Human Immunodeficiency Virus Type 1 Subtype C Molecular Phylogeny: Consensus Sequence for an AIDS Vaccine Design?

HIV-1 sequences.

Thirty-one sequences reported previously by us (37, 41) and others (21, 28, 44, 46) and 42 of 43 new, near-full-length genome sequences were used (Fig. 1 legend), for a total of 73 HIV-1 subtype C sequences. Each sequence represents a different patient. When multiple clones were available, one clone per patient was included in the analysis. All sequences used are nonrecombinant. For all 43 new sequences, phylogenetic analyses (data not shown) were performed to determine the subtype and to check for recombination across the entire genome: one sequence (00BW98MO35 [GenBank accession no. AY074891]) proved to be a C/D_env/C recombinant and was excluded from this study. All 43 new sequences were isolated in Botswana from 1996 to 2000 (as reflected in the sequence names): 2 isolates in 1996; 7 isolates in 1998; 5 isolates in 1999; and 28 isolates in 2000. Most isolates were from asymptomatic patients and were collected in Gaborone, except for a few isolates collected in Mochudi and Molepolole (reflected in the sequence names as “MC” and “MO,” respectively).

Amplification, cloning, and sequencing.

Forty-two near-full-length genome sequences from patients in Botswana were amplified, cloned, and sequenced as described previously (41) with modifications (38, 40). Briefly, isolated peripheral blood mononuclear cells (PBMC) were frozen in cell-freezing medium and transported from Gaborone, Botswana, to Boston, Mass. Thawed PBMC were short-term cocultured with donor PBMC, followed by DNA extraction. An ∼9-kb amplicon was obtained by one round of long-range PCR using the 696-9690 primer set (14). After a gel extraction of the amplicon, cloning was performed using a TOPO XL PCR cloning kit (Invitrogen, Carlsbad, Calif.). DNA plasmid purification and both-strand sequencing were performed as described previously (38, 40, 41).

Multiple alignment procedures.

The alignment procedure applied in the phylogenetic study involved the application of ClustalX (version 1.81) (50) followed by manual alignment editing using BioEdit (23). Pairwise alignment parameters were set to the dynamic “slow-accurate” programming, using 10 as the gap opening penalty and 0.1 as the gap extension penalty. Multiple alignment parameters included a gap extension penalty equal to 0.2.

Phylogenetic distances.

The pairwise evolutionary distances of nucleotide alignment were computed by DNADist with the Kimura two-parameter model (PHYLIP: phylogeny inference package [versions 3.52c and 3.572c]; University of Washington, Seattle). Pairwise distances between translated amino acid alignments were performed by PROTDist with the PAM model (PHYLIP: phylogeny inference package [versions 3.52c and 3.572c]). By using sequence distances that weighted positions in the viral genome equally, our method implicitly assumed that all positions were equally important in determining vaccine protection. If information were available on the relative structural and immunological importance of positions, then it could be incorporated into the analysis.

Consensus sequences.

Consensus nucleotide sequences were obtained using BioEdit (23). Gaps were treated as a fifth residue. The threshold frequency for inclusion of a residue in a consensus sequence was 51%. Sites where no residue exceeded the threshold were scored as missing. Consensus amino acid sequences were obtained by translating nucleotide sequences, realigning codons using ClustalX (version 1.81) (50) followed by manual editing using BioEdit (23), and then computing the consensus sequences using Consensus (version 3; available from M. Essex). In Fig. 1 to 3 and 5 the consensus sequence 73C (or 73C_cons) is based on all 73 HIV-1 subtype C sequences, 51BW (or 51BW_cons) is based on 51 sequences from Botswana, 22nonBW (or 22nonBW_cons) is based on 22 non-Botswana sequences, 9IN (or 9IN_cons) is based on 9 sequences from India, and 5ZA (or 5ZA_cons) is based on 5 sequences from South Africa.

Phylogenetic trees.

