Probabilistic mapping of thalamic nuclei and thalamocortical functional connectivity in idiopathic generalised epilepsy

Abstract It is well established that abnormal thalamocortical systems play an important role in the generation and maintenance of primary generalised seizures. However, it is currently unknown which thalamic nuclei and how nuclear‐specific thalamocortical functional connectivity are differentially impacted in patients with medically refractory and non‐refractory idiopathic generalised epilepsy (IGE). In the present study, we performed structural and resting‐state functional magnetic resonance imaging (MRI) in patients with refractory and non‐refractory IGE, segmented the thalamus into constituent nuclear regions using a probabilistic MRI segmentation method and determined thalamocortical functional connectivity using seed‐to‐voxel connectivity analyses. We report significant volume reduction of the left and right anterior thalamic nuclei only in patients with refractory IGE. Compared to healthy controls, patients with refractory and non‐refractory IGE had significant alterations of functional connectivity between the centromedian nucleus and cortex, but only patients with refractory IGE had altered cortical connectivity with the ventral lateral nuclear group. Patients with refractory IGE had significantly increased functional connectivity between the left and right ventral lateral posterior nuclei and cortical regions compared to patients with non‐refractory IGE. Cortical effects were predominantly located in the frontal lobe. Atrophy of the anterior thalamic nuclei and resting‐state functional hyperconnectivity between ventral lateral nuclei and cerebral cortex may be imaging markers of pharmacoresistance in patients with IGE. These structural and functional abnormalities fit well with the known importance of thalamocortical systems in the generation and maintenance of primary generalised seizures, and the increasing recognition of the importance of limbic pathways in IGE.

