The electroclinical phenotype of epilepsy following autoimmune encephalitis

Abstract

Objective

To describe the temporal and spatial dynamics of electroencephalographic (EEG) recordings in epilepsy following autoimmune encephalitis (AE) compared to temporal lobe epilepsy (TLE) without signs of inflammatory etiology.

Methods

We assessed electroencephalographic properties using continuous long-term (72-h) EEG recordings in combination with neuroimaging markers and cerebrospinal fluid (CSF) parameters of 25 patients with epilepsy following AE and 30 patients with TLE. We specifically investigated the presence of focal slowing (FS) and epileptiform discharges (EDs). Additionally, spectral EEG variability during wakefulness and sleep was analyzed using quantitative EEG analysis.

Results

Compared to TLE, patients with AE showed significantly more independent bitemporal FS and independent bitemporal EDs (p < 0.05), with a higher co-occurrence of both patterns (p < 0.05). Quantitative analysis of wake and sleep EEG revealed greater spectral variability in AE, particularly in the temporal regions. A composite score incorporating EEG data, MRI and CSF parameters, as well as clinical features, effectively distinguished AE from TLE (p < 0.0001). Using the same variables, principal component analysis also demonstrated clear separation between the two groups.

Significance

These results demonstrate distinct and quantifiable EEG characteristics in epilepsy following AE. The identified EEG patterns and the proposed score may serve as biomarkers for differentiating between epilepsy following AE and TLE of non-inflammatory etiology. This approach can aid clinicians in determining the need for further diagnostic testing or optimizing therapy decisions in the post-acute phase, where clinicians frequently encounter patients with new-onset or evolving epilepsy in whom a current or prior inflammatory etiology is suspected. In this scenario, a precise understanding of post-acute EEG characteristics may be clinically relevant to assess the probability of an underlying autoimmune/inflammatory process and to guide the immune-directed evaluation, antiseizure medication optimization, and epilepsy-surgical evaluation.

Highlights

• •Bilateral focal slowing and epileptiform discharges occur significantly more often in epilepsy following autoimmune encephalitis (AE) than in non-inflammatory temporal lobe epilepsy (TLE).
• •State-space EEG analysis shows markedly higher temporal lobe spectral variability in AE, indicating distinct underlying neural dynamics.
• •Mesiotemporal MRI abnormalities and CSF markers (pleocytosis, BBB dysfunction, OCBs) strongly support differentiation of AE from TLE.
• •A composite score integrating EEG, MRI, CSF, and clinical features discriminates AE from TLE with high specificity.

1. Introduction

Encephalitis, most often caused by viral infections, has an incidence ranging from 0.7 to 12.6 per 100,000 individuals [1]. Approximately one-third of cases are believed to have a non-infectious, autoimmune etiology. Autoimmune encephalitis (AE) arises from brain-specific immune responses and is typically associated with autoantibodies targeting neuronal or glial structures [2]. These autoantibodies may be directed against neuronal cell surface antigens (e.g., GABA receptor antibodies) or intracellular antigens (e.g., anti-GAD autoantibodies). However, up to 30% of AE cases lack an identifiable antibody despite strong evidence of an immune-mediated etiology [3], [4]. The most frequently detected autoantibodies include those against GABA-, LGI1-, and NMDA-receptors [5]. Autoantibodies targeting intracellular proteins are often linked to paraneoplastic syndromes, particularly anti-GAD and anti-Ta/Ma2 IgG [2].

Electrographic seizures occur in 50–70% of AE patients [6], [7], and these seizures may exacerbate localized inflammation, perpetuating disease activity [8]. Most autoantibodies against neuronal surface antigens induce neuronal hyperexcitability and disrupt synaptic function and plasticity [8], [9]. Consequently, AE with antibodies targeting neuronal surface proteins is strongly associated with seizures [10], [11]. Among these, AE with GABA receptor antibodies carries a higher risk of status epilepticus compared to AE with other antibodies such as anti-LGI1, anti-NMDAR, or anti-Caspr2 [12]. Seizures are also more frequent in younger than in older AE patients [7].

EEG abnormalities are observed in approximately 85% of AE cases in the acute phase [12]. The frequency of seizures and EEG abnormalities may vary according to the type of autoantibodies [7], [13]. This variability likely reflects differences in specific pathophysiology and tropism of the immune-mediated injury. EEG patterns termed «extreme delta brushes» have been reported for patients with anti-NMDAR- or anti-dipeptidyl-peptidase-like protein 6 (DPPX) encephalitis, but not for other AE [14], [15], [16], [17]. In anti-NMDAR encephalitis, 30% of patients show «extreme delta brushes», 35% focal slowing and 61% epileptiform discharges [15]. Anti-LGI1 AE was also shown to be associated with focal slowing [18]. We recently analyzed repetitive EEG recordings in the acute phase of AE and found spatial and temporal variability of focal slowing and epileptiform discharges as a consistent EEG feature [19].

