2.1 Participants selection

A single-center retrospective study was conducted at Xiangya Hospital, Central South University from January 2016 to December 2020. The study was approved by the Medical Ethics Committee of Xiangya Hospital, Central South University, and was carried out in compliance with the Declaration of Helsinki. Written Informed consent for inclusion in this study and publication for clinical details were obtained from each participant involved in this study.

During this period, a total of 340 adults diagnosed with suspected AE were referred to our hospital, of whom 45 patients were performed with cerebral ¹⁸F-FDG PET imaging. After reviewing their electric medical recording carefully, 10 patients were excluded by (1) 6 patients who underwent PET scanning after the acute stage of disease; (2) 2 patients had incomplete clinical information; (3) 2 patients modified the final diagnosis with Creutzfeldt–Jakob disease. Finally, 35 patients with AE were recruited for this study (Figure 1).

The inclusion criteria were as follows: (1) subacute onset of neurological symptoms (working memory deficits, seizure, decreased level of conscience, movement disorders, unexplained dysautonomia, or abnormal movements) or psychiatric symptoms (behavior changes, disturbances in mood or character, hallucinations, speech dysfunction); (2) at least one of the following abnormalities: CSF with pleocytosis (white blood cell count of more than 5 cells per mm³), CSF oligoclonal bands or elevated IgG index, CSF elevated protein, abnormal EEG, MRI features suggestive of AE; (3) taking the PET examination in a relatively acute disease course, according to previous definition of the acute phase (within 3 months) and chronic phase (over 3 months) of AE^{1, 13}; (4) reasonable exclusion of alternative causes.

All patients were further classified as (1) antibody-positive AE based on Graus criteria¹: autoimmune neuronal antibodies were detected in CSF and/or serum, including definite-AE, antibody-positive limbic encephalitis (Ab-positive-LE), and anti-NMDAR encephalitis; (2) antibody-negative AE: the diagnosis was made when all three of the following criteria have been met^{16, 17}: (i) none of the known neuronal antibody was identified in the serum or CSF sampling; (ii) ≥1 feature present: CSF-specific oligoclonal bands; intrathecal synthesis of immunoglobulin or elevated CSF IgG index; CSF pleocytosis (>5 cells/mm³) or elevated protein; brain MRI abnormalities suggestive of AE; effective immunotherapy; (iii) reasonable exclusion of alternative causes. These patients were sorted into Ab-negative-LE, probable-AE, possible-AE.¹⁸ Two experienced neurologists reviewed the medical record and made the corresponding classification (Yuwei Dai and Yuanyuan Xie).

The data concerning demographic information, clinical characteristics, associated neoplasia, treatment regimes, treatment delay as expressed in weeks between the onset of symptoms, and the duration of symptom onset and ¹⁸F-FDG PET/CT were collected by reviewing patients' charts and databases. First-line immunotherapy included intravenous methylprednisolone pulse (IVMP), intravenous immunoglobulin (IVIG), and plasma exchange (PLEX); rituximab (RTX) and cyclophosphamide served as second-line immunotherapy. The levels of consciousness of enrolled patients were evaluated based on the Glasgow Coma Scale (GCS) score on admission. GCS scores lower than 8 (GCS≤8) indicated decreased level of consciousness.¹⁹ Forty-three age-matched, gender-matched volunteers without any neurological or psychiatric illness were enrolled as healthy controls.

2.2 Exclusion of infectious etiologies

Infectious evaluations of serum and CSF were performed for all patients as a diagnostic procedure of AE. The following measures were adopted to exclude possible infectious etiologies: Gram stain and India ink stain; aerobic and anaerobic bacterial cultures; mycobacterium culture and acid-fast bacillus stain; fungal culture; immunoglobulin M (IgM) and IgG specific for parasites; polymerase chain reaction (PCR) tests for enterovirus, Epstein Barr virus (EBV), herpes simplex virus (HSV) 1 and 2, varicella-zoster virus, or human herpes virus 6. No definite evidence of infection was found in these enrolled cases.

2.3 Neuronal antibody detection

For every patient with suspected AE, comprehensive neural antibody studies were conducted in serum and CSF, including both indirect immunofluorescence assay (IIFA) fixed by rat brain tissues for unknown antibodies and commercial workup panels for known antibodies.

