Removal of Mannose-Ending Glycan at Asn2118 Abrogates FVIII Presentation by Human Monocyte-Derived Dendritic Cells

The development of an immune response against therapeutic factor VIII is the major complication in hemophilia A patients. Oligomannose carbohydrates at N239 and/or N2118 on factor VIII allow its binding to the macrophage mannose receptor expressed on human dendritic cells, thereby leading to factor VIII endocytosis and presentation to CD4+ T lymphocytes. Here, we investigated whether altering the interaction of factor VIII with mannose-sensitive receptors on antigen-presenting cells may be a strategy to reduce factor VIII immunogenicity. Gene transfer experiments in factor VIII-deficient mice indicated that N239Q and/or N2118Q factor VIII mutants have similar specific activities as compared to non-mutated factor VIII; N239Q/N2118Q mutant corrected blood loss upon tail clip. Production of the corresponding recombinant FVIII mutants or light chains indicated that removal of the N-linked glycosylation site at N2118 is sufficient to abrogate in vitro the activation of FVIII-specific CD4+ T cells by human monocyte-derived dendritic cells. However, removal of mannose-ending glycans at N2118 did not alter factor VIII endocytosis and presentation to CD4+ T cells by mouse antigen-presenting cells. In agreement with this, the N2118Q mutation did not reduce factor VIII immunogenicity in factor VIII-deficient mice. Our results highlight differences in the endocytic pathways between human and mouse dendritic cell subsets, and dissimilarities in tissue distribution and function of endocytic receptors such as CD206 in both species. Further investigations in preclinical models of hemophilia A closer to humans are needed to decipher the exact role of mannose-ending glycans in factor VIII immunogenicity.

The development of an immune response against therapeutic factor VIII is the major complication in hemophilia A patients. Oligomannose carbohydrates at N239 and/or N2118 on factor VIII allow its binding to the macrophage mannose receptor expressed on human dendritic cells, thereby leading to factor VIII endocytosis and presentation to CD4+ T lymphocytes. Here, we investigated whether altering the interaction of factor VIII with mannose-sensitive receptors on antigen-presenting cells may be a strategy to reduce factor VIII immunogenicity. Gene transfer experiments in factor VIII-deficient mice indicated that N239Q and/or N2118Q factor VIII mutants have similar specific activities as compared to non-mutated factor VIII; N239Q/N2118Q mutant corrected blood loss upon tail clip. Production of the corresponding recombinant FVIII mutants or light chains indicated that removal of the N-linked glycosylation site at N2118 is sufficient to abrogate in vitro the activation of FVIII-specific CD4+ T cells by human monocyte-derived dendritic cells. However, removal of mannose-ending glycans at N2118 did not alter factor VIII endocytosis and presentation to CD4+ T cells by mouse antigen-presenting cells. In agreement with this, the N2118Q mutation did not reduce factor VIII immunogenicity in factor VIII-deficient mice. Our results highlight differences in the endocytic pathways between human and mouse dendritic cell subsets, and dissimilarities in tissue distribution and function of endocytic receptors such as CD206 in both species. Further investigations in preclinical models of hemophilia A closer to humans are needed to decipher the exact role of mannose-ending glycans in factor VIII immunogenicity.

INTRODUCTION
Five to thirty percent of patients with hemophilia A develop inhibitory anti-factor VIII (FVIII) antibodies following replacement therapy with therapeutic FVIII (1). The reasons for the elevated immunogenicity of therapeutic FVIII, as compared to other therapeutic glycoproteins, have been intensively investigated. These include congenital (2,3) or bleeding-induced (4) chronic inflammation favoring the recruitment and activation of antigen-presenting cells (APCs) and of immune effectors, and disequilibrium in regulatory elements of the humoral or cellular arms of the adaptive immune system (5)(6)(7). A mandatory step for developing an anti-FVIII immune response is the endocytosis of the exogenously administered FVIII by APCs, such as dendritic cells (DCs) and macrophages, and the presentation of FVIIIderived peptides on major histocompatibility complex class II (MHC II) molecules by APCs to CD4+ T lymphocytes (8,9).
