Early B cell tolerance defects in neuromyelitis optica favour anti-AQP4 autoantibody production

Abstract

Neuromyelitis optica spectrum disorders (NMOSD) constitute rare autoimmune disorders of the CNS that are primarily characterized by severe inflammation of the spinal cord and optic nerve. Approximately 75% of NMOSD patients harbour circulating pathogenic autoantibodies targeting the aquaporin-4 water channel (AQP4). The source of these autoantibodies remains unclear, but parallels between NMOSD and other autoantibody-mediated diseases posit compromised B cell tolerance checkpoints as common underlying and contributing factors. Using a well established assay, we assessed tolerance fidelity by creating recombinant antibodies from B cell populations directly downstream of each checkpoint and testing them for polyreactivity and autoreactivity. We examined a total of 863 recombinant antibodies. Those derived from three anti-AQP4-IgG seropositive NMOSD patients (n = 130) were compared to 733 antibodies from 15 healthy donors. We found significantly higher frequencies of poly- and autoreactive new emigrant/transitional and mature naïve B cells in NMOSD patients compared to healthy donors (P-values < 0.003), thereby identifying defects in both central and peripheral B cell tolerance checkpoints in these patients. We next explored whether pathogenic NMOSD anti-AQP4 autoantibodies can originate from the pool of poly- and autoreactive clones that populate the naïve B cell compartment of NMOSD patients. Six human anti-AQP4 autoantibodies that acquired somatic mutations were reverted back to their unmutated germline precursors, which were tested for both binding to AQP4 and poly- or autoreactivity. While the affinity of mature autoantibodies against AQP4 ranged from modest to strong (K_d 15.2–559 nM), none of the germline revertants displayed any detectable binding to AQP4, revealing that somatic hypermutation is required for the generation of anti-AQP4 autoantibodies. However, two (33.3%) germline autoantibody revertants were polyreactive and four (66.7%) were autoreactive, suggesting that pathogenic anti-AQP4 autoantibodies can originate from the pool of autoreactive naïve B cells, which develops as a consequence of impaired early B cell tolerance checkpoints in NMOSD patients.

neuromyelitis optica spectrum disorder (NMOSD), B cells, autoantibodies, AQP4, tolerance

Introduction

Neuromyelitis optica (NMO) and neuromyelitis optica spectrum disorders (NMOSD) are autoimmune demyelinating diseases of the CNS (Wingerchuk et al., 2007; Hinson et al., 2016). Common manifestations include recurrent episodes of optic neuritis and longitudinally extensive transverse myelitis leading to visual loss and paralysis (Wingerchuk et al., 1999). Clinically, these symptoms can be difficult to distinguish from multiple sclerosis, creating challenges in differential diagnosis of the two diseases (Wingerchuk et al., 2015). A major advance that helped to distinguish NMOSD from multiple sclerosis was the discovery that a majority of patients with NMOSD (∼75%) (Waters et al., 2014) have detectable serum IgG autoantibodies that target the aquaporin-4 water channel (AQP4) on astrocytes (Lennon et al., 2004, 2005). These autoantibodies are more than 99% specific for clinically diagnosed NMOSD. Principal features of disease pathology can be reproduced both in vitro and in vivo using patient-derived monoclonal antibodies or immunoglobulin, thus confirming the pathogenic contribution of these autoantibodies to CNS injury (Bennett et al., 2009; Saadoun et al., 2010). While the importance of anti-AQP4 autoantibodies (Saadoun et al., 2010) and the characteristics of AQP4-specific B cells (Kowarik et al., 2015, 2017) have been investigated, questions regarding the development of autoantibody-producing B cells remain outstanding (Bennett et al., 2015). Parallels between NMOSD and other autoantibody-mediated diseases, such as myasthenia gravis, posit compromised B cell tolerance as a potential common underlying and contributing factor (Lee et al., 2016).

During early B cell development, B cell precursors undergo random rearrangements of their variable (V), diversity (D), and joining (J) immunoglobulin gene segments. Random V(D)J recombination generates a diverse antibody repertoire but inevitably also generates potentially harmful clones displaying self-reactivity (Wardemann et al., 2003). In healthy individuals, these self-reactive clones are removed to prevent potential development of autoimmunity (Meffre, 2011). Accordingly, B cells that express autoreactive B cell receptors (BCRs) are eliminated from the maturing repertoire by tolerance mechanisms present at two distinct checkpoints along the B cell development pathway, thereby potentially preventing the recognition of self-antigens and presentation of self-peptides to T cells. The first selection step is a central tolerance checkpoint in the bone marrow that removes self-reactive clones between the early immature and immature stages, prior to entering the periphery (Wardemann et al., 2003). Remaining autoreactive new emigrant/transitional B cells that migrate into the periphery are further selected against by the peripheral tolerance checkpoint before transitioning into mature naïve B cells (Meffre and Wardemann, 2008). A number of primarily non-neurological autoimmune diseases, such as rheumatoid arthritis, systemic lupus erythematosus, type 1 diabetes, and Sjögren’s syndrome have been shown to harbour defects in these tolerance checkpoints (Samuels et al., 2005; Yurasov et al., 2005; Menard et al., 2011a; Glauzy et al., 2017). Neurological autoimmune diseases were also reported to display impaired B cell tolerance checkpoints. Those studied to date include multiple sclerosis, which shares both an autoimmune target tissue (CNS) and clinical symptoms with NMOSD, and myasthenia gravis, which like NMOSD, is primarily autoantibody-mediated (Kinnunen et al., 2013a; Lee et al., 2016).

Consequently, it is reasonable to speculate that defects in B cell tolerance mechanisms may also contribute to autoimmune pathogenesis in NMOSD. There is evidence supporting this concept from both animal models and human studies. In animal models of NMOSD, pathogenic AQP4-specific T cell responses are normally restrained by central tolerance. However, these autoreactive T cells will develop when the antigen, which participates in counterselection, is not present (Aqp4^−/−) (Sagan et al., 2016). Similarly, B cells that can produce anti-AQP4 autoantibodies are not detected within the repertoire of wild-type mice. Rather, serum antibody responses to AQP4 in mice occur only in the absence (Aqp4^−/−) of the antigen during B cell development (Vogel et al., 2017). These findings indicate that AQP4-specific B cells may be removed from the wild-type repertoire in an antigen-dependent manner when B and T cell tolerance mechanisms are functional. Empirical support for similar mechanisms in humans is emerging. Anti-AQP4 autoantibodies can be generated by NMOSD patient-derived naïve B cells when they are differentiated in vitro into antibody-secreting cells. These findings indicate that a failure to establish naïve B cell tolerance may be associated with NMOSD (Wilson et al., 2018). These collective animal and human studies suggest that defective tolerance mechanisms may be among the key factors that initiate generation of anti-AQP4 autoantibodies and subsequent disease manifestation.

However, the fidelity of the B cell tolerance checkpoints in NMOSD patients has not been investigated. Therefore, this study was first designed to determine whether B cell tolerance mechanisms are compromised in anti-AQP4-IgG seropositive NMOSD. In addition, defects in early B cell tolerance checkpoints, known to associate with many autoimmune diseases, have not been linked to the production of autoantibodies mediating disease immunopathology. Accordingly, we next sought to understand the relationship between tolerance checkpoint defects and the development of anti-AQP4 autoantibodies by investigating whether these pathogenic autoantibodies, which are characteristic of the disease, can originate from a pool of naïve autoreactive B cells that have escaped counterselection. Finally, as a means to further explore the origins of anti-AQP4 autoantibodies, we investigated whether the naïve B cell precursors of pathogenic mature anti-AQP4 autoantibodies exhibit measurable binding affinity for this autoantigen.

