Structure-Guided Redesign Improves NFL HIV Env Trimer Integrity and Identifies an Inter-Protomer Disulfide Permitting Post-Expression Cleavage

Soluble HIV-1 envelope glycoprotein (Env) trimers are under active investigation as vaccine candidates in relevant pre-clinical models. Like SOSIPs, the cleavage-independent native flexibly linked (NFL) trimers are faithful mimics of the Env spike. Here, we analyzed multiple new designs to explore alternative modifications, informing tertiary interactions, while maintaining NFL trimer homogeneity and integrity. Accordingly, we performed a proline (P) substitution screen in the gp41 heptad repeat 1 region, identifying other trimer-enhancing Ps, including L555P. This P improved trimer integrity compared to I559P in selected properties. Next, we screened 15 structure-guided potential cysteine pairs in gp140 and found that A501C-L663C (“CC2”) forms an inter-protomer disulfide bond that demonstrably increased NFL trimer thermostability. We combined these two approaches with trimer-derived substitutions, coupled with glycine substitutions at helix-to-coil transitions, developed by our group. To increase the exposure of the fusion peptide (FP) N-terminus, we engineered an enterokinase (EK) cleavage site upstream of the FP for controlled post-expression cleavage. In combination, the redesigns resulted in highly stable and homogeneous NFL mimics derived from different clades. Following recombinant EK cleavage, the NFL trimers retained covalent linkage, maintaining a native-like structure while displaying enhanced stability and favorable antigenic features. These trimers also displayed increased exposure of neutralizing epitopes in the FP and gp120/gp41 interface, while retaining other neutralizing epitopes and occluding non-neutralizing elements. This array of Env-structure-guided designs reveals additional interactive regions in the prefusion state of the HIV Env spike, affording the development of novel antigens and immunogens.

Soluble HIV-1 envelope glycoprotein (Env) trimers are under active investigation as vaccine candidates in relevant pre-clinical models. Like SOSIPs, the cleavage-independent native flexibly linked (NFL) trimers are faithful mimics of the Env spike. Here, we analyzed multiple new designs to explore alternative modifications, informing tertiary interactions, while maintaining NFL trimer homogeneity and integrity. Accordingly, we performed a proline (P) substitution screen in the gp41 heptad repeat 1 region, identifying other trimer-enhancing Ps, including L555P. This P improved trimer integrity compared to I559P in selected properties. Next, we screened 15 structure-guided potential cysteine pairs in gp140 and found that A501C-L663C ("CC2") forms an inter-protomer disulfide bond that demonstrably increased NFL trimer thermostability. We combined these two approaches with trimer-derived substitutions, coupled with glycine substitutions at helix-to-coil transitions, developed by our group. To increase the exposure of the fusion peptide (FP) N-terminus, we engineered an enterokinase (EK) cleavage site upstream of the FP for controlled post-expression cleavage. In combination, the redesigns resulted in highly stable and homogeneous NFL mimics derived from different clades. Following recombinant EK cleavage, the NFL trimers retained covalent linkage, maintaining a native-like structure while displaying enhanced stability and favorable antigenic features. These trimers also displayed increased exposure of neutralizing epitopes in the FP and gp120/gp41 interface, while retaining other neutralizing epitopes and occluding non-neutralizing elements. This array of Env-structure-guided designs reveals additional interactive regions in the prefusion state of the HIV Env spike, affording the development of novel antigens and immunogens.
Keywords: hiV-1 env, soluble trimer, vaccines, antigens, stability, antigenicity Inter-Protomer Disulfide Permitting Post-Expression Cleavage Frontiers in Immunology | www.frontiersin.org July 2018 | Volume 9 | Article 1631 inTrODUcTiOn Despite many attempts, an effective HIV vaccine remains a daun ting scientific challenge. Robust antibody response to HIV Env, the sole target of broadly neutralizing antibodies (bNAbs), is likely required to induce such similar crossneutralizing Abs following vaccination. However, one big obstacle to develop such a vaccine is to generate a soluble Env that mimics the functional Env trimer present on the surface of the virus. Such nativelike trimers should ideally preferentially present bNAb epitopes but shield nonneutralizing epitopes to present to B cells as immu nogens in vivo.
