|Home | About | Journals | Submit | Contact Us | Français|
Simian immunodeficiency virus (SIV) infection of rhesus macaques causes immune depletion and disease closely resembling human AIDS and is well recognized as the most relevant animal model for the human disease. Experimental investigations of viral pathogenesis and vaccine protection primarily involve a limited set of related viruses originating in sooty mangabeys (SIVsmm). The diversity of human immunodeficiency virus type 1 (HIV-1) has evolved in humans in about a century; in contrast, SIV isolates used in the macaque model evolved in sooty mangabeys over millennia. To investigate the possible consequences of such different evolutionary histories for selection pressures and observed diversity in SIVsmm and HIV-1, we isolated, sequenced, and analyzed 20 independent isolates of SIVsmm, including representatives of 7 distinct clades of viruses isolated from natural infection. We found SIVsmm diversity to be lower overall than HIV-1 M group diversity. Reduced positive selection (i.e., less diversifying evolution) was evident in extended regions of SIVsmm proteins, most notably in Gag p27 and Env gp120. In addition, the relative diversities of proteins in the two lineages were distinct: SIVsmm Env and Gag were much less diverse than their HIV-1 counterparts. This may be explained by lower SIV-directed immune activity in mangabeys relative to HIV-1-directed immunity in humans. These findings add an additional layer of complexity to the interpretation and, potentially, to the predictive utility of the SIV/macaque model, and they highlight the unique features of human and simian lentiviral evolution that inform studies of pathogenesis and strategies for AIDS vaccine design.
The potential efficacy of human immunodeficiency virus type 1 (HIV-1) interventions, including vaccines, can be investigated by using the macaque/simian immunodeficiency virus (SIV) model of infection, wherein rhesus macaques (Macaca mulatta) are exposed to a macaque-adapted simian immunodeficiency virus (SIVmac). The SIV isolates in common use as vaccine and challenge strains originated in U.S. primate centers (2, 29) via accidental, incidental, or experimental transmission of retroviruses from captive sooty mangabeys (Cercocebus atys) to rhesus macaques, often followed by additional macaque passages. The endemic sooty mangabey virus (SIVsmm) is not overtly pathogenic in its natural host, but many SIVsmm-derived SIVmac strains cause AIDS in Asian macaques, as can native SIVsmm, SIVagm from African green monkeys, SIVmnd from mandrills, and SIVlho from L'Hoest monkeys, although adaptation may be required for full virulence (reviewed in reference 41). Natural SIV infections are often considered to be nonpathogenic in general, and indeed, there is little overt disease in the best-studied natural host/SIV pairings. We note, however, that the natural histories of most of the 40-some SIV strains have not been studied and are essentially unknown, that SIVcpz infection can cause AIDS-like disease in free-ranging chimpanzees (39), and that AIDS has, in fact, been reported to have occurred in a few “natural-host” African primates (reviewed in references 65 and 82), including a sooty mangabey (54). SIV-induced pathogenesis is therefore possible, albeit rare, in putatively well-adapted hosts.
The similar pathologies of HIV infection of humans and SIV infection of macaques (reviewed in references 9 and 41) are characterized by an initial high peak viremia, massive depletion of CD4+ T cells in the gastrointestinal tract during acute infection (10, 58, 83), lowering of viremia by 2 to 4 logs during chronic infection, progressive loss of CD4+ T cells, and eventual progression to AIDS. Furthermore, and critical to the relevance of vaccine studies in the macaque model, the dynamics of infection in low-dose SIV vaccine challenge models are similar to those of natural HIV infection of humans (40). These striking parallels, in both infection and disease, make SIV infection of macaques a powerful and informative animal model for HIV infection in humans. Animal models are extremely useful for vaccine studies, allowing control of the mode, dose, and timing of viral exposure, permitting any desired tissue sampling, and enabling efficacy testing via viral challenge to compare vaccine design strategies. Recent heterologous-challenge studies with macaques have provided new hope and direction for the HIV vaccine field: vaccine-elicited antibodies correlated with protection from infection (4), and persistent CD8+ effector memory cells localized in lymph nodes can provide protection from infection, in some cases, and stringent control of viremia and protection from disease in challenged animals that do become infected (20, 26). Nevertheless, despite the many similarities in the biology noted above, and the clear value of new insights the macaque vaccine model provides, simian viruses and simian immune responses differ significantly from their human counterparts (64), and the direct applicability of SIV/macaque challenge models to human vaccine studies is a matter of long-standing debate (29, 85).
