Comparison of Immunogenicity in Rhesus Macaques of Transmitted-Founder, HIV-1 Group M Consensus, and Trivalent Mosaic Envelope Vaccines Formulated as a DNA Prime, NYVAC, and Envelope Protein Boost

ABSTRACT

An effective human immunodeficiency virus type 1 (HIV-1) vaccine must induce protective antibody responses, as well as CD4⁺ and CD8⁺ T cell responses, that can be effective despite extraordinary diversity of HIV-1. The consensus and mosaic immunogens are complete but artificial proteins, computationally designed to elicit immune responses with improved cross-reactive breadth, to attempt to overcome the challenge of global HIV diversity. In this study, we have compared the immunogenicity of a transmitted-founder (T/F) B clade Env (B.1059), a global group M consensus Env (Con-S), and a global trivalent mosaic Env protein in rhesus macaques. These antigens were delivered using a DNA prime-recombinant NYVAC (rNYVAC) vector and Env protein boost vaccination strategy. While Con-S Env was a single sequence, mosaic immunogens were a set of three Envs optimized to include the most common forms of potential T cell epitopes. Both Con-S and mosaic sequences retained common amino acids encompassed by both antibody and T cell epitopes and were central to globally circulating strains. Mosaics and Con-S Envs expressed as full-length proteins bound well to a number of neutralizing antibodies with discontinuous epitopes. Also, both consensus and mosaic immunogens induced significantly higher gamma interferon (IFN-γ) enzyme-linked immunosorbent spot assay (ELISpot) responses than B.1059 immunogen. Immunization with these proteins, particularly Con-S, also induced significantly higher neutralizing antibodies to viruses than B.1059 Env, primarily to tier 1 viruses. Both Con-S and mosaics stimulated more potent CD8-T cell responses against heterologous Envs than did B.1059. Both antibody and cellular data from this study strengthen the concept of using in silico-designed centralized immunogens for global HIV-1 vaccine development strategies.

IMPORTANCE There is an increasing appreciation for the importance of vaccine-induced anti-Env antibody responses for preventing HIV-1 acquisition. This nonhuman primate study demonstrates that in silico-designed global HIV-1 immunogens, designed for a human clinical trial, are capable of eliciting not only T lymphocyte responses but also potent anti-Env antibody responses.

INTRODUCTION

One of the challenges to the development of an effective, global human immunodeficiency virus type 1 (HIV-1) vaccine is the extraordinary diversity of the virus (1–5). Nonhuman primate (NHP) studies suggest that vaccine-induced antibody responses are likely to be essential for blocking an infection, while T cell responses are important for the generation of both CD4⁺ T lymphocyte help as well as CD8⁺ T lymphocyte responses to control viremia in the setting of an infection (2, 6–9). Recent studies by Hansen et al. showing T cell-mediated viral clearance in 50% of the monkeys following a pathogenic simian immunodeficiency virus (SIV) challenge have reestablished the importance of vaccine-induced T cell responses (10). The ultimate goal of an effective HIV-1 vaccine remains to confer protection against the majority of circulating HIV-1 strains by eliciting either humoral or cell-mediated immune responses or both (11–13).

In the Merck STEP clinical trial, an adenovirus 5 (Ad5)-based single-clade HIV-1 Gag/Pol/Nef vaccine induced a median of 2 epitope-specific T cell responses in each vaccinee, with 40% of the vaccinees responding to Gag and only 32% generating both CD4⁺ and CD8⁺ T lymphocyte responses (14). The Merck STEP trial failed to show any vaccine efficacy, and in those with elevated prevaccination recombinant Ad5 (rAd5) antibody responses, the prevaccination rAd5 antibody response appeared to enhance HIV-1 acquisition (15). NHP vaccine studies where a number of vaccine-elicited T cell epitope responses were much greater than those in the STEP trial have shown that the number of vaccine-elicited T cell responses is inversely correlated with the viral load at set point (16). Thus, one of the reasons why the STEP trial failed to enable improved viral control upon infection may be that the number of vaccine-elicited T cell responses were too low and that many of the responses may not have been able to recognize the epitope variants found in the circulating strains of HIV-1 to which the study participants were exposed (17).

The first generation of centralized vaccines were designed based on either consensus sequences, wherein the most common amino acid at each site in an alignment is concatenated to reconstruct full-length but artificial proteins (1, 12, 18–23), or modeled sequences of ancestral states or the center of a tree, reconstructed from maximum likelihood phylogenies (1, 21, 24). Consensus Envs have been shown to be biologically active in that they can mediate infection when expressed as pseudoviruses (21). In a comparative study in guinea pigs where 8 single-clade, global consensus Envs and 12 natural strain Envs were used as immunogens, the M group consensus gp140 Env sequence, called Con-S, was the best immunogen in terms of eliciting neutralizing antibody responses against a panel of 36 tier 1 or tier 2 viruses (25). Con-S not only induced much higher titers of neutralizing antibody responses than the other immunogens used in the study against tier 1A and 1B viruses, but these responses also showed modest neutralization of a subset of tier 2 viruses.

Polyvalent mosaic immunogens were designed as a second-generation centralized immunogen approach, to maximize epitope coverage of a diverse population of viruses using a small set of proteins. The mosaic algorithm enables the design of sets of complementary proteins that, when used in combination, maximize the coverage of potential T cell epitopes in a circulating population of viruses and thus could potentially elicit immune responses with greater cross-clade reactivity (11). NHP vaccinations with mosaic immunogens indeed have demonstrated that combinations of mosaic immunogens induced more cellular immune responses, with greater cross-reactive potential per response than either consensus immunogens or natural immunogens (26, 27). A pair of mosaic antigens used in combination elicited more potent and cross-reactive responses than a nonmosaic pair of immunogens that represented a B and C clade combination (26). Furthermore, Ndhlovu et al. demonstrated that mosaic HIV antigens expressed by rAd26 are processed into natural epitopes that are recognized by human HIV-specific CD8⁺ T lymphocytes (28). In a rhesus monkey vaccine study, we demonstrated that an increase from 2-valent to 3-valent mosaic Envs enhanced the breadth and potency of T cell and antibody responses against natural variants (29). Finally, mosaic HIV Env vaccines delivered in adenovirus or by modified vaccinia Ankara (MVA) vector combinations had a strong protective effect against infection by subsequent simian-human immunodeficiency virus (SHIV) challenge in an NHP study where vaccinated animals had markedly improved outcomes postchallenge (6). Taken together, these studies demonstrated the potential of centralized and polyvalent vaccines to enhance the immune coverage of diverse HIV-1 strains by expanding the breadth and the depth of cellular immune responses. In contrast to HIV mosaics conferring protection in an SHIV challenge model, SIV mosaic Env designs based on a small set of available sooty mangabey natural SIV variants (30) did not show a protective effect against infection (9). The lack of protection from infection in the SIV case might have been due to basing the mosaic design on very limited sampling of a less appropriate population (for a detailed discussion, see Fischer et al. (30). Although infection was not inhibited in this challenge model, the SIV mosaic Gag-vaccinated animals had significantly lower viral loads after infection than the other groups of vaccinated monkeys. (9).

Here, we have compared 3 vaccine antigens: a natural transmitted-founder (T/F) Env, B.1059, selected because it is the natural strain with the best global potential T cell epitope coverage; Con-S, the best consensus Env immunogen we have tested in guinea pigs (25) in terms of eliciting neutralizing antibody responses; and a 3-valent global mosaic Env. We evaluated their antigenicities and their immunogenicities in a DNA prime, recombinant NYVAC (rNYVAC), and Env protein boost immunization protocol (31).

MATERIALS AND METHODS

Selection of Env immunogens.

