Brian Gaschen,¹ Jesse Taylor,¹ Karina Yusim,¹ Brian Foley,¹ Feng Gao,² Dorothy Lang,¹ Vladimir Novitsky,³ Barton Haynes,² Beatrice H. Hahn,⁴ Tanmoy Bhattacharya,¹ Bette Korber^15*

Globally, human immunodeficiency virus-type 1 (HIV-1) is extraordinarily variable, and this diversity poses a major obstacle to AIDS vaccine development. Currently, candidate vaccines are derived from isolates, with the hope that they will be sufficiently cross-reactive to protect against circulating viruses. This may be overly optimistic, however, given that HIV-1 envelope proteins can differ in more than 30% of their amino acids. To contend with the diversity, country-specific vaccines are being considered, but evolutionary relationships may be more useful than regional considerations. Consensus or ancestor sequences could be used in vaccine design to minimize the genetic differences between vaccine strains and contemporary isolates, effectively reducing the extent of diversity by half.

¹ Los Alamos National Laboratory, Los Alamos, NM 87545, USA.
² Duke University AIDS Center, Durham, NC 27710, USA.
³ Department of Immunology and Infectious Diseases, Harvard School of Public Health, Boston, MA 02115, USA.
⁴ University of Alabama at Birmingham, Birmingham, AL 35294, USA.
⁵ Santa Fe Institute, Santa Fe, NM 87501, USA.
^*   To whom correspondence should be addressed. E-mail: btk@t10.lanl.gov

Although the scale of the HIV-1 pandemic makes action imperative, there is still much to learn about the extent and immunological implications of HIV-1 sequence diversity. We do, however, have in hand the fruits of an extensive global HIV sequencing effort [currently there are 72,221 HIV sequences in the database (3)] that can provide a framework for reasoned vaccine strain selection. Optimizing selection is of the utmost urgency, as a number of human vaccine trials are being planned and initiated (5, 6), and it is difficult to change strains during the long course of vaccine development, from initial concept to human trial. Subtype C is the most prevalent HIV-1 subtype globally, and it predominates in several geographic regions where vaccines might be evaluated. In these regions, the epidemiologically unlinked prevalence of subtype C infections can exceed 30% of the adult population (7). Therefore, this exploration of the implications of HIV variation for vaccine strain selection focuses on subtype C; we believe, however, that our reasoning and findings can be extrapolated to other intra- and intersubtype scenarios.

Although there is hope that a single vaccine strain may elicit a sufficiently cross-reactive response to confer a benefit, there is great interest in attempting to optimize vaccines through considerations of diversity. There are currently two general approaches to selecting vaccine strains that attempt to contend with the high levels of HIV sequence variation (Table 1). The first is based on using isolates of a particular subtype, sometimes selected from a geographic region where the vaccine is intended for use. Examples of this approach that are under way include the development of several A- and C-subtype vaccines, as well as CRF01 vaccine reagents (5, 6). This kind of approach can be integrated with biological considerations, such as coreceptor usage, neutralization susceptibility, neutralization potency of the serum from the individual from whom the isolate was obtained, or the preferential use of isolates from recent seroconvertors (8). The second approach, rather than using actual viruses from within the population, is to construct either a consensus sequence or an ancestral sequence reconstructed on the basis of an evolutionary model. Such sequences have the advantage of being central and most similar to currently circulating strains of interest and may have enhanced potential for eliciting cross-reactive responses. They may also have economic and political advantages that merit consideration. Economic, because it is not feasible to duplicate vaccine design efforts using country-specific strains for every nation and region that needs a vaccine, and this is a way to limit the number of constructs that must be produced and tested in a way that is logical and scientifically defensible. Political, because such artificial sequences are not associated with any specific country of origin, so nations hosting vaccine trials would not need to contend with the natural concerns that arise when asked to host a vaccine trial using HIV-1 antigens with distant geographic origins. Some subtype C sequences of current interest are indicated in Fig. 1, a phylogenetic tree that includes geographic information to provide a basis for considering candidate vaccine strains. (Phylogenetic terms and concepts used in this article are defined in the legend to Fig. 1.) These two basic approaches, in different ways, directly confront the problem of viral diversity; we will explore their advantages and disadvantages in the sections that follow.

