Structural brain abnormalities in the common epilepsies assessed in a worldwide ENIGMA study

1Epilepsy is a prevalent neurological disorder, comprising many different syndromes and conditions, affecting 0.6–1.5% of the population worldwide. Approximately one-third of affected individuals do not respond to antiepileptic drug therapy. Alternative treatment options may not be appropriate, and are not always effective. The identification of shared biological disease pathways may help elucidate diagnostic and prognostic biomarkers and therapeutic targets, which, in turn, could help to optimize individual treatment. However, disease biology remains unexplained for most cases—especially in commonly occurring epilepsies.

2Epilepsy is a network disorder typically involving widespread structural alterations beyond the putative epileptic focus. Hippocampal sclerosis is a common pathological substrate of mesial temporal lobe epilepsy (MTLE), but extrahippocampal abnormalities are also frequently observed in MTLE, notably in the thalamus and neocortex. Neocortical abnormalities are also reported in idiopathic generalized epilepsies (IGE), and many childhood syndromes. Thus, common epilepsies may be characterized by shared disturbances in distributed cortico-subcortical brain networks, but the pattern, consistency and cause of these disturbances, and how they relate to functional decline, are largely unknown.

3Currently, we lack reliable data from large cross-sectional neuroimaging, brain tissue, or biomarker studies in the common epilepsies. Brain tissue is not available from large cohorts of patients: common forms of epilepsy are often unsuitable for surgical treatment, so biopsied tissues are simply unavailable in sufficient numbers for research into disease biology. Brain-wide post-mortem studies also require extensive effort for comprehensive analysis. MRI offers detailed information on brain structure, but MRI measures from groups of individuals with and without epilepsy are not always consistent. For example, MTLE is associated with hippocampal sclerosis in up to 70% of brain MRI scans. However, the effects of laterality, and the extent of extrahippocampal grey matter loss are inconsistently reported in studies of left versus right MTLE. Similarly, abnormalities of the basal ganglia, hippocampus, lateral ventricles, and neocortex have all been reported in IGE, but most alterations are non-specific, and visual inspection of clinical MRI in IGE is typically normal. Genome-wide association studies (GWAS) have identified genetic variants associated with complex epilepsies by ‘lumping’ different epilepsy types together, but MRI studies are typically of smaller scale, and have not widely explored whether distinct epilepsy syndromes share common structural abnormalities.

4There are many sources of inconsistency in previously reported MRI findings. First, epileptic seizures and syndromes are diverse; classifications are often revised and contested. Second, most cross-sectional brain imaging studies are based on small samples (typically <50 cases), limiting the power to detect subtle group differences. Third, variability in scanning protocols, image processing, and statistical analysis may affect the sensitivity of brain measures across studies.

5The Enhancing Neuro Imaging Genetics through Meta-Analysis (ENIGMA) Consortium was formed to address these issues. ENIGMA is a global initiative, combining large samples with coordinated image processing, and integrating genomic and MRI data across hundreds of research centres worldwide. Prior ENIGMA studies have identified genetic variants associated with variations in brain structure, and have reliably characterized patterns of brain abnormalities in schizophrenia, major depression (, obsessive compulsive disorder (, attention deficit hyperactivity disorder (, and many other brain illnesses. Large-scale, collaborative initiatives such as ENIGMA may improve our understanding of epilepsy, helping clinicians make more informed decisions and provide personalized treatment strategies. Thus, we formed the Epilepsy Working Group of ENIGMA (‘ENIGMA-Epilepsy’) to apply coordinated, well-powered studies of imaging and genetic data in epilepsy.

