Ventral anterior thalamic dysfunction distinguishes seizure generalization in temporal lobe epilepsy

Abstract

Objective

Focal‐to‐bilateral tonic–clonic seizures (FBTCS) in temporal lobe epilepsy (TLE) involve thalamocortical networks, yet the functional integrity and role of specific thalamic subregions in seizure generalization remain unclear. In this cross‐sectional study, we investigated whether thalamic subregion functional connectivity patterns distinguish TLE patients with and without FBTCS and tested for laterality‐specific differences between right TLE (RTLE) and left TLE (LTLE.

Methods

We analyzed resting‐state functional magnetic resonance imaging (fMRI) data from 166 patients with TLE (71 RTLE, 95 LTLE; 120 FBTCS+, and 46 FBTCS−) and 119 healthy participants. Thalamic parcellation identified eight bilateral regions of interest. We computed graph theory measures of regional segregation, connection density, and integration, as well as intrinsic connectivity contrast to characterize thalamic subregion network topology. Sensitivity analyses controlled for pathology subgroup and anti‐seizure medication burden.

Results

TLE patients exhibited reduced cross‐hemispheric functional connectivity between bilateral ventral anterior (VA) thalamic regions compared with controls. Compared with patients with LTLE, patients with RTLE showed more pronounced ipsilateral VA hypoconnectivity and more disrupted nodal topology (clustering coefficient, degree centrality, local efficiency) in the ipsilateral VAia. Within RTLE, FBTCS+ patients (vs FBTCS−) showed ipsilateral connectivity reductions and cross‐hemispheric VA reductions that increased with illness duration. Within LTLE, FBTCS+ patients (vs FBTCS−) showed bilateral connectivity increases in the ventral posterior medial subregion. Sensitivity analyses confirmed that these findings were robust across pathology subtypes and independent of anti‐seizure medication burden.

Significance

These findings provide initial evidence that intra‐thalamic functional connectivity, particularly within the bilateral VA complex, is associated with FBTCS history in TLE and varies with lateralization of seizureonset. The observations may represent a biomarker for FBTCS, but whether this is causal or an epiphenomenon of this seizure type remains to be determined.

Key points.

• Bilateral ventral anterior (VA) hypoconnectivity is the dominant thalamic finding in temporal lobe epilepsy (TLE), more pronounced in right TLE (RTLE) than in left TLE (LTLE).

• Focal‐to‐bilateral tonic–clonic seizure (FBTCS) history is associated with reduced cross‐hemispheric VA coupling in RTLE and is related to duration of epilepsy.

• Laterality of seizure onset relates to thalamic connectivity patterns and should be considered in network studies of TLE.

• VA hypoconnectivity is consistent across pathology subtypes and independent of anti‐seizure medication burden.

1. INTRODUCTION

Focal‐to‐bilateral tonic–clonic seizures (FBTCS) represent the most dangerous seizure subtype in temporal lobe epilepsy (TLE), carrying elevated risk for sudden unexpected death in epilepsy (SUDEP), cognitive dysfunction, and treatment resistance. ¹ , ² , ³ Medically refractory seizures occur in 30%–40% of patients with TLE, ¹ and those who experience FBTCS have mortality rates 2–3‐fold higher than patients without FBTCS. ³ The neural mechanisms driving seizure generalization remain poorly understood, with most evidence implicating thalamocortical circuit dysfunction. ⁴ , ⁵ TLE patients with FBTCS exhibit widespread structural ⁶ , ⁷ , ⁸ and functional alterations, ⁴ , ⁸ , ⁹ , ¹⁰ , ¹¹ consistent with the concept that they suffer from a network disorder with extensive extra‐temporal involvement. ⁵ The thalamus appears central to FBTCS, as bilateral regional thalamic atrophy is the most common extratemporal structural abnormality in TLE ¹² , ¹³ and thalamic involvement during early seizure propagation implicates a critical role in secondary generalization. ¹⁴

Despite substantial evidence linking the thalamus to FBTCS, a knowledge gap exists regarding functional connectivity (FC) abnormalities between thalamic subregions. It remains unclear which specific intra‐thalamic communication changes are associated with FBTCS. The thalamus receives output from basal ganglia and cerebellum as part of thalamocortical loops implicated in the inhibitory/excitatory imbalances driving FBTCS motor symptoms. ⁹ , ¹⁵ To address this gap, we: (1) tested for specific intrinsic FC abnormalities within the thalamus, (2) examined the association of any such abnormalities with the presence or absence of FBTCS, and (3) determined whether FC is related to laterality of the epileptogenic zone. Consistent with the broader brain impairments reported for FBTCS, ⁴ we hypothesized that TLE patients with FBTCS would exhibit greater FC abnormalities in thalamic nuclei ipsilateral to the seizure‐onset zone (SOZ), as well as signs of abnormal FC between the right and left thalami compared to patients without FBTCS. As intrinsic FC differences have been reported for right and left focal TLE, ¹⁶ we tested for lateralized differences in thalamic connectivity in association with FBTCS status.

By stratifying our analysis by SOZ laterality and FBTCS status, our goal was to isolate regional‐specific derangements that could elucidate which thalamic subregions best distinguish FBTCS status. Given prior evidence that pathology heterogeneity and class‐imbalanced designs may influence connectivity analyses, ⁴ we performed sensitivity analyses stratifying by temporal lobe pathology and testing robustness to FBTCS group imbalance. Experimental animal studies have identified key thalamic nuclei for seizure modulation, and current TLE neuromodulation strategies in humans (e.g., deep brain stimulation, ¹⁷ responsive neurostimulation, ¹⁸ and focused ultrasound ¹⁹ ) could be refined by identifying which thalamic subregions are most associated with specific clinical presentations, potentially enabling personalized targeting approaches. ¹⁷ , ²⁰

2. METHODS

This cross‐sectional observational study adhered to STROBE (Strengthening the Reporting of Observational Studies in Epidemiology) guidelines for reporting.

