Distinctive binding properties of human monoclonal LGI1 autoantibodies determine pathogenic mechanisms

Abstract

Autoantibodies against leucine-rich glioma inactivated 1 (LGI1) are found in patients with limbic encephalitis and focal seizures. Here, we generate patient-derived monoclonal antibodies (mAbs) against LGI1. We explore their sequences and binding characteristics, plus their pathogenic potential using transfected HEK293T cells, rodent neuronal preparations, and behavioural and electrophysiological assessments in vivo after mAb injections into the rodent hippocampus. In live cell-based assays, LGI1 epitope recognition was examined with patient sera (n = 31), CSFs (n = 11), longitudinal serum samples (n = 15), and using mAbs (n = 14) generated from peripheral B cells of two patients. All sera and 9/11 CSFs bound both the leucine-rich repeat (LRR) and the epitempin repeat (EPTP) domains of LGI1, with stable ratios of LRR:EPTP antibody levels over time. By contrast, the mAbs derived from both patients recognized either the LRR or EPTP domain. mAbs against both domain specificities showed varied binding strengths, and marked genetic heterogeneity, with high mutation frequencies. LRR-specific mAbs recognized LGI1 docked to its interaction partners, ADAM22 and ADAM23, bound to rodent brain sections, and induced internalization of the LGI1-ADAM22/23 complex in both HEK293T cells and live hippocampal neurons. By contrast, few EPTP-specific mAbs bound to rodent brain sections or ADAM22/23-docked LGI1, but all inhibited the docking of LGI1 to ADAM22/23. After intrahippocampal injection, and by contrast to the LRR-directed mAbs, the EPTP-directed mAbs showed far less avid binding to brain tissue and were consistently detected in the serum. Post-injection, both domain-specific mAbs abrogated long-term potentiation induction, and LRR-directed antibodies with higher binding strengths induced memory impairment. Taken together, two largely dichotomous populations of LGI1 mAbs with distinct domain binding characteristics exist in the affinity matured peripheral autoantigen-specific memory pools of individuals, both of which have pathogenic potential. In human autoantibody-mediated diseases, the detailed characterization of patient mAbs provides a valuable method to dissect the molecular mechanisms within polyclonal populations.

Keywords: LGI1; antibody; autoimmune; encephalitis; immunology.

Figure 1

Patient serum and CSF IgG binding to full-length LGI1, and to its LRR and EPTP domains in live cell-based assays. (A) Schematic representation of human full-length (fl) LGI1, plus LRR and EPTP domain constructs. All have an N-terminal signal peptide (SP) and C-terminal transmembrane domain (TM) plus intracellular EGFP. (B) Sera from 31 patients (20 male; age range onset 46–86, median 63 years), plus 11 CSFs, were analysed alongside 31 age-matched healthy controls (HC). While all patient sera bound both domains of LGI1, typically at similar levels, two CSF samples bound only to LRR (blue) or EPTP (red) domains. Unavailable CSF samples are indicated by an ‘x’. Fluorescent images show a live cell-based assay with Patient 1’s serum IgG binding (hu IgG, red) to both the (EGFP-tagged) LRR and EPTP domains of LGI1. DAPI was used as a nuclear marker. Scale bar = 10 μm. (C) Overall, serum end-point titres to full-length LGI1 correlated strongly with the LRR- and with EPTP domain-specific antibody levels (Spearman’s r = 0.88, P < 0.0001). (D) Ratios of serum LRR- and EPTP-antibody end-point titres over time are shown in 15 patients, treated with corticosteroids (n = 14), intravenous immunoglobulins (n = 8), plasma exchange (n = 5), mycophenolate mofetil (n = 3), azathioprine (n = 3), rituximab (n = 1), methotrexate (n = 1), cyclophosphamide (n = 1), thymectomy (n = 1) or no therapy (n = 1). The majority of patients had a constant relative ratio of LRR:EPTP antibody titres. The main outlier had an 8-fold higher EPTP serum end-point titre, which decreased after immunotherapy and cessation of seizures.

Figure 2

Epitope specificities and mutation frequencies of patient-derived LGI1 mAbs. (A) Unlike serum, 14 LGI1 mAbs generated from two patients were specific for either LRR or EPTP domains. (B) Median relative binding strengths (K_d), measured by quantification of IgG binding to a live cell-based assay expressing full-length LGI1, were comparable between LRR- and EPTP-specific mAbs. (C) Heat maps of two-way blocking experiments for both LRR- and EPTP-binding mAbs. (D) From C, LRR-specific mAbs clustered into two groups (1 and 2, blue), with cross-competition across mAbs derived from the two patients. EPTP-specific mAbs cross-competed with each other (Group 3, red), except mAb12 did not block mAb14. Three mAbs showed no blocking (non-blockers, yellow). (E) Sequences were aligned in IgBLAST against the IMGT reference database. Total and ratios of replacement:substitution (R:S) V region mutations were similar between EPTP- and LRR-binding mAbs.

