RESEARCH ARTICLE Antibodies Contributing to Focal Epilepsy Signs and Symptoms Score Marienke A. A. M. de Bruijn, MD ,1 Anna E. M. Bastiaansen, MD,1† Hana Mojzisova, MD,2† Agnes van Sonderen, MD, PhD,1‡ Roland D. Thijs, MD, PhD,3,4 Marian J. M. Majoie, MD, PhD,5,6,7,8 Rob P. W. Rouhl, MD, PhD,8 Marleen H. van Coevorden-Hameete, MD, PhD,1 Juna M. de Vries, MD, PhD,1 Amaia Muñoz Lopetegi, MD,9 Bob Roozenbeek, MD, PhD,1 Marco W. J. Schreurs, PhD,10 Peter A. E. Sillevis Smitt, MD, PhD,1 Maarten J. Titulaer, MD, PhD,1 and the ACES Study Group§ Objective: Diagnosing autoimmune encephalitis (AIE) is difficult in patients with less fulminant diseases such as epi- lepsy. However, recognition is important, as patients require immunotherapy. This study aims to identify antibodies in patients with focal epilepsy of unknown etiology, and to create a score to preselect patients requiring testing. Methods: In this prospective, multicenter cohort study, adults with focal epilepsy of unknown etiology, without recog- nized AIE, were included, between December 2014 and December 2017, and followed for 1 year. Serum, and if avail- able cerebrospinal fluid, were analyzed using different laboratory techniques. The ACES score was created using factors favoring an autoimmune etiology of seizures (AES), as determined by multivariate logistic regression. The model was externally validated and evaluated using the Concordance (C) statistic. Results: We included 582 patients, with median epilepsy duration of 8 years (interquartile range = 2–18). Twenty patients (3.4%) had AES, of whom 3 had anti–leucine-rich glioma inactivated 1, 3 had anti–contactin-associated protein-like 2, 1 had anti–N-methyl-D-aspartate receptor, and 13 had anti–glutamic acid decarboxylase 65 (enzyme-linked immunosorbent assay concentrations >10,000IU/ml). Risk factors for AES were temporal magnetic resonance imaging hyperintensities (odds ratio [OR] = 255.3, 95% confidence interval [CI] = 19.6–3332.2, p < 0.0001), autoimmune diseases (OR = 13.31, 95% CI = 3.1–56.6, p = 0.0005), behavioral changes (OR 12.3, 95% CI = 3.2–49.9, p = 0.0003), autonomic symptoms (OR = 13.3, 95% CI = 3.1–56.6, p = 0.0005), cognitive symptoms (OR = 30.6, 95% CI = 2.4–382.7, p = 0.009), and speech problems (OR = 9.6, 95% CI = 2.0–46.7, p = 0.005). The internally validated C statistic was 0.95, and 0.92 in the validation cohort (n = 128). Assigning each factor 1 point, an antibodies contributing to focal epilepsy signs and symptoms (ACES) score ≥ 2 had a sensitivity of 100% to detect AES, and a specificity of 84.9%. Interpretation: Specific signs point toward AES in focal epilepsy of unknown etiology. The ACES score (cutoff ≥ 2) is useful to select patients requiring antibody testing. ANN NEUROL 2021;89:698–710 View this article online at wileyonlinelibrary.com. DOI: 10.1002/ana.26013 Received Sep 25, 2020, and in revised form Jan 3, 2021. Accepted for publication Jan 4, 2021. Address correspondence to Dr Titulaer, Dr Molewaterplein 40, Room NF 321, 3015 GD Rotterdam, the Netherlands. E-mail: m.titulaer@erasmusmc.nl † A.E.M.B. and H.M. contributed equally. ‡ A.v.S. currently works at Haaglanden Medical Center, the Hague. § Members of the ACES Study Group and their institutional affiliations are listed in Table S6. From the 1Department of Neurology, Erasmus MC University Medical Center, Rotterdam, the Netherlands; 2Department of Neurology, Second Faculty of Medicine, Charles University, Motol University Hospital, Prague, Czech Republic; 3Department of Neurology, Stichting Epilepsie Instellingen Nederland, Heemstede, the Netherlands; 4Department of Neurology, Leiden University Medical Center, Leiden, the Netherlands; 5Department of Neurology, Academic Center for Epileptology, Epilepsy Center Kempenhaeghe, Heeze, the Netherlands; 6Department of Psychiatry and Neuropsychology, School for Mental Health and Neuroscience, Maastricht University Medical Center+, Maastricht, the Netherlands; 7Faculty of Health, Medicine, and Life Sciences, School of Health Professions Education, Maastricht University, Maastricht, the Netherlands; 8Department of Neurology, Maastricht University Medical Center+, Maastricht, the Netherlands; 9August Pi i Sunyer Biomedical Research Institute, Barcelona, Spain; and 10Department of Immunology, Erasmus MC University Medical Center, Rotterdam, the Netherlands Additional supporting information can be found in the online version of this article. 698 © 2021 The Authors. Annals of Neurology published by Wiley Periodicals LLC on behalf of American Neurological Association. This is an open access article under the terms of the Creative Commons Attribution-NonCommercial-NoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is non-commercial and no modifications or adaptations are made. de Bruijn et al: Antibody Screening in Epilepsy A utoimmune encephalitis (AIE) associated with neuro- nal antibodies is a severe but treatable neurological disease. Seizures occur frequently in patients with AIE from dedicated epilepsy centers in academic hospitals, and one general hospital with an epilepsy referral center. We requested all epileptologists (1) to include patients with (50–95%), often in combination with other symptoms, epilepsy with or without additional symptoms, but with- such as cognitive symptoms, behavioral changes, and auto- out suspicion of encephalitis; and (2) to exclude patients nomic dysfunction.1–4 Seizures are often resistant to anti- strongly suspected of having AIE. Patients with epilepsy seizure medication (ASM), whereas the response to with known infectious, genetic, or metabolic etiologies immunotherapy is good.5,6 Most patients have fulminant were excluded. Patients with a structural lesion on brain encephalitis with prominent seizures. magnetic resonance imaging (MRI) at inclusion were Neuronal antibodies have also been reported in excluded, whereas patients with mesial temporal sclerosis, patients with epilepsy (14–31%).7–9 Results from these or with T2/fluid-attenuated inversion recovery (FLAIR) studies have ensured that patients with less rapidly pro- hyperintensities mainly in the mesial temporal lobe, both gressive encephalitis are being recognized as well. Never- unilateral and bilateral, were not, as these MRI findings theless, in most of these studies, patients had short