Factor VIII exhibits chaperone-dependent and glucose-regulated reversible amyloid formation in the endoplasmic reticulum

Hemophilia A (HA), an X-chromosome–linked bleeding disorder affecting ∼10 000 males in the United States, results from deficiency in coagulation FVIII, a component of the intrinsic blood-clotting cascade^1,2  presently treated by protein replacement. Although development of recombinant-derived factor VIII (FVIII) significantly decreased risk of adventitious viral contamination, it greatly increased the cost of treatment, partially due to the low level of FVIII secretion from recombinant mammalian host cells.³  FVIII is synthesized and translocated into the endoplasmic reticulum (ER) lumen where only properly folded proteins traffic to the Golgi compartment. The accumulation of unfolded/misfolded proteins in the ER activates the unfolded protein response (UPR), an adaptive signaling pathway evolved to resolve ER protein misfolding.^4-6  FVIII is susceptible to misfolding in the ER and was the first native endogenous protein shown to activate the UPR by binding to the ER protein chaperone immunoglobulin binding protein (BiP)/GRP78.^7,8  FVIII expression and an unresolved UPR lead to apoptosis.⁹  A comprehensive understanding of the factors required for FVIII folding and secretion is unknown.

FVIII has the domain structure A1-A2-B-A3-C1-C2, where the B domain contains 18 potential N-linked glycosylation sites.¹  The A domains have amino acid homology to clotting factor V (FV) and ceruloplasmin, and structural homology to double β-barrel proteins, resembling the fold of cupredoxin domains.^10,11  Energy depletion, with inhibitors of mitochondrial oxidative phosphorylation and glycolysis, selectively inhibits FVIII anterograde trafficking in the early secretory pathway,^12,13  without affecting the trafficking of other proteins, including the homologous clotting FV and von Willebrand factor, even within the same cell.^12,14  A portion of FVIII traffics to the Golgi apparatus for processing by Furin to produce a heterodimer composed of an amino-terminal ∼200-kDa heavily glycosylated heavy chain (HC; A1-A2-B) in complex with a carboxy-terminal ∼80-kDa light chain (LC; A3-C1-C2) linked through 2 copper ions in each of the A1 and A3 domains¹⁵  that probably stabilize the A1 and A3 domains to promote A1-A3 interaction.^16,17  Unfortunately, the molecular basis of FVIII folding and trafficking through the secretory pathway is poorly defined. Here, we demonstrate that FVIII forms amyloid-like structures in the ER that can disaggregate and refold, and be secreted as functional FVIII in mammalian cells. In addition: (1) a short amino acid motif, which we termed Aggron, in FVIII is necessary and sufficient to seed aggregation; (2) glucose (Glc) metabolism is required to maintain FVIII solubility; and (3) 2 chaperone families, calnexin (CANX)/calreticulin (CRT) and BiP, promote solubility, the latter of which is also required for FVIII disaggregation by interaction with Aggron.

Parental Chinese hamster ovary cells (CHO-K1) and 2 CHO-K1 clones were engineered for constitutive FVIII expression (10A1),⁸  or, for inducible FVIII expression (H9),¹⁸  were previously described. COS-1 (ATCC CRL-1650), 293T (ATCC CRL-3216), and HepG2 (ATCC HB-8065) were obtained from the American Type Culture Collection (ATCC).

All reagents are specified in supplemental Methods (available on the Blood Web site).

Plasmid construction and treatment with metabolic inhibitors¹⁹  are detailed in supplemental Methods. FVIII immunohistochemistry, activity and antigen measurement, pulse-chase and immunoprecipitation (IP) analyses,¹⁸  co-IPs and western blotting, sucrose gradient sedimentation,¹⁹  and membrane filtration²⁰  were standard as detailed in supplemental Methods.

CHO-K1 or suberoylanilidehydroxamic acid (SAHA)-treated H9 cells were processed for immunogold localization of FVIII as in supplemental Methods.

Lysates from CHO-K1 or sodium butyrate (NaB)-treated H9 cells were subjected to sucrose gradient sedimentation for FVIII IP, elution, and analysis by transmission electron microscopy (TEM) and transmission electron cryomicroscopy (cryo-EM) as in supplemental Methods.

All statistical analysis used Prism software 7. P values were calculated using 1-way analysis of variance (ANOVA). P < .05 was considered significant. Statistical significance in figures and legends is denoted by asterisks (***P < .001; ****P < .0001).

