Introduction

Anti-N-methyl-d-aspartate (anti-NMDA) receptor encephalitis is an immune-mediated and treatable brain inflammation characterized by the acute onset of a constellation of symptoms.1, 2 The typical clinical manifestations include loss of consciousness, abnormal behaviour and cognition, speech disorder, memory deficit, abnormal movements, seizures, autonomic dysfunction and central hypoventilation.3 Approximately 80% of patients showed substantial recovery after timely treatments according to the low modified Rankin Scale (mRS).3 However, many studies have indicated that cognitive impairments, especially memory deficits and executive function impairments, constitute a major long-term consequence of this disorder, which are also consistent with high density of NMDA receptors within the hippocampus and frontal cortex.4-6 A former study by our team showed that early administration of intravenous second-line immunotherapy may be associated with more favourable verbal episodic memory outcomes; nonetheless, the patients still exhibited other domains of cognitive impairments of varying degrees, including working memory, information processing speed and executive function.6

Exploring the neurophysiological mechanisms underlying cognitive deficits, especially memory impairments in patients with anti-NMDA receptor encephalitis, by using multimodal functional MRI has attracted intensive attention in recent years.7-9 However, to our knowledge, fMRI studies about other domains of cognitive impairments, such as working memory, information processing speed and executive function are very few in literature, which has limited our understanding of this disease. The impairment of these cognitive functions has made great impact on the patients' study, work and life. It is necessary to pay attention to investigate these cognitive impairments and its mechanism for patients with anti-NMDA receptor encephalitis.

Working memory, information processing speed and executive function tasks in healthy individuals have been shown to engage groups of brain regions, such as the dorsolateral/ventrolateral prefrontal cortex and the lateral parietal cortex, which are integrated in the executive control network (ECN).10-12 A resting fMRI study based on hard parcellation of large-scale brain network analyses indicated that functional connectivity (FC) changes in frontoparietal networks (i.e. ECN) are impaired and associated with schizophrenia-like symptoms in anti-NMDA receptor encephalitis.13 However, this study did not focus on whether the ECN changes were related to cognitive function impairments in this population.

In addition, previous studies did not consider the significant dynamic characteristics in the brain network in anti-NMDA receptor encephalitis, as FC was supposed to be constant during resting-state functional MRI scanning. It is noteworthy that resting state FC can vary considerably at different temporal scales.14-16 Temporal dynamic characteristics can be detected in brain FC by analysing functional MRI signals and such analyses may demonstrate neural mechanisms that cannot be revealed through static resting state FC alone.15-18 The clinical relevance and potential biomarker applications of dynamic FC have been suggested in clinical studies of epilepsy,19, 20 Alzheimer's disease,21 Parkinson's disease,22 schizophrenia23 and autism.24 However, to the best of our knowledge, altered FC in the ECN in a dynamic context remains unknown in patients with anti-NMDA receptor encephalitis past the acute stage.

In the present study, we used resting-state fMRI and employed a group-level independent component analysis (ICA) to generate ECN. Then, we compared the static and dynamic FC of the ECN in patients with anti-NMDA receptor encephalitis and healthy controls (HCs). We speculated that abnormal static and dynamic metrics of ECN may exist and may be associated with cognitive performance in patients with anti-NMDA receptor encephalitis. Therefore, we aimed to determine whether static FC of the ECN and the temporal properties of dynamic FC states in the ECN could characterize the underlying nature of anti-NMDA receptor encephalitis and correlate with cognitive functions.

Methods Subjects

Twenty-one patients with anti-NMDA receptor encephalitis who were hospitalized or referred to the outpatient clinic for further counselling and treatment at the Department of Neurology, First Affiliated Hospital, College of Medicine, Zhejiang University were recruited between July 2016 and December 2019. Demographic and clinical data of the patients are shown in Table 1. The diagnosis was based on typical clinical features together with the presence of IgG antibodies for NMDA receptors.2, 25 There were no abnormalities in structural MRI for any patient. The assessment was conducted after the acute stage of the disease (at least 6 months after initial discharge from the hospital).