The neighbor-joining (NJ) and maximum-parsimony (MP) methods were employed for tree-building using the program PAUP* 4.0b8 (Sinauer Associates, Sunderland, Mass.). The character-status summary of the phylogenetic analysis was as follows: (i) all 9,601 characters had equal weight, (ii) 3,955 characters were constant, (iii) 1,856 variable characters were parsimony uninformative, (iv) the number of parsimony-informative characters equals 3,790, and (v) gaps were treated as “missing.” Parameters in the MP analysis were set to default in the PAUP* 4.0b8 software (Sinauer Associates) except for the “addition of sequences,” which was set to “random.” An MP tree is shown in Fig. 1A. The numbers above or beyond the branches correspond to the number of changes between nodes and depict the branches' length according to the scale at the bottom left of Fig. 1A. To obtain a statistical estimate of the reliability of particular groupings, a bootstrapping was employed for both NJ and MP analyses using 100 replicates and resampling of all 9,601 characters in the alignment. Bootstrap values for the delineated groups across the phylogenetic trees that were higher than 75 (at least by one of the methods used) are shown at the right of Fig. 1A. The maximum-likelihood (ML) analysis was performed on Nirvana, a 2,048-processor cluster of SGI Origin 2000 systems, and on a dual-processor Pentium III computer. Access to Nirvana was provided by the Advanced Computing Laboratory, Los Alamos National Laboratory, Los Alamos, N.Mex. The ML analysis was performed using versions of fastDNAml (20, 42) and DNArates (G. J. Olsen, S. Pracht, and R. Overbeek, unpublished data [http://geta.life.uiuc.edu/∼gary/programs/DNArates.html]) modified as described in reference 26. After stripping sites coded as gaps, missing, or uncertain in the included consensus sequences (see below), the alignment was reduced from 9,601 to 7,664 sites that were used in the ML analysis. Custom rate and category matrices were computed from the alignment. As described previously (26), the ML analysis consisted of iterated pairs of runs: a tree run to compute a new tree and then a rate run to compute a new pair of rate and category matrices; each run was based on results of the previous run. The transition/transversion ratio was set to 2.1, based on extensive analyses of full-length sequences of HIV-1 group M (C. L. Kuiken and T. Bhattacharya, unpublished data). Ten categories were used. Twelve iterations were required for convergence of the ML score of the best trees computed from the tree runs. In all tree runs, the Jumble option (PHYLIP: phylogeny inference package, version 3.3; University of Washington) was turned on, and each tree run consisted of three parallel searches for the best tree. This allowed the search algorithm to cover more of the space of all possible trees. The best of the three trees resulting from each tree run was used in the subsequent rate run. The root of the ML tree was based on a permutation test using seven different sets of outgroup sequences (49). The best ML tree with the lowest score was drawn using TreeExplorer 2.12 (K. Tamura [http://evolgen.biol.metro-u.ac.jp/TE/TE_man.html]) (Fig. 1B). The simian immunodeficiency virus (SIV) CPZGAB (accession number X52154) was used as an outgroup in MP and NJ analysis. The SIV outgroup was omitted from the ML analysis. This was done because, in an initial ML analysis, the SIV CPZGAB outgroup was connected to the subtype C clade at a node that is considered to be deeply nested within the clade, based on an ML analysis of all available full-length sequences for HIV and SIV (U. R. Smith, unpublished data). Incorrect (and often unstable) placement of individual SIV sequences with respect to HIV group M subtypes was observed also by Korber and coworkers (26, 56). We recommend that, when performing an ML analysis of any part of HIV group M, individual SIV sequences should be used as the outgroup only with extreme caution.

Statistical analysis.

Statistical analysis and basic graphical delineation were done using Microsoft Excel 2000 software (Microsoft Corp.), Splus (version 6.0; Insightful Corporation, Seattle, Wash.), and SigmaPlot 2001 (SPSS Inc.). Additional graphical presentation was performed using Adobe Illustrator software (version 8.0). Two sample t tests were used to compare the mean values of distances to the consensus sequence between sets of sequences. Corrected chi-square tests were used to compare the frequency of sequences with three NF-κB sites between sets of sequences. To assess if mean diversity between samples was different from the mean distance to consensus, we developed a new test statistic that appropriately accounted for the nonindependence of pairwise distances. For a set of n sequences, the new statistic is defined as the difference in between-sample and to-consensus-sample means, d-_b − d-_c, divided by the square root of the sum of the squared standard error of d-_b and the squared standard error of d-_c. The standard error of d-_b cannot be calculated in the standard way, because many of the m = n(n − 1)/2 pairwise distances are not independent. To compute it, we first noted that, by a direct calculation, c of the m(m − 1) pairs of pairwise distances, where c = m(m − 1) − m(n − 2) (n − 3)/2, have a positive correlation due to a common sequence in the two distances. We assumed a common linear correlation and estimated it by Pearson's correlation, r, based on all possible pairs of pairwise distances with a shared sequence. Then, the standard error of d-_b can be shown to equal the square root of [(1/m) + (c × r/m²)] multiplied by the standard deviation calculated using all pairwise distances. We compared the resulting Z statistic to a standard normal distribution to test the null hypothesis. For nucleotide distances, we calculated r to be 0.432, 0.350, 0.486, 0.696, and 0.869 for the 73 HIV-1C, 51BW, 22nonBW, 9IN, and 5ZA sequence sets, respectively. For protein distances, we calculated r to be 0.319, 0.299, 0.304, 0.413, 0.289, 0.420, 0.286, and 0.134 for the 72 or 73 HIV-1C Pol, Vif, Vpr, Tat, Rev, Vpu, Env, and Nef sequences, respectively.

