Analysis of variance (ANOVA) did not reveal significant differences between N1m baseline-to-peak amplitudes after stimulation with the vowel /a/ and N1m amplitudes after stimulation with a matched sine tone (Figure 1C and 1D, Tukey's multiple comparison test, p > 0.05 for the comparisons N1m₁(vowel) vs. N1m₁(tone), N1m₂(vowel) vs. N1m₂(tone), N1m₃(vowel) vs. N1m₃(tone) and N1m₄(vowel) vs. N1m₄(tone)). Thus, a grand average of all responses to vowels and tones was calculated for each individual subject. The combined analysis of both conditions yielded a better signal-to-noise ratio, resulting in a more robust source localisation and a more accurate calculation of the amplitude over time. All further analyses were performed with the grand average across individuals and conditions.

A repeated-measures ANOVA showed that the N1m amplitude decreased with the number of responses (Figure 2A, stimulation with four successive stimuli in a train. F = 40.14, p < 0.0001). Post hoc tests using Tukey's multiple comparison test indicated that the amplitudes of the second, third and fourth N1m were significantly smaller than the amplitude of the first N1m (p < 0.001 for each comparison). The relative amplitudes of the second, third, and fourth N1m waves (each normalised to the peak amplitude of the first N1m) are depicted in Figure 2C.

The N1m peak latency significantly increased with the number of responses (Figure 2B, Friedman test, a non-parametric repeated measures ANOVA, p < 0.0001). Post hoc tests using Dunn's multiple comparison test demonstrated that the latency of the fourth N1m was significantly longer than the latency of the first (p < 0.05) and the second N1m (p < 0.001). The relative latencies of the second, third, and fourth N1m waves are shown in Figure 2D.

The concentrations of N-acetylaspartate (NAA) (median 12 institutional units (IU), range 6–20 IU), choline-containing compounds (Cho) (median 1.4 IU, range 0.9–3.0 IU) and the glutamine/glutamate pool (Glx) (median 22, range 4–40 IU) demonstrated considerable intersubject variability.

The amplitude of the first N1m peak was significantly associated with the concentration of NAA (Figure 3, linear regression, R² = 0.369, F(1,13), p = 0.016) and Cho (Figure 4, R² = 0.287, F(1,13), p = 0.040). No significant association was found between Glx and the amplitude of the first N1m. Linear regression analysis revealed no significant associations between the relative amplitude of the second N1m response and NAA, Cho or Glx. However, there was a trend for higher Glx concentrations in individuals with higher relative N1m amplitude (R² = 0.26, F(1,13), p = 0.054) (Figure 5).

As this study included participants between 18 and 68 years, we addressed the question of whether age might influence the measured variables and the calculated statistical associations. No significant associations were found between age and the tested variables (NAA, Glx, Cho, first N1m amplitude, relative second N1m amplitude; data not shown). In addition, we reanalysed the data after the exclusion of the two oldest subjects (60 and 68 years old). For this subgroup (n = 13, mean age = 27 years), the amplitude of the first N1m peak was associated with the concentration of NAA (R² = 0.324, F(1,11), p = 0.042) and with the concentration of Cho (R² = 0.437, F(1,11), p = 0.014). However, no significant association was found between the relative amplitude of the second N1m peak and the concentration of Glx (R² = 0.259, F(1,11), p = 0.076).

In the present study, no significant differences between the vowel and the sine-tone condition were found regarding the absolute or relative amplitudes of the N1m peak. This result corroborates a previous report in which no significant differences between the amplitudes of the N1m elicited by a sine tone or the vowel /a/ were seen in a passive listening condition (watching a silent movie, as in the present study) 19. In contrast, vowels evoked a significantly stronger N1m response than tones in a study involving a stimulus detection task 20.

For the present study, we investigated associations between N1m amplitudes, but not N1m latencies, and the neurochemistry of the auditory cortex. We did not assess the influence of MRS variables on N1m latency because this latency depends, in part, on the subcortical auditory pathways, which could not be investigated with ¹H-MRS, used here.

The macroscopic and microscopic variability of primary and secondary auditory cortices has been carefully investigated 2122. In particular, the anatomy of the major generator site of the human N1m response, the planum temporale 23, is characterised by extensive interindividual variability 22. On the individual level, our AEF recordings displayed a pronounced interindividual variability of the peak amplitude and decrement (Figure 2A and 2C). This variability was expected in the light of previous experiments on auditory evoked potentials 2425.

