Differential Modes of Action of α1- and α1γ2-Autoantibodies Derived from Patients with GABAAR Encephalitis

Abstract Autoantibodies against central nervous system proteins are increasingly being recognized in association with neurologic disorders. Although a growing number of neural autoantibodies have been identified, a causal link between specific autoantibodies and disease symptoms remains unclear, as most studies use patient-derived CSF-containing mixtures of autoantibodies. This raises questions concerning mechanism of action and which autoantibodies truly contribute to disease progression. To address this issue, monoclonal autoantibodies were isolated from a young girl with a range of neurologic symptoms, some of which reacted with specific GABAA receptor (GABAAR) subunits, α1-subunit and α1γ2-subunit, which in this study we have characterized in detail using a combination of cellular imaging and electrophysiological techniques. These studies in neurons from wild-type mice (C57BL/6J; RRID:IMSR_JAX:000664) of mixed-sex revealed that the α1 and α1γ2 subunit-specific antibodies have differential effects on the GABAA receptor. Namely, the α1-antibody was found to directly affect GABAA receptor function on a short time scale that diminished GABA currents, leading to increased network excitability. On longer time scales those antibodies also triggered a redistribution of the GABAA receptor away from synapses. In contrast, the α1γ2-antibody had no direct effect on GABAA receptor function and could possibly mediate its effect through other actors of the immune system. Taken together, these data highlight the complexity underlying autoimmune disorders and show that antibodies can exert their effect through many mechanisms within the same disease.


Introduction
GABA A receptor encephalitis (GABA A RE) is a disease characterized by the presence of autoantibodies targeting the GABA A Receptor (GABA A R) and is associated with symptoms that include hallucinations, abnormal movement, epilepsy, and altered cognition, behavior, and consciousness (Ohkawa et al., 2014;Petit-Pedrol et al., 2014;Pettingill et al., 2015). Current treatment consists of a combination of corticosteroids and plasma exchange (O'Connor et al., 2019) and/or immunosuppressants (Lancaster, 2016;Shin et al., 2018). However, at present, the molecular and cellular mechanisms by which these antibodies cause disease progression are still rudimentary.
The predominant GABA A R autoantibody targets include A1, b 3, and g 2 subunits of the GABA A Rs (Ohkawa et al., 2014;Petit-Pedrol et al., 2014;Pettingill et al., 2015). GABA A Rs are ionotropic receptors involved in both phasic and tonic inhibition within neuronal circuits. They are assembled as pentamers with the most dominant receptor isoform being a1g 2b 2a1b 2 (Sigel and Steinmann, 2012;Ohkawa et al., 2014), which are expressed in a region and cell type-dependent manner (Mortensen et al., 2011). Subunit interfaces create binding sites not only for the neurotransmitter GABA but also for modulators such as benzodiazepine, which, for example, binds with high affinity to sites located between the a1, a2, a3, a5, and g 2 subunit combinations (e.g., a1g 2), increasing the affinity for GABA and receptor opening times (Sigel and Steinmann, 2012).
Initial studies on GABA A RE focused mainly on characterizing the long-term effects (.24 h) of patient-derived CSF in vitro within neuronal cultures. These studies revealed a loss of GABA A R clusters from inhibitory synapses following a 24-h or longer incubation with CSF (Ohkawa et al., 2014;Petit-Pedrol et al., 2014;Pettingill et al., 2015) as well as reduction in miniature IPSCs (mIPSCs; Ohkawa et al., 2014), indicating that such antibodies may directly increase neuronal network excitability and epilepsy by the loss of these receptors from synapses.
A major limitation to these and related studies is that patient CSF generally contains a mixture of several autoantibodies, only some of which are selective for GABA A Rs, raising questions about which are truly responsible for the observed symptoms. This has been largely resolved by isolating and cloning monoclonal antibodies from patient derived B-cells (Kreye et al., 2016) including from an eight-year-old girl with GABA A RE (Kreye et al., 2021), who presented at the hospital with fever, fatigue, reduced appetite, and apathy, which rapidly developed into severe confusion and mutism (Nikolaus et al., 2018). In the present study, we have performed a detailed cellular and molecular characterization of two of these antibodies, one that selectively recognized the a1 subunit and a second that recognized a1 and g 2 subunits (hereafter a1g 2) of the GABA A R. Their current mode of action is not well understood, but could involve receptor internalization (Petit-Pedrol et al., 2014;Pettingill et al., 2015;Jain and Balice-Gordon, 2016) or direct effects on GABA A Rs, depending on whether they bind an activation, neutral or modulatory site on these receptors (Diamond et al., 2009). This latter option is particularly attractive for the a1g 2-antibody which presumably binds the interface of these two subunits, perhaps affecting actions at the benzodiazepine site.
To address these issues, we have performed a detailed analysis of how the a1-antibody and a1g 2-antibody affect the distribution and function of GABA A Rs impact on neuronal network excitability as well as sensitivity to benzodiazepines in cultured striatal/cortical neurons. Our results show that the a1-antibody potently and rapidly affects spiking activity and GABA A R function (,5 min), independent of receptor cross-linking and internalization, and has a long-term effect (.20 h) on the distribution and function of synaptic GABA A Rs. In contrast, the a1g 2-antibody minimally effected the functionality of GABA A Rs, their sensitivity to benzodiazepine or striatal/cortical neuron excitability, suggesting its mode of action is indirect, perhaps via microglia-mediated immune response to this antibody. Together, these data indicate that autoantibodies can act through a variety of mechanisms to alter receptor and network function, contributing to the etiology of these diseases.