In this study, we examined a potential complicating factor of the SIVmac/macaque model that could affect the interpretation of heterologous-challenge results as they relate to cross-reactive protection against the extraordinary population diversity of HIV (44). It has been noted that the genetic diversity of naturally circulating SIVsmm strains is roughly comparable to the diversity of the HIV-1 M group (1), and the predictive utility of SIV heterologous-challenge models rests in part on this observation. However, SIVsmm in sooty mangabeys represents a far more ancient lineage than HIV-1 in humans: the global diversity of the current HIV-1 epidemic harks back to a common ancestor about a century ago (43, 89), while SIVsmm origins in sooty mangabey go back many millennia (90). Additionally, the immunological environments of SIVsmm and HIV-1 are quite different: in sooty mangabeys, anti-SIV antibody titers, cytotoxic T lymphocyte (CTL) responses, and immune activation are lower than in HIV-1-infected humans or SIVmac-infected macaques (21, 35, 77; see also Discussion). Thus, although the levels of protein diversity observed in the two lineages are roughly comparable, they have evolved under different selective forces over different timescales, which could in principle give rise to patterns of amino acid diversity that reflect these distinct biological histories. We therefore investigated patterns of natural selection in HIV-1 and SIVsmm, finding distinct evolutionary patterns that could potentially affect vaccine-induced immune responses. Here we present new natural-isolate SIVsmm sequences and compare signatures of natural selection (both diversifying and stabilizing selection) to a comparable set of viral sequences drawn from the global HIV-1 pandemic.
Near-complete genomes were amplified by limiting dilution, cloned, and sequenced for 20 new SIVsmm isolates (Table 1), representing seven of the nine SIVsmm lineages previously recovered from U.S. primate centers (2), and for 4 previously isolated viruses (Table 2). Of the new isolates, 17 SIVsmm strains (representing lineages 1 to 6) were amplified from tissue culture supernatants (TC) of short-term sooty mangabey peripheral blood mononuclear cell (PBMC) cultures, in which primary SIVsmm isolates were propagated (23), while the remaining 3 (representing lineage 7) were amplified from uncultured sera (see Table S1 in the supplemental material). The latter had been sampled between 1980 and 1986, most probably during chronic infection. Using standard procedures (38, 68, 69), viral RNA purified from tissue culture supernatants or serum was reverse transcribed, and single-genome amplification (SGA) was performed by limiting dilution of cDNA followed by nested PCR. Nucleotide sequences of single-half-genome amplicons (HGAs) were determined by direct sequencing. One sequence per isolate was used for the study. For the 17 isolates of lineages 1 to 6, nucleotide sequences of HGAs overlapped in integrase by 49 bp (D215, FTq, and G932) or 65 bp extending from U5 in the 5′ long terminal repeat (LTR) to U3 in the 3′ LTR, thus including the complete viral proteome excepting the carboxyl-terminal amino acid and stop codon of Nef. When multiple HGA-derived sequences were available from a single animal, we favored inclusion of sequences with identity in the overlap between corresponding 5′ and 3′ HGA sequences and with intact open reading frames. For lineage 7, only single HGAs have been obtained to date for CFU212 (3′ HGA) and CFU226 (5′ HGA). Therefore, additional attempts were made to obtain SGA-derived subgenomic fragments in gag (1.4 kb), pol (2.5 kb), and gp41-nef (1.1 kb). Individual SGA-derived amplicons are identified in Table S1 in the supplemental material.
To assemble a data set of HIV-1 sequences to compare with the SIVsmm sequence set, a subset of sequences was selected from the Los Alamos National Laboratory (LANL) HIV database's web reference alignments (2007 edition [http://www.hiv.lanl.gov/content/sequence/NEWALIGN/align.html]). Starting with sequences with complete reading frames for Gag, Pol, Env, and Nef, sequences were down-sampled to the point where simple neighbor-joining trees of the HIV-1 sequences were superficially similar in clade size and distribution to trees inferred from the SIVsmm data set. A full-genome codon-based alignment of HIV-1, SIVsmm, and SIVmac sequences was constructed using curated sequence alignments from the LANL HIV database (http://hiv.lanl.gov) as a starting point; individual genes were extracted as needed from the final alignment. Phylogenetic trees were constructed from both nucleotide and amino acid data sets using Garli (95) with GTR and WAG models, respectively; for both amino acids and DNA, site-to-site rate variation was modeled with a gamma distribution with invariant positions (GTR + Γ + I; WAG + Γ + I). Base frequencies were estimated, and amino acid frequencies were fixed; random starting trees were used, with 16 replicates for DNA trees and 4 replicates for amino acid analyses. Programs from the Newick Utilities package (33) were used to process phylogenetic trees for presentation (e.g., rooting, scaling, and branch coloring).
Codon-aligned nucleotide sequence sets were checked for recombination using genetic algorithm recombination detection (GARD) (49). Alignments with putative recombinants were divided into phylogenetically coherent partitions that were submitted individually to MEME (48, 62), FEL, and IFEL (47) analyses on the Data Monkey web server (http://www.datamonkey.org).