Materials and methods

Study approval, subjects and specimens

Specimens from three patients who met the 2015 criteria for anti-AQP4-IgG seropositive NMOSD (Wingerchuk et al., 2015) were collected at the University of Colorado Health Center, Department of Neurology after informed consent was obtained. De-identified specimens were then provided through the Guthy-Jackson Charitable Foundation International Clinical Consortium (GJCF-ICC). Peripheral blood mononuclear cells that were isolated by Ficoll-Paque (GE Healthcare) gradient fractionation from these patients were prepared for cryopreservation, then stored in liquid nitrogen until use. The 15 healthy controls presented here have either been previously reported [Controls GO and JB (Wardemann et al., 2003); Control JH (Ng et al., 2004; Tsuiji et al., 2006); Controls HD08 and HD09 (Herve et al., 2007); Control HD10 (Isnardi et al., 2008); Control HD15 (Meyers et al., 2011); Controls HD27 and HD28 (Sauer et al., 2012); Control HD29 (Romberg et al., 2013); Control HD30 (Cantaert et al., 2015; Pala et al., 2015); and Controls HD-1 and HD-2, (Lee et al., 2016)] or are currently under review (Control HD-Y541, Sng et al., 2019). For Control HD11, the mature naïve population has been reported previously (Isnardi et al., 2010), while the new emigrant/transitional population are studied here. For purposes of this study, healthy Controls GO, JB, JH and HD-Y541 have been renamed and appear hereafter as Controls HD05, HD06, HD07 and HD-3, respectively.

Cell staining and sorting

Single B cells were isolated as previously described (Menard et al., 2011b). Briefly, peripheral B cells from NMOSD patients and healthy controls were isolated through positive selection using CD20 magnetic beads (Miltenyi Biotec). Single CD19⁺CD21^loCD10⁺IgM^hiCD27⁻ new emigrant/transitional and CD19⁺CD21⁺CD10^-IgM⁺CD27⁻ mature naïve B cells were sorted on a FACSAria flow cytometer (BD) into the individual wells of 96-well PCR plates and immediately frozen on dry ice.

Recombinant antibody production

PCR reactions, cloning strategy, antibody expression and purification, including respective primer sequences and expression vectors, have been described previously (Wardemann et al., 2003; Meffre et al., 2004). Immunoglobulin sequences were analysed using either the IgBLAST tool (Ye et al., 2013) on the NCBI website (http://www.ncbi.nlm.nih.gov/projects/igblast) or the IMGT/V-QUEST tool available on the International ImMunoGeneTics (IMGT) information system website (http://www.imgt.org) (Brochet et al., 2008). Clones were selected for recombinant expression based on successful amplification of both heavy and light chain variable regions and sequence fidelity to germline variable gene segments. Each recombinant antibody was cloned and sequenced in duplicate, then each copy was separately expressed and tested for binding by ELISA.

ELISA

Cell culture supernatants containing recombinant IgG were tested for polyreactivity on microplates coated with double-stranded DNA (dsDNA), lipopolysaccharide (LPS), or recombinant human insulin (all purchased from Sigma) using an approach described previously (Wardemann et al., 2003). Purified recombinant IgGs were tested for autoreactivity on a commercially available human epithelial type 2 (HEp-2) cell lysate ELISA kit (INOVA) according to the manufacturer’s instructions with minor modifications that have been described previously (Yurasov et al., 2005).

B cell receptor repertoire sequencing analysis

Sequence data were aligned against the 2018–07–03 IMGT germline reference set using IMGT/HighV-QUEST v1.6.0 (Alamyar et al., 2012). Data were then processed and analysed using the Immcantation Framework (http://immcantation.org) with Change-O v0.4.0 (Gupta et al., 2015), Alakazam v0.2.11 (Gupta et al., 2015), and custom R scripts (R Development Core Team, 2009). Additional healthy control H-CDR3 sequences were collected from previous publications using a similar experimental protocol (Wardemann et al., 2003; Ng et al., 2004; Tsuiji et al., 2006; Herve et al., 2007; Isnardi et al., 2008, 2010; Meyers et al., 2011; Sauer et al., 2012; Romberg et al., 2013; Cantaert et al., 2015; Pala et al., 2015; Lee et al., 2016) including samples HD05, HD06, HD07, HD08, HD09, HD10, HD11, HD15, HD27, HD28, HD29 and HD30. For all H-CDR3 sequence analyses, the first two amino acids of the H-CDR3 (typically AR or AK) were removed for consistency with the sequence region available in a subset of the previously published healthy control data. H-CDR3 net charge was calculated at pH 7.4 using the Moore method (Moore, 1985) and hydrophobicity was calculated using the Kyte and Doolittle Grand Average of Hydrophobicity (GRAVY) scale (Kyte and Doolittle, 1982). All statistical tests were performed on the mean of means within subject, status, reactivity assay result, and/or cell sort using a t-test.

Reversion of AQP4-specific autoantibodies

Sequences from mature human autoantibodies known to bind AQP4 (Bennett et al., 2009; Owens et al., 2015) were reverted to their unmutated germline counterparts to produce germline monoclonal antibodies (mAb-R) as per the strategy illustrated in Supplementary Fig. 1 (Herve et al., 2005). For purposes of this study, these autoantibodies have been renamed from their original designations as follows: 7–5–10, mAb-1; 7–5–58, mAb-3; 7–5–186, mAb-5; 11–4–87, mAb-7; 10–3–18, mAb-11; and 10–3–33, mAb-13. The reversion strategy was performed as follows. Somatically hypermutated anti-AQP4 autoantibody (mAb) sequences were vetted for reversion based on short non-template (N) junction length. N nucleotides are inserted randomly at V-D, D-J and V_L-J_L junctions in the CDR3 of Ig heavy and light chains and increase CDR3 diversity. Because non-template (N) junctions are template-independent, direct alignments between mature antibody and germline sequences are not possible; thus, longer non-template junctions increase the likelihood of inaccurate reversions. To lessen this issue, an arbitrary maximum non-template junction length of 10 nucleotides was chosen and antibodies with variable regions harbouring non-template junctions longer than 10 nucleotides in length were therefore not selected for reversion. The average non-template junction length of selected candidate sequences was 3.8 ± 2.9 nucleotides. Selected sequences were then reverted by replacing missense point mutations with nucleotides dictated by corresponding germline variable gene matches hosted by the IMGT/V-QUEST tool (http://www.imgt.org).

Cloning and expression of anti-AQP4 autoantibodies and their germline reversions

Germline reverted and original anti-AQP4 autoantibody sequences were integrated into genomic block (Integrated DNA Technologies) constructs (gBlocks) designed to flank the variable region sequences with the appropriate restriction sites and primers for consequent PCR amplification and cloning. The gBlock DNA was reconstituted as per the manufacturer’s instructions and amplified via PCR in a total volume of 20 µl using 2 µl 10× PCR Buffer (Qiagen), 0.4 µl 10 mM dNTPs (Roche), 0.2 µl HotStarTaq DNA Polymerase (Qiagen), and 1.25 µl 10 µM forward and reverse primers. The reaction was carried out for 30 cycles at 94°C for 1 min, 54°C for 1 min, and 72°C for 1 min. Initial denaturation and final extension were at 95°C for 15 min and 72°C for 5 min, respectively. PCR products were digested, purified and directionally subcloned into the previously described immunoglobulin expression vectors (Wardemann et al., 2003; Meffre et al., 2004). Cloning product sequences were confirmed against the germline reverted and anti-AQP4 autoantibody sequences. Anti-AQP4 autoantibodies and germline revertants were expressed and tested for poly- and autoreactivity as experimental replicates, as described above.

