Immunosilencing a Highly Immunogenic Protein Trimerization Domain*

Background: Trimerization domains are commonly used to stabilize trimeric protein vaccines and therapeutics. Results: The GCN4-based isoleucine zipper domain induces strong antibody responses in vivo but this can be overcome by introducing glycans. Conclusion: Appropriately positioned glycans can effectively immunosilence the GCN4-based trimerization domain. Significance: Immunosilencing trimerization domains could be important for the exploitation of trimerization domains in protein vaccines and therapeutics. Many therapeutic proteins and protein subunit vaccines contain heterologous trimerization domains, such as the widely used GCN4-based isoleucine zipper (IZ) and the T4 bacteriophage fibritin foldon (Fd) trimerization domains. We found that these domains induced potent anti-IZ or anti-Fd antibody responses in animals when fused to an HIV-1 envelope glycoprotein (Env) immunogen. To dampen IZ-induced responses, we constructed an IZ domain containing four N-linked glycans (IZN4) to shield the underlying protein surface. When fused to two different vaccine antigens, HIV-1 Env and influenza hemagglutinin (HA), IZN4 strongly reduced the antibody responses against the IZ, but did not affect the antibody titers against Env or HA. Silencing of immunogenic multimerization domains with glycans might be relevant for therapeutic proteins and protein vaccines.

Several small protein domains can facilitate trimerization of proteins. The two most widely used trimerization domains in biochemical and biomedical research are the isoleucine zipper (IZ) 3 based on the GCN4 transcriptional activator from Saccharomyces cerevisae, and the foldon domain of the bacteriophage T4 fibritin protein (Fd) (1,2). IZ consists of ␣-helices in a coiled-coil heptad repeat, in which the first (a) and fourth (d) amino acid residues in each heptad repeat determine the oligomerization state of the protein (3). Isoleucines at the a and d positions facilitate trimerization while alternating isoleucine/ leucine or leucine/isoleucine can confer dimerization or tetramerization, respectively, which can be exploited for engineering dimeric, trimeric, or tetrameric proteins (1,3,4). In contrast, Fd consists of three ␤-hairpins, which assemble into a ␤-propeller-like structure (5). The Fd trimer is stabilized by hydrogen-bonding, hydrophobic interactions, and salt-bridges between each (2,5).
Despite their extensive use in preclinical studies, the immunogenicity of the yeast-derived IZ and bacteriophage-derived Fd has not been properly evaluated. When IZ and Fd would be routinely used for future clinical applications and vaccines, immune responses against IZ, and Fd could affect the effectiveness of these therapeutics and vaccines. For example, with repeated administration of therapeutic proteins, IZ-or Fd-specific antibodies (Abs) might enhance systemic clearance of the protein thereby decreasing the efficiency of the therapy (29,30).
Here we report that the IZ and Fd trimerization domains induce potent Ab responses in vaccinated animals. To dampen the anti-IZ response, we designed an IZ variant with four potential N-linked glycosylation sites (PNGS) at strategic positions in the heptad repeat domain of IZ (IZN4). Protein trimerization was not affected by the addition of four glycans to the IZ domain. Moreover, IZN4 induced significantly lower IZspecific Ab responses in vaccinated animals when fused to two different antigens, without hampering the immune response against these antigens. This novel IZN4 trimerization domain should be useful for trimeric protein vaccines and therapeutics.

EXPERIMENTAL PROCEDURES
Plasmids and Mutagenesis-The expression vector for the HIV-1 subtype B JRFL Env SOSIP.R6 (Env) with a C-terminal IZ (RMKQIEDKIEEILSKIYHIENEIARIKKLIGER) and histidine tag has been described elsewhere extensively (13). In short, the Env construct used in this study (Env-IZ) contains a disulfide between gp120 and gp41; an improved furin cleavage site (R6); an I559P mutations to improve trimerization; a codon-optimized GCN4-pII isoleucine zipper (IZ); a GGGGTGGGGTGlinker between the IZ and Env moiety; a C-terminal histidine tag (His-tag) (1,13,31,32). The codon-optimized influenza hemagglutinin (HA) sequence (GenBank TM : ACU65077.1, residues 10 -509) was synthesized (Mr. Gene) and was cloned to replace the Env sequence in the same expression vector yielding [HA]GGGGTGGGGTGRMKQI EDKIEEILSKIYHIENE-IARIKKLIGERHHHHHHHH (HA-IZ). To obtain Env-Fd and HA-Fd, the codon-optimized T4 bacteriophage Fd sequence (GYIPEAPRDGQAYVRKDGEWVLLSTFL) was cloned in place of the IZ sequence. Env-IZ variants with the different PNGS were generated using the Quickchange mutagenesis kit (Agilent). For HA-IZN4 we replaced the IZ in HA-IZ with the IZN4 (NGTGRMKQIEDKIENITSKIY NITNEIARIKKLIGNRT) sequence from Env-IZN4. For the constructs used in rabbit vaccinations (i.e. Env-IZ, Env-IZN4, HA-IZ, and Env-IZN4), a stopcodon was introduced before the His-tag to avoid the induction of antibodies against the tag.