We studied FVIII secretion in 2 stable CHO cell clones engineered to express human FVIII. 10A1 cells constitutively express high levels of FVIII without UPR activation.¹⁸  In contrast, H9 cells express a low level of FVIII that is inducible by histone deacetylase inhibitors (NaB or SAHA) that is coupled with UPR activation. In H9 cells, NaB treatment increased FVIII expression over time and activated UPR and apoptosis (Figure 1A).^8,19  Previous studies using sucrose gradient sedimentation demonstrated newly synthesized FVIII transiently aggregates in CHO cells.¹⁹  To measure FVIII aggregation in a more reproducible, robust, and convenient manner, we used filtration through cellulose acetate (CA) membranes, which selectively retains amyloid aggregates.²⁰  CA retention was normalized to the total amount of FVIII determined by western blotting or by retention on nitrocellulose (NC) membranes, which bind all cell proteins. FVIII aggregation occurred upon increased FVIII synthesis (supplemental Figure 1A-B) and inversely correlated with the level of FVIII secretion (supplemental Figure 1C). Immunofluorescence microscopy demonstrated similar diffuse ER localization of FVIII in H9 and 10A1 cells characterized by colocalization with KDEL-containing ER proteins (Figure 1Bi; supplemental Figure 2i). NaB induction of FVIII synthesis in H9 cells increased FVIII staining that also colocalized with the KDEL ER marker (Figure 1Bii), suggesting that ER retention occurs upon increased synthesis. Energy depletion, with inhibitors of mitochondrial oxidative phosphorylation and glycolysis, causes FVIII aggregation and selectively inhibits FVIII trafficking to the Golgi.^12,14  Treatment of 10A1 cells or H9 cells with 2′-deoxyglucose (2DG), an inhibitor of glycolysis, and sodium azide (NaN₃), an inhibitor of oxidative phosphorylation, to deplete cellular energy caused FVIII and KDEL-containing ER proteins to colocalize to large perinuclear structures (Figure 1Biii; supplemental Figure 2i vs ii). Thioflavin-S (Thio-S) selectively binds β-rich structures such as amyloid and increases fluorescence intensity.^21,22  Intriguingly, 2DG and NaN₃ caused colocalization of Thio-S staining with FVIII, suggesting that FVIII in the ER has amyloid-like properties, especially upon metabolic collapse. Surprisingly, upon replacing the metabolic inhibitors with Glc-containing media for 4 hours, FVIII and KDEL proteins resumed their diffuse web-like ER colocalization, and Thio-S costaining was significantly reduced in both H9 and 10A1 cell lines (Figure 1Biv; supplemental Figure 2iii), suggesting at the morphological level that FVIII aggregates disappear and may traffic the secretory pathway.

FVIII aggregation was also analyzed by immunogold TEM. There was insignificant FVIII reactivity in H9 cells not treated with primary antibody (Figure 1Ci), CHO-K1 cells that do not express FVIII (Figure 1Civ), and modest reactivity in H9 cells that were not induced to express FVIII with SAHA (Figure 1Cii,v). Upon FVIII induction by SAHA in H9 cells, FVIII was detected in clusters detected by 2 independent FVIII antibodies coupled to 12-nm gold particles (Figure 1Ciii,vi). In addition, immunolocalization of PDIA6 by secondary antibody conjugated to 18-nm gold particles demonstrated colocalization and clustering of FVIII and an ER-resident protein (Figure 1Ciii,vi [and magnified to the right of the subpanels]).

The structure of purified FVIII aggregates in NaB-treated H9 cells was analyzed by conventional TEM and transmission cryo-EM. After FVIII induction, H9 cell lysates were sedimented on sucrose gradients.¹⁹  FVIII was immunoprecipitated from each fraction, eluted from the beads, and analyzed by TEM and cryo-EM. Fibrils of ∼5-nm diameter were visible in negatively stained TEM images from the heavy-molecular-weight (HMW) sucrose gradient fractions (Figure 1D white arrows) from NaB-treated H9 cells, presumably representing aggregated FVIII. These fibrils were absent in FVIII similarly isolated from the light fractions, which would contain monomeric FVIII (not shown). Cryo-EM images of FVIII immunoprecipitated from HMW fractions displayed dense networks of fibrils with diameters of 5.1 ± 0.5 nm. The fibrils appeared to interact with each other in various ways and form amyloid-like aggregates reminiscent of Poly-Q aggregates observed by cryo-EM in mammalian cells²³  (Figure 1E). Antibody gold labeling for FVIII further confirmed that these fibrils are composed of FVIII. Importantly, no fibrils were detected in fractions similarly prepared from H9 cells that were not induced for FVIII expression (supplemental Figure 3). Therefore, FVIII forms fibrils that are morphologically similar to those reported for other amyloidogenic proteins.^24-26 