Anti-AQP4 autoantibody cell-based assay

Anti-AQP4 autoantibodies and their germline revertants were tested for AQP4 binding on a cell-based assay. After plating in a 96-well black microplate (Cell Bind, Corning) and growing to confluence, U87MG cells stably expressing the M23 isoform of AQP4 were washed once in Eagle’s minimum essential medium (MEM) and incubated in live cell blocking buffer [minimal essential media containing 10% foetal bovine serum (FBS), 2% normal goat serum (NGS), 1% bovine serum albumin (BSA), 1 mM non-essential amino acids (NEAA)] for 20 min at 37°C. Serial dilutions of anti-AQP4 autoantibodies or germline revertants were then applied in live cell block for an additional 30 min at 37°C. Cells were washed once in Eagle’s MEM, fixed in cold 4% paraformaldehyde (PFA) for 15 min, and washed three times in 1× phosphate-buffered saline (PBS). Cells were then blocked and permeabilized (10% donkey serum/1% BSA/0.1% Triton in 1× PBS) for 30 min at room temperature. The cells were incubated for 30 min at room temperature in 5% NGS/1% BSA/1× PBS with a commercial rabbit anti-AQP4 antibody (1:200) recognizing an intracellular epitope (16473–1-AP, Proteintech). Cells were then washed three times in 1× PBS, then secondary antibodies (donkey anti-human IgG Fc gamma specific Alexa Fluor® 488 Jackson Immunoresearch 709–545–098, 1:400; anti-rabbit Alexa Fluor® 594, 1:400; Life Technologies) were added for 30 min at room temperature in 2% donkey serum/1% BSA/1× PBS, followed by three washes in 1× PBS. Fluorescence intensity was read on the Infinite F200 Pro microplate reader (Tecan instruments). Binding affinity was calculated by non-linear regression using a single site total-binding equation of background subtracted red/green fluorescence intensity ratios. Each experimental condition was performed in triplicate. A strong-binding recombinant anti-AQP4 autoantibody, 07–5 #53, described previously (Bennett et al., 2009), was used in each experiment as a positive control.

Mature naïve-derived recombinant antibodies were also tested on a flow cytometric AQP4 live cell-based assay as described previously (Stathopoulos et al., 2019). Briefly, human embryonic kidney (HEK) 293T cells were transiently transfected with plasmid vectors encoding human AQP4-GFP. Forty-eight hours later the transfected cells were incubated with 1 μg/ml of recombinant antibody for 1 h at 4°C and, after a wash step, with an Alexa Fluor® 647-labelled anti-human IgG secondary antibody for 1 h at 4°C. Positivity was determined by a difference in mean fluorescence intensity (ΔMFI) index, defined as Alexa Fluor® 647 MFI in AQP4-transfected cells minus Alexa Fluor® 647 MFI in non-transfected cells.

Statistics

Reported results represent the mean of experimental duplicates derived from separately cloned and expressed recombinant immunoglobulins, unless specifically noted otherwise. Differences between patient and healthy control groups were analysed for statistical significance using Mann-Whitney tests on GraphPad Prism. Two-tailed P-values < 0.05 were considered statistically significant.

Data availability

The data that support the findings of this study are available from the corresponding authors, upon reasonable request.

Results

Study subjects

All patients met the 2015 criteria for anti-AQP4-IgG seropositive NMOSD (Wingerchuk et al., 2015). In addition, all patients were in remission sustained by standard-of-care therapies at the time of sample acquisition. Patients were selected on the basis of a clear anti-AQP4-IgG seropositive NMOSD diagnosis and the absence of aggressive immunotherapy. Additional clinical data, including demographics, disease characteristics and a detailed history of immunotherapy, are presented in Table 1 and Supplementary Table 1.

Table 1

Clinical demographics

Subject ID	Anti-AQP4 autoantibody sero status	Gender	Age, years	Disease duration, years	Past disease manifestations	Past treatment	Treatment at time of collection	Clinical/serological comorbidity
NMO-1	Positive	M	19	5	ON (left eye)	IVMP	MMF	Nonea
					LETM	PLEX
NMO-2	Positive	M	58	3	ON (right eye)	IVMP	MMF	Asthma
					APS			GERD
								EGa
NMO-3	Positive	F	44	9	ON (bilateral)	Ritux	MMF	DM II
					LETM	IVMP		GERD
						PLEX		ARMDa
						Pred
						IVIg

Three NMOSD patients were recruited for this study. Patients were diagnosed with NMOSD, as defined by the 2015 criteria for anti-AQP4-IgG seropositive NMOSD. At the time of sample collection, patients were on low-dose, maintenance treatment with mycophenolate mofetil (CellCept®). All subjects were non-carriers of the PTPN22 R620W polymorphism.

APS = area postrema syndrome; ARMD = age-related macular degeneration; DM II = type II diabetes; EG = eosinophilic gastritis; GERD = gastroesophageal reflux disease; IVIg = intravenous immunoglobulin; IVMP = intravenous methylprednisolone; LETM = longitudinally-extensive transverse myelitis; MMF = mycophenolate mofetil; ON = optic neuritis; PLEX = plasma exchange; Pred = oral prednisone; Ritux = rituximab.

^aNegative ANA testing.

Recombinant antibody production

For the B cell tolerance checkpoint fidelity studies, 130 unique NMOSD patient-derived recombinant antibodies were produced in duplicate (260 in total) from single new emigrant/transitional and mature naïve B cells (Supplementary Tables 2 and 3) and compared to 733 unique antibodies derived from 15 healthy donors. Sequence fidelity to germline variable gene segments was verified with IgBLAST or IMGT/V-QUEST (Supplementary Table 3).

Defective central and peripheral B cell tolerance checkpoints in patients with NMOSD

To determine whether B cell tolerance is properly established in anti-AQP4-IgG seropositive NMOSD patients we used a well-established approach that tests the functionality of the two early tolerance checkpoints shaping B cell development (Wardemann et al., 2003). This method measures the frequencies of polyreactive and autoreactive BCRs expressed by new emigrant/transitional and mature naïve B cells, which are downstream of the central and peripheral B cell tolerance checkpoints, respectively, thereby allowing the assessment of the extent of negative selection through these early selection steps. Recombinant antibodies were cloned from single B cells and then tested for reactivity against a panel of defined antigens (Wardemann et al., 2003). In multiple independent studies, these assays have unambiguously identified B cell tolerance checkpoint defects in patients with primary immunodeficiency disorders (Ng et al., 2004; Herve et al., 2007; Meyers et al., 2011; Kinnunen et al., 2013b; Romberg et al., 2013; Janssen et al., 2014; Menard et al., 2014; Cantaert et al., 2015, 2016) and a number of autoimmune diseases (Samuels et al., 2005; Yurasov et al., 2005; Menard et al., 2011a; Kinnunen et al., 2013a; Lee et al., 2016; Glauzy et al., 2017).

We examined 863 unique recombinant antibodies derived from a cohort of three anti-AQP4-IgG seropositive NMOSD patients (Table 1) and 15 healthy controls. All patients and controls lacked the protein tyrosine phosphatase nonreceptor type 22 (PTPN22) R620W risk variant; a polymorphism associated with a number of autoimmune diseases and which induces defects in autoreactive B cell counterselection (Menard et al., 2011a; Schickel et al., 2016). A total of 130 unique recombinant antibodies, each representing a singular BCR, were cloned from the NMOSD cohort: 62 new emigrant/transitional and 68 mature naïve BCRs (Supplementary Table 3). A total of 733 recombinant antibodies were cloned from healthy controls: 346 new emigrant/transitional and 387 mature naïve BCRs. We first assessed the functionality of central B cell tolerance by determining the frequency of new emigrant/transitional B cells that expressed polyreactive antibodies, i.e. cloned recombinant antibodies that bound all three antigens (dsDNA, LPS, and insulin) on an ELISA platform (Fig. 1A and Supplementary Fig. 2). The mean proportion of polyreactive antibodies observed in new emigrant/transitional B cells from NMOSD patients was high [28.10 ± 4.53%; mean ± standard deviation (SD)] compared to that observed (7.38 ± 2.35%) in control counterparts, revealing that central B cell tolerance is not established properly in anti-AQP4 antibody-positive NMOSD patients (P = 0.0012; Fig. 1B and C).

