Thalamostriatal disconnection underpins long-term seizure freedom in frontal lobe epilepsy surgery

Abstract Around 50% of patients undergoing frontal lobe surgery for focal drug-resistant epilepsy become seizure free post-operatively; however, only about 30% of patients remain seizure free in the long-term. Early seizure recurrence is likely to be caused by partial resection of the epileptogenic lesion, whilst delayed seizure recurrence can occur even if the epileptogenic lesion has been completely excised. This suggests a coexistent epileptogenic network facilitating ictogenesis in close or distant dormant epileptic foci. As thalamic and striatal dysregulation can support epileptogenesis and disconnection of cortico-thalamostriatal pathways through hemispherotomy or neuromodulation can improve seizure outcome regardless of focality, we hypothesize that projections from the striatum and the thalamus to the cortex may contribute to this common epileptogenic network. To this end, we retrospectively reviewed a series of 47 consecutive individuals who underwent surgery for drug-resistant frontal lobe epilepsy. We performed voxel-based and tractography disconnectome analyses to investigate shared patterns of disconnection associated with long-term seizure freedom. Seizure freedom after 3 and 5 years was independently associated with disconnection of the anterior thalamic radiation and anterior cortico-striatal projections. This was also confirmed in a subgroup of 29 patients with complete resections, suggesting these pathways may play a critical role in supporting the development of novel epileptic networks. Our study indicates that network dysfunction in frontal lobe epilepsy may extend beyond the resection and putative epileptogenic zone. This may be critical in the pathogenesis of delayed seizure recurrence as thalamic and striatal networks may promote epileptogenesis and disconnection may underpin long-term seizure freedom.


Introduction
Seizure outcome is variable in drug-resistant frontal lobe epilepsy, with seizure freedom at 1 year achieved in only 50% of patients, 1,2 while long-term seizure freedom occurs in less than 30% of cases. 3 Two different mechanisms have been proposed for early and delayed seizure recurrence: seizure recurrence within the first year is thought to be due to partial resection of the epileptogenic lesion or zone, 1 whilst a 5-8% yearly risk of seizure recurrence within 5 years postoperatively 4 may reflect ongoing epileptogenesis, despite complete resection of the primary abnormality. 4 Animal models indicate seizures spread along long-range white matter pathways. 5 Thus it is possible that seizure recurrence, despite complete resection of the epileptogenic lesion, may involve an epileptogenic network beyond the lesion whose dysfunction could be associated with seizure propagation and new or ongoing epileptogenesis. 6 How this network is composed and if the pathways involved are shared amongst patients remains poorly understood.
Seizures can be modulated by lesioning or stimulating discrete brain networks. [7][8][9] A role for thalamic networks in ictogenesis was demonstrated as early as the 1940s, as thalamic stimulation was shown to induce seizures 10 and thalamocortical dysfunction was linked to seizure propagation, 11,12 not only generalized seizures. 13 Within these networks, connections specifically from the anterior thalamic nucleus and the centromedian nucleus to the cortex are relevant, as their virtual disconnection through neuromodulation can diminish or abolish seizures. 7,[14][15][16][17] Their role in resective surgery, however, has not been addressed. Cortico-cortical and cortico-striatal connections may also be involved in frontal seizure propagation, as surgeries performed with the aim to disconnect rather than resect, such as callosotomy 18 or hemispherotomy, 19,20 can decrease seizure frequency and severity independent of lesion location or focality. As common networks within the frontal lobe have been linked to seizure modulation, a unifying hypothesis could posit that individuals with frontal lobe epilepsy may share a common epileptogenic network which affects seizure outcome somewhat irrespective of lesion localization, which could potentially become a target for disconnection.
To test this hypothesis, we retrospectively reviewed 47 patients who underwent surgery for drug-resistant frontal lobe epilepsy and applied disconnectome analyses to highlight patterns of disconnection associated with long-term seizure freedom.

Participants and seizure outcome
We performed a retrospective case collection, initially reviewing clinical data of 80 consecutive patients who underwent frontal lobe epilepsy surgery at our institution from 2007 to 2021. Of these, we identified 47 patients who met the following criteria: (i) preoperative and three-month postoperative MRI; and (ii) seizure outcome reported after at least 3 years. We evaluated lesion side, epilepsy duration, history of focal-to-bilateral tonic-clonic seizures, histology, seizure outcome and resection volume, and classified seizure outcome as (i) seizure-free (International League against Epilepsy grade I, 21 ILAE 1); or (ii) continuing seizures (ILAE 2 or higher). In those cases in whom (i) there was no visible residual lesion on postoperative MRI and (ii) histology confirmed a lesion/malformation, resection was considered complete. The study was conducted in accordance with ethical standards of the Declaration of Helsinki and was approved by the Research Ethics Committee of our institution (IRB: 22/SC/0016).

