Magnetic Resonance Spectroscopy in Epilepsy: Clinical Issues


Address correspondence and reprint requests to Dr. F. Cendes at Dept. Neurologia, FCM, UNICAMP, Cidade Universitária, Campinas, SP, Brazil 13083-970. E-mail:


Summary: Proton magnetic resonance spectroscopy (MRS) studies have shown focal reductions of N-acetyl aspartate (NAA) signal in patients with different forms of temporal lobe epilepsy (TLE), including those with normal magnetic resonance imaging (MRI), as well as extratemporal partial epilepsies. Both single-voxel and multivoxel 1H-MRS have high sensitivity for detecting low NAA, indicative of neuronal dysfunction in focal epilepsies. Decreases in NAA correlate strongly with EEG abnormalities and severity of cell loss, and may be a more sensitive measure than structural MRI. However, the NAA decrease is often more widespread than the epileptogenic focus. The results of published MRS studies suggest that in patients with partial epilepsy, there is a metabolic abnormality throughout the brain, with patterns of asymmetry and focal accentuation that are useful for noninvasive localization of epileptogenic foci. A major limitation of current proton MRS studies in epilepsy is the inability to cover the entire brain in a single acquisition, thus leading to major sampling bias. The area of maximal abnormality may reside farther away, even when there is an abnormality inside of the volume of interest used for that particular examination. Decreased NAA ratios are present even in children at the onset of their epilepsy, and evidence points to a gradual and progressive course of further reduction in NAA values. Conversely, the relative NAA concentration can recover ipsilaterally and contralaterally after successful resection of a temporal lobe focus. These observations, together with the fact of the often widespread NAA abnormality, must be taken into account for the correct and adequate interpretation of proton MRS studies in the assessment of partial epilepsies.

Magnetic resonance spectroscopy (MRS) permits chemically specific, noninvasive measurement of certain compounds in living tissue. MRS measurements in the brain of living animals became practical in 1980 (1), and the capability for human measurements was developed a few years later. By the early 1990s, applications of MRS to several neurologic disorders had been published (2–5).

The noninvasive nature of MRS means that repeated measurements can be made, so that kinetic and longitudinal studies are possible in a single subject. In addition, one can study human tissues that are inaccessible except by invasive techniques. In the human brain, phosphate energy stores, intracellular pH, lactate concentration, and the neuronal marker N-acetyl aspartate (NAA) are examples of MRS-measurable variables that are important for both clinical and scientific purposes and cannot be studied easily by using any other technique (2,4,6).


The preliminary studies using 31P-MRS have demonstrated increased inorganic phosphate (Pi) and decrease phosphomonoesters (PMEs) in the epileptogenic zone (7–9). An additional metabolite change may exist (increased pH) (7), but this remains controversial (8,9). From these preliminary findings, it appears that these changes can predict the side and site of seizure onset; however, these findings require verification in larger patient populations.


Water-suppressed, localized MR spectra of normal human brain at “long” echo times (TE, 136–272 ms) reveal four major resonances (3,4,6)(Fig. 1):

Figure 1.

Coronal T1-weighted inversion recovery magnetic resonance imaging (MRI) slices (top) and spectra from the right and left mid temporal lobes in a patient with left temporal lobe epilepsy with normal MRI, including volumetric measurements of hippocampi. Note the reduced N-acetyl aspartate (NAA) peak on the left temporal lobe (arrow). This patient underwent left temporal lobe resection and became seizure free after surgery. The postoperative histopathologic examination showed mild mesial temporal sclerosis.

  • • One at 3.2 ppm, which arises from tetramethyl-amines (mainly from choline-containing phospholipids) (Cho);
  • • One at 3.0 ppm, which arises primarily from creatine and phosphocreatine (Cr);
  • • One at 2.0 ppm which arises from N-acetyl groups (mainly NAA;
  • • One at 1.3 ppm, which arises from the methyl resonance of lactate, or, in certain pathologic conditions, methyl groups of lipids (4,10). Under normal physiologic conditions and with the small voxel size used for most in vivo MRS examinations of the adult human brain, the lactate peak is not visible above the baseline noise.

