MRI AND PATHOLOGY CORRELATIONS IN THE MEDULLA IN SUDEP: A POST-MORTEM STUDY

Aims: Sudden unexpected death in epilepsy (SUDEP) likely arises as a result of autonomic dysfunction around the time of a seizure. In vivo MRI studies report volume reduction in the medulla and other brainstem autonomic regions. Our aim, in a pathology series, is to correlate regional quantitative features on 9.4T MRI with pathology measures in medullary regions. Methods: 47 medullae from 18 SUDEP, 18 non-epilepsy controls (NEC) and 11 epilepsy controls (EPC) were studied. In 16 cases, representing all three groups, ex vivo 9.4T MRI of the brainstem was carried out. Five regions of interest (ROI) were delineated, including the reticular formation zone (RtZ), and actual and relative volumes, as well as T1, T2, T2* and MTR measurements were evaluated on MRI. On serial sections, actual and relative volume estimates using Cavalieri stereological method and immunolabelling indices for myelin basic protein, synaptophysin and Microtubule associated protein 2 (MAP2) were carried out in similar ROI. Results: Lower relative RtZ volumes in the rostral medulla but higher actual volumes in the caudal medulla were observed in SUDEP (p<0.05). No differences between groups for T1, T2, T2* and MTR values in any region was seen but a positive correlation between T1 values and MAP2 labelling index in RtZ (p<0.05). Significantly lower MAP2 LI were noted in the rostral medulla RtZ in epilepsy cases (p<0.05). Conclusions: Rostro-caudal alterations of medullary volume in SUDEP localise with regions containing respiratory regulatory nuclei. They may represent seizure-related alterations, relevant to the pathophysiology of SUDEP.


INTRODUCTION
SUDEP is the leading cause of death in young adults with intractable epilepsy and may have many causes (1). Recent clinical, neuroimaging and experimental research provides converging evidence for a central failure of autonomic regulation of cardiovascular and respiratory function in SUDEP (1,2). Brainstem autonomic dysfunction, including vital cardiorespiratory regulation in the post-ictal period, may represent a final common pathway.
The primary risk factor, based on meta-analysis of SUDEP series, is poor control of generalised tonic-clonic seizures (GS) (3,4). Following GS, irregular, shallow breathing (5) and ictal central apnoea with hypercapnia and hypoxaemia are not infrequent (6)(7)(8). SUDEP often occurs nocturnally or during sleep (9) and convergence of impaired brainstem arousal systems interacting with inadequate respiratory drive during seizures may also be relevant. In experimental models of SUDEP, spreading depolarisations to the dorsal medulla following seizures have also been shown to mediate cardio-respiratory arrest (10). In structural MRI studies, volume loss in the medulla in SUDEP has been shown in some but not all studies (11)(12)(13).
Central control of respiration is regulated by interconnected medullary nuclear groups forming the ventral respiratory column (14). One such nucleus is the pre-Bötzinger complex (pre-BötC), an essential generator of inspiratory rhythm (15). The pre-BötC forms an illdefined region in the reticular zone (RtZ) of the medulla, more specifically in the ventrolateral medulla (VLM) comprising mixed neuronal types with high levels of somatostatin and neurokinin 1 receptor (NK1R) expression (16,17). It is modulated by other brainstem nuclear groups, including the serotonergic neurones of the medullary raphe (MR) (18). In previous studies of the medulla in SUDEP, we have shown regional alterations in specific neuronal and glial subsets including serotonergic and neuropeptidergic neurones, potentially relevant to medullary dysfunction (19,20). Whether subtle cellular alterations in medulla contribute to regional volume changes or alteration of MRI signal is unknown.
In this post-mortem study, we aimed to study volume and signal changes in high resolution 9.4 T MRI of post-mortem brainstem samples from SUDEP cases compared to control groups, including in the RtZ and VLM regions. We also explored pathological correlations for any volume or signal changes using quantitative immunohistochemistry for myelin, neuronal somato-dendritic and synaptic markers in these regions.