Table 1. A summary of classes of potential vaccine strains for use in subtype C epidemic regions. Differences show typical values of the percentage of amino acid changes observed when comparing the potential vaccine strain sequences to the sets of available C clade protein sequences. The lower bound represents conserved proteins, the upper bound variable proteins. Vaccine source Differences (%) Advantages and characteristics Isolates, subtype B 10-30 Furthest along in vaccine testing Based on an actual virus, and strains can be selected on the basis of advantageous biological characteristics Isolates, subtype C 5-15 The closest natural form to C-subtype circulating strains Like subtype B isolates, based on actual viruses, and thus can be selected on the basis of biological features Consensus, subtype C 3-8 Central to C-subtype circulating strains Each amino acid is most commonly found at that position Ancestral, subtype C 3-8 Representative of the C subtype Maximum likelihood model of ancestor sequence M-group consensus 5-15 Representative of the HIV-1 epidemic Most likely to cross-react with all clades Consensus of the subtype consensus sequences

Other concepts being explored to address diversity issues can ultimately be considered in the framework of the two basic approaches to strain selection described above. For example, multivalent cocktails of proteins that include a spectrum of regional variants are being evaluated. Most of these strategies assume that the immune responses elicited by any one circulating strain will be of sufficient cross-reactivity to protect against other strains from the same subtype. Given that intrasubtype diversity in variable proteins can reach 20%, even this assumption may be too optimistic. Consensus sequences and strains from viable isolates could be combined in a polyvalent approach. A second strategy is to design modified envelopes to enhance exposure of epitopes known to be capable of inducing broadly neutralizing antibodies (9-11). There are a limited number of monoclonal antibodies that have broad, cross-clade neutralization capabilities (11-13), and these antibodies can act synergistically (13-15). Vaccines specifically designed to target these conserved epitopes, if successful, may ultimately be optimized by fine-tuning as subtype-specific vaccines, as appropriate; although there is little evidence that subtypes correspond to neutralization phenotypes, in some cases particular clades can be refractive or less susceptible to particular antibodies. For example, the three broadly neutralizing monoclonal antibodies 2F5, 2G12, and IgG1b12, raised against B clade strains, were not individually able to neutralize a C clade primary isolate, although they could in combination (14). Even if subtypes are not relevant to a particular neutralizing epitope, lineage-specific variation within the relevant antigenic domain may still be worth considering. A third strategy for contending with diversity is to use polyvalent peptides spanning a region like the V3 loop that induces strain-specific neutralizing antibodies (16), to attempt to elicit a set of responses that together confer cross-reactive protection (17, 18).

Isolate-Based Vaccines

There has been some discussion of choosing a regional strain for a vaccine, for example, an Indian strain to be used in India, and a South African strain in South Africa, and so on (8). There is little support for this in terms of sequence analysis. Subtype C sequences from Botswana and South Africa intermingle (28), and there is no obvious choice of a single sequence most representative of the diversity in these regional samples (28); however, selecting a sequence with a short branch length relative to the common ancestor (29) in C clade might be advantageous, as it would tend to be most similar to the majority of contemporary sequences represented in the tree (30). Conversely, it would be sensible to avoid selecting outliers. Indian sequences tend to form a distinct subclade (31) within the C clade, indicating that most sampled Indian viruses are descended from a single founder strain. A small number of sequences from African nations are associated with sequences from India, however, and the sampling is extremely limited relative to the scale of the epidemic in both regions. Thus, there may be continuing movement of the virus between Africa and India. The Indian clade sequences tend to have short branch lengths relative to the root of the C clade. As a consequence, a strain like IN101 (accession number AB023804) from India is closer to most African subtype C strains than African strains are to each other (32), an interesting quirk that emphasizes that it does not necessarily confer an advantage to select a strain from the country where a vaccine trial will be held.

