key: cord-0825416-q14x1wv2
authors: Fu, Zening; Tu, Yiheng; Calhoun, Vince D.; Zhang, Yuqi; Zhao, Qing; Chen, Jun; Meng, Qingtao; Lu, Zhijie; Hu, Li
title: Dynamic functional connectivity associated with post-traumatic stress symptoms in COVID-19 survivors
date: 2021-08-05
journal: Neurobiol Stress
DOI: 10.1016/j.ynstr.2021.100377
sha: 79186bb503eebd549dca34d7a5104996642d5436
doc_id: 825416
cord_uid: q14x1wv2

Accumulating evidence shows that Coronavirus Disease 19 (COVID-19) survivors may encounter prolonged mental issues, especially post-traumatic stress symptoms (PTSS). Despite manifesting a plethora of behavioral or mental issues in COVID-19 survivors, previous studies illustrated that static brain functional networks of these survivors remain intact. The insignificant results could be due to the conventional statistic network analysis was unable to reveal information that can vary considerably in different temporal scales. In contrast, time-varying characteristics of the dynamic functional networks may help reveal important brain abnormalities in COVID-19 survivors. To test this hypothesis, we assessed PTSS and collected functional magnetic resonance imaging (fMRI) with COVID-19 survivors discharged from hospitals and matched controls. Results showed that COVID-19 survivors self-reported a significantly higher PTSS than controls. Tapping into the moment-to-moment variations of the fMRI data, we captured the dynamic functional network connectivity (dFNC) states, and three discriminative reoccurring brain dFNC states were identified. First of all, COVID-19 survivors showed an increased occurrence of a dFNC state with heterogeneous patterns between sensorimotor and visual networks. More importantly, the occurrence rate of this state was significantly correlated with the severity of PTSS. Finally, COVID-19 survivors demonstrated decreased topological organizations in this dFNC state than controls, including the node strength, degree, and local efficiency of the supplementary motor area. To conclude, our findings revealed the altered temporal characteristics of functional networks and their associations with PTSS due to COVID- 19. The current results highlight the importance of evaluating dynamic functional network changes with COVID-19 survivors.

Coronavirus Disease 2019 (COVID-19) is a global pandemic caused by the novel severe acute 2 respiratory syndrome coronavirus-2 (SARS-CoV-2), which has been spreading worldwide with 3 more than 106 million confirmed cases and 2 million death as the end of March 2021 4 (Worldometer, 2021) . Although primarily considered as respiratory disease, COVID-19 has 5 become increasingly recognized as neurotropic; patients might experience some mild (e.g., 6

headaches, loss of smell, and tingling sensations), as well as severe neurological symptoms 7 (e.g., aphasia, strokes, and seizures) ( In addition to the aforementioned neurological symptoms, new research indicated COVID- 19 13 survivors might demonstrate long-term mental problems, including fatigue, sleep difficulties, 14 anxiety, depression, and more importantly, post-traumatic stress disorder (PTSD) (Huang et al., 15 2021 ). Considering the neuropsychiatric symptoms in COVID-19 survivors, it is plausible to 16 hypothesize that they may experience the disruption of widespread brain functional networks, 17 which can be investigated by examining functional connectivity between different brain 18 regions using resting-state functional MRI (fMRI) ( 7 The MRI data of all participants were collected using a GE 3.0 Tesla MR750 scanner with a 8 standard 32-channel head coil at the radiology department of the Renmin Hospital of Wuhan 9

University, between July and August 2020. Subjects were asked to stay awake and to keep their 10 heads still during the scan, with their eyes open and ears plugged. High resolution brain 11 structural images were acquired with a T1-weighted fast-spoiled gradient echo sequence 12 (repetition time = 8.16 ms, echo time = 3.18 ms, flip angle = 12°, slice thickness = 1 mm, 13 interslice gap = 1 mm, and field-of-view = 256). Eight-minute resting-state brain functional 14 images were acquired with a T2-weighted gradient echo-planar imaging (repetition time = 15 2,000 ms, echo time = 30 ms, flip angle = 90°, slice thickness = 3.5 mm, field-of-view = 256 16 mm, and 38 slices). 17 18 19 fMRI data were preprocessed using Statistical Parametric Mapping (SPM12) toolbox under the 20 MATLAB 2019 environment. The first five dummy scans were discarded before preprocessing 21 to guarantee the remaining fMRI scans were collected when magnetization achieved a steady 22 state. We performed a slice timing correction and then a functional realignment on the 23 functional images. The fMRI data were subsequently warped into the standard Montreal 24

