The study was approved by Ethics Committee of Jiangyin People's Hospital Affiliated with Southeast University and registered in the Clinical Trials.gov (NCT03946839). All participants provided written informed consent.
Participants
A total of 24 ICU sepsis survivors were initially recruited for this study and underwent neurocognitive testing and MRI scanning. Patients were recruited from the ICU of Jiangyin People's Hospital affiliated with Southeast University. 24 age-, gender-, and education-matched healthy volunteers were selected as the healthy controls through community postings and media advertising. All participants were right-handed Chinese Han individuals. After excluding participants with incomplete MRI scans, a history of epilepsy, abnormal MRI report or excessive head motion, 16 sepsis survivors and 19 healthy controls were included in the final analyses.
Inclusion and exclusion criteria
Inclusion criteria for sepsis survivors included the following: (1) sepsis or septic shock patients (diagnosed with Sepsis-3.0 [1] ) aged 18 to 79 years; (2) at least 6 years of education (able to speak, read, and write); (3) hospitalized in the ICU for more than 2 days, with survival and subsequent discharge. Exclusion criteria for the sepsis survivor group were: (1) death during hospitalization; (2) leaving against medical advice or transfer to another hospital; (3) declining to participate or withdrawing midway; (4) history of neuropsychiatric disease (e.g., cerebrovascular disease, Parkinson's disease (PD), Alzheimer's disease (AD), demyelinating disease, epilepsy, depression, schizophrenia, etc.); (5) severe brain injury; (6) sever systemic disease (e.g., hepatic encephalopathy, ketoacidosis, hyperosmolar hyperglycemic state, chronic renal failure, etc.); (7) history of drug abuse or insobriety; (8) use of psychotropic medications such as sleeping pills, selective serotonin reuptake inhibitors, etc.; (9) MRI incompatibility. Healthy controls were required to have a Mini-Mental State Examination (MMSE) score ≥ 24 [13]. Exclusion criteria for the health controls were a history of neuropsychiatric disease, head injury, alcohol and drug addiction or ferrous/electronic implants.
Neurocognitive measurements
All subjects underwent a comprehensive battery of neurological examinations 2 hours before MRI scanning, including the Montreal Cognitive Assessment (MoCA), MMSE, Complex Figure Test (CFT), Auditory Verbal Learning Test (AVLT), Digit Span Test (DST), Verbal F1uency Test (VFT), Clock Drawing Test (CDT), Symbol Digit Modalities Test (SDMT), and Trail Making Test (TMT). Each subject had 70-90 minutes to complete the neuropsychological examinations.
MRI data acquisition
MRI was performed at the Medical Imaging Center of Jiangyin People’s Hospital using a Discovery MR750w 3.0T scanner (GE, Boston, United States), equipped with a 24-channel head coil. Head motion was controlled using foam pads during the scans. All participants, wearing earplugs, were instructed to remain awake, motionless with their eyes closed, and not to think of anything in particular. Resting state BOLD images were acquired overa period of 7 minutes and 40 seconds by a gradient-recalled echo-planar imaging (GRE-EPI) sequence with repetition time (TR) = 2000 ms, echo time (TE) = 30 ms, flip angle (FA) = 90°, field of view(FOV) = 220×220 mm, matrix size = 64×64, slice thickness = 4mm, gap = 0 mm, number of slices = 35. The slices were acquired in an interleaved order (1, 3, 5 …, 35, 2, 4, 6 …, 34). T1-weighted 3D fast spoiled gradient recalled echo (FSPGR) images were collected over 4 minutes 25 seconds with TR = 7.2 ms, TE = 3.1 ms, flip angle = 8°, FOV = 256×256 mm, matrix size = 256×256, slice thickness = 1mm, gap = -0.5 mm, number of slices = 312, voxel size = 1×1×1mm3. Additionally, routine axial T2-weighted images were obtained to exclude subjects with major cerebral infarction, white matter (WM) changes, or other brain lesions.
fMRI image data preprocessing
SPM12 (http://www.fil.ion.ucl.ac.uk/spm) and DPABI (http://rfmri.org/ dpabi) were used for preprocessing rs-fMRI data in the MATLAB environment. The first 10 images were removed to allow steady state, leaving 220 functional volumes for each subject. After slice timing correction, images were realigned to the first volume to correct for head motion. After excluding 1 healthy participant and 4 sepsis patients with head motion > 2.0 mm maximum displacement in any direction (x, y, z) or 2.0° of angular motion, no significant differences in head motion were found between the two groups (p > 0.05). Then, 3D T1-weighted images were registered to the functional images and subdivided into WM, gray matter, and cerebrospinal fluid (CSF) using the new segment and DARTEL technique, followed by spatial normalizing into the Montreal Neurological Institute EPI template (voxel size 3 × 3 × 3 mm3). Other preprocessing steps included spatial smoothing with an isotropic Gaussian kernel of 6 × 6 × 6 mm, temporal bandpass filtering at 0.01-0.08 Hz, and nuisance signal regression (including WM signal, CSF signal, and Friston-24 head motion parameters).