might be associated with AIE.12 epilepsy duration, and most of them had signs and symp- International League against Epilepsy guidelines toms of encephalitis. Interestingly, some of the mentioned were used to define focal seizures, an unknown etiology of studies report patients with epilepsy without fulminant epilepsy, and drug-resistant epilepsy.13 The study was reg- encephalitis or even any sign of encephalitis. To compli- istered at ClinicalTrials.gov (NCT02802475). cate interpretation, some of these studies describe a variety Patients were classified as (1) definite AES if a of antibodies, some pathogenic, but others with question- known neuronal antibody was detected in serum and/or able clinical relevance.10,11 cerebrospinal fluid (CSF), and if results were confirmed An important category comprises neuronal with a different laboratory test; (2) probable AES if a antibody–positive epilepsy patients without other enceph- known neuronal antibody was detected in serum and/or alitis signs, because underdiagnosis is likely. It is essential CSF, but if we were unable to confirm the finding, or to recognize these patients early and to perform antibody when the diagnostic criteria for seronegative AIE were testing in preselected patients. At the same time, testing met14; (3) possible AES, if serum or CSF showed similar needs to be rigorous, confirming results using different staining patterns on immunohistochemistry (IHC), but tests, to avoid false positives or clinically irrelevant results. without known antibody; and (4) non-AES. Similarly, it is important to limit the number of patients who require testing, for the sake of specificity and cost- Procedures effectiveness. The including epileptologist or the coordinating investigator The aim of our prospective, multicenter study was (M.A.A.M.d.B.) completed a case record form at enroll- to identify neuronal antibodies in a comprehensive cohort ment, containing information about patient characteristics, of patients with focal epilepsy of unknown etiology, and epilepsy characteristics, presence of autoimmune-associated without, or with unrecognized, signs of encephalitis. We clinical signs or symptoms (disorders of memory, sleep, have developed a clinical score, based on the prospectively behavior, speech, movement, or the autonomic system), collected data of patients with focal epilepsy of unknown previous and current medication (including ASM), current etiology, that can be used to guide autoimmune etiology level of functioning (measured with modified Rankin Scale of seizures (AES) screening. This antibodies contributing [mRS]15), and results from prior ancillary testing (including to focal epilepsy signs and symptoms (ACES) score has MRI, electroencephalogram [EEG], and if available CSF subsequently been validated in a second, external cohort. analysis). Blood was drawn, and if CSF analysis was per- formed, this was collected as well. Patients were prospectively followed for 1 year (last Patients and Methods follow-up in December 2018). At 1, 4, 8, and 12 months Study Design, Participants, and Definitions after inclusion, the treating physician or coordinating In this prospective, multicenter, observational cohort investigator collected data about seizure type, monthly sei- study, adults with focal epilepsy of unknown etiology were zure frequency, types of ASM used, and mRS. Final diag- included by epileptologists between December 2014 and nosis was obtained at last follow-up. December 2017. Patients included in this study had been referred to (tertiary) epilepsy centers by neurologists who Laboratory Methods had no particular suspicion of AIE. Patients were included All samples were screened for immunoreactivity with in the Netherlands, from tertiary epilepsy centers and IHC, as previously described.16 Subsequently, all samples April 2021 699 ANNALS of Neurology with a positive or questionable IHC result were tested in-between analysis using the Fisher exact test and Mann– more extensively. A combination of laboratory techniques Whitney U test was used to visualize group differences. was used, depending on the staining pattern and the clinical The patients with probable or possible AES were phenotype. Commercial cell-based assay (CBA; not included in the analysis, and are only described on an Euroimmun, Lübeck, Germany) was used to detect anti– exploratory basis, as their etiology could not be confirmed N-methyl-D-aspartate receptor (anti-NMDAR), anti–α- with certainty. We repeated analyses assigning these amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid recep- patients to the non-AES group to assess the effects of tor (anti-AMPAR), anti–γ-aminobutyric acid B receptor exclusion. (anti-GABABR), anti–leucine-rich glioma-inactivated 1 A multivariate logistic regression model was used to (anti-LGI1), anti–contactin-associated protein- like 2 (anti- identify independent risk factors pointing toward AES CASPR2), and anti–glutamic acid decarboxylase 65 (anti- (comparing both extracellular-AES and GAD-AES to non- GAD65). Anti-GAD65 was tested with enzyme-linked AES). Antibody status was used as dependent variable. immunosorbent assay (ELISA; Medizym anti-GAD 96, Missing values (0.8% of the total number of values) were Medipan, Berlin, Germany) as well, to quantify antibody imputed. Multiple imputation was used to create 5 sample concentration. An in-house CBA with live cells was used to sets, with antibody status included as covariate. We detect anti-GABAAR, anti-GABABR, anti-AMPAR, and excluded the factor “family history of autoimmune anti–dipeptidyl-peptidase–like protein 6. Radioimmunoassay diseases,” as we identified overt recall bias (lack of reliable (RIA; VGKC Antibody Assay RIA, DLD Diagnostika, information in many patients without AES). The univari- Hamburg, Germany) was used to detect voltage-gated ate regression and multivariate regression models were potassium channel complex antibodies (anti-VGKC), and constructed using the imputed data. Variables with a immunoblotting was used to detect onconeural antibodies p value < 0.007 in univariate analysis (corrected for multi- (anti-Hu/Yo/Ri/Tr/amphyphysin/CV2/Ma1/Ma2; ple testing) were included in the multivariate logistic Euroimmun). We considered serum anti-GAD65 relevant regression model. To derive the ACES