We recruited 36 patients with unilateral mTLE (14 left- and 22 right-side) as the epilepsy group who were consulting in the Neurology Clinic and Inpatient Department of The First Affiliated Hospital of Zhengzhou University and 36 healthy volunteers as the control group. The diagnosis was based on the detailed history, neurological examination, electroencephalography (EEG) recordings, and imaging findings. Patients (22 females, 29.08 ± 10.12 years of age) who fulfilled the following criteria were included: (1) described or observed clinical semiology consistent with seizures of mesial temporal lobe origin; (2) MRI results that suggested hippocampal sclerosis of the patients; (3) unilateral interictal and ictal epileptic discharges according to scalp EEG or intracranial electrode EEG; (4) all right-handed. Exclusion criteria for the patient group are as follows: (1) patients with brain structural abnormalities (except for hippocampal sclerosis), other systemic diseases, psychiatric disorder, history of alcohol abuse, etc.; (2) Un-identified lateralization of mTLE; (3) head motion during scanning is greater than 3 mm.

Inclusive criteria for the control group (23 females, 28.61 ± 9.72 years of age) are as follows: (1) normal head MRI results; (2) with at least primary school education; (3) all right-handed. Exclusion criteria: Patients with any prior history of neurological or psychiatric diseases. Handedness was rated according to the Edinburgh Handedness Inventory.

This study was approved by the research ethical committee of The First Affiliated Hospital of Zhengzhou University, and informed consent was obtained from all subjects.

Resting-state fMRI data were acquired using a 3 T Magnetom Prisma MRI scanner (Siemens Healthcare, Erlangen, Germany) with a 64-channel head coil. All subjects were told to lie down flat, not think, open their eyes, and breathe quietly. The scanning parameters were as follows: TR/TE, 1,000/30 ms, FOV, 220 × 220 mm², slice thickness, 2.2 mm, slice gap, 0.4 mm, flip angle, 70°, and voxel size, 2.0 × 2.0 × 2.2 mm³, with 52 slices and 400 dynamics.

The Data Processing and Analysis of Brain Imaging (DPABI) toolbox¹ was used to preprocess the statistics. The following steps were performed: (1) conversion of data formats; (2) removal of the first 10 time points; (3) slice-timing correction; (4) realignment; Participants were excluded if their maximal head motion exceeded 3 mm displacement or 3°of rotation. No subjects were excluded after that. (5) spatial normalization of fMRI images to the standard EPI template and resampled to 3 × 3 × 3 mm³ resolution; (6) detrending; (7) regression of the Friston-24 motion parameters (Satterthwaite et al., 2012), cerebrospinal fluid signal, white matter signal, and global-brain signal; (8) band-pass temporal filtering (0.01–0.08 Hz); (9) calculation of the mean frame-wise displacement (FD) of each subject to assess the head movement (Power et al., 2012; Zhang et al., 2017); There was no significant difference of mean FD between the mTLE and HC groups (p = 0.159 in a Mann–Whitney U test). (10) scrubbing with cubic spline interpolation to remove the “bad” time points and their 1-back and 2-after time points on the basis of FD threshold of 0.5 mm; (11) spatial smoothing with a 6 mm full-width-at-half-maximum Gaussian kernel to improve the signal-to-noise ratio (SNR).

According to the steps described previously (Watanabe et al., 2019; Wengler et al., 2020), the intrinsic timescale of each voxel was estimated using the preprocessed fMRI data at the level of a single participant.

First, for each voxel, an autocorrelation function (ACF) of fMRI signals [time bin = repetition time (TR)] was estimated. Next, we calculated the sum of ACF values during the initial positive period. This initial positive period begins with the first lag and ends at the time point just before the first timepoint with a non-positive autocorrelation coefficient. Then the sum of ACF values was multiplied by the TR of the fMRI data to adjust for differences in temporal resolution. The above steps were achieved through the custom MATLAB code publicly available at https://github.com/RaichleLab. Finally, this final brain map was used as an intrinsic neural timescale map.

