Glucose and [¹¹C]flumazenil positron emission tomography abnormalities of thalamic nuclei in temporal lobe epilepsy

Patients and controls.

Twelve patients (6 men and 6 women, mean age 32.9 ± 12.4 years) with medically intractable TLE were included in the study (table 1). All patients underwent prolonged video-EEG recordings with scalp/sphenoidal electrodes. Ictal recordings were obtained in 11 patients, whereas only interictal recordings were available in 1 (Patient 10). MRI-based hippocampal and amygdala volumetry was performed in all cases using a previously described method.2,18 All patients underwent 2-deoxy-2-[¹⁸F]fluoro-d-glucose (FDG) as well as FMZ PET examination. Antiepileptic treatment at the time of the PET examinations included monotherapy or polytherapy with phenytoin (n = 5), carbamazepine (n = 5), lamotrigine (n = 6), gabapentin (n = 3), vigabatrin (n = 1), and valproate (n = 2).

The control group for defining the normal values for normalized glucose metabolic rates and metabolic asymmetries for the thalamic regions on FDG PET included six right-handed healthy volunteers (five women and one man, mean age 27.5 years, age range 23 to 39 years). The control group for FMZ PET also included six healthy volunteers (four men and two women, mean age 40.1 years, age range 30 to 50 years). This control group was used to define the normal values for thalamus volume as well as for normalized FMZ binding and FMZ binding asymmetries for hippocampus, amygdala, and thalamic regions. The control subjects were not taking any medication and had no history of neurologic or psychiatric disorder. All of them had normal MRI scans.

MRI procedure.

MRI studies were performed on a GE 1.5-T Signa 5.4 unit (GE Medical Systems, Milwaukee, WI). Volumetric imaging was performed using a spoiled gradient-echo (SPGR) sequence. The three-dimensional SPGR sequence generates 124 contiguous 1.5-mm sections of the entire head using a 35/5/1 (TR/TE/NEX) pulse sequence, flip angle of 35 degrees, matrix size of 256 × 256, and field of view of 240 mm. These images were obtained in the coronal plane, and the imaging time for this sequence was 9.5 minutes.

PET scanning protocol.

All FMZ PET studies were performed in accordance with policies of the Wayne State University Institutional Review Board, whereas FDG PET studies were performed for clinical indications as part of the presurgical evaluation. PET studies were performed using the CTI/Siemens EXACT/HR (Knoxville, TN) whole-body positron tomograph located at Children’s Hospital of Michigan, Detroit. This scanner has a 15-cm field of view and generates 47 image planes with a slice thickness of 3.125 mm. The reconstructed image in-plane resolution obtained is 6.5 ± 0.35 mm at full-width-at-half-maximum (FWHM) and 7.0 ± 0.53 mm in the axial direction for the FMZ PET, and 5.5 ± 0.35 mm at FWHM and 6.0 ± 0.49 mm in the axial direction for the FDG PET (reconstruction parameters: Shepp-Logan filter with 1.1 cycles/cm cutoff frequency and Hanning filter with 0.20 cycles/pixel cutoff frequency). For FMZ PET, attenuation correction was performed on all images using data from a 15-minute transmission scan of the head. For FDG PET, computed attenuation correction was applied according to the method of Bergstrom et al.19

Patients fasted for 4 hours before PET studies. Surface EEG electrodes were placed according to the International 10-20 system, and EEG was monitored throughout the PET examinations. A venous line was established for injection of FDG (0.143 mCi/kg) or FMZ (0.4 mCi/kg as a slow bolus over 2 minutes using a Harvard pump) produced using a Siemens RDS-11 cyclotron (Knoxville, TN). External stimuli were minimized by dimming the lights and discouraging interaction so that studies reflected the resting awake state. All patients had their PET performed in the interictal state. For the FMZ PET, a 60-minute dynamic PET scan of the brain was performed (sequence: 4 × 30 seconds, 3 × 60 seconds, 2 × 150 seconds, 2 × 300 seconds, 4 × 600 seconds), beginning at the time of injection. Summed images representing activity concentration between 10 and 20 minutes were used to display benzodiazepine receptor binding in brain. For the FDG PET, a 20-minutes static emission scan was initiated 40 minutes after tracer injection.