Case selection
Brainstems from 29 epilepsy cases were included in this study ( Table 1). The study has ethical approval with appropriate consent for research for all cases included in the study. Tissue from all post-mortem cases was retained with era-appropriate consent collected through the Epilepsy Society Brain and Tissue Bank (ESBTB) at UCL and Brain UK via the pathology department at Derriford Hospital, Plymouth, United Kingdom. The epilepsy cases included Dravet syndrome (DS) cases obtained from the University of Melbourne, Australia (as previously reported (21). The cause of death for epilepsy cases was categorised as SUDEP in 18 cases or a non-SUDEP death in 11 cases (Epilepsy control group, EPC) according to current criteria (22). 18 non-epilepsy controls (NEC) with no history of epilepsy were included, obtained through ESBTB and the MRC Brain and tissue Bank, Edinburgh. The clinical and neuropathology records, including epilepsy history, circumstances of death and main neuropathology findings are detailed in supplemental Table  1. Of note, both EPC and SUDEP groups included cases with recent onset of epilepsy (in the last year two years of life).

Brainstem 9.4T MRI
In 16 cases (6 SUDEP, 3 EPC, 7 NEC), MR imaging of whole or hemi-brainstems, fixed in 10% formalin for at least one week, was carried out (Figure 1a,b). In six cases (representing cases from all patient groups) in order to adequately fit the sample in the MR radiofrequency coil without distortion, just hemi-brainstems were scanned. In those cases, the sagittal plane of the section was taken to one side of the midline to ensure inclusion of central raphe nuclei (Supplemental Figure 1a, b). The samples were scanned on a 9.4T MR scanner (Agilent Technologies, Santa Clara, CA, USA) using a volume coil with a diameter of 33 mm (Rapid Biomedical GmbH, Rimpar, Germany). After manual global shimming and acquiring fast localization images, a T2-weighted (T2W) scan in addition to quantitative T1, T2, T2* and magnetization transfer ratio (MTR) mapping sequences were applied as in our previous study (23). To get an in-plane spatial resolution of 100×100 μm 2 and a slice thickness of 400 μm for all the samples, the field of view, the size of the acquisition matrix and the number of slices were adjusted according to the size of each sample. The FOVs used had a range of 34×34 mm 2 to 30×24 mm 2 and the number of slices ranged between 80 to 120. A multi-slice spin echo (SEMS) scan with TE/TR = 35/8000 ms was applied to obtain T2W images. The acquisition time (TA) was about 8 hours. For the T1 mapping, a multi-TR saturation recovery approach was applied using a SEMS sequence (TE=12ms and TRs = 0.35 / 0.5 / 0.85 / 1.5 / 3 / 9.5 s). For short TRs, several slice packages with a fewer number of slices and different offsets were acquired separately and then combined to obtain a full coverage of the sample. The number of averages was adapted for each TR to obtain a good signal-to-noise ratio. The number of slices was 10 for TRs=350/500ms (NA=8) and 40 for TRs=850/1500ms (NA=3). For the long TRs (3s and 9.5s), all the slices were acquired in a single scan. The NAs were 2 and 1 for TR of 3s and 9.5s, respectively. The acquisition time of all the saturation recovery scans depended on the number of slices and it was about 7.5 hours for 80 slices. For the T2 mapping, a multi-echo SEMS sequence with a TR = 8 s and TEs = 15/25/35/60 ms was used (TA=9 h). A multi-slice multi-gradient echo sequence with TR=5s, TE=5.5/13.5/21.5 ms and flip angle (FA) of 90 deg was applied for the T2* mapping (TA = 6h). For the MTR mapping, a multi-slice gradient echo sequence with TE/TR = 10/4200 ms and FA=40 deg (TA=7.5 h) was applied. The magnetisation transfer RF pulse was applied with a FA of 500 deg and frequency offsets of 6 kHz (MT weighted) and 100 kHz (control). The total scanning time for the entire protocol including the adjustments was approximately 39 hours. MRI Relaxation times T1, T2, T2* and MTR maps were obtained by processing the data using both an in-house developed script running under Matlab (Mathworks, Massachusetts, USA) and ImageJ (National Institutes of Health, Bethesda, MD, USA; http://imagej.nih.gov/ij/index.html) and the mean relaxation times for T1, T2, T2* were calculated for each case in different medulla regions to evaluate grey and white matter and MTR values for myelin integrity (detailed below).