6Here, in the largest analysis of structural brain abnormalities in epilepsy to date, we ranked effect sizes for 16 subcortical and 68 cortical brain regions in 2149 individuals with epilepsy and 1727 healthy controls, using harmonized image processing, quality control, and meta-analysis. First, we grouped all epilepsies together, to determine whether biologically distinct syndromes show robust, common structural deficits. Second, we assessed a well-characterized form of epilepsy: MTLE with hippocampal sclerosis, analysing patients with left- and right-sided hippocampal sclerosis as independent groups. Third, we examined another major set of epilepsy syndromes: IGE. Finally, we studied all remaining epilepsies as a combined subgroup, to understand the relative contributions of IGE, MTLE-L, MTLE-R, and all other syndromes on shared patterns of structural compromise. We tested how age at scan, age of onset, and epilepsy duration affected brain structural measures. Based on existing neuroimaging, neurophysiological, neuropathological, and genetic data, we predicted that (i) biologically distinct epilepsy syndromes would exhibit shared patterns of structural abnormalities; (ii) MTLEs with left or right hippocampal sclerosis would show distinct patterns of hippocampal and extrahippocampal structural deficits; and (iii) IGEs would also display subcortical volume and cortical thickness differences, compared to healthy controls.

7Each centre received approval from their local institutional review board or ethics committee. Written informed consent was provided according to local requirements (Supplementary Table 1).

8Twenty-four cross-sectional samples from 14 countries were included in the study, totalling 2149 people with epilepsy and 1727 research centre-matched healthy control subjects (Fig. 1 and Table 1). The locations, dates, and periods of participant recruitment are provided in Supplementary Table 1. An epilepsy specialist assessed seizure and syndrome classifications at each centre, using International League Against Epilepsy terminology. Participants were aged 18–55.

9To test for shared and syndrome-specific structural alterations, analyses included one group combining all epilepsies (‘all-epilepsies’; n = 2149), and four stratified subgroups: (i) left MTLE with left hippocampal sclerosis (MTLE-L; n = 415); (ii) right MTLE with right hippocampal sclerosis (MTLE-R; n = 339); (iii) IGE (n = 367); and (iv) all other epilepsies (n = 1028). Supplementary Table 2 lists all syndromic diagnoses included in the aggregate ‘all-epilepsies’ group. For the MTLE subgroups, we included anyone with the typical electroclinical constellation, and a neuroradiologically-confirmed diagnosis of unilateral hippocampal sclerosis on clinical MRI. Participants were included in the IGE subgroup if they presented with tonic-clonic, absence or myoclonic seizures with generalized spike-wave discharges on EEG. Participants were included in the ‘all-other-epilepsies’ subgroup if they were diagnosed with non-lesional MTLE (43.3%), occipital (1.67%), frontal (8.78%), or parietal lobe epilepsy (0.84%), focal epilepsies not otherwise specified (37.03%), or another unclassified syndrome (8.37%; Supplementary Table 2). We excluded participants with a progressive disease (e.g. Rasmussen’s encephalitis), malformations of cortical development, tumours or previous neurosurgery.

10Structural T1-weighted MRI brain scans were collected at the 24 participating centres. Scanning details are provided in Supplementary Table 3. T1-weighted images from cases and controls were analysed at each site using FreeSurfer 5.3.0, for automated analysis of brain structure. Volumetric measures were extracted for 12 subcortical grey matter regions (six left and six right, including the amygdala, caudate, nucleus accumbens, pallidum, putamen, and thalamus), the left and right hippocampi, and the left and right lateral ventricles. Cortical thickness measures were extracted for 34 left-hemispheric grey matter regions, and 34 right-hemispheric grey matter regions (68 total; Supplementary Table 4). Visual inspections of subcortical and cortical segmentations were conducted following standardized ENIGMA protocols (http://enigma.usc.edu), used in prior genetic studies of brain structure, and large-scale case-control studies of neuropsychiatric illnesses. Analysts were blind to participants’ diagnoses. Each analyst was instructed to execute a series of standardized bash scripts, identifying participants with volumetric or thickness measures greater or less than 1.5 times the interquartile range as outliers. Outlier data were then visually inspected, by overlaying the participant’s cortical segmentations on their whole-brain anatomical images. If the blinded local analyst judged any structure as inaccurately segmented, that structure was omitted from the analysis. The Supplementary material provides further information.

11All research centres tested for differences in age between individuals with epilepsy and controls using an unpaired, two-tailed t-test in the R statistics package (https://www.r-project.org). Each centre also tested for sex differences between individuals with epilepsy and controls using a chi-squared test in SPSS Statistics package (IBM Corp., Version 21.0).