Neurological Institute space using an echo-planar imaging template and were slightly 25 resampled to 3 × 3 × 3 mm 3 isotropic voxels. The spatial smoothing was finally performed to 26 smooth the data with a 6-mm full-width at half-maximum Gaussian kernel. There are three major procedures in this framework: 1) apply Neuromark pipeline to extract 10 corresponding functional regions and time-courses (TCs) for each individual; 2) calculate 11 dFNC between the ICA TCs via a sliding-window approach and perform a k-mean clustering 12 on dFNC estimates to identify reoccurring connectivity patterns (i.e., reoccurring states) across 13 subjects and time; 3) calculate the fractional rate to measure the frequency of occurrences of 14 different connectivity patterns, and calculate state-based dFNC and graph-theory measures to 15 explore transient information transmission in large-scale functional networks. 16

Besides dFNC, we also calculated static FNC (sFNC) between TCs using Pearson's correlation 17 to build static functional connectivity. 18 19 

NeuroMark is a reliable ICA-based pipeline that automatically estimates functional regions 21 adaptable to each individual subject and comparable across subjects by taking advantage of the 22 reliable brain network templates extracted from 1828 healthy controls as guidance. Two large 23 healthy datasets, i.e., the human connectome project (HCP, 24 http://www.humanconnectomeproject.org/) and the genomics superstruct project (GSP, 25 https://dataverse.harvard.edu/dataverse/GSP), were used for the construction of the templates. 26

Group ICA was performed on the GSP and HCP datasets, respectively, and the identified 27 independent components (ICs) from the two datasets were then matched by comparing their matrices, we applied a k-means clustering algorithm to classify dFNC into different groups 10 based on their spatial similarity. The cluster centroids were referred to as reoccurring "brain 11 states", in a conceptual analogy to electroencephalogram (EEG) microstates (Khanna et al. , 12 2015) . The optimal number of states was estimated by the elbow criterion, defining as the ratio 13 of within-cluster distance to between-cluster distance (Allen et al., 2014). 14 15 2.7. State occurrences and state-based dFNC/graph-theory 16 measures 17 To assess the occurrence of different dFNC states, we calculated the fractional rate of each 18 brain state by dividing the number of the total windows by the number of windows assigned to 19 each state. To investigate the dFNC pairs in each state, we calculated the state-based dFNC by 20 averaging the dFNC estimates across time windows that were assigned to the same state. 21

Graph-theory analysis was further applied to demonstrate the abnormal topological 22 organizations in functional brain networks. Node strength (i.e., the sum of weights of links 23 connected to a node), node degree (i.e., the number of links connected to a node), and local 24 efficiency (i.e., the number of edges within neighbors of a node) of the component TCs were 25 estimated using the dFNC matrices via the brain connectivity toolbox 26 (https://sites.google.com/site/bctnet/) and then averaged across windows within each state. We 27 threshold the dFNC matrices by setting the dFNC pairs to 0 if their absolute connectivity < 0. Independent-samples t-tests were applied to compare the total score of PCL-5, sub-domains of 2 PCL-5, GAD, and PHQ, between COVID-19 survivors and controls. The threshold for 3 statistical significance was corrected for multiple comparisons using the false-discovery rate 4

[FDR] procedure (i.e., PFDR<0.05). 5

The differences of sFNC between COVID-19 survivors and controls were analyzed using a 6 general linear model (GLM), controlling for age and gender. The statistically significant 7 threshold was corrected for multiple comparisons (i.e., across all pairs of sFNC, N=1378) using 8 the FDR procedure. 9

The statistical comparisons of state occurrences and state-based dFNC/graph-theory measures 10 between two groups were analyzed using GLMs, controlling for age and gender, and the 11 statistically significant threshold was FDR corrected. In addition, we calculated the partial 12 correlations between the dynamic characteristics (i.e., the state with abnormal occurrence in 13 COVID-19 survivors) and the scores of PCL-5 (i.e., the total score and four sub-domain scores), 14 controlling for age, gender, and group label. Table 1 summarized scores of self-report assessments in COVID-19 survivors and controls. 29

Using the independent-sample t-test, we found that COVID-19 survivors had significantly 30 J o u r n a l P r e -p r o o f higher PCL-5 total scores than controls (COVID-19 survivors: 15.36 ± 12.05; controls: 7.19 ± 1 5.55; t84=4.01, P<0.001). Similarly, COVID-19 survivors also showed significantly higher 2 scores in all PCL-5 sub-domains, GAD, and PHQ (Table 1) 6 We parcellated the brain into 53 spatially independent components as ROIs, which covered 7 almost the whole brain using the Neuromark pipeline. Based on their anatomical and functional 8 prior knowledge, the 53 ROIs were arranged into seven functional networks ( 

