Address correspondence to Willem M. Otte, Rudolf Magnus Institute of Neuroscience/Image Sciences Institute, University Medical Center Utrecht, Yalelaan 2, 3584 CM Utrecht, The Netherlands. E-mail: firstname.lastname@example.org
Purpose: Diffusion tensor imaging (DTI) is used increasingly to study white matter integrity in people with temporal lobe epilepsy (TLE). Most studies report fractional anisotropy (FA) decrease and mean diffusivity (MD) increase in multiple white matter regions. The disturbance of white matter integrity varies across studies and between regions. We aimed to obtain a more consistent estimate of white matter diffusion characteristics and relate these to the distance from the seizure focus.
Methods: Studies comparing diffusion characteristics of people with epilepsy with those of healthy controls were systematically reviewed and quantified using random and mixed effects meta analysis. In addition to the overall meta-analysis, pooled FA and MD differences were determined per hemisphere and white matter category separately.
Key Findings: We included 13 cross-sectional studies. The pooled FA difference for all white matter was −0.026 (95% confidence interval [CI] −0.033 to −0.019) and MD difference was 0.028 × 10−3 mm2/s (95% CI 0.015–0.04). FA was reduced significantly in people with TLE compared with healthy controls in both ipsilateral (mean difference −0.03) and contralateral white matter (−0.02). MD was significantly increased ipsilaterally and contralaterally. MD differed significantly between white matter connected to the affected temporal lobe and remote white matter.
Significance: The meta-analysis provides a better estimation of the true diffusion characteristics. White matter structural integrity in TLE is disturbed more severely in the ipsilateral than in the contralateral hemisphere, and tracts closely connected with the affected temporal lobe are most disturbed. The exact underlying mechanisms remain to be elucidated.
Temporal lobe epilepsy (TLE) is the most common form of focal epilepsy and also the form most refractory to drug treatment (Engel, 1996). Pathology in this condition was assumed to be confined to the temporal gray matter, but with the use of diffusion-weighted magnetic resonance imaging (MRI) techniques, in particular diffusion tensor imaging (DTI), it has become apparent that both temporal and extratemporal white matter structures are also affected (Chen et al., 2008; Gross, 2011).
The two most frequently measured DTI-based parameters are the mean diffusivity (MD), a measure of the extent of tissue water diffusion within a voxel, and the fractional anisotropy (FA), a measure of the preferred directionality of diffusion within a voxel ranging between 0 (isotropic diffusion) and 1 (unidirectional diffusion). FA reduction and MD increase in the ipsilateral temporal lobes of people with epilepsy have been consistently reported (Dumas de la Roque et al., 2005; Gross, 2011), but the results in the contralateral temporal lobe and distant white matter are more unequivocal. This may be due to a smaller effect size, heterogeneous research groups, small sample sizes, and differences in experimental protocols. To obtain a better estimate of white matter diffusion characteristics in TLE, we pooled several similar DTI studies in people with epilepsy to allow for the construction of more homogeneous and larger groups with a higher statistical power.
Findings suggest that changes in FA and MD are most pronounced in the ipsilateral temporal white matter, that is, close to the epileptogenic zone (Gross, 2011). The underlying mechanism of white matter changes remains unknown. We hypothesized that (1) the source of these changes is in the epileptogenic zone in the affected temporal lobe and (2) the severity of these changes will decrease with the (functional) distance from the temporal seizure focus. We pooled and statistically compared MD and FA data from ipsilateral and contralateral white matter areas, and compared MD and FA between white matter in directly connected tracts (i.e., fornix and cingulum), white matter from tracts that pass through the temporal white matter (closely connected), and white matter in remote tracts that do not originate in, or pass through, the temporal white matter (remotely connected). We hypothesized that TLE is associated with significantly different effects on the directly connected, closely connected, and remotely connected white matter regions.
Information sources and search
Studies were identified by searching electronic databases and reference lists of articles with no language restriction. This search was applied to PubMed (NCBI), ISI Web of Science (Thomson Reuters), and Embase (Excerpta Medica Database). See Appendix 1 for the full PubMed, Embase, and ISI Web of Science search strategy. The last search was run on August 13, 2011.