Figure 3

LRR-, but not EPTP-, specific mAbs bind docked sLGI1. (A) By contrast to EPTP-specific mAbs, 7/9 LRR-specific mAbs recognized LGI1 after it bound the surface ADAM22 or ADAM23 expressed on HEK293T cells (top two rows), and to the surface of live hippocampal neurons (second last row). They did not directly bind either ADAM22 or ADAM23 without sLGI1 (third row and Supplementary Fig. 1C). In hippocampal neuron cultures, binding was abolished after preadsorption of LRR-specific mAbs with HEK293T cells expressing membrane-tethered full-length LGI1 (bottom row), but not after preadsorption with untransfected HEK293T cells (not shown). (B) Rat brain immunohistochemistry shows an example of sagittal whole brain and hippocampal staining seen in 7/9 LRR-specific mAbs (top, mAb09 shown). This was absent for 3/5 EPTP-specific mAbs (middle, mAb12 shown) and for a human isotype-control mAb (Ctrl mAb, bottom). Similar results were obtained using different fixation methods (paraformaldehyde, formalin, acetone; not shown). All mAbs were tested at ≥5 μg/ml, images at 10 μg/ml. (C) Normalized end-point titres of 14 mAbs are shown across all these four detection methods. Using live cell cultures, binding was absent for all EPTP-specific mAbs and 2/9 LRR-directed mAbs (mAb05 and mAb10; open symbols). On rat brain sections, 2/5 EPTP-specific mAbs bound weakly compared to LRR-specific mAbs (e.g. D), but with comparable end-point titres (mAb11 binds at 100 ng/ml, mAb13 at 12.5 ng/ml). Representative data from one of two experiments are shown. Medians were compared using Mann-Whitney test *P < 0.05, **P < 0.01. (D) The 7/9 LRR-specific mAbs (top, mAb08 shown) and 2/5 EPTP-specific mAbs (bottom, mAb13 shown), which stained brain tissue from wild-type mice showed no binding to Lgi1-knockout mouse sections (at 3 μg/ml). Scale bars = 10 μm in A; 1 mm in B and D.

Figure 4

LRR-specific mAbs internalize LGI1 and its receptors. (A) LRR-specific mAbs caused internalization of sLGI1-ADAM22/23 complexes on HEK293T cells at 37°C after 4 h (top: example with ADAM23 staining). Bottom: An internalized pHrodo-labelled LRR-specific mAb. Images were similar to ADAM22 (not shown). (B) Gating strategy and quantification of pHrodo-labelled mAb uptake (5 μg/ml) by flow cytometry after 4 h incubations at 37°C. (C) The percentage of HEK293T cells with pHrodo fluorescence and the fold increase in pHrodo median fluorescent intensity are shown, after incubation with whole mAbs and Fab' fragments (both at 5 μg/ml). Medians of two to four experiments are shown (Mann-Whitney test **P < 0.01). (D) Corresponding decrease of surface-bound IgG over time. Graphs summarize the per cent of baseline surface-bound human IgG on sLGI1-ADAM22/23 expressing HEK293T cells after 0.5 and 4 h at 37°C, and after 4 h at 4°C. Data from each time point were compared to their own control values at baseline, and are shown across two experiments using seven LRR-directed mAbs [box plots with median, 25th and 75th percentiles, whiskers indicate 10th and 90th percentiles; repeated measures one-way ANOVA (plus Bonferroni correction); *P < 0.05, **P < 0.01, ***P < 0.001]. (E) Quantification of pHrodo fluorescence from conjugated mAbs and Fab′ (5 μg/ml) by flow cytometry with and without dynasore. Medians of two experiments are shown (Mann-Whitney test *P < 0.05, ***P < 0.001). (F) LRR-specific mAbs were internalized on live hippocampal neurons at 37°C after 96 h. Top: Extracellular (ec), surface human IgG was detected before cell permeabilization (green) and after cell permeabilization [extracellular plus intracellular (ic), red; arrow and inset]. Middle: ADAM23 staining was observed throughout the neuron (as Ohkawa et al., 2013), and the internalized LRR-directed mAb co-localized with ADAM23 (inset and arrow). ADAM22-directed commercial antibodies revealed no binding in these cultures, consistent with 10-fold lower quantities of ADAM22 mRNA compared to ADAM23 mRNA (qPCR, data not shown). Bottom: Fluorescence and bright field images of hippocampal neurons after 96-h incubation with pHrodo-conjugated LRR-directed mAbs (1 μg/ml). (G) Quantification shows the number of pHrodo-positive somatic clusters and relative fluorescence intensities per image [data obtained from nine images (three per well) per condition repeated in two separate cultures]. LRR-directed antibodies formed more somatic pHrodo-positive clusters and fluoresced more intensely than EPTP mAbs, two of which bound rodent brain sections (medians were compared using Mann-Whitney test: not significant, but P = 0.02 for pHrodo-positive somatic clusters if only considering neuron-binding LRR-specific mAbs, filled symbols). Two EPTP antibodies were not tested. (H) Representative images of internalized pHrodo-labelled mAb02 (LRR) and its corresponding Fab′ fragments (1 μg/ml) in hippocampal neurons. Arrows indicate somatic clusters. For C and G, open blue symbols identify the two LRR-specific mAbs that did not bind hippocampal neurons. Scale bars = 10 μm.