The amino-acid NAA is found almost exclusively in neurons and axons, and can be used as an indirect measure of neuronal integrity and density. Loss of NAA has been reported in a variety of disorders, e.g., in individuals with epilepsy 26, Alzheimer's disease 27 or cognitive impairment 28. Cho, the choline-containing compounds measurable with ¹H-MRS, are dominated by phosphocholine, glycerophosphocholine, and free choline 29. As precursors or degradation products of membrane phospholipids make a major contribution to the Cho signal, Cho is regarded as a membrane marker 30. In addition, choline is the rate-limiting substrate for the synthesis of the neurotransmitter acetylcholine via the enzyme choline acetyltransferase.

The magnetic dipole moment detected by MEG represents the postsynaptic currents of approximately 1 million synapses activated synchronously. Differences in brain magnetic fields are thus assumed to reflect differences in the number of activated synapses and/or in the coherence of activated neurons 31. Our data suggest that the number and integrity of neurons (NAA) and the density and the functional integrity of cell membranes (Cho) are associated with the amplitude of auditory evoked responses.

The metabolic variables reported here, however, explain only in part the interindividual differences in cortical auditory processing. The association between NAA and the N1m amplitude accounts for 37% of the measured variance, while the association between Cho and the N1m amplitude accounts for 29% of the variance.

This study revealed no significant associations between the concentrations of NAA, Cho or Glx and the reduction of the N1m amplitude after a preceding stimulus (the N1m decrement). However, there was a trend for higher Glx concentrations in individuals with higher relative N1m amplitude (i.e., smaller decrement) (Figure 5, R² = 0.26, F(1,13), p = 0.054).

Previous work on cortical processing of rapidly recurring stimuli indicated that inhibitory interneuronal networks modulate the amplitude and the decrement of cortical evoked responses 332. Several classes of interneurons with distinct functional properties regulate the inhibitory activity in the neocortex 33. The activity of interneurons appears to be mainly regulated by the major inhibitory neurotransmitter gamma-aminobutyric acid (GABA) 34 and the major excitatory neurotransmitter, glutamate 35.

The concentration of GABA could not be measured by the MRS technique used here. The spectroscopic measurement of GABA in vivo is difficult, due to the overlapping peaks of NAA, Glx, and creatine. Although MR protocols have been developed for the assessment of GABA, these techniques require a measurement volume of more than 35 cm³ (compared with 3.375 cm³ here), much larger than the auditory cortex 25.

Glx represents the concentrations of the amino-acid glutamate and its precursor glutamine. In the peripheral and central auditory system, glutamate receptors of the alpha-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) type play an important role in the rapid synaptic transmission 36. Moreover, AMPA receptors regulate the processing of rapidly successive stimuli in the peripheral auditory system 37.

The aim of the present study was to correlate electrophysiological parameters of auditory processing with the neurochemistry of the auditory cortex. This study relies on the following assumptions.

Since the advent of non-invasive techniques to investigate human brain function, such as positron emission tomography, functional magnetic resonance imaging, MRS and MEG, numerous studies have been performed using either modality. Recently, an increasing number of researchers have combined neuroimaging modalities in an attempt to overcome the limitations of each of these methods 4546.

The present study correlated the electrophysiology of auditory processing as recorded by MEG with the neurochemistry of the auditory cortex as assessed by MRS. Our study, however, has several limitations. First, the small number of participants limited the power of the statistical analyses. In addition, we acquired data from only the left hemisphere. The 37-channel MEG system used here could be positioned close to the superior temporal plane, resulting in a good signal-to-noise ratio, but could only record the neural activity of one hemisphere at a time. The processing of acoustic stimuli can differ between hemispheres, especially when speech stimuli are used 47. Similarly, the concentrations of neurochemicals, as assessed by MRS, may display interhemispheric differences 48. Therefore, we cannot rule out that MEG-MRS associations for the right auditory cortex differ from those reported here for the left auditory cortex. For future studies, a larger sample size and data collection from both hemispheres (using a whole-head MEG system) are desirable. Moreover, we performed MEG recordings while participants were watching a silent movie (i.e., in a passive listening condition). The associations between MEG and MRS parameters reported here might change when MEG is recorded under top-down attentional modulation, as attention is known to influence auditory processing 49.

In addition to the concentration of NAA and Cho, other factors have been shown to influence auditory cortical processing, such as serotonergic neurotransmission (see Background) 7. In a study combining electrophysiology and genetics, the amplitude increase of the N1/P2 component in response to increasing stimulus intensities (termed the loudness dependence) was significantly different between individuals with different variants of the serotonin transporter gene. We expect that genetic studies will considerably contribute to future research on individual auditory processing.