After decades of development, advances in soluble HIV1 Env trimer design to permit the generation of a diverse array of nativelike trimers (1,2). Development of the soluble SOSIP.664 trimers provided initial proofofprinciple (3) as these trimers assume a prefusion nativelike conformation (4)(5)(6)(7). The SOSIPs are proteolytically cleaved by cellular furins to gp120 and gp41 subunits, covalently linked by an engineered intraprotomer disulfide bond A501CT605C (SOS). These trimers also contain a I559P substitution in the gp41 heptad repeat 1 (HR1) region to generate wellordered oligomers and, as well, require expression of exogenous furin for con formational integrity (3,(8)(9)(10)(11)(12)(13)(14)(15)(16)(17). In the past several years, we developed an improved nativelike trimer design, generating wellordered soluble Env mimics that are completely cleavage independent and termed native flexibly linked (NFL) trimers. This design uses a flexible linker (two copies of Gly4Ser, "G4S") to replace the natural cleavage site (18). The linker between the preCterminus of gp120 (residue 507) and Nterminus of gp41 (residue 512) allows the uncleaved trimers to achieve a nativelike conformation without the need of furin for precur sor processing. The original NFL trimer design contains the I559P mutation as well (18) to disfavor the postfusion state (3). Both the original SOSIP and NFL designs do not form a high percentage of wellordered trimers in all Env contexts. For instance, the original NFL design is relatively inefficient in generating high yields of trimers derived from clade C strains, such as 16055 (19).
To improve NFL design, we incorporated residues from BG505 [called trimerderived (TD) residues] into the 16055 NFLs, impro ving the propensity to form nativelike trimers (19) and, as well, the elicitation of tier 2 clade C neutralizing antibodies (20,21). Further improvements on the TD design by targeted glycine sub stitutions at helixtocoil transitions to disfavor the postfusion state (TD CC+), significantly improve trimer homogeneity, yield, stability, and antigenicity, resulting in the first highresolution clade C Env crystal structure (22).
Here, we analyzed three alternative strategies to inform biophysical relationships in the prefusion state while maintain ing wellordered, homogenous, and stable NFL trimers. First, we performed a proline substitution screen across the gp41 HR1 region identifying a new proline substitution (L555P) that improves the generation of stable trimers in some contexts. Second, using structureguided design, we screened 15 cysteine pairs at different positions, identifying a new cysteine pair A501CL663C ("CC2") that forms a stabilizing interprotomer disulfide bond that is compatible with most TD CC+ substitu tions. Third, because the NFL trimers are cleavageindependent (uncleaved), there is limited exposure of the Nterminus of the gp41 fusion peptide (FP) that is recently implicated as broadly neutralizing determinant of the bNAb, VRC34 (23). To restore the exposure of the FP Nterminus, we engineered an EK cleav age site upstream of the FP for controlled postexpression cleav age, greatly enhancing VRC34 recognition. These new designs provide insights regarding critical elements and interactions in the prefusion Env soluble trimer, adding substantially to the plethora of nativelike immunogens available for hypothesis driven, but empirical, immunogenicity determination.

Design of new nFl Trimer constructs
The BG505 NFL and 16055 NFL Env sequences (18) (16055 accession numbers EF117268 and BG505 DQ208458) were used as parental templates to generate gp140 trimer mutants. The parental NFL contains a proline substitution at residue 559 (I559P) to facilitate trimer formation (16). For the proline screen ing, a panel of 36 residues spanning the NFL HR1 region (from residue 548 to 585) was individually substituted by prolines. For the disulfide bond linkage screening, a panel of 15 cysteine pairs was generated. To increase the exposure of epitopes in the FP, we engineered an EK cleavage site (amino acids DDDDK, namely D4K) upstream of the FP for controlled postexpression cleav age. Finally, we built promising proline (555P), interprotomer cysteine linkage (C501-C663), and enterokinase cleavage site into our recently reported NFL TD CC+ trimer constructs (22), namely NFL TD 2CC+ D4K (schematic representing the NFL trimer design is shown in Figure 1A). Substitutions in the Envderived NFL glycoproteins were introduced via sitedirected mutagenesis PCR (Agilent Technologies) into expression plasmids and confirmed by sequencing (Genewiz). The final constructs are shown as schematic representations in Figures 1B,C.

