Recognition of Variant HIV-1 Epitopes from Diverse Viral Subtypes by Vaccine-Induced CTL¹

The HIV-1 genome is characterized by sequence variation that is caused by mutations or viral recombination events within gene products (1). When this variation occurs in regions encoding epitopes recognized by CTL and Th lymphocytes (HTL),⁴ it provides a mechanism of escape from immune system control (2, 3, 4). Viral escape from HIV-1-specific CTL is strongly implied by data obtained from HIV-1-infected individuals that transition from acute to chronic infection (5, 6), individuals that lose control of viral replication and subsequently progress to AIDS (3, 7, 8), and in well-studied cases of mother-to-child transmission (9). Viral escape from CTL-mediated control of SIV is also attributed to mutations, primarily those that alter the amino acids used as primary anchor residues for dominant CTL epitopes (10, 11). Thus, HIV-1 variation represents a challenge to immune system-based control of viral replication within infected individuals and a significant obstacle for therapeutic and prophylactic vaccine development (1, 12, 13, 14).

Vaccine strategies developed to address HIV-1 variation include the use of subtype-based ancestral or consensus sequences (13, 15, 16) and the use of highly conserved regions or epitopes (17). The logic behind this latter approach is that these regions of the HIV-1 genome are conserved because mutations would negatively impact gene function and general viral fitness. Analyses of HIV-1 sequences demonstrate that CTL epitopes are concentrated in conserved regions, whereas regions devoid of CTL epitopes are the most variable (18). Additionally, CTL recognition across HIV-1 subtypes was reported in the setting of natural infection and vaccination (19, 20, 21, 22, 23, 24, 25), and thus, there is little doubt that conserved CTL epitopes exist and can be identified. Recognition of conserved epitopes derived from diverse HIV-1 subtypes by CD8⁺ T lymphocytes conceptually supports the development of epitope-based vaccines. Unfortunately, greater than 40% of sequenced viral isolates contain CTL epitopes from highly conserved regions without amino acid sequence identity to the most prevalent circulating HIV-1 subtypes.

TCR are known to be somewhat flexible, or promiscuous, with respect to recognition of epitope peptides bound to HLA molecules. The specificity of T lymphocyte recognition of epitope peptides bound to MHC molecules is determined by the V regions of the αβ subunits of TCR (26, 27, 28). The contact points for the TCR with CTL epitopes bound to MHC class I molecules include the α1 and α2 domains of the MHC molecule; up to 65% of the TCR contact surface is contributed by the MHC molecule (28, 29, 30). However, TCR interaction is functionally dependent on the molecular structures contributed by the epitope peptide on the surface of the MHC molecule, and there exists a hierarchy for amino acids within an individual epitope, referred to as primary and secondary TCR contact residues (31). TCR binding to variant epitopes is predicted to be relatively unaffected by amino acid substitutions at noncontact residues or for epitopes with conservative amino acid substitutions at TCR contact residues, provided binding to MHC molecules is not negatively impacted. Promiscuity of TCR binding for epitopes may be extremely high; estimates of up to 10⁶ different peptides capable of binding a single TCR have been published (32, 33, 34, 35). Thus, the TCR appears to have evolved to allow promiscuous recognition of epitope-peptide bound to MHC molecules (34).

Within the HIV-1 field, the flexibility of epitope recognition by TCR was demonstrated as CTL recognition of related, but slightly variable, epitopes by cloned CTL lines, which were developed using CTL produced following natural infection (4, 36, 37, 38). Similar flexibility of CTL epitope recognition was demonstrated using rhesus macaques and natural infection with SIV or immunization (39, 40). The ability of CTL to recognize variant epitopes most likely contributes to immune system control of acute infection, as it provides a means to directly confront a significant portion of the sequence variation and thus control viral escape. Our working hypothesis is that vaccines can be designed using epitopes that induce CTL capable of recognizing large numbers of epitope variants. In this study, we present data on CTL recognition of five sets of variant epitopes and a method for identifying epitopes capable of inducing broadly specific CTL.

Peptides were synthesized using an Applied Biosystems (Foster City, CA) 430A peptide synthesizer and FMOC chemistry. Peptide purity was determined by analytical reverse-phase HPLC, and was routinely >95%.