2.1. Human subjects

A cohort of 166 patients with refractory unilateral TLE (LTLE = 95, RTLE = 71), comprising 120 with and 46 without a lifetime history of FBTCS, was recruited between 2007 and 2019, from the Jefferson Comprehensive Epilepsy Center, together with 119 age‐ and sex‐matched healthy participants (HCs). The Jefferson Institutional Review Board approved the study, and all participants provided written informed consent. All eligible consenting patients were included; systematic non‐participation data were not collected in this consecutive cohort design. SOZ lateralization was determined through convergent evidence from clinical history and semiology, continuous scalp video–electroencephalography (EEG) monitoring, structural magnetic resonance imaging (MRI), and ¹⁸ F‐Fluorodeoxyglucose positron emission tomography (FDG‐PET), and intracranial video‐EEG when clinically indicated (see diagnostic‐yield percentages in Supplemental Methods §S1.1). No patients met criteria for temporal‐plus epilepsy. Temporal lobe pathology was classified from MRI radiology and (when available) neuropathology records into four prespecified categories: mesial temporal sclerosis (MTS), gliosis (non‐MTS), other lesional, and non‐MTS/other; MTS was present in 68 of 143 patients with available records (47.6%). Pathology distribution by laterality and FBTCS status is shown in Table 1. Sensitivity analyses confirming consistency across pathology subgroups are in §3.7 and Table S5.

TABLE 1.

Demographic, clinical, and pathology characteristics of study participants.

	TLE Total	RTLE	LTLE	FBTCS+	FBTCS−	HC
	(n = 166)	(n = 71)	(n = 95)	(n = 120)	(n = 46)	(n = 119)
Sample demographics
Age at scan, years, M (SD)	41.9 (13.8)	41.5 (13.5)	42.2 (14.1)	42.1 (13.5)	41.6 (14.6)	38.1 (13.2)
Sex, male, n (%)	85 (51.2)	35 (49.2)	50 (53.2)	62 (51.5)	23 (50.0)	51 (42.9)
Sex, female, n (%)	81 (48.8)	36 (50.8)	45 (46.8)	58 (48.5)	23 (50.0)	68 (57.1)
Education, years, M (SD)	14.5 (2.4)	14.5 (2.3)	14.5 (2.5)	14.4 (2.4)	14.7 (2.4)	16.25 (2.7)
Epilepsy Characteristics
Seizure‐onset age, years, M (SD)	25.1 (15.0)	26.4 (14.4)	24.2 (15.4)	24.4 (13.4)	27.0 (18.5)	—
Seizure duration, years, M (SD)	16.8 (15.4)	15.1 (12.9)	18.0 (16.9)	17.7 (15.3)	14.7 (15.6)	—
Number of ASMs, M (SD) *	1.6 (.5)	1.5 (.5)	1.7 (.7)	1.6 (.7)	1.5 (.8)	—
FBTCS status, n
FBTCS+ patients	120	53	67	120	—	—
Current (<1 year) ***	75	35	40	75	—	—
Remote (>1 year) ***	44	17	27	44	—	—
FBTCS– patients	46	18	28	—	46	—
Seizure frequency, sz/mo, median [IQR]	4.00 [1.00, 8.00]	4.00 [2.00, 10.00]	4.00 [1.00, 8.00]	3.00 [1.00, 8.66]	4.33 [2.00, 8.00]	—
Temporal lobe pathology, n *
MTS, hippocampal sclerosis	68	23	45	53	15	—
Gliosis, non‐MTS	38	22	16	27	11	—
Other Lesional**	9	3	6	2	7	—
Non‐MTS/Other	28	11	17	21	7	—

Abbreviations: ASM, anti‐seizure medication; EMR, electronic medical record; FBTCS, focal‐to‐bilateral tonic–clonic seizures [− without; + with]; FIA, focal impaired awareness; HC, healthy control; IQR, interquartile range; LTLE, left TLE; MTS, mesial temporal sclerosis; RTLE, right TLE; TLE, temporal lobe epilepsy.

^*Reflects = 143 patients with available EMR data (86.1% of the full 166 cohort). For categorical variables, 𝜒2 tests were carried out. Education (HC > TLE; t = 4.21, p < .001). No significant differences between RTLE and LTLE for any variable (all p's > .27). Seizure frequency is computed across all seizure types per patient‐month and is reported as the within‐group median with interquartile range. Mann–Whitney U test for FBTCS+ vs FBTCS– on seizure frequency: U = 2315, p = .251 (n.s.). No other significant differences between FBTCS+ and FBTCS− for demographics/clinical variables (all p's > .29).

^**Other Lesional includes cavernous hemangioma, tumors, encephalocele, cortical dysplasia.

^***One subject (1 RTLE FBTCS+) could not be identified as current or remote.