Cavalieri method for medulla volume estimations
Following the MRI scanning, medulla blocks were cut at 5mm intervals from obex 0mm (the caudal recess of the 4 th ventricle) in an axial plane to the upper brainstem, using a cutting guide. For all 47 cases, a single block from the medulla blocks was selected, aiming to include obex level 6 to 12 mm where available, as the putative level of the human preBötC (24).
Serial sections through each block were cut at 20μm thickness using the Tissue-Tek AutoSection automated microtome (Sakura Finetek, U.S.A. Inc) obtaining 150-200 sections per case. Every 10th section (200 μm steps) was stained with cresyl-violet (CV) as previously described (20). The obex levels of each section were determined independently by two observers (MT, SP) according to the atlas of Paxinos (25). Ten CV sections at 200 um equidistance (equivalent to 2000 μm) were selected for volume estimation, nearest to the preBötC region in the VLM. Using the Cavalieri method (StereoInvestigator software, MBF Biosciences) the volume of RtZ was measured in each case, in addition to two control regions of interest (ROI) the 12th cranial nerve nuclei (12 CNN) and inferior olive nuclei (IO) as comparisons. For whole brainstem sections, both left and right sides were analysed and averaged. The boundaries of RtZ ROI were defined as follows: medial boundary (midline of the brainstem), lateral boundary (line parallel to the midline touching the lateral end of the IO, dorsal boundary (line perpendicular to the midline at the ventral end of the solitary tract) and ventral boundary (line perpendicular to midline, at medial end of inferior olive, and following the contour of the inferior olive to meet the lateral boundary) ( Figure  1c). The RtZ ROI therefore captured all nuclei of the reticular formation (intermediate, lateral, ventral, dorsal and gigantocellular nuclei) and the ventral medullary respiratory column as well as the nucleus ambiguus (NA) and medullary raphe obscurus (25). The contours of the IO and 12 CNN were manually outlined and selected as control regions as they could be easily anatomically identified on CV sections. The actual volume (AV) of the RtZ as well as a relative volume (RV) RtZ volume/ total volume of all ROI was calculated. The RV was calculated to factor for case to case differences, differential shrinkage from fixation and processing as well as tilts in the axial plane of cut between cases. The coefficient of error was 0.05 or less in all volume measurements for each case.

MRI and alignment and quantitation
The MRI image slices were manually co-aligned with the 10 CV sections selected for the Cavalieri volume estimation (Supplemental Figure 1a,b). Distinct anatomical landmarks (folds in the olivary nuclei, vascular structures, contours of the ventricles etc.) were used independently by two observers (SP, MT) to match the MRI slices that corresponded to the range of serial CV sections selected for the Cavalieri volume method. The mean number of MRI slices per case was 8 (range 7-10). Using the freehand selection drawing tool on Image J, the following ROIs were drawn ( Figure 1d): RtZ (using the same boundaries as described for the Cavalieri volume estimation), VLM (representing the ventrolateral quadrant within the RtZ using boundaries as previously described (20)), MR (using a line parallel to midline drawn from the halfway point between the medial end of the tip of the inferior olive (IO) and the midline), IO (drawing around nucleus and hilus freehand, excluding accessory nuclei), ST (drawing around the contours of the solitary tract). The mean quantitative MRI (qMRI) values for T1, T2, T2*, and MTR values and volume estimate were obtained for each ROI across 10 levels. The relative volume of the VLM to the total volume of medullary regions for each case was obtained.

