A specialized channel for encoding auditory transients in the magnocellular division of the human medial geniculate nucleus

Abstract

Objective:

We test the hypothesis that there exists a generalized magnocellular system in the brain optimized for temporal processing. In the visual system, it is well known that the magnocellular layers in the lateral geniculate nucleus (LGN) are strongly activated by transients and quickly habituate. However, little is known about the perhaps analogous magnocellular division of the medical geniculate nucleus (MGN), the auditory relay in the thalamus.

Methods:

We measured the functional responses of the MGN in 11 subjects who passively listened to sustained and transient non-linguistic sounds, using functional MRI (fMRI).

Results:

We observed that voxels in the ventromedial portion of the MGN, corresponding to the magnocellular division, exhibited a robust preference to transient sounds, consistently across subjects, whereas the remainder of the MGN did not discriminate between sustained and transient sounds.

Conclusions:

We conclude that the magnocellular neurons in the MGN parallel the magnocellular neurons in its visual counterpart, LGN, and constitute an information stream specialized for encoding auditory dynamics.

Introduction

The brain has evolved mechanisms to efficiently encode the large bandwidth of sensory information. In the visual system, this begins in the retina, where there are multiple classes of ganglion cells tuned to specific stimulus properties. Two of the most prominent channels are the magnocellular and parvocellular streams that remain disjoint in the different layers of the lateral geniculate nucleus (LGN) in the thalamus. In the macaque, magnocellular neurons in the LGN are specialized to encode transient visual stimuli, responding quickly to abrupt onsets and rapidly habituating [1–3]. It seems to be the case that magnocellular neurons throughout the sensory and motor systems of the brain are specialized for temporal processing [4–7], but evidence for this in other sensory modalities is scarce.

The medial geniculate nucleus (MGN), the auditory relay nucleus in the thalamus, has three main subdivisions based on cellular morphology: the ventral, dorsal and medial divisions [8]. The latter is hereafter referred to as the magnocellular division based on a population of large neuron cell bodies found here [9; 10], comparable to the magnocellular neurons in the visual lateral geniculate nucleus (LGN). The magnocellular and ventral divisions of the MGN contain tonotopic maps, orderly representations of sound frequency [8; 11; 12]. Little is otherwise known about the response properties of neurons in the magnocellular division of the MGN in humans; however, the neurons here in other species habituate rapidly [13; 14] and exhibit short response latencies [15].

We sought to test the magnocellular system theory by directly measuring the activity in the magnocellular division of the MGN. We hypothesized that these magnocellular neurons might be analogous to those in the LGN and would selectively encode auditory transients.

Materials and Methods

Subjects.

11 subjects (nine female) participated, all 18–32 years old. None had any neurological disorders, and all were right‐handed. The subjects provided informed written consent under the research protocol approved by the Institutional Review Board at the University of Delaware.

Stimuli.

The stimuli were generated using MATLAB software (The MathWorks, Inc.) with the Psychophysics Toolbox 3 functions [16–18] running on a Linux computer. The stimuli were synchronized to the MRI acquisition using a trigger signal from the scanner, interfaced to the computer through a fORPs response box (Current Designs, Inc.). Auditory stimuli were presented through headphones (OptoActive II, Optoacoustics Ltd.) with real-time algorithmic, out-of-phase harmonic active noise cancelation that attenuated the background noise from the scanner. All subjects passively listened to the stimuli and reported clearly hearing them throughout the scanning procedures. Twelve blocks each of sustained or transient auditory stimuli were presented, randomly interleaved during each 312 s scanning run (Fig. 1). Blocks were 13 s in duration, consisting of approximately 9 s of sound and the remaining approximately 4 s of the block unstimulated (silent). The stimuli in each block were sampled from broadband natural sounds with no linguistic content, e.g. bells, wild animal roars, car horns, trumpets and steam engine whistles. The transient blocks consisted of a series of three to four sound bursts of the samples, separated by 0.5 s of silence and windowed with a square wave to produce abrupt onsets and offsets (Fig. 1). The duration of these bursts was either 50–150 ms for shorter bursts or 200–300 ms for longer bursts. The silent gap between bursts was 20–40 ms. This timing was chosen to approximate the typical syllabic timing in speech. The sustained blocks were composed from the same sound samples, but 0.5–2 s in duration, separated by 0.5 s and windowed with sine onset and offset ramps, each one-quarter of the sound duration, to eliminate transients. Visual stimuli were also presented, consisting of high contrast checkerboards flickering at various frequencies, with timing independent of the auditory stimuli, but these were not analyzed for the present study.

Figure. 1.