First, serum and CSF samples were screened with indirect tissue-based assays (TBAs) composed of rat cerebellum and rat hippocampus for the presence of neuronal antibodies. Then, those patients were evaluated for known antibodies in the serum and CSF, using IIFA fixed by commercial transfected HEK293 cells assays (CBAs) to test antibodies against cell surface antigens (Euroimmun, Lübeck, Germany), including NMDAR, LGI-1, contactin-associated protein-2 (CASPR2), α-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid receptor (AMPAR), γ-aminobutyric acid type B receptor (GABA_BR) and IgLON family member 5 (IgLON5), and immunoblotting (Euroimmune Ag) to detect intracellular antibodies (GAD65, SOX1, Hu, Yo, Ri, Ma2, CV2, amphiphysin). Positivity was determined by repeat testing (≥2 times per sample). The antibody testing results were interpreted by two experienced investigators using the signal of the surrounding fields as negative controls.²⁰

2.4 Evaluation and follow-up of long-term prognosis

Two neurologists (Xianghe Liu and Lili Long) retrospectively evaluated the severity of clinical symptoms by a comprehensive scale proposed recently, the Clinical Assessment Scale in Autoimmune Encephalitis (CASE).²¹ Meanwhile, the modified Rankin Scale (mRS) was also conducted to evaluate the neurological disability of all patients.²² The CASE score (range: 0–27) and mRS score (range: 0–6) were obtained at four time points (at admission, at disease peak, at discharge, and at the last follow-up) based on medical records. Responders to treatment were defined as the decrease of at least 1 point in mRS score after immunotherapy.¹⁸

The follow-up was achieved via subsequent outpatient visits. The long-term outcomes were evaluated at least 1 year from disease onset.⁶ Patients were categorized into two groups based on the mRS score at the last follow-up: good outcome (mRS: 0–2) and poor outcome (mRS: 3–6).

2.5 Review of brain MRI

Patients' brain MRIs were collected using 3.0-Tesla or 1.5-Teala MAGNETOM Verio scanner (Siemens, Germany) in Xiangya Hospital, Central South University. The following sequences were acquired: T1-weighted images (T1WI), T2-weighted images (T2WI), fluid-attenuated inversion recovery (FLAIR), diffusion-weighted images (DWI), and gadolinium-enhanced images. In the case of multiple MRI examinations, MRI data interpretation primarily focused on the results nearest to the onset of symptoms. MRI images with evidence of underlying encephalitis or inflammatory changes (T2 or T2/FLAIR hyperintensities restricted to one or both mesial temporal lobes, or multifocal in gray matter, white matter, or both in the cerebral cortex, cerebellum, striatum, brainstem) were considered as positive. Blinded interpretation of MRI results was conducted by two experienced radiologists.

2.6 ¹⁸F-FDG PET/CT scanning

All subjects accepted ¹⁸F-FDG PET/CT scanning at the PET Center of Xiangya Hospital, Central South University using a Discovery Elite PET/CT scanner (GE Healthcare, Boston, MA, United States) during the acute phase of the disease. All participants were required to fast for at least 6 h and discontinue all steroid hormones or antiseizure drugs (ASDs) for at least 24 h. Fasting blood glucose levels of all patients did not exceed 8 mmol/L before injection of ¹⁸F-FDG. ¹⁸F-FDG was intravenously injected following a dose of 3.7 MBq/kg within 1 min. Before PET scan acquisition, participants were allowed to rest quietly in a dedicated room for an hour. A low-dose CT scan (120 kV, 60–180 mA/s, slice thickness 3.75 mm) was started first, brain PET scanning was performed for 15–20 min followed by whole-body PET scanning for 35–45 min. The whole-body PET scans ranged from the top of the head down to mid-thigh, to screen occult tumors. The full width of the scan at half maximum was 5.4 mm. The brain images were reconstructed into a 192 × 192 trans-axial matrix (35 cm field of view) using the 3D VUE Point (GE Healthcare) ordered-subset expectation–maximization algorithm. CT images were obtained for attenuation correction simultaneously.