Immunohistochemistry and quantification
Further sections from the mid-obex region of each case were immunolabelled for myelin basic protein (MBP, clone SMI94 (Sternberger); 1:2000 dilution) as a measure of myelin integrity, MAP2 (clone HM-2 (Sigma M4403); 1:5000 dilution) as a neuronal soma/ dendritic marker and synaptophysin (clone DAK-SYNAP (DAKO); 1:100 dilution) for grey matter, on Leica Bond max automated IHC stainer. The UCLH + enhancer protocol was applied which included pre-treatment (ENZ1 for 10 min for MBP, ER1 for 30 min for MAP2, and ER2 for 20 min for synaptophysin), followed by 15 min incubation with primary antibody, 8 min each with post-primary and polymer, 10 min with DAB, 5 min with DAB enhancer, and 5 min with haematoxylin. The slides were washed with Bond-wash solution between all steps. Slides were scanned with a Leica SCN400F digital slide scanner (Leica Microsystems, Wetzlar, Germany) at 40x magnification. Analysis of immunostained sections was carried out with Definiens Tissue Studio 3.6 (Definiens AG, Munich, Germany). Five ROIs were outlined using co-ordinates as detailed in the MRI study ( Figure 1h) and including both sides where available. Any defects in the section, for example small tissue folds or staining artefacts, were carefully outlined and excluded from the analysis. Intensity thresholds for detection of each marker were developed through training a range on representative sections and areas, then applied to all cases and the mean labelling index (LI) for each ROI calculated averaged for both sides.

Statistical analysis and graphical representation
Statistical analysis was carried out using SPSS (IBM version 21) and Graph-Pad Prism software was used for the graphical representation of the data. Non-parametric correlation (Kendal's tau) and Mann Whitney tests were used for comparison between groups, with clinical and post-mortem factors. Uncorrected values of p<0.05 were taken as significant with corrections for multiple comparisons. Multiple regression analysis was used to evaluate effects of other variables (post-mortem delay, obex level) on measurements between the cause of death groups.

Volume estimations with 9.4T MRI and Cavalieri method
There were no significant differences in MRI volume estimations (actual and relative) between the cause of death groups (SUDEP, EP-C, NEC) ( Table 2); when comparing all epilepsy cases to NEC, significantly higher actual volumes were noted in the IO ROI in epilepsy (p<0.05). Cavalieri method volume estimates on the entire series, including cases with MRI, also showed no difference in actual or relative volume estimates of the RtZ between the three cause of death groups ( Table 2). There was also no difference in mean RtZ volume when all epilepsy cases were compared to NEC. A positive correlation with MRI and Cavalieri volume estimates of the RtZ was noted for all 16 cases (p <0.05) but with lower volume estimates using the Cavalieri method (Supplemental Figure 1c).