Table 1. Demographic and clinical data of patients. Patient Sex Age Igg NMDA receptor antibodies (CSF titre) Symptoms Hospital stay, (months) mRS Score Time between initial discharge and data acquisition (months) Initial Total Before treatment At study time point 1 F 24 1:32 Behaviour LOC, dyskinesia, seizure, behaviour, cognition 1.2 5 0 18.5 2 F 22 1:32 Behaviour, seizure LOC, dyskinesia, seizure, behaviour, cognition 1.1 5 0 6.5 3 M 28 1:32 Behaviour, dyskinesia LOC, dyskinesia, seizure, behaviour, cognition, autonomic instability 1.2 5 1 9.3 4 M 31 1:32 Seizure, LOC LOC, dyskinesia, seizure, behaviour, cognition, autonomic instability 1.4 5 0 21 5 F 15 1:32 Dyskinesia, seizure Dyskinesia, seizure, behaviour, cognition 1.2 4 1 6.1 6 F 25 1:32 Behaviour, seizure LOC, dyskinesia, seizures, behaviour, cognition 1.2 5 0 8.3 8 F 16 1:10 Behaviour Seizure, behaviour, cognition 1.2 4 0 10.6 9 F 29 1:3.2 Behaviour LOC, behaviour, cognition 0.7 4 0 6.5 10 48 1:32 Behaviour, seizure Seizure, behaviour, cognition 1.1 4 1 8.5 11 F 33 1:10 dyskinesia, Cognition Dyskinesia, behaviour, cognition 0.9 4 0 12.1 12 F 18 1:32 Behaviour dyskinesia, seizure, behaviour, cognition 1.3 4 0 6.7 13 F 18 1:32 Behaviour, LOC LOC, dyskinesia, seizure, behaviour, cognition 0.9 5 1 13.3 14 M 19 1:10 Seizure, behaviour LOC, dyskinesia, seizure, behaviour, cognition 2.1 5 1 10.3 15 F 40 1:10 Seizure, behaviour Seizure, behaviour, cognition 0.8 4 1 7.2 16 M 22 1:32 Behaviour LOC, dyskinesia, seizures, behaviour, cognition 0.8 4 0 23.9 17 F 41 1:10 Seizure, behaviour LOC, seizure, behaviour, cognition, autonomic instability 1.2 5 1 16 18 F 34 1:32 Behaviour, cognition LOC, dyskinesia, seizures, behaviour, cognition, autonomic instability 26 5 0 36.2 19 F 24 1:32 Seizure, behaviour, Seizure, behaviour, cognition 1.3 5 0 23.2 20 M 21 1:32 Behaviour, seizure LOC, seizure, behaviour, cognition 1.3 4 0 21.5 21 M 44 1:10 Behaviour Seizure, behaviour, cognition 1.2 4 1 11.2 22 M 25 1:32 Behaviour, seizure LOC, dyskinesia, seizures, behaviour, cognition, autonomic instability 1.6 5 0 47 CSF, cerebrospinal fluid; LOC, loss of consciousness; mRS, modified Rankin scale; NMDA, N-methyl-d-aspartate.

Twenty-three individuals with no history of psychiatric or neurologic disease served as HCs in the experiments. All the patients (8 males, age: 27.48 ± 9.49 years, educational level: 13.52 ± 2.34 years) and HCs (8 males, age: 26.43 ± 5.22 years; educational level: 14.65 ± 1.77 years) were evaluated using neuropsychological assessments and underwent an MRI scan.

All subjects provided written informed consent, and the project was approved by the Ethics Committee of the First Affiliated Hospital, School of Medicine, Zhejiang University.

Neuropsychological assessment

A set of comprehensive neuropsychological tests were used, which covered working memory (digit span test, DST), verbal episodic memory (Chinese auditory verbal learning test, CAVLT), executive function (Stroop color and word test) and information processing speed (symbol digit modalities test, SDMT). We also used self-rating anxiety scale (SAS) and self-rating depression scale (SDS) to evaluate emotional state.

