Fig. 1. Relationship between activation and anatomical landmarks used for parcellation of auditory cortex. A: anatomy of the surface of the right temporal lobe of 2 individual subject (2-mm slice thickness). Note, e.g., the different width and steepness of Heschl's gyrus. Labels indicate landmarks used to divide auditory cortex (AC) into different territories. CS, circular sulcus; HG, Heschl's gyrus, HS, Heschl's sulcus, PT, Planum temporale, S tr 1, first transverse sulcus, SI, intermediate sulcus. B: patterns of functional magnetic resonance imaging (fMRI) activation of the same 2 subjects obtained with pure tone stimuli vs. silence. (Pure tones of 100-ms duration varied in frequency among 0.5, 1, 2, and 4 kHz and were presented in blocks of 25-s duration with a repetition rate of 5 Hz.) The stripe-like pattern of activation in T1b follows the rostromedial slope of HG, and in T2, it is centered on HS. These 2 stripes of activation remained separated from 1 another despite the differing width of HG in the 2 subjects. Note the more patchy organization of activation in T3. C: definition of auditory territories in the same 2 subjects. T1a (yellow), T1b (red), T2 (green), T3 (blue).

Statistical analysis of these data were performed using the nonparametric tests for matched pairs by Wilcoxon and for trend analysis by Page.

Definition of auditory territories

As shown in Fig. 1 and conceptualized in the DISCUSSION, territories were approximated by wedge-shape regions of interest (ROI) in the plane of anatomical MR images using the same anatomical landmarks in each individual brain. To verify these landmarks anatomical images were fit into the three-dimensional anatomical dataset of each subject. This allowed to view the landmarks in all three perspectives and at a higher spatial resolution (1 × 1 × 1.5 mm).

T1b borders were medially the circular sulcus, anteriorly the first transverse sulcus and caudolaterally the roof of Heschl's gyrus or the intermediate sulcus when present. The caudally adjacent T2 was centered on Heschl's sulcus, thus including the posterior rim of Heschl's gyrus and a similar area behind the sulcus on the anterior planum temporale or on the second Heschl's gyrus when this was present. T3 covered the caudally adjacent intermediate planum temporale including the most rostral part of the supramarginal gyrus. T1a was defined to be rostral to the first transverse sulcus. In Fig. 1, it is shown how separate clusters of activated voxels fall into the corresponding territories.

	RESULTS

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Data on FM level dependence of the BOLD effect were obtained in two sessions with the same subjects: first, at listening levels of 48, 72, and 96 dB SPL and second, testing more extreme levels, i.e., at 36 dB (12 dB below scanner noise level), at 72 dB as a reference to the preceding session, and at 102 dB. The performance in the pure-tone detection task was 94.1 ± 2.4% at 48 dB, 95.6 ± 0.7% at 72 dB, 97.2 ± 0.2% at 96 dB, and 97.2 ± 0.2% at 102 dB, whereas at 36 dB, it dropped to 81.1 ± 6.6%.

In the first session, the sum of all voxels in the four defined territories increased significantly with increasing stimulus-level (P < 0.05) in left and right AC with a steeper increase on the left side (Fig. 2). Separate analyses of level-dependent activations in the territories T1a, T1b, T2, and T3 are summarized in Table 1.

View larger version (18K): [in this window] [in a new window]   Fig. 2. Activated volume (AV) of the combined auditory territories of each hemisphere during the 1st session. AV of both hemispheres significantly increased between 48 and 96 dB (*). Note the steeper increase of AV in the left compared with the right hemisphere indicated by ascending lines.

The correlation of AV with stimulus-level reached significance in left and right T1b (comprising primary AC) and left T2 but not in right T2 (Fig. 3). Furthermore, SI in left T1b increased significantly with stimulus level as shown in Fig. 4. Neither AV nor SI in territories T1a and T3 showed any significant increase with stimulus-level.