Mosaic sequences were generated from sets of SIV isolates derived from naturally infected sooty mangabeys (or viruses minimally removed from such isolates). Mosaics were generated using a slight modification of our previously reported method (19): instead of generating a multiple-sequence mosaic cocktail in a single step, we first generated a single mosaic sequence and then sequentially added three more sequences, at each step using the sequence(s) from the previous step as a “fixed” sequence(s), i.e., preexisting cocktail members. The intent of this alteration was to allow incremental testing of increasing mosaic cocktail size. The first generated mosaic alone would be comparable to a single consensus sequence; adding cocktail sequences in order would increase coverage in a stepwise manner. Compared to that of simultaneously generated mosaics, the incremental cocktails' 9-mer coverage was reduced very slightly (data not shown). Incremental cocktails containing 4 sequences were generated for each of gag, pol, env, and nef; coverage data are presented here for 2-sequence cocktails only.
Previous studies of SIVsmm diversity and evolution (2, 93) suggested that the levels of within-group sequence diversity were roughly comparable between SIVsmm and HIV-1 M group viruses, implying that SIVsmm vaccine efficacy models could approximate real-world conditions for HIV-1. We performed full-length sequencing of 20 new SIVsmm strains from sooty mangabeys (Table 1) and 4 existing isolates (Table 2) and combined these with previously available sequence data. This data set was assembled with the intent of exploring the evolutionary pressures on SIVsmm and of designing SIVsmm mosaic vaccine inserts (19) to enable efficacy testing of mosaic vaccines in the rhesus macaque model (mosaic proteins are artificial proteins produced by a computational design strategy that optimizes coverage of potential epitope variants in a diverse population of viruses). To exclude effects of evolution in nonnatural hosts (i.e., evolution during pathogenic passage in macaques), and so that standard laboratory stocks could serve as heterologous challenges, we excluded macaque-adapted challenge strains (such as SIVmac239, SIVmac251, and SIVmacE660) and their derivatives from the mosaic design input data sets.
For comparisons of HIV-1 and SIVsmm evolution, and of potential epitope coverage of various potential vaccines, we selected a set of HIV-1 M group sequences comparable in size and diversity to the SIVsmm set, including a cross-section of clades (see Materials and Methods). Like HIV-1, SIVsmm has a well-defined clade structure with phylogenetically distinct lineages (Fig. 1), though this may be due in part to founder effects (since U.S. primate center sooty mangabey populations represent a small, nonrandom sampling of wild African populations). Of note, a single SIVsmm sequence with a very long branch length, SMM.SL.1992.SL92B.AF334679, isolated from a sooty mangabey, is typical of SIVsmm in Env and Nef but highly divergent in Pol and Gag (Fig. 2a); it is therefore thought to have originated via recombination between an SIVsmm strain and an SIV of unknown lineage (P. Sharp, personal communication).
We inferred maximum likelihood phylogenetic trees for the Gag, Pol, Env, and Nef genes for both the SIVsmm and HIV lineages, using both DNA and amino acid data (Fig. 2a). Overall differences between the two viral groups were evident in this simple visualization, suggestive of distinct selective pressures acting on the different coding regions over evolutionary time. The inferred evolutionary rates differed between the two viruses. Moreover, the ratios of amino acid branch lengths to DNA branch lengths varied from gene to gene and between the two viral groups (Fig. 2). There are distinct patterns of relative branch length between the four different genes in the two lineages: for HIV-1, amino acid branch lengths exceeded DNA branch lengths for Gag, Nef, and Env, while Pol amino acid branch lengths were slightly shorter than DNA branch lengths. This contrasted with SIVsmm: Pol, Gag, and Env all had amino acid branch lengths shorter than DNA branch lengths (substantially so for Pol and Gag), and Nef had amino acid branch lengths only slightly longer than DNA branch lengths. At the amino acid level, as well, SIVsmm did not match HIV-1 stereotypes: Env protein diversity in SIVsmm is much lower than in HIV-1, and while HIV-1 Gag is more diverse than HIV-1 Pol, SIVsmm Pol is more diverse than SIVsmm Gag (Fig. 2b). These distinctions in phylogenetic patterns prompted us to investigate and compare the selective regimes that gave rise to the high diversity in these two retroviral lineages.
