Inhibition of Membrane-Bound BAFF by the Anti-BAFF Antibody Belimumab

B cell activating factor of the TNF family (BAFF, also known as BLyS), a cytokine that regulates homeostasis of peripheral B cells, is elevated in the circulation of patients with autoimmune diseases such as systemic lupus erythematosus (SLE). BAFF is synthetized as a membrane-bound protein that can be processed to a soluble form after cleavage at a furin consensus sequence, a site that in principle can be recognized by any of the several proteases of the pro-protein convertase family. Belimumab is a human antibody approved for the treatment of SLE, often cited as specific for the soluble form of BAFF. Here we show in different experimental systems, including in a monocytic cell line (U937) that naturally expresses BAFF, that belimumab binds to membrane-bound BAFF with similar EC50 as the positive control atacicept, which is a decoy receptor for both BAFF and the related cytokine APRIL (a proliferation inducing ligand). In U937 cells, binding of both reagents was only detectable in furin-deficient U937 cells, showing that furin is the main BAFF processing protease in these cells. In CHO cells expressing membrane-bound BAFF lacking the stalk region, belimumab inhibited the activity of membrane-bound BAFF less efficiently than atacicept, while in furin-deficient U937 cells, belimumab inhibited membrane-bound BAFF and residual soluble BAFF as efficiently as atacicept. These reagents did not activate complement or antibody-dependent cell cytotoxicity upon binding to membrane-bound BAFF in vitro. In conclusion, our data show that belimumab can inhibit membrane-bound BAFF, and that BAFF in U937 cells is processed by furin.


INTRODUCTION
BAFF and APRIL are important fitness and survival factors for B cells and plasma cells in the periphery. They exert their function through different receptors: BAFFR (BAFF receptor, TNFRSF13A) that binds to BAFF only, TACI (transmembrane activator and calcium modulator and cyclophilin ligand interactor, TNFRSF13B) that binds to BAFF and APRIL, and BCMA (B cell maturation antigen, TNFRSF17) that also binds to BAFF and APRIL [reviewed in (1)]. BAFFR transduces BAFF survival signals in transitional and naïve B cells, both of which are greatly decreased in BAFF-ko and BAFFR-ko mice (2)(3)(4)(5). TACI and BCMA are expressed either upon B cell activation and/or at later stages of B cell differentiation. For example, BCMA is expressed in plasma cells that can use APRIL and/or BAFF for survival (6). Although BAFF is synthetized as a membrane-bound protein, it can be processed to a soluble form by cleavage at a furin consensus-processing site (7,8). Furin belongs to the substilisin/kexin-like pro-protein convertase (PCSK) family of proteases, seven of which (PCSK1-2, furin and PCSK4-7) have arginines in their recognition sequences, and most of which are ubiquitously expressed. They are often redundant for substrate cleavage and they process a vast panel of targets, among others hormones, enzymes, receptors, cytokines and extracellular matrix components [reviewed in (9)]. With regard to BAFF, circulating levels are found elevated in diseases with involvement of auto-reactive B cells, including systemic lupus erythematosus (SLE) [reviewed in (10,11)]. Belimumab, a human monoclonal IgG1λ anti-BAFF antibody approved by the FDA, can improve the condition of SLE patients (12,13). Belimumab is specific for BAFF, and is more precisely described as an inhibitor of soluble BAFF (14,15). Atacicept is another BAFF inhibitor consisting of the ligand-binding portion of the receptor TACI fused to a modified Fc portion of human IgG1 to remove binding to Fc receptors and to complement. Atacicept is characterized by a broader specificity of inhibition that includes APRIL and heteromers of BAFF and APRIL (16). Atacicept is under clinical development, also for the treatment of SLE (17). Here, we genetically inactivated furin in U937 histiocytic lymphoma cells that naturally express BAFF (18) to convert these cells form BAFF shedding into membrane-bound BAFF-expressing cells, indicating that furin itself is the main BAFF-processing protease in these cells. Membrane-bound BAFF on furin-deficient U937 was bound and inhibited by belimumab, suggesting that belimumab targets membrane-bound in addition to soluble BAFF.

