SEARCH

SEARCH BY CITATION

Keywords:

  • fMRI;
  • Memory;
  • Lateralization;
  • Reorganization

Summary

  1. Top of page
  2. Methods
  3. Results
  4. Discussion
  5. Conclusion
  6. Acknowledgments
  7. References

Purpose: We investigated functional reorganization mechanisms of the human medial temporal lobe (MTL) for episodic memory, in patients suffering from medial temporal lobe epilepsy (MTLE) with hippocampal sclerosis (HS).

Methods: We used functional magnetic resonance imaging (fMRI) to measure brain activity changes during matched episodic encoding tasks of abstract words (Verbal) and line drawings (Visual), in patients with unilateral right MTLE undergoing presurgical evaluation and healthy controls.

Results: As expected, a significant interaction between material type and the side of MTL activity was present in the control group, with preferential involvement of the left hippocampus in verbal encoding and the right parahippocampal region in visual encoding. When compared with controls, right MTLE patients with intact performance activated a region in the left hippocampus more during visual encoding, which resulted in an interaction between group and hemisphere. Importantly, an effect of memory performance on visual encoding activity was observed in the patients, with greater engagement of the left MTL being associated with higher recognition scores. Interestingly, activity in the left MTL also depended on the epileptic seizure frequency, suggesting a role for this clinical parameter in the recruitment of contralateral regions.

Discussion: Taken together, these results indicate functional reorganization of the MTLs in right HS, through transfer of function from the right to the left hemisphere, and strongly suggest an adaptive role for such reorganization mechanism in supporting preserved visual memory.

Since the historical case of patient HM (Scoville & Milner, 1957), it has become a well established finding in memory research that the medial temporal lobes (MTLs) are essential for the encoding, consolidation, and retrieval of episodic memories (Squire et al., 2004). Although still a matter of debate, both regional and hemispheric specializations exist within the MTL. One reasonably consistent finding in both patient and functional imaging studies is the lateralization of MTL activity as a function of material, with the left (dominant) and right (nondominant) MTL structures mediating verbal and visual (nonverbal) memory, respectively (Smith & Milner, 1981; Frisk & Milner, 1990; Kelley et al., 1998; Golby et al., 2001; Powell et al., 2005).

Patients with medial temporal lobe epilepsy (MTLE) often exhibit hippocampal sclerosis (HS), with damage possibly extending to other medial temporal structures but more remote brain regions generally remaining intact (Duncan et al., 1996). Many of these patients have medically refractory seizures and are thus indicated for neurosurgical removal of the epileptogenic area, through anterior temporal lobe resection. Although 60–70% of such cases yield a seizure-free outcome (Wiebe et al., 2001), memory deficits may result, with a differential decline of verbal and nonverbal memory following left and right resection, respectively (Gleissner et al., 2002; Lee et al., 2002). Mapping of MTL memory functions is therefore desirable when planning the surgical resection, in order to minimize postoperative deficits while removing the seizure focus completely (Tharin & Golby, 2007).

Cases of unilateral MTLE provide a unique opportunity to study the long-term consequences of isolated brain damage in episodic memory, because the HS pathology is thought to result from an early-childhood initial precipitating injury inducing functional and structural damage to the hippocampus (Mathern et al., 2002). The fact that patients with early seizure disease onset and severe HS may not show significant changes in memory performance after surgical removal of the epileptogenic MTL suggests that other brain regions can support this function (Jokeit et al., 1999; Gleissner et al., 2002; Dulay et al., 2006). Although functional imaging investigations of individual patients often lack the statistical power to allow conclusive inferences about their MTL function, group studies in unilateral MTLE have already shown that reorganization may occur through transfer of processing to the contralateral MTL (Golby et al., 2002; Richardson et al., 2003; Janszky et al., 2005; Powell et al., 2007a). However, only a few studies have demonstrated an association between memory performance and a pattern of brain activity in MTLE patients that differed from that of normal subjects (Richardson et al., 2003; Powell et al., 2007a) and the significance of these brain activity changes in terms of functional reorganization mechanisms remains to be understood.

The aim of the present study was to investigate the patterns of MTL activity during episodic memory tasks in right MTLE patients and to test the hypothesis that functional reorganization supports preserved visual memory performance in these patients.

Methods

  1. Top of page
  2. Methods
  3. Results
  4. Discussion
  5. Conclusion
  6. Acknowledgments
  7. References

Subjects

We studied 22 healthy volunteers and 13 patients with a diagnosis of MTLE with right HS, undergoing our adult epilepsy surgery program. All subjects were right-handed and had normal or corrected to normal vision. Informed consent was obtained from each subject according to guidelines approved by our institutional ethical committee. Exclusion criteria for both patients and controls were: major depression, dementia, and other neuropsychiatric disorders. One of the healthy volunteers was excluded from the study due to abnormal structural imaging findings. Data from two of the normal subjects and one of the patients were further excluded from group comparisons due to technical problems in recording behavioral responses. Finally, a subgroup of normal subjects (Controls [CTRL]; n = 10) and a subgroup of patients (right temporal lobe epilepsy [RTLE], n = 10) were selected for analysis so that they were matched for age and behavioral performance, as measured by recognition scores and average response times. Although education and gender may have an effect in memory function, the fact that the two groups have matched behavioral performance should control for such effects. Demographic and behavioral data acquired during the scanning sessions are presented, for the selected subjects, in Table 1. Relevant clinical parameters are shown for the patients, including disease duration, seizure frequency, and antiepileptic medication. Global intelligence and memory scores from the neuropsychological evaluation of the patients are also presented.