The three HIV-1 Env immunogens used in this study, the T/F clade B Env B.1059, a group M consensus Env (Con-S), and 3-valent mosaic Env, are being evaluated in a phase I clinical trial (HVTN 106). Con-S is an older consensus sequence that has been studied intensively for many years (1, 19); it was a global consensus of subtype consensus sequences generated as described by Gaschen et al. (1). In this study, the particular global trivalent mosaic set described here and the B.1059 Env immunogens are being used as immunogens for the first time. These reagents were designed or selected using the Los Alamos HIV database global Env reference alignment, and the comparisons during the design phase were based on all sequences available in the database in the years 2008 and 2009. The alignments included 2,020 full-length Env sequences, including only one sequence per subject: 728 were from clade B, 599 were from clade C, and 693 were a combination of all other clades, circulating recombinant forms, and unique recombinants. The 3-valent Env mosaic antigens were generated as described by Fischer et al. (11). The single T/F natural protein B.1059 Env was selected, as it provided the best-ranked 9-mer coverage of the global database among all T/F virus Env sequences available as of 2008. It was a B clade variant, reflecting a bias in the database, as more B clade viruses than any other clade were available in the database alignments at the time of study. While the B clade is rare in Africa, it is common in North America, Asia, South and Central America, the Caribbean, Europe, and Australia, and thus the B clade and the C clade, which is the most common genetic subtype in southern Africa and parts of Asia, are both key epidemic lineages for vaccine considerations (12). Figures 1 and 2 show the relative 9-mer coverage of the envelope sequences from the global database compared to that of the three vaccines studied here. Figure 1 shows the average coverage per strain, for all sequences, and for clades B and C, the most heavily represented samples in the database, as well as a summary of all other clades after excluding B and C clades. As expected, Con-S coverage of 9-mers is uniform across clades, averaging ∼26% of 9-mers with a perfect match to the vaccine per natural sequence. As expected, B.1059 provided the best 9-mer coverage of B clade sequences, reaching 32%, as the comparisons are within clade. The coverage was only 16 and 18% for C clade and non-B or non-C clade sequences, respectively, with 23% 9-mer coverage overall. The 3-valent mosaic immunogens provided an ∼48% average 9-mer match to both B and C clade sequences, a 39% match on average to all other non-B and non-C clade sequences, and 45% overall coverage. The preferential 9-mer coverage of the B and C clades by the mosaic set reflects the bias in the database, as discussed above, which was a deliberate design choice, as it reflects the global epidemic. Notably, the 39% 9-mer match to non-B and non-C clade envelopes is still substantially better than the within-clade 9-mer coverage of the B clade by the optimal natural B clade strain, B.1059.

FIG 1.

Average coverage of 9-mers by three HIV-1 Env immunogens. The red bar indicates the fraction of 9-mers that were perfectly matched by a 9-mer included in the vaccine, the orange adds on the frequency of 9-mers with a match of 8 of 9, and the yellow a match of 7 of 9. This plot is alignment independent, based on splintering all M group proteins, one sequence per person, into all possible 9-mers, attending to their frequencies, and then calculating the frequency of matches and near matches in the full database to each vaccine antigen or protein cocktail. Each vaccine is summarized 4 ways. The “Total” represents all sequences in the database alignment at the time these sequences were designed (2008). “B” is the subset of sequences that are B clade, “C” is the subset of sequences that are C clade, and “Other” are the remaining M group sequences that are neither B nor C clade (all other clades and recombinants). Mosaics were optimized in such a way that they maximized the red bar on the left, the exact matches in the full or total data set, but they provide excellent coverage of B clade, C clade, and other sequences. The best single natural T/F virus, B.1059, was also selected to maximize the number of exact matches in the full database, given the constraints of selecting from among T/F viruses and using a natural sequence. The single best natural is of course a B Env, as the B clade dominates the database, and thus the B clade has the best coverage for B.1059, and clade C and other clades have markedly less coverage. Con-S, as expected, provides similar coverage across clades.

FIG 2.

Alignment-dependent coverage of 9-mers by three HIV-1 Env immunogens. Each position represents a 9-mer in the alignment of all HIV Env proteins used for this study. The number of exact, 8 of 9, or 7 of 9 matches to the vaccine for each column of 9 positions are shown as red, orange, and yellow bands, as described for Fig. 1, with the coverage broken down per 9-mer. The 9-mers are sorted so that the ones with the most vaccine matches are presented on the left, and the ones with the fewest are on the right. The upper bound (black dashed line) is the same for all three immunogens and is the sum of the frequencies of the three most common 9-mers starting from each position. It represents the maximal limit that could be achieved for epitope coverage with 3 proteins. The reason the “total 9-mers,” shown in gray, varies dramatically is because of insertions and deletions in the alignment. Because of deletions in Env, which are very common, sometimes only fragments of a particular 9-mer in the alignment exist in a given strain; thus, the fraction of Envs that carry an intact 9-mer in the alignment for a given column of 9 amino acids is indicated in gray.

The combined mosaics offered optimal 9-mer coverage for a 3-protein set (all 9-mers are considered to represent potential epitopes in the optimization algorithm [11]). The 3-mosaic combination, as we have shown previously, provides coverage that approaches the coverage achievable by summing the frequencies of the 3 most common forms of each of the 9-mers in the alignment (Fig. 2). The less than 50% coverage of 9-mers, even with 3 mosaics, is a consequence of the great diversity of the Env protein. There are two issues that are often a source of confusion regarding the mosaic algorithm, which we attempt to clarify here. First, one cannot simply include the 3 most common 9-mers spanning the alignment in the vaccines and create 3 intact mosaic proteins, as there can be contradictions in overlapping 9-mers regarding which amino acids are favored, and so the mosaic algorithm optimizes for coverage given that constraint. Second, 9-mers are used for optimization, as that is the most common length of epitopes recognized by CD8⁺ cytotoxic T cells. As optimal CD4 T cell epitopes tend to be have variable lengths, often greater than 9 amino acids long, the question of whether mosaic immunogens are likely to elicit both cross-reactive CD8⁺ T and CD4⁺ T cells has arisen. As it turns out, the optimal solution for 9-mers is also very nearly optimal for 8- to 12-mers (13), and potential epitope coverage of all nearby lengths is significantly enhanced relative to using natural strains, which include some rare amino acids that are likely to elicit more strain-specific responses (data not shown) (13). Furthermore, the key interaction region for major histocompatibility complex (MHC) class II-epitope binding is a core region that is typically 9 amino acids in length, embedded within the longer class II epitopes, so 9-mer coverage is a reasonable choice for both MHC class I- and class II-presented epitopes. Also, previous studies have demonstrated that both CD4⁺ and CD8⁺ T cell responses are improved by mosaic antigens based on 9-mer optimization (26).

We chose to use a T/F Env for the natural strain, as the T/F sequences, like T/F Envs, have somewhat distinctive immunogenic profiles in terms of neutralization antibody induction relative to chronic infection isolates (25), and we hoped to enable induction of both good B and T cell Env responses. The reagent selection was made from a set of 113 B clade and 40 C clade T/F viruses (32, 33). T/F sequences were based on within-sample consensus of large sets of single-genome amplification (SGA) sequences obtained from the first time point sampled in acute infection and are generally identical to actual SGA sequences in the sample. Among all gp160 protein sequences available, not restricted to T/F viruses, YU-2 (accession number M93258), a 1986 sample, had the best 9-mer coverage. This was not surprising, because older sequences tend to be more central in terms of the phylogeny (34) and consequently provide better 9-mer coverage of all contemporary strains (B. Korber, unpublished data). B.1059 provided 22.6% 9-mer coverage, and YU-2 provided 24.2% 9-mer coverage; B.1059 was selected over YU-2, as it had the additional advantage of being a T/F isolate. Nine-mer coverage of all natural strains ranged between 8 and 24%. The B.1059 T/F sequence was the consensus sequence obtained from the first time point sampled from an acutely infected individual sampled in 1998 in California, USA, and perfectly matches some of the sequences obtained from this time point (EU575375 to EU578926).

Construction of the mosaic recombinant NYVACs: plasmid construction.

The mosaic 13316 gp140 gene was cloned into a transfer vector using standard cloning methods. In vitro recombination (IVR) was used to transfer the insert from the transfer vector to the genome of NYVAC. This work was conducted under good laboratory practice (GLP) conditions in BSC-40 cells. Three rounds of plaque purification were done in BSC-40 cells, followed by three rounds in chicken embryo fibroblasts (CEFs). The mosaic 13317 gp140 gene was amplified by PCR to create a linear fragment that included portions of the thymidine kinase (TK) flanking regions 5′ and 3′ to the open reading frame. IVR was used to incorporate the linear PCR product into the genome of NYVAC.

Plasmid transfer vector construction.

