Participants were recruited from local community or universities by advertisements, and the initial sample included 67 datasets (32 males and 35 females; mean age = 32.79 ± 13.11; ranged from 18.59 to 64.30). All the participants underwent a detailed mental health interview by two trained psychologists using the Mini-International Neuro-Psychiatric Interview. People with a history of major neuropsychiatric illness, head injury, and alcohol and drug abuse were excluded. They were also assessed with Wechsler Adult Intelligence Scale-4th Edition (in Chinese, WAIS-IV), Schutte Self-Report Emotional Intelligence scale in Chinese Version (SSEIS), State-Trait Anxiety Inventory, Mental Health Continuum-Short Form, Emotion Regulation Questionnaire, Chinese Perceived Stress Scale, Achievement Motivation Scale, and Self-Control Scale. The institutional review board of Institute of Psychology Chinese Academy of Sciences approved this study, and written informed consent was obtained from individual participant prior to data acquisition.

The State-Trait Anxiety Inventory (STAI) (Spielberger et al., 1983) was applied to measure participants' state and trait anxiety. The scale has a total of 40 descriptive questions related to anxiety. Items 1–20 are the State Anxiety Scale (S-Al), 10 of which describe negative emotions and 10 of which describe positive emotions, which are used to assess the feelings of people at a specific and particular moment. Items 21–40 are the Trait Anxiety Inventory (T-AI), 11 of which describe negative emotions and 9 of which describe positive emotions, which are used to assess people's habitual anxiety experience and capture the dimensions of personality linked to anxiety. Each item uses a 4-point scoring method. The higher scores indicate the higher degree of anxiety. Previous studies have verified that the Chinese version of the STAI had acceptable construct validity that supported Spielberger's conception of the multidimensional nature of the S-AI and T-AI scales (Li and Lopez, 2004; Wei et al., 2015) and could be used to measure state and trait anxiety of Chinese participants. Cronbach's α coefficient for internal consistency in our sample was acceptable (state anxiety α = 0.757, trait anxiety α = 0.717).

MRI images were collected on the 3.0 T GE scanner Discovery MR750 at the Institute of Psychology Chinese Academy of Sciences. All the participants completed a T1-weighted structural MRI scan (eyes closed) with a ABI1_t1iso_fspgr sequence (TR = 6.652 ms; TE = 2.928 ms; FA = 12°; matrix = 256 × 256; slice thickness = 1 mm) and an 8-min resting-state fMRI scan (eyes open with a fixation cross) using a gradient echo EPI sequence ABI1_bold_bw_rest [TR = 2,000 ms; TE = 30 ms; FA = 90°; number of slices = 33 (interleaved); slice thickness = 3.5 mm; gap = 0.7 mm; and matrix = 64 × 64].

MRI images were preprocessed using the Connectome Computation System (http://github.com/zuoxinian/CCS) developed by our laboratory (Xu et al., 2015), which integrated several neuroimaging-related software including AFNI (Cox, 2012), FSL (Jenkinson et al., 2012), and FreeSurfer (Fischl, 2012), as well as in-house MATLAB scripts. The full pipeline of preprocessing included structural image preprocessing, functional image preprocessing, as well as quality control (Zuo et al., 2013; Xu et al., 2015). The structural preprocessing included (1) intensity inhomogeneity correction; (2) brain extraction; (3) tissue segmentation; (4) white and pial surface generation; and (5) deformation estimation between the resulting spherical mesh and a common spherical coordinate system. The functional preprocessing was the same as our previous publications (Jiang et al., 2015; Zhang et al., 2017) and included (1) excluding the first five volumes from each scan; (2) removing and interpolating of temporal spikes; (3) slice timing correction and motion correction; (4) extracting functional brain; (5) normalizing 4D global mean of image intensity; and (6) co-registration between functional and anatomical images by employing a boundary-based registration (BBR) algorithm.

