Commonalities and distinctions of static and dynamic functional connectivity density between left and right temporal lobe epilepsy

Abstract

To comprehensively examine static and dynamic functional connectivity density (FCD) in left temporal lobe epilepsy (LTLE) and right temporal lobe epilepsy (RTLE). For 46 LTLE patients, 43 RTLE patients, and 53 healthy controls (HCs), T1-weighted structural images and resting-state functional images were collected. Static FCD and corresponding temporal dynamic FCD (dFCD) obtained via a sliding window approach were measured and compared. Additionally, seed-based functional connectivity (FC) analysis was executed. Relationships between cognitive scores and FCD or dFCD values were analyzed. Compared with HCs, LTLE and RTLE patients presented reduced static FCD values in the ipsilateral lateral temporal lobe. However, for LTLE, the FCD in the lateral temporal lobe contralateral to the epileptic focus also decreased. In RTLE patients, the FCD and dFCD in the left occipital lobe increased, whereas the dFCD in the left cerebellum decreased. FC analysis (seed: left middle temporal gyrus) demonstrated that LTLE patients had a wider range of aberrant brain regions than RTLE patients did. However, the left occipital lobe-based FC reductions were detected only in the RTLE patients. FCD and dFCD revealed moderate discrimination abilities between LTLE and RTLE. This study reveals that while both LTLE and RTLE share reduced static FCD in the ipsilateral temporal lobe, indicating common global network impairment, they exhibit distinct disruption profiles: LTLE presents more bilateral and widespread alterations, whereas RTLE specifically involves the occipital lobe and cerebellum. These findings consolidate TLE as a system-level disorder with lateralized network phenotypes. Integrating static and dynamic FCD provides a comprehensive framework for elucidating whole-brain network abnormalities and the associated pathophysiological and compensatory mechanisms in TLE.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-026-37646-0.

Introduction

Epilepsy is a prevalent neurological illness involving about 1% of the worldwide population¹. The most frequent etiology of focal epilepsy is temporal lobe epilepsy (TLE), but it is now recognized that TLE could affect widespread brain networks². Plenty of seed-based functional connectivity (FC) analyses have certified the disruption of functional connectivity among the hippocampus³, thalamus⁴, temporal lobe⁵, etc. Nevertheless, seed-based FC analyses may be limited by the selection of seed points and thus neglect important and unexpected predictable discoveries. To examine the neural functional arrangement in TLE patients, whole-brain FC studies are required.

As a voxel-based, data-driven graph theory approach, functional connectivity density (FCD) was first proposed by Tomasi and Volkow in 2010⁶. It is now used to quantify the significance of network nodes and helps in the exploration of both local and global network interactions. FCD represents the overall number of effective functional connections between the voxel and every other voxel in the whole brain; thus, an elevated FCD value of a specific voxel means that the compactness of its functional connections with the whole brain is increased. On the contrary, a decreased FCD value represents that a specific voxel lost its intrinsic functional compactness and network centrality. Furthermore, considering that static FCD is not sufficient to depict the highly dynamic fluctuating functional connectivity states^7,8, the dynamic FCD (dFCD) indicator, which is based on the widely used sliding window approach, has emerged in recent years⁹. The dFCD is a novel dynamic indicator derived from FCD that reflects the temporal dynamic variability of FCD by calculating the variance of the FCD value across all segmented time windows.

The FCD and dFCD methods have been conducted in psychiatric diseases¹⁰, idiopathic generalized epilepsy¹¹, benign epilepsy with centrotemporal spikes (BECTS)¹², etc. Li et al.¹² found out that, for BECTS, both the active and chronic epileptic states could induce alterations in dynamic functional connectivity, expressed as increased dFCD in the posterior default mode network (DMN) along with reduced dFCD in the orbital frontal cortex, anterior cingulate cortex, etc. A study of idiopathic generalized epilepsy demonstrated alterations in the FCD’s temporal variation distributed in the DMN and subcortical, sensorimotor, and high-level networks¹¹. These authors speculated that these alterations might be related to driven motion, cognitive deviations, or the generation and transmission of seizure activity. However, TLE cohorts have not received much attention¹³.

In addition, due to the asymmetry of the cerebral hemisphere, alteration in the functional network might be different between left temporal lobe epilepsy (LTLE) and right temporal lobe epilepsy (RTLE). Different organization patterns of anterior and posterior hippocampal networks between mesial LTLE and RTLE have been reported by Li et al.¹⁴. Some studies have depicted more remarkable impairments in functional connectivity and white matter connectivity in LTLE^15–18. Whereas, the study of Pang et al. revealed that static and dynamic functional connectivity in LTLE both showed no significant abnormal changes, whereas RTLE exhibited complex aberrant connectivity states¹⁹. Zhao et al. also reported more wide-ranging functional connectivity abnormalities in RTLE⁵. On the other hand, approximately similar change patterns of LTLE and RTLE have been reported in other studies. Zhao et al. studied regional homogeneity (ReHo) changes in TLE and depicted a generally similar increased/decreased pattern between LTLE and RTLE, which involves extensive temporal and extratemporal regions²⁰. In the study of Chang et al.²¹, they not only reported pronounced changes in superficial white matter and cortical thinning in TLE but also discovered that the alterations in LTLE and RTLE exhibited very similar patterns. Interestingly, the superficial white matter changes were significantly lateralized to the ipsilateral hemisphere, involving frontal and temporal regions. However, cortical thinning occurred bilaterally in both left and right TLE. The effect of lateralization needs to be further verified.

Considering the controversial results, further investigations of the impact of the laterality of the epileptogenic zone from various directions are meaningful. Utilizing the approaches of static FCD and dFCD could offer novel viewpoints and more comprehensive evidence to uncover the aberrance of brain function resulting from epileptic activity. Therefore, we hypothesized that there might be some similarities and differences between LTLE and RTLE revealed by FCD and dFCD.

This study aimed to (1) investigate the differences in FCD and dFCD among healthy controls (HCs), LTLE, and RTLE; (2) explore further FC changes on the basis of the above abnormal regions; (3) examine the possible relationships between cognitive scores and aberrant FCD and dFCD variability; and (4) evaluate whether abnormal indices could provide useful information for lateralization.

