In vivo NMDA receptor function in people with NMDA receptor antibody encephalitis

[18F]GE-179 PET binding is reduced in seropositive females following NMDAR-antibody encephalitis

Five patients (mean age 28 ± 5 years, all female) with definite NMDAR-antibody encephalitis (13) were imaged at University College London 2 to 16 months from hospital discharge. Four patients (cases #1-4) who had persistent GluN1-autoantibodies in serum (titre range 1:160 to 1:320) and were scanned two to eight months after hospital discharge were classified as persistently “seropositive”. At the time of PET imaging, these four patients had mild symptoms. One completely symptom-free patient (case #5) who no longer had serum GluN1-antibodies was scanned 14 to 18 months after discharge from hospital, and was classified as “seroreverted”. The clinical MRI was normal in all patients. Patient characteristics are displayed in Table 1. For comparison, we included 29 healthy volunteers (mean age 41 ± 13 years, 8 [28%] female; 10 scanned at UCL, 19 at two additional sites).

Persistently seropositive cases #1-4 had lower [¹⁸F]GE-179 V_T in comparison to the seroreverted case #5 and to healthy volunteers (Fig. 1). Seropositive cases had lower [¹⁸F]GE-179 V_T in grey matter (estimated marginal mean corrected for age, sex, and site: 6.2, 95% confidence interval [CI] 4.4 – 8.0) compared to healthy volunteers (8.8, 95% CI 8.1 – 9.4, F=6.5, p=0.02). We performed a voxel-based analysis to address the spatial distribution of the findings (Fig. 2B), which showed reduced [¹⁸F]GE-179 V_T in persistently seropositive cases within bilateral anterior temporal lobes (left, T = 4.5, 6537 mm³, p_FWE = 0.02; right, T = 4.9, 4779 mm³, p_FWE=0.05) and a large cluster involving bilateral superior parietal lobes, paracentral lobules, left posterior cingulate gyrus, and left precuneus (T = 5.8, 22069 mm³, p_FWE < 0.001). A priori volume of interest analyses of radioligand V_T in each cerebral lobe corroborated the voxel-based analysis, showing approximately 30% reductions in the regional V_T estimates in the temporal (34%; cases 5.7 [95% CI 3.9 – 7.6] vs. controls 8.6 [95% CI 8.0 – 9.2], F=8.3, p=0.008) and parietal (31%; cases 6.2 [95% CI 4.3 – 8.2] vs. controls 9.1 [95% CI 8.4 – 9.8], F=7.3, p=0.01) lobes (Fig. 2A). V_T in white matter was comparable between seropositive cases (4.3 [95% CI 2.6 – 6.0]) and healthy volunteers (5.6 [95% CI 5.0 – 6.2], F=2.0, p=0.17 Fig. 2A). Seroreverted case #5 had V_T estimates similar to healthy volunteers in grey (9.7) and white (5.8) matter (Fig. 1).

• Download figure
• Open in new tab

Figure 1: [¹⁸F]GE-179 uptake in autoantibody “seropositive” or “seroreverted” patients with NMDAR-antibody encephalitis and healthy volunteers.

The figure shows the spatial distribution of [¹⁸F]GE-179 total volume of distribution (V_T) on brain slices and surface projections. Panel A shows mean uptake in healthy volunteers (HV, n=29). Panel B displays individual V_T distributions in persistently autoantibody “seropositive” NMDAR-antibody encephalitis cases (n=4) that were scanned 2-8 months after discharge and had elevated serum GluN1-autoantibodies (1:160 – 1:320). The mean brain V_T in these cases is shown in the last row. Panel C displays one “seroreverted” NMDAR-antibody encephalitis case scanned 16 months after discharge with undetectable GluN1-autoantibodies on the day of scanning.

• Download figure
• Open in new tab

Figure 2: Statistical analyses of [¹⁸F]GE-179 uptake in NMDAR-antibody encephalitis cases and healthy volunteers.

Panel A shows boxplots of regional [¹⁸F]GE-179 total volume of distribution (V_T) in healthy volunteers (CTRL, n=29, split by cohort) and persistently autoantibody “seropositive” cases with NMDAR-antibody encephalitis (ENC, n=4). Panel B shows whole brain voxel-wise differences in [¹⁸F]GE-179 V_T between persistently “seropositive” encephalitis cases and healthy volunteers (p<0.05 family wise error [FWE] corrected, red colours indicate decreased uptake in encephalitis cases).

There were non-significant trends towards higher grey matter [¹⁸F]GE-179 V_T with lower serum GluN1 immunoglobulin G levels and longer time from hospital discharge and from episode onset (Fig. 3A-C). Grey matter V_T did not correlate with cognitive testing results at the time of scanning (Addenbrooke’s Cognitive Examination III [ACE-III] questionnaire, range 86 to 96, Fig. 3D) nor symptom severity at discharge (CASE score at discharge, range 4 to 8, Fig. 3E), but there was little variability between patients in these test results.