Neuronal culture preparation
All experiments with animal material were designed according to criteria set by the animal welfare committee of the Charité Medical University and the Berlin state government. Cells were isolated according to the "Banker Protocol" (Banker and Goslin, 1988;Meberg and Miller, 2003) from wild-type mice (C57BL/6J; RRID:IMSR_JAX:000664) of mixed-sex. In short, striatum and cortexes of postnatal day (P)0-P2 mice were dissected and incubated in enzyme solution (DMEM, Invitrogen, Thermo Fisher Scientific; 3.3 mM cysteine, 2 mM CaCl 2 , 1 mM EDTA, and 20 U/ml papain, Worthington) for 45 min at 37°C. Subsequently, inactivation solution (10% FCS, Thermo Fisher Scientific; 38 mM BSA, Sigma-Aldrich; and 178 mM trypsin inhibitor, Sigma-Aldrich; in DMEM) was added for 5 min to stop the digestion. Neurons were isolated by gentle trituration in NBA-complete (Neurobasal-A, Thermo Fisher Scientific; 2% B27, Invitrogen #17504; 1% glutamax, Invitrogen #35050; 0.2% penicillin/streptomycin, Roche #11074440001) and plated at different densities and in various preparations, specified below. Neuronal cells for each of these preparations were cultured for 14-18 d in vitro (DIV) in NBA-complete before subsequent use.
Autaptic striatal cultures were prepared by plating single striatal neurons onto astrocyte microislands on glass coverslips, prepared as described previously (Furshpan et al., 1976;Bekkers and Stevens, 1991). In short, acidwashed coverslips were covered in a thin layer of 0.15% agarose to discourage astrocytes and neurons from adhering. Then, small dots containing 0.25 mg/ml collagen and poly-D-lysine were added to the coverslip with the help of a stamp (Rost et al., 2015), to ensure that astrocytes and neurons would only adhere in these places. Later, astrocytes were isolated from P0-P2 wild-type mice cortices and plated at a density of 10,000 cell/cm 2 for 7 d before the addition of neurons. Next, striatal neurons were plated at 500 cells/cm 2 to ensure only one neuron per astrocyte island.
For both immunocytochemistry and electrophysiological experiments cortical-striatal mass-cultures were prepared. Here, astrocytes were plated at a density of 18,500 cells/cm 2 directly into cell-culture plates and grown for 7 d. Striatal and cortical neurons, prepared on day 7 were plated at a ratio of 3:1 and at a density of 60,000 cells/cm 2 on glass coverslips that were acid washed and poly-L-lysine coated previously (Tian et al., 2010;Penrod et al., 2011;Lalchandani et al., 2013). These coverslips were decorated with paraffin dots and were then placed upside down over the astrocyte covered cell-culture plates.
For calcium imaging experiments, astrocytes were also plated 7 d before the direct addition of neurons at a density of 18,500 cells/cm 2 . Here, cells were plated directly into round dishes containing a cell location grid at the bottom (m-Dish 35 mm, high Grid-500, Ibidi), allowing the repetitive imaging of the same cells. For some calcium imaging experiments, both astrocytes and neurons were directly plated on PDL-coated coverslip. In both cases, striatal and cortical neurons were plated in a ratio of 3:1 at a high density of 120,000 cells/cm 2 directly on top of the astrocyte feeder layer. In each case, after 7 DIV, one-third of the culture media was replaced with fresh NBA-complete, which was repeated every 2-3 d until the experimental day.

Electrophysiological recordings
We performed whole-cell patch-clamp recordings in two different neuronal-culture preparations. First, we recorded from striatal autaptic cultures that were incubated with 1 mg/ml of a1-antibody, a1g 2-antibody, or control-antibody (human anti-alemtuzumab, Bio-Rad catalog #HCA175, RRID:AB_11152938) for 24 h before the onset of the experiment. Second, we also recorded neurons in cortical-striatal mass-cultures, where neurons were treated 1 h before experiment onset with 1 mg/ml of control or the a1-antibody, or 5 mg/ml of the a1g 2-antibody. In general, recordings were performed at room temperature under the control of Clampex 10.4 software (Molecular Devices), and neurons were voltage clamped at À70 mV with a Multiclamp 700B amplifier (Molecular Devices, RRID:SCR_018455). Data were sampled at 10 kHz and filtered through a 3-kHz Bessel filter. Series resistance was kept under 15 MV and were compensated to ;70%. During all recordings, the leak current was monitored and any recordings with a leak larger than À300 pA were discarded. All cells were kept in extracellular solution during recordings, which consisted of 140 mM NaCl, 2.4 mM KCl, 10 mM HEPES, 10 mM glucose, 2 mM CaCl 2 , and 4 mM MgCl 2 . Whole-cell patch-clamp recordings were obtained through borosilicate glass pipettes (3-8 MV) pulled with a micropipette puller device (Sutter Instruments). These pipettes were filled with internal solution containing 136 mM KCl, 17.8 mM HEPES, 1 mM EGTA, 0.6 mM MgCl 2 , 4 mM ATP-Mg, 0.3 mM GTP-Na, 12 mM phosphocreatine, and 50 U/ml phosphocreatine kinase. All solutions were adjusted to pH 7.4 and osmolarity of ;300 mOsm.
For 24-h striatal autapse experiments, all recordings were performed in standard intracellular and extracellular solutions, except when bath-applying NMDA, which was measured in extracellular solution containing 0 mM Mg21, 0.2 mM CaCl 2 , and 10 mM glycine. First, we evoked IPSCs through a 2-ms somatic depolarization from À70 to 0 mV to probe the synaptic GABA A R response. Subsequently, IPSCs were evoked in the presence of bicuculline (30 mM, Santa Cruz #sc-202498) to assess whether cells were indeed inhibitory neurons. From traces in control solution, miniature IPSCs (mIPSCs) were recorded and analyzed after 1-kHz low-pass filtering using a templated based detection feature Axograph X (RRID:SCR_014284). False negative events were excluded by substracting events detected in GABA A receptor antagonist bicuculline. To quantify the synaptic and extrasynaptic surface glutamate and GABA A receptor activity, responses to pulsed application of kainate (1 s, 20 mM, Tocris #0222), NMDA (1 s, 10 mM, Tocris #0114), and GABA (3 s, 5 mM, Tocris #0344) were recorded. The size of the readily releasable pool of vesicles pool was determined by applying extracellular solution that contained additionally 0.5 mM sucrose (Rosenmund and Stevens, 1996).
All electrophysiological data were analyzed with Axograph X, Excel (Microsoft), and Prism (GraphPad).

