Normalization of neuronal metabolic dysfunction after surgery for temporal lobe epilepsy

Many patients with temporal lobe epilepsy (TLE) who do not respond well to treatment suffer progressive memory as well as more diffuse cognitive impairment associated with a concurrent progressive increase of bilateral epileptiform discharges.^1-4 These observations suggest that focal epileptic discharges can eventually lead to neuronal dysfunction remote from the seizure focus, which may in turn contribute to further impairment of temporal lobe functions.

Proton magnetic resonance spectra of normal human brain at long echo times reveal three major resonances: one at 3.2 ppm, which arises primarily from choline-containing phospholipids (Cho); one at 3.0 ppm, which arises from creatine (including phosphocreatine) (Cr); and a large signal at 2.0 ppm from N-acetyl groups that originates largely from N-acetylaspartate(NAA), a compound localized exclusively in neurons and neuronal processes in adult brain.^5,6 Proton MR spectroscopic imaging(MRSI) studies have shown focal reductions of the NAA signal in one or both temporal lobes of patients with TLE.^7-11 Decreases in NAA correlate strongly with the lateralization of TLE by EEG, although bilateral NAA abnormalities are more frequent than bilateral MRI abnormalities or bilateral independent seizure onsets on EEG.^7,9,11 The decrease in NAA associated with TLE (and other chronic CNS disorders)^10,12 is often interpreted as a sign of irreversible neuronal loss. However, observations have shown that decreases of NAA are reversible in some disorders such as demyelinating lesions and mitochondrial encephalopathies,^13-15 suggesting that decreased NAA may also reflect sublethal neuronal damage or perhaps a neuronal response to functional or metabolic stresses.¹⁶ A recent report showed significant improvement toward normal Cr/NAA levels in the unoperated temporal lobe of two patients with TLE who became seizure free after surgery.¹⁷

We performed this study to determine whether there were changes in NAA, Cr, or Cho resonance intensities in the temporal lobes of patients with TLE after they underwent surgery, and whether these changes correlated with the surgical outcome.

Methods. Selection and evaluation of patients. In this prospective study we performed proton MRSI before and after temporal lobe surgery in 14 patients with medically refractory TLE. Magnetic resonance spectroscopic imaging results were compared with a group of 21 age-matched normal control subjects. Informed consent was obtained from all patients and healthy volunteers. This study was part of an ongoing research project approved by the Ethics Review Committee of the Montreal Neurological Institute and Hospital.

From our series of consecutive patients with nonlesional TLE who underwent MRSI examinations before surgery, we performed follow-up MRSIs on the first seven patients who were seizure free and the first seven who were improved but continued to have seizures (three patients had class II and four patients had class III outcomes according to Engel's classification¹⁸) for more than 1 year after surgery.

Accurate identification of the type and localization of seizures was determined by a comprehensive evaluation, including a detailed history and neurologic examination, serial EEGs with sphenoidal electrodes, intensive video EEG telemetry for recording of seizures, neuropsychological assessment, and MRI.¹⁹ Based on previous studies at our institution,^20,21 ictal EEG findings were classified as (1) unilateral left or right, if >90% of ictal onsets were unilateral or clearly lateralized to one or the other side respectively, (2) left > right or right > left, if >70% of ictal onsets lateralized to the left or right side; or (3) bilateral (left equals right) if <70% of ictal onsets lateralized to one side. The localization of interictal epileptiform abnormalities was based on the site of maximum spike voltage on referential montage or phase reversals on bipolar montage, and classified as unilateral or bilateral as described. The electroencephalographers were unaware of the MRSI results when they reported the EEG findings.