Table 1.  Demographic, clinical, neuropsychological, and behavioral data for matched right temporal lobe epilepsy patients and control subjects
Sub.Age (years)Age at seizure onset (years)Duration of seizure disorder (years)Frequency of seizures (per month)Antipileptic medicationWAIS IQWMS-R (z-score)Recog. score (%)Response time (s)
VisualVerbalVisualVerbalVisualVerbal
  1. Recognition scores = hit rate—false alarm rate. Response times are the average for all correct responses (hits and correct rejections).

  2. WAIS, Weschler Adult Intelligence Scale; WMS-R, Revised Weschler Memory Scale; CBZ, carbamazepine; CLB, clobazam; LMT, lamotrigine; LVT, levetiracetam; PHT, phenytoin; TOP, topiramate; VGB, vigabatrin; VPA, valproic acid.

Control subjects
 127  27401,7501,850
 225  30331,7981,979
 326  33521,2331,243
 427  50401,2571,480
 529  67231,3831,413
 628  17171,3801,460
 727  40332,4872,037
 840  90731,2231,297
 930  90702,1802,333
 1035  38301,3131,313
Right temporal lobe epilepsy patients
 139 633 1CLB, CBZ, LMT 71−1,23−1,5842531,5401,403
 234221260LMT, PHT−0,92−2,4052701,2831,303
 332 923 6CBZ, TOP, VPA 54−3,15−2,5730461,2131,400
 424 22210CBZ, LVT 81−2,08−1,9520371,8971,847
 530 72330CBZ 94 0,96−0,3537331,2571,117
 633 627 6CBZ, VPA 0,41−2,5720161,2271,360
 743281510CBZ, LVT105−0,92−0,8240461,3631,410
 838 632 2CLB, CBZ 77 0,01−0,27 7231,6171,700
 9362610 2PHT70601,6171,393
 1024 32110CLB, CBZ, VPA 86−1,37−1,02 7391,4601,580

All patients had active temporal lobe epilepsy with typical complex partial seizures and were on antiepileptic drug treatment (for details see Table 1). Diagnosis of right HS with normal left hippocampus was determined by structural MRI examination. This was consistent with the side of seizure focus determined based on neurological examination and video-EEG recording. All patients underwent a comprehensive presurgical evaluation protocol, including neurological examination, video-EEG recording, structural MRI, and standardized neuropsychological testing. The intracarotid amytal procedure (or Wada test) was applied only in two cases, because this is not routinely done in right MTLE patients at our center. The neuropsychological test battery included the Wechsler Adult Intelligence Scale (WAIS) and the pairs of words associative learning and visual memory tests of the Revised Wechsler Memory Scale (WMS-R). Portuguese normative data available for age- and education-matched populations were used (Baeta, 2002).

Stimuli

The visual material consisted of abstract line drawings based on the designs previously used in the “Abstract Word List (AWL) and Abstract Design List (ADL) Learning Tasks” developed by Jones-Gotman (Jones-Gotman, 1986; Jones-Gotman et al., 1997) (http://mni.mcgill.ca/cog/jonesgotman/testdev.htm). Each drawing was composed of simple shapes displayed in a complex spatial configuration, such as the ones shown in Fig. 1. A separate normative study was conducted on the nonverbal material in a group of 12 subjects. Each drawing was presented for 4 s and subjects were asked to make a categorical nameable/non-nameable rating. Results showed a median verbalization rating of 29%. The verbal material consisted of orthographic stimuli including words and pseudowords. Abstract, low-imageability words were first obtained in English from the MRC Psycholinguistic Database (Coltheart, 1981) (http://www.psy.uwa.edu.au/mrcdatabase/uwa_mrc.htm), with imageability ratings between 150 and 450 and meaningfulness ratings between 127 and 250 (Colorado Norms) (Toglia & Battig, 1978). Words were then translated into Portuguese and a selection was obtained with three to eight letters in length and two to four syllables. Pronounceable and orthographically regular pseudowords were constructed by replacing the vowels in each word by different vowels. As a result of this procedure, words and pseudowords were matched for length in a pairwise manner. Examples of these stimuli are shown in Fig. 1.

image

Figure 1. Diagrams illustrating the visual and verbal memory encoding tasks: each pair of drawing/rotated drawing and word/pseudoword is presented for 6 s. Blocks of 18 s, comprising three items to be encoded, are alternated with periods of control task of equal duration, where a black square (detection target) appears in one of two positions. In total, 10 cycles of encoding and control task periods are presented for each type of material.

Download figure to PowerPoint

Behavioral task

Deep, incidental encoding of the material was achieved by performing a visuospatial decision task (Visual) and a lexical decision task (Verbal). For the visuospatial task, subjects were requested to compare two rotated versions of the same drawing placed side by side and to select the one aligned with the vertical or horizontal axes. For the lexical decision task, subjects were required to indicate the real word. Responses were performed through a button press. A block design was employed, alternating 18 s periods of encoding task with 18 s periods of control task over 10 cycles. Each encoding period consisted of the implicit encoding of three items, presented for 6 s each, as previously described. The control task consisted in detecting the position of a black square that appeared either on the right or the left side of the screen, alternating randomly every 6 s, so that attentional and motor response components of the task were controlled for (Fig. 1).

