by Dr. Andrew Bagshaw
University of Birmingham
Centre for Human Brain Health and School of Psychology
We are able to show using EEG-fMRI that the functional connectivity of the thalamus is generally increased by sleep onset and depth, and that sleep has a differential effect on sub-divisions of the thalamus. EEG-fMRI is a vital tool to understand the impact of sleep on cortical and subcortical processing.
Since the thalamus is crucial for regulating information flow from the periphery to the cortex and between cortical regions1, alterations to thalamocortical interactions are central to the changes to information processing and sensory responsiveness that are hallmarks of sleep onset. The thalamus is also vital for sleep-wake regulation2, and a range of sleep-related electrophysiological phenomena, including alpha-attenuation with sleep onset, sleep spindles, K-complexes and slow waves. As such, understanding the intricacies of thalamic and thalamocortical interactions across the sleep-wake cycle is a necessary step in understanding the mechanisms by which the behavioural manifestations of sleep are accomplished.
In human subjects, this can only be achieved using simultaneous EEG-fMRI. Alternative approaches such as polysomnography and positron emission tomography either do not have the sensitivity to deep brain structures or the spatiotemporal resolution that are needed. EEG data is mainly used in one of two ways: (i) to define sleep stages according to standard criteria or (ii) to identify the electrophysiological transients of sleep. The most common use of this information to inform the fMRI analysis is either to examine how fMRI functional connectivity (FC, a measure of time series correlation between brain regions) is modified by sleep stage, or to perform an event-related analysis based on the timings of discharges.
To date no examination of regional variability in thalamic FC during sleep has been conducted. This is important because the thalamus is not a unitary structure, but composed of a number of sub-divisions or nuclei3. Nuclei have different functions and patterns of connectivity with the cortex and other subcortical structures, with one of the primary distinctions being whether a nucleus receives its main inputs from the sensory periphery (first order nuclei) or the cortex (higher order nuclei). Given the diverse behavioural effects of sleep on sensory and cognitive processing, regional variability in the effect of sleep onset on thalamic sub-divisions would be expected. We examined this question using a parcellation of the thalamus based on FC profiles with large cortical regions4.
The MRI scanner is a difficult environment for participants to sleep, particularly with EEG. Participant comfort, and selecting participants who are familiar with MRI and preferably EEG-fMRI, can be vital to ensuring as much sleep as possible. Even so, we restricted our analysis to compare wakefulness (W) with non-rapid eye movement sleep stages 1 and 2 (N1 and N2). We examined thalamocortical FC and intra-thalamic FC, the latter both between homologous sub-regions across hemispheres, and between sub-regions in a single hemisphere.
We initially scanned 21 healthy participants with no history of neurological, psychiatric or sleep disorders. Participants’ habitual sleep patterns were monitored for two weeks prior to the scanning session using wrist actigraphy (Actiwatch 2, Philips Respironics), and several questionnaires were used to assess participants’ levels of daytime sleepiness, fatigue and subjective sleep quality (Epworth Sleepiness Scale, Fatigue Severity Scale, Pittsburgh Sleep Quality Index).
Scanning commenced around the participants’ normal bedtime (between 22:00 and 00.00), with participants arriving at the imaging centre approximately 30 minutes prior to the proposed start time to apply the EEG. We used a 64 channel BrainAmp MR plus system, with data acquired using BrainVision Recorder at 5kHz with hardware filters at 0.016-250Hz and synchronisation with the MR scanner clock via the SynchBox. Electrode impedance were below 20kΩ. MR data were acquired with a 3T Philips Achieva scanner with a 32 channel receive head coil. To ensure comfort and minimize movement inside the MR scanner, participants’ heads were fixated with foam pads. FMRI data were acquired in multiple scans of 1250 dynamics, each taking approximately 42 minutes (3x3x4mm voxels, TR=2000ms, TE=35ms, flip angle=80°, SENSE=2). These very long individual scans of 42 minutes were used rather than more common scans of 10-15 minutes as the cessation and then commencement of the scanner noise is the most disturbing to participants, and most disruptive to their sleep. Cardiac and respiratory data were acquired using the vectorcardiogram (VCG) and pneumatic belt provided by the MR scanner. Participants were given minimal instructions, simply not to resist sleep, and to signal via the scanner call button when they wished to terminate the session.
