Improving the Expression and Purification of Soluble, Recombinant Native-Like HIV-1 Envelope Glycoprotein Trimers by Targeted Sequence Changes

ABSTRACT

Soluble, recombinant native-like envelope glycoprotein (Env) trimers of various human immunodeficiency virus type 1 (HIV-1) genotypes are being developed for structural studies and as vaccine candidates aimed at the induction of broadly neutralizing antibodies (bNAbs). The prototypic design is designated SOSIP.664, but many HIV-1 env genes do not yield fully native-like trimers efficiently. One such env gene is CZA97.012 from a neutralization-resistant (tier 2) clade C virus. As appropriately purified, native-like CZA97.012 SOSIP.664 trimers induce autologous neutralizing antibodies (NAbs) efficiently in immunized rabbits, we sought to improve the efficiency with which they can be produced and to better understand the limitations to the original design. By using structure- and antigenicity-guided mutagenesis strategies focused on the V2 and V3 regions and the gp120-gp41 interface, we developed the CZA97 SOSIP.v4.2-M6.IT construct. Fully native-like, stable trimers that display multiple bNAb epitopes could be expressed from this construct in a stable CHO cell line and purified at an acceptable yield using either a PGT145 or a 2G12 bNAb affinity column. We also show that similar mutagenesis strategies can be used to improve the yields and properties of SOSIP.664 trimers of the DU422, 426c, and 92UG037 genotypes.

IMPORTANCE Recombinant trimeric proteins based on HIV-1 env genes are being developed for future vaccine trials in humans. A feature of these proteins is their mimicry of the envelope glycoprotein (Env) structure on virus particles that is targeted by neutralizing antibodies, i.e., antibodies that prevent cells from becoming infected. The vaccine concept under exploration is that recombinant trimers may be able to elicit virus-neutralizing antibodies when delivered as immunogens. Because HIV-1 is extremely variable, a practical vaccine may need to incorporate Env trimers derived from multiple different virus sequences. Accordingly, we need to understand how to make recombinant trimers from many different env genes. Here, we show how to produce trimers from a clade C virus, CZA97.012, by using an array of protein engineering techniques to improve a prototypic construct. We also show that the methods may have wider utility for other env genes, thereby further guiding immunogen design.

INTRODUCTION

The envelope glycoprotein trimer is the target of virus-neutralizing antibodies (NAbs) that arise during infection by human immunodeficiency virus type 1 (HIV-1). This trimer, located on the surface of virus particles, is therefore a focus of vaccine design programs (1–11). Of particular relevance is a subset of NAbs with activity, in vitro, against multiple circulating HIV-1 strains. These broadly neutralizing antibodies (bNAbs) target relatively conserved regions or motifs that span most of the trimer surface. Accordingly, immunogens that present the epitopes for multiple bNAbs are being developed and evaluated as components of strategies aimed at inducing similar antibodies in test animals and, eventually, humans (3, 7, 10, 12–15). While the presence of a bNAb epitope on an immunogen does not guarantee that an appropriate immune response will be elicited, its absence means that such a response is highly unlikely.

The Env trimer is heavily glycosylated and metastable. The dense array of glycans on the trimer surface shields key conserved epitopes from the humoral immune response, particularly regions associated with binding to cell surface receptors (CD4 and CCR5 or CXCR4) (16–18). The instability of the trimer reflects its need to undergo receptor-mediated conformational changes during the process of virus-cell fusion (19–21). Both factors create challenges when making Env trimers as recombinant proteins for immunogenicity testing and as the substrates for high-resolution structural studies (11, 22–24). Most Env immunogen development programs involve producing proteins in a secreted (i.e., soluble) form to improve the yield and simplify purification strategies (8, 12, 25, 26). However, the required removal of the membrane-anchoring domain exacerbates the fragility of appropriately cleaved trimers and necessitates the use of stabilization procedures (6, 23, 27, 28). One now widely used method, the SOSIP.664 design, involves an engineered intersubunit disulfide bond (SOS) between gp120 and the gp41 ectodomain (gp41_ECTO), a gp41-stabilizing substitution (I559P), and a truncation at gp41 residue 664 (2, 6, 11, 27–29). Further SOSIP trimer design improvements have been described previously (30–32).

When used as immunogens in small animals, native-like (NL) SOSIP.664 trimers of several genotypes (for example, but not limited to, BG505 from clade A, B41 from clade B, and CZA97.012 from clade C) have elicited strong and consistent NAb responses to the autologous tier 2 (i.e., neutralization-resistant) viruses (3, 7, 33). This type of narrow-specificity response may serve as a necessary starting point for strategies intended to create neutralization breadth (7, 33–39).

For some immunization strategies, multiple NL trimers based on diverse genotypes could be useful. However, SOSIP.664 modifications are not a universal panacea for the trimer instability problem. For some Env genotypes, the association between the gp120 and gp41_ECTO subunits remains unstable, and unacceptable amounts of conformationally irregular, nonnative trimers (referred to as pseudotrimers) or trimer fragments are produced at the expense of native-like trimers (8, 22, 40). Positive-selection columns based on bNAbs such as PGT145 or PGT151 that recognize epitopes unique to NL trimers can overcome these problems but sometimes at the cost of an unacceptable reduction in yield (22). The same factors apply to negative-selection columns that remove unwanted forms of Env (8).

Here, we have investigated how to produce NL trimers more efficiently from “difficult Env genotypes,” exemplified by CZA97.012. When SOSIP.664 changes were introduced into this env gene, the resulting Env proteins included dimers, pseudotrimers, and aggregates, together with a low yield of the desired NL trimers (22). The use of a PGT151 bNAb positive-selection column, however, allowed enough NL CZA97.012 SOSIP.664 trimers to be purified and tested as immunogens in rabbits, where they induced a strong and consistent autologous NAb response (3). To build on these observations, and also to learn more about trimer production in general, we have explored various ways to increase the yield of CZA97.012 trimers with an NL conformation.

We focused our trimer redesign efforts initially on the V2-V3 regions at the trimer apex, as the limited reactivity of the PGT145 bNAb with its quaternary epitope on CZA97.012 SOSIP.664 trimers may be indicative of some instability in this area. The gp120-gp41_ECTO interface can be another unstable area, and here, a set of eight trimer-derived sequence changes (TD8) has been reported to improve stability for some trimer genotypes (40). We found that the TD8 method was not sufficient to confer an appropriate level of stability in the context of the CZA97.012 and some other genotypes, but we identified additional sequence changes that led to the stabilization of the gp120-gp41_ECTO interface. The outcome of our reengineering efforts was a fully native-like trimer, designated CZA97.012 SOSIP.v4.2-M6.IT, that can be produced in a stable CHO cell line and purified by either 2G12 or PGT145 affinity columns at acceptable yields. As similar sequence modifications to SOSIP.664 constructs from other genotypes (i.e., DU422, 426c, and 92UG037) also facilitated the production of NL trimers, the methods that we describe here may have reasonably broad utility.

RESULTS

Purification of CZA97.012 SOSIP.664 trimers via various bNAb affinity columns.

CZA97.012 SOSIP.664 trimers were first purified by a Galanthus nivalis agglutinin (GNA) lectin affinity column, which is not conformationally selective, followed by size exclusion chromatography (SEC). The Env proteins eluted from the lectin column contained a mixture of aggregates, trimers, dimers, and monomers when analyzed by blue native PAGE (BN-PAGE) (Fig. 1A). The trimer band was diffuse and indistinct, which contrasts with our experiences with the BG505 and B41 genotypes of SOSIP.664 trimers (6, 41). Similarly, when SEC was used to fractionate the lectin-purified CZA97.012 Env proteins, the trimer peak was not well separated, and only a few fractions containing pure trimers could be collected (Fig. 1B). The overall recovery of lectin/SEC-purified trimers was 20 to 25% of the total Env protein input, and trimer purity was estimated to be ∼80% (Fig. 1B and Table 1). Negative-stain electron microscopy (NS-EM) showed that the lectin/SEC-purified trimers were structurally heterogeneous, with both NL trimers and pseudotrimers being present in approximately equal proportions (Fig. 1C and Table 1).

FIG 1.