Distribution of distance to consensus.

If the consensus sequence from a sample of n sequences (one per individual) in a geographic region is used in a vaccine targeted to the region, then it is of interest to assess the distribution of the distance from a randomly sampled sequence to the observed consensus sequence. The mean distance of sequences in the population to the observed consensus was estimated by the average distance of the samples to the consensus sequence. Using the 73 HIV-1C sequences, a 95% confidence interval (CI) for the mean distance was calculated under the assumption that the distances are an independent, identically distributed random sample from a normal distribution. In addition, the 80th and 95th percentiles of the distance to the observed consensus were estimated with the sample percentiles, and the nonparametric bootstrap percentile method was used to compute 95% CI for the percentiles.

Proximity of a new sequence to the consensus.

The probability that a randomly selected sequence in a population is within a certain threshold, D%, of the observed consensus sequence was estimated by the sample proportion of sequences within the threshold. A 95% CI for the probability was calculated using the normality approximation to a binomial random variable.

Extent of the consensus change.

We considered the extent of anticipated change in the observed consensus sequence that was built on n = 73 HIV-1C sequences if it is rebuilt including a new sequence. We addressed this by evaluating how much the distribution of the distance of a randomly sampled virus to the observed consensus sequence changes in response to the introduction of a new sequence. The mean amount by which the mean distance to the observed consensus changes was estimated by the sample mean of the n numbers d_i = x-_[−i] − x-|, i = 1,…, n, where x-_[−i] is the sample mean distance to the observed consensus, which was calculated using the n sequences with the ith sequence removed, and x- is the sample mean distance to consensus for the full data set. A 95% CI for the mean change was calculated using the nonparametric bootstrap percentile method. In addition, the nonparametric bootstrap method was used to estimate 95% CIs for the mean amount by which the 80th and 95th percentiles of the distance to the observed consensus changes.

All tests were two tailed, and a cutoff P of 0.05 was used to judge statistical significance.

Accession numbers.

The 42 new, nonrecombinant HIV-1C nucleotide sequences from Botswana were deposited in GenBank under accession numbers AF443074 to AF443115.

Overview of newly generated HIV-1C sequences.

Together with previously described sequences (37, 40, 41), the 43 new sequences reported here bring the total number of near-full-length genome sequences from Botswana to 53, including two recombinants. The nonrecombinant sequences, all from different patients, are all subtype C, consistent with the 96.2% predominance of subtype C in Botswana. These sequences contained a number of previously described subtype C signatures (21, 41, 44, 46), namely, (i) an insertion at the N-terminal region of the Vpu (from 5 to 13 amino acids in 70.0% of isolates), (ii) a prematurely truncated second exon of Rev (9 amino acids; 100% of cases), (iii) a third NF-κB site seen in the majority of the isolates (see details below), and (iv) a 4-amino-acid extension at the C-terminal region of Pol in some sequences (21.4%).

Phylogenetic relationship of near-full-length HIV-1C.

The molecular phylogeny of the 73 nonrecombinant HIV-1C sequences was analyzed by applying MP, NJ (or distance), and ML methods. Figure 1A represents an MP phylogenetic tree that was found to be the most parsimonious tree and had a minimal score of 29,382 when gaps were treated as missing. Figure 1B shows an ML tree with a score of −91931.28. The MP and ML scores are not comparable.