The mosaic gp120 13315, Con-S gp120 13318, and gp120 wild-type (WT) B.1059 13319 genes were amplified from plZAW1-containing mosaic genes by PCR, using primers that would add the XhoI and PmeI sites for cloning into pCyA20 plasmid previously digested with the same restriction enzymes. The plasmid pCyA20 was used for the engineering of the recombinant NYVAC viruses expressing the HIV-1 gp120 13315, Con-S gp120 13318, or gp120 WT B.1059 13319 genes. The plasmid was designed for a blue/white plaque screening. It contains TK left and right flanking sequences, a short TK left arm repeat, a vaccinia virus E3L promoter-driven β-galactosidase (β-Gal) expression cassette, and the ampicillin gene. Between the two flanking sequences, there is a vaccinia virus synthetic early/late (E/L) promoter driving the expression of gp120 13315, Con-S gp120 13318, and gp120 WT B.1059 13319 genes. This plasmid directs the insertion of the mosaic gp120 genes into the TK locus of the NYVAC genome. After the desired recombinant virus has been isolated by screening for expression of β-galactosidase activity, further propagation of the recombinant virus leads to the self-deletion of β-Gal by homologous recombination between the TK left arm and the short TK left arm repeat that are flanking the marker.

Mosaic NYVAC recombinant virus construction.

A total of 3 × 10⁶ BSC-40 cells were infected with NYVAC-WT at a multiplicity of 0.025 PFU/cell and transfected 1 h later with 6 μg DNA of pCyA20-gp120-13315, pCyA20-gp120-13318, or pCyA20-gp120-13319 using Lipofectamine (Invitrogen) according to the manufacturer's recommendations. After 72 h postinfection, the cells were harvested, lysed by freeze-thaw cycling, sonicated, and used for recombinant virus screening. Recombinant NYVAC viruses containing gp120 13315, gp120 13318, or gp120 13319 genes and transiently coexpressing the β-Gal marker gene were selected by 3 consecutive rounds of plaque purification in BSC-40 cells stained with 5-bromo-4-chloro-3-indolyl β-galactoside (1.2 mg/ml). In the following rounds, recombinant NYVAC viruses containing mosaic gp120 genes and having deleted the β-Gal marker gene were isolated by 3 additional consecutive rounds of plaque purification screening for nonstaining viral foci in CEF cells (Charles River) in the presence of 5-bromo-4-chloro-3-indolyl β-galactoside (1.2 mg/ml). Approximately 30% of the final plaque (6 isolation steps in total, 3 in BSC-40 cells and 3 in CEF cells) was used to infect a 25-cm² flask of CEFs and amplified for approximately 48 h to derive a P1 stock. This small stock was harvested, the titer was determined, and the stock was used to infect 10 p150 plates of CEFs at a multiplicity of infection (MOI) of 0.01 to generate a P2 stock. The P2 stock was characterized by titer, expression of HIV-1 antigens (by Western blotting and percentage of positive plaques by immunoplaque assay), blood agar plate assay to detect any viable bacterial or fungal growth, and PCR assay for mycoplasma. The P2 stock was further passaged 5 additional times in CEFs at an MOI of 0.01 PFU/cell (from P3 to P7) for analysis of stability. Expression of the correct insert from 30 individual plaques was determined at P7 by Western blot analysis. The resulting NYVAC-mosaic gp120 recombinant virus P2 stock crude preparations were used for the propagation of the viruses in large cultures of BHK-21 cells followed by virus purification through two 36% (wt/vol) sucrose cushions (35).

Selection and vaccination of monkeys.

A PCR-based assay was used to select 15 adult rhesus monkeys (Macaca mulatta) that expressed the rhesus monkey MHC class I allele Mamu-A*01 (36). The reason for selecting Mamu-A*01⁺ monkeys was to ultimately study particular epitope responses across groups, but for this particular study, we instead characterized the total cross-reactive pooled responses. Of note, the single immunogens each contain a roughly comparable number of Mamu-A*01-predicted binding motifs, ∼20 (consensus scores of <1% using the Immune Epitope Database epitope prediction tool, http://tools.immuneepitope.org/mhci/ [37]). In contrast, the mosaic polyvalent set of 3 proteins, as anticipated for any MHC molecule, has many more potential epitopes between the 3 proteins (54 with consensus score of <1). This gain reflects one of the possible benefits of including antigens that represent natural diversity, as it increases the potential for cross-reactivity. Monkeys were housed at New England Primate Research Center, Southborough, MA. The animals were maintained in accordance with National Institutes of Health and Harvard Medical School guidelines. Fifteen monkeys were distributed in three experimental groups (n = 5/group). On weeks 0 and 4, the monkeys were primed with 5 mg plasmid DNA expressing either B.1059 Env, Con-S Env, or 3-valent mosaic Envs, by the intramuscular route. On weeks 20 and 24, the monkeys were boosted with intramuscular immunizations using 10⁸ PFU of rNYVAC expressing the respective antigens. A gp120 boost of the match protein antigens was given at week 62 (Table 1 shows the full vaccination schedule).

TABLE 1.

Immunization schedule

Group (n = 5 monkeys/group)	Immunogen	Schedule of:
		Prime (plasmid DNA gp160)	Boost (rNYVAC gp120)	Protein boost (gp120)
1	B.1059	Wks 0, 4	Wks 20, 24	Wk 62
2	Con-S	Wks 0, 4	Wks 20, 24	Wk 62
3	3-Valent mosaic	Wks 0, 4	Wks 20, 24	Wk 62

IFN-γ ELISpot assays.

Multiscreen 96-well plates were coated overnight with 100 μl per well of 5 μg/ml anti-human gamma interferon (IFN-γ) (B27; BD Pharmingen) in endotoxin-free Dulbecco's phosphate-buffered saline (D-PBS). The plates were then washed three times with D-PBS containing 0.1% Tween 20, blocked for 2 h with RPMI medium containing 10% fetal bovine serum (FBS) to remove the Tween 20, and incubated with peptide pools and 2 × 10⁵ peripheral blood mononuclear cells (PBMCs) in triplicate in 100-μl reaction volumes. Each peptide pool was comprised of 15-amino-acid peptides overlapping by 11 amino acids. The pools covered the entire HIV-1 Env proteins from clades A (accession no. AY371163), B (accession no. AY561237), C (AY463223), and G (accession no. AY371121) and M group consensus Env. Each peptide in a pool was present at a 1 μg/ml concentration. Following an 18-h incubation at 37°C, the plates were washed nine times with D-PBS containing 0.1% Tween 20 and once with distilled water. The plates were then incubated with 2 μg/ml biotinylated rabbit anti-human IFN-γ (U-Cytech, The Netherlands) for 2 h at room temperature, washed six times with D-PBS containing 0.1% Tween 20, and incubated for 1.5 h with a 1:500 dilution of streptavidin-AP (Southern Biotechnology, Birmingham, AL). After five washes with D-PBS containing 0.1% Tween 20 and three washes with D-PBS alone, the plates were developed with bromochloroindolyl phosphate-nitroblue tetrazolium (BCIP-NBT) chromogen (Pierce), stopped by washing with tap water, air dried, and read with an ELISpot reader (Cellular Technology, Ltd.) using ImmunoSpot Analyzer software (Cellular Technology, Ltd., Ohio).

PBMC stimulation and intracellular cytokine staining.