MRI/PET coregistration and partial-volume correction of PET images.

Matching of PET and MR image volumes was performed as previously described,17 using a multipurpose three-dimensional registration technique (MPItool) developed by the Max-Planck-Institute (Cologne, Germany).20 After converting MRI data to PET data format, the PET image volume was coregistered with the MR image volume using MPItool, and a new image volume was created with image planes corresponding to the original MR image planes. The coregistered PET and MR image volumes were transferred to an SGI OCTANE (Silicon Graphics, Mountain View, CA) workstation, and partial volume correction of the PET images was performed using an established method.17

Regions of interest.

Brain regions of interest (ROIs) for the hippocampus and amygdala were manually defined on the MRI according to a protocol described previously.2,18 ROIs were drawn for left and right hippocampus and amygdala in all planes clearly showing these structures.

In addition, ROIs for the thalamus and thalamic regions were drawn on the MRI. The whole thalamus was defined as the portion of the diencephalon located superior to the interventricular foramen and the hypothalamic sulcus. The landmarks for this region consisted of the plane of the interventricular foramen anteriorly, the wall of the third ventricle medially, the inferior margin of the central part of the lateral ventricle superiorly, the lateral margin of the thalamic gray matter laterally, the posterior margin of the pulvinar posteriorly, and the interventricular foramen, hypothalamic sulcus, and midbrain structures inferiorly. The ROIs for the DMN were drawn following the border of the whole-thalamus ROI (figure 1A). The initial section used to define the anterior boundary of the DMN was the third or fourth section posterior to the interventricular foramen, thus allowing 3 to 4.5 mm of the anterior thalamus for the anterior nucleus of the thalamus. Thereafter, the outline of the lateral margin of the DMN was drawn parallel to the lateral margin of the thalamus at half the distance from the midline to the lateral thalamic border. The posterior boundary of the DMN was defined as the section in which the body of the fornix is formed by the joining of the two crura of the fornix. This was usually between three to five sections (4.5 to 7.5 mm) anterior to the posterior boundary of the pulvinar. The LAT was defined as the area between the lateral margin of the DMN and the lateral margin of the thalamus. The LAT drawn this way included the ventral anterior, ventral lateral, ventral posterolateral, and ventral posteromedial thalamic nuclei.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1. Regions of interest (ROIs) drawn on the MRI using contrast adjustment for the thalamus (TH), dorsomedial nucleus (DMN), and hippocampus (HC) in a patient (no. 1) with right hippocampal atrophy (A). These ROIs were superimposed on the corresponding coregistered [¹¹C]-flumazenil (FMZ) PET image (B). The figure shows a prominent binding asymmetry in the HC and DMN, with lower binding on the right side (thick arrows).

Mean hemispheric activity concentration values were obtained by drawing hemispheric ROIs (including gray and white matter) directly on the partial–volume-corrected PET images to calculate normalized glucose metabolic rates as well as normalized FMZ binding for the thalamic regions. In patients with TLE, the hemisphere contralateral to the epileptic focus was used for normalization. All ROIs of all patients and control subjects were drawn by the same person (C.J.).

Thalamus volumetric measurement.

Once the ROIs for the whole thalamus had been defined, the volumes were calculated by multiplying the number of voxels in the ROI by the voxel volume to give a total volume of thalamus in cubic millimeters. The absolute volumes, thus obtained, were then “normalized” by correcting them for individual variation in head size. Each volume was presented as the ratio of the thalamic volume to the total intracranial volume (in percent).

For the assessment of thalamic volume asymmetries we used an asymmetry index (AI), that was calculated as follows: where V_C and V_I are the volumes (in cubic millimeters) of the thalamus contralateral and ipsilateral to the seizure focus, respectively.

PET image analysis.

Study design and statistical analysis.