expression and Purification of soluble Proteins
The constructs expressing the16055 and BG505 NFL trimeric Env were transiently transfected into suspension 293F cells as previ ously described (18,19,22). Env trimercontaining cell culture supernatants were harvested 4 days post transfection and puri fied by lectinaffinity chromatography (Galanthus nivalis, Vector Labs) followed by size exclusion chromatography (SEC) on a Superdex 200 16/60 or Superdex 200 10/300 GL (GE Healthcare). In most cases, the trimer peak was subjected to negative selection (NS) by the nonneutralizing mAbs GE136 or F105 to remove disordered trimers. The flowthrough from the GE136 or F105 column, containing the wellordered trimers, was resolved by a second SEC.
immunoprecipitation Immunoprecipitation (IP) of the expressed Env variants was done as described previously (18) with minor modifications. (a) Schematic depiction of the approaches used to redesign the NFL trimers using the available BG505 SOSIP.664 structure (PDB: 5CEZ). The trimer is shown in the ribbon representation with an inset of a closer view to illustrate selected approaches of NFL trimer redesign. The gp120 and gp41 in the first gp140 protomer are shown in light blue and yellow, in the second protomer in blue and green, and in the third protomer in gray. Proline substitutions screened in the gp41 heptad repeat 1 region (aa 548-585) are shown in the hot pink ribbon; the first protomer is shown in a close-up view (upper left). L555P is indicated with light pink spheres and I559P by the magenta spheres. The A501C-L663C cysteine pair forming a putative inter-protomer disulfide bond between the first and second protomers is shown in yellow and green spheres in the lower-left, close-up view. The engineered enterokinase (EK) cleavage site is shown as a solid blue line (upstream of the fusion peptide) to allow controlled post-expression cleavage.

Differential scanning calorimetry (Dsc)
The thermal transition points (Tm) of 16055 NFL and BG505 NFL variants were determined by DSC using a MicroCal VP capillary instrument (Malvern) as described previously (18,22).

Binding analyses by elisa and Biolayer interferometry (Bli)
ELISA and BLI analyses were performed as previously des cribed (18,24,25). Briefly, ELISA plates coated with 2 µg/ml antiHis mAb were used to capture NFL trimers (2 µg/ml) followed by primary mAbs (fivefold serially diluted, starting from 10 µg/ml) and a peroxidaseconjugated goat antihuman secondary Ab (1:10,000). Plates were developed using 3,3′,5,5 tetramethylbenzidine chromagen solution. The data were plotted in GraphPad Prism version 7. The BLI analyses were carried out on an Octet Red instru ment (ForteBio) with IgGs immobilized on antihuman IgG Fc capture sensors (ForteBio). The NFL trimers were assessed as free analytes in solution (PBS pH 7.4). The analytes started from 200 nM and then twofold serially diluted to a final con centration of 12.5 nM. Association and dissociation times were 2 and 2 min or 3 and 3 min, respectively. Data were analyzed using the ForteBio analysis software version 7.1 (ForteBio) and the kinetic parameters were calculated using a global fit 1:1 model.

electron Microscopy Data collection and Processing
The purified trimers were analyzed by negative stain electron microscopy (NSEM). A 3 µl aliquot containing ~0.01 mg/mL of the sample was applied for 15 s onto a carboncoated 400 Cumesh grid that had been glowing discharged at 30 mA for 30 s, then negatively stained with 0.7% uranyl formate for 45 s. Data were collected using a FEI T20 electron microscope operating at 200 kV, with an electron dose of ~45 e − /Å 2 and a magnification of 80,000× that resulted in a pixel size of 2.74Å at the specimen plane. Images were acquired with an Eagle 2k × 2k CCD camera (FEI) using a nominal defocus of 1,000 nm and the SerialEM software (26). Particles were selected from the micrographs, extracted, and a referencefree 2D class averages were obtained using RELION 2.1.0 (27).
n-glycan Profiling of env Variants With the hr1 P substitutions The five BG505 HR1 proline mutants were expressed in 293F cells and purified by the same methods as described above. The overall Nlinked glycosylation profiles of these Env variants were analyzed by hydrophilic interaction liquid chromatographyultraperformance liquid chromatography (HILICUPLC) (28). In brief, 10 µg of each of the Envs were resolved by SDSPAGE under nonreducing and reducing conditions, and the Coomassie blue stained bands cor responding to gp140 were excised and washed five times, alternatively with acetonitrile and water. The total Nlinked glycans were enzymatically released by treatment with PNGase F and the released glcyans were washed extensively in water and finally dried in SpeedVac concentrator as described earlier. The released glycans were fluorescently labeled with 2 aminobenzoic acid (2AA) and resolved on a Acquity BEH Amide column (2.1 mm × 10 mm, 1.7 μm particle size) (Waters) by HILICUPLC method as described in detail elsewhere. The raw data were analyzed by Empower 3 software. The relevant peakareas of different Nlinked oligomannose before and after EndoH digestion were integrated and normalized to calculate the percentage abundance of oligomannosetype glycans in all the Envs.