Presurgical evaluation
A workflow of presurgical evaluations at our Institution has been detailed as in Duncan 22 and descriptive surgical cases can be found in the Supplementary material. Initially patients were assessed with history of epilepsy, drug resistance, comorbidities, and had volumetric MRI, video-EEG telemetry, neuropsychiatric and neuropsychological assessments. Handedness was determined using a standardized questionnaire. 23 Language lateralization was determined using language functional MRI (fMRI) 24 or Wada test. Magnetoencephalography (MEG), EEG-fMRI, ictal single photon emission CT (SPECT) and fluorodeoxyglucose (FDG)-PET were carried out as indicated. These data were discussed by a multidisciplinary team of neurologists, neurophysiologists, neurosurgeons, neuropsychologists and neuropsychiatrists. 22 For individuals in whom results from semiology, imaging, neuropsychology and electrophysiology were not definitive, invasive EEG recordings were performed to localize the epileptogenic zone. Stereoencephalography (SEEG), surface electrodes (grid or strip electrodes) or combined depth and surface electrodes were tailored to individual patients.

Imaging acquisition and processing
Structural MRI datasets were collected preoperatively and at a minimum follow-up of 3 months on a 1.5 T or a 3 T MR imaging scanner (General Electric). Three-month postoperative T 1 MRIs were used to exclude immediate postoperative brain swelling and brain-shift.

Tractwise and atlas-based disconnectome analysis
First, we computed the disconnectome map of each patient using the BCBtoolkit (http://www.bcblab.com). This map shows, for each voxel, the probability of having disconnected white matter fibres passing through it, based on a healthy population tractography dataset resulting in the associated disconnectome maps (tractwise voxel-based analysis). 31,32 This is achieved by computing a tractogram of fibres passing through the cavity in each of 20 unrelated healthy right-handed adults from the 7 T dataset of the Human Connectome Project, processed using High Angular Resolution Diffusion Imaging tractography (spherical deconvolution). For each patient lesion, 20 tractograms are generated and overlayed to produce a probabilistic map that takes inter-individual variability into account. A sample of 20 was chosen as this accounts for more than 80% of shared variance in the overall population. 31 Specific tract disconnection was then assessed using white matter atlases 33,34 (atlas-based analysis) in Tractotron.
For disconnectome maps and tract disconnection from each cavity, a probability of disconnection above the level of chance was selected, as previously. 32 Each patient's disconnection profile was used to investigate if disruption of specific white matter pathways were associated with seizure outcome. In the tractwise voxelbased analysis, non-parametric two sample t-tests were performed on disconnection maps using FSL's Randomise with 5000 permutations, family-wise error-correction and threshold-free cluster enhancement 35 to determine which disconnection profiles were associated with seizure outcome at the different time-points. In the atlas-based analysis, a chi-square analysis with Bonferroni correction was performed using tract disconnection (disconnected versus non-disconnected) in Tractotron and seizure outcome (seizure-free versus seizure-relapsing) according to the side of resection (left-sided tracts were evaluated in left-sided resections and right-sided tracts in right hemisphere resection). A flow-chart describing the disconnectome analysis can be found in Supplementary Fig. 1 and in previous reports. 36,37 Tractography-based disconnectome analysis Preoperative tractography was acquired in seventeen of the operated patients. Tracts were selected for dissection according to the results of the previous disconnectome analysis, with dissection performed in each patient's DWI native space. 38 Preoperative MRI, postoperative MRI and the resection cavity were coregistered to the preoperative diffusion data using the Elastix registration toolbox (https://github.com/lassoan/SlicerElastix) in 3D Slicer. Virtual dissection for white matter pathways was performed manually as detailed in Rojkova and colleagues 33 using a constrained region of interest-based approach. The number of streamlines that were intersecting with the resection cavity was used to calculate a percentage of disconnection for each relevant tract. An independent t-test was performed to test percentage of disconnection associated with seizure outcome.
A secondary tractography analysis was performed on the same normative dataset of 20 healthy subjects. After relevant white matter tracts were dissected in each subject, percentages of disconnection were calculated with the patients' resection cavities normalized in MNI space. Percentages of tract disconnection were then averaged according to each of the 47 resection cavities and independent t-test was performed to test whether percentage of disconnection was associated with seizure outcome.