PBMCs were isolated from EDTA-anticoagulated blood (Amersham Biosciences, New Jersey) and cryopreserved. Cells were later thawed and allowed to rest for 4 h at 37°C in a 5% CO₂ environment. PBMCs were then incubated at 37°C in a 5% CO₂ environment for 6 h in the presence of either RPMI containing 10% fetal bovine serum (unstimulated), phorbol myristate acetate (PMA)-ionomycin as a positive control, or a pool of HIV-1 Env peptides. Again, each peptide pool contained 15-mers overlapping by 11 amino acids, spanning the entire HIV-1 Env. We used five sets of pooled HIV-1 Env peptides from 4 different clades, the 4 described above, as well as an additional clade B Env (accession number AY037269). The 5 natural strains of HIV-1 that were used for the design of peptide reagents in this study were a subset of a larger set of Env peptides that were designed based on 10 natural strains and that we have used in previous studies to assess the cross-reactivity of vaccine responses to natural variants (29, 38). All cultures contained a protein transport inhibitor, monensin (Golgi Plug BD Bioscience), as well as 1 μg/ml of anti-CD28 and 1 μg/ml of anti-CD49d (BD Bioscience). These cultured cells were then stained with a cell viability marker (yellow amine dead cell stain; Invitrogen) and pretitered quantities of anti-CD3 Pacific Blue (SP34.2; BD Biosciences), anti-CD4 PerCP-Cy5.5 (L200; BD Bioscience), and anti-CD8 allophycocyanin (APC)-Cy7 (SK1; BD Bioscience). Following fixing and permeabilization with Cytofix/Cytoperm solution (BD Biosciences), cells were stained with anti-CD69-ECD (TP1.55.3; Beckman Coulter), anti-IFN-γ phycoerythrin (PE)-Cy7 (B27; BD Biosciences), anti-tumor necrosis factor alpha (TNF-α)-fluorescein isothiocyanate (FITC) (monoclonal antibody 11 [MAb11]; BD Biosciences), and anti-interleukin 2 (IL-2) APC (MQ1-17H12; BD Biosciences) and fixed with 1% formaldehyde. Samples were collected on an LSR II instrument (BD Biosciences) and analyzed using FlowJo software (version 9.3.1). Approximately 500,000 events were collected per sample. The background level of cytokine staining varied from sample to sample but was typically below 0.07% of gated T cells. All values used in the analyses were background subtracted.

Binding assays.

Direct binding enzyme-linked immunosorbent assays (ELISAs) were conducted in 384-well ELISA plates (Costar number 3700) coated with 2 μg/ml antigen in 0.1 M sodium bicarbonate and blocked with assay diluent (PBS containing 4% [wt/vol] whey protein-15% normal goat serum-0.5% Tween 20-0.05% sodium azide). Sera were incubated for 90 min in 3-fold serial dilutions starting at 1:30 followed by washing with PBS-0.1% Tween 20. A total of 10 μl horseradish peroxidase (HRP)-conjugated goat anti-human secondary antibody (Jackson ImmunoResearch) was diluted to 1:10,000 in assay diluent without azide, incubated for 1 h, washed, and developed with 20 μl SureBlue Reserve (KPL 53-00-03) for 15 min. The reaction was stopped with the addition of 20 μl HCl stop solution. Plates were read at 450 nm.

Neutralization assays in TZM-bl cells.

Neutralizing antibody assays in TZM-bl cells were performed as described previously (39). Plasma samples were tested at eight 3-fold dilutions starting at a 1:20 dilution. Env-pseudotyped viruses were added to the plasma dilutions at a predetermined titer to produce measurable infection and incubated for 1 h. TZM-bl cells were added and incubated for 48 h before lysis, after which supernatant was measured for firefly luciferase activity by a luminometer. The data were calculated as a reduction in luminescence compared with that of virus control wells after subtraction of background luminescence in cell control wells and reported as the plasma dilution 50% inhibitory concentration (IC₅₀). All Env-pseudotyped viruses were prepared in 293T cells and titrated in TZM-bl cells as described, and analysis strategies are described below.

Statistical analyses and considerations of grouping for statistical evaluation.

Both antibody (40) and T cell immune responses to natural HIV infection (41) have enhanced within-clade cross-reactivity. These observations, together with the frequency of shared potential epitopes for the T/F B clade antigen B.1059 (Fig. 1), led us to anticipate that heterologous but within-clade peptide vaccine responses would be greater in magnitude and frequency than responses to peptides based on sequences from other clades. Thus, we analyzed responses to B clade heterologous peptides separately from to A, C, and G peptides, to determine the within-clade and between-clade cross-reactive potential based on the natural B clade vaccine, and compared those responses to the global M group Con-S and mosaic vaccines. A desirable outcome for global centralized vaccines would be to elicit immune responses that are as good or better than within-clade heterologous responses, so in this study we explored the relative immunogenicity of the B clade T/F antigen within and between clades compared to the immunogenicity of Con-S or 3 mosaics evaluated against the same targets.

Statistical analyses of the T cell responses.

To compare the magnitude of responses, responses to B clade peptides were compared to interclade A, C, and G peptides to the different vaccines using a nonparametric permutation test, as described in detail by Parrish et al. (42). Here, to test for statistical differences across the three vaccines in the magnitude of responses, we used a 1-sided permutation test comparing the vaccines in a pairwise manner (B.1059 to Con-S, B.1059 to mosaic, and Con-S to mosaic). The permutation test is a conservative, nonparametric rank-based test designed as follows: for each iteration, the labels “vaccine A” and “vaccine B” are reshuffled across monkeys. The responses are then ranked, and then the ranks from vaccine A are summed. After 99,999 iterations, a P value is computed as the percentage of times the randomized sum of ranks is greater than or equal to the observed sum of ranks. Such a P value measures the probability of observing higher responses in vaccine B than in vaccine A by chance alone. To test individual vaccine responses to the B clade peptides, we used a 1-sided paired Wilcoxon test, comparing the three vaccines pairwise. Finally, we fit a Gaussian generalized linear model (GLM) with vaccine as the sole fixed effect and clade and animal treated as random effects to compare across-clade responses from B.1059 and Con-S and from B.1059 and mosaic. The statistical software R was used for all basic statistical and graphical comparisons of ELISpot and neutralizing antibody responses. Heat maps using hierarchical clustering were generated using the Los Alamos database heat map tool (http://www.hiv.lanl.gov/content/sequence/HEATMAP/heatmap.html).

Statistical analyses of antibody binding and blocking responses.

All pairwise comparisons between groups were tested using a nonparametric Wilcoxon test at each time point. In the figures, we present the raw P values of <0.05 without multiple testing corrections; thus, the data should be viewed in terms of this cautionary note. There was a general consistency in patterns across time points, however, as well as agreement with expected outcomes (e.g., antibody binding to the vaccine strains tended to be higher than to other proteins); thus, these exploratory analyses, even with only 5 monkeys per group, suggest trends and patterns in the data that are likely to be meaningful. As we had only 5 animals per group, the study was not sufficiently powered to do a paired nonparametric Wilcoxon signed-rank test to evaluate the levels of response after the NYVAC boost (week 26) compared to after the protein boost (week 64). Therefore, we looked across the 3 groups (n = 15) for each of the binding and blocking activities to determine if the HIV-specific responses were generally increased in magnitude following the protein boost. We used the Benjamini-Hochberg false discovery rate (FDR) corrected P value to adjust for multiple tests (43).

Statistical analyses of neutralizing antibody responses.

When analyzing neutralizing antibody responses, in order to account for each animal's background, we considered responses to be positive only when they were still positive after subtracting three times the background; otherwise, we set them to the detection threshold of 10 and considered the virus resistant. This method is the traditionally used cutoff; however, such a stringent cutoff may introduce type II error and miss lower-level responses that are specific to a group and could potentially be of interest (25). Therefore, we also applied a more inclusive cutoff and considered responses to be positive that were greater than the detection threshold of 10 after subtracting the background; otherwise, we set them to the detection threshold of 10 and considered the virus resistant. We present both views of the data here.

To test for statistical differences in the magnitude of responses across the three vaccines, we used the 1-sided permutation test outlined above in the T cell response section to compare the vaccines while controlling for peptide clade. We also fit a lognormal mixed-effect GLM with vaccine as the fixed effect and animal and Env as random effects. To evaluate statistical differences across the three vaccines in terms of breadth of responses, we counted the number of responses above the threshold in each animal and used a Kruskal-Wallis test and then a Wilcoxon test to compare the 5 animals in each experimental group in a pairwise manner (B.1059 to Con-S, B.1059 to mosaic, and Con-S to mosaic). In addition, we fit a binomial mixed-effect model (with vaccine as the fixed effect and Envs and animal as random effects) for the proportion of positive responses.

RESULTS

Cross-clade magnitudes of vaccine-elicited IFN-γ ELISpot responses.