Multiple gp41 hr1 Proline substitutions improve Ordered soluble env nFl Trimer Formation
The original NFL trimer design contains the I559P mutation (18) to disfavor the postfusion state (3). However, both the original SOSIP and NFL designs do not form a high percent age of wellordered trimers in most Env contexts. Here, we introduced singlesite proline (P) substitutions in HR1 (resi dues 548-585) to identify other positions in Env that might efficiently form wellordered trimers. This screen had added interest to probe the plasticity of Env and the NFL platform in terms of accepting other P substitution in HR1 that were compatible with the prefusion state. To begin, we initiated a proline substitution screen in the BG505 context and studied their effects on NFL trimer formation by immunoprecipita tion analysis (IP) (18). In total, we interrogated 36 residues in HR1 ( Figure 1A) and found that at least five substitutions (S553P, N554P, L555P, Q562P, and Q563P) displayed favora ble features based upon trimerspecific bNAb recognition compared to that of the nonneutralizing mAb, F105, in the oligomeric mixture ( Figure S1A in Supplementary Material). Among the five substitutions, L555P resulted in ordered trimer production with slightly higher percentage of trim ers in a closed nativelike conformation (Figures S1B-D in Supplementary Material). These same P substitutions were not efficient in the BG505 SOS 501C/605C context to generate homogeneous trimers, resulting in a broad peak by SEC with no resolution of aggregates, trimers, and dimers/monomers (data not shown). Thermostability analysis by DSC of the five trimer variants revealed that the BG505 NFL L555P trimer was slightly more stable than BG505 NFL I559P trimer, displaying a 1°C increase in Tm ( Figure S1E in Supplementary Material; Table 1). BG505 NFL L555P trimers exhibited a similar anti genic profile compared to BG505 NFL I559P ( Figure S1F in Supplementary Material; Tables 2 and 3). The glycosylation profiles of the five NFL trimer variants were similar as that of BG505 SOSIP.664, but with a higher percentage of oligoman nose glycoforms (69.1-78.4%) ( Figure S2 in Supplementary Material). The highdensity of unprocessed oligomannose glycans in the gp120 subunit of the trimer is consistent with a nativelike, closed conformation of these trimers (29,30), limiting Nglycan enzymatic modifications to more complex glycans.
Because the HR1 region is relatively conserved among Env from different clades, we determined whether the P substitutions identified in the BG505 NFL context could be transferred to the clade C 16055 NFL, which is inefficient in its original I559P design in terms of yielding a high percentage of nativelike trimers (19). All five proline substitutions (S553P, N554P, L555P, Q562P, and Q563P) were compatible within the 16055 NFL backbone ( Figure S3A in Supplementary Material), displaying a homog enous trimer peak by SEC. In contrast, the 16055 NFL I559P SEC peak contained only small fractions of trimers (Figure 2A; Figure S3B in Supplementary Material), as previously reported (19). The trimer peak on SEC was slightly shifted "to the right" for 16055 NFL L555P design, consistent with improved trimer formation and yield ( Table 1). However, some trimer heteroge neity remained, so we used GE136affinity negative selection to remove nonnative trimers and other Env conformers. Following negative selection, the yield of the 16055 NFL L555P trimers were increased compared to the original I559P trimers ( Table 1). DSC analysis revealed that the L555P substitution generated more homogenous 16055 NFL trimers, with comparable ther mostability, compared to the I559P change ( Figure 2B; Figure  S3C in Supplementary Material). NSEM 2D average analysis showed that >94% of the trimers were in closed nativelike con formation ( Figure 2C; Figure S3D in Supplementary Material; Table 1). Antigenicity analysis by BLI and ELISA showed that L555P substitution improved the antigenic profile of 16055 NFL trimers, revealing increased recognition by the trimerspecific V2apexdirected bNAbs (PGT145, PGDM1400, and PG16) and V3targeting bNAb (PGT128), but littletono detectable binding by the nonNAbs (F105, GE136, 17b, and 44752D) ( Figure 2D; Figures S3E,F in Supplementary Material; Tables 2 and 3).
Taken together, these data show that the L555P substitution is comparable to, or improved, relative to the original I559P substitution regarding to form wellordered, homogenous, and stable NFL trimers.