Our final cohort consisted of 13 participants with periods of W, N1 and N2 (9 male, age 26±4 years). Three participants were excluded for failing to sleep, 2 for failing to enter N2, 2 for not having any period of wakefulness during the fMRI scanning, and one for a problem with the volume triggers from the MRI scanner which meant that gradient artefacts could not be corrected.
EEG: The role of the EEG data was to provide sleep staging information for each subject, which could then be used to examine fMRI FC according to sleep stage. To this end, the gradient and ballistocardiogram (BCG) artefacts were removed using template subtraction approaches implemented in BrainVision Analyzer 2. The R-peak markers necessary for the BCG artefact removal were taken from the VCG data5. Sleep staging was performed by an experienced neurophysiologist according to the guidelines of the American Association of Sleep Medicine6, leading to a single sleep stage being associated with each non-overlapping 30s epoch of data.
fMRI: fMRI data were preprocessed for FC analysis using a combination of FSL and custom code written in Matlab. Data were motion corrected, physiological artefacts reduced using RETROICOR, spatially smoothed (4mm Gaussian kernel) and temporally high pass filtered (>0.01Hz). White matter, ventricular, and global signals were regressed, in addition to the six motion parameters. Using a previous methodology based on FC with cortical regions of interest4, the thalamus was segmented into five non-overlapping regions predominantly functionally connected to temporal (TEM), motor-premotor (MOT), somatosensory (SOM), prefrontal (PRE) or occipital-parietal (OCC) cortical regions (Fig. 1).
Thalamocortical FC for each sleep-staged epoch was calculated as the Pearson correlation between the thalamic ROI and its corresponding cortical ROI. These values were subsequently z-transformed, with degrees of freedom corrected using the Bartlett correction factor to account for autocorrelation, and individual z-maps for each sleep stage and cortical ROI were combined across participants using a random effects analysis. In addition to this thalamocortical analysis, FC between sub-regions of the thalamus was investigated in two ways. Inter-hemispheric thalamic FC was defined between each homologous sub-region in the left and right hemispheres (i.e., OCC-OCC, etc.), while intra-thalamic FC was between pairwise sub-regions within a hemisphere and subsequently averaged across hemispheres (e.g., OCC-SOM, SOM-MOT, etc.).
A total of 705 epochs of W, 944 epochs of N1 and 414 epochs of N2 were analyzed. In terms of thalamocortical FC (Fig. 2), there was a significant main effect of region (F(4,48)=4.04, p=0.007), and a significant interaction between region and sleep stage (F(3.04,36.45)=4.05, p=0.014), indicating that sleep onset affected thalamocortical FC differentially across sub-regions.
This interaction was driven by significant increases in FC from W to N2 in SOM and MOT. For inter-hemispheric thalamic FC, there was a significant effect of region (F(1.27,15.28)=78.86, p=6.7×10-8) and of sleep stage (F(2,24)=16.23, p=3.5×10-5). For all thalamic sub-regions, inter-hemispheric FC increased in a step-wise fashion from W to N1 and then N2 (Fig. 3).
Finally, there was a significant main effect of sleep stage (F(2,24)=20.4, p=6.6×10-6) and thalamic region (F(3.25,41.13)=352.8, p=3.0×10-30) on intra-thalamic FC. Generally, intra-thalamic FC increased from W to N1 to N2, with the strongest effect observed between MOT and SOM sub-regions (Fig. 4).
Sleep stages are defined largely by changes to scalp EEG activity6, and therefore any approach to understand the effect of sleep on the brain must rely upon EEG. However, scalp EEG is not a sensitive marker of sub-cortical activity, meaning that in the context of understanding how sub-cortical structures such as the thalamus are affected by and contribute to sleep onset and sleep depth, EEG much be combined with other techniques. Integrating EEG with FC analysis of fMRI data allows a novel insight into these processes, which is difficult if not impossible to achieve with other techniques with the same level of spatiotemporal precision.