To explore the potential overall differences in pressure on different genes, we analyzed the spatial occurrences of synonymous and nonsynonymous mutations using SNAP (45, 46), which plots the cumulative occurrence of each type of substitution from start to end of a gene. In the absence of regional differences in selective pressure, the slope is linear (as is generally the case for synonymous substitutions [Fig. 3b]). Regions of positive pressure (diversifying selection) yield a steeply rising curve in the nonsynonymous plot (Fig. 3a), showing mutation accumulation, while regions under strong negative (stabilizing) selection are level. SIVsmm Pol, Env, and Gag genes have accumulated many more synonymous substitutions than their HIV counterparts (Fig. 3b), as would be expected in a much older epidemic, in which silent mutations may be nearly saturated. Nonsynonymous substitutions, on the other hand, appear to have accumulated much more rapidly, relatively speaking, in HIV-1 Env than in SIVsmm Env (Fig. 3a): compared to HIV-1 Pol, SIVsmm Pol appears to be subject to slightly reduced stabilizing selection, while SIVsmm Gag has both fewer nonsynonymous substitutions and more synonymous substitutions than its HIV-1 counterpart. In several regions of the genes in this study, overlapping reading frames impose additional selective constraints, complicating the analysis (such regions are retained but indicated with horizontal bars in Fig. 3 and and4);4); despite these complications, some potential local regional differences in the nonsynonymous mutation rates between gene regions in HIV and SIVsmm are apparent (Fig. 3a).
We next compared patterns of selection at a codon-by-codon level, measuring phylogenetically corrected ratios of nonsynonymous to synonymous changes (dN/dS ratios) by codon in SIVsmm versus HIV-1 in Gag, Pol, Env, and Nef using a mixed-effects model of evolution (MEME) developed by Kosakovsky Pond and colleagues (http://www.datamonkey.org/help/meme.php). MEME provides a statistical framework to identify positive selection in particular positions in sublineages within a phylogenetic tree even when positive selection is not evident across the entire tree (48, 62). To highlight regional differences in selection within each protein, we represent this analysis in a site-by-site graphic (Fig. 4) rather than averaging across entire genes.
Several patterns emerge from these analyses. In general, except for some regional hot spots, positive selection is far less common in SIVsmm than in HIV-1. Of the four proteins studied, Nef was the only one that had evidence of comparable selective pressures in HIV-1 and SIVsmm (Fig. 3 and and4d).4d). There are sites scattered throughout HIV-1 Gag that show recurrent selection throughout the phylogenetic tree (i.e., support for positive selection on a high proportion of branches [Fig. 4a]), while the SIV capsid gene, p27, showed no sites with statistical evidence for selection (see Discussion). Similarly, other relatively conserved proteins in these viruses (reverse transcriptase [RT] and protease) showed multiple sites evolving under positive selection in HIV but little or no evidence of selection in SIV.
In contrast, the upstream transframe (TF) region of Pol polyprotein, which precedes protease and overlaps with Gag in a different reading frame, has many sites with strong selective signals in both SIV and HIV (Fig. 4b); it is also subject to length variation. This small stretch of protein is autocatalytically cleaved by protease and is not very constrained in terms of mutational fitness costs (51), so many changes might be tolerated; however, the selective signal from MEME may be an artifact caused by overlapping reading frames (see Discussion below).
Overall, HIV-1 Env appears to be under greater selective pressure than SIVsmm Env, with apparent exceptions of V1 and two regions in the SIVsmm cytoplasmic tail (Fig. 4c). The region spanning V1 is hypervariable and in HIV-1 is more commonly mutated via insertions and deletions (indels) than via base substitutions (88). Frequent indels in V1 introduce extensive length variation and overlays of nonhomologous regions. Therefore, the apparently regional increase in variation in SIVsmm relative to HIV-1 is likely to be misleading, since the dN/dS ratio as a measure of evolutionary selective pressure is based on aligned codons and hence does not take into account indels and length variation. While SIVsmm V1 also shows length variation, it is to a much lesser degree and the region is much more readily aligned. Consequently, selective pressure on the HIV-1 V1 region is likely to be underestimated relative to SIVsmm by this measure. In contrast, the focused regions of positive selection in the cytoplasmic domain of SIVsmm gp41 are in regions that are readily aligned; like the Pol TF region, the cytoplasmic tail of Env may tolerate change, though it is subject to structural constraints (79).
It is likely that some regionally focused MEME-detected increases in positive selection are spurious: when reading frames overlap, stabilizing selection in one reading frame will suppress apparently synonymous substitutions in the other frame(s) and thereby give rise to a false signal of positive selection. Most regions of strongly elevated SIVsmm selection detected by MEME in these analyses, including the Pol TF region and the Env cytoplasmic tail, do in fact map to areas with active genes in multiple overlapping reading frames (Fig. 4), supporting this hypothesis.