To examine the fate of FVIII aggregates in cells upon bioenergetic collapse and recovery in Glc, we analyzed aggregation by radiolabel pulse-chase experiments and filtration on CA and NC membranes. Recombinant FVIII (rFVIII) was filtered as a control for properly folded FVIII and did not bind CA membranes (supplemental Figure 4A). 2DG and NaN₃ treatment decreased cellular adenosine triphosphate (ATP) in H9 cells by approximately fivefold (supplemental Figure 4B), and increased FVIII retention on CA membranes (supplemental Figure 4A) in H9 and 10A1 cells, which subsequently declined upon Glc recovery for 4 hours (supplemental Figure 4C). Treatment with autophagy inhibitor 3-methyladenine or proteasome inhibitor Velcade did not stabilize aggregates upon Glc repletion (supplemental Figure 4C), suggesting that these processes do not significantly contribute to eliminating the aggregates. 10A1 cells were pulse-labeled for 20 minutes, chased in unlabeled medium for 20 minutes, and then treated with or without 2DG and NaN₃ for 2 hours. After 2 hours, the cells were recovered in Glc-containing medium for up to 4 hours. Cell lysates and conditioned media were analyzed by FVIII IP, sodium dodecyl sulfate–polyacrylamide gel electrophoresis (SDS-PAGE), and autoradiography. The same lysates were also filtered through CA and NC membranes, which demonstrated aggregation after 2DG and NaN₃ treatment and subsequent resolution upon Glc recovery from 1 to 4 hours (Figure 2Ai). The pulse-chase demonstrated that untreated 10A1 cells efficiently secreted FVIII at 4 hours after the pulse label (Figure 2Aii lanes 1-3). In contrast, 2DG and NaN₃ treatment prevented secretion of the pulse-labeled FVIII (Figure 2Aii lane 4). Upon Glc repletion, labeled FVIII appeared in the medium from 1 to 4 hours (Figure 2Aii lanes 5-8). The total amount of labeled FVIII secreted after 4 hours closely approximated that secreted from cells not treated with 2DG and NaN₃ (100% vs 87%). Notably, recovery in Glc not only increased FVIII secretion (Figure 2Aii lane 8), but also functional FVIII activity in the medium (Figure 2Aiii). Addition of cycloheximide (CHX) to inhibit new protein synthesis during the chase recovery reduced the FVIII activity in the medium by ∼50% (Figure 2Aii lane 9), indicating that approximately one-half of the FVIII activity was previously synthesized and was likely derived from FVIII aggregates. These findings support the conclusion that FVIII intracellular aggregates were dissolved and refolded for secretion of active FVIII.

We then tested whether glycolysis and/or oxidative phosphorylation are required to prevent FVIII aggregation using specific inhibitors. Inhibition of Glc metabolism by 2DG immediately induced FVIII aggregation, where inhibition of oxidative phosphorylation by either oligomycin A (Oligo), a complex 5 inhibitor, or a combination of Oligo with rotenone (Rot), a complex 1 inhibitor, had no effect (Figure 2B; supplemental Figure 4D), indicating that Glc metabolism is required to solubilize FVIII. The role of Glc metabolism in reducing FVIII aggregates is under investigation. In addition, treatment with reducing agents dithiothreitol (DTT) or β-mercaptoethanol (β-ME) did not disrupt the aggregates, suggesting that they do not result from inappropriate disulfide bond crosslinking (supplemental Figure 4E). However, the FVIII aggregates in cell lysates were solubilized with 1% SDS prior to CA membrane filtration (Figure 2C), although long exposure demonstrated that ∼3% of the FVIII aggregates were resistant to 1% SDS, suggesting that FVIII aggregation is due to hydrophobic interactions.

We previously demonstrated that FV, which shares amino acid homology with FVIII, is very efficiently secreted from transfected mammalian cells compared with FVIII and does not significantly aggregate.²⁷  To identify the cause for inefficient FVIII secretion and aggregation, we studied FVIII/FV domain swap chimeras demonstrating that a 110-aa region (aa226-336) in the FVIII A1 domain is responsible for the inefficient secretion.²⁸  Furthermore, we identified hydrophobic motif aa291-310 in FVIII, which was absent in FV, for which mutations both enhanced (F309S mutation) or reduced (7LF>A mutation) secretion.^13,28  Importantly, the single mutation F309S, S being the homologous residue in FV (Figure 3A), completely conferred the efficient secretion of FV onto FVIII,¹³  suggesting that aa291-310 predispose to FVIII aggregation. Hence, we analyzed mutants that increase (F309S and L303E/F309S [ES]), have a modest effect (L300V/L303E/F309S [VES] and L294T/L303E/F309S [TES]), or decrease (7LF>A [7 mutations of L and F to A, underlined in Figure 3A] and F306W) FVIII secretion by CA membrane retention. Mutations that increased FVIII secretion (F309S and ES) reduced retention and conversely, those that reduced secretion (7LF>A and F306W) increased retention when expressed in COS-1 (Figure 3B-C), CHO-K1 (not shown), or HepG2 (supplemental Figure 5) cells, indicating that this region is necessary for FVIII aggregation and is independent of the host cell type.