Image acquisition and quantification
All immunocytochemically stained neurons were imaged on a spinning disk confocal microscope (Zeiss Axio Observer.Z1 with an Andor spinning disk and cobolt, omricron, i-beam laser) using a 40Â (1.3 NA) oil objective and an iXon ultra (Andor) camera controlled by iQ software (Andor). Immunocytochemistry of both time points, 24 and 1 h, was performed at the same time and imaged with the same setting across all conditions. Per condition, 10 neurons were selected based on the MAP2 signal and Zstack (Z = 4) images were taken. Images were processed using ImageJ (RRID:SCR_003070). In brief, maximum projections were created from each Z-stack after which three dendrites per neuron were selected via the MAP2 channel and analyzed with a custom-written script for ImageJ analysis. The script subtracts the background with a rolling stack of 3. Next, it opens Trackmate, a standard ImageJ plugin, wherein blob diameter is set to three and threshold is manually selected to include all puncta in either the GABA A R antibody channel or the VGAT channel. The mean fluorescent values of all detected puncta are then exported to excel file (Microsoft) and analyzed to see whether VGAT and autoantibody spots overlap. Puncta were considered "positive" for a given marker when the mean fluorescent value (au) was bigger than 2Â the background signal. Per experiment, three cultures were stained. Normalized means of intensity in puncta represent the means of all puncta normalized to the in-culture 1-h average intensity, this was done to account for variations in staining intensity between cultures.

Calcium imaging experiments
To visualize calcium influx, cortical-striatal primary co-cultures were infected on DIV 4 with a f(syn)-NES-jRCamP1b-WPRE-w lentivirus which expresses jRCamP1b under the Synapsin promoter. From DIV 7 on, one-third of the media was replaced with fresh NBA-complete every 2-3 d until experiment onset (DIV 14-18). Neurons were imaged using a Nikon Spinning Disk Confocal CSU-X microscope with a 20Â Plan Apo air-objective (NA = 0.8), controlled via the NIS-Elements software (Nikon) in the Charité AMBIO facility. For each imaging session, cells were kept at 37°C and 5% CO 2 during image acquisition and were imaged with a 561 laser, an exposure time of 50 ms, and 300 gain, at 0.2 Hz for 2 min.
To assess long-term effects, auto-antibodies were added 24 h before the onset of imaging at selected concentrations (1 mg/ml for a1-antibody, a1g 2-antibody, or control-antibody, and 5 mg/ml of a1g 2-antibody). For each experiment, three coverslips per condition were selected with three regions of interest (ROIs) per coverslip. For ROIs, we selected areas that had 10-15 neurons that showed spontaneous activity. During the imaging session, all three ROIs were imaged in succession after which all three ROIs were immediately imaged for a second time. During analysis, the averages spike activity of the two recordings was averaged per ROI to ensure a stable activity count over time. Afterward, bicuculline (30 mM, Santa Cruz #sc-202498) was added to the cultures, and after a 1-min waiting period to allow diffusion, the same ROIs were imaged for two consecutive times.
For 1-h and benzodiazepine experiments, antibodies were added 1 h before the imaging session. Subsequently, the three ROIs were imaged twice without any drugs for the 1-h experiments, then twice with diazepam (1 mM, Sigma-Aldrich #D0899) for the benzodiazepine experiments, and finally twice with bicuculline (30 mM, Santa Cruz #sc-202498). For these experiments, cells were initially imaged in 500 ml NBA-complete (2% B27 Invitrogen #17504, 1% glutamax Invitrogen #35050, 0.2% penicillin/streptomycin Roche #11074440001). Afterwards, 500 ml NBA-complete with a 2Â Diazepam concentration was added. After Diazepam imaging, 900 ml was removed from the imaging chamber and 900-ml NBAcomplete with bicuculline was added.
For acute antibody addition, only one ROI per coverslip was imaged. This ROI was imaged for 2 min in 500-ml NBA-complete solution, after which 500 ml of NBA-complete with 2 mg/ml of the a1-antibody or 4 mg/ml of the a1fab-antibody was added, resulting in a 1 or 2 mg/ml final solution, respectively. After the addition of the antibody, the same ROI was imaged for an additional 12 min, to assess its effect on neuronal spiking activity.
After data acquisition, images were analyzed with the help of OpenView software (Noam Ziv, Technion Institute, Haifa, Israel) and a script written by Noam Ziv in Excel (Microsoft). In brief, ROIs were manually selected by placing boxes of 27 Â 27 pixels over 10-15 visually identified neuronal cell somas. Only active cells were included in the analysis. To detect somatic calcium transients, time-series fluorescence values were converted into DF/F by calculating the ratio between the change in fluorescence signal intensity (d F) and baseline fluorescence (F0). The custom-written algorithm identified the timestamps of calcium transient onset, which were then averaged per minute to obtain the frequency of events.

Code accessibility
For the analysis of calcium imaging experiments, we used custom-written software ("OpenView"; available on request; analyses could also be performed using other software packages, such as ImageJ) written by Noam Ziv (Technion Institute, Haifa, Israel). For access to the code please kindly contact Noam Ziv.

Experimental design and statistical analyses
In each figure legend, details about the experimental design are mentioned. Each experiment represents three independent cultures. All statistical tests and graphs were represented with the help of GraphPad Prism (RRID:SCR_ 002798). Information on statistical tests and sample sizes can be found in each figure legend. In general, when comparing two groups, we used paired and unpaired t tests, and in the case of unequal variances, a Welch's correction was applied. When there were more than two conditions, we compared the means via ordinary one-way or repeated measures ANOVA followed by Tukey's multiple comparisons test, unless either the Brown-Forsythe or Barlett's test showed unequal variances. In that case, significance was tested with a Kruskal-Wallis test followed by Dunn's multiple comparisons. For acute calcium imaging experiment, we used repeated measures two-way ANOVA with Sidak's multiple comparisons. Means are reported with SEM. For the sake of readability, we mention only mean 6 SEM and p-values in the text, test statistics and post hoc analysis can be found in the figure legends.

