Table 1 summarizes various demographic, serological, and clinical features. Seventy-four patients (49.3%) were seronegative, whereas 76 (50.7%) had at least one well-characterized antineuronal antibody (therefore seropositive). Age and sex were well-balanced between the groups. As expected, the majority of patients with seropositive AE had antibodies targeting cell-surface antigens, mostly NMDAR antibodies (in 21 cases) and LGI1 antibodies (in 19 patients). Even though there was no significant difference in the overall number of malignancies between the groups, bronchial carcinomas were less frequent in the seronegative AE group (1 (1.4%) vs 8 (10.5%) cases, p = 0.034). As clinical manifestations, epileptic seizures (26 (36.1%) vs 41 (54.0%), p = 0.029), and focal neurological deficits (18 (24.7%) vs 31 (40.8%), p = 0.036) were less frequent in the seronegative AE group, whereas other symptoms were equally distributed. Patients with seronegative AE were statistically significantly more severely affected at the time of admission as measured by the modified Rankin Scale (mean mRS 3.3 ± 1.1 vs 2.9 ± 1.0, p = 0.038).

Regarding CSF parameters, there was a trend towards CSF-restricted OCBs being less frequent in the seronegative group (12 (17.1%) vs 21 (30.9%) in the seropositive AE cases, p = 0.059), whereas an increased albumin quotient as a marker of blood-CSF barrier dysfunction was significantly more frequent (seronegative 31 (43.1%) vs seropositive 19 (26.8%), p = 0.041). Various abnormalities on EEG were balanced between the groups. The proportion of patients with abnormal brain MRI (seronegative 57 (80.3%) vs seropositive 52 (74.3%) cases, p = 0.395), and FDG-PET (seronegative 50 (86.2%) vs seropositive 63 (91.3%) patients, p = 0.361) was not statistically different between the groups. The high rate of FDG-PET investigations (127/150 patients (84,7%)) is attributable to the generally low-threshold use of this modality in our hospital, irrespective of MRI results. In addition, most patients received both FDG-PET of the brain and the whole body; the latter for screening of malignancies. More details on CSF parameters, EEG findings, and imaging results are presented in Table 2.

The overall proportion of patients receiving immunotherapies was high in both groups, but significantly lower in patients with seronegative AE (53 (73.6%) vs 66 (86.8%) seropositive cases, p = 0.043). With regards to specific immunotherapies, seronegative AE patients were less frequently treated with intravenous immunoglobulins (IVIG, 7 (9.7%) vs 22 (29.0%), p = 0.003), rituximab (7 (9.7%) vs 20 (26.3%), p = 0.009), and azathioprine (8 (11.1%) vs 18 (23.7%), p = 0.045). There was also a trend towards seronegative AE patients being less frequently treated with high dose glucocorticoids (50 (69.5%) vs 63 (82.9%), p = 0.054). Of note, the proportion of patients being treated with therapeutic plasmapheresis was not statistically different between the groups (26 (31.1%) vs 32 (42.1%), p = 0.455). All patients with an underlying malignancy received oncological therapy according to current guidelines.

Therapy response on general impression by the treating physician was high, and not different between the groups (49 (92.5%) vs 57 (86.4%), p = 0.290) with most patients in both groups responding to first-line immunotherapies (44 (88.0%) vs 50 (80.7%), p = 0.292). However, there was a trend towards a longer length of hospital stay in the seronegative group (47.6 ± 52.3 vs 34.8 ± 36.4 days, p = 0.090). There was also a trend towards fewer patients having a favorable outcome (mRS 0-2) at discharge in the group of patients with seronegative AE (31 (43.1%) vs 44 (57.9%), p = 0.071). This difference became significant at last follow-up (33 (45.8%) vs 52 (68.4%) with mRS 0-2, p = 0.005). However, in both groups the proportion of patients with a favorable mRS of 0-2 was twice as high at last follow-up as compared to baseline (seronegative: 33 (45.8%) at last follow-up vs 16 (21.9%) at baseline; seropositive: 52 (68.4%) at last follow-up vs 25 (32.9%) at baseline). However, for the comparison of longer-term outcome, mean follow-up time was significantly shorter in seronegative AE patients (15.3 ± 24.9 vs 24.3 ± 28.1 months, p = 0.007). Three patients from the seropositive group (two during the acute phase, one during longer-term follow-up) and one patient from the seronegative group died, the latter also during the acute phase. Figure 1 shows frequency of treatment and therapy response on general impression in both groups. Further details on immunotherapies, and all clinical outcomes are depicted in Table 3.