an inter-Protomer Disulfide Bond improves the stability and antigenicity of the soluble nFl Trimers
To reduce the flexibility and increase the stability of the first generation of NFL I559P trimers, we sought to identify additional internal disulfide pairs to stabilize NFL trimers. Accordingly, we performed cysteine disulfide predictions based on the distance between Cα atoms (<5.7Å), taking into consideration a low prob ability of disrupting secondary structure. Guided by the existing SOSIP and NFL structures, we downselected 15 potential cys teine pairs, predicted to be within the 5.7Å sidechain distance, in the most likely rotamers (listed in Table S1 in Supplementary Material). We first assessed these potential new disulfide linkages in the original JRFL NFL I559P context. From the IP results, we identified a new cysteine pair (A501CL663C, designated here as "CC2") showing favorable recognition by PGT145, VRC06, and PGT151, and lowlevel recognition by F105 (data not shown).
We analyzed the CC2 cysteine pair in NFL Envs derived from different clades, JRFL (clade B), 16055 (clade C), and BG505 (clade A). All three NFL trimers containing the engineered CC2 cysteine substitutions form wellordered trimers, displaying a single sharp trimer peak by SEC ( Figure S4A in Supplementary Material). The purified trimers were resolved on Bluenative PAGE (BNPAGE), revealing a migration pattern consistent with predominantly trimeric Env. A low level of apparent dimer forms was detected for the new NFL CC2 design ( Figure S4B in Supplementary Material). Homogeneous trimer formation was confirmed by NSEM 2D class average as the CC2 trimers were highly ordered following negative staining and the EM analysis ( Figure 3A; Figure S4D in Supplementary Material; Table 1). To better confirm efficient 501-663 disulfide bond formation, we performed the gel analysis without and with reduction. Under reducing and nonreducing conditions, by SDSPAGE analysis, the NFL I559P (without CC2) trimer proteins migrated as gp140 monomer. However, the NFL CC2 proteins migrated as trimer under nonreducing conditions, whereas under reducing condi tions they migrated as gp140 monomer (Figure 3B; Figure S4C in Supplementary Material). These results were consistent with interprotomer disulfide bond formation, covalently linking adjacent protomers to form wellordered trimers.
Differential scanning calorimetry analysis revealed that the new interprotomer disulfide bond increased the thermostability of the NFL trimers, for example, the presence of the 501-663    Figure  S4E in Supplementary Material; Table 1). The antigenic profile of the NFL CC2 trimers analyzed by BLI and ELISA showed that CC2 in 16055 NFL improved trimer recognition by the trimerspecific bNAbs (PGT145, PGDM1400, and PG16) and  V3targeting bNAb (PGT128) (Figures 3C,D; Tables 2 and 3). In BG505 NFL, the improvement of the CC2 substitutions regarding antigenicity was evident, but to a lesser extent ( Figures S4F,G in Supplementary Material).
Taken together, these data indicate that the new cysteine pair A501CL663C (CC2) formed interprotomer disulfide bonds, increasing the thermostability and antigenicity of NFL trimers derived from different clades. Inter-Protomer Disulfide Permitting Post-Expression Cleavage Frontiers in Immunology | www.frontiersin.org July 2018 | Volume 9 | Article 1631 combinatorial approaches improve nFl Trimer Biophysical Properties and antigenicity As described above, the L555P substitution and the new CC2 interprotomer disulfide bond improved the NFL trimer design, separately. In addition, we recently reported that TD residue sub stitutions, glycine substitution at helixtocoil transitions, as well as targeted reduction of the inherent Env metastability facilitate the highyield production of crossclade stable soluble NFL TD CC+ trimers. Therefore, we combined these design strategies in the 16055 and BG505 NFL Env context, to generate NFL TD 2CC+ trimers. We assessed whether these combined designs were crosscompatible to yield improved, wellordered trimers. In addition, since our NFL trimers are uncleaved, there is limited exposure of the Nterminus of the gp41 FP recognized by the bNAb, VRC34, and variable accessibility to the cleavagesensitive bNAb, PGT151 depending upon the strain context (Figures S3E  and S4F,G in Supplementary Material). To increase the exposure and accessibility to the cleavagesensitive bNAbs, especially the gp41 FP Nterminus, we engineered an enterokinase (EK) cleav age site upstream of the FP. This modification would allow us to control postexpression cleavage of gp140, potentially exposing the VRC34FPdirected binding site (outlined in Figure 1). The FP was recently reported as a vulnerable target of the VRC34 and ACS202 bNAbs (23,31). The resulting trimers were designated as NFL TD 2CC+ D4K. For headtohead comparison, we generated two versions of NFL TD 2CC+ D4K trimers in 16055 and BG505 backbone, one containing the original I559P substitution and another possessing the identified L555P substitution. The 16055 and BG505 NFL TD 2CC+ D4K L555P and I559P variants were purified by lectinaffinity chromatography, fol lowed by SEC. The SEC trimer peak of 16055 NFL TD 2CC+ D4K I559P was much sharper than that of 16055 NFL TD 2CC+ D4K L555P, indicating the I559P substitution is more compatible with these modifications compared to L555P ( Figure 4A). As expected, the SEC profile revealed a single sharp trimer peak fol lowing NS. Similar SEC profiles were observed for BG505 NFL TD 2CC+ D4K L555P and I559P ( Figure S6A in Supplementary Material). Regardless of the P substitution used, the combinato rial design in 16055 dramatically increased the yield of well ordered trimers by over 13fold compared to original 16055 NFL I559P ( Table 1). Trimer formation was confirmed by BNPAGE ( Figures S5A and S6C in Supplementary Material). In addition, NSEM analysis revealed that >97% trimers in closed native like conformation (Figure 4B; Figure S6B in Supplementary Material; Table 1). Under nonreducing conditions, the NFL TD 2CC+ D4K proteins migrated as trimer on the SDSPAGE, whereas under reducing conditions these proteins migrated as a gp140 monomer, consistent with the formation of interprotomer disulfide bonds following the CC2 substitutions ( Figure 4C; Figure S6D in Supplementary Material).