Results
Long-term effects of a1-antibody and a1c2-antibody on GABA A receptor distribution Based on previous studies with patient derived CSF, it is expected that autoantibodies targeting GABA A Rs will affect receptor distribution. To explore whether the a1and a1g 2-monoclonal-antibodies also have a long-term effect on GABA A receptor distribution, we incubated primary cortico-striatal cultures with each of these antibodies for 1 or 24 h, before processing for immunocytochemistry. The intrinsic surface binding of each antibody was initially defined by incubating cultures for 1 h at 15°C with 1 mg/ml of either antibody, while their effects on the redistribution of these receptors was accomplished by incubating cultures with antibody for 24 h at 37°C. Of note, the longer time point was used to mirror the in vivo situation, as performed previously (Moscato et al., 2014;Ohkawa et al., 2014;Petit-Pedrol et al., 2014). Cultures were then fixed and immune-stained with fluorescent secondary antibodies as well as antibodies against the vesicular GABA transporter VGAT (GABA synapse marker) and MAP2 (a dendrite-specific microtubule-associated protein).

a1-antibody but not the a1c2-antibody reduces GABAergic currents after a 24-h treatment
A fundamental question is whether the observed changes in receptor redistribution also led to changes in GABA A R-mediated currents. This was initially investigated by performing whole-cell patch recording of inhibitory striatal neurons incubated for 24 h with each antibody. To look specifically at the effect of the antibody on total and synaptic GABA A R-mediated currents, we used striatal autaptic cultures, in which a single inhibitory striatal neuron is cultured on a microisland of astrocytes (Furshpan et al., 1976;Bekkers and Stevens, 1991;Fig. 3A), causing neurons to make synapses of a single type onto themselves.
For autaptic striatal neurons incubated with the a1-antibody for 24 h, we observed an ;45% reduction in evoked synaptic GABA currents (untreated 1 6 0.10, a1-antibody 0.56 6 0.15, and control-antibody 1.04 6 0.14, p = 0.0005, Kruskal-Wallis; Fig. 3B,C) with significantly longer decay times (untreated 118 6 6 ms, control-antibody 135 6 15 ms, and a1-antibody 186 6 18 ms, p = 0.0205, Kruskal-Wallis; Fig. 3D). When 5 mM GABA was bath applied for 3 s, activating both synaptic and extrasynaptic GABA receptors, GABA currents were reduced by ;40% after a1-antibody preincubation (p , 0.0001, Kruskal-Wallis; untreated 1 6 0.06, a1antibody 0.6 6 0.05, control-antibody 0.99 6 0.07, and a1g 2-antibody 0.96 6 0.09 p = 0.0054; Fig. 3E,F), whereas bath application of 20 mM kainate (Fig. 3G) or 10 mM NMDA (data not shown, p = 0.9990, Kruskal-Wallis; untreated 1 6 0.07, control-antibody 1 6 0.09, a1-antibody 0.98 6 0.07, a1g 2-antibody 1.16 6 0.14) did not significantly change their associated currents. To tease apart whether the observed reduction in GABA currents was because of fewer postsynaptic receptors or because of less presynaptic GABA release, we investigated the sucrose responses, which directly trigger the release of showing synaptic colocalization between VGAT and a1-antibody; scale bar: 10 mm. C, Over time, the a1-antibody leads to reductions in percentage of synapses covered with a1-antibody puncta, t (146.4) = 6.664, p , 0.0001, Welch's t test, while (D) the a1-antibody can be found equally at synaptic and extrasynaptic sites over time, t (121) = 0.691, p = 0.4907, Welch's t test. In addition, (E) long antibody incubation (24 h) leads to reductions in overall a1-antibody intensity, t (148.5) = 4.11, p , 0.0001, Welch's t test, (F) because of reduction in intensity at synaptic sites, t (136.4) = 3.701, p = 0.0003, Welch's t test, (G) and extrasynaptic sites t (129.8) = 2.329, p = 0.0214, Welch's t test. H, Overall, VGAT intensity is also decreased over time, t (156.1) = 2.479, p = 0.0142, Welch's t test. This is also seen in histograms when VGAT intensity is plotted against the a1-antibody intensity at 1 h (I) versus at 24 h (J), where we observed a reduction in the slope of the correlation because of a decrease in a1-antibody intensity. Each data point represents one ROI, except in I and J, where each data point represents a synapse. Error bars represent SEM. docked synaptic vesicles (Stevens and Tsujimoto, 1995) and thus only activates synaptic receptors. The sucrose response revealed that the addition of the a1-antibody was associated with a smaller charge (p = 0.0003, Kruskal-Wallis; untreated 1 6 0.15, a1-antibody 0.40 6 0.08, and control-antibody 1.21 6 0.22; Fig. 3H,I), however with no effect on release probability (Fig. 3J). These data indicate that the antibody induced changes in GABA currents are most likely because of postsynaptic alterations, e.g., change in number of functional postsynaptic GABA A Rs and/or total number of synapses. To further explore these options, we investigated characteristics of mIPSCs, which monitors the response of individual postsynaptic receptor clusters to the release of a single synaptic vesicle. Here, we observed that the a1-antibody treated striatal neurons exhibited a lower mIPSC frequency, 0.46 6 0.14 Hz, compared with cells treated with a control-antibody, 1.42 6 0.32 Hz, however the frequency was not significantly different compared with untreated cells, 0.89 6 0.20 Hz, although it did show a trend toward reduction (p = 0.0149, Kruskal-Wallis; Fig. 3K,L). In addition, a trend toward a decrease in mIPSC amplitude was observed (untreated 34 6 3 pA, control-antibody 36 6 3 pA, a1-antibody 25 6 3 pA; Fig. 3M), but no differences in rise time, half width, or decay time (data not shown) were observed. A reduction in both amplitude and frequency indicates a loss of functional receptors from the synapse, as well as perhaps the loss of functional inhibitory synapses after the 24-h antibody treatment. These conclusions are consistent with the initial observations from our immunocytochemical experiments. Intriguingly, for autaptic striatal neurons treated with 1 mg of the a1g 2-antibody for 24 h, we did not detect any changes in the total or synaptic GABA currents (Fig. 3B,C,E,F), suggesting that this antibody has little or no direct effect on GABA A Rs on these cells.
Cortical-striatal neuron network activity increases after a 24-h treatment with a1-antibody but not the a1c2-antibody Another fundamental question particularly relevant to the etiology of this form of encephalitis is whether such autoantibodies alter the properties of cortical striatal  networks. Specifically, we sought to test the hypothesis that by reducing GABA currents, these antibodies would lead to hyper-excitable network activity, explaining in part why many GABA A RE patients display seizures.
To test this concept, we designed several calcium imaging experiments in which cultured cortical/striatal neurons were initially infected at four DIV with a lentivirus expressing jRCamP1b (under the Synapsin promoter; Dana et al., 2016), a genetically encoded calcium indicator, and allowed to grow for 10 more days (14 DIV) before antibody treatment. Cultures were then treated for 24 h with either a1-antibody, a1g 2-antibody, or control-antibody, or left untreated before being subjected to live imaging to monitor neuronal spiking activity by detecting somatic calcium transients of individual cells. Subsequently, we added bicuculline (30 mM; a GABA A R antagonist) and re-imaged the activity of the same neurons for a within-cell comparison (Fig. 4A,B). This was used to reveal whether additional inhibitory drive is present in the system after antibody treatment.
When we looked at the excitability of the network, we found that the presence of the a1-antibody for 24 h led to a significantly higher spiking frequency compared with both control-antibody and untreated cultures (p , 0.0001, Kruskal-Wallis; untreated 9.58 6 0.47 Hz, a1-antibody 14.03 6 0.66 Hz, control-antibody 9.68 6 0.53 Hz; Fig. 4A, H). This is consistent with its negative impact on GABAmediated currents seen in autaptic striatal neurons (Fig.  3). This finding raised the question how much of the inhibitory drive was still present in the network after 24-h a1antibody treatment. To address this question, we added 30 mM bicuculline to the network. The addition of bicuculline led to an increase in network spiking activity in the untreated and control-antibody conditions (Fig. 4C-G), but had minimal effects on cultures treated with a1-antibody for 24 h (p , 0.0001, ANOVA; untreated 123.5 6 15.01% compared with control, a1-antibody 42.7 6 9.66%, and control-antibody 136.6 6 16.16%; Fig. 4E,I). These data indicate that most available GABA A R receptors are already affected by the a1-antibody, leading to a network with a small remaining inhibitory drive.
In cultures treated with the a1g 2-antibody, we did not observe any effect on neuronal spiking activity in these networks compared with either untreated or control-antibody-treated cultures (Fig. 4H,F,I). Since the a1g 2-antibody appears to have a different antibody avidity compared with the a1-antibody (Kreye et al., 2021), we also investigated whether a higher concentration of this antibody (5 mg/ ml instead of 1 mg/ml) would have an effect. Here again, no change in network spiking was detected (Fig. 4G-I), further supporting the concept that this antibody has no overt direct effects on GABA A Rs or neuronal spiking in these cultured neurons.

