Topiramate and its effect on fMRI of language in patients with right or left temporal lobe epilepsy

Abstract

Topiramate (TPM) is well recognized for its negative effects oncognition, language performance and lateralization results on the intracarotid amobarbital procedure (IAP). But, the effects of TPM on functional MRI (fMRI) of language and the fMRI signals are less clear. Functional MRI is increasingly used for presurgical evaluation of epilepsy patients in place of IAP for language lateralization. Thus, the goal of this study was to assess the effects of TPM on fMRI signals. In this study, we included 8 patients with right temporal lobe epilepsy (RTLE) and 8 with left temporal lobe epilepsy (LTLE) taking TPM (+TPM). Matched to them for age, handedness and side of seizure onset were 8 patients with RTLE and 8 with LTLE not taking TPM (−TPM). Matched for age and handedness to the patients with TLE were 32 healthy controls. The fMRI paradigm involved semantic decision/tone decision task (in-scanner behavioral data were collected). All epilepsy patients received a standard neuropsychological language battery. One sample t-tests were performed within each group to assess task-specific activations. Functional MRI data random-effects analysis was performed to determine significant group activation differences and to assess the effect of TPM dose on task activation. Direct group comparisons of fMRI, language and demographic data between patients with R/L TLE +TPM vs. −TPM and the analysis of the effects of TPM on blood oxygenation level-dependent (BOLD) signal were performed. Groups were matched for age, handedness and, within the R/L TLE groups, for the age of epilepsy onset/duration and the number of AEDs/TPM dose. The in-scanner language performance of patients was worse when compared to healthy controls — all p<0.044. While all groups showed fMRI activation typical for this task, regression analyses comparing L/R TLE +TPM vs. −TPM showed significant fMRI signal differences between groups (increases in left cingulate gyrus and decreases in left superior temporal gyrus in the patients with LTLE +TPM; increases in the right BA 10 and left visual cortex and decreases in the left BA 47 in +TPM RTLE). Further, TPM dose showed positive relationship with activation in the basal ganglia and negative associations with activation in anterior cingulate and posterior visual cortex. Thus, TPM appears to have a different effect on fMRI language distribution in patients with R/L TLE and a dose-dependent effect on fMRI signals. These findings may, in part, explain the negative effects of TPM on cognition and language performance and support the notion that TPM may affect the results of language fMRI lateralization/localization.

1. Introduction

The presence of negative effects of topiramate (TPM) on cognitive performance in healthy subjects [1–4] and in patients with epilepsy [5–7] is well recognized. In various comparisons, healthy controls or epilepsy patients treated with TPM exhibit lower performance not only on verbal/linguistic tests but also on overall global measures of cognitive performance. In one study, 88% of the neuropsychological tests showed lower performance for healthy controls treated with TPM when compared to non-drug-treated controls and 80% of the tests were worse when compared to controls treated with lamotrigine [3]. Another study showed that these effects are dose-dependent with up to 35% of healthy controls showing significant cognitive deterioration on a ~400 mg/day dose of TPM [1].While the negative effects of TPM on cognition, in general, and on language, in particular, are unquestionable, TPM has proven efficacy as an antiepileptic drug (AED) in focal-onset [8–10] and generalized epilepsies [11]. Despite the fact that patients with epilepsy experience language and memory difficulties at baseline [12–14], they are frequently prescribed TPM based on its efficacy. These negative effects of TPM on cognition may come into play during evaluation for possible epilepsy surgery.

In addition to the cognitive side effects of TPM, several recent publications described negative effect of TPM on the results of the intracarotid amobarbital procedure (IAP) [15–19]. These studies reported not only a higher frequency of procedure failure when patients are taking medication(s) known to inhibit carbonic anhydrase (e.g., TPM) but also negative effects of these types of medications on IAP performance, which may affect eligibility of patients with temporal lobe epilepsy to undergo surgical intervention in order to control their seizures [17,18]. Although IAP has been used for approximately 60 years for language lateralization in presurgical evaluation of epilepsy patients, its use is overall decreasing [20,21]. This decrease is, in part, related to a significant rate of IAP complications reaching up to 10.9% in some studies, the use of other procedures including neurocognitive testing and fMRI, and the recognition of its limitations [22–26]. Of importance for this study is the fact that fMRI is increasingly used in staging for epilepsy surgery with several reports showing either excellent correlation with IAP or high predictive value for postsurgical language outcomes [22,27–29].