Reagents-mAbs were gifts or purchased from the following Cell Culture and Transient Expression-HEK293T cells were maintained in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal calf serum (FCS), penicillin (100 units/ml), and streptomycin (100 g/ml). Cells were transfected with PEImax (Polysciences) and plasmid DNA expressing IZ-and Fd-fusion proteins. For the experiments shown in Fig. 3A, plasmid DNA encoding the protease furin was co-transfected in a 1:1 ratio with Env-IZ or Env-IZN4 plasmid to induce cleavage of the gp120 and gp41 ecto -IZN4 subunits of the Env-IZN4 proteins. When indicated, kifunensine (5 g/ml, Cayman Chemical) was added during transfection to block mannose trimming of N-linked glycans (33). To remove oligomannose glycans from Env-IZ and Env-IZN4, 20 l supernatant of kifunensine-treated HEK293T cells was treated with 1 l of endoglycosidase H (Sigma Aldrich) according to the manufacturer's instructions.
Anti-Env and Anti-HA ELISA-Anti-gp120 Ab titers were measured by D7324-capture ELISA as described previously (34). Anti-HA Ab titers were measured using HA-Fd immobilized on Ni-NTA HisSorb 96-well plates. Reactivity of Env proteins with mAbs or receptor mimics was measured by anti-Env-IZ ELISA immobilized on Ni-NTA plates as described above, using HRPlabeled goat-anti-human as the secondary Ab.
Immunizations-For the experiments shown in Fig. 1A four outbred NMRI mice, four Wistar rats, and four New Zealand White rabbits were immunized in the abdominal dermis with endotoxin-free Env-IZ plasmid DNA using gene gun technology at week 0, 2, and 4. For mice and rats we used 20 g of plasmid DNA and 125 g of DNA was used for rabbits. For the experiment shown in Fig. 1B, sera from rabbits immunized with Env-Fd (Env from clade B isolate YU2 with C-terminal foldon domain, described extensively elsewhere) (36) were obtained from an ongoing vaccination experiment in New Zealand White rabbits that is done in collaboration with IAVI. These rabbits were immunized with 30 g of Env-Fd protein in ISCOMATRIX TM adjuvant at weeks 0, 4, and 20. In previous vaccination experiments using similar protocols, five to six animals per group provided enough statistical power to detect significant differences between two groups in binding antibody measurements (35,37). Therefore, we used six New Zealand White rabbits for the experiments in Fig. 5, A and B that were immunized in the abdominal dermis at week 0, 2, 4, 8 with 125 g endotoxin-free DNA encoding Env-IZ, Env-IZN4, HA-IZ, and Env-IZN4 using gene gun technology (35,37). All protocols dealing with animal manipulations were in accordance with guidelines published by FELASA (Federation of European Animal Science Association) and GV-SOLAS (German Society of Laboratory Animal Science) and were reviewed by the Harlan, MFD Diagnostics, or Davids Biotechnologie Animal Care Committees, as appropriate.