a1-antibody elicits rapid effects on GABA A receptor function
The dramatic effect of the a1-antibody on corticalstriatal networks raises fundamental questions of how this antibody changes the functionality of these receptors. Possibilities include direct actions on these receptors and/or their internalization. For some neurotransmitter receptors, antibodies have been found to stimulate their internalization on a time scale of a few hours. For example, antibodies targeting NMDAR, present in the CSF of patients with NMDAR encephalitis, showed reductions in receptor surface levels in the first 2-4 h with internalization peaking after 12 h (Moscato et al., 2014). We thus examined whether changes in GABA-mediated currents by the a1-antibody seen at 24 h could also occur on shorter time scales. This was initially accomplished by performing whole-cell patch recording of mIPSCs in cortico-striatal mass-cultures 1 h after the addition of 1 mg/ml of the a1-antibody. Interestingly, at this time, we observed a dramatic decrease in the frequency (p = 0.0002, ANOVA; untreated 4.62 6 0.41 Hz, a1-antibody 2.10 6 0.41 Hz, and control-antibody 3.90 6 0.50 Hz; Fig. 5A,C), charge (p = 0.0174, ANOVA; control 707.3 6 64.83 fC, a1-antibody 475.5 6 45.19 fC; Fig. 5E) and amplitude (p = 0.0020, ANOVA; untreated 42.97 6 3.24 pA, a1-antibody 27.45 6 2.71 pA, and control-antibody 44.03 6 3.87 pA; Fig. 5B, D) of mIPSCs in cultures treated with the a1-antibody as compared with untreated or control-antibody treated cultures. Of note, there was no observed effect of the a1-antibody on mIPSC kinetics including their rise time, half-width, or decay times ( Fig. 5F-H). Reduction in both frequency and amplitude could indicate a reduction in the number of receptors at synapses, which would cause the mIPSCs to fall below detection levels. Interestingly, the observed effects on mIPSCs were not pronounced after a 24-h antibody incubation (Fig. 3K-M), reflecting possible homeostatic exchange in synaptic GABA A R subunit composition in these neurons. Of note the addition of the a1g 2antibody at 5 mg/ml for 1 h had no detectable effect on mIPSCs frequency and amplitude or kinetics (Fig. 5A-H).
To complement these recordings, we also added these antibodies to mass-cultures and monitored firing frequency by calcium imaging. Here, we found that network excitability was dramatically increased in the presence of the a1-antibody after only 1 h, but not in untreated cultures or those treated with control-antibody (p , 0.0001, Kruskal-Wallis; untreated 13.98 6 1.21 Hz, a1-antibody 32.46 6 2.58 Hz, control-antibody 17.64 6 1.93 Hz, and a1g 2-antibody 5 mg 12.87 6 1.13 Hz p , 0.0001; Fig. 6A, B). This increase in spiking activity was not reversible by washing out of the a1-antibody (Extended Data Fig. 6-1). Importantly, the addition of the a1g 2-antibody at higher concentrations (5 mg/ml) for 1 h had no effect on the network spiking activity (Fig. 6B).  Conceptually, the faster effects observed for the a1-antibody on GABA currents and network activity could be caused by two possible mechanisms. The first is that the a1-antibody has a direct effect on the receptor, either via an allosteric interaction or as a competitive antagonist to GABA. The second is that the antibody promotes receptor crosslinking and triggers internalization. To distinguish between these possible mechanisms, we performed two types of experiments.
In the first, we monitored neuronal spiking activity by calcium imaging before and then continuously for 12 min after the addition of the antibody. Here, we observed a rapid (,5 min) increase in spiking frequency of cells treated with a1-antibody that was sustained during the remaining 12 min of the imaging session (p = 0.0026,   Fig. 6-1). When the a1-antibody is added for only 12 min to cortical-striatal cultures (C, example plots), there is a significant increase in spiking frequency, already after 2 min, compared with other conditions (D) that is sustained for the full 12 min, F (1,29) = 10.82, p = 0.0026, repeated measures two-way ANOVA. The dotted vertical line indicates repeated measures two-way ANOVA; Fig. 6C,D). This increase was not seen for the control-antibody (Fig.  6C,D), suggesting that the antibody has an effect on neuronal firing already after minutes.
To rule out fast internalization through a1-antibodymediated receptor crosslinking, we employed a Fab-fragment from the a1-antibody (Kreye et al., 2021). When 2 mg/ml of the Fab-fragment was added to our cortical/ striatal cultures, we observed a rapid increase in network activity similar to that seen with the a1-antibody, reaching a peak at 4 min after antibody addition (p = 0.0002, repeated measures two-way ANOVA; Fig. 6E,F), which remained elevated for the duration of the imaging session compared with the control-antibody (Fig. 6F). These results strongly hint at a direct effect of the a1-antibody, allosteric or antagonistic, on GABA A R containing this subunit rather than loss of function by rapid receptor internalization.
To explore whether the effects of the a1-antibody are reversible, we performed both calcium imaging and electrophysiological experiments on cortical-striatal cultures incubated for only 1 h with the a1-antibody before being washed out. When neurons were stained for the antibody and VGAT, 6 h after washout, we saw that the antibody was still present on the neurons at synapses (Extended Data Fig. 6-1A). Monitoring spike frequency of untreated and a1-antibody treated cortical-striatal cultures revealed a dramatic increase in firing frequency 6 h after washout (Extended Data Fig. 6-1B,D; p , 0.0001, t test, untreated 12.24 6 0.98 Hz, a1-antibody 36.71 6 1.88 Hz), indicating the antibody remains bound and continues to repress GABA A receptor-mediated inhibition in these networks. To assess whether part of this effect is because of an ongoing block of synaptic GABA A receptor, we recorded the frequency, amplitude, charge and kinetics of the mIPSC 6 h after washout. Here, we also observed a remarkable decrease in mIPSC frequency (p = 0.0003, t test, untreated 3.35 6 0.36 Hz, a1-antibody 1.56 6 0.30 Hz), amplitude (p , 0.0001, Welch's t test, untreated 36.65 6 2.86 pA, a1-antibody 21.93 6 1.68 pA) and charge (p , 0.0001, t test, untreated 655.8 6 53.81 fC, a1antibody 345.6 6 35.69 fC; Extended Data Fig. 6-1E-H). We also detected a decrease in the decay time (p = 0.0012, t test, untreated 37.57 6 1.38 ms, a1-antibody 29.07 6 2.07 ms), half-width of these minis (p = 0.0184, t test, untreated 13.35 6 0.46 ms, a1-antibody 11.26 6 0.73 ms), as well as an increase in-rise time (p = 0.0057, Welch's t test, untreated 0.72 6 0.03 ms, a1-antibody 1.2 6 0.16 ms; Extended Data Fig. 6-1I-K). Together, these data indicate that the a1-antibody after washout remains bound to receptors and continues to repress synaptic GABA A receptor-mediated responses as well as the inhibitor drive in these networks. The change in kinetics of these minis suggests either direct effects of the a1-antibody on a1-containing receptors or the emergence/ unmasking of non-a1-containing receptors with different kinetic properties at these synapses.

