2.1 Participants

In the patient group, we included data from the database of patients with drug-resistant HS who underwent antero-medial temporal lobe resection (AMTR) at the Brno Epilepsy Center between 2016 and 2020. The diagnosis was based on presurgical evaluation including long-term video-EEG monitoring, neuropsychological assessment, MRI, and FDG–PET. MRI scans were performed with a Siemens 3-T MRI scanner. There were 47 patients (31 female, 16 male). The mean age of the patients was 40.5 (±11.3) years. None of the subjects had major visual or cognitive impairment, except for a memory deficit reflecting mesiotemporal impairment. All patients had histopathological evidence of HS; three of them had associated focal cortical dysplasia in the temporal pole. Only patients with a follow-up of more than 2 years were considered for analysis. Patients were categorized as seizure-free (ILAE 1) or non-seizure-free (ILAE 2–5) based on the outcome. Patient characteristics are listed in Table 1: Clinical data collected included age at epilepsy onset, epilepsy duration, and age at the time of examination. The control group (HC) included data from the database of patients with oncology. This group comprised 25 individuals (14 female, 11 male) who underwent a standard FDG–PET examination as part of the oncology screening protocol; all of them had normal PET findings. The mean age of the HC at the time of imaging was 53 (±18) years. The HC had no neurological or psychiatric disorders and had no pathological findings on their MRI scans. They were not taking any medications that affect brain metabolism.

2.2 Clinical and MR data characteristics

For the assessment of structural MRI data MRI, the approach outlined by Chassoix et al.²⁴ was employed. Two trained neurologists, who were blinded to clinical data, visually analyzed the patients' MRI scans to determine the extent of hippocampal atrophy (mild, moderate, or severe), temporal lobe atrophy, and signal changes (or blurring between gray and white matter) in the temporal pole. The same methodology was used for defining of electroclinical patterns of patient's seizures: pure mesial group (M) without evidence of early spread beyond the temporal lobe; widespread mesio-lateral group (WML) with wide spread of seizures to both anterior and posterior lateral temporal and perisylvian areas; antero-mesial group (AML) with anterior spread (involving the anterior lateral temporal cortex and insulo-fronto-opercular areas and bitemporal group (BT) with early contralateral temporal spread. For more details on semiological classification see Chassoix et al. study.²⁴ The classification of electroclinical patterns was based on video-EEG findings (both semiological and electrophysiological) and stereo-EEG data when available. For practical reasons, patients from the M group were compared with patients from other groups (WML, AML, BT).

Differences in clinical data between study groups (left HS vs. right HS, and ILAE1 vs. ILAE non-I) were analyzed using a Mann–Whitney test for numerical variables (age at onset, age at surgery, disease duration, seizure frequency, and number of ASMs). The Fisher exact test was used for binary variables such as sex, history of focal to the bilateral tonic–clonic seizures (FBTCSs), history of early brain trauma, febrile seizures, familial epilepsy, and outcome. The Chi-Square test was used for nominal variables such as electroclinical pattern, HS degree, temporal atrophy, and polar signal abnormality with a significance level of P < 0.05.