Following the preprocessed individual MRI images, the CCS also provided a quality control procedure (QCP) for both functional and structural images. The QCP includes the following steps (Zuo et al., 2013): (1) brain extraction or skull stripping; (2) brain tissue segmentation; (3) pial and white surface reconstruction; (4) BBR-based functional image registration; and (5) head motion during resting-state MRI. The pipeline also computed the mean frame-wise displacement (meanFD) (Power et al., 2012) and the minimal cost of the BBR co-registration (mcBBR) for the subsequent statistical tests as covariates. All participants with bad brain extraction, bad tissue segmentation, and bad surface construction will be excluded from the subsequent analysis. One participant did not complete MRI scanning, and one participant did not pass the mental health interview. Five participants were excluded because their mcBBR was >0.65. Therefore, we had 60 participants for final group analysis. The detailed participant information and intervariable correlations are shown in Table 1.

Here, we used a macroscale brain network parcellation developed by Yeo et al. (2011) as nodes to construct human brain structural and functional network, which subdivided the entire cortical surface into 51 spatially connected parcels based on resting-state FC. We excluded the parcels whose vertex number was <50, and finally 30 parcels were reserved for final group analysis, which expanded across all the Yeo-7 networks, including visual network, somatomotor network, dorsal attention network, ventral attention network, limbic network, frontoparietal (control) network, and DMN (Table 2).

Network topology means the full details of network connections, and here we mainly considered two types of topological measurements, namely, network efficiency and centrality. In more detail, we applied graph theory (Achard et al., 2006) to compute network efficiency, including global efficiency (Eglob), nodal efficiency (Enodal), and local efficiency (Elocal), as well as network centrality, including degree centrality (DC), betweenness centrality (BC), eigenvector centrality (EC), and pagerank centrality (PC), and the specific calculation tools included the Brain Connectivity Toolbox (http://www.brain-connectivity-toolbox.net) (Rubinov and Sporns 2010) and the Connectome Computation System (http://github.com/zuoxinian/CCS) scripts (Xu et al., 2015).

To investigate topological mechanisms of human brain structural and functional networks underlying state and trait anxiety, we used general linear model to calculate the partial correlations between topological properties (E, efficiency; C, centrality; CV, coefficient of variation) and anxiety. In more detail, for structural network, we considered age, gender, education, intracranial volume (ICV), and total volume for volume network (total area for area network or mean thickness for thickness network) as covariates. The detailed statistical model is shown in the following equation:

For functional network, we considered age, gender, education, mcBBR, and meanFD as covariates. The detailed statistical model is shown in the following equation:

False discovery rate (FDR, q < 0.05) correction for 30 parcels was used to control type 1 error over multiple tests, and all the statistical analyses were performed using MATLAB scripts in this study.

We used morphological similarity network (bin number = 30) of cortical volume, surface area, cortical thickness, and FC network (time point = 235) to explore whether state anxiety and trait anxiety had common or distinct topological mechanisms. For functional network, we found state anxiety (r = 0.5509, corrected p = 0.0004) and trait anxiety (r = 0.4505, corrected p = 0.0168) were both significantly positively correlated with pagerank centrality in the RH_SalVentAttn_FrOper (Figures 2A,B), which mapped to the opercular part of the inferior frontal gyrus, and the inferior and superior of the insula (Destrieux et al., 2010).

Using different spatial scales during the construction of human brain morphological similarity network, we got spatial variations of human brain structural network. Using sliding-window method, we got temporal variations of human brain functional network. The detailed topological properties for different spatial scales and different time slots are illustrated in Supplementary Figure 1.

For human brain structural network at different spatial scales, we found that the CV of degree centrality was negatively correlated with trait anxiety in the LH_Default_PFC (corresponding to PFC) (r = −0.4163, corrected p = 0.0471), and CV of nodal efficiency was significantly negatively correlated with both state anxiety (r = −0.4175, corrected p = 0.0455) and trait anxiety (r = −0.4186, corrected p = 0.0441) in the LH_DorsAttn_FEF (corresponding to frontal eyes field) of volume network (Figure 3). There were no significant results in other morphological networks.