12Each research centre tested for case-versus-control differences using multiple linear regressions (via the lm function implemented in R), where a binary indicator of diagnosis (0 = healthy control, 1 = person with epilepsy) was the predictor of interest, and the volume or thickness of a specified brain region was the outcome measure. We calculated effect size estimates across all brain regions using Cohen’s d, adjusting for age, sex and intracranial volume (ICV). ICV is a reliable, indirect measure of head size, used as a covariate in other large-scale ENIGMA collaborations. Cohen’s d effect sizes and regression beta coefficients were pooled across centres using a random-effects, restricted maximum likelihood method of meta-analysis via the R package, metafor. The Supplementary material provides additional details.

13Each centre conducted a series of linear regressions, testing the association between subcortical volume or cortical thickness, and: (i) age at onset of epilepsy; and (ii) duration of epilepsy. All centres tested for interactions between diagnosis of epilepsy (including syndrome groups) and age at time of scan. Beta values representing the unstandardized slopes of each regression were extracted for each analysis. Sex and ICV were included as covariates in all secondary analyses.

14We conducted four independent regressions (one case versus control regression, and three regressions with clinical variables) across 84 regions of interest, adjusting the statistical significance threshold to Pthresh < 1.49 × 10−4 to correct for 336 comparisons. To account for correlations between tests, we also applied a less conservative adjustment for false discovery rate (FDR), using the Benjamini and Hochberg method. For clarity, we report only P-values significant after stringent Bonferroni correction; FDR-adjusted P-values are summarized in the Supplementary material.

15Across all regions of interest, we calculated the sample sizes necessary to achieve 80% power to detect case-control differences, given the observed effect sizes at each region of interest, based on two-tailed t-tests, using G*Power Version 3.1. For each region of interest, we also estimated N80: the total number of samples required, per group, to achieve 80% power to detect group differences using a t-test at the threshold of P < 0.05 (two-tailed).

16The sample size-weighted mean age across all epilepsy samples was 34.4 (range: 26.2–40) years, and the weighted mean age of healthy controls was 33.3 (range: 25.2–42.3) years. The weighted mean age at onset of epilepsy and duration of epilepsy were 17.6 (range: 12.1–28.2) years and 17.4 (range: 8.3–28) years, respectively. Females comprised 57% of the total epilepsy sample (range: 34–75% by individual sample), and 53% of the controls (range: 31–71% by individual sample). Case-control differences in age were observed at 8 of 24 research centres, and case-control differences in sex were observed at 2 of 24 research centres (Supplementary Table 5); hence, age and sex were included as covariates in all group comparisons.

17Compared to controls, the aggregate all-epilepsies group exhibited lower volumes in the left (d = −0.36; P = 1.31 × 10−6) and right thalamus (d = −0.37; P = 7.67 × 10−14), left (d = −0.35; P = 3.04 × 10−7) and right hippocampus (d = −0.34; P = 6.63 × 10−10), and the right pallidum (d = −0.32; P = 8.32 × 10−9). Conversely, the left (d = 0.29; P = 2.14 × 10−12) and right (d = 0.27; P = 3.73 × 10−15) lateral ventricles were enlarged across all epilepsies when compared to controls (Table 2 and Fig. 2A). A supplementary analysis of all-epilepsies, excluding individuals with hippocampal sclerosis or other lesions, revealed similar patterns of volume loss in the right thalamus and pallidum, and bilaterally enlarged ventricles; however, volume differences were not observed in the hippocampus (Supplementary Table 6).

18The MTLE-L subgroup showed lower volumes in the left hippocampus (d = −1.73; P = 1.35 × 10−19), left (d = P = 2.19 × 10−11) and right thalamus (d = −0.46; P = 8.12 × 10−5), left putamen (d = −0.39; P = 1.07 × 10−6), and right pallidum (d = −0.45; P = 5.48 × 10−7). As in the overall group comparison, we observed larger left (d = 0.47; P = 1.96 × 10−7) and right lateral ventricles (d = 0.36; P = 8.95 × 10−5) in MTLE-L patients relative to controls (Table 2 and Fig. 2B).