Figure 5

EPTP-specific mAbs block the interaction of LGI1 with its receptors. (A) Human sLGI1 was preincubated with mAbs, and sLGI1 binding was detected on the surface of ADAM22/23-expressing HEK293T cells with a fluorescently-labelled LRR-specific mAb (mAb02 shown; to assess LRR-specific mAbs within cross-blocking Group 1, a fluorescently-labelled LRR-specific mAb from Group 2 was used). Preincubation with increasing concentrations of all EPTP-specific mAbs (gradient depicted in left four panels), but none of the LRR-specific mAbs (right-most panel), resulted in a complete loss of fluorescence. (B) Individual minimum titres of EPTP-directed mAbs required to achieve complete loss of fluorescence are shown for both ADAM22 and ADAM23-transfected HEK293T cells. No blocking was achieved with LRR-directed mAbs (data from one of three representative experiments are shown; medians were compared using Mann-Whitney test ***P < 0.001). (C) Representative images showing EPTP-specific, but not LRR-specific, Fab′ fragments (200 ng/ml) blocked the interaction of sLGI1 with ADAM23 in transfected HEK293T cells. (D) Representative images showing that preincubation of hippocampal neurons with an excess of each EPTP-specific mAb individually reduced LGI1 internalization, as visualized with a pHrodo-conjugated LRR-specific mAb (mAb01). Preincubation with an excess of the unconjugated LRR-specific mAb (mAb01) was used as a negative control and displaced all of the observed pHrodo-conjugated mAb01 internalization (not shown). (E) Quantification of pHrodo-positive neurons, using all five EPTP-specific mAbs, two EPTP-specific Fab′ fragments and negative controls (anti-A33 mAb, three healthy control IgG preparations and no IgG); representative data obtained from nine images (three per well) per condition repeated in two separate cultures; medians were compared using Kruskal Wallis test (plus Dunn's correction) *P < 0.05. Scale bars = 10 μm.

Figure 6

Single LGI1 mAbs are sufficient to cause pathogenicity in vivo. (A) Timeline for stereotactic injections, electrophysiology studies and behavioural assessments. (B) Representative staining of human IgG on coronal sections of animals injected unilaterally with LGI1 mAbs to demonstrate the distribution in the hippocampus. Both mAb specificities were absent from the uninjected, contralateral hippocampus, akin to control mAb injections (inset). While LRR mAbs were strongly retained in the injected hippocampus, EPTP mAbs showed only limited binding to this tissue. (C) Cell-based assays expressing membrane-tethered full-length LGI1 showed the presence of EPTP mAb (all animals tested, mAb13 shown), but not any of the LRR mAb (all animals tested, mAb02 shown) in the serum of injected animals. (D) Protein quantification and representative western blots showing a downregulation of K_v1.1α in hippocampus enriched solubilized brain lysates after injection with LRR- and EPTP-specific mAbs [LRR mAbs tested: mAb02 (n = 6), mAb06 (n = 2), mAb08 (n = 2); EPTP mAbs tested: mAb11 (n = 2), mAb12 (n = 2), mAb13 (n = 6); Ctrl mAb (n = 6)]. Data were normalized to the control mAb band on each blot to correct for inter-experimental variability. (E) Novel object recognition testing showed impaired memory in animals injected with either mAb specificity (LRR mAbs tested: mAb02, mAb06, mAb08; EPTP mAbs tested: mAb11, mAb12, mAb13; Ctrl mAb; n = 8 animals per mAb). The effect was significant only for LRR mAbs with high relative binding strengths (mAb02 and mAb06). (F) Representative traces from recordings of fEPSPs in hippocampal area CA1 following stimulation of Schaffer collaterals pre and post theta-burst stimulation (TBS). (G) Average traces showing LTP following TBS (arrow). Both mAb specificities prevented induction of LTP (control mAb: n = 6; mAb02 and mAb13: n = 5 animals per group). (H) Median fEPSP slope 50–60 min after TBS. All data shown as box plots with median, 25th and 75th percentiles, whiskers indicate 10th and 90th percentiles; Kruskal-Wallis test (plus Dunn's correction) *P < 0.05, **P < 0.01, ***P < 0.001, ****P < 0.0001. Scale bars = 1 mm in B and 10 μm in C.