Branch lengths in the MP and ML analyses are not easily compared. This is because each method uses a large number of sites not used by the other method. MP uses sites that contain gaps; ML does not. This difference may be responsible for the relatively longer “backbone” branches of the MP tree (Fig. 1A) versus the ML tree (Fig. 1B). ML uses sites that are nearly constant, differing for only one sequence; MP does not. The longer terminal branches in the ML tree (Fig. 1B), compared to the MP tree (Fig. 1A), are due in part to this difference. These sites, which represent nucleotide changes unique to a single sequence in the analysis, necessarily increase the length of the terminal branch for that sequence.

Although there are numerous well-supported “lineages” (a term that is used hereafter for discussion purposes and does not imply taxonomic standing) of sequences within HIV-1C, the backbone of the HIV-1C lineage itself is not well supported; hence, the tree topology is unstable. This instability is evident in the very different positions of lineages between the MP and ML trees (Fig. 1) and in the prevalence of relatively short branch lengths along the backbones of both trees. The sequences used in this study are nearly full-length genomes: these genomes can yield little or no further sequence data. Hence, to obtain a more-stable, reliable topology it will be necessary to find new genomes having sequences that help to support these unstable portions of the tree and/or to improve methods of phylogenetic analysis to make better use of the available sequences. For example, current ML analyses discard sites that have gaps for any sequences included in the analysis. This is unfortunate because many of these sites represent “indels” (insertions and deletions) outside the hypervariable regions of the HIV-1 genome, and many indels in other organisms are known to contain significant phylogenetic information.

Because a detailed phylogenetic analysis of HIV-1C sequence evolution was outside the scope of this study, we addressed only the general shape of the ML tree based on near-full-length genome HIV-1C sequences, as well as congruency among the ML, MP, and NJ trees. The basal pair of sequences found in the MP tree (96BW17A09 and 00BW1471.27 [Fig. 1A]) agrees with results of an explicit search for the root of the HIV-1C lineage (49). This result is reflected in the ML tree (Fig. 1B), which is assumed to have the same root as the MP tree. Note that the ML tree includes only HIV-1C sequences and that, in phylogenetic terms, all trees produced in this study are unrooted.

A number of well-supported lineages (shaded in Fig. 1) were identified within the 73 near-full-length genome HIV-1C sequences. All these lineages occur in the best trees generated by all three phylogenetic methods (Fig. 1). Topology of the samples grouped within lineages was nearly identical in MP, NJ, and ML trees (sequence 00BW1811.3 was within the 00BW2128.3-96BW16.26-00BW1773.2-96BW0502 lineage in the ML but not the NJ and MP trees). Generally the bootstrap values were highly consistent by the MP and NJ analysis, although in one case (samples 96BW0407 and 98BWMO18.d5) a relatively high bootstrap value of 83 for a lineage in the NJ tree was not supported in the bootstrap analysis of the MP tree (value of 57 only). Overall, 57 of 73 (78.1%) HIV-1 sequences were found within lineages. Eight of the identified lineages were composed of two isolates, two lineages contained three sequences, three lineages embraced four isolates, one lineage included five, and two lineages consisted of eight sequences. Eight of nine samples from India and four of five samples from South Africa formed separate lineages, apparently demonstrating relevance to the phylogenetic founder effect. Sequences from Brazil, Ethiopia, and Israel were found in one lineage, which suggested a phylogenetic relatedness of these geographically distinct isolates. The Botswana sequences were scattered across the phylogenetic tree, with 39 viral isolates (76.5%) that formed 13 distinct lineages supported by high bootstrap values (Fig. 1). The Botswana lineages did not include any of the 22 non-Botswana sequences, and none of the Botswana sequences were part of any of three non-Botswana lineages. A number of sequences outside the lineages demonstrated similar topologies in MP and ML trees (00BW3886.8, 94IN476.104, and 00BW1783.5). In the MP analysis, neither the content of each lineage nor the overall topology of the best tree was changed when additional outgroup sequences were introduced to the alignment (HIV-1 subtype A, Q23-17, accession number AF004885; HIV-1 subtype B, HXB2CG, accession number K03455; data not shown).