Materials and methods

Participants

This research was authorized by the Research Ethics Council of the First Affiliated Hospital of Zhengzhou University, and all methods were performed in accordance with the relevant guidelines and regulations of the First Affiliated Hospital of Zhengzhou University (Ethical review number: 2019-KY-232). The study was conducted in accordance with the Declaration of Helsinki. Informed consent was obtained from all participants or their guardians.

Forty-six LTLE patients (19 males, mean age: 29.02 ± 12.39 years) and 43 RTLE patients (24 males, mean age: 29.74 ± 9.91 years) from our hospital, 53 healthy volunteers (24 males, mean age: 27.02 ± 8.50 years) from the local communities were enrolled in this study.

The inclusion criteria for the TLE patients were as follows: ① right-handed; ② ≥14 years of age; and ③ received a unilateral focal TLE diagnosis by our multidisciplinary epilepsy patient management team (clinical semiology, electroencephalography, brain magnetic resonance imaging (MRI) and positron emission tomography-computed tomography (PET-CT) examinations were used to determine the diagnosis and lateralization). ④ MRI findings acquired using epilepsy protocol (including 3D T1WI and 3D T2-FLAIR sequences) were interpreted as non-lesional except for hippocampal sclerosis. The exclusion criteria were as follows: ① history of other systemic and neurologic/psychiatric illnesses (excluding epilepsy); ② abuse of addictive substances (alcohol, heroin, tobacco, etc.); ③ patients diagnosed as familial mesial TLE (with a family history of epilepsy or febrile seizures); ④ At the time of scanning, patients had been evaluated as seizure-free for at least 12 months; ⑤ unsatisfactory fMRI quality (maximum head motion ≥ 2.5 mm/2.5°); and ⑥ unconfirmed lateralization of TLE.

The inclusion criteria for the HC group were as follows: ①right-handed; ②≥14 years of age; ③no systemic or neurologic/psychiatric disorders; and ④no habituation of tobacco, alcohol, heroin, etc. The exclusion criteria were as follows: : ①abnormal cerebral MR findings; and ②unsatisfactory fMRI quality (maximum head motion ≥ 2.5 mm/2.5°).

MRI data acquisition

MRI data were collected on a 3 T scanner (MAGNETOM Prisma, Siemens Healthcare, Erlangen, Germany) via a 64-channel head coil. The participants were advised to lie still, keep eyes closed, and stop thinking. Resting-state functional MRI parameters: voxel size, 2.0 × 2.0 × 2.2 mm³; repetition time (TR)/echo time (TE), 1000 ms/30 ms; volume, 400; slice number, 52; slice thickness, 2.2 mm; slice gap, 0.4 mm; flip angle, 70°. 3D T1WI parameters: voxel size, 0.9 × 0.9 × 0.9 mm³; TR/TE/inversion time (TI), 2300 ms/2.3 ms/900 ms; and slice number, 176.

Neuropsychological tests

The auditory verbal learning test (AVLT) of the Mandarin Chinese version was selected to assess the cognition, as it is recently widely used in clinical settings to detect memory decline for its brevity and simple scoring. It includes six ratings: N1, N2, N3 (instantaneous recall), N4 (5-minute delayed recall), N5 (20-minute delayed recall), and reconfiguration. AVLT could evaluate different aspects of learning and memory, such as verbal learning capacity, immediate memory, long-term memory, and recognition memory.

A total of 42 healthy volunteers, 30 LTLE patients, and 27 RTLE patients have completed the AVLT test. The testing procedure is as follows: ① The participant was read a list of 12 words (three different types (flower, occupation, and apparel), with 4 words in each type) by a trained examiner with intervals of one second between every two words. Then the participants were asked to say all the words they can recall within one minute. This learning phase and recalling phase are repeated 3 times. The numbers of recalled words are recorded as the scores of AVLT-N1, AVLT-N2, and AVLT-N3, respectively. The N1-N3 scores represent for short-term memory. ② Then the participant takes a 5-minute nonverbal test and then recalls the 12 words again for the fourth time. The number is recorded as the AVLT-N4 score, which stands for short-delay recall. ③ Another 20-minute nonverbal task is then carried out, after which the participant recalls the words for the fifth time as the AVLT-N5 score, which represents long-delay recall. ④ Finally, the participant was read a list of 24 words (12 words of the first list plus 12 distractors) and was asked to make a yes/no response based on whether these words belong or not to the first list of words. Record the number of correct identifications as the reconfiguration score.

Data analysis

Rs-fMRI preprocessing

The Data Processing Assistant for Resting-State fMRI Analysis Toolkit (DPARSF) was utilized to preprocess the rs-fMRI data. Firstly, delete the first 10 time points. Secondly, perform slice-timing and realignment. Thirdly, the functional images were resampled to 3 × 3 × 3mm³ after being normalized to the standard EPI template. Then, spurious variances were regressed via multiple linear regression analyses, which included cerebrospinal fluid and white matter signals, linear regression factors, 24 head motion metrics, and time points with a mean frame-wise displacement (FD) greater than 0.2 mm. Lastly, the functional images were temporally band-pass filtered (0.01–0.1 Hz).

Calculation of static FCD and dynamic FCD

The FCD and dFCD analyses were both conducted via the Dynamic Brain Connectome (DynamicBC) toolbox (http://restfmri.net/forum/DynamicBC)²². As a data-driven, voxel-based approach, FCD was first proposed by Tomasi and Volkow in 2010 and is now used for measuring the significance of network nodes⁶. In the present study, the FCD of a specific voxel represents the total number of effective functional connections between that voxel and every other voxel in the whole brain. Firstly, the degree of connection between voxels was assessed via Pearson’s linear correlation, and then a functional connection matrix was gained when the computation was repeated across every voxel within the whole brain. To eliminate the weak correlations that may arise from noise, the connection between two voxels was regarded as effective if its correlation coefficient exceeded a specific threshold (r = 0.2, p < 0.001). Then, the number of effective connections in the matrix is counted as the voxel’s FCD.