However, EEG characteristics of autoimmune epilepsy (AE) in the post-acute phase remain insufficiently defined. Clinically, it is essential to consider AE as a differential diagnosis when new-onset temporal lobe epilepsy (TLE) presents in age groups that fall outside the typical demographic profile. In these situations, long-term EEG (LT-EEG) monitoring is frequently used. However, only a limited number of studies have examined electrophysiological markers capable of distinguishing post-acute AE-related TLE from non-inflammatory TLE. Although seizure patterns and semiology may provide electroclinical features suggestive of post-acute AE [12], [20], interictal LT-EEG findings often show substantial overlap between the two groups [21], [22], [23].

Here, we specifically investigate interictal findings in LT-EEG recordings combined with neuroimaging, cerebrospinal fluid (CSF) parameters, and clinical data as biomarkers for differentiating epilepsy following AE from TLE without signs of inflammatory etiology.

2. Methods

2.1. Patients' inclusion criteria

We systematically analyzed long-term EEG recordings in a cohort of 25 patients who developed epilepsy following autoimmune encephalitis (AE) and 30 patients with temporal lobe epilepsy (TLE) of non-inflammatory etiology. To focus on the post-acute epilepsy phenotype, acute symptomatic seizures occurring only during the encephalitic illness were not sufficient for inclusion. AE patients were included only when recurrent post-acute seizures (i.e. epilepsy following AE) was documented. As a main inclusion criterion for all patients (both the AE and TLE group), at least one seizure event must have been recorded as an electroclinical seizure, showing a clinical seizure time-locked to an ictal EEG pattern over the temporal region in video-EEG recording. If no seizure was detected in EEG, at least one event had to be documented based on the patient's self-reported history, witnessed seizure, or video recording where the seizure was characterized by an epigastric aura followed by oral and/or manual automatisms, as defined by Beniczky et al. [24]. These electroclinical seizures or the distinct semiology provides strong evidence for the diagnosis of temporal lobe epilepsy [24], [25].

All patients included in this study were at least 18 years old at the time of EEG recording and had a minimum of 72 h of EEG data available for analysis. All patients underwent a minimum of one cerebral MRI examination. In 92% (23/25) of AE patients and 70% (21/30) of TLE patient results of diagnostic lumbar puncture were available. In all cases, lumbar puncture was performed to evaluate a potential inflammatory etiology of epilepsy. CSF evaluation in the TLE group was obtained to exclude inflammatory etiology in a diagnostic context, for example in adult-onset or drug-resistant temporal seizures, non-lesional or otherwise unclear MRI findings, or during presurgical evaluation. The TLE group should therefore be interpreted as a clinically relevant differential-diagnostic control group, rather than as a cohort intended to represent the overall distribution of etiologies seen in temporal lobe epilepsy. We reviewed medical records of both the AE and TLE group and collected information on demographics, EEG data, seizure types, comorbidities, CSF findings, findings on neuroimaging (MRI), and treatment history (Table 1).

Table 1.

Patient characteristics. BBB, blood-brain barrier; CSF, cerebrospinal fluid; LP, lumbar puncture; OCBs, oligoclonal bands. Mean ± SD, p-values according to Mann-Whitney U test or Fisher's exact test. Significant differences in bold.

	AE	TLE	p-value
Number of Patients	25	30
A) Demographics
Female/male ratio	1.27	1	0.79
Age at onset (years)	44 ± 17	35 ± 18	0.13
Age at first seizure (years)	44 ± 17	35 ± 17	0.14
Age at EEG (years)	49 ± 17	50 ± 12	0.94
Age at LP (years)	50 ± 17	50 ± 12	0.83
Disease duration at EEG (months)	73 ± 69	190 ± 168	0.027
Disease duration at LP (months)	68 ± 72	182 ± 171	0.029
B) Types of seizures
First seizure - focal (%)	64.0%	73.3%	0.56
First seizure - tonic clonic (%)	36.0%	26.7%	1.0
Following seizures - focal only (%)	24.0%	26.7%	0.82
Following seizures - focal to tonic clonic only (%)	8.0%	6.7%	0.85
Following seizures - focal and focal to tonic clonic (%)	68.0%	70.0%	1.0
Status epilepticus (%)	20.0%	0.0%	0.015
Family history of epilepsy (%)	0%	16.0%	0.056
C) Comorbidities
Cognitive impairment (%)	64.0%	50.0%	0.41
Psychiatric disorders (%)	32.0%	43.3%	0.42
Developmental delay (%)	0.0%	16.7%	0.056
Head trauma before onset (%)	0.0%	13.3%	0.12
Focal CNS findings (%)	8.0%	10.0%	1.0
Febrile seizure (%)	0.0%	10.0%	0.24
Infectious meningitis / encephalitis before onset (%)	0.0%	10.0%	0.24
D) Treatments
Number of ASM at EEG (n)	1.7 ± 1.0	1.8 ± 0.9	0.51
Seizure freedom at EEG (%)	12.0%	0.0%	0.088
Drug-resistance at EEG (%)	36.0%	43.3%	0.78
E) EEG findings
Bilateral FS (%)	96.0%	66.7%	0.011
Bilateral EDs (%)	48.0%	16.7%	0.0011
TIRDA (%)	4.0%	13.3%	0.36
F) MRI findings
Mesiotemporal lesion (%)	76.0%	26.7%	0.00039
Hippocampal sclerosis (%)	24.0%	16.7%	0.52
Gadolinium enhancing lesion (%)	0.0%	0.0%	1.0
G) Serum and CSF parameters
Serum autoantibodies positive (%)	52.0%	0.0%	0.0000006
CSF cell counts (/μl)	4.3 ± 3.6	2.1 ± 0.8	0.38
CSF Pleocytosis (%)	32.0%	0.0%	0.00089
Total serum protein (g/L)	373 ± 104	355 ± 90	0.93
CSF albumin (mg/L)	205 ± 71	214 ± 68	0.62
Serum albumin (g/L)	42.5 ± 2.7	41.6 ± 3.0	0.47
CSF/Serum albumin ratio (x 10−3)	4.9 ± 1.6	4.8 ± 1.1	0.31
BBB dysfunction (%)	20.0%	0.0%	0.015
CSF igG (mg/L)	27.2 ± 11.1	24.4 ± 10.5	0.84
Serum IgG (g/L)	8.27 ± 1.97	8.78 ± 1.51	0.86
CSF/Serum IgG ratio (x 10−3)	2.84 ± 1.08	2.68 ± 0.87	0.75
OCBs types 2 or 3 (%)	32.0%	0.0%	0.00089