There are other subclades within the C subtype, besides the Indian subclade (33), that could be considered as vaccine candidates (Fig. 1), but such subclades tend to have much shorter defining branch lengths than subtypes and, consequently, fewer distinguishing amino acids, so the benefit of considering them each separately for vaccines diminishes. On rare occasions, geographically localized epidemics have been identified soon after the introduction of a founder virus, and prevalent viruses were highly related, as in Thailand (34) and Kaliningrad (35). It may ultimately be advantageous to develop vectors and strategies for rapid-response vaccine programs in such circumstances, when a highly similar virus is spreading explosively through a vulnerable population, but first, a working vaccine concept must be in place.

Evolutionary evidence for subtype-specific antigenicity in the envelope protein. Clearly, subtype-specific vaccines would increase the overall sequence similarity of the vaccine antigen relative to circulating viruses (Fig. 2), but this is only part of the story for antibody binding, because protein folding and exposure of antigenic domains are of great importance. To explore the hypothesis that there may be subtype-specific patterns in the exposure of antigenic domains that are able to elicit antibody responses strong enough to drive escape mutations, we compared estimates of codon-specific ratios of nonsynonymous to synonymous substitution rates (dN/dS) (36) in B clade and C clade envelope genes. (We selected B and C subtypes, as B clade vaccines are being considered for use in populations where the C subtype dominates.) High rates of diversifying selection were identified in different regions of the envelope protein (Env) in the two lineages, most strikingly in the Env V3 to C4 region. The V3 loop is less variable in the C subtype than in other subtypes (37), and as expected, the density of sites in the V3 loop with dN/dS > 1 was higher in the B clade than in the C clade. This pattern was reversed, however, in the region just proximal to the V3 loop, where multiple sites show an excess of nonsynonymous substitutions in the C clade but not in the B clade (Fig. 3). To explore the consistency of these patterns within the C clade, three subclades, or phylogenetically associated groups of sequences within the C subtype, were examined independently. Two sets had 21 subtype C sequences, and the third had 18 sequences. The results suggested substantial intraclade coherence in how selection acts on individual codons within the C subtype; the 12 strongly selected sites in the region downstream of the V3 loop (Fig. 3) had a dN/dS ratio > 1 in each of the three independent C-subtype data sets, and the tip of the V3 loop was relatively constrained with low dN/dS values. Given that immune escape is likely to be a driving force of positive selection, immune pressure may be focused on different regions of Env in the B subtype (the V3 loop) and C subtype (the COOH-terminal region beyond the V3 loop). If this interpretation of the observed differences in selection pressure in B and C subtypes is correct, there may be advantages in using a clade-appropriate vaccine strain, as the immune response to the vaccine and the circulating virus would share antigenic domains.

Artificial Sequences for Minimizing Diversity

Envelope proteins are the most difficult HIV proteins to construct artificially, as both ancestral and consensus sequences contain hypervariable domains with multiple insertions and deletions (indels). Alignments are subjective in such regions, and indels do not evolve according to the base substitution models currently assumed in deriving a maximum likelihood tree. For constructing our consensus and ancestor sequences (3), hypervariable regions are aligned by anchoring on glycosylation sites, and only minimal common elements spanning the region are retained. As both consensus and ancestral sequences are derived and not actual sequences, expression, antigenicity, and biological activity require careful characterization before use in a vaccine (39).

Although artificial sequences may not have a proper protein conformation, and this may be critical for antibody responses, it is less important for designing T cell epitopesor peptide reagents for testing T cell responses. Consensus sequences may be ideal for peptides used to explore the T cell immune response, as it would probably improve recognition compared with any single reference strain, and using sets of autologous strain peptides can be prohibitively expensive. A consensus may even be preferable to autologous peptides, as CTL escape mutations can rapidly predominate in the viral quasispecies, and important early responses (40) may go undetected through the use of peptides based on isolates from later time points that have escaped the early responses. Consensus peptides for several subtypes are available (41).

