Thalamohippocampal atrophy in focal epilepsy of unknown cause at the time of diagnosis

Participants

We identified patients with archived MRI, acquired according to a clinical epilepsy protocol, within 12 months of diagnosis of focal epilepsy of unknown cause and scanned on a 3 T GE Discovery MRI system at the Walton Centre NHS Foundation Trust, Liverpool, UK since 2015. At initial screening, 140 patients with likely NDfE and corresponding MRI for analysis were retrieved. More detailed assessment of patient clinical histories and MRI resulted in the exclusion of 58 patients due to one of the following factors: (1) first seizure with no diagnosis of epilepsy, (2) probable idiopathic generalised epilepsy, (3) symptomatic seizures (e.g. tumour, infection), (4) presence of epileptogenic lesion (e.g. focal cortical dysplasia, hippocampal sclerosis), or (5) unusable or unavailable MRI data for analysis. This resulted in 82 patients with NDfE of unknown cause with corresponding MRI data for image analysis (Table 1). All patients were diagnosed by consultant neurologists at the Walton Centre NHS Foundation Trust. All images were reported non-lesional by a neuroradiologist with expertise in the assessment of MRI for epileptogenic lesions. All patients had no history of learning disability.

Additional clinical data were obtained by searching through hospital electronic records. We obtained age at diagnosis and seizure type (focal aware seizures [FAS], focal impaired awareness seizures [FIAS] and focal to bilateral tonic-clonic seizures [FBTCS]) for patients. Fifty-one (62.2%) had undergone EEG and we recorded whether inter-ictal abnormalities were captured. Seizure status at 6 and 12 months after diagnosis, hereon referred to as seizure outcome, was obtained for 58 patients at 6 months and 48 patients at 12 months.

For comparison with patients, we used imaging data from a cohort of 40 healthy adult controls that were scanned as part of a different study ^[20] but who had the equivalent MRI scans for comparative analysis (Table 1). There was no significant difference between the age of patients (when diagnosed) and controls (at time of MRI) (t = 2.8, P = 0.13). However, there was a significant sex difference between patients and controls (X² = 4.72, P = 0.03). The North West – Liverpool research ethics committee approved this study (14/NW/0332).

MRI acquisition

The standard 3 Tesla MRI protocol for patients with a new presentation of seizures at our centre included a high in-plane resolution T1-weighted fluid attenuated inversion recovery (FLAIR) MRI acquisition of the whole brain (TE 1.5 ms, TR 2500 ms, flip angle 111°, voxel size 0.4 mm × 0.4 mm, slice thickness 3.0 mm, FOV 220 mm, matrix size 320 × 384). This sequence was used for analysis in the present study. Other sequences acquired for diagnostic purposes but not used for analysis included axial T2-weighted and coronal T2-FLAIR scans.

MRI analysis

Given the slice thickness of the T1-weighted FLAIR images, we were unable to reliably apply automated image analysis tools to extract subcortical and hemispheric volume. We therefore used rigorous manual techniques to estimate the volume of subcortical and hemispheric structures. The volume of the left and right hippocampus, thalamus, caudate nucleus, putamen, and cerebral hemispheres were quantified for all participants using the Cavalieri method of design-based stereology ^[21]. This approach has been frequently applied to MRI data in epilepsy studies ^{[15, 22-24]}, has been considered the benchmark measurement approach to which automated MRI techniques have been compared ^[23], and provides a mathematically unbiased and validated approach to estimate brain compartment volume ^{[21, 25]}.

Using Easymeasure software ^[23], each volume of interest (VOI) was estimated using a series of parallel two-dimensional (2D) MR sections set at a constant distance apart. A randomly orientated grid of pixels was overlaid on each section and points intersecting each region-of-interest (ROI) were counted separately for the left and right structures of each patient and control. The pixel size used for point-counting was altered depending on the size of the ROI (pixel sizes: hippocampus, 4; thalamus, 6; caudate nucleus, 5; putamen, 6; and cerebral hemisphere, 30) in order to optimise the sampling density ^[21]. The number of points transecting the ROI was multiplied by distance between each consecutive section to produce volume estimates. Given that nuclei (e.g. thalamic nuclei) and subregions (e.g. hippocampal cornu ammonis) of structures measured are almost indistinguishable on clinical MRI, we measured the structures as an entire complex.

