Article Text

Download PDFPDF

Research paper
Epilepsy-related cytoarchitectonic abnormalities along white matter pathways
  1. G Russell Glenn1,2,3,
  2. Jens H Jensen1,2,
  3. Joseph A Helpern1,2,3,
  4. Maria V Spampinato2,
  5. Ruben Kuzniecky4,
  6. Simon S Keller5,6,
  7. Leonardo Bonilha7
  1. 1Center for Biomedical Imaging, Medical University of South Carolina, Charleston, South Carolina, USA
  2. 2Department of Radiology and Radiological Science, Medical University of South Carolina, Charleston, South Carolina, USA
  3. 3Department of Neuroscience, Medical University of South Carolina, Charleston, South Carolina, USA
  4. 4Department of Neurology, New York University, New York City, New York, USA
  5. 5Department of Molecular and Clinical Pharmacology, Institute of Translational Medicine, University of Liverpool, Liverpool, UK
  6. 6Department of Clinical Neuroscience, Institute of Psychiatry, King's College London, London, UK
  7. 7Department of Neurology, Medical University of South Carolina, Charleston, South Carolina, USA
  1. Correspondence to Dr Leonardo Bonilha, Comprehensive Epilepsy Center, Department of Neurology, Medical University of South Carolina, 96 Jonathan Lucas St, Suite 301, Charleston 29425, SC, USA; bonilha{at}musc.edu

Abstract

Objective Temporal lobe epilepsy (TLE) is one of the most common forms of epilepsy. Unfortunately, the clinical outcomes of TLE cannot be determined based only on current diagnostic modalities. A better understanding of white matter (WM) connectivity changes in TLE may aid the identification of network abnormalities associated with TLE and the phenotypic characterisation of the disease.

Methods We implemented a novel approach for characterising microstructural changes along WM pathways using diffusional kurtosis imaging (DKI). Along-the-tract measures were compared for 32 subjects with left TLE and 36 age-matched and gender-matched controls along the left and right fimbria-fornix (FF), parahippocampal WM bundle (PWMB), arcuate fasciculus (AF), inferior longitudinal fasciculus (ILF), uncinate fasciculus (UF) and cingulum bundle (CB). Limbic pathways were investigated in relation to seizure burden and control with antiepileptic drugs.

Results By evaluating measures along each tract, it was possible to identify abnormalities localised to specific tract subregions. Compared with healthy controls, subjects with TLE demonstrated pathological changes in circumscribed regions of the FF, PWMB, UF, AF and ILF. Several of these abnormalities were detected only by kurtosis-based and not by diffusivity-based measures. Structural WM changes correlated with seizure burden in the bilateral PWMB and cingulum.

Conclusions DKI improves the characterisation of network abnormalities associated with TLE by revealing connectivity abnormalities that are not disclosed by other modalities. Since TLE is a neuronal network disorder, DKI may be well suited to fully assess structural network abnormalities related to epilepsy and thus serve as a tool for phenotypic characterisation of epilepsy.

Statistics from Altmetric.com

Request Permissions

If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.

Introduction

Temporal lobe epilepsy (TLE) is the most common form of medically intractable focal epilepsy and is frequently associated with hippocampal sclerosis (HS)1. Despite that hippocampal pathology is generally considered the primary seizure generator and principal node in a temporal epileptiform network in TLE,2 there is a sizeable literature indicating that structural abnormalities extend beyond the medial temporal lobe. Many studies have reported grey matter atrophy, white matter (WM) loss and gliosis affecting extrahippocampal and extratemporal regions.3–6 Crucially, the distribution of tissue damage in TLE is not random but follows an anatomical and functional pattern whereby the most affected regions are those directly or indirectly associated with the medial temporal lobe and the limbic system.7–9 This regular distribution of damage implies that a limited number of common pathophysiological mechanisms are responsible for brain injury in TLE. In particular, grey matter loss may be caused by cellular excitoxicity along the limbic path of seizure spread, or by deafferentation injury from loss of neural connectivity.10

However, the full extent of microstructural brain damage in TLE is still incompletely understood, and most patients with TLE demonstrate some degree of extrahippocampal abnormality.11 Importantly, seizure control after pharmacological and surgical intervention can vary significantly among patients with TLE, and there are clearly distinct phenotypes of TLE when it comes to treatment responsiveness. For this reason, it is fundamentally important to accurately assess in vivo patterns of brain injury in TLE, with special emphasis to cytoarchitectonic features of tissue damage and their anatomical distribution.