Multiple lines of evidence support the use of NAA as a neuronal marker. It is found exclusively in neurons and their processes in the mature brain (3,11,12). In human brain spectra, in vivo NAA is reduced in situations known to be associated with neuronal loss (2). When decreases in the relative NAA signal arise from neuronal or axonal degeneration, irreversible changes are expected. However, there have been observations of reversible decreases in NAA in a number of conditions, emphasizing that neuronal dysfunction or transient relative volume changes also can lead to decreased NAA (13–15). The ability to quantify specifically neuronal loss or damage is one of the most interesting potential applications of MRS in cerebral disorders.

Changes in the resonance intensity of Cho probably result mainly from increases in the steady-state levels of phosphocholine and glycerol-phosphocholine. These choline-containing membrane phospholipids are released during active myelin breakdown. Thus the resonance intensity of Cho increases in acute demyelinating lesions in humans (16), and in certain tumors, such as meningiomas (17).

Total Cr concentration is relatively constant throughout the brain and tends to be relatively resistant to change. However, large changes can be seen with destructive pathology such as malignant tumors (17,18). It is reasonable to use creatine as an internal standard to normalize resonance intensities of NAA and Cho to correct for artifactual variations in signal intensity due to magnetic field and radiofrequency inhomogeneity. However, this must be done with caution. Use of an external concentration reference can be reliable if factors such as radiofrequency field inhomogeneity and coil tuning and coupling can he adequately controlled (4,6).

Lactic acid is the end product of glycolysis and accumulates when oxidative metabolism cannot meet energy requirements. Elevation of lactic acid in cerebral neoplasms correlates approximately with relative rates of glucose uptake, for example. However, lactate is both intracellular and extracellular, and large amounts may be accumulated outside actively anaerobic tissue (e.g., in necrotic tissue or fluid-filled cysts) (3,19).


Animal studies

Brain lactate is elevated by seizure activity, as demonstrated by conventional biochemical studies of excised tissue. In vivo animal studies using proton MRS allowed more dynamic studies and a better appreciation of how persistent lactate elevation is after even brief convulsive seizure (20–22). Selective neuronal injury by kainate-induced status epilepticus in rats was associated with focal reduction of NAA, determined by proton MRS imaging (MRSI), even before T2-weighted MRI changes were observed (23).

Najm et al. (24) used proton MRS to identify specific in situ metabolic markers for seizures and seizure-induced neuronal damage in rat brains. They pretreated rats with placebo or cycloheximide 1 h before kainic acid injection and then scanned rat brains at the level of the hippocampus before, during, and 24 h after seizures. They found a significant increase in lactate ratios in kainic acid–treated rats during and 24 h after seizure onset; however, this increase was prevented by cycloheximide pretreatment. They suggest that in situ lactate increase is a marker of seizure-induced neuronal damage and that there is no significant increase of in situ lactate during seizures that do not lead to neuronal damage (24).

Human studies

An early use of proton MRS in the context of human epilepsy was analysis of extracts of samples from temporal lobe tissue resected for treatment of drug-resistant temporal lobe epilepsy (TLE) (25). This type of in vitro study is a useful source of information about concentrations of several human brain metabolites that are observable by in vivo MRS studies.

The first observations of elevated brain lactate (and also reduced NAA) in vivo by proton MRS associated with a seizure disorder was reported by Matthews et al. (26) in two patients with Rasmussen's syndrome. Another study suggests that lactate accumulation results from repetitive seizures rather than from the disease process in Rasmussen's syndrome (27).

Subsequent proton MRS studies have shown focal reductions of NAA signal in patients with nonlesional TLE (25,26,28–42) and extratemporal partial epilepsies (27,43,44) with good correlation with EEG abnormalities and severity of cell loss. The results of published MRS studies suggest that in patients with partial epilepsy, there is a metabolic abnormality throughout the brain, with patterns of asymmetry and focal accentuation that are useful for noninvasive localization of epileptogenic foci (25–39,43–46)(Table 1). The MRS findings may have prognostic value for seizure outcome as well (39,47,48).