Coverage of potential T cell epitopes based on sequence substrings (i.e., k-mers) provides a metric of sequence diversity that is likely to be both immunologically and epidemiologically relevant: the overlap of amino acid 9-mers between vaccine candidates and target viral populations provides a rough indication of the potential for broad CTL-mediated protection (19, 53). Thus, 9-mer coverage of the different genes of the SIVsmm and HIV-1 provides a metric for the possible breadth and depth of potential vaccine-induced epitope responses. As we have seen before (19), we find that in comparisons of within-clade, cross-clade, and overall lineage coverage by different putative vaccine immunogens (Fig. 5 and and6;6; see also Fig. S1 in the supplemental material), single-natural-sequence immunogens are very limited in terms of population coverage of potential epitope diversity in natural strains; polyvalent mosaic designs can provide dramatic increases in coverage of natural circulating strains. Theoretical k-mer-based estimates have consistently translated to cross-reactive potential of immune responses in macaque models, as well as in mice (5, 42, 71, 72).
As there is a particular interest in Gag and Env as potential immunogens (44), we compared potential epitope coverage for these two proteins, simulating homologous and heterologous challenge experiments in SIV and within-clade and cross-clade vaccination/exposure in HIV-1. For evaluation of coverage in challenge experiments, we used SIVmac sequences obtained following low-dose infection of macaques (40). In Fig. 5, we present 9-mer coverage distributions based on target virus populations, showing the potential coverage of a given population by a given vaccine. For HIV-1, we compared coverage of groups of B-clade and C-clade sequences and of a diverse HIV M group sequence set by, in turn, a single B-clade sequence, a single C-clade sequence, or a 2-sequence mosaic HIV cocktail (Fig. 5a and andc).c). For SIV, we calculated coverage of swarm sequences of SIVmac239/251 and of SIVmacE660, as well as our diverse SIVsmm population sequence set (as a comparison to the HIV-1 M group [Fig. 5]), evaluating SIVmac239, SIVmacE660, and a 2-sequence mosaic SIV cocktail as candidate vaccine inserts.
In HIV-1, for both Env and Gag, cross-clade coverage is poor, within-clade coverage is moderate, and mosaic coverage is superior to within-clade coverage (Fig. 5a and andc).c). For both SIVmac239 and SIVmacE660 swarm sequences, coverages of cross-clade (“heterologous challenge”) and within-clade (“homologous challenge”) are both considerably higher (and more tightly peaked) than the corresponding HIV-1 coverages: within-clade coverage is very high, cross-clade coverages are moderate, and mosaic coverage values for SIVmac239 and SIVmacE660 are between cross-clade and within-clade values. Compared to mosaic single-clade coverage values for HIV-1, SIVsmm mosaic coverage values are equivalent for Gag and much higher for Env (Fig. 5b and andd).d). SIVsmm mosaic coverage of the diverse SIVsmm data set is substantially greater than the comparable mosaic coverage of the HIV-1 M group.
The greatest potential for high-coverage optimized vaccine immunogens appears to be at moderate levels of variability, e.g., for SIVsmm and HIV-1 Gag, SIVsmm Pol, and the more conserved SIVsmm Env (see Fig. S1c, d, g, k, and l in the supplemental material). The highly conserved HIV-1 Pol gene, which is reasonably well covered by HXB2 (see Fig. S1i), is more broadly covered by the 2-sequence mosaics (see Fig. S1k), but the advantage is more dramatic for Gag and SIVsmm Pol. The much more variable HIV-1 Env and both Nef proteins are covered at only a moderate level by the mosaics (see Fig. S1h, o, and p), though there is still a substantial improvement compared to the single-sequence immunogens.
Intriguingly, the potential epitope diversity that has evolved in HIV-1 in less than 100 years is more extensive than the diversity found in SIVsmm, which has evolved over thousands of years. An SIVmacE660/SIVmac239 vaccine/heterologous-challenge pair provides better potential T cell epitope coverage for both Env (Fig. 5b, top and center) and Gag (Fig. 5d, top) than either within- or between-clade single HIV immunogens. In contrast, a 2-sequence mosaic cocktail yields potential epitope coverage above 60% for SIVmac239/251 Gag (Fig. 5d, top), a value that is typical of 2-sequence mosaic coverage of M group isolates.
Potential epitope coverage of individual viral isolates by a single natural sequence is limited; essentially, the only viral strains well covered by single natural sequences are those to which they are most closely related. In addition to the distributions of coverage values within sequence sets (Fig. 5), we also present the phylogenetic distribution of coverage for different candidate immunogens for all four genes (Fig. 6 features Gag as an example; Fig. S1 in the supplemental material also includes Env, Pol, and Nef). As expected, there are striking gene-to-gene differences in the proportion of well-covered viral isolates in terms of amino acid 9-mer coverage: Gag and Pol are much more broadly covered than Env or Nef. Notably, the high autologous coverage of natural sequences decays rapidly with increased phylogenetic distance: HXB2 has only moderate coverage of even closely related B-clade sequences and poor coverage of viruses from other clades (Fig. 6c; see also Fig. S1i, j, m, and n in the supplemental material). The 2-sequence mosaic cocktails, in contrast, maintain a moderate level of coverage over disparate clades (Fig. 6d; see also Fig. S1k, l, o, and p in the supplemental material and discussion of isolate breadth below).
