Subcortical-cortical dynamical states of the human brain and their breakdown in stroke

The mechanisms controlling dynamical patterns in spontaneous brain activity are poorly understood. Here, we provide evidence that cortical dynamics in the ultra-slow frequency range (<0.01–0.1 Hz) requires intact cortical-subcortical communication. Using functional magnetic resonance imaging (fMRI) at rest, we identify Dynamic Functional States (DFSs), transient but recurrent clusters of cortical and subcortical regions synchronizing at ultra-slow frequencies. We observe that shifts in cortical clusters are temporally coincident with shifts in subcortical clusters, with cortical regions flexibly synchronizing with either limbic regions (hippocampus/amygdala), or subcortical nuclei (thalamus/basal ganglia). Focal lesions induced by stroke, especially those damaging white matter connections between basal ganglia/thalamus and cortex, provoke anomalies in the fraction times, dwell times, and transitions between DFSs, causing a bias toward abnormal network integration. Dynamical anomalies observed 2 weeks after stroke recover in time and contribute to explaining neurological impairment and long-term outcome.


March 2021
For manuscripts utilizing custom algorithms or software that are central to the research but not yet described in published literature, software must be made available to editors and reviewers. We strongly encourage code deposition in a community repository (e.g. GitHub). See the Nature Portfolio guidelines for submitting code & software for further information.

Data
Policy information about availability of data All manuscripts must include a data availability statement. This statement should provide the following information, where applicable: -Accession codes, unique identifiers, or web links for publicly available datasets -A description of any restrictions on data availability -For clinical datasets or third party data, please ensure that the statement adheres to our policy Field-specific reporting Please select the one below that is the best fit for your research. If you are not sure, read the appropriate sections before making your selection. Source data to reproduce the main figures are provided with this paper. Raw neuroimaging and neuropsychological data are publicly available at https:// cnda.wustl.edu/data/projects/CCIR_00299 and require controlled access as they contain sensitive patients' data. The person requesting the data must sign a confidentiality agreement provided by Washington University stipulating that they will make no attempt at identifying the patients and that they will use data for research purposes only. Correspondence and requests should be addressed to M.C. (maurizio.corbetta@unipd.it).
The sample size was determined based on previous works relating to the same dataset (e.g., Corbetta et al., Neuron, 2015;Siegel et al., PNAS, 2016;Griffis et al., Cell Reports, 2019). Furthermore, we considered Leonardi et al. (Neuroimage, 2015) to ensure that the data were suitable to the dynamical analysis we performed.
As a result of the pre-processing, 114 subjects were available at 2 weeks (sub-acute), 79 at 3 months, and 64 at 12 months, 24 and 22 controls at the first and second acquisition, respectively. For the implementation of the dynamical functional analysis, only subjects with a sufficient number of good frames (300) were considered.

Patients:
We selected only patients, who participated to all the three recordings (2 weeks, 3 months, 12 months after stroke). Threfore, 47 patients were considered. Controls: The number of control subjects with sufficient frames were 20 during the first visit and 20 during the second visit. To avoid one group dominating the other in the following analysis steps, we equalized the number of controls and patients. Thus, we used all the controls' data as they were from different subjects.
Our results are based on a single experiment and have not been replicated in an independent cohort.
We controlled that our results did not depend on the choice of the parameters that we apply during the analysis. In particular the following test were performed: -we verified that our results were not dependent on the fact that we considered all the controls' data as they were from different subjects. Specifically, all the analyses have been re-run considering the averaged measures over sections for the subset of control subjects, who participated to both sections.
-we tested the quality of the dimensionality reduction process: (1) test of the similarity of the reduced FC with the unreduced FC, and check that within-network connectivity was significantly stronger than between-networks connectivity.
(2) verification that all previous results on stroke impairment in static FC were reproduced with the reduced data.
-we analyzed the impact of sliding windows width selection during the definition of the Dynamical Functional States (DFSs). From the results, we can confirm that our choice for the sliding window width did not impact our main results.
Patients and control subjects were allocated in two different groups, based on clinical evaluation.
The staff that was involved in segmenting or in reviewing the lesions was blind to the individual behavioral data. Note that full information on the approval of the study protocol must also be provided in the manuscript.

Clinical data
Policy information about clinical studies (4) Presence of other neurological, psychiatric or medical conditions that preclude active participation in research and/or may alter the interpretation of the behavioral/imaging studies (e.g., dementia, schizophrenia), or limit life expectancy to less than 1 year (e.g., cancer or congestive heart failure class IV). (5) Report of claustrophobia or metal object in body. Noise and artifact removal Volume censoring connectivity through Pearson's correlation and dynamical connectivity measures as described below in the 'Models and Analysis' section. From behavioral scores, principal components analyses (PCA) were used to decompose the behavioral data from each domain. Detailed descriptions of the behavioral testing and PCA analyses can be found in the Supplemental Material for Corbetta et al. (Neuron, 2015) and Siegel et al. (PNAS, 2016).

Resting-state
Six to eight resting state (RS) fMRI runs, each including 128 volumes (30 min total), for each subjects, and for each section. Stroke patients' data were collected at three time points: 1-2 weeks after stroke, 3 months after stroke, 1 year after stroke. Control subjects' data were recorded twice, 3 months apart.  Functional MRI data pre-processing consisted of slice-timing correction using sinc interpolation, correction of inter-slice intensity differences resulting from interleaved acquisition, normalization of whole-brain intensity values to a mode of 1000, correction for distortion via synthetic field map estimation, and within-and between-scan spatial re-alignment. BOLD data were re-aligned, coregistered to the corresponding structural images, normalized to atlas space, and resampled to 3mm cubic voxel resolution using a combination of linear transformations and non-linear warps.

MNI152
Processing steps were applied to account for non-neural sources of signal variance. Confounds related to head motion, global signal fluctuations, and non-gray matter signal compartments were removed from the data by regression of the six head motion parameters obtained from rigid body correction, along with the global GM signal and the CSF and white matter signals extracted from FreeSurfer tissue segmentations (Dale et al., Neuroimage, 1999). BOLD data were band-pass filtered (0.009 < f < 0.08 Hz) to retain low-frequency fluctuations.
A frame was censored if it exceeded a 0.5 mm framewise displacement threshold, and the succeeding frame was also censored to further reduce confounds related to motion (Power et al., Neuroimage, 2014).