Table 1.  Proton MRS studies in temporal lobe epilepsy
of pts
decreased NAA
Contralateral decreased
Lateralization in
agreement with EEG
Lateralization opposite
to the EEG
  • MRS, magnetic resonance spectroscopy; NAA, N = acetyl aspartate.

  • a

     Single-voxel MRS.

  • b

     The remaining patients had bilateral MRS abnormalities with no definite lateralization; all of whom had bitemporal EEG abnormalities.

  • c

     Patients with bitemporal temporal lobe epilepsy. One of them was operated on, on the side of the EEG lateralization, with poor outcome.

Achten et al. (45)25a24 (96%)18 (72%)24 (96%)
Breiter et al. (38)77 (100%)07 (100%)0
Cendes et al. (31)100100 (100%)54 (54%)84 (84%)b2 (2%)c
Connelly et al. (30)25a22 (88%)10 (40%)15 (60%)3 (12%)
Cross et al. (35)20a15 (75%)9 (45%)11 (55%)0
Duc et al. (76)11a11 (100%)11 (100%)0
Ende et al. (46)1616 (100%)8 (50%)16 (100%)0
Hetherington et al. (37)1010 (100%)4 (40%)10 (100%)0
Hugg et al. (29)88 (100%)8 (100%)0
Kuzniecky et al. (39)3029 (97%)7 (23%)29 (97%)1 (3%)
Ng et al. (36)2523 (92%)3 (12%)21 (84%)2 (8%)
Vainio et al. (34)7a7 (100%)7 (100%)0
Total284272 (95.8%)113 (43.8%)243 (85.6%)8 (3.1%)

In a series of 100 consecutive patients with TLE (31), the NAA/Cr values were abnormally low in at least one temporal lobe in all but one patient and were low bilaterally in 54%. The asymmetry between right and left sides of NAA/Cr lateralized 86 (92.5%) of 93 patients who had lateralization by ictal EEG. There were seven patients with no clear lateralization by EEG. The MRSI lateralization was ipsilateral to the EEG in all but two patients who had bilateral TLE and bilateral AM-HF atrophy greater on the same side as the MRSI. Twelve of 13 patients with normal MRI volumes of the hippocampus (MRI Vol) had a significant decrease of NAA/Cr within the mesial temporal lobe ipsilateral to the ictal EEG focus (Fig. 1). Seven of these underwent surgery, and the histopathologic examination showed mild mesial temporal sclerosis.

The previously mentioned study (31) showed a direct correlation between the measure of NAA/Cr and MRI Vol. However, another study with 33 patients showed no correlation between the measure of hippocampal specific Cr/NAA and MRI Vol (39). Methodologic and statistical regional analysis differences are most likely responsible for the discrepancy between studies. Some degree of disassociation between severity of NAA abnormality and hippocampal volume or cell loss is not unexpected. This includes the finding of abnormal NAA measures in normal-volume hippocampi (both ipsilateral and contralateral to site of seizure onset) and the reversibility or correction of NAA abnormalities in patients who become seizure free after surgery (14,15).

It has been demonstrated that the degree of relative hypometabolism on fluorodeoxyglucose–positron emission tomography (FDG-PET) does not correlate with either MRI Vol (49) or histopathologic measure of cell loss (50). Potentially, if NAA and glucose metabolism are related, the component of decreased NAA that is believed to be related to disturbances in the metabolism might be better understood. Here too, findings from different studies are conflicting. One study with single-voxel proton MRS (1H-MRS) measure of mesial temporal NAA/(Cho+Cr) showed a significant correlation with normalized glucose metabolism in midtemporal (combined mesial and lateral temporal cortex) and temporal polar regions (51). Another study restricted to sampling only the hippocampus showed no correlation between measurement of Cr/NAA and normalized hippocampal glucose uptake (52). One further study (45) did not look at a direct correlation between NAA measures and glucose metabolism, but rather examined the sensitivity in relationship of lateralized abnormalities in MTLE. They found that mesial temporal single-voxel MRS measurement of NAA/(Cho+Cr) was abnormal in 42 temporal lobes, whereas FDG-PET showed only 25 with decreased metabolism, a finding that demonstrates a higher sensitivity for 1H-MRS in detection of metabolic disturbance than that with FDG-PET. Again, differences in methods, and in particular, region of analysis, make it difficult to compare the findings between the studies.