Fifteen healthy volunteers (six women, nine men) with a median age of 24 years (range: 18–68 years) participated in this study. Two subjects were 60 years or older, the remaining 13 were 54 years or younger. Fourteen participants were right-handed, one was left-handed 50. All participants had a normal audiological status and were without a history of neurological or otological disorders. All individuals gave their informed consent to participate in the study. This project was reviewed and approved by the Research Ethics Board, Medical Faculty, University of Münster. The participants of this study had served as healthy controls in a previous study on metabolic and electrophysiological changes in patients with depression 18. MEG recordings of subgroups of the participants in this study were also published separately 5152.

To investigate individual brain anatomy and to exclude possible brain lesions, structural MR images were obtained from every subject. T1-weighted, three-dimensional, spoiled gradient echo MRI of the whole brain and T2- and proton density-weighted fast spin echo images in axial and coronal orientation were performed on a 1.5-Tesla scanner (Magnetom SP, Siemens, Erlangen, Germany). To assess the concentrations of brain metabolites, single voxel stimulated echo acquisition mode (STEAM) spectroscopy was used (echo time, TE = 20 ms; repetition time, TR = 2.5 s; number of scans = 128) 53. Using the individual structural MRI, the MRS voxel was centred at the left transverse gyrus of Heschl (Brodmann area 41) on the dorsal surface of the superior temporal plane containing the primary and secondary human auditory cortex. The MRS voxel volume was 1.5 × 1.5 × 1.5 cm³ (Figure 6).

The postprocessing of MRS spectra was performed as described previously 405455. Concentrations of N-acetylaspartate (NAA), choline-containing compounds (Cho) and the glutamine/glutamate pool (Glx) were quantified using a time domain fitting program, based on the linear combination (LC) model 56. Owing to the overlapping resonances of glutamate, glutamine and GABA at a field strength of 1.5 T, a combined fitting of the glutamine/glutamate pool was performed. Metabolite concentrations, acquired by the LC model and scaled to water, were corrected for coil loading 57 and partial volume effects 58. To account for variations in voxel composition (cerebrospinal fluid, grey matter and white matter), brain segmentation was performed with a semiautomatic interactive algorithm (M. Fiebich, University of Applied Sciences, Giessen, Germany) 54. Metabolites were then normalised to the grey-matter fraction in the voxels, and expressed in institutional units (IU). For robustness of results, only data with a fitting error <20% of the standard deviation was included in the final analysis.

To test for reliability of MRS data, we measured six healthy individuals twice, at a mean ± SD of 5.2 ± 0.8 months apart, in a separate study 40. Repeated-measures ANOVA yielded no significant difference between Glx concentrations of the left dorsolateral prefrontal cortex in the first and second measurement (F = 0.39, p = 0.56). Glx concentrations are usually responsible for the largest variance in MRS data (B. Pfleiderer, unpublished observation). The coefficient of variation for Glx in our previous study 40 was 12%. These values are similar to the coefficients of variation, ranging from 13% to 15%, which were previously reported for ¹H chemical shift imaging experiments 61. These results indicate that the metabolite concentrations measured here are reasonably stable over time.

To assess the reliability of AEF recordings, measurements were repeated in two participants after 4 months (Figure 8). The overall correlation between amplitudes over time in the first and the second recording were high (participant 1: r = 0.93, participant 2: r = 0.87). Most important for this study, the first N1m responses were almost identical. The reliability of later responses, in particular the second N1m, was smaller. These measurements corroborate an earlier study on the reliability of N1m recordings 14.

To test if the collected data were normally distributed, the Shapiro-Wilk normality test was performed 62. This test is widely used to test for normality in smaller samples because of its high statistical power compared with alternative tests 63. The assumption of a normal distribution was rejected if p < 0.01. According to this test, all electrophysiological and neurochemical variables were normally distributed, except the latencies of the four N1m responses. To compare the absolute and relative amplitudes of the four successive N1m waves, a repeated-measures ANOVA with Tukey's multiple-comparison post hoc test was calculated (Figure 2A). A nonparametric repeated-measures ANOVA (Friedman test with Dunn's multiple comparison post hoc test) was performed to compare the absolute and relative latencies of the four N1m waves (Figure 2B). To assess the relationship between absolute and relative N1m amplitudes and neurochemical parameters, a linear regression analysis was performed (Figures 3, 4, 5). Linear-regression models assume that the unpredicted variation (the error term ε) is normally distributed. To test for the normality of errors, a normal-probability plot of the residuals was performed (insets in Figures 3, 4, 5). If the error distribution is normal, the points on the normal-probability plot should fall close to a diagonal line. The linear-regression model was used to fit a regression line through the data and to calculate the 95% confidence intervals (Figures 3, 4, 5). Statistical analyses were performed with the statistical package R for Mac OS X 64.