View larger version (13K): [in this window] [in a new window]   Fig. 3. AV of the auditory territories of the 1st session. AV significantly increased from 48 to 96 dB in left and right T1b and left T2 indicated by ascending lines.

View larger version (21K): [in this window] [in a new window]   Fig. 4. Signal intensity (SI) in T1b and T2 of the 1st session. SI of left T1b significantly increased with increasing stimulus level (P < 0.001, ***). Lateralization of SI toward right T2 was significant at 48 and 72 dB and showed a trend at 96 dB.

A direct comparison of hemispheric differences revealed higher average SI values in right territories T1b and T2 at all levels and significant lateralization toward the right side for T1b at 48 dB and for T2 at 48 and 72 dB and as a trend at 96 dB (Fig. 4, Table 2).

An example of the activation pattern in AC of an individual subject during the first session is given in Fig. 5A.

View larger version (91K): [in this window] [in a new window]   Fig. 5. Pattern of auditory cortex activation of an individual subject during the 1st session (A) and (same subject) during the 2nd session (B). Significant activation (P < 0.05) was color coded for each territory (T1a, yellow; T1b, red; T2, green; and T3, blue). Note the strong activation at 36 dB in T1a of the left hemisphere.

In the second session, some decrease of mean AV and SI relative to the first was observed at the reference level of 72 dB SPL in the left and right hemisphere (compare Figs. 2 and 6); this was significant for SI (P < 0.01). This decrease of activation on repetition of relatively simple detection tasks is a common observation in our laboratory and may be related to decreased attention levels. This result precluded pooling of raw activation data of the two sessions.

View larger version (16K): [in this window] [in a new window]   Fig. 6. AV of the combined auditory territories of each hemisphere during the 2nd session. At 36 dB, AV of the left hemisphere was larger than at 72 and 102 dB. This difference was not significant.

Activation during the second session showed an unexpected effect at the 36 dB stimulus level: AV of the left hemisphere at this low level was larger than at 72 dB or at 102 dB (Fig. 6). Territorial analysis revealed this effect to be significant for left T1a (P < 0.05) but not for the other territories of that hemisphere (P > 0.1) (Fig. 7). Activated volume at 36 dB in left T1a was also significantly larger than in right T1a (P < 0.05)

View larger version (11K): [in this window] [in a new window]   Fig. 7. AV of the auditory territories of the 2nd session. At 36 dB, AV of left T1a was significantly larger than at 72 and 102 dB.

Analysis of hemispheric differences in SI summarized in Table 2 showed significant lateralization toward the right hemisphere for T3 at all sound intensities (Fig. 8) and for T2 at 72 dB. The laterality effect at 72 dB for T2 was already observed in the first session.

View larger version (14K): [in this window] [in a new window]   Fig. 8. Mean blood oxygen level dependent contrast (BOLD) signal intensity (SI) of T3 during the 2nd session. SI of right T3 activation was significantly larger than that of left T3.

An example of the activation pattern during the second session of the same subject as in Fig. 5A is given in B.

	DISCUSSION

TOP ABSTRACT INTRODUCTION METHODS RESULTS DISCUSSION REFERENCES

Parcellation of human auditory cortex

Because a more detailed spatial analysis of the auditory cortex is fundamental not only to the present results, it will be discussed in detail in this section. In animals, the functional parcellation of AC into multiple distinct fields in each individual is largely based on functional gradients (e.g., tonotopic organization) or areas of neuronal specialization (for reviews, see Ehret 1997; Kaas and Hackett 1998; Kaas et al. 1999; Merzenich et al. 1977; Pandya 1995; Rauschecker 1998; Scheich 1991; Schreiner 1992; Suga 1990). This type of distinction between fields by physiological properties of neurons is not available for human AC. Therefore we used an indirect approach with the following rationale: the multiple separate foci of fMRI activity seen on the surface of the temporal lobe were related to a framework of anatomical landmarks on each lobe. This allowed integration of these foci into previously described cytoarchitectural schemes of AC parcellation which involved the same landmarks. The underlying assumption was that physiologically defined functional fields each presumably remain within the boundary of a primary or secondary cytoarchitectural area even though at this stage of knowledge it is not clear whether such an area contains several fields. The phenomenon that multiple foci of metabolic activity relate to structurally and electrophysiologically distinct fields was previously noticed in animal AC with other methods that also measure energy consumption, i.e., with 2DG mapping (Budinger et al. 2000a,b; Scheich et al. 1993; Thomas et al. 1993) and with optical recording of hemoglobin signals (Hess and Scheich 1996).