The EP HIV-1090 DNA vaccine encodes 21 CTL epitopes, restricted to the HLA-A2, A3, and B7 supertypes (17). The synthetic gene was constructed using overlapping oligonucleotides in a PCR-based synthesis (41). EP HIV-1090 DNA was produced by growth in Escherichia coli in Terrific Broth (Difco, Sparks, MD) with kanamycin (25 μg/ml) and purified using Qiagen MegaPrep columns (Qiagen, Valencia, CA).

Published full-length and near full-length sequences for a set of 167 HIV-1 isolates in the HIV Sequence Database (http://www.hiv.lanl.gov/content/index) were available at the time of our initial analysis. These viral sequences included isolates from the following HIV-1 subtypes, groups, and circulating recombinant forms: 12 A, 22 B, 62 C, 4 D, 5 F, 3 G, 3 H, 2 J, 2 K, 1 U, 46 circulating recombinant forms, 1 N-group, and 4 O-group. Fifty-one of 62 HIV-1C sequences used represented viral isolates collected from HIV-1⁺ patients throughout the South African country of Botswana (15, 42, 43). Viral sequences were analyzed using motif-based search software to identify putative CTL epitopes analogous to those encoded in the EP HIV-1090 DNA vaccine (17). Synthetic peptides representing the putative epitopes were synthesized, and their binding to solubilized HLA-A*0201 and HLA-A*1101 molecules was measured (44, 45).

HLA-A*0201/K^b and HLA-A*1101/K^b transgenic mice were used in these studies as representative for the HLA-A2 and HLA-A3 HLA supertypes, respectively (46, 47). Target cells, or APC, for the CTL assays were generated by the transfection of Jurkat (HLA-A*0201/K^b) or 721.221 (HLA-A*1101/K^b) cells with same genes used to produce the HLA transgenic mice (47, 48).

Groups of six to nine HLA transgenic mice were injected with 50 μl of 10 μM cardiotoxin (Sigma-Aldrich, St. Louis, MO) in the anterior tibialis muscles of both legs, and 3–4 days later, 50 μl of 1 mg/ml EP HIV-1090 DNA vaccine was administered to the same sites. When synthetic peptides representing CTL epitopes were used, the HLA transgenic mice were injected with 50–100 μg of the selected peptide coemulsified in IFA (Sigma-Aldrich) with 140 μg of a suitable HTL peptide. Eleven to 14 days after a single immunization, mice were sacrificed and single-cell suspensions of splenocytes were prepared and pooled to measure T lymphocyte recognition of epitopes. All experiments involving HLA transgenic mice were repeated using additional mice, and therefore, the data reported were derived from a total of 12–18 animals.

Splenocytes were stimulated in vitro with synthetic peptides, representing variant epitopes, to induce recall CTL responses that were measured as a function of IFN-γ production using the in situ ELISA (49). Specifically, triplicate splenocyte cultures (2.5 × 10⁷ cells/flask) were incubated in upright 25-cm² flasks in RPMI 1640 medium with HEPES (Invitrogen Life Technologies, Grand Island, NY) supplemented with 10% FBS (Omega Scientific, Tarzana, CA), 4 mM l-glutamine, 50 μM 2-ME, 0.5 mM sodium pyruvate, 100 μg/ml streptomycin, and 100 U/ml penicillin, and containing 1 μg/ml peptide and 10⁷ irradiated spleen-derived feeder cells, previously activated for 3 days with LPS (25 μg/ml) and dextran sulfate (7 μg/ml). After 6 days in culture, cells were recovered, washed once, and tested for IFN-γ production.