Quantitative MRI, immunohistochemistry and correlations
qMRI values showed no differences for T1, T2, T2* and MTR values between the three cause of death groups for any ROI. There were also no differences when all epilepsy cases were compared to NEC (Table 2). MAP2 labelling highlighted neurones and dendritic processes in the RtZ, VLM and other ROI (Figure 1h In the 47 cases studied, SMI94/MBP LI was significantly lower in epilepsy groups than in NEC in the VLM (p<0.01), RtZ (p<0.05) but there were no significant differences between SUDEP and EPC groups (Figure 2d). MAP2 LI was significantly lower in the epilepsy groups than NEC for all ROI (p<0.05 to p<0.001), but there were no significant differences between EPC and SUDEP groups (Figure 2e). Synaptophysin LI was significantly lower in SUDEP than NEC for all ROI (p<0.05 to p<0.01) but no significant difference was observed between EPC and NEC (Figure 2f).
Clinical and neuropathological factors-There was no significant correlation between MRI or pathology measures and age at death. Age at seizure onset for the epilepsy group was not available in all cases (Supplemental Table 1), so this was not further analysed. Epilepsy cases were categorised into pathology negative cases (no significant neuropathological abnormality) or with a confirmed lesional pathology at post-mortem, for example hippocampal sclerosis, DNT, old scars, potentially relevant to cause or effect of seizures (see supplementary Table 1). There were no significant differences in medullary volume estimates or SMI94/MBP, synaptophysin, MAP2 LI between cases with or without brain pathology. There were no significant differences in recorded brain weights between cause of death groups, but a positive correlation noted between Cavalieri-estimated actual volume of the RtZ and brain weight (p<0.05).
Stratifying cases for obex levels-The obex levels of the medullary samples that were available from the tissue bank archives were not identical between cases; in many cases only a single medulla block was available. Although there was no significant difference in the mean mid-obex level between the cause of death groups (Table 1), the volume of the RtZ, as estimated by both MRI and Cavalieri methods were observed to increase with obex level (Supplemental Figure 1c). We therefore stratified the 47 cases into three groups according to the mid-obex level of the block, for the Cavalieri volume estimation and for the neuropathological measures: caudal medulla <3mm, mid medulla 4-6mm and rostral medulla >7mm (Figure 3). Further analysis showed significantly greater actual RtZ volumes in the mid-medulla in the SUDEP group (but not EPC) compared to NEC (Figure 4a). In contrast, significantly lower RtZ relative volumes were noted in the rostral medulla in SUDEP compared to NEC (p<0.05) (Figure 4b). For pathology measures significantly lower MAP2 were noted in rostral medulla ROI in epilepsy cases (p<0.03 to p<0.004) and lower MAP2 in the mid medulla in the RtZ in the SUDEP group alone compared to NEC (p<0.05). SMI94 LI was significantly lower LI in the IO ROI only in epilepsy cases (p<0.05) in the rostral medulla.
For the 16 cases with MRI, cases were stratified into two groups, caudal and rostral medulla (≤6mm / >6mm). Significantly higher VLM volumes (p<0.05) (Figure 3, 4c) (and relative VLM volumes p = 0.05, (Figure 3d)) were noted in the caudal medulla in epilepsy cases but lower VLM volumes in the rostral medulla compared to NEC (p<0.05) (Figure 4c). There was no statistical relationship between qMRI measurements and obex level or differences between cause of death groups when stratified into upper and lower medulla groups.
Post-mortem variables-The post-mortem intervals (PMI) and fixation times (total time in formaldehyde prior to tissue processing) varied between cases as these were mainly coronial post-mortem investigations carried out in different centres. Although not significantly different between the cause of death groups (Table 1) there was variability, with longer mean fixation times in epilepsy controls but shorter PMI. For all 16 cases, a positive correlation between T1 and MTR values with PMI (p<0.05) in the RtZ was noted but an inverse correlation between T1 values and fixation time (p<0.05). Similar associations were also seen for the VLM and IO regions but not for T2/T2* values in any region. There were no significant correlations between volume estimates (either on MRI or Cavalieri) or for SMI94/MBP LI with either PMI or fixation time. A negative correlation between fixation times and MAP2 LI (p<0.05) and synaptophysin LI (p<0.005) was noted.
Multivariate analysis-We carried out multiple regression analysis to factor all independent variables of fixation time, PMI and obex level effecting pathology and MRI measures. This further analysis confirmed no differences in qMRI values for any ROI between cause of death groups. MAP2 LI remained significantly lower in the rostral medulla RtZ in all epilepsy cases compared to NEC (p<0.05).

DISCUSSION
In this post-mortem study, incorporating high resolution MRI of the brain stem, we provide further evidence for volume alterations in autonomic regulatory regions in SUDEP compared to controls, when aligning for the obex level. We also observed differences in the level of medullary myelination and neuronal marker MAP2 in epilepsy cases compared to controls which could indicate brainstem neurodegeneration. Identification of alterations in the medulla in epilepsy will further the understanding of networks involved and pathways that could lead to SUDEP.

