Graduate School of Psychology
R
ESEARCHM
ASTER’
SP
SYCHOLOGYT
HESISStatus: Final draft Date: 20 / 01 / 2015
1. WHO AND WHERE
Student
Name : Joeri B.G. van Wijngaarden, BSc.
Student ID number : 5808111
Address : Eerste Ringdijkstraat 224
Postal code and residence : 1097 BC Amsterdam
Telephone number : +31 6 4108 5015
Email address : joerivanwijngaarden@gmail.com
Supervisor(s)
Within ResMas (obligatory) : Mike X. Cohen, PhD.
Specialisation : Brain and Cognition
External supervisor(s), if any : Dr. Paul F.M.J. Verschure
Second assessor : Dr. Lourens J. Waldorp
Research center / location : Synthetic, Perceptive, Emotive and Cognitive Systems (SPECS) Group, Universitat Pompeu Fabra, Barcelona
Number of credits (1 ec = 28 hrs) : 30 ec At least 25 ec
Ethics committee reference code : n/a http://ce.psy-uva.nl/
The Emergence of Thalamocortical Dysrhythmia
After Acute Ischaemic Stroke
Joeri B.G. van Wijngaarden1,3, Simon Finnigan2, and Paul F.M.J. Verschure3
1
Department of Psychology, University of Amsterdam, Amsterdam, The Netherlands
2
UQ Centre for Clinical Research, The University of Queensland, Royal Brisbane and Women’s Hospital, Australia
3
Laboratory of Synthetic Perceptive, Emotive and Cognitive Systems-SPECS, Center of Autonomous Systems and Neurorobotics, Catalan Institute of Research and Advanced Studies (ICREA) & Universitat Pompeu Fabra, Barcelona, Spain
Keywords: thalamocortical circuit, ischaemic stroke, EEG oscillations, low frequency bursts
Thalamocortical dysrhythmia (TCD) has been identified as an important underlying mechanism for a subset of neurological disorders such as Parkinson’s disease, neurogenic pain and tinnitus. Positive symptoms result from entrainment of thalamocortical loops, characterized by low frequency oscillations originating from the thalamus. This study investigated TCD as a possible mechanism for indirect sensory-related symptomology resulting from acute ischaemic stroke. Resting-state electroencephalogram (EEG) data of ischaemic stroke patients (n = 17) were compared to data of healthy controls (n = 17). Patient spectral power data displays an increase of low-frequency cortical activity together with an attenuation of the dominant power peak, which is highly characteristic of TCD. This notion is supported by our computational model of the thalamus and suggests that decreased corticothalamic activity leads to a switch of the thalamic system from tonic firing into a bursting regime. These results demonstrate the involvement of the thalamus in post-stroke symptomology and can possibly explain secondary sensory-related symptoms found in ischaemic stroke.
INTRODUCTION
The thalamus is considered to be an important gateway for subcortical structures to communicate with parts of the cerebral cortex. Nearly all of the excitatory input into the cortex originates either from the thalamus or from other cortical neurons (Jones, 1985; Douglas & Martin, 2010). The different thalamic nuclei mainly consist of relay neurons. These neurons receive most of the body’s sensory information through ascending axons – coming from sensory organs or the spinal cord (e.g., the periphery) – and relay these signals to the respective cortical region (i.e., relay retina information to the visual cortex; Alitto & Usrey, 2003).
Each modality has its own specific relay nucleus. For example, there are the lateral geniculate nucleus (LGN) and the medial geniculate body (MGB), receiving visual and auditory information respectively from the periphery, which is then relayed through cortical layer IV, projecting onto the primary visual and auditory cortex (Jones, 2002; Alitto & Usrey, 2003). Axonal-tracing and electron microscopic studies over the past decade have shown the wide-range of axons providing input for these modal-specific relay nuclei. Even though these nuclei serve as a relay hub, their input from the periphery is relatively small, with estimates ranging from 5 to 16 percent of the total input (van Horn et al., 2000; Jones, 2002; Sherman & Guillery, 1998). The remainder of excitatory input consists of descending corticothalamic axons originating from cortical layer VI, constituting around 40 to 50 percent of all arriving axons, forming recurrent corticothalamic loops. While these descending connections have a direct excitatory influence on the relay neurons, they are also indirectly connected through the thalamic reticular nucleus (RTN) comprising the remaining ~30 percent of input. This nucleus is regarded as the inhibitory “pacemaker” within the thalamic system, and excitation of the RTN results in silencing of the relay neurons (Jones, 1998).
The fine balance of corticothalamic and peripheral excitation, together with reciprocal inhibition from the RTN gives rise to an interesting feature of all thalamic neurons, that is, their ability to switch between two different firing regimes: tonic versus burst firing. This ability is dependent on T-type Calcium (Ca2+) channels in the cell’s membrane. Under relatively depolarized conditions, Ca2+ channel currents (𝐼𝐼𝑇𝑇) are inactivated, causing the neuron to fire tonically: a stream of unitary action potentials follows suprathreshold input. In contrast, relatively hyperpolarized conditions will activate the It, leading to a low threshold calcium spike (LTS) with multiple sodium spikes riding on its crest,
creating a short burst of spikes followed by an extended refraction period (Sherman, 2001; Sherman & Guillery, 2001, 2002).
During sleep or drowsiness, the thalamus is commonly in a state of burst firing, with large synchronized bursts across most of the thalamic relay cells (Steriade, McCormick, & Sejnowski, 1993; Steriade & Timofeev, 2003). This rhythmic bursting prevents the cells from performing their normal relay functions. With an LTS, there is a non-linear step-wise relationship between the neuron’s input and an excitatory post-synaptic potential (EPSP). The neuron is either silent or bursts with 4 to 6 spikes, making it less responsive to incoming action potentials. Conversely, with inactive Calcium
currents, there is a more direct and linear relationship between input and firing rate where increased input always results in more EPSP’s (Sherman, 2001). While this bursting is common and even necessary to reach a state of sleep, it has been related to several types of psychopathology when awake. These types fall under a condition known as thalamocortical dysrhythmia (TCD; Jeanmonod et al., 1996; Llinás et al., 1999; Schulman et al., 2011). TCD consists of a steady, long-lasting increase of low frequency bursting activity of the thalamus – often in the theta range (4 to 8 Hz) – that is accompanied by increased thalamocortical coherence. Due to the cortical organization, synchronized low frequency oscillations influence local interneurons at the cortical level, creating an imbalance in cortico-cortical inhibition. This asymmetry provides a ring of reduced regional inhibition, allowing for increased local and high frequency activity also called the edge-effect (Llinás et al., 1999, 2005). Depending on the affected thalamocortical circuit, this edge-effect leads to positive symptoms in different neurological disorders such as: (i) Parkinsonian tremors, (ii) tinnitus-related pain or ringing in the ears and (iii) neurogenic pain (Sarnthein & Jeanmonod, 2007, 2008, Moazami-Goudarzi et al., 2010). Identified affected regions for these disorders include primary sensory and motor areas, as well as parts of the limbic system (Jeanmonod et al., 2003; Stern et al., 2006; Moazami-Goudarzi et al., 2008, 2010).