A recognition task followed approximately 10 min after encoding. Subjects were shown the 30 previously studied targets pseudorandomly intermixed with 30 novel foils, for each material type, and were asked to make old/new judgments on each item through a button press. Responses were registered and classified as hits (H), misses (M), correct rejections (CR), and false alarms (FA). Recognition performance was assessed in terms of both accuracy and speed. Response rates were calculated as: Hrate=#H/(#H +#M); Mrate=#M/(#H +#M); CRrate=#CR/(#CR +#FA), and FArate=#FA/(#CR +#FA), where, for example, #H stands for the number of hits during the experiment. A recognition score was then calculated as Hrate–FArate and mean response times were obtained by averaging over all response types.

Image acquisition

Imaging data were collected using a 1.5 Tesla Philips Gyroscan Intera whole-body MRI system (Philips Medical Systems, Best, The Netherlands). Changes in blood-oxygenation level-dependent (BOLD) signal were measured by using gradient-echo echo-planar-imaging (GE-EPI) with TR = 3000 ms, TE = 50 ms, and 90° flip angle. The whole brain was covered with a total of ∼30 axial slices, with ∼4 mm thickness, ∼240 × 240 mm2 field of view, and a 64 × 64 acquisition matrix, yielding a voxel size of ∼3.5 × 3.5 × 4.0 mm3. A spoiled gradient recalled echo (SPGR) pulse sequence was used to collect high-resolution T1-weighted structural images in the same session, with 1-mm-thick axial slices of 240 × 240 mm2 field of view and a 256 × 256 acquisition matrix, yielding a reconstructed voxel size of ∼1 mm3. For the functional scans, stimuli were projected on a translucent screen placed at the feet of the scanner bed, at a distance of approximately 3 m from the eyes of the subject. Projected images were viewed through a mirror system, with ∼6°× 6° visual angle for each stimulus. Stimuli presentation and response recordings were accomplished using Presentation (Neurobehavioral Systems, Albany, CA, U.S.A.; http://www.neurobs.com).

Image analysis

Memory encoding functional imaging datasets were analyzed using FEAT (FMRIB Software Library; Oxford, U.K.; http://www.fmrib.ox.ac.uk/fsl) in order to detect brain activity changes based on significant changes in BOLD signal. The following preprocessing steps were applied to each BOLD time series: motion correction (Jenkinson et al., 2002); nonbrain removal (Smith, 2002); spatial smoothing (Gaussian kernel, 8 mM FWHM); mean-based intensity normalization of all volumes by the same factor; and high-pass temporal filtering (Gaussian-weighted least squares straight line fitting, 50 s cutoff). A general linear model (GLM) approach with local autocorrelation correction was used to test for encoding-related activity changes (Friston et al., 1994; Woolrich et al., 2001). Each encoding task period (Verbal, Visual) was modeled for each subject by convolving a square function of width equal to the stimulus duration with the canonical Gamma-variate hemodynamic response function (HRF) (Boynton et al., 1996). The first time derivative of the canonical HRF was also included as a regressor, in order to account for any potential variability in the delay and dispersion of the hemodynamic response across the brain and between subjects. Six motion correction parameters were further included in the GLM as covariates of no interest. Linear contrasts between the verbal and visual encoding conditions and the control baseline were then calculated for each subject, yielding statistical maps of increased brain activity during task performance (Verbal > Control, Visual > Control).

Group analyses were carried out through a mixed-effects approach using Bayesian estimation techniques implemented in FLAME (Beckmann et al., 2003; Woolrich et al., 2004). Low-resolution functional images were registered to the corresponding high-resolution structural images of each subject, which were in turn registered to the Montreal Neurological Institute (MNI) template for a standard brain (Collins et al. 1994), using linear registration tools (Jenkinson et al., 2002). The normalized individual contrast images of the verbal and visual tasks were then entered into second-level statistical analyses. The mean group effects of each condition (Verbal, Visual) and the main effects of memory encoding (Verbal & Visual) were identified by second-level one-sample t-tests for each group. Subjects were treated as a random effect and conditions as a fixed effect. The interaction between conditions was investigated by performing second-level two-sample paired t-tests (Verbal > Visual, Visual > Verbal), using each subject's mean as a covariate of no interest. Comparisons between groups of subjects were achieved by performing two-sample unpaired t-tests (CTRL > RTLE, RTLE > CTRL). Finally, the effects of relevant behavioral and clinical parameters were investigated by including them as additional covariates in one-sample t-tests for each group and each condition.

The theory of Gaussian random fields (GRF) was used to accomplish maximum-height thresholding of the z-score images at specified significance levels, p, of false-positive probabilities, corrected for multiple comparisons (Worsley et al., 1996). Image analyses were constrained to predetermined regions-of-interest (ROIs) of the MTL and a small volume correction (SVC) procedure was thus employed: the number of voxels in the ROI was divided by the number of voxels per resolution element (RESEL), which was determined by the smoothness of the data. A SVC is justified when a priori assumptions exist regarding the regions of interesting activation, which restrict to a subset the number of comparisons being made, thus rendering the correction for multiple comparisons across the whole brain too conservative.