Volume change in medulla in SUDEP and epilepsy.
Brain atrophy is accelerated in epilepsy. In the ENIGMA consortium of 2149 epilepsy patients, distinct patterns of reduced grey matter in cortical and subcortical structures were shown on MRI (26) and recent longitudinal MRI studies have confirmed progressive and accelerated regional cortical thinning in patients with focal epilepsy compared to control groups (27). None of these studies looked into pathological correlations, but cortical neuronal loss or neurodegeneration is the proposed, unproven underlying process with regional patterns suggesting a selective vulnerability. The brainstem is rarely investigated in epilepsy but is relevant to SUDEP (28). All MRI studies to date have used retrospective analysis of in vivo MRI investigations carried out at variable times prior to the SUDEP. In the report by Mueller et al., based on in vivo 4T MRI, volume loss in the brainstem, including the upper medulla, was shown in one person who went on to have SUDEP compared to non-SUDEP controls (12). This finding was supported in a subsequent study of 26 cases with 3T MRI, showing volume loss in autonomic regions of interest in the brainstem, including the midbrain periaqueductal grey, raphe nuclear regions and medulla in SUDEP (29). Limitations due to the lower resolution reachable at 3T prevented delineation of anatomical brainstem structures but nevertheless their method ensured accurate rostrocaudal evaluation of volume loss (12). A recent VBM study of 3T MRI from 25 SUDEP cases also showed volume loss in periaqueductal grey of the midbrain in addition to left thalamus, left hippocampus and posterior cingulate and cerebellum, but did not replicate findings in the medulla (11).
We did not observe differences in medulla regional volumes between the cause of death groups with either MRI or Cavalieri method for the whole cohort, but when stratifying cases according to their obex level, some significant rostro-caudal differences emerged. Decreased actual volumes specifically in the VLM region in the rostral medulla were seen in the epilepsy group on MRI and this was reinforced in our larger pathology series by decreased relative RtZ volumes in the rostral medulla and more specifically in the SUDEP group. The RtZ ROI includes both key medullary respiratory nuclear groups of the ventral respiratory column in the VLM, such as the pre-BötC, nucleus ambiguus (24,25), as well as medullary raphe which contain serotonergic neurones with chemosensory regulatory properties (14). In contrast, in the caudal medulla greater RtZ volumes were observed in the SUDEP group and greater actual and relative volumes in the VLM in all epilepsy cases on MRI. Knowledge of the connectivity of brainstem respiratory/autonomic networks is largely based on animals (14,15,30) as human fMRI studies lack the spatial resolution (31,32). There is a hierarchical organisation in a rostro-caudal direction of the ventral respiratory groups in the VLM; the Bötzinger complex and pre-BötC, that control expiratory and inspiratory respiratory rhythms respectively, are more rostral than ventral respiratory groups that coordinate output to the phrenic and spinal motor neurones (14). In the human brainstem, the pre-BötC is considered largest between obex levels 9 to 10 mm (Schwarzacher et al., 2011). Our observation of lower volumes in SUDEP cases in rostral medulla could be relevant to pathology in the pre-BötC region. However, the higher RtZ volumes noted in SUDEP in the caudal medulla could implicate pathology in VRG that determines the motor control of respiration, of potential equal importance (Figure 3).
We also aimed to address the pathological correlation of MRI medullary volume changes in SUDEP. Quantitative ex vivo 9.4T MRI has the sensitivity to detect subtle differences in neuronal density, myelination and brainstem anatomy (33). Indeed in vivo studies have demonstrated its utility to show microanatomical structures in the human midbrain (34). In previous quantitative studies of epilepsy surgical samples with similar immunomarkers, a positive correlation between T1, T2* relaxation times and neuronal marker MAP2 but an inverse correlation with MTR was observed (23). In contrast, myelin stains including SMI94/MBP show an inverse correlation with T1, T2, T2* but positive correlations with MTR in both cortex and white matter ROI (23,35). In the current study, we noted similar correlations in many but not all ROI. This could partly be explained by the more complex intermingling of grey matter nuclei and white matter tracts in the medulla compared to their relative compartmentalisation in the neocortex. We did not, however identify differences in these MRI quantitative values between the cause of death groups.
