Restriction spectrum imaging reveals decreased neurite density in patients with temporal lobe epilepsy

INTRODUCTION

Temporal lobe epilepsy (TLE) is the most common localization-related epilepsy and may be highly refractory to medication management¹. Although once considered to be a disease of the hippocampus, it is now well-appreciated that TLE is a network disorder associated with damage to multiple cortical and subcortical structures that extend well beyond the medial temporal lobes^{2; 3}. In fact, studies using diffusion tensor imaging (DTI) have consistently revealed bilateral patterns of microstructural damage to multiple long-range association tracts, with greater compromise to fiber tracts ipsilateral and proximal to the seizure focus^3–5. This is typically demonstrated by decreases in fractional anisotropy (FA) and/or increases in mean diffusivity (MD) along entire fiber tracts or within subsections of a tract, which are interpreted as reflecting axonal loss and demyelination of white matter within the affected regions.

Despite the unique insights obtained from traditional DTI, it is increasingly appreciated that FA and MD are nonspecific measures of cerebral pathology that are influenced by a number of tissue-related factors⁶. In addition to axonal loss and demyelination, decreases in FA obtained from the basic tensor model may reflect the presence of crossing fibers or increases in extracellular diffusion due to edema or inflammation^{7; 8}. Given recent data suggesting that inflammation may play a role in the pathogenesis of TLE⁹, better understanding the neurobiology of decreased FA in temporal and extratemporal regions could help to guide treatments in patients with TLE or other epilepsy syndromes⁹.

In recent years, a number of advanced diffusion methods have emerged that extend beyond the tensor model and may provide more sensitive and/or specific measures of cerebral pathology in TLE^{6; 10–13}. Diffusion kurtosis imaging (DKI), which is a statistical method that probes non-Gaussian diffusion and estimates diffusion heterogeneity¹⁴, has been applied in several studies of children¹² and adults^{10; 11} with TLE. These studies have suggested that kurtosis measures reveal a broader and more robust pattern of microstructural abnormalities in TLE compared to conventional DTI, which may reflect DKI’s greater sensitivity to multiple pathologic factors including cell loss, inflammation, and axonal and dendritic reorganization. In addition, diffusion spectrum imaging (DSI), a high-angular diffusion imaging (HARDI) technique, has recently been combined with a multi-compartment diffusion model, the neurite orientation dispersion and density imaging (NODDI) model, to estimate structural connectivity and network properties in TLE¹³. This DSI study revealed that reduced FA in the medial temporal lobes was primarily driven by reductions in intracellular diffusion, commensurate with neurite loss (i.e., cell and axonal or myelin loss), whereas reduced FA in extratemporal regions was also driven by fiber orientation changes (i.e., high orientation dispersion index; ODI), potentially reflecting disorganized fiber orientation and packing. These studies suggest that advanced diffusion techniques, which aim to resolve intravoxel tissue properties, may provide more sensitive measures of network pathology in TLE.

In this study, we apply a new multi-compartment diffusion model, restriction spectrum imaging (RSI), which combines key properties of these existing methods to resolve whether decreases in FA in patients with TLE are better explained by decreased axonal/neurite density (ND), crossing fibers (CF), and/or increases in extracellular diffusion [i.e., isotropic hindered (IH) and free water (IF) diffusion]¹⁵. This new method is an extension of traditional HARDI techniques for reconstructing CFs and modeling complex orientation and structure, but within a clinically-feasible scan frame (4–8 minutes). Using multiple diffusion weightings (b-values), RSI further distinguishes hindered and restricted diffusion pools in the extracellular and intracellular compartments, respectively, and models both spherically-restricted (within cells) and cylindrically-restricted (within axons) diffusion (see Figure 1). Isolating the anisotropic, restricted compartment within axons is important as this allows for estimates of the axonal density and orientation distribution that are distinct from changes in extracellular diffusion. Thus, RSI’s ability to probe both the scale and geometry of tissue microstructure positions it well to determine whether regional reductions in DTI-derived FA values are due to axonal loss/demyelination versus extracellular changes (e.g., CSF-filled spaces; inflammation) or CFs. We have previously shown the advantage of RSI over standard DTI for reconstructing fiber tracts in regions of peritumoral edema⁸, identifying regions of high cellularity in patients with high grade glioma¹⁶, and evaluating response to therapy in patients treated with bevacizumab¹⁷. We now extend the application of RSI to unveiling the complex cytoarchitecture of white matter injury in patients with TLE.