Cendes et al. (33) showed a significant increase in lactate/creatine+phosphocreatine (lactate/Cr) values, during and soon after complex partial seizures (CPSs), but not during or soon after absence seizures associated with generalized epilepsy (33). In patients with TLE, the NAA resonance relative to Cr (NAA/Cr) was low in one or both temporal lobes, indicating neuronal loss or damage. This was not observed in patients with primary generalized epilepsy. The regions with abnormal lactate/Cr and NAA/Cr values corresponded to the epileptogenic focus as defined by clinical-EEG investigation. There was no change in the NAA/Cr values in the temporal lobes between the interictal, ictal, or postictal states, for both individual patients and intergroup comparisons. The normal lactate levels observed during and after nonconvulsive generalized seizures are in keeping with the lack of or minimal postictal confusion in absence attacks. The lack of an increase in lactate levels after absence attacks suggests that these seizures may be less likely to result in cellular damage than are the CPSs of TLE (33). Similar results have been reported by Maton et al. (53) in individual patients between interictal and postictal studies.

Proton MRS studies (26,27,33,36,53) indicate that (a) partial seizures are associated with abnormally high lactate levels, but absence seizures are not, and (b) no short-term changes of NAA occur during or soon after CPSs (33,53) or absence seizures (33). These findings (33) may be related to the lack of postictal confusion in patients with absence seizures, as well as with the more benign course of primary generalized epilepsy with nonconvulsive attacks.

The spatial relationship between the NAA decrease and the underlying mechanisms causing neuronal damage is unclear. It has been observed that the neuronal damage as measured by NAA can extend to areas at a distance from the lesion, and that the timing of the insult may contribute to the widespread neuronal damage (54). It remains to be seen if this is relevant for the severity of epilepsy.

The NAA signal is often used as a parameter of neuronal integrity. The side of maximal NAA reduction often coincides with the side of EEG abnormality. The relationship between spiking frequency and underlying neuronal function and epileptogenic state is unclear. Serles et al. (55) showed trend toward higher interictal spike frequencies on surface EEG in regions of pronounced neuronal metabolic damage or dysfunction. This suggests that both variables parallel an underlying pathologic substrate, although the pathophysiologic processes may be distinct. Peeling and Sutherland (56) used high-resolution 1H-MRS to determine the concentrations of several metabolites (lactate, alanine, NAA, γ-aminobutyrate (GABA), glutamate, aspartate, creatine, choline, taurine, inositol, and succinate) in tissue from patients undergoing surgical treatment for intractable epilepsy. They correlated the metabolite profiles with the results of histopathologic analysis of the excised tissue and the spike activity. Surprisingly, they found no differences in metabolite levels from tissue with active spiking or nonspiking neocortical sites. Conversely, Maton et al. (57) found that bilateral Cr/NAA abnormalities were 3 times as frequent as the detection of bitemporal interictal spikes. Thus as with PET and MRI Vol, measures of interictal NAA disturbances do not appear completely linked with the EEG measure of neurophysiologic epileptiform disturbance.

Proton MRS studies (14,15,58) have shown recovery of relative NAA either ipsilaterally or contralaterally after successful temporal lobe removal. This suggests that structural or functional changes associated with seizure activity may lead to depression of NAA in the ipsilateral or contralateral temporal lobe. This preliminary observation has potentially great significance for understanding the utility of imaging NAA in the presurgical lateralization of TLE, as it suggests that reduction in NAA reflects not only the sequelae of the initial injury to temporal lobe structures, but also an effect of the seizure activity itself (or other factors associated with the ongoing epileptic state). Still, what remains most interesting is the component of NAA decrease that is not directly related to neuronal loss, atrophy, glucose hypometabolism, and neurophysiologic epileptiform disturbances as discussed earlier. Discovering the causes of reversible decreases in NAA concentration holds great promise for better understanding cellular dysfunction associated with epilepsy.

