INTRODUCTION
Neuropsychological impairment is an important co-morbidity of chronic epilepsy (Elger et al., 2004). Considerable research has examined the relationship between cognition and a variety of clinical factors including etiology, age of onset, seizure type and severity, duration, antiepilepsy medications, and other factors. (Aldenkamp & Arends, 2004; Dodrill, 2004; Helmstaedter & Kurthen, 2001; Jones-Gotman, 2000; Saling et al., 1993). In addition, modal cognitive profiles have been derived for various epilepsy syndromes (including mesial temporal lobe epilepsy), and efforts have been undertaken to identify the shared versus unique cognitive risks across epilepsy syndromes (Elger et al., 2004; Lassonde et al., 2000; Nolan et al., 2003). These approaches have provided insight into the influence of clinical seizure factors on cognition in epilepsy.
A yet untapped approach to understanding cognitive morbidity in epilepsy is taxonomic in nature. This involves addressing the question of whether empirically derived groupings of patients with similar profiles of cognitive function can be identified either within or across epilepsy syndromes. Taxonomies facilitate reliable clustering of individuals into meaningful groups, provide a common language and organizing influence in the field, and they set the stage for further investigation of clinical and neurobiological correlates. Significant progress in the epilepsies has resulted from efforts to identify and characterize the heterogeneity inherent in the disorder, perhaps the best example being the classification of seizures and epilepsy syndromes (Commission on Classification and Terminology of the ILAE, 1981, 1989; Duchowny & Harvey, 1996; Engel, 2001; Fisher et al., 2005; Luders et al., 1998).
To date, taxonomic approaches have rarely been used to advance understanding of the neurobehavioral complications of the epilepsies (Paradiso et al., 1994). That is, rather than grouping patients based on clinical seizure characteristics (e.g., seizure frequency) and examining the relationships of individual clinical seizure characteristics to cognition, one derives a grouping of patients based solely on their pattern of performance across several cognitive domains. Such an approach would identify distinct cognitive profile types, the relative proportion of patients expressing each profile type, and provide a basis for identifying the broad spectrum of demographic, clinical seizure, and other neurobiological features that may be associated with discrete cognitive typologies. This represents a different way of conceptualizing and investigating cognitive function in epilepsy.
In this study, we used cluster analytic techniques to characterize cognitive phenotypes in temporal lobe epilepsy. This objective was stimulated by our previous finding that, compared to healthy controls, patients with chronic temporal lobe epilepsy exhibited an “average” cognitive profile characterized by considerable generalized cognitive disruption, certainly more than would be expected for a syndrome of localization-related epilepsy (Oyegbile et al., 2004). However, within this context, inspection of individual subject profiles revealed considerable heterogeneity ranging from intact to severely impaired mentation, suggesting that finer groupings of neuropsychological status might exist and distinct profiles of clinical and neurobiological correlates identified. Similarly, clinical experience suggests that there may be considerable variation in mental status across persons within a specific epilepsy syndrome.
As will be shown, identification of cognitive phenotypes among patients with temporal lobe epilepsy is achievable. The identified cognitive phenotypes are distinct in nature and frequency of presentation, and are characterized by important differences in the profiles of demographic, clinical epilepsy, brain structure correlates, and cognitive course.
METHODS
Subjects
Research participants were patients with temporal lobe epilepsy (n = 96) and healthy controls (n = 82). Selection criteria for epilepsy participants have been described in detail previously (Oyegbile et al., 2004). Subjects were between 14–59 years of age with WAIS-III Full Scale > 69. Selection criteria for epilepsy patients included: (a) complex partial seizures of definite or probable temporal lobe origin based on consensus conference review of ictal and/or interictal EEG, clinical semiology, and clinical history, (b) no MRI abnormalities other than atrophy on clinical interpretation, and (c) no other neurological disorder. Controls were a friend or relative of epilepsy participants and met the following criteria: (a) no current substance abuse or medical or psychiatric condition that could affect cognitive functioning and (b) no episode of loss of consciousness greater than five minutes, developmental learning disorder, or repetition of a grade in school. This study was reviewed and approved by the University of Wisconsin School of Medicine and Public Health Human Subjects Research Committee. All subjects gave informed consent for participation in this research project.
