Five naturally SIV-infected SMs (Table 1) were sampled three times over a 2-year period. Viral RNA in plasma obtained in 3/99, 5/99, and 5/01 was measured by a real-time RT-PCR assay designed to quantitatively detect the diverse SIVsm variants [23]. Time points were chosen so that evolution could be assessed over both shorter and longer time intervals. Viral load averaged 1.5 × 10⁶ SIV RNA copies/ml plasma, and fluctuated modestly over the 2-year period (Figure 1). No clinical signs of AIDS were observed in any of the infected SMs over the study period.

Multiple V1V2 env clones (range 15–29) and p27 gag clones (range 5–19) were sampled from each animal at each time point (Genbank Accession numbers AY733102-AY733566). Env and gag were chosen for analysis since they were thought to represent the extremes of diversity in SIV populations. These genes also differ in how the immune system detects them, with env V1V2 being exposed primarily to neutralizing antibodies [38] and gag p27 being recognized mostly through cellular immune responses [39]. The number of individual viral sequences analyzed (Table 1) combined with the sampling of variants over a short time interval (2 months) and a longer time interval (2 years) exceeds that reported in previous studies of SIV diversity in natural hosts [40-43].

To characterize the overall evolutionary dynamics of natural SIV variation, we built maximum likelihood trees of both env V1V2 (Figure 2A) and gag p27 (Figure 2B) sequences. The SIVsm variants from each SM formed distinct clades in both genes, and the env and gag trees showed the same relationship between virus populations of the 5 animals. These results demonstrate that each host harbors a phylogenetically distinct population of SIVs, presumably as the result of infection with distinct viral populations and subsequent host-specific viral evolution.

The translated env aa sequences (FJo, Figure 3; data from all animals can be obtained from THV) demonstrate significant V1V2 heterogeneity, including heterogeneity in numerous predicted N-glyc sites (NXS/T, where X can be any aa but proline). Considerable V1 length variations were observed (Table 2 and for example, Figure 3), such that alignment of this region required manual adjustment, and may not represent precise homology. There were no trends in V1V2 sequence length variation over time (data not shown). Gag aa alignments (available from THV) showed significantly less aa variation reflecting its highly conserved nature.

Pairwise nt and aa diversity was calculated after removing regions of uncertain homology (gap-stripping) in V1, such that the values obtained for intra-host diversity represent minimum values. Average pairwise aa diversity was high in env V1V2 (average: 5.6%, range: 0 and 37.7%; Table 1) and low in gag p27 (average 1%; range: 0 and 7.1%, data not shown). The minimal diversity detected in gag, which was amplified under identical conditions, confirms that the observed V1V2 diversity is not the result of PCR-introduced mutation. In individual animals, the magnitude of nt and aa diversity did not change significantly over the 2-year observation period (Table 2). However, there appeared to be animal-to-animal variation in the extent of V1V2 diversity, with animals FFj and FDo exhibiting lower V1V2 nt and aa diversity than FJo and FBo (ANOVA p < 0.01, with Bonferroni adjustment). Nt and aa diversity were not correlated with viremia, suggesting that mechanisms other than or in addition to the magnitude of virus replication determine the extent of viral diversity. We cannot rule out that reduced diversity in FFj and FDo are the result of infection with less diverse virus populations.

Although the magnitude of sequence diversity did not change over time, it was likely that env sequences at later time points had diverged from those sampled earlier. To investigate the temporal pattern of sequence evolution within each animal, all available samples from all three time-points for each animal were pooled and analyzed by maximum likelihood (Fig. 4; FQi). Sixteen of the nineteen (85%) bootstrap-supported clades from FQi contain variants from a single time point only. This pattern was repeatable amongst variants from all other animals; 100%, 80%, 69%, and 63% of bootstrap supported clades consisted of a single time point in animals FDo, FFj, FJo, and FBo, respectively. In an analysis of random trees, the number of matching time-point sequences that comprise a monophyletic group showed a Poisson distribution; 86% of variants did not form monophyletic clades with any other matching time-point variant (i.e., these sequences stood alone). Thus, the observed temporal clustering of SIVsm viral populations does not occur by chance alone (Kolmogorov-Smirnov test, p < 0.01).

Temporal phylogenetic structure in V1V2 suggested that continual V1V2 diversification was occurring. To look for evidence of positive selection, dN and dS were calculated at each site and averaged over a 3-codon sliding window for VIV2 (Fig. 5A) or 30-codon sliding window for p27 (Fig. 5B). These results confirmed that dN-dS>0 (p = 0.003, t-test) in V1 (aa's 25–55) in all animals, indicating positive selection. For p27, the same test showed that dS>dN along this gene (t-test, p < 0.001), indicating that purifying selection limits its diversity. V1 was consistently found to be under significant positive selection in all animals, except FFj (data not shown). By contrast, the few aa changes in p27 sequences in the different animals over time appeared random in nature except for a single partially fixed mutation in FDo.