Figure 1

Central B cell tolerance is compromised in patients with NMOSD. Recombinant antibodies (rIgGs) derived from new emigrant/transitional B cells from three NMOSD patients were compared to those derived from 15 healthy controls. Antibodies were tested for polyreactivity on a solid-phase ELISA against three structurally distinct antigens: double-stranded DNA (dsDNA), lipopolysaccharide (LPS), and insulin. The recombinant IgGs were tested at a maximum concentration of 1.0 µg/ml (shown here) and three additional 4-fold serial dilutions (Supplementary Fig. 2). (A) Representative ELISA data from the NMOSD and healthy control groups are shown. Polyreactivity results for six individuals are summarized in the 3D plots. LPS and dsDNA absorbance values are plotted along the axes; insulin absorbance is indicated by diamond size. Each point represents a mean of experimental duplicates. Boxed area mark the positive reactivity cut-off at OD₄₀₅ 0.5. Polyreactive recombinant IgGs were defined as those that bound all three antigens (dsDNA, LPS, insulin) above the cut-off. Filled diamonds represent polyreactive recombinant IgGs; non-polyreactive recombinant IgGs are represented by open diamonds. (B) The frequency of polyreactive recombinant IgGs per subject is represented in the corresponding pie charts. Black shading indicates the polyreactive antibody frequency (%). The number in the centre of the pie chart represents the total number of individual recombinant IgGs tested. Data from the three NMOSD subjects are compared to three representative examples from the HD cohort. (C) Polyreactive antibody frequencies in the NMOSD and healthy control cohorts. The frequency of polyreactive antibodies was plotted for each subject along with the mean and standard deviation for each subject group. Statistical differences are shown when significant. (D) Frequencies of polyreactive new emigrant/transitional B cells in seven distinct autoimmune diseases and healthy control cohorts. Proportions of polyreactive antibodies expressed by new emigrant/transitional B cells were plotted for each subject group along with the mean and standard deviation for each subject group. Statistical differences are shown when significant (****P < 0.0001; ***P ≤ 0.001; **P ≤ 0.01). MG = myasthenia gravis; MS = multiple sclerosis; RA = rheumatoid arthritis; SLE = systemic lupus erythematosus; SS = Sjögren’s syndrome.

The fidelity of the peripheral B cell tolerance checkpoint was examined by determining and comparing the proportions of polyreactive mature naïve B cells in NMOSD patients and healthy controls (Fig. 2A and Supplementary Fig. 3). Frequencies of polyreactive clones in the mature naïve B cell compartment of NMOSD patients were high, with an average of 27.5 ± 3.48%, whereas the proportions of polyreactive mature naïve B cells were lower in healthy donors (8.87 ± 2.83%) (P = 0.0012; Fig. 2B and C). The functionality of the peripheral B cell tolerance checkpoint was further assessed by testing purified recombinant antibodies cloned from mature naïve B cells for autoreactivity against a HEp-2 cell lysate extract, which contains a wider range of self-antigens (Fig. 3A). We found that NMOSD patients expressed an average frequency of 46.03 ± 6.51% of HEp-2 reactive/autoreactive antibodies. In contrast, the mean proportion of autoantibodies in healthy controls was lower (20.96 ± 3.73%) (P = 0.0025, Fig. 3B).

Figure 2

Accumulation of polyreactive mature naïve B cells in the blood of patients with NMOSD. Recombinant antibodies (rIgGs) derived from mature naïve B cells from three NMOSD patients were compared to those derived from 15 healthy controls. Antibodies were tested for polyreactivity on a solid-phase ELISA against three structurally distinct antigens: double-stranded DNA (dsDNA), lipopolysaccharide (LPS), and insulin. The recombinant IgGs were tested at a maximum concentration of 1.0 µg/ml (shown here) and three additional 4-fold serial dilutions (Supplementary Fig. 3). (A) Representative ELISA data from the NMOSD and healthy control groups are shown. Polyreactivity results for six individuals are summarized in the 3D plots. LPS and dsDNA absorbance values are plotted along the axes; insulin absorbance is indicated by diamond size. Each point represents a mean of experimental duplicates. Boxed areas at OD₄₀₅ 0.5 mark the positive reactivity cut-off. Filled diamonds represent polyreactive recombinant IgGs; non-polyreactive recombinant IgGs are represented by open diamonds. (B) The frequency of polyreactive recombinant IgGs per subject is represented in the corresponding pie charts. Black shading indicates the polyreactive antibody frequency (%). The number in the centre of the pie chart represents the total number of individual recombinant IgGs tested. Data from the three NMOSD subjects are compared to three representative examples from the HD cohort. (C) Polyreactive antibody frequencies in the NMOSD and healthy control cohorts. The frequency of polyreactive antibodies was plotted for each subject along with the mean and standard deviation for each subject group. Statistical differences are shown when significant. (D) Frequencies of polyreactive mature naïve B cells in seven distinct autoimmune diseases and healthy controls. Proportions of polyreactive antibodies expressed by mature naïve B cells were plotted for each subject group along with the mean and standard deviation for each subject group. Statistical differences are shown when significant (****P < 0.0001; ***P ≤ 0.001; **P ≤ 0.01). MG = myasthenia gravis; MS = multiple sclerosis; RA = rheumatoid arthritis; SLE = systemic lupus erythematosus; SS = Sjögren’s syndrome.

Figure 3

The peripheral B cell tolerance checkpoint is impaired in patients with NMOSD. Recombinant antibodies (rIgGs) derived from mature naïve B cells from the three NMOSD patients were compared to those derived from 15 healthy controls. Purified antibodies were tested for autoreactivity on a solid-phase ELISA against human epithelial type 2 (HEp-2) cell lysate. (A) Representative ELISA data from the three NMOSD patients and healthy control groups are shown. Antibody reactivity to HEp-2 lysate is illustrated by the binding curves. ED38, a monoclonal antibody cloned from a VpreB+L+ peripheral B cell, was used as a positive control and shown by the dotted line curves. Solid line curves represent patient and control-derived antibodies. Each data point represents the mean of experimental duplicates. Dotted horizontal lines mark the positive reactivity cut-off at OD₄₀₅ 0.5. For each subject, the total number of antibodies tested and the percentage of which displayed autoreactivity, as determined through HEp-2 lysate binding, is displayed in the corresponding pie charts. (B) Autoreactive antibody frequencies in the NMOSD and healthy control cohorts. The frequency of autoreactive antibodies was plotted for each subject along with the mean and standard deviation for each subject group. Statistical differences are shown when significant. (C) Frequencies of autoreactive mature naïve B cells in seven distinct autoimmune diseases and healthy controls. Proportions of polyreactive antibodies expressed by mature naïve B cells were plotted for each subject group along with the mean and standard deviation for each subject group. Statistical differences are shown when significant (****P < 0.0001; ***P ≤ 0.001; **P ≤ 0.01). MG = myasthenia gravis; MS = multiple sclerosis; RA = rheumatoid arthritis; SLE = systemic lupus erythematosus; SS = Sjögren’s syndrome.