Unlike the aforementioned static FCD, dFCD is a measure combined with the sliding window method, which dynamically intercepts signals via windows and computes the information captured in every window. The window length has been proposed to be an open yet key parameter for resting-state dynamics analysis on the basis of sliding window approach^23,24. The window length should not only be sufficiently wide to detect fluctuations but also small enough to capture rapidly shifting signals. Hence, a window length of 50 TRs along with a window overlap of 80% were selected in the present study²⁴. We can obtain a global FCD map within every window. Subsequently, the dFCD was defined as the variance of these FCD maps across the 35 sliding windows, which characterized the temporal variability.

The dFCD variability was analyzed again via an additional window length and overlap (80 TRs and 95%) to verify the reproducibility of the dFCD metrics and the robustness of the sliding-window analysis.

Calculation of seed-based RsFC

With the hypothesis that there might be differences among the LTLE, RTLE, and HC groups, we performed statistical group comparisons of the FCD and dFCD maps. Then, to provide additional insight into FCD anomalies, the following FC analyses were conducted. The detected abnormal clusters’ peak voxels were selected as the central point, and spherical regions of interest (ROIs) with a 6-mm radius were established as seed regions for the FC analyses, which were processed by the Data Processing and Analysis of Brain Imaging (DPABI) toolbox. Firstly, the averaged time course of each ROI was extracted. Then, Pearson correlation coefficients were computed between these values and those of each voxel within the gray matter.

Finally, with the purpose of improving the signal-to-noise ratio and lessening anatomical heterogeneity among participants, all FCD, dFCD, and FC maps were z-standardized beneath the gray matter mask and then smoothed (6 × 6 × 6 mm full width at half maximum).

Statistical analysis

The Kruskal-Wallis nonparametric test and least significant difference (LSD) post hoc analysis were employed to examine differences in cognition scores (p < 0.05).

Firstly, two-sample t-tests were performed on the FCD and dFCD maps to detect regions showing significant alteration between groups of TLE and HC. Multiple comparison correction was performed using a Gaussian random field (GRF) with a threshold of p_voxelwise < 0.005 and p_{cluster−wise} < 0.05.

Secondly, we performed analysis of covariance (ANCOVA) across the subgroups (LTLE, RTLE, and HC groups) to identify brain regions with significant FCD or dFCD differences. Regions surviving multiple comparison correction (GRF, p_voxelwise < 0.005, p_{cluster−wise} < 0.05) were further examined with post-hoc pairwise comparisons using Bonferroni correction for the number of comparisons. These post-hoc analyses were considered descriptive and aimed to characterize the pattern of group differences within ANCOVA-significant regions. The covariates included mean frame-wise displacement (FD), sex, age, and years of education.

For differences in FC maps across groups, ANCOVA and post hoc LSD pairwise comparisons were employed, with mean FD, age, sex, and education years as covariates. Multiple comparison correction was performed using GRF with a threshold of p_voxelwise < 0.005 and p_{cluster−wise} < 0.05.

We subsequently extracted the values of FCD and dFCD within the ROIs (spherical, 6-mm radius, peak voxel of the aberrant clusters as the central point) and evaluated the correlations between the values and AVLT scores via Spearman correlation analysis. To control for multiple comparisons across all pairwise correlations, we applied the False Discovery Rate (FDR) correction with q < 0.05.

To investigate the possibilities of FCD and dFCD in discriminating LTLE from RTLE, receiver operating characteristic (ROC) curve analysis was performed. First, the data were randomly divided into an exploration set (80%, n = 113 (LTLE = 37, RTLE = 34, HC = 42)) and a validation set (20%, n = 29 (LTLE = 9, RTLE = 9, HC = 11)). In the exploration set, group comparisons were conducted using the same method described above to identify abnormal brain regions between LTLE and RTLE. Then, ROC analysis of the FCD or dFCD values extracted from the ROIs (spherical, 6-mm radius, peak voxel of the aberrant clusters as the central point) was performed on the independent validation set.

Results

Neuropsychological results

Compared to the HC group, cognitive deficits were found in both the LTLE and RTLE groups (p < 0.05). All AVLT scores decreased in LTLE, whereas only AVLT-N1, AVLT-N4, and AVLT-N5 scores decreased in RTLE. However, no significant difference in AVLT was found between LTLE and RTLE (p > 0.05). (Table 1)

Table 1.

Clinical characteristics and neuropsychological test data.

	LTLE (P-value)	RTLE (P-value)	HC	F-value (P-value)	df
AVLT - N1	3.80 ± 1.42* (< 0.001)	4.19 ± 1.90* (0.005)	5.36 ± 1.59	8.946 (< 0.001)	2
AVLT - N2	5.97 ± 1.99* (0.014)	6.48 ± 1.72 (0.169)	7.17 ± 2.17	3.221 (0.044)	2
AVLT - N3	7.00 ± 2.15* (0.002)	8.15 ± 2.48 (0.338)	8.67 ± 2.00	5.173 (0.007)	2
AVLT - N4	5.50 ± 3.24* (< 0.001)	6.00 ± 2.63* (0.004)	7.86 ± 1.93	8.464 (< 0.001)	2
AVLT - N5	4.47 ± 3.34* (0.001)	5.04 ± 2.59* (0.009)	6.88 ± 2.52	7.324 (0.001)	2
AVLT - Reconfiguration	19.53 ± 3.21* (< 0.001)	21.44 ± 1.89 (0.138)	22.40 ± 2.51	10.732 (< 0.001)	2
Age (years)	29.02 ± 12.39	29.74 ± 9.91	27.02 ± 8.50	/	/
Sex (M/F)	19/27	24/19	24/29	/	/
Hippocampal sclerosis (with/without)	19/27	19/24	/	/	/
Auditory syndromes (with/without)	3/43	0/43	/	/	/
Familial mTLE	None	None	/	/	/

Note: Data represent mean±standard deviation; ^* indicates that the difference in data between the patient group and HC group was statistically significant. HC, healthy control; LTLE, left temporal lobe epilepsy; RTLE, right temporal lobe epilepsy; AVLT, auditory verbal learning test; df, degree of freedom; mTLE, mesial temporal lobe epilepsy.