For demographic and clinical characteristics, differences in age and gender between the group of mTLE and HC were analyzed with two-sample t test and Chi-square test, respectively (P < 0.05). Differences in age and mean FD among subgroup I, subgroup II, and HC were analyzed with one-way analysis of variance (ANOVA) and Kruskal–Wallis ANOVA, respectively (P < 0.05). Difference in gender among subgroup I, subgroup II, and HC was analyzed with Chi-square test (P < 0.05). Difference in lateralization between subgroup I and subgroup II was analyzed with Chi-square test (P < 0.05). Differences in age and illness duration between the subgroups of LTLE and RMTLE were analyzed with two-sample t test (P < 0.05). Differences in gender and mean FD between the subgroups of LTLE and RMTLE were analyzed with Chi-square test and Mann–Whitney U-test, respectively (P < 0.05).

First, two-tailed two-sample t test was performed on the timescale maps using SPM12 toolkit to determine regions showing significantly altered INT between groups of mTLE and HC. Multiple comparison correction was performed using a Gaussian random field (GRF) where threshold of voxel-wise p < 0.001 and cluster-level p < 0.05 [the cluster-level correction refers to familywise error rate (FEW)] in DPABI toolbox (we set cluster extent threshold at 30 voxels).

Secondly, we divided patients into two subgroups according the duration of the epilepsy (subgroup I: n = 17, epilepsy duration<10 years, subgroup II: n = 19, epilepsy duration ≥ 10 years). ANOVA was used to detect the differences among the three groups (subgroup I, subgroup II, and HC) (GRF corrected, P_voxel < 0.001, P_cluster < 0.05). The ANOVA result map was applied as mask to perform secondary analysis. Then pairwise between-group differences were investigated by two-sample t test (GRF corrected, P_voxel < 0.001, P_cluster < 0.05).

Thirdly, according to the side of the epileptogenic focus, patients were divided into two subgroups: the left mTLE group (L-mTLE) and the right mTLE group (R-mTLE). We used two-sample t test to compare INT maps between the two subgroups (GRF corrected, P_voxel < 0.001, P_cluster < 0.05).

In all above comparisons, age, gender and mean FD were treated as nuisance covariates.

To rule out the effect of head motion and the quality of the imaging data on our results, we calculated the correlation of mean FD and SNR (Welvaert and Rosseel, 2013) with the altered INT values in patients with mTLE. The altered INT was defined as the mean values of peak coordinates with a spherical radius of 6 mm in brain regions with significant between-group differences. The peak coordinates of the decreased INT were in Table 2. A statistically significant threshold of p < 0.05 was set for all correlation analyses.

There were no significant differences in age and gender between the group of mTLE and HC (p > 0.05) (Table 1). In subgroup analyses, there were no significant differences among the three groups (subgroup I, subgroup II, and HC) for age and gender (p > 0.05), and there was no statistical difference in side of epileptogenic focus between subgroup I and subgroup II (p > 0.05) (Supplementary Table 1). And there were no significant differences between the subgroups of LTLE and RTLE for age, gender, and illness duration (p > 0.05) (Supplementary Table 2).

The spatial distribution maps of INT for each group are shown in Figure 1. Longer timescales were highly localized in frontal and parietal cortices and shorter timescales in sensorimotor, visual, and auditory areas. Intrinsic timescales showed a similar whole-brain pattern of distribution in patients and HC groups.

The group comparison results of INTs revealed that patients with mTLE exhibited atypically shorter intrinsic timescales in the right inferior/middle orbital frontal gyrus (OFG), right insular, and right cerebellum crus1/crus2/6 (GRF corrected, P_voxel < 0.001, P_cluster < 0.05, voxel wise ≥ 30; Table 2 and Figure 2).

No significantly differences were detected in patients in subgroup I compared with HC or subgroup II. In subgroup II, the right inferior OFG and right cerebellum crus1/6 showed reduced INT compared with HC (GRF corrected, P_voxel < 0.001, P_cluster < 0.05, voxel wise ≥ 30; Figure 3 and Supplementary Table 3).

We observed no significantly differences between subgroups of L-mTLE and R-mTLE (GRF corrected, P_voxel < 0.001, P_cluster < 0.05).