Proteins and Antibodies
Belimumab (registered trade name Benlysta), denosumab (registered trade name Xgeva), adalimumab (registered trade name Humira), and human IgG (intraveinous immunoglobulins; registered trade name Kiovig) were purchased from the Pharmacy of Lausanne University Hospital (CHUV). Atacicept was kindly provided by Henry Hess (Merck KGaA). BCMA-Fc and Fc-BAFF were produced in CHO cells and affinity purified on Protein A-Sepharose, essentially as described (19). When indicated, dimeric BCMA-Fc was used. Dimeric BCMA-Fc was obtained as a defined peak after size fractionated by gel filtration on a Superdex 200 Increase 10/30 column (GE Healthcare) equilibrated in PBS. BCMA-COMP-Flag (20) was produced by transient transfection of HEK 293T cells with the polyethyleneimide method (21). 7-day-old conditioned cell supernatants in serum-free OptiMEM medium were collected and used directly.
For the measure of membrane-bound BAFF on CHO-3296 cl7 cells, experiments were performed in 48 well plates. Untransfected CHO cells, or CHO-3296 cl7 cells (1.5 × 10 5 /well) were left to adhere for 2 h, after which time medium was removed and replaced by 200 µl of fresh medium. Wells without CHO cells contained medium only. Hundred µl of inhibitors (at 5-fold the final desired concentration), 50 µl of Fc-BAFF at 1 µg/ml (or 50 µl of medium in conditions without Fc-BAFF) and 150 µl of CFSE-labeled BAFFR:Fas reporter cells (∼2.5 × 10 5 cells) were added and incubated for 11 h. Non adherent cells were transferred into 4 ml tubes, and then pooled with adherent cells detached with trypsin-EDTA. Cells were spun for 1 min at 5,000 rpm (2,400 × g), pellets were suspended in 100 µl of PBS, 5% FCS and supplemented with 10 µl of 10 µg/ml propidium iodide before analysis (See "Flow cytometry").
CHO-3296 clone 7 cells were stained with atacicept, belimumab or denosumab as described for transfected 293T cells, except that MFI was reported for all live cells. U937 cells (wild type, furin-deficient or BAFF-deficient) were pre-incubated with human IVIGs (50 µg in 50 µl of PBS, 5% FCS) for 20 min on ice, washed and stained for 20 min on ice with the indicated amount of biotinylated atacicept or biotinylated belimumab in 50 µl of PBS 5% FCS followed by phycoerythrin-coupled streptavidin (1/500). CFSE-labeled reporter cells (co-cultured with membranebound BAFF-expressing cells, recombinant Fc-BAFF or controls) were stained with propidium iodide as described (see "Cytotoxic assays with BAFFR:Fas reporter cells") and analyzed by flow cytometry. The proportion of live reporter cells was calculated as follows: [number of live reporter cells / (number of live reporter cells + number of dead reporter cells)], where live cells were CFSE + PI − cells in the live FCS/SSC gate, and dead cells were all CFSE + cells in the dead FCS/SSC gate. This number was divided by the proportion of live cells in untreated controls to correct for spontaneous death, and multiplied by 100 to get the percentage of specific target cell viability.

BAFF ELISA
BAFF in conditioned supernatant of U937 cells was quantified using the hypersensitive hBAFF ELISA kit (Adipogen, AG-45B-0001-KI01) according to the manufacturer's protocol. Supernatants of BAFF-deficient and furin-deficient cells were measured in duplicate at a dilution of 1/2, and those of wild type cells at dilutions of 1/4, 1/40, and 1/400.  Frontiers in Immunology | www.frontiersin.org 96-wells culture plates. Fc-containing reagents were then added in 10 µl of medium, incubated for 10 min at room temperature before addition of 10 µl of untreated or heat-inactivated (30 min at 56 • C) normal human serum (from P.S.) for 2 h at 37 • C. Cell viability was monitored with PMS/MTS (see "Cytotoxic assays with BAFFR:Fas reporter cells").

Antibody-Dependent Cell-Mediated Cytotoxicity Assay
Antibody-dependent cell-mediated cytotoxicity was measured using the ADCC core kit (Promega, G7010), according to the manufacturer's protocol. CHO-3296 clone 7 cells were used as targets, adalimumab, atacicept, BCMA-Fc and belimumab at the indicated concentrations were used as mediators, and Jurkat cells expressing both human FcγRIIIa V158 and NFAT-induced luciferase were used as effectors. 25,000 target cells per well were seeded in 96 wells white plates in 100 µl of DMEM:F12 2% FCS and left to adhere overnight. Ninety-five microliter of medium was removed and replaced by 25 µl of RPMI supplemented with 4% of low Ig FCS, 25 µl of the test reagent and 25 µl of freshly defrosted effector cells. After 6 h of incubation, 75 µl of luciferase substrate was added and luminescence was monitored 15 min later.