2.2. Imaging methods

All participants underwent anatomic T1‐weighted and resting‐state functional MRI (fMRI) imaging on a Philips Achieva 3.0 T scanner. During the 5‐min resting‐state scan participants viewed a fixation crosshair without task requirements. ⁹ , ¹⁰ , ¹¹ , ²¹ , ²² , ²³ The fMRI sequence used was a single‐shot echo‐planar imaging (EPI) sequence with three discarded initial volumes for signal stabilization; T1‐weighted images were acquired with an MPRAGE sequence (1 mm³ in‐plane, 180 slices). DICOM‐to‐NIFTI conversion used dcm2niix; structural and blood oxygen level–dependent (BOLD) data were preprocessed in fMRIPrep v20.2.0 ²⁴ following standard best practices, with subsequent pipeline steps summarized in Figure S1A and detailed acquisition parameters provided in Supplemental Methods §S1.2–S1.3. Slice‐timing correction was applied as in prior work. ²² , ²³ Preprocessed images were imported into the CONN toolbox (v22a; RRID:SCR_009550) for connectivity analyses. A functional‐based parcellation defined 16 bilateral thalamic regions of interest (ROIs; eight per hemisphere) ²⁵ from the Tian Scale‐IV atlas: ventral anterior inferior‐anterior (VAia), ventral anterior inferior‐posterior (VAip), ventral anterior superior (VAs), ventral posterior medial (VPm), ventral posterior lateral (VPl), dorsal anterior medial (DAm), dorsal anterior lateral (DAl), and dorsal posterior (DP). (Detailed anatomic descriptions are in Supplemental Methods §S1.4).

2.2.1. Analytic scope

In line with our focus on intra‐thalamic connectivity, we restricted our analyses to 16 thalamic ROIs from the Tian Scale‐IV parcellation, which is a resting‐state functional connectivity‐based parcellation scheme. This provided the subnuclear scheme compatible with our 2 × 2 × 4 mm voxel size. We should note that a fixed 16‐ROI dependent vector preserved power for the FBTCS+/− contrast given our sample sizes. Cortico‐thalamic and basal‐ganglia–thalamic extensions are not within the scope of this analysis. Detailed VA subnuclear volumes and the effective point‐spread function of the acquired voxel +6 mm full width at half maximum (FWHM) smoothing are reported in Supplemental Methods §S1.4.1. Briefly, the VA complex (VAia + VAip combined ≈2704 mm³ per hemisphere) and was well above the effective PSF of approximately 6.3 × 6.3 × 7.2 mm FWHM, whereas individual VA subregions approach the PSF Therefore, we interpreted VA‐complex findings as primary and Vaia vs VAip distinctions as resolution limited. The 6 mm FWHM kernel was selected a priori for three reasons: (1) it matched the laboratory's previously validated resting state fMRI pipeline and was held constant across patients and controls to avoid analytic flexibility; (2) it sits at the lower bound of the 2 to 3 times voxel‐dimension range commonly applied to subcortical structures; and (3) it satisfied the spatial‐correlation assumptions required for Gaussian Random Field inference at the ROI level. We acknowledge that a smaller kernel could in principle improve subnuclear specificity, and we therefore report VA‐complex findings as primary and individual VAia vs VAip contrasts as resolution‐limited supportive findings (Supplemental Methods §S1.4.1).

2.3. Statistical analyses

2.3.1. Functional network connectivity

We first examined pairwise regional relationships using hierarchical clustering that optimized both Euclidean distance between ROI centroids and functional similarity metrics. We applied the functional network connectivity procedure (FNC ²⁶ ) to examine correlations between all ROI time‐series pairs using specific between‐group contrasts: (1) TLE patients vs controls, (2) seizure laterality effects (left vs right TLE), and (3) FBTCS status comparisons. We identified significant subnetwork differences with connection thresholds at p < .001 (uncorrected) and cluster thresholds at p < .05 (false discovery rate (FDR)–corrected). Post hoc pairwise tests determined which ROI pairs showed the strongest connectivity within each subnetwork.

2.3.2. Thalamic nodal topology analysis

For each ROI, three key nodal graph theory measures were estimated following implementations within the CONN toolbox (v22a) ²⁷ : clustering coefficient (CC), a measure of local segregation reflecting the proportion of interconnected neighbors around a node; degree centrality (DC), a measure of connection density reflecting the total number of direct connections a region maintains; and local efficiency (LE), a measure of regional integration reflecting information transfer efficiency at the local level. ¹⁰ We then estimated regional cross‐hemispheric interactions using a voxel‐based intrinsic connectivity contrast (ICC) representing a measure of node centrality at each voxel averaged for each ROI. ²⁸ To assess abnormal inter‐hemispheric interactions, ICC deviations were computed by subtracting each patient's regional ICC from the corresponding healthy control group mean, with z‐score transformation applied. Findings are reported using metric abbreviations (CC, DC, LE, ICC, ICCD) as primary descriptors. Interpretive labels (segregation, connection density, integration) are provided parenthetically on first mention. Further detailed description for features and analysis are provided in the Supplemental Methods.

2.3.3. Repeated‐measures multivariate analysis of variance (rmMANOVA)

For each metric, the eight thalamic parcels per hemisphere served as the dependent variables; thalamic region and hemisphere were within‐subject factors and group was the between‐subject factor. Five rmMANOVA models were run: (1) TLE vs controls, (2) LTLE vs RTLE, (3) FBTCS+ vs FBTCS−, and within (4) LTLE and (5) RTLE; full model specifications are in Supplemental Methods §S2. Effect sizes are reported as partial η², significance was set at p < .05, and analyses were performed in SPSS v29.0.2.0. We utilized a hierarchical analysis plan comprising three steps. (1) A primary rmMANOVA contrast tested for TLE abnormality by testing for a region by group interaction for (TLE vs HC). (2) Secondary contrasts compared subgroups: RTLE vs HC, LTLE vs HC, and FBTCS+ vs. FBTCS‐ overall and analyzed within the RTLE and LTLE subgroups. Four parallel nodal‐topology metrics (CC, DC, LE, and ICC/ICCD) served as dependent repeated measures. Within each rmMANOVA, post hoc pairwise tests used Bonferroni correction across the 16 thalamic ROIs. Across all nodal‐topology metrics (families), family‐level interaction p‐values were controlled with Benjamini–Hochberg FDR (q = .05). (3) Sensitivity analyses tested for robustness to other factors, namely pathology stratification and MTS‐only subgroup (Table S5), class imbalance (Table S6), anti‐seizure medication (ASM) covariate (Table S7), and FBTCS‐chronicity stratification (Table S4). Included 5000‐iteration permutation tests of the FBTCS label to control for Type I error (Table S8). A complete contrast‐by‐correction map is provided in Table S9. Permutation testing (5000 iterations) confirmed that observed effects were not artifacts of class imbalance. A pre‐specified 500‐iteration split‐half cross‐validation, stratified on TLE laterality and FBTCS status, assessed the robustness of the primary VA findings (Table S10). Secondary FBTCS‐laterality contrasts, however, were exploratory and should be interpreted as hypothesis‐generating in light of the FBTCS+/− class imbalance.