Studies were included if the following criteria were met: Study design: cross-sectional design including a control group, investigating the relation between temporal lobe epilepsy and white matter structure by means of DTI, independent of field strength and number of diffusion directions. Analysis method: white matter region-of-interest (ROI) analysis. Subjects: participants of any age with mesial or lateral temporal lobe epilepsy. Outcome measures: FA and MD, reported as mean ± standard deviation (SD) or standard error of the mean (SEM). If published studies did not provide sufficient data for inclusion in the meta-analysis, the corresponding authors were contacted to provide this information. If necessary, reported apparent diffusion coefficient (ADC) trace values were divided by 3 to obtain the corresponding MD. Studies using voxel-based methods including tract-based spatial statistics and voxel-based morphometry were excluded, as more sophisticated techniques, such as signed differential mapping (Radua & Mataix-Cols, 2009), are needed for meta-analysis of whole-brain data.
Study selection and data collection
One reviewer (WMO) performed the literature search and screened all titles and abstracts. From all potentially relevant articles, full-text versions were retrieved and screened by two reviewers (WMO and PvE). Data extraction was performed by one reviewer (WMO) and checked by another (PvE). Disagreements were resolved by consensus.
Extracted data from each study included year of publication, characteristics of study population (age at investigation, sex, age at TLE onset, duration of illness, involved hemisphere, type of pathology, and type of TLE), DTI parameters (including field strength, number of diffusion-weighted directions, and b-value), mean and SD of FA and MD values per white matter region for both the affected and contralateral hemisphere, and the total number of participants allocated in both study arms.
Methods of analysis
Mean FA and MD differences and 95% confidence intervals (CIs) between groups were calculated by fitting a random effects weighted mean difference model using restricted maximum likelihood estimation (Schmidt et al., 2009). This procedure provides a weighted average, with the weights expressing the amount of information contributed by each study. In this way, larger studies have more influence on the pooled FA and MD difference estimates. This weighted analysis includes two steps. First the heterogeneity (τ2) between the individual studies entered in the meta-analysis was calculated using the restricted maximum likelihood estimator. Next, the FA and MD difference between controls and people with epilepsy in each study was multiplied by the inverse sum of the variance (σ2) and overall heterogeneity (i.e., for study i, weighti = 1/σ2i + τ2). We used this weighted random effects model, as it accounts for interstudy variation and provides a more conservative effect than a fixed effects model. In addition, we corrected for uncertainty in the τ2 estimation using the Knapp and Hartung method (Knapp & Hartung, 2003).
The meta-analyses and creation of forest plots were performed in the rmeta (Lumley, 2004) and metafor (Viechtbauer, 2010) packages in R, version 2.13 (Team, 2011). For meta-regression we used a mixed effects model, where the modifiers included (1) hemisphere (whether a ROI was in the ipsilateral or contralateral hemisphere), (2) category (directly connected to the temporal lobe, closely connected to the temporal lobe or remote white matter) and (3) ROI. The white matter ROIs considered in (2) as directly connected were the cingulum and fornix. ROIs closely connected to the temporal lobe included the uncinate fasciculus, the inferior longitudinal fasciculus, and the arcuate fasciculus. ROIs considered as remote included the anterior and posterior corpus callosum, external capsule, posterior and anterior internal capsule, superior longitudinal fasciculus, corticospinal tract, and inferior fronto-occipital fasciculus. The effect modifier (2) was calculated only for the ipsilateral white matter regions. A p-value < 0.05 was considered statistically significant. p-values < 0.001 are indicated as ***, p-values < 0.01 as ** and p-values < 0.05 as *.
In addition to FA and MD, the DTI-derived axial and radial diffusivity also inform about white matter integrity. Too few studies reported on these measures; therefore, meta-analyses were run for FA and MD only.
Meta-analysis was performed for FA in 13 white matter ROIs (Fig. 2) and for MD in 12 white matter ROIs (Fig. 3). Values are presented as absolute FA and MD difference, with the control group as reference. White matter FA was significantly lower in people with TLE than in controls: −0.026 (95% CI −0.0328 to −0.0192; p < 0.0001). MD was significantly increased: 0.0275 (×10−3 mm2/s) (95% CI 0.0154–0.0397; p < 0.0001).