A similarity plot maps the percent similarity of a query sequence relative to a test set in a window spanning a region of a specified size that is moved progressively along an alignment. In Fig. 2, prototype vaccine reagents (a C consensus, two subtype C vaccine strains, and a subtype B isolate) are used as query sequences and compared with 23 subtype C sequences from South Africa, India, and Botswana. In every gene region, the same relative pattern holds. The spectrum of similarity scores for the C consensus sequence compared with the set of 23 C sequences is 5 to 15% greater than when any one C isolate is compared with others in the set (28). In turn, subtype C proteins are 5 to 15% more similar to the subtype C sequences than are subtype B sequences. This implies that using a B clade virus as the basis of a vaccine in a C clade-dominated epidemic may be less effective than using a C clade virus, and a C clade virus may not be as effective as a C consensus. Conserved proteins from different subtypes can be more closely related than variable proteins from the same subtype, and this fact might be exploited by using a single vaccine strain for conserved proteins and multiple clade-specific strains for variable proteins.

We have been discussing pooling sequences within a subtype to generate artificial central sequences, but it is also possible to pool the subtypes themselves. To maximize potential cross-reactivity, we have created sequences central to the M group, the diverse viruses that have contributed most to the global epidemic. The set of subtype consensus sequences was used to build an M-group consensus, thus weighting the subtypes equally. The M-group consensus and the most recent common ancestor can be very nearly identical (1, 28). Because of the nature of the HIV-1 M-group phylogeny, the average distance from HIV-1 sequences to the M-group consensus is similar to intrasubtype sequence distances between contemporary isolates (32), roughly half that of intersubtype distances (Table 2). In the Democratic Republic of the Congo (DRC) (42, 43), so many subtypes and recombinants circulate together that the extent of the regional diversity resembles the global diversity. In this setting, an M-group consensus may be helpful, or a polyvalent approach including representative strains from common subtypes along with the M-group consensus. Even in a design focusing on epitopes that are conserved across clades, an M-group consensus might be the optimal baseline sequence. Consensus and ancestral sequences for the major HIV-1 subtypes, CRFs, and the M group are available (3) and will be updated as sequences accrue. Intersubtype similarity comparisons with an M-group consensus are included in the supplementary material (31).

Table 2. Median and range of percent similarity scores between potential vaccine candidates and an alignment of C clade sequences. The similarities are shown for the p24 and gp160 proteins, representative of highly conserved and highly variable proteins. The C-subtype ancestral and consensus (con) sequences, when compared with the set of protein sequences from contemporary subtype C isolates, have comparable distributions of similar- ity scores. The M-group consensus is comparable to C clade isolate sequences. Two representative C clade isolates are shown, DU422 from South Africa and BR025 from Brazil. The last column shows the results of a subtype B strain compared with C sequences (intersubtype) for contrast. Regions with gaps were included, and either a relative insertion or deletion or an amino acid change at any position is considered one difference. Protein Consensus sequences Isolate sequences C-subtype con C ancestral M-group con C_ZA.DU422 C_BR.92BR025 B_FR.HXB2R p24 95.4 (97.4-92.8) 94.8 (97.7-92.8) 93.1 (95.1-91.1) 94.8 (97.4-92.1) 92.8 (96.4-88.9) 88.9 (91.8-86.6) gp160 87.5 (90.6-83.5) 86.1 (88.3-81.7) 81.4 (84.0-77.1) 81.3 (89.3-77.4) 83.1 (85.9-79.6) 72.1 (73.8-68.0)