TCD is thought to develop mainly due to an imbalance of excitation and inhibition within the thalamic system, where either decreased excitation or increased inhibition leads to hyperpolarized cell membranes. One possible way to reach this imbalance is deafferentation of ascending nerve fibers – as is the case with neurogenic pain or tinnitus – leading to decreased peripheral recruitment of thalamic neurons. Alternatively, with Parkinson’s disease, there is an increase of inhibition due to an overactive globus pallidus, also leading to a state of hyperpolarization. However, less is known about the third possible route: decrease descending corticothalamic excitation. In theory, decreased cortical recruitment of the relay or reticular neurons would create a state of relative hyperpolarization and upon reaching a certain threshold, cause a switch in firing regimes of thalamic nuclei.
The aim of this study was to explore this disrupted corticothalamic connectivity and its influence on the thalamus’ firing behavior. To do so, a similar approach was used compared to previous work in this field, which is to investigate electro-neurophysiological data of a specific patient group and identify neural activation patterns consistent with TCD. In this case stroke patients match the profile. With ischaemic stroke, an artery providing blood to the brain is blocked – because of narrowing of the arteries or a blood clot – resulting in dying of large ensembles of neurons (e.g., a lesion). As a result, the primary symptoms of stroke are directly related to the failing of critical brain regions, often the primary sensory areas, leading to: (i) weakness, numbness or failure of muscles in the face or other limbs, (ii) abrupt loss of vision, sensation and/or speech and (iii) severe headaches with a possible loss of consciousness. We hypothesize that when the lesion is part of a thalamocortical circuit, not only does this lead to a loss of local activity, it also severs corticothalamic connections and cause a decrease of cortical recruitment of the thalamus. When this decrease is sufficient to create a
state of relative hyperpolarization for thalamic neurons, it’s possible that thalamocortical dysrhythmia emerges. In this scenario, patients develop secondary sensory-related symptoms as an indirect result of the lesion, with TCD as an underlying mechanism. These symptoms are related to distorted sensory information processing in the form of pain in non-paretic limbs, visual flashes and sharp noises.
To investigate this hypothesis, electroencephalogram (EEG) data of ischaemic stroke patients and healthy controls were gathered. We expect these patients’ activity patterns to be highly comparable to the distinct EEG characteristics commonly found in TCD. This includes: (i) slowing and often amplification of the dominant power peak, (ii) increased power over the lower frequency ranges and (iii) increased synchrony between the thalamus and cortical regions – all relative to a healthy control group. In addition to this electrophysiological data, we utilized a computational model of the thalamus in order to explain the differences in the EEG results. Previous modeling work in this field focused on neurogenic pain in particular, where deafferentation of ascending axons leads to hyperpolarized thalamic cell membranes and cause a switch in firing regimes of thalamic nuclei (Proske et al., 2011). Our goal was to use this model to simulate a stroke, and demonstrate that introducing a cortical lesion into the system leads to similar results.
METHODS
Patients
Seventeen patients (8 male, 9 female; mean age 69 years, range 38-84) – all suffering from acute middle cerebral artery (MCA) stroke – were recruited from the Royal Brisbane and Women’s Hospital in Brisbane, Australia. Ischaemic stroke was assessed using acute computed tomography (CT) scans, followed by magnetic resonance imaging (MRI) in six cases. Time of onset for the stroke was defined as the last moment the patient did not experience any stroke symptoms. In the case of patients waking up with stroke symptoms (three cases), time of stroke onset was defined as the mid-point between bed-time and the bed-time of waking up. Patient demographic information can be found in Table 1. Data of this patient group has previously been published in Schleiger et al. (2014).
Within 30 minutes after the EEG recording session, the National Institute of Health Stroke Scale (NIHSS) was administered. The NIHSS is a systematic assessment tool, comprising of 15 items, to evaluate the neurological status of stroke patients.
Healthy Controls
The control group consisted of seventeen healthy individuals (9 male, 8 female; mean age 68 years, range 60-80) with no cognitive impairments or history of depression and/or anxiety. Data of these subjects has been previously published in Cummins et al. (2007, 2008). In short, all subjects were recruited using newspaper advertisements for a longitudinal study of amnestic Mild Cognitive Impairment (aMCI) and Alzheimer’s disease (AD). They underwent a comprehensive battery of
neuropsychological tests, including the Mini Mental State Exam (MMSE), the cognitive section of the Alzheimer’s disease Assessment Scale (ADAS-Cog), the Wechsler Memory Scale (WMS) and several verbal ability tests. Data of this control group were used as a reference to compare against the patient data.
Table 1. Clinical description of patients.
Patient ID Age Gender EEG time after
stroke onset (h) Lesioned hemisphere NIHSS
1 56 Male 74 Right MCA 0
2a 82 Female 75 Left MCA 2
3a 38 Female 57 Left MCA 1
4 45 Male 74 Left MCA 19
5 74 Female 68 Right MCA 4
6 66 Female 72 Right MCA 0
7a 74 Female 71 Left MCA 11
8 81 Male 61 Left MCA 5
9 51 Female 72 Right MCA 18
10 82 Male 71 Right MCA 16
11 79 Female 21 Right MCA 9
12 77 Female 68 Right MCA 6
13 49 Male 72 Left MCA 5
14 79 Female 99 Left MCA 7
15 75 Male 63 Right MCA 3
16 77 Male 72 Right MCA 15
17 84 Male 70 Left MCA 1
a
wake-up stroke
EEG acquisition and processing
EEG data of the patient group was acquired using continuous recordings at the patient’s bedside in the acute stroke unit, approximately 68 hours (range 21-99) after stroke onset. The system used to measure EEG signals was a NicOne Brain Monitor (Natus Medical Inc.), recording at a sampling rate of 500 Hz. A total of 19 Ag/Ag-Cl electrodes (Nicolet) were used, placed according to the international 10-20 system. Additionally, a reference electrode was placed midway between Fz and Cz together with a ground electrode on the collar bone. Electrode impedances were below 5-10 kΩ in all electrodes included for further analysis.
EEG data of the healthy controls was acquired using an elasticized quick cap with 32 Ag/Ag-Cl electrodes (Neuromedical supplies), placed according to the 32-channel positioning of the international 10-20 system. All data was digitized by a Neuroscan Synamps amplifier while recording continuously at a sampling rate of 500 Hz. The linked earlobes were used as a reference for all channels. Additionally, two bipolar channels recorded vertical and horizontal electro-oculograms (EOGs) to allow for the exclusion of ocular artefacts. One set of electrodes was placed on the outer canthus of each eye to measure horizontal artefacts, the other set below and on top of the supra-orbital ridge of the left eye to measure vertical artefacts.