Centromedian and anterior nuclei of the thalamus have been associated with initiation, propagation and maintenance of generalised SWDs (Miller & Ferrendelli, 1990;Miller, Hall, Holland, & Ferrendelli, 1989;Tyvaert et al., 2009). Although abnormal thalamocortical networks are clearly important in generating and maintaining generalised seizures, it is currently unknown whether differences in thalamocortical architecture and connectivity underlie pharmacoresistance in patients with IGE.
There have been many MRI-based investigations of thalamic alterations in patients with IGE. Frequently used research methods include thalamic volume, thalamic shape and voxel-based morphometry (VBM) analyses. While volume analysis attempts to evaluate gross thalamus volume change, shape and VBM analyses have allowed investigation of regional or subtle changes in the thalamus. Studies have reported a decrease (Boss, Abela, Weisstanner, Schindler, & Wiest, 2019;Ciumas & Savic, 2006;Du et al., 2011;Kim et al., 2014;Kim, Kim, Suh, & Kim, 2018;Lee, Seo, Lee, Kim, & Park, 2020;Mory et al., 2011;Nuyts, D'Souza, Bowden, & Vogrin, 2017;Perani et al., 2018;Pulsipher et al., 2009;Saini et al., 2013;Di Wang et al., 2016;Wang et al., 2012;Whelan et al., 2018;Zhong et al., 2018) and an increase (Betting et al., 2006;Bin et al., 2017; in thalamic volume in patients with IGE compared to healthy controls. Some studies have, however, reported no statistically significant differences in thalamic volume (Betting et al., 2010;Natsume, Bernasconi, Andermann, & Bernasconi, 2003;Seeck et al., 2005). This discrepancy may be due to a number of factors, including that thalamic nuclei are potentially differentially impacted in IGE, differential combinations of refractory and / or nonrefractory patients in statistical analyses, and different analytical approaches were adopted. Kim and colleagues reported grey matter (GM) atrophy in the anteromedial thalamus in IGE relative to controls (Kim et al., 2014). Two other studies that specifically focused on juvenile myoclonic epilepsy (JME) suggested volume reduced in the anterior (Mory et al., 2011) and medial thalamus (Saini et al., 2013) in patients relative to controls. Wang and colleagues also identified thalamic atrophy in the medial dorsal and pulvinar nuclei in patients with IGE generalised tonic-clonic seizure alone (IGE-GTCS) relative to controls (Wang et al., 2012). Localised volume reduction of the ventral thalamus has been reported in patients with refractory IGE relative to controls (Boss et al., 2019). There is therefore accumulating evidence indicating regionally specific structural alterations of the thalamus in IGE, but there is a lack of consistency with respect to the particular region of the thalamus affected.
There have been many approaches to segment thalamic nuclei using MRI; many of these necessitate the use of high field or advanced MRI sequences that may not be incorporated to clinical scanning for patients with epilepsy. A recently published thalamic nuclei segmentation method provides an opportunity to probabilistically segment the thalamus into 50 thalamic nuclei based on histological data and ex-vivo MRI and necessitates only three dimensional T1-weighted (T1w) MRI data (Iglesias et al., 2018). The approach has been recently applied to determine regionally specific thalamic alterations in neurological disorders (Bocchetta et al., 2020;Hougaard et al., 2020;Ngamsombat et al., 2020;Shin, Lee, & Park, 2019) but has not yet been applied to patients with IGE. In the present study, we use this method to determine regional structural and functional thalamic abnormalities in patients with IGE and differences according to seizure control.
Previous simultaneous EEG-fMRI studies have reported that the thalamus is activated during generalised SWDs (Centeno & Carmichael, 2014;Klamer et al., 2018;Seneviratne, Cook, & D'Souza, 2014;Tyvaert et al., 2009). Large-scale network analysis using resting-state functional MRI (rs-fMRI) has revealed altered functional connectivity in patients with IGE relative to controls, which were often found in intrinsic networks, such as the default mode and salience networks McGill et al., 2012;Wang et al., 2011;Wei et al., 2015). Few studies have identified thalamocortical functional alterations in IGE using rs-fMRI. Serving as a functional complex and relay station between different subcortical areas and the cerebral cortex, distinct thalamic nuclei may be divergently associated with functionally distinct areas in cortex (Mai & Majtanik, 2019;Noback, Strominger, Demarest, & Ruggiero, 2005;Rikhye, Wimmer, & Halassa, 2018;Yuan et al., 2016). This suggests that thalamocortical functional connectivity alterations in IGE might also be divergent depending on the involvement of particular thalamic nuclei-potentially differing between patients with refractory and non-refractory IGE. Some studies have reported reduced or increased thalamocortical rs-fMRI connectivity when the regional thalamic seed was generated based on VBM results (Kim et al., 2014;Wang et al., 2012;Zhong et al., 2018) or diffusion tensor imaging connectivity-based thalamic parcellations . However, there has been no work administered to date that has explored the functional connectivity between cortex and the delineated nuclei of the thalamus, while investigating potential differences in thalamocortical connectivity in patients with refractory and nonrefractory IGE.
There were two primary goals of the present study. First, we sought to determine regional thalamic nuclear volume alterations in patients with refractory and non-refractory IGE and healthy controls.
We hypothesised differential thalamic nuclei involvement based on whether patients were refractory or non-refractory to ASM. Second, using parcellated thalamic nuclei regions as seeds, we sought to investigate rs-fMRI thalamocortical connectivity in IGE and to determine whether alterations are associated with treatment outcomes. To our knowledge, this represents the first study to probe differences in thalamocortical architecture and connectivity in patients with refractory and non-refractory IGE.
2 | METHODS  Fischl, 2012) and a probabilistic thalamic segmentation algorithm incorporated into FreeSurfer software were applied to segment and estimate the volumes of thalamic nuclei. First, T1w images were processed with the FreeSurfer "recon-all" function to correct for nonuniformity and fluctuations in MRI intensity, remove skull, and perform an automated intensity-based segmentation for cortical and subcortical brain structures. The second step was to segment the left and right thalamus each into 25 different nuclei using Bayesian inference based on a probabilistic atlas built with histological data (Iglesias et al., 2018). The parcellated thalamic nuclei are illustrated in Figure 1;

| Participants
abbreviations for each nucleus are presented in Table 3. We divided the nuclear volume measurements by the respective total intracranial volume estimated using FreeSurfer (Buckner et al., 2004) in order to control for the effect of brain size. The following formula was used: Nuclei volume % ð Þ¼ nuclei volume mm 3 À Á =total intracranial volume mm 3 À Á À Á

Â100%
Analysis was performed using a one-way analysis of covariance approach (Benjamini & Hochberg, 1995) was applied to correct for multiple comparisons using MATLAB 2018a (The Mathworks, Inc.,