Despite the proven negative effects of carbonic anhydrase inhibitors, such as TPM, on language performance, the importance of language assessment in the presurgical evaluation of epilepsy patients and the increasing utilization of fMRI in place of IAP for presurgical mapping, we were able to identify only one relatively small study assessing the effect of these medications on fMRI activation patterns [30]. In most qualitative analyses, these authors reported less fMRI activation in the language regions of epilepsy patients taking TPM when compared to other epilepsy patients, suggesting that TPM may have an effect not only on language and IAP performance but also on fMRI performance and activation pattern(s). Thus, our goal was to [1] assess the effect of TPM on the fMRI activation patterns in patients with left and right temporal lobe epilepsy undergoing surgical staging, [2] to evaluate the differences in the fMRI lateralization patterns between epilepsy patients and healthy controls matched for age, sex and personal handedness, and [3] to conduct a preliminary analysis of the effects of TPM dose on the brain's hemodynamic response, as previous studies reported a relationship between BOLD signal and the use of medications acting via the carbonic anhydrase mechanism [31,32]. The overarching hypothesis guiding this research was that TPM may affect not only the cognitive performance but also the fMRI activation patterns in patients with TLE.

2. Methods

2.1. Subjects

Out of 81 healthy control subjects and 70 epilepsy patients enrolled in an ongoing study of language localization and lateralization in focal-onset epilepsy [29,33–37], 8 patients with right and 8 with left temporal lobe epilepsy (L/R TLE) taking TPM as part of their epilepsy medication regimen (+TPM) were selected. An additional 8 patients with RTLE and 8 with LTLE not taking TPM (−TPM) were matched to them by age, sex, and handedness (measured with Edinburgh Handedness Inventory; [38]). Also matched (1:1) as close as possible for age, sex and handedness were 32 healthy controls — these subjects were enrolled in this study only for the purpose of comparison of the language localization patterns with the epilepsy patients. Thus, a total of 64 subjects comprised the study cohort. Demographic characteristics including mean age of seizure onset, illness duration, and number of medications for each group of patients with TLE are summarized in Table 1. As previously reported, all epilepsy patients completed presurgical evaluation for possible epilepsy surgery including prolonged video/EEG monitoring, standard clinical 3 T MRI, neuropsychological testing and other tests as deemed necessary by the multidisciplinary epilepsy team during a presurgical conference [29,34,36,37]. Healthy controls were recruited by word of mouth and local advertising. Some of the subjects included in this study were also included in our previous publications. All subjects signed an informed consent form approved by the Institutional Review Board of the University of Cincinnati prior to study participation.

Table 1.

Demographic and performance characteristics for the 32 patients with temporal lobe epilepsy and 32 healthy subjects.