Properties of CZA97.012 SOSIP.664 Env proteins transiently expressed in 293F cells. (A) Env proteins purified by the indicated affinity columns were analyzed on a BN-PAGE gel. Bands corresponding to various Env forms are marked. In each case, the left lane shows molecular mass marker proteins. (B) SEC profiles of the Env proteins eluted from the same affinity columns (not done for PGT145, as the yield was too low). The trimer-containing fractions, shaded green, were collected and pooled for further study. The x and y axes record the elution volume and UV absorbance, respectively. (C) NS-EM images of the trimers purified by the four methods. The percentage of total trimer in the NL form is indicated at the bottom of each panel. The Env proteins used in this experiment contain a C-terminal His tag, which is used for ELISAs but not for purification purposes. Our general experience is that nontagged Env proteins tend to form fewer aggregates, but this can vary among genotypes.

TABLE 1.

Properties of CZA97.012 SOSIP trimer variants

Constructa	Purification methodb	% trimer yieldc	% purityd	% NL trimerse	Trimer yield (mg/liter)f
SOSIP.664	GNA lectin/SEC	20–25	∼80	50	0.4–0.6
SOSIP.664	2G12/SEC	20–25	∼80	76	0.3–0.4
SOSIP.664	PGT145	<10	>95	100	<0.1
SOSIP.664	PGT151/SEC	10–20	>95	100	0.1–0.2
SOSIP.664-M6	PGT151/SEC	10–20	>95	ND	0.1–0.2
SOSIP.664-M6	PGT145	15–25	>95	ND	0.1–0.2
SOSIP.664-M6.TD8	PGT145	20–25	>95	ND	0.2
SOSIP.664-M6.IT	PGT145	40–50	>95	ND	0.4–0.5
SOSIP.664-M6.IT	2G12/SEC	ND	ND	ND
SOSIP.v4.2-M6.IT	PGT145	40–50	>95	>95	0.4–0.6
SOSIP.v4.2-M6.IT	2G12/SEC	30–40	>95	>95	0.3–0.5
SOSIP.v4.2-M6.IT (CHO)	PGT145	60–70	>95	>95	2.0
SOSIP.v4.2-M6.IT (CHO)	2G12/SEC	40–50	>95	>95	1.2

^aThe constructs are summarized in Fig. 2. All the constructs except the one marked “(CHO)” were expressed by transient transfection of 293F cells.

^bThe column chromatography methods used to purify trimers are described in the text.

^cFor each construct, the total amount of Env protein (monomers, dimers, trimers, and aggregates) isolated from transfection culture supernatants via the GNA lectin affinity column is defined as 100%. The amount of total trimers (NL plus nonnative) purified by the specified procedure is then compared to this reference value to calculate the reported percentages.

^dThe purity of the purified trimers was assessed by BN-PAGE. Contaminants, when detected, included dimers and aggregates.

^eThe purified trimers were assessed by NS-EM and classified as NL or nonnative. The ratio of NL to total trimers (100%) was then determined. ND, not done.

^fThe values shown are estimates of the total amounts of trimers that can be purified from 1 liter of the culture supernatant derived from the 293F cell transient transfections or the stable CHO cell line. In most cases, the values approximately correspond to the yield of NL trimers, but for SOSIP.664 trimers purified by GNA lectin/SEC (∼50% in the NL form), the yield of NL trimers is proportionately lower.

We next compared various bNAb affinity columns. Our previous enzyme-linked immunosorbent assay (ELISA) studies on the antigenicity of CZA97.012 trimers showed that they bound efficiently to the PGT151 bNAb against a quaternary (i.e., specific for NL trimers) epitope at the gp120-gp41_ECTO interface but poorly to the PGT145 bNAb against a quaternary epitope at the trimer apex (22). The 2G12 bNAb was also poorly reactive in ELISAs, which may reflect the absence of the N295 glycan from the parental CZA97.012 construct (22). The ELISA data suggested that PGT151 would be the best choice to obtain NL trimers, but we also assessed 2G12 and PGT145 columns.

Purification on a 2G12 column.

A BN-PAGE gel showed that the 2G12-purified CZA97.012 Env proteins included aggregates, trimers, dimers, and monomers (Fig. 1A). This outcome reflects the limited selectivity of the 2G12 column for trimers over nontrimeric forms of Env, a selectivity that is, however, better than can be achieved on lectin columns. The trimers were separated from the aggregates via SEC but not fully from the dimers (Fig. 1B). The overall recovery of 2G12/SEC-purified trimers and their purity were comparable to those of the lectin/SEC product (Table 1). NS-EM micrographs showed that ∼76% of the trimers were NL, which is an increase over the ∼50% estimate for the lectin/SEC procedure (Fig. 1C and Table 1) (22).

Purification on a PGT151 column.

As we previously reported, PGT151 affinity purification followed by SEC (PGT151/SEC) produced almost completely pure trimers; a small amount of aggregates was visible on the BN-PAGE gel, but they were eliminated by the SEC column (Fig. 1A and B) (22). NS-EM images showed that the PGT151/SEC-purified trimers are fully NL, which is again consistent with the conformational selectivity provided by the PGT151 bNAb (Fig. 1C). The overall recovery of trimers was still, however, quite low (0.1 to 0.2 mg/liter) (Table 1).

Purification on a PGT145 column.

The Env proteins on a PGT145 column were overwhelmingly trimeric when analyzed by BN-PAGE, which is consistent with the conformational selectivity of PGT145 (Fig. 1A). Indeed, the PGT145-isolated trimers required no further purification by SEC. NS-EM micrographs revealed that >95% of these trimers were in the NL conformation (Fig. 1C). However, as predicted by the poor PGT145 reactivity of CZA97.012 trimers in an ELISA, the yield from the PGT145 column was very low (<0.1 mg/liter, which is a minor fraction of the total Env proteins present) (Table 1) (22).

Overall, these studies confirmed and extended our previous report that the PGT151/SEC method is the most suitable way to purify fully NL CZA97.012 SOSIP.664 trimers (22). However, the NL trimer yield is poor (0.1 to 0.2 mg/liter) compared to our experiences when SOSIP trimers of many other genotypes, including BG505 and B41, are also produced by transient transfection (>1 mg/liter). Thus, if NL trimers of the CZA97.012 genotype are to be produced on any practical scale, various deficiencies that affect their yield need to be addressed.

Restoring the PGT145 quaternary epitope on CZA97.012 trimers via V2 and V3 changes.

The quaternary epitope for the PGT145 bNAb at the trimer apex is a feature of many NL trimers. However, PGT145 neutralizes the CZA97.012 virus only weakly; it binds CZA97.012 SOSIP.664 trimers inefficiently in a surface plasmon resonance (SPR) assay due to a rapid off-rate, is nonreactive with these trimers in ELISAs, and cannot be used to affinity purify them (22). Accordingly, we sought to improve the presentation of the PGT145 epitope via mutagenesis, to see whether doing so would also increase the stability of the SOSIP.664 trimer apex. The various changes made to the CZA97.012 Env sequence are summarized in Table 2 and Fig. 2. The antigenicity of the resulting unpurified Env proteins was assessed by screening transfection supernatants by an ELISA and, for some constructs, confirmed using PGT151/SEC-purified trimers (Table 2 and Fig. 3A).

TABLE 2.

Antigenicity of unpurified CZA97.012 Env variants in ELISA^b

Construct	Antibody binding
	PG16	PGT145	PGT151	14e
CZA97.012 SOSIP.664a (wild type)	−	−	+++	+++
BG505-V2	−	−	+++	+++
92UG037-V2	−	−	+++	+++
CAP45-V2	−	−	+++	+++
DU422-V2	−	−	+++	+++
Q171K	−	−	+++	+++
I165L Q171K G172V	+/−	+/−	+++	+++
I165L Q171K G172V I192R	+/−	+/−	+++	+++
Q171K ΔN184	+	+	+++	+++
Q171K N187S	+	+	+++	+++
I165L Q171K G172V ΔN184 I192R (M5)a	+	+	+++	+++
I165L Q171K G172V N187S I192Ra	+	+	+++	+++
M307Ia	−	−	+++	+++
I165L Q171K G172V ΔN184 I192R M307I (M6)a	+++	+++	+++	+++
I165L Q171K G172V N187S I192R M307Ia	++	++	+++	+++

^aIn follow-up experiments, the indicated constructs were expressed in 293F cells, and the trimers were purified by the PGT151/SEC method to allow the generation of more precise, confirmatory ELISA data (Fig. 2).