Experimental design and mean activations. Transient and sustained auditory stimuli were randomly interleaved in each 312 s scanning run in separate 13 s blocks that included approximately 9 s of sound and 4 s of silence. The transient stimuli consisted of short (50–150 ms) or long (200–300 ms) pulses with a 20–40 ms gap between. The sustained stimuli were 500– 2000 ms in duration with 125–500 ms sine ramped onsets and offsets.

Neuroimaging.

Data were acquired using a 3T Siemens (Erlangen, Germany) Prisma MRI scanner with a 64-channel head coil at the Center for Biomedical and Brain Imaging at the University of Delaware. All subjects participated in two to five MRI scanning sessions. A high-resolution T₁-weighted scan was acquired for each subject in each scanning session [MPRAGE, TR = 2080 ms, TE = 4.64 ms, flip angle = 9°, 208 sagittal slices, FOV = 210 × 210 mm, acquisition matrix = 288 × 288, isotropic (0.7 mm)³ resolution, parallel imaging (iPAT) acceleration factor (GRAPPA) = 2]. In one session, we acquired 40 proton density-weighted (PD) spin-echo structural images [acquisition time 89 s, TR = 3000 ms, TE = 16 ms, flip angle = 150°, 35 coronal 1 mm thick slices covering the thalamus, FOV = 256 × 256 mm, acquisition matrix = 256 × 256, isotropic 1 mm³ resolution, iPAT GRAPPA = 2]. In the remaining sessions, we acquired six to ten runs, 209 volumes each, of functional data using a multi-band gradient echo EPI sequence [TR = 1.5 s, TE = 39 ms, flip angle = 45°, 84 horizontal 1.5 mm thick slices, FOV = 192 × 192 mm, acquisition matrix = 128 × 128, isotropic (1.5 mm)³ resolution, A → P phase encoding, partial Fourier factor = 6/8, slice acceleration factor = 6, bandwidth = 1562 Hz/Px]. The subjects’ heads were surrounded by foam padding to reduce head movements.

Anatomical regions of interest.

The location of the MGN in humans is well known from functional and anatomical studies [19–23]. The 40 PD images were registered using an affine transformation to correct for displacement between acquisitions, up-sampled to twice the resolution in each dimension and averaged to create a mean image with high signal-to-noise. These images were aligned to the T₁ and used to manually trace the anatomical extent of each MGN (Fig. 2). These anatomical ROIs were used to guide the functional ROI, and the two ROIs matched well except in one subject (Fig. 2).

Figure 2.

Region of interest selection. Three representative subjects are shown, each in a separate row. For each subject, a zoomed coronal slice is shown through the thalamus, with the subject’s structural image as the background. The columns from left to right show: 1. structural image, 2., outlines of the anatomical segmentations of the MGN and LGN, based on proton-density images, 3. functional activation to combined auditory stimuli, with the color bar indicating for each voxel the z-score of the sound vs. blank contrast, 4. overlay of the anatomical ROI on the functional activations. D, dorsal; V, ventral; R, right; L, left.

Experimental design and statistical analysis.

Individual subjects were analyzed in their native space. Functional data were preprocessed using FEAT (FMRI Expert Analysis Tool) Version 6.00, part of FSL (FMRIB’s Software Library, www.fmrib.ox.ac.uk/fsl)[24]. The functional images were realigned to correct for small head movements using FLIRT [25] and then linearly registered to each participant’s T₁. Each image was smoothed with a 2 mm full width at half maximum (FWHM) Gaussian kernel. Each subject’s data were analyzed with a general linear model [26], with two explanatory variables (EVs) accounting for the transient and sustained stimuli, with the silent periods as the baseline. The estimated motion parameters for each run were included as covariates of no interest, and the motion outliers were defined as additional confound EVs. A fixed effects analysis combined the multiple scanning runs within each subject. Because the amount of data varied among subjects (who had one to four functional scanning sessions each), the noise level among subjects also varied. Therefore, to define the functional ROI, a cluster-based significance test was used with a different height threshold for each subject, with z ∈ {0.5, 1, 1.5, 2} (the lowest necessary for the noisiest three of the 22 MGN; one other MGN was not detected), and a cluster p < .05, corrected for multiple comparisons [27]. This resulted in a functional activation distinct from the background noise and matching the size of the anatomical ROI. A transient index, (T − S)/(T + S), was computed for each activated voxel using the weights of the transient (T) and sustained (S) activations. To compare the activation in the whole MGN across subjects, the mean activations for each EV were calculated over all activated voxels in each MGN, and these mean activations were subjected to a mixed effects repeated measures model with hemisphere (left or right) and stimulus (S or T) as within-subjects fixed effects.