No significant correlation was observed between either mean FD or SNR and the altered INT values in mTLE patients (for all correlation analyses, p > 0.05). In addition, there was no significant difference between patients with mTLE and HCs either in mean FD (p = 0.159 in a Mann–Whitney U test) or in SNR (p = 0.415 in a two-sample t test) (Table 1).

Human cerebral cortex have been proven to follow a hierarchical ordering in intrinsic neural timescales (Murray et al., 2014; Demirtaş et al., 2019), and subcortical areas such as striatum, thalamus, and cerebellum also topographically mirrored this intrinsic timescale gradients (Raut et al., 2020). Consistent with previous findings, we revealed similar hierarchical timescales that longer timescales in frontal and parietal cortices and shorter timescales in sensorimotor, visual, and auditory areas. Given that INT is obtained by calculating the magnitude of autocorrelation of neural signals, the shorter intrinsic timescale in patients might indicates greater randomness and variability of brain signals (Watanabe et al., 2019). In the seizure propagation network, transient abnormal electrical activity may randomly disrupt functional connectivity within the network or with others with a consequently higher variability of brain signals (Morgan et al., 2015). Previous studies have found that patients with temporal lobe epilepsy performed atypically large signal variability in the prefrontal region (Laufs et al., 2014; Robinson et al., 2017; Tong et al., 2019) and insula (Tong et al., 2019), which mostly overlapping the regions showing decreased INT in our studies. Moreover, this temporal property in local brain signals was proposed to be in connection with the local gray matter volumes (GMVs) (Watanabe et al., 2019). Therefore, our study is supported by previous research which has revealed decreased GMVs in frontal lobe areas, limbic structures, and cerebellum in mTLE (Bernasconi et al., 2004; Riederer et al., 2008; Santana et al., 2010).

With regard to the shorter INT observed in mTLE patients both in our mainly analysis and subgroup analyses, it is unclear whether this arises from pathologic mechanisms directly related transient abnormal electrical activity (Morgan et al., 2020) or secondary changes of hippocampal sclerosis (Shuman et al., 2020). By developing a biophysical model, Chaudhuri et al. (2015) proposed that the INT hierarchy is implemented by a gradient of recurrent excitatory connections (excitatory-to-excitatory synapse strengths) that is scale by the hierarchal position of each region. Further, several studies demonstrated that imbalance of excitation-inhibition (E/I) ratio in individual brain regions could possibly result in aberrance of neural timescales (Kiebel et al., 2008; Hasson et al., 2015; Wengler et al., 2020), specially, elevated E/I ratio could lead to relative increases in neural timescales (Wengler et al., 2020). A review (Trevelyan et al., 2013) previously proposed that epileptic discharges almost completely inhibits all neuronal activity and then prevent normal processing by enhanced inhibition in surrounding areas, leading to disruption of normal information processing not only in ictal stage but also in interictal stage. Therefore, we suggest that the increases in the strength of recurrent inhibition in peripheral areas outside the seizure onset zone may lead to decreases in E/I ratio and further explaining reduced brain intrinsic timescale in our results. Besides, longer INT in cortical areas means greater ability to integrate and hold information, that is to say, new information can be maintained for a long time (Chaudhuri et al., 2015). Several studies have shown that hippocampal circuits contribute to continuous processing when they are without damage (Hannula et al., 2007; Olsen et al., 2012). And hippocampus has been reported to play an important role in the retainment of episodic memories (Hasson et al., 2015). A study in epileptic mice also showed that desynchronization of interneuronal firing between the CA1 and dentate gyrus (hippocampal subregion) significantly impacts spatial processing (Shuman et al., 2020). However, whether hippocampus is a necessary component during this process remains unclear. For instance, stimulus information in hippocampal amnesia can be retained long enough to participate in a conversation (Hasson et al., 2015), suggesting that information retention can be independent of the hippocampus. In our study, atypical INT was found in patients with mTLE whose hippocampus was structurally (Bernasconi et al., 2004) and functionally (Laufs et al., 2014) impaired, which to some extent may support the important role of hippocampus in information retention and processing in cortical areas.