a1c2-antibody does not affect GABA A receptor function by interference with their benzodiazepine site
In the experiments presented thus far, we failed to detect any remarkable effects of the a1g 2-antibody on GABA currents and/or distribution of these receptors in striatal neurons. This lack of response is even more interesting in the light of a previous study wherein the a1g 2antibody exhibited a strong epileptogenic effect in rodents when the antibody was intraventricularly delivered via an osmotic Alzet pump (Kreye et al., 2021). This raised the question as to how this antibody can affect neuronal excitability in vivo but not in the used in vitro systems. As the a1g 2-antibody binds an epitope at the a1 and g 2 subunit interface, it is conceptually possible that this antibody interferes with the binding of endogenous benzodiazepine like compounds (e.g., endozepines), known to act at a site situated between the a and g subunit of GABA receptor subtypes containing a1, a2, a3, and a5 subunits. To explore this possibility, we recorded mIPSCs in mass-cortical-striatal co-cultures, 1 h after the addition of the a1 g 2-antibody in the presence and absence of the benzodiazepine, diazepam. We chose to add the diazepam after 1 h of antibody incubation because this way we could measure whether the a1g 2-antibody would silently bind the GABA A R and thereby prevent the diazepam from binding, rather than a 24-h time point at which there could be confounding effects caused by subunit substitution or similar homeostatic effects. When mIPSCs were recorded in untreated neurons, the half-width times and the charge were observed to increase, as expected, in the presence of the diazepam (p , 0.0001, paired t test; untreated base 12.27 6 0.49 ms, untreated benzo 14.46 6 0.57 ms; Fig. 7A,C). In cultures treated with either the a1g 2-antibody, a1-antibody, or control-antibody, diazepam was also able to increase the half-width times of mIPSC [p , 0.0001, paired t test; control-antibody base 12.77 6 0.57 ms, control-antibody benzo 15.74 6 0.66 ms (Fig. 7D); p = 0.0002, paired t test; a1-antibody base 12.69 6 0.82 ms, a1-antibody benzo 15.08 6 0.75 ms (Fig. 7E); p , 0.0001, paired t test; a1g 2-antibody base 13.14 6 0.67 ms, a1g 2-antibody benzo 15.8 6 0.61 ms (Fig. 7F)]. When we compared the relative increase in mIPSCs charge or half-width (ratio of after/before addition of diazepam), we see that all conditions respond in a similar manner to the addition of diazepam (Fig. 7B). These results indicate that the actions of diazepam are not impaired by either the a1g 2-antibody or a1-antibody and that neither directly interferes with the benzodiazepine binding site.
To further investigate possible effects of these antibodies on the actions of benzodiazepine at a network level, we performed calcium imaging experiments, 1 h after continued when antibody is added to the culture. Addition of an Fab fragment from the a1-antibody (E) similarly induces an increase in spike frequency that keeps rising up till 4 min after addition, even higher than the full-length antibody, and then slowly declines, although not all the way back to control level over the course of 12 min (F), F (1,33) = 18.27, p = 0.0002, repeated measures two-way ANOVA. Each data point represents the average spiking activity of one field of view. Error bars represent SEM. antibody incubation. With the addition of 1 mM diazepam, the spiking activity of a1g 2 treated cultures went down by À72.12 6 5.05% (Fig. 8B,H) compared with cultures to which only control-antibody was added (Fig. 6B). This effect was also seen in cultures treated with the a1-antibody. However, here the average firing frequency remained higher, with only a À48.32 6 7.49% decrease compared with baseline (p = 0.0021, Kruskal-Wallis; untreated À86.29 6 3.37%, control-antibody À76.37 6 5.46%; Fig. 8B,H). As expected, adding 30 mM bicuculline triggered an increase in spiking activity for all four conditions compared with spiking activity during diazepam incubation (8D-G). Interestingly, cultures treated with the a1g 2-antibody returned to spiking levels (262.2 6 40.68%) similar to untreated (205.5 6 24.71%) and control-antibody (190.7 6 29.85%), whereas the a1-antibody showed a smaller increase in activity (82.54 6 15.76%) compared with untreated, a1g 2-antibody and control-antibody conditions (p , 0.0001, Kruskal-Wallis; Fig. 8I). These results indicate that the increases in neuronal excitability in vivo (Kreye et al., 2021) associated with the addition of a1g 2-antibody is not necessarily due its ability to block the benzodiazepinebinding-site of GABA A receptors containing either a1 or a1g 2 subunits. However, this conclusion requires some caution as the continued effect of diazepam on these antibody-treated cultures could be because of other benzosensitive GABA A receptors containing, e.g., g 2 and a2, a3, or a5 subunits. Clearly, single channel recording would be necessary to resolve this issue.