Table 1 provides demographic and clinical seizure characteristics. As expected, the temporal lobe epilepsy group had a lower mean Full Scale IQ than controls (p < .001). There were no significant differences in other sociodemographic features, although epilepsy patients exhibited a trend of slightly older age (p = .07) and fewer years of education (p = .10). Subjects in the temporal lobe epilepsy group suffered from seizures of adolescent onset (mean = 14.8 years) and long duration (mean = 21.6 years). A significant minority (39%) of the temporal lobe epilepsy patients also experienced secondarily generalized seizures. Regarding treatment, the majority of patients (64%) were treated with polytherapy (48 subjects on two medications, 11 subjects on three medications, and 2 subjects on four medications). The minority of patients (36%) was treated with monotherapy including carbamazepine [n = 20], phenytoin [n = 4], sodium valproate [n = 1], gabapentine [n = 3], lamotrigine [n = 4], phenobarbital [n = 1], and topiramate [n = 2].
Demographic and clinical characteristics
Neuropsychological Assessment
All research participants were administered a comprehensive test battery that included standard clinical measures of intelligence (Pilgrim et al., 1999; Wechsler, 1997a), language (Benton et al., 1994; Goodglass & Kaplan, 1983), visuoperceptual/spatial skills (Benton et al., 1994), immediate and delayed verbal and visual memory (Wechsler, 1997b), executive functions (Kongs et al., 2000; Wechsler, 1997a), speeded psychomotor processing, and fine motor dexterity (Reitan & Wolfson, 1993). These tests were selected in order to provide representative coverage of the major domains of higher cognitive function, focusing on measures that are commonly used in clinical practice. Table 2 depicts the cognitive domains, cognitive abilities tested, the tests administered, and the nature of the dependent measure (i.e., correct items, errors, and time).
Neuropsychological Test Battery
Raw scores for all psychometric tests for the healthy controls were converted to adjusted (age, gender, and years of education) z-scores (mean = 0, SD = 1) using multiple regression techniques. There were a very small number of outliers, defined as exceeding +/–3 SD (11 of 1633 cells or 0.6%), representing an essentially equal number of both positive (n = 5) and negative (n = 6) values. These outliers were deleted, the regression equations recomputed for the controls and then applied to the raw scores of the epilepsy patients. Adjusted z-scores take into account potential meaningful variables that are associated with cognition. In addition, by placing all test scores on a common metric, relative performance across the various cognitive domains can be compared directly and definitions of abnormality (e.g., z-score ≥ −2.0) applied uniformly across tests. Cognitive domain scores were created by computing the average adjusted z-scores of tests falling within the designated cognitive domains (see Table 2).
MRI Volumetrics
Images were obtained on a 1.5 Tesla GE Signa MR scanner. Sequences acquired for each participant included: (1) T1-weighted, three-dimensional SPGR acquired with the following parameters: TE = 5, TR = 24, flip angle = 40, NEX = 2, FOV = 26, slice thickness = 1.5 mm, slice plane = coronal, matrix = 256 × 192; (2) Proton Density (PD), and (3) T2-weighted images acquired with the following parameters: TE = 36 msec (for PD) or 96 msec (for T2), TR = 3000 msec, NEX = 1, FOV = 26, slice thickness = 3.0 mm, slice plane = coronal, matrix = 256 × 192, and an echo train length = 8. MRIs were processed using a semi-automated software package [i.e., Brain Research: Analysis of Images, Networks, and Systems (BRAINS) (Andreasen et al., 1996; Harris et al., 1999; Magnotta et al., 2002; Magnotta et al., 1999)]. MR processing staff was blinded to the clinical, sociodemographic, and neuropsychological characteristics of the participants. Variables of interest were segmented volumes of total cerebral tissue as well as segmented volumes of total cerebral gray and white matter and CSF, and total hippocampal volume. A neural network application (Magnotta et al., 1999) was used to trace the hippocampus by using established guidelines (Pantel et al., 2000) with manual correction of the traces by a technician trained to criterion. Because all measurements were obtained in the image space of the subject and not normalized, all volumes were adjusted for total ICV.
Statistical Analyses
Cluster analysis was used to identify groups of participants based on their average z-scores across cognitive domains including intelligence, language, visuoperception, immediate memory, delayed memory, executive function, and motor/psychomotor skills. Cluster analysis constructs a sequence of partitions from an object set, whereby objects that are similar become associated with one another to form meaningful clusters (Borgen & Barnett, 1987). Cluster analysis has been used by researchers in diverse disciplines such as psychology, biology, sociology, economics, engineering, and business. This technique is particularly useful when the objective of the research is classification of objects according to natural relationships (Hair & Black, 1996). In this study, Ward's hierarchical agglomerative clustering method was used, with squared Euclidean distance as the index of pair-wise similarity–dissimilarity between participant profiles.