Statistical analysis was performed using StatView statistical package (BrainPower, Inc., Calabasas, CA). Initially, we tested whether significant differences could be detected between left and right thalamic volumes, normalized glucose metabolic rates, and FMZ binding in the control group using Student’s t-test for paired comparisons. If no significant differences were found for a specific variable, the left and right values were pooled; otherwise, left and right values were analyzed separately. The whole thalamus and the three thalamic regions in each hemisphere obtained from the patient group were separated into those regions ipsilateral and those contralateral to the seizure focus. These two sets of regions were then compared with the corresponding thalamic regions obtained from the control group.

Group comparisons of mean thalamic volumes, normalized glucose metabolic rates, and FMZ binding between the three sets of regions were performed using analysis of variance. Furthermore, the left/right asymmetries for the aforementioned variables were tested using Student’s t-test for paired comparisons separately within each group. The normal AI limits were established for the hippocampus, amygdala, and thalamic regions, computed as 2.5 SD above the mean AI value obtained in the control group. To determine if regional asymmetries of the measured variables were abnormal in the individual patients, the AI values were compared with the previously established normal AI limit. Multiple regression was used to assess a possible correlation between age or epilepsy duration (independent variables) and the normalized MRI thalamic volumes, glucose metabolic rates, and FMZ binding of the thalamic regions, as well as the corresponding AIs for these variables (dependent variables). Similarly, assuming that glucose metabolic rates and MRI volumes are not independent, their thalamic AIs were compared to thalamic FMZ binding AIs for possible correlations using multiple regression. In addition, to assess a possible relationship between mediotemporal and thalamic FMZ binding abnormalities, regression coefficients were obtained for FMZ binding AIs derived from amygdala and hippocampus (independent variables) and the three thalamic regions (dependent variables). The significance of each correlation was assessed using the F-test. A p value less than 0.05 was considered significant after correction for multiple comparisons by using Bonferroni’s method.

Thalamus volumetric findings.

Mean normalized volumes of the right and left thalami in the control subjects (n = 6) were not statistically different (0.362% ± 0.019% and 0.362% ± 0.016%, respectively), which allowed us to pool data from right and left to obtain a mean value. The normalized volumes of the thalami ipsilateral to the seizure focus of patients with TLE (0.327% ± 0.033%) proved to be significantly lower than the mean volume of the control subjects (p = 0.028). Comparison of normalized volumes of the contralateral thalami (0.354% ± 0.041%) did not differ significantly from the control mean volume (p = 0.57). Paired comparison showed significantly smaller volumes on the side of the focus compared with the contralateral side (p = 0.0002).

The mean AI for the normalized volumes of the thalami in control subjects was 2.32% ± 1.76%, and thus the upper limit of normal volume asymmetry (mean ± 2.5 SD) was 6.72%. Six of 12 patients had abnormally high AIs, always with smaller volumes on the side of the focus (table 1).

Thalamic glucose metabolism.

In the control group, the normalized glucose metabolic rates of the right and left DMN and LAT were not significantly different. Thus, left and right values were averaged for comparison with the patient group. In contrast, the normalized glucose metabolic rate of the left pulvinar was significantly higher than that of the right pulvinar for control subjects (p = 0.005); therefore, we did not combine these values but used them separately to compare them with the corresponding values of patients with right (n = 6) and left (n = 6) epileptic foci. Normalized glucose metabolic rates for patients with TLE were significantly higher in the LAT bilaterally (p = 0.01 and p = 0.001 ipsilateral and contralateral to the focus, respectively) compared with the control subjects. The normalized glucose metabolic values of the other thalamic regions (DMN and pulvinar) were not significantly different from those of the control subjects. Paired comparisons showed significant metabolic asymmetry only in the DMN, with lower glucose metabolism on the side of the epileptic focus (p = 0.047).