Table 1 shows the NHP immunization regimen with the three arms of the study. The magnitude and cross-clade breadth of the vaccine-elicited cellular immune responses were evaluated by IFN-γ ELISpot responses to peptide pools comprising 15-mers overlapping by 11 amino acids spanning the entire HIV-1 envelope sequence. HIV-1 Env peptide sets representing isolates from four different clades, one each from clades A, B, C, and G, were used to stimulate peripheral blood lymphocytes (PBL) from the vaccinated monkeys for recognition of these four different sets of HIV-1 Env peptides (38). After plasmid DNA priming (week 6), low to moderate IFN-γ responses were detected in monkeys in all three immunization groups to all four peptide pools (Fig. 3, top). However, as shown in Fig. 3, robust ELISpot responses to the peptide antigens were detected in all monkeys at week 22, 2 weeks following the first rNYVAC boost (Fig. 3, middle). The 2nd rNYVAC boost did not enhance the magnitude of responses (Fig. 3, bottom); instead, responses in clade B- and mosaic-immunized monkeys diminished slightly, possibly due to development of antivector immunity in the animals.

FIG 3.

The magnitude of clade-specific T lymphocyte responses in individual monkeys in three vaccine groups as determined by IFN-γ ELISpot assay. PBL from each vaccinated monkey was evaluated for IFN-γ ELISpot responses to peptide pools of 15-mers overlapping by 11 amino acids spanning each of the 4 selected envelope sequences at week 6 (top), week 22 (middle), and at week 26 (bottom). Data are shown as spot-forming cells (SFC) per million PBL. Blue bars represent responses to the clade B peptide pool. To distinguish the B.1059 vaccine within-clade responses and between-clade responses, the B clade peptide responses are shown in blue, and the A, C, and G clade peptide sets are shown in different shades of red/orange (maroon bars for clade A, red bars for clade C, and orange bars for clade G peptide pools). Each set of 4 bars represents the ELISpot responses of one animal in a group, with a total of 5 animals per vaccine group. For the B.1059 vaccine, intraclade responses were higher than interclade responses.

Cross-clade breadth of the vaccine-elicited IFN-γ ELISpot responses was analyzed at post-prime and at post-1st rNYVAC boost time points at weeks 6 and 22, respectively. Here, we present results from week 22, as it was the sample time point with the highest level of activity. Since one of the vaccines, B.1059, expressed a natural clade B Env immunogen, we first wanted to determine if the magnitude of T cell responses to peptides based on a heterologous but within-clade isolate was greater than that of the responses to peptides based on 3 isolates from other clades. B.1059 responses were higher in B clade than A and C clade peptides in each of the 5 animals in the B.1059 vaccine group (P = 0.03 in each case by 1-sided, paired Wilcoxon test). Significance was lost when comparing B clade to G clade responses, likely due to the impact of one exceptional immunodominant response (P = 0.31, 1-sided, paired Wilcoxon test) (data not shown). Such exceptions are not unexpected, as occasionally an epitope will be better matched in a sequence from a different clade than in a heterologous sequence from the same clade. When we compared intraclade and interclade responses to the B.1059 vaccine using a Gaussian GLM, where inter/intra was the only fixed effect and animal and clade nested within inter/intra were random effects, the model yielded a nonsignificant result (P = 0.2). However, this comparison became significant when the outlier response to G clade peptide was removed (P = 0.01). After excluding that exception, the other B.1059 responses were on average 938.5 SFC/10⁶ PBL higher to B clade peptides than to peptides based on sequences from other clades.

The B.1059 Env vaccinees responded to B clade (intraclade) peptides at a level that was statistically comparable to that of the monkeys that received the Con-S or mosaic Env vaccines (Fig. 4). However, B.1059 vaccine elicited significantly lower cross-clade responses to A, C, and G clade peptides than those elicited by Con-S (P = 0.008) and mosaic vaccines (P = 0.01), as determined by a nonparametric permutation test (Fig. 4). We then fit a Gaussian GLM, with vaccine as the sole fixed effect and clades and animals treated as random effects, to compare B.1059 single-clade vaccine responses to the mosaic and Con-S responses. Mosaic vaccine responses were significantly higher than responses to the B.1059 vaccine (P = 0.03; on average, mosaic was 821.1 SFC/10⁶ PBL higher than B.1059). Con-S responses were on average 610.9 SFC/10⁶ PBL higher than B.1059 responses, but this trend was not statistically significant (P = 0.1). Con-S and mosaic vaccines were not significantly different from each other using the Gaussian GLM, although mosaics ranked higher.

FIG 4.

Magnitudes of ELISpot responses at week 22. Magnitudes of the IFN-γ ELISpot data from week 22 were analyzed in detail. (A) B clade responses across the three vaccine groups (P values are from a 1-sided, paired Wilcoxon test); (B) responses to clades A, C, and G across the three vaccine groups (P values from a 1-sided permutation test). Interclade responses to the B.1059 vaccine are significantly reduced relative to Con-S and mosaic responses, while intraclade responses to B.1059 are comparable to Con-S and mosaic responses. Horizontal bars show the respective medians.

Vaccine-elicited CD4⁺ and CD8⁺ T lymphocyte responses.

The DNA and NYVAC vectors used in the present study generate both CD4⁺ and CD8⁺ T lymphocyte immune responses (31, 44). While mosaic immunogens have been shown to improve the cross-reactivity of both CD8⁺ and CD4⁺ T lymphocyte vaccine responses (26), the mosaic advantage in earlier studies was far more pronounced for CD8⁺ T-cell responses (26, 27). Thus, we assessed the CD4⁺ and CD8⁺ T lymphocyte responses separately in all three groups using intracellular cytokine assays (ICS). PBL from all 15 monkeys from week 26 were stimulated with five different HIV-1 Env peptide pools from four different clades as described in Materials and Methods (for the ICS studies, an additional heterologous B clade Env peptide set was added to the 4 HIV Env pools used for the ELISpot experiment described above, to enrich the within-clade B.1059 response data). Following stimulation, the lymphocytes were stained with antibodies to CD3, CD4, CD8, IFN-γ, TNF-α, and IL-2 to determine the percentages of CD4⁺ and CD8⁺ T lymphocytes that were producing the three cytokines in response to the peptide stimulation. IL-2 responses in both CD4⁺ and CD8⁺ T lymphocytes were marginal from all vaccine groups (shown for CD8⁺ T cells in Fig. 5); thus, to assess IL-2 responses, we did not do a comprehensive analysis but performed a simple 1-sided Wilcoxon rank test to compare all 25 peptide responses between groups. This analysis indicated that the B.1059 vaccine IL-2 response levels were lower than Con-S or mosaic Env responses for CD8⁺ T cells (P = 0.0073 and P = 0.0035, respectively) and CD4⁺ T cells (P = 0.037 and P = 0.0012, respectively). Con-S and mosaic vaccines induced comparable levels of IL-2 responses. Next, we did a more detailed analysis of the IFN-γ and TNF-α responses, as those were much stronger than the IL-2 ones. IFN-γ and TNF-α responses were highly correlated (Spearman's ρ was 0.95 for CD4 and 0.98 for CD8). Given this, for the following analyses of the cross-clade magnitude of the cytokine-positive CD4⁺ and CD8⁺ T lymphocyte responses, we report just one parameter, IFN-γ secretion, as representative of vaccine-elicited responses. As with the ELISpot magnitude comparisons, we were interested in comparing intra- and interclade responses to B.1059 vaccination to centralized vaccines, Con-S and mosaic.

FIG 5.

Percentages of cytokine-positive CD8⁺ T lymphocyte responses in three vaccine groups. PBL from each vaccinated monkey were evaluated for secretion of IFN-γ, TNF-α, and IL-2 in an intracellular cytokine assay following stimulation with peptide pools from five selected HIV-1 envelope sequences, one each from clades A, C, and G and 2 heterologous viruses from clade B; each of the 5 strains is described in Materials and Methods. Responses to the two heterologous B clade peptides (denoted as B1 and B2) are shown in different shades of blue bars. Clade A, maroon; clade C, red; clade G, orange. Percentages of cytokine⁺ CD8⁺ T cells are shown.