Mixed effects meta-analysis results on FA and MD differences for the separate bilateral, ipsilateral, and contralateral white matter ROIs are shown in Table 2. FA reductions were most pronounced in the ipsilateral white matter.
Table 2. Mean fractional anisotropy (FA; left) and mean diffusivity (MD; right) differences between people with TLE and controls per white matter region-of-interest (ROI) are specified for the bilateral, ipsilateral, or contralateral hemispheres
Mixed effects meta-analysis for the overall ipsilateral and contralateral white matter ROIs indicated significantly decreased FA ipsilaterally: −0.0295 (95% CI −0.0384 to −0.0205; p < 0.0001), and contralaterally: −0.0171 (95% CI −0.0282 to −0.0660; p = 0.0032) (Fig. 4). MD was significantly increased ipsilaterally: 0.0722 (×10−3 mm2/s) (95% CI 0.022–0.1224; p = 0.0059) and contralaterally: 0.0838 (95% CI 0.0107–0.1569; p = 0.0258) (Fig. 4).
Figure 5 shows the mixed effects meta-analysis results for the three types of categorized ipsilateral white matter ROIs. FA was lower in people with TLE than in controls in all categories of white matter, but FA differences were most pronounced in the closely connected white matter category: −0.0335 (95% CI: −0.0499 to −0.0171; p = 0.0002) (Fig. 5A). MD was increased significantly in the closely connected white matter category: 0.1649 (×10−3 mm2/s) (95% CI 0.0887–0.2411; p = 0.0002) (Fig. 5B). In addition, significant MD differences were found between closely and remotely connected white matter (p < 0.01, Fig. 5B).
We have assessed, through meta-analysis, the possible association of TLE and changes in global white matter integrity as measured with DTI.
Our results confirm that FA is reduced significantly in the majority of white matter regions, whereas MD is significantly increased (Figs. 2 and 3). The FA decrease and MD increase are not restricted to the ipsilateral hemisphere but extend contralaterally (Fig. 4). In addition, white matter passing through the temporal lobe tends to be more affected than either remote white matter regions or directly connected white matter. Voxel-based DTI studies not included in the meta-analyses support the findings of bilaterally decreased FA and increased MD in the white matter of people with TLE (Focke et al., 2008; Riley et al., 2010; Afzali et al., 2011). If more voxel-based TLE studies become available, signed differential mapping would be the recommended statistical technique for undertaking meta-analysis of white matter group differences at a higher spatial resolution (Radua & Mataix-Cols, 2009).
DTI is much more sensitive than conventional MRI in detecting white matter abnormalities. It measures the movement of water molecules along white matter tracts. Diffusion perpendicular to the main white matter tract direction is more hindered by cell membranes and myelin layers than is diffusion along the main axis. This directionality is quantified as the FA, whereas MD measures the overall motion of water molecules independent of tissue directionality (Pierpaoli & Basser, 1996; Le Bihan et al., 2001). Changes in FA and MD could, therefore, be related to a change in perpendicular diffusivity, a change in diffusivity along the main axis, or both. Different mechanisms have been suggested that could change MD and FA: intracellular and extracellular fluid exchange, blood–brain barrier damage, neuronal cell death, or axonal demyelination (Wall et al., 2000; Song et al., 2003; Seidenberg et al., 2005). Recently more insight into the underlying structural tissue changes has been gained by two studies combining DTI with histologic white matter examination. First, a report on postsurgical fimbriae specimens from people with TLE found an increased extraaxonal fraction, reduced axonal membrane circumference, and reduced myelination (Concha et al., 2010). The presurgical FA values correlated significantly with the reduction in axonal membrane circumference and showed a (nonsignificant) trend toward decreased myelin thickness. No significant correlations were found between histology and MD values. Second, the importance of myelin structures contributing to remote white matter integrity changes in the process of TLE epileptogenesis has also been shown. Decreased staining with luxol fast blue (a myelin marker) and loss of characteristic banding was found in the corpus callosum and bilateral fimbriae before the first spontaneous recurrent seizures occurred and preceded a corpus callosum FA reduction (van Eijsden et al., 2011).