Our original purpose in sequencing this collection of SIVsmm viruses was to provide a diverse set of SIV sequences to enable the design of SIV mosaic vaccine antigens (19) to allow efficacy testing of the mosaic T cell vaccine concept in an SIV vaccine/SIV challenge animal model. Since HIV-1 itself does not infect macaques, protective-efficacy tests of the mosaic HIV-1 vaccine concept require either an SIV mosaic vaccine or the use of a SHIV (HIV Env cloned into in a SIV backbone) challenge, which is limited and only enables the study of Env vaccine antigens. The strikingly distinct evolutionary pathways of SIVsmm and HIV-1 we present here became evident during sequence analysis for the SIV mosaic design. The caveats we raise should be considered when interpreting more traditional heterologous-challenge/vaccine approaches as well as when using the SIV-mosaic strategy we were interested in testing. To assess the potential impact of distinct HIV-1 and SIVsmm evolutionary trajectories in the context of vaccines of experimental interest, we assessed, for both SIVsmm and HIV-1, the 9-mer coverage of two natural sequences and a two-sequence mosaic immunogen. For SIVsmm, we compared the common SIV vaccine/challenge strains SIVmacE660 and SIVmac239 and our newly designed SIV mosaic vaccine set against our collection of diverse natural SIVsmm isolates (the most comprehensive whole-genome sequence set currently available). For HIV-1, we compared a single natural B-clade and single natural C-clade sequences and a set of HIV-1 mosaic vaccine immunogens that are now approaching human phase I safety and immunogenicity trials against a set of HIV-1 M group isolates.
Genetic (DNA-based) distances were roughly comparable between the two viruses, excepting SIVsmm Pol, which has larger distance values (Fig. 2). At the protein level, however, the SIVsmm lineage is generally less diverse, and apparently under less positive selective pressure (or under greater fitness constraints), than the HIV-1 M group. Furthermore, inferred positions subject to positive selection were distributed differently among and within different proteins in SIVsmm and HIV. The differences in positive selection between HIV-1 and SIVsmm Gag were particularly striking: there are sites scattered throughout HIV-1 Gag that show recurrent selection throughout the phylogenetic tree (i.e., support for positive selection on a high proportion of branches [Fig. 4a]). In contrast, SIVsmm Gag was under substantially weaker positive selection (or stronger negative selection, i.e., greater fitness constraints) than Pol: the SIVsmm p27 capsid (equivalent to HIV-1 p24) showed very little variation at the amino acid level and no evidence of positive selection at all (Fig. 4a). Capsid (p24/p27) is one of the more conserved proteins in both HIV and SIV, and certain HIV p24 epitopes are targets for HLA-restricted T cell responses associated with viral control and long-term survival (67, 80). Despite the overall conservation of HIV-1 p24, these epitopes still vary under CTL-mediated immune pressure, and it has been hypothesized that immune escape from these epitopes comes at a high fitness cost for the virus, contributing to the beneficial effect associated with particular HLAs (60). The lack of inferred selection in SIVsmm p27 suggests either that the SIVsmm capsid protein is generally less tolerant of mutation than the HIV-1 capsid or that mangabeys lack the potent Gag-directed immune responses seen in humans (67, 80) and in macaques (55, 86). The latter explanation appears probable: although sooty mangabeys make CTL responses to SIV infection that are strong enough to drive viral escape (34, 36), and although Gag-directed CTL responses in particular do indeed exist in sooty mangabeys (59, 84), both the magnitude and breadth of CTL responses are reduced in SIVsmm-infected sooty mangabeys compared with those in HIV-infected humans (15). Furthermore, SIVsmm-infected sooty mangabeys have lower neutralizing antibody titers than HIV-infected humans (35, 52) or SIVmac-infected rhesus macaques. In SIVagm-infected African green monkeys (56), another “natural host,” anti-p27 antibody responses are also much weaker than in rhesus macaques; this may be a general feature of nonpathogenic primate lentiviral infections (28, 35). Finally, CD8+ T cell depletion experiments with SIVsmm-infected sooty mangabeys (6) showed smaller effects on viral load than cognate experiments with macaques (32, 57, 75) and African green monkeys (22). These combined data suggest that immunological pressure on SIVsmm p27 is much lower than on HIV-1 p24. Therefore, the observed variation in SIVsmm more likely results from gradual accrual of mutations over time (i.e., genetic drift) than from rapid selection for immune evasion. In contrast, rapid immune escape is a major aspect of positive selection in HIV-1: most early mutations are concentrated within T cell epitopes (17, 18, 25, 27), viruses within individuals continually evolve to escape from neutralizing antibodies during chronic infection (3, 61, 91), and HLA imprinting is evident in viruses circulating in different human populations (37, 67).