Stereological point counting on MR images for volume estimation of the hippocampus, thalamus, caudate nucleus and putamen is shown in Figure 1. Detailed information on the hippocampal VOI is provided elsewhere ^[24]. Moving along its longitudinal axis, the hippocampus is bound superiorly by the white, myelinated fibres of the alveus and often by an additional region of cerebrospinal fluid superior to the alveus. The hippocampus was differentiated anteriorly from the amygdala through visualisation of the alveus. The posterior boundary of the hippocampus was reached when the lateral ventricles divide into the frontal and temporal horns. The hippocampal VOI comprised the hippocampus proper, dentate gyrus, alveus, subiculum, presubiculum and parasubiculum; the amygdala, uncus, choroid plexus and grey matter above the alveus were not included in the measurements.

• Download figure
• Open in new tab

Figure 1.

T1-weighted fluid attenuated inversion recovery coronal sections through an exemplar patient showing point counting for stereology through the subcortical volumes of interest. For each structure, point counts are removed in the left hemisphere and coloured orange in the right hemisphere. A, hippocampus; B, thalamus; C, caudate nucleus; D, putamen. Zoomed sections at the bottom of each panel show point counting in a rostral (top left) to caudal (bottom right) direction.

The anterior border of the thalamus began immediately posterior to the anterior commissure and maintained a close relationship with the internal capsule laterally and the central canal of the ventricles medially; the posterior border of the thalamus was the pulvinar. Measurements ended with the formation of the atrium of the ventricles. The zona incerta formed the inferior border of the thalamus. We excluded the subthalamic nuclei, substantia nigra, and red nuclei from thalamic measurements. The lateral and medial geniculate bodies and the habenular nucleus were also excluded ^[23]. The posterior border of the caudate nucleus was considered the last slide in which the caudate tail was still superior to the lateral ventricle. Caudate nucleus and putamen measurements ended with the formation of the atrium from the temporal and frontal horns of the lateral ventricle. Neither caudate nucleus nor putamen measurements included the striatal cell bridges connecting the two nuclei or the nucleus accumbens. The medial and lateral borders of the putamen were the internal capsule and external capsule, respectively ^[22]. The posterior border of the putamen often coincides with the appearance of the medial and lateral geniculate bodies. Measurement of the entire cerebral hemispheres was also obtained which included all supratentorial grey and white matter, excluding the brainstem and cerebellum. All subcortical and hemispheric volumes were normalised using whole brain volumes (summation of left and right cerebral hemispheric volumes); the proportion of each subcortical and hemispheric volume relative to whole brain volume was calculated.

Statistical analysis

The data did not meet the assumptions of parametric tests; therefore, nonparametric tests were used to analyse the data. For patient-control analysis, all volumes were analysed using nonparametric Mann-Whitney U-tests in SPSS (version 25, www.spss.com). Given the significant sex difference between patients and controls, and the higher average age of patients compared to controls, we performed all statistical analyses on normalised volumes, with their residuals corrected for age and sex in a confound only regression model. Volumes lower than two standard deviations of the control mean were considered abnormal and suggestive of structural atrophy in individual patients. Spearman’s correlations were used to investigate relationships between volumes and age at first seizure, years between first seizure and diagnosis, and age at diagnosis. Categorical analysis of clinical and neuroimaging data was performed using Chi-squared tests. Multiple comparisons were corrected using the false discovery rate (FDR) and results were considered statistically significant at P < 0.05.

Clinical data

Two patients did not commence AED treatment. These patients did not have outcome data. For the remainder of the patient cohort, the first AED used was Lamotrigine (n = 47, 59%), Levetiracetam (n = 14, 18%), Zonisamide (n = 8, 10%), Carbamazepine (n = 6, 8%), Sodium Valproate (n = 3, 4%), Oxcarbazepine (n = 1, 1%) and Phenytoin (n = 1, 1%). There were no significant associations in the clinical data.