Previous studies have investigated alterations in WM pathways in TLE using diffusion tensor tractography.12–14 However, these studies predominantly use whole-tract analyses, which are limited as pathological changes may be concentrated in anatomically specific regions and whole-tract analyses may obstruct the detection of focal pathology. Moreover, diffusion tensor imaging (DTI) is incapable of detecting multiple, intravoxel fibre bundle orientations in complex neurological tissue, which limits its potential for tractography.15 ,16 Diffusional kurtosis imaging (DKI) extends conventional DTI by estimating the diffusion and kurtosis tensors to quantify restricted, non-Gaussian diffusion that occurs in biological tissues.17 ,18 Accordingly, DKI has demonstrated improved sensitivity for detecting neuropathology in a variety of conditions including epilepsy,19–22 stroke,23–26 Alzheimer's disease27–29 and numerous others. More recently, the advantages of DKI have been leveraged to provide more comprehensive assessment of diffusion in complex neural environments, including the characterisation of diffusion anisotropy beyond the conventional fractional anisotropy (FA)30 and computation of DKI-based WM tractography, enabling the resolution of multiple intravoxel fibre bundles.16 ,31 These advantages are improved by using DKI in conjunction with automated fibre quantification (AFQ),32 for characterisation of tissue microstructure along WM pathways, by incorporating a more comprehensive and potentially more sensitive collection of parameters for detecting disease-related pathology than does DTI. Thus, DKI is remarkably synergistic with AFQ, and the combination of the two forms a particularly effective imaging method for detecting pathological WM changes.

In this present study, we applied a novel neuroimaging approach combining the strengths of DKI and AFQ for the non-invasive characterisation of pathological WM changes in TLE. We hypothesise that cytoarchitectural abnormalities follow a crescendo gradient towards the temporal lobe with pathological effects concentrated in particular WM regions, revealing patterns of neuroarchitectural pathology associated with TLE, potentially underlying distinct phenotypical subtypes.

Methods

Subjects

This study was approved by the Institutional Review Board at the Medical University of South Carolina (MUSC). We evaluated data from 32 consecutive subjects with left TLE who were followed at the Comprehensive Epilepsy Center at MUSC. All subjects were diagnosed with left TLE in concordance with the diagnostic criteria proposed by the International League Against Epilepsy (ILAE),33 including classical mesial TLE characteristics as revealed through seizure semiological investigations by experienced epileptologists in combination with surface EEG investigations and MRI, with the majority of subjects (84%) demonstrating neuroradiological evidence of HS. The mean (±SD) age of all subjects was 44.8 (±16.7) years, and included 10 males and 22 females. A control group of 36 age-matched and gender-matched healthy individuals with no history of neurological problems was also recruited from the local community. Control subjects had a mean (±SD) age of 40.4 (±11.6) years, including 12 males and 24 females. Clinical and demographic information for the subjects with TLE included in this study is further described in the table provided in the online supplementary material. The subjects included in this study are also described in a previous study from our group using voxel-based methods without tractography.22

Supplemental material

Our cohort contained subjects with varying disease severity, including subjects with recently diagnosed TLE and subjects whose seizures were well controlled with antiepileptic drugs (AEDs). Thus, subjects in this cohort were not all surgical candidates. Subjects well controlled on AEDs were identified by having one or fewer seizures per 6 months (n=13), and subjects not well controlled on AEDs were identified by having more than one seizure per 6 months (n=19).