Based on our data and another study with similar inclusion criteria (15), the frequency of seronegative AE equaled that of seropositive cases in our tertiary care university hospital, which contrasts previous prevalence data indicating a considerably lower frequency of seronegative AE (1). Differences between study populations, methods in data acquisition (hospital-based vs population-based), and an increasing awareness towards the diagnosis over time have to be taken into account. The high proportion of seronegative cases in our cohort emphasizes the importance of identifying these patients, since the majority benefitted from immunotherapies. Even though some clinical features (epileptic seizures and focal neurological deficits) were less frequent in seronegative patients in our study, these patients are usually regarded as clinically indistinguishable from seropositive cases (5, 15, 16). Furthermore, there were no specific findings differentiating seronegative and seropositive AE cases using CSF analysis, EEG, MRI or FDP-PET imaging as reported previously (5, 8, 15–17). Hence, in the acute clinical setting, when antibody results are pending, or during further course, when all antibody results turn out to be negative, seronegative and seropositive AE patients seem to be mostly indistinguishable based on clinical and paraclinical features, yet therapeutic decisions have to be made.

In comparison to two previous studies, patients with seronegative AE in our study less frequently had seizures (36% in our study vs 81% and 65%, respectively) (16, 18), psychiatric symptoms (59% in our study) were either less frequent (76%) (18) or identical (16), and memory problems (76% in our study) were either less frequent (93%) (18) or more frequent (62%) (16). These differences are most likely due to different patient populations as well as different inclusion criteria. Of note, all three categories of symptoms were most frequent in the study by Lee et al., which might be attributable to the prospective study design (18), whereas our study and the study by Gastaldi et al. were retrospective (16).

As stated above, the vast majority of patients with seronegative AE in our study received immunotherapies, which is in line with previous publications (16–19). Relative frequency of various immunotherapies were in line with other studies with glucocorticoids and other first-line immunotherapies (IVIGs and plasmapheresis) being given to most of the patients, regardless whether they are seronegative (17–19) or seropositive (7, 16).

With regards to therapeutic response rates reported previously in patients with AE, these were generally high, yet naturally revealed some variations dependent on patients included, outcome parameter applied and time point analyzed. The thus far largest study included 577 seropositive patients with NMDAR encephalitis only (7), of whom 472 (82%) received first-line therapies. Of these, 251 (53%) showed improvement on the mRS within four weeks. With regards to seronegative cases only, the thus far largest study that included 147 patients with exclusively seronegative AE (117 with antibody-negative probable AE, 23 with seronegative limbic encephalitis, and 7 with acute demyelinating encephalomyelitis) was performed by Lee and colleagues (18). In this study, a favorable 2-year outcome, defined as a mRS 0-2, was achieved in 57%, which is roughly within the order of magnitude of our data. However, direct comparison is limited due to divergent inclusion criteria, and since in our study patients were less severely affected at baseline (median mRS 5.0 in the study by Lee et al. vs a mean mRS 3.3 in our study), follow-up time was shorter (2 years in the study by Lee et al. vs a mean of 19.9 months in our study), time till treatment initiation longer (median 8.0 days in the study by Lee et al. vs a mean of 14 weeks between primary manifestation and hospital admission in our study), and treatment less aggressive, i.e., Lee et al. treated 79% with rituximab (vs only 9.7% in our study), 40% with tocilizumab (not performed on our cohort), and 10% with cyclophosphamide (vs 1.4%). The high rate of seronegative AE patients being treated with rituximab in the study by Lee et al. at first glance seems counterintuitive (18). However, the high efficacy indicates a significant role of B cells in seronegative cases. Speculatively, these patients have antibodies against yet to be identified antigens or pathophysiologically antibody-independent B cell immunity plays a role (20–22). These data challenge current guidelines recommending use of rituximab preferably in seropositive AE (9), and could encourage treating physicians to consequently escalate therapy also in seronegative AE patients.

In the thus far only other study directly comparing treatment efficacy in patients with seronegative and seropositive AE by Gastaldi et al. the majority of patients (110/118; 93.2%) received immunotherapies, and showed clinical improvement (84.5%) (16). Yet, in contrast to our data seronegative cases responded less frequently than seropositive patients (65.6% vs 92.3%), which was at least in part attributable to delays in treatment initiation as well as less patients receiving second-line immunotherapies (13.7% vs 39.5%) in this group (16).

Two studies focused on seronegative limbic encephalitis only, and included 11 and 28 patients, respectively (17, 19). Following immunotherapies, an improvement was achieved in 54% and – dependent on the outcome parameter applied – 11-48% of cases, which was considerably lower compared to 93% clinical improvement in our study including unselected seronegative patients, and also lower compared to 81% in another study including unselected seropositive (68%) and seronegative (32%) patients (1). Even though mode and frequency of various treatments were not generally different between these studies, a generally worse prognosis in limbic encephalitis has been presumed (17–19).

The other side of the coin with rapid initiation and escalation of immunotherapy, especially in seronegative AE cases, is the potential of misdiagnosis, and unnecessarily exposing patients to side effects. In a recent retrospective study, this related to 27% of all patients, who were first diagnosed with AE, and later identified to have other disease. However, the majority of these (77%) was not fulfilling diagnostic consensus criteria, and had diagnosis other than AE, e.g., functional neurological syndromes or psychiatric disorders (23). Hence, strict adherence to consensus criteria and exclusion of differential diagnosis would have avoided these misclassifications (5).