Magnetic resonance spectroscopic imaging. Proton MRSIs were acquired using a Philips 1.5-T combined imaging and spectroscopy system(Philips Medical Systems, Best, The Netherlands). After scout imaging in axial and sagittal planes, multislice spin-echo MRIs (TR, 2,000; TE, 30) were obtained in the transverse plane along the axis of the temporal lobes and in the coronal plane perpendicular to the axis of the sylvian fissure. A large region of interest (ROI), including both temporal lobes and excluding bone, was defined for selective excitation prior to phase encoding for the MRSI. The ROI was placed in a similar position for all examinations to cover the entire extent of both hippocampi. The size of the ROI varied, depending on the head size and shape: 85 to 100 mm left-right × 75 to 95 mm anteroposterior × 20 mm craniocaudal. A water-suppressed proton MRSI was obtained from that ROI (TR, 2,000; TE, 272; field of view[FOV], 250 × 250 mm; phase-encoding steps, 32 × 32). This was followed by a short version of the proton MRSI without water suppression(TR, 850; TE, 272; FOV, 250 × 250 mm; phase-encoding steps, 16 × 16). After zero filling the latter to 32 × 32 profiles, the water-suppressed MRSI was divided by the nonwater-suppressed MRSI. The resulting time domain signal was left shifted and subtracted from itself to improve water suppression.²² This procedure reduces the amplitude of water and nearby resonances, and results in relatively high ratios of NAA/Cr. This modulation was consistent for control and patient data. To enhance spectral resolution, a Lorentzian-to-gaussian transformation was applied prior to Fourier transformation in three dimensions. The nominal voxel size in plane was approximately 8 × 8 mm before and 12 × 12 mm after K-space filtering for a voxel volume of 1.2 to 2.4 cc.

Resonance intensities in individual spectra were determined by integration of peak areas using locally developed software. The values for NAA, Cho, Cr, and NAA/Cr were determined for the middle posterior region of each temporal lobe by averaging values from all spectra (22 ± 3) in these regions. These areas included part of the body and tail of the hippocampus, and portions of gray and white matter from the intermediate and posterior temporal lobe. This excluded the anterior temporal lobe (pole) and most of the middle posterior temporal neocortex, as well as the amygdala, the head, and the initial portion of the body of the hippocampus (which are the structures removed during surgery). The region was chosen to enable the comparison of the same region before and after the operation. Thus, in each patient the voxels were selected so that tissue to be resected was not included within the region studied in the preoperative MRSI(figure 1). Spectra were excluded from the analyses if they were artifactually broadened (i.e., full width at half maximum >10 Hz) or if Cho and Cr peaks were not resolved. The operator was "blinded" to the extent that he was unaware of the side of the seizure focus at the time of the initial data processing and analysis of the surgical outcome at the time of follow-up data processing.

The intensity ratio NAA/Cr was used to simplify quantitation across patients. The observations derived from the NAA/Cr ratio do not in themselves depend on Cr being a stable internal reference. However, if Cr is stable or only slightly increased in brain regions associated with epileptogenic damage,^7,10,11 then observed decreases in the ratio NAA/Cr can be interpreted in terms of neuronal or axonal damage. This is reasonable and useful for understanding the pathogenesis of TLE.^7-11 As an independent check on this assumption, we expressed the temporal lobe metabolite resonance intensities corrected (NAA_corr, Cho_corr, and Cr_corr) by multiplying each subject's observed signal intensities by the ratio of the normal control subject's mean to each subject's central brainstem Cr intensity(Cr_brainstem). The Cr_brainstem consisted of the average Cr from three to five voxels in the central brainstem (see figure 1). This normalizes for instrumental differences, assuming that Cr_brainstem in patients and normal subjects should be similar. This assumption is reasonable since (1) the brainstem is located in the center of the ROI and is not directly affected by the epileptogenic process as far as is known, and (2) NAA/Cr, NAA/Cho, NAA/Cho+Cr, and Cho/Cr values from the brainstem of patients and normal control subjects did not differ (ANOVA, p > 0.5).

Statistical analysis. We used the t-test to compare length of follow-up and interval from surgery with MRSI between patients who became seizure free and patients who were not seizure free. We used the chi-squared test to compare outcomes in relation to the type of surgical approach.

We performed one-way ANOVA to compare the resonance intensities of NAA_corr, Cho_corr, Cr_corr, and NAA/Cr in the temporal lobes for all groups at once. This we followed by post hoc comparisons of all possible pairs of groups using Tukey's test, which corrects appropriately for multiple comparisons.^23,24 We used the following groups: controls, left and right temporal lobes; patients seizure free after surgery, temporal lobe preoperatively-ipsilateral and contralateral, and postoperatively-ipsilateral and contralateral (to the side of surgery); and patients who were not seizure free after surgery, temporal lobe preoperatively-ipsilateral and contralateral, and postoperatively-ipsilateral and contralateral (to the side of surgery).

Statistical significance was c onsidered to be present for p< 0.05. In addition, we analyzed individual values from each patient for abnormalities. For this we considered values more than 2 SD from the mean of the control group to be abnormal.

Results. Table 1 summarizes MRSI measures and clinical data for each patient. Repeat examinations of normal control subjects as well as of patients with TLE with an interval of up to 18 months were reproducible with no statistically significant differences between paired examinations.