Signal processing and analyses were performed offline using Matlab (The MathWorks, Natick, MA, USA) and its toolbox EEGLAB (http://sccn.ucsd.edu/eeglab; Delorme & Makeig, 2004) with custom in-house scripts. All data were bandpass filtered between .5 and 35 Hz (12dB/octave) and re-referenced to the common average of all electrodes. The patient data was epoched into continuous epochs lasting 4096 ms (2048 datapoints), while the control data’s epochs were half that size (2048 ms). Epochs containing EEG amplitudes of more than ±75 μV were automatically excluded, after which manual visual inspection led to the rejection of any epochs containing clear low-amplitude artefacts – mainly of ocular or muscular nature – which were removed from further analyses. Finally, the first 45 artefact-free epochs were selected. The amount of electrodes differed for the patients and controls. To maintain the same signal to noise ratio for both groups, only those 17 electrodes that overlapped were included for further analyses, which were: F3, F4, F7, F8, Fz, C3, C4, Cz, P3, P4, Pz, T3, T4, T5, T6, O1 and O2. In addition, as the amount of time points for each epoch also differed between patient and control data, only the first 1024 time points (2048 ms) of each epoch were selected.
EEG analysis
Wavelet-decomposition analysis was used to decompose the EEG data into the frequency domain, described in detail in Cohen (2014). At first, a set of complex Morlet wavelets was computed: complex sine waves within a Gaussian-window defined as 𝑒𝑒−𝑖𝑖2𝜋𝜋𝜋𝜋𝜋𝜋∗ 𝑒𝑒−𝜋𝜋2/2𝑠𝑠2, where f is frequency, t is time and s represents the number of cycles. In total, wavelets at 80 different frequencies were computed, starting at 1 Hz and scaling logarithmically up to 35 Hz. With equal amount of cycles, higher frequency wavelets would span a shorter time-window compared to lower frequencies. Hence, the amount of cycles for each wavelet (defined as 𝑠𝑠 2𝜋𝜋𝜋𝜋⁄ ) scaled logarithmically from 8 up to 15 cycles, increasing with higher frequencies. The second step was to take the fast Fourier transformation (FFT) of the EEG signal at each electrode, and multiply this with the FFT of every wavelet. Taking the inverse FFT results in an analytical signal that was used to extract power or magnitude of the signal (equation 8) and phase angle (equation 9) at each time-frequency point. Finally, spectral power was
binned and averaged into commonly used frequency bands: delta (δ; 1 to 4 Hz), theta (θ, 4 to 8 Hz), alpha (α; 8 to 12 Hz) and beta (β; 12 to 30 Hz), giving the mean spectral energy (MSE) per band.
𝑝𝑝(𝑡𝑡) = 𝑟𝑟𝑒𝑒𝑟𝑟𝑟𝑟[𝑧𝑧(𝑡𝑡)]2+ 𝑖𝑖𝑖𝑖𝑟𝑟𝑖𝑖[𝑧𝑧(𝑡𝑡)]2 (Eq. 8)
𝜑𝜑𝜋𝜋 = arctan (𝑖𝑖𝑖𝑖𝑟𝑟𝑖𝑖[𝑧𝑧(𝑡𝑡)] 𝑟𝑟𝑒𝑒𝑟𝑟𝑟𝑟[𝑧𝑧(𝑡𝑡)]⁄ ) (Eq. 9)
To investigate the topographical distribution of differences in spectral power between the patient and control group, a Wilcoxon rank sum test was applied at each electrode and each frequency-point. The result is a matrix with Z-scores, indicating differences between groups. Note that no correction for multiple comparisons was applied, hence those results should be considered qualitative and exploratory.
Among the patient group, eight individuals suffered a lesion in the left hemisphere, against nine individuals with a lesion in the right hemisphere. To prevent averaging out lesion-related effects in any topographical map, two subgroups were required. This would lead to a loss of statistical power and complicated further classification analyses. Alternatively, a more optimal approach is to mirror the values for all non-midline electrodes for one subgroup only. Hence, all power values in the left- and right-hemispheric electrodes were swapped for the group of patients with a lesion in the left hemisphere
(e.g., F3 becomes F4 and vice versa). As a result, all topographical maps should be interpreted with a lesion in the right hemisphere, meaning right-hemispheric activity is always ipsilateral and left-hemispheric activity always contralateral to the lesion.
Patient versus control classification
A common method for the classification of patients versus controls is linear discriminant analysis (LDA). This method classifies an individual to either group, using certain parameters – such as spectral power peak height and/or position – based on a linear border. The percentage correctly classified is an indication of how well the parameters differentiate between groups. We chose to opt for a more sophisticated method called Support Vector Machines (SVM). A benefit of using SVM over LDA is the choice of different kernels without the restriction of using a linear border.
In our model, a radial basis function (RBF) was selected as a kernel, following equation 10. This kernel clusters the data into two separate groups based on the distance between all data points – evaluated using an exponential rather than a linear function – and can include any number of dimensions.
𝐾𝐾(𝑥𝑥, 𝑥𝑥′) = e𝛾𝛾 ��𝑥𝑥−𝑥𝑥′�� 2 (Eq. 10)
The first step in applying SVM is training the model to attain the optimal parameters γ (defined as − 1 2𝜎𝜎⁄ 2, with σ representing the feature vector) and an error margin C. Ideally the data is divided into a separate set for training and for classification. In this case however the amount of data points and participants is rather limited. Hence, the training includes the data of all participants. Data was rescaled to [-1, +1], and 3-fold cross-validation was applied with γ{2−15, … , 23} and 𝐶𝐶{2−5, … , 215}. Cross-validation splits the data into three comparable sets, the first two to train the model and the last one to evaluate its performance. Finally, all possible combinations of different values for γ and C were tested using these first two sets and evaluated on the third. From these results, the model with the highest classification rate was selected giving the optimal values for γ and C. This optimal model was then used to classify the data into two separate groups.
Computational model architecture
The starting point for our computational model was published earlier in Proske et al. (2011). In short, the model’s main architecture includes the three different classes of nuclei found in the thalamus: a modal-specific relay nucleus (SP; comparable to the LGN in the visual circuit), a higher-order nucleus (NSP; comparable to the central lateral nucleus) and the inhibitory RTN. Each class of nuclei is represented by a group of 100 neurons, making three groups in total (Fig. 1). The model’s input consists of three random spike generators, all with a firing probability P (spikes per millisecond) that follows a Poisson distribution. Two spike train generators serve as input for the SP and NSP while another generator represents the cortical layer and serves as cortical input for the RTN.
Fig 1. Thalamic model’s main
architecture (reproduced with permis-sion from Proske et al., 2011). In total, three nuclei are represented (SP, NSP and RTN), each consisting of 100 neurons.