MRI data acquisition

All subjects were scanned with a Siemens MAGNETOM Prisma 3 T scanner (Siemens, Erlangen, Germany) at the Center for Brain Imaging Science and Technology, Zhejiang University using the standard setup for clinical studies with a 20-channel phased array head coil. Foam padding was used to minimize head movement, and earplugs were used to reduce scanner noise. High-resolution three-dimensional (3D) T1-weighted MRI scans were collected using a magnetization prepared rapid gradient-echo sequence (repetition time = 2300 msec, echo time = 2.32 msec, inversion time = 900 msec, flip angle = 8°, field of view = 240 × 240 mm2, matrix size = 256 × 256, 192 slices, slice thickness = 0.90 mm). Resting-state fMRI was performed using an echo-planar imaging sequence (repetition time = 1000 msec, echo time = 34.0 msec, flip angle = 62°, field of view = 230 × 230 mm2, 52 slices, slice thickness = 2.50 mm, acquisition time = 5 min, 300 volumes). All the subjects were instructed to lie still with their eyes closed while remaining awake during the resting-state fMRI scans. The framework of characterizing and static dynamic ECN connectivity is shown in Figure S1.

Resting-state fMRI data preprocessing

Resting-state fMRI data were preprocessed using the Graph-theoretical Network Analysis Toolkit (GRETNA)26 based on SPM12 (http://www.fil.ion.ucl.ac.uk/spm/software/spm12). The processing included the following steps. First, the first 20 volumes of functional images were discarded to allow the longitudinal magnetization to reach a steady state and to acclimate the participant to the scanning noise, and then head motion correction were performed. Head motion did not exceed 3.0 mm of maximal translation (in the x, y or z direction) or 3.0° of maximal rotation throughout the course of scanning in any of the subjects. Second, 3D T1 images were aligned to an individual averaged functional image and subsequently spatially normalized to the MNI template (3 × 3 × 3 mm) using diffeomorphic anatomical registration through exponentiated lie algebra (DARTEL) algorithm for spatial normalization. Third, the resulting images were spatially smoothed using a 6-mm full width at half-maximum (FWHM) Gaussian kernel. The acquired smoothed data were utilized in ICA.

Identification of ECN

Group spatial ICA was adopted to decompose all preprocessed data into independent components (ICs) using GIFT software27 with the following steps. First, a two-step principal component analysis was applied to reduce the data into 34 independent components, using the minimum description length criterion.28 The reliability of the infomax ICA algorithm's estimations were evaluated through a comparison against 20 iterative estimates using ICASSO implemented in GIFT.29 Second, we applied a spatially constrained approach called group information-guided ICA30 to perform back reconstruction of participant-specific spatial maps and corresponding time courses. Third, of the 34 independent components, 5 independent components of the ECN were selected by visual comparisons with previously defined maps.31, 32 No network templates were applied. The selected ECN is shown in Figure 1. Finally, additional postprocessing steps were performed on the time courses of the ECN as follows. The time courses of ECN were detrended (linear, quadratic and cubic trends). Then, the resulting images were despiked, outliers were detected, and finally, they were low-pass filtered with a cutoff frequency of 0.15 Hz.

The five independent components of executive control network identified by a group ICA. IC 3, mainly located in the ventrolateral prefrontal cortex; IC 15, mainly located in the dorsolateral and medial prefrontal cortex; IC 19 and 27, mainly located in the right frontoparietal executive network; IC 29, mainly located in the left frontoparietal executive network; ICA, independent component analysis.

Static ECN connectivity analysis

We performed pairwise Pearson correlations to create a static FC matrix of ECN by using the postprocessed time courses between the independent components over the entire scan; these were then converted to z-values using Fisher's z-transformation. Then, we used two-sample t-tests to compare the differences in ECN correlation values between patients and HCs. We used a threshold of p < 0.05 and applied a false discovery rate (FDR) correction to each analysis to correct for multiple correlation comparisons. In addition, to explore the FC between brain regions within the ECN, five region of interest (ROI) spheres (radius: 6 mm) were built around peak activations in respective five ICs of ECN. Then seed-based FC analysis was performed.