To test whether residues in the FVIII A1 domain are sufficient for aggregation, we introduced 79 residues (aa253-331) (supplemental Figure 6A) in frame into the enhanced green fluorescent protein (eGFP) fused to the human proinsulin C-peptide (Cpep) containing a cMyc tag (herein, CMy) (Figure 3D). Antibody to Cpep conveniently recognizes all of these chimeras. These 79 aa were chosen because they comprise the copper ion ligands at H267, C310, and H315 that might be important for maintaining FVIII structural integrity (supplemental Figure 6A). These chimeras have no glycosylation sites or disulfide bonds, therefore providing the ability to disentangle FVIII aggregation from posttranslational modifications. Fluorescence microscopy demonstrated significant green fluorescence in wild-type FVIII (wtFVIII)-eGFP-CMy expressing cells indicating that insertion of these 79 aa did not disrupt eGFP synthesis and folding (Figure 3E). Western blotting of cell lysates and media demonstrated that, in contrast to eGFP-CMy, insertion of the FVIII 79 aa into eGFP-CMy severely inhibited secretion that was further reduced by the multiple 7LF>A mutations (Figure 3F). Interestingly, secretion of the F309S and ES mutations in FVIII-eGFP-CMy was increased compared with wtFVIII-eGFP-CMy in 293T (Figure 3F) and HepG2 (supplemental Figure 7A-B) cells, indicating that these 79 aa are necessary and sufficient to seed aggregation. Notably, expression of FVIII-eGFP-CMy wt and all mutants caused ER stress–monitored BiP induction in transfected cells (Figure 3F). Thus, we propose that these 79 aa, termed Aggron, seed FVIII amyloid formation.

Application of the Tango algorithm, which predicts cross β-sheet propensity,²⁹  to the 79 aa identified 2 regions that can form β-sheets (supplemental Figure 8). Although C310S substitution did not alter the Tango plot, F309S or C310E substitutions or the respective 79 aa from FV eliminated the second β-sheet at F309. In contrast, F306W substitution, which further reduces secretion, enhanced the peak at F309. The F309S and C310E mutations in FVIII-eGFP-CMy were expressed in 293T (Figure 3G-H) and HepG2 (supplemental Figure 9) cells for analysis by CA membrane filtration and sucrose gradient sedimentation. In contrast to eGFP-CMy without the FVIII sequence, which migrated in light fractions 2-4, the majority of wtFVIII-eGFP-CMy migrated in heavy fractions 5-10 (Figure 3H-I). Compared to wtFVIII-eGFP-CMy, ∼10-fold more F309SFVIII-eGFP-CMy and C310EFVIII-eGFP-CMy migrated in fraction 3 and ∼20% less migrated in heavy fractions 9-10, indicating improved solubility of these mutants. These results support the validity of the Tango analysis and show that both F309 and C310 significantly contribute to FVIII aggregation. The C310 mutants C310S and C310E were both retained on CA membranes (supplemental Figure 10), suggesting that FVIII-eGFP-CMy aggregation is not mediated by an intermolecular disulfide linkage.

The aggregation properties of the 79 aa were further analyzed by removing eGFP to study wt and F309S FVIII-CMy in context of a smaller peptide (Figure 4A). For controls, CMy was expressed alone or with insertions of eGFP (eGFP-CMy) or the homologous residues from FV (FV-CMy). These polypeptides were detected in transfected 293T cells by western blotting (supplemental Figure 6B). Cpep enzyme-linked immunosorbent assay (ELISA) revealed that wtFVIII-CMy was poorly secreted, where FV-CMy and F309S FVIII-CMy secretion was approximately fivefold to ∼10-fold greater (Figure 4B). However, secretion of the unrelated controls CMy or GFP-CMy was significantly further increased over FV-CMy and F309S FVIII-CMy. Inversely correlating with secretion, wtFVIII-CMy was retained upon CA membrane filtration, whereas F309S FVIII-CMy retention was reduced to 60% (Figure 4C). CA membrane retention of eGFP-CMy and FV-CMy was not detectable (Figure 4C). Sucrose gradient sedimentation also demonstrated reduced aggregation for F309S FVIII-CMy (Figure 4D-E). We next tested whether the wtFVIII-CMy aggregates have properties similar to full-length wtFVIII aggregates. Analysis of sucrose gradient fractions 3 and 10 from 293T cells that express wtFVIII-CMy demonstrated that wtFVIII-CMy was sensitive to 1% SDS but resistant to DTT and β-ME (Figure 4F). Together, these findings confirm that the 79 aa seed aggregation of structures that appear biochemically similar to wtFVIII.