Connectivity between the different nuclei is characterized by excitatory connections from the SP and NSP projecting onto the RTN, with returning inhibitory connections from the RTN. One notable difference between the SP and NSP is the parallel arborization (one-to-one) between the SP and RTN versus a divergent arborization (one-to-many with a chance of .15) between the NSP and RTN. This follows neuroanatomical data from Steriade & Deschenes (1984). From both the SP and
NSP, excitatory connections run to the cortical layer in the model. Refer to Table 2 for an overview of all the model’s parameter values.
Table 2. Model’s architecture connectivity values for all parameters.
Connection Type 𝑖𝑖𝑠𝑠 Arborization 𝜏𝜏𝑠𝑠 Delay (ms)
Peripheral input SP Exc. .005 one-to-one 10 0
Peripheral input NSP Exc. .005 one-to-one 10 0
SP RTN Exc. .02 one-to-one 20 3 SP CRX Exc. .01 one-to-one 20 7 NSP RTN Exc. .01 one-to-one 20 3 NSP CRX Exc. .01 one-to-one 20 7 RTN SP Inh. .02 one-to-all (.15) 30 3 RTN NSP Inh. .03 one-to-all (.15) 30 3 CRX RTN Exc. .01 one-to-one 7 7
IFB neuron model
The thalamic neurons are modelled based on integrate-and-fire-or-burst (IFB) dynamics following Smith et al. (2000). The basis is the classical integrate-and-fire model from Knight (1972) with an added slow variable h that represents the inactivation of the T-type Calcium current 𝐼𝐼𝑇𝑇. The membrane potential of each neuron follows equation 1, integrating excitatory and inhibitory input over a decay variable 𝑖𝑖𝑠𝑠 and time constant 𝜏𝜏𝑠𝑠 according to equation 2 and 3. Excitatory connections have a faster time constant of 20 ms versus 30 ms for inhibitory connections. The membrane potential has a leakage over time, 𝐼𝐼𝐿𝐿, captured in equation 4, causing it to slowly return to its resting state value in the absence of input. Once the membrane potential reaches a threshold 𝑉𝑉𝜃𝜃, the neuron releases an action potential after which it is reset to 𝑉𝑉𝑟𝑟𝑟𝑟𝑠𝑠𝑟𝑟𝜋𝜋.
𝐶𝐶d𝑉𝑉d𝜋𝜋 = 𝐼𝐼𝑖𝑖𝑖𝑖𝑝𝑝− 𝐼𝐼𝐿𝐿− 𝐼𝐼𝑇𝑇 (Eq. 1) 𝐼𝐼𝑖𝑖𝑖𝑖𝑝𝑝= �𝐸𝐸𝑟𝑟𝑥𝑥𝑒𝑒/𝑖𝑖𝑖𝑖ℎ− 𝑉𝑉� ∑ 𝑖𝑖𝑖𝑖 𝑠𝑠𝑖𝑖 (Eq. 2) d𝑔𝑔𝑠𝑠 d𝜋𝜋 = −𝑔𝑔𝑠𝑠 𝜏𝜏𝑠𝑠 (Eq. 3)
𝐼𝐼𝐿𝐿= 𝑖𝑖𝐿𝐿(𝑉𝑉 − 𝐸𝐸𝐿𝐿) (Eq. 4)
The Calcium current 𝐼𝐼𝑇𝑇 follows equation 5. Under relatively depolarized conditions, meaning V is greater than 𝑉𝑉ℎ, the activation level of 𝐼𝐼𝑇𝑇, h, relaxes to zero with time constant 𝜏𝜏ℎ− (=20 ms). However, once hyperpolarization is sufficient, h relaxes to unity with time constant 𝜏𝜏ℎ+ (=100 ms), according to equation 6, meaning h grows increasingly larger until it gets activated. This activation is captured in equation 7, where 𝐻𝐻(∙) represents the Heaviside step function. Using these equations, time constant 𝜏𝜏ℎ− sets the duration of the burst, whereas 𝜏𝜏ℎ+ sets the duration of required hyperpolarization for the activation level h to reach its maximum.
𝐼𝐼𝑇𝑇 = 𝑖𝑖𝑇𝑇𝑖𝑖∞ℎ(𝑉𝑉 − 𝐸𝐸𝐿𝐿) (Eq. 5) dℎ d𝜋𝜋 = � −ℎ 𝜏𝜏⁄ ℎ− (𝑉𝑉 > 𝑉𝑉ℎ) (1 − ℎ) 𝜏𝜏⁄ ℎ+ (𝑉𝑉 < 𝑉𝑉ℎ) (Eq. 6) 𝑖𝑖∞= 𝐻𝐻(𝑉𝑉 − 𝑉𝑉ℎ) (Eq. 7)
In summary, the IFB neuron model holds two important thresholds, 𝑉𝑉𝜃𝜃 and 𝑉𝑉ℎ, that are responsible for producing an action potential. Excitatory and inhibitory input is integrated until the cell’s membrane potential reaches the first threshold 𝑉𝑉𝜃𝜃, after which it releases an action potential and is reset. Upon release, this action potential will be a single spike in case the calcium current is inactive (e.g., 𝑉𝑉ℎ is not reached) or a burst of several spikes when the calcium current has been activated. To demonstrate this difference, the membrane potential of two random IFB neurons from the model are plotted in Fig 2. When firing tonically, the cell’s membrane potential hovers around -41 to -44 mV before the release of an action potential. However, when firing with bursts, the membrane potential dips down below -50 mV, and with the Calcium current active, a burst of spikes is released. The values used for all cellular parameters in the model are shown in Table 3.
Table 3. IFB model’s cellular parameter values, similar to smith et al. (2000).
Parameter Unit RTN SP NSP 𝑉𝑉𝜃𝜃 mV -35 -35 -35 𝑉𝑉𝑟𝑟𝑟𝑟𝑠𝑠𝑟𝑟𝜋𝜋 mV -50 -50 -50 𝑖𝑖𝐿𝐿 mS / ms2 / cm2 .035 .035 .035 𝐸𝐸𝐿𝐿 mV -65 -65 -65 𝑖𝑖𝑇𝑇 mS / ms2 / cm2 .07 .07 .07
𝑉𝑉ℎ mV -64 -66 -66 𝜏𝜏ℎ− ms 40 20 20 𝜏𝜏ℎ+ ms 100 100 100 𝐸𝐸𝑇𝑇 mV 120 120 120 𝐶𝐶 μM / cm2 2 2 2 𝐸𝐸𝑟𝑟𝑥𝑥𝑒𝑒 mV 0 0 0 𝐸𝐸𝑖𝑖𝑖𝑖ℎ mV -85 -85 -85
Fig 2. The membrane potential of two random IFB neurons. On the left: a neuron firing tonically. On the right: a
neuron bursting with LTS. The bursts occur when the cell membrane is relatively hyperpolarized before an action potential – dipping down below -50 mV – before firing.