^bAll the constructs are based on the wild-type CZA97.012 SOSIP.664 sequence and include a His tag. The Env proteins were transiently expressed in 293T cells, and the culture supernatants were tested in an ELISA. In the BG505-V2, 92UG037-V2, CAP45-V2, and DU422-V2 constructs, the CZA97.012 V2 loop was replaced by the corresponding loops from those genotypes. The other mutant constructs contain the listed single and multiple point substitutions. The five changes in V2 and the six changes in V2 plus V3 that create the CZA97.012 M5 and M6 constructs, respectively, are listed. The magnitude of antibody reactivity in ELISAs is recorded semiquantitatively on a scale from − (no binding) to +++ (strong binding).

FIG 2.

Schematic representation of CZA97.012 SOSIP constructs including V2 and V3 sequence changes. (A) Schematic representations of SOSIP variants. The boundaries of the gp120 variable loops are marked by vertical dotted lines. The engineered intermolecular disulfide bond between residues C501 and C605, the I559P change in gp41_ECTO, and the hexa-arginine (R6) motif at the gp120-gp41_ECTO cleavage site are all shown in blue. The gray bars within V2 and V3 indicate the locations of the point substitutions present in all of the M6 constructs; these changes are listed in green at the bottom of the schematic. The various TD8 mutations are listed in black directly beneath the SOSIP-M6.TD8 schematic (note that no TD8-derived change was needed at position 47 in the CZA97.012 sequence, as D47 is the wild-type residue). The IT mutations are shown, also in black, beneath the SOSIP.v4.2-M6.IT schematic. The additional trimer-stabilizing changes present in the SOSIP.v4.2-M6.IT construct are marked by yellow bars and recorded in black above the schematic. (B) Sequence alignment of the V2 region and part of the V3 region of the CZA97.012 and other Env genotypes. Residues are numbered according to the HXBc2 system. The residues highlighted in boldface type and in blue, and the potential N-glycosylation (PNG) sites in the hypervariable region marked in red, are all known to influence the PGT145 epitope in some sequence contexts, as do other residues highlighted in green. (C) Alignment of the SOSIP.664 sequences of the indicated genotypes. The variable loop sequences are highlighted in gray. The gp120-gp41 interface residues selected for modification in the CZA97.012 SOSIP.v4.2-M6.IT sequence are highlighted in green. The changes are based on the corresponding BG505 and/or B41 residue.

FIG 3.

Substitutions in V2 and/or V3 restore the PGT145 epitope. (A) The ELISA binding curves show the reactivity of the indicated NAbs and a non-NAb with various PGT151/SEC-purified CZA97.012 SOSIP trimer variants. OD₄₅₀, optical density at 450 nm. (B) The SPR binding profiles show the reactivity of the indicated NAbs at 500 nM with 2G12/SEC-purified, immobilized wild-type BG505 SOSIP.664 and I307M mutant trimers. The response difference (RU) is given on the y axis as a function of time (seconds) on the x axis. The kinetic values for PGT145 Fab binding are tabulated. The table shows the mean values ± standard errors of the means of fitted parameters obtained from a conformational-change model. Means are given with two significant figures, and standard errors of the means are given with one. K_D = k_off/k_on describes the initial docking interaction, K_F = k_f/k_b describes the forward conformational change, and K_D(conf) = 1/{(k_on/k_off)[1 + (k_f/k_b)]} describes the overall interaction. S_m is the estimated stoichiometric value, which represents the number of Fab molecules per trimer at the maximum extent of binding.

Our initial approach was based on the assumption that the CZA97.012 V2 loop sequence was suboptimal for PGT145 binding and neutralization. Accordingly, we replaced the entire CZA97.012 V2 loop (residues 157 to 196) by the corresponding sequences from viruses that are potently neutralized by PGT145: BG505, 92UG037, CAP45, and DU422. However, none of the resulting chimeric SOSIP.664 Env proteins bound either PGT145 or a second trimer apex bNAb, PG16 (Table 2). Because a change as radical as a complete V2 loop exchange may create too much overall distortion in the trimer apex area, we next used a more targeted mutagenesis strategy focused on V2 residues known to influence the PGT145 epitope, which include, but are not limited to, positions 168 to 171 (42–44). The CZA97.012 V2 loop contains four residues that usually differ in the BG505, 92UG037, CAP45, and DU422 sequences: I165, Q171, G172, and I192. We therefore introduced the I165L, Q171K, G172V, and I192R substitutions, singly or in combination, into CZA97.012 SOSIP.664 trimer variants (Table 2 and Fig. 2). In addition, we assessed the effects of deleting the glycan site from V2 position 184 (ΔN184 change) or altering the location of the glycan site at position 187 (N187S); these V2 glycan changes are known to improve the neutralization of the HIV-1 LT5 strain by the trimer apex bNAbs PG9 and PG16 (45).

In general, single changes in the V2 core epitope did not improve PGT145 or PG16 binding, but some double or triple substitutions had modestly beneficial effects. The most substantial increase in PGT145 or PG16 reactivity was seen when the I165L, Q171K, G172V, ΔN184, and I192R substitutions were introduced collectively into a construct designated CZA97.012 SOSIP.664-M5 (Table 2 and Fig. 3). All of the mutants and the wild-type trimer bound the PGT151 trimer interface bNAb comparably, implying that this conformationally sensitive epitope was unaffected by the V2 sequence changes (Table 2 and Fig. 3).

The initial screening of trimer mutants by an ELISA indicated the importance of V2 residues for PGT145 binding. However, emerging knowledge on the structural influences on this epitope suggests that “breathing” at the trimer apex is essential for PGT145 binding (62). We define breathing as a relaxation of the V1/V2 and V3 packing that is an intrinsic property of Env, integral to its requirement to undergo large, receptor-induced conformational changes to drive the fusion of the virus and host cell membranes. In particular, we found residues outside the PGT145 epitope where changes influence how this bNAb binds and neutralizes, and we identified a subtle allosteric mechanism by which the stability, or the lack thereof, of the variable loops at the apex alters the antigenicity of the trimer (62). Accordingly, we took advantage of this new information when analyzing residues at the interface of V1/V2 with V3 to identify sites where substitutions might have a beneficial effect on CZA97.012 SOSIP trimers.

A comparison of the CZA97.012 V3 sequence with those from BG505, 92UG037, CAP45, and DU422 showed that M307 was present only in CZA97.012, whereas the other four strongly PGT145-reactive Env proteins contained I307 (HXBc2 numbering system) (Fig. 2B). This finding from a small sample set was consistent with an analysis of 3,794 sequences in the Los Alamos Sequence Database, where the frequency of M307 was 4.6% compared to 71.1% for I307. We then inspected the potential structural consequences of having methionine at position 307 and noted that it is positioned to make a cross-strand contact with Q170 in V1/V2, potentially stabilizing the apex (46, 62). We therefore introduced the M307I change into the CZA97.012 SOSIP.664 trimer. By itself, this point substitution had no effect on PGT145 binding, but when it was combined with the five V2 changes present in the SOSIP.664-M5 construct, PGT145 binding was further improved (Table 2 and Fig. 3). This new construct is designated SOSIP.664-M6. In an ELISA using unpurified transfection supernatants that contain multiple forms of Env, the V3 non-NAb 14e reacted strongly with all the constructs, including SOSIP.664-M6 (Table 2) (47). However, when PGT151-purified NL trimers were similarly assessed, all of the constructs bound 14e comparably and to only a modest extent (Fig. 3A). The implication is that V3 is appropriately sequestered on NL trimers of the CZA97.012 genotype.

To further examine the role of Ile-307 in PGT145 reactivity, we made the converse I307M substitution in the BG505 genotype. In an SPR analysis, the PGT145 Fab bound somewhat less well to purified BG505 SOSIP.664-I307M mutant trimers than the wild type, in that the stoichiometry was lower (0.42 versus 0.61) and the overall dissociation constant [K_D(conf)] was 2-fold higher (18 versus 9 nM) (Fig. 3B). Thus, the identity of V3 residue 307 (Ile versus Met) modestly influences the presentation of quaternary epitopes at the apex of the CZA97.012 trimer and, similarly, for its BG505 counterpart.

Residues at the gp120-gp41 interface influence trimer formation.

We found that the CZA97.012 SOSIP.664-M6 trimers could now be efficiently purified using a PGT145 affinity column, with a yield, 0.1 to 0.2 mg/liter, comparable to those of wild-type SOSIP.664 trimers isolated with the PGT151 column (Table 1). Yields on this scale are, however, still impractically low. Moreover, the trimer yield still represented only ∼15 to 25% of the total Env proteins present in the transfection supernatant, with the rest being aggregates, dimers, and monomers (Table 1). Any further increase in the trimer yield therefore requires improving the efficiency with which they form and/or their stability.