These last decades, we and others have explored the mechanisms by which FVIII is endocytosed by APCs either by targeting endocytic receptors (10,11) or by masking and mutating specific amino acid residues in FVIII (11)(12)(13). While there is evidence that residues in the C1 (R2090, K2092, and F2093) (12) and C2 (R2215 and R2220) (13) domains of FVIII are involved in the endocytosis of FVIII by human monocytederived DCs (MO-DCs), the nature of the involved endocytic receptor(s) remains unknown. In vitro experiments using an excess of mannan demonstrated the importance of mannosesensitive receptors for the endocytosis of different FVIII products by human DCs and for the ensuing presentation of FVIII-derived peptides to T cells (10,14). FVIII is a heterodimeric glycoprotein composed of a heavy chain (A1-a1-A2-a2-B domain) and a light chain (a3-A3-C1-C2 domain) linked by non-covalent binding. FVIII contains 20 N-glycosylations that are unequally distributed over the FVIII molecule: two on the A1 domain, one on the A3 and C1 domains and the remaining on the B domain (15). Both plasma-derived FVIII, recombinant fulllength (FL) and B domain-deleted FVIII (BDD-FVIII) have been reported to contain mannose-ending glycans at positions N239 and N2118 of the A1 and C1 domain, respectively (16,17).
Interestingly, both pre-incubation of DCs with an antibody toward the macrophage mannose receptor (CD206), and enzymatic removal of mannosylated glycans on FVIII, lead to reduced FVIII presentation to a human CD4 + T cell line (10). Conversely, a recombinant CD206 construct was shown to bind both the light and heavy chains of BDD-FVIII. Recombinant FL-FVIII and BDD-FVIII products, commercially available at the time of the studies, interact with CD206 (10,18). While mannose-ending glycans on foreign glycoproteins generally mediate pathogens recognition and elimination by the immune system, oligomannose carbohydrates on self-antigens and their binding to CD206 have been implicated in their catabolism (19). Here, we studied the involvement of the two mannose-ending glycans present at positions N239 and N2118 of FVIII in its immunogenicity in vitro and in vivo. Cloning of Wild-Type and Mutant B Domain-Deleted FVIII for in vivo Gene Transfer All clonings and generation of FVIII variants were performed using a BDD-FVIII coding sequence. Indeed, both FL-FVIII and BDD-FVIII demonstrate similar levels of immunogenicity in hemophilia A patients, are endocytosed by human MO-DCs through mannose sensitive pathway (14), bind to CD206 (18), and present mannose-exposed sugars at positions N239 and N2118 (17).
2FVIII contains the 30 N-terminal amino-acids of the B domain of FVIII, and hence the N-glycosylation site NAT at position 757-759 (23). The pcDNA3.1-2FVIII plasmid containing the 2FVIII cDNA was mutated using the QuickChange II XL mutagenesis kit (Stratagene, La Jolla, CA, United States). N239 and/or N2118 were mutated to Q, using the protocol provided by Stratagene. The wild-type and mutated 2FVIII cDNA were inserted into the pLIVE vector (Mirus, Madison, WI, United States).
Cloning, Production and Purification of Wild-Type and Mutant B Domain-Deleted FVIII cDNA encoding human BDD-FVIII (HSQ), containing the 14amino acid segment SFSQNPPVLKRHQR in place of the B domain, cloned in the ReNeo plasmid (24) was used as a template to generate the FVIII 239Q , FVIII 2118Q , FVIII 2118A , and FVIII 239Q/2118Q mutants by splicing-by-overlap extension mutagenesis. Presence of the mutations was confirmed by standard sequencing analysis. BHK-M cells (a kind gift from Prof P. Lollar, Emory University, Atlanta, Georgia, United States) were transfected and selected for neomycin resistant clones using Geneticin-sulfate (500 µg/ml, Sigma Aldrich, St-Louis, MO, United States). Screening of FVIII producing clones was performed by detection of the FVIII:antigen (FVIII:Ag) that refers to the amount of FVIII protein and FVIII:C that refers to the detectable pro-coagulant activity of FVIII. FVIII:Ag was detected by a sandwich ELISA using an anti-FVIII light chain specific monoclonal antibody (mAb) (Clone ESH-8, BioMedica Diagnostics, Stanford, CA, United States), and a biotinylated anti-FVIII heavy chain mAb (Clone GMA-8015, Green Mountain Antibodies, Burlington, United States), as capture and detection antibodies. FVIII:C was measured by chromogenic assay (Siemens Healthcare Diagnostic, Marburg, Germany). For quantification of FVIII in supernatants and cell lysates, a sandwich ELISA with ESH8 mAb and the biotinylated SAF8C polyclonal Ab, as capture and detection antibodies was performed. The stable expression of wild-type and mutated FVIII by BHK-M cells, and FVIII purification were performed as previously described (13). Briefly, the highest expressing clones for each FVIII was selected (according to the quantity of FVIII:C produced by 1.10 6 million cells/ml in 24 h) and were scaled up to near confluency before switching the medium to serum-free AIM-V medium (Thermo Fisher Scientific). Medium was collected every 24 h and cells were replenished with fresh AIM-V medium. FVIII purification was performed by affinity chromatography on VIII select column (GE Healthcare, Chicago, IL, United States), followed by anion-exchange chromatography on HiTrap Resource Q column (GE Healthcare), as previously described (13). Purified FVIII was analyzed by 4-12% gradient SDS-PAGE (1 µg per well) with and without activation by bovine thrombin (Sigma-Aldrich). A silver staining of the gel was performed to detect proteins. Specific activity of purified WT and FVIII mutants were evaluated by chromogenic assay and absorbance at 280 nm with molar extinction coefficient of 256,300 M −1 cm −1 .