In the control subjects, the upper limit of the AI was 6.3% for the DMN, 10.5% for the LAT, and 10.2% for the pulvinar. We found a high variability in the AI for the LAT, although there was no left/right bias. Abnormal thalamic AIs of glucose metabolism with lower values on the side of the seizure focus occurred in at least one thalamic region in 5 of the 12 patients (table 2, figure 2A). An abnormally high AI with lower glucose metabolism contralateral to the seizure focus occurred in the pulvinar of one patient (no. 12).

• Download figure
• Open in new tab
• Download powerpoint

Figure 2. 2-Deoxy-2-[¹⁸F]fluoro-d-glucose (FDG) and [¹¹C]-flumazenil (FMZ) PET images of Patient 10, who had marked left hippocampal and mild left amygdaloid atrophy. FDG PET (A) shows hypometabolism in the left lateral temporal cortex (arrows), whereas the hippocampal abnormality was less pronounced. An abnormal asymmetry of glucose metabolism was detected in the dorsomedial thalamic nucleus (DMN; asymmetry index [AI]: 16.0%; lower on the left). FMZ PET (B) showed abnormal asymmetry in the hippocampus (thin arrow; AI: 21.4%), the amygdala (AI: 15.6%), and the DMN (thick arrow; AI: 7.5%), with lower values on the left side.

[¹¹C]Flumazenil binding in thalamus.

In the control group, normalized FMZ binding in the right and left DMN, LAT, and pulvinar was not significantly different. Therefore, we again combined the values for the right and left regions, and the combined means were compared with the corresponding values determined from the patients. The mean FMZ binding of the DMN ipsilateral to the epileptic focus was significantly lower in the patients compared with control subjects (p = 0.045; table 3). In contrast, the mean FMZ binding of the LAT contralateral to the focus was significantly higher than in the control group (p = 0.031), whereas FMZ binding in the ipsilateral LAT showed a tendency toward increase (p = 0.059). FMZ binding in other thalamic regions (pulvinar ipsilateral and contralateral to the focus, as well as the DMN contralateral to the focus) did not differ significantly from the control values (all p values > 0.43). Paired comparisons showed that FMZ binding in the DMN was significantly lower on the side of the epileptic focus compared with the corresponding contralateral values (p = 0.0018).

The upper limit of the FMZ binding AI in the control group was 6.6% for the DMN, 11.3% for the LAT, and 11.3% for the pulvinar. AIs in the DMN were abnormally high in seven patients (table 1, figure 2B), including 3 patients with abnormal pulvinar asymmetry (the lower values appeared on the side of the focus in all of these cases).

There was a significant correlation between the AIs of the DMN and amygdala (r = 0.77, p = 0.009), but not be-tween those of the amygdala and other thalamic regions (p = 0.95 and 0.09 for the LAT and pulvinar, respectively). There was no correlation between the FMZ binding AI in hippocampus and thalamus (p = 0.14, 0.88, 0.58 for the DMN, LAT and pulvinar, respectively) as well as between that in hippocampus and amygdala (p = 0.17). There was no correlation between volumetric or glucose metabolic AIs and FMZ binding AIs of any thalamic regions (p values between 0.23 and 0.83).

Methodologic considerations.

A number of methodologic issues in our study should be addressed. Our patient group included people with medically intractable TLE who had concordant EEG, MRI, and PET evidence of unilateral temporal lobe seizure foci. Although extratemporal neocortical FDG (n = 5) and FMZ (n = 1) PET abnormalities were seen in some patients on visual PET evaluation, it is well known that such alterations can be found commonly in mesial TLE,3 and may represent remote effects rather than primary epileptogenic areas. Although four patients also had temporal lesions on MRI in addition to hippocampal atrophy, these were associated with ipsilateral hippocampal atrophy as well as mesiobasal seizure onset on ictal EEG. It is known that mesial and lateral temporal areas have dense, reciprocal connections with each other, and ictal discharges commonly propagate from one to the other. Thus, the individual contributions of these portions of the temporal lobe on the extratemporal functional abnormalities cannot be distinguished. Therefore, we believe that despite the relatively wide range of PET and MRI abnormalities of our patients, they represent a reasonably homogeneous population for evaluating the effect of TLE on thalamic glucose metabolism and benzodiazepine receptor binding.