We noted reduced myelin content, based on MBP labelling, in the RtZ and VLM in epilepsy cases. There is evidence from both imaging and pathology studies of compromised myelination both in the vicinity and at distance from a focal epileptogenic focus, likely reflecting network changes due to seizures (36). It is known that seizures can spread to the brain stem (37) and neuronal activity influences oligodendroglial proliferation, and myelin plasticity in epilepsy, although the pathological basis is not well understood (38). When we factored for obex level however, significantly reduced myelin labelling was only noted in the IO region in the SUDEP group. In contrast, neuronal and dendritic marker MAP2 was reduced in the rostral RtZ (> 7mm) in all epilepsy cases and the mid medulla (3-6mm) in the SUDEP group only; following multivariate analysis, factoring fixation time effects, MAP2 LI remained significantly lower in the rostral medulla RtZ in epilepsy, but not specifically in the SUDEP group. Reduced MAP2 could indicate loss of neurones and/or dendritic branches or cellular MAP2 expression. There are mixed neuronal types in the RtZ including the pre-BötC, formed by interconnecting and modulating nuclei and networks (14) and characterised by multi-transmitter systems (39); these include pacemaker-like somatostatin/neurokinin 1 receptor (NK1R)-positive cells of the pre-BotC and modulating serotonergic neurons in the MR and VLM. In a previous study we showed a reduction in somatostatin neuronal subsets in the VLM and tryptophan hydroxylase expressing serotonergic neurones in the VLM and medullary raphe, particularly in rostral medulla (obex 7-9mm), the putative level of the pre-BötC in SUDEP (20). In contrast however, Neurokinin1 Receptor positive neurones in the VLM were predominantly reduced in the caudal medulla (3-4mm) in SUDEP (20). These rostro caudal variations reflect the differential MAP2 findings and volume alterations in the current series.
In a recent study addressing medullary glial populations, we did not identify any differences in overall GFAP astroglial labelling to suggest reactive gliosis in SUDEP, although diminished populations of specialised chemosensory glial subsets were identified (19). We have also noted no significant reduction in medullary catecholaminergic neuronal populations in SUDEP (Patodia et al., under revision) and in an earlier study we also reported no significant microglial activation in the medulla, hypoxic neuronal injury or loss of integrity of the blood brain barrier in SUDEP to suggest more acute medullary pathology (40). Taken together these findings propose a chronic, regionally-selective reduction of specific types of medullary neurones and glia, which collectively could explain the subtle quantitative volume alterations reported in the current and in-vivo MRI studies (12,13). These alterations could be directly secondary to seizures or its effects. It is recognised that neuroplasticity in respiratory nuclei occurs, enabling a capacity for compensatory adaptation under physiological challenges (41) as well as disease states (14). Chronic, intermittent hypoxia and hypercarbia result in augmentation of respiratory motor outputs (42), likely involving serotonergic networks, their connections to the preBötC (43) and central chemoreception (44). For example, a recent experimental study demonstrated a reduction in the number of serotonergic and total NeuN-positive neurons in the MR and VLM following 30 days of increased CO 2 (45). It is plausible that recurrent generalised seizures with associated, albeit intermittent, episodes of hypoxia and hypercarbia induce respiratory plasticity in brainstem nuclei which could influence vulnerability to SUDEP. Further work to understand brainstem networks in epilepsy and SUDEP is clearly needed.
Limitations include that as an archival study, different levels of the medulla had been sampled and fixation times varied between cases which we have factored in the analysis. Our SUDEP group included both definite and possible/probable cases according to current criteria (22); however there were no differences in the pathology and volume measurements between these SUDEP groups (data not shown). Ex-vivo MRI methods were based on a small sample size of 16 and MTR values only achieved on one epilepsy control. Furthermore, in-vivo MRI volume analysis, such as voxel-based morphometry, include steps where whole brains are spatially 'normalised' to a template to account for individual differences in brain shape. Such a step is not available for this ex-vivo study. We calculated relative volumes in addition to actual volume estimates of the RtZ in an aim to address case differences, tissue processing effects and non-axial section planes. We also identified higher IO volumes in all epilepsy cases as an unexplained finding on MRI and we cannot exclude that this has influenced the relative volume estimates. In addition, the co-registration between MRI and pathology slices was carried out by manual alignment and we have only examined a 2mm axial length of the medulla from each case in the relatively small number that had 9.4T MRI. Further studies of brainstem regional volumes, including the pons and midbrain, are essential to evaluate any global brainstem alterations in epilepsy and SUDEP.
In summary, our study findings suggest a rostro-caudal alteration in medullary volume and neuronal composition in the RtZ in epilepsy and SUDEP. As these localise with regions containing respiratory regulatory nuclei, they could represent seizure-related alterations, as a SUDEP vulnerability factor and potential disease biomarker.