Previous researches implicated that dysfunction in temporal lobe epilepsy was related to disease duration (Robinson et al., 2017), which supporting mTLE associated with hippocampal sclerosis is a progressive disorder (Zhang et al., 2017). In addition, lots of previous studies have provided evidence that heterogeneity have identified for L- mTLE and R-mTLE (Ridley et al., 2015; Lee et al., 2018). For example, there are different patterns of functional connection (Lee et al., 2018) and different degree of GMV atrophy (Riederer et al., 2008) between the left and right mTLE. Whereas, we observed no statistically significant changes between patients with long- and short-term epilepsy duration or between L-mTLE and R-mTLE. This might be due to the lack of sufficient sample size in our study.

mTLE showed decreased INT values in the insula, an area that constitutes the core region of the salience network (SN) (Menon and Uddin, 2010). The SN plays a role in the transition between default mode network (DMN) and central executive network (CEN) (Zhou et al., 2018), and consequently subserves the mediation of information flow across its interconnected brain networks involved in attentional processing and cognition (Menon and Uddin, 2010). As such, we considered that neurodynamics changes in the insular observed in our study may affect between-network interactions and ultimately higher cognitive functions which heavily rely on coordinated activity of resting state networks (RSNs) (Burianová et al., 2017). Moreover, SN is involved in maintaining consciousness and awareness (Burianová et al., 2017). Based on this, altered timescales in the salience network might account for the reduction or loss of awareness during seizures (Luo et al., 2014). Besides, decreased INT was also observed in OFC. The insular cortex and OFC are all belong to the affective network (AN), playing a broad role in modulating affective information (Li et al., 2018). Patients with temporal lobe epilepsy has been reported to have attentional bias for negative emotional information (Lanteaume et al., 2012). Given this, we hypothesize that decreased INT in the insular and OFC found in the mTLE group may potentially be one of neural bases of emotional lability, higher anxiety level and depression, which are common comorbidities in epilepsy (Harden, 2002; Vazquez and Devinsky, 2003; Keezer et al., 2016).

Right posterior lobe of cerebellum also exhibited decreased INT values. The posterior cerebellum, such as crus I/II which show robust structural (Kelly and Strick, 2003) and functional (Krienen and Buckner, 2009) correspondence with the prefrontal cortex, is engaged in intellect, social cognition and emotion control (Schmahmann, 2019). Posterior cerebellar impairments have been observed in patients with mTLE (Riederer et al., 2008). The abnormal INT values detected in present study might imply the disruption of neuronal activities in prefrontal-cerebellar circuits underpins the higher-level cognitive and affective deficits following cerebellar damage.

In our study, almost all brain regions with INT changes were higher order regions. Cognitive processes, such as decision-making (Cavanagh et al., 2016) and memory processing (Hasson et al., 2015), unfold over time and must therefore rely on the long-term accumulation and integration of information in those areas (Murray et al., 2014; Chaudhuri et al., 2015). Hence, significantly shorter INT in higher-level areas might lead to inefficient integration of information and this may point to the underlying neuropathology of cognitive or emotional impairments in mTLE.

We also focus on the effect of head motion and SNR on our results to a certain extent. The relationship between either mean FD or SNR and the altered INT values in patients with mTLE was not significant and there was no significant difference between mTLE and HCs either in mean FD or in SNR. It proved that the effect of head movement and SNR can be avoided in our study to some extent.

There are several limitations in the current study. First, the sample size is relatively small and all participants come from a single dataset. The main reason is that our inclusion criteria is very strict, only those TLE patients with unilateral hippocampal seizure lesions were enrolled. And the recruitment of healthy volunteers was performed using the matched control pairwise approach to increase the statistical power of these analyses. Future study could use another independent dataset and increase sample size to confirm our findings. Second, since there were no clear staging criteria for epilepsy progression, we only chose epilepsy duration for dividing the patients into two groups in the subgroup analysis, which obviously cannot represent the occurrence of epilepsy comprehensively. Third, potential effects of antiepileptic drugs in TLE patients were not considered. Finally, we did not calculate the correlation analysis between degree of hippocampal sclerosis and INT shortening and we will improve it in future studies.