2018) Bioinformatics
Toolbox. An FDR-adjusted p-value of less than .05 was considered statistically significant.

| Thalamic functional connectivity
As a first step, all imaging data were manually recentered to the anterior commissure. Rs-fMRI data were preprocessed using Statisti-   (Table 3) due to shared functional topology and the small size of those seeds (Mitchell & Chakraborty, 2013;Pergola et al., 2018). The pulvinar seed was defined as a collection of PuA, PuM, PuI and PuL (Table 3). All nuclei were then spatially normalised to MNI space using SPM tools.
To compute seed-based functional connectivity, the resting-state BOLD time series for each seed ROI was averaged and correlated with BOLD time series for each GM voxel. Fisher-transformed bivariate correlation coefficients were calculated to represent the degree of functional connectivity. The seed-based connectivity maps were subsequently used for the second-level analysis of relative thalamocortical functional connectivity changes between groups. A statistical parametric map was created to characterise the differences in functional connectivity between groups using the general linear model, corrected for age and sex. Voxel-wise statistics for thalamocortical connectivity throughout the entire brain were controlled at an uncorrected level (p < .001) with an additional clusterlevel correction (p FDR < .05) based on Gaussian Random Field theory (Worsley et al., 1996) applied for FDR correction (Alonazi et al., 2019;Chumbley, Worsley, Flandin, & Friston, 2010;Fallon et al., 2016).

Identification of anatomical regions for significant clusters was based
on Harvard-Oxford atlas (Makris et al., 2006).

| Thalamic nuclei volumetric analysis
Statistically significant volume reduction was only observed in patients with refractory IGE compared to healthy controls (left AV, p FDR = .025; right AV, p FDR = .018) (Figure 3). There was no trend for volume reduction of the left or right AV in patients with nonrefractory IGE compared to controls (Table 4). Trends were observed for volume reduction of other thalamic nuclei in patients compared to controls, but these effects did not survive FDR correction (Table 4).

| Thalamocortical resting-state functional connectivity analysis
Patients with IGE had significantly altered functional connectivity between thalamic seeds-including left VLa and LGN bilaterally-and regions of the cerebral cortex, relative to controls (Figure 4a). F I G U R E 2 A framework for resting-state thalamocortical functional connectivity analysis. (a) Identifying thalamic regions of interest as per Figure 1. All nuclei were extracted and binarized into 50 separate masks using FSL. Nuclei selected for analysis were registered to ICBM 152 template. (b) rs-fMRI pre-processing based on the standard SPM protocol. The origin of T1w and rs-fMRI data was set to the anterior commissure prior to all processing. rs-fMRI data underwent corrections, normalisation and spatial smoothing; T1w data underwent tissue characteristic segmentation and spatial normalisation. (c) CONN processing for resting state functional connectivity. All pre-processing data (from a and b) were input into CONN toolbox for running functional connectivity analysis. The seed-to-voxel approach was used. Movement parameters were regressed, and noise was filtered to produce clean BOLD signals. Second-level analysis was implemented to determine significant differences of rs-fMRI thalamocortical connectivity between study groups F I G U R E 3 Significant volume reduction of the left and right anteroventral (AV) only in patients with refractory idiopathic generalised epilepsy (IGE) as revealed by one-way analysis of covariance (ANCOVA). Asterisks indicate significant differences between groups at p FDR T A B L E 4 Thalamic nuclear volume and group differences

| DISCUSSION
There were two primary goals of the present study. First, we sought to determine regional thalamic nuclear volume alterations in patients with refractory and non-refractory IGE and healthy controls. We report significant volume reduction of the AV nuclei bilaterally only in patients with refractory IGE. Patients with non-refractory IGE did not show significance or a trend (uncorrected p < .05) for volume reduction of these nuclei. Second, we sought to investigate rs-fMRI thalamocortical connectivity alterations in IGE and to determine whether alterations are associated with treatment outcomes. We report significant alterations in thalamocortical functional connectivity between patients with IGE and controls and found evidence for increased functional connectivity between the VLp nuclei bilaterally and regions of frontal and occipital cortex in patients with refractory IGE relative to patients with non-refractory IGE.