18 survivors and controls 19 Reoccurring dynamic brain states were identified by clustering the windowed dFNC estimates 20 based on their spatial similarity (Allen et al., 2014; Fu et al., 2018) . Here, the term "brain states" 21 refers to the dFNC patterns that reoccur across windows and subjects. We performed a cluster 22 number validity analysis using the elbow criteria to determine the optimal number of clusters 23 as 3, which was within the reasonable range of the number of clusters used in previous studies 24 (Allen et al., 2014; Tu et al., 2020, 2019). Three highly structured dFNC brain states are 25 displayed in Fig. 2 . The circle panel displays the functional profile of each brain state that only 26 retained strong connectivity (absolute connectivity strength > 0.2). The fractional rate was not 27 uniformly distributed across brain states. State 1 and 2 occurred less frequently, showing 28 strongly interconnected brain networks. The dFNC in state 1 showed less negative connectivity 29 between SCN and SMN/VSN but more heterogeneous between SMN and VSN (i.e., negative 30 correlations between these two networks) than dFNC in state 2. Interestingly, positive 31 J o u r n a l P r e -p r o o f connectivity between CBN and VSN, as well as negative connectivity between CBN and SMN, 1 was only observed in state 1. In state 2, strong negative connectivity between SCN and 2 SMN/VSN, as well as strong within-network connectivity in VSN, were observed. In contrast, 3

State 3 was a sparsely connected brain state with weak inter-network connectivity, but occurred 4 most frequently (i.e., accounts for >70% of all windows). Overall, the identified sparsely 5 connected brain state (i.e., state 3) was more frequent, while the more strongly connected states 6 (i.e., states 1 and 2) were less frequent, which were in line with previous findings in dynamic states did not differ between groups, we calculated the Euclidean distance from each time 19

window to the cluster centroids in COVID-19 survivors and controls, respectively. Results 20 showed no significant difference in distance from dFNC estimates to the cluster centroids 21 between groups (P > 0.10). 22

Despite the intact global spatial patterns of brain states, we found an altered temporal 23 characteristic of the dFNC state in COVID-19 survivors. Among the three identified dFNC 24 states, we observed that COVID-19 survivors showed a significantly increased occurrence 25 (measured by fractional rate) in dFNC state 1 (PFDR < 0.05). We then excluded subjects without 26 state 1 (i.e., dFNC patterns in these subjects were not clustered into state 1) and repeated the 27 statistical analysis. The occurrence of state 1 was still significantly different between the two 28 groups (PFDR < 0.05). We further investigated the relationships between this abnormal temporal 29 characteristic of dFNC state and PTSS in COVID-19 survivors. We found that with and without 30 excluding subjects without state 1, the fractional rate of state 1 was significantly correlated 31 J o u r n a l P r e -p r o o f with the PCL total score, PCL cognition score, and PCL arousal score, respectively ( Fig. 3C;  1 PFDR < 0.05). 2 3 3.5. State-based dFNC and graph-theory measures 4 Although the global dFNC patterns are intact between COVID-19 survivors and controls, we 5 hypothesized that local dFNC and topological organizations might reflect brain changes 6 induced by PTSS in COVID-19 survivors. We found that the dFNC between thalamus and 7 inferior parietal lobule (IPL) in state 3 shrank in COVID-19 survivors (closer to 0) and 8 interestingly, this dFNC was positively correlated with the PCL-5 intrusion score (Fig. 4) . 9

Other state-based dFNC did not show any significant differences between groups (P > 0.05). 10

The graph-theory analysis showed that in state 1, the node degree and strength of the 11 supplementary motor area (SMA) were significantly lower in COVID-19 survivors than 12 controls (PFDR < 0.05; the local efficiency of the SMA was also lower in COVID-19 survivors, 13 Puncorrected = 0.0014), suggesting that the function of the SMA in state 1 was disrupted in 14 COVID-19 survivors (Fig. 5) . The graph-theory measures did not show any significant 15 differences in other brain regions (P > 0.05). 16 17 

In the present study, we assessed PTSS and collected fMRI with COVID-19 survivors around 19 six months after they were discharged from hospitals. Compared to non-COVID-19 controls, 20 COVID-19 survivors had significantly higher PCL-5 total scores as well as scores of all PCL-21 5 sub-domains, GAD, and PHQ. Using a novel fMRI analytical pipeline, we found that 22

in a reoccurring brain state, which was associated with the severity of their self-report PTSS. Due to the small number of males in both groups (N=11 and N=12 in control and COVID-19 31 survivor groups, respectively), sex was not considered as a factor in the present statistical 1 analyses. Nevertheless, we observed that female survivors reported significantly higher PCL-2 5 scores than male survivors (t=4.37, P<0.001), suggesting females could be more vulnerable 3

to PTSS due to Although recent studies have demonstrated brain structural abnormalities in COVID- 19 5 survivors (Lu et al., 2020) , few previous studies tapped on the brain functional abnormalities, surface (Douaud et al., 2021) . Taken together these previous findings and our results, we 2 speculate that COVID-19 survivors with depressive symptoms might show a reduction of brain 3 volume within multiple brain regions linking with SMA, which will limit the local activities in 4 these regions. The restricted local activities will further affect the information processing and 5 integration, reflecting by transiently decreased regional efficiency. 6