19The MTLE-R subgroup showed lower volumes across a number of regions in the right hemisphere only, including the hippocampus (d = −1.91; P = 6.36 × 10−37), thalamus (d = −0.73; P = 1.6 × 10−12), and pallidum (d = −0.45; P = 3.96 × 10−7), together with increased volumes of the left (d = 0.39; P = 1.52 × 10−6) and right lateral ventricles (d = 0.44; P = 6.57 × 10−12) compared to controls (Table 2 and Fig. 2C).

20The IGE subgroup showed lower volumes in the right thalamus (d = −0.4; P = 3.6 × 10−6) compared to controls (Table 2 and Fig. 2D).

21The all-other-epilepsies subgroup showed lower volumes in the right thalamus (d = −0.31; P = 7.9 × 10−11) and the right pallidum (d = −0.24; P = 8.1 × 10−5) compared to controls. The all-other-epilepsies subgroup also showed significant enlargements of the left (d = 0.33; P = 5.1 × 10−7) and right amygdala (d = 0.22; P = 1.46 × 10−4), and the left (d = 0.2; P = 1.2 × 10−5) and right lateral ventricles (d = 0.21; P = 4.62 × 10−6) compared to controls (Table 2 and Fig. 2E).

22All volume differences can be visualized using the interactive ENIGMA-Viewer tool, at http://enigma-viewer.org/ENIGMA_epilepsy_subcortical.html (Supplementary material). Volume differences significant after FDR adjustment can also be visualized at http://enigma-viewer.org/ENIGMA_epilepsy_subcortical_fdr.html (Supplementary Tables 26–30).

23The all-epilepsies group showed reduced thickness of cortical grey matter across seven regions bilaterally, including the left (d = −0.38; P = 1.82 × 10−18) and right precentral gyri (d = −0.4; P = 8.85 × 10−20), left (d = −0.32; P = 2.11 × 10−15) and right caudal middle frontal gyri (d = −0.31; P = 2.09 × 10−9), left (d = −0.31; P = 2.05 × 10−6) and right paracentral gyri (d = −0.32; P = 2.19 × 10−9), left (d = −0.19; P = 1.29 × 10−4) and right pars triangularis (d = −0.2; P = 4.25 × 10−8), left (d = −0.28; P = 1.51 × 10−7) and right superior frontal gyri (d = −0.27; P = 4.49 × 10−6), left (d = −0.19; P = 1.05 × 10−5) and right transverse temporal gyri (d = −0.18; P = 2.81 × 10−5), and left (d = −0.23; P = 9.87 × 10−5) and right supramarginal gyri (d = −0.22; P = 5.24 × 10−5). The all-epilepsies group also showed unilaterally thinner right cuneus (d = −0.2; P = 9.68 × 10−8), right pars opercularis (d = −0.18; P = 6.48 × 10−7), right precuneus (d = −0.28; P = 2.7 × 10−5), and left entorhinal gyrus (d = −0.26; P = 2.04 × 10−5), compared to healthy controls (Table 3 and Fig. 3A). Supplementary analysis in a non-lesional epilepsy subgroup revealed a similar pattern of cortical thickness differences compared to controls, suggesting that the changes observed in our main analysis were not driven by the inclusion of patients with hippocampal sclerosis or other common lesions (Supplementary Table 7).