2.7 ¹⁸F-FDG PET/CT imaging analysis

PET imaging data were analyzed by SPM12 software (Welcome Department of Cognitive Neurology, London, United Kingdom), implanted in MATLAB 2018a (MathWorks Inc., Sherborn, MA, USA) environment. The image preprocessing steps were as follows: scans from all individuals were firstly segmented and spatially normalized into standard stereotactic Montreal Neurological Institute (MNI) space with voxel sizes of 2 mm × 2 mm × 2 mm. Then, normalized images were smoothed by an 8-mm Full-Width Half-Maximal (FWHM) 3D Gaussian kernel to increase the signal-to-noise ratio. Subsequently, PET image values were processed to intensity normalization, performed by proportional scaling to overall activity.

Once the PET images were preprocessed, a two-sample t-test was conducted to delineate metabolic pattern characteristics of single subjects with age and gender as the nuisance variables, comparing each patient with the control group. FDG uptake was compared voxel-to-voxel between patients and the control group using two-t-sample-test of SPM. We also compared AE patients with good outcomes between AE patients with poor outcome using the same way to the predictive value of ¹⁸F-FDG-PET biomarkers on long-term prognosis. Metabolic changes were considered statistically significant at the strict threshold (p < 0.001) and corrected for multiple comparisons (family-wise error [FWE] corrected, p < 0.05) with cluster size (K_E) above 20 contiguous voxels. If significant metabolic changes were not found, a more lenient threshold was set (p < 0.001, uncorrected; K_E = 20 voxels).

2.8 Statistical analysis

Continuous variables were described using means ± standard deviation (SD) or median (interquartile range [IQR]). The counting data were presented as percentages (%). Continuous variables were tested to determine whether they conformed to a normal distribution by the Shapiro–Wilk normality test first. For the univariate analysis of poor long-term outcomes, continuous variables were compared using the t-test or non-parametric Mann–Whitney U-test, and Fisher's exact tests were used for categorical variables. The variables with p values < 0.05 on univariate analysis were further selected by the least absolute shrinkage and selection operator (LASSO) regression using the R software package glmnet.²³ The features selected by LASSO regression were subsequently entered into the multivariate Cox proportional hazards regression model. The strength of association was evaluated by Hazard ratios (HR) and corresponding 95% confidence interval (CI) in the multivariate Cox regression model. A two-tailed p-value < 0.05 was considered significant. Statistical analyses were performed with SPSS 26.0 software package for Windows (IBM Corp.) and R language (version 4.3.3).

3.1 Clinical characteristics

Demographic information and clinical characteristics of 35 patients (15 males, 20 females) with AE are summarized in Table 1. The average age of disease onset was 43.89 ± 18.21 years old. Forty-three age-matched healthy subjects (mean age: 45.13 ± 16.25 years old, 20 females) served as controls.

In the acute stage of AE, the main symptoms of this cohort were seizures (20/35, 57.1%), psychiatric symptoms (19/35, 54.3%), followed by decreased level of consciousness (8/35, 22.9%), memory deficit (6/35, 17.1%), central hypoventilation (5/35, 14.3%). Other clinical symptoms included ataxia (3/35, 8.6%), dystonia (3/35, 8.6%), and speech dysfunction (2/35,5.7%). 2 subjects presented with FBDS, and they were tested positive for LGI1 antibodies. Tumors were found in 5 patients (2 with lung cancer, 1 with colon cancer, 1 with hematological malignancies, and 1 with breast cancer).

All patients had tested antibodies in both serum and CSF (35/35, 100%). Positive neural antibodies were detected in 24 patients (24/35, 68.6%), and 11 patients were classified as antibody-negative AE (11/35, 31.4%). Twenty-four patients with positive neural antibodies were further classified into 10 anti-NMDAR encephalitis, 3 Ab-positive-LE, and 11 definite AE. In 24 patients with antibody-positive AE, 21 patients (21/35, 60.0%) found positive neuronal antibodies in both serum and CSF (10 NMDAR; 5 GABA_BR; 2 LGI1;2 CASPR2; 1 D2R; 1 IgLON5), 3 patients (3/35, 8.6%) only tested positive intracellular antibodies in serum (2 SOX; 1GAD65). In 11 patients with antibody-negative AE, 6 were probable AE and 5 were possible AE. A detailed table about the clinical, laboratory, and radiological features of patients with antibody-negative AE is in Table S1.