Because 2DG treatment of H9 cells caused wtFVIII aggregation, we tested whether 2DG causes the short wtFVIII-CMy chimera to aggregate. Indeed, 2DG and NaN₃ treatment increased wtFVIII-CMy aggregation monitored by CA membrane filtration, with only a slight reduction in expression (Figure 4G; supplemental Figure 6B). These metabolic inhibitors did not significantly cause aggregation of CMy, FV-CMy, or eGFP-CMy. In addition, recovery of 2DG-treated cells in Glc-containing media decreased wtFVIII-CMy aggregation ∼50% (Figure 4G), consistent with observations of wtFVIII in H9 and 10A1 cells. This suggests that FVIII aggregation was not simply a consequence of altered glycosylation in response to 2DG because wtFVIII-CMy contains no N-linked glycans.

Finally, the potential of wtFVIII-CMy to form intermolecular heterodimers/oligomers with FVIII was investigated. wtFVIII-CMy and full-length FVIII or B-domain–deleted (BDD) FVIII, which aggregates similarly to FVIII (not shown), were coexpressed in 293T cells (supplemental Figure 11) and their interaction was analyzed by coimmunoprecipitation (co-IP). Surprisingly, Cpep IP of the short chimera wtFVIII-CMy efficiently pulled-down both wtFVIII and BDD-FVIII (Figure 4H), suggesting that these 79 aa interact in trans to alter the wtFVIII folding pathway.

Mass spectrometry of FVIII immunoprecipitated from control CHO cells and H9 cells treated with NaB identified many ER chaperones and enzymes that selectively interact with wtFVIII (supplemental Figure 12A-B). FVIII interactions with BiP, CANX, UGGT1, SEL1L, and PDIA6 were validated by FVIII IP and western blotting (supplemental Figure 12C). PDIA6 also colocalized with FVIII by immunogold TEM (Figure 1C).

The 2 most significant FVIII interactors were BiP/GRP78/HSPA5 and its cochaperones (ERdj5, ERdj3, and GRP170) and CANX/CRT with its associated factors UGGT1 and ERP57/PDIA3, raising the question of whether these chaperones influence FVIII aggregation. Complete inhibition of the CANX/CRT cycle with castanospermine (CST), an inhibitor of α-glucosidase (GS1 and GS2) activity, increased FVIII aggregation without altering FVIII expression (supplemental Figure 13A-C), suggesting that CANX/CRT prevents wtFVIII aggregation, possibly by binding soluble monoglucosylated N-linked glycans on FVIII. However, CST treatment of cells expressing the wtFVIII-CMy short chimera did not significantly increase wtFVIII-CMy aggregation (supplemental Figure 13D-E), confirming that aggregation of the 79 aa is not due to altered glycosylation because this chimera has no N-glycans.

To study the role of BiP in FVIII aggregation, we analyzed whether BiP binds to low-molecular-weight (LMW) and/or HMW FVIII species. Cell lysates from NaB-treated H9 cells were analyzed before (Figure 5Ai) and after sucrose gradient sedimentation (Figure 5Aii) by FVIII IP and western blotting. Although similar amounts of FVIII migrated in LMW fraction 5 (L) from both untreated and NaB-treated H9 cells (Figure 5Aii lanes 1, 3), there was a vast amount of FVIII in the HMW fraction 10 (H) (Figure 5Aii lane 2), with even more upon NaB induction (Figure 5Aii lane 4). FVIII IP and western blotting demonstrated detectable BiP interaction only with the LMW fraction 5 of FVIII from NaB-treated, and not in untreated, H9 cells (Figure 5Aii lanes1, 3). BiP was not detected in the HMW fraction (Figure 5ii lanes 2, 4), nor did we detect BiP in the postimmunoprecipitation (post-IP) HMW sucrose fraction supernatant (Figure 5Aii lanes 6, 8). Significantly, the majority of BiP was detected in the FVIII post-IP supernatant in the LMW fraction (Figure 5Aii lanes 5, 7). Collectively, these data suggest that BiP preferentially associates with LMW FVIII.