	LTLE +TPM (4 M, 4 F)		LTLE −TPM (4 M, 4 F)		Healthy (9 M, 7 F)				RTLE +TPM (8 M)		RTLE −TPM (8 M)		Healthy (12 M, 4 F)
	Mean	SD	Mean	SD	p-Vala	Mean	SD	p-Valb	Mean	SD	Mean	SD	p-Vala	Mean	SD	p-Valb
Age, years	34.9	13.1	34	11.6	0.89	34.5	10.7	0.89	36.9	8.7	38.1	9.3	0.78	36.7	9.4	0.98
EHI score	55	62.4	65.6	68.9	0.75	70.1	57.5	0.75	89.4	13.6	95.4	5.1	0.27	95.9	8.5	0.46
Age of seizure onset	9	8.6	10	6.6	0.80	–	–	–	14	14	16.1	15.1	0.77	–	–	–
Illness duration, years	25.9	12.2	24	17.5	0.81	–	–	–	22	8.9	22	13.5	1	–	–	–
Number of medications	2.5	0.8	2.3	0.7	0.51	–	–	–	2.1	0.4	2.1	0.6	1	–	–	–
Boston Naming Testc	49.3	8.6	50.3	7.9	0.81	–	–	–	46.7c	8.1	51.8	4.1	0.17	–	–	–
Word fluency	31.5	7.1	31.3	14.5	0.97	–	–	–	27.5	8	34.6	9.6	0.13	–	–	–
Category fluency	15.4	3.1	16.3	4.2	0.64	–	–	–	13.9	4	17.8	3.2	0.053	–	–	–
Semantic decisiond	(n=8)	9.6	(n=6)	9.7	0.72	(n=13)	7.3	0.044	(n=5)	5.9	(n=6)	11.4	0.34	(n=13)	10.7	0.018
% correct	66.8		64.8			75.6			69.6		64.2			80.2
Tone decisiond	(n=8)	14.9	(n=6)	17.9	0.20	(n=13)	4.3	0.075	(n=6)	11.4	(n=6)	14.6	0.028	(n=13)	4.8	0.0004
% correct	83.8		71.3			94.8			57.5		77.2			92.9

Note: Neuropsychological language tests (Boston Naming Test, word fluency, category fluency) were only performed for epilepsy patients.

Abbreviations: LTLE = left temporal lobe epilepsy; RTLE = right temporal lobe epilepsy; TPM = topiramate; M = male; F = female; EHI = Edinburgh Handedness Inventory.

^ap-Value for t-test of +TPM and −TPM subjects.

^bp-Value for t-test of +TPM and healthy subjects.

^cTest score was not available for one RTLE +TPM subject.

^dPerformance was not available for all subjects; number of subjects in which performance measure was available is indicated for each group.

2.2. Language assessments

All epilepsy patients were underwent presurgical language evaluation. For the Boston Naming Test (BNT), patients were asked to overtly name the picture presented to them, which tested language processing related to word-finding and semantic retrieval [39].Word fluency (WF) and category fluency (CF) tests required the patients to overtly generate as many words as they could in a given time frame that began with a particular letter or were in a given category (e.g., farm animals), respectively. Performance on each language assessment was measured using the total number of correct responses. All neuropsychological testing was conducted during admission to the epilepsy monitoring unit during which AEDs were either lowered or stopped. Two-sample t-tests were performed to examine performance differences for each language assessment between +TPM and −TPM subjects for each of the L/R TLE groups with p<0.05 considered significant.

2.3. Functional MRI language task

The semantic decision and tone decision (SDTD) fMRI language task was developed with minor adaptations based on the task previously published by Binder et al. [27,34,35,40]. Briefly, the control (i.e., tone decision) condition is performed 8 times and the active (i.e., semantic decision) condition 7 times for the total task duration of 7′15″. In this block-design task, each condition lasts for 30 s (15 s for the first tone decision block) with auditory stimuli presented every 3.75 s; non-dominant hand button presses are required in response to each stimulus. In the tone decision condition, subjects hear brief sequences of mixed 500 Hz/750 Hz tones and respond by pushing “1” for any sequence containing two 750 Hz tones or “2” for all other tone sequences. In the semantic decision condition, subjects hear names of animals and respond by pressing “1” for stimuli designating animals native to the United States and commonly used by humans or by pressing “2” in all other cases. Typical brain activation with this task includes prefrontal cortex of the inferior, middle, and superior frontal gyri, posterior cingulate gyrus and retrosplenial cortex, anterior/superior temporal sulcus and middle temporal gyrus, posterior/inferior temporal gyrus, fusiform and anterior parahippocampal gyri, anterior hippocampus, angular gyrus and posterior cerebellum (Figs. 1a, b) [35,41]. In all cases, epilepsy subjects and healthy controls practiced the task before being placed in the scanner including a mock run of 5 sets of tones followed by 5 nouns designating animals. Subjects were allowed to proceed to the scanner only if they responded correctly to all items.