The elevated frequencies of polyreactive new emigrant/transitional, and polyreactive and HEp-2-reactive mature naïve B cells from the NMOSD patients were similar to those previously reported in patients with other autoimmune diseases (Figs 1D, 2D and 3C). These collective data from multiple independent studies suggest that impaired early B cell tolerance checkpoints are commonly associated with autoimmunity (Samuels et al., 2005; Yurasov et al., 2005; Menard et al., 2011a; Kinnunen et al., 2013a; Lee et al., 2016;,Glauzy et al., 2017).

Given that NMOSD-derived mature naïve B cells include an elevated proportion of self-reactive clones, we then tested whether the recombinant antibodies produced from these cells might also bind AQP4. A live cell-based antibody assay, which is commonly used to measure clinically relevant NMOSD autoantibodies, was used to assess binding of recombinant antibodies to AQP4. Control experiments validated our cell-based assay by showing that AQP4 expressed on the surface of HEK cells was bound by a previously cloned recombinant anti-AQP4 autoantibody (mAb-3) that was derived from an NMOSD patient (Supplementary Fig. 4) (Owens et al., 2015). We found that recombinant antibodies cloned from 67 NMOSD mature naïve B cells did not bind to AQP4, irrespective of their polyreactivity or autoreactivity status (Supplementary Fig. 4 and Supplementary Table 4).

We next examined the BCR repertoire of new emigrant/transitional and mature naïve B cells of NMOSD patients and healthy controls (HDs). We first sought to identify repertoire differences between healthy controls and NMOSD patients by analysing several fundamental heavy chain complementarity-determining region 3 (H-CDR3) physicochemical properties. These properties included H-CDR3 length, hydropathy (GRAVY) index and net charge in new emigrant/transitional or mature naïve B cells. No significant differences in any of these properties were observed between the healthy controls and NMOSD patients (Supplementary Table 5). Next, we compared the H-CDR3 antibody sequences of polyreactive clones to non-polyreactive clones from both new emigrant/transitional and mature naïve compartments derived from the healthy controls and NMOSD subjects. The hydropathy (GRAVY) index was not different in any of the comparisons. However, in both the healthy controls and NMOSD subjects the H-CDR3 of the polyreactive clones was invariably longer and the net charge higher than was found in the non-polyreactive clones. Statistical significance was reached for several comparisons (Supplementary Table 6). These findings are consistent with previous observations (Wardemann et al., 2003). Finally, we compared the H-CDR3 antibody sequences of HEp-2-reactive to HEp-2-non-reactive clones from mature naïve compartments derived from the HDs and NMOSD subjects. Here, significant differences in the hydropathy (GRAVY) index and CDR3 length were not observed; however, the net H-CDR3 charge of HEp-2-reactive antibody sequences (versus HEp-2-non-reactive) was higher, and approached significance, in both HDs and NMO (Supplementary Table 7).

We conclude that anti-AQP4-seropositive NMOSD patients display defective central and peripheral B cell tolerance checkpoints and accumulate large numbers of autoreactive clones in their naïve B cell compartments. The physicochemical properties associated with polyreactive and autoreactive clones from healthy and anti-AQP4-seropositive NMOSD patients are largely consistent with findings from previous investigations (Wardemann et al., 2003).

Anti-AQP4 autoantibodies may originate from self-reactive naïve B cells

We next examined whether poly/autoreactive B cells that evade tolerance checkpoints and populate the naïve repertoire can constitute a reservoir of clones from which pathogenic, AQP4 autoantibody-secreting B cells emerge. We therefore reverted somatically mutated monoclonal autoantibodies that specifically bind AQP4 back to their unmutated germline-encoded precursors (Herve et al., 2005; Glauzy et al., 2017) and tested these unmutated revertants for both polyreactivity and autoreactivity. First, we evaluated 20 AQP4-binding monoclonal autoantibody sequences that were previously reported (Bennett et al., 2009; Owens et al., 2015). We developed a comprehensive approach aiming at generating unmutated germline revertants from mature, mutated antibodies with as much precision as possible (Supplementary Fig. 1 and ‘Materials and methods’ section). Because it is not possible to determine whether somatic mutations have been introduced in regions containing non-template N nucleotides that do not match germline immunoglobulin gene segments, we increased reversion accuracy by only selecting autoantibody sequences that displayed short (<10) non-template N-nucleotide additions in the heavy and light chain CDR3s. As a consequence, six AQP4-binding monoclonal autoantibodies were selected and their VH and VL regions including the CDR3s were reverted to their unmutated original sequences by replacing all point mutations in the V(D)J with germline nucleotides (Supplementary Figs 1 and 6).

These anti-AQP4 autoantibodies and their germline revertants were cloned, expressed in vitro and tested for polyreactivity against dsDNA, LPS, and insulin by ELISA. We found that two of the six (33.3%) mature and mutated anti-AQP4 autoantibodies, mAb-1 and mAb-13, and two of the six (33.3%) revertants (unmutated naïve precursors), mAb-11R and mAb-13R showed binding to all three antigens and were therefore polyreactive (Fig. 4A, B and Supplementary Fig. 7). The revertant from mAb-1, mAb-1R, was not polyreactive, suggesting that somatic hypermutation (SHM) may have yielded polyreactivity. The revertant from mAb-13 however, mAb-13R, was found to be polyreactive, suggesting that the polyreactive mature, mutated mAb originated from a polyreactive naïve B cell precursor (Fig. 4A, B and Supplementary Fig. 7). The second polyreactive revertant (unmutated naïve precursor), mAb-11R was derived from mAb-11, a mature, mutated clone that was not polyreactive, thereby suggesting that somatic hypermutation eliminated polyreactivity.

Figure 4

Anti-AQP4 autoantibodies and their unmutated revertants contain both polyreactive and autoreactive clones. Anti-AQP4 autoantibodies (mAb) and their unmutated revertants (mAb-R) were tested for polyreactivity on a solid-phase ELISA against three structurally distinct antigens: double-stranded DNA (dsDNA), lipopolysaccharide (LPS), and insulin. Antibodies were tested at a maximum concentration of 1.0 µg/ml (shown here) and three additional 4-fold serial dilutions (Supplementary Fig. 7). Polyreactivity results are summarized in the 3D plots. LPS and dsDNA absorbance values are plotted along the axes; insulin absorbance is indicated by diamond size. Each point represents a mean of experimental duplicates. Boxed areas at OD₄₀₅ 0.5 mark the positive reactivity cut-off. Polyreactive recombinant IgGs were defined as those that bound all three antigens (dsDNA, LPS, insulin) above the cut-off. Filled diamonds represent polyreactive recombinant IgGs; non-polyreactive recombinant IgGs are represented by open diamonds. The frequency of polyreactive antibodies is represented in the corresponding pie charts. Black shading indicates the polyreactive antibody frequency (%). The number in the centre of the pie chart represents the total number of unique antibodies tested. (A) mutated anti-AQP4 autoantibodies; (B) unmutated reverted antibodies. Purified anti-AQP4 autoantibodies (mAb) and their unmutated revertants (mAb-R) were tested for autoreactivity on a solid-phase ELISA against human epithelial type 2 (HEp-2) cell lysate. Antibody reactivity to HEp-2 lysate is illustrated by the binding curves. Solid line curves represent AQP4 mAbs and mAb-Rs. ED38, a monoclonal antibody cloned from a VpreB+L+ peripheral B cell, was used as a positive control; L50, a monoclonal antibody cloned from a mature naïve B cell of a uracil N-glycosylase-deficient patient was used as a negative control. Dotted line curves represent ED38 and L50. Each data point represents the mean of experimental duplicates. Dotted horizontal lines mark the positive reactivity cut-off at OD₄₀₅ 0.5. For both the mAbs and the mAb-Rs, the total number of antibodies tested and the percentage of which displayed autoreactivity, as determined through HEp-2 lysate binding, is displayed in the corresponding pie charts. (C) Mutated anti-AQP4 autoantibodies; (D) unmutated reverted antibodies.