Two Widely Used Protein Trimerization Domains Are Highly
Immunogenic-To determine the immunogenicity of the IZ domain, we immunized mice, rats and rabbits with plasmid DNA encoding the HIV-1 Env from strain JRFL containing an IZ domain at the C terminus (Env-IZ) (32,35). In addition, the immunogenicity of the Fd domain was studied in rabbits that were immunized in an independent study with Env from the strain YU2, containing the Fd domain at the C terminus (Env-Fd). Sera from all immunized animals were tested for IZ-and Fd-specific Ab responses by testing the response against fusion proteins in which IZ and Fd were fused to influenza hemagglutinin from strain A/Vietnam/1194/2004 (HA). We observed high levels of Abs against IZ in sera from Env-IZ-vaccinated mice, rats and rabbits (Fig. 1A). Similarly, we observed strong anti-Fd response in sera from Env-Fd-vaccinated rabbits (Fig.  1B), but not vice versa. Thus, both IZ and Fd are highly immunogenic. To analyze the relative amounts of IZ-specific versus Env-specific Abs, we performed a competition ELISA in which we depleted either Env-or IZ-specific Abs. In the four tested rabbits, 36 to 61% of the relative binding Ab response against the Env-IZ immunogen was specific for the IZ-domain (Fig.  1C). We conclude that the relatively small IZ trimerization domain (12 kDa) was highly immunogenic in rabbits in comparison with Env (ϳ400 kDa).
Design of an Immunosilenced IZ Trimerization Domain: IZN4 -Silencing of unwanted immunodominant Ab epitopes can be achieved by using N-linked glycans (N-glycans) (28,38,39). Therefore we investigated whether it would be possible to shield the immunogenic IZ protein surface with N-glycans. The IZ domain is a heptad repeat in which isoleucines on the first (a) and fourth (d) position in the ␣-helix direct trimerization ( Fig.  2A). Introduction of PNGS by means of the amino acid motif aspargine-X-serine/threonine (NX(S/T)) is only possible on the cde positions in the heptad repeat, because other options would eliminate the isoleucines responsible for oligomerization, or would result in a steric clash of the N-glycan with a neighboring ␣-helix. We chose to use NXT motifs over NXS, because NXT motifs have a higher probability of glycosylation (40). Based on these considerations we selected four positions for insertion of NXT motifs that were ultimately combined in a variant that had all four motifs (Fig. 2, B and D). Note that the first PNGS is in the linker region between Env and IZ. In silico prediction suggested that introducing four NXT motifs in IZ (IZN4) had a negligible effect on the tertiary structure of the IZ (Fig. 2B). Modeling of IZ glycans on the IZN4 trimer showed that the accessibility of underlying protein surface was dramatically reduced (Fig. 2C).
IZN4 Is Glycosylated and Trimerizes Efficiently-We sequentially introduced the NXT motifs at positions 1, 2, 3 and 4 of IZ in the context of Env-IZ (Fig. 2D): positions 1 and 4 (IZN2); positions 1, 2, and 4 (IZN3a); positions 1, 3, and 4 (IZN3b); positions 1, 2, 3, and 4 (IZN4). To verify that the newly introduced PNGS were glycosylated, we tested the glycan occupancy on IZ and IZN4 by separating gp41 ecto -IZ and gp41 ecto -IZN4 from gp120 by furin cleavage and subsequent reducing SDS-PAGE (31). gp41 ecto -IZN4 migrated slower than gp41 ecto -IZ in   (Fig. 3A, lanes 1 and 2). To confirm that the slower migration of gp41 ecto -IZN4 was due to N-glycans, we produced Env-IZ and Env-IZN4 in the presence of furin and the mannose-analog kifunensine, resulting in exclusively uniform oligomannose Man 8/9 GlcNAc 2 glycans (33). The supernatant was then treated with Endoglycosidase H (EndoH), which removes oligomannose N-glycans. Deglycosylated gp41 ecto -IZ and gp41 ecto -IZN4 showed the same fast migrating band (Fig. 3A,  lanes 3 and 4) indicating that the size difference in lane 1 and 2 was caused by glycans. The number of occupied PNGS in gp41 ecto -IZ or gp41 ecto -IZN4 was determined by reducing SDS-PAGE of the kifunensine produced furin-cleaved Env-IZ and Env-IZN4 proteins in which each glycoform of gp41 ecto -IZ and gp41 ecto -IZN4 was readily resolved (Fig. 3A, lanes 5 and 6). Gp41 ecto -IZ contains four PNGS in gp41 ecto , but only one to three PNGS were occupied by an N-glycan (Fig. 3A, lane 5), consistent with earlier findings on gp41 glycan occupancy (41). In contrast, gp41 ecto -IZN4 contains eight PNGS and we found occupancy of five to eight N-glycans (Fig. 3A, lane 6), suggesting that all four PNGS per ␣-helix in the IZN4 were occupied, showing that we successfully added a total of twelve N-glycans to the IZN4 trimer.