FCD and dFCD analyses

Alterations of FCD and dFCD between TLE and HC groups

Compared with the HC group, the FCD in the right lateral temporal lobe decreased in the TLE group. However, the dFCD revealed no significant statistical difference (GRF corrected, p_voxelwise < 0.005, p_{cluster−wise} < 0.05). (Table 2; Fig. 1).

Table 2.

Differences in FCD and dFCD between TLE and HC groups.

Measurements	Brain regions	MNI coordinates	Cluster size (Voxel)	Peak t-value
		x y z
FCD	STG_R, MTG_R, ITG_R	48 −21 −3	211	−4.252
dFCD	-	-	-	-

R, right; FCD, functional connectivity density; dFCD, dynamic functional connectivity density; HC, healthy control; TLE, temporal lobe epilepsy; STG, superior temporal gyrus; MTG, middle temporal gyrus; ITG, inferior temporal gyrus.

Fig. 1.

Brain regions with significant changes in terms of static FCD between TLE and HC groups. Colder colors represent lower FCD values in TLE.

Alterations of FCD and dFCD among LTLE, RTLE and HC groups

Firstly, the whole-brain ANOVA result of FCD revealed significant group differences in three regions: the left lateral temporal lobe, the right lateral temporal lobe, and the left occipital lobe (GRF corrected, p_voxelwise < 0.005, p_{cluster−wise} < 0.05). (Table 3; Fig. 2).

Table 3.

Brain regions showing significant group differences.

Measurements	groups	Brain regions	MNI coordinates	Cluster size (Voxel)	Peak F-value
			x y z
FCD	LTLE vs. RTLE vs. HC	STG_L, ITG_L, MTG_L	−45 3 −18	55	9.926
		STG_R, MTG_R	54 −18 −6	103	10.820
		CAL_L, SOG_L	−15 −72 27	62	10.081
dFCD	LTLE vs. RTLE vs. HC	Cerebellum_L	−36 −72 −48	75	10.923
		CAL_L, SOG_L	−15 −72 27	52	11.733

L, left; R, right; FCD, functional connectivity density; dFCD, dynamic functional connectivity density; HC, healthy control; LTLE, left temporal lobe epilepsy; RTLE, right temporal lobe epilepsy; STG, superior temporal gyrus; MTG, middle temporal gyrus; ITG, inferior temporal gyrus; CAL, calcarine cortex; SOG, superior occipital gyrus.

Fig. 2.

Brain regions with significant changes in terms of static FCD among LTLE, RTLE, and HC groups.

The whole-brain ANOVA result of dFCD revealed significant group differences in two regions: the left cerebellum and the left occipital lobe (GRF corrected, p_voxelwise < 0.005, p_{cluster−wise} < 0.05). (Table 3; Fig. 3).

Fig. 3.

Brain regions with significant changes in terms of dFCD among LTLE, RTLE, and HC groups.

Subsequently, within these significant regions, we performed post-hoc pairwise comparisons (LTLE vs. HC, RTLE vs. HC, LTLE vs. RTLE) with Bonferroni correction.

The results of FCD (Bonferroni corrected, α = 0.05/9 = 0.0056) showed that, in the left lateral temporal lobe, the LTLE had significantly lower FCD values than HCs and RTLE, while RTLE did not differ from HCs. In addition, LTLE and RTLE both had lower FCD values in the right temporal lobe than HCs. In the occipital lobe, the RTLE had higher FCD values than HCs and LTLE, while LTLE did not differ from HCs. (Table 4)

Table 4.

Post-hoc pairwise comparisons within ANOVA-significant regions.

Measurements	ANOVA-significant region	Group comparison	Significance	Peak t-value
FCD	left lateral temporal lobe	LTLE vs. HC	Yes	−4.688
		RTLE vs. HC	No	/
		LTLE vs. RTLE	Yes	−3.861
	right lateral temporal lobe	LTLE vs. HC	Yes	−4.579
		RTLE vs. HC	Yes	−4.491
		LTLE vs. RTLE	No	/
	left occipital lobe	LTLE vs. HC	No	/
		RTLE vs. HC	Yes	3.470
		LTLE vs. RTLE	Yes	−4.071
dFCD	left cerebellum	LTLE vs. HC	No	/
		RTLE vs. HC	Yes	−3.685
		LTLE vs. RTLE	Yes	4.261
	left occipital lobe	LTLE vs. HC	No	/
		RTLE vs. HC	Yes	3.553
		LTLE vs. RTLE	Yes	−4.531

The results of dFCD (Bonferroni corrected, α = 0.05/6 = 0.0083) showed that, RTLE had higher dFCD values in the occipital lobe and lower dFCD values in the cerebellum than HCs and LTLE. (Table 4)

Moreover, the validation analysis results with another window length and overlap (80 TRs and 95%) remained similar to the main findings (50 TRs and 80%).

Seed-based RsFC analyses

After statistical analysis of the FCD and dFCD maps, we found that the significantly abnormal brain regions among the LTLE, RTLE, and HC groups included the bilateral lateral temporal lobe, left occipital lobe, and left cerebellum. Thus, to give a further explanation of hypo-/hyper- brain function connectivity in patients with TLE, rsFC analysis was conducted with the aforementioned brain regions as seeds. The pairwise group comparisons results were as follows (GRF corrected, p_voxelwise < 0.005, p_{cluster−wise} < 0.05) :

When the left lateral temporal lobe (spherical, 6-mm radius, MNI (−45, 3, −18) as the central point) was selected as a seed, the results demonstrated that in contrast with the healthy controls, the LTLE patients had decreased FC in the following areas: the bilateral lateral temporal lobe, bilateral frontal lobe, bilateral precuneus and left occipital lobe. Meanwhile, LTLE revealed elevated FC in the bilateral supramarginal gyrus, insula, superior frontal gyrus, supplementary motor area, and left cerebellum. The RTLE group presented decreased FC in the bilateral lateral temporal lobe, frontal lobe, cerebellum and left angular and increased FC in the bilateral superior frontal gyrus, supplementary motor area, precentral gyrus, postcentral gyrus, and precuneus compared to HCs. Moreover, LTLE revealed higher connectivity in the bilateral thalamus, basal nuclei, and insula than RTLE. (GRF corrected, p_voxelwise < 0.005, p_{cluster−wise} < 0.05) (Table 5; Fig. 4).