24The MTLE-L and MTLE-R subgroups showed distinct patterns of cortical thickness reductions when compared to healthy controls (Table 3, Fig. 3B and C). In MTLE-R, lower cortical thickness was reported across four motor regions, including the left (d = −0.51; P = 7.67 × 10−7) and right paracentral gyri (d = −0.42; P = 6.24 × 10−11), and the left (d = −0.42; P = 4.31 × 10−6) and right precentral gyri (d = −0.52; P = 1.25 × 10−9). The MTLE-R subgroup also showed thickness changes in the left transverse temporal gyrus (d = −0.31; P = 2.15 × 10−5), and right pars opercularis (d = −0.27; P = 1.45 × 10−4) (Table 3 and Fig. 3C). By contrast, in MTLE-L, lower thickness was observed across six regions of the motor cortex, including the left (d = −0.43; P = 1.61 × 10−5) and right paracentral gyri (d = −0.38; P = 5.14 × 10−7), left (d = −0.47; P = 8.64 × 10−9) and right precentral gyri (d = −0.49; P = 2.37 × 10−10), and left (d = −0.54; P = 7.35 × 10−5) and right precuneus (d = −0.47; P = 5.16 × 10−6). The MTLE-L group also showed thickness changes across five regions of the frontal cortex, including the left (d = −0.41; P = 1.02 × 10−11) and right superior frontal gyri (d = −0.37; P = 1.44 × 10−9), left (d = −0.4; P = 7.07 × 10−9) and right caudal middle frontal gyri (d = −0.44; P = 3.61 × 10−7), and the right pars triangularis (d = −0.29; P = 2.16 × 10−6). In MTLE-L, thickness alterations were also observed in four regions of the temporal cortex, including the left temporopolar cortex (d = −0.32; P = 3.33 × 10−6), left parahippocampal gyrus (d = −0.3; P = 3.95 × 10−5), left entorhinal gyrus (d = −0.45; P = 7.35 × 10−10), and left fusiform gyrus (d = −0.36; P = 2.19 × 10−7) (Table 3 and Fig. 3B).

25The IGE subgroup showed reduced thickness in the left (d = −0.34; P = 1.75 × 10−6) and right precentral gyri (d = −0.39; P = 5.27 × 10−8), when compared to healthy controls (Table 3 and Fig. 3D).

26The all-other-epilepsies subgroup showed lower thickness across six cortical regions bilaterally, including the left (d = −0.38; P = 1.76 × 10−16) and right precentral gyri (d = −0.35; P = 1.7 × 10−14), left (d = −0.26; P = 1.34 × 10−8) and right paracentral gyri (d = −0.35; P = 1.1 × 10−14), left (d = −0.29; P = 1.32 × 10−10) and right caudal middle frontal gyri (d = −0.21; P = 2.62 × 10−6), left (d = −0.22; P = 7.27 × 10−7) and right superior parietal gyri (d = −0.22; P = 1.15 × 10−6), left (d = −0.24; P = 3.51 × 10−5) and right superior frontal gyri (d = −0.23; P = 7.15 × 10−6), and the left (d = −0.18; P = 1.34 × 10−4) and right precuneus (d = −0.24; P = 7.78 × 10−6) compared to controls. The all-other-epilepsies group also showed unilaterally reduced thickness in six right hemispheric regions, including the cuneus (d = −0.23; P = 2.15 × 10−7), lateral occipital gyrus (d = −0.21; P = 3.18 × 10−6), pars triangularis (d = −0.21; P = 3.32 × 10−6), supramarginal gyrus (d = −0.21; P = 9.95 × 10−6), transverse temporal gyrus (d = −0.18; P = 6.84 × 10−5), and lingual gyrus (d = −0.18; P = 7.12 × 10−5), compared to controls (Table 3 and Fig. 3E).

27An interactive 3D visualization of these results is available via the ENIGMA-Viewer tool, at http://enigma-viewer.org/ENIGMA_epilepsy_cortical.html (Supplementary material). Cortical thickness differences significant after FDR adjustment can also be visualized at http://enigma-viewer.org/ENIGMA_epilepsy_cortical_fdr.html (Supplementary Tables 31–35).