score, variables if the ELISA concentration was >10,000IU/ml. Only sam- were eliminated using a backward likelihood ratio method ples tested positive by ELISA and with a compatible posi- with cutoff p value of 0.05. The selected risk factors were tive staining on IHC were considered positive.17 In used on the original and imputed data to check the accu- addition, all serum samples were screened for the presence racy. Each risk factor was assigned 1 point to create the of anti–glycine receptor (anti-GlyR) with an in-house CBA ACES score (range = 0–6 points). Sensitivity, specificity, with live cells, as these antibodies cannot be visualized with positive predictive value (PPV), and negative predictive IHC.18,19 All positive results were confirmed in separate value (NPV) were calculated using the nonimputed experiments. dataset. Percentages were visualized with 95% confidence Samples scored questionable or positive on IHC, but interval (CI) assuming a Poisson distribution. without a known antibody, were tested more extensively The systematic difference in direction between the using immunocytochemistry with live hippocampal cell Antibody Prevalence in Epilepsy and Encephalopathy cultures.16 (APE), APE2 and ACES scores were compared using A subgroup of patients considered for epilepsy sur- McNemar paired test. Internal validation with boo- gery were side by side tested by CBA, ELISA, and RIA for tstrapping was used to estimate the degree of optimism in antibodies, as a part of a standardized protocol. In addi- the ACES score. In addition, Firth logistic regression was tion, patients with an ACES score of 2 or more were used. We performed a sensitivity analysis to assess the tested by commercial CBA and ELISA post hoc. validity of our model allocating the excluded patients with probable or possible AES to the non-AES group (to be Statistical Analysis conservative). Dutch Cohort Patients were divided into 3 groups, based on antibody Czech Cohort test results: (1) antibodies targeting extracellular neuronal External validation was performed in a Czech cohort con- antigens, (2) high-concentration anti-GAD65, and (3) sisting of 128 temporal lobe epilepsy patients.20 In the non-AES. Comparisons between the 3 groups (extracellu- validation cohort, samples were tested by commercial lar-AES, GAD-AES, and non-AES) were performed with CBAs for anti-NMDAR, anti-AMPAR, anti-GABABR, the Fisher–Freeman–Halton test, the Pearson Chi-square anti-CASPR2, and anti-LGI1 (Euroimmun) and immuno- test, or the Kruskal–Wallis test, when appropriate. To cor- blot (PNS 11 Line Assay, Ravo Diagnostika, Freiburg, rect for multiple testing, we only considered p values Germany), confirmed by RIA for anti-GAD65 (CentAK < 0.002 to be relevant (Bonferroni). Exploratory, post hoc anti-GAD65 M, Medipan GMBH, Berlin, Germany). All 700 Volume 89, No. 4 de Bruijn et al: Antibody Screening in Epilepsy positive and questionable samples had been tested by IHC for less than 1 year at inclusion, 6 had antibodies (9%; see (F.L.). For uniformity, all questionable and positive sam- Fig 1). ples were additionally tested by CBA, ELISA, and IHC in Three patients of all 582 patients were scored as Rotterdam (M.J.T.). All 6 risk factors used in the ACES probable autoimmune. One patient, with a neuropil score had been collected prospectively in the Czech staining pattern on IHC, fulfilled the criteria of seronega- cohort, and were retrieved from the database to calculate tive AIE,14 whereas in serum of the other 2 patients anti- the ACES score. GlyR were detected using CBA. Nine patients had a posi- The model performance in this validation cohort tive IHC, but no known antibody was identified, nor was was described by discriminatory value, using the Concor- immunocytochemistry positive in any of these samples; dance (C) statistic (area under the curve of the receiver these patients were considered possible autoimmune. operating curve) on the ACES model. Statistical analysis Serum and CSF of 3 of these 9 patients showed a similar was performed using SPSS Statistics 21 (IBM, Armonk, “LGI1-like” staining pattern on IHC. In the other NY), R statistical software version 3.6.2 (rms library), and 6 patients, IHC showed a diffuse neuropil staining pat- Prism 8.0.1 (GraphPad Software, San Diego, CA) for tern. No additional antibodies were detected in the Windows. patients with questionable IHC staining patterns and neg- ative live neurons (n = 36). In addition, 186 IHC negative Ethics Committee Approval samples were tested by commercial antibody testing, partly The institutional review board of the Erasmus MC Uni- in parallel, and often as part of screening for epilepsy sur- versity Medical Center approved the study protocol gery (n = 104). The other samples were tested because of (MEC-2014-463). Written informed consent was high ACES scores (ACES ≥2, n = 82; post hoc testing). obtained from all patients. No antibodies were found in these samples. Results Comparison between AES and Non-AES Patients: Dutch Cohort Comparison of patients with antibodies against extracellu- Informed consent was obtained in 618 patients, of whom lar antigens (LGI1, CASPR2, NMDAR; n = 7), patients 36 (6%) were excluded (Fig 1). Of 582 included patients with anti-GAD65 (n = 13), and non-AES patients with focal epilepsy of unknown etiology, 48% were male. (n = 550; Table S1), showed that patients with AES (anti- The median age at inclusion was 44 years (interquartile bodies against both extracellular antigens and GAD65) range [IQR] = 29–58, range = 18–89). Ten percent of all more frequently had drug-resistant epilepsy (p = 0.002) included patients had a history of autoimmune diseases. and uni- or bilateral T2/FLAIR hyperintensities of the Median epilepsy duration was 8 years (IQR = 2–18, mesial temporal lobe (p < 0.0001), had higher mRS at range = 0.1–75), whereas 66 patients (11%) had an epi- inclusion (p < 0.0001), and more often had cognitive lepsy duration of <1 year. Patients were treated with a symptoms (p < 0.0001) and behavioral changes median of 1 ASM at inclusion (IQR = 1–2, range = 0–5). (p < 0.0001; Fig 2). Post hoc in-between analysis showed MRI was normal in 532 patients (91%), whereas that patients with antibodies against extracellular antigens T2/FLAIR hyperintensities of the mesial temporal lobe