Supplementary Material
Refer to Web version on PubMed Central for supplementary material.

ACKNOWLEDGMENTS
UCL is part of the Center for SUDEP Research (CSR) and supported through the National Institute of Neurological Disorders And Stroke of the National Institutes of Health (Award Numbers neuropathology of SUDEP: 5U01NS090415 and SUDEP admin core grant: U01-NS090405). Epilepsy Society supports SMS, and through the Katy Baggott Foundation, supports the UCL Epilepsy Society Brain and Tissue Bank (ESBTB). The Horne Foundation Trust supported infrastructure in the ESBTB that enabled this work. This work was undertaken at UCLH/UCL who received a proportion of funding from the Department of Health's NIHR Biomedical Research Centres funding scheme. We are very grateful for provision of additional SUDEP and control material for this study from the following resources: The MRC Sudden Death Brain Bank in Edinburgh (cases detailed supplemental Table). Tissue samples were also obtained from David Hilton at Derriford Hospital as part of the UK Brain Archive Information Network (BRAIN UK) which is funded by the Medical Research Council and Brain Tumour Research. We are grateful to the help of Derek Marsden in the neuropathology laboratory for his technical expertise.   ) in cause of death groups; mean SMI94/MBP labelling index was significantly lower in SUDEP and epilepsy controls (EPC) than in nonepilepsy controls (NEC) in the VLM and RtZ but there were no significant differences between SUDEP and EPC. E. In all 47 cases for all obex levels, scatter graphs of MAP2 labelling index; mean labelling index was significantly lower in SUDEP and EPC groups than NEC for all ROI but there were no significant differences between EPC and SUDEP. F. In all 47 cases for all obex levels, scatter graphs of synaptophysin (syn) labelling index; mean labelling index was significantly lower in SUDEP than NEC for all ROI but no significant difference was observed between EPC and NEC. Bars represent mean values in D-F and whiskers standard deviation.

Figure 3. Diagrammatic representation of obex levels, respiratory autonomic nuclei and volume alterations in SUDEP and epilepsy
On the left hand side, the key respiratory nuclear regulatory groups are outlined and their approximate obex levels in the human medulla and position indicated. On the right side, the volume changes noted in SUDEP and epilepsy cases compared to controls when cases were grouped according to the obex level (rostral, mid or caudal). Volume contraction in the reticular zone (RtZ) of the rostral medulla and expansion in the caudal medulla was observed with the Cavalieri method in SUDEP cases (blue boxes) and similar volume changes in the ventrolateral medulla (VLM) region in epilepsy cases with MRI (white boxes). A. Cavalieri volume estimates stratified into three groups, according to the mid-obex level (caudal medulla <3mm, mid medulla 4-6mm and rostral medulla >7mm) showed significantly greater actual RtZ volumes in the mid-medulla regions in the SUDEP group compared to non-epilepsy controls (NEC). B. Cavalieri relative volume estimates stratified into three groups, according to the mid-obex level (caudal medulla <3mm, mid medulla 4-6mm and rostral medulla >7mm) showed significantly lower RtZ relative volumes in the rostral medulla in SUDEP compared to NEC. C. Medullary volume estimates for the 16 cases with 9.4T MRI stratified into two groups (caudal (≤6mm) and rostral medulla (>6mm)) and significantly higher actual VLM volumes were noted in the caudal medulla in epilepsy cases (All Ep) (and SUDEP group alone) but lower VLM volumes in the rostral medulla in all epilepsy cases compared to NEC D. MRI relative VLM volumes also showed significant differences between epilepsy cases and NEC in the caudal medulla.

Summary of clinical details for the four groups studied.
Detailed case information is provided in supplemental Table 1). PMI = post mortem interval (time between death and post mortem examination), Fixation time = time brain fixed in formalin prior to processing. Differences between groups with non-parametric tests; significant differences shown in bold.