Having addressed a phylogenetic relationship among consensus sequences, we compared the topology of the consensus sequence that represented the entire set of the 73 available HIV-1C sequences in the study to the topology of the consensus sequences that represented different subsets comprising 51 Botswana isolates, 22 non-Botswana sequences, 9 sequences from India, and 5 sequences from South Africa. The origin of the consensus sequences was reflected in their phylogenetic relationships (Fig. 1). While an observed closeness of the consensus sequences to the tree root in MP (Fig. 1A), NJ (data not shown), and ML (Fig. 1B) trees was not unusual, note that the consensus sequences were placed closer to the root of the tree than was any particular sequence from which they originated. As expected, the consensuses of the Indian and South African subsets clustered within the corresponding groups of sequences. Interestingly, a consensus sequence of the 22 non-Botswana isolate subsets, which included both subsets from India and South Africa, was closer to the root of the tree than was 5ZA_cons or 9IN_cons. An obvious nearness between consensuses representing 51 sequences from Botswana and the entire set of 73 HIV-1C sequences was not surprising, because Botswana sequences were dominant in the entire set. However, a relative closeness between consensuses of 51 sequences from Botswana and 22 non-Botswana sequences was striking. Assuming that sequence homology between the vaccine candidate and the infecting or challenging virus is essential, the overall topology of the consensus sequences and their phylogenetic relationship with corresponding sequences might suggest that candidate AIDS vaccines incorporating the consensus sequence have a greater potential than those incorporating any particular isolate sequence.

In terms of pairwise genetic distances between consensus sequences, the entire set of 73 HIV-1C was almost identical to the 51BW set (Fig. 2). This result is due to the predominance of Botswana sequences in the entire set. Nucleotide distance between 51BW and 22nonBW consensuses was only 0.48%. The consensus sequences for nine sequences from India and five sequences from South Africa differed from the 51BW consensus by 2.08 and 1.84%, respectively, while the distance between Indian and South African consensus sequences was 3.5%. Overall, pairwise genetic distances (Fig. 2) demonstrated a remarkable closeness between consensus sequences of different HIV-1C subsets.

HIV-1C genetic distances.

The pairwise nucleotide distances and corresponding statistics among the entire set of 73 near-full-length genome HIV-1C sequences are shown in Fig. 3A. While the mean value of diversity between samples reached 8.93% (range, 2.9 to 11.7%), the diversity to the consensus sequence was significantly lower, with a mean value of 4.86% (range, 3.3 to 7.2%; P < 0.0001). The results of a similar analysis and accompanying statistical data among subsets of sequences are presented in Fig. 3B. The diversity between samples was significantly higher than the distances to the consensus sequences in all groups analyzed (P < 0.0001 for 51BW and 22nonBW, P = 0.0049 for 9IN, and P = 0.0076 for 5ZA), highlighting the notion of a potential advantage for the consensus sequence to be used in the vaccine design. It is noteworthy that the subset of sequences from Botswana (51BW) demonstrated the highest level of pairwise diversity between samples (mean value of 9.17%), as well as distances to the consensus sequence (mean value of 4.92%) that were significantly higher than the corresponding mean value among subsets of 22nonBW, 9IN, and 5ZA (t tests comparing mean distances to consensus: P = 0.0014 for 22nonBW and P < 0.001 for 9IN and 5ZA).

HIV-1C Gag extended consensus.

Figure 4 delineates the extended version of the consensus sequence for the HIV-1C Gag p17, p24, and p2/p7/p1/p6 based on the amino acid alignment of 73 subtype C sequences. The invariable amino acid residues within the HIV-1C Gag p17 were observed at 37 out of 129 positions (28.7%). In addition to the invariable amino acids, 57 positions (44.2%) across HIV-1C Gag p17 were relatively conserved by showing less than 10% diversity at a particular amino acid residue in the consensus sequence. The number of variable residues in the consensus that had a frequency of 90% or less was 35 (27.1%), which was at the level of invariable amino acids. The most variable amino acid residues that had frequencies of less than 50% in the consensus sequence were seen at four positions within p17. The alternative characteristics of amino acid residues (i.e., by charge) at positions 15 (K₄₆, T₃₀, A₁₂), 90 (E₃₆, A₂₇, K₂₁), 91 (G₃₈, K₂₉, N₁₁, E₁₀), and 119 (K₄₅, E₄₄) might suggest a substantial difference in the biological properties of different p17 proteins (subscripts indicate the percentage of a residue's frequency at that particular position in the alignment). Positions 15, 28, 62, 90, 91, 93, 111, 115, 118, and 119 might be considered the most variable within p17 by virtue of accommodating from seven to nine different amino acids at each position. A few sequences demonstrated indels in the C-terminal part of p17.