A set of eight trimer-derived (TD8) sequence changes has been reported to increase the propensities of JR-FL (clade B) and 16055 (clade C) Env proteins to form NL trimers (40). We therefore assessed whether these mutations would work in the CZA97.012 sequence context, with the SOSIP.664-M6 mutant described above serving as the template (Fig. 2A). When the resulting SOSIP.664-M6.TD8 mutant was produced by transient transfection, the yield of trimers from the PGT145 column was not improved compared to SOSIP.664-M6 (Table 1). Thus, either the TD8 changes do not work in the context of the CZA97.012 sequence or their influence is insufficient to overcome other instabilities in the resulting trimers.

To gain further insights into trimer formation, we compared the CZA97.012 sequence with those of B41 and BG505 (Fig. 2C). The latter two sequences produce SOSIP.664 trimers efficiently and without contamination by nonnative forms (6, 26, 41). Sequence alignment allowed us to identify residues in the CZA97.012 sequence that differed from those of BG505 and/or B41 and that potentially affected the gp120-gp41_ECTO interface, as revealed from the structure of the BG505 SOSIP.664 trimer. Because of the large number of sequence differences, it was not practical to carry out a systematic analysis of their influence on an individual basis, but pilot-scale 293T cell expression experiments showed that incorporating several B41 residues into the CZA97.012 sequence increased the trimer yield by severalfold. Taking the in silico analyses and pilot-scale studies into account, we elected to introduce a total of 26 sequence changes simultaneously into the CZA97.012 SOSIP.664-M6 sequence to make the SOSIP.664-M6.IT construct. The locations of these changes are as follows: V31A, G32A, N33K, M34K, T46K, K49E, T58A, R63T, and V65K in the gp120-C1 domain; G429R, R432Q, I489V, E490K, L491I, K492E, G500A, and E507Q in the gp120-C4 and -C5 domains; and Q543N, T578A, K588R, S618T, Q619I, T620N, N624D, E630Q, and T651N in gp41 regions interactive with gp120 (Fig. 2A and C). Among these 26 changes, K49E, V65K, E106T, I165L, G429R, and R432Q are derived from the previously described TD8 substitutions (40); the I165L substitution is one of the M6 changes that collectively enhance PGT145 binding (Table 2 and Fig. 2); and the Q543N substitution is a component of the SOSIP.v4 design improvements (32). We designate this set of 26 changes IT, to reflect the improved trimerization of the resulting construct.

This new construct and earlier intermediates were transiently expressed in 293T cells, and the Env proteins in the supernatant were studied by BN-PAGE followed by Western blotting with 2G12. Compared to SOSIP.664 and several other comparator proteins, the new SOSIP.664-M6.IT construct produced far fewer monomers, dimers, and aggregates, and the trimer content was proportionately increased (Fig. 4A). The SOSIP.664-M6.IT construct was then expressed in 293F cells, and the Env proteins were purified using a PGT145 column, which yielded almost completely pure trimers (Fig. 4B). The outcomes were 3- to 4-fold and ∼10-fold increases in the yields of fully NL trimers, compared to those of SOSIP.664-M6 and SOSIP.664, respectively (Table 1). In an additional analysis, when the same Env proteins were purified using a 2G12 column, BN-PAGE analysis showed that the SOSIP.664-M6.IT construct yielded a higher percentage of trimers than did SOSIP.664 or SOSIP.664-M6 (Fig. 4B). Thus, residues at or near the gp120-gp41_ECTO interface indeed modulate the efficiency of soluble-trimer formation, but the most beneficial ones may need to be identified on a case-by-case basis.

FIG 4.

The CZA97.012 SOSIP.664-M6.IT construct forms trimers efficiently. (A) The indicated CZA97.012 SOSIP constructs (all containing a C-terminal His tag) were transiently expressed in 293F cells. The Env proteins present in the culture supernatants were fractionated on a BN-PAGE gel, and a Western blot of the gel was then probed with the 2G12 bNAb. In the rightmost lane, SOSIP.v4.2-M6.IT proteins derived from the stable CHO cell line were purified on a PGT145 affinity column and similarly analyzed. (B) The CZA97.012 SOSIP.664, SOSIP.664-M6, and SOSIP.664-M6.IT Env proteins, expressed in 293F cells, were isolated on a 2G12 affinity column and then analyzed on a BN-PAGE gel. The SOSIP.664-M6.IT trimers were also purified using a PGT145 column. In both panels, the bands corresponding to various Env forms are indicated, with the left lane of panel B showing molecular mass marker proteins (M).

SOSIP.v4.2 changes further improve CZA97.012 trimers.

We used an ELISA to gain initial insights into the antigenicity of the CZA97.012 SOSIP.664-M6.IT trimers (Table 3). Like the similarly purified SOSIP.664 parental trimers, PGT151/SEC-purified SOSIP.664-M6.IT trimers bound efficiently to all the tested bNAbs but not to any of the non-NAbs (Table 3). However, sCD4 was able to induce conformational changes in the SOSIP.664-M6.IT trimers that expose the CD4-induced (CD4i) epitope for the 17b non-NAb (Table 3). We have previously shown that the exposure of V3 and CD4i non-NAb epitopes on SOSIP.664 trimers can be reduced via targeted sequence changes in C1 and V3 that create the SOSIP.v4 constructs (33). We therefore introduced the H66R and T316W substitutions to create the CZA97.012 SOSIP.v4.2-M6.IT construct (Fig. 2A). These modifications had the desired effect on antigenicity; compared to SOSIP.664-M6.IT trimers, the binding of 14e and the ability of sCD4 to induce the 17b epitope were reduced for SOSIP.v4.2-M6.IT (Table 3). We also found that SOSIP.v4.2-M6.IT trimers were antigenically comparable whether purified via PGT145 or PGT151 columns (Table 3). The yields of the PGT145-purified trimers were in the range of 0.4 to 0.6 mg/liter, a substantial increase over that with the parental construct (Table 1).

TABLE 3.

Binding of antibodies to CZA97.012 SOSIP trimers in ELISA^a

Antibody	Binding to CZA97.012 trimers
	SOSIP.664	SOSIP.664-M6.IT	SOSIP.v4.2-M6.IT	SOSIP.v4.2-M6.IT (PGT145/SEC)
2G12	+	++	++	++
PG16	−	+++	+++	+++
PGT145	−	+++	+++	+++
PGT151	+++	+++	+++	+++
35022	+++	ND	ND	+++
3BC315	+	ND	ND	++
PGT122	+++	+++	+++	+++
PGT128	+++	ND	ND	+++
VRC01	+++	+++	+++	+++
CD4-IgG2	++	ND	ND	++
b6	−	ND	ND	−
F105	−	−	−	−
14e	+	+	−	−
17b	−	−	−	−
sCD4 + 17b	++	++	+	+

^aThe various CZA97.012 SOSIP trimers (all His tagged) were purified by the PGT151/SEC method except when indicated (right column, SOSIP.v4.2-M6.IT purified by PGT145/SEC). The binding of NAbs (highlighted in boldface type) or non-NAbs was assessed in an ELISA. The magnitude of antibody reactivity is recorded semiquantitatively on a scale from − (no binding) to +++ (strong binding).

We next used SPR to obtain more detailed antigenicity data (Fig. 5). Compared to the parental SOSIP.664 trimer, the SOSIP.v4.2-M6.IT variant bound better to the 2G12, PG16, PGT145, 35O22, and 3BC315 bNAbs, in some cases markedly so, while there was no difference in PGT122, PGT128, and PGT151 reactivity. For PGT145, the rate of dissociation from the SOSIP.v4.2-M6.IT trimer was initially lower than that from SOSIP.664 (see additional studies described below). There was modestly slower binding of CD4-IgG2 to the new variant than to SOSIP.664, and the extent of VRC01 binding was also reduced, which are effects attributable to the sequence changes that convert SOSIP.664 trimers to SOSIP.v4.2 (33). Of the non-bNAbs to V3 epitopes, there was greater binding of 14e to SOSIP.664 than to SOSIP.v4.2-M6.IT, while the converse was true for 39F and 19b. However, in each case, the extent of 14e, 39F, and 19b binding was very low compared to those of the various bNAbs. Similarly, there was only minimal to undetectable binding of the 17b non-NAb to its CD4i epitope on either trimer. Finally, the CD4bs non-NAb b6 did not bind significantly to either trimer (Fig. 5).