3. RESULTS

3.1. Demographical and clinical characteristics

Each metric (FNC, CC, DC, LE, ICC, and ICCD) was evaluated separately, with findings reported holistically addressing our hypotheses: (1) TLE vs controls; (2) RTLE vs LTLE (SOZ laterality‐specific vulnerability); and (3) FBTCS+ vs FBTCS− within laterality subgroups. We further assessed robustness of cross‐hemispheric abnormalities, clinical associations, and potential confounding of pathology, ASM, or sample imbalance.

Table 1 displays sample demographic, clinical, pathology, and seizure characteristics. No significant differences were observed between TLE patients and controls for age or sex (all p's > .05). TLE patients demonstrated lower education levels compared to controls (p < .001). Among 143 patients with available electronic medical records (EMRs), temporal lobe pathology comprised MTS (47.6%), gliosis (26.6%), other lesional (6.3%), and non‐MTS/other (19.6%). Pathology distribution did not differ by laterality (χ ²  = 6.17, p = .104) but differed between FBTCS groups (χ ²  = 12.40, p = .006), driven by over‐representation of other lesional pathology in the FBTCS− group. No other demographic or clinical variables differed between RTLE/LTLE or FBTCS+/FBTCS− groups (all p's > .27). Among FBTCS+ patients, median seizure frequency was .08 seizures per year. (interquartile range IQR\ .00–.41). The mean number of ASMs (1.6 ± .7) did not differ between FBTCS groups (p = .98; see Table S7).

3.2. Ventral anterior thalamic sub‐network findings

Hierarchical clustering of the 16 thalamic ROIs based on spatial proximity and functional similarity yielded a six‐cluster solution (Figure 1B). Of these, only the bilateral VA subnetwork (comprising VAia and VAip connectivity) exhibited significant between‐group hypoconnectivity after FDR correction in TLE patients compared to HCs (F = 7.74, p‐FDR = .003; Figure 1). This translated to an average asymmetric reduction in the nodal connectivity of each ventral anterior region (left VAia = 15.5%; right VAia = 35.7%), with effect sizes ranging from medium to large (η²p = .022–.070). Relative to controls, this significant hypoconnected subnetwork was more pronounced in right (F = 7.92, p = .003) compared to left TLE patients (F = 4.35, p = .08). Direct FNC comparisons revealed no significant FBTCS+/− group differences either in TLE patients as a whole or within right or left TLE groups (p > .25).

FIGURE 1.

Thalamic FNC reveals a single subnetwork containing hypoconnected bilateral ventral anterior complex. (A) Thalamic functional parcellation map in MNI‐space ²⁵ (L: Left (light gray); R: Right (darker gray); D‐V: Dorsal‐ventral). (B) Six candidate clusters emerged from hierarchical clustering of 16 thalamic ROIs based on anatomic proximity and functional similarity; only the bilateral VA subnetwork (highlighted) survived FDR‐corrected between‐group testing (connection threshold p < .001; cluster threshold p < .05 FDR). ANOVA interaction, between‐subject contrast. Inset color bar reflects increased vs reduced connectivity following pairwise post hoc tests. Reduced paired connections between the bilateral ventral anterior thalamic complex were observed (VAia and VAip: blue = negative paired ROI t‐statistic; connection P‐FDR < .01, LTLE > HC approaching significance). Within both RTLE and LTLE subgroups separately, no significance was determined comparing patients with FBTCS(+) to patients without FBTCS(−). Abbreviations: ANOVA; analysis of variance; FBTCS, focal‐to‐bilateral tonic–clonic seizures; FDR, false discovery rate; FNC, functional network connectivity; HC, healthy controls; LTLE, left TLE; MNI, Montreal Neurological Institute; RTLE, right TLE; TLE, temporal lobe epilepsy; VAia, ventral anterior inferior‐anterior; VAip, ventral anterior inferior‐posterior.

3.3. Thalamic nodal topology

rmMANOVA comparing TLE patients to controls revealed significant region × group interactions for all nodal measures (CC: p = .008; DC: p = .005; LE: p = .009; detailed model statistics in Tables S1A–C), indicating that group differences varied by thalamic region, with the most pronounced abnormalities localized to bilateral ventral anterior nuclei. Post hoc pairwise comparisons identified regional deficits (p < .001–.043, Figure 2A). Right VAia exhibited the most severe deficits in TLE compared to controls: CC (β = −.166 [−.26, −.07], p < .001), DC (β = −.518 [−.83, −.21], p = .001), and LE (β = −.183 [−.28, −.08], p < .001). Left VAia showed similar patterns: CC (β = −.162 [−.26, −.07], p = .002), DC (β = −.392 [−.70, −.09], p = .012), and LE (β = −.167 [−.27, −.06], p = .002). Conversely, left VPm demonstrated increased LE (β = .111 [.02, .20], p = .019) and CC (β = .098 [.01, .18], p = .026).

FIGURE 2.