28A secondary analysis identified significant associations between duration of epilepsy and several affected brain regions in the all-epilepsies, MTLE-R, and all-other-epilepsies groups. In the all-epilepsies group, duration of epilepsy negatively associated with volume measures in the left hippocampus (b = −8.32; P = 8.16 × 10−13), left (b = −13.58; P = 3.52 × 10−15), and right thalamus (b = −12.25; P = 1.58 × 10−13), and right pallidum (b = −2.67; P = 1.78 × 10−7), in addition to bilateral thickness measures in the left (b = −0.003; P = 2.99 × 10−11) and right pars triangularis (b = −0.002; P = 4.24 × 10−9), left (b = −0.003; P = 1.61 × 10−15) and right caudal middle frontal gyri (b = −0.003; P = 1.65 × 10−17), left (b = −0.003; P = 1.77 × 10−13) and right supramarginal gyri (b = −0.003; P = 2.58 × 10−19), left (b = −0.003; P = 5.84 × 10− 12) and right precentral gyri (b = −0.003; P = 2.54 × 10−24), left (b = −0.004; P = 1.94 × 10−12) and right superior frontal gyri (b = −0.003; P = 4.65 × 10−11), left (b = −0.004; P = 1.05 × 10−10) and right transverse temporal gyri (b = −0.003; P = 8.24 × 10−10), and left (b = −0.002; P = 5.22 × 10−6) and right paracentral gyri (b = −0.002; P = 5.63 × 10−6). Duration of epilepsy also negatively associated with unilateral thickness measures in the right precuneus (b = −0.003; P = 6.03 × 10−21), right pars opercularis (b = −0.003; P = 5.59 × 10−13), and right cuneus (b = −0.002; P = 1.1 × 10−9; Supplementary Table 8). In the MTLE-R subgroup, duration of epilepsy negatively associated with volume measures in the right hippocampus (b = −22.42; P = 1.1 × 10−7), and the right thalamus (b = −18.11; P = 1.84 × 10−5), and thickness measures in the left transverse temporal gyrus (b = −0.007; P = 8.39 × 10−5; Supplementary Table 8). In the all-other-epilepsies subgroup, duration of epilepsy negatively associated with bilateral thickness measures in the left (b = −0.003; P = 3.39 × 10−7) and right caudal middle frontal gyri (b = −0.003; P = 6.91 × 10−8), left (b = −0.003; P = 1.36 × 10−9) and right superior frontal gyri (b = −0.003; P = 3.16 × 10−7), and the left (b = −0.003; P = 3.17 × 10−5) and right precuneus (b = −0.003; P = 5.01 × 10−9), in addition to unilateral thickness measures in the right precentral gyrus (b = −0.004; P = 1.16 × 10−12), right cuneus (b = −0.003; P = 8.57 × 10−8), right pars triangularis (b = −0.003; P = 5.16 × 10−7), and right supramarginal gyrus (b = −0.003; P = 2.24 × 10−7). Duration of epilepsy also showed a positive association with the size of the left lateral ventricle in the all-other-epilepsies group (b = 13.6; P = 1.17 × 10−5).

29In the all-epilepsies group, age at onset of epilepsy negatively associated with thickness measures in the left (b = −0.003; P = 2.66 × 10−15) and right superior frontal gyri (b = −0.003; P = 9.77 × 10−10), left (b = −0.003; P = 2.78 × 10−9) and right pars triangularis (b = −0.003; P = 6.51 × 10−7), right pars opercularis (b = −0.003; P = 5.4 × 10−14), left transverse temporal gyrus (b = −0.003; P = 1.03 × 10−8), and right cuneus (b = −0.001; P = 4.9 × 10−6). In the all-other-epilepsies subgroup, age at onset negatively correlated with thickness measures in the left (b = −0.003; P = 3.21 × 10−8) and right superior frontal gyri (b = −0.002; P = 1.18 × 10−4), left (b = −0.002; P = 8.42 × 10−6) and right precuneus (b = −0.002; P = 7.23 × 10−5), right pars triangularis (b = −0.003; P = 2.53 × 10−5), and right supramarginal gyrus (b = −0.002; P = 2.38 × 10−6). Age at onset also positively associated with the size of the right lateral ventricle in the all-other-epilepsies subgroup (b = 57.73; P = 1.62 × 10−7).

30Age at onset negatively associated with other regional volumetric and thickness measures in the all-epilepsies, IGE, MTLE-L, MTLE-R, and all-other-epilepsies groups, but these associated areas showed no significant structural differences in the primary case-control analysis (Table 1 and Supplementary Table 8).

31There were no interaction effects between age and syndromic diagnosis in the all-epilepsies, MTLE-L, MTLE-R, IGE, or all-other-epilepsies groups.

32In our sample of 2149 individuals with epilepsy and 1727 healthy controls, we had 80% power to detect Cohen’s d effect sizes as small as d = 0.091 at the standard alpha level of P < 0.05 (two-tailed), and 80% power to detect Cohen’s d effect sizes as small as d = 0.149 at the study’s stringent Bonferroni-corrected threshold of P < 1.49 × 10−4.