There was no clear statistical support for a within-clade advantage in the CD8⁺ T cell IFN-γ responses measured by ICS, but a trend was evident. In a pairwise nonparametric comparison, the B.1059 vaccine elicited higher responses to clade B than to clade A with modest significance (P = 0.03) but not to clades C or G; this trend is consistent with the intrasubtype advantage evident in the ELISpot data presented above. Modeling the responses to B.1059 using a GLM revealed no statistical support for an intraclade advantage within the B.1059 vaccine group. When B clade responses were compared across all vaccine groups, both Con-S and mosaic vaccines elicited significantly higher IFN-γ-secreting CD8⁺ T lymphocytes than B.1059 vaccines: Con-S elicited the highest B clade responses (P = 0.004, nonparametric permutation test), but the mosaic vaccine also improved the response relative to B.1059 (P = 0.02) (Fig. 6A). Similarly, as shown in the right panel of Fig. 6B, for clades A, C, and G, both Con-S and mosaic vaccines had higher responses than B.1059, with Con-S being significantly higher (P = 0.008; permutation test) and mosaic marginally higher (P = 0.09; permutation test). When responses between Con-S and mosaic vaccines were compared to each other, across all clades, no significant difference was found. We then fit a lognormal GLM to support the results obtained with the permutation test. The model had the vaccine as the sole fixed effect and clade and animal as random effects. When looking at only B clade responses (in this case, the random effects were animal and peptide), the vaccine effect was highly significant (P = 0.0047), with Con-S estimated to yield on average 0.98 more CD8 T cells than B.1059, and mosaic 0.74 more than B.1059 (both increased percentage values are on a log₁₀ scale). When looking at clade A, C, and G responses, the GLM yielded a significant result when comparing Con-S to B.1059 (P = 0.0016, with Con-S eliciting 0.94 more CD8 T cells on average than B.1059, on a log₁₀ scale) but only marginal when comparing mosaic to B.1059 (P = 0.06, with mosaic yielding 0.65 more CD8 T cells than B.1059 on average, on a log₁₀ scale).

FIG 6.

Magnitudes of IFN-γ⁺ CD8⁺ and CD4⁺ T cell responses. B clade responses of CD8⁺ T cells (A) and CD4⁺ T cells (C) compared across the three vaccine groups are shown (two B clade peptides were used, denoted here as B1 and B2). Panels B and D show responses to peptides from all other clades for CD8⁺ T cells and CD4⁺ T cells, respectively. Horizontal bars show the respective medians. All P values were calculated using a 1-sided permutation test. The color scheme is the same as that in Fig. 5. The values shown are the log₁₀ of the percentage of IFN-γ⁺ CD8⁺ and CD4⁺ T cells. Immunizations with Con-S and mosaic immunogens elicited significantly higher responses than B.1059 immunogen across all clades.

We repeated the analyses described above for CD4⁺ T lymphocyte populations, to compare the vaccine groups and the different peptide sets (Fig. 6C and D). Again, there was a slight suggestion of a preferential within-B-clade response in B.1059-vaccinated animals (P = 0.03 for A, P = 0.06 for both C and G). But in contrast to CD8⁺ T cells, there was no statistical difference between the vaccine groups assessed with either the nonparametric resampling strategy or the GLM approach.

Antigenicity of natural B.1059, Con-S, and 3-valent mosaic Env gp120 proteins.

Antigenicity of the mosaic and consensus Env proteins was evaluated by surface plasmon resonance (SPR) measurements of binding to sCD4 and CD4i epitope MAb (17b) and to a panel of broadly neutralizing (bnAb) and nonneutralizing MAbs (Table 2). Mosaic 3.3 NYVAC had an L121Q and I224V mutation compared to the DNA mosaic 3.3 sequence that arose during manufacture for a phase 1 human trial. Thus, we made both the wild-type and mutated L121Q and I224V gp120s and determined that the mutations did not affect antigenicity (Table 2). In immunizations described below, the mutated L121Q and I224V mosaic 3.3 gp120 was used as a protein boost. CD4 did not bind to mosaic 3.1 but bound well to mosaic 3.2 and to both the original and the mutated 3.3, as well as to Con-S and B.1059 gp120s. While mosaics 3.2 and 3.3 were infectious in a pseudotyping assay, mosaic 3.1 was not, consistent with its inability to bind CD4. The potent CD4bs bnAb VRC01, however, bound to mosaics 3.1 and 3.2 in approximately micromolar affinity but bound in low nanomolar affinities to mosaic 3.3, Con-S, and B.1059 gp120s (Table 2). CD4-induced binding of the 17b MAb was strong for both of the mosaic Env proteins that bound CD4 and was similar to the strong CD4i upregulation observed for Con-S and the transmitted founder B.1059 Env. The V3-N³³² oligomannose glycan-dependent bnAb PGT128 also bound to each of the Env proteins with nanomolar dissociation constants (K_d) and also bound well to the V3-loop MAb 19b. Glycan bnAb 2G12 bound all Envs at low nanomolar K_d and the PG9 V1V2 bnAb bound in a range from 25 nM (Con-S) to 540 nM (mosaic 3.3 L121Q/I224V). Thus, the mosaic and consensus Env Con-S gp120s were highly antigenic for a variety of neutralizing and nonneutralizing antibodies and were similar to the WT B.1059 Env gp120.

TABLE 2.

Antigenicity of T cell mosaic envelopes

Antibody/epitope	Kd (nM)a
	(TCM) M.mos 3.1 gp120	(TCM) M.mos 3.2 gp120	(TCM) M.mos 3.3 gp120	(TCM) M.mos 3.3 L121Q/I224V	Con-S gp120	B.Con 1059 gp120
CD4	NB	83.5	12.4	22.6	23.3	∼10
17b (CD4i)	NA	+++b	+++b	+++b	++b	++b
A32 (C1)	30.4	60.0	<1c	<1c	0.1	<1c
VRC01 (CDbs)	∼μM	∼μM	3.9	5.7	3.7	3.3
19b (V3)	53.4	8.2	24.6	42.1	17.3	9.4
PG9 (V1/V2)	183	218	371	540	25	55.3
CH01 (V1/V2)	NB	NB	NB	NB	NB	NB
697D (V2)	120	28.1	3.2	12.4	173	1160
2G12 (CHO)	7.1	2.1	6.6	12.2	4.7	10.4
CH58 (V2)	65.2	NB	NB	NB	371	NB
PGT128 (CHO)	171	97.5	5.9	5.9	8.0	7.2

^aK_d, dissociation constant; CHO, glycan; TCM, T cell mosaic; NB, no binding detected; NA, not tested due to lack of CD4 binding.

^bRelative to the Con-S gp120 ratio of 17b binding to Env gp120 captured; ++, ratio is the same as that for reference Con S gp120 (ratio = 0.35 − 0.44); +++, ratio is higher than that for Con S gp120 (ratio > 0.5).

^cBiphasic fit (K_d estimated from fast components).

Vaccination-induced binding antibodies.

Binding antibody responses were not seen to mosaic Envs until 2 weeks after the second NYVAC immunization at week 26 (Fig. 7). Mosaic immunization induced antibodies that bound somewhat better to mosaic Envs than WT Env B.1059 (Fig. 7), and there was consistent statistical support for this finding. Encouragingly, mosaic vaccine responses were higher than those of B.1059 to all three components of the mosaic cocktail; though not conclusive, this is consistent with the expectation that each of the three components was contributing to the response of the trivalent immunogen. Immunization with B.1059 Env or Con-S Env also induced slightly higher levels of antibodies to the immunizing Env than to heterologous Envs (Fig. 8). Interestingly, immunization with either mosaic or consensus Envs first induced early Con-S Env antibodies, 2 weeks after the second DNA immunization at week 6 (Fig. 7 and 8); in contrast, B.1059 Env immunization began to induce autologous B.1059 antibodies at week 6 but not Con-S binding antibodies (Fig. 8).

FIG 7.

Binding antibody responses to mosaic proteins. Log area under the curve (AUC) values of binding antibody responses to three separate components of the 3-valent mosaic immunogen, mos 3.1, mos 3.2, and mos 3.3, are shown at time points tested as shown on the x axis. Responses from monkeys immunized with B.1059 Env are shown by blue circles, Con-S by red squares, and mosaic by green triangles. Panels A to C show responses to gp120 forms of mos 3.1, mos 3.2, and mos 3.3, respectively. Panels D to F show responses to gp140 forms of mos 3.1, mos 3.2, and mos 3.3. All P values of <0.05 are shown, using a 2-sided Wilcoxon test to do a pairwise comparison of each group.

FIG 8.

