Altered Dynamic Functional Connectivity of Cuneus in Schizophrenia Patients: A Resting-State fMRI Study

Objective: Schizophrenia (SZ) is a functional mental condition that has a significant impact on patients’ social lives. As a result, accurate diagnosis of SZ has attracted researchers’ interest. Based on previous research, resting-state functional magnetic resonance imaging (rsfMRI) reported neural alterations in SZ. In this study, we attempted to investigate if dynamic functional connectivity (dFC) could reveal changes in temporal interactions between SZ patients and healthy controls (HC) beyond static functional connectivity (sFC) in the cuneus, using the publicly available COBRE dataset. Methods: Sliding windows were applied to 72 SZ patients’ and 74 healthy controls’ (HC) rsfMRI data to generate temporal correlation maps and, finally, evaluate mean strength (dFC-Str), variability (dFC-SD and ALFF) in each window, and the dwelling time. The difference in functional connectivity (FC) of the cuneus between two groups was compared using a two-sample t-test. Results: Our findings demonstrated decreased mean strength connectivity between the cuneus and calcarine, the cuneus and lingual gyrus, and between the cuneus and middle temporal gyrus (TPOmid) in subjects with SZ. Moreover, no difference was detected in variability (standard deviation and the amplitude of low-frequency fluctuation), the dwelling times of all states, or static functional connectivity (sFC) between the groups. Conclusions: Our verdict suggest that dynamic functional connectivity analyses may play crucial roles in unveiling abnormal patterns that would be obscured in static functional connectivity, providing promising impetus for understanding schizophrenia disease.


Introduction
Schizophrenia (SZ) is a mental condition in which individuals experience hallucinations [1], delusions, and profoundly disorganized thinking and behavior as a result of their erroneous interpretation of reality. Positive symptoms such as hallucinations, delusions, and flight of ideas are common in SZ patients, as are negative symptoms such as apathy, emotionlessness, lack of social functioning, and cognitive symptoms such as difficulties concentrating and paying attention, as well as memory deficits [2]. Literature suggests that a person's risk of developing the illness is increased by a mix of physical, genetic, psychological, and environmental variables. A stressful or emotional life experience may precipitate a psychotic episode in some persons who are predisposed to SZ. Approximately half of SZ patients have a history of drug abuse disorders [3], and nearly all of them smoke [4]. An Omega-3 polyunsaturated fatty acids (PUFAs) dietary imbalance could aggravate its course while also raising the risk of metabolic complications [5]. Despite the availability of more anterior parts of the cuneus, though usually for higher-level operations such as object perception and spatial processing, such as attention [45]. Different analysis tools have been used to describe brain FC using rsfMRI data, including seed-based [11,38], independent component analysis (ICA) [15,[46][47][48], principal component analysis (PCA) [49], clustering [50,51], fisher discriminant dictionary learning (FDDL) [52], centrality [53], multivariate pattern analysis (MVPA) [54,55], and graph theory [56,57]. In this study, we used seed-based analysis, in which the connectivity patterns are based on a selected seed region of interest (ROI). The advantage of this method is that the results are focused on particular ROIs and, hence, could be easier to understand in relation to neuropathology [58].
With rsfMRI data of 72 SZ patients and 74 HC groups from the COBRE dataset, sFC and dFC were used to analyze the whole brain in both groups, and the dynamic characteristics (standard deviation (SD) and mean strength (Str)) were used to estimates the dFC. We used the proposed seed-based FC method to obtain the FC between the cuneus and other brain regions and compare whether sFC and dFC might provide factual information by identifying dFC indicators in order to understand SZ neural mechanisms better.