Neuropsychological assessment
Expressive language (Graded Naming Test), 39 executive function (phonemic fluency and semantic fluency) 40,41 and memory function (BIRT Memory and Information Processing Battery (BMIPB) 42 were assessed preoperatively and 12 months after surgery. The Graded Naming Test consists of oral naming of 30 black and white pictures. 39 In phonemic and semantic fluency 40,41 tasks, the subject produces as many words as possible in 1 min according to category (phonemic fluency: words starting with the letter 'S'; semantic fluency: 'animals'). In the memory task, participants are read a list of 15 words and then are asked to recall them. The task comprises five trials with performance measured by the sum of recalled words. 42 To evaluate neuropsychological outcome we used reliable change indices (RCIs) using a 90% confidence interval [43][44][45] and grouping performance binarily (deteriorated or not).

Statistical analysis
Statistical analysis was performed using SPSS Statistics (28, IBM, Armonk, NY, USA). Normality of variables distribution were evaluated using a Kolmogorov-Smirnov test. A chi-square analysis with Bonferroni correction was used to evaluate the association between seizure outcome and histology or focal-to-bilateral tonic-clonic seizures in the different timepoints. An independent two sample t-test was used to test whether resection volumes were associated seizure outcome. An atlas-based chisquare analysis with Bonferroni correction was used to evaluate tracts involved in seizure outcome or neuropsychological outcome at the different time-points. Non-parametric voxel-based two-sample t-tests were performed with FSL's Randomise using threshold-free cluster enhancement and corrected for family-wise error rate 35 to evaluate each voxel involved in seizure outcome at the different time-points. Independent two sample t-tests were used to assess how percentage of tract disconnection was associated with seizure outcome or neuropsychological outcome.

Data availability
Anonymized data that support the findings of this study are available from the corresponding authors upon reasonable request.

Presurgical evaluation
Data on presurgical non-invasive and invasive investigations are summarized in Table 2. Example surgical cases are included in the Supplementary material. In total, 13/47 (27%) of patients underwent resection without invasive recordings. Of these, 11 underwent fMRI, 5 FDG-PET, 3 ictal SPECT and 1 MEG. Thirty-four (73%) underwent invasive recording (10 patients had SEEG, 9 had surface electrodes in the form of grid or strip electrodes and 15 had combined surface and depth electrodes). Of these, 30 underwent fMRI, 22

Disconnectome analysis Tractwise voxel-based disconnectome analysis
A non-parametric two-sample t-test showed that after 3 years, seizure freedom was associated with disconnection of fibres involving the left anterior limb of the internal capsule, including the anterior thalamus, mediodorsal thalamus and striatum, the subthalamic nucleus and the dorsomesial, dorsolateral and ventrolateral prefrontal cortices ( Fig. 2A). A similar result was also present at 5 years (Fig. 2B).
To rule out whether worse long-term seizure outcome was caused by partial resection, we performed a second analysis to test the influence of cortico-thalamostriatal disconnection on longterm seizure outcome in those 29 patients for whom total resection of the lesion was achieved (18/29 patients at the 5-year follow-up). At the 3-year follow-up, only one of the 12 patients in whom the anterior thalamocortical projection was resected had seizure