Consensus and ancestral sequences conserve CTL epitopes. Experimentally defined CTL epitopes in the HIV Immunology Database (3) cluster more densely in conserved regions of HIV proteins (44). The peptides spanning variable regions used to detect CTL responses can be quite different from the infecting strain that elicited the response, no doubt contributing to the paucity of defined epitopes in variable domains, but, in addition, an enrichment of features that could contribute to CTL escape can be discerned in variable domains (45). Either way, regions where defined epitopes are concentrated are likely to be key for cross-reactive CTL responses (44). The epitopes in the database have primarily been defined for B clade responses; however, the C clade peptides that trigger immunodominant responses tend to be localized in these same regions (44, 46). Thus, in contrast to Fig. 2 and Table 2, where whole proteins were analyzed, we focused on protein regions where CTL epitopes have been found in order to create Fig. 4, which shows the average sequence distances from potential vaccine strains to immunogenic regions in subtype C proteins, by country of origin. Three proteins were selected, representing the spectrum of variability: highly conserved p24, variable p17, and highly variable envelope (subunit gp160). In the immunogenic regions analyzed, C-subtype consensus and ancestral sequences had the fewest amino acid changes relative to contemporary C-subtype protein sequences. Within-subtype comparisons of single C-subtype viral strains and the M-group consensus sequences gave comparable numbers of amino acids changes, roughly half the number of changes relative to B subtype interclade comparisons.

Consensus and ancestral sequences conserve predicted immunoproteasome cleavage sites. For viral proteins to be recognized by CTL they must be processed, and each step of epitope processing has potential constraints imposed by sequence specificity. Immune escape due to mutations in epitope flanking regions demonstrates escape from immune suppression through cleavage abrogation (47) and shows that epitope processing is sensitive to the surrounding sequence, although a simple cleavage signal is not readily discernable. If the tendency to be cleaved at a relevant site is markedly different in a vaccine strain and a challenge strain, the immunological priming induced by the vaccine will be ineffective. This problem is difficult to resolve experimentally, so we addressed it computationally by means of NetChop (48), a neural net prediction program for immunoproteasome cleavage (32, 45).

The median cleavage prediction scores for subtypes B and C were correlated, but although many sites preserved their relative tendency to be cleaved, there were many exceptions, positions with high cleavage prediction scores in subtype B but not in C, or vice versa (32). This suggests that the predilection for cleavage of many sites would be altered in the two subtypes, which could result in diminished breadth of cross-reactive responses. C clade sequences and the M-group consensus gave cleavage prediction patterns that were similar when compared with the median scores for the C clade alignment, and they performed better than sequences from the B clade (32). Scores predicted for the C-subtype consensus cleavage correlated most strongly with the median scores for the subtype C population (32), suggesting that it would be processed at any given position similarly to most of the subtype C strains, and so it may have the greatest potential for eliciting cross-reactive immune responses at the population level. The complete analysis is provided in the supplemental information (32), but in summary, the linear correlation coefficients (r² values) for comparisons of the median C clade cleavage scores to vaccine candidate strains are as follows for positions in the Envelope protein: B clade, 0.65; the M-group consensus, 0.79; the M-group ancestor, 0.80; specific sequences from subtype C isolates, 0.79 to 0.81; the subtype C ancestor, 0.88; and the subtype C consensus, 0.92.

Consensus or reconstructed ancestor? One might assume that an ancestral sequence would resemble more closely a real viral protein than a consensus. It is statistically extremely unlikely, however, that an ancestor corresponds to an ancestral sequence of a clade as complex and diverse as an HIV-1 subtype. Furthermore, reconstruction greatly depends on assumptions inherent in building maximum likelihood trees; for example, if positions are not evolving independently or there are undetected recombination events, the ancestral reconstruction would be influenced and incorrect. Thus, it is highly improbable that an ancestor of a subtype ever existed precisely as reconstructed. Ancestor and consensus sequences are subject to different sampling biases and will change from year to year as sequences accrue. An ancestor is influenced by sequences external to the subtype of interest and will tend to be slightly more distant from available sequences within a subtype than a consensus sequence (Table 2), as well as slightly closer to sequences of other subtypes. The inclusion of a new outlier that branches near the basal node of a subtype could have a strong influence on the ancestral node (see, for example, Fig. 1, where the ancestral and consensus sequences are separated mainly because of one single outlier virus from South Africa), but as a single sequence it would have little bearing on the consensus. In contrast, a consensus will be influenced by the sampling of sequences from within subclades. For example, if many sequences were obtained from the Indian subclade during the next year, the next C consensus would be more like the Indian subset, but the shift in sampling would have less impact on the subtype C ancestor, unless the new sequences substantially altered the evolutionary model.