• Download figure
• Open in new tab

Figure 3: Correlation of [¹⁸F]GE-179 uptake with patient characteristics and MRI-based voxel-based morphometry.

Scatter plots (panels A to E) show the association of [¹⁸F]GE-179 total volume of distribution (V_T) in NMDAR-antibody encephalitis cases (n=5, both “seropositive” and “seroreverted” cases) with serum GluN1-specific immunoglobulin G (IgG, panel A), time from discharge to scan (panel B), time from episode onset to scan (panel C), cognitive testing with Addenbrooke’s Cognitive Examination III (ACE-III, panel D), and symptom severity at discharge measured with the CASE score (panel E). Panel F displays voxel-wise differences in grey matter volume between encephalitis cases and healthy volunteers (p<0.05 family wise error [FWE] corrected, blue colours indicate atrophy in encephalitis cases).

Gray matter atrophy and neurotransmitter levels assessed using MRI

We assessed grey matter atrophy using MRI-based voxel-based morphometry (VBM). We found reduced grey matter volume in the cerebellar hemispheres (left, T = 4.6, 7388 mm³, p_FWE = 0.001; right, T = 4.3, 4512 mm³ voxels, p_FWE = 0.008; Fig. 3F). There was no overlap between areas of decreased grey matter volume and regions with significantly reduced [¹⁸F]GE-179 V_T (Fig. 2B).

For all subjects imaged at UCL, we evaluated whether changes in radiotracer uptake correlated with hippocampal neurotransmitter levels, as assessed by the glutamate plus glutamine (Glx) to total creatine (Glx:tCr) ratio from magnetic resonance spectroscopy (MRS). The Glx:tCr ratio was no different between seropositive cases and healthy controls (2.6 [95% CI 0.5 – 4.6] vs. 1.5 [95% CI 0.5 – 2.4], F=1.0, p=0.35) but Glx:tCr ratio in the hippocampi negatively correlated with [¹⁸F]GE-179 uptake in grey matter (Pearson’s partial correlation coefficient -0.63, p=0.04).

Study population

We screened consecutive patients with NMDAR-antibody encephalitis that were hospitalised or under regular follow-up in two tertiary referral centres in the UK (John Radcliffe Hospital, Oxford University Hospitals NHS Foundation Trust, and St. George’s University Hospital NHS Foundation Trust London). We included patients with definite disease according to established diagnostic criteria (13) and positive GluN1-autoantibodies in CSF at diagnosis who were (i) between 18 and 65 years old; (ii) had minimal or no symptoms and thus could tolerate a 70-minute PET-MR scan; and (iii) did not take any medication that could interfere with NMDA receptors, in particular antipsychotics or antidepressants, permitting immunotherapy and anticonvulsants. The included patients were divided into a group scanned two to eight months after hospital discharge with mild symptoms and persistent “seropositivity” (cases #1-4) that had detectable GluN1-autoantibodies in serum on the day of scanning, and a “seroreverted” participant (case #5) that did not have detectable GluN1-autoantibodies in serum on the day of scanning, was scanned > 1 year after hospital discharge and was completely symptom free. All patients were scanned at University College London (UCL site), UK. Clinical data were retrospectively extracted from medical records according to previously described procedures (1). We calculated the CASE score, a measure of autoimmune encephalitis severity, at discharge (range 0 to 27 points; higher scores indicate more severe disease) (24). Cognitive testing with Addenbrooke’s Cognitive Examination III (ACE-III; range 0 to 100 points; higher scores indicate better cognitive function; scores > 89 typically considered normal) and assessment with Beck’s Depression Inventory (BDI; range 0 to 63 points; higher scores indicate more depressive symptoms; scores < 10 typically considered normal) were performed shortly before the scan.

We also included 29 healthy volunteers that had a [¹⁸F]GE-179 PET scan at one of three UK sites: 10 at the UCL site; 10 at Addenbrooke’s Hospital in Cambridge (Cambridge site); and 9 at Hammersmith Hospital in London (Hammersmith site). Healthy volunteers did not have any relevant neurological or psychiatric disorders and were not taking any regular medication.

The study was approved by the local research ethics committee and the local National Health Service Trust research office. All participants gave written informed consent.

Data acquisition

All participants underwent dynamic emission PET scans after intravenous injection of a target dose of 185 MBq of [¹⁸F]GE-179. [¹⁸F]GE-179 binds specifically to the phencyclidine site inside the NMDA receptor channel (25). Tracer binding requires the opening of the receptor channel complex. Successful blocking experiments were performed in vivo for GE-179 in rats following activation of NMDA receptors using electrical stimulation (14) and for the structural analogues CNS-5161 and GMOM in non-anaesthetised rodents (26, 27), confirming the specificity of the tracer to the phencyclidine site within the open NMDA receptor ion channel. Because anaesthesia is expected to reduce the baseline opening probability of NMDA receptors, blocking experiments in anaesthetized rodents were not successful (15).