Fig. 1.

Statistical maps for the SDTD task in (a) healthy subjects and (b) epilepsy subjects. Warm colors in (a) and (b) represent activation greater for semantic than tone decisions, while cool colors represent greater activation for tone than for semantic decisions. Regression analyses were also performed to directly compare patients with (c) LTLE and (d) RTLE who were treated (+TPM) and not treated (−TPM) with TPM and (e) the effect of TPM dose on task activation in +TPM patients. Warm colors in (c) and (d) represent overall increased activation in TPM-treated patients compared to untreated patients (+TPM > −TPM), while cool colors represent greater activation for untreated relative to TPM-treated patients (+TPM < −TPM). In (e), warm colors represent a positive association and cool colors represent a negative association between TPM dose and activation during the SDTD task. Correlation analyses revealed a positive relationship between performance on semantic decision making and activation in the left cingulate gyrus of patients with LTLE (c1), while negative associations were found between RTLE performance on the BNT and activation in the right superior frontal gyrus (d1) and the left visual association cortex (d2). On the scatterplots, data points in orange are +TPM subjects and in blue are −TPM subjects. Activated regions in statistical maps are significant at p<0.05 corrected for multiple voxel comparisons, with a cluster size of at least 60 contiguous voxels. Each statistical map is presented in radiological convention with left in the image as right in the brain and superimposed on an average T₁-weighted image generated from all healthy subjects (a) or subjects with TLE (b–e). The 12 axial slices selected for each display panel range in Talairach coordinates from Z = −17 mm (top left) to Z = +27 mm (bottom right).

Performance on the SDTD task was measured using the percentage of correct responses separately for the semantic and tone decision conditions. Two-sample t-tests were performed using SPSS version 18.0 (SPSS Inc., Chicago, IL, USA) to examine performance differences for each condition between +TPM and −TPM subjects as well as between +TPM and matched healthy subjects for each of the L/R TLE groups.

2.4. Functional MRI

After passing the standard Department of Radiology MRI screening procedures, all subjects underwent fMRI with either a 3 T Bruker Biospec 30/60 (Bruker Medizintechnik, Karlsruhe, Germany) or a 4 T Varian (Oxford Magnet Technology, Oxford, UK) scanner. The proportion of scans acquired between the two scanners was similar between TLE (27 at 4 T; 5 at 3 T) and HC (26 at 4 T; 6 at 3 T) and between +TPM (14 at 4 T; 2 at 3 T) and −TPM (13 at 4 T; 3 at 3 T) subjects. Patterns of fMRI activation during the SDTD task were similar between scanners and consistent with previously published data [34,35]. Thus, given that the overall data quality, blood oxygenation level-dependent (BOLD) signal distribution and subject performance between scanners were comparable as described previously [34,35], the 3 T and 4 T datasets were combined. However, scanner type was used as a covariate in all analyses to minimize confounding effects of this variable. Since the details of the procedures from the 3 T and 4 T scanners are provided elsewhere [42], only a brief description will be given here. The procedure at 3 T included ${\mathrm{T}}_2^{*}$-weighted gradient-echo echo-planar imaging (EPI) pulse sequence (TR/TE = 3000/38 ms, FOV = 25.6 × 25.6 cm, matrix = 64 × 64 pixels, slice thickness = 5 mm, flip angle = 90°) for functional scans, and a high-resolution T₁-weighted 3-D anatomical scan obtained using a modified driven equilibrium Fourier transform (MDEFT) sequence protocol (TR = 15 ms, TI = 550ms, TE = 4.3 ms, FOV = 25.6 × 19.2 × 16.2, flip angle = 20°; spatial resolution of 1 × 1 × 1.5 mm) to provide images for anatomical localization. For the scans obtained at 4 T, the procedure included ${\mathrm{T}}_2^{*}$-weighted gradient-echo EPI pulse sequence (TR/TE = 3000/25 ms, FOV = 25.6 × 25.6 cm, matrix = 64 × 64 pixels, slice thickness = 4 mm, flip angle array: 85/180/180/90) for the functional scans, and a high-resolution T₁-weighted 3-D anatomical scan obtained using a MDEFT protocol (TR = 13 ms, TE = 6 ms, FOV = 25.6 × 19.2 × 15.0, flip angle array of 3: 22/90/180 with the voxel size of 1 × 1 × 1 mm) for anatomical localization. Each scanner was equipped with an audiovisual system for presentation of task stimuli (3 T: SV 4120 system from Avotech Systems Inc., Jensen Beach, FL; 4 T: Resonance Technologies, Inc., Northridge, CA). After placing the subjects in the scanner and assuring appropriate level of comfort with foam padding, subjects were given a button box for generating responses and to alert the MRI technologist in case of a problem.