FIG 5.

Antigenicity of CZA97.012 SOSIP.v4.2-M6.IT trimers. The CZA97.012 SOSIP.664 and SOSIP.v4.2-M6.IT His-tagged trimers, expressed in 293F cells, were purified by the PGT151/SEC and PGT145/SEC methods, respectively. SPR was used to assess the binding of the indicated NAbs and non-NAbs to the immobilized trimers. The response difference (RU) is given on the y axis as a function of time (seconds) on the x axis. Note that the scales on the y axes are not identical, with top values of 200, 400, or 600 RU on various plots.

Overall, the SPR and ELISA data suggest that all the bNAb epitopes that are present on parental CZA97.012 SOSIP.664 trimers are preserved or improved on the modified version, except for the modest reduction in the exposure of the VRC01 epitope. The presentation of non-NAb epitopes remains low or undetectable.

Increased yield of CZA97.012 SOSIP.v4.2-M6.IT trimers from a stable CHO cell line.

The multiply modified CZA97.012 SOSIP.v4.2-M6.IT trimer can be produced by transient transfection and purified via a PGT145 column, is fully native-like, and has an appropriate antigenicity profile. However, the trimer yield of ∼0.5 mg/ml is still at the low end of the range that we consider to be practically useful. Our general experience based on making BG505 and B41 stable CHO cell lines is that this method of trimer production can increase yields by severalfold over those achieved by transient transfection (26). Accordingly, we made a stable line expressing the CZA97.012 SOSIP.v4.2-M6.IT construct and, for comparison, one expressing the SOSIP.664 construct, using our established method (26). Unlike the trimers that were produced by transient transfection, the CHO cell line-expressed versions did not include a His tag. The stable cell line produced fully cleaved SOSIP.v4.2-M6.IT trimers that could be purified via either PGT145/SEC or the conformationally less selective 2G12/SEC method at yields that were in the range of 1.5 to 2.0 mg/liter or 1.0 to 1.5 mg/liter, respectively (Table 1 and Fig. 6A to D).

FIG 6.

Expression of CZA97.012 SOSIP.v4.2-M6.IT trimers in a stable CHO cell line. (A) SEC profiles of Env proteins isolated using a PGT145 (left) or a 2G12 (right) affinity column, with the trimer peaks indicated and shaded in green. BN-PAGE analysis of various fractions from the trimer peak is shown below each profile. (B) BN-PAGE gel analysis of the Env proteins purified by the indicated bNAb affinity column. The migration positions of molecular marker proteins (M) are shown in the first lane. (C) Trimers purified by the indicated bNAb affinity column followed by SEC were analyzed on SDS-PAGE gels under nonreducing (without DTT [−]) and reducing (with DTT [+]) conditions, with molecular mass marker proteins in the first lane. The arrows indicate the gp140 (− DTT lanes) and gp120 (+ DTT lanes) bands. (D) Yields of 2G12/SEC- or PGT145/SEC-purified SOSIP.664 trimers produced by transient transfection of 293F cells and of SOSIP.v4.2-M6.IT trimers expressed by the same method or in the stable CHO cell line.

We then used SPR to compare the CHO cell line-expressed PGT145/SEC-purified SOSIP.v4.2-M6.IT trimers and PGT151/SEC-purified SOSIP.664 trimers. The two trimers bound 2G12 and PGT128 to comparable extents, but only SOSIP.v4.2-M6.IT bound strongly to PGT145 (Fig. 7A). Note that in this SPR format (i.e., trimer as a functionally monovalent analyte, injected at 10 nM), the minimal binding of PGT145 to the SOSIP.664 trimer is compatible with the low-affinity interaction that yielded the much larger response seen when bivalent PGT145 IgG was injected at 500 nM to His-tagged immobilized trimers (Fig. 5) (47). As expected, the V3 non-NAb 14e bound the SOSIP.664 trimer but not the engineered variant (Fig. 7A). We also compared PGT145/SEC-purified SOSIP.v4.2-M6.IT trimers produced either in the CHO cell line or by transient transfection of 293F cells, to assess whether any subtle differences in glycosylation might affect reactivity with bNAbs PG16 and PGT145 against V2-glycan-dependent epitopes. Both bNAbs bound both versions of the trimer, with the CHO cell line-derived version being the more antigenic of the two (Fig. 7B). Hence, there are no indications that SOSIP trimer expression in CHO cells is problematic from the antigenicity perspective. This assessment is consistent with reports that producer cell variation has only a subtle influence on the glycosylation of native-like trimers (48–50).

FIG 7.

Properties of CHO cell line-expressed CZA97.012 SOSIP.v4.2-M6.IT trimers. (A) SPR analysis of PGT151/SEC-purified SOSIP.664 and PGT145/SEC-purified SOSIP.v4.2-M6.IT trimers derived from the stable CHO cell lines. The sensorgrams show the binding of the trimers, added at 10 nM, to immobilized bNAbs 2G12, PGT128, and PGT145 and non-NAb 14e. The response difference (RU) is given on the y axis as a function of time (seconds) on the x axis. (B) SPR analysis of PGT145/SEC-purified SOSIP.v4.2-M6.IT trimers derived from 293F or stable CHO cells. The sensorgrams show the binding of the trimers, added at 10 nM or 30 nM, to immobilized bNAb PG16 or PGT145. (C) Analysis of sera from five rabbits immunized with 293F cell-expressed, PGT151/SEC-purified CZA97.012 SOSIP.664 trimers (3). The sera were titrated against immobilized His-tagged trimers in an ELISA, and the signals derived at a 1:100 dilution are plotted. The ELISA antigens were CZA97.012 SOSIP.664 or SOSIP.v4.2-M6.IT trimers, which were expressed in 293F cells and purified by the PGT151/SEC and PGT145/SEC methods, respectively. The sera also bound comparably to the two trimers at other dilutions from 1:300 to 1:10,000 (not plotted). Similar data were obtained using sera from 5 rabbits immunized with only CZA97.012 SOSIP.664 trimers in an ongoing experiment (not plotted). (D) EM images of SOSIP.v4.2-M6.IT trimers produced in 293F or CHO cells and purified as indicated. (E) Thermostability profiles, measured by DSC, of SOSIP.v4.2-M6.IT trimers purified from CHO cells by the indicated method.

We also assessed whether rabbit sera raised against the CZA97.012 SOSIP.664 trimer would still recognize the modified version of the trimer in an ELISA. The 5 sera were derived from an immunization experiment in which the CZA97.012 trimers were used to boost animals that had previously received the other genotype of trimer; 3 of the 5 animals developed NAbs that neutralized autologous CZA97.012 (3). The sera bound to both trimers in an ELISA roughly in proportion to their neutralizing capacity although somewhat less well to the SOSIP.v4.2-M6.IT trimers than to the SOSIP.664 immunogen (Fig. 7C). The reduced reactivity probably reflects the decreased exposure of non-NAb epitopes on the V3 region of the modified trimers (see, for example, Fig. 7A).

We confirmed that the PGT145-purified SOSIP.v4.2-M6.IT trimers, whether produced transiently in 293F cells or from the stable CHO cell line, were fully NL when viewed by NS-EM (Fig. 7D). Finally, the 2G12/SEC- or PGT145/SEC-purified, CHO cell line-derived SOSIP.v4.2-M6.IT trimers had similar melting temperatures (T_m) of 62.3°C and 61.9°C, respectively, as determined by differential scanning calorimetry (DSC) (Fig. 7E). Overall, we conclude that this line produces an acceptable amount of antigenically appropriate, thermostable, and native-like trimers based on the CZA97.012 sequence.

Location of the sequence changes to the SOSIP.v4.2-M6.IT trimer.

When improving the CZA97.012 trimers, we introduced 26 amino acid changes into the gp120-gp41_ECTO interface to produce SOSIP.664-M6.IT and then the SOSIP.v4.2-M6.IT construct. In all probability, not all of these 26 IT changes will be beneficial to trimer formation, and some may even be detrimental. A very large number of mutants would need to be made and analyzed to obtain more specific information, and we have not pursued that option. However, we studied the locations of the 26 IT sequence changes and the positions of the M6 changes that restore PGT145 binding on the BG505 trimer structure (no corresponding CZA97 structure is yet available) (Fig. 8).

FIG 8.