Participants
Data used in this study were from COBRE [59], consisting of 72 SZ patients (fourteen females and fifty-eight males) and 74 HCs (twenty-three females and fifty-one males). Participants were screened and ruled out if they had a history of neurological illness, mental retardation, extreme head trauma resulting in more than 5 min of loss of consciousness, or history of drug abuse or dependence during the previous 12 months. Diagnostic information was collected using the Structured Clinical Interview used for DSM Disorders (SCID) [59]. The inclusion was of men and women of 37.0 ± 12.78 years of age. Collected T1-weighted (MPRAGE) image parameters were: voxel size = 1 mm × 1 mm × 1 mm flip angle = 7 • , echo time (TE) = 1. 64, 3.5, 5.36, 7.22, 9.08, repetition time (TR) = 2530, field of view (FoV) = 256 mm × 256 mm, inversion time (TI) = 900 ms, thickness = 176 mm, matrix = 256 × 256 × 176. Functional MRI data were gathered with: TR/TE = 2 s/29 ms, 33 slices, voxel size: 3 × 3 × 4 mm 3 , matrix = 64 × 64, and 150 time points. All participants were asked to remain relaxed and keep their eyes open during the scan.
The steps for data preprocessing were as follows: one subject was discarded because of having too-few time points in image data (67 instead of 150). For the rest of subjects, we removed the first five volumes, leaving 145 volumes; corrected slice-timing and realigned images; manually reoriented the T1 and functional images; co-registered T1 images into functional images then segmented images to gray matter (GM), white matter (WM), and cerebrospinal fluid (CSF); nuisance covariates were regressed (including Friston 24 head motion parameters [63] and white matter and cerebrospinal fluid signals), functional images were then normalized into MNI space DARTEL Method [64] and resliced to 3.0 × 3.0 × 3.0 mm 3 ; spatial smoothing of normalized images was performed at 6 mm FWHM; band-pass filtering images at 0.01-0.08 Hz as it was found that resting-state signals are consistent low-frequency fluctuations in this range [65]. To address head motion in fMRI, image volumes with mean framewise displacement FD (Jenkinson) > 0.5 mm were scrubbed to reduce its effect using cubic spline interpolation [66,67]. Subjects were not excluded as they did not exceed the head transition < 3 mm, rotation < 3 • [68].

Regions of Interest Definition and Network
To identify differences in rsFC between SZ and HC groups, regions obtained from the automated anatomical labelling template (AAL) [69] were applied to determine the FC based on the region of interest (ROI). Except the cerebellum [70], the 90 ROIs of AAL were divided into prefrontal regions, other regions of frontal lobe, parietal regions, occipital regions, temporal regions, and subcortical regions. Finally, mean time courses from all 90 ROIs were extracted to calculate functional connectivity.

Static Resting-State Functional Connectivity
With the remaining 145-image volumes, we estimated the Pearson's correlation coefficient of Fisher's z transformation between each pair of the mean time course in 90 ROIs, to enhance the normality of correlation distribution [70] and for the construction of the static rsFC brain network.

Dynamic Resting-State Functional Connectivity
The sliding window method [50] was applied, implementing a window size at 30 TRs (60 s) being that 30 s to 60 s is considered good enough [15,50,71]. We selected a sliding temporal window length of 30 TRs (60 s) to 145 data length (290 s). Rectangular sliding windows unconvolved with a Gaussian kernel were then used to capture sharper transitions that could be undetected in a tapered window [72]. By sliding the window by the 2 s step size 1 TR, 116 temporal windows (145 − 30 + 1) were generated. Finally, 116 Fisher's z-transformed Pearson's correlation maps (90 × 90 matrix size, for each window) for each subject were obtained as the dynamic functional connectivity maps ( Figure 1). scrubbed to reduce its effect using cubic spline interpolation [66,67]. Subjects were not excluded as they did not exceed the head transition < 3 mm, rotation < 3° [68].

Regions of Interest Definition and Network
To identify differences in rsFC between SZ and HC groups, regions obtained from the automated anatomical labelling template (AAL) [69] were applied to determine the FC based on the region of interest (ROI). Except the cerebellum [70], the 90 ROIs of AAL were divided into prefrontal regions, other regions of frontal lobe, parietal regions, occipital regions, temporal regions, and subcortical regions. Finally, mean time courses from all 90 ROIs were extracted to calculate functional connectivity.