The anti-AQP4 autoantibodies and their germline revertants were then quantitatively tested for autoreactivity by HEp-2 ELISA. Six of six (100%) mutated anti-AQP4 autoantibodies showed binding to the HEp-2 cell lysate (Fig. 4C). The cross-reactivity for self-antigens expressed by HEp-2 cells that we observed with the anti-AQP4 autoantibodies has also previously been reported for some anti-influenza virus antibodies and is also frequently associated with antibodies expressed by IgG+ memory B cells (Tiller et al., 2007; Kaur et al., 2015; Schickel et al., 2016). In addition, four of six (66.7%) unmutated revertants derived from mAb-1, mAb-11, mAb-7 and mAb-13 also showed binding to the HEp-2 cell lysate (Fig. 4D). Therefore, pathogenic B cells expressing anti-AQP4 autoantibodies in NMOSD patients can originate from autoreactive naïve clones expanded in the blood of these subjects.

Unmutated revertants derived from anti-AQP4 autoantibodies do not exhibit specificity toward AQP4

To determine if the naïve precursors of anti-AQP4 autoantibodies exhibited AQP4 reactivity, we measured the binding affinity (K_d) of the mutated anti-AQP4 autoantibodies and their revertants to assess if they display detectable binding to AQP4. While all mutated anti-AQP4 autoantibodies bound to AQP4 with a range of affinities as expected, none of the unmutated revertants displayed any measurable AQP4 binding in the cell-based assay (Fig. 5 and Table 2). We conclude that somatic hypermutation can play an essential role in the generation of anti-AQP4-specific autoantibodies in NMOSD patients.

Table 2

AQP4 binding affinities

Antibody	Kd, nM	SE, Kd, nM	Binding to AQP4
mAb-1	24.2	7.9	Strong binder
mAb-3	57.09	9.7	Fairly strong binder
mAb-5	15.2	2.4	Strong binder
mAb-7	56.28	90.05	Fairly strong binder
mAb-11	559	NA	Weak binder
mAb-13	218.3	59.8	Weak binder
mAb-1R	NBD		NBD
mAb-3R	NBD		NBD
mAb-5R	NBD		NBD
mAb-7R	NBD		NBD
mAb-11R	NBD		NBD
mAb-13R	NBD		NBD

The binding affinities for AQP4 of anti-AQP4 autoantibodies (mAb) and their unmutated revertants (mAb-R) were measured via cell-based assay. Shown here are the measured equilibrium dissociation constants (K_d, nM) with their standard errors (SE), where appropriate, as well as a qualitative assessment of AQP4 binding affinity for each tested antibody. NA = not available; NBD = no binding detected.

Figure 5

Unmutated precursors of the anti-AQP4-specific autoantibodies do not bind to AQP4. Anti-AQP4 autoantibodies (mAb) and their unmutated revertants (mAb-R) were tested for surface binding to the AQP4 M23 isoform on AQP4-transfected U87MG cells. Representative binding curves from which affinity values were calculated for two autoantibodies (mAb) and their respective unmutated revertants are shown. The y-axis represents ratios of bound antibody to cell surface AQP4 (rAb/AQP4) as means with their respective standard errors. The x-axis represents the various concentrations (nM) at which mAbs and mAb-Rs were tested. Data are fit using a single site total binding model. Detailed K_d results for all of the autoantibodies and their unmutated revertants are presented in Table 2.

Discussion

The contribution of B cell tolerance defects to NMOSD immunopathology

Anti-AQP4 autoantibodies play a key role in the pathogenesis of NMOSD. The AQP4 water channel is highly expressed on astrocyte end-feet, and autoantibodies can influence internalization of the antigen and activate complement (Saadoun et al., 2010; Hinson et al., 2012). Importantly, features of NMOSD pathology have been replicated in animal models through the transfer of human monoclonal antibodies or patient-derived immunoglobulin, demonstrating the direct role these autoantibodies play in the disease (Bennett et al., 2009; Kinoshita et al., 2009; Asavapanumas et al., 2014). In addition to producing autoantibodies, B cells can make further contributions to NMOSD immunopathology. This is highlighted by treatment with the B cell depleting therapeutic rituximab. Patients with NMOSD respond well to this treatment (Kim et al., 2013), but their improved clinical outcomes are not always directly associated with diminished anti-AQP4 autoantibody, as high titres of circulating anti-AQP4-IgG can persist (Pellkofer et al., 2011). This observation may point toward a heterogeneous population of anti-AQP4-binding autoantibodies, including those that are pathogenic and those that are not, the latter of which may contribute to the persistent titre. In addition, this observation could indicate that B cells participate in the inflammatory cascade that leads to demyelination and tissue injury through antibody-independent means (reviewed in Krumbholz and Meinl, 2014; Bennett et al., 2015). For instance, AQP4-reactive B cells may internalize this self-antigen recognized by their B cell receptors, and then function as antigen-presenting cells (APCs), leading to T cell activation and initiation of T cell-mediated responses. Indeed, it has been demonstrated that such MHC class II-dependent APC function of B cells contributes to autoimmunity (Serreze et al., 1998; Noorchashm et al., 1999; Shlomchik et al., 2001) and is necessary for models of CNS autoimmunity (Molnarfi et al., 2013; Parker Harp et al., 2015). Anti-AQP4-specific B cells also likely produce inflammatory cytokines that contribute to disease pathology. Similar antibody-independent roles for B cells (Li and Bar-Or, 2018) are thought to explain the marked benefit of B cell depletion therapy in multiple sclerosis (Hauser et al., 2008, 2017) where no specific target autoantigen has been identified to date (reviewed in Willis et al., 2015).

Given these multidimensional roles, defects in early B cell tolerance checkpoints that result in increased autoreactive B cells in naïve populations could contribute to NMOSD pathology through both autoantibody production and a number of antibody-independent B cell effector functions. In agreement with this scenario and using a well established method to determine frequencies of autoreactive clones in the naïve B cell populations of NMOSD patients, we identified increased frequencies of new emigrant/transitional B cells and mature naïve B cells that expressed polyreactive and/or autoreactive BCRs in patients compared to healthy controls. These results demonstrate that both central and peripheral tolerance mechanisms are defective in NMOSD, a finding consistent with previous reports describing the role of functional deletional tolerance in preventing humoral anti-AQP4-directed autoimmunity in mouse models (Vogel et al., 2017). The results of our study are also consistent with findings in acetylcholine receptor (AChR) and muscle-specific tyrosine kinase (MuSK) autoantibody-positive myasthenia gravis. Like NMOSD, myasthenia gravis is an autoimmune disease where autoantibodies directly contribute to pathology (albeit without the requirement of entry into the protected CNS compartment) and where both early B cell tolerance checkpoints are compromised (Lee et al., 2016). These B cell selection defects in NMOSD and myasthenia gravis may contribute to reported co-existence of these diseases in the same patients (Leite et al., 2012; Bibic et al., 2018). However, multiple sclerosis also shares significant disease similarity with NMOSD (Wingerchuk et al., 2007), although most multiple sclerosis patients were shown to have functional central, but defective peripheral B cell tolerance checkpoints (Kinnunen et al., 2013a). Since the conspicuous effect size associated with the assays makes the evaluation of small patient cohorts sufficient to establish B cell tolerance checkpoint defects (Figs 1D, 2D and 3C), we only studied a limited number of patients, although we tested the reactivity of 130 unique NMOSD-derived recombinant antibodies. However, we leave open the possibility that different impairments in B cell selection may be associated with NMOSD clinical subtypes. Larger studies, which include anti-myelin oligodendrocyte glycoprotein (MOG)-autoantibody seropositive and autoantibody seronegative NMOSD patients, will determine if defective central and peripheral B cell tolerance checkpoints can be identified in all NMOSD patients. Nevertheless, NMOSD, myasthenia gravis, and multiple sclerosis patients are characterized by the accumulation of large numbers of autoreactive mature naïve B cells in their blood that may favour the development of their neurological autoimmune disease.