Locations of the M6 and IT sequence changes on the BG505 trimer. Mutated residues are shown as spheres using the crystal structure of BG505 SOSIP.664 in complex with bNAbs (PDB accession number 5CEZ; antibodies were removed for clarity) (46). The M6 changes, best studied in the top view, are represented by green, red, and yellow spheres, while the IT changes (see the side view) are shown as blue and purple spheres. SOSIP.664.v4.2 mutations are shown in cyan and orange. Figures were made using PyMOL (The PyMOL Molecular Graphics System, version 1.8; Schrödinger, LLC).

The M6 changes are all located at the trimer apex, which is consistent with their influence on the PGT145 epitope in this area (Fig. 8, top and side views). The locations of the H66R and T316W sequence changes incorporated into the SOSIP.v4.2 trimer design are also shown (33). The T316W change in V3 may also affect the trimer apex and its associated PGT145 epitope (Fig. 8). The IT changes are clustered near the gp120-gp41_ECTO interface, where they are positioned in a way that may strengthen the association between these subunits of the trimer (Fig. 8, side view). Any such stabilization effects would be expected to increase the yield of appropriately folded trimers. Whether this outcome is attributable to an increase in trimer formation per se (i.e., at the expense of dimers and monomers) or whether there are also (or instead) beneficial effects on the overall folding and assembly of the gp140 protein remains to be determined experimentally. Most likely, both factors are relevant.

Influences on the PGT145 epitope on CZA97.012 SOSIP trimers.

The restoration of the highly conformationally sensitive PGT145 epitope on the CZA97.012 SOSIP trimer background enabled us to investigate what sequence changes in V2 and V3 most influenced its presentation. Accordingly, we purified SOSIP.664, SOSIP.v4.2-M6.IT, and four intermediate trimers via PGT151 columns and used SPR to compare their reactivities with a PGT145 Fab. A conformational-change model was fitted to binding data derived from Fab titrations (Fig. 9). The Fab bound approximately equally well to the SOSIP.664, SOSIP.v4.2, and SOSIP-M307I trimers [K_D(conf) = 36 to 67 nM] but had a higher affinity for SOSIP.v4.2-M6.IT [K_D(conf) = 2.6 nM]. The corresponding K_D(conf) values for SOSIP.v4.2-M5 and SOSIP.v4.2-M6 were intermediate, at 16 to 18 nM. Thus, among the CZA97.012 trimers that were compared, SOSIP.v4.2 had the lowest affinity for Fab PGT145, which is likely to be caused by the reduced conformational flexibility conferred on the trimer apex by the H66R and T316W changes that create the SOSIP.v4.2 construct (33). Overall, the stronger PGT145 binding to the SOSIP.v4.2-M6.IT trimer reflects favorable although subtle improvements in all four kinetic constants compared to the other trimers in the study. The stoichiometric values were generally similar for the comparator trimers (S_m, 0.45 to 0.51), with the exception of SOSIP.664-M6, for which the S_m was somewhat higher at 0.65. For comparison, the stoichiometry of PGT145 binding to the BG505 SOSIP.664 trimer has been reported to be in the range of 0.78 to 0.88 (47).

FIG 9.

PGT145 Fab binding kinetics for CZA97.012 SOSIP trimer variants. PGT145 Fab binding to the listed SOSIP trimer variants was analyzed by SPR. The response difference (RU) is given on the y axis as a function of time (seconds) on the x axis. A conformational-change model was fitted to the data. The constants in the table are defined in the legend to Fig. 3. Means and standard errors of the means are given with two and one significant figures, respectively.

Improving SOSIP.664 trimers of other genotypes.

We reported that SOSIP.664 trimers of the 92UG037.8 (clade A) genotype also yield a mixture of NL trimers and pseudotrimers when purified by 2G12/SEC (22). The DU422 clade C SOSIP.664 trimers are fully in the NL form after 2G12/SEC purification, but their yield is low (0.2 mg/ml) compared to that of BG505 or B41, and they cannot be purified efficiently on PGT145 columns (51). In preliminary experiments, we found that another clade C SOSIP.664 construct, 426c, also produces PGT145-purifiable NL trimers inefficiently (Fig. 10A). The limitations of the SOSIP.664 design for these genotypes generally involve the poor formation or stability of NL trimers, which compromises the resulting yield. We therefore assessed whether what we had learned when creating the CZA97.012 SOSIP.v4.2-M6.IT construct could be applied to improving the yield of fully NL trimers based on the above-described three env genes. Specifically, we made the three SOSIP.v4-M6.IT constructs (see Materials and Methods for sequence changes) and used a PGT145 column to purify NL trimers from 293F cell transfection supernatants. In each case, the outcome was a marked increase in trimer yield for the SOSIP.v4-M6.IT constructs compared to that of SOSIP.664, by 3-, 8-, and 2.5-fold for 426c, DU422, and 92UG037.8, respectively (Fig. 10A). All three genotypes of the modified trimers were predominantly NL when viewed by EM (Fig. 10B). We have not yet made stable CHO cell lines producing these three trimers, a step that may further increase yields from the present levels of 0.2 to 0.3 mg/liter. We also have not yet tested this trimer improvement method with additional genotypes.

FIG 10.

SOSIP.v4-M6.IT trimers of various genotypes. (A) Yields of SOSIP.664 and SOSIP.v4-M6.IT trimers (SOSIP.v4.1-M6.IT for 92UG037.8 and SOSIP.v4.2-M6.IT for DU422 and 426c). Trimers derived from transiently transfected 293F cells and purified via a PGT145 affinity column are shown for the indicated genotypes. (B) The purified SOSIP.v4-M6.IT trimers were imaged by NS-EM, with the percentage in the NL form indicated below each panel.

DISCUSSION

Native-like SOSIP trimers are now widely used for structural studies, and they represent a platform for immunogen design improvements (11, 52, 53). However, many env genotypes do not yield SOSIP trimers efficiently for various reasons, including low expression and/or instability (11, 22, 51). Thus, in some cases, the trimers can aggregate or dissociate into dimers and monomers, or they may adopt unwanted, nonnative configurations (54). The CZA97.012 clade C genotype exemplifies several problems that can be encountered. Thus, although we could produce SOSIP.664 trimers from this sequence, only ∼50% of them adopted an appropriate NL configuration; the remaining trimers were irregular and splayed out and resembled uncleaved CZA97.012 gp140_UNC-Fd-His proteins of an earlier design (4, 22, 25). The NL SOSIP.664 trimer subset could be positively selected using a PGT151 bNAb affinity column although with only a low overall yield. These purified CZA97.012 SOSIP.664 NL trimers were immunogenic in rabbits, inducing autologous tier 2 NAbs to the corresponding tier 2 virus (3). In contrast, the earlier-generation CZA97.012 uncleaved gp140 pseudotrimers lacked this property when tested as immunogens (4, 25).

We have now modified the CZA97.012 SOSIP.664 sequence to increase the expression of stable, fully NL trimers that now display the quaternary structure-dependent apex epitope for the PGT145 bNAb. Together, the various modifications overcome some of the deficiencies associated with the SOSIP.664 version of this genotype (22). The presence of the PGT145 epitope at the apex of a soluble trimer may be a good indicator of native-like structure and facilitates an alternative trimer purification strategy via a PGT145 affinity column. The first part of our optimization strategy involved the predicted core of the PGT145 epitope in the strand C region of the V2 loop. However, changes here were not sufficient for PGT145 to bind strongly to the resulting trimers. Various other substitutions of relatively uncommon residues in the CZA97.012 V2 sequence were also not helpful in this regard. In contrast, the M307I substitution in V3, when made together with the I165L, Q171K, G172V, ΔN184, and I192R sequence changes in V2, allowed the creation of the NL SOSIP.v4.2-M6.IT trimer that bound strongly to PGT145. How the presence of an Ile residue at V3 position 307 improves the binding of a trimer apex-specific bNAb to such an extent remains to be determined. One possibility is that the change may increase the conformational mobility of the apex in a way that better accommodates the CDR3 loops of PGT145 and other bNAbs that recognize this region of the trimer, which can be highly flexible (55).