To identify an optimal grouping of participants in the clustering hierarchy, the agglomeration schedule was examined to find a late stage in the hierarchy, with a relatively small number of participant clusters, in which the error sum of squares coefficients increased dramatically at subsequent stages in the hierarchy after relatively small increases at previous stages (Berven & Hubert, 1977). The stage producing four clusters of participants had relatively small increases of 34, 56, and 60 in the error sum of squares from one stage to the next at the three stages preceding the four-cluster stage, compared with increases of 110, 118, and 678 at the three subsequent stages in the hierarchy. Thus, cluster homogeneity dropped substantially after the four-cluster stage, however, two of the clusters in the four-cluster solution had very small sample sizes (n = 11 and n = 16) and differed primarily in the severity of abnormality in executive and motor domains. In the three cluster solution, these two groups were merged to form a larger cluster. The three-cluster solution provided a reasonable compromise between maximizing cluster homogeneity, deriving a cluster solution resulting in groups of epilepsy patients that could be meaningfully interpreted and with appropriate sample sizes for subsequent statistical analyses.
RESULTS
The mean cognitive performance for the three cluster groups is provided in Table 3. A graphic representation of the cognitive profiles is presented in Figure 1. Mean performance for the controls (mean = 0 and SD = 1) is represented on the x-axis. A 1-way MANOVA was used to compare the three cluster groups across the seven cognitive domains. A significant effect of group was obtained, Hotelling T = 4.13, df = 21,491, p < .001, and univariate effects were significant across all cognitive domains. A description of the cognitive profile associated with each of the cluster groups is provided later.
Means and standard deviations of the cognitive measures for the clusters
Mean cluster performance across cognitive domains.
Cluster 1: minimally impaired group
Cluster 1 was composed of 44 participants (47% of temporal lobe group) who exhibited a pattern of minimal cognitive impairment compared to controls. There were no significant differences in intelligence, perception, and immediate memory. However, Cluster 1 exhibited statistically significant lower scores than controls in the domains of language (p = .002), delayed memory (p = .01), executive function (p = .025) and cognitive/psychomotor speed (p = .006). Within group pairwise contrasts indicated that the delayed memory (p = .011), language (p = .001) and executive function (p = .006) domains were significantly lower than mean IQ.
Cluster 2: memory impaired group
Cluster 2 was composed of 23 participants (24% of temporal lobe group) who exhibited prominent memory impairment (approximately two standard deviations below the average scores of the controls) in the context of mild-moderate cognitive impairment in the remaining domains. Although the Cluster 2 group obtained significantly lower scores on intelligence (p < .001) and visuoperception (p < .001) than the Cluster 1 group, memory (both immediate and delayed) was most significantly affected. Within group pairwise contrasts confirmed that only immediate (p = .034) and delayed (p = .023) memory were significantly lower than mean IQ. The memory domains were also significantly lower compared to all other cognitive domains. Furthermore, Cluster 2 subjects scored similar to Cluster 1 subjects on the executive and speed domains (p's > .05).
Cluster 3: memory, executive, and speed impaired group
Cluster 3 was composed of 27 participants (29% of temporal lobe group) who exhibited a pattern of moderate to severe cognitive impairment. Cluster 3 subjects performed significantly (p < .001) worse than controls across all cognitive domains, and also performed significantly worse than both Cluster 1 and 2 groups across all cognitive domains (all p's < .007). In the context of this generalized impairment, within group pairwise comparisons indicated that compared to IQ, Cluster 3 subjects exhibited significant impairments in memory (p < .004), speeded psychomotor ability (p < .001), and executive function (p < .001).
Cluster Characteristics
Demographic and clinical seizure features
The three cluster groups were compared on demographic characteristics (age, gender, and education), clinical epilepsy features (age of onset, duration of disorder, number of antiepilepsy medications), and ICV-adjusted quantitative MRI variables (total cerebral gray and white matter, total cerebral CSF, total hippocampal volume). These comparisons were conducted using one-way analysis of variance with post-hoc pair-wise comparisons. Table 4 provides group means for the demographic and clinical seizure variables, and Figure 2 provides a depiction of ICV adjusted volumetric measurements across groups.