The mean time interval between surgery and postoperative MRSI was 20± 8.5 months. The mean length of postoperative follow-up was 37± 8.7 months. Patients underwent either selective amygdalohippocampectomy (n = 7) or anterior temporal lobe resection, including the amygdala, anterior hippocampus, and temporal pole neocortex (n= 7). There was no difference in outcome in relation to the type of surgical approach. There were no differences in length of follow-up or interval from surgery to MRSI between patients who became seizure free and those who were not seizure free.

Group comparisons. N-acetylaspartate/Cr values varied significantly among the 10 subgroups (as defined earlier) (p < 10^-14) (table 2). Pairwise post hoc comparisons showed that NAA/Cr values of normal control subjects from the right and left temporal lobes did not differ (p = 0.999). Preoperative NAA/Cr values were significantly lower in the temporal lobe ipsilateral to the main epileptic focus (and the side of subsequent surgery) for the patients compared with normal control subjects (p < 0.0001). Patients who continued to have seizures after surgery had preoperative NAA/Cr values significantly lower than control subjects in the temporal lobe contralateral to the surgery as well (p = 0.0002), but patients who became seizure free did not (p > 0.05). In keeping with this, preoperative studies of patients who became seizure free after surgery showed significantly higher NAA/Cr in the temporal lobe contralateral to the intended surgery compared with the ipsilateral side (p = 0.002), while the mean NAA/Cr values were not significantly different for the two sides in the group of patients who were not to become seizure free after surgery (p = 0.1). Postoperative NAA/Cr values from both the ipsilateral and contralateral temporal lobes of patients who became seizure free did not differ from control subjects (p > 0.3), but patients who did not become seizure free continued to have NAA/Cr values significantly lower than control subjects (p < 0.002) on both sides.

A comparison between postoperative versus preoperative MRSI showed a statistically significant increase of NAA/Cr values after surgery on the ipsilateral side (p = 0.0003) in the group that became seizure free. There was no consistent change in the NAA/Cr values for the group of patients who were not seizure free after surgery (p > 0.9).

Since a decrease in the NAA/Cr ratio could occur with changes in either metabolite signal (NAA or Cr), we used an intrinsic normalization procedure based on brainstem values (which we assume to be unaffected in TLE) to evaluate the changes in each metabolite signal intensity independently. Similar ANOVA and post hoc comparisons among the 10 subgroups were performed for the individual metabolite intensities "corrected" by the brainstem creatine values (NAA_corr, Cho_corr, and Cr_corr), as defined earlier. As suggested by the NAA/Cr metabolite ratios, we found significant differences between NAA_corr values among the patient subgroups (p < 0.0001). These post hoc comparisons yielded results similar to those from comparisons of NAA/Cr values, but with higher p values. There were no statistical differences for Cho_corr(p = 0.57) and Cr_corr (p = 0.09) group values (seetable 2).

Individual patient comparisons. The predictive value of the changes described for the patient groups was tested by comparison of data from individual patients with the mean values from the normal control group.

Preoperative MRSI examinations. N-acetylaspartate/Cr values were significantly decreased (more than 2 SD below the mean of the control subjects) ipsilateral to the side of surgery in all patients, except in one patient who did not become seizure free. N-acetylaspartate/Cr values were also decreased in the contralateral temporal lobe in five of seven patients who were to continue to have seizures after surgery and in one of seven patients who became seizure free (see table 1). Significantly low NAA_corr values were seen in the ipsilateral temporal lobe of nine of 14 patients, and in the contralateral temporal lobe of six of 14 patients.

Creatine_corr values were higher than normal (more than 2 SD from the mean of the normal control subjects) in the ipsilateral temporal lobe of six of 14 patients (three of whom became seizure free and three who did not). Creatine_corr was also high in the contralateral temporal lobe of one patient who did not become seizure free after surgery. The increases in Cr_corr values were far less pronounced than the magnitude of the decrease in NAA_corr values. This can be appreciated by the fact that the difference of Cr_corr values across groups did not reach statistical significance, and by the degree of increase (considering the number of SD) in the six patients who had this abnormality. By contrast, the NAA_corr abnormalities were more intense and more frequent, and were statistically significant across groups. Therefore we assumed that the increases in the Cr relative resonance intensity accounted for only a portion of the decrease in the NAA/Cr ratios. Choline_corr values in all patients were within 2 SD from the mean of the control subjects.

