Research Statement SIGNIFICANCE Memory impairments are common to several neurological and
psychiatric disorders, including Alzheimer's disease and depression, and these impose a
heavy burden on patients, families and society (Dickerson and Eichenbaum, 2007). Novel
treatment and diagnostic strategies are needed, and these may arise from a deeper
understanding of the brain basis of episodic memory (Tulving, 1983).
Group-averaged neuroimaging studies have revealed that a distributed network, known as
the 'default network' (DN), increases activity during the recollection of past events
(Buckner et al. 2008). This network occupies regions including posteromedial cortex
(PMC), posterior parietal cortex (PPC), and the medial temporal lobe (MTL), as well as
lateral temporal and lateral and medial prefrontal cortices. Building on recent advances
in functional magnetic resonance imaging (fMRI; Poldrack et al., 2015; Laumann et al.,
2015), recent evidence has shown that when functional anatomy is defined in individuals,
the DN comprises at least two juxtaposed networks, named DN-A and DN-B for convenience
(Figure 1). This finding forces us to reconsider the role of the DN in episodic processes
(see also: Dastjerdi et al., 2011; Andrews-Hanna et al., 2010). Here we propose
experiments to deepen our understanding of these networks using a multimodal approach
that provides high spatiotemporal resolution and whole-brain network definition. We will
combine within-individual fMRI mapping with intracranial electroencephalography (iEEG)
and electrical brain stimulation (EBS). We will directly record local field potentials
from precisely mapped network regions, and apply electrical stimulation with millimeter
precision. This will provide novel information regarding episodic memory in two domains
that cannot be gathered by fMRI alone: i) characterizing fast temporal dynamics of
network recruitment during episodic recollection, and ii) establishing causal
interactions between brain regions during recollection.
INNOVATION Methodologically, this project will provide proof of principle that precision
fMRI mapping can be performed in a clinical population and successfully combined with
invasive recordings and stimulation. Theoretical innovation will be obtained through a
deeper understanding of the task-response dynamics, coupling, and causal relationships
between regions of distributed networks, including how neural engagement changes during
memory recollection. Finally, this proposal provides translational innovation by directly
testing whether precision-fMRI guided intracranial stimulation can be used modulate
memory performance.
APPROACH General methods: Participants in the proposed experiments will be neurosurgical
patients with presumed focal epilepsy that are to undergo implantation with intracranial
electrodes for localizing seizure foci. The proposal will be carried out at the
Northwestern University Feinberg School of Medicine. Patients scheduled for intracranial
seizure monitoring will be invited to enroll in the study and will undergo 1 to 4
sessions of fMRI prior to surgical implantation of electrodes. After surgery, patients
are typically monitored for ~7 days in the Northwestern Memorial Hospital Comprehensive
Epilepsy Centre (CEC), during which they will be invited to participate in the proposed
experiments. All subjects must provide informed consent before participating.
Enrollment: A minimum of 40-50 patients are expected to be monitored at the CEC over the
next 3 years. Electrode locations are determined by the clinical needs of the patient.
60-70% of patients are typically implanted with dense coverage of the medial temporal
lobes achieved through depth electrodes with trajectories that allow sampling of lateral
temporal cortices. A small number of electrodes are also typically implanted in posterior
cingulate, lateral inferior parietal and ventromedial prefrontal cortex. Due to the
distributed nature of the networks under investigation, which contain regions in multiple
cortical zones, it is likely that we will have coverage over relevant brain regions in
many cases. Some patients are also likely to be implanted with broader cortical coverage
using subdural grids. Preliminary results have shown that even when a patient is
implanted only with depth electrodes, which are not placed on the cortical surface but
penetrate into the brain, coverage of different candidate network regions was often
achieved along the electrode trajectory. With conservative estimates, 20-30 subjects will
be good candidates for the project aims outlined below. Given the high signal-to-noise
ratio of iEEG (usually a 200-300% task-evoked increase in signal from baseline; Parvizi
and Kastner, 2017), reliable effects can typically be found within individuals. All
proposed analyses will be carried out within individuals, hence multiple subjects are
required to generalize the findings, not increase statistical power. Therefore, a small
number of subjects (as low as n = 12) would be sufficient (e.g. Braga and Buckner, 2017;
Foster et al., 2013).
Neuroimaging acquisition: MR scans will be collected in 1-4 sessions from each patient.