We applied the Graus criteria for diagnosis of autoimmune encephalitis (suppl. Table 1) [3]. Inclusion criteria were 1. the diagnosis of definite autoimmune encephalitis with detection of an associated antibody (classified as antibody positive AE) or 2. autoantibody-negative but probable autoimmune encephalitis (classified as antibody-negative AE). For antibody-negative cases, besides fulfilling established consensus criteria, cases had to be categorized as antibody-negative AE in the clinical workup at our center based on the broader clinical context (syndrome, disease course, and treatment response). We acknowledge that - particularly in antibody-negative cases - robust and specific biomarkers are lacking and diagnostic certainty therefore relies on a structured criteria-based approach combined with comprehensive clinical judgement.

We intentionally included the full spectrum of AE entities to mirror real-world clinical practice. In a specialized epilepsy center, referrals are commonly driven by clinical suspicion of any immune-mediated epilepsy rather than a delineated AE subtype, and diagnostic labels often become clear only as MRI, CSF, EEG and serological testing data accumulate. A spectrum-based approach therefore most faithfully reflects the heterogeneous, often diagnostically challenging population in which EEG findings are interpreted. This inclusive approach also addresses the repeatedly discussed issue of potential under-recognition of immune-mediated epilepsy.

TLE cases of non-inflammatory and non-infectious etiology were included and diagnosed according to International League Against Epilepsy (ILAE) criteria [26]. The cohort reflects real-world variations in disease etiology, severity, and treatment history, ensuring a representative sample of patients encountered in routine clinical practice at a tertiary epilepsy center, including cases where an inflammatory etiology is considered and subsequently excluded. We did not intended to represent the full epidemiological spectrum of TLE or focus on the more common hippocampal-sclerosis/mesial-temporal-sclerosis TLE phenotype. To reduce the likelihood of an underlying inflammatory etiology in the TLE group, electronic medical records were reviewed for clinical, laboratory, and imaging findings. Patients with any evidence of current or past systemic inflammatory processes were excluded. This study was approved by the Local Ethics Committee (KEK number 2019–00890).

2.2. Patients' characteristics

All AE patients fulfilled the Graus criteria for definite autoimmune limbic encephalitis or autoantibody-negative but probable autoimmune encephalitis (suppl. Table 1). All AE patients in the study met the respective criteria for subacute onset and no previous seizure disorders. Among AE patients, in 52% (13/25) a specific AE-associated antibody was detected, mostly anti-GAD (53.8%, 7/13, 11280 to 12,560 in serum and 1:256 to 1:32 in CSF), followed by anti-CASPR-2 (23.1%, 3/13, 1,12′800 to 1100′000 in serum and 1:64 to 1:2000 in CSF), anti-LGI1, anti-NMDA-R, and anti-Ta/Ma (7.7%, 1/13 each). Twelve patients were characterized as AB-negative AE, meeting the criteria for autoantibody-negative but probable autoimmune encephalitis. Antibodies were detected by cell-based assays, immunoblot and mouse brain section as tissue-based assay. All patients with autoimmune encephalitis (AE) underwent a thorough diagnostic evaluation, revealing no evidence of neoplastic disease, with the exception of a single case involving anti-CASPR-2 antibodies, in which a squamous cell carcinoma of the tongue was identified.

None of the 30 TLE patients met the Graus criteria for definite autoimmune limbic encephalitis, possible AE or autoantibody-negative but probable AE. Specifically, none of the TLE patients had a Graus-compatible subacute encephalitic syndrome, CSF pleocytosis, CSF-restricted oligoclonal bands, elevated CSF IgG index, or neural autoantibodies. Psychiatric diagnoses in the TLE group did not represent acute or subacute encephalitic psychiatric manifestations. Among TLE patients, 26.7% (8/30) showed CNS lesions, and 16.7% (5/30) met MRI criteria for hippocampal sclerosis (HS). The comparatively low proportion of HS reflects the intended diagnostic control design rather than a classical mesial TLE cohort with HS.