The pathophysiologic mechanisms underlying the bilateral changes in white matter structure in people with TLE are yet unknown. The widespread changes could be explained by the initial lesion or epileptogenic process that underlies and precedes the seizure onset; the direct effect of axonal damage and associated secondary wallerian degeneration in ipsilateral bundles due to repetitive seizure spread and distal effects of frequent interictal spikes propagating through white matter parts of the epileptic network; the systemic effects of seizures such as hypoxia and vasoconstriction; antiepileptic drug treatment; altered brain development; and plasticity-related reorganization of local and global white matter networks. It could be, however, that these different mechanisms are not equally involved in damaging all white matter regions. For example, the meta-analytic results on ipsilateral MD changes show a significantly larger MD increase in the white matter passing though the temporal lobe than in remote white matter (Fig. 5). No statistical differences were found between white matter regions that are directly and remotely connected with the affected temporal lobe.
Clearly, additional clinical and translational studies are needed to identify directly contributing factors and underlying mechanisms. A better understanding of white matter changes in people with TLE could eventually lead to surrogate markers in assessing treatment outcome and to an improved selection of epilepsy surgery candidates.
Our study has limitations. We have corrected for statistical heterogeneity in the meta-analyses. Nevertheless there was a substantial variability in ages, age at onset of epilepsy, and duration of epilepsy. DTI parameters were also variable. The ROIs are defined by different neuroanatomists and will probably differ in exact location and size between studies. In addition, the accuracy of measurements of FA (differences) in white matter is dependent on slice thickness (thicker slices potentially result in more partial volume of gray matter) and number of diffusion-weighted directions (more directions increase the signal-to-noise ratio and robustness of the diffusion tensor model fit). Differences in the DTI parameters are probably one of the causes of the heterogeneity of measured FA and MD differences between studies.
Another potential source of variation in our meta-analyses results is the combined analysis of data from both the left and right sides of seizure onset. There is evidence from recent studies that changes in white matter integrity are not identical for left and right TLE (Ahmadi et al., 2009; Kemmotsu et al., 2011). A significant lower FA was found in the uncinate fasciculus and inferior longitudinal fasciculus in left-sided TLE as compared to right-sided TLE (Kemmotsu et al., 2011). We were unable to take this potential difference between the hemispheres into account in the meta-analyses, as not all the included studies separated patients based on side of seizure onset.
Finally, we were unable to identify the white matter integrity changes for mesial or neocortical TLE separately, as most studies combined these patient populations. The combined analyses of both mesial and neocortical TLE data in our meta-analyses could possibly be the cause of the lack of difference in MD between the white matter tracts directly connected to the temporal lobe and the closely connected white matter. Exact information on the epileptogenic source could be useful in tracking down the relationship between remote white matter and temporal lobe focus and should be addressed in future work.
We thank Dr. D.W. Gross, L. Concha, J.C. Schoene-Bake, S. Rodrigo, and C. Oppenheim for kindly providing the requested data for our study. We thank Dr. G.S. Bell for her valuable comments on the manuscript. JWS and JSD are based at UCLH/UCL Comprehensive Biomedical Research Centre, which receives a proportion of funding from the Department of Health’s NIHR Biomedical Research Centres funding scheme. JWS is supported by the Epilepsy Society and the Dr Marvin Weil Epilepsy Research Fund. KPJB and WMO were supported by the Dutch National Epilepsy Fund (NEF 08-10).
None of the authors has any conflict of interest in relation to this work to disclose. 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.
Search strategy: PubMed (NCBI)
(partial epilepsy[text] or partial epilepsy[mesh] or temporal lobe epilepsy[mesh] or temporal lobe epilepsy[text]) and (diffusion magnetic resonance imaging[mesh] or diffusion MRI[text] or diffusion tensor imaging[mesh] or diffusion tensor imaging[text] or diffusion weighted imaging[text] or dwi[text] or dti[text]).
Search strategy: Embase (Ovid)
“temporal lobe epilepsy”/exp and (“diffusion tensor imaging”/exp or “diffusion tensor imaging”/syn or “diffusion weighted imaging”/exp).
Search strategy: ISI Web of Science (Thomson Reuters)
Topic=((temporal lobe epilepsy or partial epilepsy) and (diffusion tensor imaging or diffusion weighted imaging or dti or dwi)).