Our study has several limitations. First, due to the retrospective design, some data was missing in patients, and the length of follow-up was highly variable, notably shorter in seronegative AE patients. However, the majority received standardized diagnostic work-up with clinical as well as neuropsychological testing, CSF analysis, antibody testing in both serum and CSF, and imaging with both MRI and FDG-PET. Nonetheless, electronically recorded data were insufficient in the retrospective setting to calculate the CASE score (Clinical Assessment Scale in Autoimmune Encephalitis) during follow-up, which had been specifically developed as an objective measure to evaluate outcome in patients with AE (24). Instead, a qualitative and subjective measure (“general impression”) as well as the quantitative, but not AE-specific mRS were applied, since these were easily available and had been used by others (7, 16, 18). In addition, data on the premorbid status of our patients were insufficient. Hence, it might be that the higher mRS in the seronegative group at admission was due to a higher premorbid mRS. However, this is rather speculative. Second, not all patients (n = 32, 21%) were fulfilling consensus criteria (5). However, cases were extensively discussed, and still included if satisfying recent “best practice recommendations for diagnosis and acute management” of AE (9). These recommendations comprise a broader clinical spectrum in comparison to the consensus criteria, e.g. patients with cerebellitis or cerebellar degeneration and encephalomyelitis. According to these recommendations, CSF-restricted OCBs instead of pleocytosis (0.7% of our cases) and FDG-PET instead of MRI alterations (3.3% of our cases) might support an inflammatory etiology in the right clinical setting and after exclusion of alternative diagnosis. Additionally, patients fulfilling all criteria except for the subacute presentation within three months were included (6.7%). The reason, why a significant proportion of our patients was not fulfilling this criterion was potentially due to a high proportion of patients with a pure psychiatric syndrome, who tend to not present within the acute stage of the disease. Finally, a significant number of patients (10.7%) had antibodies against unknown neuronal antigens, which according to the criteria “strongly support the diagnosis of autoimmune encephalitis” (5). According to a recent publication analyzing “autoimmune encephalitis criteria in clinical practice”, in 538 patients diagnosed at the Mayo clinic, 177 (33%) were not fulfilling consensus criteria, a significant number having inflammatory CSF alterations other than pleocytosis (e.g. OCBs), presenting outside the 3-months-interval or having antibodies against unknown neuronal antigens (25). Third, due to the monocentric design the question of generalizability arises. However, our tertiary care university hospital has a large catchment area of approximately 2,500,000 inhabitants and the inclusion of both neurologic and psychiatric patients warrants inclusion of the majority if not all patients who had been diagnosed with AE in Southwestern Germany within a ten-year period. In addition, the monocentric design made sure, that all patients received identical diagnostic work-up and had access to the same therapeutic options. In this regard, previous studies showed good consistency with respect to glucocorticoids being used as first choice. However, the use of other therapeutic options (e.g., plasmapheresis, rituximab) showed great variability (1, 7, 16–19), probably due to personal preferences and experiences. Fourth, due to the continuous identification of novel target antigens (22), the classification of patients into the two categories seronegative and seropositive naturally changes over time. However, for reasons of consistency and comparability with previous studies we used definitions of currently available consensus criteria and guidelines (5, 9). Fifth, since the group of seronegative patients was presumably heterogenic, potentially including patients with low titer antibodies, antibodies against yet to be identified antigens, with cell-mediated immune processes, but also with non-immunogenic etiologies in some cases, expected therapy response also would be heterogenic. The high response rate to immunotherapies at first glance indicates an immunogenic pathophysiology in most cases. However, a clear diagnostic marker – as with antibodies in all seropositive cases – is lacking. Hence, improvement in some cases might also reflect spontaneous disease course or response to concomitant medication (e.g., anticonvulsants or neuroleptics), rather than immunotherapy response. Therefore, further research on the underlying pathophysiology is urgently needed. Presuming heterogenous pathophysiological processes in patients with seronegative AE, we decided not to subdivide patients with seropositive AE into those with antibodies against cell-surface and intracellular antigens. Since the latter are known to show poor response to immunotherapies, this would have introduced a relevant selection bias.

In conclusion, the vast majority of patients with seronegative AE benefitted from immunotherapies. Of note, treatment response was not different in comparison to seropositive cases. This indicates that international consensus criteria for the diagnosis of AE, that were deliberately not reliant on antibody results (5), and current guidelines (9) as well as rigorous exclusion of alternative diagnosis if an autoimmune etiology is presumed, seem to be sufficient in identifying patients benefitting from immunotherapies, regardless whether they are seronegative or seropositive. This is of particular clinical importance, since seronegative patients are otherwise clinically and paraclinically mostly indistinguishable from seropositive AE cases, and therapeutic decisions have to be made often before antibody results are available. For more reliable conclusions prospective multicenter studies would be necessary. The pathophysiological mechanisms underlying seronegative AE (e.g., novel neuronal antibodies or T-cell mediated processes) should be analyzed in parallel.