Image acquisition and analysis

Image acquisition was performed on a 3 Tesla Magnetom Verio MRI scanner (Siemens Medical, Erlangen, Germany) and included a DKI dataset and T1-weighted images. DKI analysis included characterisation of mean diffusivity (MD) and FA from the diffusion tensor and corresponding mean kurtosis (MK) and kurtosis fractional anisotropy (KFA).30 DKI-derived tractography16 ,31 was performed using diffusional kurtosis estimator software (https://www.nitrc.org/projects/dke/). DKI was incorporated into the AFQ image processing pipeline (https://github.com/jyeatman/AFQ) using fully automated in-house scripts, which included along-the-tract characterisation of the fimbria-fornix (FF), parahippocampal WM bundle (PWMB), arcuate fasciculus (AF), inferior longitudinal fasciculus (ILF), cingulum bundle (CB) and uncinate fasciculus (UF).

The effects of seizure burden and seizure control with AEDs were tested in the PWMB and CB, as these limbic pathways are crucial for the progression of disease,12 neuropsychological manifestations of TLE14 and differentiation of TLE subtypes by treatment response including surgical outcomes34 ,35 and pharmacoresistance.36 Seizure burden was defined as equal to log10 (frequency×duration), with the logarithm being applied to accommodate subjects with very high seizure frequency, and the effects were assessed using Pearson's product-moment correlation coefficient.

A summary of the image analysis steps for a single subject is given in figure 1, and a detailed description of our image acquisition protocol and image analysis steps is given in the online supplementary material.

Figure 1

AFQ with DKI. (A) DKI uses multiple diffusion weighting b values and diffusion encoding directions to characterise non-Gaussian diffusion which occurs in vivo. The images shown include an average b=0 image along with images with diffusion weightings of b=1000 and 2000 s/mm2 for a single diffusion-encoding direction. (B) Images in the DKI dataset are combined to estimate the DT and KT, which characterise the 3D intravoxel diffusion dynamics based on physical properties of water diffusion. (C) The diffusion and kurtosis tensors are then analysed to generate scalar, quantitative parameter maps that can be used to characterise tissue microstructure. (D) The diffusion and kurtosis tensors are combined to perform DKI-based tractography, which can improve tractography relative to DTI by enabling the resolution of multiple intravoxel fibre bundles in complex neural tissue. (E) AFQ performs a series of automated steps to segment fibre groups from standardised WM ROIs and then isolates each fibre group's tract core for analysis of the diffusion parameters. Each subject generates tract profiles for each diffusion metric along each tract core, which can be compared to investigate individual and group-wise differences. AFQ, automated fibre quantification; DKI, diffusional kurtosis imaging; DT, diffusion tensor; DTI, diffusion tensor imaging; KT, kurtosis tensor; ROIs, regions of interest; WM, white matter.

Statistical analysis

Individual tract profiles were averaged over five regions of interest (ROIs), and a two-sample t test was performed to determine the significance of group-wise differences. Significance levels were corrected for multiple comparisons using the false discovery rate (FDR) procedure.37 For correlations with seizure burden, statistical significance was corrected for multiple comparisons with the FDR procedure, and the effects of pharmacoresistance were tested using the well-controlled and not well-controlled groups using a two-sample t test. Cohen's d parameter was used to quantify the effect size. The ROIs used in this study are illustrated in figure 2.

Figure 2

The location of WM ROIs is defined from the reconstructed fibre tracts. The insert for each fibre group in the upper right-hand corner illustrates WM tracts identified by AFQ and DKI for a single subject, overlaid on the corresponding b=0 image. The solid black line indicates the core of each tract used in generating the individual tract profiles. Tract cores identified for all subjects in this study are averaged and overlaid on an anatomical MRI template to illustrate the group-wise representation of each fibre group. Each fibre group is divided into five ROIs, with increasing ROI numbers indicating regionally specific locations in each tract. The ROIs in this figure correspond to the ROIs used in the tables included in this study. AFQ, automated fibre quantification; DKI, diffusional kurtosis imaging; ROIs, regions of interest; WM, white matter.

Results

Group-wise tract profiles for all fibre groups are shown in figure 3. The tract profiles demonstrate similar along-the-tract variation of the diffusion metrics between subjects and controls and between the ipsilateral and contralateral hemispheres. Importantly, these results demonstrate that epilepsy-related abnormalities can be restricted to specific regions of each tract, which would be undetected by methods that group all data from one tract into a single value. The results in figure 3 are tabulated in the online supplementary material.