Regions-of-interest (ROI)

Six anatomical ROIs within an extended MTL were considered: left and right hippocampus, left and right amygdala, and left and right parahippocampal region, including the parahippocampal gyrus and extending into the fusiform gyrus (Lavenex & Amaral, 2000). Each ROI was manually defined on the high-resolution structural image of the MNI standard brain, based on published anatomical criteria and brain atlas (Duvernoy, 1999). Left MTL and right MTL ROIs were then defined, comprising the respective hippocampus, amygdala, and parahippocampal region. The most anterior and inferior portions of the ROIs were masked out, in order to avoid the regions critically suffering from susceptibility artifacts in our EPI scans. The predefined MTL ROIs were then used to mask the group analysis z-score statistical images before thresholding.

Functional ROIs were defined at the group level as the clusters of significant MTL activity obtained through each contrast (p < 0.05 uncorrected) and were then registered into the subjects' individual brain spaces. The percent signal change associated with each contrast was averaged across all voxels within the respective ROI, providing a measure of encoding-related activity for each subject, in each of the two memory conditions. These measures were finally entered into ANOVAs in order to test for material type, hemisphere, and group effects. The effects of recognition performance, seizure frequency, and age of epilepsy onset were also investigated through correlation and linear regression analyses.

Results

  1. Top of page
  2. Methods
  3. Results
  4. Discussion
  5. Conclusion
  6. Acknowledgments
  7. References

Behavioral results

A summary of the behavioral results obtained during the recognition task for each subject is shown in Fig. 2. Repeated measures ANOVA of the response rates revealed no significant main effects or interactions, except for a mild interaction between response type and material (F{1,54}= 4.199, p = 0.037) (Fig. 2A). Repeated measures ANOVA of the recognition scores yielded no significant effects of group, material type, or the interaction between them (Fig. 2B). Although a trend for different visual recognition scores between patients and controls is apparent, an independent samples t-test revealed no significant difference (t{18}= 1.515, p = 0.147). For the response times, no significant main effects or interactions were found, except for a marginally significant interaction between response type and group (F{1,39}= 4.308, p = 0.050) (Fig. 2C). Repeated measures ANOVA of the mean response times yielded no significant effects of group, material type or their interaction (Fig. 2D).

image

Figure 2. Behavioral data of the control (CTRL) and patient (RTLE) groups. (A) Response rates for each response type (CR, FA, H, M) and material type (Verbal, Visual); (B) Recognition scores, defined as H rate minus FA rate, for each material type. (C) Response times for each response type (CR, FA, H, M) and material type (Verbal, Visual); (D) Mean response times, averaged over all response types, for each material type. In all cases, bars represent averages over all subjects in each group (CTRL, RTLE). Error bars represent standard errors of the mean.

Download figure to PowerPoint

Imaging results

For all contrasts studied, peaks of encoding-related activity, obtained by second-level analyses restricted to the MTL, in the CTRL and RTLE groups, or comparisons across them, are displayed in Table 2.

Table 2.  Peaks of encoding activity in the medial temporal lobe for the control group and the right temporal lobe epilepsy patient group
ContrastCTRL groupRTLE group
RegionHem.MNI coords.Z-scorep corr.RegionHem.MNI coords.Z-scorep corr.
  1. For each contrast, brain region, hemisphere, and MNI coordinates (x y z [mm]) of voxels with local maximum z-score, within the medial temporal lobe, are presented by decreasing z-score, with respective statistical significance level, p, after small volume correction (p corr.). No other comparisons between groups produced significant results at corrected p < 0.05.

  2. n.s., not significant; n.a., not applicable; CTRL, control subjects; RTLE, right temporal lobe epilepsy.

Verbal > ControlHippocampusL−24−32 −84.5p < 0.005 HippocampusL−28−18−184.8p < 0.005 
HippocampusR+24−32 −84.5p < 0.005 HippocampusR+26−18−163.4p < 0.05  
Visual > ControlParahippocampusR+32−34−226.3p < 0.005 ParahippocampusR+36−32−223.6p < 0.05  
HippocampusR+18−32 −64.7p < 0.005   
Effects of memory and material type
Verbal & VisualParahippocampusR+32−34−226.5p < 0.0001HippocampusL−28−20−165.1p < 0.0001
HippocampusR+18−34 −66.0p < 0.0001AmygdalaL−24 −4−284.9p < 0.0001
HippocampusL−24−30 −85.6p < 0.0001HippocampusR+22 −8−243.8p < 0.05  
Verbal > VisualHippocampusL−28−18−144.2p < 0.005  n.s. 
Visual > VerbalParahippocampusR+32−36−224.2p < 0.005  n.s. 
Correlations with behavior and clinical parameters
Verbal: recognition scoreHippo./Parahippo.L−36−16−303.6p < 0.05  HippocampusL−18−30−104.1p < 0.05  
Visual: recognition score n.s. HippocampusL−28−38 +83.5p < 0.05  
Verbal: seizure frequency n.a. n.s. 
Visual: seizure frequency n.a. HippocampusL−20−40 +43.6p < 0.05  
Group comparisons
 RTLE-High > CTRLRTLE-High > RTLE-Low
Visual > ControlHippocampusL−28−22−183.6p < 0.05  HippocampusL−28−38−103.8p < 0.05  