The described approach frames multiple foci of fMRI activation as candidates of functional fields using the following landmarks identifiable by structural MR: Heschl's gyrus (HG), circular sulcus (CS), the first transverse sulcus rostromedial to HG, and Heschl's sulcus (HS) caudolateral to HG (Fig. 1). There is large gross-anatomical variability of the human temporal lobe (Leonard et al. 1998; Morosan et al. 2001; Penhune et al. 1996; Rademacher et al. 1993, 2001). But this seems to chiefly concern the cortical space covered by these landmarks and the geometric relationships between them (see Fig. 1) rather than their mere presence. A large body of histological work on human AC relates histologically defined areas to these landmarks (Beck 1930; Braak 1978; Brodmann 1909; Flechsig 1908; Galaburda and Sanides 1980; Hopf 1954; Rademacher et al. 1993; Rivier and Clarke 1997; von Economo and Horn 1930). Because of the reliable presence of these landmarks, it may be assumed that they are meaningful entities for structural parcellation. An empirical support for their relevance seems to be that focal patterns of fMRI activation geometrically relate to these landmarks (see Fig. 1). Consequently the whole approach solves at least one major problem concomitant with multiple foci of activation in the presence of anatomical variability namely to establishing correspondence between activation foci across individuals.

In summary, areas that are differentiable histologically may also be distinct functionally. Assuming they are, the relationship between histological areas and gross-anatomical landmarks should resemble the relationship between functionally distinct areas and landmarks. There is certainly intersubject variability in the spatial relationship between histologically defined areas and gross anatomical landmarks. Nevertheless as a backbone of parcellation, two major relationships appear to hold fairly uniformly across subjects: koniocortex is bounded rostromedially by the first transverse sulcus and root and belt regions of AC form parallel bands flanking the "stripe-like" koniocortex. These major relationships form the basis for the parcellation strategy of the present study.

Our imaging strategy was to optimally relate anatomical landmarks to fMRI activation patterns by orienting the slices in each subject parallel to the dorsal surface of the temporal lobe in both hemispheres. With this imaging strategy, most activation patterns in AC fall into four stripe-like or wedge-shaped territories emanating medially from the circular sulcus (Fig. 1). In the following, a rationale for the preliminary dissection of AC into four territories is provided.

One stripe-like activation on HG attributed to primary AC in this and previous work from our laboratory was characteristically located on the rostromedial slope of HG. This appears to be a significant feature because the most stringent definition of primary AC in this location was already given by Flechsig (1908) and Pfeifer (1920). Their studies of early ontogenetic myeloarchitecture revealed exclusively the thalamic projection unconfounded by the later developing association systems. Similar topographic definitions were provided by cytoarchitectonic features [BA41 (Brodmann 1909), TC₁ (von Economo and Horn 1930), KAm (Galaburda and Sanides 1980)]. Tonotopic single-unit data from recordings in patients were obtained in this restricted medial area (Howard et al. 1996). The span of the tonotopic gradient suggests that HG must contain at least one additional functional area along its rostrolateral course, also suggested by cytoarchitectural work (Morosan et al. 2001). A territory covering activation throughout the rostromedial slope of HG was termed T1b.