Figure 3

Mean tract profiles (±SEM) for ipsilateral and contralateral fibre groups demonstrate regional group-wise differences in diffusion metrics between subjects and controls. Group-wise differences are tested over bins indicated by the shaded bars and summary statistics for group-wise comparisons are given in the online supplementary material. Comparisons marked with an asterisk (*) have p values <0.05, and a double asterisk (**) indicates p values <0.005, after correction the significance level for multiple comparisons using the FDR procedure. The vertical bins correspond to the ROIs illustrated in figure 2, with increasing ROI number corresponding to increasing tract section number. MD is in units of μm2/ms, while the remaining parameters are dimensionless. FDR, false discovery rate; MD, mean diffusivity; ROIs, regions of interest.

In general, MD is higher in subjects with TLE relative to controls in all ROIs and all fibre groups with the exception of one ipsilateral ROI (ROI 3 in the UF) and eight contralateral ROIs (ROIs 1 and 5 in the FF, ROI 1 in the AF, ROIs 2 and 3 in the UF and ROIs 3–5 in the right ILF), although the observed changes were not found to be statistically significant. FA tended to be lower in subjects with TLE relative to controls, with statistically significant reductions being found in ROIs 4 and 5 of the ipsilateral AF.

MK demonstrated significant reduction in the ipsilateral FF, PWMB and UF in multiple ROIs. In the ipsilateral FF and UF, this reduction was more pronounced with increasing ROI number (further anteriorly within the temporal lobe). MK showed statistically significant reductions in all ROIs in the bilateral AF and ILF, except for ROI 1 in the contralateral AF and ROI 3 in the contralateral ILF, with the ipsilateral side tending to demonstrate a stronger effect size.

The location and relative significance of the observed differences are illustrated in the section-wise t score plots in figure 4. Qualitatively, the abnormal t scores demonstrated a crescendo effect increasing in significance into the temporal lobe. Similar to the tract profiles, the section-wise t score plots demonstrated a slight, but general increase in MD and decrease in FA in subjects relative to controls. With MK, the changes can be seen bilaterally, with the effect being the largest within the ipsilateral temporal lobe.

Figure 4

Section-wise t score plots summarise the observed differences in the tract profiles. Section-wise t scores are calculated from the tract profiles illustrated in figure 3. These are overlaid on a WM template at positions indicated by the average of the tract cores for all participants included in this study. Section-wise t scores provide a visual representation of where pathological changes occur, with dark red indicating greater group-wise reductions in the subject versus control groups and dark blue indicating greater group-wise increases in the subject versus control groups. WM, white matter.

Correlations with seizure burden are illustrated in table 1. Significant correlations were found in the PWMB and CB, with MD demonstrating significant correlations on the ipsilateral hemisphere and MK and KFA demonstrating bilateral limbic effects. In the ipsilateral PWMB, significant correlations were found for MD, MK and KFA in ROI 3, with the correlations extending further along the tract anteriorly and posteriorly with MD and KFA. In the ipsilateral CB, significant correlations were found in ROI 5 for MD, ROIs 2–5 for MK and all ROIs for KFA. On the contralateral side, significant correlations with MK were found in ROI 3 of the PWMB and ROIs 2–5 of the CB, and with KFA in ROIs 3 and 4 of the PWMB and ROI 5 of the CB.

Table 1

Correlations with seizure burden for the PWMB and CB

Comparisons between AED responsive and unresponsive groups are illustrated in table 2. Uncorrected p values <0.05 were found in comparing subjects well controlled with AEDs with those poorly controlled with AEDs for the ipsilateral PWMB in ROI 3 in MD and ROIs 3–4 in KFA, as well as for the ipsilateral CB in ROI 5 in MD and all ROIs with the anisotropy parameters, FA and KFA. Uncorrected p values <0.05 were also found for the contralateral CB in MK in ROI 2 and KFA in ROI 5. While none of these attained statistical significance following FDR correction, they may be indicative of trends that would warrant further investigation with a larger sample size. For example, the not well-controlled group demonstrated a 21% reduction in KFA in ROI 2 of the ipsilateral CB compared with the well-controlled group with a Cohen's d parameter of −1.262, suggesting a potentially large effect.