Differential scanning calorimetry analysis revealed that the Tms of these trimers were over 80°C for both NFL TD 2CC+ D4K L555P and I559P, displaying an increase of over 21°C for 16055 NFL TD 2CC+ D4K and over 14°Cfor BG505 NFL TD 2CC+ D4K, relative to the "first generation" NFL I559P trimers (Figure 4D; Figure S6E in Supplementary Material). There was no significant difference of thermostability between NFL TD 2CC+ D4K L555P and I559P, but there was over a 3°C gain for NFL TD 2CC+ D4K compared to their corresponding NFL TD CC+ trimers, indicat ing the addition of CC2 increased thermal stability ( Table 1). The NFL TD 2CC+ D4K L555P and I559P trimers were highly stable in solution, displaying no significant degradation at 37°C for 30 h indicated by gel analysis (data not shown). We used the previously described panel of bNAbs and nonneutralizing Abs to assess the antigenicity of 16055 NFL TD 2CC+ D4K L555P and I559P trimers by ELISA and BLI, revealing that both trimer variants were recognized comparably by the trimerspecific bNAbs with no detectable recognition by the nonneutralizing Abs tested (Figure 4E; Figure S5  Taken together, these analyses demonstrated that the combi nation of L555P, CC2, TD CC+, and engineered postexpression cleavage site, preserve the prefusion state of the NFL trimers with improved trimer formation, biophysical properties, and antigenicity.

Post-expression cleavage of nFl TD 2cc+ D 4 K Trimers increases the exposure of cleavage-sensitive epitopes
Next, we assessed the impact of postexpression cleavage on the highly stable NFL TD 2CC+ D4K trimers regarding their structure, biophysical properties, and antigenicity. Following cleavage by rEK, 16055 NFL TD 2CC+ D4K L555P and I559P trimers showed single sharp trimer peaks on SEC (Figure 5A) with >94% of the trimers in a closed nativelike conformation as resolved by NSEM (Figure 5B). Similar results were observed for BG505 NFL TD 2CC+ D4K L555P and I559P trimers ( Figures  S7A,B in Supplementary Material). Under native conditions, cleaved 16055 NFL TD 2CC+ D4K proteins migrated as trimer on BNPAGE, similar to their uncleaved counterparts ( Figure S5A in Supplementary Material). To test the efficiency of postexpression cleavage, we performed SDSPAGE analysis. Under reducing conditions, cleaved 16055 NFL TD 2CC+ D4K L555P and I559P proteins migrated as two bands, gp120 and gp41, whereas the uncleaved proteins migrated as a single gp140 band (Figure 5C), indicating the trimers were completely cleaved by rEK. Similar results were obtained for BG505 NFL TD 2CC+ D4K L555P and I559P trimers ( Figure S7C in Supplementary Material).