The HIV-1 M group has likely been evolving in the human population for on the order of 100 years (43, 89). The much greater age of the “natural host” epidemics in African monkeys, and in sooty mangabeys in particular (90), is consistent with the greater extent of silent mutations we observed in SIVsmm, and it would furthermore allow the evolution of higher degrees of viral tolerance (virus/host coevolution for reduced pathogenicity). Sooty mangabeys, like African green monkeys, rarely progress to AIDS despite high-intensity SIVsmm infection; one mechanism by which tolerance is achieved in mangabeys is reduced immune activation (8, 15, 16, 74). Rates of progression to AIDS are associated with chronic immune activation (24), and the failure to downregulate interferon-stimulated genes after acute infection appears to be restricted to the pathological infections that occur in macaques and humans (8). In any case, the combined immunological and natural selection data argue strongly that the nonpathogenicity of SIVsmm in mangabeys is not due to a more effective immune response, and SIVsmm diversity appears to have evolved under lower immune pressure than HIV-1 diversity.
Diversity is of course a central problem in HIV vaccine design. A vaccine that protects against (or mitigates) HIV-1 infection must induce cross-reactive immunity against pandemic variants. We have proposed mosaic vaccines as a strategy for inducing sufficient numbers of T cell immune responses that cross-react with many different HIV-1 variants (19).
To quantify T cell cross-reactivity, we previously distinguished between breadth and depth of T cell responses (5), breadth referring to the summed quantities (counts) of different epitope loci recognized and depth referring to the recognition of multiple variants at individual epitope loci. In previous experimental work (70, 71, 72), we assessed the cross-reactivity of vaccine-induced responses by comprehensively testing peptides derived from diverse HIV-1 isolates. We here introduce the theoretical counterpart, the concept of “isolate breadth,” meaning the proportion of viral isolates from a given population for which a vaccine candidate has potential epitope coverage above a given threshold. A vaccine immunogen (or immunogen set) with high isolate breadth will match large portions of many sequences in a target population, and hence the responses it induces are more likely to cross-react with epitopes of an infecting virus. As expected based on earlier findings, a stark contrast in isolate breadth is evident between the 2-mosaic immunogen cocktails and the single “standard strain” sequences (Fig. 6; see also Fig. S1 in the supplemental material). The contrast is consistent within both viruses and all four genes tested, not only in the highly variable Env and Nef genes but also in the relatively more conserved Pol and Gag. Since multiple vaccine-induced epitope responses per gene appear to be correlated with protection against stringent heterologous challenge in macaques (55, 86), and the potential epitope overlap of any single natural sequence with any likely infective strain is low (72), optimized high-coverage immunogens such as mosaics are much more likely than single-sequence natural immunogens to have protective efficacy against HIV-1 exposure “in the wild.”
The importance of sufficient diversity in SIV heterologous-challenge studies must be emphasized: although SIVmac239 and SIVmac251 have been described by some authors as heterologous challenge pairs, an SIVmac251 challenge of SIVmac239-vaccinated animals (for examples, see references 14 and 31), or vice versa, cannot be considered heterologous. SIVmac239 and SIVmac251 originated in the same individual macaque (Mm251-79) and are far more closely related than circulating HIV-1 sequences even within the same clade (Fig. 1); SIVmac239 is a clone isolated after additional macaque passages of SIVmac251 (11, 30, 63). For a heterologous challenge to resemble the distances observed in the HIV circulating population, more distant viruses must be used as vaccine/challenge pairs (for instance, SIVmac239/SIVmac251 vaccination followed by SIVmacE660 challenge [7, 50, 66, 92]). SIVmacE660 is phylogenetically distinct from the SIVmac251/239 lineage (Fig. 1), and the genetic distance between the SIVmacE660 and SIVmac239/251 lineages is roughly comparable to the distances between HIV-1 strains circulating in the human population (93); the criteria of phylogenetic distinctness and genetic distance should always be evaluated for proposed vaccine/challenge pairs.
Using natural variation from SIV-infected sooty mangabeys (1) enables comparisons between polyvalent design strategies that attempt to more comprehensively address diversity and other, more traditional, polyvalent or monovalent natural-strain immunogens. Based on this premise, we gathered and sequenced novel SIVsmm strains, combined the results with existing data, and, using the full data from this augmented set, designed an SIVsmm polyvalent mosaic vaccine, which will be tested for efficacy in macaques (G.J.N., study in progress). Mosaic vaccines have been shown to induce greater numbers of epitope responses and enhanced cross-reactivity of responses (5, 71, 72) compared to those of monovalent immunogens; the mosaic design presented here will enable us to test whether either or both of these improvements in immunological response may result in effective T cell responses to pathogenic heterologous viral challenge.