were older at disease onset, were more frequently male, were observed in 14 patients (2%), and 36 patients (6%) had shorter epilepsy durations, and more frequently had had mesial temporal sclerosis. The EEG showed epileptic autonomic symptoms than patients with anti-GAD65 or discharges in 389 patients (67%). In 362 of these non-AES patients. Patients with high-concentration anti- 389 patients, EEG reports were available. In 23% GAD65 were all female and more often had other autoim- (n = 83), seizures had a multifocal onset, whereas in 77% mune diseases, speech problems (word finding difficulties, (n = 279) there was a focal seizure onset, including tempo- non-fluent speech), and muscle stiffness than patients with ral (39%, n = 108), frontotemporal (39%, n = 110), and antibodies against extracellular antigens or non-AES extratemporal (20%, n = 57) localization. patients. Results of Laboratory Studies ACES Score Serum was collected from all patients, and CSF was col- Eight variables showed a statistically significant associa- lected from 46 patients (8%). In 20 of all 582 patients tion with antibody status in univariate analysis: drug- (3.4%), neuronal antibodies related to AES were detected resistant epilepsy, autoimmune diseases, behavioral (definite AES), including anti-LGI1 (n = 3), anti-CASPR2 changes, cognitive symptoms, muscle stiffness, speech (n = 3), anti-NMDAR (n = 1), and high-concentration problems, autonomic symptoms, and MRI hyper- anti-GAD65 (n = 13). Of the 66 patients with epilepsy intensities of the mesial temporal lobe (Table 1 and April 2021 701 ANNALS of Neurology FIGURE 1: Flowchart with follow-up diagnosis of all included patients. *This patient had focal onset epilepsy with sporadically occurring focal to bilateral tonic–clonic seizures, and had a recently discovered esophagus carcinoma. Eleven months after inclusion, his seizure frequency increased, and he developed psychotic symptoms. Cerebrospinal fluid was positive for anti–N- methyl-D-aspartate receptor (NMDAR), confirmed with live hippocampal neurons. #According to the criteria.14 AES = autoimmune etiology of seizures; CASPR2 = contactin-associated protein-like 2; GAD65 = glutamic acid decarboxylase 65; GlyR = glycine receptor; LGI1 = leucine-rich glioma-inactivated 1; MRI = magnetic resonance imaging. Fig 2). The multivariate model included the following remained significant and independent risk factors (data factors: autoimmune diseases, behavioral changes, cogni- not shown). tive symptoms, speech problems, autonomic symptoms, The external validation cohort consisted of 128 tem- and MRI hyperintensities of the mesial temporal lobe poral lobe epilepsy patients, of whom 7 had neuronal anti- (see Table 1). The internally validated C statistic was bodies (5.5%, compared to 3.4% in the Dutch cohort, 0.95. To correct for optimism, penalized modeling was p = 0.30). Data of the Dutch and Czech cohort are performed showing similar odds ratios for all factors; all compared in Table S2, and results of the Czech 702 Volume 89, No. 4 de Bruijn et al: Antibody Screening in Epilepsy FIGURE 2: Overview of the characteristics, signs, and symptoms that occurred more often in patients with neuronal antibodies. *p between 0.05 and 0.005, **p between 0.005 and 0.0001, ***p < 0.0001. The top p values correspond to the values visualized in Table S1. The lighter-colored lines visualize significance of the post hoc in-between analysis. (A) Factors that differed significantly between autoimmune etiology (AE) of seizures and non-AE. (B) Values that differed between the extracellular antigen group and the non-AE group. (C) Values that differed between the GAD65 and non-AE groups. In order of appearance, the y-axis shows: (1) years (age at onset), (2) cumulative percentage (all bar diagrams), (3) months and years (0–12, 1–80 respectively; epilepsy duration). In the bar diagrams, the x-axis shows the raw values. CASPR2 = contactin-associated protein-like 2; GAD65 = glutamic acid decarboxylase 65; LGI1 = leucine-rich glioma-inactivated 1; MRI = magnetic resonance imaging; mRS = modified Rankin Scale; NMDAR = N-methyl-D-aspartate receptor. [Color figure can be viewed at www.annalsofneurology.org] April 2021 703 ANNALS of Neurology TABLE 1. Results from Univariate and Multivariate Analysis Univariate Analysis Multivariate Analysis Odds Ratio (95% CI)a pa Adjusted Odds Ratio (95% CI)a pa Drug-resistant epilepsy 7.1 (2.1–24.5) 0.002 — — Autoimmune diseases 8.2 (3.2–20.7) <0.0001 13.3 (3.1–56.6) 0.0005 Behavioral changes 11.0 (4.1–29.3) <0.0001 12.6 (3.2–49.9) 0.0003 Cognitive symptoms 31.8 (4.4–229.9) 0.001 30.6 (2.4–382.7) 0.009 Muscle stiffness 7.3 (1.9–28.7) 0.004 — — Speech problems 12.8 (4.1–40.0) <0.0001 9.6 (2.0–46.7) 0.005 Autonomic symptoms 22.1 (6.4–75.5) <0.0001 23.3 (3.8–143.3) 0.001 Temporal MRI hyperintensities 16.9 (4.6–62.1) <0.0001 255.3 (19.6–3,332.2) <0.0001 The table only shows the data of patients with autoimmune etiology of seizures (n = 20), and with no evidence for autoimmunity (n = 550). a Numbers shown are from the cohort data after imputation. CI = confidence interval; MRI = magnetic resonance imaging. TABLE 2. Testing Properties of the ACES Score Using Different Cutoff Values Cutoffa n (%)b PPV (95% CI) NPV (95% CI) Sensitivity (95% CI) Specificity (95% CI) ≥1 point 305 (53%) 6.5 (2.8–13.1) 100 (81.4–100) 100 (81.4–100) 48.2 (36.5–63.6) ≥2 point 103 (18%) 19.4 (12.2–29.7) 100 (81.4–100) 100 (81.4–100) 84.9 (67.9–100) ≥3 point 22 (4%) 54.5 (41.4–70.5) 98.5 (80.5–100) 60 (45.8–77.2) 98.2 (80.5–100) ≥4 point 1 (0.2%) 100 (81.4–100) 96.7 (78.7–100) 5 (1.6–11.7) 100 (81.4–100) a The six factors include: autoimmune diseases, behavioral changes, cognitive symptoms, speech problems, autonomic symptoms, temporal magnetic resonance imaging hyperintensities. b N = 570. CI = confidence interval; NPV = negative predictive value; PPV = positive predictive value. antibody-positive patients are shown in Table S3. The CI = 67.9–100; Table 2). No patient with an ACES scor- C statistic of our model for the external validation cohort e ≥ 2, but without AES (n = 83), was identified by com- was 0.92, showing that the (overall) discriminative perfor- mercial tests, outside the planned screening algorithm (see mance is similar in the validation cohort. Fig 3B). In