It has been a matter of dispute whether recurrent seizures can cause neuronal loss in human TLE, and whether TLE is a progressive disease. Studies using MRSI have produced seemingly conflicting results. Vermathen et al. (59) studied a group of patients with non–temporal neocortical epilepsy and showed that hippocampal NAA/Cr was not reduced, in contrast to that in patients with unilateral TLE. They argued that seizures did not cause secondary hippocampal damage. Garcia et al. (60) found a negative correlation between NAA and seizure frequency in patients with both frontal and temporal epilepsy, although no correlation with duration. Tash et al. (61) found that ipsilateral and contralateral NAA/Cr were negatively correlated with duration of TLE. Frequency of CPSs was not correlated with MRS or MRI Vol. Patients with frequent generalized tonic–clonic seizures had lower NAA/Cr bilaterally and smaller hippocampal volumes ipsilaterally than did patients with none or rare generalized tonic–clonic seizures (61).

Preliminary data suggest that MRS may be useful for the evaluation of other forms of partial epilepsies (40,43,44,54), including those associated with cortical developmental malformations (CDMs) (41,42). Li et al. (41) demonstrated that different types of CDM show different degrees of decrease of NAA. In cortical dysplasia, the relative NAA signal was very low. This disorder appears to result from abnormal neuronal and glial cell differentiation and proliferation, and the lesion contains structurally abnormal neurons with abnormal synaptic activity and connectivity, thus explaining the reduced NAA values. In polymicrogyria, in which the NAA values were normal or slightly abnormal, the malformation is due to an abnormal cortical organization caused by a postmigrational insult, and the neurons are mature. Heterotopia consists of a large number of neurons that failed to initiate or complete the migration process. In heterotopia, because of the high number of neurons, one would expect a relative increase of the NAA signal. This assumption is based on histopathologic studies showing normal-appearing neurons and on early FDG-PET studies showing patterns of glucose uptake similar to those of normal cortex. However, proton MRS studies have shown NAA signal intensity to be variably normal or abnormal in patients with heterotopia. This suggests that at least some of these apparently normal neurons are dysfunctional.

In another study at high field (4.1 T), Kuzniecky et al. (42) found that patients with focal cortical dysplasia had significant metabolic abnormalities in correspondence with the structural lesions, whereas patients with heterotopia and polymicrogyria demonstrated no subcortical MRSI abnormalities. They showed significant correlation between the metabolic abnormalities and the frequency of seizures but not with the degree of interictal EEG discharges. Quantitative neuronal and glial cell counts revealed no statistically significant correlation between cell loss and the abnormal metabolic ratios in those who underwent surgery. Their findings suggest that MRSI-based metabolic abnormalities in patients with CDM are variable and are likely to be associated with complex cellular mechanisms involving the regulation of NAA, total Cr content, and Cho, and perhaps with seizure activity (42). Another study in a patient with a giant heterotopia (62) showed changes in NAA and creatine-phosphocreatine (Cr) levels, reflecting alterations in energy metabolism and neuronal dysfunction in the area of heterotopia and in regions of the ipsilateral hemisphere that appeared normal on MRI. It is still unclear whether the decreases in NAA are related only to the abnormal structure of the dysgenic cortex, or to the ongoing epileptogenicity, or to both. However, studies have failed to find a correlation between the degree and extent of EEG abnormalities and NAA values. The differences in relative NAA signals among the different types of CMD discussed earlier appear to reflect more the type of malformation than the amount of interictal EEG abnormality.


MRS has been applied to children with epilepsy for the past 6 years and is continuing to advance our understanding of pediatric epilepsy in three areas.