2.3 PET acquisition and processing

Patients were imaged as outpatients in the interictal state. Subjects were prepared by fasting for 6 hours before the scan and resting in a quiet, darkened room for 50–60 minutes after FDG administration. EEG monitoring was not performed during the scan and patients were asked to report any seizures experienced on the day of the scan. The dose of FDG administered was 170–200 MBq per subject with no weight differentiation. The PET images were acquired using a Siemens mCT Flow PET/CT scanner in 3-D mode using the “Brain” protocol. The intrinsic spatial resolution of the scanner was 6.3 mm at full width at half maximum (FWHM) 1 cm from the center of the FOV, and 6.7 mm at FWHM 10 cm from the center of the FOV. The emission acquisition time in 3-D mode was 10 minutes. Forty-seven tomographic slices with a 3 mm slice thickness were reconstructed with a 128 × 128 iteration matrix with 6 iterations and 16 subsets, and a 6 mm FWHM Gaussian filter was applied. Spatial pre-processing and statistical analysis were performed using SPM12 (Wellcome Department of Cognitive Neurology, Institute of Neurology, University College London, UK). All the FDG–PET images were spatially normalized into the standardized stereotactic Montreal Neurological Institute space using our in-house FDG–PET template. This template was created following instructions introduced by Soma et al. and described in a study by Kojan et al.^{19, 20} A three-dimensional isotropic Gaussian kernel with 8 mm FWHM was used for smoothing all spatially normalized images. The regional standard uptake value (SUV) ratio was then calculated using the cerebellum as the reference area according to the following equation: SUVR = (SUV ROI)/(SUV total cerebellum).²¹ The subject's regional SUV ratio was extracted for 20 ROIs based on an automated anatomical labeling atlas 1 (AAL1).²²

The procedure described by Huang et al. was used to calculate the individual MC matrix.²³ Briefly, individual MC is calculated by combining group connectivity and an individual weighting matrix. The PET image is used to extract signals from selected areas based on the atlas, from which the mean metabolic activity and group correlation in the healthy control group are derived. The individual MC is obtained by combining the group connectivity with the individual weighting matrix, which is determined based on the difference between the individual metabolic activity and the mean group activity.

Individual connectivity matrices were compared between HC, left HS, right HS, seizure-free HS, and non-seizure-free HS groups using a two-sample t-test. Sex and age were included as nuisance variables. To facilitate comparison and interpretation of results, patient data were unilateralized, i.e., images of patients with right HS were flipped on the right–left axis, while the images of patients with left HS remained unchanged. In the unilateralized data, the hippocampus with the lesion was consistently on the left side. Multiple testing correction via false discovery rate (FDR) was performed in all the results.

To examine the correlation between clinical factors, including the epilepsy onset, duration, frequency of seizures, electroclinical patterns, the degree of hippocampal sclerosis (HS), temporal atrophy, and polar signal abnormalities, with MC values that were significantly different in the ILAE-1 and ILAE-non1 groups identified in the previous analysis, the Spearman correlation coefficient was employed.

3.1 Clinical, electroclinical, and neuroimaging data characteristics

3.2 Changes in MC in HS compared with HC

Compared to HC, patients with HS had significantly decreased MC in both temporal and extratemporal areas. No increase in MC was observed. In the temporal areas, decreased MC was found in the mesiotemporal structures, temporal pole, and temporal lateral and inferolateral cortex (inferior, middle, and superior temporal gyrus). In the mesiotemporal structures, the MC was decreased between both hippocampi and parahippocampal gyri. The most significant changes in MC were found in the ipsilesional (on the same side as the lesional hippocampus) inferolateral temporal cortex (inferior temporal gyrus), which showed decreased MC in both insulas, orbitofrontal cortices, and ipsilesional mesiotemporal structures. The MC of this area showed even more changes than the lesional hippocampus. Decreased MC was also found in the ipsilesional temporal pole, which decreased particularly in contralesional areas (mesiotemporal, temporal pole, and insula).

From the extratemporal areas, the majority of changes were seen in the orbitofrontal cortex, particularly ipsilesional. The ipsilesional orbitofrontal cortex showed decreased MC, particularly in the contralesional (contralateral to the lesional hippocampus) insula and the temporal inferolateral and lateral cortex (inferior, middle, and superior temporal gyrus). The contralesional orbitofrontal cortex showed decreased MC prevalently with the ipsilesional temporal lateral cortex. Finally, the operculo-insular regions showed most MC changes in both insulas.

In general, more extensive MC changes were found in the right HS. The main difference was found in the ipsilesional temporal lateral cortex. Decreased MC between the ipsilesional lateral (and inferolateral) temporal cortex and other areas (both insulas, orbitofrontal cortices, and contralateral temporal cortex) was more pronounced in right HS than in left HS. None of the MC changes correlated with clinical features. Significant MC changes in left HS and right HS as compared with HC are shown in Figures 1 and 2.