Table 5.

Significant differences in RsFC among LTLE, RTLE and HC groups.

Seed	Groups	Brain regions	MNI coordinates	Cluster size (Voxel)	Peak t-value
			x y z
Left lateral temporal lobe	LTLE vs. HC	MSFG_B, SFG_B, SMA_B, MTG_L, Rectus_B	−39 24 −33	1309	−7.651
		MTG_R, ITG_R, STG_R, OIFG_R	36 24 −33	429	−4.792
		MTG_L	−48 −27 −6	77	−4.701
		MTG_R	42 −33 −3	88	−4.586
		IOG_L, MOG_L, Fusiform_L, Lingual_L, CAL_L	−27 −90 −6	154	−3.582
		Angular_R, MTG_R, STG_R	54 −57 27	130	−4.612
		Angular_L, MTG_R, STG_R	−48 −57 27	122	−4.553
		Precuneus_B	−9 −51 33	116	−4.512
		SMG_L, STG_L	−60 −30 39	129	4.012
		SMG_R, STG_R	54 −30 30	129	4.304
		Insula_L, ROL_L	−33 −21 −3	232	5.726
		Insula_R, Putamen_R	36 0 15	187	4.957
		SFG_R, MCG_B, SMA_R	21 −6 57	347	4.779
		SFG_L, SMA_L	−18 −9 54	72	3.718
		Cerebellum_L	−33 −42 −51	80	4.206
	RTLE vs. HC	MTG_L, STG_L, OIFG_L	−36 24 −30	168	−7.595
		MTG_R, STG_R, ITG_R, OIFG_L	45 29 −26	510	−6.973
		Cerebellum_L	−24 −78 −39	121	−3.650
		Cerebellum_R	9 −75 −27	67	−3.728
		MTG_L	−48 −27 −6	44	−4.036
		MTG_R, STG_R	48 −27 −6	53	−3.540
		OIFG_L, TIFG_L	−45 30 −3	89	−3.870
		Angular_L	−54 −54 39	96	−3.431
		OIFG_R, PreCG_R	48 9 30	42	4.034
		Precuneus_L, SPG_L, SOG_L	−12 −60 54	296	4.223
		PoCG_L	−30 −39 60	105	3.493
		PoCG_R, SPG_R, Precuneus_R	18 −66 57	509	3.962
		SFG_L, PreCG_L	−21 −9 48	126	3.745
		SFG_R, PreCG_R	27 −9 63	162	4.107
		SMA_B	−6 −15 54	57	3.515
	LTLE vs. RTLE	Thalamus_B, Putamen_B, Pallidum_B, Insula_R	12 −12 9	393	4.799
Right lateral temporal lobe	LTLE vs. HC	STG_L, MTG_L, ITG_L	−39 18 −27	152	−3.955
	RTLE vs. HC	OIFG_L, TIFG_L	−42 33 −6	111	−3.814
	LTLE vs. RTLE	-	-	-	-
Left occipital lobe	LTLE vs. HC	-	-	-	-
	RTLE vs. HC	Cerebellum_B	0 −75 −36	291	−4.497
		MTG_R, STG_R	39 24 −33	182	−5.713
	LTLE vs. RTLE	Cerebellum_B, Vermis	0 −75 −36	619	5.577
		PreCG_R, PoCG_R, SMA_R	6 −18 69	455	−3.965
		PreCG_L, PoCG_L, PCL_L	−33 −15 66	252	−4.640
Left cerebellum	LTLE vs. HC	SMG_R, PoCG_R, ROL_R	60 −12 24	149	4.002
		PreCG_R, MFG_R	45 0 57	122	4.927
	RTLE vs. HC	SMG_R, PoCG_R, ROL_R	66 −24 33	233	4.139
	LTLE vs. RTLE	Hippocampus_R, PHG_R, MTG_R, Amygdala_R, Pallidum_R	18 −6 −24	144	−4.363

L, left; R, right; B, bilateral; HC, healthy control; LTLE, left temporal lobe epilepsy; RTLE, right temporal lobe epilepsy; ACG, anterior cingulate gyrus; CAL, calcarine cortex; HES, Heschl’s gyrus; IOG, inferior occipital gyrus; ITG, inferior temporal gyrus; MCG, median cingulate gyrus; MFG, middle frontal gyrus; MOG, middle occipital gyrus; MSFG, medial superior frontal gyrus; MTG, middle temporal gyrus; OIFG, orbital inferior frontal gyrus; OMFG, orbital middle frontal gyrus; PCL, paracentral lobule; PHG, parahippocampal gyrus; PoCG, postcentral gyrus; PreCG, precentral gyrus; ROL, Rolandic operculum; SFG, superior frontal gyrus; SMA, supplementary motor area; SMG, superior marginal gyrus; SOG, superior occipital gyrus; SPG, superior parietal gyrus; STG, superior temporal gyrus; TIFG, triangular inferior frontal gyrus.

Fig. 4.

Brain regions with significant changes between the TLE and HC groups in terms of FC (seed: left lateral temporal lobe). A: LTLE vs. HC; B: RTLE vs. HC; C: LTLE vs. RTLE. Colder (warmer) colors represent lower (higher) indicator values.

When the right lateral temporal lobe (MNI (54, −18, −6) as the central point) was selected as a seed, decreased FC in the left temporal neocortex in the LTLE and decreased FC in the left inferior frontal lobe in the RTLE were demonstrated compared with those in HCs. (GRF corrected, p_voxelwise < 0.005, p_{cluster−wise} < 0.05) (Table 5; Fig. 5).

Fig. 5.