Although several high-resolution Env structures, including those of BG505 SOSIP.664 trimers in complex with PGT145, are now available, the mechanism underlying the subtle breathing of the trimer apex remains to be fully elucidated. We show here that most single point substitutions in V2 have little or no impact on PGT145 binding, but when introduced in combination, much more substantial effects can be seen. These influences on PGT145 reactivity arise even though the various sequence changes are located outside the epitope itself; indeed, only the changes at residues 171 and 172 could be considered to be close to it (62). Each sequence change may make a minor contribution to the overall presentation of the PGT145 epitope, reinforcing how the antigenic profile of the conformationally flexible trimer can be altered in subtle and unexpected ways. On a more pragmatic level, our improvement of the PGT145 epitope now allows the use of this bNAb as an affinity column that is completely selective for NL CZA97.012 SOSIP trimers. The same methodology may be applicable to other env genes.

Some, and perhaps most, of the instability problems associated with the inefficient formation of certain SOSIP.664 trimers are rooted in the interface between the gp120 and gp41_ECTO subunits. In this context, eight sequence changes (the TD8 strategy) at the gp120-gp41_ECTO interface have been identified as a method for improving trimers of some genotypes (40). We found that either the TD8 substitutions did not work in the context of the CZA97.012 sequence or their influence was insufficient to overcome other instabilities in the resulting trimers. We have also observed that the TD8 changes were insufficient to improve SOSIP.664 trimers based on the clade C genotypes 426c, ZM331, ZM3678, ZM4248, ZM3827, and ZM282TF but were useful when applied to the clade C ZM3618 construct. Guided by the BG505 SOSIP.664 trimer structure, we therefore identified additional residues in the C1, C4, and C5 domains of gp120 and in gp41_ECTO that differed between the CZA97.012 and the BG505 and/or B41 sequences and that plausibly affected the interface. As the BG505 and B41 genotypes of the SOSIP.664 trimer are fully native-like and express trimers efficiently, we made a total of 26 sequence changes to the gp120-gp41_ECTO interface of the CZA97.012 SOSIP.664-M6 trimer that introduced BG505 and/or B41 residues. The outcome was an improved yield of trimers at the expense of unwanted dimers and monomers. The resulting sequence-modified SOSIP.664-M6.IT trimers could be purified in a fully native-like form not just via the PGT145 column but also via the less conformationally selective 2G12/SEC method. It is possible, and indeed likely, that not all of the 26 IT changes that we introduced into the CZA97.012 sequence are essential, but we note that when the same changes were made in the 426c, DU422, and 92UG037 SOSIP.664 contexts, there was a consistent improvement in the yields of NL trimers. Different genotypes may, however, need different solutions on a case-by-case basis.

We have not explored the individual impact of all 26 IT sequence changes. To do so would require 26 new trimer mutants to be made and analyzed (52 for a more thorough, bidirectional analysis). In our opinion, the most likely outcome would be that no single sequence change has a dramatic effect and perhaps not even enough of an impact to quantify with any precision. More likely is that multiple sequence changes work in concert, but identifying which ones (by making double, triple, and quadruple mutants, etc.) would expand the work scope to an impractical extent. We note that the description of the TD8 mutations (n = 8) did not include an investigation of the impact of all the individual changes (40).

Together, the V2/V3 and gp120-gp41_ECTO interface modifications created the CZA97.012 SOSIP.664-M6.IT trimer, which has desirable antigenicity properties with respect to the presentation of most bNAb epitopes. The further introduction of the v4.2 changes (H66R and A316W) to create the SOSIP.v4.2-M6.IT variant drove additional stability and antigenicity improvements, consistent with our findings on other trimer genotypes (33). We noted that modifications to the CZA97.012 sequence that improved PGT145 reactivity also increased the binding of the 3BC315 bNAb to its gp41_ECTO epitope. This epitope becomes more accessible when trimers undergo the transient conformational transitions often now referred to as breathing (56, 57). Clearly, the efficiency of trimer formation depends heavily on how efficiently gp41_ECTO associates with gp120 elements from the C1 and C5 regions. These interactions initiate and orchestrate the trimerization process, but the stability of the trimer as a whole is also influenced by other interactions that include ones involving the apex. A more precise understanding of these stabilizing interactions may be useful for improving trimer stability in general and particularly in the context of germ line bNAb-targeting trimers that may need conformational adjustments to confer the desired antigenicity profile (30, 58, 59).

We were able to further increase the expression of fully NL CZA97.012 SOSIP.v4.2-M6.IT trimers by making a stable CHO cell line, a technique that we have found to be generally valuable from this perspective. A stable line is, of course, also highly useful for producing much larger amounts of trimer more conveniently than can be achieved by transient transfection (26). By various criteria, therefore, these new clade C trimers are markedly superior to the original SOSIP.664 version, which was in turn a substantial improvement over the nonnative uncleaved gp140_UNC-Fd Env protein of this genotype that has been studied extensively in animals and humans (4, 25). In a recent report, low-resolution cryo-EM images of a construct expressing the latter uncleaved CZA97.012 gp140 protein as a fusion protein with a large, trimeric scaffold at the C terminus were interpreted as showing that the gp140 component had characteristics of native-like trimers (60). In our opinion, those images were not of a sufficient quality to support that assessment. There was no additional information on the characteristics of the soluble nonscaffolded, nonnative gp140 protein that was the basis of the improvements described here and in our previous report (22). Whether the now fully native CZA97.012 SOSIP.v4.2-M6.IT trimers are worth pursuing as an immunogen is something to now consider, noting that the earlier SOSIP.664 trimers of this genotype induce autologous, tier 2 NAbs in rabbits (3).

MATERIALS AND METHODS

Env construct design.

We made the CZA97.012 (clade C), 92UG037.8 (clade A), 426c (clade C), and DU422 (clade C) SOSIP.664 constructs by cloning the corresponding env genes into the pPPI4 expression vector and then making the following changes, as described previously (22, 51). The natural leader sequence was replaced with the tissue plasminogen activator leader; the A501C, T605C, and I559P substitutions were added; the REKR cleavage site at residues 508 to 511 was replaced with the RRRRRR sequence, and gp41_ECTO was terminated at residue 664. The 92UG037.8 construct also contains an I535M substitution, and the CZA97.012 construct contains the L535M and Q567K changes, which we found to improve trimer formation and expression (22, 61).

The SOSIP.664 design can be improved by introducing point substitutions that create the SOSIP.4.1 (E64K plus T316W) or SOSIP.v4.2 (H66R plus T316W) constructs. Here, we used the SOSIP.v4.2 changes with clade C trimers and SOSIP.v4.1 for trimers of other genotypes, based on our previous report (32). The sequence changes used to make the CZA97.012 SOSIP.664-M6.IT and then the SOSIP.v4.2-M6.IT constructs are listed in Results. The SOSIP.v4-M6.IT constructs of the other genotypes incorporate the following sequence changes to the SOSIP.664 background: G32A, N33K, M34K, E64K, V65K, T316W, R490K, K500A, E507Q, S618T, E619I, R620N, E624D, and K651N for 92UG037.8 SOSIP.v4.1-M6.IT; N31A, L32A, D33K, L34K, K49E, K63T, V65K, H66R, I192R, T316W, E429R, R432Q, E490K, K492E, K500A, G507Q, L519F, Q543N, T578A, K588R, S618T, L619I, and G620N for DU422 SOSIP.v4.2-M6.IT; and V31A, G32A, N33K, L34K, K49E, K63T, V65K, H66R, E106T, D160N, I165L, T169K, R170Q, ΔN184, T316W, E429R, K432Q, D448N, D500A, E507Q, Q543N, T578A, K588R, S618T, K619I, E620N, and E624D for 426c SOSIP.v4.2-M6.IT.

Unless otherwise specified, a single glycine and then a 6-histidine (His) tag were added to all of the various constructs immediately after residue 664 in gp41_ECTO, to facilitate enzyme-linked immunosorbent assay (ELISA) and surface plasmon resonance (SPR) assay procedures (47). From here on, we refer to the presence of the His tag only when it is relevant to understanding the experiment.

Site-directed mutagenesis and chimeric clones.

Point substitutions in the env genes described above were made using the QuikChange site-directed mutagenesis kit (Agilent Technologies). Briefly, a primer pair was designed for each mutation or two adjacent mutations that could be accommodated in a single primer without compromising its melting temperature. Two overlapping primer pairs were used to generate the chimeric clones: one pair for amplifying the vector backbone of CZA97.012 SOSIP.664 expressing the recipient plasmid and the second for amplifying a short fragment from an env donor plasmid (BG505, 92UG037.8, CAP45, or DU422). These two PCR fragments, with homologous sequences at both ends, were joined using an In-Fusion cloning kit (CloneTech). The chimeric clones grown from the colonies on LB plates were verified by sequencing.