Next, we calculated the Pearson correlation between abnormal ECN connectivity and executive control performance in patients and HCs, respectively, to explore the potential relationship. The correlation analyses were performed using SPSS 16.0 for Windows (SPSS Inc., Chicago, IL).

Dynamic ECN connectivity

The processing used the dynamic FC toolbox in GIFT to perform the following steps. First, we used a sliding window approach to explore time-varying changes in FC in five independent components of the ECN during resting-state fMRI scans. The window size was 30 TR convolved with sigma 3 TR of Gaussian, based on previous reports that a window size range of 30 sec to 1 min is a reasonable choice for obtaining dynamic patterns in FC.15, 16, 33 The window was moved in steps of 1 TR, resulting in 250 consecutive windows across the entire scan. A covariance matrix with windowed data was calculated to measure the dynamic FC between the components in the ECN. Then, the changes in FC between components in the ECN for each participant as a function of time were estimated by a 250 × 10 array.

To estimate reoccurring dynamic FC patterns (states), we used k-means clustering methods to cluster the 250-window FC matrices for all subjects. We measured the squared Euclidean distance to estimate the similarity between each FC matrix and the cluster centres. The k-means algorithm was iterated 500 times and repeated 150 times on the exemplars of all patients and controls to obtain the group cluster centroids (functional dynamic ECN states). We used the elbow criterion to estimate the optimal number of centroid states (which was determined to be k = 4), where each centroid state represented a functional dynamic ECN state.

Then, we examined whether the patients and HCs presented different temporal properties of different functional dynamic ECN states during resting-state fMRI. To count the occurrences of one state, we used only participant data where at least 10 windows belonged to that state. We calculated the following measures in each subject, including: (1) mean dwell time in each state, measured by averaging the number of consecutive windows belonging to one state before changing to another state, which represents how long a subject stayed in a certain state; (2) fractional windows, assessed by the proportion of time spent in each state; and (3) number of transitions, represented state stability over time, and they were measured by the number of times a subject switched from one state to the other. We performed a two-sample t-test (p < 0.05, FDR corrected) to examine the group differences in mean dwell time and the fractional window in each state and number of transitions between patients and HCs.

Next, we calculated the Pearson correlation between the abnormal temporal properties of dynamic ECN and executive control performance in patients and HCs, respectively, to explore any potential relationship. The correlation analyses were performed using SPSS 16.0 for Windows (SPSS Inc.).

Results Demographic data and neuropsychological assessment

The patients and HCs did not differ significantly with regard to the age (t = −0.445, p = 0.659) or educational level (t = 1.79, p = 0.081). The neuropsychological test results assessed by a two-sample t-test are shown in Table 2. The Bonferroni correction was applied for multiple comparisons to reduce the risk of type 1 errors. Hence, a stricter threshold of 0.00625 (0.05/8) as the level of statistical significance was used to explain the results of these two-sample t-tests. The patients showed significant impairments in the SDMT (t = 4.27, p = 0.000) and DST (backward, t = 4.44, p = 0.000). Uncorrected results showed additional impairments in the Stroop test (dots, t = −2.44, p = 0.019) and CAVLT (immediate, t = 3.110, p = 0.003; delayed, t = 3.356, p = 0.002). However, the scores on SAS (t = −1.53, p = 0.133) and SDS (t = −0.81, p = 0.425) were not significantly different between the two groups.