We used three measures of thalamic FC and found regionally specific effects of sleep in all of them, indicating that FC is able to uncover diverse changes to the thalamocortical system relating to sleep. Most notable of our results is the observation that the thalamus becomes a more functionally homogeneous structure as sleep depth increases, potentially representing a shift from externally-driven, regionally specific connectivity patterns to a mode suggestive of more internally driven dynamics. This would be consistent with the behavioural and electrophysiological signatures of sleep, and opens up new avenues of investigation for studying the sleeping human brain. Intra-thalamic FC in particular is an intriguing metric. Electrophysiological studies have suggested that thalamic nuclei do not have direct connections between them, but that interactions are facilitated by an indirect route involving the inhibitory thalamic reticular nucleus (TRN)7. This suggests that intra-thalamic FC may be a marker of TRN function, which could prove useful in neurological and psychiatric disorders such as epilepsy and schizophrenia8,9, as well as in normal functions such as attention10.
There is considerable scope for future methodological improvements. For example sleep stages, while useful for clinical and research applications, are an oversimplification11. Within a 30s epoch a diverse and complex range of dynamic processes are occurring which more detailed sleep staging12 can help to uncover. This would be facilitated by faster fMRI sequences such as simultaneous multi-slice, which can reduce the time needed to acquire a single volume by a factor four or more. Similarly, the spatial dimension could clearly be improved, and methods based on thalamic anatomy, identified using a range of MR sequences13–15, would be beneficial. These have the advantage that they do not rely on the connectivity profile of thalamic sub-regions, which is diverse and complex even for first order nuclei. The EEG data could also be used more fully, for example to examine alterations to EEG-based FC in relation to sleep stage, thalamic fMRI FC etc.
EEG-fMRI is a vital tool to understand the impact of sleep on cortical and subcortical processing, providing a level of spatiotemporal resolution that cannot be accessed with other techniques. In particular, questions about the role of the thalamus in sleep that were previously the domain of invasive electrophysiology can be addressed, but with the advantage of whole brain coverage in human subjects..
 Sherman, S. M. & Guillery, R. W. The role of the thalamus in the flow of information to the cortex. Philos. Trans. R. Soc. Lond. B. Biol. Sci. 357, 1695–1708 (2002).
 McCormick, D. & Bal, T. Sleep and arousal: thalamocortical mechanisms. Annu. Rev. Neurosci. 20, 185–215 (1997).
 Sherman, S. M. The thalamus is more than just a relay. Curr. Opin. Neurobiol. 17, 417–422 (2007).
 Hale, J. R. et al. Comparison of functional thalamic segmentation from seed-based analysis and ICA. Neuroimage 114, 448–465 (2015).
 Mullinger, K. J., Morgan, P. S. & Bowtell, R. W. Improved Artifact Correction for Combined Electroencephalography / Functional MRI by Means of Synchronization and Use of Vectorcardiogram Recordings. J. Magn. Reson. Imaging 616, 607–616 (2008).
 Iber, C., Ancoli-Israel, S., Chesson, A. & Quan, S. The AASM manual for the scoring of sleep and associated events. Rules, terminology and technical specifications. (American Association of Sleep Medicine, Westchester, IL, 2007).
 Crabtree, J. W. Intrathalamic sensory connections mediated by the thalamic reticular nucleus. Cell. Mol. Life Sci. 56, 683–700 (1999).
 Steriade, M. Sleep , epilepsy and thalamic reticular inhibitory neurons. Trends Neurosci. 28, 317–324 (2005).
 Ferrarelli, F. & Tononi, G. The Thalamic Reticular Nucleus and Schizophrenia. Schizophr. Bull. 37, 306–315 (2011).
 Halassa, M. M. et al. State-Dependent Architecture of Thalamic Reticular Subnetworks. Cell 158, 808–821 (2014).
 Müller, B., Gäbelein, W. D. & Schulz, H. A Taxonomic Analysis of Sleep Stages. Sleep 29, 967–974 (2006).
 Ogilvie, R. D. The process of falling asleep. Sleep Med. Rev. 5, 247–270 (2001).
 Tourdias, T., Saranathan, M., Levesque, I. R., Su, J. & Rutt, B. K. Visualization of intra-thalamic nuclei with optimized white-matter-nulled MPRAGE at 7T. Neuroimage 84, 534–545 (2014).
 Gringel, T. et al. Optimized high-resolution mapping of magnetization transfer (MT) at 3 Tesla for direct visualization of substructures of the human thalamus in clinically feasible measurement time. J. Magn. Reson. Imaging 29, 1285–1292 (2009).
 Traynor, C. R., Barker, G. J., Crum, W. R., Williams, S. C. R. & Richardson, M. P. Segmentation of the thalamus in MRI based on T1 and T2. Neuroimage 56, 939–950 (2011).