Effects of memory and material type in the MTL

The clusters of activity obtained in the control group for the main effects of memory (Verbal & Visual), as well as for the specific effects of each material type (Verbal > Visual and Visual > Verbal), are shown in Fig. 3. In controls, both verbal and visual encoding tasks elicited activity in the posterior hippocampus bilaterally, as well as in the right parahippocampal region and the right amygdala (Fig. 3A). Specifically, verbal encoding involved a region in the left anterior hippocampus more than visual encoding (Fig. 3B), while visual encoding involved a region in the right posterior parahippocampal region more than verbal encoding (Fig. 3C). In RTLE patients, both verbal and visual encoding tasks elicited activity in the anterior hippocampus bilaterally, in the right parahippocampal region and the left amygdala. However, no differential effects of memory were found as a function of material type, in contrast to the control group (Table 2).

image

Figure 3. Statistical parametric maps of MTL encoding activity in the control group (p < 0.001 uncorrected) and respective ROI mean signal changes in controls (CTRL) and patients (RTLE). On the left, sagittal, coronal, and axial views of the z-score maps are overlaid upon the MNI template brain. On the right, bars represent averages over all subjects, error bars represent standard errors of the mean and significant differences are indicated. (A) Main effects of memory (Verbal & Visual), showing bilateral MTL (peaks +18, −34, −6; Z = 6.03; p < 0.00005 corrected and −24, −30, −8; Z = 5.59; p < 0.0005 corrected): controls and patients show similar levels of activity in this ROI. (B) Specific effects of verbal encoding (Verbal > Visual), showing a region in the left anterior hippocampus (peak −28, −18, −14; Z = 4.2; p < 0.005 corrected): while a significant difference between verbal and visual conditions exists for controls, this is only a trend in patients due to relatively increased activity on the left in the visual condition. (C) Specific effects of visual encoding (Visual > Verbal), showing a region in the right parahippocampal and fusiform gyri (peak +32, −26, −22; Z = 4.18; p < 0.005 corrected): while a significant difference between verbal and visual conditions exists for controls, this is not present in patients due to relatively decreased activity on the right in the visual condition.

Download figure to PowerPoint

ROI analysis of the effects of material type, hemisphere, and group

The signal changes measured in the functional ROIs are shown next to the graphs illustrating the location of the respective group activity clusters in Fig. 3, averaged across all CTRL and RTLE subjects, for both the verbal and visual tasks. Both controls and patients significantly activated the bilateral posterior hippocampus ROI during verbal and visual encoding (Fig. 3A). The left anterior hippocampus ROI was significantly activated by both controls and patients during verbal encoding, but only the patients exhibited a significant activation of this ROI during visual encoding, while it was in fact deactivated relative to the visual detection task in controls (Fig. 3B). Finally, the right parahippocampal ROI activated by controls during visual encoding was significantly less activated by patients during this task, and it was also significantly less activated by both controls and patients during verbal encoding (Fig. 3C).

A repeated measures ANOVA of the left and right MTL ROIs mean signal change showed a significant interaction between material type (Verbal, Visual) and hemisphere (Left, Right) (F{1,18}= 41.199, p < 0.001), with no main effects of either factor. A significant interaction between group (CTRL, RTLE), hemisphere, and material type was also evident (F{1,18}= 10.862, p < 0.004), which was related to the interaction between group and hemisphere (F{1,18}= 5.929, p < 0.026), with no significant interaction between group and material type. Consistently, paired t-tests revealed significant differences between verbal and visual conditions in both the left and right MTL ROIs in controls (t{9}= 4.150, p < 0.002; t{9}=−4.651, p < 0.001) but not in the patients. Furthermore, two-sample t-tests showed significant differences between controls and patients in the visual encoding task, with reduced right hemisphere activity (t{18}=−2.261, p < 0.02) and increased left hemisphere activity(t{18}= 2.568, p < 0.02) in patients relative to controls (Fig. 3B, C).

Correlations of MTL activity with behavioral and clinical parameters

The fact that both patients and controls exhibited a wide range of recognition score values allowed us to investigate the effects of performance on MTL activity. Both groups showed a significant effect of verbal recognition score on verbal encoding activity in a region within the left MTL (Table 2). However, only the patients exhibited an effect of visual recognition score on visual encoding activity in a cluster over the left posterior hippocampus and parahippocampal gyrus (Table 2; Fig. 4). Consistently, the ROI mean signal change showed a significant positive correlation with visual recognition score (ρPearson= 0.79, p < 0.01) and a significant fit by linear regression (R2= 0.63, p < 0.006).

image

Figure 4. Statistical parametric maps of MTL clusters where visual encoding activity is correlated with recognition score in the patient group (p < 0.01 uncorrected) and respective ROI mean signal changes plotted against recognition score, for both patients (RTLE) and controls (CTRL). On the left, sagittal, coronal, and axial views of the z-score map are overlaid upon the MNI template brain. On the right, significant correlation coefficients, ρPearson, as well as the linear regression fit and goodness of fit values, R2, are indicated. A region over the left posterior hippocampus and parahippocampal gyrus (peak −28, −38, +8; z = 3.48; p < 0.05 corrected) exhibits a significant correlation between visual encoding activity and recognition score in RTLE patients, but not controls.