Considerable thought has been applied to evaluating the differences between the macaque and human immune responses to immunodeficiency viruses in the context of vaccine development (reviewed in reference 76). In contrast, we are here attempting to highlight the potential importance of the viral component in assessing the applicability of the SIV/macaque model. The differences between the immune responses of both macaque and human nonnatural hosts are of course significant, but while macaques' immune responses to SIVsmm may indeed be more human-like than those of mangabeys, the diversity against which they are challenged appears to be derived from a much different set of selective pressures, and such differences may affect heterologous challenge. For instance, as we discuss above, SIVsmm capsid (p27), Pol RT, and Env gp120 appear to have evolved under low immunological pressure in sooty mangabeys, and they show reduced signatures of selection. The corresponding HIV-1 proteins, in contrast, have been subjected to intense immune pressure, and HIV-1 population diversity has adapted, to some degree, to the immunological diversity of its human host populations (67, 73). Vaccine-induced CTL responses in macaques (e.g., to the SIVmac CM9 Gag epitope ) can be correlated with reduced viral load; furthermore, the overall number of Gag-directed CTL responses is correlated with viral control (55, 86), and CTL responses can control viral replication in the absence of neutralizing antibodies (87). However, the relative benefits of a Gag-directed response against SIV in macaques (e.g., versus Nef- or Env-directed responses) might not be paralleled in a human population: whether particular benefits of particular responses to particular proteins observed in SIV/macaque heterologous-challenge models will translate to HIV in humans remains an open question. This is because the sequence diversity in SIVsmm-derived heterologous macaque challenges is not adapted to the macaque immune system, so even if the immune responses were qualitatively similar between macaques and humans, the viral diversity is not. Identical considerations apply to neutralizing antibody responses, since SIVmac Env protein diversity has largely evolved in the absence of strong antibody responses.
In short, the origin of interstrain diversity in macaque-adapted SIV strains is different from the origin of interclade and within-clade HIV diversity. HIV-1 diversifies under immune pressure and selection for rapid transmission; as we discuss above, SIVsmm appears to have diversified under reduced immune pressure, as well as long-term selection for low pathogenicity (78). Throughout the spread of the HIV-1 pandemic, the initial low diversity of small viral founder populations has expanded dramatically (both within and outside Africa) via continuous passage in hosts with highly active immune systems. Laboratory SIVmac variants originated from sooty mangabey-adapted lineages, with a history of reduced immune pressure, and they have not been continuously passaged in a host with strong immune responses. Therefore, even if sequence diversity between two SIVsmm or SIVmac isolates arithmetically approximates the diversity between two HIV-1 isolates, it may not be functionally equivalent in terms of induced immune responses, and the degree of heterology against which a vaccine might be protective (i.e., the breadth of protection) may not be easily predicted from a macaque model.
Nonhuman primate models are nevertheless an invaluable tool for studying immune responses to HIV, and careful use of the various SIV/macaque models will advance the imperative goal of a protective HIV vaccine. However, the differences in immunology, pathogenesis, and diversity (29, 85), as well as the distinctive evolutionary pressures in the SIVsmm and HIV-1 M groups discussed here, should be considered when extrapolating from SIV/macaque experimental results to HIV/human vaccine applications.
Many of the full-length SIVsmm sequences listed in Table 1 and in Table S1 in the supplemental material were generated by the late Matthias Kraus; other viral isolates (Table 2) were kindly provided by Martine Peeters and Preston Marx. We thank Michael Worobey and Paul Sharp for useful comments on the manuscript.
This research was supported by the National Institutes of Health via the following grants: AI-067854 (CHAVI) from the Division of AIDS, NIAID (W.F., B.T.K., G.M.S., B.H.H., M.L.S., and Y.L.), R01 AI-065325 (C.A., I.P., and R.G.), and R37 AI-50529, R21 AI-087383, and P01 AI-088564 (G.M.S., B.H.H., M.L.S., and Y.L.), as well as by the Los Alamos National Laboratory Directed Research and Development program (W.F. and B.T.K.), the Bill and Melinda Gates Foundation (G.M.S., B.H.H., M.L.S., and Y.L.), and the intramural research program of the Vaccine Research Center, NIAID, NIH, through NIH/DOE interagency agreement NIH Y1-A1-8309 (W.F. and B.T.K.).
Published ahead of print 10 October 2012
Supplemental material for this article may be found at http://jvi.asm.org/.
This article is dedicated to the memory of Norm Letvin, whose many questions originally stimulated our efforts on this project.