Tractography-based disconnectome analysis
First, we performed tractography dissections in a subgroup of 17 patients who also underwent preoperative tractography to examine patterns of disconnection linked to patient-specific anatomy.
Results from the tractography analysis confirmed those from the voxel-based and atlas-based analyses: at three years, patients with anterior thalamic disconnection showed a trend towards better seizure outcomes (11/17 patients were seizure free) with an average of 41% of streamlines resected in the seizure-free group versus 16% resected in those with seizure recurrence [t(15) = 2.08; P < 0.056], which was significant if the five patients with partial resections were excluded [8/12 patients seizure free; 41% versus 10% streamlines resected in seizure-free/seizure-recurrence groups, t(10) = 2.44; P < 0.035]. Only five patients had tractography at the 5-year follow-up, precluding statistical analysis. We next ran a similar tractography analysis using normative data. This also showed disconnection of the anterior thalamic radiation was significantly associated with seizure outcome after 3 years [85% versus 64% streamlines resected in seizure-free/ seizure-recurrence groups; t(45) = 3.125; P < 0.003] and after 5 years [87% versus 61% streamlines resected in seizure-free/seizurerecurrence groups; t(31) = 3.142; P < 0.004]. Disconnection of anterior corticostriatal projections was significantly associated with Figure 2 Tractwise voxel-based analysis of disconnectome maps. (A) A non-parametric two sample t-test shows that disconnection of thalamic, putaminal and subthalamic regions as well as in the anterior prefrontal cortex is associated with a better seizure outcome 3 years after surgery. (B) A nonparametric two sample t-test shows that disconnection of thalamic, putaminal and subthalamic fibres projecting to the anterior prefrontal cortex is associated with a better seizure outcome 5 years after surgery. Cortical terminations: ALIC = anterior limb of the internal capsule; pTRI = pars triangularis; mSFG = middle superior frontal gyrus. seizure freedom at the 3 year follow-up [74% versus 54% streamlines resected in seizure-free/seizure-recurrence groups; t(45) = 2.488; P < 0.017], and there was a trend towards a better seizure outcome at 5 years, although this did not reach significance [72% versus 55% streamlines resected in seizure-free/seizure-recurrence groups; t(31) = 1.684; P = 0.102]. Notably, disconnection of the anterior thalamic radiation was also associated with seizure freedom when selecting only patients who underwent complete resections [after 3 years: 88% versus 66% streamlines resected in seizure-free/ seizure-recurrence groups, t(27) = 3.470; P < 0.002; after 5 years: 88% versus 64% streamlines resected in seizure-free/seizure-recurrence groups, t(16) = 2.846; P < 0.012], with disconnection of anterior corticostriatal projections being significantly associated with seizure freedom at 3 years [76% versus 54% streamlines resected in seizure-free/ seizure-recurrence groups, t(27) = 2.114; P < 0.044] but only showing a trend towards a better seizure outcome at 5 years [69% versus 55% streamlines resected in seizure-free/seizure-recurrence groups, t(16) = 0.971; P = 0.346]. Importantly, there was no significant association between cortico-thalamostriatal disconnection and seizure outcome in patients with partial resections. As there was no significant difference in the amount of cortico-thalamostriatal streamlines disconnected between complete versus partial resections, these analyses may potentially suggest that disconnection supporting a better seizure outcome may not arise from just disconnection per se but may work combined with complete resections. Patterns of disconnection associated with long-term seizure outcome are displayed in Fig. 4, and data with disconnection in the individual patient and normative data are provided in Supplementary Tables 2 and 3.

Neuropsychological assessment
Neuropsychological data of our cohort of patients can be found in Table 3. We investigated whether disconnection of anterior thalamic and cortico-striatal projections was associated with a poorer neuropsychological outcome in patients tested for language, executive and memory function using atlas-and tractography-based disconnectome analyses. There was no significant association between disconnection of these pathways and a worse neuropsychological outcome in language, executive (phonemic and semantic fluency) and memory one year after surgery (Supplementary material).

Discussion
In models of focal drug-resistant epilepsy, the epileptogenic zone is envisioned to be a localized cortical region and surgery aims for complete removal of this region which should lead to seizure freedom. 46 However, data from neuromodulation studies in animals and humans, 9,14,17 electrographic patterns of seizure spread 5 and evidence from disconnective surgery 18,20,47 all indicate a wider epileptogenic network exists whereby white matter connections may facilitate seizures. 6,48 In the present study, we analysed retrospective data from 47 patients operated for frontal lobe epilepsy to investigate common patterns of disconnection that may be relevant for long-term seizure freedom. Our disconnectome analyses indicated that disconnection of anterior thalamic and corticostriatal projections is significantly associated with seizure freedom at 3 and 5 years, suggesting these white matter pathways may play a role in preventing delayed seizure recurrence. These findings indicate that cortico-thalamostriatal disconnection may be linked to longterm seizure freedom and are in agreement with previous evidence from animal 49 and stimulation studies, 50 which suggests that these fibres may modulate epileptogenesis. Our findings offer a novel framework to understand delayed seizure recurrence in focal frontal lobe epilepsy with direct implications for resective epilepsy surgery and long-term seizure freedom.