To measure IFN-γ, 96-well ELISA plates (Costar, Cambridge, MA) were coated with 100 ng/well rat mAb specific for murine IFN-γ (clone R4-6A2; BD Pharmingen, San Diego, CA) in 0.1 M NaHCO₃ buffer, pH 8.2, overnight at 4°C. Before use, plates were blocked for 2 h at room temperature with 10% FBS in PBS. The activated spleen cells were tested using a serial dilution scheme starting with 4 × 15⁵ cells/well and 6 × 4-fold dilution steps; the final concentrations of cells tested were approximately 1 × 10⁵, 2.5 × 10⁴, 6.25 × 10³, 1.6 × 10³, and 4 × 10²/well. The activated spleen cells were cultured for 20 h with peptide epitopes and 10⁵ Jurkat A2.1/K^b cells, as APC, at 37°C with 5% CO₂. The following day, the cells were washed, and the amount of IFN-γ secreted into the wells and bound by the Ab was detected using a sandwich ELISA format. The detection rat mAb is also specific for murine IFN-γ and is biotinylated (clone XMG1.2; BD Pharmingen). HRP coupled to streptavidin (Zymed Laboratories, San Francisco, CA) and 3,3′,5,5′tetramethylbenzidine and H₂O₂ (ImmunoPure TMB Substrate Kit; Pierce, Rockland, IL) were used, according to the manufacturer’s directions. This substrate requires that absorbance be read at 450 nm; we used a Labsystems Multiskan RC ELISA plate reader (Labsystems, Helsinki, Finland).

The in situ IFN-γ ELISA data were evaluated using secretory units (SU), which are based on the number of cells that secrete 100 pg of IFN-γ in response to a particular peptide, corrected for the background amount of IFN-γ in the absence of peptide (49). To determine whether a culture was positive, splenocytes from unimmunized mice were used in the assays and stimulated with the epitope peptides. The significance values for CTL responses to individual epitopes were established using the resulting responses, calculated as a one-tailed t test (p ≤ 0.05) or by adding 2 SDs to the mean SU. Responses obtained using splenocytes from immunized HLA transgenic mice were classified as positive if the mean SU values from triplicate cultures were greater than the significance value calculated for the same epitopes using splenocytes from unimmunized mice.

Standard flow cytometry using mAbs conjugated with FITC (BD Pharmingen) and specific for Vβ2 (B20.6), Vβ3 (KJ25), Vβ4 (KT4), Vβ5 (MR9-4), Vβ6 (RR4-7), Vβ7 (TR310), Vβ8 (F23.1), Vβ9 (MR10-2), Vβ10 (B21.5), Vβ11 (RR3-15), Vβ12 (MR11-1), Vβ13 (MR12-3), and Vβ14 (14-2), and pan-TCR (H57-597) in conjunction with a PE-conjugated monoclonal specific for mouse CD8a was used to measure Vβ expression on murine CD8⁺ splenocytes following restimulation with peptides.

Study subjects were selected from a cohort of individuals followed in the Adult Infectious Diseases Group Practice at the University of Colorado Health Sciences Center. The study was approved by the University of Colorado Health Sciences Center Institutional Review Board, and all study subjects participated voluntarily and gave informed consent. HLA typing was performed at the University of Colorado Health Sciences Center Clinical Immunology HLA Laboratory. Subjects were receiving combination antiretroviral therapy with plasma HIV-1 RNA levels of 400-1000 HIV-1 RNA copies/ml. PBMC were obtained from the heparinized blood of each study subject by gradient density centrifugation and were immediately cryopreserved. Three HIV-1-infected, HLA-A*0201/A*0301 subjects were included in this study.

Responses to CTL epitopes were evaluated using an IFN-γ ELISPOT assay (50). Briefly, membrane-based 96-well plates (Millipore, Bedford, MA) were coated overnight at 4°C with the murine mAb specific for human IFN-γ (clone 1-D1k; Mabtech, Cincinnati, OH) at the concentration of 5 μg/ml. After washing with PBS, RPMI 1640 + 10% heat-inactivated human AB serum (Omega Scientific) was added to each well and incubated at 37°C for at least 1 h to block membranes. The CTL epitope peptides were diluted in AIM-V medium (Invitrogen Life Technologies) and added to triplicate wells in a volume of 100 μl at the final concentration of 10 μg/ml. Cryopreserved PBMC were thawed, resuspended in AIM-V at a concentration of 1 × 10⁶ PBMC/ml, and dispensed in 100 μl vol into test wells. The assay plates were incubated at 37°C for 40 h, after which they were washed with PBS + 0.05% Tween 20. To each well, 100 μl of biotinylated mAb specific for human IFN-γ (clone 7-B6-1; Mabtech) at the concentration of 2 μg/ml was added, and plates were incubated at 37°C for 2 h. The plates were again washed, avidin-peroxidase complex (Vectastain Elite kit) was added to each well, and the plates were incubated at room temperature for 1 h. Finally, the plates were developed using 3-amino-9-ethyl-carbazole (Sigma-Aldrich), washed, and dried. Spots are counted using the Zeiss KS ELISPOT reader (Carl Zeiss MicroImaging, Thornwood, NY).