Discussion
Autoimmune encephalitis is a devastating disorder associated with antibodies targeting neural-specific antigens. While these antibodies target one primary antigen, evidence exists that polyclonal antibodies react to multiple targets/receptors with different subunit compositions in a single patient (Kreye et al., 2021). This complexity makes it difficult to define causality between a specific antibody and a patient's symptoms, and/or whether they operate via similar or distinct modes of action. To bring some clarity to these issues, we have characterized the molecular and functional modes of action of two recently isolated monoclonal antibodies (Kreye et al., 2021) from a GABA A RE patient (Nikolaus et al., 2018), one targeting the a1 subunit and another one targeting the a1g 2 subunits  Here, diazepam leads to a ;75% reduction in spiking activity compared with cultures without diazepam (Fig. 6B), with the exception to the a1-antibody condition where the reduction was significantly smaller (;50%) H (3) = 14.67, p = 0.0021, Kruskal-Wallis; untreated À86.29 6 3.37% versus a1-antibody À48.32 6 7.49% p = 0.0024, control À76.37 6 5.46% versus a1-antibody p = 0.0148. I, Quantification of percent changes in spiking activity following the addition of 30 mM bicuculline. This leads to an ;200% increase in network activity compared with baseline ( Fig. 6B) for all groups except for a1-antibody condition, where the increase is only 80%, H (3) = 22.77, p , 0.0001, Kruskal-Wallis; untreated versus a1-antibody p = 0.0007, control versus a1-antibody p = 0.0141, a1-antibody versus a1g 2-antibody p , 0.0001. Error bars represent SEM.
of GABA A Rs. The former was found to have long-term effects on receptor distribution and fast inhibitory effects on GABA A R function that did not require internalization by antibody cross-linking. Intriguingly, we could detect no overt functional effects on GABA A Rs by the a1g 2-antibody, although in vivo, it was previously found to elicit seizure activity (Kreye et al., 2021), implying it may have indirect effects on receptor function via, e.g., microglia. These observations highlight diverse modes of action of autoantibodies, contributing mechanistically to autoimmune encephalitis.

Multimodal effects of the a1-antibody
Our analysis of the a1-antibody indicates that it acts on several levels of GABA A R function, including both longterm and short-term effects. Long-term addition of the a1-antibody (;24 h) to primary cortical/striatal co-cultures promotes a loss of immune-positive receptors from GABAergic VGAT-labeled synapses (Fig. 1). We found that under basal conditions ;80% of GABAergic synapses between cortical/striatal neurons possess a1-containing GABA A Rs. This was reduced to 50% following 24-h antibody incubation. Autoantibody-mediated redistribution of receptors has also been reported for other antigen/receptors (Prüss, 2021) and is consistent with antibody induced movement of receptors away from the synapse and/or their internalization (Ohkawa et al., 2014;Petit-Pedrol et al., 2014;Pettingill et al., 2015).
Using whole-cell patch recording of striatal autaptic neurons, we also explored the functional impact of the a1-antibody on both synaptic and extrasynaptic GABA A Rmediated inhibitory currents following a 24-h treatment.
Here, we observed a dramatic decrease in total inhibitory GABA currents, with no effect on excitatory currents (Fig.  3). This was associated with a remarkable decrease in evoked IPSCs as well as an increase in decay kinetics. These latter data are consistent with loss of synaptic a1-GABA A Rs and a possible change in synaptic subunit composition. Sucrose and mIPSC recordings support these conclusions, as they also reveal a decrease in total inhibitory synaptic currents (sucrose response) and number of functional synapses (frequency) with no change in mIPSC amplitude of the remaining synaptic receptors (Fig. 3).
To distinguish between the real-time effects of the a1antibody and homeostatic changes such as receptor subunit composition, we assessed changes in synaptic GABA A R function following a shorter 1-h incubation with cortical/ striatal co-cultures. Here, we observed a dramatic decrease in mIPSC frequency, amplitude, and charge ( Fig. 5C-E) with no change in rise time, half-width, or decay times (Fig. 5F-H), indicating a possible direct inhibition of GABA A Rs. These latter data further support the concept that long-term exposure of neurons to the a1-antibody triggers homeostatic changes to neurons, including possible changes in receptor subunit composition to a2-containing, a3-containing, or a5containing receptors. Of note these other GABAAR subunits have slower decay kinetics than the a1 subunit (Gingrich et al., 1995). As such redistribution of a1-containing receptors away from the synapse at later time points could account for the observed increase in decay times after the 24-h antibody incubation, which have yet to occur after 1 h of incubation. Interestingly, changes in receptor subunit composition are also observed in networks after long-term exposure to allosteric agonists such as benzodiazepine (Vinkers et al., 2012).
Further hints to the mode of impact of the a1-antibody came from examining its effects on network spiking activity of cortical-striatal co-cultures using calcium imaging. Consistent with its ability to decrease GABA mediate currents, we observed dramatic increases in spiking frequency following the addition of the a1-antibody for 1 or 24 h (Figs. 4, 6A,B). Another interesting observation is that after the 1-h treatment, these networks were less sensitive to benzodiazepine regulation. This is also observed in GABA A RE patients, who generally present with status epilepticus (O'Connor et al., 2019) that does not respond to treatment with benzodiazepines (Rossetti and Lowenstein, 2011). Remarkably, when the a1-antibody was added acutely to the network, we could detect real-time changes in neuronal spiking activity in as little as 2-3 min (Fig. 6C,D), indicating that this antibody can directly modify a1-containing GABA A Rs function in a fast manner. This appears to be independent of a fast antibody-mediated cross-linking and receptor internalization as the acute application of a noncross-linking Fab-fragment of the a1-antibody also triggered a realtime increase in spiking activity in ;2 min ( Fig. 6E,F) that is not quickly washed out (Extended Data Fig. 6-1). Future, biophysical studies and pHluorin-based imaging of tagged receptors will be necessary to resolve these options.
Taken together, these data indicate that the a1-antibody can function on multiple levels to affect excitatory/ inhibitory balance within neuronal networks of patients with GABA A RE. The first appears to operate in real time to alter the functionality of individual receptors, while a second acts on a slower time scale possibly by removing synaptic receptors through antibody-mediated receptor endocytosis. While not examined in detail here, previous studies have explored the role of autoantibodies on receptor internalization. In particular, it has been shown that IgG1-isotype autoantibodies can lead to cross-linking of receptors causing their internalization (Crisp et al., 2019;Dalmau and Graus, 2022). For example, it was shown that NMDAR IgG1-type auto-antibodies promote loss of synaptic receptors in ;4-h, reaching maximal internalization rates by 12 h (Moscato et al., 2014). Similarly, glycine receptor (GlyR) autoantibodies promoted the endocytosis of receptor in ;2 h in HEK293 cell and 16 h in motor neurons (Crisp et al., 2019). How much the latter applies to the actions of the a1-antibody is unclear. Our own results revealed a dramatic loss of synaptic receptors following a 24 h antibody incubation, indicating that antibody-mediated endocytosis could contribute to the long-term effects of these antibodies, similar to both GlyR and NMDARs (Kreye et al., 2016;Crisp et al., 2019).
The rapid effects of the a1-antibody, modifying GABA A receptor function within minutes, is not unique and has also been observed for IgG1 autoantibodies targeting inhibitory GlyR and GABA B receptors (Nibber et al., 2017;Dalmau and Graus, 2022). For example, Fab fragments of antibodies targeting the GlyRs could reduce mIPSC frequency in as little as 15 min following its addition to dissociated spinal cord cultures (Crisp et al., 2019). Mechanistically, how might the a1-antibody elicit such rapid changes in GABA A R function? One possibility is that it triggers a conformational change that promotes rapid receptor internalization. Alternatively, it could act on GABA binding sites, as a competitive antagonist, similarly to gabazine (Johnston, 2013;Ghit et al., 2021). Ultrastructural epitope mapping of the a1-antibody binding site and initial electrophysiological recording from transfected cells support this concept (Noviello et al., 2022), yet more detailed biophysical studies are needed to define the actual molecular mechanism.