addition, we reviewed data of 55 Dutch anti- The 6 independent risk factors were used to create NMDAR, 43 anti-LGI1, 19 anti-GAD65, and 14 anti- the ACES score, each factor assigning 1 point. Evaluating CASPR2 patients with seizures. All of them, except different cutoff values for the ACES score (Fig 3) in the 1 patient with anti-LGI1 (ACES score = 1), had an ACES original cohort, 19.4% (95% CI = 11.4–29.6) of the score of 2 or more. Using the same cutoff for the ACES patients with a score ≥ 2 had antibodies (PPV), whereas score, 37 patients in the Czech validation cohort had an none of the patients with < 2 risk factors had antibodies, ACES score ≥ 2 (29%). All patients with AES were identi- resulting in an NPV of 100% (95% CI = 81.4–100). Sen- fied (see Fig 3C).20 sitivity of an ACES score of ≥ 2 was 100% (95% Results from the sensitivity analysis (allocating the CI = 81.4–100), and specificity was 84.9% (95% 12 patients with probable or possible AES [3 and 704 Volume 89, No. 4 de Bruijn et al: Antibody Screening in Epilepsy scored better in patients with chronic epilepsy (p = 0.0015; Fig 4). Clinical Characteristics of the Dutch Cohort Looking very carefully, additional subtle signs or symp- toms of encephalitis/encephalopathy were present at inclu- sion in 19 of 20 patients with AES. All patients with anti- LGI1 had subtle characteristic signs, including cognitive symptoms, insomnia, hyponatremia, and behavioral changes, and 2 of 3 patients had faciobrachial dystonic sei- zures (FBDS).22 Two of 3 patients with anti-CASPR2 had other core signs or symptoms,23 albeit mild, including cognitive symptoms, peripheral nerve hyperexcitability, insomnia, autonomic symptoms, and cerebellar symptoms, and 1 patient developed neuropathic pain after inclusion. In both the anti-LGI1 and anti-CASPR2 patients, some signs were identified in retrospect (insomnia, FBDS, peripheral nerve hyperexcitability, cerebellar symptoms), whereas others were observed but remained unrecognized as AES-related (because of their subtleness, and concomi- tant start of ASM). During follow-up, 1 patient with focal epilepsy with sporadically occurring tonic–clonic seizures and recently diagnosed esophagus carcinoma developed hallucinations, refractory epilepsy, and psychosis. Anti- NMDAR was detected in his CSF 11 months after inclu- sion, whereas his serum was negative at baseline, but also at 11 months after inclusion. All 13 women with anti-GAD65 had a serum con- centration > 10,000 IU/ml (median serum concentration = 466,000 IU/ml, median CSF concentration = 4,600IU/ml). Five anti-GAD65 patients had diabetes mellitus type 1 FIGURE 3: Autoimmune etiology of seizures (AES) patients as (38%). All patients had focal onset seizures, most frequently distributed by ACES score. The number of AES patients by ACES score (A) are provided for both the original, Dutch with sensory onset (déjà vu episodes; 9/13). Interestingly, in cohort (B) and the Czech validation cohort (C). Only the data 4 patients, temporal seizures were clearly provoked by of the patients with antibodies targeting extracellular music. Oligoclonal bands unique to the CSF were tested in antigens, anti–glutamic acid decarboxylase 65 (GAD65), and 8 patients, of whom 5 tested positive. no evidence for autoimmunity are shown. The numbers in the bar diagrams correspond with the numbers of patients of In the “probable autoimmune” group, 1 patient each specific group. All patients with an ACES score of ≥2 developed frequent seizures. Shortly afterward, he suffered without AES were tested by commercial cell-based assay and from abdominal symptoms and weight loss, which turned enzyme-linked immunosorbent assay post hoc, and all were negative. MRI = magnetic resonance imaging. [Color figure out to be Crohn’s disease. He then continued to deterio- can be viewed at www.annalsofneurology.org] rate, with progressive cognitive symptoms. He was treated with steroids and adalimumab, leading to (slow) cognitive improvement and cessation of seizures. He met the “sero- 9 patients, respectively] to the non-AES group) showed negative AIE” criteria.14 The 2 patients with anti-GlyR comparable odds ratios (a sensitivity of 100% and specific- were a middle-aged male with drug-resistant epilepsy and ity of 84.7%). a young woman with sporadic seizures without other Comparing the ACES score (of both cohorts) to signs. the prior published APE/APE2 score,8,21 showed that In the “possible autoimmune” group, all 3 women the APE/APE2 score detected 8 of 10 patients with anti- with a similar “LGI1-like” pattern on IHC had daily bodies against extracellular proteins, and 7 of 17 patients drug-resistant focal onset seizures. In 2 of these 3 women, with anti-GAD65. All scores performed well in patients ASMs were withdrawn because of ineffectiveness. The 9 with a short history of epilepsy, but the ACES score patients with neuropil staining on IHC had variable April 2021 705 ANNALS of Neurology FIGURE 4: Comparison of the ACES score with the Antibody Prevalence in Epilepsy and Encephalopathy (APE) and APE2 score. (A) Visualization of the epilepsy scores per patient, combining the Dutch and Czech cohorts, shows that all patients are identified by the ACES score, whereas a considerable percentage is not identified by the APE or APE2 score. (B) Dissection by duration of epilepsy shows that the difference in performance is caused by the patients with epilepsy for >1 year. * p < 0.05, **p < 0.01. The colors refer to the antibodies identified: red, contactin-associated protein-like 2; orange, leucine-rich glioma-inactivated 1; green, N-methyl-D-aspartate receptor; blue, glutamic acid decarboxylase 65. clinical phenotypes. All clinical characteristics of the became seizure-free at 13 months, just outside the study patients with definite, probable, and possible AES are period), as did the anti-NMDAR encephalitis patient. Ten shown in Tables S4 and S5. of the 13 anti-GAD65 patients were treated with immu- In 550 patients, there was no evidence for autoim- notherapy; 5 of them had a remarkable decrease in their munity at last follow-up. Of these 550 patients, 96% were seizure frequency (see Fig 5C; 50–95%), deviating from diagnosed as focal epilepsy with unknown etiology, their seizure frequency the year before, whereas in the whereas in the other