The UCL site used a Siemens Biograph mMR combined PET-MR scanner for acquisition of 70-minute dynamic emission scans with a 25.8 cm axial field of view. To make the scanning procedure more easily applicable in clinical practice, as requested by the study funder (Medical Research Council), at the UCL site the scan duration was shortened from 90 to 70 minutes and an image-derived input function (IDIF) was used instead of arterial sampling.

Our group showed that shortening the scan duration to 70 minutes provides V_T estimates that correlate excellently (Spearman’s rank correlation coefficient 0.99, 95% CI 0.98 - 0.99, p<0.001) with those obtained from 90 minute scans and the results were robust for all studied subregions (7 distributed regions, Spearman’s rank correlation coefficient ranging from 0.98 to 1.00) (28). List-mode data were initially processed using an in-house motion detection algorithm based on a principle component analysis of the dynamic PET signal (29). Frame timing was individually adapted to exclude the typically short (duration between seconds to max. 1 min) signal alterations and reduce the effect of interframe sudden head movement.

This resulted in a variable number of frames (45 to 54). Data were reconstructed with 3-dimensional filtered backprojection to voxel sizes of 1.40 × 1.40 × 2.03 mm. Venous blood samples were obtained 7, 12, 22, 42, and 62 minutes after [¹⁸F]GE-179 injection to provide plasma-to-whole blood and parent fraction corrections to the IDIF (30). The MR protocol included a high-resolution volumetric T1-weighted sequence (MPRAGE) with 2000 ms repetition time, 2.92 ms echo time, 256 × 256 matrix, and 1.1 × 1.1 × 1.1 mm voxels. To facilitate the IDIF we also performed an arterial time-of-flight magnetic resonance angiography (TOF-MRA) sequence that included the carotid arteries with 22ms repetition time, 4.17ms echo time, 256 × 256 matrix, and 0.625 × 0.625 × 2mm voxels. The magnetic resonance spectroscopy (MRS) acquisition (PRESS, repetition time, 2000ms; echo time 30ms) involved the manual placement and estimation of a single voxel covering the hippocampi (left and right separately) of each subject with approximate dimensions 30 mm (Anterior—Posterior), 16 mm (Left—Right), and 16 mm (thickness). One-hundred-twenty transients were collected per measurement.

The Cambridge site used a GE Discovery 690 combined PET-CT scanner for acquisition of 90-minute dynamic emission scans with a 15.4 cm axial field of view. Data were reconstructed with 3-dimensional filtered backprojection to 58 frames with voxel sizes of 2.00 × 2.00 × 3.27 mm. Arterial blood from the radial artery was continuously sampled for the first 6.5 minutes of the scan with discrete arterial samples taken at 0.5, 4.5 10, 15 and 20 minutes post-injection, and at 10-minute intervals thereafter. Each volunteer underwent 3T brain MRI on a Siemens Trio or Skyra system (Siemens Healthineers, Erlangen, Germany) including a high-resolution volumetric T1-weighted sequence (MPRAGE).

The Hammersmith site used a Siemens/CTI ECAT EXACT HR+ model 962 PET scanner for acquisition of 90-minute dynamic emission scans with a 15.5 cm axial field of view (25).

Data were reconstructed with Fourier rebinning combined with 2-dimensional filtered back-projection to 34 frames with voxel sizes of 2.09 × 2.09 × 2.42 mm. Arterial samples were obtained continuously for the first 15 minutes, and discretely at baseline and at eight further times throughout the remainder of the scan. Each volunteer underwent 3T MRI on a GE Signa HDx system acquiring a T1-weighted coronal SPGR sequence with a voxel size of 0.938 mm × 1.100 mm × 0.938 mm.

Data preprocessing

We performed data pre-processing following a previously described and validated protocol (30). The Cambridge and Hammersmith sites used arterial sampling, with the input function corrected by plasma parent fraction and plasma-to-whole blood ratio determined from the arterial samples. The UCL site used an IDIF from the carotid arteries corrected by plasma parent fraction and plasma-to-whole blood ratio estimated from venous samples. The IDIF involved segmentation of the carotid arteries on TOF-MRA scans followed by extraction and partial volume correction of the radioactivity concentration in these arteries from the dynamic PET data. We previously validated this method by showing that the combination of an IDIF with venous samples provides reliable [¹⁸F]GE-179 V_T estimates compared to arterial sampling (Pearson’s correlation coefficient 0.95, p<0.001) (30).