2.3. Developmental delay

For the purposes of this study, the term “developmental delay” refers to patients with a documented clinical diagnosis of developmental delay, global developmental delay, or intellectual disability in the medical record.

2.4. Albumin quotient

Albumin quotient [x 10^3.] was calculated as mg/L in CSF / g/L in serum. An elevation was defined based on age-dependent standard values calculated by albumin quotient_age = (4 + age / 15) × 10⁻³, as previously reported [27]. By definition, blood-brain barrier dysfunction is present, if albumin quotient_measured > albumin quotient_age.

2.5. Mesiotemporal lobe signal abnormalities

Mesiotemporal lobe signal abnormalities were defined by the presence of a T2-weighted fluid-attenuated inversion recovery (FLAIR) MRI signal confined to either medial temporal lobe, or by the detection of any other well-characterized structural lesion affecting either temporal lobe.

2.6. EEG analysis

We reviewed long-term-EEG recordings with 23-channel scalp 10–20 montage (recording time at least 72 h, sampling rate 256 Hz). All long-term EEG data was reviewed first by an epileptologist and reevaluated by a study physician (S.N.). Focal slowing (FS) in EEG was defined as equivalent to “focal slow activity” and encompasses both focal “intermittent slow activity” and/or “continuous slow activity,” as described by Kane [28]. This includes any focal abnormal EEG signals with a frequency lower than the alpha range, specifically within the theta and delta bands. The term “epileptic discharges” (EDs) is defined as equivalent to “epileptiform patterns” as described by Kural [29].

2.7. Mathematical model of sleep EEG analysis (state space model)

We additionally applied a previously established mathematical model of EEG analysis (state space model) [30]. In this approach, the continuous EEG recordings for each patient were segmented into 5-s epochs and analyzed using a multitaper method (1 Hz bandwidth) to estimate power spectral densities. Each epoch was then represented as a point in the two-dimensional state space defined by the ratio of the 8.6–19.3 Hz band power and the 1.0–10.9 Hz band power (Ratio 1), and ratio of the 11.5–20.3 Hz band power and the 17.9–31.5 Hz band power (Ratio 2). Dynamical aspects of sleep and wake EEG were quantified by calculating the Euclidean distance of consecutive epochs in the ratio space to derive the “velocity” measure, and in log-ratio space to derive the “log velocity”. The time evolution of all measures was smoothed using a Hanning window of 50 epochs to diminish outlier effects and short-term fluctuations. These velocity measures quantitatively capture spectral variability of the EEG time series. Automatic classification of epochs into the two functional states (sleep and wake) was performed by clustering the log-transformed and smoothed ratios using a Gaussian Mixture Model. The analysis based on the state space model could only be performed on a subset of patients (23 TLE, 22 AE) because of incompatible channel montages or data formats.

2.8. Scores

Variables that exhibited significant differences between AE and TLE were incorporated into a compound score. First, a score (A) was derived from parameters that were significantly higher in AE compared to TLE. Second, a separate score (B) was constructed based on parameters that were significantly elevated in TLE relative to AE. The final score was calculated by subtracting score B from score A.

2.9. Statistics

Statistical comparisons between the two subsets of patients were conducted using the Fisher's exact test or the Mann-Withney U test. Statistical significance was defined as two-sided p-values <0.05. Principal component analysis was performed using GraphPad Prism 10.

3. Results

3.1. Demographics

Between AE and TLE, there were no significant differences in sex distribution (female/male ratio, p = 0.79), age at disease onset (p = 0.13), age at first seizure (p = 0.14), age at electroencephalography (EEG) (p = 0.94), or age at lumbar puncture (LP) (p = 0.83) (Table 1A, Supplementary Table 2). In contrast, disease duration at the time of EEG and LP was significantly longer in TLE compared to AE (p = 0.027 and p = 0.039, respectively) (Table 1A, Supplementary Table 3). A family history of epilepsy was observed only in patients with non-lesional TLE (22.7%, 5/22). Mean disease duration at EEG or LP was shorter for AE than TLE: Disease duration at EEG 73 ± 69 months for AE, compared to 137 ± 120 months for TLE (p = 0.027); disease duration at LP 68 ± 72 months for AE, compared to 127 ± 123 months for TLE (p = 0.039) (Table 1A, Supplementary Table 3).

3.2. Seizure characteristics

In both AE and TLE, the initial seizure type was predominantly focal rather than focal to bilateral tonic–clonic (Supplementary Table 1B, Supplementary Table 4): focal seizures occurred in 92.0% (23/25) of AE cases and 93.3% (28/30) of TLE cases.