First, in localization-related epilepsies, early biochemical changes identified by MRS can provide localization of the “epileptogenic” focus and identification of subjects for potential surgical therapies (35,63). Single-volume MRS techniques have demonstrated changes in metabolites before anatomic changes are observed on MRI (64). MRSI is better suited to clinical application, and several investigators have identified metabolic changes within the suspected epileptic region as well as in neighboring areas. These changes have been correlated with both cognitive function and outcome. MRSI of GABA and glutamate have been developed (65), and application of these MRSI methods to localization-related epilepsies will provide further insight into their pathogenesis. Further prospective studies using MRSI on extratemporal epilepsies and cerebral dysgenesis are needed.

Second, in generalized epilepsies, MRS can identify progressive neurometabolic disorders as a cause of childhood epilepsies (66). Changes in GABA and other amino acid neurotransmitters are being identified and may have potential for segregating childhood epilepsies by their biochemical profiles (67,68). A recent study demonstrated a difference in brain GABA levels in children with idiopathic versus cryptogenic generalized epilepsies (67). Subjects with idiopathic generalized epilepsy did not have significantly lower cortical GABA levels than did controls [1.5 + 0.1 mM, n = 24 (controls) vs. 1.1 + 0.4 mM, n = 15; p = 0.2]. However, subjects with cryptogenic generalized epilepsy had significantly lower GABA levels in the occipital cortex than did controls (0.8 + 0.3, n = 12; p < 0.001). With recent advances in the genetics of epilepsies, MRS studies can now be combined with molecular methods to provide genotype–phenotype correlation and to begin to provide an understanding of the mechanisms of specific gene defects.

Third, MRS is particularly well suited to monitor antiepileptic therapy. Serial measurements of brain GABA and homocarnosine by MRS can both identify subjects who may potentially respond to drugs targeted at the GABAergic system and monitor their response to therapy (66,69). MRS is providing further insights into the mechanisms of action of the ketogenic diet. Both proton and phosphorus MRS have identified changes in neuroenergetics and other biochemical processes in the brain (70). MRS studies will continue to play an important role in our understanding of the mechanisms of actions of antiepileptic therapies and permit close monitoring of the response to these therapies in both adults and children.


MRS has the potential of serving as a surrogate marker for neuronal dysfunction in epilepsy independent of structural or electrophysiologic changes. Thus MRS is a tool of great interest to study neuropsychologic dysfunction associated with the epileptogenic process. A number of studies have begun to shed some important data on this subject.

In a study of 22 children with focal, intractable TLE, Gadian et al. (71) found that verbal IQ correlated selectively and significantly with left mesial temporal NAA/Cr+Cho ratios, whereas nonverbal IQ correlated selectively and significant with right mesial NAA/Cr+Cho. Martin et al. (72) later examined cognitive function and hippocampal Cr/NAA in 46 adults with focal TLE. In this series, left hippocampal Cr/NAA correlated significantly with measures of both episodic and semantic verbal memory, whereas right hippocampal Cr/NAA correlated with a measure of facial recognition. These studies conferred a “cognitive validity” to data acquired from 1H-MRS.

Three published studies used 1H-MRS to examine a specific cognitive process. Incisa della Rocchetta et al. (73) found that patients with right TLE and a contralateral (i.e., left temporal) abnormality on 1H-MRS performed significantly worse on measures of episodic verbal memory than did those patients with right onset but without a contralateral abnormality. These data provide a potential physiologic substrate for the seemingly paradoxic finding of verbal memory deficits in the patient with right TLE. Sawrie et al. (74) used 1H-MRS to examine the role of the speech-dominant hippocampus in visual confrontation naming, a cognitive ability thought to be underpinned primarily by the function of the lateral temporal lobe. Their results provide evidence that the speech-dominant hippocampus is a significant component of the overall neuroanatomic network of visual confrontation naming. Finally, Sawrie et al. (75) used 1H-MRS to examine the possibility of nonlinear trends in the relation between hippocampal Cr/NAA and verbal memory. The authors suggested the possibility of a physiologic reserve for memory function, in which one must reach a certain level of abnormality in hippocampal Cr/NAA before declining in memory. These findings become more interesting when considered within the context of recent evidence suggesting that NAA hippocampal abnormalities might be progressive (61), thereby slowly exhausting memory reserve through prolonged duration of illness.