Model manipulation & data simulation
All simulations were done using IQR (http://iqr.sourceforget.net), free and open-source software created to run detailed neural simulations (Bernardet & Verschure, 2010). The model ran at steps of 1 ms, with values starting at zero. Data recording started after the model had settled in a steady mode of activity, around five to eight seconds after initialization. In total, data were recorded for 10.000 steps, equaling ten seconds of activity. The original model files and parameter settings can be accessed at
http://specs.upf.edu/thalamus_model.
Different bio-physiological states of the model were reached through the manipulation of the input generators. The starting point is the thalamic system in an awake state, with sufficient excitatory input (𝑃𝑃𝑠𝑠𝑝𝑝,𝑖𝑖𝑠𝑠𝑝𝑝 = 0.5, 𝑃𝑃𝑒𝑒𝑟𝑟𝑥𝑥 = 0.15) to put the system in a tonic firing mode. While in this mode, the SP nucleus shows the greatest activity, where the NSP and RTN are relatively inactive, displayed in Fig 6A. To simulate a stroke, deafferentation of the cortical layer in the model was applied. When a neuron is deafferentated, its output has no effect on the post-synaptic neuron. Different sizes of
deafferentation were explored to investigate the influence on the activity of the RTN and NSP/SP, starting small with 5 percent and increasing in steps of 5.
To quantify rhythmicity of firing rates of the model, a fast Fourier transformation was applied to a measure of the system’s local field potential (LFP). This LFP is calculated as the combined membrane potentials of all 300 neurons at each time point. Data for this FFT spanned all 10.000 data points of the model in its steady state of firing.
RESULTS
EEG Spectral Power
The resting-state EEG power spectra for each individual showed similar shape and size across time-points and across electrodes. Hence, data was averaged over time and over electrodes to create a single spectral power plot for each individual, shown in Fig 3. Across the entire frequency spectrum, the patient group showed distinct increased power over the lower frequencies (1 to 8 Hz) and decreased power over the higher frequencies (12 to 30 Hz). To test for significant differences between both groups, the Wilcoxon rank sum test was used with a bonferroni corrected α of 0.13. The patients showed significantly increased power in the delta band (p = .013) and significantly decreased power in the beta band (p < .001). The increased power for patients in the theta band shows a clear trend (p = .046) while the difference in the alpha band does not (p = .158).
Fig 3. Average spectral power. Individual spectra (grey lines) and group averages (colored lines) for patients (left) and controls (right). Patients show a slowing of the peak compared to controls. The asterisks mark the dominant power peaks between 6 and 12 Hz.
Each individual’s spectral power shows a dominant peak in the alpha frequency range. The height and position of these peaks are shown in Fig 4. Peaks were calculated as the highest power value in the range of 6 to 12 Hz. The patients’ peaks proved to be at a significantly lower frequency compared to the controls (p < .001). There is no significant difference however in the height of the peaks between both groups (p = .47).
Fig 4. Individual spectral power peaks. Patient (red) power peaks are at a lower frequency compared to controls (blue). There is no significant difference in height between both groups.
Topography of spectral energy
The topographical distribution of differences between patients and healthy controls for each frequency band are shown in Fig 5. Note that all non-midline electrodes have been swapped laterally for patients with a left hemispheric lesion, making the right hemisphere always ipsilateral and left hemisphere contralateral to the lesion. In de delta band, patients showed high positive Z-scores (e.g., an increase of activity) in occipital electrodes (strongest contra-lateral in O1) and around the central lateral electrodes (T3, T4, T6, C4 and F8) which is mainly concentrated ipsi-lateral to the lesion. Differences in the theta band are qualitatively similar, though slightly smaller in size. The alpha band shows nearly no differences at all, with most Z-scores below a value of 2. Finally, patients show high negative Z-scores (e.g., a decrease of activity) distributed across the entire scalp.
Classification
Two different models were trained in order to classify the data into two separate groups. For the first model, the feature vector only included dominant peak frequency as the dominant peak height did not significantly differ between groups. Cross-validation resulted in an optimal model with the value of 2−1for γ and 2−5 for C. Using this model, the entire dataset was classified with a correct classification rate of 79 percent, based solely on peak frequency.
The second model used a 4-dimensional feature vector and included all mean spectral power values of the four different frequency bands. Cross-validation resulted in the optimal values of 2−13 for γ and 27 for C, and applying this model to the entire dataset resulted in a near perfect correct classification rate of 94 percent, with one false-positive and one false-negative.
Fig 5. Z-score matrix of comparisons between patients versus controls at each electrode and frequency-point using the Wilcoxon rank sum test. On the right side: topographical distribution of these Z-scores for the four different frequency bands.
Clinical scales
The patient’s scores on the sub-acute NIHSS are shown in table 3. Of the 15 items, three are of particular interest (scales 3, 8 and 11) which focus on sensory, mainly visual information processing and neglect. In order to relate these type of symptoms with our EEG results, a correlation matrix was computed (Table 4), correlating frequency band power values with the scores on the 3 sensory scales. To correct for multiple-comparisons, a family-wise error rate (α) of .013 was maintained for each family-wise comparison for the three scales separately. As expected, reasonably high correlations of .58 and .66 were found in the delta (p < .01) and theta (p = .014) range, although only with scale 11.
None of the other correlations proved to be significant. Scale 11 is related to inattention or extinction to bilateral simultaneous stimulation in one or more modalities.
Table 4. Pearson’s correlation values of MSE and NIHSS subscales 3, 8 & 11.
Delta Theta Alpha Beta
Visual loss (3) .30 .17 -.12 .03 Sensory loss (8) .16 .11 -.23 -.17 Neglect (11) .66** .58* -.01 .08 * Trend with p = .014 ** Significant with p < .01
Computational model
To investigate the influence of a cortical lesion on the network behavior, deafferentation of different sizes were applied to the spike generator that represents the cortical layer and provides excitatory input for the RTN. Small lesions below 30 percent (30 out of 100 neurons) did not have a direct noticeable impact on the global network behavior, with the system remaining in a tonic mode of firing. However, once deafferentation reached 30 percent, the system made a switch from tonic to burst firing, shown in Fig 6B. When the lesion reached this size, excitatory input for the RTN decreased enough for the neuron’s membrane potential to reach a hyperpolarized state. From this point, the Calcium currents were activated and nearly all action potentials released were a burst rather than singular spikes.
The switch from tonic into burst firing of the entire system occurred gradually over a time period of several hundreds of milliseconds. At first, a small portion of reticular neurons release a burst of activity. This activity spreads mainly to the NSP through their divergent connectivity and leads to an increased number of NSP neurons becoming activated. Next, a greater portion of NSP neurons release a burst, again mainly aimed at the RTN. Over a short period of time, this alternating and reciprocal bursting of the NSP and RTN grows and causes the entire system to enter its state of burst firing, resulting in large synchronized and alternating bursts of the different nuclei. From this point onwards, the system remains in this mode until external input is manipulated or the model is stopped.