Cluster characteristics
Mean ICV adjusted z-scores of quantitative volumetric measurements across cluster groups.
There were no significant differences between the three cluster groups on gender, age of epilepsy onset, education, and overall seizure frequency (daily, weekly, monthly, and yearly). A subset of the study sample underwent ictal monitoring, which identified patients with unilateral left (n = 24) or right (n = 21) temporal lobe onset. The distribution of the left and right temporal lobe groups across the cluster groups did not differ (χ2 = 3.6, df = 2, p = .16). However, significant group (cluster) effects were evident for age (F = 4.5, df = 2,91, p = .014), duration of epilepsy (F = 7.1, df = 2,91, p < .001), and number of AEDs (F = 3.4, df = 2,91, p = .037). Cluster 3 (Memory, Executive, and Speed impaired) was significantly older (p = .01), had a longer epilepsy duration (p = .001), and was taking more AEDs (p = .03) compared to Cluster 1 (Minimally impaired).
Nonsignificant trends were also evident for history of status epilepticus (χ2 = 4.6, df = 2, p = .09) with a smaller proportion of positive cases in Cluster 1 (5.5%) compared to Clusters 2 and 3 (21.7% and 23.8% respectively); history of an infectious initial precipitating injury (χ2 = 5.07, df = 2, p = .079) with a greater proportion of positive cases in Cluster 3 (23.8%) compared to Clusters 1 and 2 (14.7% and 5.5% respectively); and history of >50 lifetime generalized tonic-clonic seizure (χ2 = 4.38, df = 2, p = .11) with 22%, 33%, and 57% of subjects in Clusters 1 through 3 with such histories.
Quantitative MR volumetric characteristics Significant differences between the cluster groups were also evident on several MRI volumetric measures including total ICV adjusted cerebral tissue (F = 10.1, df = 3,139, p < .001), cerebral gray matter (F = 2.8, df = 3,138, p = .04), cerebral white matter (F = 8.9, df = 3,138, p = <.001), cerebral CSF (F = 12.3, df = 3, 139, p < .001) and total hippocampal volume (F = 15.5, df = 3, 131, p < .001).
Total cerebral tissue volume Cluster 2 (p = .004) and Cluster 3 (p < .001) but not Cluster 1 (p = .30) had significantly smaller total cerebral volumes compared to healthy controls. Within the epilepsy groups, Cluster 2 (p = .06) and Cluster 3 (p < .001) had smaller volumes than Cluster 1.
Total cerebral gray matter Clusters 2 (p = .06) and Cluster 3 (p = .07) but not Cluster 1 (p = .52) had a trend of smaller gray matter volumes compared to the healthy controls. Within the epilepsy groups both Cluster 2 (p = .028) and Cluster 3 (p = .032) had smaller volumes compared to Cluster 1.
Total cerebral white matter Cluster 1 (p = .031), Cluster 2 (p = .099), and Cluster 3 (p < .001) had smaller volumes compared to controls. Within the epilepsy groups, Cluster 3 (p = .004) but not Cluster 2 (p = .94) had a smaller volume than Cluster 1.
Total cerebral CSF volume Cluster 2 (p = .001) and Cluster 3 (p < .001) but not Cluster 1 (p = .23) had a larger CSF volume than the healthy controls. Within the epilepsy groups, both Cluster 2 (p = .02) and Cluster 3 (p < .001) had larger CSF volumes than Cluster 1.
Total hippocampal volume All clusters exhibited significantly smaller total hippocampal volume compared to controls (all p's < .008). Within the epilepsy groups, Cluster 1 had larger hippocampal volumes compared to Cluster 2 (p = .07) and Cluster 3 (p = .007).
Lobar volumetrics
Differences between cluster groups and controls in total lobar gray and white matter volumes were analyzed by MANCOVA with age and ICV as covariates. There was a significant overall effect of clusters, Hotelling T = .501, F = 2.69, df = 24,386, p < .001. Inspection of univariate effects revealed significant effects only for white matter in the frontal (p = .013), parietal, (p < .001), and temporal (p < .001) lobes. Post hoc pairwise comparisons (LSD) revealed that Clusters 1 and 2 exhibited a selective reduction in temporal lobe white matter compared to controls (p < .001), with no difference between them (p = .69). In contrast, Cluster 3 exhibited more diffuse reduction in white matter volume compared to controls including frontal (p = .002), temporal (p < .001), and parietal (p < .001) lobes. Cluster 3 also exhibited greater reduction in white matter compared to Cluster 1 in the parietal (p < .001) and temporal (p = .034) lobes and Cluster 2 in the parietal lobe (p = .009). In summary, Clusters 1 and 2 differed from controls only in temporal lobe white matter and did not differ from each other in any lobar region. Cluster 3 showed the temporal lobe effect compared to controls but with extension of white matter reduction into frontal and parietal lobe white matter.