| Biological and clinical implications
VBM studies have not consistently reported atrophy of the anterior regions of the thalamus in IGE, although some have described this in patients with JME (Kim, Kim, Seo, Suh, & Koh, 2013;Mory et al., 2011). On the contrary, one study reported increased volume in the anterior thalamus in patients with IGE relative to controls, which was more obvious in patients with absence seizures, who had worse seizure control (Betting et al., 2006). Our findings that the AV nuclei are specifically affected in patients with refractory IGE is a new finding; the fact that these nuclei are affected in both cerebral hemispheres is consistent with the bihemispheric nature of IGE. The anterior thalamic nuclei complex is a part of the limbic system and an important relay of the Papez circuit (Papez, 1937). The nuclei receive output from the hippocampus via the fornix or mammillary bodies via the mammillothalamic tract, and project to the cingulate cortex; information travels through the cingulate bundle and returns to the hippocampus to complete the circuit (Shah, Jhawar, & Goel, 2012;Weininger et al., 2019). The anterior thalamus has been suggested to relate to multiple cognitive processing tasks such as memory, executive function and spatial navigation (Jankowski et al., 2013  Bihemispheric thalamocortical alterations seeded from the ventral lateral nuclei were prominent findings in the present study. We observed significantly increased functional connectivity between these nuclei and cortex in all patients compared to controls (anterior), refractory patient's relative to controls (anterior), and refractory patients relative to non-refractory patients (anterior and posterior).
F I G U R E 4 Significant differences in thalamocortical functional connectivity between groups. Boxplots of mean Fischer transformed correlation coefficients illustrate correlation (positive) and anti-correlation (negative) relationship of individual thalamocortical functional connectivity. (a) Functional connectivity alteration with thalamic seeds in patients with IGE relative to controls. Cluster colour indicates the direction of contrast (red: control > IGE; blue: IGE > control); (b) Functional connectivity alteration with thalamic seeds in patients with nonrefractory IGE relative to controls. Cluster colour indicates the direction of contrast (red: Control > nonREF-IGE; blue: nonREF-IGE > Control); (c) Functional connectivity alteration with thalamic seeds in patients with refractory IGE relative to controls.  (Mai & Majtanik, 2019). Hyperconnectivity was observed between these nuclei and primary motor and frontal lobe regions (in addition to superior occipital regions), which is consistent with the known pathophysiology of generalised seizures and hyperactivation and hyperconnectivity during cognitive processing in IGE Vollmar et al., 2011). Key to our findings is that increased functional connectivity between ventral lateral nuclei and cortex could represent an imaging marker that can help discriminate between refractory and non-refractory given that such hyperconnectivity was observed in patients with refractory IGE compared to non-refractory IGE. However, whether this represents a mechanistic marker of pharmacoresistance or is an epiphenomenon of a clinical difference between patient groups (e.g., prevalence of GTCS) remains to be elucidated. To our knowledge, this is the first demonstration of functional nuclei-cortical connectivity differences between patients with refractory and non-refractory IGE. Similar to our structural findings, that these findings were bihemispheric is in keeping with the generalised nature of the disorder.
Both patients with refractory IGE and patients with non-refractory IGE showed decreased functional connectivity between the right CM nucleus and areas of the medial frontal cortex relative to controls. On the basis that the CM-parafascicular complex contributes to neuromodulation within a basal ganglia and motor network (Mclardy, 1948;Sadikot & Rymar, 2009), the CM nucleus is the DBS target for refractory primary generalised seizures (Valentín et al., 2013;Zangiabadi et al., 2019). CM DBS yields over 50% reduction in seizure frequency across various types of epilepsy and patients with medically refractory generalised epilepsy manifesting primary tonic-clonic or absence seizures benefiting the most (Alcala-Zermeno et al., 2021;Cukiert et al., 2009;Ilyas, Pizarro, Romeo, Riley, & Pati, 2019;Klinger & Mittal, 2018;Son et al., 2016;Valentín et al., 2013;Velasco et al., 2007;Velasco, Velasco, Jimenez, Velasco, & Marquez, 2002 we are not aware of previous work reporting the significance of these nuclei in the generation or modulation of primary generalised seizures. Interconnectivity with the thalamic reticular nucleus may provide insights (Uhlrich, Manning, & Feig, 2003); the thalamic reticular nucleus has been suggested to be involved in generalised SWDs in absence seizures (Huguenard, 2019). A recent study has also proposed that increasing neuronal firing in the thalamic reticular nucleus may enhance the inhibitory activity of the dorsal LGN (Campbell, Govindaiah, Masterson, Bickford, & Guido, 2020).