33N80, the number of cases and controls required to achieve 80% power to detect group differences using a two-tailed t-test at P < 0.05, ranged from N80 = 6, to detect group effects in the right hippocampus in our MTLE-R group, to N80 = 503, to detect group effects in the right pars opercularis in our ‘all epilepsies’ group (Tables 2 and 3).

34In the largest coordinated neuroimaging study of epilepsy to date, we identified a series of quantitative imaging signatures—some shared across common epilepsy syndromes, and others characteristic of selected, specific epilepsy syndromes. Our sample of 2149 individuals with epilepsy and 1727 controls provided 80% power to detect differences as small as d = 0.091 (P < 0.05, two-tailed), allowing us to identify subtle, consistent brain abnormalities that are typically undetectable on visual inspection, or overlooked using smaller case-control designs. This international collaboration addresses prior inconsistencies in the field of epilepsy neuroimaging, providing a robust, in vivo map of structural aberrations, upon which future studies of disease mechanisms may expand.

35In the first of five cross-sectional MRI analyses, we investigated a diverse aggregation of epilepsy syndromes, putative causes, and durations of disease. This all-epilepsies group exhibited shared, diffuse brain structural differences across several regions including the thalamus, pallidum, precentral, paracentral, and superior frontal cortices. With the exception of hippocampal volume and entorhinal thickness differences (Supplementary material), these structural alterations were not driven by any specific syndrome or dataset (Supplementary Figs 3 and 7). Our findings suggest a common neuroanatomical signature of epilepsy across a wide spectrum of disease types, complementing recent evidence for shared genetic susceptibility to a wide spectrum of epilepsies. Some structural and genetic pathways may be shared across syndromes, despite the heterogeneity of epilepsy and seizure types. This shared MRI signature underpins the contemporary shift towards the study of epilepsies as network phenomena.

36In MTLE, as expected, we observed hippocampal volume abnormalities ipsilateral to the patient’s side of seizure onset. Neither MTLE-L nor MTLE-R showed significant contralateral hippocampal volume reductions, confirming that sporadic, unilateral MTLE is not routinely underpinned by bilateral hippocampal damage. Both MTLE groups showed extrahippocampal abnormalities in the ipsilateral thalamus and pallidum, with widespread reductions in cortical thickness, supporting a growing body of literature indicating that MTLE, as an example of a specific disease constellation in the epilepsies, is also a network disease, extending beyond the mesial temporal regions. Disruption of this network, notably in the thalamus and thalamo-temporal white matter tracts, may be associated with postoperative seizure outcome in MTLE.

37Patients with left and right MTLE showed distinct patterns of structural abnormalities when compared to controls, resolving conflicting findings from smaller studies, some reporting an equal distribution of structural differences, and others indicating more diffuse abnormalities, either in left MTLE or in right MTLE. The structural differences observed in the present study may reflect a younger age at onset of epilepsy in left MTLE, which occurred, on average, 1.2 years earlier than those with right MTLE (Supplementary Table 20). Independent, large-scale studies of MTLE patients have confirmed a significantly earlier age at onset in left, compared to right, MTLE. Duration-related effects were also observed in right, but not left, MTLE, pointing to possible biological distinctions between the two.

38In IGE, a clinically and biologically distinct group of epilepsies typically associated with ‘normal’ MRI on clinical inspection, we identified reduced volume of the right thalamus, and thinner precentral gyri in both hemispheres, supporting prior reports of structural, electroencephalographic, and functional abnormalities in IGE. These IGE cases were considered typical by reviewing neurologists, suggesting that this common type of epilepsy is also associated with quantifiable structural brain abnormalities.

39The precentral gyri, site of the primary motor cortex, showed bilateral structural deficits across all epilepsy groups (all-epilepsies, IGE, MTLE-L, MTLE-R, and all-other-epilepsies), without detectable inter-cohort or between-disease heterogeneity (Supplementary Figs 3–12). Atrophy of the motor cortex has been linked to seizure frequency and duration of epilepsy in MTLE; here, we observed a negative correlation between precentral (and postcentral) grey matter thickness and duration of epilepsy in the aggregate all-epilepsies group.