Cloning and Site-Directed Mutagenesis of the Light Chain of FVIII
A N2118A mutated light chain of human FVIII was amplified using appropriate primers from the commercially available cDNA (pSP64-VIII, ATCC, Manassas, VA, United States). The mutated light chain fragment was cloned in the pNUT vector, with the signal sequence of human IgG kappa. The mutated light chain was produced in DMEM/F12, 5% fetal calf serum (FCS) by chinese hamster ovary (CHO) K1 stably transfected, and purified by ion-exchange chromatography. Briefly, a Resource S column column (GE Healthcare) was equilibrated with 10 mM histidine-HCl, 5 mM CaCl 2 , 250 mM NaCl, 0.01% tween 80 (pH 6.0). The culture supernatant in 10 mM MES, 5 mM EDTA, 0.01% tween 80, pH 6.0 was injected on the column. The recombinant light chain was eluted with a NaCl gradient, dialyzed in RPMI for 2 h at 4 • C and quantified by sandwich ELISA using the human monoclonal anti-C2 conformational epitope-specific IgG (clone BO2C11, gift from JM Saint Rémy, Katholieke Universiteit Leuven, Leuven, Belgium), and ESH8 monoclonal antibody. Human recombinant FVIII was used as a standard. The wild-type light chain was a kind gift from Dr. E. L. Saenko (Center for Vascular and Inflammatory Diseases, University of Maryland School of Medicine, Baltimore, United States) and had been prepared from human FVIII as described (25).

FVIII Gene Transfer to FVIII-Deficient Mice and Correction of Bleeding Time
FVIII-deficient mice were injected in the tail vein with pLIVE plasmid (100 µg) encoding wild-type FVIII ( 2FVIII) (22), or the N239Q ( 2FVIII 239Q ), N2118Q ( 2FVIII 2118Q ) or N239Q/N2118Q ( 2FVIII 239Q/2118Q ) mutants (Figure 1A), in a large volume (10% of body weight) of 0.9% NaCl within 5 s (26). Four days later, blood was collected. FVIII:C and FVIII:Ag were quantified in citrated plasma using a chromogenic assay (Siemens Healthcare Diagnostics) and an Asserachrom kit (Stago, Asnières-sur-Seine, France), respectively. On day five, mice were anesthetized and 3 mm of the distal tail were cut, blood was collected during 30 min in 50 ml of saline buffer at 37 • C. Erythrocytes were then pelleted at 1500 g and lysed in H 2 O. The amount of released hemoglobin, proportional to blood loss, was determined by measuring the optical density at 416 nm, using a standard curve prepared upon lysis of 20 to 100 µl of mouse blood. The half-life of endogenously produced FVIII was performed as previously described (27). Briefly, 6 days after the hydrodynamic injection, mice were injected intravenously with 500 µg of biotin-N-hydroxysuccinimide ester (Pierce, Rockford, United States) dissolved in saline buffer. Blood was collected on citrated tubes at different time points after injection and residual biotinylated FVIII was measured by ELISA, with a polyclonal anti-human FVIII antibody (SAF8C, Kordia, Leiden, Netherlands) as capture antibody and streptavidin-HRP for detection.