To further analyze how BiP impacts FVIII aggregation, we used the Shiga-toxic Escherichia coli virulence factor SubA₂₇₂B_mt (SubAB), which specifically and acutely cleaves at a single site in the hydrophobic linker that connects the BiP amino-terminal ATPase domain with the substrate-binding domain (Figure 5B).^30,31  As control, cells were treated with a protease-mutant SubAB. Treatment of cells with 1 or 2 μg/mL of the active protease acutely cleaved ∼35% and ∼85% of BiP, respectively, and increased FVIII aggregation detected by retention on CA membranes (Figure 5C), suggesting that intact BiP, but not cleaved BiP, prevents FVIII aggregation. BiP cleavage was associated with UPR activation by induction of phosphorylated eukaryotic initiation factor 2 α (P-eIF2α) and CHOP (Figure 5C), consistent with previous findings.³² 

Although these results indicate that BiP and CANX/CRT prevent FVIII aggregation, they do not address their role in disaggregation. Hence, we treated H9 cells with NaB for 18 hours with the last 2 hours in the presence of 2DG and NaN₃ to induce maximal FVIII aggregation. Then, fresh media with Glc were added for 4 hours in the presence or absence of SubAB protease or CST. Although 2DG and NaN₃ reduced FVIII expression, BiP cleavage prevented the Glc-dependent dissociation of FVIII aggregates (Figure 5D). In contrast, inactivation of the CANX/CRT cycle by CST did not significantly prevent disaggregation. Collectively, we conclude that BiP, and not CANX/CRT, is required to disaggregate polymerized FVIII.

To evaluate the role of BiP in preventing FVIII aggregation, BDD-FVIII, for which aggregation is induced by BiP inactivation similarly to wtFVIII (supplemental Figure 11), was coexpressed with Flag-tagged BiP (BiP-F), mutant V461FBiP-Flag (BiPmt-F), which is peptide-binding defective,³³  or Myc-tagged BiP (BiP-M).³³  Forced expression of BiP-F or BiP-M did not alter BDD-FVIII expression, although they significantly reduced BDD-FVIII aggregation measured by retention on CA membranes (supplemental Figure 14). In addition, the V461F mutant BiP was partially defective in preventing BDD-FVIII aggregation. Thus, we conclude that BiP requires its peptide-binding site to efficiently promote FVIII solubility in the ER.

We next tested whether BiP cleavage by SubAB impacts aggregation of the wtFVIII-CMy chimera. 293T cells expressing wtFVIII-CMy were treated with mt or wtSubAB. After 2.5 hours, ∼40% to 50% of BiP was cleaved without affecting wtFVIII-CMy expression (Figure 5E top). Compared with mtSubAB treatment control, wtFVIII-CMy aggregation increased ∼60% upon BiP cleavage (Figure 5E bottom). Although SubAB inhibits secretion of immunoglobulin,³⁴  secretion of eGFP-CMy was not reduced (supplemental Figure 15A-B), indicating that the secretory pathway remained intact. Collectively, these findings show that intact BiP prevents aggregation of wtFVIII-CMy through interaction with the BiP peptide-binding site.

We finally tested whether BiP binds to the 79 aa by coexpression of wtFVIII-CMy with BiP-F or the peptide-binding defective BiPmt-F.³³  Indeed, Flag IP of BiP coimmunoprecipitated wtFVIII-CMy (Figure 5F), suggesting a direct interaction between BiP and the 79 aa of FVIII. In contrast, Flag IP of mutant BiPmt-F did not co-IP. However, treatment with 2DG and NaN₃ did not significantly increase the BiP-wtFVIII-CMy interaction, consistent with the notion that SubAB toxin cleavage of BiP induces wtFVIII-CMy aggregation, as we could not detect BiP binding to HMW FVIII aggregates (Figure 5A). However, mutations that reduce FVIII-CMy aggregation did not detectably alter BiP interaction measured by co-IP (not shown). It is likely that BiP binds multiple sequences so mutations that reduce aggregation do not alter BiP interaction, as previously suggested.^35,36  These findings support the conclusion that BiP binds 79 aa in wtFVIII-CMy to prevent aggregation.