For post-acute seizures in AE, 24.0% (6/25) were focal only, 8.0% (2/25) were focal to bilateral tonic–clonic only, and 68.0% (17/25) included both types. In TLE, the distribution was similar: 26.7% (8/30) focal only, 6.7% (2/30) focal to bilateral tonic–clonic only, and 66.7% (20/30) both. Status epilepticus occurred exclusively in AE (20.0%, 5/25; p = 0.015 compared to TLE). No significant differences were observed in comorbidities (Table 1C, suppl. Table 5), including cognitive impairment (64.0%, 16/25 for AE, and 50.0%, 15/30 for TLE) and psychiatric disorders (32.0%, 8/25 for AE, and 43.3%, 13/30 for TLE). However, developmental delay (16.7%, 5/30), head trauma before onset (13.3%, 4/30), and febrile seizures (10.0%, 3/30) occurred in TLE patients only. Between AE and TLE, there were no significant differences in the number of anti-seizure medication, seizure freedom rates, or drug resistance at the time of EEG registration (Table 1D, suppl. Table 6).

3.3. Focal slowing (FS) and epileptiform discharges (EDs)

The occurrence of bilateral FS in long-term EEG recordings was significantly higher in AE compared to TLE: 96.0% (24/25) for AE versus 66.7% (20/30) for TLE (p = 0.0068) (Fig. 1, Table 1E, suppl. Table 7 A). Similarly, the frequency of bilateral EDs was higher in AE than in TLE: 48.0% (12/25) for AE versus 16.7% (5/30) for TLE (p = 0.012). Furthermore, AE patients more frequently exhibited concomitant bilateral FS and bilateral EDs compared to TLE patients: 48.0% (12/25) for AE versus 16.7% (5/30) for TLE (p = 0.012). Among AE patients, those with associated autoantibodies exhibited a higher incidence of bilateral EDs compared to those without autoantibodies (69.2%, 9/13 vs. 25.0%, 3/12; p = 0.027). Regarding localization, bilateral FS in AE mainly reflected independent bitemporal temporal slowing, observed in 24/25 AE patients versus 20/30 TLE patients; the only AE patient without this pattern showed central midline slowing. In TLE, other FS patterns included unilateral temporal focal slowing in two patients, unilateral temporal slowing with additional extratemporal slowing in two patients, right temporal with bifrontal slowing and left temporal with right frontal slowing, and bifrontal plus bitemporal slowing in one patient. Independent bitemporal EDs were observed in 12/25 AE patients and 5/30 TLE patients. In AE, EDs were bitemporal only in 8/25 patients, bitemporal with additional frontal EDs in 3/25, and unilateral temporal EDs in 4/25 (two left and two right). In TLE, EDs were restricted to temporal regions, with Independent bitemporal EDs in 5/30 and unilateral temporal EDs in 17/30 (eight left and nine right). TIRDA was uncommon and did not differ significantly between groups, occurring in 1/25 AE patients and 4/30 TLE patients (p = 0.36).

Fig. 1.

Differences in bilateral focal slowing (bFS) and epileptiform discharges (bEDs) between autoimmune encephalitis (AE) and temporal lobe epilepsy (TLE). The occurrences of bilateral FS and EDs were higher in AE than in TLE. AE patients more often showed concomitant bilateral FS and bilateral EDs. P-values according to Chi-squared test.

3.4. State space analysis

State space analysis for 23 EEGs of the TLE cohort and 22 EEGs of the AE cohort revealed overall higher spectral EEG variability (“EEG velocities”) in AE compared to TLE (Fig. 2). In wakefulness, the spectral EEG variability was most prominently elevated in bilateral temporal posterior regions. In NREM sleep, spectral EEG variability was also higher in AE patients, but most pronounced over the posterior mid-line regions.

Fig. 2.

State space model. Measurements of spectral variability in wakefulness (upper panel) and NREM sleep (lower panel) by means of state space velocity in long term EEG of patients with temporal lobe epilepsy (TLE, n = 23) and patients with epilepsy following autoimmune encephalitis (AE, n = 22). The figure to the right shows the subtraction of both graphical representations (positive values indicating higher velocity in AE).

3.5. Neuroimaging (MRI)

Mesiotemporal lobe signal abnormalities were significantly more frequent in AE than in TLE (Table 1E, suppl. Table 7B): 76.0% (19/25) for AE versus 26.7% (8/30) for TLE (p = 0.00039, Fig. 3). No significant difference was found between AE patients with positive and negative autoantibodies (p = 0.16). The prevalence of hippocampal sclerosis (HS) did not differ significantly between groups: 24.0% (6/25) for AE versus 16.7% (5/30) for TLE (p = 0.52), and no difference was found between AE patients with and without autoantibodies (p = 0.41). In both AE and TLE, no gadolinium-enhancing lesions were detected.

Fig. 3.

Discrimination between patients with autoimmune encephalitis (AE) or temporal lobe epilepsy (TLE). A) Violin graphs of score obtained by patients with AE or TLE using 11 variables. Score A included occurrence of bilateral FS and EDs, mesiotemporal lesions, CSF pleocytosis, blood brain barrier-dysfunction, and oligoclonal bands types 2 and 3 (variable 0/1). Subtracted was score B, which included history of febrile seizure, head trauma and/or infectious meningitis / encephalitis before disease onset, and developmental delays (variables 0/1). The score A- B highly significantly discriminate between AE and TLE. Mann-Whitney U test. B) Principal component analysis using the 11 variables above. PC1/PC2 clearly discriminated between AE (red circles) and TLE (black circles). Red and black dotted circles represent the respective median ± interquartile ranges. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

3.6. CSF cell counts

The cerebrospinal fluid cell count was not significantly different (Table 1, suppl. Table 8): 4.3 ± 3.6 cells/μl for AE versus 2.1 ± 0.8 cells/μl for TLE (p = 0.38). However, pleocytosis (defined as ≥4 CSF cells/μl) was observed only in AE: 32% (8/25) for AE versus 0.0% (0/30) for TLE (p = 0.00089). No significant difference was found between AE patients with positive and negative autoantibodies (p = 0.89).