Static Resting-State Functional Connectivity
With the remaining 145-image volumes, we estimated the Pearson's correlation coefficient of Fisher's z transformation between each pair of the mean time course in 90 ROIs, to enhance the normality of correlation distribution [70] and for the construction of the static rsFC brain network.

Dynamic Resting-State Functional Connectivity
The sliding window method [50] was applied, implementing a window size at 30 TRs (60 s) being that 30 s to 60 s is considered good enough [15,50,71]. We selected a sliding temporal window length of 30 TRs (60 s) to 145 data length (290 s). Rectangular sliding windows unconvolved with a Gaussian kernel were then used to capture sharper transitions that could be undetected in a tapered window [72]. By sliding the window by the 2 s step size 1 TR, 116 temporal windows (145 -30 + 1) were generated. Finally, 116 Fisher's z-transformed Pearson's correlation maps (90 × 90 matrix size, for each window) for each subject were obtained as the dynamic functional connectivity maps ( Figure 1).

Figure 1.
Dynamic network analysis flow-chart. A total of 90 ROIs times series from AAL template were extracted from fMRI data (145 TR), from which a rectangular sliding window (W = 30 s) was used to generate dynamic rsFC maps.

Measurements of the Dynamic Characteristics
Mean strength and temporal variability (standard deviation and the amplitude of low-frequency fluctuation (ALFF)) were measured in order to demonstrate the dynamic characteristics, as follows: where: z(i, j) t represents ROI i and j FC value at window t, z(i, j) m ; mean/average of z(i, j) t over the total of T windows [73], dFC-Str; strength and dFC-SD as temporal variability.
For the ALFF analysis, it was carried out using DPABI. The power spectrum was produced by transforming the filtered time series of each voxel into the frequency domain using a Fast Fourier Transform. We calculated ALFF for each voxel by taking the square root of the signal at 0.01-0.08 Hz [74]. The ALFF of each voxel was further divided by the global mean of ALFF values for each participant within the default brain mask from the DPABI, with background and other non-brain tissue signals discarded, for uniformity and to limit the influence of individual variance in ALFF values. This resulted in a standardized ALFF map for the entire brain.

Characterization of dFC States' Property
We examined the mean dwelling time (MDT), which measures the amount of time an individual spends in a specific state, to see if SZ had any effect on the features of transient states [24,75].

Statistical Analyses
We applied two-sample t-tests to test whether the mean strength (dFC-Str) and temporal variability (dFC-SD and ALFF) within each temporal windowed connectivity map present any significant differences between SZ and HC groups. False Discovery Rate (FDR)corrected activations with p < 0.05 [76,77] were conducted, with confounding variables such as age, gender, IQ, and handed-ness (Table 1) as covariates in the GRETNA toolbox [78]. Permutation tests (p < 0.05) were also used to evaluate the differences between groups in state properties (MDT).  Table 1 shows subjects' clinical information, in which no significant difference was seen between age, sex, handed-ness, and education years of the two groups (p > 0.05) by two-sample t-tests. However, significant differences were seen in IQ (p < 0.05) using two-sample t-tests. The PANSS scales were reported in the patients' group and the primary diagnosis information for the SZ patient group and the HC group are presented in Table 2. Other information available in COBRE INDI Additional data in Supplementary Materials.

Static Functional Connectivity
No significant difference (p > 0.05) was apparent in static rsFC between schizophrenia patients and healthy controls (Table 3 at 290 s), suggesting that static rsFC might be insensitive in capturing neural functional abnormalities underlying schizophrenia.