FVIII is poorly secreted due to retention in the ER. Here, we provide significant insight into the mechanism for FVIII retention in the ER. We demonstrated that FVIII forms amyloid-like fibrils in the ER when expression is constitutively high and, to a greater extent, upon increased synthesis. In addition, inhibition of Glc metabolism, but not oxidative phosphorylation, causes FVIII aggregation and reduces secretion. The following characterize the demonstrated amyloid fibril properties of these aggregates: (1) sucrose gradient fractionation of cell lysates, FVIII pulse-chase labeling and IP, negative-stain TEM, and cryo-EM, which identified ∼5-nm fibrils and fibril networks only in the very HMW gradient fractions; (2) FVIII colocalization with Thio-S in the ER; (3) FVIII retention on CA membranes; (4) immunogold TEM showing FVIII clustering with the ER-localized PDIA6; and (5) sensitivity to 1% SDS, although a small fraction (∼3%) was resistant, possibly analogous to prefibrillar conversion to fibrillar amyloid structures that undergo transition from SDS-sensitive to SDS-resistant forms.^37,38  Other than immunoglobulin LC and HCs,³⁹  this represents a unique example of amyloid-like aggregates formed in the ER. Remarkably, unlike other amyloids, FVIII aggregates were reversible upon Glc supplementation and were recovered as functional FVIII secreted into the medium. To our knowledge, this is the first example of reversible amyloidogenesis leading to proper folding of a protein in metazoans.

Mass spectrometry demonstrated FVIII interaction with BiP and CANX/CRT. Here, we show that CANX/CRT interaction reduces wtFVIII aggregation, similar to other glycoproteins.^40,41  BiP preferentially binds to LMW FVIII and not HMW FVIII aggregates. BiP inactivation increased FVIII aggregation, and forced BiP expression reduced FVIII aggregation. Thus, BiP is required to prevent FVIII aggregation and/or to mediate disaggregation. It was shown that increased BiP expression inhibits wtFVIII secretion.¹⁴  However, contrary to what we expected, increased BiP expression prevented wtFVIII aggregation and retained soluble LMW FVIII in the ER. From these findings, we propose a model for BiP function in preventing FVIII amyloid formation (Figure 6A).

To further characterize FVIII aggregation, we identified 79 aa in the A1 domain that seed β-aggregation. Expression of a chimeric polypeptide (wtFVIII-eGFP-CMy or wtFVIII-CMy) recapitulated the major aspects of wtFVIII aggregation including: (1) wtFVIII-CMy aggregated in response to glycolysis inhibition and (2) wtFVIII-CMy aggregated in response to BiP cleavage and inactivation by SubAB cytotoxin. In addition, we show that the 79-aa motif interacts in trans with wtFVIII and BDD-FVIII as well as with the peptide-binding site on BiP. We propose that BiP binding to the FVIII aggregate-seeding region, Aggron, prevents aggregation, consistent with previous reports of other proteins.^35,42,43  As BiP prevents FVIII aggregation, it is interesting to note that 3 BiP cochaperones, ERdj3, ERdj5, and GRP170, the latter 2 of which interact with aggregation-prone sequences,³⁵  were also identified as significant interactors with wtFVIII. Although previous studies linked cochaperone interactions with ER-associated degradation, FVIII is not significantly degraded by ER-associated degradation, but rather is secreted. Present studies are testing the significance of these cochaperone interactions. Importantly, without the complication from glycosylation sites and disulfide bonds, wtFVIII-CMy did not aggregate upon inhibition of the CANX/CRT cycle, implicating the glycans as important for wtFVIII solubility. Finally, these observations on FVIII aggregation were cell-type independent as they were observed in CHO, 293T, COS-1, and HepG2 cells.

At least 3 HA gene therapy clinical trials are in progress delivering BDD-FVIII in adeno-associated viral vectors (AAVs) to hepatocytes (https://clinicaltrials.gov/ct2/results?term =hemophilia+a+gene+therapy). One report demonstrated significant success in 9 men with severe HA at 1 year after virus delivery to obtain 100% FVIII levels in the blood.⁴⁴  However, a more recent update indicated the FVIII levels in the high-dose cohort (6e13 vector genomes per kilogram) decreased ∼50% after year 2,⁴⁵  which is unusual compared with the stable expression observed in other hepatocyte-directed AAV gene-therapy trials.^46,47  All of the properties of wtFVIII misfolding and aggregation also exist for BDD-FVIII that is presently used in HA gene-therapy clinical studies.⁹  It should be considered that the decline in FVIII expression may reflect toxicity of aggregated BDD-FVIII in hepatocytes. In addition, endogenous FVIII is not likely expressed in hepatocytes, but rather in Kupffer cells and sinusoidal endothelial cells,^48-50  which may harbor machinery to efficiently fold FVIII. We also observed aggregation of BDD-FVIII in hepatocytes of mice that received BDD-FVIII DNA expression vectors (data not shown). In addition, ER stress was induced in livers upon DNA delivery of BDD-FVIII to mice in vivo,⁹  consistent with ER stress detected upon AAV delivery of BDD-FVIII.^51,52  Recently, it was shown that ER protein misfolding in hepatocytes with subsequent inflammation caused by a high-fat diet in mice can initiate nonalcoholic fatty liver disease with progression to nonalcoholic steatohepatitis and hepatocellular carcinoma.⁵³  Thus, it will be important to monitor liver function in the HA gene-therapy studies.