Anti-AQP4 autoantibody production

We sought to potentially identify a link between early defects in B cell tolerance and the generation of pathogenic anti-AQP4 autoantibodies. Using anti-AQP4-specific monoclonal autoantibodies isolated from NMOSD patient CSF plasmablasts and plasma cells (Bennett et al., 2009; Owens et al., 2015), we removed mutations in these antibodies to generate unmutated revertants that may correspond to BCRs expressed by the naïve B cell precursors of these anti-AQP4-IgG producing cells. The polyreactivity and self-reactivity associated with many of the unmutated revertants suggest that mutated anti-AQP4 autoantibodies may originate from the pool of autoreactive mature naïve B cells that escape deletion at early B cell tolerance checkpoints. However, none of the unmutated revertants displayed any measurable binding to AQP4 using a highly sensitive live cell-based assay, which has clinical relevance since it is preferentially used for diagnosis of anti-AQP4-IgG seropositive NMOSD patients (Waters et al., 2012, 2014). Somatic hypermutations therefore play an essential role in increasing antibody self-reactivity towards AQP4. These findings are consistent with a number of studies that have investigated the binding of autoantibody unmutated revertants. In pemphigus vulgaris, somatically hypermutated pathogenic autoantibodies recognize desmoglein-3, whereas unmutated revertants did not bind this self-antigen (Di Zenzo et al., 2012). Autoantibodies directed towards double-stranded DNA or the extractable nuclear antigen found in patients with systemic lupus erythematosus lost their specificity when somatic hypermutations were removed (Wellmann et al., 2005; Mietzner et al., 2008). Similarly, unmutated revertants from anti-cytokine autoantibodies in either AIRE-deficient patients (Meyer et al., 2016) or pulmonary alveolar proteinosis patients (Piccoli et al., 2015) did not bind their antigen. The behaviour of these autoantibody-derived unmutated revertants contrasts that of several well-investigated antibodies that develop toward exogenous (non-self) antigens during a normal immune response. Examples of such include viral infections (Corti et al., 2010, 2011; Pappas et al., 2014), wherein unmutated revertants of virus antigen-specific monoclonal antibodies retain binding activity.

These collective findings point toward a critical role for somatic mutations in introducing autoantigen-specific reactivity during an ongoing autoimmune response. It is unclear at this point if the autoreactive naïve B cell precursors of anti-AQP4 pathogenic autoantibodies may have very weak affinity below the detection of our method for measurement or if anti-AQP4 autoantibody-producing B cell clones may be initially activated by an antigen other than AQP4. Indirect evidence may support the latter scenario, because T cell clones (Varrin-Doyer et al., 2012) that recognize the dominant AQP4 T cell epitope display cross-reactivity to a homologous peptide sequence within a protein of Clostridium perfringens, an indigenous intestinal bacterium over-represented in patients with NMOSD (Cree et al., 2016). While there is no evidence for anti-AQP4 autoantibody cross-reactivity to other antigens available at this time, the observation that AQP4-specific T cells may be influenced by other antigens leaves open the possibility that B cell antigens other than AQP4 could participate in NMOSD pathogenesis. Alternatively, some autoreactive naïve B cells may recognize self-antigens, instead of an alloantigen, through their unmutated BCRs. In support of this hypothesis, defects in the peripheral B cell tolerance checkpoint in AIRE-deficient patients, where thymic T cell tolerance mechanisms are impaired, result in the emergence of autoreactive mature naïve B cells that contain clones expressing unmutated BCRs with measurable micromolar affinity for insulin and IL-17, two self-antigens often targeted in this syndrome (Sng et al., 2019).

In addition, the peripheral accumulation of autoreactive mature naïve B cells is often associated with dysfunctional regulatory T cells (Tregs) (Sauer et al., 2012; Kinnunen et al., 2013a, b; Romberg et al., 2013; Janssen et al., 2014; Pala et al., 2015; Cantaert et al., 2016). While defects in the function of Tregs in NMOSD have not yet been described, AQP4-specific Tregs are present at a reduced frequency in NMOSD patients (Varrin-Doyer et al., 2012). Hence, a functionally compromised Treg compartment lacking AQP4-specific T cell clones may contribute to first, a defective peripheral tolerance checkpoint leading to failure to counterselect autoreactive naïve B cell precursors of anti-AQP4 clones, and second, the production of pathogenic anti-AQP4 autoantibodies in the absence of Treg-mediated suppression of T cell help in NMOSD (Mitsdoerffer et al., 2013; Vogel et al., 2017).

Taken together, our data provide details for a model (Fig. 6) of anti-AQP4 autoantibody production in the presence of defective B cell tolerance checkpoints. We suggest that autoreactive naïve B cells resulting from defective early B cell tolerance checkpoints in NMOSD patients likely contribute to the production of pathogenic anti-AQP4 autoantibodies, and naïve clones expressing BCRs with initial low specificity for AQP4 may improve their self-reactivity through somatic hypermutation and affinity maturation regulated by T cells.

Figure 6

Schematic diagram illustrating the potential consequence of defective B cell tolerance checkpoints in the development of NMOSD autoantibodies. During early B cell development, immunoglobulin variable region gene segments are stochastically recombined to generate functional antibodies (B cell receptors) that are expressed on the cell surface. This process is fundamental for the generation of the wide diversity of the immunoglobulin repertoire but also generates self-reactive B cells (red cells) alongside those that comprise the non-self-reactive naïve repertoire (green cells). To evade the development of an immune response against self, two separate tolerance mechanisms remove autoreactive B cells during their development. The first is a central tolerance checkpoint in the bone marrow between the early immature and immature B cell development stages, which removes a large population of B cells that express self-reactive/polyreactive antibodies (shown as red cells). The second checkpoint selects against self-reactive new emigrant/transitional B cells before they enter the long-lived mature naïve B cell compartment. Deficiencies in the integrity of these tolerance mechanisms can be demonstrated through quantifying the frequency of both polyreactive and self-reactive B cells downstream of each checkpoint. A number of autoimmune diseases, including NMOSD, have central and peripheral B cell checkpoints that fail to enforce B cell tolerance and proper counterselection. Thus, these patients include an abnormally high frequency of polyreactive and/or self-reactive new emigrant/transitional and mature naïve B cells. Our study findings suggest that the reservoir of new emigrant/transitional or mature naïve B cells that develop in the presence of defective B cell tolerance checkpoints can supply clones that become pathogenic anti-AQP4 autoantibodies after acquiring somatic hypermutations.

Defects in B cell tolerance checkpoints are a common denominator in autoimmunity

Given the large number of autoimmune diseases in which B cell tolerance checkpoints have been studied (Samuels et al., 2005; Yurasov et al., 2005; Menard et al., 2011a; Kinnunen et al., 2013a; Lee et al., 2016;,Glauzy et al., 2017), it is becoming increasingly clear that an essential characteristic of autoimmunity is that either the central, the peripheral, or both B cell tolerance checkpoints are impaired (Figs 1D, 2D and 3C) and result in the accumulation of autoreactive B cells in the blood of patients with rheumatoid arthritis, systemic lupus erythematosus, type 1 diabetes, multiple sclerosis, myasthenia gravis, Sjögren’s syndrome and NMOSD (reviewed by Meffre, 2011). Although common among many autoimmune conditions, it is not understood how these fundamental defects underlie the development of autoimmune diseases of divergent pathology. While type 1 diabetes, systemic lupus erythematosus, myasthenia gravis and NMOSD all have B cell tolerance checkpoint defects, why does a systemic autoimmune disease, such as systemic lupus erythematosus, occur in some individuals while an autoimmune disease with a well defined autoantigen, such as NMOSD, occur in others? B cell tolerance checkpoint defects appear to allow development of a reservoir of naïve B cells that escape counterselection and may present self-antigens, thereby potentially initiating autoimmune responses. Then, in genetically predisposed individuals with specific MHC favouring the presentation of self-antigens, altered specific T cell subsets including Tregs, coupled with environmental influences, may lead to the development of autoreactive B cells of distinct specificity.