Up to 10 N-glyc sites are contained within the SIVsm V1V2 regions sequenced in this study. In multiple locations overlapping consensus motifs (aa's 42–44, 52–54, and 95–107) are present, such that the exact site of glycosylation varies (Fig. 3). These overlapping consensus motifs are in particularly diverse regions of V1V2 and in regions of strong positive selection.

V1V2 clones from the five SMs contained variable numbers of N-glyc sites, ranging from 3 to 10. The average number of N-glyc sites among all animals was 7.2. There was no clear pattern of increased or decreased V1V2 env glycosylation with time. However, the mean number of N-glyc sites for FFj and FDo (7.8 and 8.2, respectively) was significantly higher than the other animals (average between 6.5 and 6.9; ANOVA, Tukey B, p < 0.001). An additional N-glyc site is found in V1 in the majority of sequences in FFj and FDo at position 45, but not in the other animals. There was also a smaller range of N-glyc sites per set of sequences in FFj and FDo (6–9) compared to other animals (3–10). As described, the FDo and FFj SIVsm populations were less diverse and had lower average dN compared to the virus populations found in the other 3 animals (Table 2). A significant inverse correlation between the mean number of N-glyc sites and both pairwise nt diversity and nonsynonymous substitutions was observed when combining data from all five SMs (p < 0.001, Fig. 6).

Diversification of the HIV genome in humans underlies its success in evading pharmacologic and immunologic selection pressures, and likely facilitates human-to-human transmission events. It has also been suggested that extensive virus diversification actually drives disease progression and the destruction of the immune system [44,45]. To compare the SIVsm genome diversity observed in natural hosts with that of HIV-1 in humans, longitudinally sampled env aa sequences from proviral DNA representing 9 untreated, chronically HIV-infected humans [46] were compared to our plasma RNA-derived SIVsm env data. Two time points were chosen from both the SM and the human dataset so that the interval between observations was approximately 2.5 years.

For the comparison of nucleotide sequence diversity, homologous regions surrounding V1V2 were aligned and gap-stripped. Average pairwise nucleotide diversity was calculated separately in each host at both time points (Figure 7A). Measures of SIVsm and HIV-1 nt diversity were not significantly different from each other within each time point (Figure 7B; p > 0.05, Mann-Whitney U test). Thus SIVsm V1V2 sequence diversity in the natural SM host is at least as great as, if not greater than that observed in HIV-1-infected humans, especially given that the archival nature of proviral sequences may overestimate the diversity of the actively replicating viral RNA population [47-49].

Env adapts not only through raw nt sequence variability, but also through variation in both sequence length and N-glyc site density and position. Substantial changes in these phenotypic parameters will affect the ability of env to utilize different co-receptors [50,51], evade neutralizing antibodies [52,53] and establish new infections in naïve hosts [54,55]. To elucidate differences in SIVsm and HIV-1 V1V2 sequence length and N-glyc site density variation, a pooled estimate of variance within each species was compared. Neither the variances of sequence length nor glycosylation density differed significantly between species at time point 1 although although humans had a greater variance in both parameters at time point 2 (F_max test, p < 0.01). The variance of sequence length of SIVsm V1 decreased between the two time points (F_max test, p < 0.005) suggesting that the magnitude of selection in SMs shifts over time, while in humans the variance remained stable (Figure 7C). The variation in glycosylation density (Figure 7D-E) remained relatively stable over time within both species except for a slight but non-significant expansion of variance in humans at time point 2.

Five age-matched, naturally SIV-infected SMs from the colony at the Yerkes National Primate Research Center, Atlanta, GA were chosen for study. Individual animals were between 8 and 12 years of age and were estimated to have been infected for approximately 3 to 9 years, based on available HIV-2 seroconversion data. Thus, all animals were born in, and acquired their SIVsmm infection in, captivity. Group housing of the animals confounds identification of potential donor-recipient pairs. Plasma from animals FQi, FJo, FFj, FDo, and FBo was obtained on 3-13-99, 5-12-99, and 5-10-01 and viral RNA was extracted and quantified using a real-time RT-PCR assay designed to quantitatively detect the diverse SIVsmm variants [23]. Viral RNA was diluted such that approximately 2500 copies of viral RNA were used in a Superscript™ First-Strand Synthesis System for RT-PCR (Invitrogen Corporation, Carlsbad, CA.), following the protocol provided, primed by random hexamers. 2 μL of cDNA from the RT-PCR was used for PCR amplification of both env V1V2 and gag p27 with Qiagen HotStar Taq (Qiagen Inc., Chatsworth, CA.). The env V1V2 region was amplified with the forward primer V1V2DF (5'-TTTGATGCNTGGAAYAAYAC-3') corresponding to bp 6774–6792 of the SIVsmmH4 genome (GenBank accession no. X14307), and the reverse primer V1V2DR (5'-CATAGCATCCCARTARTGCTT-3') corresponding to bp 7217–7238 of the SIVsmmH4 genome. The primer pair amplified a 421 bp fragment spanning the V1–V2 hypervariable region of envelope. The gag region was amplified using shortgagF1 (5'TTAAGTCCAAGAACATTAAATGC-3') and shortgagR (5'GTAGAACCTGTCTACATAGCT-3') which correspond to bp 1493–1515 and 19371957 of SIVsmmH4, respectively, yielding a 421 bp product of the 5' end of the p27 capsid protein. Primers were designed by choosing highly conserved regions from an alignment of all SIV and HIV2 env and gag sequences from the HIV sequence database [81]. Conditions for each reaction were 30 min. at 50°C, 15 min. at 95°C, followed by 40 cycles of 94°C for 1 min., 52°C for 30 s, and 72°C for 1 min. A final extension time was carried out for 5 min. at 72°C. No-template controls and negative controls from the RNA extraction were used in each set of reactions, both RT and PCR, to ensure that no cross contamination occurred at either step. RT-PCR sensitivity was determined to be = 500 copies per reaction.