Table 2

AED responsiveness for the PWMB and CB

Discussion

In this study, we employed a novel neuroimaging technique that combines DKI and AFQ for the in vivo characterisation of cytoarchitectonic abnormalities along WM pathways which are physiologically relevant for TLE. In accordance with the previous literature, we detected pathological changes in several extrahippocampal and extratemporal WM tracts in subjects with TLE. The important novel findings of this study pertain to the superior sensitivity of DKI-based tractography to identify and localise intrapathway structural connectivity abnormalities in TLE. These observations complement our initial reports of increased sensitivity of DKI in scalar diffusion voxel-based maps of subjects with epilepsy.22 This is the first study to use DKI-based tractography combined with AFQ, demonstrating how DKI tractography can overcome limitations imposed by fibre crossing and unveil epilepsy-related abnormalities. Our data indicate that group-wise reductions in MK are observed in regionally specific areas of the ipsilateral FF, UF and PWMB, as well as more diffuse bilateral abnormalities in the ILF and AF (figure 3). We also report significant effects of seizure burden on MD, MK and KFA of ipsilateral limbic pathways. MK and KFA indicated additional correlations with seizure burden in contralateral pathways (table 1). The overall salience of these findings hinges on the technical innovations of these new forms of tractography and the critical need to better define phenotypic characterisations of subjects with epilepsy.

Technical innovations

This is the first study to combine DKI and AFQ for the fully automated detection of cytoarchitectonic alterations along WM fibre pathways, which may be a particularly sensitive method for assessing WM tissue microstructure. With scalar, voxel-based data, it is not always clear which pathways are compromised. For example, an abnormal voxel in an ROI corresponding to the ILF may be related to transverse fibres in the same region. By defining which specific tracts are abnormal, one can develop a more detailed understanding of the distribution of cytoarchitectonic abnormalities. The methodological benefits of these approaches are further enhanced when augmented with along-the-tract measures, which identify the structurally compromised tracts and additionally have the capability to localise specific abnormalities within the long axis of a tract. Moreover, the tract cores analysed can preserve a significant amount of intersubject anatomical tract variability while still enabling group-wise comparisons, which can help avoid normalisation errors that complicate conventional voxel-wise techniques. This is further improved by using DKI, which characterises higher order diffusion dynamics compared with DTI and can thus describe more complex diffusion profiles. Consequently, DKI enables the detection of crossing WM fibre bundles for diffusion tractography and provides a more comprehensive collection of quantitative parameters, which may enhance the detection of disease-related abnormalities. Thus, the combination of DKI and AFQ creates an effective tool for characterising WM pathways, enabling further insights into patterns of neuroarchitectural pathology that occur in numerous neurological and psychiatric disorders.

Towards a phenotypic microstructural connectivity characterisation of TLE

Increasingly, advanced neuroimaging techniques have demonstrated localised and networked cytoarchitectonic abnormalities in TLE with limbic alterations potentially underlying various clinicopathological features of the disorder, including the pathological mechanisms that lead to medically intractable TLE,12 neuropsychological impairments,14 AED response36 and surgical outcomes.34 ,35 In the present study, we recruited a cohort of 32 consecutive subjects diagnosed with left TLE, which comprised subjects with various disease severities. DKI in combination with AFQ detected pathological WM alterations consistent with our understanding of TLE as a network disease having tissue abnormalities concentrated in the temporal lobe of the brain. Moreover, statistical trends were observed in limbic structures between subjects whose seizures were well controlled with AEDs and those who had worse AED control (table 2), which could be an important clinical prognosticator. Interestingly, KFA in the ipsilateral PWMB and CB correlated with seizure burden, and we observed trends for differences in tract characteristics between subjects who had well-controlled seizures and those who did not, despite no detectible group-wise differences in this region with normal controls. A similar trend was seen between subjects who had well-controlled seizures with AEDs and those who did not in FA in the ipsilateral CB. A possible explanation for this is that distinct mechanisms may underlie AED response compared to pharmacoresistance, with AED responders having higher than normal diffusion anisotropy and subjects whose seizures were not well controlled having lower than normal diffusion anisotropy in these limbic structures. This also supports the need for the improved sensitivity in detecting patterns of neuroarchitectural alterations in TLE afforded by DKI. Moreover, DKI detected contralateral changes in MK that were not apparent in analysis of the conventional diffusivity-based parameters of MD and FA.