Table 2. Results of cognitive tests of patients and healthy controls. Item HCs (mean ± SD) Patients (mean ± SD) t p Age (years) 26.43 ± 5.22 27.48 ± 9.49 0.445 0.659 Education (years) 14.65 ± 1.77 13.52 ± 2.34 1.79 0.081 SDMT 59.26 ± 7.57 47.81 ± 10.13 4.27 0.000** DST (forward) 9.30 ± 0.97 8.33 ± 1.74 2.31 0.026 DST (backward) 7.48 ± 1.31 5.57 ± 1.43 4.61 0.000** Stroop test (colour dot) 12.72 ± 2.04 14.44 ± 2.73 −2.44 0.019* Stroop test (colour word) 24.27 ± 5.96 31.07 ± 1.26 −2.33 0.025 CAVLT (immediate memory following interference) 14.04 ± 0.82 12.29 ± 2.57 3.11 0.003* CAVLT (delayed recall) 13.83 ± 1.03 11.90 ± 2.53 3.36 0.002* CAVLT (recognition) 14.87 ± ````0.34 14.43 ± 0.93 2.13 0.039 SAS 22.09 ± 2.54 23.33 ± 2.85 −1.53 0.133 SDS 21.91 ± 3.18 22.61 ± 2.58 −0.81 0.425 DST, Digit Span Test; HCs, healthy controls; SAS, Self-Rating Anxiety Scale; SDMT, Symbol-Digit Modalities Test; SDS, Self-Rating Depression Scale; CAVLT, Chinese auditory verbal learning test. Static functional connectivity of ECN

The five ICs selected as ECNs of interest (Fig. 1) were primarily located in association cortex areas and underwent further analysis of FC changes. IC 3 and IC 15 were mainly located in the prefrontal cortex. IC 3 was mainly located in the ventrolateral prefrontal cortex, whereas IC 15 was mainly located in the dorsolateral prefrontal cortex and medial prefrontal cortex. IC 19, IC 27 and IC 29 were mainly located in frontoparietal association cortex areas: IC 19 and IC 27 were mainly located in the left frontoparietal executive network, whereas IC 29 was mainly located in the right frontoparietal executive network. For the FC between brain regions within the ECN, the results were not significant between the patients and HCs.

Further analysis indicated that, compared to HCs, increased FC was observed between IC 15 and IC 27, IC 15 and IC 29 in patients with anti-NMDA receptor encephalitis (FDR corrected, p < 0.05; Table 3 and Fig. 2). Uncorrected results showed increased FC was observed between IC 27 and IC 29 (uncorrected p < 0.05; Table 3). Among the patients, when controlling for age, gender and education in a partial correlation, a linear correlation analysis showed a significant positive correlation between IC 15-IC 29 FC and DST (backward) performance (r = 0.704, p = 0.003, Fig. 3A) and between IC 15-IC 27 FC and SDMT performance (r = 0.712, p = 0.003, Fig. 3B). No significant correlations between the FC of the ECN and subjects' performances in the DST (forward) test, CAVLT or Stroop test were observed. No significant relationship between static ECN connectivity and cognitive performance was found for HCs.

Table 3. Results of static functional connectivity among independent components of ECN. Functional connectivity t p IC 3-IC 15 −0.1151 0.9089 IC 3-IC 19 0.0049 0.9961 IC 3-IC 27 0.5391 0.5926 IC 3-IC 29 1.534 0.1325 IC 15-IC 19 −0.5958 0.5545 IC 15-IC 27 3.0567 0.0039* IC 15-IC 29 3.2336 0.0024* IC 19-IC 27 −0.7767 0.4417 IC 19-IC 29 0.2891 0.7739 IC 27-IC 29 2.5718 0.0137** ECN, executive control network; IC 15, mainly located in the dorsolateral and medial prefrontal cortex; IC 19 and 27, mainly located in the right frontoparietal executive network; IC 29, mainly located in the left frontoparietal executive network; IC 3, mainly located in the ventrolateral prefrontal cortex.

Results of functional connectivity among independent components of ECN. ECN, executive control network; HCs, healthy controls; IC 15, mainly located in the dorsolateral and medial prefrontal cortex; IC 27, mainly located in the right frontoparietal executive network; IC 29, mainly located in the left frontoparietal executive network. *p