Brain regions with significant changes between the TLE and HC groups in terms of FC (seed: right lateral temporal lobe). A: LTLE vs. HC; B: RTLE vs. HC. Colder (warmer) colors represent lower (higher) indicator values.

When the left occipital lobe (MNI (−15, −72, 27) as the central point) was selected as a seed, the RTLE revealed decreased FC in the bilateral cerebellum and right lateral temporal lobe compared with that in HCs. FC increased in the bilateral cerebellum but decreased in the bilateral precentral gyrus and postcentral gyrus in LTLE compared with RTLE. (GRF corrected, p_voxelwise < 0.005, p_{cluster−wise} < 0.05) (Table 5; Fig. 6).

Fig. 6.

Brain regions with significant changes between the TLE and HC groups in terms of FC (seed: left occipital lobe). A: RTLE vs. HC; B: LTLE vs. RTLE. Colder (warmer) colors represent lower (higher) indicator values.

When the left cerebellum (MNI (−36, −72, −48) as the central point) was selected as a seed, the LTLE and RTLE both revealed decreased FC in the right superior marginal gyrus, postcentral gyrus, and Rolandic operculum. In contrast with RTLE, LTLE showed decreased FC in the right mesial temporal lobe. (GRF corrected, p_voxelwise < 0.005, p_{cluster−wise} < 0.05) (Table 5; Fig. 7).

Fig. 7.

Brain regions with significant changes between the TLE and HC groups in terms of FC (seed: left cerebellum). A: RTLE vs. HC; B: LTLE vs. RTLE. Colder (warmer) colors represent lower (higher) indicator values.

Correlation analyses

We extracted FCD and dFCD values from four statistically significant clusters (the left lateral temporal lobe, the right lateral temporal lobe, the left occipital lobe, and the left cerebellum) and performed correlation analyses with the AVLT scores (N1, N2, N3, N4, N5, and reconfiguration). However, none of the correlations reached statistical significance after FDR correction with q < 0.05. See Supplementary Table S1 for the full set of tested correlations.

ROC analyses

The ANOVA and post hoc pairwise comparisons result of the exploration set showed that, the FCD values of LTLE decreased in the left temporal lobe while elevated in the bilateral cerebellum compared to RTLE. Meanwhile, LTLE revealed lower dFCD values in the left occipital lobe and higher dFCD values in the bilateral cerebellum.

Then, the indicators’ values of the independent validation set were extracted from the ROIs (spherical, 6-mm radius, peak voxel of the aberrant clusters as the central point) and used in the ROC analysis. The results demonstrated that the area under the curve (AUC) of FCD in the left temporal lobe and dFCD in the left occipital lobe were higher than 0.7. However, the statistical significance was not achieved in the current small sample (p>0.05). Detailed performance metrics are shown in Table 6; Fig. 8.

Table 6.

Summary of ROC analysis Performance.

Test variable	AUC	significance	95% CI
FCD of left temporal lobe	0.765	0.058	0.524 ~ 1.000
FCD of left cerebullum	0.432	0.627	0.151 ~ 0.713
FCD of right cerebullum	0.605	0.453	0.331 ~ 0.879
dFCD of left cerebullum	0.617	0.402	0.345 ~ 0.889
dFCD of right cerebullum	0.543	0.757	0.265 ~ 0.821
dFCD of left occipital lobe	0.753	0.070	0.516 ~ 0.990

Fig. 8.

ROC curves for differentiating LTLE from RTLE in the validation set.

Discussion

In the present study, we comprehensively investigated the changes in FCD, ranging from static to dynamic, in LTLE and RTLE patients compared with HCs. We further conducted seed-based FC analysis to provide a more thorough explanation of the abnormalities in brain network connections. As hypothesized, we identified both convergent and divergent patterns of functional connectivity disruption between LTLE and RTLE, providing novel insights into the neural mechanisms underlying these clinically distinct TLE subtypes. The main findings are: (1) a shared core deficit characterized by reduced static FCD in the ipsilateral temporal neocortex in both LTLE and RTLE groups; (2) a more extensive and bilateral disruption in LTLE, involving the contralateral temporal lobe and broader extratemporal regions; and (3) a unique pattern in RTLE, featuring specific static FCD or dFCD alterations in the occipital lobe and cerebellum. These results not only affirm TLE as a disorder of large-scale brain networks but also crucially highlight that the lateralization of the epileptogenic zone dictates distinct profiles of network compromise, which may underlie their differential cognitive and clinical presentations.

Convergent abnormalities: ipsilateral Temporal neocortex dysfunction

In the present study, both LTLE and RTLE patients exhibited reduced static FCD in the ipsilateral temporal neocortex. The temporal lobe neocortex, also known as the lateral temporal lobe, mainly includes the superior, middle, and inferior temporal gyrus, along with the temporal pole. The middle and inferior temporal gyrus, which belong to the auxiliary visual area, are responsible for the processing of visual information and play important roles in language processing²⁵. Also, the superior temporal gyrus consists of important functional areas such as the auditory cortex and the Wernicke zone. In general, the lateral temporal lobe is not only tightly connected to auditory and vision but also with higher-order cognition and mental activities, for example, verbal memory and emotion²⁶. Therefore, its functional connective status has a profound influence on the cognition of epilepsy patients. As the most common comorbidity in TLE patients, cognitive impairment might seriously affect quality of life²⁷.

Over the years, converging researchers have noted the functional abnormalities of the temporal neocortex and their correlation with cognitive deficits in unilateral TLE patients^28–30. In line with it, the present study identified a concurrent pattern of significantly lower auditory verbal learning test scores, reduced FCD in the temporal neocortex, and widespread FC reductions in frontal and temporal lobes. This triad of findings suggests an association between learning or memory impairment and large-scale network dysfunction in TLE, where functional disruptions in temporal regions may collectively contribute to cognitive decline.