Network construction and topological analysis
Preprocessed rs-fMRI data were used to construct the whole brain functional connectivity network for each subject. We used automated anatomical labeling (AAL) atlas [14] to identify 116 functional regions of interest (ROIs) throughout the brain, including the cerebrum and cerebellum. Each brain region was considered as a network node. The time series of all voxels in each ROI were extracted and subsequently averaged to obtain a representative time series. Pearson’s correlation coefficients between the mean time series of all possible pairs of the 116 regions were computed, which were considered as the edges of the network. To improve the distribution of data for group analysis, Pearson correlation coefficients (r) were standardized by Fisher’s z transformation, resulting in a 116×116 correlation weight matrix for each subject.
Topological properties of networks were analyzed using GRETNA software (http://www. nitrc. org/projects/gretna/) based on graph theory. Only positive correlations were involved in the subsequent network metrics analysis to minimize potential confounding effects of global signal regression. A network sparsity (S) was applied to produce binary undirected functional networks, and a wide range of sparsity threshold was identified, ranging from 0.05 to 0.4 with an interval of 0.01. The global and local metrics of brain functional network were estimated for each brain region at each selected sparsity threshold. Global metrics included small-world parameters (clustering coefficient (Cp), characteristic path length (Lp), normalized clustering coefficient (γ), normalized characteristic path length (λ), and small-worldness (σ)) and network efficiency (global efficiency (Eg) and local efficiency (Eloc)). Specifically, σ = γ/λ,and σ>1 was used as an indication of a small-world organization of the network. Two nodal centrality properties were employed for regional nodal network analysis: DC and NE. For further statistical comparison, area under the curve (AUC) for each network metric was calculated, providing a summarized scalar for topological parameters, avoiding the error caused by a single threshold.
GCA
GCA is a statistical concept of causality that based on the multiple variant auto regression (MVAR) model [12], commonly used as a reliable method in neuroscience to estimate EC, which characterizes directional functional connections among brain regions [15]. In this study, GCA was conducted based on those brain regions showing significant DC or NE changes between sepsis survivors and healthy controls. Those brain regions were selected and defined as ROIs for seeds using WFU-pick-atlas (https://www.nitrc.org/ projects/wfu_pickatlas), then resampled to 3 mm×3 mm×3 mm. Based on pre-processed data, the REST_v1.8 software package (http// www.restfmri.net) was applied to compute the causal effects of the time series x of selected seed points and the time series y of each voxel over the whole brain. In the Granger causality model, a value of 0 indicates no causal connection from x to y, a value of 1 indicates strong positive causality, and a value of -1 indicates strong negative causality. The causality analysis was performed twice on each ROI: from the seed point to whole-brain voxels (x to y) and from whole-brain voxels to the seed point (y to x). The obtained effective EC graph was transformed by Fisher’s z to improve distribution normality, resulting in a Z-map of GCA.
Statistical analysis
1. Demographic and neurocognitive data: statistical analyses were performed using SPSS software (version 27.0.1, IBM SPSS Inc., Chicago, IL, USA). Categorical variables were compared between groups using the Chi-squared test or Fisher’s exact test, as appropriate. Continuous variables, including demographic data and cognitive scores, were analyzed using independent t-tests. A p-value less than 0.05 was considered statistically significant.
2. rs-fMRI data: The topological properties of brain functional networks were analyzed using GRETNA software. Between-group comparisons of the AUC for these topological properties were conducted using one-way ANOVA in R software (version 4.3.2; https://www.r-project.org/), with age, gender, and years of education included as covariates. Multiple comparisons were corrected using the false discovery rate (FDR) method.
3. GCA: Granger causal influence measures (Z-EC values) derived from healthy control subjects and sepsis survivors were compared using two-sample t-tests, with age, gender, and years of education as covariates. Statistical significance was determined using Alphasim correction with a voxel-level threshold of p < 0.01 and a cluster-level threshold of p < 0.01 (two-tailed).
4. Correlation analysis: Pearson correlation analyses were performed to assess the relationships between cognitive test scores and topological properties, controlling for age, gender, and years of education. These analyses were conducted using SPSS software, and a p-value less than 0.05 was considered statistically significant.