Deep brain stimulation increases motor cortical 1/f

Deep brain stimulation increases motor cortical 1/f noise and decouples high gamma
amplitude from beta phase
Scott R. Cole, Erik J. Peterson, Coralie de Hemptinne, Philip A. Starr, Bradley Voytek
Deep brain stimulation (DBS) of the subthalamic nucleus is a common and effective treatment
for Parkinsonian motor signs, including bradykinesia and rigidity. Recently, it was discovered
that Parkinsonโ€™s Disease patients show pathological overcoupling between the beta phase and
high gamma amplitude in the primary motor cortex (M1). DBS-induced decoupling may underlie
improvement of Parkinsonian motor signs. In frontal and temporal regions, phase-amplitude
coupling decreases with age. Empirical findings and computational simulations support the
hypothesis that this reduction in coupling is caused by decreased synchrony in population spike
timing. This desynchronization is also associated with increased 1/f noise, manifested as a
flattening of the slope of the power spectral density.
We test the hypothesis that the neurophysiological mechanism of high-frequency DBS of the
STN is the desynchronization of M1 spiking. This hypothesis predicts that the decrease in M1
coupling is the result of population spike desynchronization, and thus the DBS-induced
reduction in coupling should be predicted by the DBS-induced increase in 1/f noise.
Electrocorticography recordings were obtained over M1 in Parkinsonโ€™s Disease patients before
and during DBS. We observed that DBS caused: 1) significant decreases in coupling; 2)
significant flattening of the power spectrum, and most importantly; 3) a correlation between the
magnitude of the changes in (1) and (2) such that increased 1/f noise predicts the drop in
coupling. DBS-induced spiking desynchronization seems to decrease pathological overcoupling
allowing M1 to respond more dynamically to signals from frontal executive areas.
Exploring the Neural Basis of the Electrophysiological Power Spectrum
Richard Gao & Bradley Voytek
The power spectrum of meso- and macro-scale brain electrical recordings in the forms of the
local field potential (LFP), electrocorticogram (ECoG), and electroencephalogram (EEG) are
often described to be following an inverse power law relationship, given by ๐‘ƒ = ๐ด๐น!!, where F is
frequency, P is power, and A and ๐œ’ are free parameters characterizing the power law. In the
log-log domain, this relationship is represented by a linear trend with a negative slope of โ€“ ๐œ’ and
a y-intercept of ๐‘™๐‘œ๐‘”๐ด. This phenomenon has been well documented in empirical data, noting
changes in A and ๐œ’ during various perceptual and motor tasks. In addition, recent computational
models using population-spiking neurons have attributed these parameters to different biological
mechanisms.
While the power law formulation of the spectrum has proven fruitful, two key observations are
unsatisfactorily accounted for. First, there have been reports of an increase in strictly high
gamma power (>60Hz), resulting in a curling of the spectrum, without changes in the slope or
intercept. Second, rotations of the spectrum resulting from a change in slope are observed to be
about a frequency of ~25Hz, instead of 1Hz, which would be the case if only a change in ๐œ’ were
to occur.
Here, we argue that a strict inverse power law model is an incomplete description of the
underlying processes giving rise to the power spectrum, and propose the addition of an additive
term, i.e. ๐‘ƒ = ๐ด๐น!! + ๐ต. Furthermore, we postulate that B, a broadband signal akin to white noise,
arises from de-correlated (Poissonic) population firing engaged in local computation. Using a
Poisson population model, we demonstrate that an increased firing rate leads to both an
increase in gamma power (high frequency curling) and a rotation of the spectrum about ~25Hz.
In addition, we validate our model by demonstrating an improved fit of the power spectrum
derived from human ECoG and rat LFP data. In summary, the new formulation has both
explanatory powers over the data and a sound neurophysiological basis, improving our
understanding of the power spectrum of electrophysiological data.
Influencing visual target detection with oscillatory phase-specific stimulus presentation
Robert J. Gougelet, Thomas Donoghue, Matthew Piper, Alric Althoff, Thomas P. Urbach,
Bradley Voytek
We investigated the extent to which ongoing electroencephalography (EEG) neural oscillations
emanating from fronto- parietal and occipital scalp regions in the theta (3-7 Hz) and alpha (8-12
Hz) frequency ranges contribute to the detection of visual stimuli in humans. We first observed
how subject-specific alpha phase affects the detection of a cued, near-threshold visual target
such that the visual cortical alpha phase at the time of visual stimulus onset biased target
detection, replicating previous reports.