Binding antibody responses to clade B immunogen, consensus immunogen, and gp70B.CaseA V1V2 protein. Log AUC values of binding antibody responses to clade B immunogen B.1059, consensus immunogen Con-S, and gp70 B.CaseA V1V2 protein are shown at time points tested as shown on the x axis. Responses from monkeys immunized with B.1059 Env are shown by blue circles, Con-S by red squares, and mosaic by green triangles. Responses to B.1059 gp120 (A), B.1059 gp140 (B), Con-S gp120 (C), Con-S gp140 (D), and gp70B.CaseA V1V2 (E) are shown. All P values of <0.05 are shown, using a 2-sided Wilcoxon test to do a pairwise comparison of each group.

Antibodies to the gp70B.CaseA V1V2 protein construct correlated with decreased transmission risk in the RV144 ALVAC/AIDSVAX vaccine efficacy trial (45). Thus, we compared the responses to the three immunogens using this assay. We found that Con-S gave significantly lower responses to this peptide than the other 2 groups and that B.1059 and the trivalent mosaics were comparable (Fig. 8).

We next compared the ability of antibodies raised with the different immunogens to block the binding of sCD4 and various Env antibodies. Immunization with Con-S induced significantly higher levels of antibodies that blocked the binding of the bnAb 1b12 to the B.JRFL gp140, both 2 weeks after the second NYVAC boost and 2 weeks following a gp120 B.1059 Env boost, than of other antibodies (Fig. 9A). Both Con-S and mosaic Envs, but not B.1059, induced low levels of antibodies that blocked the glycan-specific antibody, 2G12; this was most evident 2 weeks after the gp120 protein boost at week 64 (Fig. 9D). Similarly, Con-S and mosaic vaccines induced antibodies that were able to block sCD4 binding, to a limited extent, and B.1059 did not. After the gp120 protein boosts, plasma from all three groups were able to block the C1-binding A32 antibody, particularly the mosaic vaccine, though the groups were not significantly different.

FIG 9.

Vaccine-elicited blocking antibody responses. Percent blocking of soluble CD4 and monoclonal antibodies binding to Envs by plasma from three different vaccine groups is shown. Responses from monkeys immunized with B.1059 Env are shown by blue circles, Con-S by red squares, and mosaic by green triangles. Blocking of the binding of 1b12 to B.JRFL (A), CD4 to JRFL (B), A32 to A244 D11 gp120 (C), and 2G12 to B.JRFL (D) is shown. All P values of <0.05 are shown, using a 2-sided Wilcoxon test to do a pairwise comparison of each group.

To determine if any other glycan-reactive Env antibodies could be blocked by these plasma, we tested the ability of the plasma to block the V1V2 glycan bnAbs CH01 and PG9 and of V3-glycan antibodies PGT125 and PGT128 to bind to AE.A244 gp120 (Fig. 10). We found that after the protein boost, mosaic and Con-S immunization, but not B.1059, induced antibodies that blocked the V3-glycan binding bnAb PGT128 (P = 0.008, 2-sided Wilcoxon test), with mosaic immunization inducing the highest level of MAb-blocking Abs. Mosaic immunization induced higher levels of PGT125-blocking antibodies than either Con-S or B.1059 (P = 0.02). Mosaic immunization also gave the highest levels of CH01 and PG9 blocking activity (Fig. 10C and D), reaching borderline significance compared with the B.1059-immunized group (P = 0.03). These data demonstrate the presence of antibodies that can sterically block these glycan antibodies but do not prove that glycan bnAbs were induced. Rather, specific glycan-reactive antibodies will need to be isolated to probe the vaccine-induced repertoire to address this point.

FIG 10.

Vaccine-elicited blocking antibody responses. Percent blocking of glycan-reactive antibodies to different Envs by plasma from three different vaccine groups is shown. Responses from monkeys immunized with B.1059 Env are shown by blue circles, Con-S by red squares, and mosaic by green triangles. Percent blocking of V3-glycan antibody PGT125 (A) and PGT128 (B) to bind AE.A244 gp120 is shown. Percent blocking of V1V2 glycan antibody CH01 (C) and PG9 (D) to bind AE.A244 gp120 is shown. All P values of <0.05 are shown, using a 2-sided Wilcoxon test to do a pairwise comparison of each group.

Antigen binding (Fig. 7 and 8) responses declined but generally remained detectable at week 62, 38 weeks after the second NYVAC boost. The protein boost at week 62 was able to bring both binding responses (Fig. 7 and 8) and antibody-blocking responses (Fig. 9 and 10) to higher levels than were seen after the second NYVAC boost. To statistically assess this, a Wilcoxon signed-rank test was used for the primary analysis, with a Benjamini-Hochberg corrected P value provided to adjust for multiple tests. The 15 animals from the three groups were evaluated together to determine if the general increase in antibody levels was statistically supported, and indeed an enhanced response was strongly supported in all cases shown in Fig. 7 to 10, except for CD4 blocking and gp70B.CaseA binding. The results from the statistical analyses are shown in Tables 3 and 4.

TABLE 3.

Comparison of antigen binding levels at week 26 and week 64^a

Antigen	Figure no. to which data correspond	P value
		Wilcoxon signed-rank test	FDR
M.mos 3.1 gp120	8	0.00012	0.0039
M.mos 3.2 gp120	8	0.000061	0.0027
M.mos 3.3 gp120	8	0.000061	0.0027
M.mos 3.1 gp140	8	0.00012	0.0039
M.mos 3.2 gp140	8	0.00012	0.0039
M.mos 3.3 gp140	8	0.000061	0.0027
B.1059 gp140	9	0.000061	0.0027
B.1059 gp120	9	0.000061	0.0027
CON-S gp140	9	0.000061	0.0027
CON-S gp120	9	0.000061	0.0027
gp70 B.CaseA V1V2	10	0.015	0.13

^aA Wilcoxon signed-rank test was used for the primary analysis, with a Benjamini-Hochberg corrected P value provided to adjust for multiple tests (FDR P value).

TABLE 4.

Comparison of levels of antibody/sCD4 blocking activity at week 26 to week 64^a

Antigen	Figure no. to which data correspond	P value
		Wilcoxon signed-rank test	FDR
1b12 × JRFL	11	0.0020	0.063
2G12 × JRFL	11	0.0057	0.063
A32 × A244 Δ11 gp120	11	0.00012	0.0085
CD4 × JRFL	11	0.19	0.37
CH01 × A244 Δ11 gp120	12	0.0020	0.063
PGT125 × A244 Δ11 gp120	12	0.00012	0.0085
PGT128 × A244 Δ11 gp120	12	0.00012	0.0085

^aA Wilcoxon signed-rank test was used for the primary analysis, with a Benjamini-Hochberg corrected P value provided to adjust for multiple tests (FDR P value).

Magnitude and breadth of vaccine-elicited neutralizing antibody responses.

Plasma samples from monkeys were tested in a TZM-bl assay against a panel of 16 viruses: 4 tier 1A, 3 tier 1B, and 9 tier 2 viruses. As a cautionary note, although BR025.9 is a primary isolate, it is quite neutralization sensitive and at the border of tier 1B and tier 2 classification; we call it tier 2 here, as described by Hraber et al. (40), but it is among the most sensitive tier 2 viruses. Seven were clade B, 2 were CRF01 (AE, recombinants, subtype E in Env), and 7 were clade C. Figure 11 shows the heat maps of 50% infective dose (ID₅₀) responses of plasma samples from three vaccine groups in the TZM-bl assay (panel A shows the inclusive cutoff, whereas panel C shows the more conservative, traditional cutoff). We analyzed responses calculated using both cutoffs described above, the traditional and conservative one and the more inclusive one. When three vaccine groups were compared for magnitude of antibody responses, Con-S yielded significantly more potent responses than B.1059 vaccine using the inclusive threshold, but a trend was evident only with the conservative 3× threshold (P = 0.012, inclusive; P = 0.084, conservative, permutation test). The mosaic was marginally higher than B.1059 (P = 0.045, inclusive; P = 0.027, conservative). Mosaic and Con-S were not statistically different using either threshold (Fig. 12, the top panel shows the traditional cutoff and the bottom panel the more inclusive one). The lognormal GLM confirmed the significant difference across vaccines (P = 0.012 with the inclusive threshold, P = 0.008 with the conservative), with Con-S yielding magnitudes on average 2.2 times higher than B.1059, and mosaics 2 times higher than B.1059. When we looked at the breadth of these neutralizing antibody responses, comparing the number of positive and negative responses per monkey in each group, a Kruskal-Wallis test suggests no one group was different than the others.