Download figure to PowerPoint

A significant effect of seizure frequency was also found for visual encoding activity of the RTLE patients in a region over the left posterior hippocampus and parahippocampal gyrus which was largely overlapping with the region showing correlation with visual recognition performance (Table 2). Consistently, the ROI mean signal change showed a significant positive correlation with seizure frequency (ρPearson= 0.84, p < 0.01) and could be fit by linear regression (R2= 0.71, p < 0.002). No correlations were found between memory performance and seizure frequency, showing that the respective effects are independent. No significant effects of age of epilepsy onset, or disease duration, were found on MTL encoding activity of RTLE patients.

Group comparisons

Image group analyses comparing the RTLE and CTRL groups yielded no significant differences in MTL activity, but there was a trend for a region in the left anterior hippocampus to show greater activity in patients than controls (peak −28, −20, −18; z = 3.3; p < 0.07 corrected). As there was a trend for a significant difference in visual memory performance between patients and controls (Fig. 2B), this could have hindered a meaningful comparison of encoding activity between the two groups. In order to overcome this problem, and to verify whether the critical measure of recognition performance was influencing the observed results, the patients were split according to whether their visual recognition score was above or below a cutoff value of 30%. This value was chosen, based on the distribution of visual recognition scores of the patients, so as to divide the patients in two subgroups with equal numbers of subjects, yielding a high-performance subgroup (RTLE-High, n = 5) and a low-performance subgroup (RTLE-Low, n = 5). The best performing patients showed very similar performance to the controls, while the worst performing patients exhibited significantly lower recognition score in the visual task (t{13}= 3.380, p < 0.005).

Two-sample group analyses were then conducted to compare encoding activity within the MTL between the two patient subgroups and the control group. Significant results were found during visual encoding, with the better performing patients showing greater activity in a region within the left anterior hippocampus relative to controls (Table 2, Fig. 5) and the better performing patients showing greater activity in a region within the left posterior hippocampus and parahippocampal region relative to the lower performance patients (Table 2). Consistently, an ROI analysis of the mean signal change in the visual encoding task showed a significant difference between the RTLE-High and CTRL groups (t{13}= 4.121, p < 0.002) (Fig. 5). The difference between the better and worse performing patients reflects, and is consistent with, the effect of visual recognition score on visual encoding MTL activity observed in approximately the same region (Table 2, Fig. 4).

image

Figure 5. Statistical parametric maps of MTL clusters where visual encoding activity is greater for high performance patients (RTLE-High) compared with controls (CTRL) (p < 0.001 uncorrected) and respective ROI mean signal changes in both groups. On the left, sagittal, coronal, and axial views of the z-score maps are overlaid upon the MNI template brain. A region in the anterior left hippocampus (−28, −22, −18; Z = 3.63; p < 0.05 corrected) shows significantly greater visual encoding activity in patients than controls. On the right, bars represent averages over all subjects, error bars represent standard errors of the mean and significant differences are indicated.

Download figure to PowerPoint

Discussion

  1. Top of page
  2. Methods
  3. Results
  4. Discussion
  5. Conclusion
  6. Acknowledgments
  7. References

Our study found modifications of normal brain activity patterns of episodic encoding in temporal lobe epilepsy patients with right HS, which were related with memory performance and with their seizure frequency. This finding indicates functional reorganization and suggests an adaptive role for such reorganization.

Relation with previous evidence for MTL hemispheric specialization

We first demonstrated the hemispheric specialization of the normal MTL for episodic memory encoding of matched verbal and nonverbal material, through several approaches. First, asymmetric brain activity patterns were found for the verbal and visual encoding tasks. Second, paired comparisons revealed dissociation between material type and side of MTL activity, with greater right MTL activity for the visual compared with the verbal task and greater left MTL activity for the verbal compared with the visual task. Furthermore, an ROI analysis of the mean signal change measured in left and right MTL showed a strongly significant interaction between material type and hemisphere.

Several functional imaging studies have employed both verbal and visual memory paradigms, showing left and right lateralized activity patterns, respectively (Kelley et al., 1998; Golby et al., 2001; Powell et al., 2005). The stimuli used in those experiments included words, faces, line-drawings of nameable objects, pictures of scenes and abstract patterns. While verbal material was consistently found to preferentially engage the left MTL, data concerning the right MTL are not so clear. In particular, encoding of unfamiliar faces produced right-lateralized MTL activation in two studies (Kelley et al., 1998; Powell et al., 2005) and symmetrical MTL activation in another study (Golby et al., 2001). In the latter work, it was shown that the lateralization of encoding processes depends on the verbalization rate of the material, with abstract patterns yielding the most right-lateralized activation results. Our results support this finding, by showing right hemisphere dominance during episodic encoding of abstract line drawings in controls.

In previous imaging studies of MTL memory function lateralization, no systematic attempt was made to use well-matched memory tasks for clearly separable types of material and different stimuli yielded variable performance levels, which may have biased the brain activation results. In order to overcome this problem, we used two matched learning tasks of lists of abstract words and abstract designs, which had been previously shown to yield a partial functional dissociation between side of excision and type of material in selective amygdalohippocampectomy patients (Jones-Gotman, 1986; Jones-Gotman et al., 1997). A retention deficit on the verbal task in individuals with left MTL excision and a learning deficit on the visual task in individuals with right temporal-lobe dysfunction were observed using this material. By employing similar tasks, we were indeed able to obtain well-matched recognition accuracy and reaction times between the verbal and visual encoding tasks, as desired.