Following rEKmediated cleavage, DSC analysis of putative trimers revealed single narrow symmetric thermal transition pro files, indicating that the trimers were homogeneous (Figure 5D; Figure S7D in Supplementary Material). The Tms of the cleaved 16055 NFL TD 2CC+ D4K L555P and I559P trimers were 82.8 and 82.6°C, respectively, displaying 2.6 and 2.5°C increases compared to their uncleaved counterparts. The Tms of cleaved BG505 NFL TD 2CC+ D4K L555P and I559P trimers were 81.6 and 81.0°C, respectively, with 0.7 and 0.6°C increases over their uncleaved counterparts ( Table 1). We used a panel of bNAbs and  Table 1. (e) Comparison of ELISA binding properties of selected mAbs to 16055 NFL TD 2CC+ D4K I559P and L555P trimers. The EC50 are summarized in Table 2 nonneutralizing Abs to assess the antigenicity changes of 16055 NFL TD 2CC+ D4K L555P and I559P trimers after cleavage by ELISA and BLI. Following rEK cleavage, the 16055 NFL TD 2CC+ D4K L555P and I559P trimers displayed increased rec ognition by the cleavagesensitive bNAbs VRC34 and PGT151, while retaining similar levels of recognition by bNAbs targeting  Table 1. (e) BLI comparison for the interactions of uncleaved and cleaved (w/ rEK) 16055 NFL TD 2CC+ D4K trimers with selected mAbs. The kinetic parameters are summarized in Table 3. 12 other epitopes (2G12, VRC01, and PGT128) (Figure 5E; Figure  S8A in Supplementary Material). In addition, there was no recognition by the nonneutralizing Abs (F105, GE136, 17b, and 44752D). Similar antigenic profile was observed for BG505 NFL TD 2CC+ D4K L555P and I559P trimers after cleavage as well ( Figures S7E and S8B in Supplementary Material).
Overall, these data indicate that the CC2 covalently linked the rEKcleaved trimers to maintain nativelike structure with enhanced stability and increased exposure of epitopes in the FP, gp120/gp41 interface.

DiscUssiOn
Here, we describe two NFL redesign strategies, namely proline sub stitution in HR1 and introduction of an interprotomer disulfide bond, to generate soluble NFL trimers with improved biophysi cal properties. By single proline substitution in HR1, we identify several positions that favor the prefusion state of Env, similar to 559P. In terms of wellordered soluble NFL trimer forma tion, the L555P substitution is the most comparable to, and in some context, even superior to the I559P substitution. Based on structureguided cysteine pair analysis, we found that A501C L663C (CC2) forms an interprotomer disulfide bond that stabi lizes the original uncleaved NFL trimer. We combined these two approaches to generate NFL TD CC+ trimers, containing the CC2 interprotomer linkage in combination with the intraprotomer 201-433 CC disulfide bond. To restore more efficient recognition by the FPdirected bNAb, VRC34, we substituted an EK cleavage site upstream of the FP for controlled postexpression cleavage. We show that following cleavage, the NFL TD 2CC+ D4K trimers were covalently linked by CC2 to maintain nativelike structure, displaying enhanced stability and increased exposure of epitopes in the FP and gp120/gp41 interface. Thus, Envstructureguided trimer redesign results in homogenous crossclade immunogens with the potential to advance vaccine development, Ab isolation, and Env structural analysis.
The ability of multiple proline substitutions in HR1 to generate varying degrees of wellordered NFL trimers in the oligomeric population is interesting and consistent with the concept that this region contributes to metastability. In fact, one approach to gener ate cleavageindependent nativelike trimers modifies the FP and portions of HR1 to generate stable, wellordered UFO trimers (32). Guenaga et al. reported that glycine substitutions at key coil tohelix HR1 transitions favor trimer formation in the prefusion state (22). Another study shows that the interface between α6 in HR1 and α9, and the intersubunit βsheet are critical for trimer stability (33). This is likely related to HIV gp41 metastability in the context of the native spike. Following gp120 engagement with receptor/coreceptor, the spike undergoes conformational changes from a putative highenergy prehairpin conformation to a lowenergy, stable postfusion sixhelix bundle (6HB) conformation, mediating viraltohost cell membrane fusion. Mutations K574R and I535M in HR1 are reported to increased Env stability in different clades (34,35), and V570D and I573D can destabilize the 6HB formation (36), favoring the prefusion state on the cellsurface, independent of the I559P substitution. The SOSdefined I559P substitution is an important modification to stabilize soluble trimer mimics in the prefusion conformation for SOSIPs (3,(8)(9)(10)(11)(12)(13)(14)(15)(16)(17)33), singlechain gp140 (37), DSSOSIP (38,39), and the NFL trimers (18,19,22). Here, the L555P identi fied in our P screen is comparable to or even better than I559P in some NFL contexts but is not compatible with the BG505 SOS backbone in this regard, indicating additional advantages of the NFL design to accept specific engraftments, perhaps since it lacks the engineered 501-605 disulfide.