3.7. CSF/Serum albumin quotient and blood-brain barrier dysfunction

Between AE and TLE, no significant difference was observed for total serum protein (p = 0.93), CSF albumin levels (p = 0.62), serum albumin levels (p = 0.47) or CSF/serum albumin quotient (Table 1G, suppl. Table 9). However, elevated serum albumin quotient x 10³ (a surrogate marker for blood-brain barrier dysfunction) was detected in AE patients only (20.0%, 5/25, p = 0.015, suppl. Table 9). There was no difference between AE antibodies positive and negative patients (p = 0.16).

3.8. CSF/Serum IgG quotient and oligoclonal bands

Between AE and TLE, no significant difference was observed for CSF IgG (p = 0.84), serum IgG (p = 0.86), or CSF/Serum IgG index (corrected for Q Lim albumin) (Table 1G, suppl. Table 10). However, oligoclonal bands (OCBs) types 2 and 3 were detected in AE patients only (32.0%, 8/25. p = 0.0089). There was no difference between AE antibodies positive and negative cases (p = 0.47).

3.9. Score A

The following variables (0/1 values) were used to calculate a summarized exploratory score for features suggesting AE: occurrence of bilateral FS and EDs, mesiotemporal lobe signal abnormalities, CSF pleocytosis, BBB-dysfunction, and OCBs types 2 and 3 (score max. 6). The score A was significantly higher in AE than in TLE (mean ± SD, 3.0 ± 0.8 in AE, and 1.1 ± 0.7 in TLE, p = 5.1 × 10⁻⁹).

3.10. Score B

To refine this score, family history of epilepsy and specific comorbidities, such as history of febrile seizure, head trauma and/or infectious meningitis / encephalitis before disease onset, and developmental delays can be considered (score max. 5). This score B was higher in TLE than in AE (0.67 ± 0.76 in TLE, and 0.0 ± 0.0 in AE, p = 1.9 × 10⁻⁴).

3.11. Final score = A – B

Score A was subtracted from Score B to create a composite measure (Score A–B), which discriminated strongly between AE and TLE. This score was significantly higher in AE than in TLE (mean ± SD: 3.0 ± 0.8 for AE vs. 0.43 ± 0.89 for TLE; p < 0.001 Fig. 3A). No significant difference was observed between AE patients with positive versus negative autoantibodies (p = 0.20). A threshold of ≥2 yielded a specificity of 90.9% for AE.

As an alternative, we calculated a simplified score using only the two EEG variables by summing the presence of bilateral FS (0/1) and bilateral EDs (0/1) (Supplementary Fig. 1). This alternative score also significantly discriminated between AE and TLE (p < 0.005).

3.12. Principal component analysis

To develop a model for differentiating epilepsy following AE from TLE, we applied a principal component analysis (PCA) approach. Using the 11 variables included in Scores A and B, the model demonstrated strong discrimination between AE and TLE (Fig. 3B). Coefficients and variable contributions are detailed in Supplementary Table 11. Principal Component 1 (PC1) < 0 achieved a specificity of 92.7% for AE.

4. Discussion

In this study, we identified distinct electrographic features that differentiate epilepsy following autoimmune encephalitis (AE) from temporal lobe epilepsy (TLE). Two individual EEG markers were strongly associated with AE: Bilateral focal slowing was significantly more prevalent in patients with AE than in those with TLE (p < 0.05). In addition, the occurrence of bilateral epileptiform discharges (EDs) was markedly higher in the AE group (p < 0.05). Notably, the co-occurrence of bilateral FS and bilateral EDs was more frequently observed in AE patients compared to those with TLE (p < 0.05). In addition, quantitative state-space analysis of wake EEG revealed higher temporal EEG velocity in AE compared to TLE, indicating greater short-timescale variability of the spectral composition in temporal regions. Finally, a composite score integrating visual EEG findings with MRI, cerebrospinal fluid (CSF) data, and clinical parameters, proved highly effective in distinguishing AE from TLE (p < 0.0001). In this context, principal component analysis identified a robust model capable of reliably differentiating between the two patient cohorts.

