We recruited 42 treatment-naive patients with first-episode schizophrenia (Patients-0W) and 38 education-, gender-, and age- matched healthy controls in the Second Affiliated Hospital of Xinxiang Medical University in China from 2012.12 to 2014.01, China. This fMRI dataset has been used previously by our group (22, 23). Experienced psychiatrists diagnosed patients according to the Structured Clinical Interview for DSM-IV-TR. The illness duration of all the patients in this study was less than 12 months. SCID-non-patient edition was used to scan controls to confirm that they had no history of other psychiatric or neurological disorders. In this observational cohort study, patients were treated with risperidone, and then followed for 8 weeks.

The study was approved by the Ethics Committee of Henan Mental Hospital and Second Xiangya Hospital of Central South University (protocol code S088, 2012). It was conducted in accordance with the Declaration of Helsinki. Written informed consent was acquired from all the participants involved in the study.

The initial dose of risperidone was 2 or 1 mg per day within the first week. After 1 week, a slow titration was then utilized, and the risperidone dose was increased at 1-week intervals until patients showed satisfactory improvement. After 4 weeks, patients who did not exhibit obvious improvement were prescribed up to 6 mg/day of risperidone. All patients were treated with risperidone monotherapy for 8 weeks. Patients were also treated with benzhexol if they had extrapyramidal symptoms. During the 8-week intervention, some patients (n = 19) were given at least one dose (ranging from 1 to 2 mg) of lorazepam; however, this medication could not be taken within 1 week before the second scan. Patients were also not allowed to use antidepressants or mood stabilizers during the intervention period, as these medications are potential synergists. Treatment safety was evaluated weekly through clinical interviews with doctors.

Symptom severity was assessed both before and after 8 weeks of treatment on the day of MRI scanning by using the Positive and Negative Syndrome Scale (PANSS) (24), which primarily evaluates positive (PANSS-P, Items P1–P7) and negative (PANSS-N, Items N1–N7) symptom dimensions.

The participants underwent T1 weighted (T1W) imaging and fMRI on a 3.0T Siemens MRI scanner (Verio) at the Magnetic Imaging Centre of the Henan Mental Hospital. A standard 16-channel head coil was utilized. We scanned patients both at baseline and at the 8-week follow-up, while healthy controls only underwent the baseline scans. T1W images were obtained sagittally with a gradient echo pulse sequence. The detailed parameters were echo time (TE) = 2.52 ms, repetition time (TR) = 1,900 ms, field of view (FOV) = 250 mm × 250 mm, flip angle (FA) = 9°, slice gap = 0 mm, slice thickness (ST) = 1.0 mm, and number of slices = 176. Resting-state fMRI data were acquired with a gradient-echo echo-planar imaging sequence. The parameters were TE = 30 ms, TR = 2,000 ms, FOV = 240 mm × 240 mm, FA = 90°, time point = 240, matrix size = 64 × 64, ST = 3 mm, voxel size = 3.75 mm × 3.75 mm × 3 mm, and slice gap = 0. Patients were required to relax and keep their eyes closed while staying awake during the scanning.

Four patients withdrew from the follow-up scans. We therefore obtained imaging data from 42 patients at baseline (patients 0W) and 38 patients at the 8-week follow-up (patients-8W) as well as 38 healthy volunteers.

We used the Data Processing Assistant for Resting-State fMRI (DPARSF,¹ DPARSF5.2) (25) to preprocess the fMRI images. The DPARSF is a toolbox for MATLAB (R2019b) and Data Processing and Analysis of Brain Imaging (DPABI)² (26). Slice timing and EPI undistortion correction were performed. Any subject with head motion >1.5° rotation or >1.5 mm translation in any direction was removed. We regressed out the motion parameters, white matter, and CSF from the time series. All subjects’ functional images were normalized to the Montreal Neurological Institute (MNI) space with the standard EPI template (resampled voxel size 3 × 3 × 3 mm³). We also conducted T1 coregistration analysis to preprocess the imaging data. The resampled images were then smoothed with the Gaussian kernel and detrended to remove the linear drift of BOLD signals.

The baseline fMRI data of four patients failed to meet the standards of quality control, and these four patients withdrew from the follow-up imaging scans.

We imported the AAL template and extracted the time series of the 3 ROIs (putamen, pallidum, and caudate) in DPABI. The MNI coordinates and the position of the 3 ROIs in the Automated Anatomical Labeling (AAL) template are shown in Supplementary Figure 1. DPABI was used to extract the time series of the 3 ROIs and all other brain regions. We averaged the BOLD signal intensity of all the time points (fMRI time frames) of each ROI.

With the goal of extracting striatum-related spontaneous CAPs, we first chose the top 15% of time points based on previous evidence (14), which demonstrated that the similarity (spatial associations) between the ROI-seeded association map utilizing all time points and the averaged signals of the selected time points rapidly increased while including more time points by decreasing the selection threshold; the spatial associations then reached a plateau after including ∼ 15% of all time points. Then, we selected the top 36 (240 × 15% = 36) time points for the 3 ROIs as the highly activated state. The similarity between our approach and a previous study (14) is the setting of the threshold (top 15% time points), while the difference lies in the selection of the seed-regions.