SDS-PAGE
The Env, Env-IZ, Env-IZN2/3a/3b/4 constructs were then expressed in a mammalian cell-line (HEK293T) and analyzed by native PAGE analysis. The addition of up to four N-glycans to IZ had no negative effect on the trimerization propensity of Env-IZ, but might slightly increase the trimer proportion in the supernatant. In contrast, removing the IZ trimerization domain resulted in formation of Env monomers, dimers, trimers, and aggregates ( Fig. 3B) (13). We conclude that the introduction of four PNGS did not disrupt IZ trimerization, but allowed the attachment of four N-glycans per IZ ␣-helix.
Antigenic Epitopes Are Shielded by Glycans in IZN4 -To examine whether the introduced N-glycans can shield immunogenic epitopes of the IZ domain, we tested sera of Env-IZ vaccinated rabbits for binding to Env-IZ or Env-IZ-glycan variants in ELISA. Introduction of N-glycans at the N and C terminus of the IZ domain (Env-IZN2) decreased binding and the binding was further reduced by the addition of N-glycans on position 2 (Env-IZN3a). Env-IZN3b and Env-IZN4 showed inefficient binding to the sera of Env-IZ vaccinated rabbits (Fig. 4A). The remaining binding can be attributed to Env-directed Abs.
To verify that glycosylation of the IZ domain did not affect the folding of the Env antigen, we compared the antigenic structure of Env-IZ and Env-IZN4 using antibodies that bind to four distinct antigenic sites on Env: the CD4 binding site (VRC01), the CD4-induced epitopes (48d), the membraneproximal external region (MPER), located adjacent to the IZ domain (2F5), and the glycan-dependent supersite centered around N332 (2G12 and PGT121) (Fig. 4B). Binding of these antibodies was similar for both Env-IZ and Env-IZN4, indicating that the antigenic structure and glycosylation of Env was unaltered by glycosylation of IZ.
IZN4 Is Immunosilenced but Preserves Antigen-specific Responses-To investigate whether IZ glycosylation resulted in decreased anti-IZ Ab responses, a vaccination experiment was conducted using Env and influenza HA as the model antigens. Rabbits were immunized with Env-IZ, Env-IZN4, HA-IZ, or HA-IZN4. After 10 weeks we determined the Ab binding titers to IZ and IZN4 for all four vaccine groups using the reciprocal proteins, i.e. the anti-IZ and anti-IZN4 responses for the Env-IZ and Env-IZN4 vaccinated rabbits were measured by using immobilized HA-IZ and HA-IZN4 and vice versa. Animals vaccinated with Env-IZ and HA-IZ raised 3-4-fold higher IZ-specific Ab titers compared with animals vaccinated with Env-IZN4 and HA-IZN4, independent of the fused antigen (Fig. 5, A  and B). Moreover, IZN4-vaccinated rabbits did not induce a higher IZN4-specific Ab response, showing that the addition of N-linked glycans to the IZ decreases its overall immunogenicity without shifting to an IZN4-specific response (Fig. 5, A and B). The decreased immunogenicity of IZN4 did not alter the Ab response against the Env and HA immunogens. The gp120 binding titers for Env-IZ and Env-IZN4 vaccinated animals were similar (Fig. 5C). HA-IZN4 vaccinated animals induced slightly higher levels of HA Ab compared with HA-IZ, but this was not statistically significant (Fig. 5D). The responses against gp120 were at least 20-fold lower than those against HA, consistent with observations that the Env is a relatively poor immunogen (42). Collectively, these results show that IZN4 was effectively immunosilenced, without compromising the antigen-specific response.  2) and presence (lanes 3-6) of kifunensine to block synthesis of complex glycans (33). The kifunensine-treated 293T cell supernatants were subsequently treated with EndoH to deglycosylate the protein (lanes 3 and 4). The ladder between lanes 5 and 6 marks all possible nine glycoforms of gp41 ecto -IZ and gp41 ecto -IZN4 that could be readily resolved in kifunensine produced proteins. B, BN-PAGE analysis of Env and different Env-IZ variants, followed by blotting with 2G12. Monomeric, dimeric, trimeric Env forms, and aggregates are indicated by arrows.