A fast Fourier transformation was applied to the data to quantify rhythmicity of firing rates in both the awake and stroke model, with the results plotted in Fig 7. The awake model shows FFT power values across many different frequencies following the 1 𝜋𝜋⁄ power law, with small peaks around 4 and 6 Hz. The stroke model is bursting at a highly reliable rate of 8.5 bursts per second, with a clear FFT power peak at 8.5 Hz and a resonance peak at 17 Hz.
Fig 6. Different modes of firing of the thalamic model. The bottom bar graph is binned activity over 10 steps,
with SP/NSP activity in dark gray and all three combined in light gray. (A) The model in its balanced awake state, with neurons firing tonically and uncorrelated. (B) The model in its bursting regime after introducing a 30 percent lesion in the cortical layer, causing reduced cortical recruitment of the RTN.
Fig 7. Smoothed fast Fourier transformation of the local field potential (LFP) of all three nuclei combined. The
FFT of the model in its balanced state (left) follows the1 𝜋𝜋⁄ power law, with small peaks around 4 and 6 Hz. In the model with a cortical lesion of 30 percent (right), bursting occurs in a rhythmic manner at 8.5 Hz, with a resonance peak at 17 Hz.
DISCUSSION
Resting-state EEG
This is the first study marking thalamocortical dysrhythmia as a possible mechanism for sensory-related post-stroke symptomology. In order to identify TCD, electro-neurophysiological data of ischaemic stroke patients were compared to those of healthy controls. In this data, the neural activation patterns of stroke patients proved to be consistent with TCD and similar to those of other patients with an established TCD syndrome – including neurogenic pain, tinnitus or Parkinson’s disease patients. The spectral power of stroke patients showed the typical characteristics of TCD, which is a slowing or attenuation of the dominant power peak and an increase of power over the lower frequency ranges.
Patients displayed increased power across the delta (1 to 4 Hz) and theta (4 to 8 Hz) band compared to healthy controls. A topographical map indicated that these lower frequency differences were localized contralateral to the lesion over the occipital region and mainly ipsilateral over the central lateral regions. Across the alpha range (8 to 12 Hz), patients did not differ in terms of mean spectral energy, but one of the most characteristic features of TCD emerged: the dominant power peaks were attenuated, shifting to a lower frequency. A model using support vector machines indicated that this attenuation was able to classify every individual relatively well, with a correct classification rate of 79 percent. This is a small improvement compared to previous results using LDA, with classification rates of 73 percent for peak frequency (Sarnthein et al., 2006).
Contrary to previous EEG studies on TCD, stroke patients displayed a decrease rather than an increase of activity over the higher frequencies (Schulman et al., 2011; Sarnthein et al., 2006; Stern et al., 2006). A possible explanation for this opposite effect is related to the brain’s response to the traumatic event of a stroke. During an ischaemic episode, the brain reacts by releasing great amounts of GABA, an inhibitory neurotransmitter (Green et al., 2000). This overwhelming increase of GABA causes a shutdown of the cortex in an attempt to limit the tissue damage. The EEG data was recorded within several days of the stroke, meaning cortical regions had drastically lowered their local activity. This decrease of local activity is reflected in the data through the wide-spread decrease of beta band activity across all cortical regions.
Computational model
Utilizing a computational model of the human thalamic system, we demonstrated that inducing a lesion in the cortical layer of the system leads to a switch in firing regimes of thalamic neurons: nuclei fire with large synchronized bursts rather than decorrelated tonic spikes. This switch is a result of decreased cortical recruitment of the thalamic reticular nucleus, leading to hyperpolarized membrane potentials. Under these conditions, Calcium currents in the cell’s membrane are activated and suprathreshold excitatory input now causes a low threshold calcium spike: an action potential with several sodium spikes riding on its crest.
This switch in firing regimes within the model starts after a cortical lesion reaches the size of 30 percent. Smaller lesions have little effect on the network behavior of the thalamic system. From 30 percent and larger, a small group of RTN neurons reach a state of relative hyperpolarization. This causes the neurons to send out bursts of spikes, increasing inhibition mainly of NSP neurons through divergent connectivity between both nuclei. This increased inhibition leads to hyperpolarization of NSP neurons, which in return start bursting and activate a greater portion of RTN neurons. Over a time period of hundreds of milliseconds, these nuclei enforce each other through growing reciprocal and alternating activity until the entire thalamic system is bursting at a highly reliable rate of 8.5 bursts per second. Such oscillations can entrain corticothalamic loops resulting in increased coherence between the regions. At the cortical level, low frequency oscillations have an influence on local interneurons, creating an imbalance in cortico-cortico inhibition. This imbalances allows for an increase of local activity that results in the positive symptoms of TCD (Llinás et al., 1999; Schulman et al., 2011).
As it currently stands, the computational model cannot yet fully explain the differences in spectral power for the patient group. A cortical lesion in the model results in a switch in firing regimes of thalamic nuclei, but it does not yet explain the shift of the dominant spectral power peak nor the increase of power over the entire lower frequency range. One possible reason is the inability of the
model, using our set parameter values, to burst rhythmically in the delta and low theta (1-6 Hz) range. Instead, it focused on the thalamus as a generator of low alpha oscillations, taking a dominant role in the production of 8.5 Hz oscillations, both in the model and in the spectral power peaks of the EEG data. However, the thalamus is also known to be a generator of sleep spindles and delta waves during sleep or drowsiness (Steriade et al., 1993; De Gennaro & Ferrara, 2003). Given our results it seems plausible that with TCD, the thalamus could be oscillating in the delta range while awake, which would explain the increase of spectral power over the lower frequency range (1-8 Hz). Intracranial EEG measurements could offer further explanations, but have not been possible up to date due to the invasiveness of this measure.
One additional affair to take into consideration is the clear boundary we set between bursting and tonic firing and consider the former pathological when awake while the latter is healthy. In practice this distinction is less clear, which is demonstrated by several animal studies. For instance, both burst and tonic firing of thalamic neurons has been marked as beneficial to task performance and sensory information processing for cats and macaque monkeys (Guido & Weyand, 1996; Weyand et al., 2001; Ramcharan et al., 2000). Conversely, studies on TCD in humans using micro-electrode implants in thalamic nuclei often demonstrate the opposite: increased thalamocortical coherence and low frequency bursting of the thalamus is pathological and results in positive symptoms in TCD (Sarnthein et al., 2003; Sarnthein et al., 2007, 2008). Both lines of research however do agree on the fact thalamic nuclei are never in either one mode or the other, where often both tonic and bursting circuits affect the other. For example, Proske and colleagues (2011) report the interference of unaffected circuits on the re-depolarization in affected circuits, leading to changes in bursting frequency of the affected circuit.