40The right thalamus also showed evidence of structural compromise across all epilepsy cohorts, re-emphasizing the importance of the thalamus as a major hub in the epilepsy network. Loss of feed-forward inhibition between the thalamus and its neocortical connections may be epileptogenic, and thalamocortical abnormalities have previously been reported in IGE and MTLE. These findings support prior ‘system epilepsies’ hypotheses of pathophysiology, suggesting that a broad range of common epilepsies share vulnerability within a thalamocortical structural pathway involved in, and likely affected by, seizures. Given this study’s cross-sectional design, we cannot determine if these are causative changes, consequences of recurrent seizures, prolonged drug treatment, or a combination of factors. The epilepsies, as a broad group, may involve progressive structural change, indicating the need for large-scale longitudinal studies.

41A heterogeneous subgroup of individuals without confirmed diagnoses of IGE or MTLE with hippocampal sclerosis showed similar patterns of structural alterations to those observed in the aggregate all-epilepsies cohort. The findings included enlarged ventricles, smaller right pallidum and right thalamus, and reduced thickness across the motor and frontal cortices. Hippocampal abnormalities were not observed in this subgroup, suggesting that the patterns of reduced hippocampal grey matter observed in the aggregate group were driven by the inclusion of MTLEs with hippocampal sclerosis. Unlike the IGE, MTLE, and aggregate epilepsy cohorts, this subgroup also showed bilateral enlargement of the amygdala—a phenomenon previously reported in non-lesional localization-related epilepsies and non-lesional MTLEs. Non-lesional MTLEs formed a large proportion of this ‘all-other-epilepsies’ cohort (43.3%; 445 individuals), but the subgroup included many other focal and unclassified syndromes, potentially obscuring specific biological interpretations. Future, sufficiently powered studies will stratify this cohort into finer-grained subtypes to delineate syndrome-specific effects.

42Despite its international scale, our study has limitations. All results were derived from cross-sectional data: we cannot distinguish between historical acute damage and progressive abnormalities. We cannot disentangle the relative contributions of environmental and treatment-related factors, including antiepileptic medications, seizure types and frequencies, disease severity, language dominance, and other initial precipitating factors. On average, duration of epilepsy was at least 10 years; longitudinal investigations of new-onset and paediatric epilepsies will provide a more comprehensive understanding. Despite using standardized image processing protocols, quality control, and statistical techniques, some brain measures showed a wide distribution of effect sizes across research centres, which may reflect sample heterogeneity and differences in scanning protocols (Supplementary material).

43We observed modest thickness differences across the majority of cortical regions; Cohen’s d effect sizes ranged from small to moderate (d = 0.2–0.5), with some very small effects (d < 0.2) noted in the right pars opercularis, bilateral pars triangularis, and bilateral transverse temporal gyri of the aggregate all-epilepsies group. Other large-scale ENIGMA studies have reported similarly modest (albeit less widespread) cortical abnormalities in psychiatric illnesses including major depression and bipolar disorder. Although epilepsy is characterized by an enduring predisposition to generate abnormal excessive or synchronous neuronal activity in the brain (, our findings indicate that common epilepsies are associated with widespread, but relatively subtle, structural alterations of the neocortex. Replication in independent MRI cohorts, complemented by advanced imaging modalities and large-scale gene expression datasets, will help elucidate how these cortical abnormalities relate to underlying disease processes.

44Overall, in the largest neuroimaging analysis of epilepsy to date, we demonstrate a pattern of robust brain structural abnormalities within and between syndromes. Specific functional interpretations cannot be inferred from grey matter differences, but lower volume and thickness measures may reflect tissue loss, supporting recent observations that the common epilepsies cannot always be considered benign. The study provides a macroscopic neuroanatomical map upon which neuropathological work, animal models, and further gene expression studies, can expand. Our consortium plans to investigate more specific neuroanatomical traits and epilepsy phenotypes, explore sophisticated shape and sulcal measures, and eventually conduct genome-wide association analysis of brain measures, to improve our understanding and treatment of the epilepsies.

45All image processing, quality assurance, and statistical analysis protocols for this study can be downloaded from the ENIGMA website, at: http://enigma.usc.edu/ongoing/enigma-epilepsy/enigma-epilepsy-protocols/.