Statistics
Replicate measurements are shown as mean ± SEM. EC 50 of titration curves were determined after normalization of cell viability using the "non linear regression (curve fit)" followed by the "log(agonist) vs. normalized response-variable slope" function of Prism 7 (GraphPad Software).

RESULTS
Belimumab and Atacicept Bind to and Inhibit BAFF at the Surface of Transfected 293T or CHO Cells In a FACS-based assay, both atacicept and belimumab stained 293T cells transfected with full length BAFF, but not mocktransfected cells. BAFF with a mutated furin consensus sequence (SRNKRAVGP → SRNKLQGP) (BAFF FS) or BAFF lacking the stalk region linking the transmembrane domain to the extracellular receptor-binding domain, a deletion that included the furin consensus sequence (BAFF stalk) were also recognized by atacicept and belimumab, while the unrelated antibody denosumab stained none of these cells ( Figure 1A). Binding of atacicept or belimumab to membrane-bound BAFF prevented subsequent binding of recombinant BCMA, one of the  BAFFR:Fas reporter cells were exposed to undiluted conditioned medium of the indicated U937 cells that were grown alone, but for the same time (17 h) and otherwise identical conditions to those of U937 cells used in panels C and D. Medium and Fc-BAFF at 50 ng/ml in medium pre-incubated alone for 17 h in the same conditions were used as controls. Viability of reporter cells was monitored with the PMS/MTS assay. Mean of triplicates ± SEM. Experiment performed twice. receptors for BAFF, suggesting that the binding of either atacicept or belimumab inhibits membrane-bound BAFF (Figure 1B).
In order to measure the activity of membrane-bound BAFF, we generated a stable clone of CHO cells expressing BAFF stalk, i.e., uncleavable membrane-bound BAFF. These cells stained with both atacicept and belimumab, but not with the control antibody denosumab (anti-RANKL) (Figure 2A) and did not release detectable levels of soluble BAFF activity in their supernatants ( Figure 2B). Unlabeled CHO cells expressing uncleavable BAFF were co-cultured with CFSE-labeled BAFFR:Fas reporter cells. In reporter cells, the extracellular domain of BAFFR is fused to the transmembrane and intracellular domains of the death receptor Fas, so that they are killed upon engagement of BAFFR, and so that BAFF inhibitors can protect them from BAFF-mediated killing. CHO cells expressing uncleavable BAFF readily killed BAFFR:Fas reporter cells, a killing that could be inhibited in a dose-dependent manner by both atacicept and belimumab, although atacicept was about 10 times more efficient than belimumab in this respect (Figure 2C). Reporter cells also died in response to soluble recombinant BAFF, but in this case inhibition by atacicept or belimumab was equally good (Figure 2D). Controls showed that atacicept, belimumab or untransfected CHO cells had no deleterious effect on reporter cells (Figure 2E), and that untransfected CHO cells did not prevent atacicept and belimumab from inhibiting soluble BAFF ( Figure 2F). Taken together, these results show that binding of atacicept and belimumab to membrane-bound BAFF is inhibitory.

Processing of Membrane-Bound BAFF to Active Soluble BAFF Is Mediated by Furin in U937 Cells
In order to determine the action of atacicept and belimumab on endogenous BAFF, we used monocytic U937 cells that naturally produce soluble BAFF (18,25). Atacicept and belimumab did not stain wild type and BAFF-deficient U937 cells by flow cytometry (Figures 3A,B). However, CRISPR/Cas9-mediated deletion of furin with a construct targeting the active site of the protease abrogated >98% of the release of both BAFF protein ( Figure 3C) and BAFF activity ( Figure 3D) in supernatants of U937 cells, while increasing cell surface levels of BAFF that became detectable by staining with both atacicept and belimumab (Figures 3A,B).