FIG 11.

Heat maps of neutralizing antibody responses. (A) Heat map of ID₅₀ responses from the TZM-Bl assay, using the inclusive cutoff. The color intensity is based on the log₁₀ values, and white indicates values that were lower than the baseline responses. Vaccinated animals are organized by group, and Env sensitivity is organized by behavior using hierarchical clustering. (B) Heat map of the A3R5 assay results, also using the inclusive cutoff; (C) heat map of the TZM-Bl assay results, this time using the traditional cutoff; (D) heat map of the A3R5 assay results, using the traditional cutoff.

FIG 12.

Magnitude of neutralizing antibody responses. Vaccinated animals are organized by group. Each vertical column of points represents the neutralizing potency against a panel of 16 Envs evaluated by the TZM-bl assay using the traditional threshold (A) and using the more inclusive threshold instead (B) for each animal. The magnitude of the log₁₀ ID₅₀ from the sera of each vaccinated animal against each protein is plotted on the y axis. Tier 1A Env ID₅₀s are indicated by dark-red dots, tier 1B by red dots, and tier 2 by blue dots. The dashed horizontal line represents the cutoff for detection; all values below that line are simply below background, and they are tiered stepwise to help visually distinguish the vaccine groups.

The results with the A3R5 assay were consistent with what we found using the TZM-bl assay. The A3R5 assay is still relatively new, and how to best interpret the levels of neutralization seen using this assay is still being evaluated; in particular, weak neutralizing responses can sometimes be detected in this assay, but the biological value of such responses is not yet well understood (46). In this study, the results were similar between the two assays, TZM-bl and A3R5 (Fig. 11 and 12). Con-S yielded significantly higher responses than B.1059 (P = 0.008, inclusive; P = 0.004, conservative), and Con-S responses were not significantly different than mosaic responses. Mosaic responses were statistically significantly different from B.1059 responses only when the conservative threshold was used (P = 0.15, inclusive; P = 0.004, conservative). The GLM confirmed the significant difference across vaccines (P = 0.001, inclusive; P = 0.0005, conservative). A Kruskal-Wallis test evaluating the number of positive responses per animal indicated at least one of the 3 groups was different than the others (P = 0.02); therefore, we went on to do pairwise comparisons. With an inclusive cutoff, Con-S yielded significantly more responses than both B.1059 (P = 0.009) and mosaic (P = 0.02), while mosaic was not significantly different than B.1059. In contrast, with a conservative cutoff, mosaic and Con-S groups both had significantly more responses (0.003 and 0.004, respectively) than B.1059. The vaccine group was a significant predictor of the number of responses through a binomial GLM (P = 0.002), with Con-S eliciting 1.9 times more positive responses than B.1059, and mosaic 1.6 times more responses than B.1059. In summary, by either assay, Con-S generally ranked the best in terms of neutralizing antibody responses, and both Con-S and mosaics performed better than B.1059 in terms of eliciting potent tier 1 responses and some low-level tier 2 neutralizing activity. The full ID₅₀ neutralizing responses are provided in Table S1 in the supplemental material.

DISCUSSION

In this study, we have demonstrated that both mosaic and group M consensus Con-S immunogens improved T cell and antibody responses relative to B.1059, a wild-type transmitted founder Env. Moreover, for induction of HIV neutralizing antibody responses, we found that the centralized gene Envs Con-S gave the best results, followed by trivalent mosaics; both were superior to wild-type B.1059 gp120 Env. In terms of induction of nonneutralizing antibodies against the V2 region, B.1059 and the mosaic vaccines gave significantly higher responses than Con-S.

For this rhesus macaque study, we chose optimized immunogens from each of three classes of HIV Env antigens: a natural transmitted-founder virus, a consensus, and a set of 3 complementary mosaics designed to be used as a polyvalent set that could provide optimal global coverage of potential T cell epitopes. Con-S was our choice of consensus, as it is global and has outperformed other consensus and ancestral sequences in vaccine evaluations conducted in small mammals (25). B.1059 was our choice of a T/F natural Env, as at the point of antigen design, it ranked best among T/F viruses in terms of global coverage of potential T cell epitopes. The global mosaics were designed based on the full data set of envelopes at the Los Alamos database, at the point of antigen design, and, like B.1059, are newly presented here. We used 3 rather than 2 mosaics in this study, as 3 mosaic sets both theoretically (11) and experimentally (29) showed better breadth of coverage; the number of mosaics included in a vaccine, however, needs to weigh production cost against the level of improved coverage as the number of antigens in the mix increases.

The cellular immune responses elicited by the 3-valent mosaic immunogens and Con-S had statistically improved magnitudes of responses across all clades than the responses elicited by the natural clade B B.1059 immunogen, and the mosaic immunogens ranked best among the three in terms of median response levels (Fig. 4). Responses with greater magnitudes across clades may mediate greater cross-recognition of diverse circulating forms of HIV-1 at the time of exposure to the virus. In addition, mosaic antigens are designed to include the most common variants of each potential T cell epitope, and these common variants likely represent the most fit immune escape variants. Thus, in theory, responses with greater cross-reactivity may also be able to block common and fit immune escape routes for the virus, if a vaccinee becomes infected.

The Con-S Env gp120 (this study, in macaques) or gp140 (a large study comparing many Env antigens in guinea pigs [25]) has induced detectable but low-level virus-neutralizing Abs against a subset of primary isolate viruses. This response was consistent across guinea pigs in the Con-S group in a study by Liao et al. (25), and the response of one animal in the Con-S group in this study, 40-11, was particularly promising (Fig. 12). The BR025.9 virus, the tier 2 virus with the highest susceptibility to Con-S-elicited antibodies, is a primary isolate that has been classified as tier 2 (40) but is near the boundary for tier 1b and tier 2 classification and is quite sensitive. Thus, the moderate level of response to this virus by Con-S-vaccinated monkey 40-11 should be interpreted with the overall sensitivity of this virus in mind. Highly potent tier 1 virus-neutralizing Abs were raised using either mosaics or Con-S as immunogens. The current notion is that potent bnAbs are subdominant clonal lineages of cells that will require repeated immunizations with immunogens that target both the early and late members of subdominant bnAb clonal lineage members (47). Nonetheless, there is recent evidence that tier 1 virus-neutralizing common Abs can indeed mediate immune pressure and neutralize a spectrum of easier-to-neutralize virus isolates that arise in chronic infections (48) (M. A. Moody, B. Korber, and B. F. Haynes, personal communication). Thus, Con-S Env remains an attractive component of a polyvalent Env immunization regimen.

The importance of vaccine-induced anti-Env antibody responses in preventing HIV-1 acquisition is gaining prominence post-RV144 Thai trial (49). It is notable that of the three immunization regimens tested in this study, the T/F B.1059 Env ranked as the highest in inducing the gp70V1V2 binding antibodies, although the responses in the B.1059 and mosaic groups were statistically indistinguishable, and both groups gave significantly higher responses than Con-S (Fig. 9). These types of antibodies were found to be a correlate of decreased infection risk in the RV144 vaccine protection trial in Thailand (45). It will be of interest to see if this regimen can confer any protection of challenged NHPs immunized with B.1059 Env.

Although mosaic immunogens are designed to optimize for contiguous T lymphocyte epitope population coverage responses, and indeed both Con-S and mosaics confer a greater breadth of T cell responses than a natural immunogen, it is interesting that these vaccine constructs also elicited improved HIV-1 neutralizing antibody responses. However, mosaic Env-induced antibodies potently neutralized only the easy-to-neutralize tier 1 viruses, with only weak sporadic responses to more difficult-to-neutralize tier 2 viruses; Con-S did slightly better against tier 2 viruses. Since tier 2 viruses are representative of the T/F viruses that are responsible for the global AIDS epidemic (50), it is likely that additional immunization regimens will be needed to generate a robust tier 2 HIV-1 neutralizing antibody response. Nonetheless, the profound impact of the increase from single-valent natural or consensus to 2- to 3-mosaic polyvalent vaccines on the strength of the tier 1 antibody response to diverse strains is, however, intriguing. It is possible that exposure of antibodies to epitope variants during their clonal evolution and affinity maturation may improve their breadth.