As the objective of the present study was the investigation of MTL memory function reorganization, and taking into account time limitations, it was chosen not to perform a speech-functional magnetic resonance imaging (fMRI) experiment in order to analyze language dominance. Nevertheless, the patients studied had unilateral right MTLE and were right handed, and therefore atypical language lateralization was not expected. Moreover, our data concerning the lexical task showed that normal left hemisphere lateralization was preserved in the RTLE patients during verbal encoding, which indicates normal language dominance.

Relation with other episodic memory paradigms

The success of fMRI studies of episodic memory function has often been hindered by the critical localization of the MTL, especially its anterior portion, in the basal and medial temporal regions of the brain, where echo-planar images suffer from severe static susceptibility artifacts leading to geometric distortions and local signal loss and consequently to reduced BOLD sensitivity (Ojemann et al., 1997). The efficiency of the paradigm design hence becomes of fundamental importance, so that activity detection sensitivity can be optimized (Friston et al., 1999). Although event-related designs have frequently been preferred in order to allow trial-by-trial variations in encoding success to be used to detect memory-related activation (Richardson et al., 2003; Powell et al., 2005), some studies have shown that these benefits may not be sufficient to overcome the sensitivity advantages of blocked designs (Binder et al., 2005; Narayan et al., 2005). On the other hand, one study found differences in the location of MTL activation obtained by event-related and block paradigms, with the former eliciting activity in more anterior regions (Powell et al., 2005). These findings could in fact partly explain the lack of sensitivity of event-related designs in studies of MTL function. More recently, the same authors have employed both paradigms in MTLE patients and found meaningful results with the block design as well as the event-related task, suggesting that block designs are indeed appropriate to investigate memory related activity in these patients (Powell et al., 2007a).

In our study, we have chosen to employ a simple boxcar block design, in order to optimize the sensitivity of our BOLD measurements at 1.5 T. We also observed that patients were more comfortable with this type of design, while they had difficulties in completing long encoding sequences of trials. Consistent with published data (Powell et al., 2005), we found main effects of encoding activity mostly in posterior locations within the MTL. However, specific effects of material type could be found in more anterior portions of the hippocampus. Moreover, group differences between patients and controls were also found in an anterior cluster within the left hippocampus. We therefore conclude that, although event-related designs are best suited for the direct identification of subsequent memory effects, block designs may be advantageously used in patients to investigate differential effects of memory between conditions or groups, or as a function of relevant behavioral or clinical parameters.

Evidence for adaptive functional reorganization

Our investigation of MTL hemispheric specialization in MTLE patients produced evidence of redistribution of activity relative to controls. While normal left hemisphere lateralization was preserved during verbal encoding, the right hemisphere dominance observed in controls during visual encoding was abolished in the patient group. Consistently, an ROI analysis revealed an interaction between hemisphere and group, with greater left MTL activity in the patients relative to controls. Although no significant activity differences were initially found by performing image analyses comparing the control and patient groups, this was probably hindered by the fact that some patients performed considerably worse than others. By applying a cutoff value of 30% to the visual recognition score, the patients with higher scores were very well matched with the controls and comparison between them indeed showed a region within the left hippocampus that was more engaged by these better performing patients.

Most interestingly, we found an effect of memory performance within the left MTL of RTLE patients, such that better visual recognition scores were associated with greater activity in the left MTL during visual encoding. No such correlation was found in controls. The extent of left MTL recruitment was also correlated with the number of epileptic seizure events per month. Overall, these results suggest that preserved visual memory performance in the functionally intact patients is achieved by recruiting the MTL contralateral to the pathology in functions normally performed by the damaged MTL. Besides providing evidence for the redistribution of activity within the MTL in MTLE patients, our study therefore indicates that redistribution processes are associated with preserved memory function. This observation is strongly supportive of an adaptive mechanism leading to functional reorganization of memory circuits towards the left hemisphere in cases of right HS.

A correlation between the degree and extension of unilateral atrophy and the BOLD-fMRI activity could not be established in this study. However, the interpretation of our evidence of adaptive functional reorganization should not be affected. In fact, although a positive correlation between BOLD-fMRI activity and the volume of the sclerotic hippocampus has been reported in previous studies, no such structure-function correlations were found for the contralateral MTL (e.g., Powell et al., 2007a). Furthermore, we have focused on the correlation between performance and activity, regardless of the fact that the asymmetric patterns of MTL activity observed in our patients could have been biased by the atrophy of the ipsilateral hippocampus. Indeed, our main results concern the contralateral MTL activity, its comparison with controls, and its correlation with the patients' memory performance and seizure frequency. For example, better performing patients showed greater activity in a region within the left anterior hippocampus relative to controls (Table 2, Fig. 5), which cannot be explained by atrophy of the sclerotic hippocampus. Furthermore, the patients exhibited an effect of visual recognition score on visual encoding activity in a cluster over the left MTL (Table 2; Fig. 4). Because no significant atrophy is expected and no structure–function relationships have been reported for the contralateral MTL, these results should hold.