CSF samples were collected during hospitalization (35/35, 100%). Eight patients had increased intracranial pressure (≥180 mmH₂O). White blood cell (WBC) counts in the CSF of 12 subjects were mildly increased (12/35, 34.3%). 9 (9/35, 25.7%) had elevated cerebrospinal fluid protein levels CSF protein. Among 25 patients who were tested for CSF oligoclonal bands, 60% (15/25, 60.0%) reported a positive result. Twenty patients completed 24-h dynamic or video EEG monitoring (20/35, 57.1%), and 17 showed abnormal EEG results, mainly revealing slow-waves or slow-sharp waves (17/20, 85.0%).

All subjects were treated with first-line immunotherapy (35/35,100.0%). Eleven of them received additional second-line immunotherapy based on first-line immunotherapy during the acute phase (11/35, 31.4%). Twenty-one patients (21/35, 60.0%) had a good response to treatment. Six patients experienced at least one relapse (6/35, 17.1%), of which 4 were antibody-positive AE. The median interval between disease onset and first recurrence was 15.5 months (IQR: 9.5–33). The median mRS scores at admission, at disease peak, at discharge, at the last follow-up were 5.0, 5.0, 3.0, and 2.0. The median CASE scores at admission, at disease peak, at discharge, at the last follow-up were 5.0, 7.0, 4.0, and 3.0.

3.2 PET imaging analysis

The details of imaging results (both MRI and ¹⁸F-FDG PET/CT) for individual patients were summarized in Table S2. The MRIs were performed at a median of 17 days (IQR: 12.0–33.0) after disease onset and 23 days (IQR:15.0–35) for ¹⁸F-FDG PET/CT. The median time interval between MRI and PET was 3 days (IQR:3.0–5.0). There was no difference in time interval from the initial symptom onset until imaging examination. For MRI imaging results, 22 individuals showed normal or unspecific MRI results (22/36, 62.9%). Abnormal findings were observed in 13 patients (13/35, 37.1%), in which 3 patients presented with abnormal T2/FLAIR signal change in the bilateral medial temporal lobe (MTL). Meanwhile, 33 patients presented abnormal metabolism patterns on ¹⁸F-FDG PET/CT (33/35, 94.3%). ¹⁸F-FDG PET/CT demonstrated higher sensitivity than MRI (94.3% vs. 37.1%, p < 0.05). In particular, the detection rate of ¹⁸F-FDG PET/CT was 90.9% in patients with normal MRI (20/22, 90.9%), and 13 patients with abnormal MRI results also detected abnormal brain metabolism.

Figure 2 demonstrated corresponding MRI images, abnormal cerebral glucose metabolism patterns, and ¹⁸F-FDG PET/CT results based on SPM12 of 4 representative patients (No. 12, No. 15, No. 16, No. 35). Compared to the healthy control group, brain metabolic images of each subject were performed with semi-quantitatively analysis (p < 0.05, FWE corrected). Thirty-three patients presented mixed metabolism patterns, showing cortical hypometabolism combined with focal hypermetabolism (33/35, 94.3%). For groupwise analysis, patients with AE mainly presented mixed metabolism patterns compared to the matched controls, demonstrating hypermetabolism of cerebellum, BG, MTL, midbrain, pons, medulla oblongata, insula, middle frontal gyrus, paracentral lobule, precuneus, precentral gyrus and relatively hypometabolism in the frontal cortex, occipital cortex, middle temporal gyrus, superior temporal gyrus, right superior parietal gyrus, middle parietal gyrus, precuneus, left cingulate gyrus (p < 0.05, FWE corrected; Figure 3). Mixed metabolism patterns were a representative feature of patients with AE in the acute stage.

3.3 Group analyses of individual patients in different long-term outcome subgroups

3.3.3 Prognostic value of PET imaging

Further semi-quantitative group analysis of metabolic patterns based on SPM found that patients with AE with poor outcomes showed decreased metabolism of the right superior frontal gyrus along with increased metabolism of the middle and upper brainstem, mainly located in bilateral pons, compared with those with good outcomes (p < 0.001, uncorrected, Figure 4).