MRI and fMRI post-processing was performed using Cincinnati Children's Hospital Image Processing Software (CCHIPS) developed in the IDL software environment (IDL 8.1; Research Systems Inc., Boulder, CO). Multi-echo reference scans were used to correct geometric distortion due to B0 field inhomogeneity [43]. Data were first co-registered and motion corrected using a pyramid iterative algorithm [44] and then affine-transformed into the Talairach reference frame prior to statistical analysis [45]. First-level statistical analysis was performed using the general linear model (GLM) to calculate the fMRI blood oxygen level-dependent (BOLD) response differences between the active and control conditions of the SDTD task for each subject [46]. The first 15 s (i.e., tone decision block) of the SDTD task were not included in the GLM to allow for equilibration of the MR signal. The GLM did include motion correction parameters as covariates to optimize fMRI results [47], a set of cosine basis functions to account for low-frequency signal drift, and modeled the BOLD response as 30-second blocks of the active condition and 30-second blocks of the control condition. The GLM contrasted activation during semantic decision-making relative to tone decision-making to determine task-related fMRI activation during semantic processing.

Individual Z-score maps resulting from the GLM contrast were combined to create the group composite datasets for second-level analysis. We performed one-sample t-tests to create statistical maps to examine group effects for task-related activation for the TLE and healthy subjects. We then performed random-effects regression analysis to directly compare task-related activation between +TPM and −TPM patients within each LTLE and RTLE group. We also performed random-effects regression analysis in +TPM patients to determine the overall relationship between BOLD signal and the dose of TPM. Results of the EHI and scanner type were used as covariates in the regression analyses comparing +TPM and −TPM task activations, and an additional covariate of epilepsy focus lateralization was used in examining the effect of TPM dose on task activation in +TPM patients. A 4-mm full-width half maximum (FWHM) Gaussian filter was applied to each of the resultant datasets, and activated clusters with a minimum size of at least 60 contiguous voxels were statistically significant at a corrected threshold of p<0.05, as determined by an alpha probability algorithm in CCHIPS to correct for multiple voxel comparisons [48].

2.5. Relating task activation and language performance in patients with TLE

Activation clusters showing significant differences resulting from the direct comparison of +TPM and −TPM subjects for each of the LTLE and RTLE groups were defined as regions of interest (ROIs). For each ROI, the Z-score value was extracted from each subject's GLM map and plotted against performance on the semantic decision blocks, the Boston Naming Test, the word fluency test and the category fluency test. Associations between language performance and recruitment of ROIs during the SDTD task were assessed using Pearson's correlation coefficient.