It is possible that a consensus sequence based on contemporary isolates may be more likely to reflect escape variants relevant to the host population than an ancestral sequence. For example, a CTL escape mutant in an epitope presented by a human leukocyte antigen molecule common in a certain population may be selected for and may be more likely to be represented in the consensus sequence than a reconstructed ancestor sequence. If most viruses in the circulating population had already lost the original epitope because of immune escape, and if the epitope elicited a dominant response upon vaccination with a strain that carried it, then the consensus sequence would have an advantage. On the other hand, if the wild-type form of the epitope was still circulating, even infrequently, and if the epitope was particularly potent, there might be an advantage in using the ancestor. In the end, both concepts need to be tested experimentally, both in terms of B cell and T cell responses.

Applying Evolutionary Principles to Vaccine Strain Selection

Most vaccines are intended to elicit polyclonal responses to multiple epitopes, so even if they differ in some antigenic domains from a given virus, in others they may be cross-reactive. Selecting a clade-appropriate vaccine for a regional trial would tend to increase the number of potentially cross-reactive epitopes by increasing the level of similarity between the vaccine and the population, and the use of consensus and ancestors would enhance the cross-reactive potential. The difference in selection pressure on B and C clade envelopes is indicative of lineage-specific antigenicity, further supporting the use of subtype-appropriate vaccines to maximize the probability that the vaccine elicits immune responses to domains that are antigenic in the circulating viruses.

We could see no compelling advantage in further subdividing the C clade by country of origin, although this is often a consideration for vaccine design (8). Our analysis supports the recommendations of the international meeting on candidate vaccines for the developing world sponsored by the Vaccine Research Center of the United States, National Institute of Allergy and Infectious Diseases (51), indicating that, although there may be advantages to a subtype-specific vaccine, a promising subtype-specific vaccine candidate could be used in many different geographic locations without compromising the potential for success. This does not mean that there would never be an advantage in tailoring a vaccine further by selecting a sequence from an interior subclade within a subtype. For example, there might be an advantage in using an Asian, not African, CRF01 in Thailand, or an Indian C clade sequence in India. But within-clade differences tend to be subtle and represent far fewer amino acid changes than between-subtype differences.

In regions where an epidemic is dominated either by a particular subtype or CRF, it makes sense to use that dominant lineage for a vaccine and to consider the use of a consensus or ancestor. Although we cannot know if even the use of central sequences will be enough to contend with HIV diversity, this kind of strategy can potentially enhance the cross-reactivity and breadth of a vaccine response relative to any single strain. In regions where two or three subtypes and multiple recombinants are cocirculating, to include each of the prevalent subtypes could improve the potential coverage not only of those subtypes, but of the variety of recombinant forms that stem from them (52). Finally, nations with very diverse viral populations, like the DRC, might be best served by developing polyvalent vaccines including a spectrum of natural forms combined with an M-group consensus. An M-group consensus or ancestor is central not only to the major subtypes, but to recombinant forms involving the subtypes. Even if a single subtype predominates in a country, combining an M-group consensus with a regionally dominant subtype might be advantageous in an urban context where people of many nationalities mingle.

REFERENCES AND NOTES

Abstract of this Article Reprint (PDF) Version of this Article dEbates: Submit a response to this article Related articles in Science Similar articles found in: SCIENCE Online Supplemental Data Search Medline for articles by: Gaschen, B. || Korber, B. Alert me when: new articles cite this article Download to Citation Manager     Collections under which this article appears: Medicine/Diseases

Related articles in Science:

THERAPIES:
Confronting the Limits of Success

Jon Cohen
Science 2002 296: 2320-2324. (in News Focus) [Summary] [Full Text]