The thalamic ROIs were drawn on high-resolution MRI using a protocol we believe is reliable for outlining the borders of the three thalamic regions, including the DMN. The signal intensity of this major thalamic nucleus is also different from the surrounding areas on the MRI (figure 1). Both the region of the pulvinar and the LAT include several smaller nuclei that could not be identified separately. The anterior thalamus could be drawn typically in only two or three MRI planes and showed very low activity on PET images, with high interindividual variability. Thus, although we included this region in the volumetric measurement, we excluded it from PET analysis.

Visual PET evaluation was done on non–partial-volume-corrected images and failed to find thalamic asymmetries in several cases. For FMZ PET, the low thalamic activity made it even more difficult to assess localized asymmetries; this may explain the low sensitivity of visual evaluation of thalamic abnormalities (table 1). These findings further emphasize the necessity of partial volume correction in the analysis of small brain structures. The volume of thalamic regions (usually ranging between 1 and 2 cm³) is comparable with that of the hippocampal subregions that have been identified reliably in previous PET studies.17 The low interindividual variabilities resulting from the use of normalized glucose metabolic rates and FMZ binding instead of absolute measures were favorable for group comparisons. We used mean hemispheric activity values for normalization obtained from the hemisphere contralateral to the focus because it would be less likely to be affected by localized metabolic or benzodiazepine receptor changes.

Glucose metabolism was higher in the left compared with the right pulvinar in all six normal subjects. The defined ROIs in this region might also include the lateral posterior nucleus, and this metabolic asymmetry might be attributed to the previously reported structural and functional asymmetry of the human posterior thalamus with left-sided dominance.22 This is thought to be caused by the asymmetric pattern of cortical connections in this thalamic region, which is probably related to the language specialization of the dominant thalamus.22 Thus, we accepted this normal asymmetry and did not combine the left and right metabolic values for group comparisons, but used them separately for statistical comparison with the pulvinar of patients with left- and right-sided TLE foci.

Our patients were taking anticonvulsant medication, some of which probably decreased global brain glucose metabolism.23,24 The use of normalized metabolic values, however, diminished this effect. It is also unlikely that general metabolic changes selectively affected some parts of the thalamus while leaving other regions unaffected. The metabolic asymmetries also cannot be attributed to drug effects. Thus, it is reasonable to assume that our findings reflect disease-related abnormalities rather than drug effects.

PET abnormalities in the dorsomedial nucleus of the thalamus.

One of the main findings in the current study is the demonstration of localized decreased FMZ binding in the DMN ipsilateral to the seizure focus in patients with TLE. This asymmetry has a strong lateralization value for the seizure focus and can be useful for clinical purposes: its presence may verify the side of the epileptic focus.

Altered benzodiazepine receptor binding in the thalamus has been reported previously in patients with generalized epilepsy25 and in TLE.3 Our findings show that decreased FMZ binding is common and can be especially prominent in the DMN. This abnormality was not detected using statistical parametric mapping,4 but recent studies using ROIs and correction for partial-volume effects proved that benzodiazepine binding abnormalities are not confined to the mediotemporal structures in TLE associated with hippocampal atrophy.17 The finding of thalamic volume loss is consistent with the recent findings of DeCarli et al.,24 who demonstrated thalamic atrophy in patients with complex partial seizures of left temporal origin. Although loss of GABA receptor–containing cells can be a potential cause of decreased benzodiazepine receptors in the DMN, the lack of correlation between the degree of thalamic volume loss and decreased FMZ binding suggests that these PET abnormalities cannot be explained simply by neuronal loss.