We fitted arterial (Cambridge and Hammersmith sites) and venous (UCL site) radioactivity concentrations, plasma parent fractions, and plasma-to-whole-blood ratios as described previously (30). Site-specific spatial smoothing was applied to adjust the PET images from all sites to the same spatial resolution. Dynamic PET images were corrected for head motion using a post-hoc frame-to-frame realignment method and the realigned PET images were linearly co-registered with the T1-weighted MRI using Statistical Parametric Mapping 12 (SPM12, Wellcome Centre for Human Neuroimaging, London, https://www.fil.ion.ucl.ac.uk/spm/software/spm12/). We parcellated brain structures on T1-weighted MRIs using an algorithm based on Geodesic Information Flows (GIF) (31), and adjusted the PET images for partial volume effects using the iterative Yang method (32) implemented in the PET-PVC toolbox (33). Regional partial volume-corrected time-activity curves were extracted and modelled using a 2-brain-compartment 4-rate-constant kinetic model with a variable blood volume component as described previously (25). Total V_T in mL/cm³ was estimated for overall grey and white matter and lobar regions of interest as a metric of radiotracer binding.

Additionally we computed V_T for each scan on a voxel level using Logan graphical analysis (34). We used the Computational Anatomy Toolbox (CAT12, http://www.neuro.uni-jena.de/cat/) in SPM12 to segment and nonlinearly register all T1-weighted MRI images to a standard template. During this process we also extracted parametric grey matter volume images. We applied the same transformations to co-registered V_T images. We performed median filtering and gaussian smoothing with an 8 mm full-width-at-half-maximum kernel. For all subjects imaged at the UCL site, MRS data were phase and frequency corrected, before being summed into a single spectrum. A Hankel singular value decomposition filter was applied to remove the residual water signal and metabolites were quantified using a simulated basis set in Tarquin (http://tarquin.sourceforge.net). The average of the glutamate plus glutamine (Glx) to total creatine (Glx:tCr) ratio in the left and right hippocampus was extracted for each subject.

The live cell-based assays for serum GluN1-autoantibody titre measurements were performed as follows. HEK293T cells were cultured at 37°C in Dulbecco’s Modified Eagle Medium (DMEM) supplemented with foetal calf serum and antibacterial/antimycotic solution for 30 hours on glass cover slips, then transfected with plasmid DNA containing GluN1 subunit for 15 hours. Sera were tested following incubation for a further 24 hours in 1.4 μL/ml ketamine supplemented culture medium. Serum, previously thawed to room temperature, was diluted 1:20 with DMEM supplemented with 1% bovine serum albumin and 200mM 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES). 250μL diluted samples were tested alongside positive and negative controls. Samples were incubated with transfected cells for 1 hour at room temperature. The diluted samples were then aspirated, each well was washed 3 times with DMEM-HEPES, each cover slip was then fixed with 4% formaldehyde for 5 minutes and then washed three times with DMEM-HEPES. The cells were then incubated with secondary antibody (Alexa Fluor 594-conjugated donkey anti-human IgG Fc-gamma; Jackson 709-585-098) in the dark for 45 minutes. This was aspirated, and each well was washed twice in DMEM-HEPES then twice in PBS. Each coverslip was individually mounted in 4′,6-diamidino-2-phenylindole (DAPI)-supplemented mounting medium on a glass slide. Following review of positive and negative control cover slips, test samples were read at 40x zoom on a Leica DM2000 fluorescence microscope. Positivity was determined by the presence of a characteristic coronal cell surface deposition of fluorescent reporter antibody on multiple cells. All positive samples were repeated on an independent assay and serially diluted to establish an end-point dilution, the dilution at which the positive signal was still unambiguously present. The threshold for seropositivity was set at > 1:20.

Statistical analysis

We compared regional V_T estimates and hippocampal MRS data between patients with NMDAR-antibody encephalitis and healthy volunteers using the general linear model adjusting for age, sex, and site (i.e. scanning equipment). We analysed the four “seropositive” patients and one “seroreverted” patient separately. We report V_T and Glx:tCr mean values as estimated marginal means that were adjusted for age, sex, and, for V_T, site. We calculated the correlation between hippocampal MRS data and grey matter [¹⁸F]GE-179 V_T using Pearson’s partial correlation adjusting for sex and age.

To analyse the subregional distribution of the findings, we assessed the spatial distribution of differences between groups for voxel-wise V_T estimates. To address the contribution of grey matter atrophy to our findings, we performed voxel-based morphometry (VBM) using segmented, normalised and modulated parametric grey matter volume images. Both voxel-wise datasets were assessed using general linear models corrected for age, sex, and site. We report voxel-wise p-values at a threshold of <0.05 on a cluster-level family-wise error corrected for multiple comparisons (p_FWE).