The more conserved p24 protein demonstrated a lesser extent of amino acid variation (Fig. 4B). One hundred forty invariable amino acid residues (60.6%) and 69 relatively conserved amino acids that had less than 10% variation (29.9%) determined a strong conservation of the HIV-1C p24. However, 22 amino acid positions (9.5%) showed equal or more than 10% diversity in the consensus sequence of HIV-1C Gag p24. Although amino acids at some positions (252, 256, and 357 in Gag) were close to the frequency of 50% in the consensus sequence, only one (position 223 of Gag, V₄₇, I₄₆, N₄, and A₃) has reached a frequency lower than 50%. In contrast to the variations in p17, substitutions within HIV-1C p24 occurred between relatively similar amino acids, i.e., I₈₅ and V₁₅ at positions 159 (I-to-V, or I159V), V223I, E230D, V256I, D260E, K286R, D312E, R335K, T342S, etc. No indels were found within the HIV-1C p24.

As shown in Fig. 4C, HIV-1C Gag p2/p7/p1/p6 was comprised of 45 invariable amino acid residues (34.2%), 56 relatively conserved residues with less than 10% diversity in the consensus (42.4%), and 31 variable residues with a frequency of 90% or less in the consensus sequence (23.5%). The frequency of amino acid residues at positions 372 and 373 was less than 50% in the consensus sequence: N₄₉, S₄₀, —₈, T₁, Q₁, G₁ and T₄₂, —₁₆, A₁₅, I₁₀, V₅, P₅, S₃, M₃, G₁, respectively (where “—” denotes a gap introduced to improve alignment). The asparagine and serine at position 451 were observed at the frequency of 50% and 47%, respectively. The positions 373, 389, and 478 were represented by eight or nine different amino acid residues each. Although the insertions were not rare across p2/p7/p1/p6, most amino acid residues in the insertions were seen at low frequency. However, the frequency of some amino acid residues within the insertion between positions 455 and 456 reached the level of 30% (Fig. 4C).

Table 1 displays the amino acid distances across HIV-1C Gag, p17, p24, and p2/p7/p1/p6 for the entire set of 73 HIV-1C sequences (73C) and for the subsets. A high diversity between samples in the p17 (mean, 14.8%) and in the p2/p7/p1/p6 (mean, 12.7%), together with a low diversity within the p24 region (mean, 5.1%), resulted in the overall mean diversity of 9.5% across the entire HIV-1C Gag. The amino acid diversity among isolates from Botswana was slightly higher than the diversity among non-Botswana samples for the entire Gag and its subregions. Statistical significance of the differences depended on the presence of sequences from India in the group of non-Botswana samples, which changed from highly significant to nonsignificant if sequences from India were excluded. No significant differences were found between the subsets of Botswana and non-Botswana sequences for the amino acid distances to the consensus sequences (P > 0.10). However, the subset of sequences from Botswana demonstrated significantly higher amino acid diversity than the subset from India for the entire HIV-1C Gag and for any of the Gag regions in both between-sample (P < 0.0001) and to-consensus (P < 0.0001) comparisons. The subset of HIV-1C sequences from South Africa showed lower diversity within the p24 region than Indian sequences, although the difference was not statistically significant (P = 0.40).

For the 73 HIV-1C sequences, the 51BW sequences, and the 22nonBW sequences, the mean amino acid diversity measured between samples was significantly higher than the mean distance measured to the consensus sequence for all Gag regions (P < 0.001 in each test). For the 9IN sequences there were significant differences for Gag (P = 0.018), p17 (P = 0.046), and p24 (P = 0.028), but not for p2/p7/p1/p6 (P = 0.069). Similarly, for the 5ZA sequences there were significant differences for Gag (P = 0.014) and p24 (P = 0.044), with a trend toward significance for p17 (P = 0.052) and for p2/p7/p1/p6 (P = 0.11). We believe that the lack of statistical significance in some subregions was due to a small sample size and propose that between-sample diversity in the populations of sequences is consistently larger than the to-consensus diversity in all Gag subregions. Overall the diversity to the consensus was 1.9 times lower (range, 1.6 to 2.3) than the diversity between sequences for the entire HIV-1C Gag and its subregions.