Relation with previous evidence of functional reorganization

Consistent with our results, functional reorganization in MTLE has been shown in previous studies by shifts in memory-related fMRI MTL activation toward the hemisphere contralateral to the epileptogenic focus (Bellgowan et al., 1998; Detre et al., 1998; Jokeit et al., 2001; Golby et al., 2002; Richardson et al., 2003, 2006; Powell et al., 2007a). Moreover, some studies have shown that preoperative fMRI measures of MTL asymmetry are predictive of memory decline following anterior temporal lobe resection, with relatively greater ipsilateral compared to contralateral MTL activity being correlated with greater memory decline (Richardson et al., 2004; Powell et al., 2007b). However, only a few studies have reported an association between task performance and a pattern of brain activity that differs from that of normal subjects. An investigation of verbal memory fMRI activity in MTLE patients with left HS showed increased right hemisphere activation to be associated with preserved verbal memory performance (Richardson et al., 2003). Our results extend those findings by showing an increase in left MTL activity as a result of right HS that is positively related to preserved visual memory performance. Taken together, these studies suggest that recruitment of the contralateral MTL in cases of unilateral HS constitutes an adaptive functional reorganization mechanism supporting preserved episodic memory function. Similar evidence has been obtained for language function reorganization in left TLE following left anterior temporal lobe resection, where recruitment of regions in the right hemisphere was associated with high reading ability (Noppeney et al., 2005). However, in a recent report, both left and right MTLE patients underwent verbal and nonverbal memory fMRI and, although shifts in memory activity to the contralateral MTL were observed in each patient group, these were not positively related with performance (Powell et al., 2007a). In fact, the authors found a positive correlation with memory recall in the damaged MTL and a negative correlation in the contralateral MTL. We speculate that functional reorganization to the contralateral hemisphere only occurs if no viable tissue is available in ipsilateral MTL to maintain performance. This shift might depend on disease severity and extent, as suggested by our finding of a correlation between contralateral MTL activity and seizure frequency.

In terms of the factors affecting the observed functional reorganization mechanisms in MTLE, the finding that visual encoding activity in the left MTL increased with the number of seizures per month is in agreement with previous studies indicating the influence of epileptic activity on reorganization mechanisms. One study showed that the bilateral independent epileptiform discharges occurred more often in patients with symmetric visuospatial memory fMRI patterns than in patients with typical memory lateralization to the side contralateral to the epileptic focus (Janszky et al., 2004). Another study showed that a shift of fMRI language representation to the right hemisphere was associated with higher interictal spiking frequency in left MTLE patients (Janszky et al., 2006). Overall, these results support the view that functional factors, such as epileptic activity, can affect brain reorganization mechanisms. On the contrary, and consistent with several related studies, we found no effect of age of epilepsy onset, or disease duration, on the memory activity patterns of MTLE patients (Janszky et al., 2006; Powell et al., 2007a).

Remaining questions

One potential confound in our results was the inclusion of a larger number of female participants in the RTLE group compared with the CTRL group. In fact, it was not possible to match the final groups according to gender because a considerable number of subjects had to be excluded from the study for different reasons. In view of the limitations in the choice of controls, preference was given to age matching due to its well-known influence on memory function. Although some studies have in fact suggested that memory lateralization may also be influenced by gender (Frings et al., 2006), no such effect could be found in our control group. We compared MTL activity during our visual memory task between males and females in the control group and found no significant differences, indicating that these could not explain the differences observed between the patient and control groups. Nevertheless, it should be further investigated what the effects of gender may be on memory lateralization and its reorganization.

More data is necessary in order to better understand the functional reorganization mechanisms involved in cases of unilateral MTLE and their significance. Future work using larger patient populations should address the role of a number of factors that may have an effect on memory reorganization, such as the seizure frequency, the age of seizure onset/disease duration and the nature, extent and severity of the initial precipitating injury. Moreover, it remains to be investigated what the possible contribution of ipsilateral regions may be, particularly along the longitudinal axis of the hippocampus and within the neocortex in the parahippocampal region. Ultimately, a definite demonstration of reorganization would require the longitudinal investigation of functional anatomy changes in the course of the disease, which is critically dependent on the development of more robust and reproducible fMRI paradigms for memory studies of individual patients.

Conclusion

  1. Top of page
  2. Methods
  3. Results
  4. Discussion
  5. Conclusion
  6. Acknowledgments
  7. References

In summary, we found evidence for adaptive functional brain reorganization in patients with right HS, who showed greater left MTL engagement than controls, which was related with their recognition performance, as well as the frequency of their epileptic activity. We believe that these results are relevant not only to our understanding of functional reorganization mechanisms in the human brain but also for presurgical planning in right MTLE.

Acknowledgments

  1. Top of page
  2. Methods
  3. Results
  4. Discussion
  5. Conclusion
  6. Acknowledgments
  7. References

We acknowledge the Portuguese Foundation for Science and Technology for financial support through the grants POCTI/NSE/46438/2002 and POCI/PSI/56325/2004 and Fundação BIAL for the grant Nb 16/2004. We thank Alda Pinto and Cláudia Azevedo for assistance with scanning and Raquel Lemos for help with the clinical data.

Conflict of interest: We confirm that we have read the Journal's position on issues involved in ethical publication and affirm that this report is consistent with those guidelines. The authors declare to have no conflicts of interest regarding the work presented in the manuscript.

References

  1. Top of page
  2. Methods
  3. Results
  4. Discussion
  5. Conclusion
  6. Acknowledgments
  7. References