In June 2014, Typhoon Rammasun struck Hainan Province and caused at least 14 deaths. More than a thousand people were trapped and almost drown in the worst hit area. From November 2014 to January 2015, we ultimately recruited 70 trauma-exposed survivors. This study was conducted according to the declaration of Helsinki and was granted permission by the ethics committee of Jinling Hospital, People's Hospital of Hainan Province and the Second Xiangya Hospital. Before the examination, all the participants did not undergo any anti-depressant or psychotherapy. The inclusion criteria of the current study were as followed: age 18–65 years; right-handedness; no use of psychiatric medication or substances abuse; without MR imaging contraindications. Subjects with any history of head trauma, loss of consciousness, long-term significant physical conditions, neurologic or psychiatric disorders except for depression and anxiety should be excluded. After considering the rigid requirements, we totally excluded 10 trauma-exposed subjects for failing to obtain complete imaging data (n = 3), excessive head movement (n = 3), brain infarction (n = 1), denture-artifacts (n = 2), and pregnancy (n = 1). Thirty demographically matched healthy controls (HC) without trauma-exposure were also enrolled in our study. Every participant provided written informed consent prior to MRI scan and neuropsychological assessments.

All the typhoon survivors were screened with PTSD checklist-civilian version (PCL_C) (25), a 17-item self-report instrument designed to assess symptoms of PTSD (26). Those subjects who PCL_C scored more than 35 were further administrated with Clinician-Administrated PTSD Scale (CAPS) to estimate the frequency and intensity of each sub-symptom including re-experience, avoidance, and hyper-vigilance (27). The remaining subjects scoring <30 were considered as the trauma-exposed controls (TECs). The comorbidities with other psychiatric disorders were confirmed via the Structured Clinical Interview for DSM-IV (28). Additionally, emotional assessments including Self-Rating Anxiety Scale (SAS) (29) and Self-Rating Depression Scale (SDS) (30) were conducted on all participants to estimate emotional status.

The resting-state functional MR imaging was acquired with a 3.0 Tesla MR scanner (Skyra, Siemens Medical Solutions, Erlangen, Germany) equipped with the standard head coil. Each subject was instructed to lay supine, rest, and keep his or her eyes closed with the head still during MRI scanning. Firstly, the routine diagnostic T1 weighted image and T2 fluid-attenuated inversion-recovery image acquisitions were conducted to rule out subjects with structural brain lesions. The resting-state fMRI data were then acquired using a single-shot, gradient-recalled echo planar imaging (250 volumes, repetition time [TR]/echo time [TE] = 2,000/30 ms, flip angle = 90°, image matrix = 64 × 64, FOV = 230 × 230 mm², slice thickness = 3.6 mm, 35 axial slices with no intersection gap). Each volume was whole-brain coverage and aligning along the anterior-posterior commissure. Additionally, high-resolution T1-weighted 3-Dimension anatomical images were obtained in the sagittal orientation using a rapid gradient-echo sequence (TR/TE = 2000/1.97 ms, flip angle = 9°, image matrix = 256 × 256, FOV = 256 × 256 mm², slice thickness = 1 mm, 176 slices).

Initial data preprocessing was conducted using the Data Processing Assistant for rs-fMRI advanced edition (DPARSFA, http://www.restfmri.net) (31) based on MATLAB (The Math Works, Inc., Natick, MA, USA) platform. The first 10 volumes of each fMRI data were removed for the signal equilibrium. Then, the subsequent procedures were performed on the remaining 240 volumes, including slice timing, realignment, and co-registered with the anatomical scan. The co-registered data was further segmented into gray matter, white matter (WM), and cerebrospinal fluid (CSF) and normalized into standard Montreal Neurological Institute (MNI) space with a final voxel size of 3 × 3 × 3 mm³. Additionally, preprocessing for temporal lagged analysis included spatial smoothing by convolution with an isotropic Gaussian kernel of 8 mm, removal of linear trends to correct for general signal drift and band-pass filtering (0.01–0.08 Hz) to reduce low-frequency noise (32, 33). The final set of nuisance covariates including the six head motion parameters, average signals from WM and CSF, and the time series averaged over the brain (34) were regressed. Moreover, we applied frame censoring to each group by using the DVARS (differentiated rms variance) measure (35) with a threshold of 0.5 % root mean square frame-to-frame intensity change (36). Therefore, the criteria removed 8.10 ± 2.79, 9.93 ± 3.16, and 4.36 ± 1.07% of frames per individual, respectively, in PTSD, TEC, and HC group. Subjects with < 195 frames should be excluded. There were no statistically significant differences in the amount of censored time points between groups. Lastly, de-noising was consequently conducted to improve “cosmetic” using a combination of strategies similar to previous studies (37–40).

Our method for computing lags between time series was according to previously published literature (19). Considering the temporal features of intrinsic brain activity in its latency structure (Supplementary Figure 1), we evaluated the lagged cross-covariance functions according to the following formula (19): (1)Cxixj(τ)= 1T∫xi(t+τ)×xj(t)dt  where τ represent the time lag. The value of τ where Cxixj (τ) exhibited an extremum defines the temporal lag between signals xi and xj. T is the interval of integration. BOLD time series are aperiodic (41). Accordingly, it would almost always generate a single, well-defined extremum when calculated by the cross-covariance functions, typically in the range ±1 s.