Long-term seizure freedom and disconnection of anterior cortico-thalamostriatal projection
Late seizure recurrence occurs frequently after epilepsy surgery. 1,4 This occurs predominantly within the first 5 years after surgery and is reportedly unrelated to the completeness of the lesion resection, 1 compared with early seizure recurrence within the first 6 months, which is linked to partial lesion resection. 1,4 This suggests structures outside the epileptogenic zone may have a role in supporting novel seizure generation. This hypothesis may be supported by our results which indicate that disconnecting anterior thalamocortical and anterior corticostriatal projections may contribute to prevent seizure recurrence after 3 and 5 years. A role for thalamocortical connections in supporting seizure spread has long been established, 10 and thalamocortical dysregulation has been shown to be critical in focal-to-bilateral tonic-clonic seizures, 11 absence seizures 51,52 and other seizure types in genetic generalized epilepsies. 14,53,54 Extra-thalamic projections have been proposed to impact seizure severity, with the basal ganglia-including the subthalamic nucleus 8 -possibly involved in modulating seizure activity. 8,9 Stimulation of the anterior thalamic nucleus (ANT) in generalized and focal epilepsy may decrease or abolish seizures. 16,17 Stimulation of diffuse thalamocortical projections through the centromedian nucleus (CM) modulates cortical activity 15 and may decrease seizure frequency and severity. 14 Projections within the anterior thalamic radiation may overlap with thalamic projections of the ANT, the CM and the medio-dorsal nucleus of the thalamus, as well as with striatal projections from the subthalamic nucleus, putamen and caudate to the orbitofrontal, dorsolateral and ventrolateral prefrontal cortex. 6,55,56 Thus, it is tempting to speculate that disconnection of anterior thalamic and striatal projections may affect the same networks modulated in deep brain stimulation. Modulatory networks arising from the thalamus and other subcortical regions may thus play a role not only in seizure severity and maintenance of epileptic activity, 6,14,17 but also in delayed seizure recurrence.

Surgical relevance
Our results suggest that complete resection of the epileptogenic zone alone may be insufficient if coexisting networks involved in seizure generation and propagation are not disconnected. This is relevant for future developments in epilepsy surgery, as it suggests that tailored disconnection together with complete resection may increase chances of long-term seizure freedom. It is important to note that disconnection alone was not associated with long-term seizure freedom. This may suggest that cortico-thalamostriatal disconnection may help to prevent seizure recurrence in individuals in whom the epileptogenic zone has been completely excised. 13 This approach may not only prevent alteration of the healthy brain 6 but also disrupt a reverberating network that may promote seizure recurrence. 13 Disconnection, however, can also cause dysfunction. Recent evidence from resections has highlighted that damage to the white matter, rather than the cortex, is more likely to cause permanent neurological deficits. [57][58][59] Therefore, if disconnection may be advocated on the basis of a better seizure outcome, the latter has to be balanced with the need to preserve function. 60 Disconnection of anterior thalamic or corticostriatal projections was not associated with language, executive and memory deficits in this cohort. However, a more comprehensive neuropsychological and neuropsychiatric battery may be required to fully assess the effect of this pattern of disconnection, particularly when considering that some of these pathways, such as the anterior thalamic radiation, are critical components of the Papez circuit that mediate memory and emotion 61,62 and their disconnection has been linked to cognitive control deficits. 36 Accordingly, while our results suggest that disconnection should be considered together with lesion resection in frontal lobe epilepsy surgery, planned disconnection of these pathways must be tailored to each patient's history, neuropsychological data and expectations, prioritizing the need to offer the optimal balance between seizure control and preservation of function.

Limitations
This study has a number of limitations. Our cohort of 47 patients is heterogeneous in terms of aetiology, and confirmation of our results is needed in a larger cohort, although this is the largest sample with advanced imaging analysis that has assessed long-term seizure outcome in frontal lobe epilepsy. Similarly, different pathologies may impact differently on brain reorganization, 45 and future work has to disentangle cognitive alterations specific to distinct etiological and pathological entities.

Conclusions
Epilepsy is increasingly considered to be a network disorder. In this study, we show that disconnection beyond ablation of the epileptogenic zone may prevent delayed seizure recurrence, and cortico-thalamic and cortico-striatal disconnection may play a critical role in retaining seizure freedom in those patients with complete resection of the epileptogenic lesion. These connections may constitute an epileptogenic network common to patients with frontal lobe epilepsy, irrespective of lesion location. Tailored disconnection of this network, taking into account the individual functional status and cognitive reserve, may complement extant surgical resection practices and lead to improved seizure outcomes.