PBMC were obtained by leukapheresis from healthy male and female donors and used as the source of CD8⁺ T lymphocytes. The donors were screened for common infectious diseases and HLA typed, first using serological testing (One Lambda, Canoga Park, CA), followed by molecular subtyping (Forensic Analytical Laboratory, Hayward, CA).

The PBMC were purified using standard Ficoll-Paque (Amersham Biosciences, Uppsala, Sweden) density gradient sedimentation and cryopreserved at 50 × 10⁶ cells/ml. Dendritic cells (DC), derived from monocytes, were used as the APC. Monocytes were obtained from PBMC by adherence by plating 10 × 10⁶ PBMC in 3 ml of complete medium (RPMI 1640 + 5% human AB serum, nonessential amino acids, sodium pyruvate, l-glutamine, and antibiotics) in each well of a six-well plate. After 2 h at 37°C, the nonadherent cells were removed, and 3 ml of complete medium containing 50 ng/ml GM-CSF and 1000 U/ml IL-4 (R&D Systems, Minneapolis, MN) was added. On day 7, the DC were collected, washed, and pulsed with 40 μg/ml peptide at a cell concentration of 1–2 × 10⁶/ml in the presence of 3 μg/ml human β₂-microglobulin for 4 h at 20°C. These APC were irradiated (42 Gy) before use.

The CD8⁺ T lymphocytes were isolated by positive selection with immunomagnetic beads (MACS; Miltenyi Biotec, Auburn, CA), according to the manufacturer’s instructions. To induce the expansion of precursor CTL to mature cells, 0.25 ml of medium containing 2 × 10⁶ CD8⁺ T lymphocytes/ml was cocultured with 0.25 ml of cytokine-generated DC, a total of 2.5 × 10⁴ DC/well, in the presence of 10 ng/ml human rIL-7 (R&D Systems). Human rIL-10 was added 24 h later to a final concentration of ng/ml, and human rIL-2 was added on day 2 at the final concentration of 10 IU/ml. Cultures were restimulated on days 7 and 14 using adherent monocytes, pulsed with 10 μg/ml peptide in the presence of 3 μg/ml β₂-microglobulin in RPMI 1640 + 5% human AB serum for 2 h at 37°C. Cultures were again supplemented with human rIL-10 (10 ng/ml) 24 h later, followed by the addition of human rIL-2 (50 IU/ml) on days 2 and 3.

Seven days after the day 14 restimulation step, CTL activity was assessed as a function of IFN-γ secretion using the in situ ELISA. The assay is similar to that used to measure responses in HLA transgenic mice modified using monoclonal specific for human IFN-γ (BD Pharmingen). CD8⁺ T lymphocytes were cultured for 48 h at concentrations of 1–10 × 10⁴/well with peptide epitopes, using a concentration titration starting at 5 μg/ml and 1–10 × 10⁴ HLA-transfected .221 cells as the APC at 37°C with 5% CO₂. The background levels of response were determined using an irrelevant peptide in the assay. After 48 h, the wells were washed to remove cells and the amount of IFN-γ secreted into the wells using a sandwich ELISA format and human rIFN-γ to establish a standard curve. A culture was considered positive if the measured cytokine level was twice the background level, determined using an irrelevant peptide and the same cell line.