2.6. Language lateralization assessments

Language lateralization indices (LIs) were calculated using methods previously described [29,49,50]. First, we designed functional ROIs based on the combined activation map from all 32 healthy subjects and then applied them to the epilepsy subjects. These primary ROIs included the left lateral frontal, the left lateral posterior temporal/parietal (“posterior”) regions and the corresponding right homologues. A global language ROI was created by combining the frontal and temporo-parietal ROIs. Next, we calculated the LI for all subjects based on the individual Z-score maps with only voxels greater or equal to the mean Z-score within an ROI included in the calculation. This method is based on the previously published toolbox with the Z-scores of the included voxels typically falling between 2 and 2.5 [50]. Subjects with LI scores ≥0.2 were considered to have typical left-lateralized language function, while others were considered to have atypical language lateralization [29]. Fisher's exact tests were used to determine the distribution patterns of language lateralization for the patients with TLE. All analyses were performed using SPSS version 18.0; considering the exploratory nature of this study, corrections for multiple comparisons were not performed.

3. Results

3.1. Subject demographics and performance

Demographic and performance variables for the patients with TLE and matched healthy subjects are summarized in Table 1. Overall, patients with LTLE had an earlier onset of epilepsy (9.5 vs. 15.0 years; p = 0.004) and a longer duration of epilepsy (24.9 vs. 22.5 years; p = 0.014) than the patients with RTLE, but the age at enrollment was similar between groups (34.4 vs. 37.5; p = 0.14). We also observed negative correlations between age of epilepsy onset (n = 32) and language lateralization in frontal ROI LI (Pearson's r = −0.41; p = 0.02) and in global ROI LI (Pearson's r = −0.46; p = 0.007) but not in posterior ROI LI (Pearson's r = −0.25; p = 0.17) indicating that earlier onset of epilepsy is associated with more atypical (i.e., right hemisphere) language lateralization [51]. Finally, the average number of AEDs per patient was similar in all groups ranging from 2.1 to 2.5 (Table 1); the doses of TPM in patients with L/R TLE were also similar (p = 0.803) with the average dose of TPM in the patients with LTLE of 312.5 mg±170 (minimum dose 100 mg/day; maximum dose 500 mg/day) and 393.8 mg±269 (minimum dose 100 mg/day; maximum dose 1000 mg/day) in the patients with RTLE.

Due to technical difficulties, some of the intra-scanner behavioral data were not available (see Table 1 for details regarding missing data). Nevertheless, as expected, we observed negative effects of TLE on intra-scanner language performance when compared to healthy controls whether in semantic decision (all p≤0.044) or in tone decision (all p≤0.075). The performance on BNT, WF and CF in the −TPM patients was overall similar or better than in the +TPM patients with L/R TLE with a more pronounced performance difference in patients with RTLE, particularly with CF approaching significance at p = 0.053.

3.2. Functional MRI activation patterns

Statistical parametric maps for the healthy subjects revealed language-related activation consistent with the results of previous studies that utilized this version of the SDTD task (Fig. 1a) [35,40,41]. Similar to healthy controls, subjects with left or right TLE (Fig. 1b) showed typical activation lateralized to the left frontal and temporal language areas. The patterns of activation in healthy controls and epilepsy patients included in this study are similar to the patterns presented previously from a mixed group of epilepsy patients and will not be discussed here in detail [29,34] except for slightly lower intensity of the activation in the epilepsy patients which is consistent with the literature [30].

The comparison between patients with LTLE taking and not taking TPM for the treatment of epilepsy (Fig. 1c) revealed greater activation in the midline left cingulate gyrus (centroid:−10, 33, 26; cluster volume: 60 voxels; Brodmann's area (BA) 32) and decreased activation in the left superior temporal gyrus (centroid: −47, −23, 8; cluster volume: 85 voxels) in the +TPM relative to −TPM patients. Overall, more +TPM patients with LTLE had atypical language lateralization than −TPM patients in the frontal, temporo-parietal or global ROIs, but none of the differences were statistically significant (all p>0.32). Similar comparisons performed in the patients with RTLE (Fig. 1d) revealed increased BOLD signal in the bilateral (right > left) superior frontal gyrus (centroid: 12, 53, 10; cluster volume: 147 voxels; BA 10) and left lingual/occipital gyrus (centroid: −26, −69, −5; cluster volume: 74 voxels) as well as decreased signal in the left inferior frontal gyrus (centroid:−41, 25,−6; cluster volume: 66 voxels; BA 47) in the +TPM relative to −TPM patients. Again, more +TPM patients with RTLE had atypical language lateralization in the frontal, temporo-parietal and global ROIs with the temporo-parietal ROI difference reaching significance at p = 0.041 and the global ROI approaching significance at p = 0.077.