MATERIALS AND METHODS

RESULTS

Table 1 displays demographic and clinical variables for the patient and control groups. There were no statistically significant differences among the controls, LTLEs, or RTLEs in age, (F [2,29] = .634, p > .05). However, there was a group difference in years of education (F [2,29] = 4.02, p<.05) with controls achieving a higher level of education than either patient group. The distribution of gender across the three groups was comparable (χ²[2] = 1.90, p >.05). There were no statistically significant differences between the two patient groups in disease duration (t[17] = 1.63, p >.05), age at seizure onset (t [17]= −.77, p > .05), seizure frequency (t[17] = .47, p >.05), or the number of patients with MTS (χ²[1] = 1.56, p > .05). However, the volume of the ipsilateral hippocampus was slightly smaller in patients with LTLE compared to those with RTLE (t[19] = −2.44, p = .025). Contralateral hippocampal volumes (HCVs) did not differ between the patient groups (t[19] = .51, p > .05). Both patient groups’ ipsilateral HCVs were smaller when compared to HCVs on the corresponding side for the controls (t[19] = 4.09, p = .001, and t[20] = 2.50, p = .04, LTLE and RTLE, respectively), but their contralateral HCVs were not statistically different from the controls (t[19] = .508, p > .05, and t[20] = .93, p > .05, LTLE and RTLE, respectively).

DISCUSSION

In this study, we demonstrate the value of a new advanced diffusion imaging technique, RSI, for better elucidating the nature of white matter injury in patients with TLE compared to traditional DTI. RSI’s advantage in this scenario stems from its ability to separate intracellular from extracellular diffusion while simultaneously modeling complex fiber structure (i.e., CFs). This ability to probe both the scale and geometry of tissue microstructure likely allows for a more precise estimate of axonal and/or myelin loss. In a group of patients with refractory TLE, we found reductions in ND that generally corresponded to decreases in FA. However, decreases in ND were greater in magnitude and showed a broader pattern of decreases compared to FA. This finding was demonstrated in both our voxelwise and tract-based analysis and is commensurate with a small, but growing literature demonstrating improved sensitivity of advanced diffusion imaging for delineating white matter pathology in epilepsy. Similar to studies indicating that DKI-derived measures (e.g., mean kurtosis) may provide a more sensitive measure of pathology^{10; 12; 28–30}, we found the pattern of reduced ND to be more robust and congruent with known network abnormalities in TLE, which are most prominent within medial temporal/limbic networks ipsilateral and proximal to the seizure focus⁵ (Figure 3). However, it is important to note that whereas reduced kurtosis indicates a loss of diffusion heterogeneity—a metric that should correlate with a loss of diffusion restriction, and therefore, white matter pathology—RSI is a multi-compartment model that can more directly probe the nature of this pathology. That is, RSI quantifies the degree to which diffusional heterogeneity within a voxel is driven by decreases in intra-axonal water (i.e., loss of ND) versus increases in extracellular diffusion (i.e., increased IF and IH) while accounting for complex fiber orientation. Thus, similar to DSI with NODDI, RSI provides a biophysical model for separating different tissue compartments and modeling the geometry of the neurite compartment. However, RSI takes this one step further by also characterizing CFs and volume fractions of fibers in multiple directions. Whereas the ODI obtained from NODDI provides a single estimate of the dispersion of fibers within a voxel, RSI can further characterize the nature of dispersion (i.e., whether high dispersion reflects crossing fibers or just faned out fibers) and within a clinically-feasible time frame (i.e., DSI typically takes ~30 minutes to acquire compared to ~4–7 minutes for RSI).

The increased precision of this measure may be reflected by the fact that decreased ND, but not FA, was associated with longer disease duration in our patient cohort in the UNC. This association is particularly interesting given the proposed role of the UNC in seizure propagation³¹ as well as the vulnerability of this white matter tract in TLE³². Whether or not loss of neurites in the UNC is due to years of refractory seizures versus other treatment- or disease-related factors requires testing in longitudinal series. We propose that RSI-derived ND could provide a better means for capturing these changes than current DTI-based metrics.