Our findings confirm that the thalamus, and especially the DMN, may play an important role in human TLE, one that probably involves propagation pathways and regulation of the spread of epileptic discharges.26 This is consistent with findings from animal studies showing that the blockade of excitatory transmission in the DMN blocks limbic motor seizures induced by pilocarpine.27 Furthermore, the amygdala is important for ipsilateral seizure propagation in the limbic system, including the DMN,12 and the correlation between amygdala and DMN FMZ binding abnormalities suggests that these structures interact strongly in human TLE. Although benzodiazepine receptor binding abnormalities in the DMN may be primary changes supporting seizure propagation, altered benzodiazepine receptor density can also be a response to recurrent abnormal electrical activity. In support of the second possibility, animal models of limbic seizures demonstrated that prolonged seizures can cause progressive pathologic changes in the DMN,14,15 similar to the thalamic changes in patients dying after prolonged status epilepticus.16

The thalamic glucose metabolic asymmetry is consistent with findings from several previous studies showing lower glucose metabolism in the thalamus ipsilateral to the epileptic focus.3,28 A very recent study29 showed that hippocampal cell loss causing decreased efferent neuronal activity correlates with thalamus hypometabolism in TLE, thus supporting a primary role for mesiotemporal structures in functional changes of the thalamus. Although this correlation was seen bilaterally, our results show that thalamic asymmetry of glucose metabolism can be more widespread than the FMZ binding abnormality and can affect any part of the thalamus. Nevertheless, our findings strongly support the pivotal position of the DMN in the limbic seizure network of human TLE.

Increased [¹¹C]flumazenil binding and metabolism in the lateral thalamus.

Increased glucose metabolism and FMZ binding was confined to the LAT. Although widespread hypermetabolism can occur during seizures or in the postictal period,30 all of our PET studies were performed in the interictal state as verified by EEG, and opposite changes in other thalamic regions (see earlier) also suggested that these increases could not be attributed to transient ictal or postictal alterations. Although the increased metabolism and FMZ binding in the LAT was an unexpected finding, previous studies using various animal models of epilepsy have shown that repeated seizures can induce an increased number of benzodiazepine receptor binding sites in different parts of the brain. For example, increased GABA/benzodiazepine receptor number in dentate granule cells of the hippocampus31 after repeated seizures is associated with a long-lasting augmentation of messenger RNAs of different GABA(A) receptor subunits.32 An increased density of γ-hydoxybutyrate binding sites was also reported in the lateral thalamus in an animal model of absence seizures.33 The lateral thalamus in humans consists of several functionally different nuclei, including the thalamic relay nuclei, and it appears to be preferentially involved in the generation of spike and wave discharges.34 This thalamic region is also a part of an intrathalamic GABAergic inhibitory circuit that modulates the strength of afferent input into the thalamus.35 In animal studies, both complex partial and generalized convulsive seizures could be elicited by injecting carbachol into the lateral but not into the medial thalamus.36 In another study, amygdala kindling resulted in signs of long-term membrane remodeling that continued even 2 weeks postseizure in the lateral dorsal thalamus,37 suggesting long-lasting changes in brain morphology and chemistry in this part of the brain of kindled animals. Although these changes were unilateral, it was suggested that contralateral changes may occur at later stages of kindling, as had been demonstrated with regard to c-fos expression during evolution of kindling.38 In patients with TLE, it can be hypothesized that the concomitant FMZ binding increase and hypermetabolism of the LAT indicate enhanced synaptic activity due to an upregulated GABAergic system. These increases could represent an activated inhibitory circuit as a response to the epileptic process, but also a compensatory receptor upregulation due to the progressive loss of afferent thalamic GABAergic input. The existence of such compensation was suggested by previous studies showing interictal hypermetabolism in the temporal lobe contralateral to the seizure focus.39 Moreover, Franceschi et al.40 reported bilateral interictal hypermetabolism in the cortex and in subcortical structures in drug-naive patients with TLE. A recent study also demonstrated that remodeling of neuronal circuits can spread beyond the sclerotic hippocampus and is related to the duration of epilepsy.41

Our findings provide in vivo evidence that different thalamic nuclei undergo different functional alterations in human TLE. These changes are especially important when considering the role of the thalamus in regulation of cortical excitability and seizure propagation. Whether these patterns of abnormal glucose metabolism and benzodiazepine receptor binding in the thalamus are specific for TLE or also are present in extratemporal epilepsies requires further study.