Finally, we performed random-effects regression analysis to determine the relationship between BOLD signal and the dose of TPM (Fig. 1e), as one of the aims of this study was to test whether TPM itself exerts an effect on the brain’s hemodynamic response. Blood oxygenation level-dependent signal increases associated with increased dose of TPM were noted in the basal ganglia/anterior thalamus (centroid: −2, 0, 14; cluster volume: 61 voxels), while negative associations were observed in the anterior cingulate cortex (centroid: 1, 38, 14; cluster volume: 70 voxels; BA 32) and lingual gyrus (centroid: 1, −77, 2; cluster volume: 85 voxels; BA 18) suggestive of a variable effect of TPM on cerebral blood flow/perfusion.

3.3. Associations between task activation and language performance

Activation for the patients with LTLE (Fig. 1c1) in the left cingulate gyrus showed a significant positive relationship with performance on the semantic decision blocks (r = 0.59, p = 0.026). The patients with RTLE exhibited negative associations between BNT performance and activation in both the left lingual/occipital gyrus (r = −0.56, p = 0.03; Fig. 1d1) and the right superior frontal gyrus (r = −0.44, p = 0.10; Fig. 1d2).

4. Discussion

The goal of this study was to assess the effect of TPM on the fMRI activation patterns, to evaluate epilepsy and healthy control group differences in language lateralization patterns, and to conduct a preliminary analysis of how TPM may influence the brain's hemodynamic response in patients with TLE. In the course of the analyses, several findings emerged. First, while the overall group fMRI activation patterns between healthy controls and patients with L/R TLE were similar (Figs. 1a, b), significant differences between matched patients taking and not taking TPM for the control of their epilepsy became apparent. These differences were observed in both patients with right and left TLE onset despite the fact that their performances on neuropsychological measures and intra-scanner performance were relatively similar. Second, as expected, as a group, patients with epilepsy performed worse on intra-scanner performance when compared to healthy controls (Table 1). Third, we observed significant associations between neuropsychological measures and BOLD signal changes in the areas of differences between patients with L/R TLE (Figs. 1c1, d1, d2). Finally, we noted significant associations between BOLD signal changes and TPM dose suggestive of a dose-dependent effect of TPM on fMRI signal (Fig. 1e). All these findings warrant a detailed discussion.