Application and treatment
Treatments for TCD are rather invasive and fall into three different categories: (i) deep brain stimulation (DBS) with Parkinson’s disease patients (Deuschl et al., 2006), (ii) neurosurgical removal of the affected thalamic nuclei for neurogenic pain patients (Jeanmonod et al., 2001), or (iii) the use of electrical or magnetic stimulation of cortical regions – in the form of repetitive transcranial magnetic stimulation (rTMS) or transcranial direct current stimulation (tDCS) – for stroke or tinnitus patients (Kleinjung et al., 2007; Marcondes et al., 2008). Of these three types, rTMS and tDCS have already been applied to the treatment of stroke as a supportive method, increasing the effectiveness of training tasks to enhance memory performance or language learning. Conversely, by placing the electrodes over primary sensory regions – such as visual, auditory or motor cortices instead of prefrontal or temporal regions – it might be possible to increase neuronal activity in these areas with increased cortical recruitment of the thalamus as a result. A comparable procedure was used in an attempt to relieve neurogenic pain patients of their pain symptoms by applying rTMS with electrodes over the motor cortices, with reasonable success (Lefaucheur et al., 2004; Cruccu et al., 2007).
Another interesting outcome is related to our use of support vector machines. The model that included the average power values for the different frequency bands into the feature vector managed to reach a near perfect classification rate of 94 percent. Up to date, many studies have tried to improve or sophisticate EEG as a diagnostic tool in hospitalized use (Jordan, 2004; van Putten, 2004) or relate neuronal data with stroke symptoms and recovery (Finnigan et al., 2004, Schleiger et al., 2014). While our method is still rudimental in its application – using small group sizes and few data points – it has the potential to build a functional and complete model that can assist in the diagnosis of ischaemic stroke, together with the assessment of post-stroke symptoms as well as the recovery process.
Conclusion
Combining electro-neurophysiological data of stroke patients together with a computational model of the thalamic system, we carefully conclude that an ischaemic stroke leads to the emergence of thalamocortical dysrhythmia. We found the most distinct characteristics of TCD in the resting-state EEG data of stroke patients and support this notion with our modelling results. Taken together, this is the first study linking thalamocortical dysrhythmia with acute ischaemic stroke as a possible mechanism in the development of sensory-related symptoms. While much more work remains to be done, sensory-related post-stroke symptomology should be viewed in the light of this mechanism and might explain some of the indirect neurological deficits resulting from a lesion.
Acknowledgements
The authors would like to thank several people for their contribution to this thesis. First and foremost, M. Cohen as a supervisor for providing important feedback and assistance regarding most EEG analyses and writing of the manuscript. Secondly, A. Duff and H. Proske for helping with the development and analysis of our computational model. Thirdly, L. Waldorp for assisting in the development and application of our support vector machines. Finally, F. van Arendonk and M. Boffo who were of great support with critical comments in the earlier stages of the project and during the writing of this manuscript.
Abbrevations
LGN, lateral geniculate nucleus; MGB, medial geniculate body; MCA, middle cerebral artery; RTN, reticular thalamic nucleus; SP, specific nucleus; NSP, non-specific nucleus; TCD, thalamocortical dysrhythmia; LTS, low threshold calcium spike; IFB, integrate-and-fire-or-burst neuron; 𝐼𝐼𝑇𝑇, Calcium (Ca2+) current; EEG, electro-encephalogram; MSE, mean spectral energy; FFT, fast Fourier transformation; LDA, linear discriminant analysis; SVM, support vector machines.
References
Alitto, H. J., & Usrey, W. M. (2003). Corticothalamic feedback and sensory processing. Current opinion in neurobiology, 13(4), 440-445.
Bernardet, U., & Verschure, P. F. (2010). iqr: A tool for the construction of multi-level simulations of brain and behaviour. Neuroinformatics, 8(2), 113-134.
Churchill, L., & Kalivas, P. W. (1999). The involvement of the mediodorsal nucleus of the thalamus and the midbrain extrapyramidal area in locomotion elicited from the ventral pallidum. Behavioural brain research, 104(1), 63-71.
Cohen, M. X. (2014). Analyzing neural time series data. Cambridge: MIT Press.
Cruccu, G., Aziz, T. Z., Garcia‐Larrea, L., Hansson, P., Jensen, T. S., Lefaucheur, J. P., ... & Taylor, R. S. (2007). EFNS guidelines on neurostimulation therapy for neuropathic pain. European Journal of Neurology,14(9), 952-970.
Cummins, T. D., & Finnigan, S. (2007). Theta power is reduced in healthy cognitive aging. International Journal of Psychophysiology, 66(1), 10-17.
Cummins, T. D., Broughton, M., & Finnigan, S. (2008). Theta oscillations are affected by amnestic mild cognitive impairment and cognitive load. International Journal of Psychophysiology, 70(1), 75-81. Delorme, A., & Makeig, S. (2004). EEGLAB: an open source toolbox for analysis of single-trial EEG dynamics
including independent component analysis. Journal of neuroscience methods, 134(1), 9-21.
Deuschl, G., Schade-Brittinger, C., Krack, P., Volkmann, J., Schäfer, H., Bötzel, K., ... & Voges, J. (2006). A randomized trial of deep-brain stimulation for Parkinson's disease. New England Journal of Medicine, 355(9), 896-908.
Douglas, R. J., & Martin, K. A. C. (2010). Canonical cortical circuits. Handbook of Brain Microcircuits, 15-21. De Gennaro, L., & Ferrara, M. (2003). Sleep spindles: an overview. Sleep medicine reviews, 7(5), 423-440. Finnigan, S. P., Rose, S. E., Walsh, M., Griffin, M., Janke, A. L., McMahon, K. L., ... & Chalk, J. B. (2004).
Correlation of Quantitative EEG in Acute Ischemic Stroke With 30-Day NIHSS Score Comparison With Diffusion and Perfusion MRI. Stroke, 35(4), 899-903.
Green, A. R., Hainsworth, A. H., & Jackson, D. M. (2000). GABA potentiation: a logical pharmacological approach for the treatment of acute ischaemic stroke. Neuropharmacology, 39(9), 1483-1494.
Guido, W., & Weyand, T. (1995). Burst responses in thalamic relay cells of the awake behaving cat. Journal of Neurophysiology, 74, 1782-1782.
Jeanmonod, D., Magnin, M., & Morel, A. (1996). Low–threshold calcium spike bursts in the human thalamus Common physiopathology for sensory, motor and limbic positive symptoms. Brain, 119(2), 363-375. Jeanmonod, D., Magnin, M., Morel, A., & Siegemund, M. (2001). Surgical control of the human thalamocortical
dysrhythmia: I. Central lateral thalamotomy in neurogenic pain. Thalamus & Related Systems, 1(01), 71-79.
Jeanmonod, D., Schulman, J., Ramirez, R., Cancro, R., Lanz, M., Morel, A., ... & Llinás, R. (2003). Neuropsychiatric thalamocortical dysrhythmia: surgical implications. Thalamus & Related Systems, 2(2), 103-113.
Jones, E. G. (1985). The thalamus. New York: Plenum Press.