It is possible that a link exists between B cell tolerance checkpoint defects and autoimmunity that arises after cancer treatment with checkpoint blockade immunotherapy. This treatment approach tilts the balance toward immune activation by blocking the interaction of inhibitory receptors with their ligands, promoting the generation of efficient tumour-eliminating T cells. Two well studied immune checkpoint targets are cytotoxic T lymphocyte-associated protein 4 (CTLA-4) and programmed cell death protein 1 (PD-1). The number of reports describing emerging autoimmunity following the use of checkpoint inhibitor immunotherapy continues to grow rapidly. Interestingly, different autoimmune characteristics appear in these patients, including myasthenia gravis, diabetes, inflammatory arthritis, myositis and vasculitis (Tocut et al., 2018). However, it remains unclear how these autoimmune conditions precisely arise and why particular antigens are targeted in some cases and other antigens in other cases. While only a fraction of checkpoint inhibitor immunotherapy patients develop autoimmunity, it is tempting to consider whether they correspond to those who display B cell tolerance checkpoint defects such as in asymptomatic individuals carrying the PTPN22 R620W risk variant (Menard et al., 2011a). A coordinated study of autoimmunity development in cancer patients prior to and after checkpoint inhibitor immunotherapy may offer a unique opportunity to investigate autoimmunity mechanisms and perhaps offer a predictive biomarker that would inform clinical expectations and decisions.

Current therapeutic strategies do not target B cell tolerance defects

The connection between dysfunctional tolerance and pathogenic autoantibody production has important implications for the treatment of NMOSD patients with therapies that specifically target B cells, such as rituximab and ocrelizumab. The new emigrant/transitional and mature naïve B cell population both express CD20; consequently, they are depleted by treatment with either rituximab and ocrelizumab. However, through investigating B cell tolerance checkpoint fidelity before and after rituximab-mediated B cell depletion, we previously demonstrated that defective B cell tolerance checkpoints present in patients with type 1 diabetes prior to treatment are not reset, as evidenced by the elevated proportion of autoreactive naïve B cells in the repopulated repertoire (Chamberlain et al., 2016). In the current study, one of three NMOSD patients had received rituximab more than 4 years prior to sample acquisition, while the other two NMOSD patients had not. B cell tolerance checkpoint defects were present, however, in all three patients, thereby confirming the findings of the type 1 diabetes study and that early B cell tolerance checkpoint defects are not likely to be affected by immunotherapies commonly applied to patients with NMOSD including rituximab (Supplementary Table 1). As a consequence, the durability of B cell depletion in NMOSD may be fragile, given that the repopulated naïve B cell repertoire is expected to contain newly generated polyreactive and self-reactive B cells that evaded counterselection at both the central and peripheral checkpoints and may contribute to anti-AQP4 autoantibody production. However, this issue may be circumvented if B cell depletion is maintained through recurrent treatment, such as the current approach using ocrelizumab in some multiple sclerosis patients (Hauser et al., 2017) or rituximab in some NMOSD patients (Kim et al., 2015). Alternatively, strategies that directly target early defects of B cell tolerance may offer a better risk-to-benefit ratio, as they would target the development of autoimmunity rather than the autoimmune features that have already emerged. Indeed, inhibition of PTPN22, a phosphatase with a genetic variant associated with many autoimmune diseases, can prevent the production of autoreactive B cells in the bone marrow and might be explored in preclinical studies (Schickel et al., 2016). Moreover, inebilizumab, an anti-CD19 agent currently being tested as a treatment for NMOSD (NCT02200770), could be more successful in restoring B cell tolerance checkpoints because, in contrast to rituximab, it can theoretically eliminate CD19+ B cell precursors (Chen et al., 2014).

Conclusion

This study first demonstrates that the integrity of the central and peripheral B cell tolerance checkpoints in anti-AQP4 autoantibody-positive NMOSD patients is compromised. This places NMOSD on the growing list of autoimmune diseases that are characterized by an impaired counterselection of developing autoreactive B cells. Second, pathogenic anti-AQP4 autoantibodies can originate from the pool of naïve B cells that escape B cell tolerance due to dysfunctional selection processes. This finding provides a highly novel link between the defective naïve B cell repertoire due to early tolerance defects, and the production of pathogenic autoantibodies. Third, the unmutated precursors of B cells secreting anti-AQP4 autoantibodies do not bind the autoantigen, revealing that anti-AQP4 specificity and the generation of pathogenic autoantibodies requires affinity maturation and the acquisition of somatic hypermutations, a process known to involve T cells that may also be improperly selected during their development in the thymus. These collective findings provide new insight into the important role played by B cells in NMOSD immunopathology.

Abbreviations

Abbreviations
 
• BCR

  B cell receptor

 
• NMO

  neuromyelitis optica

 
• NMOSD

  neuromyelitis optica spectrum disorders

Acknowledgements

The authors thank Karen Boss for providing editorial assistance; the Alice S. Powers Trust and Anna Shubik Sweeney for providing support to the O’Connor Laboratory; and Jacinta Behne, Dr. Terry J. Smith, and Dr. Michael R. Yeaman of the Guthy-Jackson Charitable Foundation for providing the opportunity for the project to commence.

Funding

This work was supported by the Guthy-Jackson Charitable Foundation through grants to K.C.O. and J.L.B., and through its CIRCLES research and biorepository program. Support was also provided by the National Institute of Allergy and Infectious Diseases of the National Institutes of Health through a grant award to K.C.O., under award number AI114780, to S.H.K. under award number AI104739, to M.L. under award number AI130548, to E.M. under award numbers AI071087 and AI061093, and to J.L.B. under award number AI110498. Support was also provided by the National Eye Institute of the National Institutes of Health through a grant award to J.L.B., under award number EY022936. J.N.S. is a trainee of the Medical Scientist Training Program at the University of Colorado Anschutz Medical Campus (MSTP T32 GM008497). P.S. was partly supported by the Onassis Foundation under the Special Grant and Support Program for Scholars' Association Members (Grant R ZO 006/2018-2019).

Competing interests

K.C.O. has received personal compensation for educational activities from Genentech and New England Biolabs, for consulting services from Proclara Biosciences, Momenta Pharmaceuticals and Editas Medicine and has received research support from Ra Pharma. All of these activities, which occurred during the last 4 years, were conducted outside of the work reported in this manuscript. J.L.B. has received personal compensation for consulting services to Clene Nanomedicine, Viela Bio, Chugai Pharmaceuticals, EMD Serono, Equillium, Alexion, and Frequency Therapeutics; has received research support from Mallinckrodt; and serves on the editorial boards of the Multiple Sclerosis Journal, Neurology: Neuroimmunology & Neuroinflammation, and the Journal of Neuro-Ophthalmology. All of these activities were conducted outside of the work reported in this manuscript. The other authors report no potential conflicts of interest.

Author notes

Elizabeth Cotzomi, Panos Stathopoulos, Eric Meffre and Kevin C. O’Connor authors contributed equally to this work.