Preliminary data has shown that in this clinical population 2-3 MRI sessions are
desirable to allow exclusion of non-compliant runs (e.g. those containing excess head
motion). We will collect 6-8 runs of fMRI data per session, resulting in between 42 - 224
mins of fMRI data per patient. This will allow robust and reliable estimates of network
topography. Subject sleepiness will be monitored through an in-scanner eye-tracking
camera. Compliance may be improved by allowing patients to watch movies inside the
scanner when needed, with pilot analyses showing comparable maps are obtained using movie
and visual fixation task data. Hence both tasks will be administered to improve
compliance.
Network definition within individuals: Networks will be defined within individuals using
two methods to ensure robustness. MRI preprocessing will be performed using a custom
pipeline 'iProc' that optimizes within-subject alignment and minimizes blurring.
Individual seed regions will be hand-selected and correlation maps will be thresholded at
r > 0.2 to remove regions of low certainty. The networks of interest, DN-A and DN-B, will
be targeted and identified using the expected anatomical distribution of each network
(described in detail in Braga and Buckner, 2017). Once candidate seed regions are
selected, definition of networks will be performed again in each individual using
data-driven clustering, which reduces potential experimenter bias. Networks from the
clustering analysis that most closely match up with the networks defined by hand will be
selected and labelled as DN-A and DN-B. Network maps will be used to label electrode
contacts (each 'electrode' can have multiple 'contacts' along its shaft or grid) by their
approximate location within or near each network.
Electrode localization: Electrode locations will be determined using a computerized
tomography (CT) scan. Estimates of the center of each contact in CT space will be
obtained using BioImage Suite. The CT image will be registered to the anatomical T1 image
(containing brain tissue locations) using a linear transform, allowing coordinates of
each contact to be projected to the T1 space. Preliminary data has shown that the
inter-rater error in this localization process is typically ~1mm. A 2-mm radius sphere
will be generated centered on each contact coordinate to approximate the sampling volume
of each contact, which is extended due to tissue conductance. Contacts that are
predominantly sampling white matter will be removed by excluding contacts whose sphere
does not overlap with the gray matter ribbon (estimated using FreeSurfer). The overlap
between spheres and gray matter will be used for surface-based and volume-based
functional connectivity (FC) analyses. FC maps will be created for each contact, and the
resulting maps will be visualized. If a contact fails to produce a FC map with distant
regions of high correlation (indicating that the contact is sampling a distributed
network), the contact will be excluded. If the contact's FC map resembles DN-A and DN-B,
as defined using the clustering and manually defined seed-based analyses, this contact
will be labelled as sampling DN-A and DN-B and included for further analysis. Two nearby
electrodes, one situated in DN-A and one in DN-B, will be selected a priori in two
different cortical zones (e.g. PMC vs. PPC, based on coverage).
iEEG processing: All contacts within the epileptic zone or corrupted by external noise
will be removed from further analysis. Raw signals will be notch filtered at 60, 120 and
180 Hz to remove electrical noise and harmonics. Notch-filtered signals will be
re-referenced by subtracting the common average, after removal of pathogenic or spiky
signals, as well as those presenting as clear outliers in power spectra plots. Data will
be bandpass filtered to extract amplitude and phase information at different frequency
bands. The high-frequency broadband (HFB; 70-140 Hz) signal is an important surrogate for
local neuronal population activity and corresponds to low-frequency correlations of the
blood oxygenation-level dependent signal (Logothetis et al., 2001). HFB band-limited
power will be calculated and low-pass filtered at <0.1 Hz. Pair-wise correlations in HFB
power will be used to estimate functional connectivity.
Direct cortical stimulation: Risks associated with the research stimulation protocol are
considered incremental and are further reduced by carrying out the stimulation under
supervision of a clinical researcher, when patients are on antiepileptic medication, and
keeping stimulation to within safety limits. Low frequency (1 Hz) single pulse
stimulation will be applied to regions of DN-A and DN-B to map cortico-cortical evoked
potentials (CCEPs). This will be used to estimate the strength, as well as provide data
on the directionality of connections between regions. In a departure from original plans,
based on recent findings (Hermiller et al. 2019), theta-burst stimulation (gamma-band
stimulation applied intermittently at theta frequencies) will be applied to regions of
DN-A regions in lateral temporal, posteromedial and prefrontal cortices during a
recollection task to test if stimulation of distant DN-A regions can lead to improvements
in hippocampus-mediated episodic memory recollection. Currents will be administered at a
threshold below that which causes after-discharges (usually around 6-8 mA).