Prospective cognitive course
We examined the implications of cluster membership for prospective cognitive course. Of the original sample, 45 epilepsy and 64 controls have completed prospective cognitive reassessment 4 years after baseline assessment. Regression based z-scores (Sawrie et al., 1996; Hermann et al., 1996; Martin et al., 2002) were calculated and the 3 cluster groups were compared by MANOVA. Negative z-scores reflect lower obtained than expected retest scores. All three cluster groups showed a poorer cognitive course compared to controls across the cognitive domains (Fig. 3). However, Cluster 3 exhibited a significantly poorer course than Clusters 1 and 2 across all cognitive domains except intelligence, whereas Clusters 1 and 2 did not differ from each other on any of the cognitive domains. Thus, there is an epilepsy effect compared to controls across domains with an additive cluster group effect. The cluster groupings therefore have predictive utility for cognitive prognosis.
Regression-based z-scores for change for the cluster groups.
Supplementary analyses
Initial grouping of tests into cognitive domains was determined clinically or rationally. A series of supplemental analyses were conducted using a quantitative approach to construction of cognitive domains with subsequent cluster analysis of derived scores. A principal components analysis with varimax rotation and Kaiser normalization was performed on the test measures and a three factor solution was derived (speed/nonverbal factor, memory factor, and intellectual/problem solving factor). It is worth noting that all memory indices loaded on the memory factor regardless of whether they tapped immediate versus delayed recall or verbal versus visual presentation. Subsequent cluster analysis of the factor scores resulted once again in a three group solution, again with a “normal” group, a primary memory impaired group, and a predominantly executive/problem speed impaired group, although there were differences in the proportion of patients represented in each group. Thus, there is considerable symmetry in cluster outcome whether the cognitive tests are reduced to domains rationally (on the basis of clinical groupings) versus quantitatively (factor analysis). In addition, external validation of the cluster solution, based on factor analysis of the test measures, revealed a very similar pattern of findings for demographic, clinical, volumetric, and prospective test findings. Limitations of this approach, however, include combining epilepsy and controls together to maintain optimal subject to test ratio and unusual loadings of tests within and between factor scores.
DISCUSSION
Patients with chronic temporal lobe epilepsy have been shown previously to exhibit more cognitive dysfunction than expected in association with a focal epileptogenic lesion. Specifically, a mean pattern of relatively generalized cognitive dysfunction has been described with poorer performance compared to controls across all tested cognitive domains including memory (Oyegbile et al., 2004). This same pattern of generally adversely affected cognition was demonstrated previously among patients with the syndrome of mesial temporal lobe epilepsy with histopathologically confirmed hippocampal sclerosis (Hermann et al., 1997). Whereas informative, characterization of the average neuropsychological profile of patients with chronic temporal lobe epilepsy does not provide insight into the possible distinct groupings or cognitive typologies that may exist within the overall group.
In this investigation, cluster analysis identified three distinct cognitive subgroups or phenotypes; (1) minimally impaired, (2) memory impaired, and (3) memory, executive, and speed impaired (Table 3 and Fig. 1). Furthermore, validation of these distinct groups was provided by comparisons across demographic factors, clinical seizure characteristics and quantitative MRI volumetric variables, and cognitive course (Table 4 and Figs. 2 and 3). Thus, distinct cognitive phenotypes characterized by the nature, pattern, and severity of cognitive complications were identified in this sample of temporal lobe epilepsy subjects and, to our knowledge, this represents the first empirical demonstration of distinct cognitive phenotypes in temporal lobe epilepsy.