Discussion
This is one of the few dFC studies of schizophrenia based on the COBRE public dataset [79][80][81][82][83]. In this study, we investigated differences in functional connectivity (FC) dynamics at rest between schizophrenia (SZ) patients and healthy controls (HC). Sliding windows were used to construct dynamic resting-state functional connectivity maps whose windowed cuneus-connectivity properties were assessed. We also explored the group differences in large time-scale connectivity (static perspective), which were computed as correlations of fMRI time series over the full range of scanning time. We found that, while there were no significant differences in static functional connectivity (sFC), mean strength (dFC-Str) analyses presented SZ patients with decreased cuneus-connectivity findings that could not be identified in static analyses. However, the variability of cuneus connectivity did not demonstrate any significant difference between the groups in all temporal correlation-windows.
The cuneus is delimited by the parieto-occipital sulcus and the calcarine sulcus in the occipital lobe, known for its role in both primary and secondary visual processing. With respect to decreased functional connectivity, fairly similar findings to our observations were reported in previous studies such as decreased gray-matter volumes in the cuneus and lingual gyrus as compared to healthy participants [84,85]. Another study illustrated that when glaucoma patients were compared to healthy people, the functional connectivity of the cuneus decreased [36,86]. The cuneus also indicated substantial reduction in brain neural homogeneity in people with acute open-globe injury [87] and retinal detachment [88] in another study. Some other studies emphasize that smaller volumes of gray matter in the right cuneus [89] and lower activity in the cuneus and precuneus [90] were associated with suicide attempters. Tohid and colleagues found reduced activation in the right cuneus and fusiform gyrus during episodic memory retrieval in SZ patients compared with HCs [91]. In a first-episode psychosis study, patients with poor insight had decreased gray matter volume in the cuneus and medial gyrus of the right occipital lobe as compared to those with good insight [92]. Together, these findings suggest that patients with schizophrenia are characterized by altered cuneus connectivity in large scale brain networks.
Previous investigations, on the other hand, were unable to identify reduced connectivity in the cuneus, but did find it in the occipital lobe. For example, Glahn et al. (2008) found a reduced gray-matter density in SZ patients relative to HCs [93], lower global connectivity [94], and decreased connectivity strength [95]. The previous study by Phillips and colleagues revealed reduced external white matter fractional anisotropy, especially in bilateral occipital lobes, in SZ patients compared to their relatives [96]. Hallucination was also associated with a reduced volume of the occipital lobe in SZ patients [97], reduction in the overall volume of occipital lobe in SZ [97][98][99], lower white-matter volume in first-episode patients, and decreased activation in fusiform gyrus [100][101][102]. No cuneusconnectivity attenuation was reported by the authors in all these studies. One reason could be the limitations of analysis techniques or because alterations in brain activity linked to illnesses are not always supported by large-scale evidence. Rather, they are visible across a small time-scale of seconds, necessitating short time-scale investigations to capture those deficit patterns.
In our study, SZ patients exhibited decreased cuneus mean strength connectivity (dFC-Str) in several brain areas, such as the left cuneus and right calcarine, the right cuneus and right calcarine, the right cuneus and right lingual gyrus, and between the right cuneus and right-middle temporal gyrus. These changes were all observed in dFC and not in sFC, this may be because some brain disorders need short-time-scale analyses in order to capture subtle deficits., Previous and current rsfMRI researches on SZ have increasingly focused on dFC [1,22,23], especially in specific time intervals (windows) [15,23,[103][104][105][106][107]. Our results are in accordance with Damaraju et al. in SZ and Nguyen et al. in Bipolar Disorder studies, who demonstrated how dynamic FC has more predictive accuracy than static FC [15,108]. We also observed similar connections in different time. The evidence in Table 3 shows how the connections were lost in bigger window sizes (Table 3 at 75 TR) and at full-length (145 TR). Together, these findings point to the conclusion that information that is likely to be lost in static FC (bigger window sizes approaching to full-length) can be trapped in the dynamic FC.
Although we only found a decrease in cuneus connectivity in this study, previous research has found that SZ can also cause an increase in cuneus connectivity. Notably, increased dynamic ReHo was also discovered in the super temporal gyrus, postcentral gyrus, thalamus, middle cingulum cortex, and cuneus in SZ patients [106]. Other regions have also been linked to an increased functional connectivity in SZ illnesses, such as an increased global-brain functional connectivity in the left superior frontal gyrus [109].
Even though the majority of SZ studies have emphasized the decreased FC patterns, the lack of increased connection patterns in these studies and our study could be linked to the type of SZ patients regarding the stage of illness and medication [110]. The brain compensatory mechanism's inability to recruit more resources to retain, maintain, or restore cognitive functioning, or in response to cognitive demands, can be viewed as the main indication of decreased connectivity in SZ. Since it has long been assumed that SZ is a spectrum of diseases caused by several pathophysiological processes rather than a single disease [111], other factors such as diverse samples and different preprocessing approaches could also lead to this inconsistency.
Even though the cuneus is believed to be associated with cognitive functions such as behavior engagement [112], working memory [113], and cognitive control [43]. It is thought to be largely engaged in visual processing. The cuneus, in particular, appears to be involved in detecting changes in ocular position as a result of spontaneous eye movements [114]. Impairments in smooth-pursuit eye movements have been repeatedly recorded, especially in SZ patients [115]. These studies reveal that SZ patients, or those who are at risk of developing the condition, may have problems monitoring their own eye movements and this could be because of reduced connectivity in the cuneus. Evidenced by previous works, the SZ decreased connectivity in visual areas could also be because those affected typically perform worse in visual motion [116] and visual contrast [117]. Whether the cuneus anatomical abnormalities seen in the SZ subjects in this study are also related to abnormal eye tracking is a topic worthy of future discussion.