Inhibition of Membrane-Bound BAFF in U937 Cells by Atacicept and Belimumab
Inhibition of non-mutated membrane-bound BAFF was tested in furin-deleted U937 cells. Atacicept and belimumab on their own were not toxic for reporter cells (Figure 4A). On the contrary, they both protected reporter cells from a lethal dose of soluble recombinant BAFF at roughly stoichiometric ratio ( Figure 4B). BAFF-deficient U937 cells were harmless to BAFFR:Fas reporter cells (Figure 4C), but furin-deficient U937 cells, which express membrane-bound BAFF, efficiently killed them (Figure 4D). This killing was inhibited in a dosedependent manner and with similar efficacy by both atacicept and belimumab, while the control antibody denosumab did not protect against membrane-bound BAFF (Figure 4D). An additional control showed that BAFF released into supernatant of furin-deficient U937 cells during the 17 h co-culture period could only kill about 30% of reporter cells (Figure 4E), indicating that the remaining toxicity of furin-deficient U937 cells on BAFFR:Fas reporter cells was due to membrane-bound BAFF ( Figure 4D). Taken together, these results indicate that both endogenous membrane-bound BAFF and residual soluble BAFF in U937 cells are bound and inhibited by atacicept and belimumab.

Opsonized Atacicept and Belimumab Do Not Induce Complement or Cell-Mediated Toxicity
IgGs that are opsonized on a cell surface activate complement or antibody-dependent cell-mediated cytotoxicity via their fragment crystallisable (Fc) regions. Atacicept was used as a negative control because its mutated Fc region activates neither of these pathways. Neither atacicept nor belimumab activated complement when bound to CHO cells expressing uncleavable BAFF (Figure 5A). In contrast, BCMA-Fc, a BAFF-binding decoy receptor with a wild type human IgG1 Fc portion, decreased viability of target cells in the presence of normal human serum, but not in de-complemented serum and not in CHO cells expressing no BAFF ( Figure 5A). Adalimumab, an anti-TNF antibody with reported ability to activate complement (26) had no activity in the absence of its target (Figure 5A). It is noteworthy that a purely dimeric preparation of BCMA-Fc, obtained after gel filtration, displayed the same activity as "crude" BCMA-Fc ( Figure 5A).

Antibody-dependent
cell-mediated cytotoxicity was monitored in a surrogate assay in which the activation of Fcγ receptor IIIa over-expressed in Jurkat T cells drives NFAT (nuclear factor of activated T cells)-dependent expression of luciferase. Neither atacicept nor belimumab activated reporter cells when bound to membrane-bound BAFF on CHO cells, while the positive control BCMA-Fc did ( Figure 5B). We conclude from these data that atacicept and belimumab inhibit their target cytokine(s) in soluble or membrane-bound forms, but do not kill BAFF-expressing cells, at least in vitro.