patients ancillary testing led to other 5 patients no seizure frequency reduction was another diagnosis, including a symptomatic lesion on observed (see Fig 5D). Three patients refused immuno- revised or new MRI (n = 13), generalized epilepsy (n = 3), therapy (reasons: less severe disease [n = 2] or fear of side dementia and seizures (n = 2), and no epilepsy (n = 2). effects [n = 1]). Immunotherapy Responses Discussion All 6 patients with anti-LGI1 and anti-CASPR2 encepha- In this prospective, multicenter cohort study, we show litis were treated with immunotherapy (Fig 5A, B) and that a small, but relevant, proportion of patients with focal became seizure-free (1 anti-LGI1 encephalitis patient epilepsy of presumed unknown etiology have neuronal 706 Volume 89, No. 4 de Bruijn et al: Antibody Screening in Epilepsy FIGURE 5: Individual treatment responses of patients with anti–leucine-rich glioma-inactivated 1 (LGI1), anti–contactin-associated protein-like 2 (CASPR2), and anti–glutamic acid decarboxylase 65 (GAD65). The figures visualize the seizure frequency over time per antibody: (A) anti-LGI1, (B) anti-CASPR2, (C) anti-GAD65 with treatment response, (D) anti-GAD65 without treatment response (both black), or untreated (gray). The Roman numerals correspond to the numbers shown in Table S4. All patients were treated with antiseizure medications (ASM). The figure only shows ASM changes that led to a decrease in seizure frequency. AED = antiepileptic drug; IVIg = intravenous immunoglobulin; −T12 = 12 months before inclusion, T0 = inclusion date, T1 = 1 month after inclusion, T4 = 4 months after inclusion, T8 = 8 months after inclusion, T12 = 12 months after inclusion. [Color figure can be viewed at www.annalsofneurology.org] antibodies. All patients with AES had unrecognized signs be side effects of ASM. In addition, patients were only of encephalitis. Identifying these patients is crucial, included in the AES group when substantial evidence for because seizures respond better to immunotherapy than to autoimmunity for the detected antibodies was available. In ASM, which is reflected by seizure freedom in all patients a smaller prospective cohort of patients with temporal lobe with antibodies against extracellular proteins after immu- epilepsy, the same antibodies were identified as in the notherapy. For recognition of AES patients, we have pro- ACES study,20 whereas others found a greater variability vided and validated a simple clinical score that helps of antibodies.7,8 Some of the described antibodies are physicians in selecting patients who should be screened pathogenic, like anti-NMDAR, and anti-GABABR, but for neuronal antibodies. occurred in highly selected patients with clear encephalitis. In the current study, neuronal antibodies were found However, experiments to confirm antibody results were in 3.4% of patients, whereas other studies describe higher lacking in some studies, although it is an essential neces- numbers of >10%.7,8 The most important issue explaining sity when screening large cohorts, to avoid false positive this discrepancy is the difference in patient selection. The results. Prior studies also describe antibodies with a debat- patients included in our study were referred to epilepsy able role in neuroinflammation, including low- centers by neurologists who had no suspicion of AIE. In concentration anti-GAD65, and double negative anti- retrospect, most of the patients with AES had subtle signs VGKC, nowadays considered clinically irrelevant by or symptoms of AIE, but these signs were unrecognized as most.10,11 being related to AIE. For example, insomnia, muscle stiff- The Czech study20 was used to externally validate ness, or mild cognitive problems were often considered to the results. This study was chosen because it was of a April 2021 707 ANNALS of Neurology considerable size, investigated chronic epilepsy as well testing should be expanded, taking into account the clini- (median duration even longer), and did not included cal phenotype. patients referred for overt encephalitis. In contrast to our We identified 6 independent risk factors pointing study, all patients had temporal epilepsy, and they more toward AES. The assessed risk factors were predefined often had mesiotemporal sclerosis. Otherwise, the epide- before the study started. The ACES score was created miological and clinical characteristics were largely similar. assigning all factors 1 point instead of creating a value- Reassuringly, the antibody frequency and type of anti- weighted score. The easier the score, the more likely it is bodies (anti-LGI1, anti-CASPR2, anti-GAD65) were to be used in clinical settings. In addition, as our score is within the same range. meant to identify those who need antibody testing, our We found anti-LGI1, anti-CASPR2, and anti- primary aim was to create a sensitive score. Although a GAD65. This is a comprehensible observation, because weighted score would provide better model performance, the clinical phenotypes of patients with these syndromes it would not increase the sensitivity (already optimal if at can be more prolonged and less aggressive, and diagnosis least 2 factors of the ACES score were present). Lastly, is more likely to be delayed or missed than in patients some of the odds ratios showed large confidence intervals with clear encephalitis.1,12,22 Although clinical phenotypes due to the low frequency in the control group of the origi- related to these antibodies can be less serious, immuno- nal cohort. By not assigning a weighted value, we lower therapy is superior to ASM,5,17 which makes diagnosis in the risk of overfitting. Using a cutoff value of 2 for the an early phase important. Thirteen women had high anti- ACES score, all antibody-positive patients were identified, GAD65 concentrations. Earlier performed laboratory but antibody testing was unrevealing in 14%. It should studies were unable to reveal the pathogenicity of anti- therefore not be used to diagnose AES, but to guide selec- GAD65.23 However, patients with high concentrations tion of patients for antibody screening. have overlapping and well-defined neurological syn- The factors included in the ACES score partially dromes, and comparable seizure characteristics, not overlap with those used in the APE/APE2 score.8,21 Nev- observed