Thalamic nodal topology alterations in TLE patients with and without FBTCS. (A) Thalamic network abnormalities in TLE patients compared to healthy controls. Brain schematics show bilateral thalamic parcellation with color‐coded regions indicating significant alterations. Bar plots display effect sizes (β) with 95% CIs for clustering coefficient (local segregation), degree centrality (hubness), and local efficiency (regional integration). TLE patients demonstrated significant bilateral ventral anterior thalamic dysfunction across all three network measures. Right VAia showed the most pronounced deficits in local segregation, hubness, and regional integration. Left VAia exhibited similar patterns with moderate effect sizes across measures (η²p = .022–.034, power = .71–.87). Conversely, left VPm demonstrated increased connectivity metrics, particularly in hubness. RTLE patients showed more severe ipsilateral ventral anterior reductions compared to LTLE patients, with right VAia hubness exhibiting the largest effect size and conversely increases in left DAm across all measures. (B) FBTCS‐specific thalamic network alterations within TLE patient subgroups. FBTCS+ patients exhibited distinct laterality‐specific organization patterns. In overall TLE comparisons, only right DP regional integration showed significant reduction. RTLE patients with FBTCS demonstrated widespread posterior thalamic reductions, including right VPm segregation and integration, right VPl integration, and right DP measures. Conversely, right DAm hubness increased significantly, while left VPl hubness decreased. Conversely, LTLE patients with FBTCS showed increased bilateral VPm integration and integration. *Color coding reflects effect size magnitude within each measure: Blue indicates negative effects (decreased effect sizes) and red indicates positive effects (increased effect sizes), with intensity representing distance from zero. Asterisks denote statistical significance: *p < .05, **p < .01, **p < .001. Error bars represent 95% CIs. Y‐axis scales are fixed within each network measure across all comparisons to enable direct magnitude comparison. Abbreviations: CI, confidence interval; DAm, dorsal anterior medial; DP, dorsal posterior; FBTCS, focal‐to‐bilateral tonic–clonic seizures; HC, healthy control; LTLE, left TLE; RTLE, right TLE; TLE, temporal lobe epilepsy; VAia, ventral anterior inferior‐anterior; VAip, ventral anterior inferior‐posterior; VPl, ventral posterior lateral; VPm, ventral posterior medial.

rmMANOVA comparing RTLE and LTLE with controls revealed significant region × hemisphere interactions for LE and DC (p < .013; Tables S1A–C). RTLE patients exhibited more pronounced ipsilateral thalamic reductions than LTLE patients, particularly in ipsilateral VA regions (VAip: p = .023; VAia: p < .001; Figure 2A). Ipsilateral VAia demonstrated the most severe reductions in RTLE: CC (β = −.175 [−.29, −.06], p = .004), DC (β = −.701 [−1.09, −.32], p < .001), and LE (β = −.195 [−.32, −.07], p = .002). In addition, ipsilateral VAip DC was similarly compromised (β = −.506 [−.94, −.07], p = .023). Conversely, relative to the LTLE patients, the RTLE patients showed increases in contralateral DAm across all measures: CC (β = .107 [.00, .21], p = .045), DC (β = .761 [.29, 1.23], p = .002), and LE (β = .130 [.02, .24], p = .024). In contrast, LTLE patients exhibited relatively preserved thalamic function, with only ipsilateral VPm CC increases (β = .161 [.04, .28], p = .004). Key ventral anterior thalamic connectivity features across subgroup contrasts are summarized in Table 2.

TABLE 2.

Key ventral anterior thalamic connectivity features by subgroup comparison.

Comparison	Right VAia				Left VAia
	CC	DC	LE	ICCD	CC	DC	LE	ICCD
	Segregation	Hubness	Integration	Cross‐hemispheric	Segregation	Hubness	Integration	Cross‐hemispheric
TLE vs HC	−.166***	−.518**	−.183***	—	−.162**	−.392*	−.167**	—
	[−.27, −.07]	[−.83, −.21]	[−.29, −.08]		[−.26, −.06]	[−.70, −.09]	[−.27, −.07]
RTLE vs HC	−.175**	−.701***	−.195**	—	NS	NS	−.132*	—
	[−.29, −.06]	[−1.10, −.30]	[−.32, −.08]				[−.26, .00]
LTLE vs HC	NS	NS	NS	—	NS	NS	NS	—
RTLE vs LTLE	−.175**	−.701***	−.195**	−.032***	NS	NS	NS	−.009
	[−.29, −.06]	[−1.10, −.30]	[−.32, −.08]	[−.05, −.02]				[−.03, −.01] (NS)
RTLE: FBTCS+ vs. FBTCS−	NS	NS	NS	−.024*	NS	NS	NS	.006
				[−.05, .00]				[−.02, .04] (NS)
LTLE: FBTCS+ vs. FBTCS−	NS	NS	NS	.003	NS	NS	NS	.004
				[−.03, .03] (NS)				[−.02, .03] (NS)

Note: Values are unstandardized β coefficients [95% CI] from repeated‐measures MANOVA post hoc pairwise comparisons with Bonferroni correction. Negative β indicates reduced connectivity in the first‐named group.

Abbreviations: CC, clustering coefficient (local segregation); DC, degree centrality (connection density); FBTCS, focal‐to‐bilateral tonic–clonic seizures; HC, healthy control; ICC, intrinsic connectivity contrast (cross‐hemispheric nodal centrality); ICCD, ICC deviation from HCs (cross‐hemispheric abnormality); LE, local efficiency (regional integration); LTLE, left TLE; NS, not significant (p > .05); dash entries (–) in ICCD columns denote undefined values. ICCD is computed relative to HC; therefore, both the right‐ and left‐VAia ICCD column is undefined for HC contrasts. RTLE, right TLE; TLE, temporal lobe epilepsy; VAia, ventral anterior inferior‐anterior thalamic nucleus (Tian et al., 2020 Scale IV parcellation).