| Methodological considerations
The patient cohort included different subtypes of IGE. Previous studies have suggested that thalamic alterations may be exhibited differently in IGE with different underlying seizure conditions (Betting et al., 2006). We were unable to analyse individual IGE subtypes in this study given the overall sample size; future work should try to replicate our findings separately in patients with refractory and nonrefractory absence epilepsy, juvenile myoclonic epilepsy, and epilepsy with primary generalised tonic-clonic seizures alone. Our primary goal was to prospectively recruit patients according to whether It is plausible to assume that administration of different ASMs have differential effects on thalamocortical functional connectivity in patients. Previous studies suggested the chronic use of topiramate and carbamazepine-less common ASMs in our cohort-may have effects on BOLD signal measures and negatively impact on cognition in patients with focal epilepsy, although valproic acid, lamotrigine or levetiracetam were unlikely to yields significant effects (Haneef, Levin, & Chiang, 2015;van Veenendaal et al., 2015van Veenendaal et al., , 2017. Although there is limited evidence indicating that typically prescribed ASMs used for primary generalised epilepsy affect resting-state fMRI networks in patients with IGE, it would be prudent to design a study that systematically examines whether this is the case. Although the probabilistic method of thalamic nuclei segmentation used in the present study corresponds well to a histological atlas of the thalamus (Iglesias et al., 2018), the approach does not delineate nuclear subfields and is inherently constrained by the limited spatial resolution and contrast of T1w MRI. Optimising sequences to improve tissue contrast (Iglesias et al., 2018;Su et al., 2019) and high-field delineation of thalamic regions (Kanowski et al., 2014;Liu, D'Haese, Newton, & Dawant, 2020;Plantinga et al., 2014) may increase the reliability of segmentations. The accuracy of nuclei segmentations and limited spatial resolution may also impact rs-fMRI connectivity; the time series from adjacent nuclear regions may impinge on some of the observed thalamocortical functional correlations. Furthermore, we have suggested that regional thalamic structural and functional alterations shown in our study may be relevant for cognitive impairments previously observed in patients with IGE. However, we urge caution with this interpretation given that we did not collect cognitive data on our participants and therefore we were unable to investigate this potential correlation directly.
Finally, it is important to be mindful that interictal epileptiform discharges may impact on regional glucose metabolism, rs-fMRI data and resultant functional networks in patients with IGE (Aghakhani et al., 2015;Cheng, Yan, Xu, Zhou, & Chen, 2020;Dahal et al., 2019;Gotman et al., 2005;Kim, Im, Kim, Lee, & Kang, 2005;Lv, Wang, Cui, Ma, & Meng, 2013). We did not perform simultaneous EEG-fMRI to control for interictal epileptiform discharges in this study and this will be an important for future resting-state functional network studies in IGE. Moreover, our study would also have been strengthened by the inclusion of ambulatory EEG or video EEG data. Ambulatory EEG data is not routinely collected in patients with IGE in the United Kingdom.
Finally, given that, the LGN is one of the smallest nuclei, our method of probabilistic segmentation may have overestimated the LGN region of interest and functional correlations may have been confounded by adjacent nuclei. Further work is needed to determine whether functional connectivity alterations of the LGN are due to the disorder or methodological confounds.

| CONCLUSION
Atrophy of the anterior thalamic nuclei and resting-state functional hyperconnectivity between ventral lateral nuclei and cerebral cortex may be imaging markers that help discriminate between refractory and non-refractory patients with IGE. These structural and functional abnormalities fit well with the known importance of thalamocortical systems in the generation and maintenance of primary generalised seizures, and the increasing recognition of the importance of limbic pathways in IGE.

ACKNOWLEDGMENTS
This work was supported by an ERUK project grant (grant number 1085) and UK Medical Research Council grants (grant numbers MR/S00355X/1 and MR/K023152/1) awarded to SSK. Open access funding enabled and organized by Projekt DEAL.

CONFLICT OF INTEREST
None of the authors has any conflict of interest to disclose.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available on request from the corresponding author. The original data are not publicly available due to ethical restrictions.