State 2 was a regionally densely connected dFNC state characterized by highly positive within-7 network connectivity (i.e., subcortical, auditory, sensorimotor, and visual networks) and 8 negative between-network connectivity (i.e., between subcortical-sensorimotor networks). 9

Although this brain state has been identified abnormal in many neuropsychiatric disorders, There are several limitations in this study. First, we only recruited participants from Wuhan 25 city, the center of the outbreak of COVID-19 at the beginning of 2020. Cohort studies from 26 other regions and countries are necessary for validation. Second, we only recorded brain 27 imaging data in one session. Longitudinal observations will be insightful to observe the 28 trajectories of PTSS and brain dysfunctions in COVID-19 survivors. Another limitation of this 29 study is that we only investigated the associations between dynamic features and 30 PTSS/GAD/PHQ. The investigation of correlations between dynamic features and other 31 behavioral scores will provide more comprehensive information on COVID-19. For example, 32 1 (Colebatch et al., 1991) , and higher motor processing functions (Rao et al., 1993) . In future 2 work, we would like to collect motor performance data for COVID-19 survivors to examine 3 whether COVID-19 influences motor performance and whether it is associated with brain 4 abnormalities in SMA. Finally, clinical records of COVID-19 survivors were not retrievable. 5 Thus, we were not able to examine the relationship between clinical characteristics (e.g., the 6 severity of the disease) and their PTSS/brain connectivity after 6 months. 7

In conclusion, we recorded PTSS and brain MRI in patients who survived from COVID- 19 8 infection. We found that COVID-19 survivors had more severe PTSS than controls, and the 9 PTSS were not associated with the sFNC, but with the disrupted dFNC state. Given the 10 dynamic nature of brain activity and connectivity, we would like to highlight the great potential State-based dFNC between thalamus and IPL was associated with post-traumatic stress 8 disorder checklist for DSM-5 (PCL-5) intrusion score. Asterisks * indicate significant group 9 difference after false-discovery rate (FDR) correction. represents an individual's value, and the color represents the group label. The green box 16 represents 95% of the standard deviation of the mean, and the yellow box represents the 17 standard deviation. Asterisks * indicate significant group difference after false-discovery rate 18 (FDR) correction. Only subjects who had the state 1 were used for the statistical analyses. 19 

Evolution and resolution of brain involvement associated with

SARS-CoV2 infection: A close Clinical -Paraclinical follow up study of a case

Dynamic Functional Network Connectivity States

Tracking whole-brain connectivity dynamics in the resting state

Multifocal laminar cortical brain lesions: a consistent MRI finding 13 in neuro-COVID-19 patients

Regional cerebral blood flow during voluntary arm and hand movements in human 13 subjects

Dynamic functional connectivity analysis reveals transient states 17 of dysconnectivity in schizophrenia

Coronavirus Disease (COVID-19)

Andersson, 22 relationship with dynamic functional connectivity: An application to schizophrenia

Transient increased thalamic-sensory connectivity and decreased whole-brain 4 dynamism in autism

Functional connectivity in the 7 resting brain: A network analysis of the default mode hypothesis

COVID-19 in patients discharged from hospital: a cohort study

Dynamic functional connectivity: Promise, issues, and interpretations

Posttraumatic 21

Stress Disorder in Patients After Severe COVID-19 Infection. JAMA Psychiatry. 22 study of time-varying functional connectivity in resting fMRI. Netw. Neurosci

COVID-19 survivors: Role of inflammatory and clinical predictors

Functional role of the supplementary and pre-7 supplementary motor areas

Disruptions of functional connectivity in the default mode network of comatose patients

COVID-19-13 associated acute disseminated encephalomyelitis (ADEM)

The emerging 24 spectrum of COVID-19 neurology: Clinical, radiological and laboratory findings

Dissociated large-scale functional connectivity 28 networks of the precuneus in medication-naïve first-episode depression

Lifting the mask on neurological manifestations of COVID-1 19

Mental Health and the Covid-19 Pandemic

The dynamic functional connectome

State-of-the-art and perspectives

Functional magnetic resonance imaging of complex human 10 movements

Functional 12 connectivity of supplementary motor area during finger-tapping in major depression

A brief measure for assessing 15 generalized anxiety disorder: The GAD-7

Distinct thalamocortical network dynamics are associated with the 20 pathophysiology of chronic low back pain

Thalamocortical dysrhythmia detected by 1 machine learning

Durability of Responses after SARS-CoV-2 mRNA-1273 Vaccination

Coronavirus Update (Live): Cases and Deaths from COVID-19 Virus 10

Impact of COVID-19 pandemic on mental 13 health in the general population: A systematic review

The authors declare no competing interests.