Internal engineered disulfide bonds clearly increase trimer protein stability. For example, the respiratory syncytial virus F protein trimers (dSCAV) and influenza virus HA soluble trimers provide proofofprinciple (40). For HIV Env, the well documented intraprotomer SOS bond covalently linking gp120 residue 501 to gp41 residue 605 impacts favorably on trimer formation (3,7,16,17). A second intraprotomer disulfide bond linking gp120 residues 201-433, stabilizes the bridging sheet in its prefusion state and thereby reducing conformational changes and potential CD4induced exposure of nonneutralizing Abs epitopes in both NFL and SOSIP trimers (19,39). Two addi tional disulfide bonds have been introduced into BG505 SOSIP and generate hyperstable nativelike trimers (BG505 SOSIP. v6) but results in lower trimer yields (41). Here, we identified a new cysteine pair called CC2 (A501CL663C) that forms an interprotomer disulfide bond and increases the thermostability of multiple NFLs, augmenting trimer formation while maintain ing desired antigenicity. Although the engineered disulfide is efficiently formed, it does generate some "offpathway" dimers for reasons that are not yet clear.
Size exclusion chromatography analyses of NFL TD 2CC+ D4K trimers reveals that I559P is more compatible with the CC2 and TD CC+ modifications than is either the L555P or Q563P substitutions (data not shown), indicating that further optimization is needed to render the L555P more compatible in combination with the CC2 and TD CC+ substitutions. To approach this issue, we performed NS to remove heterogeneous, nonnativelike trimers from all NFL TD 2CC+ D4K variants. Following the NS step, all trimer variants are homogenous and highly stable (>80°C Tm), without degradation at 37°C for 30 h. Following postexpression cleavage by rEK, the thermostabil ity of trimers increases as much as 2.6°C, while maintaining a favorable antigenic profile. Uncleaved NFL trimers exhibit isolatespecific effects regarding the exposure of the PGT151 and VRC34 epitopes, and particular stabilizing modification also affect the accessibility of these epitopes in the uncleaved context. However, following postexpression cleavage by rEK, trimer recognitions are greatly increased for VRC34 and PGT151 rela tive to most uncleaved states, indicating increased exposure of these binding sites. Although rEK postcleavage adds another processing step for the NFL trimers, this approach does allow such trimers to be cleaved in a controlled manner to expose the Nterminus of the FP for improved presentation of the VRC34 epitope on NFL interprotomer disulfidelinked immunogens, if so desired.
In sum, by the multifold modifications described here, we generated highly stable prefusion, closed, and cleavageindependent Env trimers from multiple strains with favorable antigenic and biochemical properties that are amenable for HIV vaccine development. The design strategies may also be applicable to stabilize fusion proteins derived from other enveloped viruses that undergo conformational changes in transitioning from pre fusion to postfusion states as candidate vaccines or for structural analysis.
aUThOr cOnTriBUTiOns LY, SS, and RTW designed research studies. LY, SS, CC, RW, AJB, and NV performed experiments and data analysis. LY and RTW wrote the paper. All authors reviewed the results and approved the final version of the manuscript.

acKnOWleDgMenTs
We thank Malcolm Wood and Theresa Fassel of the Electron Microscopy Core at TSRI for their assistance in sample prepara tion, image collection of trimer EM field views, and helpful in discussions. The work was supported by HIVRAD grant P01 AI104722 (SS, CC, RW, and RTW), grant P01 AI124337 (JG and RTW), the Scripps CHAVIID AI100663 (LY and RTW), grant R01 AI0988602, and funding from the International AIDS Vaccine Initiative (IAVI) (KT and RTW  Table 3.  Table 2. (F) BLI measurements for trimers interaction with selected mAbs. The kinetic parameters are summarized in Table 3.  Table 2.
(g) BLI measurements for BG505 and 16055 NFL CC2 trimers with selected mAbs. The fitting curves are shown in green color, and the kinetic parameters are summarized in Table 3.  Table 2. (c) BLI measurements of 16055 NFL TD 2CC+ D4K I559P and L555P trimers without rEK cleavage. The fitting curves are shown in blue color, and the kinetic parameters are summarized in Table 3.  Table 2. (g) BLI measurements of BG505 NFL TD 2CC+ D4K I559P and L555P trimers without cleavage. The fitting curves are shown in blue color and the kinetic parameters are summarized in Table 3.  Table 3.  Table 2. reFerences