3.3 Changes in MC in HS dependent on surgical outcome

3.3.2 Decreased MC in seizure-free patients

From another perspective, the seizure-free patients showed significantly decreased MC (compared to the non-seizure-free patients) in both orbitofrontal cortices and the ipsilesional temporal lobe (temporal pole, inferior temporal cortex, and lateral cortex). Specifically, MC was decreased between the ipsilesional gyrus rectus and the ipsilesional temporal pole, the inferior and middle temporal gyrus, and between the contralesional gyrus rectus and the ipsilesional middle temporal gyrus. Significant changes in MC according to surgical outcome are summarized in Table 3 and Figure 3.

TABLE 3. Summary of results of intergroup metabolic connectivity correlated with surgical outcome.

Decreased MC
I Hippocampus	I Frontal operculum (inferior frontal gyrus – pars opercularis)
	C Insula
C Hippocampus	C Insula
Increased MC
I Lateral temporal cortex (middle temporal gyrus)	I Orbitofrontal cortex (gyrus rectus)
	C Orbitofrontal cortex (gyrus rectus)
I Inferolateral temporal cortex (middle temporal gyrus)	I Orbitofrontal cortex (gyrus rectus)
	C Orbitofrontal cortex (gyrus rectus)
Temporal pole (middle temporal gyrus – pole)	I Orbitofrontal cortex (gyrus rectus)

• Note: The significant differences in metabolic connectivity between seizure-free (ILAE 1) and non-seizure-free (ILAE non-1) patients with hippocampal sclerosis after surgery. Results after the unilateralization, depicted according to the side of the lesional hippocampus to the ipsilesional and contralesional side. Metabolic connectivity between pairs of regions of interest is shown as significantly decreased or increased in the non-seizure-free group as compared to the seizure-free group of patients.
• Abbreviations: C, contralesional hemisphere; I, ipsilesional hemisphere; MC, metabolic connectivity.

4.1 MC in HS

Both patients with left HS and right HS showed decreased MC between bilateral mesiotemporal structures (hippocampus or parahippocampal gyrus), probably reflecting the long-term effects of the epileptogenic hippocampus on the non-affected hippocampus, as other studies^{24, 27} have shown that bitemporal PET hypometabolism correlates with epilepsy duration. Decreased MC was also found in other areas, typically in the lesional hemisphere: temporal pole, lateral temporal cortex, orbitofrontal cortex, and especially in the inferior temporal gyrus. These are all areas of decreased metabolism in HS.^{17, 20, 24} In the contralateral hemisphere, most changes were found in the insula, orbitofrontal cortex, and temporal lateral cortex. The finding of an altered metabolic network reaching these contralesional areas is novel. However, it is not surprising as these are typical areas for the propagation of ictal activity in MTLE seizures and supports the findings of the FC studies showing widespread bilateral changes in patients with HS.²⁸

4.2 Different MC patterns in left and right HS

Our results showed that the changes in MC were more pronounced in right HS than in left HS. This asymmetry between left and right HS cannot be explained by asymmetric clinical features, as there were no significant differences in clinical features between these groups. Metabolic asymmetry in HS is a well-known phenomenon. The largest PET study investigating left HS and right HS separately showed that the metabolic changes (hypometabolism or hypermetabolism) are more widespread in right HS, which is in agreement with our study.¹⁷ However, another study showed the opposite results, with greater metabolic changes in left HS.²⁹ Functional network changes are typically greater in the left HS.^{10, 11} Regarding changes in MC, no study has shown such an asymmetry because previous studies have not compared left and right HS separately.^{8, 9} Compared to left HS, the most significant decrease in MC in right HS was found in the ipsilesional temporal lateral (especially inferolateral) cortex; other findings were bilateral insulas, orbitofrontal cortices, and contralateral temporal cortex. This is consistent with PET and MRI studies showing more pronounced changes in some of these regions in the right HS, but the clinical significance of these findings is unclear.^{12, 17} We did not find a worse surgical outcome in RTLE than in LTLE. Therefore, the clinical impact of more pronounced changes in the right HS is uncertain. In the unilateralized PET data, we found that surgical outcome correlated with decreased FC within a network of the hippocampi and operculoinsular regions. We could assume that the MC of these areas was not altered in RTLE (compared with LTLE) enough to affect the outcome.