Postoperative MRSI examinations. N-acetylaspartate/Cr values in the temporal lobe ipsilateral to the surgery remained abnormally low in all patients who continued to have seizures, but increased into the normal range (i.e., within 2 SD of the mean of the control subjects) in all patients who became seizure free. The NAA/Cr values from the contralateral temporal lobe were abnormally low in five of seven of the patients who continued to have seizures and in none of those who became seizure free.

N-acetylaspartate_corr values remained abnormally low after surgery in the ipsilateral temporal lobe of four of seven patients who did not become seizure free postoperatively, but was abnormally low in only one of seven patients who became seizure free. N-acetylaspartate_corr values were significantly lower after surgery in the contralateral temporal lobe for only one of seven patients who continued to have seizures. Creatine_corr values remained high in three of seven patients who continued to have seizures (in both temporal lobes in one patients, in the contralateral temporal lobe in one patient, and in the ipsilateral temporal lobe in the third patient), and in none of the patients who became seizure free (see table 1). Choline_corr values in all patients were within 2 SD from the mean of the control subjects.Figure 2 shows representative spectra from a normal control subject (figure 2A), from a patient who became seizure free (figure 2B), and from patient who continued to have seizures (figure 2C) after surgery.

EEG findings. The localization and lateralization of the EEG findings are summarized in table 1. Patients with bilateral EEG abnormalities had more bilateral proton MRSI abnormalities than patients with unilateral interictal and ictal EEG abnormalities. Overall there was good concordance between localization and lateralization of EEG and MRSI findings, except for Patient 14 (see table 1). This patient had MRSI lateralization (both for the middle and posterior temporal regions) to the right temporal lobe, in addition to right hippocampal atrophy. However, scalp EEG investigation failed to lateralize, and intracranial EEG investigation revealed clear predominance of seizure onsets from the left mesial temporal structures. This patient underwent a left amygdalohippocampectomy with poor outcome after 22 months of follow-up. Her postoperative EEGs continued to show epileptiform abnormalities over both temporal lobes.

Discussion. We used proton MRSI to measure relative NAA, Cr, and Cho signal intensity in the temporal lobes of seven patients with TLE who continued to have seizures after surgical treatment and seven patients who became seizure free after surgery. N-acetylasparate/Cr resonance intensity was low preoperatively in one or both temporal lobes of all 14 individuals. In general it was low ipsilaterally in the patients who became seizure free and bilaterally in the patients who did not. Postoperatively it increased to the normal range on the side of surgery in all patients who became seizure free. In the one patient who became seizure free in whom NAA/Cr was low bilaterally before surgery, the NAA/Cr values in the contralateral temporal lobe also increased to the normal range. By contrast, NAA/Cr values did not change in the patients who continued having seizures after surgery. These findings are in keeping with a recent report by Hugg et al.,¹⁷ showing significant contralateral improvement toward normal Cr/NAA levels in two patients with TLE who became seizure free after surgery.

Low NAA cannot be explained solely by loss of mesial temporal neurons. Our results confirm that neuronal damage or dysfunction as determined by the relative NAA resonance intensity is widespread in the temporal lobe ipsilateral to the seizure focus in patients with TLE, that it extends beyond the regions of abnormality demonstrated by conventional MRI, and that it is frequently bilateral.^7-9,11,25 An important observation that expands that of Hugg et al.¹⁷ is that decreases in NAA can be reversible in both ipsilateral and contralateral temporal lobes after successful control of seizures by surgical removal of the epileptogenic area. The fact that resection of temporal lobe structures results in increases in relative NAA resonance intensity ipsilateral to the surgery clearly demonstrates that the decrease in NAA observed in patients with TLE cannot result only from Wallerian degeneration of axons as a consequence of mesial temporal sclerosis. This is consistent with previous clinical and experimental observations suggesting that seizure "foci" do not exist in isolation, but are associated with and may even depend on abnormal surrounding and possibly remote neurons.^1,26-28 Elimination of the "focus" may result in these surrounding and remote neurons becoming less abnormal.^27,29

As discussed by DeStefano et al.,¹³ three factors could be responsible for the reversible component of decreases in relative NAA signal intensity: (1) changes in the NMR relaxation times of the protons in NAA, (2) changes in the relative volume of NAA-containing neurons in the ROI, or (3) changes in the concentration of NAA in individual neurons. The first hypothesis that the observed changes in NAA are due to changes in NMR relaxation times seems unlikely, as large changes in relaxation times would need to occur selectively for NAA and not for Cr or Cho. (The absence of conventional image contrast changes shows that this does not occur for water protons.)