In addition to our main finding of significant reductions in ND, increases in IF were found proximal to the epileptic focus in both the ipsilateral FX and PHC, whereas increases in IH extended into the ipsilateral CING and contralateral PHC in the fiber ROI analysis. In general, IF and IH appeared to be highly related in TLE, but with additional extratemporal increases seen for IH. However, increases in extracellular water did not appear to be a dominant source of pathology in the other fiber tracts. Increases in the hindered and free water fraction may reflect a number of pathological processes, including increased CSF in spaces surrounding axon/myelin loss and/or local tissue inflammation. The former is a more simplistic explanation and one that is supported by analysis of excised fornix tissue scanned at 7T³³. However, surgical specimens from patients with TLE also suggest an increase in inflammatory infiltrates in excised temporal lobe tissue³⁴ and recent PET studies have shown increases in biomarkers of inflammation (i.e., translocator protein; TSPO) in the bilateral temporal lobes in TLE³⁵, suggesting ongoing local inflammation in patients with refractory seizures. Studies using advanced diffusion imaging in patients with multiple sclerosis³⁶ and schizophrenia⁷ have recently linked an increase in the isotropic restricted and isotropic free water compartments, respectively, to neuroinflammation. However, histological validation of these in vivo diffusion metrics is primarily based on preclinical data³⁷ whereas validation in humans is still limited (but see Wang et al.³⁶). Although we are not able to provide unique biological explanations for increased IH versus IF in this study, future work with RSI will address whether these extracellular components can differentiate local tissue inflammation/edema from CSF-filled pockets, respectively.

We also found group differences in CFs in both the ipsilateral FX and PHC. The greater effect of CFs in these regions likely reflects a more complex fiber structure in controls (see Figure 4). CFs are ubiquitous in the brain and it is well-known that they contribute to FA decreases in some regions more than others. In the medial temporal lobe, the PHC is crossed by fibers of the perforant path and adjacent pathways, which run from the entorhinal cortex into the hippocampus³⁸ and are known to be affected in TLE³⁹. Figure 4 shows a voxel taken from the PHC in a patient and control. As can be seen, FA is lower in the control (i.e., represented by a more spherical tensor) due to the presence of CFs in this region, whereas FA is not reduced in the patient due to simplified fiber structure. Although the reason for this simplified fiber structure in TLE is unknown, it is possible that the dying off of perpendicular axons within the perforant path and adjacent fibers bundles leads to this simplified structure in patients, driving up FA values in the PHC in patients moreso than in controls. Like HARDI models, RSI can model this complex fiber structure, producing measures of ND and IH/IF fractions that are not affected by CFs.

Our study demonstrates the utility of RSI for delineating the nature of white matter pathology in TLE. Other strengths of our study include an analysis of white matter microstructure at both the voxelwise and tract-based levels and our use of an advanced image registration method (i.e., ANTS) that has been shown to be robust for groupwise diffusion-based analysis²³. However, there are several limitations to our study that should be noted. First, the sample size is relatively small and we may not have had the power to detect subtle group differences in some fiber tracts that may be revealed with a larger cohort. However, it is important to note that we were able to capture many robust associations with ND with large effects sizes even in a small cohort. This could translate to needing fewer patients enrolled in clinical trials to demonstrate an effect when RSI- vs DTI-derived metrics are used. Second, we included patients with and without MTS in the same analysis. There is now considerable evidence that DTI-derived abnormalities are more pronounced in patients with radiological evidence of MTS^{2; 40}. In our small sample, it is of note that patients with MTS demonstrated a trend toward lower ND in the PHC relative to those without MTS (p = .054) and this likely contributed to the stronger patterns in FA and ND loss seen in the VBA maps in LTLE (the group with a greater number of patients with MTS) relative to RTLE. However, our group of MTS+ (N = 8) and MTS− (N = 13) were too small and unbalanced to analyze separately in this initial study. Future research with a larger patient sample will test the utility of RSI-derived measures in each patient cohort separately. Although the RSI acquisition scheme was selected to optimize scanning and statistical efficiency, the number of parameters estimated is on the cusp given the available data. Additional information may be gleaned from collecting more data. However, this would result in a longer scan time, likely decreasing its clinical feasibility without significantly improving the signal to noise^{18; 19}. Finally, although we believe that our multi-compartment diffusion model offers advantages over many existing models for better elucidating the nature of white matter changes within a clinically feasible time frame, as with most diffusion techniques, histological validation of the RSI-model is primarily limited to preclinical data¹⁵ and will require further validation in patients with TLE. As numerous promising advanced diffusion techniques are now emerging, there is also a significant need to directly compare these techniques to one another and to determine which method(s) provides the most clinicially-meaningful information for patients with TLE and other epilepsy syndromes.

Supplementary Material