^* p < .05.

^** p < .01.

^*** p ≤ .001.

3.4. FBTCS‐specific network organization

rmMANOVA comparing FBTCS+ and FBTCS− revealed significant region × group interactions for LE (p = .019) and DC (p = .043; Tables S1A–C). In post hoc comparisons, FBTCS+ patients exhibited decreased right dorsal posterior LE (β = −.137 [−.26, −.01], p = .035; Figure 2B). Within LTLE, the FBTCS+ patients displayed enhanced bilateral VPm function: CC (left VPm: β = .222 [.05, .39], p = .011; right VPm: β = .162 [.01, .32], p = .042) and LE (left VPm: β = .252 [.08, .43], p = .005; right VPm: β = .196 [.03, .37], p = .025). Conversely, within RTLE, the FBTCS+ patients displayed more widespread effects: reduced ipsilateral VPm CC/LE (β range: −.276 to −.264, p < .012), VPl LE (β = −.212 [−.41, −.01], p = .038), DP CC/LE (β range: −.236 to −.214, p < .029), and contralateral VPl DC (β = −.857 [−1.70, −.01], p = .047). However, ipsilateral DAm DC increased substantially in FBTCS+ RTLE patients (β = .932 [.14, 1.72], p = .022).

3.5. Cross‐hemispheric interactions ‐ intrinsic connectivity contrast (ICC) and ICC deviation from HP

ICC rmMANOVA revealed a significant multivariate region × hemisphere × group interaction in TLE compared to HCs (F = 2.598, p = .013; Supplemental Table S2A), driven by reduced ICC in the right VAia (β = −.023 [−.04, −.01], p = .001; Figure 3). Significant region × hemisphere × group interactions were also observed across SOZ hemisphere (F = 3.338, p < .001) and FBTCS status models (F = 1.897, p = .024). RTLE patients showed more severe R‐VAia reductions (β = −.040 [−.06, −.02], p < .001) compared to controls, and FBTCS+ patients showed reduced R‐VAia ICC relative to FBTCS− patients (β = −.025 [−.04, −.01], p < .001; further model statistics in Tables S2A and S2B).

FIGURE 3.

Regional ICC relative to HC. Average ICC maps were characterized by the strength of connectivity between a given voxel and the rest of the thalamus. Comparing regional ICC deviations to HCs allow inference of intra‐ vs cross‐hemispheric interactions. Positive deviations compared to HCs indicate effects ipsilateral to the seizure‐onset zone (SOZ; intra‐hemispheric connections), whereas negative deviations indicate effects contralateral to the SOZ (cross‐hemispheric connections). ²⁸ ICC deviations from HCs tested for multivariate interactions across each patient subgrouping * p < .05 Bonferroni corrected; RTLE_FBTCS+ > RTLE_FBTCS‐ for the right VAia thalamus representing −.047 mean difference (±.012, 95% CI). All remaining thalamic parcels yielded no other significant deviations from HC.

Aligned with our FNC analysis, cross‐hemispheric interaction (as measured by ICC deviation from HCs) was found only in RTLE, specifically in the FBTCS+ vs FBTCS− comparison, involving the ipsilateral VAia (β = −.024 [−.05, .00], p = .044; Figure 3).

3.6. Association between connectivity and clinical variables

In FBTCS+ RTLE patients, longer seizure duration was associated with increased ipsilateral VAia thalamic ICC and ICC deviation from HCs (R‐VAia: r = −.522, p = .026). Conversely, FBTCS− RTLE patients demonstrated no such association (R‐VAia: r = .10, p = .48). No significant clinical‐connectivity associations emerged when analyzing all TLE patients collectively (all p's > .05).

3.7. Sensitivity analyses

The key cross‐hemispheric deviation metric (ICCD R‐VAia) showed no pathology subgroup effect across four pathology categories within FBTCS+ patients (p = .718; Table S5). Within MTS‐only patients (n = 68; FBTCS+ = 53, FBTCS− = 15), the ICCD R‐VAia difference preserved the expected direction and approached significance (p = .051; Table S5). Permutation testing (5000 iterations) of FBTCS labels for ICCD R‐VAia yielded a non‐significant univariate result (observed difference = −.014, permutation p = .205; Table S6), consistent with the conservative nature of univariate tests relative to our primary multivariate rmMANOVA models. Internal split‐half cross‐validation showed 100% sign‐consistency for the bilateral‐VA FNC effect and 97.4% for the R‐VAia ICC effect across 500 random splits, with mean F and β estimates found to be within the 95% confidence intervals (CIs) of the full‐sample analysis (Table S10). ASM burden did not differ between FBTCS+ and FBTCS− groups (p = .98) and did not confound connectivity–FBTCS relationships (Table S7). FBTCS timing analysis (Current <1 year vs Remote >1 year vs Never) showed a numerical trend toward graded ICCD R‐VAia disruption (Never: −.014; Remote: −.031; Current: −.026; p = .097; Table S4), suggesting that VA abnormalities reflect chronic network reorganization rather than the effects of a recent/current seizure state.

4. DISCUSSION

This study provides the first demonstration of intrinsic thalamic subregion functional connectivity differences associated with FBTCS in TLE. For all TLE subjects, the primary finding was bilateral ventral anterior thalamic (or VA) hypoconnectivity. Reduced cross‐hemispheric coupling between the right and left VAia and VAip was the intra‐thalamic feature most consistently associated with FBTCS history, with this association strongest in RTLE. This VA dysfunction was the only sub‐network to survive FDR‐corrected testing. VA dysfunction was present across multiple graph‐theory metrics (CC, DC, and LE) and was not attributable to a single connectivity measure. Whether the reduced connectivity is causal for FBTCS or represents an epiphenomenon cannot be determined from this analysis.