Env protein expression and trimer purification.

Transient transfection of 293F cells in serum-free medium was performed essentially as previously described (22, 41). Env proteins were purified from the resulting culture supernatants using affinity columns containing GNA lectin or bNAb 2G12, PGT145, or PGT151 (see Results), which were made by using CNBr-activated Sepharose 4B resin (GE Healthcare) as previously described (6, 22, 41). In each case, the culture supernatant was flowed through the column at a constant rate of 1 ml/min, the beads were washed with buffer (20 mM Tris-HCl, 500 mM NaCl [pH 8]), and Env proteins were eluted with 3 M MgCl₂ (pH 7.2). The eluted proteins were immediately buffer exchanged into a solution containing 10 mM Tris-HCl and 75 mM NaCl (pH 8) and concentrated using a 100-kDa-cutoff Vivaspin column (GE Healthcare). A Superdex 200 26/60 size exclusion chromatography (SEC) column and the same buffer were then used to isolated trimer fractions, which were pooled, concentrated, and stored at −80°C.

To produce culture supernatants containing unpurified SOSIP.664 Env proteins for ELISA or Western blot assessments (see Results), the transfection procedures involved 293T cells, and the culture medium contained 5% fetal bovine serum (FBS).

Production of stable CHO cell lines.

We made stable CHO cell lines expressing the untagged CZA97.012 SOSIP.664 or SOSIP.v4.2-M6.IT construct using an established procedure (26). Briefly, the env gene construct was cloned into an expression plasmid containing an Flp recombination target (FRT) site linked to the hygromycin resistance gene. This procedure allows Flp recombinase-mediated integration to occur and facilitates the selection of a stable cell line expressing the Env protein under the control of the human cytomegalovirus (CMV) promoter; the furin protease gene is included in the same expression vector under the control of the EFI Alpha promoter. Flp-In CHO cells (Invitrogen, Carlsbad, CA) were then cotransfected with the resulting pAM/C CZA97.012 SOSIP.664 or SOSIP.v4.2-M6.IT plasmid and the pOG44 recombinase (Invitrogen) using Lipofectamine 2000 (Invitrogen) according to the manufacturer's instructions. The transfected cells were cultured in the selection medium until hygromycin-resistant foci were observed. Cells were harvested from these foci, transferred to a new 48-well plate, and expanded in the presence of hygromycin. The clones that expressed the highest levels of Env protein, as assessed by dot blotting using bNAb 2G12, were then selected and expanded.

SDS-PAGE, BN-PAGE, and Western blotting.

Env proteins purified using an affinity column followed by SEC were analyzed using a 4 to 12% Bis-Tris NuPAGE SDS-PAGE gel (Invitrogen) followed by Coomassie blue staining. To assess cleavage, the proteins were incubated for 10 min at room temperature with 0.1 mM dithiothreitol (DTT) before loading onto SDS-PAGE gels. Env protein-containing cell culture supernatants, affinity-purified Env proteins, or fractions derived from SEC columns were analyzed using a BN-PAGE gel system (Invitrogen). After Coomassie blue staining, the proteins were transferred to a nitrocellulose membrane (Invitrogen), at 32 to 40 V for 1 to 1.5 h. The membranes were blocked with 10% goat serum for 1 h at room temperature and incubated for 3 h with bNAb 2G12. The membrane was washed 3 times with Tris-buffered saline (TBS) before the addition of a 1:5,000-diluted solution of horseradish peroxidase (HRP)-conjugated goat anti-human IgG followed by the Western Lightning plus-ECL enhanced luminol reagent (PerkinElmer, Waltham, MA).

Antibodies.

The following monoclonal antibodies (MAbs) were obtained as gifts, or purchased from, the indicated sources: VRC01 and F105 from John Mascola; PG16, PGT122, PGT128, PGT145, PGT151, and b6 from the International AIDS Vaccine Initiative; 2G12 from Polymun Scientific (Vienna, Austria); 3BC315 from Michel Nussenzweig; 17b, 14e, 19b, and 39F from James Robinson; and 35O22 from Mark Connors.

Surface plasmon resonance assays.

SPR assays were carried out as described previously (47). Briefly, purified His-tagged trimers were captured onto CM5 chips by an anti-His MAb (GE Healthcare) to immobilization levels (R_L values) of 500 RU (response units). The CD4-IgG2 protein and all the MAbs were then injected at a concentration of 500 nM. The flow rate was 50 μl/min throughout the MAb association and dissociation phases of 300 s and 600 s, respectively, except for the 17b and CD4-IgG2 injections, for which the association-monitoring period was 200 s and the flow rate was reduced to 30 μl/min. After each cycle of MAb association and dissociation, the anti-His surface was regenerated by injecting a single pulse of 10 mM glycine (pH 2.0) for 60 s at a flow rate of 30 μl/min. We also compared the binding of the CZA97.012 SOSIP.664 and SOSIP.v4.2-M6.IT trimers, produced in the stable CHO cell line, and the CZA97.012 SOSIP.v4.2-M6.IT trimer (His tagged although used here as an analyte), produced by transient transfection of 293F cells, to immobilized Env-specific MAbs. In this approach, anti-human IgG (Fc) antibody was conjugated to CM5 chips for the capture of the Env-specific MAbs to immobilization levels of 300 RU. The trimers, as analytes, were then injected at a concentration of 10 nM or 30 nM at a flow rate of 50 μl/min. Background binding in control channels (anti-Fc) amounted to <5% of the binding to the MAbs. In both assay formats, control channel and zero-analyte signals were subtracted, giving the response difference (RU) (47).

Enzyme-linked immunosorbent assay.

Briefly, purified trimers were captured via their His tags onto preblocked Ni-nitrilotriacetic acid (NTA) 96-well plates (Qiagen) by incubation for 2 h at 0.5 μg/ml in TBS containing 5% FBS. Culture supernatants from transfected 293T cells, which contain 5% FBS, were diluted 1:4 in TBS before addition to the ELISA wells. After the capture stage and washing, test MAbs, CD4-based reagents, or rabbit sera were added for 1 h in the same buffer. Bound MAbs were detected using an appropriate HRP-conjugated secondary antibody and the 3,3′,5,5′-tetramethylbenzidine (TMB) substrate system with an optical density endpoint at 450 nm (Bio-Rad). The 50% binding values (50% effective concentrations [EC₅₀]) for MAb binding were calculated by plotting the nonlinear regression curves using Prism software, version 5.0 (22). The rabbit sera used were derived from a previously published study approved by the Institutional Animal Care and Use Committee (IACUC).

Differential scanning calorimetry.

The thermal denaturation of purified trimers, previously dialyzed into phosphate-buffered saline (PBS) and added to the test cell at a concentration of 250 μg/ml, was probed using a MicroCal VP-Capillary DSC calorimeter (Malvern Instruments) as described previously (6). The scan rate for the temperature increase was 60°C/h. Buffer correction, normalization, and baseline subtraction procedures were applied before the data were analyzed using Origin 7.0 software. The resulting data were fitted using a non-two-state model, as the asymmetry of some of the thermal denaturation peaks suggested that unfolding intermediates were present.

Negative-stain electron microscopy.

Purified trimers were prepared for negative-stain EM analysis as previously described (6, 33, 41). Data were collected either on an FEI Tecnai T12 electron microscope operating at 120 keV, with an electron dose of ∼25 e⁻/Ų and a magnification of 52,000× that resulted in a pixel size of 2.05 Å at the specimen plane, or on an FEI Talos electron microscope operating at 200 keV, with an electron dose of ∼25 e⁻/Ų and a magnification of 73,000× that resulted in a pixel size of 1.98 Å at the specimen plane. Images were acquired with a Tietz TemCam-F416 CMOS camera (FEI Tecnai T12) or a FEI Ceta 16M camera (FEI Talos) using a nominal defocus range of 1,000 to 1,500 nm.

Data processing methods were adapted from ones used previously (6, 22, 33, 41). The resulting two-dimensional (2D) class averages were visually inspected, and discernible aggregates, dimers, and monomers (based on relative mass/size) were computationally removed before an additional round of 2D classification. The final 2D classes were labeled as having either NL trimer phenotypes similar to those of BG505 or B41 SOSIP.664 (6, 41) or nonnative phenotypes indicative of malformed trimers (54). The NL trimer percentage is defined as the total number of particles in classes labeled as native-like divided by the total number of particles in the final round of 2D classification.