Amino acid diversity within other HIV-1C proteins is shown in Table 2 together with P values and ratios for the comparison of between-sample and to-consensus distances. The between-sample distances ranged from 6.42% in Pol to 25.20% in Vpu, while to-consensus sequences were in the range of 3.45% in Pol to 13.68% in Vpu. Importantly, to-consensus distances were significantly lower than between-sample distances for each HIV-1C protein. Amino acid diversity in Env demonstrated the highest ratio of 2.4 in comparison of between-samples and to-consensus distances. Table 3 summarizes the diversity among lineages across HIV-1C proteins. For most cases amino acid distances are shorter than the average distances for the entire set of 73 HIV-1C, although there are a number of examples of equal or even higher diversity within lineages.

We addressed the predictive power of the consensus sequence for new hypothetical (not-yet-sequenced) HIV-1C isolates. Assuming that sequence homology between the vaccine candidate and the infecting virus is essential for the vaccine design, it might be important to know (i) the distribution of the distance of a randomly sampled sequence to the observed consensus (i.e., how far do individual sequences tend to be from the observed consensus sequence?); (ii) the probability that an arbitrary new sequence in the population will be within a certain threshold, D%, of the observed consensus sequence; and (iii) the degree of perturbation of the observed consensus sequence if a new sequence is used in its recalculation (i.e., how does the distribution of the distance of a randomly sampled virus to the observed consensus sequence change as a result of the introduction of the new sequence?). If the consensus sequence is used for the vaccine antigen, then analyses of points i, ii, and iii address the extent of vaccine coverage.

Based on the set of 73 HIV-1C sequences, we estimated that the mean distance to the observed consensus sequence was 4.86%, with a 95% CI of 4.69 to 5.02%. The estimated 80th percentile was 5.45% (95% CI, 5.21 to 5.52%), and the 95th percentile was 5.80% (95% CI, 5.51 to 6.79%). Thus, on average HIV-1C sequences are within about 5% of the observed consensus, and with high confidence, 95% of viruses in the sampled population are within 6.79% of the observed consensus sequence.

The observed proportion of the 73 HIV-1C sequences within D% of the observed consensus, with D being equal to 4, 5, 6, and 7%, respectively, was 8.22% (95% CI, 1.92 to 14.52%), 54.79% (95% CI, 43.48 to 66.21%), 97.26% (95% CI, 93.52 to 100.00%), and 98.63% (95% CI, 95.96 to 100.00%). Thus, fewer than 15% of HIV-1C viruses are inferred to be within 4% of the observed consensus, between 43% and 66% of viruses are inferred to be within 5%, and at least 96% are inferred to be within 7% of the observed consensus.

We found that a single sequence does not appreciably alter the distribution of distances to the observed consensus. Based on the 73 HIV-1C viruses, we calculated that the mean of the 73 numbers d_i = |^-x_[−i] − ^-x| was 0.135%. The bootstrap 95% CI about the mean change in the mean distance to the observed consensus was 0.121 to 0.159%. The estimated mean change in the 80th percentile of the distance to the observed consensus was 0.126% (95% CI, 0.111 to 0.135%), and for the 95th percentile the estimated mean change was 0.117% (95% CI, 0.075 to 0.197%).

HIV-1C LTR promoter-enhancer region.

Figure 5 depicts the extent of conformation to the GGGRNNYYCC consensus within NF-κB sites among the 73 HIV-1C sequences and 5 consensus sequences. The number of NF-κB sites within the promoter-enhancer region of HIV-1C varied from one (isolate 00BW1880.2) to three or more (isolates 96BW0502 and 96BWMO3.2), while the consensus sequences for the entire set or any subset of HIV-1C isolates demonstrated three NF-κB sites that conformed to the GGGRNNYYCC consensus. Within the subsets the HIV-1C sequences from Botswana demonstrated three NF-κB sites in 32 of 51 (62.7%) of cases, while only 10 of 22 (45.5%) non-Botswana sequences had the third NF-κB site (P > 0.10 [chi-square test]). The potential or prospective NF-κB sites represent a region that does not comply with the GGGRNNYYCC consensus but, in fact, is relatively close to it and might become a new NF-κB site due to a few point mutations. Viral isolates from Botswana that did not have three NF-κB sites demonstrated a potential/prospective NF-κB site more often than non-Botswana subtype C sequences (25.5 versus 4.5%, respectively; P < 0.001), which suggests that the promoter-enhancer region might be a hot zone within the evolving HIV-1C.