This study also extends the work of Concha et al,38 where along-the-tract measures were assessed in the ILF, AF and UF using a manual segmentation routine with DTI in subjects with medically intractable TLE. In that work, it was argued that the changes in diffusion metrics could reflect astrogliosis and microstructural alterations related to the occurrence of seizures with potential effects of postictal vasogenic oedema. In the present study, the reduction in MK reflects a net loss in the complexity of microstructural tissue compartmentalisation, which is also consistent with subtle pathological denervation. By including a more comprehensive assessment of along-the-tract diffusion abnormalities, the proposed technique may provide an important step towards a better understanding of the neuroarchitectural alterations that occur in TLE, as well as the development of fully automated imaging biomarkers for the separation of TLE subtypes based on clinically important distinctions.

Limitations

By focusing this study on tract profiles within the AFQ-identified tract cores and using only a subset of the possible DKI-derived diffusion metrics, we have substantially restricted the scope of our analysis. This is a potential limitation of this study, as there may be important disease-related differences missed outside of the tract cores. Moreover, the quantitative parameters employed in this study depict physical properties of water diffusion which may be differentially influenced by multiple, distinct factors.15 To address this limitation, DKI-based WM modelling techniques can be applied, which may improve the specificity of the observed changes.39 The subject cohort included in this study comprised individuals with left TLE, as left and right TLE may have intrinsically different pathological effects on temporal lobe structures.40 Thus, we were not able to assess the effects of right-sided TLE. This study also comprised individuals with varying disease severity, including recently diagnosed and chronic TLE, subjects with and without neuroradiological signs of HS and subjects whose seizures were well controlled and not well controlled with AEDs. We do observe slight reductions in seizure frequency, seizure burden and along-the-tract diffusion abnormalities in the ipsilateral FF in subjects with TLE without signs of HS; however, with only five subjects without MRI evidence of HS, we were unable to test the statistical significance of this observation. Well-controlled and intractable TLE may also represent distinct pathological mechanisms; so by including both groups, sensitivity may be lost in characterising regionally specific distinctions. Nevertheless, combining DKI with AFQ revealed distinct patterns of cytoarchitectonic abnormalities, which highlights the sensitivity as well as the potential applicability of the proposed technique.

Conclusion

There are measurable differences in WM tissue that are not routinely considered in the clinical assessment of subjects with unilateral TLE. We have described a diffusion MRI-based image analysis technique that, by combining the strengths of DKI and AFQ, can quantify cytoarchitectonic abnormalities in specific, WM fibre pathways. The proposed technique is shown to detect group-wise pathological changes, with the largest effect sizes lateralising to the ipsilateral temporal lobe and extending along the tracts from the ipsilateral temporal lobe and including the contralateral side of the brain. Microstructural changes are also found to correlate with seizure burden in specific limbic pathways and trends are found towards detecting differences between subjects with well-controlled and not well-controlled TLE. Combining DKI and AFQ may be a particularly effective neuroimaging technique for detecting microstructural alterations along physiologically relevant WM pathways that could provide further insights into the variable clinical course of TLE, as well as a wide array of other neuropsychological conditions affecting the structural organisation of the human brain.

References

Footnotes

  • Contributors GRG, LB and JHJ conceived the project. LB and MVS collected the data. GRG, LB, JHJ and SSK designed experiments, planned comparisons to be made and conceived statistical analyses. GRG wrote software for incorporating DKI and AFQ and analysed the data. GRG, JHJ, JAH, MVS, RK, SSK LB interpreted the data, drafted the manuscript and approved the final version to be published. GRG is a guarantor.

  • Funding This work was supported in part by the National Institutes of Health grant T32GM008716 (to P. Halushka) and by the Litwin Foundation (to JAH).

  • Competing interests None declared.

  • Ethics approval Institutional Review Board at the Medical University of South Carolina.

  • Provenance and peer review Not commissioned; externally peer reviewed.