Voxels were defined by dividing the gray matter mask in atlas space into 6-mm isotropic cubic regions. Given the time series {x1(t), x2(t), …, xn(t)}, extracted from all voxels (n = 5,797 in current study), finding all τi, j corresponding to the extrema of Cxixj (τ) yields the antisymmetric, time-delay (TD) matrix: (2)TD=[τ1,1⋯τ1,n⋮⋱⋮−τ1,n ⋯τn,n] Then, group level latency projections were obtained by calculating the projections that computed as the mean across the columns of TD matrix at subject level and then averaging. All the results were represented in three-dimensional image formats using the BrainNet Viewer (http://www.nitrc.org/projects/bnv/).

Besides, we evaluate the voxel-wise whole-brain functional connectivity strength (FCS) to offer a reference for comparison with the results of RS-LA. Conventional FCS was defined by computing the average functional connectivity between a given voxel and all other voxels in the brain (42). The procedures of data preprocessing were almost the same as RS-LA, except that data smoothing was implemented after the Pearson correlation coefficients were converted using Fisher r-to-z transformation. All the processes were calculated by using DPARSFA. Voxels with high FCS (> mean) were identified as functional hubs, which indicated that they were highly connected to the rest of the brain.

SPSS 22.0 (SPSS INC, Chicago, IL, USA) was used to analyze the demographic and clinical data. The Chi-square test was applied to evaluate gender difference among three groups. Normally distributed material expressed as mean ± standard deviation was assessed by one-way analysis variance (ANOVA) and the homogeneity of variance in these data was examined by the Bartlett test. When the ANOVA analysis revealed significant differences, post-hoc analysis was employed for inter-group comparisons.

Due to the group differences in education level, RS-LA and FCS differences among the three groups were analyzed by using ANOVA with educational level as covariates, followed by post-hoc t-test to confirm the between-group differences. Regardless of the trauma effect, we also compared the group differences between PTSD patients and all control subjects with two-sample t test. The above analyses were respectively based on the Statistical non-Parametric Mapping software for RS-LA (SnPM13, http://warwick.ac.uk/snpm) and Statistical Parametric Mapping software for FCS (SPM8, http://www.fil.ion.ucl.ac.uk/spm/). All the maps were multiple compared and corrected with AlphaSim Program, with the threshold set at P < 0.01 and the voxel numbers of the cluster larger than 75, which corresponded to P < 0.05.

Finally, to investigate the association between latency structures and PTSD symptoms, the average latency values extracted from the clusters with significant lagged differences were correlated with the clinical measurements. Spearman correlation analysis with the significant level of P < 0.05 (two-tail test, Bonferroni corrected).

Ultimately, 27 drug-naïve patients with PTSD (48.41 ± 10.32 years; 7 males, 20 females), 33 TEC (48.45 ± 7.48 years, 7 males, 26 females), and 30 HC (49.87 ± 6.11 years, 7 males, 23 females) underwent data analysis. The detailed demographics and clinical characteristics are shown in Table 1. There were no group differences for age and gender distribution. However, the education level of HC group was higher than TEC and HC group. Besides, there were also significant differences of SAS and SDS among the three groups, which the scores in PTSD and TEC groups are much higher than HC group, and the scores of PTSD were also higher than TEC group.

Latency projections results obtained from the PTSD, TEC, and HC group are displayed in Figure 1. All group level latency maps span 1 s between the earliest and latest brain regions. It exhibited that time latency pattern of the early and late brain structures were highly symmetrical and spatially distinct in each group. Additionally, we could found significant group differences among the three groups.

The group differences among three groups are showed in Table 2 and Figure 2, The changes in time latency of PTSD patients were the prominent shift toward later in left superior temporal gyrus (STG), posterior cingulate cortex/precuneus (PCC/PCu), middle prefrontal cortex (mPFC), right angular, and left pre- and post-central cortices (Pre/PostCG). Most of these differences could be found when comparing the PTSD group with HC group, while only the STG was the exception as the result of comparison between PTSD and TEC group. Besides, we observed similar time latency in right angular in TEC group compared to HC group. There was no significant difference in brain regions with time latency shifting toward earlier among the two trauma groups.

The time latency values with significant group differences were extracted and displayed in the form of bar chart at the group level (Figure 3A). The results of correlations between clinical measurements and RS-LA values of all significant clusters in both PTSD and TEC were reported in the Supplementary Table 1. In PTSD group, our further analysis found negative correlation between the time latency values of mPFC and the PCL_C scores (r = −0.578, P = 0.002 < 0.05, Figure 3B), which is absence in the TEC group (r = 0.030, P = 0.868).

Conventional resting-state functional connectivity strength (RS-FCS) differences of the three groups are shown in Figure 4. In the PTSD group, visual inspection reveals aberrant brain hyper-activity in bilateral parahippocampus/hippocampus, as well as the hypo-activity in right middle and inferior frontal gyrus, PCC/PCu, middle cingulate cortex, bilateral cuneus, and left Pre/PostCG. It should be noted that most of these alterations could also be observed in the TEC group. Additionally, the noticeable brain regions of RS-LA and FCS comparison between PTSD patients and all controls did not show the specialness (Supplementary Figure 2).

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.