Comparison between L +/− TPM groups revealed BOLD signal increases in the midline left cingulate gyrus and a decrease in the left superior temporal gyrus (Fig. 1c), while the comparison between R +TPM and R −TPM groups includes decreased BOLD signal in the left inferior frontal gyrus and increases in the right > left superior frontal gyri and left lingual/occipital gyrus. Further, +TPM patients more frequently showed atypical language representation when compared to the −TPM patients (difference not significant). Taken together, in view of similar performances on neuropsychological language tests and intra-scanner performance measures, these differences indicate that compensatory mechanisms may play a role in subjects' adjustment to the effects of TPM on cognitive performance. Certainly, BOLD signal differences between +TPM and −TPM groups of mixed groups of patients with epilepsy were already observed in one study [30]. In agreement with our findings, other studies also observed different and possibly compensatory fMRI activation patterns between patients with left or right TLE, or mixed epilepsy and healthy controls who were performing verbal memory/recognition tasks [33,52,53]. Of importance is that while TLE is overall associated with substantial impairment of cognitive function, such an impairment includes attention, concentration or executive functions only in a minority of patients [13,54]; poorly controlled TLE may be associated with neuroimaging evidence of ongoing bilateral and multifocal injury [55,56]. Thus, the observed differences between +TPM and −TPM patients are unlikely to be related to the epilepsy syndrome, and other reasons for these differences must be considered. Since the main difference between the groups with TLE was the use of TPM for the management of epilepsy, it is likely the cause of the observed BOLD signal differences. Therefore, since TPM is known to cause executive and working memory dysfunctions and may affect emotional well-being, the observed differences in medial frontal BOLD signal changes, i.e., in cingulate and bilateral superior frontal gyri, produced by TPM (Figs. 1c, d) may be an explanation for the previously reported neuropsychological testing results [3,6]. Further, previous neuroimaging studies noted semantic or material-specific retrieval and retrieval of familiar information to be associated with prefrontal fMRI activation [57–59]. Thus, differences in activation in this area associated with differences in neuropsychological performance suggest inefficient compensatory mechanisms employed by the +TPM subjects in order to perform the semantic decisions required by the SDTD task. Finally, the difference observed in the left inferior frontal region (decreased activation in +TPM patients with RTLE) is the likely explanation for the lower performance on linguistic testing in +TPM patients [3,60]. Overall, these neuroimaging findings provide some explanation for lower performance of +TPM subjects, whether healthy or with epilepsy, on neuropsychological measures of language and verbal memory and provide supportive evidence for the previously reported difficulties in obtaining valid lateralized language testing results in patients undergoing IAP [15–18].

While the etiology of the cognitive deficits in +TPM patients is not entirely clear, several studies postulated that it may be the carbonic anhydrase effect (and sulfa moiety) of TPM or other medications with similar effect that contributes to the observed cognitive findings [15,16,60,61]. Certainly, the results of the dose/BOLD signal level analysis support such a notion and are further corroborated by the results of a previous study in which fMRI signal was measured before and after administration of acetazolamide, a potent vasoactive drug that exerts its effect via carbonic anhydrase inhibition [31]. These authors reported cortical and subcortical increases and decreases in BOLD signal in response to carbonic anhydrase inhibition. Similar effects were recently described in a review focusing on TPM effects on verbal fluency (Fig. 1; other information not provided; [32]). Certainly, the pattern of BOLD signal changes observed in our study (Fig. 1e) is somewhat similar to that observed in studies of resting state networks in that increases in BOLD signal are observed in thalami and other deep brain structures and decreases in frontal/frontopolar and retrosplenial/occipital head regions [62–65]. This finding is consistent with a recent abstract indicating a negative effect of TPM on the default mode network in patients with epilepsy and healthy controls [66]. These authors postulate that the negative effects of TPM on verbal fluency may be, in part, related to failure to deactivate the resting state network while the subjects are performing a verbal fluency task. Certainly, the deactivations in frontal lobes associated with increased doses of TPM observed in our study are in agreement with this report and with a previous study of dose-dependence of the negative effects of TPM on cognition [1].

Limitations of this study should be mentioned. These include a moderate number of subjects with TLE (n = 32) which may not allow for detailed analysis of the effects of TPM and/or other AEDs on language function or detection of subtle effects, the retrospective nature of the neuropsychological data collection, lack of neuropsychological profiles of healthy controls, lack of post-operative outcomes data (language and seizure) in patients with TLE, and combining fMRI data from two different scanners. The severity of epilepsy and the lack of seizure control in patients included in this study are additional limitations that do not allow generalization of our findings to patients with milder forms of TLE. Further, the analysis of the dose effects of TPM is preliminary and will need to be evaluated in a study where the pre-TPM and on-TPM resting state data are collected. We are aware of at least one such ongoing study that was recently presented in abstract form [66].

To summarize, we observed negative effects of TPM on BOLD signal distribution and language lateralization in a group of patients with temporal lobe epilepsy as well as an overall dose-dependent effect on what appears to be involved in the resting state of the brain. As fMRI is being used more commonly in clinical practice, especially in the presurgical evaluation of epilepsy patients, medication effects on BOLD signal need to be considered when addressing patients' suitability for surgical management.