The findings of visual and quantitative EEG analyses provide complementary insights into the divergent electrophysiological patterns in patients with epilepsy following autoimmune encephalitis (AE) compared to those with temporal lobe epilepsy (TLE) in the absence of an inflammatory etiology. The significantly higher incidence of bilateral focal slowing (FS) and bilateral epileptiform discharges (EDs) in AE patients relative to TLE suggests a more pervasive and pronounced disruption of neural activity, particularly within the temporal lobes. This disparity likely reflects the influence of an immune-mediated or inflammatory pathophysiological process or immune-mediated injury in AE, leading to more dissipated and severe cortical dysregulation. Similarly, the significantly higher state-space velocities in temporal regions during wakefulness in AE indicate increased spectral variability. Namely, state-space velocity reflects the distance between consecutive 5-s-long EEG epochs in a space of spectral ratios and therefore captures how rapidly the relative spectral composition changes over time. Higher velocities therefore suggest stronger short-timescale fluctuations of the temporal EEG dynamics and a more complex, less stable state of neural activity in AE.

Together, the increased prevalence of FS and EDs on visual EEG analysis and the higher state-space velocities on quantitative EEG analysis point to a more profound and dynamic perturbation of brain function in AE, potentially driven by immune-mediated mechanisms, as compared to TLE with other etiologies. The quantitative EEG findings, although distinct, are compatible with the visually determined EEG abnormalities: FS and EDs identify clinically recognizable abnormal patterns, whereas state-space velocity captures spectral variability of the EEG dynamics. Importantly, this agreement is also topographical: visual EEG showed predominantly bitemporal abnormalities, while quantitative EEG analysis in wake showed increased spectral variability also predominantly in temporal regions. The state space model, by offering a quantitative and objective measure of spectral variability, provides a more nuanced understanding of the underlying neural disturbances, complementing the qualitative EEG abnormalities (FS and EDs) and enhancing our appreciation of the neurophysiological alterations associated with AE. Nevertheless, epochs containing intermittent slowing, epileptiform discharges, or shifts in the dominance of different frequency components are more likely to change their position in spectral-ratio space. Therefore, patients with more frequent visually recognizable EEG abnormalities may also show higher state-space velocities. This shared dependence on changes in the spectral composition of the EEG signal may therefore explain why both findings occur together in AE: visual EEG abnormalities and increased state-space velocities in the temporal regions likely represent two different expressions of a more dynamically changing electrophysiological pattern.

A possible explanation for the higher rates of bilateral EEG abnormalities and spatial variability of EEG findings over time in patients with AE compared to TLE might be the dynamic differences in the severity and spatial distribution of immune processes over time, namely the expression of autoantigens and focal accumulation of immune mediated tissue damage. It is well established that the affected brain regions in AE are dependent on the tropism of the immune-mediated injury which is usually associated with specific autoantibodies. For example, GAD is located in nerve endings at the presynaptic terminal [31], while CASPR-2 is a membrane protein of myelinated axons [32]. While the classical diagnostic criteria for AE do not specifically differentiate between unilateral and bilateral EEG abnormalities [3], our findings suggest that long-term EEG monitoring provides essential insights into the neural dynamics of these two conditions, even in the chronic phase.

In addition to EEG findings, other diagnostic markers such as MRI, CSF analysis, and clinical data significantly aid in distinguishing AE from TLE. A score including EEG data, MRI and CSF parameters, as well as clinical data allowed to distinguish between AE and TLE (p < 0.001). Using the same variables, principal component analysis proposed a model discriminating well between these two groups. Nevertheless, it is important to note that a score using EEG variables only, namely bilateral FS and bilateral EDs, already discriminated well between AE and TLE in our study, underscoring the importance of EEG in the diagnostic process. Integrating EEG findings with MRI and CSF parameters enhances diagnostic accuracy. While several predictive scores have been developed to aid in the selection of patients for neural antibody testing, many of these scores overlap with existing diagnostic criteria for AE, such as the presence of pleocytosis, raising questions about their additional clinical utility [33]. In this context, our aim was not to a establish a novel diagnostic score to replace current diagnostic standard, but the description of accessible long-term EEG features that are not fully captured by current criteria. Specifically, independent bitemporal slowing and independent bitemporal EDs may provide additional diagnostic information when interpreted together with clinical information, MRI, CSF, and antibody results.

In the diagnostic workup of suspected epilepsy, particularly in cases of new-onset temporal lobe epilepsy (TLE) presenting in atypical age groups, the consideration of autoimmune epilepsy (AE) as a differential diagnosis is of considerable clinical relevance. Although long-term EEG is routinely employed in the post-acute phase of new-onset epilepsy, there is a notable lack of data on electrophysiological markers that reliably distinguish AE-related TLE from non-inflammatory TLE during this period. Current diagnostic criteria highlight the relevance of bilateral MRI and EEG abnormalities [3]. However, prior EEG studies conducted in the post-acute phase have not demonstrated higher prevalence of bilateral EEG features in AE-related TLE [22]. In our cohort, bilateral EEG abnormalities appeared more frequently in AE-related than in non-inflammatory TLE. Although our study cohort is heterogeneous in terms of underlying etiologies, these simple EEG features may still contribute to diagnostic accuracy and help to guide clinical diagnostic decision-making in the post acute phase of AE.