A computer-based algorithm to predict which epitopes should induce CTL capable of recognizing the largest number of variant epitopes was developed. Known HLA epitope-binding sequence motifs for the HLA-A2, HLA-A3, and HLA-B7 supertypes were used, and variation of amino acid sequences that were consistent with the motifs was allowed. All conservative amino acid substitutions at non-HLA anchor positions were allowed regardless of their number in an individual epitope. Conservative substitutions were defined based on a previously developed method for determining the degree of similarity between amino acids that incorporates hydrophobicity, side chain volume, and the Dayhoff PAM250 score (51). Not all amino acids can substitute in an equally reciprocal manner. For example, substitutions of G for T, D, or P were considered to be conservative changes, whereas the substitution of G by the others was not. Similarly, S and T can substitute for N, but the reverse substitution, N for either S or T, is considered to be only semiconservative, based on the Dayhoff PAM250 score (51). Using the established definitions for conservative, semiconservative, and nonconservative amino acid substitutions, each variant epitope was compared with its parent epitope, and the likelihood of recognition by CTL specific for the parent epitope was predicted as either positive or negative. The predictions of CTL recognition for variant epitopes were compared with the experimental results produced using the HLA transgenic mice and human PBMC, and subsequently used to predict CTL recognition of variant forms of the 21 CTL epitopes encoded in the EP HIV-1090 DNA vaccine (17).

Five HLA-A2- and HLA-A3/A11-restricted CTL epitopes encoded in an experimental plasmid DNA vaccine (EP HIV-1090) were selected for study. The parent epitope was defined as the amino acid sequence encoded in the vaccine, and variant epitopes, defined by the presence of an HLA-binding motif in the analogous position, were identified in the 167 HIV-1 sequences (Fig. 1). The degree of conservation for the five parent epitopes in these sequences varied significantly. The parent form of the Env 134 epitope (KLTPLCVTL) was the most highly conserved, present in 80% of the sequences (Fig. 1,A), whereas the Pol 98 epitope (VTIKIGGQLK) was the least conserved and present in only 11% of the sequences (Fig. 1 D). The Env 47 (VTVYYGVPVWK), Vpr 62 (RILQQLLFI), and Gag 386 (VLAEAMSQV) epitopes were present in 59, 51, and 32%, respectively.

A total of 19 variant forms of the Env 134 epitope was identified in 32 HIV-1 sequences (Fig. 1,A). Most variant epitopes contained a single amino acid substitution. The epitope was not identified a single viral sequence, indicating an amino acid change that disrupted the epitope-binding motif. Ten of the 19 variant forms of the epitope were recognized by CTL produced by immunization of HLA-A*0201/K^b transgenic mice with the DNA vaccine; a response was considered positive if the SU value was greater than the significance value calculated for that epitope. Thus, 157 of the 167 HIV-1 virus isolates represented in the database could be recognized by CTL specific for the parent Env 134 epitope. Similarly, the Gag 386 (Fig. 1,B), Vpr 62 (Fig. 1,C), and Env 47 (Fig. 1 E) epitopes, in the parent or variant forms, were identified in a high percentage of viral isolates, 97% (162 of 167), 96% (160 of 167), and 82% (138 of 167), respectively. CTL specific for the parent forms of these epitopes typically recognized greater than 95% of the variant epitopes examined.

The Pol 98 epitope represented a significantly different case because parent and variant epitopes were identified in only 71% (118 of 167) of the HIV-1 sequences (Fig. 1 D). Total loss of the epitope in a viral sequence reflected multiple changes that disrupted the HLA-A*0301/*1101-binding motif. The variant epitopes from 100 viral sequences were represented in 30 sequences with up to four amino acid changes. Fifteen of the 17 (88%) variant epitopes with one or two amino acid changes were recognized by CTL produced in HLA-A*1101/K^b transgenic mice. Although 7 of 13 (54%) epitopes with three or four amino acid changes were recognized to measurable levels, the levels of IFN-γ were low, ≤10% of that produced to the parent epitope for five of these. Despite the highly variable nature of this epitope, CTL specific to the parent epitope recognized the variant epitopes in 50% (84 of 167) of the viruses represented in the database.

Epitopes with only a single amino acid change were likely to be recognized by CTL specific to the parent epitope, and recognition rates decreased as the numbers of substitutions increased; this is best demonstrated as shown with the Pol 98 epitope (Fig. 1 D). Single changes within anchor positions were generally tolerated provided the change did not abolish epitope binding to HLA molecules. Outside of the primary anchor positions, conservative and semiconservative amino acid substitutions were most commonly tolerated. For example, the K at position 1 (P1) of Env 134 could be substituted by Q or R, and the P at P4 could be substituted by F or L without significant loss of CTL recognition.