Activation of FVIII-Specific T Cells
As a source of FVIII-specific T-cells, the murine FVIII-specific T-cell hybridoma 1G8-A2, which was generated by immunizing SureL1 mice with BDD-FVIII (see Supplementary Material) (18), and the human FVIII-specific T-cell line D9E9 (from Dr. M. Jacquemin) (29) were used. Five day old MO-DCs from DRB1 * 0101 healthy donors or mitomycin C-treated splenocytes from SureL1 mice were incubated for 24 h with the FVIII-specific T-cell hybridoma 1G8-A2 (10,000 MO-DCs or 200,000 splenocytes for 100,000 T cells) with FVIII in X-VIVO 15 medium (Lonza). Levels of secreted interleukin-2 (IL-2) were assessed using BD OptEIA mouse IL-2 ELISA set (BD Biosciences). When the human FVIII-specific T-cell line D9E9 was used, 5,000 T cells were incubated with FIGURE 1 | In vivo production of wild-type and mutated B domain-deleted FVIII lacking the N-glycosylation sites at 239 and 2118. (A) Human BDD-FVIII was modified by site-directed mutagenesis. The four glycosylation sites are depicted with pin symbols, the full pins represent N239 and N2118, on A1 and C1 domain, respectively. Four constructs were generated from the 2FVIII sequence: 2FVIII wild-type FVIII ( 2FVIII WT ), single N239Q mutant ( 2FVIII 239Q ), single N2118Q mutant ( 2FVIII 2118Q ) and double mutant FVIII ( 2FVIII 239Q/2118Q ). (B) The different pLIVE constructs were injected to FVIII-deficient mice (4 to 11 mice per group) and blood was recovered after 4 days. The graph depicts as boxes and whiskers the levels of FVIII in plasma, as measured by ELISA (FVIII:Ag, empty bars) and by functional chromogenic assay (FVIII:C, full bars) using normal human plasma as a standard. Results are a pool of two independent experiments. (C) FVIII-deficient mice received hydrodynamic injections of pLIVE constructs encoding wild-type (six mice, empty squares) and double mutant FVIII (six mice, empty triangles) FVIII. Six days later, activated NHS-biotin was injected to each mouse, and residual biotinylated FVIII was measured at indicated time points. Data represent the percentage of residual biotinylated FVIII (mean ± SEM) quantified at t = 0 (15 min after biotin injection) which was a set at 100% for each mouse. For calculation of half-life, the best fit was obtained with non-linear regression with two exponential decay (Prism v5.0b, GraphPad Software, Inc.). (D) Four days after hydrodynamic injections of pLIVE constructs encoding wild-type (five mice, empty squares) and double mutants FVIII (five mice, empty triangles), FVIII-deficient mice were subjected to tail clip experiments and blood loss was measured. Untreated FVIII-deficient mice (8 mice, empty circles) and wild-type C57BL/6 mice (eight mice, full circles) were used as controls. Horizontal bars depict medians. Statistical analysis was performed using the non-parametric two-tailed Mann-Whitney U test with a confidence interval of 95%. **P < 0.01; ***P < 0.001; ****P < 0.0001; ns: not significant.

Administration of FVIII and Characterization of the Anti-FVIII Immune Response
To evaluate the half-life of recombinant FVIII, FVIII-deficient mice were administered intravenously with FVIII HSQ or FVIII 2118Q (100 µl, 10 nM). Plasma was collected at different time points and FVIII:Ag was measured by ELISA using the antilight chain mAb ESH-8, and an anti-FVIII heavy chain mAb GMA-8015, as coating and detection antibodies, respectively, and human plasma as a standard (Siemens Healthcare Diagnostics).
To investigate the immunogenicity of recombinant FVIII, FVIIIdeficient mice or VWF/FVIII-deficient mice were injected intravenously with 0.5 µg FVIII HSQ or FVIII 2118Q , once a week for 4 weeks. Endotoxin levels in the different recombinant FVIII were below the accepted threshold (i.e., <0.01 ng endotoxin/20 g mouse weight) as assessed using the ToxinSensor Chromogenic LAL Endotoxin Assay Kit (Genscript, Piscataway, NJ, United States). Blood was collected 5 days after the last FVIII injection. ELISA plates were coated with FVIII (1 µg/ml, Recombinate, Baxter, Maurepas, France) overnight at 4 • C. After blocking with PBS-1% BSA, plasma was incubated for 1 h at 37 • C. Bound IgG were revealed with a HRP-coupled polyclonal goat anti-mouse IgG antibody (Southern Biotech, Anaheim, CA, United States) and the OPD substrate. The mouse monoclonal anti-FVIII IgG mAb6 (from Dr. J. M. Saint-Remy) was used as a standard. FVIII inhibitors were measured by incubating heat-inactivated mouse plasma with human standard plasma for 2 h at 37 • C. FVIII residual pro-coagulant activity was measured by chromogenic assay. Results are expressed in Bethesda titers (BU/ml) that correspond to the reciprocal dilution of the mouse plasma that yields 50% residual FVIII activity.
The half-lives of the endogenously produced FVIII were evaluated following injection of activated NHS-biotin. FVIII clearance in FVIII-deficient mice followed a double-exponential distribution. Half-lives of 2FVIII WT in the fast and slow phases of elimination were 30 and 260 min, respectively, with 38% of the FVIII being removed in the fast phase (R 2 = 0.91, Figure 1C). Half-lives of 2FVIII 239Q/2118Q in the fast and slow phases of elimination were 48 and 276 min, respectively, with 50% of the FVIII being removed in the fast phase (R 2 = 0.96). Thus, the half-lives of 2FVIII WT and 2FVIII 239Q/2118Q were similar.