A second stripe-like activation caudal to that on HG is found in HS mostly reaching on the rostral planum temporale (see also Dhankhar et al. 1997; Di Salle et al. 2001). This location seems to be in BA42 (Brodmann 1909), in Area TBC (von Economo and Horn 1930), and in parakoniocortex (Galaburda and Sanides 1980). The corresponding territory centered on HS was termed T2. It may correspond to the lateral belt area of macaques (Kaas and Hackett 1998 Kaas et al. 1999; Morel et al. 1993; Rauschecker 1998).

A functional dissociation of areas presumably corresponding to T1b and T2 has been found in a study of middle latency auditory evoked potentials to 1-kHz tone bursts through intracerebral recording in auditory cortex of humans (Liegeois-Chauvel et al. 1994).

The large remaining surface of the planum temporale caudal to T2 was tentatively defined as territory T3. It includes the posterior wall of the sylvian fissure, which presumably exhibits the same cytoarchitecture as the most caudal and lateral part of the superior temporal gyrus (temporoparietal cortex: Galaburda and Sanides 1980). It may therefore correspond to the parabelt area in macaques. The fMRI activation even though it may reach on the lateral surface of the superior temporal gyrus is rarely stripe-like but typically shows several clusters of activity. This may relate to functional specialization as already shown for a small motion-sensitive area within T3 (Baumgart et al. 1999) and similarly for voice-selective areas on the superior temporal gyrus (Belin et al. 2000).

An additional area on the rostral surface of the superior temporal lobe anterior of Heschl's gyrus showed distinct fMRI activation (Gaschler-Markefski et al. 1998; Scheich et al. 1998) and has been termed T1a. It may correspond to the rostral prokoniocortex of Galaburda and Sanides (1980) and to medial belt areas in macaques.

Increase of activation with sound level

Increasing activation of human AC with increasing pure tone level has been described with different methods, including PET (Lockwood et al. 1999), fMRI (Hall et al. 2001; Jäncke et al. 1998) and event-related 100-ms deflections in electroencephalographic (EEG) and magnetencephalographic (MEG) recordings (Hegerl et al. 1994; Vasama et al. 1995). With "sparse sampling" fMRI Hall et al. (2001) found a bilateral increase of activation (AV and SI) for a pure tone (300 Hz) and with respect to SI for a harmonic tone (f₀ = 186 Hz) as a function of level between 66 and 91 dB SPL defining the whole superior temporal gyrus as region of interest. A standard EPI fMRI study by Jäncke et al. (1998) analyzed level-dependent activation in two AC areas at sound pressure levels of 75, 85, and 95 dB for pure tones and consonant-vowel syllables. Activations were analyzed in two ROIs attributed to Brodmann areas 41/42 together and BA 22 using the Talairach space. For the combined activation of BA 41 and BA 42, they described a level-dependent bilateral increase of activated volume but not of BOLD SI for tones and syllables.

In contrast to previous studies, the present study exploited the high spatial resolution of fMRI to differentiate four AC areas. With regard to the increase of AV, the result of Jäncke et al. (1998) is in accordance with the level-dependent increase of AV in two of our territories, in T1b (presumably corresponding to BA41) and T2 (presumably corresponding to BA42) using a different type of stimuli. But by analyzing T1b and T2 separately, we additionally found level-dependent BOLD SI and also hemispheric differences of level dependence in these territories. SI in left T1b but not in left T2 increased significantly with sound level. Furthermore activated volume was level dependent in left but not right T2.

A recent fMRI-study by Goodyear and Menon (1998) analyzed activated volume and BOLD SI in human visual cortex in response to varying luminance contrast. They found that activated volume in primary and secondary fields increased with increasing luminance contrast but only primary visual cortex activation additionally showed an increase in BOLD SI. This finding in the visual domain is analogous to our results in the auditory system showing that primary (T1b) and secondary (T2) auditory cortex can be differentiated on the basis of level-dependent increase in BOLD SI.