Clinically, these EEG features are not intended to establish AE diagnosis in isolation. Their utility lies in risk stratification and in guiding next diagnostic and therapeutic steps. In a patient with new-onset or evolving temporal lobe epilepsy, independent bitemporal temporal slowing or independent bitemporal EDs may justify repeat MRI review, CSF analysis, neuronal antibody testing, and neuroimmunological consultation. Conversely, a pattern dominated by unilateral temporal EDs with negative inflammatory workup may support a conventional TLE pathway. These findings also have implications for epilepsy surgery: EEG features of AE do not preclude presurgical evaluation, but they should prompt careful assessment of whether seizures arise from a stable unilateral epileptogenic zone or from a bilateral/dynamic inflammatory network. Active or relapsing inflammation should be treated and stabilized before irreversible epilepsy-surgical decisions are made, whereas chronic post-inflammatory drug-resistant epilepsy may still be considered for resective, ablative, or neuromodulatory approaches when electroclinical, imaging, and neuropsychological data are concordant.

CSF analysis plays a crucial role in differentiating autoimmune encephalitis from temporal lobe epilepsy, as evidenced in our cohort. Specifically, 20% of AE patients exhibited blood-brain barrier dysfunction, indicated by an elevated albumin quotient, a finding not observed in TLE patients. Although CSF/serum IgG levels and the IgG index did not show significant differences between AE and TLE, the presence of oligoclonal bands (types 2 and 3) in 32% of AE patients is noteworthy. These CSF-restricted OCB types, which serve as a sensitive marker of intrathecal immune activity, were present in both autoantibody-positive and autoantibody-negative AE patients, but were absent in TLE patients. This finding suggests increased intrathecal B-cell activity and may serve as a marker of disease severity or prognosis in AE [34]. Antibodies directed against the CD20 epitope on the surface of B cells are an important and well-established second-line treatment of AE [35]. In our cohort of AE patients, 38% displayed pleocytosis, while 62% did not, underscoring the critical point that a normal CSF cell count does not rule out AE. This finding is consistent with previous studies [36], highlighting the need for clinicians to consider AE even in the absence of pleocytosis.

A key limitation of our study is the small sample size, which reduced statistical power and limits the precision and generalizability of our estimates. In addition, we pooled all AE entities in a single analysis; while this mirrors real-world practice in a specialized center, it may significantly obscure subtype-specific EEG patterns and introduce heterogeneity that cannot be fully accounted for in a cohort of this size. The TLE comparator group should not be interpreted as representative of TLE in general or as a hippocampal-sclerosis-enriched surgical mTLE cohort. Rather, it reflects the clinically relevant tertiary diagnostic scenario in which inflammatory etiologies have been considered and excluded. This approach increases clinical relevance for the differential diagnosis of post-acute AE-related epilepsy, but it limits generalizability to population-based TLE cohorts. Finally, the composite score and PCA were exploratory and require independent external validation before clinical implementation. The quantitative state-space analysis should also be interpreted cautiously, as higher spectral variability might not necessarily be specific for autoimmune mechanisms but may reflect any brain process that increases short-timescale spectral variability.

In summary, CSF findings such as pleocytosis, oligoclonal bands (types 2 and 3), and an elevated albumin quotient may indicate an underlying inflammatory process and immune-mediated involvement of the central nervous system. Additionally, bilateral focal slowing or bilateral epileptiform discharges in LT-EEG, and mesiotemporal lesions on MRI may further support an immune-mediated hypothesis. Conversely, certain clinical features may suggest a non-inflammatory cause of temporal lobe epilepsy, such as a history of developmental delay, prior infectious meningitis or encephalitis, head trauma, family history of epilepsy, or febrile seizures. These factors point toward structural, genetic, or post-infectious mechanisms. These findings underscore the critical importance of systematic history-taking, comprehensive EEG analysis, and the integration of clinical parameters in evaluating immune-mediated etiologies in temporal lobe epilepsy, offering clinicians a practical and accessible diagnostic framework.

5. Conclusion

This comparative cohort study demonstrates objective and quantifiable LT-EEG properties in epilepsy patients following AE. The distinctive EEG patterns and the proposed score could serve as biomarkers for differentiating between epilepsy following AE and TLE of non-inflammatory etiology. Our findings may aid clinicians in determining the need for further diagnostic testing and in optimizing therapy decisions, but they require external validation and should be interpreted in conjunction with established clinical, MRI, CSF, and antibody criteria.

CRediT authorship contribution statement

Stephan Neidhart: Writing – original draft, Investigation, Formal analysis, Data curation, Conceptualization. Tena Dubcek: Writing – review & editing, Methodology, Formal analysis, Data curation. Debora Ledergerber: Writing – review & editing. Oona Kohnen: Writing – review & editing, Methodology, Investigation. Lukas Imbach: Writing – review & editing, Writing – original draft, Conceptualization.

Informed consent and patient details

Informed consent was obtained from all patients involved in the study.

Ethical statement

All procedures performed in this study were conducted in accordance with the ethical standards of the responsible institutional and national research committees and with the 1964 Helsinki Declaration and its later amendments. Ethical approval was obtained from the Local Ethics Committee (KEK number 2019–00890). Informed consent was obtained from all individual participants included in the study. Patient data were anonymized prior to analysis to ensure confidentiality, and no identifiable information is reported. The authors declare that they have no conflicts of interest related to this work.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary data

Supplementary material