The VA hypoconnectivity associated with FBTCS+ status reflected diminished coherence of activity between thalamic homologues and could represent either a reduction in structural connections between thalami or a reduction in physiological activity with intact connections. In contrast, TLE patients without FBTCS exhibited preserved VA cross‐hemispheric connectivity comparable to healthy controls, suggesting that intact VA networks are associated with a lower likelihood of secondary generalization of seizures. Whether VA integrity is causally protective, a marker of less‐severe underlying network pathology, or both, cannot be resolved by the present cross‐sectional analysis and would require a different type of analysis. The bilateral nature of VA dysfunction in patients with unilateral temporal lobe epilepsy suggests the possibility of thalamic subnetwork involvement beyond the side of seizure onset. ⁴ , ⁵ , ⁷ The reductions in cross‐hemispheric VA coupling were associated with greater seizure burden, appeared laterality‐dependent, and worsened with longer epilepsy duration; these patterns are consistent with that interpretation, and key findings are summarized in Table 2.

The laterality‐specific patterns we observed reveal the complexity of thalamic reorganization in TLE. Patients with right TLE exhibited more‐pronounced ipsilateral VA connectivity reductions than left TLE patients, suggesting that the effects of reduced connectivity depend on the hemispheric laterality of the SOZ. In right TLE, patients with FBTCS displayed ipsilateral reductions in ventral posterior and dorsal posterior connectivity, along with cross‐hemispheric VAia reductions that correlated with illness duration. In left TLE, patients with FBTCS showed bilateral increases in ventral posterior medial CC and LE. These opposing patterns suggest that VA thalamic nuclei may have different relay functions depending upon the hemisphere of seizure origin. In right TLE, VA dysfunction may permit disinhibition through reduced gating; in left TLE, it may facilitate excessive excitatory output. ⁹ , ¹⁵ , ²⁹ Hemispheric differences in language and cognitive organization as well as plasticity mechanisms may contribute to these laterality‐specific effects, although the precise mechanisms underlying these laterality‐dependent patterns remain to be determined. Stratification by both FBTCS status and SOZ laterality proved essential for detecting these connectivity patterns, underscoring the importance of clinical phenotype in epilepsy network studies. ⁴ , ⁵ , ¹⁶

Normal VA connectivity in TLE patients without FBTCS was evident in patients with either RTLE or LTLE, suggesting that VA network integrity might help prevent focal‐to‐bilateral seizure propagation; this hypothesis would require testing in a separate study. Our analyses further confirmed that VA connectivity abnormalities were robust and not confounded by pathology type, ASM burden, or FBTCS class imbalance. The bilateral VA‐complex hypoconnectivity finding was present in both LTLE‐vs‐HC (trend, p = .08) and RTLE‐vs‐HC (p = .003), representing the one effect shared across the cohort. The ipsilateral VAia nodal‐topology disruption, the R‐VAia ICC/ICCD reductions, and the FBTCS+/− cross‐hemispheric VA contrast were observed only in RTLE. The apparent absence of analogous LTLE‐specific effects may reflect (i) a true biological asymmetry; (ii) reduced power for within‐RTLE FBTCS+/− contrasts owing to the small RTLE FBTCS‐ subgroup (n = 18), which limits sensitivity for any parallel comparison attempted in LTLE; or (iii) potential masking by the dense left‐hemisphere language and executive networks, which share thalamic input territories with VA subregions and may obscure FBTCS‐related connectivity perturbations in LTLE resting‐state fMRI signal. These possibilities cannot be resolved by our data. Therefore, we frame the laterality patterns as RTLE‐specific in this cohort and explicitly hypothesis‐generating for LTLE (Supplemental §S5).

These findings suggest investigating a novel approach for neuromodulation strategies in drug‐resistant TLE. Current thalamic stimulation protocols primarily target the thalamic anterior nuclei (ANT, as in the SANTE trial), ³⁰ centromedian nucleus (CM), ³¹ and pulvinar nuclei. ³² The VA complex is anatomically and functionally distinct from these targets, receives basal ganglia output, and projects to premotor and motor cortical areas, ¹⁷ , ¹⁹ , ²⁰ making it mechanistically relevant to the tonic–clonic motor component of FBTCS. Given the cross‐sectional association between aberrant bilateral VA connectivity and FBTCS in TLE, the VA complex could be considered as a target for neuromodulation of FBTC seizures. Animal data showing that VA lesions reduce experimental seizure frequency ³³ support a functional role for VA in seizure networks. This could be explored with either direct VA targeting or modulation of structures projecting to VA. ¹⁸ Because FBTCS confer elevated risk for SUDEP and injury, ³⁴ connectivity‐guided approaches to reducing this seizure subtype merit focused attention.