However, previous studies have predominantly employed seed-based FC analyses to investigate the dysfunction of the temporal lobe. Such approaches are inherently limited by their dependence on a priori ROI selection and their inability to capture whole-brain connectivity patterns. To overcome these constraints, the present study utilized the FCD method—a data-driven method that could quantify voxel-wise connectivity across the entire brain without requiring predefined hypotheses⁶. This approach enables a more comprehensive and unbiased examination of network alterations in TLE. Given these methodological advantages, the temporal lobe abnormalities identified in our study likely represent more robust and reliable findings.

FCD can be simply defined as the amount of functional connections between a specific voxel and every other voxel of the entire brain. A reduction of FCD in the temporal neocortex indicates a disruption of its functional connectivity to the whole brain and a diminishing of its core status in the brain network. Thus, on the basis of the implications embodied in FCD, we prudently speculated that the network hypoconnectivity or centrality decline of the lateral temporal lobe is probably one of the key characteristics of TLE, which may be a potential reason for cognitive impairment.

Moreover, we observed lateralized dysfunction in the temporal neocortex for both LTLE and RTLE groups, with abnormalities predominantly ipsilateral to the epileptic focus. A meta-analysis of Jiang et al. suggested that the reduced FC in temporal structures might be related to the epileptic network, which is responsible for the generation and spread of discharges³¹. Previous single photon emission computed tomography (SPECT) studies related to TLE also verified the importance of the temporal lobe in seizure propagation and disturbance of consciousness^32,33. Complementary to this view, the “surround inhibition hypothesis” assumed that the areas around the epileptogenic focus may act as an adaptive mechanism to cut down seizure propagation^32,34. Thus, given that initial epileptic activity is derived from the temporal lobe in TLE, we supposed that functional impairment and adaptive adjustment might first be concentrated on the ipsilateral side. Previous researchers have proposed that recurrent interictal epileptic activity, which may be responsible for lasting cognitive disturbance, is the most likely explanation for anterior temporal lobe aberrance in TLE patients³⁵. Although seizure discharges are transient, the impairment of brain function induced by prolonged recurrent discharge is cumulative. Therefore, the present result of aberrant functional connectivity decline in the ipsilateral temporal neocortex probably is an indication of prolonged malfunction linked to prolonged epileptic processes.

It is worth mentioning that nearly half of the patients in our TLE group were diagnosed as TLE with hippocampal sclerosis (HS). But our primary findings were concentrated in the lateral temporal lobe, with no abnormal changes observed in the mesial temporal lobe. We speculated that it may be attributed to the heterogeneity of the patient cohort, which might have diluted the between-group statistical power. Another consideration is that the sensitivity and orientation of different resting-state fMRI metrics vary. Unlike amplitude of low-frequency fluctuation (ALFF), which reflects local intensity of neural activity and is more sensitive to abnormalities in regions like the hippocampus, the FCD may be more indicative of chronic, long-term network alterations. Pathological increases in connectivity, potentially originating from the mesial temporal lobe, might be more transient and state-dependent, becoming more prominent during periods with interictal epileptiform discharges (IEDs) or seizures. So regarding the above considerations, it is meaningful to achieve an electroencephalogram-functional magnetic resonance imaging (EEG-fMRI) study with a more homogeneous cohort in the future.

Divergent abnormalities: more widespread disruption in LTLE, occipital and cerebellar changes in RTLE

While the aforementioned convergent patterns are noteworthy, more illuminating were the divergent network alterations we observed between the two TLE subtypes. Firstly, for LTLE, the reduction in FCD in the temporal neocortex was bilateral, and the left temporal neocortex-based FC analysis revealed a more pronounced and widely involved alteration of functional connectives, and the left temporal neocortex-based FC analysis revealed a more pronounced and widely involved alteration of functional connectives. Whereas RTLE showed a relatively more confined ipsilateral pattern with FCD reduction. This asymmetry may be mechanistically linked to the dominant hemisphere language network³⁶. The left temporal neocortex, especially in classic language regions, is a critical node within a highly integrated, bilateral language network. Epileptiform activity originating in this strategically located hub may therefore propagate more efficiently along these pre-existing, high-traffic pathways, leading to contralateral functional disturbances³⁶. Since all the patients enrolled were right-handed and their dominant hemisphere was on the left hemisphere, we prudently speculated that LTLE patients with epileptic foci on the left hemisphere would probably suffer more severe functional disruption caused by epileptic processes. Our findings provide partial support for this view. While both LTLE and RTLE groups exhibited significant cognitive decline compared to controls, the LTLE group showed deficits in more cognitive indices.

Consistent with the present findings, some studies have depicted more remarkable impairments in functional connectivity and white matter connectivity in LTLE^15–18,37. However, the impact of brain hemispheric asymmetry on TLE and the distinctions between LTLE and RTLE are controversial. In the research of Zhao et al.⁵, more wide-ranging functional connectivity abnormalities in RTLE were reported, which seems to diverge from our findings. Whereas, a closer examination of their detailed results reveals a convergence with our results. They corroborated the ipsilateral temporal lobe as a core hub of network abnormalities in both groups and further demonstrated that FC reductions in LTLE were predominantly bilateral, whereas in RTLE they were more unilateral. On the other hand, approximately similar change patterns of LTLE and RTLE have been reported in other studies^20,21. In the study of Chang et al.²¹, the superficial white matter changes in both LTLE and RTLE were focused on the ipsilateral hemisphere, whereas cortical thinning was bilaterally distributed. In the present study, there are also discrepancies between the findings of FCD and FC. For RTLE, although the FCD did not reduce in the left temporal neocortex, the FC analysis based on the left temporal neocortex revealed abnormal connectivities, and the distributions are kind of similar to the LTLE. These variable results might be due to the distinct analytical indices, whose implications are different. However, the LTLE group exhibited more pronounced and extensive FC abnormalities, which still indicated the presence of a distinct difference between the two groups. So we prudently speculated that it underscores that network alterations in TLE involve both distinct asymmetric features and shared pathological processes. The convergence and divergence between FCD and seed-based FC results collectively paint a more complete picture. The effect of lateralization needs to be further verified and summarized.