Using dynamic functional networks acquired from sliding window, we found both state and trait anxiety were positively correlated with median of pagerank centrality (state anxiety: r = 0.5121, corrected p = 0.0019; trait anxiety: r = 0.4674, corrected p = 0.0096) and median of degree centrality (state anxiety: r = 0.447, corrected p = 0.0187; trait anxiety: r = 0.4406, corrected p = 0.0228) in the RH_SalVentAttn_FrOper (Figure 4). We also found both state and trait anxiety were positively correlated with median of betweenness centrality (state anxiety: r = 0.3991, corrected p = 0.0381; trait anxiety: r = 0.4322, corrected p = 0.0295) and mode of degree centrality (state anxiety: r = 0.4289, corrected p = 0.0326; trait anxiety: r = 0.4308, corrected p = 0.0308) in the RH_SalVentAttn_FrOper (Supplementary Figure 3); trait anxiety was positively associated with median (r = 0.3991, corrected p = 0.0381) and mode (r = 0.4265, corrected p = 0.0325) of pagerank centrality in the LH_SalVentAttn_FrOper, median (r = 0.4104, corrected p = 0.0381) and FWHM (r = 0.4242, corrected p = 0.0360) of betweenness centrality in the LH_SalVentAttn_FrOper, and mode (r = 0.4909, corrected p = 0.0325) of pagerank centrality in the RH_SalVentAttn_FrOper (Supplementary Figure 2). Besides the above parcel-wise network, we also examined vertex-wise functional network and got similar results as shown in Supplementary Table 1. We also tried group comparisons between high-anxiety and low-anxiety groups, and found that both state and trait anxiety were associated with global efficiency of area similarity network (Supplementary Tables 2–8).

At the human brain functional level, we found that pagerank centrality of the anterior insula, which mostly occupied the right frontal operculum (RH_SalVentAttn_FrOper), was significantly positively correlated with state and trait anxiety in both static and dynamic functional networks. This was in line with previous studies indicating the activations of the insula in healthy participants' anxiety provocation tasks (Tian et al., 2016; Geng et al., 2018; Dammann et al., 2020). The insula has been proved to play an important role in subjective emotional processing and internal body awareness (Critchley et al., 2004; Craig, 2009; Singer et al., 2009; Tranel et al., 2009; Ernst et al., 2013), anticipation of future uncertain threat (Geng et al., 2018) and unpleasant negative stimuli (Herwig et al., 2007; Simmons et al., 2011; Lutz et al., 2013), risk and uncertainty in decision-making (Grinband et al., 2006; Clark et al., 2008), as well as visual and auditory awareness (Bushara et al., 2001; Kondo and Kashino, 2007). The positive correlations between topological properties of the insula and anxiety indicated that healthy participants with higher state and trait anxiety might be prone to pay more attention and awareness to internal body feelings and anticipate the risky and uncertain future events or threats due to the activations of the insula. Meanwhile, the role of the insula was also reflected in anxiety disorders, including social anxiety disorder (Duval et al., 2018; Atmaca et al., 2021) and generalized anxiety disorder (Shah et al., 2009; Cui et al., 2020), where they found more interoceptive body awareness, deficits in attentional control and emotion modulation, as well as hyperactive to general emotional images in patients compared to healthy controls. All these studies together demonstrated the connections between the insula and anxiety.

While at the human brain structural level, we found that spatial variations (CV) of nodal efficiency in the left frontal eyes field (FEF) was negatively correlated with state and trait anxiety, which reflected that decreased information transfer ability and efficiency of the FEF contributed to higher state and trait anxiety. The FEF, located in the superior precentral sulcus of the superior frontal sulcus corresponding to Brodmann's area 6 (Vernet et al., 2014), was not only involved in preparing and triggering various eye movements (Nyffeler et al., 2004; Nagel et al., 2008; Yang and Kapoula, 2011; Jaun-Frutiger et al., 2013), but also important in cognitive processes, including attentional orienting (Muggleton et al., 2003; Grosbras et al., 2005), visual awareness (Smith et al., 2005; Quentin et al., 2013), and perceptual modulation (Thompson et al., 1996). The negative correlations indicated participants with high state and trait anxiety had weakened ability and efficiency of attention control as well as perceptual modulation, and might show a tendency to increase more attention to unpleasant threats and stimuli, which was consistent with the attention control theory (Eysenck et al., 2007).