Jones, E. G. (1998). Viewpoint: the core and matrix of thalamic organization. Neuroscience, 85(2), 331-345. Jones, E. G. (2002). Thalamic circuitry and thalamocortical synchrony. Philosophical Transactions of the Royal
Society of London. Series B: Biological Sciences, 357(1428), 1659-1673.
Kleinjung, T., Eichhammer, P., Landgrebe, M., Sand, P., Hajak, G., Steffens, T., … & Langguth, B. (2008). Combined temporal and prefrontal transcranial magnetic stimulation for tinnitus treatment: a pilot study. Otolaryngology--Head and Neck Surgery, 138(4), 497-501.
Lefaucheur, J. P., Drouot, X., Menard-Lefaucheur, I., Zerah, F., Bendib, B., Cesaro, P., ... & Nguyen, J. P. (2004). Neurogenic pain relief by repetitive transcranial magnetic cortical stimulation depends on the origin and the site of pain. Journal of Neurology, Neurosurgery & Psychiatry, 75(4), 612-616.
Llinás, R. R. (1988). The intrinsic electrophysiological properties of mammalian neurons: insights into central nervous system function. Science, 242(4886), 1654-1664.
Llinás, R. R., Ribary, U., Jeanmonod, D., Kronberg, E., & Mitra, P. P. (1999). Thalamocortical dysrhythmia: a neurological and neuropsychiatric syndrome characterized by magnetoencephalography. Proceedings of the National Academy of Sciences, 96(26), 15222-15227.
Llinás, R., Urbano, F. J., Leznik, E., Ramírez, R. R., & van Marle, H. J. (2005). Rhythmic and dysrhythmic thalamocortical dynamics: GABA systems and the edge effect. Trends in neurosciences, 28(6), 325-333.
Llinás, R. R., & Steriade, M. (2006). Bursting of thalamic neurons and states of vigilance. Journal of neurophysiology, 95(6), 3297-3308.
Marcondes, R. A., Sanchez, T. G., Kii, M. A., Ono, C. R., Buchpiguel, C. A., Langguth, B., & Marcolin, M. A. (2010). Repetitive transcranial magnetic stimulation improve tinnitus in normal hearing patients: a double‐blind controlled, clinical and neuroimaging outcome study. European Journal of Neurology, 17(1), 38-44.
Moazami-Goudarzi, M., Sarnthein, J., Michels, L., Moukhtieva, R., & Jeanmonod, D. (2008). Enhanced frontal low and high frequency power and synchronization in the resting EEG of parkinsonian patients. Neuroimage, 41(3), 985-997.
Moazami-Goudarzi, M., Michels, L., Weisz, N., & Jeanmonod, D. (2010). Temporo-insular enhancement of EEG low and high frequencies in patients with chronic tinnitus. QEEG study of chronic tinnitus patients. BMC neuroscience,11(1), 40.
Proske, J. H., Jeanmonod, D., & Verschure, P. F. (2011). A computational model of thalamocortical dysrhythmia. European Journal of Neuroscience, 33(7), 1281-1290.
Putten, M. J. van, & Tavy, D. L. (2004). Continuous quantitative EEG monitoring in hemispheric stroke patients using the brain symmetry index.Stroke, 35(11), 2489-2492.
Ramcharan, E. J., Gnadt, J. W., & Sherman, S. M. (2000). Burst and tonic firing in thalamic cells of unanesthetized, behaving monkeys. Visual neuroscience, 17(01), 55-62.
Sarnthein, J., Morel, A., Von Stein, A., & Jeanmonod, D. (2003). Thalamic theta field potentials and EEG: high thalamocortical coherence in patients with neurogenic pain, epilepsy and movement disorders. Thalamus & Related Systems, 2(3), 231-238.
Sarnthein, J., Stern, J., Aufenberg, C., Rousson, V., & Jeanmonod, D. (2006). Increased EEG power and slowed dominant frequency in patients with neurogenic pain. Brain, 129(1), 55-64.
Sarnthein, J., & Jeanmonod, D. (2007). High thalamocortical theta coherence in patients with Parkinson's disease. The Journal of neuroscience, 27(1), 124-131.
Sarnthein, J., & Jeanmonod, D. (2008). High thalamocortical theta coherence in patients with neurogenic pain. Neuroimage, 39(4), 1910-1917.
Schleiger, E., Sheikh, N., Rowland, T., Wong, A., Read, S., & Finnigan, S. (2014). Frontal EEG delta/alpha ratio and screening for post-stroke cognitive deficits: The power of four electrodes. International Journal of Psychophysiology, 94(1), 19-24.
Schulman, J. J., Cancro, R., Sandlin Lowe, F. L., Walton, K. D., & Llinás, R. R. (2011). Imaging of thalamocortical dysrhythmia in neuropsychiatry. Frontiers in human neuroscience, 5, 69.
Sherman, S. M., & Guillery, R. W. (1998). On the actions that one nerve cell can have on another: distinguishing “drivers” from “modulators”. Proceedings of the National Academy of Sciences, 95(12), 7121-7126. Sherman, S. M. (2001). Tonic and burst firing: dual modes of thalamocortical relay. Trends in
neurosciences, 24(2), 122-126.
Sherman, S. M., & Guillery, R. W. (2001). Exploring the thalamus. Elsevier.
Sherman, S. M., & Guillery, R. W. (2002). The role of the thalamus in the flow of information to the cortex. Philosophical Transactions of the Royal Society of London. Series B: Biological Sciences, 357(1428), 1695-1708.
Smith, G. D., Cox, C. L., Sherman, S. M., & Rinzel, J. (2000). Fourier analysis of sinusoidally driven thalamocortical relay neurons and a minimal integrate-and-fire-or-burst model. Journal of Neurophysiology, 83(1), 588-610.
Steriade, M., & Deschenes, M. (1984). The thalamus as a neuronal oscillator. Brain Research Reviews, 8(1), 1-63.
Steriade, M., McCormick, D. A., & Sejnowski, T. J. (1993). Thalamocortical oscillations in the sleeping and aroused brain. Science, 262(5134), 679-685.
Steriade, M., & Timofeev, I. (2003). Neuronal plasticity in thalamocortical networks during sleep and waking oscillations. Neuron, 37(4), 563-576.
Stern, J., Jeanmonod, D., & Sarnthein, J. (2006). Persistent EEG overactivation in the cortical pain matrix of neurogenic pain patients. Neuroimage, 31(2), 721-731.
Van Horn, S. C., Erişir, A., & Sherman, S. M. (2000). Relative distribution of synapses in the A‐laminae of the lateral geniculate nucleus of the cat. Journal of Comparative Neurology, 416(4), 509-520.
Weyand, T. G., Boudreaux, M., & Guido, W. (2001). Burst and tonic response modes in thalamic neurons during sleep and wakefulness. Journal of neurophysiology, 85(3), 1107-1118.