in low-concentration patients.17 Almost 75% of ertheless, there are important differences, which makes the high-concentration anti-GAD65 women had déjà vu sense as, judging by the published papers of Dubey et al, episodes. In addition, one-quarter had musicogenic epi- the scores serve different populations. Our study included lepsy, meaning temporal seizures clearly triggered by spe- patients with focal epilepsy, all without overt encephalitis. cific music or songs, whereas the prevalence of In addition, our ACES score was validated in another musicogenic epilepsy in other non-AES patients is very chronic epilepsy cohort, and showed very good discrimina- low. Musicogenic epilepsy has been described before in tion. In this study, we confirm that the APE/APE2 score anti-GAD65 encephalopathy.24 Concerning immunother- is useful to detect patients in a (sub)acute setting with apy responses, in contrast to the patients with anti-LGI1 antibodies against extracellular neuronal antigens. How- or anti-CASPR2, no patient became seizure-free. How- ever, almost 70% of chronic patients would have been ever, in one-half of the treated patients, a long-term epi- missed using this score, underlining the importance of a lepsy frequency reduction of up to 95% was observed, score that is applicable to patients with variable disease although these patients had been refractory to ASMs for a courses. significant time. This makes regression to the mean or Some patients were classified as “probable or possi- natural history as the explanation for the epilepsy fre- ble autoimmune.” Data for these patients were not used quency reduction highly unlikely. In support, seizure in the main analysis, but inclusion would not have chan- response was accompanied by a serological response. ged the results. Two patients categorized as “probable The finding that only anti-LGI1, anti-CASPR2, and autoimmune” had anti-GlyR. Clinical characteristics of anti-GAD65 occurred suggests that screening for these these 2 patients showed no notable similarities. Testing antibodies only is generally sufficient in patients with focal serum of these patients consistently showed a positive epilepsy of unknown etiology without clear encephalitis. If result on live CBA, but we were unable to analyze CSF or anti-LGI1 is detected in serum and the phenotype fits, to confirm our findings with additional techniques. Except diagnosis can be made, and treatment can be started.12 If for these 2 epilepsy samples, we have not identified any serum is positive for anti-CASPR2, CSF should be tested positive sample in 1,206 patients outside published clini- as well.22 If the anti-GAD65 ELISA concentration is cal phenotypes.19,25 Although anti-GlyR can be directly >10,000IU/mL, this can be considered clinically rele- pathogenic,18 we prefer to be careful in describing vant.17 This is roughly comparable to an RIA titer of unconfirmed findings. The “possible autoimmune” cate- 20nmol/L8 or 2,000U/mL.24 In case of deterioration of gory contained 3 patients with similar clinical characteris- symptoms or suggestive signs or symptoms, antibody tics and drug-resistant epilepsy. Both serum and CSF 708 Volume 89, No. 4 de Bruijn et al: Antibody Screening in Epilepsy showed an unknown pattern on IHC. We are currently analyzing these samples using immunoprecipitation and Acknowledgments mass spectrometry. In 6 other patients, IHC showed a dif- This work was supported by the Dutch Epilepsy Founda- fuse neuropil staining However, live hippocampal neuron tion, project numbers 14-19 and 19-08. It is generated staining was negative, CSF was not available for confirma- within the European Reference Network for Rare Immu- tion, and patients had different clinical phenotypes, all nodeficiency, Autoinflammation, and Autoimmune Dis- questioning the clinical relevance. eases. H.M., and M. Elisak and P. Marusic of the ACES Our study has limitations. Due to the small number Study Group, were supported by the Charles University of antibody-positive patients, there is a risk of overfitting. project GA UK No 746120. Therefore, we also assessed a penalized model. In addition, We thank all patients participating in this study; the external validation in a second, independent, foreign epileptologists from Kempenhaeghe and Stichting Epi- cohort showed comparable test characteristics. Concerning lepsie Instellingen Nederland for their cooperation; and A. data extraction, this was performed thoroughly; however, Doets for her help with R. it is possible that recall bias occurred, especially in the var- iables that are not discussed in the standard intake. An Author Contributions example is “family history of autoimmune diseases.” We M.A.A.M.d.B., R.D.T., M.J.M.M., and M.J.T. therefore decided to exclude this variable from the analy- contributed to the conception and design of the study; all sis. Another limitation was the use of only serum in most authors contributed to acquisition and analysis of data; patients. Although highly sensitive to test for anti-LGI1,12 M.A.A.M.d.B., A.E.M.B., H.M., R.D.T., P.A.E.S.S., and anti-CASPR2,22 anti-GAD65, and anti-GABABR,26 it is M.J.T. contributed to drafting the text and preparing the somewhat less sensitive to screen for anti-NMDAR27 or figures. anti-AMPAR.28 Testing only serum poses the risk of miss- ing a small proportion of patients. However, both anti- Potential Conflicts of Interest NMDAR and anti-AMPAR encephalitis tend to present Nothing to report. more fulminantly,2 and the syndromes would probably have revealed themselves during follow-up, as exemplified by the only patient who was antibody negative (in serum) at inclusion, but who developed a panencephalitis References 1. Irani SR, Alexander S, Waters P, et al. Antibodies to Kv1 potassium 11 months after inclusion with anti-NMDAR in CSF. In channel-complex proteins leucine-rich, glioma inactivated 1 protein retrospect, it is unclear, because of the lack of CSF at and contactin-associated protein-2 in limbic encephalitis, Morvan’s syndrome and acquired neuromyotonia. Brain 2010;133:2734–2748. inclusion, whether this patient had anti-NMDAR, causing his seizures before deterioration, or if he coincidentally 2. Titulaer MJ, McCracken L, Gabilondo I, et al. Treatment and prog- nostic factors for long-term outcome in patients with