It could be asked whether the MRSI changes could be related only to changes in brain parenchyma due to surgically resected tissue, thus changing the relative volume of NAA-containing neurons in the ROI (second hypothesis). However, the increase of NAA/Cr intensity cannot be due to more normal brain tissue replacing resected tissue in the ROI because: (1) the postoperative increase in NAA/Cr occurred both ipsilateral and contralateral to the side of surgery; (2) it was significantly more pronounced in those patients who became seizure free, but small or absent in those who continued to have seizures; (3) the surgical approach and amount of tissue resected was similar in both groups of patients; and (4) the ROI contained the same structures pre- and postoperatively, since the middle and posterior temporal lobe structures in the ROI were behind the area of resection.

The hypothesis that functional changes in neurons affecting oxidative metabolism may give rise to reversible changes in the relative concentration of NAA therefore seems to be the most reasonable explanation for the recovery of NAA/Cr and NAA_corr This hypothesis is supported by clinical and experimental studies.^13,16,30-32 Axonal morphologic changes have been shown to accompany TLE.^26-29 A recent experimental study by Rango et al.³¹ has shown that deafferentation in the CNS may lead to a (possibly reversible) transsynaptic decrease of NAA.

Structural and functional changes associated with seizures lead to dynamic changes in NAA and Cr. The possibility that seizure activity leads to structural or functional changes and widespread depression of NAA/Cr measurements in the ipsilateral temporal lobe is important for understanding the emerging role of NAA/Cr in presurgical lateralization of TLE. Previous investigations^{7-9,11,25,33-35} have shown that quantification of asymmetric decreases in NAA/Cr between the two temporal lobes may be capable of correctly lateralizing the vast majority of patients with TLE being evaluated for surgery. The data reported here demonstrated that reduction in NAA/Cr ipsilateral to the epileptogenic focus in TLE reflects more than the(irreversible) pathologic changes associated with mesial temporal sclerosis. Since the decrease in NAA/Cr is largely reversible with cessation of seizures following surgical treatment, it follows that the seizure activity itself, or associated changes, must depress NAA/Cr also. This effect of seizures is crucial for understanding why asymmetrically low NAA/Cr in the temporal lobes is useful in determining the lateralization of temporal lobe seizures.

The total Cr signal can be used as an internal standard for normalization of signal intensities in pathologies where Cr remains constant. Creatine increased slightly in the temporal lobes of our patients with respect to brainstem Cr (which is unlikely to be affected by pathologic changes associated with TLE), which served to enhance the decreases in NAA/Cr associated with temporal lobe seizures. The increases in Cr were much too small to account for the entire observed decrease in NAA/Cr. Thus, decreases in NAA/Cr can still be interpreted as reflecting neuronal damage or dysfunction.

A modest increase in Cr has been reported previously in patients with TLE and interpreted to arise from irreversible gliotic changes.^7,8,10,11 Our observation of reversible changes in Cr suggests that the explanation may be more complex. Although dynamic changes in glial cell density are not impossible, the phenomenon would be without precedent to our knowledge. An alternative possibility is that changes in Cr reflect local alterations in glial or neuronal metabolism.³⁶

The MRSI protocol used in the present study provided a limited coverage of the cerebral hemispheres on each examination, and excluded metabolite ratios in the outer cortical regions (see figure 1A) to avoid artifacts from the skull. Potentially, the sensitivity of proton MRSI for the investigation of TLE, and other forms of epilepsy, could be enhanced by improvements in the technology, including three-dimensional MRSI, efficient lipid suppression, and fully automated postprocessing and quantification of MRSI data. Together with other MR techniques, proton MRSI is certain to improve the understanding of many epileptic syndromes and become central to guiding the management of some of them.

The observation that the relative NAA concentration in the temporal lobe increases after surgical therapy in patients who become seizure free indicates that decreased NAA ipsilateral to the seizure focus does not simply reflect loss of axons associated with mesial temporal sclerosis. Rather, decreases in NAA appear to provide a dynamic marker of physiologic responses to the primary epileptic activity, as well as to secondary epileptogenic damage or dysfunction.

Acknowledgment

D.L.A. is grateful to the Killam Trust for personal support.