Volumetric changes in NDfE

Volumetric descriptive statistics are provided in Table 2. Volume of the left hippocampus (U = 1232, P_(FDR-corr) = 0.04), and left (U = 1002, P_(FDR-corr) = 0.002) and right (U = 1228, P_(FDR-corr) = 0.04) thalamus were significantly smaller in patients compared to controls (Figure 2). There was a trend for the right hippocampus to be smaller in patients relative to controls (U = 1311, P_(FDR-corr) = 0.09). There were no significant differences (P_(FDR-corr) < 0.05) or trends for differences in volume of the left or right caudate nucleus, putamen or whole cerebral hemisphere between patients and controls. In patients, average volume of the left hippocampus was decreased by 6.9%, right hippocampus by 6.5%, left thalamus by 6.4%, and right thalamus by 4.1%, relative to controls. Individual volumetric analysis revealed abnormal volume of the left hippocampus in 9 (11%) patients, the right hippocampus in 4 (4.9%) patients, the left thalamus in 14 (17.1%) patients, and the right thalamus in 7 (8.5%) patients.

• Download figure
• Open in new tab

Figure 2.

Scatterplots with minimum, 25th percentile, median, 75th percentile, and maximum normalised and corrected volume measurements of brain structures in seizure free patients, patients with persistent seizures, and healthy controls. Statistically significant between groups: *(P_(FDRcorr) < 0.05), **(P_(FDRcorr) < 0.01).

Correlations with clinical variables

There were no associations between volumes and seizure outcome at 6 or 12 months, EEG finding, loss of awareness during seizures, history of FBTCS, age at first seizure, years between first seizure and diagnosis or age at diagnosis. Table 3 presents the descriptive and statistical comparisons between outcome groups at 12 months. There were also no clinically significant differences between the individual patients who had significant loss of hippocampal and thalamic volume and those who did not.

Biological and clinical implications

Of the limited number of quantitative MRI studies that exist in NDfE, the focus has been on the hippocampus. Our results are in support of those studies that reported hippocampal atrophy at diagnosis ^{[14, 15, 26]} and are in contrast to those that did not report atrophy ^[16]. Patients with chronic epilepsy have shown a 13 to 16% reduction in hippocampal volume compared to controls ^[26]; we identified a 6.5 to 6.9% decrease in hippocampal volume which may indicate hippocampal atrophy is present at time of diagnosis and may worsen with the progression of epilepsy. The present study also sought to determine whether extrahippocampal subcortical atrophy was present at diagnosis. To our knowledge, the present study is the first to identify thalamic atrophy in adults with a new diagnosis of focal epilepsy. A previous study in a small sample of patients with NDfE did not report thalamic atrophy ^[16]. Thalamic atrophy has been reported to be almost as common as hippocampal atrophy in a meta-analysis of voxel-based morphometry studies of refractory TLE ^[27], and is observed in a range of longstanding focal and generalised epilepsy disorders ^{[2, 28]}. Thalamic volume loss has also been reported in children with NDfE ^[29] and patients with new-onset GGE ^[19]. Consistent with our findings, volume loss of the thalamus in both hemispheres is frequently reported in patients with chronic focal epilepsy ^{[2, 28, 30-32]}. Taken together, our results suggest that thalamohippocampal atrophy is likely established prior to the onset of habitual epilepsy; further hippocampal and thalamic damage may occur as the disorder becomes longstanding, particularly in refractory cases ^[33-35].

There are very few existing studies that have attempted to predict pharmacoresistance from the point of diagnosis of focal epilepsy using advanced imaging in a way that resembles work predicting surgical outcome in focal epilepsy ^{[31, 36]}. This is an unmet need in the early stages of the disorder ^{[7, 8, 37]}. Having a reliable imaging biomarker of the health issues patients will experience (e.g. uncontrolled seizures, memory impairment) from diagnosis will provide clinicians and patients with realistic expectations and could serve to assist the patient management pathway (e.g. earlier use of adjunctive / alternative therapies in patients likely to be pharmacoresistant) ^[37]. In the present study we have reported that gross neuroanatomical volume of subcortical structures are not related to seizure control at 6 or 12 months after diagnosis. (We note the relatively brief follow up; however, 12-month outcome is highly predictive of pharmacoresistance in NDE ^[12].) The advantage of investigating potential imaging biomarkers using MRI acquired as part of standard diagnostic evaluation is the applicability of findings to routine clinical practice. However, the disadvantage is that highly variable gross brain morphology is unlikely to be sensitive enough to identify markers of pharmacoresistance. It is likely that these markers, if found, will be microstructural, functional, or metabolic. Interestingly, in patients with longstanding focal epilepsy, MR spectroscopy hippocampal N-Acetylaspartate/Creatine measurements have been related to seizure control ^[38].