Limitations
Our study, however, presented some limitations; first, 41 subjects (more than half of SZ) were of Paranoid type, suggesting that it is unclear whether the observed results were of all SZ patients or were influenced by the large paranoid-type data (57%). Second, patients with schizophrenia may be using a variety of antipsychotic medications, which may influence their cognitive functions. A previous study revealed that SZ patients showed synchronized brain activity and decreased integration function across brain networks after undergoing six-week treatment with second-generation antipsychotic medications [118]. These antipsychotic medications may also affect their brain activities. For example, olanzapine treatment increased default mode network (DMN) connectivity in patients with SZ [119]. Hence, attention should be paid while interpreting our results. Third, although we did not achieve statistical significance in variability, it is possible that a better-designed study could provide more definite results. Fourth, the lack of neuropsychological tests for SZ and HC may also limit the study since the data used in this study were obtained online (COBRE), resulting in no correlation analysis between altered rsFC and neuropsychological evaluation in both groups. Thus, intergroup difference in neuropsychological evaluation also remains unknown. Fifth, the authors of this work acknowledge that the sliding window technique has a number of drawbacks [25,120,121], such as that multiple parameters must be specified including window function, length, and step size, but due to a shortage of ground truth in resting-state fMRI data, the ideal settings are still uncertain. We confirm that our work, just like the vast majority of resting-state-dFC studies have employed this technique, mainly due to its simplicity [71]. Lastly, our biggest challenge was to obtain the dataset, and the main reason was the COVID-19 pandemic, which limited us only to use publicly available dataset.

Conclusions
In this study, we used seed-based method to obtain the FC between the cuneus and other brain regions in individuals with schizophrenia. Results revealed that patients with SZ indicated decreased mean strength connectivity in short-time-based analysis (dynamic) in the cuneus, relative to their matched healthy control, while presenting no significant differences in variability and full-time (static) sense. This suggests that strong insights in understanding brain deficits underlying schizophrenia can be provided by dynamicconnectivity analyses of the cuneus. This hypothesis will need to be investigated further, utilizing various structural, functional, and metabolic methodologies. To address the suppressing effect associated with the sliding windows, the authors of this work are looking forward to use windows with a flattened spectrum within the bandwidth such as modulated rectangular (mRect) [122] in our future studies.