DISCUSSION
It was previously reported that belimumab inhibits BAFF, but not APRIL. In addition, belimumab does not inhibit any of the BAFF-APRIL heteromers, while atacicept inhibits BAFF, APRIL and heteromers thereof (16). Another difference is that soluble BAFF can exist as 3-mers, or can assemble as 60mers (27). Belimumab cannot inhibit BAFF 60-mers unless they first dissociate into 3-mers, whereas atacicept and another anti-BAFF antibody, tabalumab, can inhibit BAFF 60-mers (28)(29)(30). Belimumab is often described in the literature as an inhibitor of soluble BAFF. This conclusion probably finds its origin into the original description of the antibody, where belimumab could not stain IFNγ-stimulated human monocytes by flow cytometry, while another antibody called 12D6 could (14). This was later interpreted as " [. . . ], belimumab specifically recognizes the soluble, biologically active form of BLyS" (31). Our present results question this conclusion and re-enforce previous results obtained with over-expressed BAFF in CHO cells showing that belimumab can inhibit membrane-bound BAFF (28). The monoclonal antibodies 12D6, 2E5, and 9B6 can all stain human monocytes (7,14,18). It would be interesting to monitor their binding to wild type, furin-deficient and BAFF-deficient U937 cells side by side with belimumab to determine if they are really better binders of membrane-bound BAFF, or whether there might be other causes for their binding to monocytes.
It has long been recognized that BAFF is processed at a furin consensus site (7,8). Here we provide the first indication that in U937 cells, BAFF is cleaved by furin, with no or little contribution of other members of the pro-protein convertase subtilisin/kexin family. However, we did not investigate whether this is also true for other cell lines or primary cells. Like Baff -ko mice, mice with non-cleavable BAFF have few mature B cells, indicating that membrane-bound BAFF, unlike soluble BAFF, does not support naïve B cells (32). But when follicular B cells were restored by administration of recombinant soluble BAFF 3-mer, full maturation of B cells, measured by CD23 expression, required membrane-bound BAFF (32). In mice with non-cleavable BAFF, membrane-bound BAFF co-localized with medullary fibroblastic reticular cells in lymph nodes. These cells make more direct contacts with plasma cells than any other cell type, and they can support survival of plasma cells in vitro, alone or together with macrophages, while other lymph node cell types cannot (33). This suggests that membrane-bound BAFF may be more important for plasma cells than for naïve B cells, possibly through activation of the receptors TACI and BCMA, and that inhibition of membrane-bound BAFF might be relevant to the mechanism of action of belimumab in SLE patients.
In this study, we have observed that belimumab binds normally to BAFF stalk overexpressed in CHO cells but was less efficient than atacicept to inhibit it. This was however not the case in furin-deficient U937 cells in which both atacicept and belimumab similarly inhibited full-length membrane-bound BAFF (and residual soluble BAFF). Thus, BAFF binding and BAFF inhibitory activities did not correlate well only for the pair of belimumab -BAFF stalk, but not for the pairs belimumabfull length BAFF, atacicept-BAFF stalk or atacicept-fulllength BAFF. How can these results be reconciled in a molecular model? We have previously observed that atacicept binds and inhibits BAFF even when it is densely packed as 60-mers (30). It is therefore not surprising that atacicept can bind and inhibit membrane-bound BAFF irrespective of the length of the stalk region, as observed experimentally. We have also observed that, in contrast to intact belimumab, Fab fragments of belimumab did not or only very poorly inhibit the activity of BAFF 3-mers, probably because monovalent Fab can be displaced from BAFF by BAFFR (30). This would support a hypothetic model in which the first monovalent binding of belimumab to membrane-bound BAFF would allow cell staining by flow cytometry but would In contrast, the engineered stalk-less BAFF is much closer to the membrane. Atacicept binds BAFF from the side opposite to the membrane and has free access regardless of the length of the stalk. Belimumab is not only bulkier but also binds more on the side of BAFF along the entire height of the TNF homology domain. The first binding of belimumab to membrane-bound BAFF is probably always easy, but insufficient to inhibit the biological activity of BAFF. Inhibition only takes place upon binding of the second arm of the antibody to membrane-bound BAFF. Movements of the antibody to reach its second binding site might be compromised for steric hindrance reasons if BAFF is located too close to the cell membrane in BAFF stalk. be poorly inhibitory. Membrane-bound BAFF would only be inhibited upon binding of the second arm of the antibody that increases binding avidity, preventing efficient displacement by BAFFR. The molecular "gymnastic" that belimumab must accomplish to reach its second binding site on membrane-bound BAFF might be partially impaired for steric hindrance reasons in the specific case of BAFF stalk that is positioned very close to the cell membrane ( Figure 6).
Interestingly, we find that belimumab does not activate complement and Fc receptors in vitro. If this holds true in vivo, belimumab should not act by depleting cells expressing membrane-BAFF, in line with the current view of the mechanism of action of belimumab.
Belimumab was the first targeted treatment, approved in 2011, for the treatment of SLE, but its clinical efficacy is limited and not all patients benefit from treatment; this also holds true for other BAFF-blocking agents that have been tested in clinical trials (34). Loss of tolerance and development of auto-immunity is multifactorial, and SLE likely encompasses a spectrum of conditions with distinct etiologies that may be more or less dependent on BAFF (35,36). This diversity is also reflected in the various spontaneous or induced mouse models of the disease (37). Different forms of BAFF and APRIL may trigger BAFFR, TACI, and BCMA in distinct manners with effects that could both promote (e.g., stimulation of auto-reactive B cells) or counteract (e.g., stimulation of regulatory B cells) development of symptoms. Indeed, in a mouse model of lupus, intricate positive or negative actions of the different receptors on disease severity have been described (38). In this context, it might not be superfluous to know the fine specificity of belimumab and other BAFF inhibitors.

DATA AVAILABILITY
The authors declare that the data supporting the finding of this study are available within the article, in its Supplementary Information File and as a dataset doi: 10. 5281/zenodo.1481232.

AUTHOR CONTRIBUTIONS
PS and MV conceived experiments. CK-Q, DC, LW, and PS performed experiments with help of MV and CJ. PS wrote the paper. All authors reviewed the results and approved the final version of the manuscript.