anti-NMDA developed anti-NMDAR encephalitis later on. The pres- receptor encephalitis: an observational cohort study. Lancet Neurol ence of esophageal carcinoma in this case underlines the 2013;12:157–165. importance of screening for tumors in these patients.29 3. Lancaster E, Lai M, Peng X, et al. Antibodies to the GABA(B) recep- tor in limbic encephalitis with seizures: case series and characterisa- Samples were screened with immunohistochemistry, and tion of the antigen. Lancet Neurol 2010;9:67–76. questionable and positive samples were tested more thor- 4. Petit-Pedrol M, Armangue T, Peng X, et al. Encephalitis with refrac- oughly (CBAs, ELISA, live hippocampal neurons). This tory seizures, status epilepticus, and antibodies to the GABAA recep- approach was chosen because of the high sensitivity of tor: a case series, characterisation of the antigen, and analysis of the effects of antibodies. Lancet Neurol 2014;13:276–286. IHC.12,17,22,26–28,30,31 Although we cannot guarantee that 5. de Bruijn M, van Sonderen A, van Coevorden-Hameete MH, et al. we have identified all patients, previous publications sup- Evaluation of seizure treatment in anti-LGI1, anti-NMDAR, and anti- port our claim that this approach seems most reliable and GABABR encephalitis. Neurology 2019;92:e2185–e2196. sensitive to test large cohorts of patients with low prior 6. Thompson J, Bi M, Murchison AG, et al. The importance of early chances, as also enforced by testing 186 samples negative immunotherapy in patients with faciobrachial dystonic seizures. Brain 2018;141:348–356. by IHC with additional CBA and ELISA without reveal- 7. Brenner T, Sills GJ, Hart Y, et al. Prevalence of neurologic autoanti- ing additional antibodies. bodies in cohorts of patients with new and established epilepsy. To conclude, patients with AES present differently Epilepsia 2013;54:1028–1035. and have specific, although not always discriminative charac- 8. Dubey D, Alqallaf A, Hays R, et al. Neurological autoantibody preva- teristics. Recognition of these patients can be difficult, but is lence in epilepsy of unknown etiology. JAMA Neurol 2017;74: 397–402. important for treatment decisions. An ACES score ≥ 2 can 9. Malter MP, Helmstaedter C, Urbach H, et al. Antibodies to glutamic be used to identify patients with higher risk of having AES, acid decarboxylase define a form of limbic encephalitis. Ann Neurol and to select those who require antibody testing. 2010;67:470–478. April 2021 709 ANNALS of Neurology 10. van Sonderen A, Schreurs MW, de Bruijn MA, et al. The relevance of 21. Dubey D, Pittock SJ, McKeon A. Antibody prevalence in epilepsy VGKC positivity in the absence of LGI1 and Caspr2 antibodies. Neu- and encephalopathy score: increased specificity and applicability. rology 2016;86:1692–1699. Epilepsia 2019;60:367–369. 11. Lang B, Makuch M, Moloney T, et al. Intracellular and non-neuronal 22. van Sonderen A, Arino H, Petit-Pedrol M, et al. The clinical spectrum targets of voltage-gated potassium channel complex antibodies. of Caspr2 antibody-associated disease. Neurology 2016;87:521–528. J Neurol Neurosurg Psychiatry 2017;88:353–361. 23. Gresa-Arribas N, Arino H, Martinez-Hernandez E, et al. Antibodies to inhibitory synaptic proteins in neurological syndromes associated 12. van Sonderen A, Thijs RD, Coenders EC, et al. Anti-LGI1 encephali- with glutamic acid decarboxylase autoimmunity. PLoS One 2015;10: tis: clinical syndrome and long-term follow-up. Neurology 2016;87: e0121364. 1449–1456. 24. Falip M, Rodriguez-Bel L, Castaner S, et al. Musicogenic reflex sei- 13. Kwan P, Arzimanoglou A, Berg AT, et al. Definition of drug resistant zures in epilepsy with glutamic acid decarbocylase antibodies. Acta epilepsy: consensus proposal by the ad hoc task force of the ILAE Neurol Scand 2018;137:272–276. commission on therapeutic strategies. Epilepsia 2010;51:1069–1077. 25. Hutchinson M, Waters P, McHugh J, et al. Progressive encephalomy- 14. Graus F, Titulaer MJ, Balu R, et al. A clinical approach to diagnosis elitis, rigidity, and myoclonus: a novel glycine receptor antibody. of autoimmune encephalitis. Lancet Neurol 2016;15:391–404. Neurology 2008;71:1291–1292. 15. van Swieten JC, Koudstaal PJ, Visser MC, et al. Interobserver agree- 26. van Coevorden-Hameete MH, de Bruijn M, de Graaff E, et al. The ment for the assessment of handicap in stroke patients. Stroke 1988; expanded clinical spectrum of anti-GABABR encephalitis and added 19:604–607. value of KCTD16 autoantibodies. Brain 2019;142:1631–1643. 16. Ances BM, Vitaliani R, Taylor RA, et al. Treatment-responsive limbic 27. Gresa-Arribas N, Titulaer MJ, Torrents A, et al. Antibody titres at encephalitis identified by neuropil antibodies: MRI and PET corre- diagnosis and during follow-up of anti-NMDA receptor encephalitis: lates. Brain 2005;128:1764–1777. a retrospective study. Lancet Neurol 2014;13:167–177. 28. McCracken L, Zhang J, Greene M, et al. Improving the antibody- 17. Munoz-Lopetegi A, de Bruijn M, Boukhrissi S, et al. Neurologic syn- based evaluation of autoimmune encephalitis. Neurol Neuro- dromes related to anti-GAD65: clinical and serologic response to immunol Neuroinflamm 2017;4:e404. treatment. Neurol Neuroimmunol Neuroinflamm 2020;7:e696. 29. Titulaer MJ, Soffietti R, Dalmau J, et al. Screening for tumours in par- 18. Carvajal-Gonzalez A, Leite MI, Waters P, et al. Glycine receptor anti- aneoplastic syndromes: report of an EFNS task force. Eur J Neurol bodies in PERM and related syndromes: characteristics, clinical fea- 2011;18:19–e3. tures and outcomes. Brain 2014;137:2178–2192. 30. Muñoz-Lopetegi A, de Bruijn MAAM, Boukhrissi S, et al. Neurologi- 19. McKeon A, Martinez-Hernandez E, Lancaster E, et al. Glycine recep- cal syndromes related to anti-GAD65: clinical and serological tor autoimmune spectrum with stiff-man syndrome phenotype. response to treatment. Neurol Neuroimmunol Neuroinflamm 2020;7: JAMA Neurol 2013;70:44–50. e696. 20. Elisak M, Krysl D, Hanzalova J, et al. The prevalence of neural anti- 31. Hoftberger R, van Sonderen A, Leypoldt F, et al. Encephalitis and bodies in temporal lobe epilepsy and the clinical characteristics of AMPA receptor antibodies: novel findings in a case series of seropositive patients. Seizure 2018;63:1–6. 22 patients. Neurology 2015;84:2403–2412. 710 Volume 89, No. 4