Approximately 50% of patients with NDfE exhibit impairment in at least one cognitive domain ^[13]. Although we did not assess cognition in the current study, atrophy of subcortical structures may be related to cognitive impairment in patients with NDfE. Previous research has shown hippocampal volume loss is associated with poorer verbal memory performance in NDfE ^[14] and higher seizure frequency after AED withdrawal in TLE ^[39]. Lower thalamic volume and higher seizure frequency have been observed in cognitively impaired children with NDE ^[29]. Future research of imaging and cognitive biomarkers of pharmacoresistance in NDfE would be beneficial in establishing the most effective treatment pathways for patients likely to experience refractory epilepsy ^[37].

Methodological issues

The discrepancy between our findings and those of a previous study that did not report thalamic atrophy in patients with NDfE ^[16] is difficult to reconcile. We have previously demonstrated good agreement between stereology and Freesurfer analysis – the method used by the previous study – for volume estimation of the thalamus ^[23]; therefore image analysis approach is unlikely to be a factor. The substantially larger sample size of the present study is also unlikely to explain this discrepancy as the previous study reported a small trend for increased thalamic volume in patients with NDfE ^[16]. Both studies investigated only patients with non-lesional NDfE. The only remaining difference between studies is type of MRI scans. The previous study used conventional three-dimensional (3D) T1-weighted scans with isotropic voxels (1 mm × 1 mm × 1mm), which offers limited differentiation between thalamic grey matter and adjacent white matter ^[23]. We used non-isotropic T1-weighted FLAIR with high resolution in-plane coronal sections (0.4 mm × 0.4 mm), which provides superior delineation between the thalamus and white matter (Figure 1).

The standard MRI protocol for patients with a new presentation of seizures at our institution does not include 3D volume scans that would be amenable to automated segmentation techniques. It was necessary for us to apply manual volumetric analysis due to the 3 mm slice thickness of the coronal 2D T1-weighted FLAIR MRI scans. Despite this being a time-inefficient way of obtaining morphometric data, there were distinct advantages to our approach. Firstly, manual measurement of brain regions is considered gold standard and stereology provides a mathematically unbiased and validated approach to estimate brain compartment volume ^{[21, 25]}. Secondly, despite the non-isotropic voxel size, the high in-plane resolution and contrast of the scans provided excellent grey-white matter differentiation, even in regions where the grey matter and white matter borders are difficult to establish (e.g. the thalamus).

Following diagnostic MRI most patients with NDfE will not undergo further investigation, which is usually undertaken in those with a refractory course, for who more precise localisation of the seizure focus is more likely as more seizures are witnessed and following increasingly detailed imaging, EEG and neuropsychological evaluation. Our sample of patients with NDfE is therefore likely to be clinically heterogenous in terms of seizure foci, despite commonalities in a new diagnosis of focal epilepsy and non-lesional MRI. We suggest that what is lost through the inclusion of a highly phenotyped group of patients is gained through a pragmatic approach to studying all non-lesional patients with a new diagnosis of focal epilepsy. Indeed, studying patients with NDfE pragmatically may yield common brain abnormalities, and potentially biomarkers of treatment outcome, given that; (i) widespread alterations in brain network structure can give rise to a clearly localised focal onset in one brain region ^[40]; (ii) particular anatomical circuits act as critical modulators of seizure generation and propagation, and seizure activity does not spread diffusely throughout the brain but propagates along specific anatomical pathways, regardless of the localisation of the brain insult ^{[41, 42]}; (iii) pathological structural connectivity causes disturbances to common large scale functional brain networks regardless of the localisation of the epileptogenic zone in patients with refractory focal epilepsy ^[43]; and (iv) subcortical structures - such as the thalamus and striatum - that play a crucial role in the clinical manifestation of seizures in the epilepsies ^[44], and anatomically support widespread distributed cortico-subcortical networks ^[45], are structurally and physiologically abnormal in both hemispheres in patients with longstanding focal and generalised epilepsy disorders ^{[2, 28, 44]}.