We further analyzed the capacity of endogenously produced 2FVIII WT and 2FVIII 239Q/2118Q to correct blood loss in tail clipping experiments. We first confirmed that wild-type nonhemophilic mice lose significantly less blood than FVIII-deficient mice (P = 0.001, Figure 1D). The hydrodynamic injection of 2FVIII WT or 2FVIII 239Q/2118Q -encoding plasmids into FVIII-deficient mice corrected blood loss as compared to untreated mice (P = 0.004 and P = 0.013, respectively). Interestingly, blood loss in FVIII-deficient mice treated with the 2FVIII 239Q/2118Q construct was not significantly different from that of wild-type non-hemophilic mice. The fact that FVIII-deficient mice treated with the 2FVIII WT -encoding plasmid bled significantly less than mice treated with the 2FVIII 239Q/2118Q -encoding plasmid (P = 0.010) is probably due to the fact that circulating levels of FVIII were significantly greater in the former group of mice (373 ± 35 IU/ml, mean ± SEM, Figure 1B) than in 2FVIII 239Q/2118Q -treated mice (69 ± 8 IU/ml).
the light chain of the thrombin-digested FVIII 2118Q mutant, as compared to the A1 domain and light chain of the thrombindigested FVIII HSQ confirmed the removal of the N-linked glycosylation sites ( Figure 2C). Besides, the similar molecular weights observed for the A1 fragment (digested FVIII) in the case of FVIII HSQ and FVIII 2118Q and the digested light chain of the FVIII 239Q and FVIII HSQ suggests that site directed mutagenesis at N239 or/and N2118 did not affect the N-glycosylations at positions N41 and N1810, respectively. MS analysis would confirm this observation. Further, a digestion of FVIII HSQ and FVIII 2118Q with the N-glycosidase F was performed to confirm that the lower molecular weight of the light chain is due to the removal of N-linked glycosylation at N2118 and not due to an uncontrolled proteolytic cleavage (Supplementary Figure S1).
The purified FVIII mutants exhibited specific activities between 4600-6200 IU/mg ( Figure 2D) similar to that of FVIII HSQ . Furthermore, FVIII HSQ and the FVIII mutants bound similarly to immobilized VWF and phosphatidylserine (Supplementary Figure S2). In order to decipher whether the removal of N239 or/and N2118 affects the production and/or secretion of FVIII, we measured secreted (supernatant) and intracellular (cell lysate) FVIII after 24 h of culture of BHK-M cells. Intracellular FVIII levels were similar for the FVIII 239Q , FVIII 2118Q and FVIII HSQ transfected cells (106 ± 18 ng/10 6 cells; 195 ± 31 ng/10 6 cells; 161 ± 28 ng/10 6 cells, respectively). In contrast, the intracellular FVIII levels were significantly decreased in the case of FVIII 239/2118Q (70 ± 9 ng/10 6 cells/24 h, P < 0.05, Figure 2E). Comparison of the ratios of secreted versus intracellular FVIII levels ( Figure 2E, inset) confirms that the N239Q mutation, either alone or in combination with the N2118Q mutation, is associated with a hampered secretion of FVIII.

Removal of the N2118 Glycosylation Abrogates FVIII Presentation by Human Dendritic Cells
To investigate the potential of the purified FVIII HSQ and mutants to be presented to CD4+ T cells by human APCs, we generated a mouse T-cell hybridoma specific for human FVIII and restricted to the HLA-DRB1 * 01:01 allele, referred to as 1G8-A2. The activation of 1G8-A2 by FVIII-loaded human MO-DCs from a HLA-DRB1 * 01:01 donor was evaluated by measuring IL-2 secretion. 1G8-A2 is specific for the 2013 LFLVYSNKC 2022 core peptide, located in the C1 domain of FVIII light chain (Figure 3A and Supplementary Figure S3). As previously reported (10), T-cell activation was drastically reduced when MO-DCs were pre-incubated with mannan (1 mg/ml) or EDTA (5 mM) prior to incubation with FVIII HSQ (Figure 3B). Likewise, removal of the N2118 glycosylation site resulted in a drastic reduction in 1G8-A2 activation by MO-DCs (39.5 ± 13.4 pg/ml, mean ± SD) as compared to the use of FVIII HSQ (572.9 ± 151.9 pg/ml, P < 0.0001 at 10 nM FVIII, Figure 3C). A different FVIII variant with an N2118A instead of an N2118Q mutation also failed to activate 1G8-A2 (35.3 ± 2.1 pg/ml, P < 0.001 as compared to FVIII HSQ ). In contrast, co-incubation of cells with the FVIII 239Q mutant yielded a significant though less marked reduction in IL-2 secretion (377.7 ± 96.8 pg/ml, P = 0.002 as compared to FVIII HSQ ). Of note, removal of both N239 and N2118 glycosylation sites did not have an additional effect on T-cell activation over the removal of the N2118 site alone (Supplementary Figure S4).