The three-dimensional brain maps of the 36 highly activated time points from the whole-brain voxel level were extracted and normalized by subtracting the mean BOLD signal intensity at each time point and dividing that value by the standard deviation. We averaged the normalized BOLD signal intensity of the highly activated time points on a regional level. The 3 ROI-related spontaneous CAPs in the control group are shown as an example in Supplementary Figure 2.

We compared the differences in striatum-related CAPs between Patients-0W and controls as well as between Patients-0W and Patients-8 by performing independent samples t-tests and paired t-tests, respectively. GRF correction was performed as well, with voxel level P < 0.01, cluster level P < 0.05, and voxel number >50.

To extract further information about the dynamic variations in longitudinal ROI-related CAPs after treatment, after extracting striatum-related spontaneous CAPs, we then classified the 36 detected highly activated time points (in the ROI-related CAPs) according to their spatial similarity utilizing the k-means clustering approach. The above approach, i.e., clustering the top 15% of time points, has been applied previously (14). This process was performed on Dynamic Brain Connectome (Dynamic BC),³ a MATLAB toolbox (Supplementary Figure 3). The CAPs of both Patients-0W and Patients-8W were classified into three states. The brain maps in each state (CAP 1, CAP 2, and CAP 3) were extracted and averaged for Patients-0W and Patients-8W. We used paired sample t-tests to compare the longitudinal changes in BOLD signal intensity in each state, with GRF correction, voxel level P < 0.01, cluster level P < 0.05, and voxel number >50.

Overlapping areas between regions with case-control differences and those with longitudinal differences were viewed as effective treatment targets. We used Pearson’s correlation analysis to compute the correlations between patients’ clinical symptom improvement and longitudinal changes in imaging markers of the overlapping regions, including longitudinal alterations in CAPs and dynamic CAP states.

We detected no significant differences in the between-group comparisons of age, gender, handedness, and education data of patients and controls (Ps > 0.05; Supplementary Table 1). Patients showed significant clinical improvement in positive symptoms (Ps < 0.001; Supplementary Table 2), while their negative symptoms showed no significant alterations (P > 0.05; Supplementary Table 2).

In the putamen-related CAPs, patients showed significantly increased intensity in the bilateral thalamus, bilateral supplementary motor area (SMA), bilateral medial, and paracingulate gyrus, left paracentral lobule, left medial superior frontal gyrus, and left anterior cingulate gyrus compared with healthy controls (GRF correction, voxel level P < 0.01, cluster level P < 0.05, and voxel number >50, Figure 1A and Supplementary Table 3). There were no significant between-group differences in the caudate- or pallidum-related CAPs.

In pallidum-related CAPs, patients had significantly increased intensity in the left lingual, fusiform, middle temporal, superior temporal, and angular gyrus after treatment (GRF correction, voxel level P < 0.01, cluster level P < 0.05, and voxel number >50, Figure 1B and Supplementary Table 4). In the putamen-related CAPs, patients had significantly decreased intensity in the left lingual gyrus (GRF correction, voxel level P < 0.01, cluster level P < 0.05, and voxel number >50, Figure 1C and Supplementary Table 4). Regarding the caudate-related CAPs, patients showed no significant longitudinal alterations.

There were no overlapping regions between areas with case-control differences and those with 8-week longitudinal differences in patients. We therefore did not further analyze the correlations between patients’ clinical symptom improvement and the longitudinal changes in neuroimaging measurements.

We detected significant longitudinal differences in the three subregion-related dynamic CAP states (GRF correction, voxel level P < 0.01, cluster level P < 0.05, and voxel number >50), except the putamen-related CAP 2 (see Table 1).

By comparing the differential brain regions of Patients-0W vs. Controls (Figure 2A) and that of Patients-0W vs. Patients-8W (Table 1), we found that thalamus (detected in the putamen-related CAP 1, Table 1) and medial and paracingulate gyrus (detected in the putamen-related CAP 3, Table 1) were the overlapping brain regions. Thalamic signals in the putamen-related CAP 1 showed a significant increase (GRF correction, voxel level P < 0.01, cluster level P < 0.05, and voxel number >50, Figure 2A), while the signals of the medial and paracingulate gyrus in the putamen-related CAP 3 significantly decreased after treatment (GRF correction, voxel level P < 0.01, cluster level P < 0.05, and voxel number >50, Figure 2B).

The increase in thalamic signal intensity in the putamen-related CAP 1 showed significantly positive correlations with the percentage reduction (Patients-0W–Patients–8W/Patients-0W) of PANSS_P (r = 0.5302, P = 0.017 with Bonferroni correction, Figure 2C). We did not detect significant relations between the percentage reduction of PANSS_P and the signal intensity decrease of the medial and paracingulate gyrus in the putamen-related CAP 3 (P > 0.05 with Bonferroni correction).