a1c2-antibody, an unresolved puzzle
Initial studies on the a1g 2-antibody suggested that it could elicit strong neuronal network dysfunction since P21 rats receiving a cerebroventricular infusion showed increases in ictal events similar to the a1-antibody (Kreye et al., 2021). Thus, our initial expectation was that similar to the a1-antibody, the a1g 2-antibody would alter GABA A R distribution and/or functional properties. However, although this antibody decorated ;40% of GABAergic VGAT-positive synapses, there was no significant redistribution of these receptors after a 24-h treatment (Fig. 2). If anything, this antibody accumulated, which may be related to its low affinity (Kreye et al., 2021;Fig. 2D-F). The reason why this antibody only decorated 40% of the synapses could be because of the relative low expression of a1g 2containing receptors in the striatum which is around 50% (Boccalaro et al., 2019). However, low labeling because of technical reasons cannot completely be ruled out. Surprisingly, incubating striatal autaptic neurons for 24 h did not affect total GABA-mediated currents (Fig. 3F), evoked IPSC amplitudes (Fig. 3C), or mIPSC frequency and amplitude ( Fig. 3K-M). Furthermore, increasing the concentration of the a1g 2 antibody to 5 mg/ml did not affect spiking frequency of cortical-striatal cultures compared with control-antibody or 1 mg/ml a1g 2 both after a 24-or 1-h incubation (Figs. 4, 6A,B, respectively). Together these data raise a fundamental question of how such an antibody could increase ictal events when infused into the brains of rats (Kreye et al., 2021), but not cortical-striatal neurons in cultures.
Two major differences between these two preparations exist: (1) the absence of naturally occurring allosteric modulators, such as endozepines or neurosteroids, which normally enhance GABA A R function to maintain E/I balance (Olsen, 2015) as well as (2) microglia. Regarding the former, the ability of the a1g 2-antibody to bind both subunits hints that this antibody might occupy the benzodiazepine binding site. However, we failed to detect changes in the response to diazepam, following a1g 2-antibody incubation (Figs. 7,8). This could be because of a mixture of benzo-sensitive receptor-subtypes on these cells and/ or that the a1g 2-antibody blocks the action of other modulators not investigated in this study. Future detailed biophysical studies of single-channel conductance of a1g 2-containing GABA A Rs could help resolve this issue.
Although seldom considered as a mechanism associated with autoimmune encephalitis, there is a growing appreciation that microglia and cellular immunity play important roles in the etiology of a number of brain disorders with focal lesions, epilepsy and cognitive decline, including Alzheimer's disease (Sarlus and Heneka, 2017). Here, it is worth noting that both the a1 and a1g 2-antibodies are of the human IgG1 (hIgG1) subtype with binding sites for both complement and Fc-g receptors (Bournazos et al., 2017). This raises the possibility that microglia in situ could actively strip hIgG1-antibodies/GABA A receptor complexes from the surfaces of neurons, increasing network excitability as a consequence of reduced inhibitory drive, a mechanism worth exploring.

General directions for autoimmune encephalitis
This study demonstrates that pathogenic autoantibodies can act through many mechanisms within the same disease or sometimes even within the same autoantibody and highlights the complexity underlying GABA A R encephalitis and related diseases. It is becoming clear that these mechanisms can range from acute effects on receptors, through long-term receptor loss, to silent binding and possible recruitment of immune system components. This could also mean that each patient has a unique disease signature based on the combination of autoantibodies detected in CSF. This complexity makes it challenging to implement new treatment strategies other than general immunosuppressants; however, further understanding of the type of autoantibodies present in patients and their modes of action could lead to a more individualistic approach to treatment.