We extended these observations by presenting visual targets at specific phases of the ongoing
oscillatory alpha using a real-time oscillatory phase tracking system. We implemented online
phase tracking in two ways, and compare the efficacy of them both. In the first, we sampled the
ongoing EEG datastream and peak filtered it using a phase- and group-delay compensated
Parks-McLelland FIR digital bandpass filter, centered at the pre-determined maximum amplitude
and center frequency of the subject-specific occipital alpha oscillation. Only periods when the
ratio of alpha/broadband power spectra reached a predetermined threshold were isolated for
phase detection. Peaks and troughs of the filtered datastream were then extracted to determine
the periodic timing of the ongoing alpha phase.
We extrapolated the timing characteristics of the detected peaks to predict peak and trough
phase intervals beyond the causal window of the datastream, and presented stimuli during such
intervals in real- time. In the second, we recorded a few minutes of resting EEG to identify
individual visual cortical alpha center frequency. These data also allow us to identify alpha
peaks and troughs using a simple thresholding procedure wherein the top and bottom 0.1% of
the sorted amplitude values of the raw, ongoing EEG reflect, with very high accuracy (> 95%),
individual alpha peaks and troughs. Given the stability of the alpha occipital rhythm over short (<
1 cycle) time frames, we attempted to present stimuli during specific phases of the dominant
oscillatory alpha as well.
Neural network properties can be inferred from electrophysiological power spectral
geometry
Torben Noto, Richard Gao, Erik Peterson, Bradley Voytek
The study of the biophysical and cognitive role of neural oscillations has become a cornerstone
of modern neuroscience. These oscillations are inferred from the power spectral density (PSD)
of the neurophysiological signal of interest. The PSD of electrophysiological neural activity
assumes a general 1/f form. Although changes of power in narrow frequency bands (alpha, beta,
etc.) have been related to a variety of cognitive and behavioral states, and the broadband power
(the offset) of this process has been shown to reflect aggregate population spiking activity, there
is scant evidence for how other global properties of power spectral geometry relate to the
underlying neural network activity.
Treating the neural power spectra as a holistic entity affords the application of different analyses
that may provide novel insights that would not be evident in narrow bands. Presumably, neural
networks produce characteristic changes in the full spectrum (aside from narrowband
oscillations) under different operational modes, so it may be possible to estimate certain
features of the network from its geometry. Concurrent single cell and local field recordings from
rat hippocampus and recordings directly from human cortex allow us to probe these
relationships. The slope of the broadband spectrum (10-100 Hz) had a positive correlation with
spike count (r=0.35) and a negative correlation with the fano factor of the inter-spike interval (r=
-.35). Additionally, the slope of high gamma (80-125 Hz) negatively correlated with
phase/amplitude coupling (r = -0.1). These results draw a link between spectral geometry,
network properties, and neurobiology, and support the idea that the power spectrum should be
considered as a holistic entity that contains a wealth of information about the network that
produces it.
Spike-field coupling does not imply spike-spike coupling
Erik Peterson & Bradley Voytek
The origin and function of oscillatory activity remains a major outstanding question in
neuroscience. One prominent hypothesis for the functional role of gamma oscillations is
'communication through coherence'. This theory posits that regional coherence enhances
communication by increasing the precision of spike timing, i.e. spike-spike coupling. The focus
on coherence has lead to computational investigations of already oscillating populations. While
important in establishing coherence as useful for communications, and in showing how
information flow is maximized when coherence between oscillating pairs is maximized, these
studies skip over a basic question: is spike-time precision enhanced by the onset and amplitude
of gamma oscillations?
By definition, oscillating neural populations have repeating periods of decreased firing. If all else
is held equal, these periods of relative silence would mean a decrease in information flow. As
firing declines so does information. If oscillations increase information flow, they must alter
spiking to overcome these 'silent costs'. Keeping with the idea that oscillations alter spike timing,
and using Hodgkin-Huxley neurons in classic excitatory- inhibitory configurations, we simulated
the effect of gamma onset and amplitude on spike precision and on information flow.
Our simulations suggest a much larger range of parameters can generate gamma oscillations,
compared to only a narrow range of parameters that can actually increase precision and
information transmission in excitatory neurons. From a theoretical perspective, our results
suggest the 'communication through coherence' hypothesis may require fairly stringent
biophysical constraints to function as proposed. When aggregating over all models, gamma
power does not statistically predict spike precision, nor does a change to spike-field coupling
imply a change in spike-spike coupling. In sum these results suggest gamma oscillations, when
driven solely by excitatory-inhibitory interactions, reflect mostly silent periods rather than the
spike-time shifting necessary for enhanced precision.