PCR products from each sample were run on a 1.5% low-melt agarose gel. The resulting 425 bp V1V2 or 421 bp gag product was extracted and cloned into the pCR4-TOPO vector (TOPO TA Cloning Kit, Invitrogen). From Rodrigo et al. [82] it was determined that if 2500 copies of viral RNA are used in the RT-PCR reaction, 20 clones picked from the PCR product will be unique. Therefore, approximately 20 clones from V1V2 and 10 from gag (due to lower expected diversity in this conserved gene) at each time point and each animal were randomly selected and sequenced using the M13F and M13R primers using the dye terminator cycle sequencing method with an MJ Research automated sequencer.

Sequences were aligned using the program CLUSTAL X [83], followed by manual adjustment using MacClade 4.0 [84] and BioEdit Sequence Alignment Editor [85]. Non-aligned regions of length variation in V1 and V2 were removed (corresponding to nucleotides 6932–6974), and sequences containing internal stop codons or frame shifts were also excluded from analysis as these are thought to be PCR artifacts [86].

For tree construction, the Modeltest program [87] was used to construct and evaluate the DNA substitution models used. Based on the Modeltest results phylogenetic analysis on sequences obtained from successive time points during the acute infection was performed by maximum likelihood (ML) using the program Treefinder [88]. The general-time-reversible model, which allows for rate variation between sites [89-91], was used, and the shape parameter (α) of the gamma distribution used in this model was estimated, as were base frequencies and substitution rate parameters. Bootstrap support was determined with 1,000 resamplings of the ML tree using distance methods in PAUP4.0b10*, incorporating the estimated rate parameters. Phylogenetic trees were constructed from all clones obtained from V1V2 and gag and also separately on V1V2 and gag sequences obtained from each animal at each time point by maximum likelihood (ML) using the program Treefinder.

The cumulative number of nonsynonymous (dN) and synonymous (dS) nucleotide substitutions was estimated using SNAP, Synonymous/Non-synonymous Analysis [81] which calculates rates of nucleotide substitution based on the method of Nei and Gojobori [92], and incorporating a statistic developed in Ota and Nei [93]. Viral diversity at each time point was determined by calculating the pairwise nucleotide distances for each of the clones using the method of Tamura and Nei [94], and pairwise amino acid distances using the Gamma distance method in the program MEGA 2.1 [95]. Average dN and dS were calculated using the modified Nei-Gojobori method in MEGA 2.1. Phylogenetic trees constructed with synonymous or nonsynonymous sites only were constructed in MEGA 2.1 using distance methods, incorporating the Tamura-Nei model of nucleotide substitution with gamma-distributed rates. All statistics were computed using SYSTAT 10.

In order to show that viral populations do not vary randomly through time, random trees of all variants from each animal were generated and the number of matching time-point sequences that formed a monophyletic clade was counted for each random tree. For the random trees, the number of matching time-point sequences that comprise a monophyletic group are Poisson distributed. The Kolmogorov-Smirnov test was used to compare our observed trees with those built from randomly sampled sequences.

Env nt sequences from 9 patients of a study of 10 HIV-infected patients [46] were compared to our SIVsm env data with respect to nt diversity, sequence length variation, and predicted N-linked glycosylation site diversity. For V1V2 nt diversity comparisons, sequences from both SMs and patients were aligned and stripped of gaps. Pair-wise estimates of intra-host nt diversity were calculated using Mega 2.1 [96]. For sequence length variation, alignments (including gaps) of both SIVsm and HIV-1 were pared down to the V1V2 region as defined by the flanking regions of extreme conservation. For this test, homology of each amino acid site was not as important as the overall homology of the region. Mean-squared error variance was determined by ANOVA in R [97] for both glycosylation density and sequence length in each species at each time point. Variances were compared manually using an F_max test.

Genbank Accession Nos: AY733102-AY733566