To confirm this observation, we used a different experimental set-up (different HLA context and using a T-cell line with a different epitope specificity), in which we compared the presentation of a N2118A mutated or wild-type FVIII light chain (25) to the HLA-DRB1 * 15:01-restricted human D9E9 T-cell line which is specific for the 2144 IIARYIRLHPTHYSIRST 2164 C1 domain peptide (29) (Figure 3A). As a source of APCs, we compared HLA-matched MO-DCs to the human BO2C11 B-cell line (30). The activation of D9E9 T cells by MO-DCs incubated with the wild-type FVIII light chain was drastically reduced in the presence of mannan (Figure 3D), thus reproducing the data obtained with complete FVIII HSQ (Figure 3B). In contrast, BO2C11 B cells activated D9E9 T cells when incubated in the presence of either wild-type or mutated light chains (Figures 3E,F), indicating that the absence of the mannoseending glycan on the C1 domain does not alter the conformation of the BO2C11 target epitope on the FVIII C2 domain. It also confirmed that FVIII light chain endocytosis by BO2C11 B cells, which is mediated by the B-cell receptor, is insensitive to the presence of mannan (Figure 3E). Since removal of the glycans at position N2118 was sufficient to significantly reduce FVIII presentation to CD4+ T cells, we restricted the rest of the study to FVIII 2118Q .

Mannose-Ending Glycans at Position 2118 do Not Modulate FVIII Endocytosis and Presentation by Mouse APCs
We have shown earlier that mouse BM-DCs, in contrast to human MO-DCs, do not endocytose FVIII through mannosesensitive endocytosis pathways (31). Here, we compared the ability of human MO-DCs and HLA-DRB1 * 01:01-transgenic Sure-L1 mouse DCs to endocytose and present FVIII HSQ and FVIII 2118Q to the FVIII-specific T-cell hybridoma 1G8-A2. While a 50% reduction of endocytosis by human MO-DCs was observed with FVIII 2118Q compared to FVIII HSQ (Figure 4A), mouse BM-DCs internalized FVIII HSQ , and FVIII 2118Q in a similar manner ( Figure 4B). Likewise, mouse splenic APCs activated 1G8-A2 in a similar manner irrespective of the source of FVIII used: FVIII HSQ or FVIII 2118Q (Figure 4D) whereas only around 10% of T-cell activation by human MO-DCs was measured with FVIII 2118Q (Figure 4C).

The Absence of N2118 Mannose-Ending Glycan Does Not Reduce FVIII Immunogenicity in Mice
We then investigated the consequence of removing the N2118 glycosylation site on FVIII immunogenicity in FVIII-deficient mice. First, we confirmed that kinetics of FVIII elimination in both the slow and fast phases did not differ statistically between FVIII HSQ and FVIII 2118Q (Figure 5A), confirming the observations obtained with the endogenous expression of 2FVIII 239Q/2118Q (Figure 1C). FVIII-deficient mice were injected with 10 nM FVIII 2118Q or FVIII HSQ once a week for 4 weeks, and anti-FVIII IgG and inhibitory titers were measured in serum. The anti-FVIII IgG titers were similar whether mice were treated with FVIII 2118Q or FVIII HSQ (17362 ± 9632 AU versus 18758 ± 3166 AU, respectively, mean ± SEM, Figure 5B). Likewise, the inhibitory titers in the serum of FVIII 2118Q -treated mice were similar to that in the serum of FVIII HSQ -treated mice (810 ± 527 BU/ml versus 639 ± 455 BU/ml, respectively, Figure 5C). Next, we investigated the immunogenicity of FVIII 2118Q and FVIII HSQ in the absence of VWF in double FVIII/VWF-deficient mice. As seen in the case of FVIII-deficient mice, neither anti-FVIII IgG titers nor inhibitory titers differed between double deficient mice treated with FVIII HSQ (2712 ± 2599 AU; 174 ± 195 BU/ml, Figure 5D) or FVIII 2118Q (3979 ± 4384 AU; 180 ± 175 BU/ml, Figure 5E). Taken together, our results indicate that oligomannose carbohydrates at N2118 do not play a major role in FVIII immunogenicity, at least in mice.