Multiple conservative changes were more likely to be deleterious than individual changes involving the same substitutions and positions. For example, in the Pol 98 epitope, substitution of K by R at P4 and I by V at P5 almost totally abolished CTL recognition when the substitutions occurred together, but not as individual substitutions. Similarly, for the Gag 386 epitope, substitution of the Q at P8 by R, K, or H reduced recognition by CTL when the substitution was concurrent with the change of V to A at P9, whereas these changes were not deleterious when they occurred individually. The HLA-binding affinity of these variant epitopes was not significantly altered by these double substitutions, which indicates the TCR contact points were probably altered.

Recognition of variant epitopes could be mediated by different subpopulations of CTL, or individual CTL may recognize variants through promiscuous TCR binding of epitopes.

To distinguish between these options, Env 134- or Gag 386-specific CTL lines were generated by multiple in vitro restimulations of splenocytes from HLA transgenic mice immunized with the EP HIV-1090 DNA vaccine. These lines were characterized for changes in Vβ TCR expression and tested for T cell avidity on both the parent and variant epitopes. The magnitude of responses increased for the parent and most variant epitopes following five restimulations, indicating enrichment in the proportion of responding cells (Table I). The Vβ TCR usage for the Gag 386-responding cells narrowed to Vβ6 after the multiple restimulations, indicating clonal or oligoclonal expansion of CTL (data not shown). The Vβ TCR usage for the Env 134-responding cells could not be determined, but all Vβ isoforms detected before restimulation decreased to background levels following restimulation, consistent with clonal or oligoclonal expansion of CTL expressing a TCR not recognized using available mAbs.

Recognition of variant epitopes by these two CTL lines was assessed using a peptide titration assay format as a relative measure of TCR affinity for the epitopes (Fig. 2). Typically, only the parental forms of the epitopes were recognized at maximal levels using lower concentrations of peptides. We interpret these data to indicate that the highest affinity TCR binding was to the parent epitopes and that recognition of variant epitopes occurred through promiscuous and lower affinity TCR binding.

The structure and function of the murine and human TCR are analogous, and, as such, we predicted that human CTL should be capable of recognizing multiple variant epitopes. To assess this directly, we identified three HLA-A2/A3 volunteers who were chronically infected with HIV-1 and had CTL specific for epitopes Pol 98 and Env 47; these epitopes were selected for study because of their different levels of sequence variation. The subtype origin of the existing HIV-1 infection in these volunteers was not known, but was assumed to be subtype B. The epitope specificity of responding cells was determined using individual peptides and an ELISPOT assay. Recognition of many of the variant epitopes was readily demonstrated, and the patterns of response were similar to those obtained using the HLA transgenic mice; data obtained from a single volunteer that responded to two epitopes are shown (Fig. 3). Specifically, for the Pol 98 epitope (Fig. 3,A), variants with single amino acid changes were recognized at higher rates than those with multiple substitutions. The majority of the Env 47 (Fig. 3 B) variants were recognized.

CTL lines specific for Env 134 and Vpr 62 were produced in vitro, using PBMC from HIV-1-uninfected donors, and tested for their ability to recognize variant forms of these same epitopes. Similar to the CTL produced in response to HIV-1 infection, recognition of variant epitopes was readily demonstrated, and the patterns of response were similar to those of the HLA transgenic mice (Fig. 4).

An algorithm was developed to predict recognition of variant epitopes by CTL specific to epitopes in EP HIV-1090. The five epitopes for which data was generated in the HLA transgenic mice (Fig. 1) were evaluated first. For Env 134, we predicted that 13 of the variant epitopes should be recognized by specific CTL, while 6 should not be recognized. The immunological data and the predictions were matched for 14 of the variant epitopes and incorrect for 5; we predicted negative results for 2 epitopes that were recognized and positive results for 3 epitopes that were not recognized. For the 101 total variant epitopes in our data set, immune recognition for 68 was correctly predicted (67%). The discordant data were evenly split between epitopes incorrectly predicted negative (15) or positive (18). Because the prediction program treated all substitutions independently and did not take into account the number of substitutions, we hypothesized that prediction of single substitutions would be more accurate. Indeed, CTL recognition of 38 of 47 single substitution epitope variants (80%) was correctly predicted.