Validation of these cognitive phenotypes was provided by examination of their profiles of demographic features (age), clinical seizure features (duration of epilepsy, AED polytherapy) and brain volumetrics (segmented whole brain and lobar tissue volumes, CSF, and hippocampus), and 4-year cognitive course. Table 3 and Figure 2 reveal that the most cognitively impaired group (Cluster 3) was older, had the longest duration of epilepsy, took more medications, had more abnormal brain volumes (total, white matter and CSF), and showed the most adverse cognitive course than the other groups, especially Cluster 1. There were also meaningful but statistically non-significant trends in regard to other clinical seizure features. Cluster 3 group had the highest proportions of patients with histories of >50 lifetime generalized tonic-clonic seizures, status epilepticus, and initial precipitating injuries. Thus, this appears to be a group that is most likely to have incurred an earlier neurodevelopmental insult along with a more protracted and severe course of epilepsy (Hermann et al., 2002). We have previously argued that it is important to consider the cognitive impairment of temporal lobe epilepsy patients in the context of a neurodevelopmental perspective, and this point appears to be best exemplified in Cluster 3. Age of onset of epilepsy did not differ significantly across the groups, but the current study sample focused on epilepsy subjects with childhood/adolescent onset and this restriction in range may have contributed to the lack of a significant effect. Overall, these resulting profiles of demographic, clinical, and volumetric factors provide a heuristic overview of the correlates of each of the cognitive subgroups in this taxonomy.
The issue of cognitive course and progression is a topic of considerable interest in epilepsy and these findings indicate that the prospective cognitive status is different for the three cluster groups over a 4-year interval. The group with the greatest cognitive impairment at baseline (Cluster 3) showed the greatest prospective cognitive decline, whereas the remaining cluster groups showed a smaller adverse effect over time. This pattern of findings differs from that observed following anterior temporal lobectomy for treatment of intractable seizures. In that situation, subjects with the least cognitive impairment in memory function have been reported to exhibit the greatest decline, whereas those with more impaired pre-surgical performance exhibit less post surgical decline.
It could be argued that the three cluster groups identified in the current study merely represent a continuum of severity of cognitive impairment rather than distinct cognitive phenotypes. We do not believe that the findings merely reflect an ordinal level of impairment effect. For example, Cluster 2 exhibits a distinct profile of marked memory impairment despite similar findings for demographic (age, education) and clinical seizure characteristics (duration, number of AED medications). Furthermore, whereas Clusters 1 and 2 show similar degrees of total cerebral white matter and gray matter volume abnormality, Cluster 2 (memory impaired) showed greater total hippocampus volume abnormality than Cluster 1. Further, Cluster 3, while generally cognitively impaired, exhibits especially impaired memory, executive, and speeded performances relative to other cognitive domains. Thus, there are rather selective and parallel cognition-brain volume patterns that characterize the differences between the cluster groups.
We have only examined patients with temporal lobe epilepsy and it is necessary to determine if a similar phenotype classification is evident in other epilepsy syndromes, or if there are different characteristic cognitive profiles. Whereas a predominantly memory impaired group was observed in this sample of subjects with temporal lobe epilepsy, it is conceivable that syndrome-specific typologies may be identified in other localization-related epilepsy (e.g., profiles of impaired executive function in frontal lobe epilepsy) and primary generalized epilepsies (Elger et al., 2004; Lassonde et al., 2000; Nolan et al., 2003).
More generally, these results suggest that it is possible to derive meaningful neurobehavioral phenotypes of patients with temporal lobe epilepsy. Classification systems (i.e., seizure syndromes) have served the epilepsies well, and cognitive and neurobehavioral taxonomies might prove to be a useful addition for both clinical and research purposes.
Although cluster analysis is a powerful technique for simplifying a complex data set (Borgen & Barnett, 1987), the relatively small sample size examined here may limit the representativeness of patients with temporal lobe epilepsy. Additional phenotypes of patients with temporal lobe epilepsy may be obtained with larger and more representative samples. Further, the reproducibility of cognitive phenotypes across samples varying in patient characteristics, administered test batteries, data reduction procedures, and other methodological details will speak to the robustness of specific cognitive phenotypes across cohorts of epilepsy patients.
ACKNOWLEDGMENTS
This work was supported by 2RO1 NINDS 37738 and MO1 RR 03186 (GCRC). This investigation could not be completed without the help of Drs. Brian Bell and Jana Jones as well as Michelle Szomi who was responsible for recruiting patients and Erika Johnson and Christian Dow for psychological testing. The authors sincerely thank Drs. Fred Edelman, Raj Sheth, Jack Jones, Brad Beinlich, and Kevin Ruggles for referring their patients with temporal lobe epilepsy to this study.