DISCUSSION
Missense mutations in 580 (42%) of the 1384 amino acids that encompass activated FVIII lead to moderate, mild or severe hemophilia A (HAMSTeRS database November 2016) 1 . This illustrates the critical relationship between the structure and function of proteins in general, and of FVIII in particular, which prompted us to estimate the repercussion of removing N-glycosylation sites of FVIII on its pro-coagulant activity. The recombinant FVIII 239Q/2118Q demonstrated specific activity similar to that of its respective non-mutated counterpart. Accordingly, 2FVIII 239Q/2118Q protected mice from experimentally induced blood loss, and FVIII 239Q/2118Q bound normally to VWF and phosphatidylserine. These observations indicate that removal of oligomannose carbohydrates at N239 and N2118 does not alter the pro-coagulant activity of FVIII.
Circulating levels of 2FVIII in vivo and levels of production of FVIII HSQ in vitro were similar or 2-fold lower when the molecules were mutated at N2118 and 9 to 14-fold lower when mutations were located at N239. In support of our data, Wei et al. have shown that, in contrast to mutation at N2118, mutation at N239 drastically reduces FVIII production (32). At the cellular level, the efficient transport of FVIII from the endoplasmic reticulum (ER) to the Golgi apparatus requires the interaction of FVIII with the mannose-binding lectin LMAN1 (33). The interaction of FVIII with the carbohydrate recognition domain of LMAN1 implicates either high mannose-containing N-linked glycosylations, or protein-protein interactions (34). Our observation of a low ratio of secreted versus intracellular FVIII in the case of FVIII 239Q and FVIII 239/2118Q suggests an essential role for the N239 glycan in FVIII export. Comparatively, the N2118 glycan plays a negligible role in FVIII production and secretion. We hypothesize that removal of the N239 glycosylation site, which is highly conserved among orthologous FVIII molecules (32), synergizes with the absence of N-glycosylation sites of the B-domain that is missing in FVIII HSQ and 2FVIII, resulting in inefficient intracellular transport of FVIII. Of note, the in vivo half-life of 2FVIII was not affected by the removal of the 239 and 2118 N-glycosylation sites. Accordingly, the half-lives of recombinant FVIII HSQ and FVIII 2118Q were also similar. Taken together, our data suggest that the oligomannose carbohydrates at N239 and N2118 of FVIII, that mediate binding to CD206 expressed on human macrophages and dendritic cells (10,18), and to CLEC4M expressed on human liver and lymphatic sinusoidal endothelial cells (35), or to the soluble lectin Galectin-1 (36) are not involved in FVIII catabolism, at least in mice.
The first step of the primary immune response to therapeutic FVIII is its uptake by APCs. Our earlier work had documented the importance of mannose-ending glycans on FVIII for its recognition and internalization by MO-DCs, used as a model of APCs, and its subsequent presentation to CD4+ T cells (10). The phenomenon was true for both full-length and BDD-FVIII, thus highlighting the contribution of mannoseending glycans outside the B domain for FVIII recognition by human APCs (10,14). In humans, the presence of high mannose glycans is not frequent on secreted proteins (37). The deliberate mannosylation of non-self-antigens has been shown to enhance their endocytosis by human DCs and subsequent presentation to antigen-specific T lymphocytes (38,39). Indeed, human DCs express several endocytic C-type lectin receptors, including CD206, DC-SIGN and dectin 2, which bind exposed mannose residues on glycoproteins through their carbohydrate recognition domains (40,41). Interestingly, CD206 is not only expressed by in vitro derived DCs, but also by some subsets of DCs in the skin, tonsils and blood (42). The present work demonstrates the prevalent role of mannosylated glycans at N2118 in mediating FVIII uptake by MO-DCs, in the context of Half-life of wild type and N2118Q mutated FVIII. FVIII HSQ and FVIII 2118Q were administered to FVIII-deficient mice and the residual FVIII:Ag was measured at different time points (n = 6 mice per time point) using a sandwich ELISA. The data is plotted as percentage of initial FVIII activity (measured 5 min after administration) versus time (mean ± SEM). Representative of two independent experiments. Inset. The slow and fast phases of FVIII clearance were determined by fitting the data to a two phase decay curve. The immunogenicity of FVIII 2118Q was evaluated in FVIII-deficient mice: (B,C), seven mice for both FVIII HSQ and FVIII 2118Q group and VWF/FVIII double-deficient mice; (D,E) 8 and 6 mice for FVIII HSQ and FVIII 2118Q group, respectively. One week after the fourth injection, blood samples were collected and anti-FVIII IgG titers (B,D) quantified using a purified mouse monoclonal anti-FVIII IgG (Mab6) as a standard. Results are expressed in arbitrary units (AU). Inhibitory titers toward FVIII (C,E) were measured by chromogenic assay. Statistical analysis was performed using the non-parametric two-tailed Mann-Whitney U test. ns, not significant. the absence of the heavily glycosylated B domain. Mutations of N239 had only a marginal effect on T-cell activation by MO-DCs, which may relate to the fact that glycans at N239 contain highmannose, as well as hybrid and complex structures, while glycans at N2118 exclusively contain high-mannose structures (Man 5-9) (15)(16)(17).