The group analysis was computed in standard space in the same manner as the individual analyses, with the following differences. Registration from the T₁ to standard space was calculated using FNIRT nonlinear registration [28]. A mixed effects (FLAME 1+2) analysis was conducted to compute the group EVs from each subject’s individual results [29]. The group transient index was computed for each voxel using the weights of the group EVs. The ROIs of the MGN in standard space were defined using the Jülich histological atlas [30].

Results and Discussion

The MGN was segmented in each subject using structural imaging, and the mean volume across subjects was 134.2 ± 7.0 mm³ (mean ± SEM). Using functional magnetic resonance imaging (fMRI), we measured the relative responses to transient and sustained auditory stimuli within the MGN. We imaged subjects as they passively listened to transient or sustained non-linguistic sounds, using a block design (Fig. 1). The MGN in all subjects were well activated by the combined auditory stimuli (Fig. 2), bilaterally in all but one subject. In the whole MGN, i.e., the mean of the activated voxels in each MGN in the native spaces of each subject, we found that subjects exhibited greater activation to the transient than sustained stimuli (F_1,7.8 = 259.3, p < .0001, Fig. 3B), with no difference between the left and right MGN (F_1,9.9 = 0.26, p = .6202).

Figure 3.

Group results. A. A schematic of the divisions of the human MGN, based on cellular morphology [9]. The medial division is also known as the magnocellular division (D, dorsal; V, ventral; M, medial; L lateral). B. The mean transient (T) and sustained (S) activations across subjects are shown, collapsed across left and right MGN in the native space of each subject. C. A zoomed region of the brain is displayed covering the thalamus on four adjacent coronal slices, arranged anterior (A) to posterior (P). The transient index, (T − S)/(T + S), is displayed for each voxel in the MGN in standard space, computed over all subjects, where T and S are the transient and sustained group activations, respectively. The blue voxels with indices near zero showed no preference between the transient and sustained stimuli. The red and yellow voxels with indices closer to one were more strongly activated by the transient stimuli. Voxels with large transient indices were largely confined to the ventromedial portion of the MGN, the magnocellular division (L, left; R, right; M, midline).

To visualize the distribution of activations to the transient (T) and sustained (S) stimuli within the MGN volume, we computed a normalized transient index (T − S)/(T + S) for each voxel. In both the group in standard space (Fig. 3C) and in every subject individually in their native spaces (Fig. 4), we observed strong responses to the transient stimuli (high transient indices) along the entire ventral surface of the MGN, corresponding to the expected size and location of the magnocellular division as expected from the anatomy (Fig. 3A). There were also small clusters of increased activation to transient vs. sustained stimuli on the dorsolateral surface of the MGN. There were no areas in the cortex selective for the transient stimuli.

Figure 4.

Transient indices for four representative subjects in native space. Each pair of columns shows the right (R) and left (L) MGN for a single subject, arranged relative to the midline (M), with six coronal slices per MGN, ordered anterior (A) to posterior (P). The transient index is displayed for each activated voxel in the MGN in each subject. The blue voxels have a transient index near zero, indicating that there was little or no difference in response to the transient or sustained stimuli, whereas red and yellow voxels exhibited a stronger response to the transient stimuli.

Across subjects, we observed a strong and remarkably consistent preference for transient compared to sustained auditory stimuli in the ventromedial portion of the MGN, corresponding to its magnocellular division. The remainder of the MGN responded equally to sustained or transient stimuli. Our results validated the magnocellular systems hypothesis, that a common encoding strategy is used across modalities in the brain, with one channel specialized for temporal precision.

Throughout the sensory and motor systems in the brain, there are neurons with large cell bodies that are specialized for temporal processing [4; 5; 7], comprising a magnocellular “system”. Although the magnocellular division in the human MGN contains a variety of different neuron types [9], the functional response seems to be dominated by the magnocellular neurons, as found in the magnocellular divisions in the other auditory relay nuclei [6].

Auditory transients are important for speech comprehension, wherein small differences in perceived timing distinguish different phonemes that convey important meaning, and object recognition [31]. Malfunction of the magnocellular system has been hypothesized to be a core deficit in dyslexia, a common reading-specific disorder [4; 5; 7].

Although the whole MGN can be readily segmented using anatomical imaging, and the location of the magnocellular division is generally known, we found that a contrast between transient and sustained stimuli can be used as an effective functional localizer for the magnocellular division.

Conclusion

We found that the magnocellular division of the MGN selectively encoded auditory transients, whereas the other divisions in the MGN responded similarly to transient and sustained sounds. Comparing the magnocellular division of the MGN to the that in the LGN, we conclude that different sensory systems in the brain employ similar encoding strategies to efficiently convey information.