We expanded our analysis to investigate the impact of the one-way preferred substitution patterns for amino acids using the HLA-A2-restricted Gag 271 epitope (Fig. 5). The variant common to subtype B sequences is MTNNPPIPV, whereas the sequence MTSNPPIPV is most common in subtype C. This epitope was selected for study because the substitution of serine (S) at P3 by asparagine (N), glycine (G), or alanine (A) was considered to be conservative, but substitution of the N by these other amino acids was considered only semiconservative. We therefore predicted that MTSNPPIPV would induce CTL capable of recognizing the other variants. This prediction was tested using HLA-A2 transgenic mice immunized with synthetic peptides representing the subtype B and C forms of the epitopes. As predicted, the MTSNPPIPV form of the epitope induced CTL capable of recognizing 11 of the 14 epitope variants tested, representing 152 of 167 HIV isolates in the database. In contrast, MTNNPPIPV induced CTL capable of recognizing only three of the variant forms of the epitope, representing 39 of 167 HIV-1 isolates. The CTL recognition for the variant forms of this epitope was predicted correctly at a rate of ∼85%.

The ability to predict epitopes capable of inducing broadly specific CTL has important implications for vaccine development because it provides a means to address HIV-1 variation. We assessed the potential level of recognition by CTL of variant epitopes in the multiple subtypes represented in our database, assuming CTL were induced using the parent form encoded in the EP HIV-1090 DNA vaccine. The results are very encouraging; as predicted, CTL recognition of variant epitopes is significantly greater than the level of amino acid conservation for all epitopes studied (Table II).

The TCR is known to be promiscuous with respect to its ability to bind to epitopes with different amino acid sequences (35, 52). In fact, promiscuity of TCR binding for epitopes may be extremely high; estimates of up to 10⁶ different peptides capable of binding a single TCR have been published (32, 33, 34, 35). For HIV-1, this promiscuity was demonstrated as CTL recognition of related epitopes by cloned CD8⁺ T lymphocytes from an HIV-1-infected individual (36, 37, 38). Similar promiscuity of CTL epitope recognition was demonstrated using rhesus macaques infected with SIV or following immunization (39, 40). In this study, we provide data that promiscuous recognition of HIV-1 epitopes by CTL is a common occurrence that can be partially predicted and potentially applied to the design of vaccines.

The epitopes used for this study are encoded by an experimental DNA vaccine, designated EP HIV-1090 (17). They were identified in HIV-1 sequences using an approach based on epitope-HLA-binding motif scanning combined with a selection bias toward conserved amino acid sequences. Recognition of these epitopes by CTL from HIV-1 infected was used to document their generation through natural processing as an occurrence of infection. Although these epitopes are likely to be significantly less variable than those identified using an individual selection criterion, such as simple epitope prediction algorithms or recognition by CD8⁺ T lymphocytes from HIV-1-infected patients, numerous variant forms of the epitopes were identified. This variation provided a means to evaluate the breadth of CTL recognition for naturally occurring HIV-1 epitopes.

The tolerated amino acid changes were generally conservative in nature and included those that did not disrupt the HLA-binding motif. These findings are consistent with tolerated amino acid substitution patterns that we defined as part of previous efforts to identify alterations within tumor Ag epitopes that increased their immunogenicity (51). Increased immunogenicity of modified HLA-A2-restricted CTL epitopes from HIV-1 reverse transcriptase was also recently described (53). In this study, the increased immunogenicity of the modified forms of the epitope was readily demonstrated using standard in vitro assays, while the benefit of the resulting CTL responses was also demonstrated in vivo, using an infectious challenge model based on recombinant vaccinia virus encoding HIV-1 reverse transcriptase. Increased recognition of variant CTL epitopes was not a common occurrence in our study, when measured as a function of magnitude of response. However, the one-way preferred substitution patterns for amino acids did provide a means to identify epitopes likely to induce CTL responses with the greatest breadth. For the example shown, Gag 271, use of the epitope with S at P3 rather than N resulted in CTL that recognized more than twice the naturally occurring variants of this epitope in our database. It is interesting to note the extent to which a single amino acid change altered the immunogenicity of this epitope and the recognition patterns of the CTL specific for these epitopes. Although not directly tested, we believe it feasible to use site-directed mutagenesis to alter epitopes within current generation DNA or viral vectored vaccines to increase their utility for use in certain viral subtypes or to address continuing viral evolution.