We and others have recently documented the crucial role played by charged residues in the C1 and C2 domains of FVIII for its recognition and endocytosis by MO-DCs (11,13). Mutation to alanine residues of R2215 and R2220 in the protruding loops of the C2 domain of FVIII HSQ (referred to as FVIII R2215−20A ) as well as of R2090, K2092 and F2093 in the protruding loops of C1 (referred to as FVIII C1 ) independently reduced FVIII endocytosis by ≥50% and presentation to T cells by ≥75%. Importantly, FVIII R2215−20A and FVIII C1 contain the N-linked high-mannose carbohydrates at position N2118. Conversely, FVIII 2118Q contains the native amino acids at positions 2092, 2093, 2090, 2215 and 2220, but is not presented to T cells by MO-DCs. These results point toward a critical role played by the C domains, and particularly the C1 domain, in the recognition and internalization of FVIII by MO-DCs. They also suggest that the engagement of the three FVIII entities -charged residues in the C1 domain, charged residues in the C2 domain and glycans at N2118, is required for FVIII uptake, and pave the way toward the identification of the surface receptors involved in the process.
Removal of the N-linked glycosylation site at position 2118 did not reduce the immunogenicity of FVIII in FVIII-deficient mice. This is reminiscent of the lack of reduced immunogenicity of the FVIII R2215−20A and FVIII C1 in hemophilic mice, despite a drastic reduction in endocytosis by DCs (13). Interestingly, the immunogenicity of FVIII C1 was reduced as compared to that of FVIII HSQ in double FVIII/VWF-deficient mice, thus highlighting a role for VWF in impairing the potential beneficial effect of mutations in C1 on a reduction of FVIII immunogenicity in vivo. We had previously identified CD206 as one of the receptors implicated in the mannose-sensitive uptake of FVIII by MO-DCs. The activation of T cells was partially blocked when MO-DCs and FVIII were incubated in the presence of an anti-CD206 antibody, and FVIII was demonstrated to interact with soluble and immobilized CD206 (10). Importantly, the interaction between FVIII and CD206 was inhibited in a dosedependent manner by VWF (43). A role for endogenous VWF in preventing the reduction of FVIII 2118Q immunogenicity in FVIII-deficient mice is, however, not probable owing to the fact that there was no reduction in FVIII 2118Q immunogenicity in double FVIII/VWF-deficient mice.
The mutation of charged residues in the C1 and C2 domains inhibits endocytosis both by human MO-DCs and by mouse BM-DCs (12,13). In contrast to human MO-DCs, however, mouse BM-DCs do not internalize FVIII through mannosesensitive pathways (31). Accordingly, FVIII 2118Q and FVIII HSQ were captured and presented to a similar extent to T cells when mouse BM-DCs or splenic APCs were used. Of note, we observed, as reported previously, that CD206 is expressed by mouse macrophages from the red pulp of the spleen (44), and by a subset of mouse splenic dendritic cells (45) that are present in freshly isolated splenocytes from Sure-L1 mice (data not shown), while in human spleen CD206 has been reported to be expressed by lining venous sinus cells and not by macrophages (46,47). Overall, the organ-specific expression of C-type lectin receptors is different between mice and human (48). This is of importance in view of the fact that we have previously shown the accumulation of exogenously administered FVIII at the level of metallophilic macrophages in the marginal zone of the spleen of FVIII-deficient mice (49). The identification of preclinical animal models that better mimic the distribution of C-type lectin receptors in the human is necessary to confirm the reduced immunogenicity of FVIII 2118Q in vivo, and its potential safety for replacement therapy in patients with hemophilia A.

DATA AVAILABILITY STATEMENT
The datasets generated for this study are available on request to the corresponding author.

ETHICS STATEMENT
Ethical review and approval was not required for the study on human participants in accordance with the local legislation and institutional requirements. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements. The animal study was reviewed and approved by the Charles Darwin, Sorbonne Université, #2058.04.