Oscillatory visual cortical alpha disruptions in age-related working memory impairments
Tammy Tran, Nicole Hoffner, Bradley Voytek
Neural oscillations in the visual cortex play an important role in attention and memory.
Oscillations, in particular 8-12 Hz alpha activity, support interregional communication, and
ongoing fluctuations in alpha power and phase bias perception and cognition. While these
phenomena are fairly well characterized individually, here we examine the overlap between
visual attention and working memory and how these oscillatory processes are affected by
healthy aging.
We used electroencephalographic (EEG) recordings to compare behavioral performance and
alpha activity of younger and older adults (20-30 and 60-70 years old) during a lateralized,
visual working memory task. In this task, subjects were first presented with a non-informative,
foveally-presented alerting cue indicating the beginning of a trial. This cue was followed by brief
presentation of a visual working memory array and a delay period. Subjects then reported if a
second, test array was the same as or different from the memory array. In this task, older adults
showed increased reaction times overall and decreased accuracy in high memory load trials.
Both groups showed decreased contralateral delay activity with increasing memory load.
Analysis of extrastriate alpha power revealed decreased contralateral power during the delay
period in both groups. While older adults also showed decreased ipsilateral alpha power,
younger adults showed increased ipsilateral power instead, revealing significant lateralized
alpha power differences as a function of age. Analysis of alpha phase revealed that, while
memory array presentation equally increased extrastriate phase-resetting (or intertrial
coherence, ITC) in both groups, younger adults showed strong alerting-cue-induced ITC that
was nearly absent in older adults. These results suggest that in performing this task, older
adults make less use of the alerting cue than do younger adults, and older adults may rely more
heavily on bilateral visual cortical attention systems than do younger adults.
Auditory attention modulates frontal and temporal oscillatory dynamics in humans:
Evidence from electrocorticography
Roemer van der Meij, Aurélie Bidet-Caulet, Josef Parvizi, Nathan Crone, Edward Chang, Robert
T. Knight, Bradley Voytek
The brain needs to flexibly route information in distributed neuronal networks to meet the needs
of rapid environmental changes. This selective communication between neuronal populations
could be achieved via oscillatory dynamics. One such oscillatory phenomenon is phaseamplitude coupling (PAC), which reflects the hierarchical modulation of oscillations at different
frequencies, but also the phase-coupled modulation of local neuronal spiking activity. The latter,
manifested as high gamma activity (HG; 70-250 Hz) is modulated by oscillations at multiple
frequencies dependent on task demands. HG can be observed as broadband increased power
in the power spectrum of micro- to mesoscale measurements in human electrocorticographic
data obtained from subdural recordings (ECoG).
Recent evidence suggests that the power spectrum, which reflects both oscillatory and nonoscillatory processes, can provide insight into the dynamics supporting goal-directed behavior.
Notably, an upward rotation (flattening) of the power spectrum could reflect a shift in neuronal
resources from a regime of tight oscillatory inhibition towards one of increased sensitivity to
incoming signals. Local HG activity has been shown to be modulated by low frequency rhythms
that are coherent over broader regions suggesting that power spectral changes could be
coordinated in a similar manner. To investigate this, we obtained ECoG recordings from 8
epilepsy patients undergoing resective surgery while they performed a dichotic attention task.
They had to detect deviant sounds within a relevant stream while ignoring an irrelevant acoustic
stream. A third condition (control condition) was added in which all sounds received the same
amount of attention.
We found that fronto-temporal HG activity increased, frontal theta increased, and temporal
alpha decreased as a function of attention. Additionally, we also observed upward rotations (a
flattening) of the power spectrum at temporal electrodes, which captures both the decrease of
temporal alpha, and the increase of temporal HG activity. Moreover, this spectral flattening was
found to be phase-locked to distributed frontal theta rhythms, suggesting temporally coordinated
changes in neuronal recruitment. Taken together, these results show the modulation of local
neuronal activity by distributed oscillatory rhythms as a function of task demands. Importantly,
this modulation occurred (1) on timescales as short as several 100ms, and (2) between distinct
regions including auditory cortex and frontal areas. The results provide evidence that oscillatory
dynamics provide a key mechanism for routing of information in the brain.