Simple TCR-based recognition of variant epitopes by T lymphocytes does not always lead to appropriate cellular activation and effector functions. Rather, the TCR often interacts with variant epitopes with reduced affinity, and this can result in decreased cellular activation, mediated through the TCR and measured as reduced levels of phosphorylation of the TCR ζ-chain (54). Reduced levels of CD4⁺ or CD8⁺ T lymphocyte activation can result in reduced cytokine production or loss of proliferation and cytolytic capabilities (31, 55, 56, 57, 58, 59). In our studies, we demonstrated that maximal recognition of the variant CTL epitopes often required significantly higher concentrations of peptides, which most likely indicates a decreased affinity of TCR binding to these variant forms.

Variant forms of epitopes that interact with TCR in a suboptimal manner can sometimes serve as antagonists and actually suppress T lymphocyte responses, including the ability of the T lymphocyte to respond to the immunizing epitope (31, 60, 61, 62). Natural variants of both HTL and CTL epitopes from viral pathogens, including HIV-1, with antagonistic properties have been identified, demonstrating the biological significance of altered epitope recognition (63, 64, 65, 66). Reported antagonism of CD4⁺ T lymphocyte activity in a vaccine study is of particular concern because the antagonism was identified in a patient that became infected following vaccination; this phenomenon may therefore represent a potential mechanism for vaccine failure.

In our studies, splenocytes from immunized HLA transgenic mice were restimulated in vitro with the parent or variant peptides before measuring IFN-γ production in response to both the parent and variant epitopes. Antagonism may explain the failure of a small number of epitopes with single conserved amino acid substitutions, which did not significantly reduce binding affinity of the peptides to HLA, to restimulate CTL in our assay system. For example, two variants of the Env 134 epitope, parent epitope (KLTPLCVTL) and variant epitopes (KLTPFCVTL and KLTPLCITL), bind to HLA-A2 with high affinity, 1.7 and 87.3 nM, respectively, but failed to restimulate CTL in splenocyte preparations. However, these types of responses were not commonly observed, and simple failure of the CTL to recognize the variant epitopes because of lack of TCR binding may also explain the data. Additional studies will be required to address this question of TCR antagonism for HIV-1 CTL epitopes.

It is widely accepted that the sequence variation of HIV-1 gene products represents one of the most significant obstacles for vaccine development, and we believe the use of multiple epitopes in a single vaccine holds promise for addressing this issue. Vaccines composed of multiple CTL epitopes have evolved over the past decade from simple constructs, for example an experimental vaccine encoding two LCMV CTL epitopes restricted by two different mouse MHC haplotypes (67, 68), to more complex vaccines encoding 10 or more epitopes (69, 70, 71, 72, 73). In studies completed in mice, the utility of epitope-based experimental vaccines for simultaneously inducing responses against numerous epitopes restricted to different MHC allelic products was readily demonstrated, as was protection from experimental infections. Within the HIV-1 vaccine field, the immunogenicity of multiple CTL epitopes delivered using DNA vaccines or viral vectors has been documented using mice and nonhuman primates (17, 74, 75, 76), and phase 1 clinical trials to document the safety of current generation vaccines are ongoing. Thus, multiepitope vaccines are approaching the stage of development in which their efficacy can be tested.

In our present studies, we demonstrated that variation within HIV-1 epitopes does not necessarily abrogate their recognition by CTL specific for related epitopes. In fact, our data indicate that tolerated, or even preferred, amino acid substitutions can be predicted and that epitopes can be selected or modified to increase the breadth of CTL specificity for variant epitopes. These data also demonstrate that individual epitopes cannot be considered as equal or interchangeable with respect to the specificity of CTL they induce, and they provide a basis for identifying the most suitable vaccine epitopes.

We gratefully acknowledge the technical assistance of David Brown, Ajesh Maewal, David Seiber, and Corazon Gajudo, and the participation of human subjects through the University of Colorado Hospital Infectious Diseases Group practice.