Our study has several limitations. First, our cross‐sectional design precludes causal inference regarding whether thalamic dysfunction predisposes to FBTCS, results from ongoing seizures, or reflects an underlying pathological process driving both. ⁴ , ⁶ , ⁷ The association between longer epilepsy duration and greater VA connectivity reductions in FBTCS+ RTLE patients may reflect progressive degradation, developmental effects related to age at onset, or inherent differences in VA connectivity at epilepsy onset. ³⁵ , ³⁶ Longitudinal studies are needed to disentangle these relationships. ¹⁴ Second, FBTCS classification relied on clinical history and may have introduced recall bias, although chronicity‐stratified analyses showed no effect of timeframe (Table S4). Third, our cohort had drug‐resistant TLE, limiting generalizability to other epilepsy types; heterogeneous pathology was deliberately included to broaden applicability beyond hippocampal sclerosis, and sensitivity analyses confirmed VA findings across pathology subgroups (Table S5). The FBTCS+/− class imbalance reduced rather than inflated power, and permutation testing confirmed effects were not artifacts of unequal group sizes (Table S6). Fourth, the single 5‐min resting‐state acquisition, while consistent with validated protocols, ⁹ , ¹⁰ , ¹¹ , ²² , ²³ may limit statistical power; multi‐run, longer‐duration acquisitions integrated with structural imaging are warranted. ⁴ , ¹² , ¹³ , ³⁷ Fifth, restricting analyses to 16 thalamic ROIs preserves subnuclear resolution and power for the FBTCS+/− contrast but precludes inferences about cortical and basal ganglia/cerebellar contributions in seizure propagation. ²³ , ²⁹ Sixth, our 2 × 2 × 4 mm voxel and 6 mm FWHM smoothing approached the linear extent of individual VA subnuclei; the bilateral VA‐complex finding is robust, but distinctions between VAia and VAip should be interpreted as suggestive pending higher‐resolution acquisition. Finally, although internal split‐half cross‐validation supported robustness within the cohort, no independent external replication was available; multi‐site validation in cohorts such as ENIGMA‐Epilepsy will be needed before VA connectivity can be considered a candidate biomarker of FBTCS status.

5. CONCLUSION

In conclusion, bilateral VA hypoconnectivity was the predominant functional‐connectivity derangement between thalamic nuclei in TLE, and preserved VA cross‐hemispheric connectivity in patients without FBTCS was associated with a lower likelihood of secondary generalization. The effects of FBTCS on thalamic connectivity were laterality‐dependent, with more pronounced abnormalities in RTLE. These findings extend prior work on thalamic structural abnormalities to the functional connectivity domain ⁵ , ⁶ , ⁷ , ¹² , ¹³ , ³⁷ and provide initial, hypothesis‐generating evidence that VA connectivity is a candidate marker of seizure‐generalization vulnerability in TLE. Longitudinal multimodal imaging with prospective seizure monitoring, ideally in independent cohorts, is required to determine whether VA connectivity has prognostic, predictive, or therapeutic‐targeting utility.

AUTHOR CONTRIBUTIONS

S.N.H.: Conceptualization, methodology, formal analysis, and writing (original draft and all revisions). M.R.S.: Investigation, resources, and writing (review & editing). H.A.: Methodology, software, and supervision. J.I.T.: Conceptualization, funding acquisition, supervision, and writing (review & editing).

FUNDING INFORMATION

This work was supported by National Institutes of Health/National Institute of Neurological Disorders and Stroke (Joseph I. Tracy, PI), R01 NS112816‐01.

CONFLICTS OF INTEREST STATEMENT

None of the authors has any conflict of interest to disclose.

ETHICS STATEMENT

We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this report is consistent with those guidelines.

Supporting information

Figure S1. Patient Recruitment/Stratification & Preprocessing Pipeline and Functional network connectivity (FNC) cluster‐level analysis method. Patient Recruitment/Stratification. All 166 patients meeting unilateral TLE criteria determined by multidisciplinary epilepsy‐surgery‐conference consensus were included. The chronicity stratification (Current <1 yr., Remote >1 yr., Never) was classifiable for 154/166 patients; FBTCS+ chronicity was unrecorded for 1 patient (one subject (1 RTLE FBTCS+) could not be identified as current or remot) and FBTCS− chronicity was unrecorded for 11 patients. The chronicity‐row totals (Current 75 + Remote 44 + Chronicity‐unrecorded 1 = 120 FBTCS+; Never 35 + Chronicity‐unrecorded 11 = 46 FBTCS−) reconcile to the FBTCS+/− totals shown in the row above. The chronicity‐unrecorded patients were excluded only from the §S4.5 chronicity analysis; all primary analyses use the full n = 166. No exclusions post‐consensus. FBTCS status based on lifetime history; laterality determined by convergent evidence from clinical history, video‐EEG, structural MRI, and FDG‐PET. (A) Preprocessing pipeline using fMRIPREP (v20.2.0)⁴, CONN⁵ (RRID:SCR_009550) release 22a and SPM⁶ (RRID:SCR_007037) release 12.7771. (B) Hierarchical clustering based upon ROI‐to‐ROI anatomical proximity and functional similarity metrics (all ROI pairs examined for within‐ and between‐intrinsic network connectivity) was utilized to yield functional network connectivity (FNC) sets that survived multivariate parametric general linear model testing and familywise error control with the connection threshold set at p < .05 false discovery rate (FDR)–corrected and the cluster threshold set at p < .05 FDR¹⁵.

Table S1. (A) rmMANOVA Results for Clustering Coefficient (CC). (B) rmMANOVA Results for Degree Centrality (DC). (C) rmMANOVA Results for Local Efficiency (LE).

Table S2. (A) rmMANOVA Results for Intrinsic Connectivity Contrast (ICC). (B) rmMANOVA Results for Intrinsic Connectivity Contrast deviation from HP (ICCD).

Table S3. Significant Thalamic Post‐hoc Findings from rmMANOVA.

Table S4. Group Comparisons of Connectivity Features Across FBTCS Chronicity.

Table S5. Pathology Subgroup Sensitivity Analyses.

Table S6. Class Imbalance Robustness.

Table S7. ASM and Covariate Analyses.

Table S8. FBTCS Frequency–Connectivity Correlations.

Table S9. Pre‐specified analysis plan and multiple‐comparison corrections.

Table S10. Split‐half cross‐validation of primary VA findings.

Table S11. Effect generalizability across SOZ laterality.

DATA AVAILABILITY STATEMENT

Anonymized data not published in this article will be made available by request from any qualified investigator, subject to institutional data‐sharing agreements. All code and de‐identified processed data can be made available upon request.