An interpretation of such fMRI results in the light of electrophysiological studies in animals is possible if neuronal activity and BOLD signal are correlated. First evidences for such a correlation between average spiking activity in monkey visual cortical areas and BOLD signal of homologue human areas have recently been demonstrated qualitatively (Heeger et al. 1999) and quantitatively (Rees et al. 2000), even though this relationship is not yet fully understood.

At present, there are two hypotheses of level coding within the auditory cortex which are not mutually exclusive (for review, see Clarey et al. 1992). First, increasing SPL recruits an increasingly larger proportion of neurons with monotonic rate-level functions and increases the discharge rate in responsive cells that have not yet reached saturation. Within a given fMRI voxel, both the increase in discharge rate and the recruitment of neurons would presumably contribute to an increase in BOLD SI. Additionally, due to the recruitment phenomenon, more activated voxels may be expected. The second hypothesis relies on neurons with nonmonotonic rate-level functions with different best SPL and smaller dynamic range compared with monotonic neurons. Thus stimuli with different SPL would lead to a recruitment of different populations of such neurons. The BOLD SI of a given voxel including only such neurons would presumably not change dramatically. From electrophysiological studies, it is known that primary and nonprimary auditory cortical fields can have different proportions of monotonic respectively nonmonotonic neurons (Phillips and Orman 1984; Recanzone et al. 2000). If there exists a correlation of average spiking rate and BOLD signal, the preceding considerations on level coding of neurons might explain the difference in level-dependent activation of T2 compared with T1b.

In conclusion, the results of the present study show that primary (T1b) and secondary fields (T2) in human AC can be differentiated on the basis of sound-level-dependent BOLD SI.

Level-independent activation of some territories

No systematic changes in activation above 36 dB (neither for SI nor AV) were observed by trend analysis in the anterior territory T1a and T3 on the planum temporale. To be sure that these territories did not exhibit any increase in activation with increasing level, we directly compared AV and SI activation parameters at 48 and 96 dB. This did not reveal significantly stronger activation at 96 compared with 48 dB for these territories. Thus both T1a and T3 might not be strongly concerned with the processing of this general stimulus feature i.e., sound pressure level.

Jäncke et al. (1998) found a level-dependent increase of AV on the planum temporale in both hemispheres for consonant-vowel syllables but not for tones. It is possible that in parts of the planum temporale more neurons are recruited with increasing loudness of speech that contain many complex acoustic parameters. A level-invariant representation of linear FM sweeps as found here is thus not excluded by these results.

According to the hypothesis formulated in the INTRODUCTION level invariant activation of these areas might point to a special role of these areas for other aspects of FM like sweep direction, steepness, or bandwidth. This hypothesis followed an analogous argument used in a PET study of word processing comparing activation on Heschl's gyrus and in Wernicke's area (Frackowiak 1994). There was a linear increase in activation with word repetition rate on Heschl's gyrus versus activation in Wernicke's area that was independent of repetition rate (Price et al. 1992; but see also Dhankhar et al. 1997). The latter was taken as a sign of specialization for word processing invariant of segmentation speed.

Activation invariant to a nonspecific stimulus parameter like intensity or repetition does not necessarily mean that such an area is specialized for other stimulus parameters but might serve as an indication.

Hemispheric differences

We describe systematic left-right differences of level-dependent activation with FM that were not reported in the study by Jäncke et al. (1998) with levels of tones and syllables. We found a steeper overall increase of activated volume in the left hemisphere compared with the right hemisphere. The increase in AV of T2 was significant in the left but not in the right hemisphere. Furthermore the increase of AV at 36 dB was only seen in territories of the left hemisphere with a significant effect in left T1a. The overall increase of BOLD SI with level in the primary area T1b was only significant on the left side and steeper than on the right side. Together these results suggest that left auditory cortex is more concerned with level coding of FM. This fine tuning to levels in the left AC is contrasted by other activation results in favor of right AC areas. SI for right T2 was stronger than for left T2 for sound levels of 48 and 72 dB. The same right lateralized effect was significant for T3 at all stimulus levels of the second session.