In addition, a pattern unique to the RTLE group emerged in our analysis, which was represented as additional elevations in both FCD and dFCD within the left occipital cortex. Furthermore, our use of dFCD successfully captured decreased temporal variation in the cerebellum in RTLE, a finding missed by static analyses, highlighting the added value of temporal metrics in uncovering state-specific network pathologies. Our results are further strengthened by the subsequent observation that alterations in occipital-based functional connectivity were exclusive to the RTLE group. We therefore prudently speculate that the left occipital lobe is a key region that differs between LTLE and RTLE. This speculation is cautiously informed by preliminary ROC analysis, which revealed a moderate, albeit statistically non-significant (p > 0.05), discriminative capacity (AUC > 0.70) for dFCD in the left occipital lobe within the current small sample of the validation set. However, further investigation and validation with larger samples are needed.

As part of the visual cortex and the visual attention network, the calcarine cortex within the occipital lobe performs key functions in the processing of vision information and attention, as well as long-term memory³⁸. The existence of visual impairment, abnormality in the visual attention network, and its effect on TLE patients’ cognitive function has been verified^39–41. The dFCD is a novel dynamic indicator based on the widely used sliding window approach, which reflects the temporal dynamic variability by calculating the variance of the FCD value across all segmented time windows⁹. It represents the variation in cerebral collaboration over a short period of time, which, aimed at depicting instability, could further delineate abundant dynamic features⁴². Hyperexcitability, hypersynchronism, and instability are considered characteristics of epilepsy⁴³.The dFCD could represent the capability of information integrating, so the increase in FCD and dFCD in the occipital lobe suggested hyper-connectivity and hyper-variability stations during interactions with other parts of the brain. A study by Voets et al. suggested that pathological hyperconnectivity and hypoconnectivity both reflected dysfunction⁴⁴. The hyper-connectivity has been supported by prior studies. Zhang et al.³⁹ reported increased functional coherence in the primary visual cortex in TLE, and they cautiously interpreted this aberrance as a hyperfunction caused by epilepsy activity. Similarly, Shi et al.⁴¹ discovered enhanced functional connectivity in the lingual gyrus in TLE, which may correlate with visual aura and memory deficits. But contrary to the interpretation of Zhang et al.³⁹, they proposed that such enhanced functional coordination may serve to compensate for broader brain network dysfunction. In addition, accumulating evidence has indicated the reasonableness of enhanced functional connectivity in TLE patients, which may be caused by repeated seizures or inappropriate reorganization to compensate for functional disconnection elsewhere^37,45. Thus, the enhancive and overactive connectivity of the visual cortex in the present study might be interpreted as an unstable functional compensatory reorganization after impairment.

Beyond the occipital lobe, cerebellar abnormality is another important feature of RTLE. The cerebellum is a key brain region related to motor regulation and cognitive processes since it is widely interconnected with cerebral structures⁴⁶. There is evidence that the cerebellum and the hippocampus have bidirectional functional connections, which indicates that the cerebellum can affect the activity of the hippocampus and vice versa⁴⁷. Dysfunction of the cerebellum has been verified from functional and structural aspects^48,49, and our previous study revealed reductions in the degree and variability of the cerebellum’s functional connectivity related to RTLE⁵⁰. Therefore, the observed reduction of dFCD variability in the cerebellum might point to a diminished capacity for flexible network integration, ultimately reflecting a functional abnormality within the cerebellocerebral circuitry. However, the complex functional changes in TLE induced by lateralization are significantly more sophisticated than we currently comprehend. Further exploration is needed.

Limitation

This study has several limitations. Firstly, the sample size was not large enough, and only some of the patients completed the cognitive assessment, which may have led to sampling bias and limited the statistical power. Meanwhile, the TLE cohort was clinically heterogeneous, comprising TLE with and without hippocampal sclerosis. Although we excluded patients with macroscopic structural lesions, the underlying etiologies might influence network dysfunction patterns. Therefore, to distinguish network changes unique to TLE subtypes, larger sample sizes and phenotyped cohorts are required in future research. Secondly, the connection between interictal discharge states and brain function was not explicitly investigated in our study. Such correlations may be clarified by future EEG-fMRI studies, such as investigating whether aberrant activation in medial temporal lobe regions occur during interictal epileptiform discharges. Thirdly, in this study, only ROC analysis was conducted to assess the discrimination ability of the dFCD. A subsequent study based on a larger cohort and classifiers is necessary. Fourthly, it is important to note that the voxel-wise analysis, despite using stringent GRF correction, inherently involves a trade-off between controlling type I errors (false positives) and type II errors (false negatives). Thus, independent replication in larger cohorts is necessary to validate and expand our findings, possibly using complementary statistical techniques or multimodal imaging.

Conclusion

In conclusion, this study demonstrates that LTLE and RTLE are characterized by both overlapping and divergent patterns of functional network disruption. The principal convergent finding across LTLE and RTLE was a significant reduction in static FCD within the ipsilateral temporal neocortex, which indicates a common impairment in global network communication originating from the epileptogenic hemisphere. In contrast, the divergent patterns revealed two distinct profiles: bilateral and more widespread network disruption in LTLE and altered connectivity specifically involving the occipital lobe and cerebellum in RTLE. These findings advance our understanding of TLE from a focal disorder to a system-level condition with lateralized network phenotypes. The synergistic use of static and dynamic FCD could offer novel viewpoints and more comprehensive evidence to uncover the aberrance of the whole-brain functional network from epileptic activity, which helps deepen our understanding of the pathophysiological network impairment and adaptive compensatory mechanism in TLE.

Supplementary Information

Below is the link to the electronic supplementary material.

Author contributions

Chengru Song: Writing - original draft, Conceptualization, Formal analysis, Data curation. Xiaonan Zhang: Formal analysis, Data curation. Keran Ma: Data curation. Yajun Lian: Data curation. Jingliang Cheng: Project administration. Yong Zhang: Writing - review & editing, Project administration.

Funding

This research did not receive any specific grant from funding agencies in the public, commercial, or not-for-profit sectors.

Data availability

The datasets presented in this article are not readily available because of the privacy of all the participants. The data that support the findings of this study are available from the corresponding author on reasonable request.

Declarations

Competing interests

The authors declare no competing interests.