4.3 Correlation of metabolic network with surgical outcome

Comparing non-seizure-free patients (ILAE 2–5) with seizure-free patients (ILAE 1) after surgery, unfavorable surgical outcomes correlated with significantly decreased MC of the lesional hippocampus and both the ipsilateral frontal operculum and the contralateral insula. It also correlated with decreased MC between the contralateral hippocampus and the contralateral insula. The insula is a typical area for the propagation of temporal seizures and PET hypometabolism in HS.^{17, 24} PET studies have shown that both ipsilesional and contralateral insula hypometabolism may be markers of unfavorable outcomes.^{7, 20} However, an SEEG study showed that insula involvement in temporal seizures was not a negative prognostic factor for surgical outcomes.³⁰ The change in MC could be a consequence of the preferential spread of epileptic activity or it could reflect the existence of a temporo-operculo-insular epileptogenic network in HS. The involvement of the contralateral hippocampus in the pathological metabolic network is predictable, as contralateral temporal hypometabolism has already been identified as a negative prognostic factor.²⁰ Such hypometabolism could also reflect the spread of epileptic activity contralaterally or the possible presence of bitemporal epilepsies among our patients.²⁴ Analogously, all of these changes in the metabolic network in non-seizure-free patients, extending beyond the pathological hippocampus to the orbitofrontal cortex and operculo-insular region, could also indicate the presence of temporal-plus or temporal neocortical epilepsy in our patients with MTLE.³¹ Our study shows that both the hippocampi and the operculo-insular regions could be part of a pathological metabolic network, probably reflecting a long-term preferential spread of epileptic activity. We hypothesize that standard AMTR did not sufficiently disrupt this network and thus caused the failure of the surgery. This hypothesis of long-term consequences of preferential spreading is supported by the finding that MC changes in the contralesional insula and hippocampus correlate with a longer duration of epilepsy. It also highlights the importance of early surgery in patients with DRE with HS to prevent possible pathological network changes.

A favorable surgical outcome correlated with a relatively decreased MC in non-seizure-free patients (compared to seizure-free patients) between both orbitofrontal cortices and the ipsilesional temporal lobe (temporal pole, temporal inferior, and lateral cortex). The correlation between relative hypometabolism and positive surgical outcomes in focal epilepsy is problematic because hypometabolism is the usual marker of an epileptogenic focus.^{5, 7, 32, 33} A better outcome has been described with relative hypometabolism in the ipsilesional orbitofrontal cortex and temporal pole. Dupont hypothesized that the hypometabolism of these structures reflects a preferential path for seizure propagation to these structures. After AMTR, this pathway is disrupted and therefore these patients are seizure-free.⁵ We confirm this hypothesis but add the temporal lateral and inferolateral temporal cortex involvement to this network. We also found that these changes are probably a long-term effect of such a preferential spread, as it positively correlates with the earlier epilepsy onset.

MC changes relevant to surgical outcome did not correlate with electroclinical patterns. According to PET studies, hypometabolism may differ between different electroclinical patterns.^{7, 24} The non-significant correlations in our study may be due to the small sample size. An alternative explanation could be that MC reflects more network changes than hypometabolism alone, making comparison with FC studies more appropriate. The correlation between semiology and network (FC changes) is problematic and still a matter of research.³⁴ We also found no correlation of MC changes relevant to surgical outcome with any of the structural characteristics. This result supports the findings of FC studies that network changes may be independent of structural abnormalities.^{35, 36}