Thus as argued in the preceding text, the right hemisphere may harbor mechanisms for the analysis of FM-specific parameters independent of the general sound parameter level. Note that our experimental design of level variation included a large mix of upward and downward modulations of FM in different frequency ranges. This is favorable for activating various FM-sensitive and -selective mechanisms of neurons in large parts of AC and consequently for detecting any invariance formation for level.

The reasoning on possible right AC specialization for FM but not level is in line with other observations. There is deteriorated discrimination of rising versus falling pitch contour in patients with right but not left temporal lobe lesions (Zatorre 1988). Furthermore it was shown in a magnetencephalographic study of linear frequency sweeps (1 octave per 3, 30, 300 ms) at the beginning or end of a constant pure tone that the N100 responses to the fast and intermediate FM at tone end were stronger and earlier over right AC (Pardo et al. 1999). An fMRI-study by Binder et al. (2000) showed bilateral activation for stepwise varying tone frequencies (referred to as FM tones). This stepwise variation of tone frequencies is discussed as one important difference to speech sounds that are characterized by rather continuous FMs. Thus the bilateral symmetric activation by these stepwise varying tone frequencies in that study is not in contrast to a possible right AC specialization for continuously frequency modulated tones as used in the present study.

Several animal and human studies from this laboratory support that right AC is more specialized for FM parameters than left AC. In the Mongolian gerbil, it was found that right but not left AC lesions impaired discrimination learning of rising versus falling FM and the recall of trained discrimination (Wetzel et al. 1998). Additionally, in right gerbil AC physiological mechanisms pertaining to the categorization of these stimulus classes were found (Ohl et al. 2001). A recent fMRI study showed lateralized right human AC activation for a directional discrimination task with a similar stimulus set as used in the gerbil (Brechmann and Scheich 2000).

Effects at stimulus level below background noise

The significantly stronger activation of left T1a at 36 dB compared with 72 and 102 dB is in support of our previous hypothesis of an involvement of this area in foreground background decomposition tasks (Scheich et al. 1998). The present experimental design has some formal equivalence to the tasks in the cited study. Here the occurrence of tone targets (in the pattern of the FM tones) had to be detected in the presence of a continuous background sound from the scanner that was louder than the foreground. Thus the conditions at 36 dB were not simple masking, but a foreground-background decomposition of different auditory patterns as reflected in the lower hit rate at 36 dB during this otherwise very simple detection task. That rostral temporal lobe areas might be involved in cocktail party problems (Cherry 1953) has long been suggested by corresponding deficits in patients with rostral temporal lobe lesions (Efron et al. 1983). Such deficits are not necessarily interpretable in terms of impaired sound localization because various acoustic cues can be used to solve a cocktail party problem only one of which is the spatial characteristic of sound (Yost et al. 1996).

We also hypothesized a stronger activation of T1a in the right hemisphere, which was not the case. This might also be related to the preceding discussed specialization of the right hemisphere in FM processing because the task was not detection of FM but of deviant pure tones. But this has to be confirmed in future studies. Furthermore, all other territories of the left AC were strongly activated at 36 dB although not significantly. The difficulty of the task to monitor the stimuli at a level below the scanner noise suggests a connection to attention mechanisms (for reviews, see Alho 1992; Woldorff 1999). In event-related potentials and event-related magnetic fields, attention effects have been found as early as 30-40 ms after stimulus onset and to peak before the N 100. The late attention effects coincide with the P 200 and their sources were localized ~1 cm anterior to those of the early effects (Arthur et al. 1991; Rif et al. 1991). The separable early and late phase and the spatial relationships seem compatible with the prominent reflection of the 36 dB effect by T1a, i.e., in anterior AC. Thus general attention mechanisms might be one reason for the global increase of activation in the left hemisphere at 36 dB. However, the strong reflection of this level effect in left T1a together with the level independence of T1a at higher stimulus intensities points to an involvement in particular auditory problem solutions.