In this study, we conducted systematical studies to confirm that state and trait anxiety shared common brain network topological mechanisms. We found topological properties of the insula and the FEF were both correlated with state and trait anxiety, respectively, at human brain functional and structural level, which suggested that healthy participants who tended to feel anxious in daily life (trait anxiety) might share the same brain network topological patterns during anxious events and threats evocation (state anxiety). Previous studies have demonstrated that there was an interaction and association between state anxiety and trait anxiety (Mathews, 1990; Williams et al., 1996). More importantly, Takagi et al. (2018) have directly demonstrated that trait and state anxiety shared the same brain network and had a substantial biological interrelationship in a unified analytical framework. They hypothesized that participants with high trait anxiety would become increasingly anxious when facing unanticipated anxiety-related events and further enhance their levels of state anxiety. In contrast, participants with low trait anxiety would show a resistant and defensive response to the threats to reduce their levels of state anxiety. In more detail, the insula and the FEF were both involved in preliminary cognitive processing stage awareness (whether external or internal; visual or auditory), which indicated that state and trait anxiety shared the same brain topology mechanisms in alarming and sensing anxiety-related signals. Meanwhile, we also found that spatial variations (CV) of degree centrality in the PFC were only negatively correlated with trait anxiety but not state anxiety. The PFC generally acts as a role in emotion regulation and generation (Dixon et al., 2017), as well as integrates internal and external stimuli, to influence behavioral reactions (Bickart et al., 2012; Lindquist et al., 2012). We inferred that state and trait anxiety might use different human brain network topological mechanisms in the subsequent processes of coping with anxiety-related threats or events.

To the best of our knowledge, we are the first to not only explore the topological properties underlying state and trait anxiety using MRI technique but also verify that the common topological properties of human brain structural and functional networks underlying state and trait anxiety had temporospatial robustness. Human brain is one of the most complicated systems in the world, and it must coordinate complicated information at different spatial and temporal scales. Based on temporal correlations of BOLD time series, functional network could be used to detect human brain functional organization at different temporal scales (Friston, 2011; Allen et al., 2014; Calhoun Vince et al., 2014). Traditional FC method employed correlation coefficient of the time series of the entire scan across brain regions without capturing the temporal variations, while dynamic FC could quantify temporal alterations in FC metrics across multiple temporal scales (Hutchison et al., 2013b). Dynamic FC was proved to be a sensitive and specific marker of mental illness (Calhoun Vince et al., 2014; Yao et al., 2017). In more detail, previous studies have applied dynamic FC to various diseases such as generalized anxiety disorder (Yao et al., 2017; Chen et al., 2020), schizophrenia (Damaraju et al., 2014; Du et al., 2017), and bipolar disorder (Rashid et al., 2016; Nguyen et al., 2017). All these validated that dynamic FC was useful and efficient in characterizing human brain functional organization at different temporal scales. Corresponding to functional network, we also changed bin number (spatial scales to measure morphological distribution) and gained different morphological similarity resolutions to investigate spatial variations of human brain structural network underlying state and trait anxiety. Previous studies have demonstrated that morphological similarity might be the most accurate and robust method to reflect information transfer between brain regions (Seidlitz et al., 2018; Li et al., 2021), and different spatial resolutions of measuring morphological similarity could detect spatial robustness of human brain structural network underlying state and trait anxiety. This was the first study to examine spatial robustness of human brain networks and characterized the dependence of topological mechanisms of anxiety on spatial resolution. The above temporal and spatial robustness studies would enhance our understanding of temporal and spatial organization of human brain networks. In summary, our study systematically examined the topological mechanisms of state and trait anxiety from both structural and functional networks perspectives, as well as from their spatial and temporal variations perspectives. Our results indicated that state and trait anxiety shared common brain network topology mechanisms, and the common topological mechanisms exhibited high temporospatial robustness.

In conclusion, we explored topological properties underlying state and trait anxiety based on human brain structural and functional network, as well as their spatial and temporal variations. We found state and trait anxiety were both positively correlated with pagerank centrality in the right insula, median and mode of pagerank, and degree centrality in the right insula and negatively correlated with CV of nodal efficiency in the left frontal eyes field of volume network. Our results demonstrated that state and trait anxiety shared common brain network topological mechanisms with high temporospatial robustness and would enhance our understanding of spatial and temporal organization of human brain.

Several limitations should be taken into consideration for this study. First, our sample size was small but across a large age span, and we need to increase the number of participants in future studies. Second, our study only analyzed human brain information of cerebral cortex but did not involve subcortical data. Anxiety as a negative emotional state, subcortical information such as amygdala, hippocampus, and caudate should be considered in future explorations. Finally, we only conducted inter-regional correlations and intra-regional vertex-wise analysis, and whole-brain vertex-wise analysis required immense memory storage or efficient computational capacity.