DISCUSSION
Heterologous trimerization domains are used in many (pre)clinical studies, but an extensive analysis of their immunogenicity has never been reported. We found that two of the most widely used trimerization domains, IZ and Fd, induced potent Ab responses. This is relevant for a number of therapeutic proteins as well as protein subunit vaccines that utilize these domains (4, 6 -27, 43).
An important question is whether an immune response against a trimerization domain affects efficacy of the therapeutic protein or vaccine, and/or might be harmful. Several key issues need to be considered. First, Abs against IZ or Fd might facilitate immune complex formation, which may result in an adjuvant effect and could actually be beneficial for vaccines (44). For therapeutic (self) proteins the formation of immune complexes could break tolerance to the endogenous protein and result in an autoimmune response. Immune complex formation might also alter clearance rates of the therapeutic proteins and vaccines thereby affecting their efficacy (30,45). Second, the introduction of immunodominant epitopes on a trimerization domain could potentially act as a decoy for more relevant epitopes on the vaccine antigen (46). Third, when trimerization domains would become more widely used in humans, pre-existing immunity against such domains might decrease the efficacy of therapeutic proteins and vaccines.
To negate any potential adverse effects of immunogenic trimerization domains, we designed a novel IZ-based trimerization domain where immunodominant epitopes are shielded by glycans (IZN4). It might also be possible to use this strategy to immunosilence the Fd domain, but the ␤-propeller-like structure of Fd might be less amendable to the addition of N-glycans. IZN4 was occupied by twelve N-linked glycans per trimer, trimerized efficiently and did not impact the antigenic structure of a model Env antigen. When IZN4 was fused to two different antigens (Env and HA) and used in immunization experiments, the levels of IZ-directed Abs were considerably reduced. We also considered that IZ glycosylation might impact the response against the attached antigen. Some studies have shown that silencing of an immunodominant epitope by adding N-glycans can redirect the immune response to other antigenic regions (28,38), but other studies failed to show such an effect (39,47). In our study, the decrease in Ab response against the immunodominant IZ domain did not significantly increase the Env-or HA-specific response.
Although the IZN4 response was 4-fold lower than IZ, IZN4directed Abs were still detected (Fig. 5, A and B). This indicates that IZN4 is weakly immunogenic, possibly because of the high number of charged hydrophilic amino acids at the C terminus of IZN4 (48). However, we cannot completely rule out that the . Glycosylation of IZ decreases binding of Env-IZ sera and preserves antigenicity of Env-IZ. A, sera (week 12) of four Env-IZ vaccinated rabbits, described previously (35), were tested for their ability to recognize Env-IZ, Env-IZN2, Env-IZN3a, Env-IZN3b, and Env-IZN4 in ELISA. The bars represent the midpoint binding titers relative to that of Env-IZ. Statistical analyses were performed using the Kruskal-Wallis test followed by Dunn's Multiple Comparison test. *: p Յ 0.05; **: p Յ 0.01. B, representative ELISA curves of Env-IZ and Env-IZN4 with representative Abs against four different epitopes on Env: V3 glycans (2G12 and PGT121), MPER (2F5), CD4 binding site (soluble CD4 and VRC01), and the CD4 induced state (48 d).
FIGURE 5. IZN4 is effectively immunosilenced without compromising antigen-specific responses. Rabbits were immunized on week 0, 2, 4, and 8. Midpoint binding titers of week 10 sera against IZ and IZN4 of rabbits that were immunized with (A) Env-IZ and Env-IZN4; (B) HA-IZ and HA-IZN4. C, gp120 binding titers and (D) HA binding titers of the sera used in A and B, respectively. Statistical analysis was performed using a Mann-Whitney test. *: p Յ 0.05; **: p Յ 0.01. residual binding comes from Abs binding to the linker regions between Env/HA and IZ, that were similar in the constructs used for immunization and detection. To further reduce the immunogenicity of IZN4 one could attach more than four N-glycans to IZ, although such a high density of PNGS might hamper efficient glycan attachment to all sites and/or affect the trimerization propensity. Alternatively, one could remove B cell epitopes by replacing the charged residues on the surface of IZ by neutral amino acids (48). Furthermore, one could predict and silence immunodominant T cell epitopes on IZ to dampen T cell help (49).
In conclusion, we found that commonly used protein trimerization domains can be highly immunogenic, but they can be immunosilenced by the addition of N-glycans. The immunosilenced IZN4 domain might be a useful tool for protein vaccines and therapeutics.