Importantly, why does FVIII harbor an Aggron in the A1 domain? Intriguingly, the Aggron includes Cys310, which ligands Cu⁺ in functional wtFVIII.¹⁵  Of all amino acid variants that reduce FVIII aggregation and improve secretion, F309S was the most effective. Previously, it was demonstrated that F309S FVIII exhibits clotting factor–specific activity, thermal denaturation, and thrombin activation indistinguishable from wtFVIII.¹³  However, F309S FVIII was 10-fold more sensitive to EDTA inactivation, which presumably extracts Cu⁺. The F309S mutation substitutes a large, hydrophobic side chain with a smaller, more polar, one, and may increase flexibility to relax the C310 interaction with Cu⁺ (Figure 6B). Although the potential of neoantigenicity of F309S mutation has not been addressed, F309 is buried within the FVIII 3-dimensional structure^16,17  and may not be accessible to the immune response. Because F309S FVIII and F309S BDD-FVIII function identically to wtFVIII and BDD-FVIII, respectively, display lesser aggregation and lesser UPR activation, and are secreted more efficiently, they may be considered an alternative choice to wtFVIII or BDD-FVIII for HA gene therapy. Finally, along these lines, because increased plasma FVIII levels associate with venous thrombosis,^54,55  our findings should encourage evaluating whether F309S FVIII mutation associates with thrombosis in humans.

The authors thank David Ron for kindly providing BiP expression vectors, and Peter Arvan and Ming Liu for providing the pTargetT-hProCpepMyc and pTarget-hProCpepeGFP vector. Baxter Corp (Deerfield, IL) kindly provided FVIII antibody-conjugated sepharose beads. The authors gratefully thank Michael Cunningham for performing the TEM of FVIII fibrils (Figure 1D), Lisa Giles for analyzing immunofluorescence and establishing the CA filter-binding assay (Figure 1B and supplemental Figure 4), and Jyoti Malhotra for performing pulse-chase radiolabeling and IP (Figure 2Aii). Immunogold labeling and TEM was performed at the University of California San Diego (UCSD) Cellular and Molecular Medicine Electron Microscopy Facility. The authors thank the University of Michigan Transmission Electron Microscopy Laboratory for analysis of negative-stained purified FVIII fibrils. The authors thank Peter Arvan for critical review of this manuscript.

This work was supported by the following National Institutes of Health grants: R01HL052173 (National Heart, Lung, and Blood Institute; R.J.K.); R37DK042394, R01DK113171 (R.J.K.), R01DK103185 (R.J.K.), and R24DK110973 (R.J.K.) (all 4 from the National Institute of Diabetes and Digestive and Kidney Diseases); R01CA198103 (National Cancer Institute; R.J.K.); R01AG062190 (National Institute on Aging; R.J.K.); and P30 CA030199 (National Cancer Institute SBP Cancer Center). The cryo-EM studies were supported by National Institutes of Health grants R01GM119948 (National Institute of General Medical Sciences; D.H., N.V.) and S10-OD012372 (Office of the Director; D.H.).

R.J.K. is a member of the UCSD Diabetes Research Center (P30 DK063491 [National Institute of Diabetes and Digestive and Kidney Diseases]) and is an adjunct professor in the Department of Pharmacology, UCSD.

Contribution: J.P. performed all studies on FVIII-CMy and FVIII-eGFP-CMy, designed the experiment, and wrote the manuscript; Z.C. provided expert consultation and assisted in manuscript preparation; J.Y. performed sucrose gradient analyses and BiP interaction and prepared the manuscript; A.R.C. performed mass spectrometry analyses and bioinformatics; A.P. performed mass spectrometry analyses and experiments to validate the significance of FVIII-interacting proteins; J.Y., M. Swift, D.H., and N.V. performed the cryo-EM sample preparation and analysis; D.H. and N.V. provided financial support for the cryo-EM studies; J.L.-A. provided technical support; M. Siirin performed studies involving CA and NC membrane filtration and analyses of FVIII secretion and the intracellular ATP level; A.W.P. and J.C.P. provided essential reagents; and R.J.K. directed the studies, assured validity of findings, wrote the manuscript, and provided financial support for the studies.

Conflict-of-interest disclosure: The authors declare no competing financial interests.

Correspondence: Randal J. Kaufman, Degenerative Diseases Program, SBP Medical Discovery Institute, 10901 N. Torrey Pines Rd, La Jolla, CA 92037; e-mail: rkaufman@sbpdiscovery.org.